y is a local diffeomorphism. Thus we conclude that S(f,
is open (Foias and Temam [1977] give a direct proof of
openness of 0 rather than using the implicit function theorem). One knows, also from elliptic theory that S(f,
an
using a transversality analogue of the Smale-Sard theorem (see Supplement 3.6B), as was pointed out by A. J. Tromba. We leave the precise formulation as a project for the reader.
fl:n
~ SUPPLEMENT 3.6B The Parametric Transversality Theorem
3.6.22 Density of Transversal Intersection
Let P, M, N be Ck manifolds, seN a
submanifold (not necessarily closed) and F: P x M ~ N a Ck map, k ~ 1. Assume (i) M is finite dimensional (dim M
=m)
and that S hasfinite codimension q in N.
(ii) P x M is Lindelof (iii) k > max(O, n--<()
(iv) F Iii S
Then Iii(F, S) := {p
E
pi Fp : M ~ N is transversal to S at all points of S} is residual in P.
The idea is this. Since F Iii s, p-l(S) c P x M is a submanifold. The projection p-l(S) ~ P has the property that a value of then apply Sard's theorem to
7t
7t:
is a regular value iff Fp is transverse to S. We
7t.
A main application is this: consider a family of perturbations f: ]-1, I[ x M
~
N of a
given map fo: M ~ N, where f(O, x) = fo(x). Suppose f Iii S. Then there exist t's arbitrarily close to zero such that ft Iii S; i.e., slight perturbations of fo are transversal to S. For the proof we need two lemmas. 3.6.23 Lemma Let E and F be Banach spaces, dim F = n, prJ: E x F
~
E the projection onto
the first factor, and GeE x F a closed subspace of codimension q. Denote by p the restriction of prJ toG. Then p is a Fredholm operator of index n-q. Proof Let H = G + ({OJ x F) and K = G n ((OJ) x F). Since F is finite dimensional and G is closed, it follows that H is closed in Ex F (see Exercise 2.2M(ii». Moreover, H has finite codimension since it contains the finite codimensional subspace G. Therefore H is split (see Exercise 2.2N) and thus there exists a finite dimensional subspace seE x {OJ such that Ex F == H ® S. Since KeF, choose closed subspaces Go c G and Fo
C
{OJ x F S1\ch that G = Go
® K and (OJ x F = K ® Fo. Thus H = Go ® K ® Fo and Ex F = Go ® K ® Fo ® S. Note that prl I Go ® S : Go ® S ~ E is an isomorphism, K = ker p, and pr1(S) is a finite
Section 3.6 The Sard and Smale Theorems
233
dimensional complement to peG) in F. Thus p is a Fredholm operator and its index equals dim(K) - dimeS) = dim(K
(B
Fo) - dimeS
(B
Fo). Since K
(B
Fo = (D) x F and Fo
(B
S is a
complement to G in Ex F (having therefore dimension q by hypothesis), the index of p equals n-q. T 3.6.24 Lemma In the hypothesis of Theorem 3.6.22, let V = p-\S). Let 7t': P x M ~ P be
the projection onto the first factor and let 7t = 'It I V. Then 7t is a Ck Fredholm map of constant index n - q. Proof By (iv), V is a C k submanifold of P x M so that 7t is a Ck map. The map T(p,m)7t: T (p,m) V
~
T
l
is Fredholm of index n - q by lemma 3.6.23: E is the model of P, F the model
of M, and G the model of V. T We shall prove below that p is a regular value of
Proof of 3.6.22
7t
if and only if
Fp in
s.
If
this is shown, since 7t: V ~ P is a Ck Fredholm map of index n - q by lemma 3.6.24, the codimension of V in Ex F equals the codimension of S in N which is q, k > max (0, n - q), and V is Lindelof as a closed subspace of the LindelOf space P x M, the Smale-Sard theorem 3.6.15 implies that in(F, S) is residual in P. By definition, (iv) is equivalent to the following statement: (a)
for every (p, m) E P x M satisfying F(p, m) E S, T(m,p)F(Tl x TmM) + T F(p,m)S = TF(p,m)N and (T(m,pl)-I(T F(p,m)S) splits in
Tl x TmM .
Since M is finite dimensional, the map m E M 1-+ F(p, m) E N for fixed pEP is transversal to S if and only if (b) for every mE M satisfying F(p,m)E S, TmFp(Tm M )+ TF(p,rn)S=TF(p,rn)S, Since
7t
is a Fredholm map, the kernel of T7t at any point splits being finite dimensional (see
Exercise 2.2N). Therefore p is a regular value of 7t if and only if (c)
for every mE M satisfying F(p, m) E S and every v E
Tl,
there exists u E TrnM
such that T(rn,pl(v, u) E T F(p, rn)S, We prove the equivalence of (b) and (c). First assume (c), take m E M, PEP such that F(p, m) E S and let WE TF(p,rn)S, By (a) there exists v E TpP, u l E TrnM, that T(rn,pl(v, Therefore,
Ul)
Z
E TF(p,rn)S such
+ Z =w. By (c) there exists liz E TrnM such that T(rn,pl(v, u 2 ) E TF(p,rn)S, w
T(rn,p)F(v, u l ) T(rn,pl(O, u l
-
-
T(rn,pl(v, u2 ) + T(rn,pl(v, u 2 ) + Z u 2 ) + T(rn,pl(v, u 2 ) +
Z
T(rn,pl(O, u) + Z' E TrnFp(Trn M) + TF(p,rn)S, Where u = u l - u2 and z' = T(rn,pl(v, liz) + Z E T F(p,rn)S, Thus (b) holds. Conversely, let (b) hold, take PEP, mE M such that F(p, m) E S and let v E TpP. Pick u l E TrnM, E
TmM and
z2
zl
E TF(p,rn)S and define w = T(rn,pl(v, uI) +
E T F(P,m)~.such that w = T rnFp(u 2) +
z2'
zl'
By (b), there exist liz
Subtract these two relations to get
Chapter 3 Manifolds and Vector Burulles
234
i.e., T(m,pl(v, u l -ll:l) = z2 - zi
E
T F(m,p)S and therefore (c) holds. •
There are many other very useful theorems about genericity of transversal intersection. We refer the reader to Golubitsky and Guillemin [1974], Hirsch [1976], and Kahn [1985) for the finite dimensional results and to Abraham [1963b) and Abraham-Robbin [1968) for the infinite dimensional case and the situation when P is a manifold of maps.
Exercises 3.6A Construct a C~ function f:]R ~]R whose set of critical points equals [0, 1). This shows that the set of regular points is not dense in general. 3.6B Construct a
~
function f:]R
~]R
which has each rational number as a critical value.
(Hint: Since Q is countable, write it as a sequence {qo I n = 0, I, ... }. Construct on the
closed interval [n, n+ 1) a
C~
function which is zero ncar n and n+ 1 and equal to qo on
an open interval. Define f to equal fo on [n, n+ 1).) 3.6C Show that if m < n there is no C I map of an open set of ]Rm onto an open set of ]R0. 3.6D A manifold M is called ck-simply connected, if it is connected and if every Ck map f: S I ~ M is Ck-homotopic to a constant, i.e., there exist a Ck-map H: )-£, 1 + £[ X SI ~ M such that for all s E S I, H(O, s) = f(s) and H(l, s) = m o' where mo E M. (i) Show that the sphere So, n ~ 2, is Ck-simply connected for any k ~ 1. (Hint: By Sard, there exists a point x E So \ f(S I). Then use the stereographic projection defined by x.) (ii) Show that So, n ~ 2, is CO-simply connected. (Hint: Approximate the continuous map g: SI ~ So by a CI-map f: SI ~ So. Show that one can choose f to be homotopic to g.) (iii) Show that SI is not simply connected. 3.6E Let M and N be submanifolds of ]R0. Show that the set {x E ]R0 I M intersects N + x transversally} is dense in ]R0. Find an example when it is not open. 3.6F Let f:]R° ~]R be C2 and consider for each a E ]R0 the map fa(x) = f(x) + a • x. Prove that the set {a E ]R0 I the matrix [a2fa(xo)/axiihi) is nonsingular for every critical point Xo of fa} is a dense set in ]R0 which is a countable intersection of open sets. (Hint: Use Supplement 3.68; when is the map (a, x) 1-+ Vf(x) + a transversal to (O}?)
Section 3.6 The Sard and Smale Theorems
235
3.6G Let M be a C2 manifold and f: M -t lR a C2 map. A critical point mo of f is called non-degenerate, if in a local chart (U,
Show that the notion of non-degeneracy is chart independent. Functions all of
(ii)
whose critical poins are nondegenerate are called Morsejunctions . Assume M is a C2 submanifold of lRO and f: M -t lR is a C2 function. For a
E
lRo define fa: M -t lR by fa(x) = f(x) + a • x. Show that the set {a
E
lRo
I
fa is a Morse function} is a dense subset of lRo which is a countable intersection of open sets. Show that if M is compact, this set is open in lRo. (Hint: Show first that if dim M = m and (x I, ... , XO) are the coordinates of a point x E M in lRo, there is a neighborhood of x in lRo such that m of these coordinates define a chart on M. Cover M with countably many such neighborhoods. In such a neighborhood U, consider the function g: U C lRO -t lR defined by g(x) = f(x) + am+lxm+1 + ... + aOxo. Apply Exercise 3.6F to the map fa,(x) = g(x) + alxl + ... + amx m, a' = (a I, ... , am) and look at the set S = {a E lRO I fa is not Morse on U}. Consider S n (lRm x (am+l , ... , an)) and apply Lemma 3.6.2,) (iii) Assume M is a C2-submanifold of lRO. Show that there is a linear map a: lRo -t lR such that a I M is a Morse function. (iv) Show that the "height functions" on So and 'f" are Morse functions.
3.6H Let E and F be Banach spaces. A linear map T: E -t F is called compact if it maps bounded sets into relatively compact sets. (i) Show that a compact map is continuous. (ii) Show that the set K(E, F) of compact linear operators from E to F is a closed subspace of L(E, F). (iii) If G is another Banach space, show that L(F, G) 0 K(E, F) C K(E, G), and that K(E, F) L(G, E) C K(G, F). (iv) Show that if T E K(E, F), then T* E K(F*, E*). 0
3.61 (F. Riesz) Show that if K E K(E, F) where E and F are Banach spaces and a is a scalar (real or complex), then T = Identity + aK is a Fredholm operator. (Hint: It suffices to prove the result for a = -1. Show ker T is a locally compact space by proving that K(D) where D is the open unit ball in ker T. To prove that T(E) is closed and finite dimensional show that dim(E / T(E» = dim(E / T(E»* = dim(ker T*) = dim(ker(Identity K*» < 00 and use Exercise 2.2H.)
= D,
3.6J Show that there exist Fredholm operators of any index. (Hint: Consider the shifts (xl' x2' •.• ) 1-+ (0, ... , 0, xI' x2 ) and (xI' x2 ... ) 1-+ (x o' xo+I ' ... ) in F(lR).)
Chapter 3 Manifolds and Vector Bundles
236
3.6K Show that if T E L(E, F) is a Fredholm operator, then T* E L(F*, E*) is a Fredholm operator and index(T*) = -index(T). 3.6L (i)
Let E, F, G be Banach spaces and T E L(E, F). Assume that there are S, S' E L(F, E) such that SoT - Identity E K(E, E) and To S' - Identity E K(F, F). Show that T is Fredholm. (Hint: Use Exercise 3.61.)
(ii)
Use (i) to prove that T E L(E, F) is Fredholm if and only if there exists an operator S E L(F, E) such that (S
0
T - Identity) and (T 0 S - Identity) have finite dimensional
range. (Hint: If T is Fredholm, write E = ker T EB Fa, F = T(E) EB Fa and show that To = T I Eo : Eo
~
T(E) is a Banach space isomorphism. Define S E L(F, E) by
SI T(E) = TO-I, SIFo=O.) (iii)
Show that if T E L(E, F), K E K(E, F) then T + K is Fredholm.
(iv)
Show that if T E L(E, F), S E L(F, G) are Fredholm, then so is SoT and that index(S
0
T) = index(S) + index(T).
3.6M Let E, F be Banach spaces. (i)
Show that the set Fred/E, F) = {T E L(E, F) I T is Fredholm, index(T) = q} is open in L(E, F). (Hint: Write E = ker T EB Eo, F = T(E) EB Fa and define T: E EB Fa ~ FEB ker T by T(z EB x, y) = (T(x) EB y, z), for x E Eo' z E ker T, y E Fa. Show that T E CiL(E EB Fa, FEB ker T). Define p : L(E EB Fa, F EB ker T) ~ L(E, F) by p(S) = 1t 0 S
0
i, where 1t: FEB ker T
~
F is the projection and i: e E E f--+ (e, 0) E
E EB Fa is the inclusion. Show that p is a continuous linear sUijective map and hence open. Prove p(CiL(E EB Fa, FEB ker T) (ii)
C
Fredq(E, F), p(T) = T.)
Conclude from (i) that the index map from Fred(E, F) to Z is constant on each connected component of Fred(E, F) = {T E L(E, F) I T is Fredholm}. Show that if E = F = ~2(lR) and T(t)(x 1 , x2' ... ) = (0, tx 2, x3 ,
... )
then index(T(t» equals 1, but
dim(ker(T(t) and dim( ~2(lR) / T(t)( ~2(JR») jump at t = O. (iii)
Homotopy invariance of the index Show that if q> : [0, 1]
~
Fred(E, F) is
continuous, then index(q>(O» = index(q>(l». (Hint: Let a = sup{t E [0,1]1 s < t implies index(f(s» = index(f(O»}. By (i) we can find
£
> 0 such that I b - a I < £
implies index(f(b» = index(f(a». Let b = a - £/2 and thus index(f(O» = index(f(b» = index(f(a». Show by contradiction that a = 1.) (iv) (v)
If T E Fred(E, F), K E K(E, F), show that index(T + K) = index(T). (Hint: T +
K(E, F) is connected; use (ii).) The Fredholm alternative Let K E K(E, F) and a ~ O. Show that the equation K(e) = ae has only the trivial solution iff for any vEE, there exists u E E such that K(u) = au + v. (Hint: K - a(ldentity) is injective iff (l/a)K - (Identity) is injective. By (iv) this happens iff (l/a)K - (Indentity) is onto.)
3.6N Using Exercise 3.58, show that the map 1t: SL(JRm,JRk) x IL(JRk, JRn) ~ 8t(m, n; k), where k:5 min(m, n), defined by 1t(A, B) = BoA, is a smooth locally trivial fiber bundle
Section 3.6 The Sard and Smale Theorems
237
with typtical fiber CiL(lltk ). 3.60 (i) Let M and N be smooth finite dimensional manifolds and let f: M ~ N be a C 1 bijective immersion. Show that f is a C 1 diffeomorphism. (Hint: If dim M < dim N, then f(M) has measure zero in N, so f could not be bijective.) (ii) Formulate an infinite dimensional version of (i).
Chapter 4 Vector Fields and Dynamical Systems This chapter studies vector fields and the dynamical systems they determine. The ensuing chapters will study the related topics of tensors and differential forms. A basic operation introduced in this chapter is the Lie derivative of a function or a vector field. It is introduced in two different ways, algebraically as a type of directional derivative and dynamically as a rate of change along a flow. The Lie derivative formula asserts the equivalence of these two defmitions. The Lie derivative is a basic operation used extensively in differential geometry, general relativity, Hamiltonian mechanics, and continuum mechanics.
§4.1 Vector Fields and Flows This section introduces vector fields and the flows they determine. This topic puts together and globalizes two basic ideas we learn in undergraduate calculus: the study of vector fields on the one hand and differential equations on the other.
4.1.1 Definition Let M be a manifold. A vector freld on M is a section of the tangent bundle TM of M. The set of all C vector fields on M is denoted by Xr(M) and the C~ vector fields by X~(M) or X(M). Thus a vector field X on a manifold M is a mapping X : M
~
TM such that X(m) E
TmM for all mE M. In other words, a vector field assigns to each point of M a vector based (i.e., bound) at that point.
4.1.2 Example Consider the force field determined by Newton's law of gravitation. Here the manifold is ]R3 minus the origin and the vector field is F(x, y, x) = -
mMG
-3-
r
r,
Section 4.1 Vector Fields and Flows
239
where m is the mass of a test body. M is the mass of the central body. G is the constant of gravitation. r is the vector from the origin to (x. y. z). and r = (x 2 + y2 + z2)1!2; see Fig. 4.1.1. •
\
"-
I
/
\
/
/""
..-
-.
,
"Figure 4.1.1 The study of dynamical systems. also called flows. may be motivated as follows. Consider a physical system that is capable of assuming various "states" described by points in a set S. For example. S might be 1R3 x 1R 3 and a state might be the position and momentum (q. p) of a particle. As time passes. the state evolves. If the state is So E S at time A and this changes to s at a later time t. we set
and call Ft.A. the evolutWn operator; it maps a state at time A to what the state would be at time t; i.e .• after time t - A has elapsed. "Determinism" is expressed by the law Ft.t 0 Ft.A.
= F~.A.
Ft,t
= identity.
sometimes called the Chapman-Kolmogorov law. The evolution laws are called time independent when Ft,A. depends only on t - A; i.e .• F t.A. =FS,1l
if
t-
A = s -11.
Setting Ft = Ft,o' the preceding law becomes the group property.' Ft 0
F~ =
Ft+'t'
Fo = identity.
We call such an F t aflowand Ft A. a time-dependent flow. or as before. an evolution operator. If the system is nonreversible. tha~ is. defined only for t 2': A. we speak of a semi-flow. It is usually not Ft,A. that is given. but rather the laws of motion. In other words. differential equations are given that we must solve to find the flow. These equations of motion have the form
240
Chapter 4 Vector Fields and Dynamical Systems
ds
dt =
Xes),
s(O) = So
where X is a (possibly time-dependent) vector field on S. 4.1.3 Example The motion of a particle of mass m under the influence of the gravitational force field in Example 4.1.2 is determined by Newton's second law:
ir
m - =F;
dt2 i.e., by the ordinary differentatial equations d 2x
mMGx
dt2
r3
d 2y
mMGy
dt2
r3
d2z
mMGz
dt2
r3
m--=----
m--=----
m--=----
Letting q = (x, y, z) denote the position and p = m(dr/dt) the momentum, these equations become dq
dt=
£.. m'
dp
dt =
F(q)
The phase space here is the manifold (JR.3\{O)) x JR.3, i.e., the cotangent manifold of JR.3\{Oj. The right-hand side of the preceding equations define a vector field on this six-dimensional manifold by X(q, p) = «q, p), (p 1m, F(q))). In courses on mechanics or differential equations, it is shown how to integrate these equations explicitly, producing trajectories, which are planar conic sections. These trajectories comprise the flow of the vector field. • Let us now turn to the elaboration of these ideas when a vector field X is given on a manifold M. If M = U is an open subset of a Banach space E, then a vector field on U is a map X: U ---) U x E of the form X(x) = (x, Vex)). We call V the principal part of X. However, having a separate notation for the principal part turns out to be an unnecessary burden. By abuse of notation, in linear spaces we shall write a vector field simply as a map X: U ---) E and shall mean the vector field x ....... (x, X(x)). When it is necessary to be careful with the distinction, we shall be. If M is a manifold and
C
M ---) veE is a local coordinate chart for M, then a
Section 4.1 Vector Fields and Flows
241
vector field X on M induces a vector field X on E called the local representative of X by the formula X(x) = Tq>· X(q>-I(x». If E =]!tn we can identify the principal part of the vector field X with an n-component vector function (XI(x), ... , xn(x». Thus we sometimes just say "the vector field X whose local representative is (xi) = (Xl, ... , xn)." Recall that a curve c at a point m of a manifold M is a Cl-map from an open interval I of ]!t into M such that 0 E I and c(O) = m. For such a curve we may assign a tangent vector at each point c(t), tEl, by c'(t) = Ttc(l). 4.1.4 Definition Let M be a manifold and X E X(M). An integral curve of X at m
E
M is a
curve c at m such that c'(t) = X(c(t» for each tEl. In case M = U
C
E, a curve c(t) is an integral curve of X : U ...... E when
c'(t) = X(c(t» , where c' = dc/dt. If X is a vector field on a manifold M and X denotes the principal part of its local representative in a chart q>, a curve c on M is an integral curve of X when de
dt (t) = where e = q>
0
X(e(t»,
c is the local representative of the curve c. If M is an n-manifold and thc local
representatives of X and c are (Xl, ... , xn) and (c l , ... , cn) respectively, then c is an integral eurve of X when the following system of ordinary differential equations is satisfied
n
dc (t ) = Xn( c I (t, ) dt
n(t». ... , c
The reader should chase through the definitions to verify this assertion. These equations are autonomous, corresponding to the fact that X is time independent. If X were time dependent, time t would appear explicitly on the right-hand side. As we saw in Example 4.1.3, the preceding system of equations includes equations of higher order (by their Usual reduction to first-order systems) and the Hamilton equations of motion as special cases. 4.1.5 Local Existence, Uniqueness, and Smoothness Theorem Let E be a Banach space
and suppose X: U C E ...... E is of class C~. For each Xo E U, there is a curve c : I ...... U at Xo Such that c'(t) = X(c(t» for all tEl. Any two such curves are equal on the intersection of their domains. Furthurmore, there is a neighborhood U o of the point
Xo E
U, a real number a> 0,
Chapter 4 Vector Fields and Dynamical Systems
242
and a Cu
C~
mapping F: Vo x I ~ E, where I is the open interval ] - a, a[, such that the curve
: I ~ E, defined by cu(t)
=F(u, t)
is a curve at u E E satisfying the differential equations c'u(t)
= X(cu(t» for all tEl.
4.1.6 Lemma Let E be a Banach space, veE an open set, and X : veE
map; i.e. there is a constant K > 0 such that
II X(x) -
X(y)
II ~
=
~
E a Lipschitz
II x - y II for all x, y E U. Let E III x - Xo II ~ b} lies in V, and
K
Xo E V and suppose the closed ball of radius b, Bb(x O) {x E II X(x) II ~ M for all x E Bb(xO)' Let to E IR and let a. = min(1/K, bIM), Then there is a unique C 1 curve x(t), t E [to - A., to + a.] such that x(t) E Bb(x O) and
{
X'(t) = X(x(t» x(to) = Xo .
Proof The conditions x'(t) = X(x(t», x(to) = Xo are equivalent to the integral equation
x(t) = Xo +
f
X(x(s» ds .
10
Put xo(t) = Xo and define inductively
Clearly xn(t) E Bb(x O) for aU n and t E [to - a., to + a.] by definition of a.. We also find by induction that
Thus xn(t) converges uniformly to a continuous curve x(t). Clearly x(t) satisfies the integral equation and thus is the solution we sought. For uniqueness, let y(t) be another solution. By induction we find that MK n It - to In+1/(n + I)! ; thus,letting n ~
00
II xn(t) -
y(t)
II ~
gives x(t) = y(t). T
The same argument holds if X depends explicitly on t and/or on a parameter p, is jointly continuous in (t, p, x), and is Lipschitz in x uniformly in t and p. Since xn(t) is continuous in (x o' to' p) so is x(t), being a uniform limit of continuous functions; thus the integral curve is jointly continuous in (x o' to, p). 4.1.7 Gronwall's Inequality Let f, g: [a, b[ ~ IR be continuous and nannegative. Suppose
that for all t satisfying a ~
t ~
b,
243
Section 4.1 Vector Fields and Flows
f(t) ::; A + f f(s)g(s) ds, for a constant A
~ O.
Then f(t) ::; A exp(f g(s) dS) for all t E [a, b[ .
Proof
First suppose A > O. Let h(t)
=
A + f f(s)g(s) ds;
thus h(t) > O. Then h'(t) = f(t)g(l)::; h(t)g(t). Thus h'(t)/h(t) ::; get). Integration gives
h(t) ::; A exp(f g(s) dS) .
This gives the result for A> O. If A = 0, then we get the result by replacing A by E > 0 for every E > 0; thus h and hence f is zero. T
4.1.8 Lemma Let X be as in Lemma 4.1.6. Let Ft(x o) denote the solution (= integral curve) of x'(t) = X(x(t», x(O) = xo' Then there is a neighborhood V of Xo and a number E> 0 such that for every y t E [-E, E],
E
V there is a unique integral curve x(t) = Ft(y) satisfying x'(t) = X(x(t» for all
and x(O) = y. Moreover,
Proof Choose V = B b/2 (x O) and E = min(1/K, b/2M). Fix an arbitrary y Bb(x o) and hence II X(z)
II ::; M for all z E
E
V. Then B b!2(Y) c
Bb1iY). By lemma 4.1.5 with Xo replaced by y, b
by b/2, and to by 0, there exists an integral curve x(t) of x'(t) = X(x(t»
for t E [-E, E] and
satisfying x(O) = y. This proves the first part. For the second, let f(t) = II Ft(x) - Ft(y) II. Clearly f(t)
=
III:
[X(Fs(x» - X(Fs(y»] ds + x - y
II ::; IIx -
yll + K
I:
f(s) ds ,
so the result follows from Gronwall's inequality. T This result shows that Ft(x) depends in a continuous, indeed Lipschitz, manner on the initial condition x and is jointly continuous in (t, x). Again, the same result holds if X depends explicitly on t and on a parameter p is jointly continuous in (t, p, x), and is Lipschitz in x uniformly in t and p; (Ft,l)P(x) is the unique integral curve x(1) satisfying x'(t) = X(x(t), t, p) and x(A.)
= x.
By the remark following lemma 4.1.6, (Ft)P(x) is jointly continuous in the
variables (A., t, p, x), and is Lipschitz in x, uniformly in CA., t, pl. The next result shows that
244
Chapter 4 Vector Fields and Dynamical Systems
Ft is Ck if X is, and completes the proof of 4.1.5. For the next lemma, recall that a Cl-function is locally Lipschitz. 4.1.9 Lemma Let X in Lemma 4.1.6 be of class Ck, 1$ k $ 00, and let Ft(x) be defined as before. Then locally in (t, x), Ft(x) is of class Ck in x and is Ck+1 in the t-variable.
Proof We define 'I'(t, x)
E
L(E, E), the set of continuous linear maps of E to E, to be the
solution of the "linearized" or "first variation" equations: d dt 'I'(t, x) = DX(Ft(x»
where DX(y) : E DX(Ft(x»
0
~
0
'I'(t, x), with '1'(0, x)
= identity,
E is the derivative of X taken at the point y. Since the vector field 'I' f-+
'I' on L(E, E) (depending explicitly on t and on the parameter x) is Lipschitz in '1',
uniformly in (t, x) in a neighborhood of every (to' xo)' by the remark following 4.1.8 it follows that 'I'(t, x) is continuous in (t, x) (using the norm topology on L(E, E». We claim that DFt(x)
= 'I'(t, x).
To show this, fix t, set 9(s, h)
= Fs(x + h)
- Fs(x), and
write 9(t, h) - 'I'(t, x) . h =
=
S: {X(Fs(x + h» - X(Fs(x»} ds - S: [DX(Fs(x)
0
'I'(s, x)] . h ds
S: DX(Fs(x»' [9(s, h) - 'I'(s, x) . h] ds +
S: {X(Fs(x + h» - X(Fs(x» - DX(Fs(x»
Since X is of class C 1, given I: > 0, there is a/» term is dominated in norm by
. [Fs(x + h) - Fs(x)]} ds .
0 such that II h II < /) implies the second
s: 1:11 Fs(x + h) - Fs(x) II ds , which is, in tum, smaller than AI: II h II for a positive constant A by lemma 4.1.8. By Gronwall's inequality we obtain II 9(t, h) - 'I'(t, x) . h II :5 (constant) I: II h II. It follows that DFt(x)· h = 'I'(t, x) . h. Thus both partial derivatives of Ft(x) exist and are continuous; therefore Ft(x) is of class C 1• We prove Ft(x) is Ck by induction on k. Begin with the equation defining Ft :
so
Section 4.1 Vector Fields and Flows
245
and
Since the right-hand sides are Ck -
I,
so are the solutions by induction. Thus F itself is
Ck . • Again there is an analogous result for the evolution operator (Ft,A)P(x) for a time-dependent vector field X(x, t, p), which depends on extra parameters p in a Banach space P. If X is Ck , then (Ft.,,)P(x) is Ck in all variables and is Ck+1 in t and A.. The variable p can be easily dealt with by suspending X to a new vector field obtained by appending the trivial differential equation
p'
=0;
this defines a vector field on Ex P and 4.1.5 may be applied to it. The flow on E x P
is just F/x, p)
= (Fl(x), p).
For another more "modem" proof of 4.1.5 see Supplement 4.1C. That alternative proof has a technical advantage: it works easily for other types of differentiability assumptions on X or on Ft, such as HOlder or Soholev differentiability; this result is due to Ebin and Marsden [1970] . The mapping F gives a locally unique integral curve
Cu
for each u
E
U o' and for each t E
I, F t = F I (U o x {t}) maps U o to some other set. It is convenient to think of each point u being allowed to "flow for time t" along the integral curve Cu (see Fig. 4.1.2 and our opening motivation). This is a picture of a U o "flowing," and the system (Uo' a, F) is a local flow of X, or flow box. The analogous situation on a manifold is given by the following.
Figure 4.1.2 4.1.10 Definition Let M be a manifold and X a
cr
vector field on M, r ~ 1. Aflow box of
X at m E M is a triple (U o' a, F), where (i) U o C M is open, m E U o' and a E lR., where a> 0 or a = + 00; (ii) F: U o x Ia ~ M is of class Cr, where Ia = j-a, a[;
(iii) for each u
E
U o'
Cu :
Ia ~ M defined by cu(t) = F(u, t) is an integral curve of X at
Chapter 4 Vector Fields and Dynamical Systems
246
the point u; (iv)
if Ft : Uo ~ Ft is a
M is defined by F/u) = F(u, t), then for tEla' Ft(U O) is open, and
cr diffeomorphism onto its image.
Before proving the existence of a flow box, it is convenient first to establish the following, which concerns uniqueness. 4.1.11 Global Uniqueness Suppose c l and c2 are two integral curves of X at m E M. Then c l = c2 on the intersection of their domains.
Proof This does not follow at once from 4.1.5 for c l and c2 may lie in different charts. (Indeed. if the manifold is not Hausdorff, Exercise 4.1M shows that this proposition is false.) Suppose c I
n
: II ~ M and c2 : 12 ~ M. Let I = II 12, and let K = {t E I I cl(t) = c2(t)}; K is closed since M is Hausdorff. We will now show that K is open. From 4.1.5, K contains some neighborhood of O. For t E K consider CIt and c 2t. where des) = c(t + s). Then CIt and c2t are integral curves at c,(t) = c 2(t). Again, by 4.1.5 they agree on some neighborhood of O. Thus some neigborhood of t lies in K, and so K is open. Since I is connnected, K = I. • 4.1.12 Proposition Suppose (Uo' a, F) is a triple satisfying (i), (ii), and (iii) of 4.1.10. Thenfor t, sand t + s E la' Ft+, = Ft 0 F, = F, 0 Ft and Fo is the identity map, whenever the compositions above are defined. Moreover, if U t = Ft(U o) and Ut n Uo 'I- 0, then Ft I U_ t Uo : U_t Uo ~ Uo Ut is a diffeomorphism and its inverse is F_t I Uo
n
n
n
n
U t•
Proof Ft+S
Ft+, = Fs 0 Ft, observe that Ft+s = Fs+t = Fs 0 Ft. Since cu(t) is a curve at u, cu(O) = u, so Fo is the identity. Finally, the last statement is a consequence of Ft 0 F_t = F_ t 0 Ft = identity. Note, however, that Ft(Uo) n Uo = 0 can occur. • 4.1.13 Existence and Uniqueness of Flow Boxes Let X be a cr vector field on a manifold M. For each mE M there is aflow box of X at m. Suppose (Uo' a, F), (Uo', a', F') are two flow boxes at mE M. Then F and F' are equal on (U o
n Uo' ) x (Ian Is')'
Proof (Uniqueness). Again we emphasize that this does not follow at once from 4.1.5, since Uo and Uo' need not be chart domains. However, for each point u E Uo n Uo' we have F I {u} x 1= F' I {u} x I, where I = I. n Is" This follows from 4.1.11 and 4.1.10(iii). Hence F = F' on
Section 4.1 Vector Fields and Flows
247
the set (U o n Uo' ) x I. (Existence). Let (U,
and let F: Uo x Ia ~ M; Since F is continuous, there is abE ]0, a[
(u, t) C
f-+
JR and Vo
C
U o open, with mE Yo' such that
C Uo. We contend that (V o' b, F) is a flow box at m (where F is understood as the restriction of F to Vo x Ib). Parts (i) and (ii) of 4.1.10 follow by construction and (iii) is a consequence of the remarks following 4.1.4 on the local representation. To prove (iv), note that
F(Vo x Ib)
for t E Ib, FI has a Cr inverse, namely, F_I as VI n U o = VI' It follows that FI(V o) is open. And, since FI and F_I are both of class Cr , FI is a Cr diffeomorphism. _ As usual, there is an analogous result for time- (or parameter-) dependent vector fields. The following result shows that near a point m satisfying X(m) '" 0, the flow can be transformed by ; change of variables so that the integral curves become straight lines.
4.1.14 Straightening Out Theorem Let X be a vector field on a manifold M and suppose at mE M, X(m) '" O. Then there is a local chart (U, 0; (ii)
Proof Since the result is local, by taking any initial coordinate chart, it suffices to prove the result
in E. We can arrange things so that we are working near 0 E E and X(O) = (0, 1) E E = G EB JR where G is a complement to the span of X(O). Letting (U o' b, F) be a flow box for X at 0 where U o = Vox j-e, e[ and Vo is open in G, define
But Dfo(O, 0) = Identity since aFI(O,O) at
I 1=0
=X(O)=(O,l)
and
Fo = Identity.
By the inverse mapping theorem there are open neighborhoods V x Ia C Vo X Ib and U = fa(V x Ia) of (0, 0) such that f = fa I V x Ia: V x Ia ~ U is a diffeomorphism. Then II : U
248 ~
Chapter 4 Vector Fields and Dynamical Systems
V x Ia can serve as chart for (i). Notice that c =f I {y} x I: I ~ U is the integral curve of X
through (y,O) for all y
E
V, thus proving (ii). Finally, the expression of the vector field X in
this local chart given by II is Df-I(y, t)· X(f(y, t» = Df-I(c(t»· c'(t) = (110 c)'(t) = (0, I), since (f- I 0 c)(t) = (y, t), thus proving (iii) .• In §4.3 we shall see that singular points, where the vector field vanishes, are of great interest in dynamics. The straightening out theorem does not claim anything about these points. Instead, one needs to appeal to more sophisticated normal form theorems; see Guckenheimer and Holmes [1983). Now we turn our attention from local flows to global considerations. These ideas center on considering the flow of a vector field as a whole, extended as far as possible in the t-variable. 4.1.15 Definition Given a manifold M and a vector field X on M, let of (m, t)
E
M x JR. such that there is an integral curve c : I
~
'Dx. C
M x JR. be the set
M of X at m with t
E
I. The
vector field X is complete if 'iJx = M x R A point m E M is called a-complete, where a = +, -, or ±, if 'iJx ({m) x JR.) contains all (m, t) for t > 0, < 0, or t E JR., respectively.
n
Let 'f+(m) (resp. l(m» denote the sup (resp. inj) of the times of existence of the integral curves through m; 'f+(m) resp. l(m) is called the positive (negative) lifetime of m. Thus, X is complete iff each integral curve can be extended so that its domain becomes = 00 and l(m) =- 0 0 for all m E M.
)-00, 00[; i.e., 'f+(m)
4.1.16 Examples A For M
=JR.2,
let X be the constant vector field, whose principal part is (0, 1). Then
X is complete since the integral curve of X through (x, y) is t B On M
=JR.2\ {OJ.
1-+
(x, y + t).
the same vector field is not complete since the inegral curve of X
through (0, -1) cannot be extended beyond t = 1; in fact as t ~ 1 this integral curve tends to the point (0,0). Thus 'f+(O, -1)
= I,
while 1(0, -1)
=
-00.
C On JR. consider the vector field X(x) = 1 + x2. This is not complete since the integral curve c with c(O) = 0 is c(a) = tan a and thus it cannot be continuously extended beyond -n12 and n12; i.e., T ±(O) = ± n!2 .• 4.1.17 Proposition Let M be a manifold and X E Xr(M), r ~ 1. Then (i) 'iJx:J M x {O}; (ii) 'iJx is open in M x JR.; (iii) there is a unique Cr mapping Fx: 'iJx ~ M such that the mapping t an integral curve at m for all m E M; (iv) for (m, t)
E
'iJx ' (Fx(m, t), s) E 'iJx iff (m, t + s) Fx(m, t + s) =Fx(Fx(m, t), s).
E
1-+
Fx(m, t) is
'iJx ; in this case
Section 4.1 Vector Fields and Flows
249
Proof. Parts (i) and (ii) follow from the flow box existence theorem. In (iii), we get a unique map Fx : 'Dx
~
M by the global uniqueness and local existence of integral curves: (m, t) E 'Dx when
the integral curve m(s) through m exists for s E [0, tl. We set Fx(m, t) = m(t). To show Fx is
cr,
note that in a neighborhood of a fixed mo and for small t, it is
cr
by local smoothness.
To show Fx is globally C r , first note that (iv) holds by global uniqueness.
Then in a
neighborhood of the compact set (m(s) I s E [0, tl} we can write Fx as a composition of finitely many
cr
maps by taking short enough time steps so the local flows are smooth. _
4.1.18 Definition Let M be a m1lnifold and X E Xr(M), r ~ 1. Then the m1lpping Fx is called
the integral of X, and the curve t
1-+
Fx(m, t) is called the maximal integral curve of X at m.
In case X is complete, Fx is called the flow of Thus, if
X
x.
is complete with flow F, then the set (Ft
ItE
lR.)
is a group of
diffeomorphisms on M, sometimes called a one-parameter group of diffeomorphisms. Since Fn
= (Fj)n (the n-th power), the notation P is sometimes convenient and is used where we use Fe 0 Fs = F t+ s wherever it is defined. Note that F/m) is The reader should write out similar definitions for the
For incomplete flows, (iv) says that F t defined for t E IT-(m), T+(m)[.
time-dependent case and note that the lifetimes depend on the starting time to. 4.1.19 Proposition Let X be Cr , where r
~
1. Let c(t) be a m1lximal integral curve of X
such that for every finite open interval la, b[ in the dOm1lin Ij(c(O», T+(c(O»[ of c, c(la, b[) lies in a compact subset of M. Then c is defined for all t
E
R
Proof It suffices to show that a E I, bEl, where I is the inteval of definition of c. Let Tn E la, b[. tn ~ b. By compactness we can assume some subsequence c(tn(k) converges, say, to a point x in M. Since the domain of the flow is open, it contains a neighborhood of (x, 0). Thus, there are e > 0 and
't
> 0 such that integral curves starting at points (such as c(tn(k)) for large
k) closer than e to x persist for a time longer than 'to This serves to extend c to a time greater than b, so bEl since c is maximal. Similarly, a E I. _ The support of a vector field X defined on a manifold M is defined to be the closure of the set (m E M I X(m) ;I'oO}. (4.1.20 Corollary. A Cr vector field with compact support on a manifold M is complete. In
particular, a Cr vector field on a compact m1lnifold is complete. Completeness corresponds to well-defined dynamics persisting eternally. In some circumstances (shock waves in fluids and solids, singularities in general relativity, etc.) one has to live with incompleteness or overcome it in some other way. Because of its importance we give two additional criteria. In the first result we use the notation X[f] = df . X for the derivative of f in
250
Chapter 4 Vector Fklds and Dynamical Systems
the direction X. Here f: E
~
lR and df stands for the derivative map. In standard coordinates
on lR n , n
af af) ~ iaf df(x) = ( - I ' ... , -n and X[f] = LJ X - i . ax ax i= 1 ax 4.1.21 Proposition Suppose X is a C< vector field on the Banach space E, k ~ I, and f: E ~
if {xn} is any sequence in E such that f(x n) ~ a, then there is a convergent subseqence (xn(i)}' Suppose there are constants K, L ~ 0 such that
lR is a C l proper map; that is,
I X[f](m) I ~ K I f(m) I + L for all
m E E.
Then the flow of X is complete. Proof From the chain rule we have (a/at)f(Ft(m»
=X[f](Ft(m»,
so that
f(Ft(m» - f(m) =
1: X[f](F~(m»
d't .
Applying the hypothesis and Gronwall's inequality we see that I f(Ft(m» I is bounded and hence relatively compact on any fmite t-interval, so as f is proper, a repetition of the argument in the proof of 4.1.19 applies . • Note that the same result holds if we replace "properness" by "inverse images of compacl sets are bounded" and assume X has a uniform existence time on each bounded set. This version is useful in some infmite dimensional examples. 4.1.22 Proposition Let E be a Banach space and X a Cr vector field on the Banach space E, r
~
1. Let cr be any integral curve of X. Assume II X(cr(t» II is bounded on finite t-intervals.
Then cr(t) exists for all t E R Proof. Suppose II X(cr(t» II < A for tEla, b[ and let tn
II
cr(~) - cr(~) II ~
tm II cr'(t)II dt
=
~
b. For tn < tm we have
It,,, II X(cr(t»11 dt ~ AI ~ - ~I .
Hence cr(tn) is a Cauchy sequence and therefore, converges. Now argue as in 4.1.19 .• 4.1.23
Examples A Let X be a Cr vector field, r ~ I, on the manifold M admitting aftrst integral, i.e. a function f: M ~ lR such that X[f] = O. If all level sets f -I(r), r E lR are compact, X is
251
Section 4.1 Vector Fields and Flows
complete. Indeed, each integral curve lies on a level set of f so that the result follows by 4.1.19. B Newton's equations for a moving particle of mass m in a potential field in JRO are given
by q(t).= -(l/m) V'V(q(t», for V: JRO ~ JR a smooth function. We shall prove that if there are constants a, b E JR, b ~ 0 such that (l/m)V(q) ~ a - bll q 11 2, then every solution exists for all time. To show this, rewrite the second order equations as a first order system -V'V(q) and note that the energy E(q, p) = (112m)
q= (l/m)p, p=
II p 112 + V(q) is a first integral. Thus, for any
solution (q(t), pet»~ we have ~ = E(q(t), pet»~ = E(q(O), p(O» ~ V(q(O». We can assume ~ > V(q(O», i.e. p(O) of. 0, for if pet) "" 0, then the conclusion is trivially satisifed; thus there exists a
to
for which p(t o) of. 0 and by time translation we can assume that to = O. Thus we have
II q(t) II .;:; II q(t) - q(O) II + II q(O) II';:; II q(O) II + J~ II q(s) II ds II q(O) II + J~ .;:; II q(O) II + J~
J1~ --k
V(q(S»] ds
.j 2(~ - a + b II q(s) 112)
ds
or in differential form :t
II q(t) II .;:;
J 2(~
- a+b
II q(t)1I 2)
whence
1
du
IIq(t) II
t
> -
IIq(O)1I
(I)
,J 2(~ -
a + bu 2)
Now let ret) be the solution of the differential equation d2r(t)
d
dt2
dr
-- = - -
2
(a - br let) = 2br(t) ,
which, as a second order equation with constant coefficients, has solutions for all time for any initial conditions. Choose reO)
= II q(O) II,
[r(O )]2
=2(~ -
a + bll q(O) 112)
and let ret) be the corresponding solution. Since
d (1. I ret) 2+ a - br(t)2)
dt
= 0 ,
it follows that (1/2) r(t)2 + a - br(t)2 = (112) r(0)2 + a - br(O)2 = [3, i.e.
252
Chapter 4 Vector Fields and Dynamical Systems
t-l
whence
-
du
r(t)
IIq(O)1I
(2)
.J 2(~ - a + ~U2)
Comparing the two expressions (1) and (2) and taking into account that the integrand is > 0, it follows that for any finite time interval for which q(t) is defined, we have II q(t) II :<=; ret), i.e., q(t) remains in a compact set for finite t-intervals. But then q(t) also lies in a compact set since IIq(t) II:<=; 2(~ - a + bll q(s) 11 2 ). Thus by 4.1.19, the solution curve (q(t), pet»~ is defined for any t ~ O. However, since (q( -t), p( -t»
is the value at t of the integral curve with initial conditions
(-q(O), -p(O», it follows that the solution also exists for all t:S: O. (This example will be generalized in §8.1 to any Lagrangian system on a complete Riemannian manifold whose energy function is kinetic energy of the metric plus a potential, with the potential obeying an inequality of the sort here). The following counterexample shows that the condition V(q) ~ a - bll q 112 cannot be relaxed much further. Take n = I and V(q) = _e2q2+(4/E)/8, e > O. Then the equation q= e(e + 2)ql+(4/E)/4 has the solution q(t)
= l/(t -
I)E/2,
which cannot be extended beyond t
= I.
C Let E be a Banach space. Suppose A(x) = A . x + B(x) , where A is a bounded linear operator of E to E and B is sublinear, Le., B : E r
~
-t
E is
cr
with
1 and satisfies II B(x) II :<=; KII x II + L for constants K and L. We shall show that X is
complete by using 4.1.22. (In JR.n, 4.1.21 can also be used with f(x) = II x 11 2). Let x(t) be an integral curve of X on the bounded interval [0, T]. Then
x(t) = x(O) + Hence IIx(t) II :S:
II x(O) II
f: f: (
(A . xes) + B(x(s») ds
II A II + K) II xes) II ds + Lt.
By Gronwall's inequality, II x(t) II :S: (LT + II x(O) lI)e(lIAII+K)t Hence x(t) and so X(x(t» remain bounded on bounded t-intervals. +
4.1.24 Proposition Let X be a c r vector field on the manifold M, r ~ 1, mo E M, and T+(mo)(T-(m o the positive (negative) lifetime of mo. Then for each e > 0, there exists a neighborhood V of mo such that for all mEV, T+(m) > T+(m o) - e (respectively, l(mo) <
»
Section 4.1 Vector Fields and Flows
253
l(m o) + E). [One says T+(m o) is a lower semi-continuous function of m.] Proof Covcr the scgmcnt (F/m o) I t E [0, r+(m o) - E]) with a finite number of neighborhoods U o' ... , Un' each in a chart domain and such that O. Pick points Pi E U i n Vi+1 and let s(i) be such that Fs(i)(m O) = Pi' i = 0, ... , n - 1, s(O) = 0, Po = rna, s(i) < s(i + 1), s(i + 1) - t(i) < o.(i), t(i + 1) - s(i + 1) < o.(i + 1), s(i + 1) - s(i) < min(o.(i), o.(i+ 1»; sce Figure 4.1.3.
o
Figure 4.1.3 Let WI = Uo n VI' Since s(2) - s(l) < 0.(1) and any integral curve starting in WI C VI exists for time at least 0.(1), the domain of Fs(2)-s(1) includes WI and hence F s(2}-s(1)(W I) makes sense. Define the open set W 2 = F s(2}-s(1)(W I) n U I n U2 and use s(3) - s(2) < 0.(2), W 2 C U 2 to conclude that the domain of F s(3)-s(2) contains U 2. Define the open set W3 = Fs(3)-s(2/W2) n V 2 n V3 and inductively define Wi = Fs(i}-s(i-l)(W i- l ) n V i_ 1 n Vi' i = 1, ... , n, whieh are open sets. Since s(1) < 0.(0) and WI C Va' the domain of F -5(1) includes WI and thus VI = F_s(I)(W t> n Uo is an open neighborhood of mo' Since s(2) - s(1) < 0.(1) and W 2 C VI' the domain of F_ s(2)+s(l) contains W 2 and so F-s(2)+s(1/W 2) C WI makes sense. Therefore F-s(2lW2) = F-s(I)(F- s(2)+s(l)(W 2 C F_s(l)(W 1) exists and is an open neighborhood of mo' Put V2 = F_ s(2)(W 2) n Va. Now proceed inductively to show that F_s(i)(W) C F-s(i_l)(W i_ l ) makes sense and is an open neighborhood of rna, i = 1,2, ... , n; see Figure
»
n Uo' i = 1, ... , n, open neighborhoods containing mo' Any integral curve e(t) starting in Vn exists thus for time at least s(n) and Fs(n)(V n) C Wn C Un' Now consider the integral curve starting at c(s(n» whose time of existence is at least o.(n). By uniqueness, c(t) can be smoothly extended to an integral curve which exists for time at least s(n) + o.(n) > tn = r+(m o) - E. • 4.1.4. Let Vi = F_s(ilW)
Chapter 4 Vector Fields and Dynamical Systems
254
Figure 4.1.4 The same result and proof hold for time dependent vector fields depending on a parameter.
cr
4.1.25 Corollary Let ~, i.e.
~
mo such that any m
V has integral curve existing for time t E
equilibrium of
E
be a
~(ffio)
time-dependent vector field on M, r
~
I, and let mo be an
= D, for all t. Thenfor any T there exists a neighborhood V of [- T,
TJ.
Proof Since T+(mo) = +00, l(mo) =-00, the previous proposition gives the result. _
I@"
Supplement 4.1A
Product Formulas A result of some importance in both theoretical and numerical work concerns writing a flow in terms of iterates of a known mapping. Let X E X(M) with flow Ft (maximally extended). Let Kt(x) be a given map defined in some open set of [D, oo[ x M containing (D) x M and taking values in M, and assume that (i) Ko(x) = x and (ii) Kt(x) is C' in £ with derivative continuous in (£, x). We call K the algorithm. 4.1.26 Theorem Let X be a Cr vector field, r consistent with X in the sense that
~
l. Assume that the algorithm Kt(x) is
255
Section 4.1 Vector Fields and Flows
X(x) =
Then,
if
aa£
Kf(x)
I
f={)
(t. x) is in the domain of Ft(x). (Kt/n)n(x) is defined for n sufficiently large and
converges to Ft(x) as n
~
"". Conversely.
if
(Kt/n)n(x) is defined and converges for 0:<0; t:
T. then (T. x) is in the domain of F and the limit is Kt(x). In the following proof the notation O(x). x
E
lR is used for any continuous function in a
neighborhood of the origin such that O(x)/x is bounded. Recall from §2.1 that o(x) denotes a continuous function in a neighborhood of the origin satisfying limx --> oo(x)/x = O. Proof. First. we prove that convergence holds locally. We begin by showing that for any
xo'
the iterates (Kt/n)n(x O) are defined if t is sufficiently small. Indeed. on a neighborhood of xO' Kf(x)
= x + 0(£).
so if (KtI)i(x) is defined for x in a neighborhood of xO' for j
then (K,/n)n(x) - x
= «Kt/n)n x -
(Kt/n)n-l x )
= 1..... n -
1.
+ «K,/n)n-1 - (K t/n )n-2x ) + ... + (Kt/n(x) - x)
= O(t/n) + ... + O(t/n) = O(t).
This is small. independent of n for t sufficiently small; so. inductively. (Kt/n)n(x) is defined and remains in a neighborhood of Xo for x near xO' Let ~ be a local Lipschitz constant for X so that
II Ft(x) -
Ft(y)
II
:<0; e J3ltl 1l x - y
II.
Now
write Ft(x) - (K,;n)n(x)
= (Ft/n)n(x) - (K,;n)n(x) = (Ft/n)n-IFt/n(x) - (Ft/n)n-lK,;n(x)
+ (Ft/n)n-2Ft/n (YI) - (Ft/n)n-2Kt/n(YI) + ... + (Ft/n)n-kFt/n(Yk_l) - (Ft/n)n-kK,;n(h_l) + ... + Ft/n(Yn-l) - K,;n(Yn-l)
n
II Fb) -
(Kt/n)n(x)
II
:<0;
L.
e J31 t l(n-k)/n
II Ft/n(Yk-l) -
Kt/n(Yk-l)
II
k= 1
:<0; ne J31 t lo(t/n) ~ 0 as n ~ "". since Ff(y) - Kf(y) = 0(£) by the consistency hypothesis. Now suppose Ft<x) is defined for 0:<0; t :<0; T. We shall show (K,'n)n(x)
converges to
Ft(x). By the foregoing proof and compactness. if N is large enough. FtiN = limn --> ~ (KtlnN)n uniformly on a neighborhood of the curve t 1-+ Ft(x). Thus. for 0:<0; t :<0; T.
Chapter 4 Vector Fields and Dynamical Systems
256
By uniformity in t. FT(x) = lim (KTI/(x). J~~
Conversely. let (Ktln)n(x) converge to a curve c(t). 0:-:; t:-:; T. Let S = {t I Ft(x) is defined and c(t) = Ft(x)}. From the local result. S is a nonempty open set. Let t(k)
E
S. t(k)
-t
t.
Thus Ft(k)(x) converges to c(t). so by local existence theory. Ft(x) is defined. and by continuity. Ft(x) = c(t). Hence S = [0. T] and the proof is complete. _ 4.1.27 Corollary
Thenfor x
E
Let X. Y
E
X(M) with flows Ft and G t. Let St be the flow of X + Y.
M.
The left-hand side is defined iff the right-hand side is. This follows from 4.1.26 by setting KE(x) = (FE
0
GE)(x). For example. for n x n matrices A and B. 4.1.27 yields the classical
formula e(A+B) = lim (eAlneB/n)n. n~~
To see this. define for any n x n matrix C a vector field Xc
E
X (]Rn) by Xc(x) = Cx. Since
Xc is linear in C and has flow Ft(x) = etCx. the formula follows from 4.1.27 by letting t = I. The topic of this supplement will continue in Supplement 4.2A. The foregoing proofs were inspired by Nelson [1969] and Chorin et al. [1978].
~
fl:n
Supplement 4.lB Invariant Sets
If X is a smooth vector field on a manifold M and N
C
M is a submanifold. the flow of
X will leave N invariant (as a set) iff X is tangent to N. If N is not a submanifold (e.g .. N is an open subset together with a non-smooth boundary) the situation is not so simple; however. for this there is a nice criterion going back to Nagumo [1942]. Our proof follows Brezis [1970]. 4.1.28 Theorem Let X be a locally Lipschitz vector field on an open set U
C
E. where E is a
Banach space. Let G cUbe relatively closed and set d(x. G) = inf{ II x - y III y following are equivalent: (i) (ii)
E
G). The
limh~o(d(x + hX(x). G)/h) = 0 locally uniformly in x E G (or pointwise if E = ]Rn); if x(t) is the integral curve of X starting in G. then x(t) E G for all t ~ 0 in the domain of x(·).
257
Section 4.1 Vector Fields and Flows
Note that x(1) need not lie in G for t ~ 0; so G is only + invariant. (We remark that if X is only continuous the theorem fails.) We give the proof assuming E = IRn for simplicity. Proof Assume (ii) holds. Setting x(t) = Ft(x), where Ft is the flow of X and x E G, for small h we get d(x + hX(x), G)
~
II x(h) - x - hX(x) II
=
- x - X(x) II I h I II -x(h) -h
from which (i) follows. Now assume (i). It suffices to show x(t)
E
G for small t. Near x = x(O)
E
G, sayan a
ball of radius r, we have and
We can assume II Ft(x) - x II < r/2. Set
II Ft(x) - Yt II for some Yt E G. (In the general Banach space case an approximation argument is needed here.) Thus, II Yt - X II < r. For small h, II Fh(y t) - x II < r, so that
small t,
= infzE
G
~ infz E G
II Ft+h(x) -
II
(II Ft+h(x) - Fh(y t) II + II Fh(y t) - Yt - hX(y t) II + II Yt + hX(y t) - zlll
= II Ft+h(x) ~ e Kh
Z
Fh(y t) II + II Fh(y t) - Yt - hX(y t) II + d(Y t + hX(y t), G)
II Yt - Ft(x) II + II Fh(y t) - Yt - hX(y t) II + d(Y t + hX(y t), G)
or
II Fh(Yt)h - Yt - X(Yt) II + d(Yt + hX(Yt), G)/h.
Hence . I1m
suPh~o
~
K
As in Gronwall's inequality, we may conclude that
so
4.1.29 Example Let X be a C~ vector field on IRn , let g : IRn ~ IR be smooth, and let A E IR be a regular value for g, so g-l(A) is a submanifold; see Figure 4.1.5. Let G = g-l(]_oo, AD and suppose that on g-l(A). (X, grad g) ~ o.
258
Chapter 4 Vector Fields and Dynamical Systems
Figure 4.1.5 The set G = g-l(]_oo , AD is invariant if X does not point strictly outwards at aGo Then G is + invariant under Ft as may be seen by using 4.1.28. This result has been generalized to the case where aG might not be smooth by Bony [1969]. See also Redheffer [1972] and Martin [1973]. Related references are Yorke [1967], Hartman [1972], and Crandall [1972]. (lJJ
~ Supplement 4.1C A second Proof of the Existence and Uniqueness of Flow Boxes
We shall now give an alternative "modem" proof of Theorems 4.1.5 and 4.1.13, namely, if X E Xk(M), k ~ 1, then for each m E M there exists a unique Ck flow box at m. The basic idea is due to Robbin [1968] although similar alternative proofs were simultaneously discovered by Abraham and Pugh [unpublished] and Marsden [1968b, p. 368]. The present exposition follows Robbin [1968] and Ebin and Marsden [1970]. Step 1 Existence and uniqueness of integral curves for Cl vector fields. Proof Working on a local chart, we may assume that x: 0/0) - t E, where 0/0) is the opcn disk at the origin of radius r in the Banach space E. Let U = 0r/2(0), 1= [-1, 1] and define
by <J>(s, y)(t) =
~; (t) -
sX(y(t)) ,
Section 4.1 Vector Fields and Flows
259
where Ci(I, E) is the Banach space of Ci-maps of I into E, endowed with the II . IIi-norm (see Supplement 2.4B), Cia (I , E) = (f E Ci(l, E) I f(O) = OJ is a closed subspace of Ci(I, E) and Cio(l, U) = (f E Cio(l, E) I f(l)
C
UJ is open in Cio(l, E). We first show that <1> is a Cl-map.
The map d/dt : Clo(l, E) ~ C°(l, E) is clearly linear and is continuous since II d/dt II ~ 1. Moreover, if dy/dt = 0 on I, then y is constant and since )'(0) = 0, it follows y= 0; i.e., dldt is injective. Given 0 E C°(l, E), y(t) =
5:
o(s) ds
defines an element of Cio(l, E) with dy/dt = 0, i.e., dldt is a Banach space isomorphism from Clo(l, E) to C°(l, E). From these remarks and the Q-lemma (2.4.18), it follows that <1> is a Cl-map. Moreover, Dy<1>(O, 0) = d/dt is an isommorphism of Clo(I, E) with C°(l, E). Since <1>(0, 0) = 0, by the implicit function theorem there is an £ > 0 such that <1>(£, y) = 0 has a unique solution YE(t) in Clo(I, U). The unique integral curve sought is )'(t) = y£(I/£), -£ ~ t ~ £. T The same argument also works in the time-dependent case. It also shows that y varies continuously with X. Step 2 The local flow of a Ck vector field X is Ck. Proof First, suppose k = 1. Modify the definition of <1> in Step 1 by setting 'P: lEt x U x
Clo(l, U) ~ C°(I, E), 'P(s, x, y)(t) = 1 '(t) - sX(x + )'(t». As in Step I, 'P is a CI-map and Dy'P(O, 0, 0) is an isomorphism, so 'P(£, x, 1) = 0 can be locally solved for 1 giving a map HE: U ~ Clo(I, U), £ > O. The local flow is F(x, t) = x + HE(x)(t/£), as in Step 1. By 2.4.17 (differentiability of the evaluation map), F is C 1• Moreover, if vEE, we have DFt(x)· v = v + (DHE(x) . v)(I/£), so that the mixed partial derivative
exists and is jointly continuous in (t, x). By exercise 2.4G, D(dFt(x)/dt) exists and equals (dldt)(DFt<x». Next we prove the result for k ~ 2. Consider the Banach space F = Ck-l(cl(U), E) and the map ffiX: F ~ F; 11 f-+ X 011. This map is C I by the Q-lemma (remarks following 2.4.18). Regarding ffix as a vector field on F, it has a unique Cl integral curve 11t with 110 = identity, by Step 1. This integral curve is the local flow of X and is Ck - I since it lies in F. Since k ~ 2, 11t is at least CI and so one sees that D11 t = u t satisfies d~/dt = DX(11 t) . ut' so by Step 1 again, u t lies in Ck - I. Hence 11t is Ck •
260
Chapter 4 Vector Fields and Dynamical Systems
The following is a useful alternative argument for proving the result for k = 1 from that for k ~ 2. For k = 1. let xn ~ X in C 1• where xn are C2. By the above. the flows of X" are C 2 and by Step 1. converge uniformly i.e. in Ca. to the flow of X. From the equations for
Drtt".
we likewise see that
Drtt"
converges uniformly to the solution of d~/dt = DX(11 t) . ut' U o
= identity. It follows by elementary analysis (see Exercise 2.4J or Marsden [I974a. p. 109]) that
11t is Cl and D11 t = ut.• This proof works with minor modifications on manifolds with vector fields and flows of Sobolev class H' or Holder class Ck +u ; see Ebin and Marsden [1970] and Bourguignon and Brezis [1974]. In fact the foregoing proof works in any function spaces for which the O-Iemma can be proved. Abstract axioms guaranteeing this are given in Palais [1968].
tl:n
Exercises 4.1A Find an explicit formula for the flow Ft :]R2 ~]R2 of the harmonic oscillator equation
it + Q)2 x
= O. Q) E
]R a constant.
4.1B Show that if (Uo' a. F) is a flow box for X. then (U o' a. F _) is a flow box for -X. where F_(u. t) = F(u. -t) and (-X)(m) = - (X(m». 4.1C Show that the integral curves of a C vector field X on an n-manifold can be defined locally in the neighborhood of a point where X is nonzero by n equations 'l'i(m. t) = c i = constant. i = 1..... n in the n+ 1 unknowns (m. t). Such a system of equations is called a local complete system olintegrals. (Hint: Use the straightening-out theorem.) 4.1D Prove the following generalization of Gronwall's inequality. Supppose v(t)
v(t) $ C +
IJ~
p(s) v(s) ds
I.
where C ~ 0 and P E Ll Then
v(t)
~ C exp( IS: p(s) ds I).
Use this to generalize 4.1.23(iii) to allow A to be a time-dependent matrix.
~
0 satisfies
261
Section 4.1 Vector Fields and Flows
4.1.E Let Ft = etX be the flow of a linear vector field X on E. Show that the solution of the equation
x= X(x) + f(x) with initial conditions Xo satisfies the varilltion of constants formula
4.1F Let
F(m. t) be a
C~
mapping of M x IR to M such that Ft+s = F t 0 Fs and Fo =
identity (where Ft(m) = F(m. t». Show that there is a unique
C~
vector field X whose
flow is F. 4.1G Let o(t) be an integral curve of a vector field X and let g: M = g(o(,(t»). Then show t
f-+
~
lR. Let ,(t) satisfy 't'(t)
o(,(t» is an integral curve of gX. Show by example that
even if X is complete. gX need not be. 4.1H (i) Gradient Flows
Let f: 1R" ~ IR be C i and let X = (Jf/Jx i •...• Jf/Jx") be the
gradient of f. Let F be the flow of X. Show that f(Ft(x)
~
f(F,(x» if t ~ s.
(ii) Use (i) to fmd a vector field X on 1R" such that X(O) =
o. X'(O) = O.
globally attracting; i.e .• every integral curve converges to 0 as t ~
00.
yet 0 is
This exercise
continues in 4.3K. 4.11 Let c be a locally Lipschitz increasing function. c(t) > 0 for t
~
0 and assume that the
differential equation r'(t) = c(r(t» has the solution with r(O) = ro ~ 0 existing for time t E [0. TJ. Conclude that r(t) ~ 0 for t E [0. TJ. Prove the following comparison lemmas. (i)
If h(t) is a continuous function on [0. TJ, h(t)
~
O. satisfying h'(t)::; c(h(t» on
[0. TJ, h(O) = roo then show that h(t)::; r(t). (Hint: Prove that
l
h(t) ~
'0
< t = lr(t) ~ c(x) '0 c(x)
and use strict positivity of the integrand.) (ii) Generalize (i) to the case h(O)::; roo (Hint: The function h(t) = h(t) + ro - h(O) ::; h(t) satisfies the hypotheses in (i).) (iii) If f(t) is a continuous function on [0. Tl. f(t)
f(t) ::; ro +
f:
~
O. satisfying
c(f(s» ds
Chapter 4 Vector Fields and Dynamical Systems
262
on [0, TJ, then show that f(t):<; ret). (Hint:
h(t) = ro +
J:
c(f(s» ds
~ f(t)
satisfies the hypothesis in (i) since h'(t) = c(f(t» :<; c(h(t».) (iv) If in addition dx - =+00 , c(x) show that the solution f(t)
~
0 exists for all t ~ O.
(v) If h(!) is only continuous on [0, T] and
h(t) :<; ro +
J:
c(h(s» ds ,
show that h(t):<; ret) on [0, T]. (Hint: Approximate h(t) by a C1-function get) and show that get) satisfies the same inequality as h(t).).
4.IJ
(i) Let X = y2(J/(Jx and Y = x2(J/(Jy. Show that X and Y are complete on ]R2 but X + Y is not. (Hint: Note that x 3 - y3 = constant and consider an integral curve with x(O)
=y(O).)
(ii) Prove the following theorem: Let H be a Hilbert space and X and Y be locally Lipschitz vector fields that satisfy the following: (a) X and Y are bounded and Lipschitz on bounded sets; (b) there is a constant
~ ~
0 such that
( Y(x), x) :<; ~II
X
112 for all x E H;
(c) there is a locally Lipschitz monotone increasing function c(t) > 0, t that
and if x(t) is an integral curve of X, d
dt II x(t) II :<; c(1I x(t) II) . Then X, Y and X + Y are positively complete.
~
0, such
263
Section 4.1 Vector Fields and Flows
II X(X O) II <; cOl Xo II) in (c) instead of (d/dt) II x(t) II II). (Hint: Find a differential inequality for (l/2)(d/dt) II u(t) Ie,
Note: One may assume <; c(1I x(t)
where u(t) is an integral curve of X + Y and then use Exercise 4.I(iii).) 4.1K Prove the following result on the convergence offlows: Let Xa be locally Lipschitz vector fields on M for a in some topological space. Suppose the Lipschitz constants of Xa are locally bounded as a
-7
a(O) and Xa -7
Xa(O) locally uniformly. Let c(t) be an integral curve of Xa(O)' 0 <; t <; T and £ > O. Then the integral curves ca(t) of Xa with ca(O) = c(O) are defined for the interval t E
[0, T - £] for a sufficiently close to a(O) and ca(t)
T - £] as a
-7
a(O). If the flows are complete Ft a
-7
-7
c(t) uniformly in t
E
[0,
Ft locally uniformly. (The
vector fields may be time dependent if the estimates are locally t-uniform.)
(Hint:
Show that
and conclude from Gronwall's inequality that ca(t) second term
-7
-7
c(t) for a
-7
a(O) since the
O. This estimate shows that ca(t) exists as long as c(t) does on any
compact subinterval of [0, T[.) 4.1L
Prove that the Cr flow of a Cr +l vector field is a C l function of the vector field by utilizing Supplement 4.1C. (Caution. It is known that the Ck flow of a Ck vector field cannot be a C l function of the vector field; see Ebin and Marsden [1970] for the explanation and further references).
4.1.M Nonunique integral curves on non-Hausdorff manifolds Let M be the line with two origins (see Exercise 3.SH) and consider the vector field X: M
-7
TM which is defined
by X([x, iJ) = x, i = 1,2; here [x, i] denotes a point of the quotient manifold M. Show that through every point other than [0,0] and [0, 1], there are exactly two integral curves of X. Show that X is complete. (Hint: The two distinct integral curves pass respectively through [0, 0] and [0, 1].) 4.1N Give another proof of Theorem 4.1.5 using Exercise 2.SJ. 4.10
Give examples of vector fields satisfying the following conditions: (i) (ii) (iii) (iv)
on]R and SI with no critical points; generalize to ]Rn and 1I"" on S 1 with exactly k critical points; generalize to 1I"'; on]R2 and lRJP'2 with exactly one critical point and all other orbits closed; on the Mobius band with no critical points and such that the only integral curve inter-
264
Chapter 4 Vector Fields and Dynamical Systems
seeting the zero section is the zero section itself; (v)
on S2 with precisely two critical points and one closed orbit;
(vi)
on S2 with precisely one critical point and no closed orbit;
(vii)
on S2 with no critical points on a great circle and nowhere tangent to it; show that
(viii)
on 1['2 with no critical points, all orbits closed and winding exactly k times around
any such vee tor field has its integral curves intersecting this great circle at most once;
1I'2. 4.1 P Let 1t: M ~ N be a surjective submersion. A vector field X if T1t
0
X = O. If IK is the Klein bottle, show that 1t:
I{ ~
E
X(M) is called 1t-vertical
st given by 1t([a, b]) = e 21tia
is a surjective submersion; see Figure 4.1.6. Prove that TIC is a non trivial SLbundle. (Hint: If it were trivial, there would exist a nowhere zero vertical vector field on IK. In Figure 4.1.6, this means that arrows go up on the left and down on the right hand side. Follow a path from left to right and argue by the intermediate value theorem that the vector field must vanish somewhere.)
Figure 4.1.6 The Klein bottle as an SLbundle.
~4.2
Vector Fields as Differential Operators
In the previous section vector fields were studied from the point of view of dynamics; that is in terms of the flows they generate. Before continuing the development of dynamics, we shall treat some of the algebraic aspects of vector fields. The specific goal of the section is the development of the Lie derivative of functions and vector fields and its relationship with flows. One important feature is the behavior of the constructions under mappings. The operations should be as natural or covariant as possible when subjected to a mapping. We begin with a discussion of the action of mappings on functions and vector fields. First, recall some notation. Let C'(M, F) denote the space of Cr maps f: M ~ F, where F is a Banach space, and let Xr(M) denote the space of Cr vector fields on M. Both are vector spaces with the obvious operations of addition and scalar multiplication. For brevity we write .r(M) == C"'(M, JR.), .rr(M) == Cr(M, JR.) and X(M) == X~(M). Note that .rr(M) has an algebra structure; that is, for f, g
E
.rr(M) the product fg defined by (fg)(m) == f(m)g(m) obeys the usual
algebraic properties of a product such as fg == gf and f(g + h) == fg + fh. 4.2.1 Definitions (i)
Let
(ii) Iff' is a
cr
diffeomorphism and X
E
Xr(M), the push-forward of X by
defined by
Consider local charts (U, X), X: U ~ U' and let (TX
0
C
E on M and (V, \jf),
\jf:
V ~ V'
C
F on N,
X 0 X-I )(u) == (u, X(u», where X: U' ~ E is the local representative of X. Then
from the chain rule and the definition of push-forward, the local representative of
Where v == (\jf 0
266
Chapter 4 Vector Fields and Dynamical Systems
where y = q>(x). We can interchange "pull-back" and "push-forward" by changing q> to * = (-1 and the pull-back of a vector field Y on N is
0
Y 0 q> (Fig.
4.2.1). Notice that
pull back
Figure 4.2.1
4.2.2 Definition Let -related,denoted X -
X
E
Note that if
T-----:.--.-l . T
X
M --------- N
Section 4.2 Vector Fields as Differential Operators
267
4.2.3 Proposition
(i) Pull-back and push-forward are linear maps, and
(ii) If
X; -", Yj, i = 1,2, and a, b E JR, then aX J + bX2 -", aY J + bY2 .
(iii) For
~
N and '1': N
~
P, we have
(iv) If X E X(M), Y E X(N), Z E X(P), X
-q>
Y and Y
-1jI
Z, then X
-1jI
0
",Z.
In this proposition it is understood that all maps are diffeomorphirns with the exception of the pull-back offunctions and the relatedness of vector fields. Proof (i) This consists of straightforward verifications. For example, if Xj
-q> Y j, i = 1,2, then T aY J +
bY 2 · (ii)
These relations on functions are simple consequences of the definition, and the ones on
X(p) and X(M) are proved in the following way using the chain rule:
T('I' °
replace real-valued functions by F-valued functions. The behavior of flows under these operations is as follows: 4.2.4 Proposition
Let
Xr(N). Let FXt and F\ denote the flows of X and Y respectively. Then X -q> Y iff
of Y is
which is equivalent to T
268
Chapter 4 Vector Fields and Dynamical Systems
d(<jloc)(t) (dC(t)) dt = T<jl """"d't = T<jl(X(c(t))) = Y«<jl
says that <jl
0
0
c)(t))
c is the integral curve of Y through <jl(c(O)) = <jl(m). By uniqueness of integral
curves, we get (<jl 0 FXt)(m) = (<jl 0 c)(t) = F\(<jl(m)). The last statement is obtained by taking <jl FX s for fixed s. • We call <jl 0 Ft 0 <jl-l the push-forward of Ft by <jl since it is the natural way to construct a diffeomorphism on N out of one on M. See Fig. 4.2.2. Thus, 4.2.4 says that the flow of the
push-forward of a vector field is the push forward of its flow. Next we define how vector fields operate on functions. This is done by mcans of lhe directional derivative. Let f: M
~
JR., so Tf: TM
~
T JR. = JR. x R Recall that a tangent vector
to lR. at a base point A E JR. is a pair (A, 11), the number 11 being the principal part. Thus we can write Tf acting on a vector v E TmM in the form Tf· v = (f(m), df(m)· v).
N
c = integral curve of X
Figure 4.2.2 This defines, for each mE M, the element df(m) E Tm *M. Thus df is a section of T*M, a
covector field, or one-form. 4.2.5 Definition The covector field df: M ~ T*M defined this way is called the differential of f. For F-valued functions, f: M ~ F, where F is a Banach space, a similar definition gives df(m) E L(TmM, F) and we speak of df as an F-valued one-form. Clearly if f is Cr , then df is Cr - 1. Let us now work out df in local charts for f E .'r(M). If <jl: U C M ~ veE is a local chart for M, then the local representative of f is the map f : V ~ JR. defined by f = f 0 <jl-l. The local representative of Tf is the tangent map for local
Section 4.2 Vector Fields as Differential Operators
269
manifolds: Tf(x, v)
= (f(x), Df(x) . v).
Thus the local representative of df is the derivative of the local representative of f. In particular, if M is finite dimensional and local coordinates are denoted (x 1,
of df are (df);
... ,
x n ), then the local components
af
= -, .
ax'
The introduction of df leads to the following.
4.2.6 Definition Let f E .'Fr(M) ami X E Xr-l(M), r ~ 1. Define the directional or Lie derivative of f along X by Lxf(m) '=' X[£](m) = df(m) . X(m),
for any mE M. Denote by X[£] = df(X) the map mE M
1--+
X[£](m) E lR.. If f is F-valued,
the same definition is used, but now X[£] is F-valued. The local representative of X[£] in a chart is given by the function x 1--+ Df(x) . X(x), where
f and X are the local representatives of f and X. In particular, if M is finite dimensional then we have X[fj,= Lxf=
Evidently if f is
cr
and X is
cr- 1
~
af '
i=l
ax
.L..J - i X'.
then X[£] is Cr -
1.
Let us observe that from the chain rule, d(f 0 cp) = df 0 Tcp where cp: N of manifolds, r
~
~
M is a
cr map
1. For real-valued functions, Leibniz' rule gives
d(fg) = fdg + gdf. (If f is F-valued, g is G-valued and B: F x G
---t
H is a continuous bilinear map of Banach
spaces, this generalizes to d(B(f, g)) = B(df, g) + B(f, dg).
4.2.7 Proposition (i) Suppose cp: M ---t N is a diffeomorphism. Then push-forward by <po That is Jor each f E .'F(M),
or the following diagram commutes:
Lx is natural with respect to
270
Chapter 4 Vector Fields and Dynamical Systems
.'J(M)
• .'J(N)
~j
..
.1{M)
jL•.
x
.1{N)
IV; or if I V : .'J (M)
--t .'J (V)
denotes restriction to V. the
following diagram commutes:
.'J(M)
IV
--~----I• .1{V)
j~,"
~j
.. .1{V) .'J(M) - - - - - - -
IV
Proof For (i). if n
E
N then
L q>.x(
0
X 0
= df(
=(
the Lie derivative is also natural with respect to pull-back by <po This
has a generalization to
E
Xr-I(M) and Y
E
Xr--I(N). If X
then Lx(
[~
I
l~
Cr-1(N, F ) - - - - - -.....-cr-I(M. F)
E
M.
-'I'
Y.
Section 4.2 Vector Fields as DijJerentWl Operators
271
Lx(
Next we show that Lx satisfies the Leibniz rule. 4.2.9 Proposition (i)
The mapping Lx: O(M, F) ~ Cr--1(M, F) is a derivation. That is Lx is JR.-linear arul for f
E
O(M, F), g
E
Cr(M, G) arul B : F x G
~
H a bilinear map
Lx(B(f, g» = B(Lxf, g) + B(f, Lxg). In particular,for real-valued functions, Lx(fg)
=gLxf + fLxg.
(ii) If c is a constant function, Lxc = o. Proof (i) This follows from the Leibniz rule for d and the definition Lxf. Part (ii) results from the definition. _ The connection between Lxf and the flow of X is as follows. 4.2.10 Lie Derivative Formula for Functions Let f
X has aflow
E
O(M, F), X E ,Xr-1(M) arul suppose
Ft' Then
Proof By the chain rule, the definition of the differential of a function and the flow of a vector field, * d dFt(m) d dt (Ft f)(m) = dt (f 0 Ft)(m) = df(Ft(m» . -d-t= df(Ft(m» . X(Ft(m» = df(X)(Ft(m» = (Lxf)(Ft(m» = (F7 Lxf)(m) . •
As an application of the Lie derivative formula, we consider the problem of solving a partial differential equation on JR.n+1 of the form
af at (x, t)
n
=
~;
af ax
L..J X (x) - ; (x, t) ;=1
(P)
With initial condition f(x, 0) = g(x) for given smooth functions X;(x), i = 1, ... , n, g(x) and a
272
Chapter 4 Vector Fields and Dynamical Systems
scalar unknown f(x, t). 4.2.11 Proposition Suppose X
= (Xl, ... , xn)
has a complete flow Ft' Then f(x, t)
= g(Ft(x»
is a solution of the foregoing problem (P). (See Exercise 4.2Cfor uniqueness.) Proof
Thus one can solve this scalar equation by computing the orbits of X and pushing (or "dragging along") the graph of g by the flow of X; see Fig. 4.2.3. These trajectories of X are called
characteristics of (P). (As we shall see below, the vector field X in (P) can be time dependent.)
leyel curves of g
0
F,
Figure 4.2.3 4.2.13 Example Solve the partial differential equation
:
= (x
+ y)
(~! - ~) ,
with initial condition f(x, y, 0) = x2 + y2. Solution The vector field X(x, y) = (x + y, - x - y) has a complete flow Ft(x, y) = «x + y)t + x, - (x + y)t + y), so that the solution of the previous partial differential equation is given by
Now we tum to the question of using the Lie derivative to characterize vector fields. We will prove that any derivation on functions uniquely defines a vector field. Because of this, derivations
Section 4.2 Vector Fields as Differential Operators
273
can be (and often are) used to define vector fields. (See the introduction to §3.3.) In the proof we shall need to localize things in a smooth way. hence the following lemma of general utility is proved first. 4.2.13 Lemma Let E be a C Banach space, i.e., one whose norm is Cr on E \ (OJ, r ~ 1. Let
V, be an open ball of radius r, about Xo and oz an open ball of radius rz. r, < r z. Then there is a C function h : E ~ lR such that h is one on V, and zero outside Oz. We call h a bump function. Later we will prove more generally that on a manifold M. if
oz are two open sets with V, and is zero outside U z. V, and
c1(U,)
U z. there is an hE .rr(M) such that h is one on
C
Proof By a scaling and translation. we can assume that U, and U z are balls ofradii 1 and 3 and centered at the origin. Let 9: lR
~
9(x)
lR be given by
=exp(-l/(l -I x IZ»
and set
if I x I < 1 if I x I ~ 1.
9(x) = O. (See the remarks following 2.4.15.) Now set
r~
9(t) dt 9,(s)=---
1~ 9(t) dt C~ function and is 0 if s < -1. and 1 if s> 1. Let 9 z(s) = 9,(2 - s). so 9 2 is a function that is 1 if s < 1 and 0 if s > 3. Finally. let hex) = 9z(1I x II) . •
so 9 1 is a C~
The norm on a real Hilbert space is
C~
away from the origin. The order of differentiability
of the norms of some concrete Banach spaces is also known; see Bonic and Frampton [1966]. Yamamuro [1974]. and Supplement 5.5B. 4.2.14 Corollary Let M be a Cr manifold modeled on a C Banach space. If urn E Tm *M. then there is an f E .rr(M) such that df(m) = urn' Proof If M = E. so TmE ;: E. let f(x) = um(x). a linear function on E. Then df is constant and equals urn' The general case can be reduced to E using a local chart and a bump function as follows. Let
274
Chapter 4 Vector Fields and Dynamical Systems
on
U
on M\U. It is easily verified that f is 0' and df(m) = 4.2.15 (i)
~.•
Proposition Let M be a 0' manifold modeled on a 0' Banach space. The collection of operators Lx for X E Xr(M), defined on Cr(M, F) and taking values in 0'-l(M, F) forms a real vector space and an :fr(M)-module with (fLx)(g) = f(Lxg), and is isomorphic to Xr(M) as a real vector space and as an :fr(M)-module. In particular, Lx = 0
iff X
=0; (ii)
and LfX =fLx. Let M be any 0' manifold. If Lxf = 0 for all f E 0'(U, F), for all open subsets U
of M, then X =
o.
Proof (i) Consider the map cr: X f-+ Lx. It is obviously IR and .'fr(M) linear; i.e.
To show that it is one-to-one, we must show that Lx = 0 implies X = O. But if Lxf(m) = 0, then df(m) . X(m) = 0 for all f. Hence, um(X(m» = 0 for all ~
E
Tm *M by 4.2.14. Thus X(m)
= 0 by the Hahn-Banach theorem.
(ii) This has an identical proof with the only exception that one works in a local chart, so it is not necessary to extend a linear functional to the entire manifold M as in 4.2.14. Thus the condition on the differentiability of the norm of the model space of M can be dropped. _ 4.2.16 Derivation Theorem (i)
If M is finite dimensional and
C~,
the collection of all derivations on j\M) is a real
vector space isomorphic to X(M). In particular,for each derivation (ii)
e
there is a
unique X E X(M) such that e = Lx. Let M be a C~ manifold modeled on a C~ Banach space E, i.e., E has a C~ norm away from the origin. The collection of all (lR-linear) derivations on C""(M, F) (for all Banach spaces F) formv a real vector space isomorphic to X(M).
Proof We prove (ii) first. Let e be a derivation. We wish to construct X such that e = Lx. First of all, note that e is a local operator; that is, if h E C~(M, F) vanishes on a neighborhood V of m, then e(h)(m) = O. Indeed, let g be a bump function equal to one on a neighborhood of m and zero outside V. Thus h = (l - g)h and so e(h)(m) = e(l - g)(m) . hem)
+ e(h)(m)(l -
gem»~
= o.
(I)
Section 4.2 Vector Fields as Differential Operators
C~(U,
If U is an open set in M, and f E
275
F) define (9 I U)(f)(m) = 9(gf)(m), where g
is a bump function equal to one on a neighborhood of m and zero outside U. By the previous remark, (9 I U)(f)(m) is independent of g, so 9 I U is well defined. For convenience we write
9 = 9 I U. Let (U,
C~(M,
F) where
~
U'
C
E; we can
write, for x E U' and a =
(
=
(
f ~t
=
(
f
(
D(
This formula holds in some neighborhood
(2)
Jb D(
a)] dt. Applying 9 to (2)
at u = m gives 9f(m) = gem) . (9
(3)
Since 9 was given globally, the right hand side of (3) is independent of the chart. Now define X on U by its local representative X<jl(X) = (x, 9(
where x =
C~(M,
F), the local representative of Lxf is D(f 0
= D(f
0
= 9f(u).
Hence Lx = 9. Finally, uniqueness follows from 4.2.15. The vector derivative property was used only in establishing (1) and (3). Thus, if M is finite dimensional and 9 is a derivation on
:r (M),
we have as before n
feu)
= f(m) + g(u) . (
L
(
i=l
where !; E
:r (V)
and a = (a 1
, .•. ,
an).
Chapter 4 Vector Fields and Dynamical Systems
276
Hence (3) becomes 8f(m) =
~ . ~ a . £..J gi(m) 8(
i= I ax
and this is again independent of the chart. Now define X on U by its local representative (x. 8(
Lx
and 8 coincide on C-functions on M. In Supplement 4.2D we will prove that Lx and 8
are in fact equal on er-I-functions. but the proofrequires a different argument.. For finite-dimensional manifolds. the preceding theorem provides a local basis for vector fields. If (U.
o
These derivations are linearly independent with coefficients in .1\U). for if
. a
n L f ' - i =0. then i=1 ax
(
. a . . L f'-i J (x)=f)=O i= I ax n
for all j= 1..... n.
since (a/axi)xi = &ii' By 4.2.16. (a/ax i) can be identified with vector fields on U. Moreover. if X
E
X(M) has components Xl ..... X" in the chart
Lxf = X[f] =
i
i=1
Xi afi = ax
(i i=1
xi
-;J ax
f.
i.e .. X =
i
i=1
Xi
~. ax'
Thus the vector fields (ala xi). i = 1..... n form a local basis for the vector fields on M. It should be mentioned however that a global basis of X(M). i.e.. n vector fields. XI' .... Xn E X(M) that are linearly independent over l' (M) and span X(M). does not exist in general. Manifolds that do admit such a global basis for X(M) are called parallelizable. It is straightforward
Section 4.2 Vector Fields as DifJerentilll Operators
to
277
show that a finite-dimensional manifold is parallelizable iff its tangent bundle is trivial. For
example. it is shown in differential topology that
S3
is parallelizable but
S2
is not (see
Supplement 7.SA). This completes the discussion of the Lie derivative on functions. Turning to the Lie derivative on vector fields. let us begin with the following. 4.2.17 Proposition If X and Yare Cr vector fields on M. then
is a derivation mapping cr+I(M. F) to Cr-I(M. F). Proof More generally. let 9 1 and 9 2 be two derivations mapping C r+1 to Cr and C r to cr- I . Cr+1 to C r- I. Also. if f E Cr+I(M.
Clearly [91' 9 2 l = 9 1 09 2 - 92 0 9 1 is linear and maps F). g E Cr+I(M. G). and BE L(F. G; H). then [91' 92 l (B(f. g»
= (9 1 0 9 2)(B(f. g» - (9 2 0 9 1)(B(f. g»
= 9 1 (B(9l0. g) + B(f. 9 2 (g»}
- 9 2 {B(9 1(0. g) + B(f. 9 1(g»}
= B(9 1(9lO>. g) + B(9l0. 9 1(g» + B(9 1(0.9 2 (g»
+ B(f. 9 1(9 2 (g») - B(92 (9 1(0). g) - B(9 1(0. 9 2 (g» - B(9l0. 9 1(g» - B(f. 9 2 (9 1(g))) = B([91' 92 l (0. g) + B(f. [91' 9 2 l (g» .•
Because of 4.2.16 the following definition can be given. 4.2.18 Definition Let M be a manifold modeled on a C~ Banach space and X. Y E *~(M).
Then [X. Yl is the unique vector field such that L[x. Yj = [Lx Lyl. This vector field is also denoted Lx Y and is called the Lie derivative of Y with respect to X. or the Jacobi-Lie bracket of X and Y. Even though this definition is useful for Hilbert manifolds (in particular for finite-dimensional manifolds). it excludes consideration of Cr vector fields on Banach manifolds modeled on nonsmooth Banach spaces. such as LP function spaces for p not even. We shall. however. establish an equivalent definition. which makes sense on any Banach manifold and works for cr vector fields. This alternative definition is based on the following result. 4.2.19 Lie Derivative Formula for Vector Fields
and let X have (local) flow Ft' Then
Let M be as in 4.2.18. X. Y E *(M).
Chapter 4 Vector Fields and Dynamical Systems
278
(at those points where Ft is defined). Proof. If t = 0 this formula becomes
d -d t
It = 0 F *Y=LxY.
(4 )
t
Assuming (4) for the moment •
.! (F*Y) = dt t
I
I
d F*t+. Y = F*t-d d * * Y. -d F.Y=FtLx s ,=0 s .=0
Thus the formula in the theorem is equivalent to (4). which is proved in the following way. Both sides of (4) are clearly vector derivations. In view of 4.2.16. it suffices then to prove that both sides are equal when aeting on an aribtrary function f E C"'(M. F). Now
by 4.2.7(i). Using 4.2.10 and Leibniz' rule. this becomes X[Y[f]](m) - Y[X[f]](m) = [X. Y][f](m). _ Since the formula for
LxY
in Eq. (4) does not use the fact that the norm of E is
c~
away
from the origin. we can state the following definition of the Lie derivative on any Banach manifold
M. 4.2.20 Dynamic Definition of Jacobi-Lie bracket If X. Y
E
Xr(M). r ~ 1 and X hasflow
Ft' the Cr - 1 vector field Lx Y = [X. Y] on M defined by
d [X. Y] = dt
I
*Y)
t=O (Ft
is called the Lie derivative of Y with respect to X. or the Lie bracket of X and Y. From the point of view of this more general definition. 4.2.19 can be rephrased as follows. 4.2.21 Proposition Let X. Y
E
Xr(M). 1 :0::; r. Then
[X. Y] = Lx Y is the unique C r- 1 vector
field on M satisfying [X. Y][f] = X[Y[f]] - Y[X[f]]
Section 4.2 Vector Fields as Differential Operators
for all f
E
279
Cr+1(U, F), where U is open in M.
The derivation approach suggessts that if X, Y E Xr(M) then [X, Yj might only be Cr- 2 , since [X, Yj maps Cr + 1 functions to
cr- 1 functions, and differentiates them twice.
However
4.2.20 (and the coordinate expression (6) below) show that [X, Yj is in fact C r- 1• 4.2.22 Proposition The bracket [X, Yj on X(M), together with the real vector space structure X(M), form a Lie algebra. That is, (i) [, j is lR bilinear;
(ii) [X, Xj = 0 for all X
E
X(M);
(iii) [X, [Y, Zll + [Y, [Z, Xll + [Z, [X, Yjj = 0 for all X, Y, Z
E
X(M) (Jacobi identity).
The proof is straightforward, applying the brackets in question to an arbitrary function. Unlike X(M), the space Xr(M) is not a Lie algebra since [X, Yj Xr(M). (i) and (ii) imply that [X, Yj = -[Y, Xl. since [X
E
Xr-l(M) for
X, Y
E
+ Y, X + Yj = 0 = [X, Xj + [X, Yj +
[Y, Xj + [Y, Yj. We can describe (iii) by writing Lx as a Lie bracket derivation: Lx[Y' Zj = [ Lx Y, Zj + [Y, LxZj.
Strictly speaking we should be careful using the same symbol Lx for both definitions of Lxf and Lx Y. However, the meaning is generally clear from the context. The analog of 4.2.7 on the vector field level is the following. 4.2.23 Proposition (i)
Let
E
X(M). Then Lx : X(M) ~ X(M) is
natural with respect to push-forward by <po That is,
or [
X(M)
(ii)
--------XCN)
Lx is natural with respect to restrictions. That is,for U [X, Yjl U; or thetollowing diagram commutes:
C
M open, [X I u, Y I Uj =
Chapter 4 Vector Fields and Dynamical Systems
280
IV
X(M) --.....;...--....·-X(V)
!~"
Lx!
X(M) - - - - -.....-X(V)
IV Proof For (i). let f
E
<.F (V). V be open in N. and
E
V. By 4.2.7(i). for any Z
E
X(M) ((
=[X. Y][f
0
= X[Y[f 0
= X[(
0
= (
= [
= D(Y[f])(x) . X(x) - D(X[f])(x) . Vex). Now Y[f](x) = Df(x) . Vex) and its derivative may be computed by the product rule. The terms involving the second derivative of f cancel by symmetry of J)2f(x) and so we are left with Df(x)· {DY(x)· X(x) - DX(x)· Vex»). Thus the local representative of [X. Y] is [X. Y] = DY . X - DX . Y.
(5)
If M is n-dimensional and the chart
Section 4.2 Vector Fields as DifferentUd Operators
281
(6)
i.e. [X, Y] = (X . V)Y - (Y . V)X. Part (i) of 4.2.22 has an important generalization to cp-related vector fields. For this, however, we need first the following preparatory proposition. 4.2.24 Proposition Let cp : M ~ N be a cr map of c r manifolds, X E Xr-I(M), and X' E Xr-I(N). Then X
-'I'
X'
iff (X'[f]) cp = X[f cp] 0
0
for all f E 1'1(V), where V is open in
N. Proof
By definition, «X'[f])
0
cp)(m) = df(cp(m» . X'(cp(m». By the chain rule,
X[f 0 cp](m) = d(f 0 cp)(m) . X(m) = df(cp(m» . Tmcp(X(m». If X -'I' X', then Tcp
0
X = X' 0 cp and we have the desired equality. Conversely, if X[f 0 cp] =
(X'[f]) 0 cp for all f E 1'1(V), and all V open in N, choosing V to be a chart domain and f the pull-back to V of linear functionals on the model space of N, we conclude that an' (X' 0 cp)(m) = an' (Tcp 0 X)(m), where n = cp(m), for all an E Tn *N. Using the Hahn-Banach theorem, we deduce that (X' 0 cp)(m) = (Tcp 0 X)(m), for all mE M.• It is to be noted that under differentiability assumptions on the norm on the model space of N (as in 4.2.16), the condition "for all
f E 1'1(V) and all V
C
N" can be replaced by "for all
f E 1'1(N)" by using bump functions. This holds in particular for Hilbert (and hence for finite-dimensional) manifolds. 4.2.25 Proposition Let cp : M ~ N be a c r map of manifolds, X, Y E Xr-I(M), and X', Y' E Xr-I(N). If X -'I' x' and Y -'I' Y' , then [X, Y]
-'I'
Proof By 4.2.24 it suffices to show that ([X', Y'][f])
0
[X', Y']. cp = [X, Y][f 0 cp] for all f E 1'1(V),
where V is open in N. We have ([X', Y'][fJ) 0 cp = X'[Y'[f]]
0
cp - Y'[X'[f]]
0
cp = X[(Y'[f]) 0 cp] - Y[(X'[f]) 0 cp]
= X[Y[f 0 cp]]- Y[X[f 0 cp]] = [X, Y][f 0 cpl .• The analog of 4.2.9 is the following. 4.2.26 Proposition For every X E X(M), Lx is a derivation on (1'(M), X(M». That is, Lx is '!ft.-linear and Lx(fY) ='(LxOY + f(LxY). Proof For g E
C~(U,
E), where U is open in M, we have
282
Chapter 4 Vector Fields and Dynamical Systems
[X. ty][g] = Lx( LfYg) - LfYLxg = Lx(fLyg) - fLyLxg = (Lxf)Lyg + !Lx Lyg - fLyLxg· [X. ty] = (Lxf)Y + f [X. Y] by 4.2.1S(ii). •
so
Commutation of vector fields is characterized by their flows in the following way. 4.2.27 Proposition Let X. Y E Xr(M). r ~ I. and let Ft' G t denote their flows. The following are equivalent. (i) [X. Y] = 0;
(ii) Ft*Y = Y; (iii) Gt*X = X; (iv) F t 0 G s = G s 0 Ft· (In (ii) - (iv). equality is understood, as usual, where the expressions are defined.) Proof Ft 0 G s = G s 0 F t iff G s= F t 0 G s 0 Ft-I. which by 4.2.4 is equivalent to Y = Ft*Y; i.e .. (iv) is equivalent to (ii). Similarly (iv) is equivalent to (iii). If Ft*Y = Y. then
d [X. Y]=-d t
I t=O
F t*Y=O.
Conversely. if [X, Y] = Lx Y = O. then
so that Ft*Y is constant in t. For t = 0, however. its value is Y, so that Ft*Y = Y and we have thus showed that (i) and (ii) are equivalent. Similarly (i) and (iii) are equivalent. • Just as in 4.2.10. the formula for the Lie derivative involving the flow can be used to solve special types of first-order linear n x n systems of partial differential equations. Consider the first -order system:
with initial conditions yi(x. 0) = gi(x) for given functions )(i(x). gi(x) and scalar unknowns yi(x. t). i = 1•...• n. where x = (xl ..... xn). 4.2.28 Proposition Suppose X = (Xl •.... xn) has a complete flow Ft. Then letting Y = (yl . .... yn) and G = (gl •...• gn). Y = Ft*G is a solution ojthejoregaing problem (Pn). (See Exercise 4.2C for uni~ueness J.
Section 4.2 Vector Fields as Differential Operators
283
Proof
since Ft*X = X and Y = Ft*G. The expression in the problem (Pn ) is by (6) the i-th component of [X. YJ . • 4.2.29 Example Solve the system of partial differential equations:
ayl
ayl
ayl
I
2
at =
(x
ay2
ay2 ay2 (x + y) ~ - (x + y) ay + yl + y2
at = with initial conditions yl(x. y. 0)
+ y)
= x.
~ - (x
+ y)
ay - y
- Y
y2(x. y. 0) = y2. The vector field X(x. y)
= (x + y. -
x-
y) has the complete flow Ft(x. y) = «x + y)t + x. - (x + y)t + y). so that the solution is given by Vex. y. t)
=Ft*(x. y2);
i.e .•
=«x + y)t + x)(1 - t) - try - (x + y)tJ2 y2(x. y. t) = t«x+ y)t + x) + (t + 1)[y - (x + y)tJ2 .• yl(x. y. t)
In later chapters we will need a flow type formula for the Lie derivative of a time-dependent vector field. In §4.1 we discussed the existence and uniqueness of solutions of a time-dependent vector field. Let us formalize and recall the basic facts. 4.2.30 Definition A C time-dependent vector field is a C map X: lR x M
~
TM such that
X(t. m) E TmM for all (t. m) E lR x M; i.e .• Xt E Xr(M). where ~(m) = X(t. m). The time-dependentjlow or evolution operator Ft •s of X is defined by the requirement that t ...... Ft.s(m) be the integral curve of X starting at m at time t = s; i.e .. d dt Ft.s(m) = X(t. Ft.s(m»
and
Fs.s(m)
=
m.
By uniqueness of integral curves we have F t.s 0 Fs.r = Ft,r (replacing the flow property Ft+s = F t 0 F s)' and Ft.t = identity. It is customary to write Ft = Ft.D• If X is time independent. Ft.s = Ft_s' In general Ft*~ #~. However. the basic Lie derivative formulae still hold. 4.2.31 Theorem Let Xt E Xr(M). r 2': 1 for each t and suppose X(t. m) is continuous in (t. m). Then F t.s is of class C andfor f E Cr+I(M. F). and Y E Xr(M). we have
284
Chapter 4 Vector Fields and Dynamical Systems d F t*sf = F t*s(Lx f), and -d t· . (
(i)
(ii)
Proof That Ft.s is
cr
was proved in §4.1. The proof of (i) is a repeat of 4.2.10:
d * d dFts(m) -dt(F·t sf)(m) = -dt (f 0· Ft s)(m) = df(Ft.s(m» = df(Ft .s(m» . Xt(Ft .s(m» t
d
For vector fields, note that by 4.2.7(i),
(F~sY)[fl = F~s(Y[F~sfl) since F s.t = Ft.s-I. The result (ii) will be shown to follow from (i), (7), and the next lemma. 4.2.32 Lemma The/ollowing identity holds:
Proof Differentiating Fs.t 0 Ft.s =identity in t, we get the backward differential equatWn: d dt Fs.t=-TFs.t
0
X
Thus d Fsi(m) * dt = -df(Fs.t(m» . TFs.t(Xt(m»
= - df(f 0 Fs.t) . Xt(m) = -Xt[f 0 Fs.t](m). '" Thus from (7) and (i),
By 4.2.21 and (7), this equals (Ft.s*[X t, Y])[f] . • If f and Y are time dependent, then (i) and (ii) read
(8)
and
Section 4.2 Vector Fields as DifferentUd Operators
285
(9)
Unlike the corresponding fonnula for time-independent vector fields. we generally have
Time-dependent vector fields on M can be made into time-independent ones on a bigger manifold. Let t E X(IR x M) denote the vector field which is defined by t(s, m) T(s.mlIR
X
M) ;: Ts IR
where X/(t, m)
= «t,
X
TmM. Let the suspension of X be the vector field X'
l), X(t, m»
and observe that X'
= t + X.
Since b: I
~
=«s, 1), Om) E E
X(IR x M)
M is an integral
curve of X at miff b/(t) = X(t, bet»~ and b(O) = m, a curve c: I ~ IR x M is an integral curve of X' at (0, m) iff c(t) = (t,
bet»~.
Indeed, if c(t)
= (a(t), b(t»
then c(t) is an integral curve of
X' iff c/(t) = (a/(t), b/(t» = X/(c(t»; that is a/(t) = 1 and b/(t) = X(a(t),
bet»~.
Since a(O) = 0,
we get aCt) = t. These observations are summarized in the following (see Fig. 4.2.4).
integral curve
of X
Figure 4.2.4 4.2.33 Proposition
Let X be a Cr-time-dependent vector field on M with evolution operator
Ft,s' The flow F t of the suspension X'
E
Xr(IR x M) is given by Ft(s, m) = (t + s, F t + s,,(m».
Proof In the preceding notations, bet) = Ft,o(m), c(t) = Ft(O, m) = (t, Ft,o(m», and so the statement is proved for s = O. In general, note that Ft(s, m) = Ft + s(O, Fo sCm»~ since t 1-+ Ft+s(O, FO,s(m» is the integral curve of X', which at t = 0 passes through f.s(O, FO,s(m» = (s, (Fs,o 0 Fo)m» = (s, m). Thus Ft(s, m) = Ft + s(O, Fo,,(m» s, F, + s,s(m». _
= (t + s, (F, + s,O 0 Fo)(m»
= (t +
Chapter 4 Vector Fields and Dynamical Systems
286
II@f'
Supplement 4.2A
Product formulas for the Lie bracket This box is a continuation of Supplement 4.1A and gives the flow of the Lie bracket [X, Y] in terms of the flows of the vector fields X, Y E ~(M). 4.2.34 Proposition Let X, Y
E
~(M) have flows F t and Gt" If B t denotes the flow of [X,
YJ, thenfor x E M,
Bt(x) = lim (G_{ii;
0
F_vtt;;'
0
G-rv;;
0
F{ii; )n(x), t ~
o.
n->~
Proof Let
The claimed formula follows from 4.1.24 if we show that
aaE K£(x) I£=0 = [X, Y](x) for all x E M. This in turn is equivalent to
aea K£*f I£= 0 = [X, Y](x) for any f E JrM). By the Lie derivative formula,
By the chain rule, the limit of this as E.l- 0 is half the
a/a·olE-derivative of the parenthesis at
E=
O. Again by the Lie derivative formula, this equals a sum of 16 terms, which reduces to the expression [X, Y][f]. • For example, for n x n matricess A and B, 4.2.34 yields the classical formula e[A,BJ = lim (e- NVn e-B/Vn e NVn eB/..Jn)n , n->~
where the commutator is given by [A, B] = AB - BA. To see this, define for any n x n matrix C a vector field Xc E ~(lRn) by Xc(x) = Cx. Thus Xc has flow Ft(x) = etCx. Note that Xc
Section 4.2 Vector Fields as Differential Operators
287
is linear in C and satisfies [X A. XB ] = - X IA . BI as is easily verified. Thus the now of [X B• XA] is ellA. BI and the formula now follows from 4.2.34. The results of 4.1.27 and 4.2.34 had their historical origins in Lie group theory. where they are known by the name of exponential formulas.
The converse of 4.1.25. namely
expressing elAe tB as an exponential of some matrix for sufficiently small t is the famous Baker-Campbell-Hausdorff formula (see e.g .. Varadarajan [1974. §2.15]). The formulas in
4.1.25 and 4.2.34 have certain generalizations to unbounded operators and are called Trotter product formulas after Trotter [1958]. See Chorin et. al. [1978] for further information. IlJJ
~ Supplement 4.2B The Method oj Characteristics
The method used to solve problem (P) also enables one to solve first-order quasi-linear partial differential equations in jRn. Unlike 4.2.11. the solution will be implicit, not explicit. The equation under consideration in jRn is n
L i=1
i I n af I n X (x ..... x • f) - . = Y(x ..... x • f). ax'
(Q)
where f = f(xl ..... xn) is the unknown function and Xi. Y. i = 1..... n are 0- real-valued functions on jRn+l. r ~ 1. As initial condition one takes an (n-I)-dimensional submanifold [" in jRn+1 that is nowhere tangent to the vector field n
L i= I
called the characteristic vector field of (Q). Thus. if [" is given parametrically by
this requirement means that the matrix Xl
Xn
Y
axl
ax n
af
at-;-
at-;-
at;"
axl dt n _ 1
ax n af atn_1 at n _ 1
288
Chapter 4 Vector Fields and Dynamical Systems
has rank: n. It is customary to require that the determinant obtained by deleting the last column be #
O. for then. as we shall see. the implicit function theorem gives the solution. The function f
i~
found as follows. Consider Ft' the flow of the vector field Li =1 ..... nXia/axi + ya/af in ]Rn+1 and let S be the manifold obtained by sweeping out r by Ft. That is. S = U (Ft(r) I t E lR}. The condition that Li =1 ..... nXia/axi + ya/af never be tangent to r insures that the manifold r is "dragged along" by the flow F t to produce a manifold of dimension n. If S is described by f = f(x I•.... xn) then f is the solution of the partial differential equation. Indeed. the tangent space to S contains the vector Li =1 ..... nXia/axi + ya/af; i.e .• this vector is perpendicular to the normal (af/ax l •... af/axn. - 1) to the surface f= f(xl ..... xn) and thus (Q) is satisfied. We work parametrically and write the components of F t as xi = xi(tl' ...• tn_I' t). i = 1• ...• n and f = f(tl' ...• tn_I' t). Assuming that
0#
Xl
Xn
axl
ax n
~
~
axl alo_ 1
ax n alo_ 1
ax n
axl
at
at
axl
ax n
~
at;""
axl
ax n alo_ 1
alo_ 1
one can locally invert to give t = (xl •... , xn). ti = ti (xl •...• xn). i = 1, ...• n - 1. Substitution into f yields f = f (x I•.... xn). The fundamental assumption in this construction is that the vector field Li =1 ..... nJ0a/axi + ya/af is never tangent to the (n - I)-manifold r. The method breaks down if one uses manifolds r. which are tangent to this vector field at some point. The reason is that at such a point. no complete information about the derivative of f in a complementary (n - I)-dimensional subspace to the characteristic is known. 4.2.35 Example Consider the equation in ]R2 given by
with initial condition r = «x. y. f) I x = s. y = (I/2)s2 -
l a~ ds~I=II as
s
1 s-I
S.
f = s). On this one-manfold
1=-1#0
so that the vector field a/ax + fd/ay + 3a/df is never tangent to r. Its flow is Ft(x. y. f) = (t + X. (3/2)t2 + ft + y. 3t + f) so that the manifold swept out by r along Ft is given by x(t. s) = t + s. y(t. s) = (3/2)t 2 + st + 0/2)S2 - s. f(t. s) = 3t + s. Eliminating t. s we get
Section 4.2 Vector Fields as Differential Operators
f = x-I
±
289
J1 - 2x2 + 4x + 4y.
The solution is defined only for 1 - 2x2 + 4x + 4y ~
o. +
Another interesting phenomenon occurs when S can no longer be described in terms of the graph of f; e.g .• S "folds over." This corresponds to the formation of shock waves. Further information can be found in Chorin and Marsden [1979], Lax [1973]. Guillemin and Sternberg
fl:JJ
[1977], John [1975], and Smoller [1982].
~ Supplement 4.2C Automorphisms of Function Algebras
The property of flows corresponding to the derivation property of vector fields is that they are algebra preserving
In fact it is obvious that every diffeomorphism induces an algebra automorphism of
'.F (M).
The
following theorem shows the converse. (We note that there is an analogous result of Mackey [1962] for measurable functions and mesureable automorphisms.) 4.2.36 Theorem Let M be a paracompact second· countable finite dimensinal manifold. Let a:
'.F (M) ~ '.F (M) be an invertible linear mapping that satisfies (M). Then there is a unique C'" diffeomorphism
<jl:
M
~
a(fg) = a(f) a(g) for all f. g
M such that a(f) = f 0
E
'.F
<jl.
Remarks
A There is a similar theorem for paracompact second-countable Banach manifolds; here we assume that there are invertible linear maps a p : C~(M. F)
~
C"'(M. F) for each Banach space F
such that for any bilinear continuous map B : F x G ~ H we have aH(B(f. g» = B(ap(f). aG(g» for f E C~(M. F) and g <jl:
M
~
E
C~(M. G). The conclusion is the same: there is a C~ diffeomorphism
M such that aF(f) = f 0
<jl
for all F and all f E
C~(M.
G). Alternative to assuming this
for all F. one can take F = 1R and assume that M is modelled on a Banach space that has a norm that is
C~
away from the origin.
We shall make some additional remarks on the
infinite-dimensional case in the course of the proof. B Some of the ideas about partitions of unity are needed in the proof. Although the present proof is self-contained. the reader may wish to consult §S.6 simultaneously. C In Chapter 5 we shall see that finite-dimensional paracompact manifolds are metrizable. so by 1.6.14 they are automatically second countable. + Proof (Uniqueness). We shall first construct a
C~
function X : M
~
lR which takes on the
290
Chapter 4 Vector Fields and Dynamical Systems
values 1 and 0 at two given points m\, m 2 E M, m]
*
m2. Choose a chart (U,
function as in Lemma 4.2.13. Then the function X : M ~ IR given by
X={9
o
0,
on U on M\U
is clearly C~ and x(m]) = 1, x(m 2 ) =O. Now assume that
*
M. Then there is a point m E M such that
P(f)
E
M such that
= f(m).
*
Proof (Following suggestions of H. Bercovici). Uniqueness is clear, as before, since for m] m 2 there exists a bump function f E J'(M) satisfying f(m]) = 0, f(m 2) = 1. To show existence, note first that P(I) = 1. Indeed P(1) = p(12) = P(l)P(1) so that either
p(1) = 0 or p(1) = 1. But P(l) = 0 would imply P is identically zero since P(f) = p(1 . f) = pel) . P(f), contrary to our hypotheses. Therefore we must have p(1) = 1 and thus P(c) = c for c E R For mE M, let Ann(m) = (f E J'(M) I f(m) = OJ. Second, we claim that it is enough to show that there is an m E M such that ker P = (f E J'(M) I P(f) = 0 j = Ann(m). Clearly, if P(f) = f(m) for some m, then ker P = Ann(m). Conversely, if this holds for some mE M and f i"
ker p, let c = P(f) and note that f - c E ker P = Ann(m), so f(m) = c and thus P(f) = f(m) for all f E J'(M). To prove that ker P =Ann(m) for some mE M, note that both are ideals in J'(M); i.e., if f E ker P, (resp. Ann(m)), and g E J'(M), then fg E ker P (resp. Ann(m)). Moreover, both of
*
them are maximal ideals; i.e., if I is another ideal of J'(M), with I J'(M), and ker pel, (resp. Ann(m) c I) then necessarily ker P = I (resp. Ann(m) = I). For ker P this is seen in the following way: since IR is a field, it has no ideals except 0 and itself; but P(I) is an ideal in IR, so P(I) = 0, i.e., 1= ker P, or P(I) = IR = P(J'(M)). If P(I) = IR, then for every f E J'(M) there exists gEl such that f - g E ker pel and hence f E g + I C I; i.e., I = J'(M). Similarly, the ideal Ann(m) is maximal since the quotient .'l{M)/Ann(m) is isomorphic to R Assume that ker P * Ann(m) for every mE M. By maximality, neither set can be included in the other, and hence for every m E M there is a relatively compact open neighborhood Urn of m and frn E ker P such that frn I Urn > O. Let Vrn be an open neighborhood of the closure, cl(Um ). Since M is paracompact, we can assume that (Vrn I mE M) is locally finite. Since M is second countable, M can be covered by (VrnG) I j EN}. Let fj =frnG ) and let Xj be bump
Section 4.2 Vector FiRlds as DijJerentinl Operators
291
functions which are equal to 1 on cl(Um(j) and vanishing in M \ Vm(j)' If we have the inequality an < 1/[n 2 sup {xn(m)f/(m) 1m E M}], then the function f = n", anXnfn2 is C~ (since the sum is finite in a neighborhood of every point), f> 0 on M, and the series defining f is uniformly convergent, being majorized by L.n", n-2. If we can show that 13 can be taken inside
L
the sum, then 13(f) = O. This construction then produces f E ker 13, f> 0 and hence 1 = (lff)f E ker 13; i.e., ker 13 = .r(M), contradicting the hypothesis 13 '# O. To show that 13
can be taken inside the series, it suffices to prove the following
"g-estimate:" for any g E 1(M), II3(g)1 :::; sup{ I gem) II mE M}. Once this is done, then
as n
~
00
by uniform convergence and boundedness of all functions involved. Thus 13(f) =
L.n", l3(anXnfn 2). To prove the g-estimate, let A > sup {I gem) II m EM} so that A ± g '# 0 on M; i.e., A ± g are both invertible functions on M. Since 13 is an algebra homomorphism, 0 I3(A ± g) = A ± l3(g). Thus ± l3(g)
'#
'#
A for all A > sup {I gem) II m EM}. Hence we get the
estimate II3(g) I:::; sup{ Ig(m) II mE M}. T Remark For infinite-dimensional manifolds, the proof of the lemma is almost identical, with the following changes: we work with 13:
C~(M,
F)
~
F, absolute values are replaced by norms,
second countability is in the hypothesis, and the neighborhoods V m are chosen in such a way that fm I V m is a bounded function (which is possible by continuity of fm ) .• Proof of existence in 4.2.36 For each m E M, define the algebra homomorphism 13m : .r(M) lR. by
I3m (f) =
a(f)(m). Since a is invertible, a(1)
'#
~
0 and since a(1) = a(l2) = a(1) a(l),
we have a(1) = 1. Thus 13 m '# 0 for all mE M. By 4.2.37 there exists a unique point, which we call
~
Supplement 4.2D
Derivations on
cr Functions
This supplement investigates to what extent vector fields and tangent vectors are characterized by their derivation properties on functions, if the underlying manifold is finite
Chapter 4 Vector Fields and Dynamical Systems
292
dimensional and of a fmite differentiability class. We start by studying vector fields. Recall from 4.2.9 that a derivation 9 is an lR-linear map from j"k+I(M) to j"k(M) satisfying the Leibniz
rule, i.e., 9(fg) = meg) + ge(f) for f, g
j"k+I(M), if the differentiability class of M is at least
E
k + 1. 4.2.38 Theorem (A. Blass) Let M be a Ck+2 finite-dimensional manifold, where k ~ O. The
collection of all derivations 9 from j"k+I(M) to j"k(M) is isomorphic to Xk(M) as a real vector space. Proof By the remark following 4.2.16, there is a unique Ck vector field X with the property that 9 I j"k+2(M)
=Lx I j"k+2(M).
Thus, all we have to do is show that 9 and Lx agree on Ck +1 functions. Replacing 9 with 9 - Lx, we can assume that 9 annihilates all Ck +2 functions and ChI functions. As in the proof of 4.2.16, it suffices
we want to show that it also annihilates all
to work in a chart, so we can assume without loss of generality that M = lRn. Let f be a Ck+1 function and fix p E lRn. We need to prove that (9f)(p)
= O.
For
simplicity, we will show this for p = 0, the proof for general p following by centering the following arguments at p instead of O. Replacing f by the difference between f and its Taylor polynomial of order k + 1 about 0, we can assume f(O)
=0,
and the first k + 1 derivatives
vanish at 0, since 9 evaluated at the origin annihilates any polynomial. We shall prove that f = g
+ h, where g and h are two ChI functions, satisfying g I U are open sets such that 0
I V = 0, (9h)(0)
E
c1(U)
n c1(V).
whence by continuity 9g I c1(U)
=0
=0
and h I V
= 0,
where U, V
Then, since 9 is a local operator, 9g I U = 0 and 9h
=0
and 9h I c1(V)
= O.
Hence (9f)(0)
= (9g)(0) +
and the theorem will be proved.
4.2.39 Lemma Let
C
lRn -t lR be a c~ function and denote by 1t: lRn \ (0) -t sn-I,
1t(x) = x / II x II the radial projection. Thenfor any positive integer r, DT(
0
1t)(x) = (0/ 0 1t)(x IIxliT
for some C~ function \jI: sn-I -t Us(lRn; lR). In particular Dk(
O.
Proof For r = 0 choose
v
D 1t(x) . V = - - - DII . II (x) . v + II X 112 II x II
and so D (
0
1t)(x) . v = _1_ D
(I -
_1_DII' II II xII
(X») . v.
II
293
Section 4.2 Vector Fields as Differential Operators
~'(x) = (1~1 xlDOIl .
But the mapping
II (x) satisfies
uniquely determined by ~ = ~' I S"-l. Hence O(ep Oep (y) . (I - ~(y», y
0,
°
E
0
~'(tx) = ~'(x) for all t>
°
so that it is
7t)(x) = (1/ IIxID(1jI 0 7t)(x), where 'V(y) =
S"-l. Now proceed by induction.
T
Returning to the proof of the theorem, let f be as before, i.e., of class Ch1 and Oif(O) =
~
i
~
k + I, and let
°
see that Oif(x) = lemma,
O\f' (
L i+I
so that Oi(f· (
0
7t»,
o( Ilxll k+1-i )0(lIxll- l ) = o(lIxll)k+ 1 -i)
=i
°
~ i ~ k + 1, can be continuously extended to 0, by making them vanish
at 0. Therefore f· (g o7t) is Ck+1 for all }R". Now choose the Coo function
n
on an open set 02 of S"-l, 01 02 = 0. Then the continuous extension g of f· (ep 0 7t) to }R" is zero on U = 7t- 1(01) and agrees with f on V = 7t- 1(02)' Let h = f - g and thus f is the sum of two Ck+1 functions, each of which vanishes in an open set having in its closure. This
°
completes the proof. • We do not know of an example of a derivation not given by a vector field on a C1-manifold. In infinite dimensions, the proof would require the norm of the model space to be Coo away from the origin and the function IjI in the lemma bounded with all derivatives bounded on the unit sphere. Unfortunately, this does not seem feasible under realistic hypotheses. The foregoing proof is related to the method of "blowing-up" a singularity; see for example Takens [1974] and Buchner, Marsden, and Schecter [1982]. There are also difficulties with this method in infinite dimensions in other problems, such as the Morse lemma (see Golubitsky and Marsden [1983] and Buchner, Marsden, and Schecter [1983]). 4.2.40 Corollary Let M be a C k+1 finite-dimensional manifold. Then the only derivative/rom ,rk+1(M) to ,rk(M), where I ~ k < 00, is zero. Proof By the theorem, such a derivation is given by a Ck- 1 vector field X. If X", 0, then for some f E ,rk+1(M), X[f] is only C k- 1 but not C k.• Next, we tum to the study of the relationship between tangent vectors and germ derivations. On
,r k(M)
consider the following equivalence relation: f - m g iff f andg agree on some
neighborhood of m
E
M. Equivalence classes of the relation -mare called germs at m; they
form a vector space denoted by ,rkm(M). The differential d on functions clearly induces an lR-linear map, denoted by d m on ,rkm(M) by dmf = df(m), where we understand f on the left hand side as a germ. It is straightforward to see that d m : ,r k m(M) ~ Tm *M is }R-linear and satisfies the Leibniz rule. We say that an lR-linear map em: ,rkm(M) ~ E, where E is a
294
Chapter 4 Vector Fields and Dynamical Systems
Banach space, is a germ derivaiWn if 8m satisfies the Leibniz rule. Thus, d m is a Tm *M-valued genn derivation. Any tangent vector vm
E
TmM defines an lR-valued genn derivation by vm[f] = < df(m),
vm ). Conversely, localizing the statement and proof of Theorem 4.2.16(i) at m, we see that on a Coo finite-dimensional manifold, any lR-valued germ derivation at m defines a unique tangent vector, i.e., T mM is isomorphic to the vector space of lR-valued germ derivations on ,'Fm(M). The purpose of the rest of this supplement is to show that this result is false if M
is a
Ck-manifold. This is in sharp contrast to Theorem 4.2.38. 4.2.41 Theorem (Newns and Walker [1956]) Let M be afinite dimensional Ck manifold, 1< k < 00. Then there are lR-valued germ derivations on ,'Fkm(M) which are not tangent vectors. In
fact, the vector space of all lR-valued germ derivations on ,'Fkm(M) is card(lR)-dimensional, where card(lR) is the cardinality of the continuum. For the proof, we start with algebraic characterizations of T mM and the vector space of all genn derivations. 4.2.42 Lemma
Let,'Fk m,O(M) = (f E ,'Fkm(M) I df(m) = O}. Then ,'Fk m(M) / ,'Fk m,O(M) is
isomorphic to T m*M. Therefore, since M is finite dimensional (,'Fkm(M) / ,'Fkm,O(M» * is isomorphic to TmM. Proof The isomorphism of ,'Fkm(M)/ ,'Fkm,o(M) with Tm*M is given by c1assof (f)l-+df(m); this is a direct consequence of Corollary 2.4.14. " 4.2.43 Lemma Let ,'Fkm,iM)
= span {I, fg.lJ, g E
,'Fkm(M), f(m)
= gem) = O}.
Then the space
of lR-linear germ derivations on ,'Fkm(M) is isomorphic to (,'Fkm (M) /,'Fkm,d(M»*.
= 0 and 8m(fg) = 0 for any f, g E ,'Fkm(M) with f(m) = gem) = 0, so that 8m defines a linear functional on ,'Fkm(M) which vanishes on the space ,'Fk m,iM). Conversely, if A is a linear functional on ,'Fkm(M) vanishing on ,'Fk m,iM), Proof Clearly, if 8m is a genn derivation 8m(l)
then A is a genn derivation, for if f, g E ,'Fk m(M), we have fg = (f - f(m»(g -
gem»~
+ f(m)g + g(m)f - f(m)g(m)
so that A(fg) = f(m) A(g) + gem) A(f), i.e., the Leibniz rule holds. " 4.2.44 Lemma All germs in ,'Fkm,iM) have k + 1 derivatives at m (even though M is only of
class Ck).
Section 4.2 Vector Fklds as Differential Operators
Proof Since any clement of .'Tkm,iM) is of the form a + bfg, f,g
E
295
.'T k m(M), f(m) = gem) = 0,
a, b E JR, it suffices to prove the statement for fg. Passing to local charts, we have by the Leibniz rule for k -I
I. (Dif)(Dk-ig) . i=1
Clearly
°
(M), but arc not in
.'T
k m,iM)
by 4.2.44, since they cannot be differentiated k + I times at m.
Therefore, .'T km,iM) is strictly contained in
.'T k m,O(M) and thus TmM is a strict subspace of the
vector space of germ differentations on .'T k m(M) by 4.2.42 and 4.2.43. Thc functions I xi Ik+f , < £ < 1 are linearly independent in .'TkmM modulo .'Tkm,d(M),
°
because only a trivial linear combination of such functions has derivatives of order k + I at m. Therefore, the dimension of
.'T km(M) / .'T km,iM) is at least card(JR), Since card(.'T km(M))
=
card(JR), it follows that dim(.'T km(M) / .'T k m,iM)) = card(JR). Consequently, its dual, which by
4.2.43 coincides with the vector space of germ-derivations at m, also has dimension card(JR). _
Exercises 4.2A
(i) On JR2, let
= (x, y; y, -
X(x, y)
x). Find the flow of X.
(ii) Solve the following for f(t, x, y):
if f(O, x, y)
4.28
=y sin x.
(i) Let X and Y be vector fields on M with complete flows Ft and 0t' respectively.
If [X, Yj = 0, show that X + Y has flow H t = Ft 0 0t' Is this true if X and Y are time dependent? (ii) Show that if [X, Yj
=
° for all Y
E
X(M), then X = 0. (Hint: From the local
formula conclude first that X is constant; then take for Y linear vector fields and apply the Hahn-Banach theorem. In infinite dimensions, assume the conditions hold locally or that the model spaces are
C~).
296
Chapter 4 Vector Fields and Dynamical Systems
4.2C Show that, under suitable hypotheses, that the solution [(x, t) = g(Ft(x)) of problem (P) given in 4.2.11 is unique. (Hint: Consider the function E(t) =
1 Rn
I fl (x, t) - fz(x,t) I z dx,
where fl and fz are two solutions. Show that dE/dt
:0;
aE for a suitable constant a
and conclude by Gronwall's inequality that E = O. The "suitable hypotheses" are conditions that enable integration by parts to be performed in the computation of dE/dL) Adapt this proof to get uniqueness of the solution in 4.2.28. 4.2D Let X, Y E X(M) have flows Ft and Gt' respectively. Show that d dd [X, Yj = d t
s
I
(F-t
0
Gs
0
t.s=O
F t )·
4.2E Show that SO(n) is parallelizable. See Exercise 3.SS for a proof that SO(n) is a manifold. (Hint: SO(n) is a group.)
4.2F Solve the following system of partial differential equations. ayl
ayl
at = (x + y) ax- + (4x ayZ
ay2
at = (x + y) a;- + (4x with initial conditions yl(x, y, 0) vector field (x + y, 4x - 2y) is (x, y)
1-+
ayl 2y)
= x + y,
ay - y
- y
ayZ 2y)
ay - 4Y
yZ(x, y, 0)
= x2
2
1
1
+ 2Y
Z
(Hint: The flow of the
1 -3t 1 Zt 4 -3t 1 Zt) ( S(x - y)e + S (4x + y) e ,- S (x - y) e + '5 (4x + y) e .
4.2G Consider the following equation for f(x, t) in divergence/orm:
ataf + ~ ax (H(f)) = 0
x
where H is a given function of f. Show that the characteristics are given by = -H'(f). What does the transversality condition discussed in Supplement 4.2B become in this case?
Section 4.2 Vector Fields as Differential Operators
297
4.2H Let M and N be manifolds with N modeled on a Banach space which has a Ck nonn away from the origin. Show that a given mapping
~
T(TM) satisfying sM
0
sM = identity on T(TM), TtM 0 sM = 'tTW and 'tTM 0 sM
= T'tw where 't M : TM~ M and 'tTM : T(TM)
~
TM are the canonical tangent bundle
projections. Let X, Y E X(M) and denote by TX, TY : TM ~ T(TM) their tangent maps. Prove the following fonnulae for the Lie bracket:
where the right hand sides, belonging to T X(m)(TmM) and T Y(m/TmM) respectively, are thought of as elements of T mM. (Hint: Show that Tm'tM of the right hand sides is zero which proves that the right hand sides are not just elements of T X(m)(TM) and T Y(m)(TM) respectively, but of the indicated spaces. Then pass to a local chart.)
~4.3
An Introduction to Dynamical Systems
We have seen quite a bit of theoretical development concerning the interplay between the two aspects of vector fields. namely as differential operators and as ordinary differential equations. It is appropriate now to look a little more closely at geometric aspects of flows. Much of the work in this section holds for infinite-dimensional as well as finite-dimensional manifolds. The reader who knows or is willing to learn some spectral theory from functional analysis can make the generalization. This section is intended to link up the theory of this book with courses in ordinary differential equations that the reader may have taken. The section will be most beneficial if it is read with such a course in mind. We begin by introducing some of the most basic terminology regarding the stability of fixed points. 4.3.1 Definition Let X be a C l vector field on an n-manifold M. A point m is called a critical
point (also called a singular point or an equilibirum point) of X if X(m) = O. The linearization of X at a critical point m is the linear mnp
defined by X/em) . v =
d~
I
(TFt(m)' v) t=O
where Ft is the flow of X. The eigenvalues (points in the spectrum) of X'(m) are called characteristic exponents of X at m. Some remarks will clarify this definition. Ft leaves m fixed: Ft(m) = m. since c(t);;; m is the unique integral curve through m. Conversely. it is obvious that if Ft(m) = m for all t. then m is a critical point. Thus TmFt is a linear map of TmM to itself and so its t-derivative at O. producing another linear map of T mM to itself. makes sense. 4.3.2 Proposition Let m be a critical point of X and let (U.
]Rn. Let x = (xl ..... xn) denote coordinates in ]Rn and Xi(xl •...• xn). i = 1....• n. the
components of the local representative of X. Then the matrix of X/em) in these coordinates is
299
Section 4.3 An Introduction to Dynamical Systems
Proof This follows from the equations
after differentiating in x and setting x =xo' t =O. • The name "characteristic exponent" arises as follows. We have the linear differential equation
and so
Here the exponential is defined. for example. by a power series. The actual computation of these exponentials is learned in differential equations courses. using the Jordan canonical form. (See Hirsch and Smale [1974], for instance.) In particular. if 111' ...• I1n are the characterisitc exponents of X at m. the eigenvalues of T mFt are elf!l •...• el).l.".
The characteristic exponents will be related to the following notion of stability of a critical point. 4.3.3 Definition Let m be a critical point of X. Then
if for any neighborhood U of m, there is a neighborhood Vof m such that if m' E V. then m' is + complete and Ft(m') E
(i)
m is stable (or Liapunov stable)
(ii)
m is asymptotically stable if there is a neighborhood V of m such that if m'
U for all t
~
O. (See Fig. 4.3.1(a).)
then m is + complete, Ft(V)
C
E
V.
Fs(V) if t > s and
lim Ft(V) = {m}, t---40+«>
i.e. Jar any neighborhood U of m. there is a T such that Ft(V)
C
U if t
~
T. (See
Fig. 4.3.1(b).) It is obvious that asymptotic stability implies stability. The harmonic oscillator x = -x giving a flow in the plane shows that stability need not imply asymptotic stability (Fig. 4.3.1(c». 4.3.4 Liapunov's Stability Criterion Suppose X is C 1 and m is a critical point of X.
Assume the spectrum of X'(m) is strictly in the left half plane. (In finite dimensions, the characteristic exponents of m have negative real parts.) Then m is asymptOlically stable. (In a similar way. if Re(l1i) > O. m is asymptotically unstable, i.e., asymptotically stable as t ~ -00.)
300
Chapter 4 Vector Fields and Dynamicol Systems
y
x
(b) Asymptotically stable
(a) Stable
(c) Harmonic oscillator
Figure 4.3.1 The proof we give requires some spectral theory that we shall now review. For the finite dimensional case, consult the exercises. This proof in fact can be adapted to work for many partial differential equations (see Marsden and Hughes [1983, Chapters 6, 7, and page 483 D. Let T: E
~
E be a bounded linear operator on a Banach space E and let o(T) denote its
{A Eel T - AI is not invertible on the complexification of E}. Then A E o(T), I A I ~ II T II. Let r(T) denote its spectral radius, defined by r(T) = sup { I A II A E o(T)} .
spectrum; i.e., o(T) =
o(T) is non-empty, is compact, and for
4.3.5 Spectral radius formula. r(T) = lim n-+oo
II ~ lI"n
The proof is analogous to the formula for the radius of convergence of a power series and can be supplied without difficulty; cf. Rudin [1973, p. 355]. The following lemma is also not difficult and is proved in Rudin [1973] and Dunford and Schwartz [1963].
4.3.6 Spectral Mapping Theorem. Let f(z) =
L 3nzn n=O
be an entire junction and define
Then o(f(T»
=f(o(T».
4.3.7 Lemma Let T : E
~
E be a bounded linear operator on a Banach space E. Let r be any
number greater than r(T), the spectral radius of T. Then there is a norm I I on E equivalent to the original norm such that I T I
~
r.
Proof From the spectral radius formula, we get sUPn ~ o( II Tn II Ir") < 00, so if we define I x I =
301
Section 4.3 An Introduction to Dynamical Systems
SUPD 2: O( II T D(X) II /rD), then I I is a norm and II x II ~ I x I ~ (SUPD 2: o( II TOil IrD» II x II. Hence I T(x) I = SUPD 2: o( II TD+I(x) II IrD) = r SUPD 2: o( II p+l(x) II I~+I) ~ r I x I .• 4.3.8 Lemma Let A: E ~ E be a bounded operator on E and let r > o(A) (i.e., Re(A.) > r). Then there is an equivalent norm
I I on E such that for t ~ 0, I etA I
Proof Using Lemma 4.3.6, ert is ~ spectral radius of etA; i.e., e rt ~ limn-->
00
if
A.
E
o(A),
~ e't.
II eDtA III/D. Set
and proceed as in Lemma 4.3.7.• There is an analogous lemma if r < o(A), giving
I etA I
~ e rt •
4.3.9 Lemma Let T: E ~ E be a bounded linear operator. Let oCT) oCT)
C
C
{z I Re (z) < O} (resp.
{z I Re (z) > 0 D. Then the origin is an attracting (resp. repelling) fixed point for the flow
{z IRe (z) < O} , there is an r < 0 with oCT) < r, as oCT) is compact. Thus
C
by lemma 4.3.8,
I etA I
~ e rt ~ 0 as t ~ +00. •
Proof of Liapunov's Stability Criterion (4.3.4) We can assume that M is a Banach space E and that m = O. As above, renorm E and find e > 0 such that II etA II ~ e-Et, where A = X'(O). From the local existence theory, there is an r-ball about 0 for which the time of existence is uniform if the initial condition Xo lies in this ball. Let R(x) = X(x) - DX(O) . x. Find r 2 ~ r such that
II x II ~ r2 implies II R(x) II
~
all x II, where a
=e 12.
Let D be the open r2/2 ball about O. We shall show that if Xo Curve starting at Xo remains in D and
~
0 exponentially as t
~
E
D, then the integral
+ 00. This will prove the
result. Let x(t) be the integral curve of X starting at xo. Suppose x(t) remains in D for 0 ~ t
< T. The equation x(t) = xo+
f:
X(x(s» ds = Xo +
f:
[Ax(s) + R(x(s»] ds
gives, by the variation of constants formula (Exercise 4.1E),
x(t) = etAxo + and so
f:
e(t-s)A R(x(s» ds ,
Chapter 4 Vector Fields and Dynamical Systems
302
II x(t) II :5 e-1£ II Xo II + a Letting f(t) = el£ II x(t)
£
e-(1-5)EII xes)
II ds.
II, the previous inequality becomes f(t):511 Xo II + a fol f(s) ds,
and so, by Gronwall's inequality, f(t):5
Hence x(t) E D,
II Xo II eat. Thus
°
:5 t < T, so as in 4.1.19, x(t) may be indefinitely extended in t and the
foregoing estimate holds. • One can also show that if M is finite dimensional and m is a stable equilibrium, then no eigenvalue of X'(m) has strictly positive real part; see Hirsch and Smale [1974, pp. 187-190] and the remarks below on invariant manifolds. See Hille and Phillips [1957] and Curtain and Pritchard [1974] for the infinite dimensional linear case. Another method of proving stability is to use Liapunov functions. 4.3.10 Definition Let X E Xr(M), r ~ 1, and let m be an equilibrium solution for X, i.e. X(m) = 0. A Liapunov function for X at m is a continuous function L: U ~ lR defined on a
neighborhood U of m, differentiable on U\ {m}, and satisfying the following conditions:
°
(i) L(m) = and L(m' ) > (ii) X[L]:50 on U\{m};
°if
m' #- m;
(iii) there is a connected chart
~
E where mE V
satisfying B/O) = {x E E III x II :5 e} inf {L(
C
C
U,
III x II
=
° e' <
°
:5 e,
e'} > 0.
The Liapunov function L s said to be strict, if(ii) is replaced by (ii') X[L] <
°
in U\ {m}.
Conditions (i) and (iii) are called the potential well hypothesis. In finite dimensions, (iii) follows automatically from compactness of the sphere of radius e' and (i). By the Lie derivative formula, condition (ii) is equivalent to the statement: L is decreasing along integral curves of X. 4.3.11 Theorem Let XE Xr(M), r ~ 1, and m be an equilibrium of X. If there exists a Liapunov function for X at m, then m is stable. Proof Since the statement is local, we can assume M is a Banach space E and m = 0. By
Section 4.3 An Introduction to Dynamical Systems
303
Lemma 4.1.8. there is a neighborhood V of 0 in E such that all solutions starting in V exist for time t E [ - O. 01. with 0 depending only on X and V. but not on the solution. Now fix E > 0 as in (iii) such that the open ball D[(O) is included in V. Let p(E) > 0 be the minimum value of L on the sphere of radius E. and defme the open set V' = {x E D[(O) I L(x) < p(E)}. By (i). V' ;f. 0. 0 E V'. and by (ii). no solution starting in V' can meet the sphere of radius E (since L is decreasing on integral curves of X). Thus all solutions starting in V' never leave D[(O) C V and therefore by uniformity of time of existence, these solutions can be extended indefinitely in steps of 0 time units. This shows 0 is stable .• Note that if E is fmite dimensional. the proof can be done without appeal to 4.1.8: since the closed E-ball is compact. solutions starting in V' exist for all time by 4.1.19. 4.3.12 Theorem Let X
E
*r(M). r ~ 1. m be an equilibrium of X. and L a strict Liapunov
junction for X at m. Then m is asymptotically stable
if anyone of the following conditions
hold: (i) M isfinite dimensional; (ii) solutions starting near m stay in a compact set (i.e., trajectories are precompact);
(iii) in a chart
--?
E satisfying (iii) in definition 4.3.10 the following inequality is
validfor some constant a> 0 X[L](x)
~
- a
II X(x) II .
Proof We can assume M = E. and m = O. By 4.3.11. 0 is stable. so if tn is an increasing ~ --?
sequence.
00. and x(t) is an integral curve of X starting in V' (see the proof of 4.3.11),
the sequence (x(tn)} in E has a convergent subsequence in cases (i) and (ii). Let us show that under hypothesis (iii). (x(tn)} is Cauchy. so by completeness of E it is convergent. For t > s, the inequality L(x(t» - L(x(s» = f X[L] (X(A» dA
~-
af
II X(X(A» II < 0,
implies that L(x(s» - L(x(t»
~
af
II X(X(A» II dA
~
a Ilf X(X(A» dA II = all x(t) - x(s)
II •
which together with the continuity of A ...... L(X(A» shows that (x(tn)} is a Cauchy sequence.
+00 such that x(~) --? Xo E D[(O), D[(O) being =O. Since L(x(t» is a strictly decreasing function of t by (ii'). L(x(t» > L(xo) for all t > O. If Xo;f. O. let c(t) be the solution of X starting at xo' so that L(c(t» < L(x o)' again since t ...... L(x(t» is strictly decreasing. Thus. for any solution c(t) starting close to xo. L(c(t» < L(x o) by continuity of L. Now take 7:(0) = x(~) for n large to get the contradiction L(x(tn + t» < L(xo)' Therefore Xo = 0 is the only limit Thus. in all three cases, there is a sequence tn
--?
given in the proof of 4.3.11. We shall prove that Xo
Chapter 4 Vector Fields and Dynamical Systems
304
point of {x(t) I t ~ O} if x(O)
E
V', i.e., 0 is asymptotically stable. _
The method of Theorem 4.3.12 can be used to detect the instability of equilibrium solutions. 4.3.13 Theorem Let m be an equilibrium point of X
E
Xr(M), r ~ 1. Assume there is a
continuous function L: V ~ M defined in a neighborhood of V of m, which is differentiable on V\ {m}. and satisfyies L(m) = 0, X[L] ~ a > 0 (respectively ~ a < 0) on V\ {m}. [fthere exists a sequence mk ~ m such that L(mk ) > 0 (respectively < 0), then m is unstable. Proof. We need to show that there is a neighborhood W of m such that for any neighborhood V of m, V C V, there is a point mv whose integral curve leaves W. Since m is an equilibrium, by Corollary 4.1.25, there is a neighborhood WI C V of m such that each integral curve starting in WI exists for time at least l/a. Let W = {m E WI I L(m) < 1/2}. We can assume M
=E,
and m
=O.
Let cn(t) denote the integral curve of X with initial condition
~ E
W. Then
so that
i.e. cn(l/a) Ii!' W. Thus all integral curves starting at the points mn E W leave W after time at most l/a. Since mn ~ 0, the origin is unstable .• Note that if M is finite dimensional, the condition X[L]
~
a > 0 can be replaced by the
condition X[L] > 0; this follows, as usual, by local compactness of M. 4.3.14 Examples A The vector field X(x, y) = (-y - x5)(afi)x) + (x - 2y3)(a/ay) E X(lR2) has the origin as an isolated equilibrium. The characteristic exponents of X at (0, 0) are ± i and so Liapunov's Stability Criterion 4.3.4 does not give any information regarding the stability of the origin. If we suspect that (0, 0) is asymptotically stable, we can try searching for a Liapunov function of the form L(x, y) =ax2 + by2, so we need to determine the coefficients a, b, 0 in such a way that
*'
X[L] < O. We have X[L]
=2ax(-y -
x5 ) + 2by(x - 2y3)
=2xy(b -
a) - 2ax 6 - 4by4,
*'
so that choosing a = b = I, we get X[L] =-2(x6 + 2y4) which is strictly negative if (x, y) (0, 0). Thus the origin is asymptotically stable by Theorem 4.3.12. B Consider the vector field X(x, y) =(-y + x5 )(afi)x) + (x + 2y3)(a/ay) with the origin as an isolated critical point and characteristic exponents ± i. Again 4.3.4 cannot be applied, so that we search for a function L(x, y) = ax 2 + by2, a, b 0 in such a way that X[L] has a definite
*'
Section 4.3 An Introduction to Dynamical Systems
305
sign. As above we get X[L] = 2ax(-y + xs) + 2by(x + 2y3) = 2xy(b - a) + 2ax 6 + 4by 4,
so that choosing a = b = 1, it follows that X[L]
= 2(x6 + y4)
> 0 if (x, y) *- (0. 0). Thus by
Theorem 4.3.13, the origin is unstable. These two examples show that if the spectrum of X lies on the imaginary axis, the stability nature of the equilibrium is determined by the nonlinear terms.
C Consider Newton's equations in ]R3,
it = v, v= -(I/m)V'V(q)
q =-(1/m)V'V(q)
written as a first order system
and so define a vector field X on]R3 x ]R3. Let (qo' Yo) be an
. equilibrium of this system, so that Vo = 0 and V'V(qo) = O. In Example 4.1.23B we saw that the total energy E(q, v)
= (l/2)m II v 112 + V(q)
is conserved, so we try to use E to construct a
Liapunov function L. Since L(qo' 0) = 0, define
1 2 L(q, v) = E(q,v)-E(qo,O) = Imllvll +V(q)-V(qo),
which satisfies X[L]
=0
by conservation of energy. If V(q) > V(qo) for q *- qo' then L is a
Liapunov function. Thus we have proved Lagrange's Stability Theorem: an equilibrium point (qo' 0) of Newton's eqUfJtions for a particle of mass m, moving under the influence of a potential
V, which has a local absolute minimum at qo' is stable. D Let E be a Banach space and L: E ~]R be C2 in a neighborhood of
o. If DL(O) =0
and there is a constant c > 0 such that D2L(0)(e, e) > c II e 112 for all e, then 0 lies in a potential well for L (i.e. (ii) and (iii) of 4.3.10 hold). Indeed, by Taylor's Theorem 2.4.15,
Thus, if 0 > 0 is such that for all
II h II < 0, 10(h2) 1 ~ cll h 112/4, then L(h) - L(O) > cll h 112/4,
i.e. inf ll h II = e[L(h) - L(O)] ? cE/4 for £ < O.• In many basic infinite dimensional examples, some technical sharpening of the preceding ideas is necessary for them to be applicable. We refer therader to LaSalle [1976], Marsden and Hughes [1983, §6.6], Hale, Magalhaes, and Oliva [1984] and Holm, Marsden, Ratiu, and Weinstein [1985] for more information. Next we turn to cases where the equilibirum need not be stable. A critical point is called hyperbolic or elementary if none of its characteristic exponents has zero real part. A generalization of Liapunov's theorem called the Hartman-Grobman theorem shows that near a hyperbolic critical point the flow looks like that of its linearization. (See Hartman [1973, Ch.9] and Nelson [1979, Ch. 3]. for proofs and discussions.) In the plane, the possible hyperbolic flows near a critical point are summarized in the table below and shown in Fig. 4.3.2. (Remember that for real systems, the characteristic exponents occur in conjugate pairs.)
306
Chapter 4 Vector Fields and Dynamical Systems
Eigenvalues
Real Jordan Form
Name
Phase portrait in a coordinate system (yl, y2) adapted to the Jordan form (Figure 4.3.2)
Al < A2 < 0 Al
= A2 < 0
Al
=~ < 0
Al
= a + ib, a < 0
~
= a-ib,
Al = ib, A2 b;toO
saddle
(a)
stable focus
(b)
stable node
(c)
[AI 0] 1 A2
stable improper node
(d)
[: ~b]
stable spiral sink
(e)
center
(f)
[AI 0] o A2
Al < 0 < A2
b;toO
=-
ib,
[~
-;]
Table 4.3.1
In cases 1 to 5, all arrows in the phase portraits are reversed and "stable" is replaced by "unstable," if the signs of AI'~' are changed. In the original coordinate system (x I ,x2) all phase portraits in Figure 4.3.2 are rotated and sheared.
(al
(el (dl
Figure 4.3.2
Section 4.3 An Introduction to Dynamical Systems
307
To understand the higher dimensional case, a little more spectral theory is required.
°
4.3.15 Lemma Suppose o(T) = 1 U 02 where d(0l' (2) > O. Then there are unique T-invariant subspaces E\ and Ez such that E = E\ EB E 2. o(Ti) = 0i' where Ti = T I Ei ; Ei is called the generalized eigenspace of 0i' The basic idea of the proof is this: let 'Yj be a closed curve with OJ in its interior and Ok' k +'= j, in its exterior; then
Note that the eigenspace of an eigenvalue A is not always the same as the generalized eigenspace of A. In the finite dimensional case, the generalized eigenspace of T is the subspace corresponding to all the Jordan blocks containing A in the Jordan canonical form.
4.3.16 Lemma Let T, 01' and 02 be as in Lemma 4.3.15; assume d(exp(o\), exp(02» > O. Thenfor the operator exp(tT), the generalized eigenspace of exp(tT;> is Ei. Proof Write. according to Lemma 4.3.15. E
_
-
=EI EB Ez.
Thus
[~ tnTn L.. ,e\. n=O
n.
From this the result follows easily .• Now let us discuss the generic nonlinear case; i.e. let m be a hyperbolic equilibrium of the vector field X and let Ft be its flow. Define the inset of m by In(m) = (m' E M I Ft(m') -t m as t -t + oo} and similarly, the outset is defined by Out(m) = (m' E M I Ft(m') -t m as t -t + -oo}.
In the case of a linear system,
x= Ax,
where A has no eigenvalue on the imaginary axis (so the
origin is a hyperbolic critical point), In(O) is the generalized eigenspace of the eigenvalues with
308
Chapter 4 Vector Fields and Dynamical Systems
negative real parts, while Out(O) is the generalized eigenspace corresponding to the eigenvalues with positive real parts. Clearly, these are complementray subspaces. The dimension of the linear subspace In(O) is called the stability index of the critical point. The Hartman-Grobman linearization theorem states that the phase portrait of X near m is topologically conjugate to the phase portrait of the linear system = Ax, near the origin, where A = X'(m), the linearized vector field at m. This means there is a homeomorphism of the two domains, preserving the oriented trajectories of the respective flows. Thus in this nonlinear hyperbolic case, the inset and
x
CO submanifolds. Another important theorem of dynamical systems theory, the stable manifold theorem (Smale [1967]) says that in addition, these are smooth, injectively immersed submanifolds, intersecting transversally at the critical point m. See Fig. 4.3.3 for an illustration showing part of the inset and outset near the critical point.
outset are
It follows from these important results that there are 'up to topological conjugacy) only a few essentially different phase portraits, near hyperbolic critical points. These are classified by the dimension of their insets, called the stability index, which is denoted by S(X, m) for m an equilibrium. as in the linear case.
OUI(m o )
\
Figure 4.3.3 The word index comes up in this context with another meaning. If M is finite dimensional and m is a critical point of a vector field X. the topological index of m is + 1 if the number of eigenvalues (counting multiplicities) with negative real part is even and is -1 if it is odd. Let this index be denoted I(X, m). so that I(X, m) = (-l)s(X, m). The Poincare-HopI index theorem states that if M is compact and X only has (isolated) hyperbolic critical points, then
L
I(X. m) = X(M)
m is a critical
point of X
Section 4.3 An Introduction to Dynamical Systems
309
where X(M) is the Euler-Poincare characteristic of M. For isolated nonhyperbolic critical points the index is also defined but requires degree theory for its definition--a kind of generalized winding number; see Section 7.5 or Guillemin and Pollack [1974, p. 133]. We now illustrate these basic concepts about critical points with some classical applications.
4.3.17 Examples A The simple pendulum with linear damping is defined by the second-order equation
x+ c it + k sin x = 0
(c > 0).
This is equivalent to the following dynamical system whose phase portrait is shown in Fig. 4.3.4:
v= - cv - k sin x.
x=v,
The stable focus at the origin represents the motionless, hanging pendulum. The saddle point at (k1t, 0) corresponds to the motionless bob, balanced at the top of its swing.
Figure 4.3.4 B Another classical equation models the buckling column (see Stoker [1950, ch. 3, §IO]):
Or equivalently, the planar dynamical system 3
. cv a,x a3 x x=v, v = - - - - - - m m m
310
Chapter 4 Vector Fields and Dynamical Systems
with the phase portrait shown in Fig. 4.3.5. This has two stable foci on the horizontal axis, denoted m) and m2, corresponding to the column buckling (due to a heavy weight on the top) to either side. The saddle at the origin corresponds to the unstable equilibrium of the straight, unbuckled column.
Figure 4.3.5
Figure 4.3.6 Note that in this phase portrait, some initial conditions tend toward one stable focus, while some tend toward the other. The two tendencies are divided by the curve, In(O, 0), the inset of
Section 4.3 An Introduction to Dynamical Systems
311
the saddle at the origin. This is called the separatrix, as it separates the domain into the two disjoint open sets, In(m o) and In(m 1). The stable foci are caIled attractors, and their insets are caIled their basins. See Fig. 4.3.6 for the special case + it - x + x 3 = O. This is a special
x
case of a general theory, which is increasingly important in dynamical systems applications. The attractors are regarded as the principal features of the phase portrait; the size of their basins measures the probability of observing the attractor, and the separatrices help find them .• Another basic ingredient in the qualitative theory is the notion of a closed orbit, also called a limit cycle ). 4.3.18 Definition An orbit )'(t) for a vector field X is called closed when y(t) is not afixed
point and there is a 't > 0 such that )'(t + 't) = )'(t) for all t. The inset of y, In(y), is the set of points m E M such that Ft(m) ~ y as t ~ +00 (i.e., the distance between Ft(m) and the (compact) set {y (t) I 0 ~ t ~ 't} tends to zero as t ~ 00.) Likewise, the outset, Out(y), is the set of points tending to y as t ~ -00. 4.3.19 Example One of the earliest occurrences of an attractive closed orbit in an important application is found in Baron Rayleigh's modelfor the violin string (see Rayleigh [1887, vol. 1, §68aj),
or equivalently,
with the phase portrait shown in Fig. 4.3.7.
Figure 4.3.7
312
Chapter 4 Vector Fields and Dynamical Systems
This has an unstable focus at the origin. with an attractive closed orbit around it. That is. the closed orbit y is a limiting set for every point in its basin (or inset) In(y). which is an open set of the domain. In fact the entire plane (excepting the origin) comprises the basin of this closed orbit. Thus every trajectory tends asymptotically to the limit cycle y and winds around closcr and closer to it. Meanwhile this closed orbit is a periodic function of time. in the sense of Definition 4.3.6. Thus the eventual (asymptotic) behavior of every trajectory (other than the unstable constant trajectory at the origin) is periodic; it is an oscillation. This picture thus models the sustained oscillation of the violin string. under the influence of the moving bow. Related systems occur in electrical engineering under the name van der Pol equation. (See Hirsch and Smale [1974]. Chapter 10 for a discussion) .• We now proceed toward the analog of Liapunov's theorem for the stability of closed orbits. To do this we need to introduce Poincare maps and characteristic multipliers.
4.3.20 Definition Let X be a
cr
vector field on a manifold M, r ~ 1. A local transversal
section of X at m E M is a submanifold SCM of codimension one with m E S and for all s E S. X(s) is not contained in T.S.
Let X be a cr vector field on a manifold M with cr flow F: 1>x C M x lR 4 M, Y a closed orbit of X with period 't, and S a local transversal section of X at m E y. A Poincare map of y is a cr mapping 8: Wo 4 WI where: PM1 W 0' WI C S are open neighborhoods of m E S, and 8 is a Cr diffeomorphism; PM2 there is a cr function 15: Wo 4 lR such that for all SEW0' (S, 't - 15(s» E
1>x, PM3
and 8(s) =F(s, 't - 15(s»; and finally,
if t E ]0, 't - 15(s)[, then F(s, l) e W 0 (see Fig. 4.3.8).
Figure 4.3.8
Section 4.3 An Introduction to Dynamical Systems
313
4.3.21 Existence and Uniqueness of Poincare Maps (i) If X is a Cr vector field on M, and y is a closed orbit of X, then there exists a Poincare map of y. (ii) If 8 : W0 ~ WI is a Poincare map of y (in a local transversal section S at m E y) and 8' also (in S' at m' E y), then 8 and 8' are locally conjugate. That is, there are open neighborhoods W2 of m E S, W2' of m' E S', and a C'-difJeomorphism H : W2 ~ W2', such that W2 c Wo n WI' diagram commutes:
W' 2
Wz' CWo' n WI'
and the following
.. S'
*
Proof (i) At any point mE y we have X(m) 0, so there exists a flow box chart (U, cp) at m with cp(U) = V x I C lRo- 1 X lR. Then S = cp-I(V x (O}) is a local transversal section at m. If F: 'Dx C M x lR ~ M is the integral of X, 'Dx is open, so we may suppose U x [-'t, 'tJ C 'iJx' where 't is the period of y. As F~(m) = m E M and F~ is a homeomorphism, Uo = F~-I(U)
n
U is an open neighborhood of mE M with FiUo) C U. Let Wo = S
W2 = F~(Wo). Then W2 is a local transversal section at mE M and diffeomorphism (see Fig. 4.3.9).
s
F~:
Wo
n Uo and
~
W2 is a
u
lJo
Figure 4.3.9 If U2 = F~(Uo)' then we may regard Uo and U2 as open submanifolds of the vector
bundle V x lR (by identification using cp) and then F~: Uo ~ U2 is a cr diffeomorphism mapping fibers into fibers, as cp identifies orbits with fibers, and F~ preserves orbits. Thus W2
314
Chapter 4 Vector Fields and Dynamical Systems
is a section of an open subbundle. More precisely, if 1t: V x I --+ V and p : V x I --+ I are the projection maps, then the composite mapping is
cr and
has a tangent that is an isomorphism at each point, and so by the inverse mapping theorem, it is a
cr
diffeomorphism onto an open submanifold. Let WI be its image, and 8
the composite
mapping. We now show that 8: Wo --+ WI is a Poincare map. Obviously PM 1 is satisfied. For PM 2, we identify U and V x I by means of q> to simplify notalions. Then 1t: W 2 --+ WI is a diffeomorphism, and its inverse (1t I W 2>-1
:
WI --+ W 2
C
WI
X
lR is a section corresponding
to a smooth function 0': WI --+ JR. In fact, 0' is defined implicitly by
or p 0 F~(wo» = 0' 0 1tF~(wo)' Now let Ii: Wo --+ lR be given by Wo 1-+ 0' 0 F t (w o) which is cr. Then we have
Finally, PM 3 is obvious since (U, q» is a flow box. (ii) The proof is burdensome because of the notational complexity in the defmition of local
conjugacy, so we will be satisfied to prove uniqueness under additional simplifying hypotheses that lead to global conjugacy (identified by italics). The general case will be left to the reader. We consider first the special case m assume sUS'
C
=m '.
Choose a flow box chart (U, q» at m, and
U, and that S and S' intersect each orbit arc in U at most once, and that
they intersect exactly the same sets of orbits. (These three conditions may always be obtained by shrinking S and S'.) Then let W 2 = S, W 2' = S', and H: W 2 --+ W2' the bijection given by the orbits in U. As in (i), this is seen to be a
cr
diffeomorphism, and H 08= 8' 0 H.
Finally, suppose m *- m'. Then Fa(m) = m' for some a E ]0, 't[, and as 'Dx is open there is a neighborhood U of m such that U x {a} C 'Dx. Then Fa(U n S) = s" is a local transversal section of X at m' E Y, and H = Fa effects a conjugacy between 8 and 8" = Fa 0 80 Fa-Ion S". By the preceding paragraph, 8" and 8' are locally conjugate, but conjugacy is an equivalence relation. This completes the argument. _ If Y is a closed orbit of X E ~(M) and m E 'Y, the behavior of nearby orbits is given by a
Poincare map 8 on a local transversal section S at m. Clearly Tm8 E L(TmS, TmS) is a linear approximation to 8 at m. By uniqueness of 8 up to local conjugacy, T m,8' is similar to Tm8, for any other Poincare map 8' on a local transversal section at m' E 'Y. Therefore, the spectrum of Tm8 is independent of me 'Y and the particular section S at m.
Section 4.3 An Introduction to Dynamical Systems
4.3.22 Definition
315
If y is a closed orbit of X E X(M). the characteristic multipliers of X at y
are the points in the spectrum of Tme. for any Poincare map
e
at any m
Another linear approximation to the flow near y is given by E y and 't is the period of y. Note that F/(X(m»
=X(m).
TmF~
E
y.
E L(TmM. TmM) if m
so TmF~ always has an eigenvalue
1 corresponding to the eigenvector X(m). The (n - 1) remaining eigenvalues (if dim(M)
=n)
are in fact the characteristic multipliers of X at y. 4.3.23 Proposition If y is a closed orbit of X E X(M) of period 't and cy is the set of
characteristic multipliers of X at y. then cyu {I} is the spectrum of
TmF~.
for any mE y.
Proof We can work in a chart modeled on E and assume m = O. Let V be the span of X(m) so E = T mM e V. Write the flow Ft(x. y) = (FtI(x. y). (Ft2(x. y». By definition. we have
Thus the matrix of TmF~ is of the form
where A
=DIF~2(m).
From this it follows that the spectrum of T mF~ is {I} U cy .
•
If the characteristic exponents of an equilibrium point lie (strictly) in the left half-plane. we
know from Liapunov's theorem that the equilibrium is stable. For closed orbits we introduce stability by means of the following definition. 4.3.24 Definition Let X be a vector field on a manifold M and y a closed orbit of X. An orbit Ft(mo) is said to wind toward y
if mo is + complete and for any local transversal section S to
X at mE y there is a teO) such that Ft(O)(mo) E S and successive applications of the Poincare map yield a sequence ofpoints that converges to m. If the closed orbit y has a neighborhood U
such that for any mo E U. the orbit through mo winds towards y. then y is called asymptotically stable. In other words. orbits starting "close to y. "converge" to y; see Fig. 4.3.10.
4.3.25 Proposition If y is an asymptotically stable periodic orbit of the vector field X and mo to> 0 Such that for all t ~ to. Ft(m) E v.
e U. the neighborhood given in 4.3.24, then for any neighborhood Vof y. there exists
316
Chapter 4 Vector Fields and Dynamical Systems
Figure 4.3.10 Proof Derme mk = ek(IIlo), where 8 is a Poincare map for a local transversal section at m to 'Y containing mo' Let t(n) be the "return time" of n, i.e., t(n) is defined by Ft(o)(n) E S. If 't denotes the period of 'Y and ~ = t(mk), then since mk -+ m, it follows that 'tk -+ 't since t(n) is a smooth function of n by 4.3.21. Let M be an upper bound for the set {I'tkll kEN}. By smoothness of the flow, F.(mk) -+ F.(m) as k -+ 00, uniformly in s E [0, MJ. Now write for any t > 0, Ft(m o) = FT(t)(mk(t», for T(t) E [0, MJ and observe that k(t) -+ 00 as t -+ 00. Therefore, if W is any neighborhood of FT(t)(m O) contained in V, since Ft(mo) = FT(t)(mk(t» converges to FT(t)(m) as t -+ 00 it follows that there exists to > 0 such that for all t ~ to, Ft(m o) Ewe V. • 4.3.26 Liapunov Stability Theorem for Closed Orbits Let 'Y be a closed orbit of X E 'Y lie strictly inside the unit circle. Then 'Y is asymptotically stable.
*(M) and let the characteristic multipliers of
The proof relies on the following lemma. 4.3.27 Lemma Let T: E -+ E be a bounded linear operator. Let a(T) lie strictly inside the unit circle. Then limn -+ ~ TOe =0 for all e E E. Proof By lemma 4.3.7 and compactness of a(T), there is a norm I I on E equivalent to the original norm on E such that I T I :;;; r < 1. Therefore I TOe I :;;; rD I e I -+ 0 as n -+ 00• • 4.3.28 Lemma Let f: S -+ S be a smooth map on a manifold S with f(s) = s for some s. Let the spectrum of Tl lie strictly inside the unit circle. Then there is a neighborhood U of s such that if s' E U, f(s') E U and fY'(s') -+ s as n -+ 00, where fY' = f 0 f 0 . . . 0 f (n times).
Section 4.3 An Introduction to Dynamical Systems
317
Proof We can assume that S is a Banach space E and that s = O. As above. renorm E and find such that 1 T 1 :0:;; r. where T = Df(O). Let E> 0 be such that r + E < 1. Choose a neighborhood V of 0 such that for all x E V
o< r < 1
1 f(x)
- Tx
1 :0:;;
E 1x 1
which is possible since f is smooth. Therefore. 1 f(x) 1:0:;; 1 Tx 1
+ E 1 x 1 :0:;; (r + E) 1 x I·
Now. choose 0 > 0 such that the ball U of radius 0 at 0 lies in V. Then the above inequality implies 1 f"(x) 1 :0:;; (r + E)n 1 x 1 for all x E U which shows that f"(x) E U and f"(x) --+ O. • Proof of 4.3.26 The previous lemma applied to f = 9. a Poincare map in a transversal slice S to y at m. implies that there is an open neighborhood V of m in S such that the orbit through every point of V winds toward y. Thus. the orbit through every point of U = {Ft(m) I t ~ O} ::::> y winds toward y. U is a neighborhood of y since by the Sraightening Out Theorem 4.1.14. each point of y has a neighborhood contained in {Ft(9(U» It> -E. E> O} .• 4.3.29 Definition If X E *(M) and y is a closed orbit of X. y is called hyperbolic if none of the characteristic multipliers of X at y has modulus 1. Hyperbolic closed orbits are isolated (see Abraham-Robbin [1967. Ch. 5]). The local qualitative behavior near an hyperbolic closed orbit. y. may be visualized with the aid of the Poincare map. 9: Woe S --+ WI C S. as shown in Fig. 4.3.8. The qualitiative behavior of this map. under iterations. determines the asymptotic behavior of the trajectories near y. Let m E y be the base point of the section. and s E S. Then In(y). the inset of y. intersects S in the inset of m under the iterations of e. That is. s E In(y) if the trajectory Ft(s) winds towards y. and this is equivalent to saying that eI«s) tends to m as k --+ + 00. The inset and outset of m E S are classified by linear algebra. as there is an analogue of the linearization theorem for maps at hyperbolic critical points. The linearization theorem/or maps says that there is a CO coordinate chart on S. in which the local representative of 9 is a linear map. Recall that in the hyperbolic case. the spectrum of this linear isomorphism avoids the unit circle. The eigenvalues inside the unit circle determine the generalized eigenspace of contraction i.e.• the inset of m E S under the iterates of 9. The eigenvalues outside the unit circle similarly determine the outset of m E S. Although this argument provides only local CO submanifolds. the global stable manifold theorem improves this: the inset and outset of a fixed point of a diffeomorphism are smooth. injectively immersed submanifolds meeting transversally at m. Returning to closed orbits. the inset and outset of y c M may be visualized by choosing a section Sm at every point m E y. The inset and outset of y in M intersect each section Sm in submanifolds of Sm' meeting transversally at m E y. In fact. In(y) is a cylinder over y. that is.
318
Chapter 4 Vector Fields and Dynamical Systems
a bundle of injectively immersed disks. So, likewise, is Out(y). And these two cylinders intersect transversally in y, as shown in Fig. 4.3.11. These bundles need not be trivial.
Figure 4.3.11 Another argument is sometimes used to study the inset and outset of a closed orbit, in place of the Poincare section technique described before, and is originally due to Smale [1967]. The flow F t leaves the closed orbit invariant. A special coordinate chart may be found in a neighborhood of y. The neighborhood is a disk bundle over y, and the flow Ft, is a bundle map. On each fiber, F t is a linear map of the form ZteRt, where Zt is a constant, and R is a linear map. Thus, if s =(m, x) is a point in the chart, the local representative of F t is given by the expression
called the Floquet normal/orm. This is the linearization theorem/or closed orbits. A related result, the Floquet theorem, eliminates the dependence of Zt on t, by making a further (time-dependent) change of coordinates (see Hartman [1973, ch. 4, §6], or Abraham and Robbin [1967]). Finally, linear algebra applied to the linear map R in the exponent of the Floquet normal form, establishes the CO structure of the inset and outset of y. To get an overall picture of a dynamical system in which all critical elements (critical points and closed orbits) are hyperbolic, we try to draw or visualize the insets and outsets of each. Those with open insets are attractors, and their open insets are their basins. The domain is divided into basins by the separatrices, which includes the insets of all the nonattractive (saddle-type) critical elements (and possible other, more complicated limit sets, called chaotic attractors, not described here.) We conclude with an example of sufficient complexity, which has been at the center of dynamical system theory for over a century.
Section 4.3 An Introduction to Dynamical Systems
319
4.3.30 Example The simple pendulum equation may be "simplified" by approximating sin x by two tenns of its MacLaurin expansion. The resulting system is a model for a nonlinear spring with
linear damping. X
. = v. v = - cv - kx
+"3k x3.
Adding a periodic forcing tenn. we have
X=
. k 3 v. v = - cv - kx + "3 x + F cos rot.
This time-dependent system in the plane is transfonned into an autonomous system in a solid ring by adding an angular variable proportional to the time. 9 = rot. Thus. x=v,
. V
=-
k 3 . cv - kx + "3 x + F cos 9. 9 = roo
Although this was introduced by Baron Rayleigh to study the resonance of tuning forks. piano strings. and so on. in his classic book Theory a/Sound [1877]. this system is generally named the
Duffing equation after Duffing who obtained the first important results (see Duffing [1908] and Stoker [1950]). Depending on the values of the three parameters (c. k. F) various phase portraits are obtained. One of these is shown in Fig. 4.3.12. adapted from the experiments of Hayashi [1964]. There are three closed orbits: two attracting. one of saddle type. The inset of the saddle is a cylinder topologically. but the whole cylinder revolves around the saddle-type closed orbit. This cylinder is the separatrix between the two basins. For other parameter values the dynamics can be chaotic (see for example. Holmes [1979a.b] and Ueda [1980]).
Figure 4.3.12
Chapter 4 Vector Fields and Dynamical Systems
320
For further information on dynamical systems, see Abraham and Shaw [1985] and Guckenheimer and Holmes [1983].
Exercises 4.3A
(i)
Let E --+ M be a vector bundle and m E M an element of the zero section. Show
(ii)
If ~ : M --+ E is a section of E, and ~(m) =0, define ~'(m) : TmM --+ Em to be
(iii)
the projection of Tm~ to Em' Write out ~'(m) relative to coordinates. Show that if X is a vector field, then X'(m) defmed this way coincides with
that TmE is isomorphic to TmM E9 Em in a natural, chart independent way.
Definition 4.3.1. 4.38 Prove that the equation 9 + 2k 9 - q sin 9
9
=0 (q > 0, k > 0)
has a saddle point at 9
=0,
= O.
4.3C Consider the differential equations
r= ar3 - br,
{) = 1 using polar coordinates in the plane.
(i)
Determine those a, b for which this system has an attractive periodic orbit.
(ii)
Calculate the eigenvalues of this system at the origin for various a, b.
4.3D Let X E X(M), cp: M --+ N be a diffeomorphism, and Y =CP.X. Show that (i) m E M is a critical point of X iff cp(m) is a critical point of Y and the characteristic exponents are the same for each; (ii) y c: M is a closed orbit of X iff cp(y) is a closed orbit of Y and their characteristic multipliers are the same. 4.3E The energy for a symmetric heavy top is 2
1 2 2 2·2 P\j1 ft H(9, po) = - - 2 - {I'lv (b - cos 9) + AI sm 9} + -J- + Mg~cos 9 21 sin 9 where I, J > 0, b, P\j1' and Mgt> 0 are constants. The dynamics of the top is described by the differential equations {) = aH/dpo' Po = -aHJi)9. (i) Show that 9 = 0, Po = 0 is a saddle point if 0 < P\j1 < 2(MgtI)112 (a slow top). (ii) Verify that cos 9 = 1 - Ysech2«~y)1I2/2), where y =2 - b2~ and ~ =2Mgt/I describe both the outset and inset of this saddle point. (This is called a homoclinic orbit. ) (iii) Is 9 = 0, Po = 0 stable if P\j1> (MgtI)l12? (Hint: Use the fact that H is constant along the trajectories.)
321
Section 4.3 An Introduction to Dynamical Systems
4.3F Let A
E
L(lRn. JR.n) and suppose that a < Re(l..i ) < b. for all eigenvalues Ai' i = 1•...• n
«. »
of A. Show that JRn admits an inner product that a III x 1112 S
with associated norm
III· III
such
«Ax. x» S b III x IIf .
Prove this by following the outline below. (i)
If A is diagonalizable over C. then find a basis in JRn in which the matrix of A has either the entries on the diagonal. the real eigenvalues of A. or 2 x 2 blocks of the form
Choose the inner product
«.» on JRn such that the one and two-dimensional
invariant subspaces of A defined by this block-matrix are mutually orthogonal; pick the standard JR2-basis in the two-dimensional spaces. (ii)
If A is not diagonalizable. then pass to the real Jordan form. There are two kinds of Jordan k j x kj blocks: I..j 0 0 1 I..j 0 0 1 I..j
0 0 0
0 if I..j
E
0 0 0 I..j
JR. or
0
0
0
~j
0
0 12
~j
0 0
o.
~j
0
12
0
12 ~j
0 where ~. J
= [ a·J -bj].
bj
aj
12
=[~ ~J
.
if I..j = aj + ibj• aj• bj E JR. bj '" O. For the first kind of block. choose the basis e l ' ..... '1,;(j)' of eigenvectors of the diagonal part D. Then. for £ > 0 small. put e/ = e/fer - I • r = 1..... k(j) and define ( • )t on the subspace span{e l t ..... '1,;(j)t}
C
JRn to be the Euclidean inner product given by this basis. Compute the
marix of A in this basis and show that
322
Chapter 4 Vector Fkids and Dynamical Systems
( Ax. x)£ Dx . x ..;....-......::. -+ - - 2 - ' ( x. x)£ II x II Conclude that for
E
as E -+ O.
small the statement holds for the first kind of block. Do the
same for the second kind of block. 4.3G Let A
E
L(]RR.]RR). Show that the following are equivalent.
(i)
All eigenvalues of A have strictly negative real part (the origin is called a sink in
(ii)
For any norm I· I on ]RR. there exist constants k > 0 and E> 0 such that for all t ~ o. I etA I ~ ke-t£ .
(iii)
There is a norm III· ilion ]RR and a constant 15 > 0 such that for all t ~ o. III etA III ~ e-t&. (Hint: (ii) => (i) by using the real Jordan form: if every solution
this case).
of
x= Ax
tends to zero as t -+ + 00. then every eigenvalue of A has strictly
negative real part. For (i) => (iii) use Exercise 4.3E and observe that if x(t) is a
i = Ax. then we have
solution of
(d/dt) III x(t) III =
that we get the following inequality: at
~
«x(t). Ax(t) » / III x(t) III. so
log III x(t) III/log III x(O) III
a = min{Re (Ai) I i = 1•...• n} • and b = max{Re (Ai) -E =
I
~
bt. where
i = 1•...• n}. Then let
b.) Prove a similar theorem if all eigenvalues of A have strictly positive real
part; the origin is then called a source. 4.3H Give a proof of Theorem 4.3.4 in the finite dimensional case without using the variation of constants formula (Exercise 4.1E) and using Exercise 4.3F. (Hint: If A = X'(O) locally show Iimx -+ 0
«X(x) - Ax. x » / III x 1112 = O.
Since
«Ax. x» ~
-E
III x 111 2 ,
E
= max Re
{Ail i = 1, ... , n, Ai eigenvalues of A}, there exists 15 > 0 such that if III x III ~ 15, then
X(x), x» ~ - C III x
111 2 ,
«
for some C > O. Show that if x(t) is a solution curve in the
closed &-ball. t E [0. TJ. then dill x(t) 111/ dt ~ - CIII x(t) III. Conclude III x(t) III ~ 15 for all t E [0. TJ and thus by compactness of the 15-ball. x(t) exists for all t ~ O. Finally. show that III x(t) III ~ e-tE III x(O) III.) 4.31 An equilibrium point m of a vector field X
E
X(M) is called a sink, if there is a 15 > 0
such that all points in the spectrum of X'(m) have real part < -15. (i)
Show that in a neighborhood of a sink there is no other equilibrium of X.
(ii)
If M =]RR and X is a linear vector field. Exercise 4.3G shows that provided Iimt-+~
m(t) = 0 for every integral curve m(t) of X. then the eigenvalues of X
have all strictly negative real part. Show that this statement is false for general vector fields by finding an example of a non-linear vector field X on ]RR whose integral curves tend to zero as t -+ 00. but is such that X'(O) has at least one eigenvalue with zero real part. (Hint: See Exercise 4.3C). 4.3J Hyperbolic Flows An operator A E L(JRR. ]RR) is called hyperbolic ifno eigenvalue of A has zero real part. The linear flow x ~ etAx is then called a hyperbolicflow.
Section 4.3 An Introduction to Dynamical Systems
(i)
323
Let A be hyperbolic. Show that there is a direct sum decomposition jRD = E S EEl Ell, A(ES) C ES, A(Ell) C Ell, such that the origin is a sink on ES and a source on Ell; ES and Ell are called the stable and unstable subspaces of jRn. Show that the decomposition is unique. (Hint: ES is the sum of all subspaces defined by the real Jordan form for which the real part of the eigenvalues is negative. For uniqueness, if jRn = E's EEl E'll and vEE's, then v = x + y, X E E S, y E Ell with elAv ~
o
as t ~
00,
so that elAx ~ 0, elAy ~ 0 as t ~
source on E' ll,
II elAy II ~ etl: II y II for some
4.3G(ii).)
'* 0,
£
00.
But since the origin is a
> 0 by the analogue of Exercise
II elAx II ~
(ii)
Show that A is hyperbolic iff for each x
(iii)
Conclude that hyperbolic flows have no periodic orbits.
00
as t ~ ± 00.
4.3K Gradient flows (this continues Exercise 4.tH) Let f: jRD ~ jR be C I and let X = (af/ax I, 00.' af/ax D). Show that at regular points of f, the integral curves of X cross the level surfaces of f orthogonally and that every singular point of f is an equilibrium of X. Show that isolated maxima of f are asymptotically stable. (Hint: If Xo is the isolated maximum of f, then f(x o) - f(x) is a strict Liapunov function.) Draw the level sets of f and the integral curves of X on the same diagram in jR2, when f is defined by f(x I, x2) = (xl _ 1)2 + (x2 _ 2)2(x2 _ 3)2. 4.3L Consider
u+ u+ u3 = O. Show that solutions converge to zero like c/l't as H(u, u) = (u + u)2 + u 3 + u 4. (See Ball and Carr [1976]
considering
t~
00
by
for more
information.) 4.3M Use the method of Liapunov functions to study the stability of the origin for the following vector fields:
(iii)
X(x, y) = -(3y + x3)(a;ax) + (2x - 5y3) (a/ay) (asymptotically stable); X(x, y) = -xy4(a;ax) + x6y (a/ay) (stable; look for L of the form x4 + ay6); X(x, y) = (xy - x3 + y)(a/ax) + (x 4 - x2y + x3)(a/ay) (stable; look for L of the
(iv)
X(x, y) = (y + x7 )(a/ax) + (y9 - x) (a/ay) (unstable);
(v)
X(x, y, z) = 3y(z + l)(a;ax) + x(z + l)(a;ay) + yz (a/az) (stable); X(x, y, z) = (-x 5 + 5x6 + 2y3 + xz2 + xyz)(a/ax) + (-y - 2z + 3x6 + 4yz + xz +
(i) (ii)
form ax 4 + by2);
(vi)
xy2)(a/ay) + (2y - z - 2x 8 - y2 + xz2 + xy3)(a/az) L(x, y, z) = (l/2)x 2 + 5(y2 + z2».
(asymptotically stable; use
4.3N Consider the following vector field on jRD+I; X(s, x) = (asN + f(s) + g(s, x), Ax + F(x) + h(s, x» where s E jR, X E jRn, f(O) = 00. = r 0,
324
Chapter 4 Vector Fwlds and Dynamical Systems
then the origin is unstable; if N is odd and a < O. the origin is asymptotically stable. (Hint: for N even. use L(s. x) = s - a II x 112/2 and for N odd. L(s. x) = (s2 - a II x 112)/2; show that in both cases the sign of X[L] near the origin is given by the sign of a.) ~ L(E. E) a continuous map. Let Ft.s denote the evolution operator of the time--dependent vector field X(t. x) = A(t)x on E.
4.10 Let E be a Banach space and A: JR
(i) (ii)
Show that Ft.s E GL(E). Show that II Ft.s 11<::; e(t-s)<x. where a = SUP"-E[S. tl II A(A.) II. Conclude th~t the vector field X(t. x) is complete. (Hint: Use Gronwall's inequality and the time dependent version of 4.1.22.)
Next assume that A is periodic with period T. i.e. A(t + T) = A(t) for all t E R (iii)
Show that Ft+T. s+T = Ft.s for any t. s E R (Hint: Show that t ...... Ft+T. s+T (x) satisfies the differential equation = A(t)x.)
(iv)
Define the monodromy operator by M(t) = Ft+T.t" Show that if A is independent of t. then M(t) = eTA is also independent of t. Show that M(s) = Ft.s 0 M(t) 0 Fs.t
x
for any continuous A: JR ~ L(E. E). Conclude that all solutions of
x= A(t)x
are
of period T if and only if there is a to such that M(tO> = identity. Show that the eigenvalues of M(t) are independent of t. (v)
(Floquet) Show that each real eigenvalue A. of M(to) determines a solution. denoted c(t; A. to) of = A(t)x satisfying c(t + T; A.. to) = AC(t; A.. to)' and also each complex eigenvalue A. =a + ib. of M(to)' where a. b E JR. determines a pair of solutions denoted cr(t; A. to) and ci(t; A.. to) satisfying
x
[
cr(t + T; A.. to)] c'(t + T; A. to)
=
[a -bJ [cr(t; A. to)]. b a
(Hint: Let c(t; A. to) denote the solution of
c'(t; A. to)
x= A(t)x
with the initial condition
c(to; A.. to> = e. where e is an eigenvector of M(to) corresponding to A; if A is complex. work on the complexification of E. Then show that c(t + T; A.. to) Ac(t; A. to) satisfies the same differential equation and its value at to is zero since (vi)
c(to + T; A.. to) = M(to) c(to; A. to) = Ac(to; A.. Lo).) (Floquet) Show that there is a nontrivial periodic solution of period T of = A(t)x if and only if 1 is an eigenvalue of M(to) for some to E R (Hint: If c(t)
x
is such a periodic solution. then c(to) = c(to+ T) = M(to)c(to).) (vii) (Liapunov) Let P: JR ~ GL(E) be a C 1 function which is periodic with period T. Show that the change of variable y = P(t)x transforms the equation = A(t)x into the equation = B(t)y. where B(t) = (P'(t) + P(t)A(t»p(t)-1 If N(t) is the monodromy operator of y = B(t)y. show that N(t) = P(t)M(t)p(t)-I.
y
4.1P
(i)
x
(Liapunov) Let E be a finite dimensional complex vector space and let A: JR ~
Section 4.3 An Introduction to Dynamical Systems
325
L(E, E) be a continuous function which is periodic with period T. Let M(s) be
x
the monodromy operator of the equation = A(t) x and let B E L(E, E) be such that M(s) = eTB (see exercise 4.10). Define pet) = etBF s•t and put yet) = P(t)x(t). Use (vii) in the previous exercise to show that yet) satisfies y = By. Prove that pet) is a periodic CI-function with period T. (Hint: Use (iii) of the previous exercise and eTB = M(s) = Fs+T. s)' Thus, for complex finite dimensional vector spaces, the equation = A(t)x, where A(t + T) = A(t), can be transformed via yet) =P(t)x(t) into the constant coefficient linear equation =By. Since the general solution of y = By is a linear combination of vectors
x
y
(ii)
exp(tA)t'(i)u i• where AI' ... , Am are the distinct eigenvalues of B, ui E E, and 1 ~ ~(i) ~ multiplicity of J.-.. conclude that the general solution of = A(t)x is a linear combination of vectors exp(tAi)t'(i) p(t)-I ui where pet + T) =pet). Show
x
that the eigenvalues of M(s) =eTB are exp(TAJ Show that Re(A) < 0 (> 0) for
x
all i = 1. ..., m if and only if all the solutions of = A(t)x converge to the origin as t ~ +00 (-00), i.e. if and only if the origin is asympotically stable (unstable).
~4.4
Frobenius' Theorem and Foliations
The main pillars supporting differential topology and calculus on manifolds are the Implicit Function Theorem, the Existence Theorem for ordinary differential equations, and Frobenius' Theorem, which we discuss briefly here. First some definitions. 4.4.1 Definition Let M be a manifold and let E
C
TM be a subbundle of its tangent bundle;
i.e., E is a distribution (or a plane field) on M. (i) We say E is involutive iffor any two vector fields X and Y defined on open sets of M and which take values in E, [X, Yj takes values in E as well.
(ii) We say E is integrable iffor any
mE
M there is a (local) submanifold N
C
M, called
a (local) integral manifold of E at m containing m, whose tangent bundle is exactly E restricted to N.
The situation is shown in Figure 4.4.1.
Figure 4.4.1 4.4.2 Examples A Any subbundle E of TM with one dimensional fibers is involutive; E is also integrable, which is seen in the following way. Using local bundle charts for TM at III E M with the subbundle property for E, find in an open neighborhood of m, and a vector field that never vanishes and has values in E. Its local integral curves through m have as their tangent bundles E restricted to these curves. If the vector field can be found globally and has no zeros, then through
Section 4.4 Frobenius' Theorem and Foliations
327
any point of the manifold there is exactly one maximal integral curve of the vector field. and this integral curve never reduces to a point. B Let f: M ~ N be a submersion and consider the bundle ker Tf C TM. This bundle is involutive since for any X. Y E ~(M) which take values in ker Tf. we have Tf([X. Y]) =0 by 4.2.25. The bundle is integrable since for any m E M the restriction of ker Tf to the submanifold f-I(f(m» coincides with the tangent bundle of this submanifold (see §3.5). C Let 'Jl"I be the n-dimensional torus. n ~ 2. Let 1 '5 k '5 n and consider E = {(v I' ...• vn) E TIn I Vk + 1= ... = vn = OJ. This distribution is involutive and integrable; the integral manifold through (tl' ...• tn) is 'lI'k X (~+ I' ...• tn). D E =TM is involutive and integrable; the integral submanifold through any point is M itself. E An example of a noninvolutive distribution is as follows. Let M =80(3). the rotation group (sce Exercise 3.5S). The tangent space at I = identity consists of the 3 x 3 skew symmetric matrices. Let
a two dimensional subspace. For Q E 80(3). let
Then E = U {EQ I Q E 80(3)} is a distribution but is not involutive. In fact. one computes that the two vector fields with p = 1. q = 0 and p = O. q = 1 have a bracket that does not lie in E. Further insight into this example is gained after one studies Lie groups (a supplementary chapter). F Let E be a distribution on M. Suppose that a collection 'E of smooth sections of E spans E in the sense that for each section X of E there are vector fields Xi' ...• Xk in 'E and smooth functions a l ••.•• ak on M such that X = aiXi. Suppose 'E is closed under bracketing; i.e .• if X and Y have values in 'E. so does [X. YJ. We claim that E is involutive. To prove this assertion. let X and Y be sections of E and write X = aiXi and Y = biYj • where ai and bi are smooth functions on M and X; and Yj belong to 'E. We calculate "
where and Thus [X. YJ is a section of E. so E is involutive. • Frobenius' Theorem asserts that the two conditions in 4.4.1 are equivalent.
Chapter 4 Vector Fklds and Dynamical Systems
328
4.4.3 The Local Frobenius Theorem A subbundle E of TM is involutive if and only if it is integrable.
Proof Suppose E is integrable. Let X and Y be sections of E and let N be a local integral manifold through m E M. At points of N, X and Y are tangent to N, so define restricted vector fields XI N, YIN on N. By 4.2.25 (on cp-relatedness of brackets) applied to the inclusion map, we have
Since N is a manifold, [XI N, YIN] is a vector field on N, so [X, Y] is tangent to N and hence in E. Conversely, suppose that E is involutive. By choosing a vector bundle chart, one is reduced to this local situation: E is a model space for the fibers of E, F is a complementary space, and U x V c: E x F is an open neighborhood of (0, 0), so U x V is a local model for M. We have a map f: U x V -+ L(E, F) such that the fiber of E over (x, y) is E(x.y)
= {(u,f(x,y),u)luE E) C:ExF,
and we can assume we are working near (0, 0) and f(O, 0) =O. Let us express involutivity of E in terms of f. For fixed u E E, let Xu(x, y) = (u, f(x, y) . u). Using the local formula for the Lie bracket (see formula (5) in §4.2) one finds that [XUl ' Xu21(x, y) = (0, Df(x, y) . (ul' f(x, y) . ul) . u2 (1)
- Df(x, y) . (u2' f(x, y) . u2) . ul) By the involution assumption, this lies in description of
E(x, y)
E(x. y)'
Since the first component vanishes, the local
above shows that the second must as well; i.e., we get the following identity:
Df(x, y) . (up f(x, y) . u 1) . u 2
= Df(x, y) . (u2' f(x, y) . ~) . u 1 .
(1')
Consider the time dependent vector fields Xt(x, y)
= (0, f(tx, y) . x)
and
~.
u(x, y)
= (u, f(tx, y) . tu)
,
so that by the local formula for the Jacobi-Lie bracket, [~,
Xt. u](x, y) = (0, tD 2f(tx, y) . (f(tx, y) . x) . u - tDf(tx, y) . (u, f(tx, y) . u) . x - f(tx, y) . u) (0, - tD1f(tx, y) . x . u - f(tx, y) . u) ,
(2)
Section 4.4 Frobenius' Theorem and Foliations
329
where the last equality follows from (1'). But a~. jat equals the negative of the right hand side of (2). i.e .•
which by 4.2.31 is equivalent to
(3) where F t =Ft. 0 and Ft. s is the evolution operator of the time dependent vector field Xt. Since ~(O.
0) = O. it follows that F t is defined for 0:5 t :5 1 by 4.1.25. Since Fo is the identity. relation (3) implies that
(4) Let N = F I (E x {O
n. a submanifold of E x F. the model space of M. If (x. y) = F I(e. 0).
the
tangent space at (x. y) to N equals {T(e.oll(u.O»luE E} = (T(e.O)FI(Xo.u(e.O»I UE E} (by (4»
(XI. u(FI(e. 0» I u E E}
=
{(u. f(x. y). u) I u E E}
=E(x.y)'
•
The method of using the time-one map of a time-dependent flow to provide the appropriatlcoordinate change is useful in a number of situations and is called the method of Lie transforms. An abstract version of this method is given in 5.4.7; we shall use this method again in Chapter 6 to prove the Poincare Lemma and in Chapter 8 to prove the Darboux Theorem. Note: The method of Lie transforms is also used in singularity theory and bifurcation theory (see Golubitsky and Schaeffer [1985]). For a proof of the Morse Lemma using this method. see 5.5.S. which is based on Palais [1969] and Golubitsky and Marsden [1983]. For a proof of the Frobenius Theorem from the Implicit Function Theorem using manifolds of maps in the spirit of Supplement 4.1 C. see Penot [1970]. See Exercise 4.4F for another proof of the Frobenius Theorem in finite dimensions. The Frobenius Theorem is intimately connected to the global concept of foliations. Roughly speaking. the integral manifolds N can be glued together to form a "nicely stacked" family of submanifolds filling out M (see Figure 4.4.1 or Example 4.4.2A). 4.4.4 Definition Let M be a manifold and ell = {La} a E A a partition of M into disjoint connected sets called leaves. The partition ell is called afoliotion if each point of M has a chart (U. cp). cp : U ~ U' x V' C E EEl F such thatfor each La the connected components (U n La)P
330
Chapter 4 Vector Fields and Dynamical Systems
of V n La are given by cp«V n La)~) =V' x {ca~}' where ca~ E F are constants f or each a. E A and ~. Such charts are called foliated (or distinguished) by «1>. The dimension (respectively, codimension) of the foliation «I> is the dimension of E (resp., F). See Figure 4.4.2.
Figure 4.4.2 Note that each leaf La is a connected immersed submanifold. In general, this immersion is not an embedding; i.e., the induced topology on La from M does not necessarily coincide with the topology of La (the leaf La may accumulate on itself, for example). A differentiable structure on La is induced by the foliated charts in the following manner. If (V, cp), cp: V ~ V' x V' C E E£l F is a foliated chart on M, and X: E E£l F ~ F is the canonical projection, then X 0 cp restricted to (V n La)~ defines a chart on La. 4.4.5 Examples
A The trivial foliation of a connected manifold M has only one leaf, M itself. It has codimension zero. If M is finite dimensional, both M and the leaf have the same dimension. Conversely, on a finite-dimensional connected manifold M, a foliation of dimension equal to dim(M) is the trivial foliaton. B The discrete foliation of a manifold M is the only zero-dimensional foliation; its leaves are all points of M. If M is finite dimensional, the dimension of M is the codimension of this foliation. C A vector field X that never vanishes on M determines a foliation; its leaves are the maximal integral curves of the vector field X. The fact that this is a foliaton is the Straightening Out Theorem (see §4.1). D Let f: M ~ N be a submersion. It defines a foliation on M (of codimension equal to dim(N) if dim(N) is finite) by the collection of all connected components of f-I(n) when n varies throughout N. The fact that this is a foliation is given by Theorem 3.5.4. In particular, we see that E EE> F is foliated by the family {E x (fll f E F. E In the preceding example, let M = JR3, N = JR, and f(x l , x2, x3) = cp(r2)exp(x3), where cp: JR ~ R is a C"" function satisfying cp(O) = 0, cp(1) = 0, and cp '(s) < 0 for s> 0,
Section 4.4 Frobenius' Theorem and Foliations and where r2
= (xl)2 + (x2)2.
Since df(x I, x2, x3)
331
= exp(x 3)[2
0, then f-I(c) = {(xl, x2, log (c/
0, are asymptotically tangent to the cylinder f-I(O). Finally, since
I, for c < 0 the leaves given by f-I(c) = {(xl, x2,log (c/
I} are diffeomorphic to the plane minus the closed unit disk, i.e., they are diffeomorphic to cylinders. As before, note that the cylinders f-I(c) are symptotically tangent to f-I(O); see Figure 4.4.3.
--_ ... ,, ,
,,
'....... __ ... "
,,
,
....
Figure 4.4.3 F The Reeb Foliation on the Solid Torus and the Klein Bottle (Reeb [1952]). We claim that in the previous example, the leaves are in some sense translation invariant. The cylinder r-I(O) is invariant; if c *- 0, invariance is in the sense f-I(c) + (0, 0, log t) =f-I(tc), for any t > O. Consider the part of the foliation within the solid cylinder (xl)2 + (x2)2 ~ 1 and form the solid torus from this cylinder: identify (ai, a2, 0) with (b l , b2, 1) iff ai = bi , i = 1,2. The foliation of the solid torus so obtained is called the orientable Reeb foliation. Out of the cylinder one can form the Klein bottle (see Figure 1.4.2) by considering the equivalence relation which identifies (ai, a2, 0) with (b l , b 2, 1) iff a l = b l , a2 = - b 2. In this way one obtains the nonorientable Reeb foliation. (This terminology regarding orientability will be explained in §6.S). G The Reeb Foliation on S3 (Reeb [1952]). Two orientable Recb foliations on the solid torus determine a foliation on S3 in the following way. The sphere S3 is the union of two solid tori which are identified along their common boundary, the torus y2, by the diffeomorphism taking meridians of one to parallels of the other and vice-versa. This foliation is called the Reeb foliation of S3; it has one leaf diffeomorphic to the torus y2 and all its other leaves are diffeomorphic to ]R2 and accumulate on the torus. Below we describe, pictorially, the
332
Chapter 4 Vector Fields and Dynamical Systems
decomposition of S3 in two solid tori. Remove the north pole. (1. O. o. 0). of S3 and stereographically project the rest of S3 onto ]R3. In the plane (x2• x4) draw two equal circles centered on the x2 -axis at the points a and b. where - a = b. Rotating about the xCaxis yields the solid torus in ]R3. Now draw all of the circles in the (x 2• x4) plane minus the two discs. centered on the x4-axis and passing through a and b. Each such circle yields two connected arcs joining the two discs. In addition. consider the two portions of the x2-axis: the line joining the two discs and the two rays going off from each disk separately. (See Figure 4.4.4.) Now rotate this figure about the x4-axis. All arcs joining the disks generate smooth surfaces diffeomorphic to ]R2 and each such surface meets the solid torus along a parallel. Only the two rays emanating from the disks generate a surface diffeomorphic to the cylinder. Now add the north pole back to S3 and pull back the whole structure via stereographic projection from ]R3 to S3: the cylinder becomes a torus and all surfaces diffeomorphic to ]R2 intersect this torus in meridians. Thus S3 is the union of two solid tori glued along their common boundary by identifying parallels of one with meridians of the other and vice-versa.
Figure 4.4.4 4.4.6 Proposition Let M be a manifold and <1> = {LalaE
A
be afoliation on M. The set
is a subbundle of TM called the tangent bundle to the foliation. The quotient bundle, denoted v(<1» = TM!f(M, <1», is called the normal bundle to the foliation <1>. Elements of T(M. <1» are
called vectors tangent to the foliotion <1>. Proof Let (U. cp), cp: U ~ U' x V' ~ E EEl F be a foliated chart. Since Tucp(TuLa) =E x {Ol for every u E un La' we have Tcp(TU n T(M;<1») = (U' x V') x (E x {O}). Thus the standard tangent bundle charts induced by foliated charts of M have the subbundle property and naturally induce vector bundle charts by mapping vm E Tm(M. <1» to (cp(m). Tmcp(vm E U' X V') x
»
Section 4.4 Frobenius' Theorem and Foliations
333
(E x (OJ) . •
4.4.7 The Global Frobenius Theorem Let E be a subbundle of TM. Thefollowing are equivalent: (i) There exists afoliation «1> on M such that E = T(M, «1». (ii) E is integrable. (iii) E is involutive. Proof The equivalence of (ii) and (iii) was proved in 4.4.3. Let (i) hold. Working with a foliated chart, E is integrable by 4.4.6, the integral submanifolds being the leaves of «1>. Thus (ii) holds. Finally, we need to show that (ii) implies (i). Consider on M the family of (local) integral manifolds of E, each equipped with its own submanifold topology. It is straightforward to verify that the family of finite intersections of open subsets of these local integral submanifolds defmes a topology on M, finer in general than the original one. Let {La} a E A be its connected components. Then, denoting by (La n U)IJ the connected components of La n U in U, we have by definition E I (U U La)/J = (E restricted to (U U La)lJ) equals T«U n La)/J). Let (..-l(U), 'If), U C M be a vector bundle chart of TM with the subbundle property for E, and let cp : U ~ U 1 be the induced chart on the base. This means, shrinking U if necessary, that cp : U ~ U' x V' C E e F and 'If(..-l(U) n E) = (U' x V') x (E x (O}). Thus cp«U n La)lJ) = U' x {ca lJ } and so M is foliated by the family «1> = {La}a EA' Because E I (U n La)/J = T«U n La)IJ, we also have T(M, «1» = E. •
There is an important global topological condition that integrable subbundles must satisfy that was discovered by Bolt [1970]. The result, called the Bott Vanishing Theorem, can be found, along with related results, by readers with background in algebraic topology, in Lawson [1977]. The leaves of a foliation are characterized by the following property. 4.4.8 Proposition Let «1> be afoliation on M. Then x and yare in the same leaf if and only if x and y lie on the same integral curve of a vector field X defined on an open set in M and which is tangent to the foliation «1>. Proof Let X be a vector field on M with values in T(M, «1» and assume that x and y lie on the same integral curve of X. Let L denote the leaf of «1> containing x. Since X is tangent to the foliation, X I L is a vector field on L and thus any integral curve of X starting in L stays in L. Since y is on such an integral curve, it follows that y E L. Conversely, let x, y ELand let c(t) be a smooth non-intersecting curve in L such that c(O) = x, c(1) = y, c'(t) ~ O. (This can always be done on a connected manifold by showing that the set of points that can be so joined is open and closed.) Thus c : [0, 1] ~ L is an immersion, and hence by compactness of [0, 1], c is an embedding. Using definition 4.4.4, there is a neighborhood of the curve c in M which is diffeomorphic to a neighborhood of [0,1] x {OJ x {OJ in lR x F x G for Banach spaces F and G such that the leaves of the foliation have the local
334
Chapter 4 Vector Fields and Dynamical Systems
representation R x F x {w}, for fixed
WE
G, and the image of the curve c has the local
representation [0,1] x {O} x {O}. Thus we can find a vector field X which is defined by c'(t) along c and extends off c to be constant in this local representation. _ Let R denote the following equivalence relation in a manifold M with a given folation <1>: xRy if x, y belong to the same leaf of <1>. The previous proposition shows that R is an open equivalence relation. It is of interest to know whether M/R is a manifold. Foliations for which R is a regular equivalence relation are called regular foliations. (See §3.5 for a discussion of regular equivalence relations.) The following is a useful criterion. 4.4.9 Proposition Let <1> be afoliation on a manifold M and R the equivalence relation in M
Lm of M TmLm ffi Tm(M, <1» =
determined by <1>. R is regular ifffor every me M there exists a local submanifold such that
Lm
intersects every leafin at most one point (or nowhere) and
TmM. (Sometimes
L.n
is called a slice or a local cross-section for the foliation.)
Proof Assume that R is regular and let 1t: M -+ M/R be the canonical projection. For
Lm
choose the submanifold using the following construction. Since 1t is a submersion, in appropriate charts (U,
L.n
Lm = ",'({O}
x V'), we
satisfies the two required conditions.
Conversely, assume that each point me M admits a slice Lm. Working with a foliated chart, we are reduced to the following situation: let U, V be open balls centered at the origin in Banach spaces E and F, respectively, let L be a submanifold of U x V, (0,0) E L, such that T(O.O)L =F, and L () (U x {v}) is at most one point for all v E V. If P2 : E ffi F -+ F is the second projection, since P2 1L has tangent map at (0, 0) equal to the identity, it follows that for V small enough, P2 I L : L -+ V is a diffeomorphism. Shrinking L and V if necessary we can assume that L () (U x {v}) is exactly one point. Let q: V -+ L be the inverse of P2 and define the smooth map s: U x V -+ L by s(u, v) =q(v). Then L () (U x {v}) = {q(v)}, thus showing that L is a slice in the sense of Lemma 3.5.25. Pulling everything back to M by the foliated chart, the prior argument shows that for each point m E M there is an open neighborhood U, a submanifold
I.m of U, and a smooth map s: U -+ I.m such that Lu () I.m
= {s(u)},
where Lu is the leaf containing u E U. By the argument following Lemma 3.5.25, the equivalence relation R is locally regular, i.e., Ru = R () (U x U) is regular. If U' = 1t-'(1t(U» where 1t: M -+ M/R is the projection, the argument at the end of Step 1 in the proof of Theorem 3.5.24 shows that R is regular. Thus, all that remains to be proved is that U can be chosen to equal U'. But this is clear by defming s': U' -+ ~ by s'(u') =s(u), where u E U L"., Lu' being the leaf containing u'; smoothness of s' follows from smoothness of s by composing it locally with the flow of a vector field given by Proposition 4.4.8. _
n
To get a feeling for the foregoing condition we will study the linear flow on the torus.
Section 4.4 Frobenius' Theorem and FolioJions
335
4.4.10 Example On the two· torus ']["2 consider the global flow F: lR x ']["2 --+ ']["2 defined by F(t, (s\, S2»
=(sle2Itit, s2e21ti(lt)
for a fixed number a E [0, 1[. By Example 4.4.5C this defines
a foliation on ']["2. If a E Q, notice that every integral curve is closed and that all integral curves have the same period. The condition of the previous theorem is easily verified and we conclude that in this case the equivalence relation R is regular; ']["2/R = SI. If a is irrational, however, the situation is completely different. Let cp(t) = (e2Itit , e2ltiat) denote the integral curve through (1, 1). The following argument shows that c1(cp(lR» = ']["2; that is, cp(lR) is dense in ']["2. Let p = (e 2Itix , e2ltiy ) E ']["2; then for all mE Z,
where y = ax + z. It suffices to show that C = {e2Itim(l E Sl I m E Z} is dense in Sl because then there is a sequence mkE Z such that exp(21timka) converges to e 21tiz. Hence, cp(x + mk) converges to p. If for each k E Z+ we divide Sl into k arcs of length 21t/k, then, because {e2Itim (l E S I I m = 1, 2, ... , k + I} are distinct for some 1 S nk < mk ~ k + 1, exp(21tim ka) and exp(21tinka) belong to the same arc. Therefore, I exp(21tim ka) - exp(21tinka) I < 21t/k, which implies
I exp(21tiqka) - 11 < 27t/k, where qk = mk - nk. Because U {e21ti(lS E Sl Is E li~, (j + 1)~]} j e
=
Sl ,
z+
every arc of length less than 27t/k contains some exp(21tijqk)' which proves c1(C) = Sl. Thus any submanifold L m , mE ']["2 not coinciding with the integral curve through m will have to intersect cp(lR) infinitely many times; the condition in the previous theorem is violated and so R is not regular. • Remark Novikov [1965] has shown that the Reeb foliation is in some sense typical. A foliation Cl> on M is said to be transversally orlentable if TM = T(M, Cl» E9 E, where E is an orientable subbundle of TM (see Exercise 6.5N for the definition.) A foliation on a three dimensional manifold M is said to have a Reeb component if it has a compact leaf diffeomorphic to ']["2 or K and if the foliation within this torus or Klein bottle is diffeomorphic to the orientable or non-orientable Reeb foliation in the solid torus or the solid Klein bottle. Novikov has proved the following remarkable result. Let Cl> be a transversally orientable C2 codimension one foliation oj a compact three dimensional manifold M. If 1t)(M) isjinite, then Cl> has a Reeb component,
'*
which mayor may not be orientable. If 1t2(M) 0 (with no hypotheses on 1t)(M» and Cl> has no Reeb components, then all the leaves of Cl> are compact with finite fundamental group. We refer the reader to Camacho and Neto [1985] for a proof of this result and to this reference and Lawson [1977] for a study of foliations in general. Even though foliations encompass "nice" partitions of a manifold into submanifolds, there
Chapter 4 Vector Fields and Dynamical Systems
336
are important situations when foliations are inappropriate because they are not regular or the leaves jump in dimension from point to point.
Consider, for example,
]R2
as a union of concentric
circles centered at the origin. As this example suggests, one would like to relax the condition that MfR be a manifold, provided that M/R turns out to be a union of manifolds that fit "nicely" together. Stratifications, another concept allowing us to "stack" manifolds, tum out to be the natural tool to describe the topology of orbit spaces of compact Lie group actions or non-compact Lie group actions admitting a slice (see, for instance, Bredon [19721, Burghelea, Albu, and Ratiu
[19751, Fischer [19701, and Bourguignon [1975]). We shall limit ourselves to the definition in the finite-dimensional case and some simple remarks. 4.4.11 Definition Let M be a locally compact topological space. A stratification of M is a
partition of M into manifolds {Mala E A called strata, satisfying the following conditions: 51 Ma are manifolds of constant dimension; they are submanifolds of M if M is itself
52
a manifold. Thefamity {Maal of connected components of all the Ma is a locally finite partition of M; i.e.Jor every m
53
E
M, there exists an open neighborhood U of m in M
intersecting only finitely many Mb~' If Ma a n cl(Mb~)"# 0 for (a, a) "# (b, 13), then Ma a c Mb~ and dim(M a) < dim(Mb)·
54
cl(Ma)\Ma is a disjoint union of strata of dimension strictly less than dim(Ma)'
From the defmition it follows that if Maa n cl(Mb~) "# 0 and if me Maa c d(Mb~) has an open neighborhood U in the topology of Ma a such that U necessarily Maa
C
C
Ma a
n
cl(Mb~)' then
Mb~ and thus dim(M.) < dim(M b). To see this, it is enough to note that the
given hypothesis makes M.a n cl(Mb~) open in Maa. Since it is also closed (by definition of the relative topology) and Ma a is connected, it must equal M.a itself, whence Maa
C
cl(Mb~) and
Mb~ and dim(Ma) < dim(Mb). . For nonregular equivalence relations R, M/R is often a stratified space. The intuitive idea is
by 53, Ma a
C
that it is often possible to group together equivalence classes of the same dimension, and this grouping is parametrized by a manifold, which will be a stratum in MfR. A simple example is partitioned by circles (the equivalence classes for
]R2
R). The circles of positive radius are
parametrized by the interval 10, oc[. Thus M/R is the stratified set [0, oc[ consisting of the two strata {Ol and 10, oc[.
Exercises 4.4A Let M be a manifold such that TM = E\ EEl ... EEl E p' where Ei is involutive. Show that
there are local charts (U,
Section 4.4 Frobenius' Theorem and Foliotions
4.4B
337
In]R4 consider the family of surfaces given by x2 + y2 + z2 - t2 = const. Show that these surfaces define a stratification. What part of ]R4 should be thrown out so as to obtain a regular foliation?
4.4C
Let f: M -+ N be a Coo map and Cl> a foliation on N. The map f is said to be
transversal to Cl>, denoted f r1i Cl>, if for every mE M, Tmf(TmM) + Tf(m/N, Cl»
=
Tf(m)N and (TmO-l(Tf(m)(N, Cl>)) splits in TmM. Show that if (LalaE A are the leaves of Cl>, the connected components of 11(La) are leaves of a foliation (denoted by f* (Cl>)) on M, and if Cl> has finite codimension in N, then so does the foliation r*(Cl» on M and the two codimensions coincide. 4.4D
(Bourbaki (1971)). Let M be a manifold and denote by M' the manifold with underlying set M but with a different differentiable structure. Show that the collection of connected components of M' defines a foliation of M iff for every m E M, there exists an open set
V in M, mEV, a manifold N, and a submersion p: V -+ N such that the submanifold p-l(n) of V is open in M' for all n E N. (Hint: For the "if' part use 3.3.5 and for the "only if' part use Exercise 3.2F to define a manifold structure on the leaves; the charts of the second structure are (V 4.4E
n La)P
-+ V'.)
On the manifold 80(3), consider the partition LA = (QA I Q is an arbitrary rotation about the z-axis in ]R31, A E 80(3). Show that Cl> = {LA I A E 80(3)} is a regular foliation and that the quotient manifold 80(3)IR is diffeomorphic to S2.
4.4F
(Hirsch and Weinstein). Give another proof of the Frobenius Theorem as follows:
Step 1
Prove it for the abelian case in which all sections of E satisfy [X, Yj
=0
by choosing a local basis X!' ... , Xk of sections and successively Ilowing out by the commuting flows of XI' ... , Xk. Step 2 Given a k-dimensional plane field, locally write it as a "graph" over ]Rk. Choose k commuting vector fields on
]Rk and lift them to the plane field. If E is
involutive, the bracket of two of them lies in E and, moreover, since the bracket "pushes down" to ]Rk (by "relatedness"), it is zero. (This is actually demonstrated in fonnula (1». Now use Step I.
ChapterS Tensors In the previous chapter we studied vector fields and functions on manifolds. Now these objects are generalized to tensor fields, which are sections of vector bundles built out of the tangent bundle. This study is continued in the next chapter when we discuss differential forms, which arc tensors with special symmetry properties. One of the objectives of this chapter is to extend the pull-back and Lie derivative operations from functions and vector fields to tensor fields.
§5.1 Tensors on Linear Spaces Preparatory to putting tensors on manifolds, we first study them on vector spaces. This subject is an extension of linear algebra sometimes called "multilinear algebra." Ultimately our constructions will be done on each fiber of the tangent bundle, producing a new vector bundle. As in Chapter 2, E, F, ... denote Banach spaces and Lk(Et' ... , ~; F) denotes the vector space of continuous k-multilinear maps of El x ...
X ~
to F. The special case L(E, JR.) is
denoted E *, the dual space of E. If {el''''' en} is an ordered basis of E, there is a unique ordered basis of E*, thedualbasis (e1, ... ,en), such that (ei,e)=&ii where &ii=l if j=i and 0 otherwise. Furthermore, for each e E E, n
e=
L
(e i, e) ei
i= 1
and for each o.E E*, n
0.=
L (a, e) e
i ,
i= 1
where ( , ) is the pairing between E and E*. Employing the summation convention whereby summation is implied when an index is repeated on upper and lower levels, these expressions become e = (e i , e) ei and a = (a, e) e i . As in Supplement 2.4C, if E is infinite dimensional, by E* we will mean another Banach space weakly paired to E; it need not be the full functional analytic dual of E. In particular, E** will always be chosen to be E. With these conventions, tensors are defined as follows.
339
Section 5.1 Tensors on Linear Spaces
5.1.1 Definition For a vector space E we put Fs(E) = U+S(E*, ... , E*, E, ... , E; JR.) (r copies of E* and s copies of E). Elements of T's(E) are called tensors on E, contravariant of order r and covariant of order s; or simply, of type (r, s). Given tl
t2
E
E
T' sl(E)
and t2
E
T2 '2(E), the tensor product
of tl and t2 is the tensor tl 0
T,+r2 sI+s 2(E) defined by (tl 0 t2)(~I, ... , ~r" yl, ... , yr2, fl' ... , fs" gl' ... , gs)
tl(~I, ... , ~r" fl' ... , fs, ) t2(yl, ... , yT2, gl' ... , gs)
=
Replacing JR. by a space F gives Fs(E; F), the F-valued tensors of type (r, s). The tensor product now requires a bilinear form on the value space for its definition. For JR.-valued tensors, 0 is associative, bilinear and continuous; it is not commutative. We also have the special cases TIO(E)
= E,
~I(E)
= E*, T0 2(E) = L(E; E*),
and TII(E)
= L(E;
E) and make the
convention that "J'lo(E; F) = F.
5.1.2 Proposition Let E be an n dimensional vector space. If {e l , ... , en} is a basis of E and {e l , ... , en} is the dual basis, then
is a basis of Fs(E) and thus dim (Fs(E))
=nr+s.
Proof We must show that the elements
of Fs(E) are linearly independent and span Fs(E). Suppose a finite sum
Apply this to (e k" ... , e\ et , ' ... , et) to get tk, ... k , t, ... t, = O. Next, check that for t we have t = t(e\ ... , e i" e)" , ... , e)· ) ei 0 ... 0 ei 0 ri' 0 ... 0 ri, .• 1
s I r
The coefficients
and called the components of t relative to the basis {e l , ... , en}'
E
Ts(E)
340
Chapter 5 Tensors
5.1.3 Examples A If t is a (0. 2)·tensor on E then t has components tij = t(ei• e}. an n x n matrix. This is the usual way of associating a bilinear form with a matrix. For instance. in JR2 the bilinear form
(where x =(xl' x2) and y =(Yl' Y2» is associated to the 2 x 2 matrix
[~ ~J
.
B If t is a (0. 2)-tensor on E. it makes sense to say that t is symmetric; i.e .• t(ei• ej) = t(ej • eJ This is equivalent to saying that the matrix [tij ] is symmetric. Symmetric (0. 2)-tensors t can be recovered from their quadratic form Q(e) = t(e. e) by t(el' e 2) = [Q(e!+~) - Q(e!- e 2)]/4. the polarization identity. If E =JR2 and t has the matrix
then Q(x) = Ax!2 + 2Bx! x 2 + CX2 2 . Symmetric (0. 2)-tensors are thus closely related to quadratic forms and arise. for example. in mechanics as moment of inertia tensors and stress tensors. C In general. a symmetric (r. O)-tensor is defmed by the condition
for all permutations (J of{I ..... rJ, and all elements o.! ..... o.re E*. One may associate to t a homogeneous polynomial of degree r. P(o.) = t(o. ..... 0.) and as in the case r = 2. P and t determine each other. A similar definition holds for (0. s)-tensors. It is clear that a tensor is
symmetric iff all its components in an arbitrary basis are symmetric. D An inner product ( • ) on E is a symmetric (0. 2)-tensor. Its matrix has components Thus ~j is symmetric and positive definite. The components of the inverse matrix are written gij.
~j
=(ei• ej ).
E The space Lk(EI' .... ~; F) is isometric to Lk(EO(!)' .... EO(k); F) for any permutation (J of (I ..... kJ, the isometry being given by A ...... A'. where A'(e o(1)' .... eO(k) = A(el' .... ~). Thus if t e T's(E; F). the tensor t can be regarded in C(r +s. s) (the number of ways r + s objects chosen s at a time) ways as an (r + s)-multilinear F-valued map. For example. if t E T2!(E). the standard way is to regard it as a 3-linear map t: E* x E* x E ~ lR. There are two more ways to interpret this map. however. namely as E* x E x E* ~ lR and as E x E* x E* ~ JR. In finite dimensions. where one writes the tensors in components. this distinction is important and is reflected in the index positions. Thus the three different tensors described above are written
341
Section 5.1 Tensors on Linear Spaces
F In classical mechanics one encounters the notion of a dyadic (cf. Goldstein [1980]). A dyadic is the formal sum of a finite number of dyads. a dyad being a pair of vectors ei' e 2 E JR.3 written in a specific order in the form elCz. The action of a dyad on a pair of vectors. called the double dot product of two dyads is defmed by
where· stands for the usual dot product in JR.3. In this way dyads and dyadics are nothing but (0.
'Pl2(JR.3). by identifying (JR.3)* with JR.3.
2)-tensors on JR.3; i.e .• e le 2 = e l ® e 2 E
G Higher order tensors arise in elasticity and Riemannian geometry. In elasticity. the stress tensor is a symmetric 2-tensor and the elasticity tensor is a fourth-order tensor (see Marsden and Hughes [1983]). In Riemannian geometry the metric tensor is a symmetric 2-tensor and the curvature tensor is a fourth-order tensor (see Section 5.6).• The interior product of a vector vEE (resp. a form /3 E E*) with a tensor t E P.(E; F) is the (r. s-l) (resp. (r -1. s» type F-valued tensor defmed by (iv t)(/3I •...• /3r• vi' ...• vs- I) =t(/3I •...• /3r• v. vi' ...• vs- I ) (i 13t)(/3 I•...• /3r-l. v l'
v.) = t(/3. /3 1•...•
...•
13'-1. v l' ...• v.).
Clearly. iv: F.(E; F) ~ P._I(E; F) and i 13 : P.(E; F) ~ P-I.(E; F) are linear continuous maps. as are v 1-+ iv and /31-+ i13 • If F =JR. and dim(E) =n. these operations take the following form in components. If <\ (resp.
ek) denotes the k-th basis (resp. dual basis) element of E. we
have
By 5.1.2 these formulas and linearity enable us to compute any interior product. Let dim(E) = n. The contraction of the k-th contravariant with the ~-th covariant index. or for short. the (k. ~)-contraction, is the family of linear maps Ck , : P.(E; F) ~ p-I._I(E) defined for any pair of natural numbers r. s ~ 1 by ck,(til ... i·il ...i,eil
® ... ® eia ®
= tll··.ik_lpik+I··.i...
dl ®
h···J'-lPJ'+l
... ® d')
. e· ···jr
11
® ... ®
e·it ® ... ® e·I. ® dl ® ... ® ~I ®
Where {e" ...• en} is a basis of E. {e I •...• en} is the dual basis in E*, and
... ®
d,.
over a vector or
342
Chapter 5 Tensors
covector means that it is omitted. It is straightforward to verify that Ckl so defined is independent of the basis. If E is infinite dimensional. contraction is not defined for arbitrary tensors. One introduces the so-called contraction class tensors. analogous to the trace class operators. defines contraction as above in terms of a Banach space basis and its dual. and shows that the contraction class condition implies that the definition is basis independent. We shall not dwell upon these technicalities. refer to Rudin [1973] for a brief discussion of trace class operators. and invite the reader to model the concept of contraction class along these lines. For example. if E* = E = ~2(lR). and ei = ei equals the sequence with 1 in the i-th place and zero everywhere else. then
n=O
n=O
The Kronecker delta is the tensor 1) E Tl1(E) dermed by 1)(a. e) = (a. e). If E is finite dimensional. 1) corresponds to the identity I E L(E; E) under the canonical isomorphism Til (E) == L(E; E). Relative to any basis. the components of 1) are the usual Kronecker symbols 1)ij• i.e .• 1) = 1)ijei ® ei. Suppose E is a finite-dimensional real inner product space with a basis {el' .... en} and corresponding dual basis {e 1..... en} in E*. Using the inner product. with matrix denoted by [gij]' we get the isomorphism (, : E ~ E* given by x - (x •. ) and its inverse
#:
E* ~ E.
The matrix of (, is [gij]; i.e.• and of
#
is [gij]; i.e .•
where xj and a j are the components of e and a. respectively. We call (, the index lowering operator and # the index raising operator. These operators can be applied to tensors to produce new ones. For example if t is a tensor of type (0.2) we can defme an associated tensor t' of type (1.1) by t'(e. a)
=t(e. at).
The components are (t')ij = gjktik (as usual. sum on k). In the classical literature one writes tii for giktik • and this is indeed a convenient notation in calculations. However. contrary to the impression one may get from the classical theory of Cartesian tensors. t and t' are different tensors.
Section 5.1 Tensors on Linear Spaces
343
5.1.4 Examples Let E be a finite-dimensional real vector space with basis {el •...• en} and dual basis {e I•...• en}. A If t E T21(E) and x = xiei• then
Thus. the componcnts of ixt are xPtkl p. The interior product of the same tensor with Cl = ClpC P takcs the form
B If t E T2 3(E). the (2.1)-contraction is given by "" em) -_ tij jIm ei "" I "" m . e 2I(tij kim ci "" "" Cj "" "" ek "" "" e I "" '01 e '01 e e An important particular example of contraction is the trace of a (1. 1)-tcnsor. N amcly. if t E TII(E). then trace(t) = ell (t) = tii• wherc t = tij ei ® ei. D The components of the tensor associated to g by raising the second index are gjkgik = gjkgki
= Oii' E Let
Thcn t has quite a fcw associated tensors. depending on which index is lowered or raised. For examplc tijk Im-- gmp tijk Ip' tijkl m = gja gkb gmc tiab Ic ~jklm = ~a gjb glc gmd tabkcd ~jk 'm = ~a gkb glc tBjbem
and so on. F The positioning of the indices in the components of associated tensors is important. For example. if t E J'l2(E). wc saw carlier that tij = gjk tik . However. t ji = gij tki • which is in gencral different from t~ whcn t is not symmetric. For example. if E = 1R3 with gij = Oij and the nine components of t in the standard basis are tl2 = I. ~I = -1. tij = 0 for all other pairs i. j. then t~ = tij' tji = tji • so that tl2 = tl2 = 1 while t21 = ~I =-l. G The trace of a (2. O)-tensor is defmed to be the trace of the associated (1. 1) tensor; i.e .• if t = tijei ® cj• then trace(t) = ti i = ~tik. The question naturally ariscs whether we get the same answer by lowering the first index instead of the second. i.e .• if we consider t{ By symmetry of ~j we have ti i = gkitik = tkk. so that the definition of the trace is independent of which index is
344
Chapter 5 Tensors
lowered. Similarly, if t E 'J'l2(E), trace(t)
=tii =gik tik =tkk.
In particular trace(g)
=g/ =gik ik =
dim(E) .• Now we turn to the effect of linear transformations on dual spaces. If
transpose of
E
E
L(E, F), thc
L(F*, E*) is defmed by (
and e E E. Let us analyze the matrices of
computed by mUltiplying v on the left by A, the matrix of
E
L(F*, E*). In the dual ordered bases, (
transpose of the matrix of
5.1.5 Definition If
E
L(T's(E), T's(F» by
where t
E
T'.(E), ~I, ... , WE F*, and fl' ...• fs
E
F.
We leave the verification that
5.1.6 Proposition Let
~
G be isomorphisms. Then
(i) ('I' 0
E
T'\(E) and t2 E T'\(E), then
Section 5.1 Tensors on Linear Spaces
345
Proof For (i). "'*(
=
=('" where yl •...• yr
0
G*. gl' ...• g.
E
E
G. and t
transposes and inverses satisfy ('" 0
=
0
E
F.(E).
",* and ('" 0
We have used the fact that the
=
0
",-I. which the reader
can easily check. Part (ii) is an immediate consequence of the definition and the fact that i* i-I = i. Finally. for (iii) we have
0
= i*
0
=i
and
(
the identity on T'.(E). Hence (iii) follows. Finally (iv) is a
straightforward consequence of the defmitions. _ Since (
5.1.7 Proposition Let
...•
en }
and
Then the components of
Proof We have e·1 = tn-I(tn(e.» TTl
= tn-I(A".f)a = Aa.tn-I(f) ]"1" a = Aa. Biae·J'
whence Bia A". l =l &. for all I I i.j. Similarly. one shows that Abi Bi. = Sba• so that [A"i l- = [Bi.l. We have T
1
(
»
= t(
346
Chapter 5 Tensors
To prove the second relation. we need the matrix of (
E
L(E*. F*). We have
so that (
E
L(E*. F*) is the transpose of the inverse of the matrix ofcp.
The assumption that
L(E. F) (not necessarily an isomnrphism). define the pull-back cp*
E
E
L(Tls(F) Tls(E» by
where t E "Pl s(E) and
el' ...• e s E E.
The next proposition asserts that
=
0
\jf*.
If i: E ~ E is the identity. then so is i* E L("Pls(E). TOs(E».
(iii) If
.
(IV)
If tl
E
°
°
T sl(F) and t2 E T S2 (F). then
Finally the components of cp*t are given by the following. 5.1.10 Proposition Let E and F be finite-dimensional vector spaces and
E
L(E. F). For
ordered bases {el' ...• en} of E and {fl' ...• fm} of F. suppose that
tE
T~(F) have components tbl ...b,' Then the components of
Section 5.1 Tensors on Linear Spaces
347
Proof
(
Abl.1I ... A b,J,. •
5.1.11 Examples A On]R2 with the standard basis {el' e 2}. let t E T2o(]R2) be given by t = e l ® e l + 2e l ® e 2 - e 2 ® e l + 3e2 ® e2 and let
A=[~
a
Then
8
=
[1-IJ
-1 2 .
According to 5.1.7. the components of T =
so that TI2 = 811B21t11 + BIIB22tl2 + BI2B21t21 + B1 2B\t22 = J.(-I)·1 + 1·2·2 + (-1)·(-1)-(-1) + (-1)·2·3 =-4 T21 = B2lB l l t 11 + B21Bl
i
2 + B22Bllt21 + B22BI2t22
=(-1)·1·1 + (-1)-(-1)·2 + 2· J.(-I) + 2·(-1)·3 =-7 Til = BIIB l l t 11 + BIIBI2tl2 + BI2Bllt21 + BI2BI2t22
= 1·1·1 + 1·(-1)·2 + (-1)·1·(-1) + (-1)-(-1)·3 =3
i
T22 = B21B 21t11 + B\ B2
2 + B22B 21t21 + B22B 22t22 = (-1)·(-1)·1 + (-1)·2·2 + 2·(-1)-(-1) + 2·2·3 = 11.
8 Let t = e l ® e2 - 2ez ® e 2 E TII(lR2) and consider the same map
348
Chapter 5 Tensors
A=[~
a
Again we shall compute
Exercises
=
5.1A Compute the interior product of the tensor t e l ® e l ® e2 + 3e2 ® e2 ® e l with e + 2e2 and <X = 2e l + e2 . What are the (1, 1) and (2, 1) contractions of t?
=- e l
5.1D Compute all associated tensors of t = e l ® e2 ® e 2 + 2e2 ® e l ® e2 - e2 ® e2 ® e l with respect to the standard metric of JR.2. 5.1C Let t
=2e l ® e l -
by the matrices
e 2 ® e l + 3e l ® e 2 and
E
L( ]R2, ]R2), 'I' E L(]R3, ]R2) be given
[-12 lIJ' [01 01-IJ2 .
Compute: trace(t),
~5.2
Tensor Bundles and Tensor Fields
We now extend the tensor algebra to local vector bundles, and then to vector bundles. For U
C
E (open) recall that U x F is a local vector bundle. Then U x Trs(F) is also a local vector
bundle in view of 5.1.2. Suppose
where t E Trs(F). Before proceeding, we shall pause and recall some useful facts concerning linear isomorphisms from 2.5.4 and 2.5.5. 5.2.2 Proposition Let GL(E, F) denote the set of linear isomorphisms from E to F. Then GL(E, F) C L(E, F) is open. 5.2.3 Proposition Let .J'l: L(E, F) --+ L(F*, E*);
Smoothness of .J'l is clear since it is linear. 5.2.4
Proposition If
isomorphism for all u E U, then
350
Chapter 5 Tensors
product rule, (cp)* is smooth. _ This smoothness can be verified also for finite-dimensional bundles by using the standard bases in the tensor spaces as local bundle charts and proving that the components of cP*t are
C~
functions. We have the following commutative diagram, which says that cp* preserves fibers: cp* U x Fs(F) - - - - - -....- U' x Fs(F')
"!
U -------~--------4.-
CPo
5.2.5 Definition Let
1t:
!.
U'
E ~ B be a vector bundle with Eb = 7r 1(b) denoting the fiber over
1t" : Trs(E) ~ B by 1tr s(e) = b where e E Fs(Eb). Furthermore,for a given subset A of B, we define Fs(E) I A = UbEAF,(Eb)' If 1t' : E' ~ B' is another vector bundle and (cp, CPo) : E ~ E' is a vector bundle mapping with CPb =
the point bE B. Define T,,(E) = U bEB Fs(Eb) and
cP I Eb an isomorphism for all bE B, let cp*: F.(E)
~
Fs(E') be defined by cp* I F.(Eb) =
(CPb)* . Now suppose that (E I U, cp) is an admissible local bundle chart of
1t,
where U
C
B is an
open set. Then the mapping cp* I [Fs(E) I UJ is obviously a bijection onto a local bundle, and thus is a local bundle chart. Further, (CP*)b = (~)* is a linear isomorphism, so this chart preserves the linear structure of each fiber. We shall call such a chart a natural chart of Fs(E). 1t : E ~ B is a vector bundle, then the set of all natural charts of B is a vector bundle atlas.
5.2.6 Theorem If ~
1tr. :
T rs(E)
Proof Condition VB1 is obvious. For VB2, suppose we have two overlapping natural charts, cP* and '1'*. For simplicity, let them have the same domain. Then (l = 'V 0 cp-I is a local vector bundle isomorphism, and by 5.1.6, '1'* 5.2.4. _
0
(cp*)-I = (l*, is a local vector bundle isomorphism by
1t",
This atlas of natural charts called the natural atlas of generates a vector bundle structure, and it is easily seen that the resulting vector bundle is Hausdorff, and all fibers are isomorphic Banachable spaces. Hereafter, Fs(E) will denote all of this structure. 5.2.7 Proposition If f: E ~ E' is a vector bundle map that is an isomorphism on each fiber, then f* : T'.<E) ~ T'.(E') is also a vector bundle map that is an isomorphism on each fiber. Proof Let (U, cp) be an admissible vector bundle chart of E, and let (V, 'V) be one of E' so
Section 5.2 Tensor Bundles and Tensor Fields
that f(U)
C
351
V and fep'll ='I' 0 f 0
atlas. we see that
5.2.8 Proposition Let f: E
~E'
and g: E'
~
E" be vector bundle maps that are isomorphisms
on each fiber. Then so is go f. and
(i) (g 0 f)* =g* 0 f*; (ii) if i : E ~ E is the identity. then i*: T'.(E)
~
T'.(E) is the identity.
(iii) if f: E ~ E' is a vector bundle isomorphism. then so if f* and (f*>-I
=(r-I)*.
Proof For (i) we examine representatives of (g 0 f)* and g* 0 f*. These representatives are the same in view of 5.1.6. Part (ii) is clear from the definition. and (iii) follows from (i) and (ii) by the same method as in 5.1.6.• We now specialize to the case where 7t: E ~ B is the tangent vector bundle of a manifold. 5.2.9 Definition Let M be a manifold and 'eM : TM ~ M its tangent bundle. We call T'.(M) = T'.(TM) the vector bundle ojtensors contravariant order r and covariant order s. or simply of type (r. s). We identify TIO(M) with TM and call ~I(M) the cotangent bundle of M also denoted by 'e* M : T*M ~ M. The zero section of Tr.(M) is identified with M. Recall that a section of a vector bundle assigns to each base point b a vector in the fiber over b and the addition and scalar multiplication of sections takes place within each fiber. In the case of T'.(M) these vectors are called tensors. The C'" sections of 7t : E ~ B were denoted r"'(7t). or f""(E). Recall that 1" (M) denotes the set of mappings from Minto JR. that are of class C'" (the
standard local manifold structure being used on JR.) together with its structure as a ring; namely. f + g. cf. fg for f. g E 1" (M). c E JR. are defined by (f + g)(x) = f(x) + g(x). (cf)(x) = c(f(x» and (fg)(x) = f(x) g(x). Finally. recall that a vector field on M is an element of X(M) = r"'(TM). 5.2.10 Definition A tensor field ojtype (r. s) on a manifold M is a C'" section of T'.(M). We denote by 'I\(M) the set r"'(T'.(M» together with its (infinite-dimensional) real vector space structure. A co vector field or differential one-jorm is an element of X*(M) = TJI(M). If f E 1{M) and t E T.(M). let ft: M ~ T'.<M) be defined by m ~ f(m)t(m). If Xi E X(M). i = 1..... s. a) E X*(M). j = 1..... r. and t' E 'Tr'.. (M) define
t(a l ..... ar. Xl ..... X.) : M ~ JR. by m 1-+ t(m)(al(m) ..... X.(m» and
t ® t' : M ~ T'+<'.+., (M) by m 1-+ t(m) ® t'(m). 5.2.11 Proposition With f. t. Xi' a i • and t' as in 5.2.10. ft and t ® t' E 'Tr+<'.+.' .
E
'Tr.(M). t(a l ..... X.)
E
1"(M).
352
Chapter 5 Tensors
Proof The differentiability is evident in each case from the product rule in local representation. _ For the tangent bundle TM. a natural chart is obtained by taking T
Now we tum to the expression of tensor fields in local coordinates. Recall that a/iJx i = (T
-"1-" i,) E 1"(U).
... , {} aX
Applying 5.1.6(iv) at every point yields the coordinate expression of an (r. s)-tensor field:
To discuss the behavior of these components relative to a change of coordinates. assume that Xi: U --+ JR, i = 1•...• n is a different coordinate system. We can write a(dxi = a;ia/axi, since both are bases of ~(U). Applying both sides to Xk yields aii = aXi/ax i i.e.. (a/ax i) = (aXi/axi)(a/aXi). Thus the dxi. as dual basis change with the inverse of the Jacobian matrix [aXi/ax i]; i.e .• dxi = (CJxi/aXi)dXi. Writing t in both coordinate systems and isolating equal terms gives the following change of coordinate formula for the components:
This formula is known as the tensoriality criterion: Asetof nr+'functions til ... i'il ...i, definedforeachcoordinate system on the open set U of M locally define an (r. s)-ensor field iff changes of coordinates have the aforementioned effect on them. This statement is clear since at every point it assures that the nr+s functions are the components of an (r, s)-tensor in TuU and conversely. The algebraic operations on tensors, such as contraction. inner products and traces. all carry over fiberwise to tensor fields. For example. if 15m E Tt 1(Tm M) is the Kronecker delta, then /): M --+ Tl 1(M); m ...... 15m is obviously C~. and /) E 'T 1 1(M) is called the Kronecker delta. Similarly. a tensor field of type (0. s) or (r.O) is called symmetric. if it is symmetric at every point. A basic example of a symmetric covariant tensor field is the following. 5.2.12 Definition A weak pseudo-Riemannian metric on a manifold M is defined to be a tensor field g E rz1l2(M) that is symmetric and weakly nondegenerate. i.e .• such that at each mE M.
Section 5.2 Tensor Bundles and Tensor Fields
353
g(m)(vrn • wrn) =0 for all wrn E TrnM implies vrn =O. A strong pseudo-Riemannian metric is a 2-tensor field that. in addition is strongly nondegenerate for all m E M; i.e,. the map vrn ..... g(m)(vrn •. ) is an isomorphism of T mM onto Tm *M. A weak (resp. strong) pseudo-Riemannian
metric is called weak (resp. strong) Riemannian if in addition g(m)(v m• vrn) > 0 for all vm
E
TrnM. vm #0. A strong Riemannian manifold is necessarily modeled on a Hilbertizable spllce; i.e .• the model space has an equivalent nonn arising from an inner product. For finite-dimensional manifolds weak and strong metrics coincide: indeed TrnM and T m*M have the same dimension and so a one-to-one map of TmM to Tm *M is an isomorphism. It is possible to have weak metrics on a Banach or Hilbert manifold that are not strong. For example. the L2 inner product on M = CO([O. IJ.JR) is a weak metric that is not strong. For a similar Hilbert space example. see Exercise S.2e. Any Hilbert space is a Riemannian manifold with a constant metric equal to the inner product. A symmetric bilinear (weakly) nondegenerate two-fonn on any Banach spaee provides an example of a (weak) pseudo-Riemannian constant metric. A pseudo-Riemannian manifold used in the theory of special relativity is JR4 with the Minkowski pseudo-Riemannian metric
where x. v. w
E
JR4.
As in the algebraic context of §5.1. pseudo-Riemannian metrics (and for that matter any strongly nondegenerate bilinear tensor) can be used to define associated tensors. Thus the maps " become vector bundle isomorphisms over the identity b: TM ~ T*M. ': T*M ~ TM; , is
b
the inverse of b. where vmb = g(m)(v m•. ). In particular. they induce isomorphisms of the spaces of sections b: X(M) ~ X*(M). ': X*(M) ~ X(M). In finite dimensions this is the operation of raising and lowering indices. Thus fonnulas like the ones in Example S.I.4E should be read pointwise in this context. There is. however. a particular index raising operation that requires special attention. 5.2.13 Definition Let M be a pseudo-Riemannian n-manifold with metric g. For f E .r(M). grad f =(dO' E X(M) is called the gradient of f. To find the expression of grad f in local coordinates. we write gjj=g ( -a, • -a) , • X= ax' axJ so we have
x ja -,. axJ
and Y=Y ja -, • ax'
(a a) "
' , _n " J. (X b . Y) = g(X. Y) = X'yJg -,. - , = X'yJg jj • i.e.. x· = X'gjjdx ax' axJ
If (X E X*(M) has the coordinate expression (X
= (Xjdx j•
we have (X' = (Xjgija/OXj where [gijJ
Chapter 5 Tensors
354
is the inverse of the matrix
[~l
Thus for
IX
= df. the local expression of the gradient is
.. af a grad f = g'! - . - . ax! ax'
. .. af (grad f)' = g'J - . . ax!
or
If M =]Rn with standard Euclidean metric gij = /iij' this formula becomes
grad f=
~ ~;
(4..... ~).
i.e .• grad f=
ax' ax'
ax
axn
the familiar expression of the gradient from vector calculus. Now we tum to the effect of mappings and diffeomorphisms on tensor fields.
5.2.14 Definition If
t E T's(M). let
0
t0
5.2.15 Proposition If
n
E
T.(M).
Proof (i) The differentiability is evident from the composite mapping theorem. together with
5.2.4. The other three statements are proved fiberwise. where they are consequences of 5.1.6.• As in the algebraic context. the pull-back of covariant tensors is defined even for maps that are not diffeomorphisms. Globalizing 5.1.8 we get the following.
5.2.16 Definition If
The next proposition is similar to 5.2.15 and is proved by globalizing the proof of 5.1.9.
5.2.17 Proposition If
t
E
Tl'
Section 5.2 Tensor Bundles and Tensor Fields (iv) if
E
'I os,(N), t2
E
355
'I 0s2(N), then
For finite-dimensional manifolds the coordinate expressions of the pull-back and push-forward can be read directly from 5.1.7 and 5.1.10, taking into account that T
=
denote by yi and j
the local expression of
= dim(M)
(i) If t E 'I's(M) and
are
If
t E 'I°s(N) and
(ii) If t E ryD'
Notice the similarity between the formulas for coordinate change and pull-back. The situation is similar to the passive and active interpretation of similarity transformations PAP-I in linear algebra. Of course it is important not to confuse the two.
= (x + 2y, y) and let t = 3x(a/ax) '1,\ ()R2). The matrix of
5.2.19 Examples A Let
E
2J [ a
aa~~'] -_ [01 21J'l'
1
--
In other words, a)
a
= ax
(a) a a '
[01 -12 J.
356
Chapter 5 Tensors
Noting that q>-I(x, y) = (x - 2y, y), we get
B With the same mapping and tensor, we compute q>*t. Since
q>*(dx) = dx + 2 dy, and q>*(dy) = dy, we have q>*t = 3(x+ 2y)q>*(
;J
® q> *(dy) + q> *( ;y) ® q>*(dy)
o ® dy + = 3(x + 2y) ox
(0 0) 0 0 - 2 ox + oy ® dy = (3x + 6y - 2) ox ® dy + ay ® dy.
C Let q>: JR.3 ~ lR.2, q>(x, y, z) = (2x + z, xyz) and t = (u + 2v)du ® du + (u)2 du ® dv E
tzi!2 (JR.2). Since q>*(du) = 2 dx + dz and q>*(dv) = yz dx + xz dy + xy dz, we have q>*t = (2x + z + 2xyz)(2 dx + dz) ® (2 dx + dz) + (2x + z)2 (2 dx + dz) ® (yz dx + xz dy + xy dz)
= 2[4x + 2z + 4xyz + (2x + z)2 yz] dx ® dx + 2(2x + z)2
XZ
dx ® dy
+ 2[2x + z + 2xyz + (2x + z)2 xy] dx ® dz + [4x + 2z + 4xyz + yz(2x + z)2]dz ® dx + xz(2x + z)2 dz ® dy + [2x + z + 2xyz + xy(2x + z)2 ]dz ® dz. D If q> : M
~
N represents the deformation of an elastic body and g is a Riemannian
metric on N, then C =q>*g is called the Cauchy-Green tensor, in coordinates
Thus, C measures how q> deforms lengths and angles. • Finally, we describe an alternative approach to tensor fields. Suppose .'J(M) and X(M) have been defined. With the "scalar multiplication" (f, X) 1-+ fX defined in 5.2.10, X(M) becomes an .'J(M)-module. That is, X(M) is essentially a vector space over .'J(M) , but the "scalars" .'1(M) form only a commutative ring with identity, rather than a field. Define
Section 5.2 Tensor Bundles and Tensor Fields
357
the .1(M)-linear mappings on X(M). and similarly T'.(M) = U+sreM) (X*(M) • .... X(M); J-""(M) the J-""(M)-multilinear mappings. From 5.2.10. we have a natural mapping 'I\(M) -+ Ts(M) which is J-""(M)-linear.
5.2.20 Proposition Let M be afinite-dimensional manifold or be modeled on a Banach space with norm C~ away/rom the origin. Then 'Tr,(M) is isomorphic to the J-""(M)-multilinear maps/rom X*(M) x ... x X(M) into J-""(M). regarded as spaces. In particular. X*(M) is isomorphic to x*(M).
J-""(M)-modules or as real vector
Proof Consider the map Ts(M) -+ U+s reM) (X*(M) • .... X(M); J-""(M» given by ~(a,I, ... , a,'. Xi' ...• Xs)(m) = ~(m)(a,I(m), ...• Xs(m». This map is clearly J-""(M)-linear. To show it is an isomorphism. given such a multilinear map ~. define t by t(m)(a,I(m) •...• Xs(m» = ~(a,I •...• X,)(m). To show this is well-defined we must show that. for each Vo E Tm(M). there is an X E X(M) such that X(m) = vO' and similarly for dual vectors. Let (V.
and all components of t are
C~
by hypothesis. _
The preceding proposition can be clearly generalized to the Ck situation. One can also get around the use of a smooth norm on the model space if one assumes that the multilinear maps arc localizable. that is. are defined on 3£*(V) x ... x X(V) with values in .1(V) for any open set V in a way compatible with restriction to V. We shall take this point of view in the next section. The direct sum 'T(M) of the 'T'.(M). including 'Too(M) = .'J(M). is a real vector space with <8> -product. called the tensor algebra of M. and if : M -+ M is a diffeomorphism. <1>* : 'T(M) -+ 'T(N) is an algebra isomorphism.
Chapter 5 Tensors
358
The construction of 'T(M) and the properties discussed in this section can be generalized to vector bundle valued (r, s)-tensors (resp. tensor fields), i.e., elements (resp. sections) of L(T*M EB ... EB T*M E9 TM EB ... EB TM, E ), the vector bundle of vector bundle maps from T*M (£) ... E9 TM (with r factors of T*M and s factors of TM) to the vector bundle E, which cover the identity map of the base M.
Exercises 5.2A Let
a
a
t = x ax ® dx ® dy + y i)y ® dy ® dy.
Show that
5.2D Let
~5.3
The Lie Derivative: Algebraic Approach
This section extends the Lie derivative Lx from vector fields and functions to the full tensor algebra. We shall do so in two ways. This section does this algebraically and in the next section. it is done in terms of the flow of X. The two approaches will be shown to be equivalent. We shall demand certain properties of Lx such as: if t is a tensor field of type (r. s). so is Lxt. and Lx should be a derivation for tensor products and contractions. First of all. how should Lx be defined on covector fields? If Y is a vector field and a is a covector field. then the contraction a· Y is a function. so Lx(a· Y) and Lx Y are already defined. (See §4.2.) However. if we would like to have the derivation property for contractions. namely Lx(a· Y)
=(Lxa)· Y + a· (LxY)'
which forces us to define Lxa by (Lxa) . Y =Lx(a . Y) - a . (LxY) for all vector fields Y. Since this defines an !F(M)-lincar map. Lxa is a well-defined covector field. The extension to general tensors now proceeds inductively in the same spirit. 5.3.1 Definition A differential operator on the full tensor algebra T(M) of a manifold M is a
collection {nT.(U)} of maps of TT.
M. any of which we denote merely that 001 n
n
~
0 and each open set U
C
(the r. s and U are to be inferredfrom the context), such
is a tensor derivation. or n commutes with contractions, i.e., n is JR.-linear and if t
E
'IT.(M),
al' .... a r
E
X*(M). and XI' .... X. E X(M).
then
r
+
L teal' .... naj' .... aT' X" ~,
002
n
•
.... X.) +
L teal' .... ar' X, ..... nx
k .....
X.).
k-'
is local. or is natural with respect to restrictions. That is,for U eye M open sets,
and IE T.CY)
Chapter 5 Tensors
360
(D t) I V = D (t I V)
E
'T',(V)
or thefollowing diagram commutes
Iu .. Ts(U)
7\(V)
jv
vj 'TTS(V)
IV
.. T,(V)
Note that we do not demand that D . be natural with respect to push-forward by diffeomorphisms. Indeed. several important differential operators. such as the covariant derivative. arc not natural with respect to diffeomorphisms. although the Lie derivative is. as we shall see. 5.3.2 Theorem Suppose for each open set V C M we have maps Eu: ![ (U) ~ ![(V) and 1"u: X(V) ~ X(V). which are (lR-linear) tensor derivations and natural with respect to restrictions. That is
(i) :Eu(f ® g) =(:Euf) ® g + f ® Eug for f. g E ![ (V); (ii) for f E ![ (M). :Eu(f I U) = (:EMf) I V; (iii) 1"u(f ® X) = (:Euf) ® X + f ® 1"uX for f E ![(V). and X E X(V);
(iv) for X E X(M). 1"uf(X I V)
=(1"MX) I U.
Then there is a unique differential operator D on 'T(M) that coincides with :Eu on ![ (V) and with 1"u on X(U).
Proof Since D must be a tensor derivation. define D on X*(U) by the formula (Da)' X = D(a· X) - a· (DX) = :Eu(a . X) - a· 1"uX for all X E X(V). By properties (i) and (iii). Da is ![(M)-linear and thus by the remark following 5.2.20. D so defined on X*(V) has values in X*(V). Note also that D(f ® a) = (:Ef) ® a + f ® (Da) for any a E X*(U). f E ![(U). This shows that D exists and is unique on X*(V) (by the Hahn-Banach theorem). Define Dt.; on 'TTS(V) by requiring 001 to hold: (Dut)(al' ...• aT' Xl' ...• Xs) = :Eu(t(al' ...• aT' Xl' ...• Xs» s
T
-L ~I
t(al •...• Daj •...• aT' XI ....• Xs) -
L
t(al •...• aT' XI' ...• 1"U X k'
...•
Xs)·
k=1
From (i). (iii). and 001 for Du on X*(U). it follows that Dut is an ![ (M)-multilinear map. i.c .. that Dut E 'T".(V) (see the comment following 5.2.20). The definition of Du on 'T",(U) uniquely determines Du from the property 001. Finally. if V is any open subset of U. by (ii) and (iv) it follows that Dv(t I V) =(Dut) I V. This enables us to define D on ![(M) by (Dt)(m) = (Dut)(m). where V is any open subset of M containing m. Since Du is unique. so is D.
Section 5.3 The Lie Derivative: Algebraic Approach
361
and so 002 is satisfied by the construction of :0 .• 5.3.3 Corollary We have (i) :O(t\ ® It. ) = :Ot\ ® t2 + t\ ®:Olt. ,and (ii) :00 = 0, where 0 is Kronecker's delta.
Proof (i) is a direct application of 001. For (ii) let a
E
3£*(U) and X E 3£(U) where U is an
arbitrary chart domain. Then (:Oo)(a, X) = :O(o(a, X» - o(:Oa, X) - o(a, :OX) =:O(a· X)-:Oa· X-a·:OX =0.
Again the Hahn-Banach theorem assures that :00 = 0 on U, and thus by 002, :00 = O. • Taking :Eu and :F'u to be Lx Iu we see that the hypotheses of theorem 5.3.2 are satisfied. Hence we can define a differential operator as follows. 5.3.4 Definition 1/ X E 3£(M), we let Lx be the unique differential operator on 'T(M), called the Lie derivative with respect to X, such that Lx coincides with Lx as given on .'T(M) and 3£(M) (see 4.2.6 and 4.2.20). 5.3.5 Proposition Let cp : M ~ N be a diffeomorphism and X a vector field on M. Then Lx is natural with respect to push-forward by cp; that is,
or the/ollowing diagram commutes: T ' , ( M ) - - - - - - - - - T',(N)
Lx!
r~'
'I\(M»----------'Tr,(N)
Proof For an open set U
C
M define
where we use the same symbol cp for cp I u. By naturality on .'T(U) and X(U), :0 coincides
Chapter 5 Tensors
362
with LxIU on !F (U) and Jt(U). Next, we show that D is a differential operator. For 001, we use the fact that *(t(al' .... <X.' Xl' ... , Xs» =(*t)(",al' ... , ",<X.. ",X I..... *Xs)' which follows from the definitions. Then for X. X\, ... , Xs E Jt(U) and al' .... a r
E
Jt"'(U).
r
+
L (",t)(*a l , ... , Lcp.x ",aj' ... , *a
r,
",X I ..... ",X.)
j= I
•
+
L (*t)(",al' ... , ",a
r,
",X I , ... , Lcp.x"'X k ,
... ,
",X s )].
k=1
by 001 for Lx. Since <1>* = (<1>-1)* by 5.2.14. thls becomes r
(Dt)(a l , ... , a r , XI' .... X.) +
L
(aI' .... Da j .... , a r , XI' ... , X.)
j= I
+
L•
t(al, ... ,a r , XI' .... DX k ,
....
X.).
k=1
For 002, let t E 'Tr'<M) and write Dt I U = [(",r l Lcp.x",tll U = (",rl[Lcp.x",tll U -I
=
(<1>"')
Lcp.xlu ",t I U
=
D(t I U).
(by 002 for Lx)
The result now follows by 5.3.2. • Using the same reasoning. a differential operator that is natural with respect to diffeomorphisms on functions and vector fields is natural on all tensors. Let us now compute the local formula for Lxt where t is a tensor field of type (r, s). Let <1>: U C M ~ veE be a local chart and let X' and t' be the principal parts of the local representatives, *X and ",t respectively. Thus X': V ~ E and t ': V ~ T'.(E). Recall from §4.2 that the local formulas for the Lie derivatives of functions and vector fields are:
Section 5.3 The Lie Derivative: Algebraic Approach (Lxf)'(x) = Df(x) . X'(x)
363
(1)
where f' is the local representative of f and (LxY)'(x) = DY'(x) . X'(x) - DX'(x) . Y'(x).
(2)
In finite dimensions these become (1')
and (2')
Let us first find the local expression for Lxa where a is a one-form. By 5.3.5. the local representative of Lxa is
which we write as Lx,a' where X' and a' are the principal parts of the local representatives. so X': V ~ E and a': V ~ E*. Let VEE be fixed and regarded as a constant vector field. Then as Lx is a tensor derivation. Lx, (a' . v) = ( Lx' a') . v + a'(Lx.v)· By (1) and (2) this becomes D(a'· v) . X' = (Lx' a') . v - a' . (DX' . v). Thus (Lx'(x') . v = (Da' . X') . v + a' . (DX'·v).
In the expression (Dcx'· X')· v. Da'· X' means the derivative of a' in the direction X'; the resulting element of E* is then applied to v. Thus we can write Lx,a' = Da'· X' + a'· DX'.
(3)
In finite dimensions. the corresponding coordinate expression is
. aai
..
axj
·
(Lxa)· VI = _ . XJVI + a· _ . VI; I axJ J axl
i.e.•
(3')
Now let t be of type (r. s). so t': V ~ L (E ........ E .... E •...• E; JR). Let a l •...• a r be (constant) elements of E* and v I' ...• V5 (constant) elements of E. Then again by the derivation
Chapter 5 Tensors
364
property. Lx' [t'(a' •...• a'. vI' ...• v.)] = (Lx,t')(a' •...• a'. vI' ...• v.)
.
,
+
L t'(a' •...• Lx,a
i •...•
a'. v, •...• v.) +
i='
L t'(a' •...• a'. v, •...• Lx,vj' ...• v.). j='
Now using the local fonnula (1)-(3) for the Lie derivatives of functions. vector fields. and one-fonns. we get (:I)t'· X')· (a' •...• a'. vI' ...• v.) = (Lx,t')(a' •...• a'. vI' ...• v.)
.
, + Lt'(a' •...• a i i =,
DX' •...• cl.v, . .... v,)+ Lt'(a' •...• a', v" ... ,-DX'·vj' ... , vs ). j='
.
Therefore, (Lx,t')(a', ... , a'. vI' ...• v s)
=(:I)t'· X')(a', ... , a', vI' ... , vs)
,
-L
s
t'(a', ... , a i
.
DX' , ... , a" v" ...• v.) +
i='
L
t'(a', ... , a', v, •... , DX' . Vj' ... , v s)·
j='
In components. this reads
i .. .i (Lxt) I '. . h···J,
=
k a i .. .i axil li .. .i . . X - p '. . - - - t 2 '. . - (all upper mdlces) axk J•...J. axl h···J,
aXm
..
+ -ax. - t'I···" mh. ...J,. + (all lower indices) JI
(4)
We deduced the component fonnulas for Lxt in the case of a finite-dimensional manifold as corollaries of the general Banach manifold fonnulas. Because of their importance, we shall deduce them again in a different manner, without appealing to 5.3.5. Let _ il .. .i,.J
t - t
.....£...... ®
J
I···. ax'i
... ®
....£...... ® ax',
_jl ... ® dxj, E L rrf dxs(U),
where U is a chart domain on M. If X = xka/axk, the tensor derivation property can be used to compute Lxt. For this we recall that
Section 5.3 The Lie Derivative: Algebraic Approach
365
and that
by the general formula for the bracket components. The formula for Lx(dxk) is found in the following way. The relation Oki = dxk(d/dx i ) implies by 001 that
Thus
k (dXk) Lx(dx)= --. dx'..
so
dX'
Now one simply applies 001 and collects terms to get the same local formula for Lxt found in (4). Note especially that L
~ (~).=o ax'
dx)
and
L~ ax'
(dxi) = 0, for all i,j.
5.3.6 Examples A Compute Lxt, where
Solution. Metfwd 1 Note that (i) Lxt = La/ilx + xa/ay t = La/ax t + La/ilyt (ii) La/ax t = La/ax { x
= La/ax (x Now note
-#y ® dx ® dy + y ~ ~ ® dx ® dy) +
® dy ® dY}
La/ax( Y ~ ® dy ® dy) = ;y ® dx ® dy + 0 .
Chapter 5 Tensors
366
a = 0, LxiJ/iJy ax a = - LiJ/iJx(xa a LxiJ/iJy ay dy) = - { l'a ay + x . 0} = - ay LxiJ/iJy dx
= 0,
and LxiJliJy dy
,
= dx.
thus (iii) LxiJliJy t
= LxiJ/iJy { x
~ ® dx ® dy + y ~ ® dy ® dY}
= (O+O+O+X~®dX®dX) + (x
~
® dy ® dy + 0 + y
~
® dx ® dy + Y ;y ® dy ® dx ).
Thus, substituting (ii) and (iii) into (i), we find a
a
a
Lxt = dy ® dx ® dy + x ay ® dx ® dx + x dy ® dy ® dy + y = (y + 1)
a
a
dy ® dx ® dy + x dY ® dx ® dx + x
a a ay ® dx ® dy + Y dy ® dy ® (
a a ay ® dy ® dy + Y ® dy ® dx.
dY
Method 2 Using component notation, t is a tensor of type (1,2) whose nonzero components are t2l2 = x and t 222 = y. The components of X are Xl = 1 and X2 = x. Thus, by the component formula (4),
The nonzero components are (Lx t)2 12 = 1 - 0 + y + 0 = 1 + y;
(Lxt)\2 = X - 0 + 0 + 0 = x;
(LXt)211 =O-O+O+x=x;
(Lx t)221 = 0 - 0 + 0 + y = y,
and hence a a a a Lxt = (y + 1) ay ® dx ® dy + x ® dx ® dx + x ay ® dy ® dy + y dy ® dy ® dx.
ay
The two methods thus give the same answer. It is useful to understand both methods since they both occur in the literature, and depending on the circumstances, one may be easier to apply than the other. B In Riemannin geometry, vector fields X satisfying Lxg = 0 are called Killing vector fields; their geometric significance will become clear in the next section. For now, let us compute the system of equations that the components of a Killing vector field must satisfy. If X = Xii}Ii}x i , and g = &jdxi ® dxj, then
Section 5.3 The Lie Derivative: Algebraic Approach
367
Lxg = (LXgi} dx i ® dxi + gij(LXdx i) ® dxj + gijdx i ® (Lxdx j) k ag··. . axi k . . axj k = X -') dx' ® dx-l + g .. dx ® dx-l + g··dx' ® dx aXk ') aXk ') aXk axk aXk} i j k agij = { X - + gk· --. + gik--. dx ® dx. ai ) ax' ax) Note that Lxg is still a symmetric (0. 2)-tensor. as it must be. Hence X is a Killing vector field iff its components satisfy the following system of n partial differential equations. called Killing's equotions
C In the theory of elasticity. if u represents the displacement vector field. the expression Lug is called the strain tensor. As we shall see in the next section. this is related to the Cauchy-Green tensor C = 2(M) is the associated tensor with components t ij = gi/j' We have from (4) :> i
.
k
i k ot j k ax' i ax (Lxt)· = X - t . - - + t k --. ) akk ) axk ax)
•
and so
kJ
k at t j k ax I I ax (LXt)ij = git ( X - k - t j -k- + t k --. . ax ax ax) But also from (4)
k a I k ax' I axk =X -(g·,t·)+g'kt·--. +g·,tk--. axk ' ) ) ax' ' ax)
a :> I t k k gil I k ot j k ax I ax = X - - t· + X g., + g'k t . - - . + g·,t k - .. ax k ) 'axk ) ax' ' ax) Thus. to have equality it is necessary and sufficient that
368
Chapter 5 Tensors
for all pairs of indices (i. j). which is a nontrivial system of n2 linear partial differential equations for gij' If X is a Killing vector field. then
which substituted in the preceding equation. gives zero. The converse statement is proved along the same lines. In other words. a necessary and sufficient condition that Lx commute with the formation of associated tensors is that X be a Killing vector field for the pseudo-Riemannian metric g.• As usual. the development of Lx extends from tensor fields to F-valued tensor fields.
Exercises 5.3A Let t = xya/ax ® dx + ya(dy ® dx + a(dx ® dy E TIl (R2). Define the map q> as follows: q>: {<x. y) I y > o} ~ {(x. y) I x> O. x2 < y} and q>(x. y) = (yex, y2e2x + y). Show that q> is a diffeomorphism and compute trace(t). q>*t. q>*t. Lxt. Lxq>*t. and Lq>.xt. for X = ya(dx + x2a/ay. 5.38 Verify explicitly that Lx(t&) 1= (Lxtf where & denotes the associated tensor with both indices lowered. for X and t in Exercise 5.3A. 5.3C Compute the coordinate expressions for the Killing equations in R3 in rectangular. cylindrical. and spherical coordinates. What are the Killing vector fields in ]Rn? 5.3D Let (M. g) be a finite dimensional pseudo-Riemannian manifold. and g# the tensor g with both indices raised. Let X E X(M). Calculate (Lxg#f - Lxg in coordinates. 5.3E If (M. g) is a finite-dimensional pseudo-Riemannian manifold and f E calculate Lx(Vf) - V(Lxf).
:r (M).
X E X(M).
5.3F Nijenhuis tensor (i) Let t E 'I\(M). Show that there is a unique tensor field Nt 'I\(M). skew-symmetric in its covariant indices. such that
E
for all X E X(M). where the dots mean contractions. Le .• (t· X)i = t ijxj. (t· S)ij = t ikSkj. where t. s E 'I\(M). and Nt' X = NijkXk. where Nt = Nijk a/ax i ® dxj ® dxk. Nt is called the Nijenhuis tensor. Generalize to the infinite dimensional case. (Hint: Show that
Section 5.3 The Lie Derivative: Algebraic Approach N i jk -- tt kti j,t - tt jti k,t + ti t tt kj
-
369
ti t tt j,k' )
(ii) Show that Nt = 0 iff
[t ' X, t ' Yj - t . [t . X. Y] = t . [X. t . Yj - t2 . [X. Yj for all X. Y E X(M). where t2 E
'TIL (M)
is the tensor field obtained by the composition
tot. when t is thought of as a map t : X(M) ~ X(M).
~5.4
The Lie Derivative: Dynamic Approach
We now tum to the dynamic interpretation of the Lie derivative. In §4.2 it was shown that Lx acting on an element of .'F(M) or X(M), respectively, is the time derivative at zero of that
element of .'F(M) or X(M) Lie dragged along by the flow of X. The same situation holds for general tensor fields. Given t E (P.(M) and X
E
X(M), we get a curve through t(m) in the
fiber over m by using the flow of X. The derivative of this curve is the Lie derivative.
5.4.1 Lie Derivative Theorem Let X
E
Xk(M), t
E
'T'.(M) be of class Ck, and FA be the
flow of X. Then on the domain of the flow (see Fig. 5.4.1) we have
Fig 5.4.1
Proof It suffices to show that
Indeed, if this is proved then
Define 9 x : 'T(M)
~
'T(M) by
I
d A= 0 (FAt)(m) * 9x(t)(m) = df.. .
Note that 9 x (t) is a smooth tensor field of the same type as t, by smoothness of t and FA. (We
Section 5.4 The Lie Derivative: Dynamic Approach
371
suppress the notational clutter of restricting to the domain of the flow.) Let us apply Theorem 5.3.2. Clearly ex is lR-linear and is natural with respect to restrictions. It is a tensor derivation from the product rule for derivatives and the relation (cp*t)(cp*a1••••• cp*ar • cp*X 1'
••.•
CP*Xs) = cp*(t(a1••••• a\ Xi' ...• Xs»
for cp a diffeomorphism. Hence ex is a differential operator. It remains to show that ex coincides with Lx on !T(M) and *(M). For f E !T(M). and X E *(M). we have
by 4.2.10 and 4.2.19. respectively. Thus by 5.3.2 and 5.3.4. ext =Lxt for all t E 'T(M) .• This theorem can also be verified in finite dimensions by a straightforward coordinate computation. See Exercise 5.4A. The identity in this theorem relating flows and Lie derivatives is so basic. some authors like to take it as the definition of the Lie derivative (see Exercise 5.4C). 5.4.2 Corollary If t E 'T(M). Lxt = 0 iff t is constant along the flow of X. That is. t = F"*1. As an application of 5.4.1. let us generalize the naturality of Lx with respect to diffeomorphisms. As remarked in §5.2. the pull-back of covariant tensor fields makes sense even when the mapping is not a diffeomorphism. It is thus natural to ask whether there is some analogue of 5.3.5 for pull-backs with no invertibility assumption on the mapping cpo Of course. the best one can hope for. since vector fields can be operated upon only by diffeomorphisms. is to replace the pair X. cp*X be a pair of cp-related vector fields. 5.4.3 Proposition Let cp: M ~ N be C"'. X E *(M). Y E *(N). X -",Y and t E 'T°s(N). Then Cp*(Lyt) = Lx'P*t.
Proof Recall from 4.2.4 that X -", Y iff G" 0 cp =cp 0 F". where F" and G" are the flows of X and Y. respectively. Thus by 5.4.1.
=
dl dA.
,,=0
cp *G"*t = cp*dl dA.
,,=0
G"* t
=
cp *(Lyt) .•
As with functions and vector fields. the Lie derivative can be generalized to include time-dependent vector fields.
372
Chapter 5 Tensors
5.4.4 First Time-dependent Lie Derivative Theorem Let X" E Xk(M), k;,. 1, for A E R and suppose that X(A, m) is continuous in (A, m). Then
if
F", ~ is the evolution operator for
X"' we have
where t
E
'T's(M) is of class C k •
Warning. It is not generally true for time-dependent vector fields that the right hand-side in 5.4.4 is
Proof As in 5.4.1, it is enough to prove the formula at A = 1.1 where F).,). = identity, for then
As in 5.4.1,
is a differential operator that coincides with
9x). = Lx).
Lx. on
1"(M) and on
X (M) by 4.2.31. Thus by 5.3.2,
on all tensors. _
Let us generalize the relationship between Lie derivatives and flows one more step. Call a smooth map t: R x M ~ 'T's(M) satisfying t).(m) =teA, m) field. Theorem 5.4.4 generalizes to this context as follows.
E
(TmMYs a time-dependent tensor
5.4.5 Second Time-dependent Lie Derivative Theorem Let
t).
be a Ck time-dependent
tensor, and X). be as in 5.4.4, k;,. I, and denote by F", ~ the evolution operator of X". Then
Proof By the product rule for derivatives and 5.4.4 we get
5.4.6 Examples A If g is a pseudo-Riemannian metric on M, the Killing equations are Lxg = 0 (sec Example 5.3.68). By 5.4.2 this says that F" *g = g, where F). is the flow of X, i.e., that the
Section 5.4 The Lie Derivative: Dynamic Approach
373
flow of X consists of isometries. B In elasticity the vanishing of the strain tensor means. by Example A. that the body moves as a rigid body. • We close this section with an important technique based on the dynamic approach to the Lie derivative. called the Lie transform method. It has been used already in the proof of the Frobenius theorem (§4.4) and we shall see it again in 6.4.14 and 9.1.2. The method is also used in the theory of normal forms (cf. Takens [1974]. Guckenheimer and Holmes [1983], and Golubitsky and Schaeffer [1985]). 5.4.7 The Lie Transform Method Let two tensor fields to and t\ be given on a smooth manifold M. We say they are locally equivalent at mo E M if there is a diffeomorphism
equivalent is to join them with a curve t(A) satisfying t(O) local diffeomorphisms <1>1. such that
= to • t(l) = t\
and to seek a curve of
If this is done.
for X~. If this is possible. let
so that
= to. If we choose X~ so X~(mo) = O. then
4.1.25 and
374
Chapter 5 Tensors
5.4.8 The Morse-Palais-Tromba Lemma Let E be a Banach space arui (,) a weakly noruiegenerate, continuous, symmetric bilinear form on E. Let h: U -+ JR be Ck , k ~ 3, where U is open in E, arui let Uo E U satisfy h(u o) =0, Dh(uo) =O. Let B = D2h(uo) : Ex E -+ JR. Assume that there is a linear isomorphism T: E -+ E such that B(u, v) = (Tu, v) for all u, vEE and that h has a Ck- I gradient (Vh(y),u) = Dh(y) . u. Then there is a local Ck- 2 diffeomorphism
Proof Symmetry of B implies that T is self-adjoint relative to (,). Let f(y) = (l!l)B(y - uO' y - uo), hi =h, and h.. =f + A.p, where p(y) =h(y) - (l!2)B(y - U o ' y - u o) is Ck and satisfies p(u o) =0, Dp(u o) =0, and D2p(u O) = O. We apply the Lie transform method to h... Thus we have to solve the following equation for a Ck-2 vector field X..
(1)
Then
Dp(x) . e = (Vp(x),e ) = =
<10
DVp(uo + t(x - Uo» . (x - uo)dt, e)
(T(x - uo), R(x) . e)
where R : U -+ L(E, E) is the Ck-2 map given by
R(x) = T- I
f
Dp(uo + t(x - uo» . (x - uo)dt
which satisfies R(uo) =O. Thus p(y) has the expression
p(y) =
f
Dp(uo + t(y - Uo»· (y - Uo)dt = - (T(y - uo), X(y»
where X: U -+ E is the Ck -2 vector field given by
(2)
Section 5.4 The Lk Derivative: Dynamic Approach
375
which satisfies X(uo) = 0 and DX(uo) = I. Therefore (Lx,h,)(y) = Dh,,(y) . X,,(y) = B(y - uo. X,,(y» + ADp(y) . X,,(y) = (T(y - uo). (I
+ AR(y» . X,,(y»
(by (2))
so that the equation (1) becomes (T(y - uo). (I + AR(y» . X,,(y» = (T(y - uo). X(y» .
(by (3»
Since R(uo) = O. there exists a neighborhood of uo. such that the nonn of AR(y) is < 1 for all A E [0.1). Thus for y in this neighborhood. I + AR(y) can be inverted and we can take X,,(y)
=
(I + AR(y»-1 X(y) which is a Ck-2 vector field defined for all A E [0.1) and which satisfies X,,(uo)
= o.
DX,,(uo) = O. Differentiating the relation (dldA)q>,,(U)
using q>,,(uo)
= Uo
yields (d/dA)Dq>,,(uo)
= DX,,(q>,,(uo»
0
= X,,(q>,,(u)) in = DX,,(uo)
Dq>,,(uo)
0
u at Uo and Dq>,,(uo)
= O.
i.e .• Dq>,,(uo) is constant in A E [0.1). Since it equals I at A = O. it follows that Dq>,,(uo) = I. _
5.4.9 The Classical Morse Lemma Let h: U ~]R be Ck. k;?: 3. U open in ]Rn. and let u E U be a nondegenerate critical point of h. i.e., h(u) = O. Dh(u) = 0 and the symmetric bilinear form D 2h(u) on ]Rn is nondegenerate. Then there is a local Ck-2 diffeomorphism 'I' of]Rn fixing u such that h('I'(x»
={l/2)[(x 1 _ u 1)2 + ... + (x _ un- i)2 _ (x n- i+1 _ un- i+1)2 _ ... _ (xn _ un)2).
Proof. In 4.5.8 take ( .) to be the dot-product in ]Rn to find a local Ck-2 diffeomorphism on ]Rn fixing Uo such that h(q>(x))
= (1I2)D 2h(u)(x -
u. x - u). Next. apply the Gram-Schmidt
procedure to find a basis of ]Rn in which the matrix of D2h(u) is diagonal with entries ±I (see 6.2.9 for a review of the proof of the existence of such a basis). If i is the number of -I's (the
index). let q> be the composition of 'I' with the linear isomorphism detennined by the change of an arbitrary basis of ]Rn to the one above. _
Exercises 5.4A Verify Theorem 5.4.1 by a coordinate computation as follows. Let F,,(x) = (yl(A. x) •...• yn(A. x» so that ifyi/aA = Xi(y) and ayi/axi satisfy the variational equation
Then write
Chapter 5 Tensors
376
Differentiate this in A. at A. =0 and obtain the coordinate expression (4) of Seclion 5.3 for Lxt. 5.4B Carry oUl the proof outlined in Exercise 5.4A for lime-dependenl veclor fields. 5.4C Slarling with Theorem 5.4.1 as the definition of Lxt, check that Lx satisfies 001, 002 and the properties (i)-(iv) of 5.3.2. 5.4D Let C be a contraction operator mapping 'T's(M) to 'Tr-1s_1(M). Use both 5.4.1 and 001 to show that Lx(Ct) =C(Lxt). 5.4E Extend Theorem 5.4.1 to F-valued tensors. 5.4F Let f(y) = (1/2)y2 - y3 + y5. Use the Lie transform method to show that there is a local diffeomorphism
(x, y) =
~1
~
~
1~12
1~3
£.. Ii xn Yn and h(x) = "2 £.. Ii x" -"3 £.. xn . n=l
n=l
Show that h vanishes on (0, 0, ... , 3/2n, 0, ... ) which
n=l ~
0 as n ~
00,
so the conclusion
of the Morse lemma fails. What hypothesis in 4.5.8 fails? 5.4H (Buchner, Marsden and Schecter [1983]). In the notation of Exercise 2.40, show that f has a sequence of critical points approaching 0, so the Morse lemma fails. (The only missing hypothesis is that Vh is Cl.)
§s.s
Partitions of Unity
A partition of unity is a technical device that is often used to piece smooth local tensor fields together to form a smooth global tensor field. Partitions of unity will be useful for studying integration; in this section they are used to study when a manifold admits a Riemannian metric.
5.5.1 Definition If t is a tensor field on a manifold M, the carrier of t is the set of m E M
for which t(m)"# 0, and is denoted carr t. The support of t, denoted supp t, is the closure of carr t. We say t has compact support if supp t is compact in M. An open set U
called a 0 carrier if there exists an f E 1'r(M), such that f ~ 0 and U
=carr f.
C
M is
A collection of
subsets (Ca ) of a manifold M (or, more generally, a topological space) is called locally finite if for each mE M, there is a neighborhood U of such that
un Ca =0
except for finitely many
indices a. 5.5.2 Definition A partition of unity on a manifold M is a collection (U i' gi)}' where (i) (U;) is a locally finite open covering of M;
(ii) gi E
l' (M), gi(m) ~ 0 for all m E M, and supp gi
(iii) for each mE M, L;gi(m)
If A
= {(Va' <Pa)}
= 1.
C U i for all i; (By (i), this is afinite sum.)
is an atlas on M, a partition of unity subordinate to A
is a
partition a/unity (Ui' g;)} such that each open set U i is a subset of a chart domain Va(i)' If any atlas A has a subordinate partition of unity, we say M admits partitions of unity. Occasionally one works with Ck partitions of unity. They are defined in the same way except gi are only required to be Ck rather than 0".
5.5.3 Patching Construction Let M be a manifold with an atlas A = (Va' <Pa)} where <Pa:Va~V'acE is a chart. Let ta bea Ck tensor field, k~l, affixed type (r,s) defined on V'a for each a, and assume that there exists a partition of unity (U i' gi») subordinate to A. Let t be defined by
afinite sum at each mE M. Then
t
is a Ck tensor field of type (r,s) on M.
Proof Since {U;} is locally finite, the sum at every point is a finite sum, and thus t(m) is a type (r, s) tensor for every m E M. Also, t is Ck since the local representative of t in the chart (Va(i)' <Pa(i» is :E/gi 0 <Pa(j)-l )ta(j) , the summation taken over all indices j such that Va(i) n Va(j)"# 0 ; by local finiteness the number of these j is finite. _
378
Chapter 5 Tensors
Clearly this construction is not unique; it depends on the choices of the indices a.(i) such that Vi C Va(i) and on the functions gi' As we shall see later. under suitable hypotheses. one can always construct partitions of unity; again the construction is not unique. The same construction (and proof) can be used to patch together local sections of a vector bundle into a global section when the base is a manifold admi tting partitions of uni ty subordinate to any open covering. To discuss the existence of partitions of unity and consequences thereof. we need some topological preliminaries. 5.5.4 Definition Let S be a topological space. A covering (Va} of S is called a refinement of
a covering (Vi} iffor every Va there is a Vi such that Va C Vi' A topological space is called paracompact if every open covering of S has a locally finite refinement of open sets, and S is Hausdorff· 5.5.5 Proposition Second-countable, locally compact Hausdorff spaces are paracompact. Proof By second countability and local compactness of S. there exists a sequence 0" ...• On' ... of open sets with c1(On) compact and UneNO n = S. Let Vn = 0, U ... U On' n = 1.2, ... and put V, =V,. Since (Vn} is an open covering of S and c1(V,) iscompact.
Put then c1(V2 ) is compact. Proceed inductively to show that S is the countable union of open sets Vn such that cl(Vn) is compact and cl(Vn) C Vn+,. If Wa is a covering of S by open sets. and K" = c1(Vn) \ Vn_" then we can cover K" by a finite number of open sets. each of which is
+,.
and is disjoint from c1(Vn_2). The union of such collections contained in some Wan V n yields the desired refinement of (Wa }. _ Another class of paracompact spaces are the metrizab1c spaces (see 5.5.15 in Supp1cmen' 5.5A). In particular. Banach spaces arc paracompact. 5.5.6 Proposition Every paracompact space is normal. Proof We first show that if A is closed and u E S \ A, there are disjoint neighborhoods of u and A (regularity). For each v E A. let V u ' Vv be disjoint neighborhoods of u and v. Let Wa be a locally finite refinement of the covering (Vv' S \ A I v E A}, and V = U Wa' the union over those a. with WunA", 0. A neighborhood Vo of u meets a finite number of WU" Let V denote the intersection of Vo and the corresponding Vu' Then V and V are the required neighborhoods. The case for two closed sets proceeds somewhat similarly, so we leave the details for the reader. _
Section 5.5 Partitions o/Unity
379
Later we shall give general theorems on the existence of partitions of unity. However, there is a simple case that is commonly used, so we present it first. 5.5.7 Theorem Let M be a second-countable (Hausdorff) n-manifold. Then M admits partitions of unity.
Proof The proof of 5.5.5 shows the following. Let M be an n-manifold and {Wa} be an open covering. Then there is a locally fmite refinement consisting of charts (Vi' cp) such that CPi(V) is the disk of radius 3, and such that CPi-I(DI(O» cover M, where DI(O) is the unit disk, centered at the origin in the model space. Now let A be an atlas on M and let {(Vi' CPi)} be a locally finite refinement with these properties. From 4.2.13, there is a nonzero function 11; E l' (M) whose support lies in Vi and hj
~
O. Let
(the sum is finite). These are the required functions. _ If { Va} is an open covering of M, we can always find an atlas A = {(Ui, cp)} such that {U;} is a refinement of {Va} since the atlases generate the topology. Thus, if M admits partitions of unity, we can find partitions of unity subordinate to any open covering. The case of CD-partitions of unity differs drastically from the smooth case. Since we are
primarily interested in this latter case, we summarize the topological situation, without giving the proofs. 1 If S is a Hausdorff space, the following are equivalent: (i) S is normal; (ii) Urysohn's lemma For any two closed nonempty disjoint sets A, B there is a continuous function f: S -+ [0, 1] such that f(A) = 0 and feB) = 1 (iii) Tietze extension theorem For any closed set A c: S and continuous function g: A -+ [a, b], there is a continuous extension G : S -+ [a, b] of g 2 A Hausdorff space is paracompact iff it admits a CO partition of unity subordinate to any open covering. It is clear that if {(Ui, g)} is a continuous partition of unity subordinate to the given open covering {Va}' then by definition lUi} is an open locally finite refinement. The converse--the existence of partitions of unity--is the hard part; the proof of this and of the equivalences of (i), (ii)
and (iii) can be found for instance in Kelley [1975] and Choquet [1969; sec. 6]. These results are important for the rich supply of continuous functions they provide. We shall not use these topological theorems in the rest of the book, but we do want their smooth versions on manifolds. Note that if M is a manifold admitting partitions of unity subordinate to any open covering, then M is paracompact, and thus normal by 5.5.6. This already enables us to generalize (ii) and
380
Chapter 5 Tensors
(iii) to the smooth (or Ck ) situation. 5.5.8 Proposition Let M be a milnifold admitting smooth (or Ck ) partitions of unity. If A and
B are closed disjoint sets then, there exists a smooth (or Ck)function f: M f(A)
=0
and f(B)
-7
[0, 1] such that
= 1.
Proof As we saw, the condition on M implies that M is normal and thus there is an atlas {(Va'
n A", 0 implies Va n B = 0. Let {(Vj, gj») be a subordinate C< partition on unity and f = L~, where the sum is over those i for which Vj n B '" 0. Then f is Ck , is
(jla») such that Va
one on B, and zero on A. • 5.5.9 Smooth Tietze Extension Theorem Let M be a milnifold admitting partitions of unity,
and let
1t:
E
-7
M be a vector bundle with base M. Suppose (J: A
-7
E is a C k section
defined on the closed set A (i.e., every point a E A has a neighborhood V. and a Ck section (Ja: Va -7 E extending (J). Then (J can be extended to a Ck global section l:: M -7 E. In particular, if g : A
-7
F is a Ck function defined on the closed set A, where F is a Banach
space, then there is a Ck extension G : M -7 F; if g is bounded by a constant R, i.e.,
II g(a) II :0;
R for all a E A. then so is G.
Proof Consider the open covering {Va' M\A I a E A) of M, with Va given by the definition of smoothness on the closed set A. Let {(V j , gj») be a Ck partition of unity subordinate to this open covering and define (Jj: Vj
-7
E, by (Jj = (J. I Vj for all Vj and (Jj ;;; 0 on all Vj disjoint
from V.' a E A. Then gpj: Vj
-7
E is a Ck section on Vj and since supp(~(Jj)
C
supp(gj)
C
V j , it can be extended in a manner to M by putting it equal to zero on M \ V j • Thus gpj: M -7 E is a Ck-section of 1t : E -7 M and hence l: =Lj~(Ji is a Ck section; note that the sum is Ck
finite in a neighborhood of every point m
E
M. Finally, if a E A
i.e., l: I A = (J. The second part of the theorem is a particular case of the one just proved by considering the trivial bundle M x F
-7
M and the section (J defined by (J(m) = (m, g(m». The boundedness
statement follows from the given construction, since all the
~
have values in [0, 1] .•
Before discussing general questions on the existence of partitions of unity on Banach manifolds, we discuss the existence of Riemannian metrics. Recall that a Riemannian metric on a Hausdorff manifold M is a tensor field g E 'T° 2(M) such that for all m E M, g(m) is symmetric and positive definite. Our goal is to find topological conditions on an n-manifold that are necessary and sufficient to ensure the existence of Riemannian metrics. The proof of the necessary conditions will be simplified by first showing that any Riemannian manifold is a metric space. For this, define for m, n E M,
381
Section 5.5 Partitions o/Unity
d(m, n) = inf { ~(y)
I y: [0, I) ~ M is a continuous piecewise C l
curve with 1'(0) = m, y (I)
=n } .
Here ~(y) is the length of the curve y, defined by
~(y) = where )'(t)
=dy /dt
f
II Y(t) IIdt,
is the tangent vector at y(t) to the curve y and
11111 = [~t) (){t),
){t»)I12
is its length. 5.5.10 Proposition d is a metric on each connected component 0/ M whose metric topology is
the original topology of M. If d is a complete metric, M is called a complete Riemannian manifold. Proof Clearly d(m, m)
= 0,
d(m, n)
= d(n, m),
and d(m, p):5 d(m, n) + d(n, p), by using the
definition. Next we will verify that d(m, n) > 0 whenever m *- n. Let mE U
C
M where (U,
=U' C
E. Then for any u E
D, g(u)(v, v)I12, defined for v E TuM, is a norm on TuM. This is equivalent to the norm on E, under the linear isomorphism Tm
is the local expression for g, then g'(u')
defines an inner product on E, yielding equivalent norms for all u' E U'. Using continuity of g and choosing U' to be an open disk in E, we can conclude that the norms g'(u')I12 and g'(m)1/2, where m'
=
satisfy: a g'(m')I!2:5 g'(u')I!2:5 bg'(m')I12 for all u' E U', where
a and b are positive constants. Thus, if 1'\: [0, I) ~ U' is a continuous piecewise C l curve, then
1
~ ag'(m') (fa ,,(t)dt,
L1
,,(t)dt
)112
~ ag'(m')(1'\( I) -
1'\(0),1'\(1) - 1'\(0»112.
Here we have used the following property of the Bochner integral:
II
f
f(t)dt
I
:5
f
II f(t) II dt,
valid for any norm on E (see the remarks following 2.2.7). Now let y: [0, 1) ~ M be a continuous piecewise C l curve, )'(0)
= m,
)'(1) = n. m E U,
Where (D,
382 which "((c)
Chapter 5 Tensors
n
E III x II
=r}) '" 0.
Then ~(y) ~ ~(yIlO, cD ~ ag(m')«
0
y)(c), (
0
y)(C»1/2 ~ ar. Thus we conclude d(m, n) ~ ar > O.
The equivalence of the original topology of M and of the metric topology defined by d is clear if one notices that they are equivalent in every chart domain U, which in turn is implied by their equivalence in
it is second countable. Hence for Hausdorff n-manifolds (not necessarily connected) paracompactness and metrizability are equivalent. Proof If M is second countable, it admits partitions of unity by 5.5.7. Then the patching construction 5.5.3 gives a Riemannian metric on M by choosing in every chart the standard inner product in JRn . Conversely, assume M is Riemannian. By 5.5.10 it is a metric space, which is locally compact and first countable, being locally homeomorphic to JRn. By 1.6.14. it is second countable. _ The main theorem on the existence of partitions of unity in the general case is as follows. 5.5.12 Theorem Any second-countable or paracompact manifold modeled on a separable Banach
space with a C k norm away from the origin admits Ck partitions of unity. In particular paracompact (or second countable) manifolds modeled on separable Hilbert spaces admit partitions of unity.
C~
5.5.13 Corollary Paracompact (or second countable) Hausdorffmanifolds modeled on separable real Hilbert spaces admit Riemannian metrics. Theorem 5.5.12 will be proved in the following two supplements. There are Hausdorff nonparacompact n-manifolds. These manifolds are necessarily nonmetrizable and do not admit partitions of unity. The standard example of a one-dimensional nonparacompact Hausdorff manifold is the "long line." In dimensions 2 and 3 such manifolds arc constructed from the Priifer manifolds.
Since nonparacompact manifolds occur rarely in
applications. we refer the reader to Spivak [1979. vol. 1, Appendix AJ for the aforementioned examples. Partitions of unity are an important technical tool in many proofs. We illustrate this with the sample theorem below which combines differential topological ideas of §3.5. the local and global existence and uniqueness theorem for solutions of vector fields. and partitions of unity. appplications of this sort can be found in the exercises.
More
Section 5.5 Partitions of Unity
383
5.5.14 Ehresmann Fibration Theorem A proper submersion f: M ~ N of finite dimensional manifolds with M paracompact is a locally trivial fibration, i.e.Jor any pEN there exists an open neighborhood V of u in N and a diffeomorphism cp: V x f-I(p) ~ f-I(V) such that f(cp(x, u» = x for all x E V and all u E r-I(p). Proof Since the statement is local we can replace M, N by chart domains and, in particular, we can ssume that N =]Rn and p = 0 E ]Rn. We claim that there are smooth vector fields XI' ... , Xn on M such that Xi is f-related to a/ax i E ~(lRn). Indeed, around any point in M such vector fields are easy to obtain using the implicit function theorem (see 3.5.2). Cover M with such charts, choose a partition of unity subordinate to this covering, and patch these vector fields by means of this partition of unity to obtain Xl' ... , Xn , f-related to a/ax I , ... , i)/()x n , respectively. Let F\k) denote the flow of X k with time variable t(k), k = 1, ... , n and let t = (t(I), ... , t(n»
E
jRn. If
II t II < C, then then integral curves of each Xk starting in r-l( (u E jRn III u II $; E jRn III v II $; 2Cj, since by 4.2.4
CD stay in f-I(v
(f a Pt(k»(Y) = (fl(y), ... , fk(y) + t(k), ... , fY'(y».
(1)
Therefore, since f is proper, 4.1.19 implies that the vector fields XI' ... , Xn are complete. Finally, let cp: ]Rn x r-1(O) ~ M be given by cp(t(1), ... , t(n), u) = (F\(l)
a ••• a
Pt(n»(u)
and note that cp is smooth (see 4.1.17). The map cp-I: M ~]Rn x r-I(O) given by
0 ••• 0
FI_t(I»(m» is smooth and is easily checked to be the inverse of <po Finally,
(f 0 cp)(t, u) = t by (1) since f(u) = O. •
~
SUPPLEMENT 5.5A
Partitions of Unity: Reduction to the Local Case We begin with some topological preliminaries. Let S be a paracompact space. If (U 13 ) is
an open covering of S, it can be refined to a locally finite covering (W13). The first lemma below will show that we can shrink this covering further to get another one, (Va) such that cl(V a )
C
Wa with the same indexing set. A technical device used in the proof is the concept of a well-ordered set. An ordered set A in which any two elements can be compared is called well-ordered if every subset has a smallest element (see the introduction to Chapter 1).
5.5.15 Shrinking Lemma Let S be a normal space and (Wa}aeA a locally finite open covering of S. Then there is a locally finite open refinement (Va)aeA (with the same indexing set) such that cl(Va ) C Wa. Proof Well-order the indexing set A and call its smallest element a(O). The set Co defined as
Co
384
Chapter 5 Tensors
= S , U a>a(O) Wa is closed.
so by normality
there exists an open set
Va(O) such that Co C
cl(Va(O» C Wa(O) . If Vy is defined for all y < a. put Ca = S, {(UyaW y)} and by normality find Va such that Ca C cl(Va ) C Wa' The collection {Va} ae A is the desired locally finite refinement of {Wa}aeA' provided we can show that it covers S. Given s E S. by local finiteness of the covering {Wa} ae A'
S
belongs to only a finite collection of them. say WI'
W 2' ... • Wn • corresponding to the elements a I' '" • an of the index set. If ~ denotes the maximum of the elements a!' ...• an' then s
eo
Wy for all y >~.
so that if in addition
s
eo
Vy
for all y<~. then s E CfJ C VfJ' i.e.• s E VfJ'.
5.5.16 Lemma (A. H. Stone) Every pseudometric space is paracompact. Proof Let {Va} aeA be an open covering of the pseudometric space S with distance function d. E Va I d(x. S, Va) ~ l/2n }. By the triangle inequality we have the inequality
Put V n.a = (x
. d(Vn,a' S, Vn+I,a) ~ l/2n - l/2n+1 = l/2n+l
•
Well-order the indexing set A and let Vn,a =
. Vn,a' Ujka Vn+I,fJ' If y. Ii
E A. we have Vn,y C S, Vn+I,S' if y< Ii. or Vn,S C S, Vn+I,y if Ii < y. But in both cases we have d(Vn,1' Vn,S) ~ l/2n+l . Define W n,a = (s E S I d(s. Vn,a) < 1/2n+3 }. and observe that d(Wn,a' Wn,W ~ l/2n+2 . Thus for a fixed n. every point s E S has a
neighborhood intersecting at most one member of the family {Wn,a I a E
N. a
E
E
A}. Hence {Wn,a I n
A} is a locally finite open refinement of {Va}' •
Let us now turn to the question of the existence of partitions of unity subordinate to any open covering.
5.5.17 Proposition (R. Palais) Let M be a paracompact manifold modeled on the Banach space E. The/ollowing are equivalent: (i) M admits Ck partitions o/unity;
(ii) any open covering 0/ M admits a locally finite refinement by Ck carriers.
° .°
1 2 such that cl(OI) CO2 , there exists a ck carrier V such that 01 eve 02' (iv) every chart domain 0/ M admits Ck partitions 0/ unity subordinate to any open covering; (v) E admits Ck partitions o/unity subordinate to any open covering 0/ E.
(iii) for any open sets
Proof ((i) ~ (ii». If {(Vi' gi)) is a Ck-partition of unity subordinate to an open covering. then clearly carr ~ forms a locally finite refinement of the covering by
ck carriers.
((ii) ~ (iii». Let {Va}aeA be a locally finite refinement of the open covering (02' S\ cl(OI)} by Ck carriers and denote by fa E ,'Fk(M). the function for which carr fa = Va' Let B = {a E A I Va CO 2}, Put V = nfJeB Vp. f= ~eBfp and remark that 01 eve 02' carr f= V. ((iii) ~ (iv». Let V be any chart domain of M. Then V is diffeomorphic to an open set in E which is a metric space. so is paracompact by Stone's theorem 5.5.16. Let {Va} be an arbitrary open covering of V and {VfJ} be a locally finite refmement. By the shrinking lemma we
Section 5.5 Partitions ojUnity
may assume that cl(V13)
C
V. Again by the shrinking lemma. refine further to a locally finite
covering {W p} such that cl(W13) 013
C
V 13' and so {Op}
385
C
V13' But by (iii) there exists a C<-carrier 013 such that W13
C
is a locally finite refinement of {Va} by Ck-carriers. whosc
corresponding functions we denote by fp. Thus f =Lpfp is a Ck map and {(V13' fp /f)} is a Ck partition of unity subordinate to {Va}' ((iv) =) (v». Consider now any open covering {Va} aEA of E and let (V. cp) be an arbitrary chart of M. Refine first the covering of E by taking the intersections of all its elements with all translates of cp(V). Since E is paracompact. refine again to a locally finite open covering {V p}. The inverse images by translations and cp of these open sets are subsets of V. hence chart domains. and thus by (iv) they admit partitions of unity subordinate to any covering. Thus every V13 admits a
C< partition of unity subordinate to any open covering. for example to {V13 n
Va I a
A}; call it {gp}. Then g = Lj.~P is a Ck map and the double-indexed set of functions gjp/g forms a Ck partition of unity of E.
E
«v)
=)
(iv». If E admits Ck partitions of unity subordinate to any open covering. then so
does every open subset by the (already proved) implication (i)
=)
(ii) applied to M =E. which is
paracompact by 5.5.16. Thus if (V. cp) is a chart on M. V admits partitions of unity. since cp(U) does. Finally. we show (iv) implies (i). Choosing a locally finite atlas. this proof repeats the one given in the last part of (iv) =) (v) . • As an application of this proposition we get the following. 5.5.18 Proposition Every paracompactn-manijold admits C"'-partitions ojunity. Proof By 5.5.17(ii) and (v) it suffices to show that every open set in ]RR is a C'" carrier. Any open set V is a countable union of open disks OJ. By 4.2.13. OJ = carr fj• for some Cfunction fi:]RR ~ R Put M j =sup{lI okfj(x) III x E ]RR. k ~ i} and let
f=
L fj /2 jMj. j=l
By Exercise 2.4J. f is a C- function for which carr f = V clearly holds. • In particular. second-countable n-manifolds admit partitions of unity. recovering 5.5.7.$:JJ
~ SUPPLEMENT 5.5B Partitions of Unity: The Local Case
Theorem 5.5.17 reduces the problem of the existence of partitions of unity to the local one.
386
Chapter 5 Tensors
namely finding partitions of unity in Banach spaces. This problem has becn studied by Bonic and Frampton [1966] for separable Banach spaces. 5.5.19 Proposition (R. Bonic and I. Frampton [1966]). Let E be a separable Banach space. The following are equivalent. (i) Any open set of E is a Ck carrier. (ii) E admits C< partitions of unity subordinate to any open covering of E. (iii) There exists a bounded nonempty Ck carrier in E.
Proof By 5.5.17, (i) and (ii) are equivalent since E is paracompact by 5.5.16. It remains to be shown that (iii) implies (i), since clearly (ii) implies (iii). This proceeds in several steps. First, we show that any neighborhood contains a Ck carrier. Let U be any open set and let carr f C D.(O) be the bounded carrier given by (iii), f E Ck(E), f?: O. Let e E U, fix eo E carr f, and choose £ > 0 such that Die) Ck(E), g?: 0 by
C
U. Define g E
g(v) = f(K(v - e) + eo)' K > 0, where K remains to be determined from the condition that carr g C DE(e). An easy computation shows that if K > (r + II eo ID/£, this inclusion is verified. Since e E carr g, carr g is an open neighborhood of e. Second, we show that any open set can be covered by a countable locally finite family of Ck carriers. By the first step, the open set U can be covered by a family of Ck carriers. By Lindelofs theorem, 1.1.6, U = Un Vn where Vn is a Ck carrier, the union being over the positive integers. We need to find a refinement of this covering by Ck carriers. Let fn E Ck(E) be such that carr fn = Vn. Define Un = (e EEl fn(e) > 0, fi(e) < lin for all i < n). Clearly U I
= VI and inductively Un = Vn n [ni 0, fi(e) < lIN for all i < N}. Thus, the sets Un cover U. This open covering is also locally finite for if e E Vn and N is such that f(e) > lIN, then the neighborhood (u E U I fn(e) > lIN) has empty intersections with all Urn for m > N. Third we show that the open set U is a Ck-carrier. By the second step, U = UnUn , with Un a locally finite open covering of U by Ck carriers. Then f = Lnfn is Ck, f(e)?: 0 for all eE E and carrf=U. _ The separability assumption was used only in showing that (iii) implies (i). There is no general theorem known to us for nonseparable Banach spaces. Also, it is not known in general
Section 5.5 Partitions o/Unity
387
whether Banach spaces admit bounded Ck carriers, for k ~ 1. However, we have the following. 5.5.20 Proposition If the Banach space E has a rumn Ck awayjrom its origin, k~I, then E
has bounded Ck-carriers. Proof By 4.2.13 there exists cp: lR -+ JR, C- with compact support and equal to one in a neighborhood of the origin. If lI'II:E\{O}-+lR is Ck,k~I, then cpolI'II:E\{O}-+lR is a nonzero map which is Ck , has bounded support II· 1I-1(supp cp), and can be extended in a Ck manner to E . • Theorem 5.5.12 now follows from 5.5.20, 5.5.19 and 5.5.17. The situation with regard to Banach subspaces and submanifolds is clarified in the following proposition, whose proof is an immediate consequence of 5.5.19 and 5.5.17. 5.5.21 Proposition (i) If E is a Banach space admitting Ck partitions of unity then so does any
closed subspace, (ii) If a manifold admits Ck partitions of unity subordinate to any open covering, then so does any submani/old. We shall not develop this discussion of partitions of unity on Banach manifolds any further, but we shall end by quoting a few theorems that show how intimately connected partitions of unity are with the topology of the model space. By 5.5.19 and 5.5.20, for separable Banach spaces one is interested whether the norm is Ck away from the origin. Restrepo [1964] has shown that a separable Banach space has a C 1 norm away from the origin if and only if its dual is separable. Bonic and Reis [1966] and Sundaresan [1967] have shown that if the norms on E and
E*
are
differentiable on E \ {O} and E* \ {O}, respectively, then E is reflexive, for E a real Banach space (not necessarily separable). Moreover, E is a Hilbert space if and only if the norms on E
and E* are twice differentiable away from the origin. This result has been strengthened by Leonard and Sundaresan [1973], who show that a real Banach space is isometric to a Hilbert space if and only if the norm is C 2 away from the origin and the second derivative of e ~ II e 112 f2 is bounded by 1 on the unit sphere; see Rao [1972] for a related result. Palais [1956b] has shown that any paracompact Banach manifold admits Lipschitz partitions of unity. Because of the importance of the differentiability class of the norm in Banach spaces there has been considerable work in the direction of determining the exact differentiability class of concrete function spaces. Thus Bonic and Frampton [1966] have shown that the canonical norms on the spaces LP(lR), ~P(lR), P ~ I, P < 00 are C- away from the origin if p is even, cP-\ with D(II . II p-\) Lipschitz, if p is odd, and C[p) with D[p)(II' II P) Holder continuous of order p[p1, if p is not an integer. The space Co of sequences of real numbers convergent to zero has an equivalent norm that is C- away from the origin, a result due to Kuiper. Using this result,
Frampton and Tromba [1972] show that the A-spaces (closures of C- in the H51der norm) admit a norm away from the origin. The standard norm on the Banach space of continuous real valued
c-
Chapter 5 Tensors
388
functions on [0, 1) is nowhere differentiable. Moreover, since CO([O, 1), JR) is separable with nonseparable dual, it is impossible to find an equivalent norm that is differentiable away from the origin. To our knowledge it is still an open problem whether CO([O, I), JR) admits C'" partitions of unity for k
~
1.
Finally, the only results known to us for nonseparable Hilbert spaces are those of Wells [1971), [1973), who has proved that nonseparable Hilbert space admits C2 partitions of unity.
The techniques used in the proof, however, do not seem to indicate a general way to approach this problem.
IbJ ~
SUPPLEMENT55C
Simple Connectivity of Fiber Bundles The goal of this supplement is to discuss the homotopy lifting property for locally trivial continuous fiber bundles over a paracompact base. This theorem is shown to imply an important criterion on the simple connectedness of the total space of fiber bundles wi th paracompact base.
5.5.21 Homotopy Lifting Theorem Let 7t: E -t B be a locally trivial CO fiber bundle and leI M be a paracompact topological space. If h: [0, 1) x M -t B is a continuous horrwtopy and f: M -t E is any continuous mLlP satisfying 7t 0 f =h(O, .), there exists a continuous homotopy H: [0, 1) x M -t E satisfying 7t 0 H = h and H(O,·) = f. If in addition. h fixes some point m E M, i.e .. h(t, m) is constant for t in a segment d of[O, 1). then H(t, m) is also constant for t Ed.
Remark The property in the statement of the Theorem is called the homotopy lifting property. A
Hurewiczjibration is a continuous surjective map 7t: E -t B satisfying the homotopy lifting property relative to any topological space M. Thus the theorem above says that a locally trivial
CO
fiber bundle is a Hurewicz fibration relative to paracompact spaces. See Steenrod [1951) and Huebsch [1955) for the proof.
5.5.22 Corollary Let 7t: E -t B be a CO locally trivial fiber bundle. If the base B and the fiber F are Simply connected. then E is Simply connected. Proof Let c: [0, 1) -t E be aloop, c(O) = c(l) = eo' Then d = 7t
0
c is a loop in B based at
7t(e o) = bo- Since B is simply connected there is a homotopy h: [0, 1) x [0, 1) -t B such that E [0, 1). and h(s. 0) = h(s, 1) = bo for all s E [0, 1). By the homotopy lifting theorem there is a homotopy H: [0. 1) x [0, 1)-t E such that 7t 0 H = h, H(O, . ) =c, and H(s. 0) = H(O, 0) = c(O) = eo' H(s, 1) = c(l) = eo. Since (7t 0 H)(l, t) = h(I. t) = b o ' it follows that t 1-+ H(1, t) is a path in ,..-1(b o) starting at H(l, 0) = eo and ending also at H(l, 1) = eo. Since 7t- 1(bo) is simply connected, there is a homotopy k: [1,2) x [0, 1) -t
h(O, t) = d(t), h(I, t) = bo for all t
Section 5.5 Partitions oJ Unity
,.-1(b o) such that k(1, t)
= H(I, t),
k(2, t)
=eo
389
for all t E [0,1) and k(s, 0)
all s E [1,2). Define the continuous homotopy K: [0, 2) x [0, 1)
1
~
=k(s, 1) =eo
for
E by
H(s, t), if s E [0,1]
K(s, t) =
k(s, t),
and note that K(O, t)
=H(O, t) =c(t),
K(2, t)
if s E [1,2]
= k(2, t) = eo
for any t E [0, 1), and K(s, 1) =eo
for any s E [0,2]. Thus c is contractible to eo and E is therefore simply connected. _
(/:;J
Exercises 5.5A (Whitney) Show that any closed set F in JR.n is the inverse image of
°
by a C"" real-
valued positive function on JR.n. Generalize this to any n-manifold. (Hint: Cover JR.n \ F with a sequence of open disks Dn and choose for each n a smooth function Xn
~
0, satis-
fying Xn I Dn> 0, with the absolute value of Xn and all its derivatives ::; 2n. Set X =
L n 2:0 Xn .) 5.5 B In a paracompact topological space, an open subset need not be paracompact. Prove the following. (i)
If every open subset of a paracompact space is paracompact, then any subspace is
paracompact. (ii)
Every open submanifold of a paracompact manifold is paracompact. (Hint: Vse chart domains to conclude metrizability.)
5.5C Let 1t: E
~
M be a vector bundle, E'
C
E a subbundle and assume M admits Ck parti-
tions of unity subordinate to any open covering. Show that E' splits in E, i.e., there exists a subbundle En such that E
=En ® En.
(Hint: The result is trivial for local bundles. Con-
struct for every element of a locally finite covering {VJ a vector bundle map fj whose kernel is the complement of E' I Vj. For {(V j, gj») a Ck partition of unity, put f = Ljgjfj and show that E = E' ® ker f.) 5.5D Let 1t: E ~ M be a vector bundle over the base M that admits Ck partitions of unity and with the fibers of E modeled on a Hilbert space. Show that E admits a Ck bundle metric, i.e., a Ck map g: M ~ T0 2(E) that is symmetric, strongly non degenerate, and positive definite at every point m E M. 5.5E Let E ~ M be a line bundle over the manifold M admitting Ck partitions of unity subordinate to any open covering. Show that Ex E is trivial (Hint: Ex E =L(E*, E) and construct a local base that can be extended.)
Chapter 5 Tensors
390
5.5F Assume M admits Ck partitions of unity. Show that any submanifold of M diffeomorphic to SI is the integral cuve of a Ck vector field on M. 5.5G Let M be a connected paracompact manifold. Show that there exists a COO proper mapping f: M ~ ]Rk. (Hint: M is second countable. being Riemmannian. Show the statement for k
= I. where f=1:i21 iCPi' and (cp;l is a countable partition of unity.)
5.5H Let M be a connected paracompact n-manifold and X E ~(M). Show that there exist h E 1{M). h > 0 such that Y = hX is complete. (Hint: With f as in Exercise 5.5G put h = exp {_(X[f))2) so that I Y[f) I :s 1. Hence (f 0 c)(]a. b[) is bounded for any integral curvc c of Y and ]a. b[ in the domain of c.) 5.51 Let M be a paracompact. non-compact manifold.
(i) Show that there exists a locally fmite sequence of open sets {Vi lie Z} such that Vi
n Vi+1 '" 0
unless j = i-I. i. i + I. and each Vi is a chart domain diffeomorphic (by the chart map CPi) with the open unit ball in the model space of M. See Fig. 5.5.1. (Hint: Let '11 be a locally finite open cover of M with chart domains diffeomorphic by their chart maps to the open unit ball such that no finite subcover of '11 covers M. and no two elements of '11 include each other. Let VO' VI be distinct elements of 0/, Vo
n VI '" 0. Let V_I E '11 be such that V_I n Vo '" 0; V_I exists by local finiteness of 0/. Now use induction.)
n VI = 0;
such a V_I
(ii) Vse (i) to show that there exists an embedding of ]R in M as a closed manifold. (Hint: Let Co be a smooth curve in Vo diffeomorphic by CPo to )-1. 1[ connecting a point in V_I n Vo to a point in Vo n VI. Next. extend Co n VI smoothly to a curve c!' diffeomorphic by CPI to ]0. 2[ ending inside VI n V 2 ; show that c I n U o extends the curve Co n VI inside Vo n VI. Now use induction.) (iii) Show that on each non-compact paracompact manifold admitting partitions of unity there exists a non-complete vector field. (Hint: Embed lR in M by (ii) and on lR consider the vector field ~ = x2 . Extend it to M via partitions of unity.)
Figure 5.5.1
Section 5.5 Partitions o/Unity
391
5.5J Show that every compact n-manifold embeds in some IRk for k big enough in the following way. If {(Vj,
X
= (X 0
X
where fj =0 outside Vj' Show that fj =(f1(m), ... , fN(m), X(
~;R0,
]RN, \jI(m)
is
C~
and that \jI:
is an embedding.
5.5K Let g be a Riemannian metric on M. (i) Show that if N is a submanifold, its g-normal bundle vg(N) = {v v
~
E
ToM I n
E
N,
ToN} is a subbundle of TM.
(ii) Show that TM I N = v g(N)
(f)
TN.
(iii) If h is another Riemannian metric on M, show that v g(N) is a vector bundle iso--
morphic to vh(N).
5.5L Show that if f: M ~ N is a proper surjective submersion with M paracompact and N connected, then it is a locally trivial fiber bundIe. (Hint: To show that all fibers are diffeomorphic, connect a fixed point of N with any other point by a smooth path and cover the path with the neighborhoods in N given by 5.5.14.)
Chapter 6 Differential Forms Differential k-fonns are tensor fields of type (0, k) that are completely antisymmetric. Such tensor fields arise in many applications in physics, engineering, and mathematics. A hint at why this is so is the fact that the classical operations of grad, div, and curl and the theorems of Green, Gauss, and Stokes can all be expressed concisely in tenns of differential fonns. However, the examples of Hamiltonian mechanics and Maxwell's equations (see Chapter 8) show that their applicability goes well beyond this. The goal of the chapter is to develop a special calculus of differential fonns, due largely to E. Cartan [1945]. The exterior derivative operator d plays a central role. The properties of d and the expression of the Lie derivative in tenns of it will be developed.
~6.1
Exterior Algebra
We begin with the exterior algebra of a vector space and extend this fiberwise to a vector bundle. As with tensor fields, the most important case is the tangent bundle of a manifold, which is considered in the next section. We first recall a few facts about the pennutation group on k elements. Proofs of the results that we cite are obtainable from virtually any elementary algebra book. The permutation group on k elements, denoted Sk' consists of all bijections the fonn of a table
0:
{I, ... , k} -+ {I, ... , k} usually given in
1 ... k) (0(1) ... o(k) ,
together with the structure of a group under composition of maps. Clearly, Sk has order k!. Letting {-I, I} have its natural multiplicative group structure, there is a homomorphism denoted sign: Sk -+ (-1, I); that is, for called even when sign
0
0,
't E Sk' sign(o
0
't) = (sign o)(sign 't). A pennutation
= + 1 and odd when sign (j = -1.
(j
is
This homomorphism can be described
as follows. A transposition is a pennutation that swaps two elements of { 1, ... , k}, leaving the remainder fixed. An even (odd) pennutation can be written as the product of an even (odd) number of transpositions. The expression of (j as a product of transpositions is not unique, but the number of transpositions is always even or odd corresponding to (j being even or odd. If E and F are Banach spaces, an element of Tlk(E, F) = Lk(E; F); i.e., a k-multiljnear
Section 6.1 Exterior Algebra
393
continuous mapping of Ex··· x E -t F is called skew symmetric when
for all el •...• ek E E and 0 E Sk' This is equivalent to saying that t(el ..... ek) changes sign when any two of el •...• ek are swapped. The subspace of skew symmetric elements of Lk(E; F) is denoted Lka(E; F) (the subscript a stands for "alternating"). Some additional shorthand will be useful. Namely. let AO(E. F) = F. AI(E. F) = L(E. F) and in general. Ak(E. F) = Lka(E; F). be the vector space of skew symmetric F-valued multilinear maps or exterior F-valued k-forms on E. If F = JR. we write AO(E) = JR. AI(E) = E* and Ak(E) = Lka(E; JR); elements of Ak(E) are called exterior k-Jorms. Some authors write Ak(E*) where we write Ak(E). To form elements of Ak(E. F) from elements of TOk(E; F). we can skew-symmetrize the latter. For example. if t E T02(E). the two tensor At defined by
is skew symmetric and if t is already skew. At coincides with 1. More generally. we make the following definition. 6.1.1 Definition The alternation mapping A: TOk(E. F) -t TOk(E. F) (for notational Simplicity we do not index the A with E. F or k) is defined by
At( el •...•ek) =
:! L
(sign 0) t( ecr(l) •...• ecr(k»'
crESk
where the sum is over all k! elements of Sk'
6.1.2 Proposition and A 0 A =A.
A is a linear mapping onto Ak(E. F). A I Ak(E. F) is
the identity.
Proof Linearity of A is clear from the definition. If IE Ak(E. F). then
At(el •...• ek) =
:! L
~~
(sign 0) t (ecr(I)' ...• ecr(k» =
:! L
t(el' ...• ek) = t(el •...• ek)
~~
since Sk has order k!. This proves the first two assertions. and the last follows from them . • From A = A 0 A. it follows that II A II:C:; II A U2. and so. as A"# 0.11 A II ~ 1. From the definition of A. we sec II A II:C:; 1; thus II A II = 1. In particular. A is continuous.
Chapter 6 Differential Forms
394 6.1.3
Definition
If a E 'fOk(E) and
~
E TO,(E). define their wedge product a
1\
~
E
Ak+'(E) by a
1\
(k + ~)! ~ = - - A(a®~). k!~!
For F-valued forms. we can also define
1\.
where ® is taken with respect to a given
bilinear form BE L(F 1• F 2; F3)' Since A and ® are continuous. so is possible conventions for defining the wedge product
1\.
1\.
There are several
The one here conforms to Spivak [1979).
and Bourbaki [1971) but not'to Kobayashi and Nomizu [1963) or Guillemin and Pollack [1974). See Exercise 6.1G for the possible conventions. Our definition of
a
1\
~
is the one that
eliminates the largest number of constants encountered later. The reader should prove that for exterior forms
0) where the sum is over all (k. ~) shuffles cr. The reason for the name "shuffles" is that these are the kind of permutations made when a deck of k + I cards is shuffled. with k cards held in one hand and ~ in the other); that is. permutations cr of {I. 2 •...• k + ~} such that
cr(1) < ... < cr(k)
and cr(k + 1) < ... < cr(k + ~). Formula (1) is a convenient way to compute wedge products. as we see in the following examples.
6.1.4 Examples A If
a is a two-form and
~
is a one-form. then
Indeed the only (2. 1) shuffles in S3 are
of which only the second one has sign -1. B
If a and
~
are one-forms. then
since S2 consists of two (1. 1) shuffles. + 6.1.5 Proposition For a E TOk(E). 13 E 'fO,(E). and 'YE TDm(E). we have (i) a 1\ 13 = Aa 1\ 13 = a 1\ AI3;
395
Section 6.1 Exterior Algebra
(ii) /\ is bilinear; (iii) a /\
13 = (-Itt 13 /\ a;
(iv) a /\
(13/\ y) =
(a /\
13) /\ Y =
Proof For (i). first note that if cr A(crt) =(sign cr)At. because
=
E
1
k'
(k+ ~ +m)'
k'~'m'
13 ® y).
A(a ®
Sk and we define crt(et' ...•
~.
£.
~)
= t(eO"(I)' ...• eO"(k)' then
.
(Sign cr) (sign pcr)t(epO"(I)' ...• epO"(k)
PE~
=
(sign cr)At(el' ...• ek)
since p 1-+ pcr is a bijection. Therefore. since
L
At = :,
(sign cr)crt.
0" ES k
we get A(Aa ®
13)
=
L (sign 't)('ta (13)]
A (:,
~ESk
= :,
L
(sign 't)A('ta ®
13)
(by linearity of A)
~E~
:, L
(sign 't')A't'(a ®
13)
~E~
where 't' E Sk+t is defined by 't'(I •...• k •...• k+~) = ('to) •...• 't(k). k + 1•...• k + ~) so sign 't = sign 't' and 'ta ® becomes
13 ='t'(a ® 13).
:, L (sign 't')(sign 't')A(a
Thus the preceding expression for A(Aa ®
®
13) = A(a ® 13) :,
'tESk
Thus A(Aa ®
13) =A(a ® 13)
L
1 = A(a ®
13)
13)·
'tESk
which is equivalent to (Aa) /\
13 = a
/\
13.
The other equality in (i)
is similar. (ii) is clear since ® is bilinear and A is linear. For (iii). let cr o E Sk+t be given by cro(l, ...• k + ~) = (k + 1•...• k + ~. 1•...• k). Then
Chapter 6 DilJerentiol Forms
396
(a ® ~)(el' ...• t\+I) = (~ ® a)(ecrJ 1)'
••••
eCSo(k+I»' Hence. by the proof of (i). A(a ® ~)
= (sign
ao)
A(~ ® a). But sign a o = (-1)kt. Finally. for (iv).
a
1\
(~
1\
1) =
(k + ~ + m)! k!(~
A( a ® (~
+ m)!
1\ 1»
(k + ~ + m)! (~+ m)1 ..:....-.:..... A(a ® k!( ~ + m)! ~!m!
...:..---~
=
(k + ~
+ m)!
k!~!m!
A(~
®
1»
A
A( a ® ... ® 1)
since A(a ® A~) = A(a ® ~). which was proved in (i). and by associativity of ®. We calculate (0.1\
~)
1\
1 in the same way . •
Conclusions (i)-(iii) hold (with identical proofs) for F-valued forms when the wedge product is taken with respect to a given bilinear mapping B. Associativity can also be generalized under suitable assumptions on the bilinear mappings. such as requiring F to be an associative algebra under B. Because of associativity. 0.1\ ~ 1\ 1 can be written with no ambiguity.
6.1.6 Examples A If a i• i = 1•...• k are one-forms. then (0. 1 1\
••• 1\
a k) (el' ...• ek) =
L (sign a)al(ecs(l)"'ak(eCS(k»
=
det[ai(ej)]'
cs Indeed. repeated application of 6.1.5 (iv) gives
(2)
where a i is a deform on E. In particular. if a i is a one form. (2) gives (2')
which yields the stated formula after using the definition of A. If el' .... en and e I •...• en are dual bases. observe that as a special case. (e 1 1\ ... 1\ ek) (el' ...• ek ) = 1. B If at least one or a or 13 is of even degree. then 6.1.5 (iii) says that 0.1\ 13 = 13 1\ a. If both are of odd degree. then 0.1\ 13 = -13 1\ a. Thus. if a is a one-form. then a 1\ 0.= 0. But if a is a two-form. then in general a 1\ 0.*0. For example. if 0.= e l 1\ e 2 + e 3 1\ e4 E A2(lR4) where e l • e2 • e3 • e4 is the standard dual basis of ]R4. then 0.1\0. =2e l l\ e2 1\ e3 1\ e4 *0.
397
Section 6.1 Exterior Algebra
C The properties listed in 6.1.5 make the computations of wedge products similar to polynomial multiplication. care being taken with commutativity. For example. if a l ..... a s are one forms on IRs. and
then the wedge product a a
A
~
a = 2a I
A
a 3 + a 2 A a 3 - 3a3
~ = - al
A
a2
A ~
as + 2a l
A
A
a3
A A
a 4 E A 2(IRS) a4
E
A(lRS).
is computed using the bilinearity and commutation properties of
= -2(a l A ( 3 ) A (a l A a 2 A as) -
A
a2
A
as)
as) + 4(a l
A ( 3) A
(a l
A
a3
A ( 4)
6(a3
A ( 4) A
(a l
A
a3
A ( 4)
(a2
+ 3(a3 A
( 4) A
(a l
A
a2
A
+ 2(a2 A
( 3) A
(a l
A
a3
A ( 4) -
A ( 3) A
(a l
A:
To express the wedge product in coordinate notation. suppose E is fmite dimensional with basis e p
... ,
en' The components of t E 'P'k(E) are the real numbers (3)
For t
E
Ak(E). (3) is antisymmetric in its indices i l ..... ik. For example. t E A2(E) yields ~j' a
skew symmetric n x n matrix. From definition 6.1.1 of the alternation mapping and (3). we have
(At)il"'~ =
:! L
(sign a)tO(i1)'''O(ik);
OE~
i.e., At antisymmetrizes the components of
If a
E
1.
For example. if t E 'P'2(E). then
Ak(E) and ~ E A'(E). then (1) and (3) yield
where the sum is over all the (k. ~)-shuff1es in Sk+"
6.1.7 Definition The direct sum of the spaces Ak(E) (i = O. 1.2 ... ) together with its structure of real vector space and multiplication induced by A. is called the exterior algebra of E. or the Grassmann algebra of E. It is denoted by A(E). Thus A(E) is a graded associative algebra. i.e .• an algebra in which every element has a
Chapter 6 Differenliol Forms
398
degree (a k-form has degree k), and the degree map is additive on products (by 6.1.2 and 6.1.3). Elements of A(E) may be written as finite sums of increasing degree exactly as one writes a polynomial as a sum of monomials. Thus if a,b,ce JR, aeAI(E) and ~eA2(E) then a+ba + c~ makes sense in A(E). The one-form a can be understood as an element of AI(E) and also of A(E), where a is identified with 0 + a + 0 + 0 + .... 6.1.8 Proposition Suppose E is finite dimensional and n
=dim E.
Thenfor k> n, Ak(E)
=
{O}, while for 0 < k ~ n, Ak(E) has dimension n!/(n-k)!k!. The exterior algebra over E has
dimension 2n. If {et' ... , en} is an (ordered) basis of E and {e l , of Ak(E) is
... ,
en} its dual basis, a basis
(4) Proof First we show that the indicated wedge products span Ak(E). If a e Ak(E), then from
5.1.2, a = a(ei l' ... , eik)eil ® ... ® e~, where the summation convention indicates that this should be summed over all choices of it' ... , ik between 1 and n. If the linear operator A is applied to this sum and (2) is used, we get a
· · 1 ·
.
ll ® ... ® elk) = a(ei ' ... , ei ) -kl ell /\ ... /\ elk = Aa = a(ei I ' ... , e;.)A(e -. I k.
The sum still runs over all choices of the il' ... , ik and we want only distinct, ordered ones. However, since a is skew symmetric, the coefficient in a is 0 if it' ... , ik are not distinct. If they are distinct and 0 e Sk' then
since both a and the wedge product change by a factor of sign 0. Since there are k! of these rearrangements, we are left with
a =
L
a(ei l' ... , eik)eil /\ ... /\ eik.
i 1 < ...
This shows that (4) spans Ak(E). Secondly, we show that the elements in (4) are linearly independent. Suppose that
For fixed i'l' ... , i'k' let
j'k+l' ... ,
j'n denote the complementary set of indices,
j/k+1
< ... < j'n'
Section 6.1 Exterior Algebra
399
Then
However. this reduces to
But e l /\ ... /\ en "* O. as (e l /\ ... /\ en)(el' ...• en) = 1 by Example 6.1.6A. Hence the coefficients are zero. • 6.1.9 Corollary If dim E = n. then dim An(E) = 1. If {al •...• an} is a basisfor E*. then a l /\ ... /\ an spans An(E). Proof This follows from 6.1.8• • 6.1.10 Corollary Let a l •...• a k an spans An(E).
E
E*. Then a l •...• a k are linearly dependent iff a l
/\ ... /\
Proof If a I•...• a k are linearly dependent, then ai=
I, cjaj j ..i
for some i. Since a /\ a = 0, for a a one-form, we see that a l /\ ... /\ a k = O. Conversely, if a l /\ ... /\ a k = 0, then by 6.1.9, ai, ... , a k is not a basis for span{a l , •.• , a k }. Therefore k > dim(span{a l •...• a k )) and so ai, ... , a k are linearly dependent. • 6.1.11 Corollary Let 9 E AI(E) and a
E
Ak(E). Then 9/\ a = 0 iff there exists ~
E
Ak-I(E)
such that a = 9 /\ ~.
Proof Clearly, if a =9 /\~, then 9/\ a = O. Conversely, assume 9/\ a choose a basis {eJiEI of E such that for some k E I, ek =9. If
=0,
9"* 0 and
from 9 /\ a = 0 it follows that all summands not involving ek are zero. Now factor ek oul of the remaining terms and call the resulting (k-l)-form ~. •
400
Chapter 6 Differentiol Forms
6.1.12 Examples A Let E = JR2, {el' e 2} be the standard basis of JR2 and {e l , e2 } the dual basis. Any element ffi of A I (JR2) can be written uniquely as ffi = ffile l + ffi 2e 2, and any element ffi of A(JR2) can be written uniquely as ffi = ffi12 e l /\ e 2. 8
Let E=]R3, {el'e2'~} be the standard basis, and {e l ,e2,e3 } the dual basis. Any E AI(]R3) can be written uniquely as ffi = ffile l + ffi 2e 2 + ffi3e3. Similarly, any
element ffi
elements 11 E A2(lR3) and ~ E A3(JR3) can be uniquely written as 11 = 1l12e l /\ e2 + ll\3el /\ e 3 + 1l23e2 /\ e 3 and ~ = ~123el /\ e2 /\ e3. • Since JR3, A I(JR3), and A2(JR3) all have the same dimension, they are isomorphic. An isomorphism ]R3 ;: AI(]R3) = (JR3)* is the standard one associated to a given basis: ei ...... e i , i = 1,2,3. An isomorphism of AI(JR3) with A2(JR3) is determined by
This isomorphism is usually denoted by *: AI(JR3) 1-+ A2(]R3); we shall study this map in general in the next section under the name Hodge star operator. The standard isomorphism of JR3 with A I(JR3) = (JR3)* is given by the index lowering
action b of the standard metric on JR3; i.e., b (ei ) = ei . Then *
0
b :
JR3 -+ A2(JR.3) has the
following property:
(5) for all v, w
E
JR.3, where x denotes the usual cross-product of vectors; i.e.,
The relation (5) follows from the definitions and the fact that if a = aIel + ~e2 + ~e3 and ~ = ~Iel + ~2e2 + ~3e3, then
Exercises 6.1A Compute a /\ a, a /\~, ~ /\ ~,and ~ /\ a /\ ~ for a = 2e l /\ e 3 - e2 /\ e3 E A2(JR3) and ~=-el +e2 _2e 3 where {e l ,e2,e3} is a basis of (JR.3)*. 6.18 If k! is omitted in the definition of A (6.1.1), show that /\ fails to be associative. 6.1C Let v\, ... , vk be linearly dependent vectors. Show that a(vl' ... , vk ) = 0 for all a Ak(E).
E
Section 6.1 Exterior Algebra
401
6.1D Let E be finite dimensional. Show that Ak(E*) is isomoprhic to (Ak(E»*. (Hint: Define
1\ ••• /\
(iii) dim(A(E,
e\)fj I il < ... < ik} is a basis of Ak(E, F) and thus dim (A k(E,
F» = (n:;k!;
F» = m2n;
(iv) if B E LOR, F; F), where B(t, f) = tf and /\ is the wedge product defined by B, regarded as a map /\: A1(E) x Ak(E, F) -+ Ak+I(E, F) show that
compute a. /\ ~. 6.1F Let (el' ... ,ek } and (fl' ... , fk} be linearly independent sets of vectors. Show that they span the same k-dimensional subspace iff fl /\ ... /\ fk = ae l /\ ... /\ ~,where a O.
'*
(Give a definition of fl /\ ... /\ fk as part of your answer.) Show that in fact
is determined by
+ 1) = c(1, 1)",0). Show llial
c(k, ~) = ",(k + ~)""(k)",( ~). Deduce that c(k, ~) = c(~, k); i.e., /\' satisfies a. /\' ~ automatically.
=(-lft~ /\' a.
(iii) Show that c given by (iii) yields an associative wedge product. (",(k) = 11k! converts our wedge product convention to that of Kobayashi and Nomizu [1963 D.
~6.2
Determinants, Volumes, and the Hodge Star Operator
According to linear algebra. the detenninant of an n x n matrix is a skew symmetric function of its rows or columns. Thus. if xl' .... xn E ]Rn. then m(x l ..... xn) = det[x l ..... xnl denotes the n x n matrix whose colums are xl' .... xn' is an element of An(~n). We also recall from linear algebra that det[x l ..... xnl is the oriented volume of the parallelipiped spanned by xl' .... xn (Fig. 6.2.1) and that if xi has components xii' the detenninant is given by
det[xl' .... xnl =
L (sign cr)~(l) ..... x~(s). O'ESn
~------------~
p
x,
Figure 6.2.1 Volume(P) =det[xl' x2 • x3 l
In this section detenninants and volumes are approached from the point of view of exterior algebra. Throughout this section E is assumed to be a finite-dimensional vector space and wc shall denote its dimension by dim E = n. If cp : ]R3 -t]R3 is a linear transfonnation. it is shown in linear algebra that det cp is the oriented volume of the image of the unit cube under cp (see Fig. 6.2.2). In fact det cp is a measure of how cp changes volumes. In advanced calculus. this fact is the basis for introducing the Jacobian detenninant in the change of variables fonnula for multiple integrals. This background will lead to the development of the Jacobian detenninant of a mapping of manifolds. Recall that the pull-back cp*a of a E ~k(F) by cp E L(E. F) is the clement of TOk(E) defined by (cp*a)(el' .... ~) =a(cp(e l )..... cp(ek If cp E G.L(E, F), then cp* =(cp-l)* denotes
».
the push-forward. The following proposition is a consequence of the definitions and 5.1.9. (The same results hold for Banach space valued fonns.)
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
403
p
Figure 6.2.2 6.2.1 Proposition Let cp E L(E. F). and", E L(F. G) (i) cp*: Tlk(F) -+ Tlk(E) is linear. and cp*(A(F» (ii) ('II' 0 cp)* =cp* 0 ",*. (iii)
(iv)
C
Ak(E).
If cp is the identity, so is cp*. If cp E aL(E.F). then cp* E aL(Tlk(F). Tlk(F». (cp-I)* = cp*. and (cp*)-I = (cp-Ih;
if 'II' E (v) If a
E
aL(F. G). then ('II' 0 cp)* = "'* 0 cp*. Ak(F). and ~ E A '(F). then cp*(a,,~) = cp*a" cp*~.
For example. if
and cp
E
L(E. F) is given by the matrix [A"i l . i.e .• relative to ordered bases {el' ...• eo} of E
and {fl' ...• fm} of F. one has cp(ei ) = A8l8. then (cp*~)
= ~8
=
I
···8
cp*(1"I)" ... "cp*(1"k). (sum over al < ... < ak)
k
~81"'"kA8ljleil" ... " A8\eik
= A ... "'·1
A 81 .... A8k. ei" ···"eik Jl
at
Jk
'
That is (cp*~)
= ~8
=
A
cp*(1"I)" ... "cp*(1"k). (sum over al < ... < ak)
... 8
I
"'8 1
k
...
= A
8
A 81.eil" .. '"A8k.eik Jk k JI
"'8 1"'"k
A 8 1 ..•. A 8 k. _i" ... " -ik JI Jkl:' I:' •
= k!~8 ... 8 A 8IJ.••. A8kJ. ei l " ... "eik• 1
k
1
k
jl < ... <jk·
Recall that <1'*: AO(E) -+ AO(E) is a linear mapping and AO(E) is one-dimensional.
Thus. if COo
404
Chapter 6 DifjerenJiol Forms
is a basis and 0> = ColO' then = c
o = bo> for some constant b. clearly unique. 6.2.2 Definition Let dim(E)
=n
and
N(E) --+ An(E) satisfies =(det cp)o> for all 0> e N(E) is called the determinant of cpo The defmition shows that the determinant does not depend on the choice of basis of E. nor does it depend on a norm on E. To compute det cpo choose a basis {el' .... en} of E with dual basis {el ..... en}. Let cpe L(E.E) have the matrix [Aiil;thatis. cp(e i) = ~lsisnAiiei' By Example 6.1.6A.
Since (e l /\ ... /\ en)(el' .... en) = 1 we get det cp = det[Aiil. the classical expression of the determinant of a matrix with xl' .... xn as columns. where xi has components Ai i. Thus the definition of the determinant in 6.2.2 coincides with the classical one. From properties of pull-back. we deduce corresponding properties of the determinant. all of which are well known from linear algebra. 6.2.3 Proposition Let cpo '" e L(E. E). Then (i) (ii) (iii)
det(cp 0
"')
=(det cp)(det "');
if cp is the identity, det cp'= 1; cp is an isomorphism iff det cp *- O. and in this case det(cp-I)
Proof For (i). (cp
0
"')*0>
= det(cp
0
",)0>; but (cp
0
"')*0>
= (",*
0
=(det cp)-I.
Cp*)O>. Hence. (
0
"')*0>
=
",*(det cp)o> = (det ",)(det and (i) follows. (ii) follows at once from the definition. For (iii). suppose cp is an isomorphism with inverse cp-I. Therefore. by (i) and (ii). 1 = det(cp
0
cp-I)
=
(det cp)(det cp-I). and. in particular. det
. we have (cp*o»(el' .... en) = 0>(0. cp(e2 ) ..... cp(en = O. Hence. det cp = O. •
»
In Chapter 2 we saw that if E and F are finite dimensional. one convenient norm giving the topology of L(E. F) is the operator norm:
1
11 cp(e) II II cp II = sup{1I cp(e)1I i II e II = I} = sup II ell where II e II is a norm on E. (See §2.2.) Hence. for any e
E
E.
Ie*-O)
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
405
II q>(e) II ~ II cp 1111 e II. 6.2.4 Proposition The 17Ulp del: L(E, E) -+ lR is continuous. Proof This is clear from the component formula for det, but let us also give a coordinate free proof. Note that II ro II
= sup { I ro(e" ... , en) 1111 e l II = ... =II en II = 1} = sup { I ro(e l ,
... ,
en) 1111 e l II .. · II en lei' ... , en
is a norm on An(E) and I ro(e l , ... , en) I
~
'* o}
II ro 1111 e l 11 .. ·11 en II. Then, for cp, '" E L(E, E),
I det cp - det '" III ro II = II cp*ro- ",*ro II
= sup{
I ro(cp(e l ),
... ,
»- ro(\jf(e
cp(en
l ), ... ,
»
\jf(en 1111 e l II
= ... = II en II = I}
»
~ sup { I ro(cp(e l ) - ",(e l ), cp(e2 ), ... , cp(en 1+'"
»
+ I ro(\jf(e l ), \jf(~), ... , cp(en) - \jf(en 1111 e l II
= ... = II en II = I}
~ II ro 1111 cp - \jf II {II cp IIn-1 + II cp IIn- 2 II '" II + ... + II \jf lin-I} ~ II ro 1111 cp - \jf 11<11 cp II + II \jf 11)0-1.
Consequently, I det cp - det \jf I ~ II cp - '" 11<11 cp II + II \jf lI)n-1 from which the result follows. _ In Chapter 2 we saw that the set of isomorphisms of E to F form an open subset of L(E, F). Using the determinant, we can give an alternate proof in the finite-dimensional casco
6.2.5 Proposition Suppose E and F are finite-dimensional and let G.L(E, F) denote those cp E L(E, F) tlult are isomorphisms. Then G.L(E, F) is an open subset of L(E, F). Proof If G.L(E, F) =0, the conclusion is true. Otherwise, there is an isomorphism \jf E G.L(E, 0 cp is also. This happens precisely when dct(",1 0 cp) O. Therefore, G.L(E, F) is the inverse image of lR \ {O} under the map taking cp to det(",1 0 cpl. Since this is continuous and lR \ {O} is open, G.L(E, F) is also open. _ F). A map cp in L(E, F) is an isomorphism if and only if ",-I
'*
The basis elements of An(E) enable us to define orientation or "handedness" of a vector space. 6.2.6 Definition The nonzero elements of the one-dimensional space An(E) are called volume elements. If ro l and COz are volume elements, we say ro l and COz are equivalent iff there is a c
406
Chapter 6 Differential Forms
> 0 such that COl = cco 2 . An equivalence class [co] of volume elements on E is called an orientation on E. An oriented vector space (E. [co]) is a vector space E together with an orientation [co] on E; [-co] is called the reverse omntation. A basis {el' .... en} of the oriented vector space (E. [co]) is called positively (resp. negatively) oriented, if ro(e l •.... en) > 0 (resp. < 0).
The last statement is independent of the representative of the orientation [co]. for if co'
E
[co], then co' = cco for some c > O. and thus co'(e l •...• en) and co(e l •.... en) have the same sign. Also note that a vcctor space E has exactly two orientations: one given by selecting an arbitrary dual basis {e l •...• en} and taking [e l /\ ... /\ en]; the other is its reverse orientation. This definition of orientation is related to the concept of orientation from calculus as follows. In JR.3. a right-handed coordinate system like the one in Fig. 6.2.1 is by convention positively oriented. as are all other right-handed systems. On the other hand. any left-handed coordinate system. obtained for example from the one in Fig. 6.2.1 by interchanging
XI
and x2 • is by
convention negatively oriented. Thus one would call a positive orientation in JR.3 the set of all right-handed coordinate systems. The key to the abstraction of this construction for any vector space lies in the observation that the determinant of the change of ordered basis of two right-handcd systems in JR.3 is always strictly positive. Thus. if E is an n-dimensional vector space. define an equivalence relation on the set of ordered bases in the following way: two bases {el' ...• en} and {e l ' •...• en'} are equivalent iff det
O. where
7'
0
since co 7' 0 in N(E). Changing the sign of e l if necessary. we can find a basis that is positively oriented. Let {e l ' •... , en'} be an equivalent basis and 0 such that co'
=
cco. so we get
That is, [co] uniquely determines the equivalence class of {e l , ...• en}' Conversely. let {el' ... , en} be a basis of E and let co = e l /\ ... /\ en, where {el, ... , en} is the dual basis. As before, co'(e l ' •... ,en') > 0 for any co' E [co] and {e l ', ... , en'} equivalent to {e l , ... , en}; thus, the equivalence class of the ordered basis {el' ... ,en} uniquely determines the orientation [co]. _
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
407
Next we shall discuss volume elements in inner product spaces. An important point is that to get a particular volume element on E requires additional structure. although the determinant does not. The idea is based on the fact that in JR3 the volume of the parallelipiped P = P(x l • x2• x3) spanned by three positively oriented vectors xI' x2• and x3 can be expressed independent of any basis as
where [(xi' x)l denotes the symmetric 3 x 3 matrix whose entries are (xi' x). If xi' x 2. and x3 are negatively oriented. det[ (xi' x)1< 0 and so the formula has to be modified to (I)
This indicates that besides the volumes. there are quantities involving absolute values of volume elements that are also important. This leads to the notion of densities.
6.2.8 Definition Let a be a real number. A continuous mapping p : Ex··· x E -+ R (n factors of E for E an n-dimensional vector space) is called an a-density if p(q>(v l ) •...• q>(v n =I det q> In p(vi' ...• vn),for all VI' ...• Vn E E and all q> E L(E. E). Let I A In(E) denote the a-densities on E. With a = 1. I-densities on E are simply called densities and I A II(E) is denoted by I A I(E).
»
The determinant of q> in this definition is taken with respect to any volume element of E. As we saw in 6.2.2. this is independent of the choice of the volume element. Note that I A In(E) is one-dimensional. Indeed. if p I and P2 then P2(e l •...• en)
= aPI(el'
q>(e). defining q>
L(E. E). Then
E
E
I A In(E). p I
;to
...• e n). for some constant a
O. and {ei' ...• en} is a basis of E. E
JR. If v\' ...• vn
E
E. let Vi
=
i.e .• P2 = aPI· Alpha-densities can be constructed from volume elements as follows. If ro E An(E). define I ro In
E
I A In(E)
by
I ro In(ei' ...• en) = I ro(e l •...• en) In where el' ...• en
E
E. This
association defines an isomorphism of An(E) with IAln(E). Thus one often uses the notation lrol n for a-densities. We shall construct canonical volume elements (and hence a-densities) for vector spaces carrying a bilinear symmetric nondegenerate covariant two-tensor. and in particular for inner product spaces. First we recall a fact from linear algebra.
6.2.9 Proposition Let E be an n-dimensional vector space and g = (.) E 'J"
408
Chapter 6 Differentiol Forms
r
g = LCie®ei• i=)
where ci = ±1 and r ~ n. or equivalently, the matrix of g is
o
C2
0
o
0
o
o
0
o
o
0
o o
000 000
o
0
o
000
o
0
o
This basis (e) •...• en} is called a g-orthonormal basis. Moreover, the number of basis vectors for which g(ei• ei) = 1 (resp. g(ei• ei) = -1) is unique and equals the maximal dimension of any subspace on which g is positive (resp. negative) definite. The number s = the number of + I's minus the number of -1 's is called the signature of g, The number of -I's is called the index of g and is denoted Ind(g), Proof (Gram-Schmidt argument) Since g is symmetric. the following polarization identity holds: g(e. 0 = [(g(e + f. e + 0 - g(e - f. e - 0)]/4. Thus if g"" O. there is an e) E E such that g(e\, e) "" 0, Rescaling. we can assume c) = g(e\, e)=±1. Let E) be the span of e) and E 2 = feE Elg(e\,e)=O}, Clearly E) n E2 = {O}, Also. if z E E. then z - c)g(z. e)e) E Ez so that E =E) + Ez and thus E =E) EB Ez, Now if g "" 0 on E 2• there is an e 2 E E2 ~uch that g(e 2• e 2) = c 2 = ±l. Continue inductively 10 complete the proof, For the second part. in the basis Ie\, "'. en} just found. let E) = span{ei I g(e i• e i) = I}. E2 = span {e;l g(ei• e) = -1} and ker g = {e I g(e. e') = 0 for all e' E E}, Note that ker g = span (ei I g(ei• ei) = O} and thus E = E) EB E2 EB ker g, Let F be any subspace of E on which g is positive definite, Then clearly F n ker g = {O}. We also have Ez n F = {O} since any v E E2 n F. V"" O. must simultaneously satisfy g(v. v) > 0 and g(v. v) < 0, Thus F n (E2 EEl ker g) = {O} and consequently dim F ~ dim E), A similar argument shows that dim Ez is the maximal
Section 6.2 Determinants, Volumes, and the H oelge Star Operator
409
dimension of any subspace of E on which g is negative definite. _ Note that the number of ones in the diagonal representation of g is (r + s)/2 and the number of minus-ones is Ind(g)
= (r - s)/2.
Nondegeneracy of g means that r =n. In this case
e E E may be written e = Li = I..... n [gee. e)/c i1ei' where ci = g(ei• ei) =± 1 and (eJ is a g-orthonormal basis. For g a positive definite inner product. r = n and Ind(g) = 0; for g a Lorentz inner product r = n and Ind(g) = 1. 6.2.10 Proposition Let E be an n-dimensional vector space and g E T>2(E) be nondegenerate and symmetric. (i) If [001 is an orientation of E there exists a unique volume element IL = lL(g) E [001. called the g-volume, such that lL(el' .... en) = I for all positively oriented g-orthonormal bases {el' ...• en} of E. Infact. if {e l ..... en} is the dual basis, then IL =
e l " ... " en. More generally, if {fl' .... fn} is a positively oriented basis with dual basis {fl, ...• fR}. then IL = I det[g(fi• 911112 fl " ... " fRo
(ii) There is a unique a-density III la. called the goa-density, with the property that III la(el' .... en) = 1 for all g-orthonormal bases {el' .... en} of E. If {e l , .... en} is the dual basis, then III la = leI" ... "en la. More generally, if vI' .... vn
E
E,
then
Proof First we establish a relation between the determinants of the following three matrices: [g(ei• ej)1 = diag(cl' ...• cn) (see 6.2.9), [g(fi• 91 for an arbitrary basis {fl' .... fn}, and the matrix representation of
g(fi' fj)=
(t
cpeP ® epJ (A\ek' A'je,) =
Cp~Pk~P,AkiA'j= cpAPiAPj . (sum on p)
p=1
Thus det[g(fi• 9 1 = (det
(2)
By 6.2.9. I det[g(e i• e)11 = 1. (i) Clearly if {el' .... en} is positively oriented and g-orthonormal. then lL(el' .... en) = I uniquely determines IL E [001 by multilinearity. Let {fl' .... fn} be another positively oriented g-orthonormal basis. If
410
Chapter 6 DijJerentiol Forms
For the third statement of (i). note that by (2)
(ii) follows from (i) and the remarks following 6.2.8. •
A covariant symmetric nondegenerate two-tensor g on E induces one on Ak(E) for every k = 1•...• n in the following way. Let
(sum over i l < ... < ik ) and let
(sum over all ji' ...• jk) be the components of the associated contravariant k-tensor. where [gkj) denotes the inverse of the matrix with entries
~j
=g(ei, ej).
Then put
(3)
If there is no danger of confusion, we will write (a, P)
= g(k)(a, Pl.
We now show that this
definition does not depend on the basis. If {fl' ... , fn} is another ordered basis of E, let a = a'ai '" akf'1/\ ... /\ f'k and P = P'a l'" atl /\ ... /\ f'k. Then the identity map on E has matrix representation relative to the bases {el' ... , en} and {fl' ... , fn} given by ei = A"i fa' If B = A-I we have by 5.1.7,
So defined. g(k) is clearly bilinear. It is also symmetric since
where [gij] = [gij)-I, and ~j = g(ei, e)- Notice that g(k) is also nondegenerate since if g(k)(a. P) = 0 for all PEA k(E), choosing for 13 all elements of a basis, shows that ai I ... ik = 0, i.e .• that a = O. The following has thus been proved.
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
411
6.2.11 Proposition A nondegenerate symmetric covariant two-tensor g = (,) on the finite-dimensional vector space E induces a similar tensor on Ak(E) for all k = 1 •...• n. Moreover, if Ie\, ...• en} is a g-orthonormal basis of E in which n
~
i
i
g = £.Jcie ® e •
Ci =
±1.
i=1
then the basis
is orthnormal with respect to
g(k) =( .). and
(4) With the aid of the g-volume
~
on E we introduce the Hodge star operator.
6.2.12 Proposition Let E be an oriented n-dimensional vector space and g = (.) E TO2(E) a given symmetric and nondegenerate tensor. Let ~ be the corresponding volume element of E. Then there is a unique isomorphism'" : Ak(E) -t An--l«E) satisfying
(5) If {e l •...• en} is a positively oriented g-orthonormol basis of E and {e l •...• en} is its dual basis, then *(eCJ(I) 1\
••. 1\
eCJ(k»
=cCJ(1) ... cCJ(k) (sign o)(eCJ(k+I) 1\ .•• 1\ eCJ(n)
(6)
where 0(1) < ... < o(k) and o(k+ 1) < ... < o(n). Proof First uniqueness is proved. Let * satisfy Eq. (5) and let ~ = eCJ(I)
1\ ••• 1\
eCJ(k) and a
be one of the g-orthornomal basis vectors
By (5).
a
= 0 unless (i\, ...• ik) = (0(1) •...• o(k». Thus. *~ = aeCJ(k+I) 1\ ... 1\ eCJ(n) for a But then ~ 1\ * ~ = a sign(o)Jl and by (4). (~.~) = cCJ(1) ... cCJ(k). Hence a = cCJ(I) ...
1\ '"
constant a.
~
cCJ(k)sign(o) and so '" must satisfy (6). Thus * is unique. Define * by (6). recalling that e CJ (1) 1\ ... 1\ eCJ(k) for (0) < ... < o(k) forms a g(k)_ orthonormal basis of Ak(E). As before. (5) is then verified using this basis. Clearly * defined by (6) is an isomorphism. as it maps the g-orthonormal basis of Ak(E) to that of An--l«E) . •
Chapter 6 Differential Fonns
412
6.2.13 Proposition Let E be an oriented n-dimensional vector space, g = (, >
E T02 (E)
symmetric and nondegenerate of signature s, and ~ the associated g-volume of E. The Hodge star operator satisfies the following propertiesfor a, ~ E Ak(E): a/\*~
*1 **a (a, ~>
= ~ /\ * a =(a, ~>~, = ~, * ~ =(_l)Ind(g), = (-1 )Ind(g)(-1 j'(n-k)a, = (_l)Ind(g)( * a, * ~>.
(7)
(8)
(9) (10)
Proof Equation (7) follows from (5) by symmetry of (a, ~>. Equations (8) follow directly from (6), witb k = 0, n, respectively, and 0
=identity
(note tbat c 1 ... c n =(_l)Ind(g». To verify ('.I),
it suffices to take a = e°(1) /\ ... /\ e°(k). By (6), *(e0(k+I) /\ ... /\ eo(n»
=be0(I) /\ ... /\ e°(k)
for a constant b. To find b use (5) witb a = ~ = eo(k+I) /\ ... /\ eo(n) to give (see (4» " be0(k+I) /\ ... /\ eo(n) /\ e0(1) /\ ... /\ eo(k) -- co(k+I) ... co(n),""' Hence b
=co(k+I) ... co(n)(-lj'(n-k) sign 0. Thus (6) implies **(e°(1) /\ ... /\ eo(n» =CO(I) ... co(k) (sign 0) *(eo(k+l) /\
... /\ eo(n»
= cO(1)" ,cO(k)co(k+I) .. ,co(n)(sign 0)2(_Ij'(n-k)eO(I) /\ ... /\ e°(k)
=(-I )Ind(g)(-I )k(n-k)e0(1) /\
... /\ eo(k).
Finally for (10), we use (7) and (9) to give (*a, *~>~
= *a
/\ **~
= (_l)Ind(gl(_Ij'(n-k) * a
/\ ~
=(_l)Ind(g)(a, ~>~.
•
= (_l)Ind(g)~ /\ * a
6.2.14 Examples A
The Hodge operator on A I(JR3) where JR 3 has the standard metric and dual basis is given from (6) by * e l =e 2 /\ e3, * e 2 =-e l /\ e3, and * e3 =e l /\ e 2. (This is tbe isomorphism considered in 6.1.11B). BUsing (5), we compute * in an arbitrary oriented basis. Write i i i ... i
*(e1/\ ..• /\ek)
=
(sum over jk+1 < ... < jn) and apply (5) witb
c,l
-Jk+l
.. k.
In
_i
_i
l""k+l /\ .•• /\l""n
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
413
where (jl' ...• jk} is a complementary set of indices to (jk+I' ...• jn}' One gets
Hence .
*(e'l
.
1\ .•• 1\
e'k) = I det [gijll
(1 ...
1I2~
n) .. ...
"",signlj I ... jn g"J, ... g'kJ~k+'
where the sum is over all (k. n - k) shuffies
G~ ... ...
1\ ••• 1\
.
do.
(11)
~) . In
C In particular. if k = I. (11) yields n
I .. I . * e'. = I det ([gij]) I1/2~' "",(-IY- g'Je 1\ ••. 1\ r} 1\ •.• 1\ en
j=1 since sign(jl' j2' ...• jn) = (-I~-I. for j2 < ... < jn' jl = j. and where deleted.
(12)
ej
means that ei is
D From B we can compute the components of * a. where a E Ak(E). relative to any oriented basis: write a = ~ , ... ikei, 1\ ••• 1\ e ik (sum over i l < ... < ik) and apply (11) to give
Hence
n)
~I ... . (*(l)iJ.+,... j•-_ I det[gijll 1/2 "'" £..P; ,... ikg i,~ ... gikjk'sIgn. I ... In
(13)
for jk+1 < ... <jn and where the sum is over all complementary indices jl < ... <jk' E Consider lR,4 with the Lorentz inner product. which in the standard basis [e\, e 2 • e 3 • ) e4 of R4 has the matrix
o
o o
0
o
0
0
0
0
0 0
0
-1
Let {el. e2• e3 • e 4 } be the dual basis. The Hodge operator on A 1(lR4) is given by
414
Chapter 6 Differential Forms
If ]R4 had been endowed with the usual Euclidean inner product, the formulas for * e4 , *(e l /\ e4), *(e z /\ e 4), and *(e3 /\ e4) would have opposite signs. The Hodge * operator on A3(L~4) follows from the formulas on AI(]R4) and the fact that for ~ E AI(]R4), ** ~ == ~ (from
formula (9». Thus we obtain
F
If ~ is a one form and v!' v Z' ... , vn is a positively oriented orthonormal basis, then
This follows from (5) taking ex = vI, the first element in the dual basis.
Exercises 6.2A
Let (e l , eZ, e 3) be the standard dual basis of ]R3 and ex = e l /\ e Z - 2e z /\ e 3 E AZ(]R3), ~ = 3e l - e2 + 2e 3 E Al (]R3), and q> E L(]Rz, ]R3) have the matrix
Compute q>*ex. With the aid of the standard metrics in ]R2 and ]R3, compute * ex, *~, *(q>*ex), and *(q>*~). Do you get any equalities? Explain. 6.2B A map q> E L(E, F), where (E, Ol),
(F,~)
are oriented vector spaces with chosen volume
clements, is called volume preserving if q>*~ = Ol. Show that if E and F have the same (finite) dimension, then q> is an isomorphism. 6.2C A map q> E L(E, F), where (E,
[~])
and (F, [Ol)) are oriented vector spaces, is called
orientation preserving if q>*~ E [Oll. If dim E = dim F, and q> is orientation preserving, show that q> is an isomorphism. Given an example for F = E = ]R 3 of an orientation-preserving but not volume-preserving map. 6.2D Let E and F be n-dimensional real vector spaces with nondegenerate symmetric two-tensors, g E T0 2 (E) and hE T0 2 (F). Then q> E L(E, F) is called an isometry if h(
Section 6.2 Determinants, Volumes, and the Hodge Star Operator
415
(i) Show that an isometry is an isomorphism. (ii) Consider on E and F the g- and h-volumes Il(g) and Il(h). Show that if
orientation-preserving isometry, then
*
Ak(F)
·-1
Ak(E)
*
•
An-k(F)
..
N-k(E)
I·'
If
6.2E Let g be an inner product and {fl' f2' f3} be a positively oriented basis of 1R3. Denote by £> and # the index lowering and raising actions defined by g. (i) Show that for any vectors u, v E JR.3 £>
[*(u
1\
v
£> #
»)
=
sign
(1 23) I i j k
det[g(f., ~b»)1
1/2 . . kl
u'vJg f,. <
(ii) Show that if g is the standard dot-product in 1R3 the formula in (i) reduces to the
cross-product of u and v. (iii) Generalize (i) to define the cross-product of n-1 vectors ul' ... , lin-I in an oriented
n-dimensional inner product space (E, g), and find its coordinate expression.
6.2 F Let E be an n-dimensional oriented vector space and let g E "f02(E) be symmetric and non degenerate of signature s. Using the g-volume, define the Hodge star operator * : Ak(E; F) ~ An-k(E; F), where F is another finite-dimension vector space by
where exi E Ak(E), {fl , ... , fm } is a basis of F and ex = exi fi. Show the following. (i) The definition is independent of the basis of F. (ii) * * = (_l)(n-s)/2+k(n-k) on Ak(E; F). * h'(f, ex) =h'(f, * ex). (iv) If /\ is the wedge product in A(E; F) with respect to a given bilinear form on F, then for ex, /3 E Ak(E; F),
(iii)
If hE "f02(F) and if we let h'(f, ex)
(* ex) /\
/3 = (* /3) /\ ex
=(* exi)h(f, fi), then
and ex /\ (*
/3) = /3 /\ (* ex).
Chapter 6 DifjerentWJ Forms
416
(v) Show how g and h induce a symmetric nondegenerate covariant two-tensor on Ak(E; F) and find formulas analogous to (7)-(10). 6.2G Prove the following identities in JR.3 using the Hodge star operator:
II u x v If = II u 11211 V 112 - (u . v)2
and
u x (v x w) = (u· w)v - (u· v)w.
6.2H (i) Prove the following identity for the Hodge star operator:
where (ii)
(X E
Ak(E) and ~
E
An-k(E).
Prove the basic properties of
*
using (i) as the definition.
6.21 Let E be an oriented vector space and S c: 1'l2(E) be the set of nondegenerate symmetric two-tensors of a fixed signature s. (i) Show that S is open. (ii)
Show that the map vol: g ...... l1(g) assigning to each g
E
S its g-volume element is
differentiable and has derivative at g given by h ...... (trace h)l1(g)!2.
~6.3
Differential Forms
The exterior algebra can now be extended from vector spaces to vector bundles and in particular to the tangent bundle. First of all, we need to consider the action of local bundle maps. As in Chapter 3, V x F denotes a local vector bundle, where V is open in a Banach space E and F is a Banach space. From V x F, we construct the local vector bundle V x Ak(F). Now we want to piece these local objects together into a global one. 6.3.1 Definition Let
each fiber. Then define
~
V' x F' is a local vector bandle map that is an isomorphism on
each fiber, then so is
if
Proof This is a special case of 5.2.4. • 6.3.3 Definition Suppose 1t: £
~
B is a vector bundle. Define
where A is a subset of B and £b = 7r 1(b) is the fiber over bE B. Let Ak(£) I B = Ak(£) and define Ak(1t): Ak(£) ~ B by Ak(1t)(t) = b if t E Ak(£b)' 6.3.4 Theorem Assume {£ I V j ,
I Vj ~
V'j
X
F'j' Then {Ak(£) I V j,
where
=(
0
0
~
B induces naturally a vector
bundle struclure on Ak(£) ~ B.
We now specialize to the important case when 1t: £ ~ B is the tangent bundle. If 'tM : TM ~ M is the tangent bundle of a manifold M, let Ak(M) = Ak(TM), and AkM = Ak('tM), so AkM : Ak(M) ~ M is the vector bundle of exterior k forms on the tangent spaces of M. Also, let
418
Chapter 6 DifjerentiJJI F onns
6.3.5 Proposition Regarding Tlk(M) as an !reM) module. d«M) is an !reM) submodule;
i.e.. Ok(M) is a subspace of Tlk(M) and if f E !reM) and a E Ok(M). then fa E Ok(M). Proof If t l • t2 E Ok(M) and f E !reM). then we must show
ftl
+ lz E Ok(M). This follows
from the fact that for each m E M. the exterior algebra on T mM is a vector space. _ 6.3.6 Proposition If a E Uk(M) and ~ E O'(M). k.
t =O.
1•...• define a /\ ~ : M ~
Ak+'(M) by (a /\ ~)(m) = a(m) /\ ~(m). Then a /\ ~ E Ok+'(M). and /\ is bilinear and
associative. Proof First. /\ is bilinear and associative since it is true pointwise. To show a /\ ~ is of class C~. consider the local representative of a /\ ~ in natural charts. This is a map of the form (a /\
~)q>
=B
0
(aq> x ~q»' with acp. ~q>' ~ and B
=/\.
which is bilinear and continuous. Thus (a
/\ ~)q> is ~ by the Leibniz rule . • 6.3.7 Definition Let OeM) denote the direct sum of Ok(M). k = O. 1•...• together with its
structure as an (infinite-dimensional) real vector space and with the multiplication /\ extended componentwise to OeM). (If dim M = n < 00. the direct sum need only be takenfor k = O. 1•...• n.) We call U(M) the algebra of exterior differentialforms on M. Elements of Ok(M) are called
k-fonns. In particular. elements of
X*(M) are called one-fonns.
Note that we generally regard OeM) as a real vector space rather than an !reM) module (a~ with 'T(M». The reason is that !reM) = OO(M) is included in the direct sum. and f /\ a
=f ® a
= fa. 6.3.8 Remarks and Examples
e on a manifold
M assigns to each mE M a linear functional on T mM.
A
A one-form
B
A two-form c.o on a manifold assigns to each m E M a skew symmetric bilinear map
C
For an n-manifold M. a tensor field t E 'Ps(M) has the local expression t(u) =
j ... j
(l
, .....
Jt
J.
a
a
J'
j
(u)-. ® ... ® - . ® dx I ® ... ® dx' axIl axl,
where u E U. (U. q» is a local chart on M. and
Section 6.3 Differential Forms
t'i "'ir.
.
J,'" J,
i ... dx i r (u) = t( dx' '"
a
- - •••
axi, ,
419
-
a )(u) .
'axil
The proof of 6.1.8(ii) gives then the local expression for
CJ) E
Ak(M), namely
where
D
In Q(M), the addition of forms of different degree is "purely formal" as in the case M = E. Thus, for example, if M is a two-manifold (a surface) and (x, y) are local coordinates on U
C
M, a typical element of Q(M) has the local expression f + g dx +
h dy + k dx /\ dy, for f, g, h, k E
E
.1"(U).
As in §6.1, we have an isomorphism of vector bundles
* : A I(J!t3) ~ A2(J!t3) given by
On the other hand, the index lowering action given by the standard Riemannian metric on
J!t3 defines a vector bundle isomorphism £>: T(J!t3) ~ T\J!t3)
=A 1(J!t3).
These two
isomorphisms applied pointwise define maps
and Then 6.1.12C implies *[(X x Y?l
for any vector fields X, Y
E
=xl> /\ y&
X(J!t3), where X x Y denotes the usual cross-product of
vector fields on J!t3 from calculus. That is,
F
= Xi -a.
i
a .
and Y = Y - . , 1 = I, 2, 3. ax' ax' The wedge product is taken in Q(M) in the same way as in the algebraic case. For example, if M = J!t3, (X = dx 1 - x1dx 2 E Ql(M) and P= x2dx 1 /\ dx 3 - dx 2 /\ dx 3, then (X /\ J3 =(dx 1 - x 1dx 2 ) /\ (x2dx 1 /\ dx 3 - dx 2 /\ dx 3) for X
=0-
X lx 2 dx 2 /\
= (x 1x2
_
dx I
/\
dx 3 - dx I
l)dx 1 /\ dx 2 /\ dx 3. •
/\
dx 2
/\
dx 3 + 0
420
Chapter 6 D;gerentiJll Forms
6.3.9 Definition Suppose F: M ~ N is a C'" mapping of manifolds. For F*o> : M ~ Ak(M) by F*O>(m) = (TmF)* 00>0 F(m); i.e .•
0> E
nk(N). define
where vI' ...• vk E TmM; for g E nD(N). F*g =g 0 F. We say F*o> is the pull-back of F. (See Fig. 6.3.1.)
0>
by
6.3.10 Proposition Let F: M ~ N and G : N ~ W be C'" mappings of manifolds. Then (i) F*: nk(N) ~ nk(M); (ii) (G 0 F)* = F*
0
G*;
(iii) if H : M ~ M is the identity. then H* : Uk(M) ~ Uk(M) is the identity;
(iv) if F is a diffeomorphism. then F* is an isomorphism and (F*)-I = (F-I)*; (v) F*(a 1\ ~) =F*a 1\ F*~ for a E nk(N) and ~ E n'(N). F*
----
......
~~-
F'Ol
M
Figure 6.3.1 Proof Choose charts (U. cp). (V. 'II) of M and N so that F(U) C V. Then the local representative F",'I1 = 'II 0 F 0 cp-I is of class C-. as is 0>'11 = (T",)*o 0> 0 ",I. The local representative of F*o> is
which is of class C'" by the composite mapping theorem. For (ii). note that it holds for the local representatives; (iii) follows from the definition; (iv) follows in the usual way from (ii) and (iii); and (v) follows from the corresponding pointwise result. _ We close this section with a few optional remarks about vector-bundle-valued forms. As before. the idea is to globalize vector-valued exterior forms.
421
Section 6.3 Differentiol Forms
6.3.11 Definition Let
1t:
E ~ B, p: F ~ B be vector bundles over the same base. Define
the vector bundle with base B of vector bundle homomorphisms over the identity from Ak(E) to F. If E = TB, Ak(TB; F) is denoted by Ak(B; F) and is called the vector bundle of F-valued k-forms on M. If F = B x F, we denote it by Ak(B, F) and call its elements vector-valued k-forms on M. The spaces of sections of these bundles are denoted respectively by nk(E; F),
nk(B; F) and nk(B; F). Finally, n(E; F) (resp. n(B; F), n(B, F» denotes the direct sum of nk(E; F), k = 1,2, ... , n, together with its structure of an infinite-dimensional real vector space and ~(B)-module. Thus, a E nk(E; F) is a smooth assignment to the points b of B of skew symmetric k-linear Eb ~ F b. In particular, if all manifolds and bundles are finite dimensional, then a E nk(M, )RP) may be uniquely written in the form a = ~i = I, "', pa i ej , where aI, ... , a P E nk(M), and {el' ... , e p} is the standard basis of )RP. Thus a E nk(E, )RP) is written in
maps
~:
Eb x ...
X
local coordinates as
for i l < ... < ik. Proposition 6.3.10(i)-(iv) and its proof have straightforward generalizations to vector-bundle-valued forms on M. The wedge product requires additional structure to be defined, namely a smooth assignment b 1-+ gb of a symmetric bilinear map gb: Fb x Fb ~ Fb for each b E B. With this structure, 6.3.10(v) also carries over.
Exercises 6.3A Show that for a vector bundle 1t: E ~ B, Ak(E) is a (smooth) subbundle of 'flk(E). Generalize to vector-bundle-valued tensors and forms. 6.3B Let
=uv du "dv E n2(IR2), compute a,,~,
= (x 2, yz).
For a
= v2du + dv E nl()R2)
and ~
6.3C (E. Cartan's lemma). Let M be an n-manifold and aI, ... , a k E n 1 (M), k:5 n, be pointwise linearly independent. Show that ~I, ... , ~k E nl(M) satisfy EISi~ai" ~i =0 iff there exist C~ functions ai i E ~(M) satisfying aii = aij such that ~ = aii~i. (Hint: work in a local chart and show first that a i can be chosen to be dxi; the symmetry of the matrix [aiil follows from anti symmetry of " and the given condition.) 6.3D A (strong) bundle metric g on 1t: E ~ B is a smooth section of L 2.(E; lR) such that g(b) is an inner product on Eb for every bE B which is (strongly) nondegenerate, i.e., ~ E
422
Chapter 6 DifferenJiol Farms
Eb 1-+ g(b)(<;, •. ) E Eb'" is an isomorphism of Banach spaces. (i) Show that the model of the fiber of E is a Hilbertizable space. (ii) If FeE is a subbundle of E, show that Fl.
= UbEBFbl.
is a subbundle of E,
where we define Fbl. = {<;, E Eb I g(b)(<;,. f~ = 0 for all fb E F b} (iii) Show that E = F €a Fl..
6.3E Assume the vector bundle 1t: E ~ B has a strong bundle metric. (i) If a : B ~ F is a smooth nowhere vanishing section of E, let F b =span (a(b)}. F = UbeBFb. Show that F is a subbundle of E which is isomorphic to the trivial bundle EIB = 1R x B. Conclude from 6.3D that EI €a (EI)l. = E.
(ii) Show that a manifold M is parallelizable if and only if TM is isomorphic to a trivial bundle. (iii) Assume that M is a strong Riemannian manifold. that M admits a nowhere vanishing
vector field and that TM ED ElM is isomorphic to a trivial bundle. Let N be another manifold of dimension ~1 such that TN ED EIN is trivial. Show that M x N is parallelizable. (Hint: Use (i) and pull everything back to M x N by the two projections.) (iv) Show that if dim N = O. the conclusion of (ii) is false. (Hint: It is know that the only odd dimensional spheres with trivial tangent bundle are S I, S3 and S7. Show that Ts2n-1 has a nowhere vanishing vector field.) (v)
Show that S·(I) x ... x s·(n) is parallelizable provided that a(i) ~ 1. i = 1..... n and at least one a(i) is odd. (Hint: Use (iii) and Exercise 3.4C.)
~6.4
The Exterior Derivative, Interior Product, and Lie Derivative
Here we extend the differential of functions to a map d: nk(M) -t nk+I(M) for any k. This operator turns out to have marvelous algebraic properties. After studying these we shall show how d is related to the basic operations of div. grad. and curl on JR3. Then we develop formulas for the Lie derivative. We first develop the exterior derivative d for finite-dimensional manifolds. Later in the section the infinite-dimensional case is discussed. 6.4.1 Theorem Let M be an n-dimensional manifold. There is a unique family of mappings dk(U) : nk(U) -t nk+I(U) (k = O. 1.2 •...• n. and U is open in M). which were merely denote by d. called the exterior derivative on M. such that (i) d is a A-antiderivation. That is, d is JR linear andfor a E nk(U) and ~ E nt(U).
d(a A~) =da A ~ + (-Ita A d~ (product rule) (ii) If f E !F(U). then df is as defined in 4.2.5; (iii) d 2 = dod = 0 (that is, dk+I(U) 0 dk(U) = 0); (iv) d is natural with respect to restrictions; that is, if U eVe M are open and a
E
nk(V). then d(a I U) =(da) I U. or the following diagram commutes:
IU As usual, condition (iv) means that d is a local operator.
Proof We first establish uniqueness. Let (U.
=(x I•.... xn).
be a chart and
If k = O. the local formula da = (dajdxi)dx i applied to the coordinate functions xi, i = I, ...• n shows that the differential of x i is the one-form dxi. From (iii). d(dxi) = O. so by (i)
Chapter 6 Differential Forms
424
Thus. again by (i).
a<\ ... i ax'
k
..
da = - - .- dx'
A
dx"
.
A .•. A
dx\
(sum over i 1 < ... < ik ).
(I)
and so d is uniquely determined on V by properties (i)-(iii). and by (iv) on any open subset of M.
For existence. define on every chart (V.
and (i) is verified. For (iii). symmetry of the second partial derivatives shows that
Thus. in every chart (V.
E
nk(V). where VeE is open. Then k
dro(u)(vo •...• vk) = L(-I)iDro(U)·Vi(VO •...• Vi' ...• vk) i=O
(2)
where Vi denotes that Vi is deleted. Also. we denote elements (u. v) of TV merely by v for brevity. (Note that Dco(u)·v E Lk.(E.lR) since co: U --t Lk.(E.lR).)
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
425
Proof Since we are in the finite dimensional case. we can proceed with a coordinate computation. (An alternative is to check that d defined by (2) satisfies properties (i) to (iv). Checking (i) and (iii) is straightforward but lengthy.) Indeed. if the local coordinates of u are (x I•...• xn). ro(u) = roo
. (u)dx il /\ ... /\ dx~
'I' .. ··it
(sum over
i l < ... < ik). then
(where the sum is over all
and i l < ...
aroi'" i dro(vo •...• vk) = __1_._k dxj axJ
/\
dxil /\ ... /\ dxik(vO' .... vk) (2')
».
(where the sum is over all i l < ... < ik' j and a's with aU) < a(i l ) < ... < a(i k hand side of (2) is
The right
(2")
».
(where the sum is over i l < ... < ik' j. i. and Tl's with Tl(i l ) < ... < Tl(ik Writing a as a product of a permutation moving j to a designated position and a permutation Tl. we see that (2') and (2") coincide. _ If M is finite dimensional and a
E
nk(M) has the local expression a= ~1".ikdXik /\ ... /\
dxik • for i l < ... < ik. it is straightforward to check that the strict components of da are given by k a. . . . a.. . (d ) . . = ~ (-l)p-I" JI"'Jp-IJp+I"'lk+1 (_l)k'\~. a JI"'lI+1 £...J U J' + u" J' ,JI < ... <Jk+I' FI ax p ux~ 6.4.3 Examples A On R2. let a = f(x. y)dx + g(x. y)dy. Then da = df /\ dx + fd(dx) + dg /\ dy + gd(dy) by linearity and the product rule. Since d2 = O. af dx + ay af ag + ay-dy a g ) /\ dy. da = df /\ dx + dg /\ dy = ( ax dy) /\ dx + ( dx'dx
426
Since dx
Chapter 6 Differential Forms
1\
dx = 0 and dy 1\ dy = 0, this becomes
df
da = 'dYdy
1\
dg
dx + dX dx
1\
dy =
ax - dydf)dx
(dg
1\
dy.
B On lR,3, let f(x, y, z) be given. Then
df
df
=
df
df
dX dx + dy dy + dz'dz,
so the components of df are those of grad f. That is, (grad f)b =df, where b is the index lowering operator defined by the standard metric of lR,3 (see §5.1). C On JR.3, let fb = Fl(x, y, z)dx + F2 (x, y, z)dy + F3 (x, y, z)dz. Computing as in Example A yields .....h
Qt'-
I 3 (dF 2 - -dF I ) dx 1\ dy - (dF - -dF ) dx 1\ dz + (dF - 3 - -dF 2 ) dy 1\ dz. ~ dy ~ ~ dy ~
= -
Thus associated to each vector field G = Gli + Gzi + G3k on and to this the two-form *(Gb) by
JR.3
is the one-form Gb
where * is the Hodge operator (see §6.2); it is clear the dFb = *(curl F)b. D The divergence is obtained from d by d * fb
=(div F)dx 1\ dy 1\ dz;
i.e., * d * fb
=div F.
Thus, by associating to a vector field F on ntJ the one-form fb and the two-form d * Fb, gives rise to the operators curl F and div F. From dFb = *(curl F'I' it is apparent that ddfb =0 =d *(curl F)b
=(div curl F)dx 1\ dy 1\ dz.
That is, d 2 = 0 gives the well-known vector identity div curl F = O. Likewise, ddf = 0 becomes d(grad f)b = 0; i.e., *(curl grad f)b =O. So here d 2 = 0 becomes the identity curl grad f= O. We summarize the relationship between the operators in vector calculus and differential forms in the table on the next page.
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
427
VECTOR CALCULUS AND DIFFERENTIAL FORMS
1 Sharp and Flat (a) (b)
(Using standard coordinates in JR.3)
yb = yldx + y 2dy + y 3dz y3 e3 CI.#
= one fonn corresponding to the vector
=Cl.lel + Cl.2e2 + Cl.3e3 =vector corresponding to the one fonn
y
CI.
= ylel
+ y 2e2 +
=Cl.ldx + Cl.2dy +
Cl.3 dz
2 Hodge Star Operator
(Equation (6), §6.2)
(a)
* 1 = dx /\ dy /\ dz
(b)
*dx
(c)
*(dy /\ dz)
(d)
*(dx /\ dy /\ dz) = 1
=dy
1\
dz, *dy
=dx,
=-dx/\ dz, *dz =dx /\ dy dz) =-dy, *(dx dy) =dz
*(dx
1\
1\
3 Cross Product and Dot Product (a)
v x w = [*(v b /\ w b )]#
(b)
(v· w)dx /\ dy
1\
dz = vb
1\
*(w b )
4 Gradient Vf = grad f = (df)#
5 Divergence V· F = div F = *d(*f b )
6 Curl V x F = curl F = [*(dFb)]#
The effect of mappings on d can now be considered. Recall that Q(M) is the direct sum of all the Qk(M). Let F: M ~ N be a CI map. As F* : Qk(N) ~ Qk(M) is JR.-linear, it induces a mapping on the direct sums, F* : Q(N) ~ Q(M). Theorem Let F: M ~ N be of class CI. Then F*: Q(N) ~ Q(M) is a homomorphism of differential algebras; that is, (i) F*('I' 1\ c.o) = F*'I' /\ F*c.o, and (ii) d is natural with respect to mappings; that is, F*(dc.o) = d(F*c.o), or the following diagram commutes:
6.4.4
428
Chapter 6 Differentiol Forms
Proof Part (i) was established in 6.3.10. For (ii), we shall show that if m neighborhood V of me M such rhat d(F*co I V)
=(F*dco) I IV,
E
M, rhen there is a
which is sufficient, as F* and
d are both natural with respect to restriction. Let (V,
..
a
and so dco = a'o'CO;, "'i dx'o /\ ... /\ dx\ where a'o' = - . , k ax'o
(sum over io and i 1 < ... < ik)
and by (i) * * *. * . i)F dx" /\ ... /\ F dx\ F co I V = (F COi'" , k If 'V E QO(N) , rhen d(F*'V)
= F*d'V
by rhe composite mapping rheorem, so by (i),
* * *. *. * d(F CO I V) = F (dCO;, ... i) /\ F dx" /\ ... /\ F dx'k = F (dco) IV. 1
k
•
6.4.5 Corollary The operator d is natural with respect to push-forward by diffeomorphisms.
That is,
if F : M ~ N is a diffeomorphism. then F*dco = dF*co, or the following diagram
commutes: nk(M)
F*
..
Qk(N)
jd
dj
..
nk+l(M)
Qk+l(N)
F*
Proof Since F* = (F-l)*, the result follows from 6.4.4(ii). • 6.4.6 Corollary Let X E X(M). Then d is natural with respect to Lx. That is. for CO E W(M)
we have Lxco E W(M) and
dLxCO = Lxdco. or the following diagram commutes:
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
429
Proof Let Ft be the (local) flow of X. Then we know that d (F t*O)(m) Lxffi{m) = dt
I
.
1=0 ~
Since Ft*O) e nk(M). it follows that LxO) e nk(M). Now we have Ft*dO) = d(F/O). Then. since d is lR-linear. it commutes with dldt and so dLxO) = LxdO). • In Chapter 5 contractions of general tensor fields were studied. For differential forms. contractions playa special role. 6.4.7 Definition Let M be a TTUlnifold. X e X(M). and
0)
e n k+1(M). Then define ixO) e
tfOk(M) by
0) e nO(M). we put ixO) = O. We call ixO) the interior product or contraction of X and (Sometimes X J 0) is written for ixm.)
If
W.
6.4.8 Theorem We have ix: nk(M) -+ n k- 1(M). k = 1•...• n. and if a e n,,(M). P e n'(M). and f e nO(M). then (i) ix is a /\ -antiderivation; that is. ix is lR-linear and we have the identity ix(a/\ P) = (ixa ) /\ P + (-I;a/\ (ixP); (ii) (iii) (iv) (v)
ifXa = fixa; ixdf = Lxf; Lxa = ixda + dixa; LfXa = fLxa + df /\ ixa.
Proof That ixa e n k- 1(M) follows from the definitions. For (i). lR-linearity is clear. For the second part of (i). write
and k
ixa/\ P + (-I) a/\ ixP =
(k+t-I)! (k-t)!t!
A(ixa ® P) + (-I)
k(k+t-I)I . A(a ® ixP)· k!(t-I)!
Now write out the definition of A in terms of permutations from 6.1.1. The sum over all permutations in the last term can be replaced by the sum over 000' where 00 is the permutation (2. 3•...• k + I. 1. k + 2 •... k + t) .... 0.2.3 •...• k + t) whose sign is (_l)k. Hence (i) follows. For (ii). we note that a is .'I{M)-multilinear. and (iii) is just the definition of Lxf. For (iv) we proceed by induction on k. First note that for k = O. (iv) reduces to (iii). Now assume that (iv) holds for k. Then a (k + I)-form may be written as l:dfi /\ O)i' where O)i is a k
Chapter 6 DifferentWl Forms
430
fonn, in some neighborhood of mE M. But Lx(df /\
=Lxdf /\ 0)
0)
+ df /\ LxO) since Lx is a
tensor derivation and commutes with A (or from the definition of Lx in terms of flows), and so ixd(df /\
0)
+ diX
= -ixdf /\ dO) + df /\ ixdO) = df /\ LxO) + dLxf /\
+ dixdf /\ 0) + ixdf /\ dO) + df /\ dixffi
0)
by the inductive assumption and (iii). Since dLxf =Lxdf, the result follows. Finally for (v) we have
Note that the proofs of (i), (ii) and (iii) are valid without change on Banach manifolds. Fonnula (iv)
L,p. = ixda + dixa
(3)
(a "magic" fonnual of Cartan) is particularly useful. It can be used in the following way. 6.4.9 Examples A If a is a k-form such that da = 0 and X is a vector field such that dixa = 0, then Fl*a = a, where Fl is the flow of X. Indeed,
so Fl*a is constant in t. Since Fa = identity, Fl*a = a for all t. B Let M
so dix a
=1R3 ,
suppose div X = 0, and let a
= d * x£> = (div X)a = O.
= dx /\ dy /\ dz.
Thus da
Thus by Example A, Fl*(dx /\ dy /\ dz)
= O.
Also
= dx /\ dy /\ dz.
As
we shall see in the next section in a more general context, this means that the flow of X is volume preserving. Of course this can be proved directly as well (see for example Chorin and Marsden [1979]). For related applications to fluid mechanics, see §8.2.
•
The behavior of contractions under mappings is given by the following proposition. (The statement and proof also hold for Banach manifolds.) 6.4.10 Proposition Let M and N be manifolds and f: M ~ N a C I mapping. If X E X(N), Y E X(M), and Y is f-related to X, then
0)
E nk(N),
Section 6.4 The Ex/erWr Derivative, InierWr Product, and Lie Derivative
431
In particular, if f is a diffeomorphism, then
That is, inerior products are natural with respect to diffeomorphisms and the following diagram commutes:
Similarly for Y E 3£(M) we have the following commutative diagram: nk(M)
..
f.
nk(N)
,y)
)I,.y
.
nk-I(M) f.
Proof Let vI' ... , vk_1
E
nk-I(N)
Tm(M) and n = f(m). Then
iyc*ro(m).(vl' ... , vk_l ) = c*ro(m)·(Y(m), vI' ... , vk _ l )
= ro(n)·«Tf 0 Y)(m), Tf(v l ),
... ,Tf(v.,_I»
= ro(n)«X
Tf(vk _ l
0
f)(m), Tf(v l ),
••• ,
»
ixro(n)·(Tf(v I)' ... , Tf(vk _ I » c*ixro(m).(v l ,···, vk _ I)· • The next proposition expresses d in terms of the Lie derivatives (see Palais [1954]).
6.4.11 Proposition Let ~ E 3£(M), i =0, ... , k, and ro E nk(M). Then we have k
(i) (LXoro)(X 1, ... ,X.,) = LXo(ro(X 1, ... , X.,» -
~ro(XI' i=1
and
... , LxoX" ... , Xk )
432
Chapter 6 DjfferentioJ Forms k
(ii) d O)(X O' Xl' ...• X k) = L(-o'LX/(O)(X O•...•
X, ..... X k »
'=0 + "'" "'-' (-1) i+j O)(Lxj(Xj). XO' ...• Xl' ...• Xj' ...• X k ) A
A
OSi<jSk
where
X,
derwtes that X, is deleted.
Proof Part (i) is condition 001 in 5.3.1. For (ii) we proceed by induction. For k
is merely d ro(Xo) = LXoO). Assume the formula for k - 1. Then if
0) E
=O.
it
nk(M). we have.
by Cartan's formula (3) d O)(Xo. Xl' ...• X k) = (iXodO)(X l •...• X k)
= (~O)(XI' ...• X k) - (d(ixoO))(X l •...• X k) k
»- LO)(x\ •...• LxoXt ..... X k)
= LXo(ro(X\ ..... X k
t=\
(by (i». But iXoO) E nk-I(M) and we may apply the induction assumption. This gives. after a permutation
Substituting this into the foregoing yields the result. _ Note that the proof of (i) and of the first formula in the next corollary holds as well for infinite-dimensional manifolds. 6.4.12 Corollary Let X. Y E X(M). Then
In particular. ix 0
Lx =Lx i x· 0
Proof It is sufficient to check the first formula on any k-form E
X(U) for any open set U of M. We have by 6.4.11(i)
0) E
nk(U) and any Xl' ...• X k _ 1
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
(iyLxro)(XI' ...• Xk_l )
433
= (Lxro)(Y, XI' ... , Xk_ l ) k-I
= Lx(ro(Y, XI' .... Xk_ l
»-Lro(Y. XI' ... , [X, X,], ... , Xk-I) -
ro([X, Y], XI' ... , Xk - I)
t=1 k-I
=
»- L(iyro)(X I, ... , [X,
Lx«iyro)(X I, ... , Xk_I
X,], .... Xk _ l )
-
(i[X,Y]ro)(X I , ... , Xk _ l )
t=1
= (LXiyro)(X I , ... , Xk_l »- (i[X.Y]ro)(X I, ... , Xk_ I), One proves [Lx' Ly]
=L[X,Yl
using the first relation and Cartan's formula (3) . •
~ SUPPLEMENT 6.4A The Exterior Derivative on Infinite-dimensional Manifolds
Now we discuss the exterior derivative on infinite-dimensional manifolds. Theorem 6.4.1 is rather awkward, primarily because we cannot, without a lot of technicalities, pass from. for example, one-forms to two-forms by linear combinations of decomposable two-forms, i.e .. two-forms of the type a /\ 11 However, there is a simpler alternative available. 1. Adopt formula 6.4.11(ii) as the definition of d on any open subset of M. Note that at
first it is defined as a multilinear function on vector fields and note that Lx is already defined. 2. In charts, 6.4.11(ii) reduces to the local formula (2). This or a direct computation shows that d : nk(M) -+ nk+I(M) is well defined, depending only on the point values of the vector fields. 3. One checks the basic properties of d. This can be done in two ways: directly, using the local formula, or using the definition and the following lemma, easily deducible from the Hahn-Banach theorem: if a k-form ro is zero on any set XI' ... , Xk E X(U) for all open sets U in M, then ro == O. This second method is slightly faster if one first proves formula (3), which in tum implies 6.4.12. Proof of Formula (3) Let a be a k-form and XI' ... , Xk a set of k vector fields defined on some open subset of M. Writing Xo = X, (ixda + dixa)(X I , ... , Xk) = da(X, XI' ... , Xk) + d(ixa)(X I, ... , Xk) k
=
L(-o'Lx;(a(X o.... , X"~ t=1 + "" £J (-1) i+j a(Lx;Xj , OSi<j,;k
... , Xk» •...• Xo, ... , Xi' , Xj'
... , Xk)
Chapter 6 Differential Forms
434
k
=
»
LXo(a(X I ..... Xk +
L(-I~a(LxoXj' Xl'
...• Xj' ...• Xk)
j=l =
(Lxa)(X I •.••• Xk ) by 6.4.1100 . •
This and 6.4.12 will allow us to give a proof of the infinite-dimensional version of 6.4.6: Lxo d = d o Lx. For funetions f this formula is proved as follows. By 6.4.11(i). (LxdI)(Y) = Lx(df(Y» - df([X. Y]) = X[Y[f)] - [X. Y][f]
= Y[X[f)] = d(X[f)(Y) = (dLxl)(Y). Inductively. assume the formula holds for (k -I)-forms. Then for any k-form a and any vector field Y defined on an open subset of M. dLXiya = LXdiya. Thus by 6.4.12. iydLxa = LyLxa - diyLxa = LxLya - L[x,yla + di[X,y]a - dLxiya = LxLya - Lxdiya - i[x,Ylda = Lxiyda - ipc,Ylda = iyLxda. Hence d 0 Lx = Lx 0 d. Next. the remaining properties of d are checked in the following way. lR-linearity and 6.4.1(iv) are immediate consequences of the definition. For 6.4.1(ii). note that df(X) = ixdf = Lxf - dixf= Lxf= X[f]. To show that d2 = O. first observe that ix 0 dod = Lx 0 d - d 0 ix 0 d = d 0 Lx - d 0 Lx + dod 0 ix = dod 0 ix ' so that for any k-form a and any vector fields Xl' .... Xk +2 ' we have
The antiderivation property of d is proved by induction using (3) and the antiderivation property for the interior products. Finally. the formula r* 0 d = d 0 r* for a map f follows by definition and the properties r*(Lxm) = Lx(r*m). (r*m)(XI' .... Xk) = r*(m(X'1' .... X'k»' r*[X. Y] = [X'. Y'] if Xi -r X'. i = 1..... k. X -r X' and Y -r Y'.
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
435
Thus. with the preceding procedure. d is defined on Banach manifolds and satisfies all the key properties that it does in the finite-dimensional case. These key properties are summarized in Table 6.4.1 below.
tbJ
For vector-valued k-forms. we adopt. as in the preceding box. Palais' formula. 6.4.1(ii). as the definition on an open subset of M. Note again that this definition uses the fact that Lx is defined for vector-valued tensors. and again one has to prove that the local formula 6.4.2 holds. Then all properties in Table 6.4.1 are verified in the same manner as previously. For vector-valued forms we have however an additional formula on nk(M; F) doA=Aod for any A E L(F. P). If F is finite dimensional. the definition and properties of d become quite obvious; one notices that if n
c.o =
L c.o/j
E
nk(M; F). where c.oj
E
nk(M).
j=l then dc.o is given by n
dc.o= Ldc.o/j j=l and this formula can be taken as the definition of d in this case. This method does not work for vector-bundle-valued forms. Additional structure on the bundle is required to be able to lift Lx. Next we discuss the important concepts of closed and exact forms and the Poincare lemma. This lemma is a generalization and unification of two well-known facts in vector calculus: 1. if curl F = O. then locally F = 'Vf;
2. if div F = O. then locally F = curl G. 6.4.13 Definition We call c.o nk-1(M) such that c.o = da.
E
nk(M) closed if dc.o = O. and exact
if there is an a E
6.4.14 Theorem (i) Every exact form is closed. (ii) Poincare Lemma If c.o is closed, thenfor each mE M. there is a neighborhood U of m for which c.o I U E nk(U) is exact. Proof Part (i) is clear since dod = O. Using a local chart it is sufficient to consider the case c.o E nk(U). where U C E is a disk about 0 E E, to prove (ii). On U we construct an lR-linear mapping H : nk(U) --+ nk-l(U) such that doH + Hod is the identity on nk(U). This will give the result. for dc.o = 0 implies d(Hc.o) = c.o. For el' ...• ek
E
E. define
Chapter 6 DijJerenJiol Forms
436
Then by 6.4.2
i=1
~
= ",,-(-1)
i=1
i+1 rl k-I Jr t co(tu)(ei' el' ... , ei' ... , ek)dt A
0
k
I
",,~, ... , ek)dt. i=1 0 (The interchange of D and f is permissible, as co is smooth and bounded as a function of t E [0, 1).) However, we also have, by 6.4.2 + L(-I)i+li tkDco(tu)'ei(u, el'
Hdro(u)'(el' ... , ek) = Ia\kdro(tU)(U, el' ... , ek)dt I
=
k
I
f. tkD co(tu)'u(el' ... , ek)dt + L(-I)if. tkD ro(tu)'ei(u, el' ... , ei' ... , ek)dt. o i=1 0
Hence
There is another proof of the Poincare lemma based on the Lie transform method 5.4.7. It will help the reader master the proof of Darboux' theorem in §8.1 and is similar in spirit to the proof of Frobenius' theorem, 4.4.3.
Alternative Proof of the Poincare Lemma Again let U be a ball about 0 in E. Let, for t > 0, Ft(u) = tu. Thus Ft is a diffeomorphism and, starting at t = 1, is generated by the time-dependent vector field ~(u)
=u!t;
that is, F1(u) =u and dFt(u)/dt =~(Ft(u». Therefore, since co is closed,
For 0 < to :5 1, we get
Section 6.4 The Exterior Derivative, Interior Product, and Lk Derivative
Letting
to --+ 0,
we get
0)
437
= d~, where
Explicitly,
(Note that this fonnula for
~
agrees with that in the previous proof.) _
It is not true that closed forms are always exact (for example, on ]R2\ {(O, 0») or on a sphere). In fact, the quotient groups of closed fonns by exact fonns (called the deRham cohomology groups of M) are important algebraic objects attached to a manifold; they are discussed in §7.6. Below we shall prove that on smoothly contractible manifolds, closed fonns are always exact. Let r ~ 1. Two C' maps f, g : M --+ N are said to be (properly) Cr-homotopic, if there exists e> 0 and a Cr (proper) mapping F: ] - e,l + e[ x M --+ N such that F(O, m) = f(m), and F(l, m) = g(m) for all mE M. The manifold M is called
6.4.15 Definition
C'-contractihle if there exists a point mo E M and Cr-homotopy F of the constant map m 1-+ mo with the identity map of M; F is called a C'-contraction of M to mo. The following theorem represents a verification of the homotopy axiom for the deRham cohomology.
a E nk(N) a closed k-form (with compact support) on N. Then g*a - r*a E nk(M) is an exact k-form on M (with compact support). 6.4.16 Theorem Let f, g: M --+ N be two (properly) Cr-homotopic maps and
The proof is based on the following. 6.4.17 Deformation Lemma For a C'-manifold M let the C' mapping it: --+ ]-e, 1 + e[ x M
be given by it(m) =(t, m). Define H: (lk+I(] -e, 1 + e[ x M) --+ (lk(M) by
Then doH
0
d = i : - i ;.
Proof Since the flow of the vector field a/at E Xr(]-e, 1 + e[ x M) is given by FA.(s, m) = (s + A., m), i.e., is+A. = FA. 0 is, for any fonn 13 E nt(l-e, 1 + e[ x M) we get
Chapter 6 Differential Forms
438
'*L A Is a/iltf'
.* d
= Is dA.
I 1..=0
F*A d "y = dA.
I 1..=0
'*F*A d "y = dA.
Is
I
.* A
d '*A
Is+f..f' = dIs f'. 1..=0 S
Therefore, since the integrand in the formual for H is smooth, d and the intcgral sign commute, so that by Cart an's formula (3) and the above formula we get
Proof of Theorem 6.4.16 Define G
=H
0
F*, where H is given in the Deformation Lemma
6.4.17 and F is the (proper) homotopy between f and g. Since F* commutes with d we get doG + God = g* - r*, so that if the term a. E Qi«N) is closed (and has compact support), (g*
- r*) (a.) = d(Ga.) (and Ga. has compact support). _ 6.4.18 Poincare Lemma for Contractible Manifolds Any closed form on a smoothly contractible manifold is exact. Proof Apply the previous theorem with g = identity on M, and f(m) =
mo. _
The naturality of the exterior derivative has been investigated by Palais [1959]. He proves the following result. Let M be a connected paracompact n-manifold and assume that A : np(M) ~ nq(M)
is a linear operator commuting with pull-back, i.e.,
diffeomorphism cp: M
~
A
0
cp* = cp* 0 A for any
M. Then
0, if 0 $; P $; n, 0 < q $; n, q;t p, p+ I,
A =
1
a (Identity), if 0 < q = P $; n,
bd, if 0
$;
P $; n-l, q = p + I,
for some real constants a, b. If M is compact, then in addition we have
I
0, if q = 0, 0 < p < n c (Identity), if p = q = 0
A=
0, if q = 0, p = n, M is non-orientable or orientable and reversible,
df M' if q = 0, p = n,
M is orientable and non-reversible,
for some real constants c, d. (Orientability and reversibility will be defined in the next section whereas integration will be the subject of Chapter 7.)
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
~
439
SUPPLEMENT 6.4B
Differential Ideals and Pfaffian Systems This box discusses a refonnulation of the Frobenius theorem in tenns of differential ideals in the spirit of E. Cartan. Recall that a sub bundle E c: TM is called involutive if for all pairs (X, Y) of local sections of E defined on some open subset of M, the bracket [X, Y] is also a local section of E. The subbundle E is called integrable if at every point m
E
M there is a local
submanifold N of M such that TmN = Em' Frobenius' theorem states that E is integrable iff it is involutive (see §4.4). Before starting the general theory let us show by a simple example how fonns and involutive subbundles are interconnected. Let
0) E
02(M) and assume that Em = {v E TM I ivO) = o} is a
subbundle of TM. If X and Y are two sections of Em' then
Thus, Em is involutive iff ixiydO) = O. In particular, if
0)
is closed, then Em is involutive. In
this supplement we shall express conditions such as ixiydO) = 0 in tenns of one forms for subbundlcs E of TM that are not necessarily defined by exterior fonns. For any subbundle E, the k-annihilator of E defined by
is a subbundle of the bundle Ak(M) of k-fonns. Denote by r(U, E) the C~ sections of E over the open set V of M and notice that
is an ideal of O)(M); i.e., if I(E).
0)1'
Wz
E
I(E) and P E O(M), then
0)1
+ Wz
E
I(E) and P 1\
0)1 E
6.4.15 Proposition The subbundle E of TM is involutive iffor all open subsets V of M and all 0) E r(V, E 0(1», we have dO) E reV, E 0(2». If E is involutive, then 0) E reV, E o(k» implies dO) E r(V, E o(k + 1). Proof For any a
E
reV, E 0(1» and X, Y E reV, E), 6.4.11(ii) yields
da(X, Y)
=X[a(Y)]- Y[a(X)]- a([X, YD =- a([X, YD.
Thus E is involutive iff da(X, Y) = 0; i.e., da
E
reV, E 0(2» . •
440
Chapter 6 DifferentWl Forms
The Frobenius theorem in terms of differential forms takes the following form.
6.4.20 Corollary The subbundle E c: TM is integrable iffor all open subsets U of M, ro
E
nU, E°(\» implies dro E nU, EO(2». The following considerations are strictly finite dimensional. They can be generalized to infinite-dimensional manifolds under suitable splitting assumptions. We restrict ourselves to the finite-dimensional situation due to their importance in applications and for simplicity of presentation.
6.4.21 Definition Let M be an n-manifold and I c: OeM) be an ideal. We say that I is locally generated by n - k independent one-/orms, if every point of M has a neighborlwod U and nk pointwise linearly independent one-forms rol' ... , ron-k E OI(U) such that: n-k
(i) if ro E I, then ro I U =
L9
i 1\
roi for some 9i E OeM);
i=1
(ii) if ro E OeM) and M is covered by open sets U such that for each U in this cover, n-k
ro I U =
L 9i
1\
roi for some 9i E OeM),
i=1
then ro E I. The ideal I c: OeM) is called a differential ideal if dI c: I Finitely generated ideals of OeM) are characterized by being of the form I(E). More precisely, we have the following.
6.4.22 Proposition Let I be an ideal of OeM) and let n = dim(M). The ideal I is locally generated by n - k linearly independent one-forms iff there exists a subbundle E c: TM with k-dimensionalfiber such that 1= I(E). Moreover. the bundle E is uniquely determined by I Proof If E has k-dimensional fiber, let Xn_k+I' ... , ~ be a local basis of nU. E). Complete this to a basis of X(U) and let roi E OI(U) be the dual basis. Then clearly rol' ... , ron _ k are linearly independent and locally generate I(E). Conversely, let rol' ... , ron _k generate lover U and define Em = {v E TmM I roi(m)(v) = 0, i = 1, ... , n - k}. Em is clearly independent of the generators of lover U so that E = nmeMEm is a subbundle of TM. It is straightforward to check that I= I(E). Finally, E is unique since E *- E' implies I(E) *- I(E') by construction. _ Differential ideals are characterized among finitely generated ones by the following.
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
441
6.4.23 Proposition Let J be an ideal of O(M) locally generated by n - k linearly independl<1I1 forms 0>1' ...• O>n_k
E
01(U). n
= dim(M). and
let 0>1
O>n_k
A ... A
= 0> E
on-k(u). Then the
following are equivalent: (i) J is a differential ideal;
(ii) do> = Ij':.70>ij A O>j for some O>;j
E
01(U) and for every U as in the hypothesis;
(iii) dO>i A 0> =0 for all open sets U. as in the hypothesis; (iv) there exists
eE
01(U) such that do> =e A 0> for all open sets U. as in the
hypothesis.
Proof That conditions (i) and (ii) are equivalent and (ii) implies (iv) follows from the definitions. Condition (iv) means that n-k
L (-l)id
0>;
A
0>1
A ••• A
&; A
••• A
O>n-k =
eA
0>1
A ••• A
O>n-k.
i=1 so that multiplying by O>i we get (iii). Finally. we show that (iii) implies (ii). Let 0>1' ...• O>n 01(U) be a basis such that 0>1' ...• O>n_k generates J over U. Then
dO>i = L<XijtO>j j
A
o>t. where <Xijt
E
E
.'F(U).
But
o = d O>i A 0> =
L
<X;jtO>j A O>t
A
0>1
A ••• A
O>n-k
n-k<jd
and thus <Xijt
=0
for n - k <j <
t.
Hence
Assembling the preceding results. we get the following version of the Frobenius theorem. 6.4.24 Theorem Let M be an n-manifold and E c TM be a subbundle with k-dimensional jiber, and J(E) the associated ideal. Thefollowing are equivalent: (i) E is integrable;
(ii) E is involutive;
(iii) J(£) is a differential ideal locally generated by n - k linearly independent one-forms 0>1' ... , O>n_k E 01(U); (iv) for every point of M there exists an open set U and 0>1' ... , O)n_k generating J(E) such that
E
01(U)
Chapter 6 Differentiol Forms
442
n-k
dOlj =
L Cl>ij /\ Cl>j for some Cl>ij
E
I
n (U);
j=1 (v) same as (iv) but with the condition on Cl>i being: dCl>i /\
Cl1 /\ ... /\ Cl>n_k = 0;
(vi) same as (iv) but with the condition on Cl>i being: there exists
= e /\ CI>,
where CI>
eE
nl(U) such that dCl>
=Cl>1 /\ ... /\ Cl>n_k'
6.4.25 Examples A In classical texts (such as Cartan [1945) and Flanders [1963)), a system of equations
Cl>1 = 0, ... , Cl>n-k = 0, where Cl>i E nl(U) and U
C
]Rn
is called a Pfaffian system. A solution of this system is a k-dimensional submanifold N of U given by xi = xi(u I,
... ,
uk) such that if one plugs in these values of xi in the system, the result is
identically zero. Geometrically, this means that Cl>1' ... , Cl>n_k annihilate TN. Thus, finding solutions of the Pfaffian system reverts to finding integral manifolds of the subbundle E = {v E TV I Cl>i(v) = 0, i = 1, ... , n - k} for which Frobenius' theorem is applicable; thus we must have dCl>i /\ Cl>1 /\ ... /\ Cl>n_k = O. This condition is equivalent to the existence of smooth functions a jj , bj on U such that n-k Cl>i = Lajjdbj. j=1 To see this, recall that by the Frobenius theorem there are local coordinates bl' ... , bn on U such
= constant, ... , bn_k = constant, so that db i, i = 1, ... , n - k annihilate the tangent spaces to these submanifolds. Thus the ideal I generated by
that the integral manifolds of E are given by b l
db l ' ... , dbn_k ; i.e., Cl>i = Lj=l ..... n-k aij db j for some smooth functions a ij on U. B Let us analyze the case of one Pfaffian equation in ]R2. Let CI> =P(x, y)dx + Q(x, y)dy
E
nlo~2) using standard (x, y) coordinates. We seek a solution to CI> = O. This is equivalent to
dy/dx = -P(x, y)/Q(x, y), so existence and uniqueness of solutions for ordinary differential equations assures the local existence of a function f(x, y) such that f(x, y) = constant give thc integral curves y(x). In other words, f(x, y)
= constant is an integral manifold of
CI>
= O.
The
same result could have been obtained by means of the Frobenius theorem. Since dCl> /\ CI>
E
n3(]R2), we get dCl> /\ CI> =0, so integral manifolds exist and are unique. In texts on differential equations, this problem is often solved with the aid of integrating factors. More precisely, if CI> is not (locally) exact, can a function f and a function g, called an integrating factor, be found, such that gCl> = df? The answer is "yes" by 6.4.24(iii) for choosing f as above, E Thus g is found by solving the partial differential equation,
= ker df locally.
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative
443
This always has a solution and the connection between g and f is given by I df
g =
P
dX
f(x, y) = constant being the solution of ro = O. C Let us analyze a Pfaffian equation ro = 0 in jRn. As in 8, we would like to be able to
write gro = df with df;t 0 on U
C
jRn, for then f(xl, ... , xn) = constant gives the (n-
I)-dimensional integral manifolds; i.e., the bundle defined by ro is integrable. Conversely, if the bundle defined by ro is integrable, then by 8, gro = df. Integrability is (by the Frobenius theorem) equivalent to dro /\ ro = 0, which, as we have seen in 8, is always verified for n = 2. For n
~
3, however, this is a genuine condition. If n = 3, let ro = P(x, y, z)dx + Q(x, y, z)dy +
R(x, y, z)dz. Then droI\ro=[p(dR _
ay
dQ)+Q(dP _dR)+R(dQ _ dP)]dX/\dY/\dz; dZ dZ dX dX ay
so ro = 0 is integrable iff the term in the square brackets vanishes. D The Frobenius theorem is often used in overdetermined systems of partial differential
equations to answer the question of existence and uniqueness of solutions. Consider for instance the following system of A. Mayer [1872] in jRP-+
i=I, ... ,p,
a=I, ... ,q.
We ask whether there is a solution y = f(x, c) for any choice of initial conditions c such that f(O, c) = c. The system is equivalent to the following Pfaffian system:
Since the existence of a solution is equivalent to the existence of p-dimensional integral manifolds, Frobenius' theorem asserts that the existence and uniqueness is equvalent to dro u = ~=I ..... qroU13 /\ ro13 for some one-forms ro u13. A straightforward computation shows that U a j k dA u i i 13 dro = C jkdx /\ dx + --13- dx /\ ro , iJy
where a
C
jk =
iJA Uj iJx k
iJA U k
-
iJA Uj
13
iJA uk
13
a;;J + ay13 A k - ayll A J"
Chapter 6 Differential Forms
444
Since dxl •...• dx p • ro l •...• roq are a basis of nl(lRP+q). we see that dron = 1:13=1 •...• qron13 Thus the Mayer system is integrable
A
rob for some one-/orms ron13
iff
C"jk
=O.
iff
C"jk
= O.
•
In §S.4 and §S.S we shall give some applications of Frobenius' theorem to problems in constraints and control theory. Many of these applications may alternatively be understood in terms ofPfaffian systems; see for example. Hermann [1977. Ch. 181.
b
IDENTITIES FOR VECTOR FIELDS AND FORMS 1. Vector fields on M with the bracket [X. Yl form a Lie algebra; that is. [X. Yl is real bilinear. skew symmetric. and Jacobi's identity holds: [[X. Yl. Zl + [[Z. Xl. Yl + [[Yo Zl. Xl = O. Locally. [X. Yl = DY·X-DX·Y and on functions. [X. Y][f] = X[Y[f]l - Y[X[f]l. 2. For diffeomorphisms <1>. ",. *[X. Yl = [*X. <1>* Yl and (
0
",)*X = *"'*X.
3. The forms on a manifold are a real associative algebra with A as multiplication. Furthermore. a A 13 = (-lf l l3 A a for k and t-forms a and 13. respectively. 4. For maps <1>. ",. *(a A 13) = *a A <1>*13. ( 0 ",)*a = "'**a. 5. d is a real linear map on forms and dda = O. d(a A 13) = da A 13 + (-If a 6. For a a k-form and
"0..... Xk
A
dl3 for a a k-form.
vcctor fields:
k
da(X o..... Xk)= L(-I)iXi[a(x o..... Xi' .... Xk)l i=O
+
L
(-l)i+ja([x i• Xjl. Xo .....
~ .....
Xj' .... X k)
OSi<jSk
If M is finite dimensional and a = <Xj 1'" iIt.dx i1
A ••• A
dx\ i l < ... < ik • then
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivative k
(da)j j
I'" k+1
=
L
p-I
(-0
p=1
a
- . a J· a)k
X
I
... )·
)'
p-I p+1
"')'
)k
hi
a
+ (-1 . - aJ· ... )· a h+1 I k
•
445
for jl<"'<jk+I'
X
Locally. k
dro(x)(vo •...• Vk) = L(-I )iDro(X)'Vi(VO' ...• vi' ...• Vk)' i=O
7. 8.
For a map
9.
ixa is real bilinear in X. a and for h : M -+ JR. ihXa = hixa = ixha. Also ixixa = O. and
for a a k-form. 10. For a diffeomorphism
Locally. k
(Lxa)x'(vi' ...• vk) = Dax'X(x)'(vl' ...• vk) + Lax(v! •...• DXx·vj •...• vk)' i=1
15. The following identities hold: (a) LfXa = fLxa + df /\ ixa (b) L[X.y]a = LxLya - LyLxa
(c) i[X,y]a=LXiya-iyLxa (d) Lxda = dLxa (e) Lxixn = ixLxa. 16. If M is a fmite dimensional manifold. X = Xta/dxt and a = ai,'" ikdxi, /\ .. '/\ dx\ i, < ... < ik • the following local formulas hold:
Chapter 6 DijJerentioJ Forms
446
da
Exercises 6.4A Compute the exterior derivative of the following differential forms on
a
= x3dx + y3dx /\ dy + xyzdx /\ dz;
~
=3exdx /\ dy + 8cos(xy)dx /\ dy /\ dz.
]R3:
6.48 Using 6.4.3 and the properties of d and *. prove the following formulas in ]R3 --t]R and F. G E *(]R3): (i) (ii)
grad(fg) = (grad f)g + f(grad g) curl(fF) = (grad f) x F + f(curl F)
(iii)
div(fF)
(iv) (v) (vi)
div(F x G) = G·curl F - F ·curl G L ..G = (F -V)G - (G·V)F = F div G - G div F - curl(F x G) curl(F x G) =(div G)F - (div F)G + (G ·V)F - (F ·V)G
(vii)
curl(curl F)
]R3
for f. g :
=grad(f) ·F + f div F
= grad(div F) -
~F. whcre (~F)i
= a 2p/ax2 + a 2p/ay 2 + a 2Fi/az2
is the usual Laplacian (viii)
V(F·F) = 2(F·V)F + 2F x curl F
6.4C
Let
6.4D
On SI find a closed one-form (y dx - X dy)/(x 2 + y2).)
a
that is not exact. (Hint: On
]R2\
{OJ consider
a
=
6.4E Show that the following properties uniquely characterize ix on finite-dimensional manifolds
Section 6.4 The Exterior Derivative, Interior Product, and Lie Derivath'e
447
(i) ix: Qk(M) --> Qk~I(M) is a /\ antiderivation. (ii) ixf = 0 for f E ,'RM); (iii) ixw = w(X) for WE QI(M); (iv) ix is natural with respect to restrictions. Usc these properties to show i[X,YI = LXiy - iyLx. Finally, show ix
0
ix = O.
6.4F Show that a derivation mapping Qk(M) to Qk+I(M) for all k ~ 0 is zero (note that d and ix are antiderivations).
6.4G Let s: T2M --> T2M be the canonical involution of the second tangent bundle (see exercise 3.4D). (i)
If X is a vector field on M, show that s 0 TX is a vector field on TM.
(ii)
If F t is the flow of X, prove that TFt is a flow on TM generated by sO TX.
(iii)
If I-l is a one-form on M, I-l': TM --> TIt is the corresponding function, and w E T2M, then show that dl-l'(sw)
= dl-l(tTM (w), TtM(w)) + dl-l'(w).
6.4H Prove the following relative Poincare lemma. Let and let N
C
W
be a closed k-form on a manifold M
M be a closed submanifold. Assume that the pull-back of
Then there is a (k - 1)-form a vanishes on N. If
w to N is zero. w and a
on a neighborhood of N such that da =
w vanishes on N, then a can be chosen so that all its first partial
derivaties vanish on N. (Hint: Let S I; X 1-+ e ix be the exponential map. Show that 'Y induces a isomorphism TS I ~ SI X TIt. Let M be a manifold and let w be an "angular varbblc," that is a smooth map w: M --> SI. Define dw, a one-form on M by taking the TIt-projection of Tw. Show that (i) if w: M --> S I, then ddw is smooth, then r*(dw) = d(r*w), where r*w =
w
0
= 0; and (ii) if
f: M --> N
f.
6.4J Prove the identity
6.4K
(i)
Let X
= (Xl,
X2, 0) be a vector field defined on the plane S
TIt} in TIt 3. Show that there exists Y
E
= ((x, y, 0) I x, y E
XCR3) such that X = curl Y on S. (Hint:
Let Vex, y, z) = (zX2(x, y), -zXI(x, y), 0).) (ii)
XeS). Show that there exists Y E X(TIt 3) such that X = curl Y on S. (Hint: By 5.5.9 extend X to X E X(TIt3) and put (j) = * XE>. Locally find a such that da = w by (i). Use a partition of unity {
E
(iii) Generalize this to forms on a closed submanifold of a manifold admiting Ck-partitions
Chapter 6 Differential Fonns
448
of unity. 6.4L
E Qk(M). If 't M : TM ~ M denotes the tangent bundle Qk(TM). A k-form r on TM for which there is an a E
Let M be a manifold and a projection, let a'
='t*a E
nk(M) such that a' = r is called basic. A vector field x T'tM 0 X = O. Show that r
E
E
X(TM) is said to be vertical if
Qk(TM) is basic if and only if ixr = 0, Lx r = 0 for any
vertical vector field X on TM. Conclude that if r is closed it is basic if and only if ixr = 0 for every vertical vector field X on TM. (Hint: Since X and the zero vector field on
M are 'tM-related, if r is vertical, the two identities follow. Conversely, if f t is the flow = r. Define a E Qk(TM) by a(m)(vl' ... , vk) = r(u)(V 1 ' ... , V k),
of X, then Ft*r
where 't(u) = m, Tm'tM(V i) = Vi' i = 1, ... , k. Show that this definition is independent of the choices of u, VI' ... , Vk in the following way. Let 't(u') = m, Tm'tM(V() = Vi' i = I, ... , k, w =u - u'. Consider the local flow F t in a vector bundle chart of TM containing T mM which occurs only in the fibers and which on T mM itself is translation by tw. The vector field it generates is vertical so that Tt*r = r and F1(u) = u'. Let T u' F1(V i') = V"i E Tu(TM) and show Tu't(V"i) = Vi since 't 0
~6.S
Orientation, Volume Elements, and the Codifferential
This section globalizes the setting of §6.2 from linear spaces to manifolds. All manifolds in this section are finite dimensional. (For infinite-dimensional analogues of orientability, see Elworthy and Tromba [1970bj.) 6.5.1 Definition A volume/orm on an n-manifold M is an n{orm
~
E Q"(M) such that
~(m)
i= 0 for all mE M; M is called orlentable if there exists some volume form on M. The pair (M, ~)
is called a volume manifold. Thus, ~ assigns an orientation, as defined in 6.2.5, to each fiber of TM. For example, :1'
has the standard volume form
~
= dx 1\ dy
1\
dz.
6.5.2 Proposition Let M be a connected n-manifold. (i)
M is orientable iff there is an element may be written
(ii)
v
=f~ for some
f
E
~
E Q"(M) such that every other v E Q"(M)
.'J(M).
If M is orientable then it has an atlas {(Uj' cp)}. where CPj: Uj ~ U{ c ;R.", such that the Jacobian determinant of the overlap maps is positive (the Jacobian determinant being the determinant of the derivative, a linear mapfrom ]R" into ]R"). The converse is true if M is paracompact.
Proof For (i) assume first that M is orientable, with a volume form ~. Let v be any other element of Qn(M). Now each fiber of A"(M) is one-dimensional, so we may define a map f: M ~]R
by v '(m) = f(m)~'(m) where
and similarly for v'. Since
~'(m)
*" 0
Conversely, if Q"(M) is generated by
~(m)
= ~'(m)dxl
for all m E M, f(m) ~,
then
~'(m)
*" 0
1\ ... 1\
dx"
= v'(m)/~'(m)
is of class
C~.
for all m E M since each fiber is
one-dimensional. To prove (ii), let {(Uj' CPj)} be an atlas with CPj(U) = U{ c ]R". We may assume that all U{ are connected by taking restrictions if necessary. Now cpt~ = fjdx 1
1\ ... 1\
dx" = fj~O'
where ~o is the standard volume form on ]R". By means of a reflection if necessary, we may assume that fj(u') > 0 (fj *" 0 since ~ is a volume form). However, a continuous real-valued function on a connected space that is not zero is always > 0 or always < O. Hence, for overlap maps we have
450
Chapter 6 Differential F onns
But \jI*(u)(a l /\ ... /\ an) = D\jI(u)*·a l /\ D\jI(u)*·a 2 /\
... /\
D\jI(u)*.an•
where D\jI(u)* ·al(e) = al(D\jI(u) ·e). Hence. by definition of determinant.
(We leave as an exercise the fact that the canonical isomorphism L(E. E) = L(E*, E*). used before docs not affect determinants.) For the converse of (ii). suppose {(Va' \jIa)} is an atlas with the given property. and let {(Ui'
=
= ~(m)IlJm) if mE Ui and ~i = 0 if m i" Ui' Since supp gi cUi' we have ~i E nn(M). Let 11 = l:i~i' Since this sum is finite in some neighborhood of each point. it is clear
and let ~Jm)
from local representatives that 11 E nn(M). To show that 11 is a volume form on M. notice that on Ui n Uj "" 0 we have Ilj =
0
0
0
l /\ ... /\ dx n) = [det D(
0
where aji E .'T(Ui n U). aji > O. By local finiteness of the covering (Ui). a given point mE M belongs only to a finite number of open sets Uio ' U i, •...• UiN' Thus
since Ilio(m) "" 0 and each
~kio(m)
> O. It follows that Il(m) "" 0 for each mE M . •
Thus. if (M. 11) is a volume manifold we get a map from nn(M) to .'T(M); namely. for each v E nn(M). there is a unique f E JiM) such that v
= fll.
6.5.3 Definition Let M be an orientable manifold. Two volume forms III and Ilz on Mare
called equivalent if there is an f E JiM) with f(m) > 0 for all mE M such that III = f1l 2 . (This is clearly an equivalence relation.) An orientation of M is an equivalence class [Ill of volume
Section 6.5 Orientation, Volume Elements, and CodjJJerentiol
451
forms on M. An oriented manifold (M, [11]), is an orientable manifold M together with an orientation [Ill on M. If [Ill is an orientation of M, then [-Ill, (which is clearly another orientation) is called the reverse orientation. The next proposition tells us when [Ill and [-Ill are the only two orientations.
6.5.4 Proposition Let M be an orientable manifold. Then M is connected iff M has exactly two orientations. Proof Suppose M is connected, and 11, v are two volwne forms with v
=fll.
Since M is
connected, and f(m)"* 0 for all mE M, f(m) > 0 for all m or else f(m) < 0 for all m. Thus v is equivalent to 11 or -11. Conversely, if M is not connected,let U"* 0 or M, be a subset that is both open and closed. If 11 is a volume form on M, define v by letting v(m) = Il(m) if m E U and v(m) = -Il(m) if m Ii! U. Obviously, v is a volwne form on M, and v Ii! [Ill U [-Ill. _ A simple example of a nonorientable manifold is the Mobius band (see Figure 6.5.1 and Exercise 6.SL). For other examples, see Exercises 6.SK, M.
Figure 6.5.1
6.5.5 Proposition The equivalence relation in 6.5.3 is natural with respect to mappings and diffeomorphisms. That is, if f: M ~ N is of class C~, IlN and v N are equivalent volume forms on N, and r*(IlN) is a volume form on M, then r*(v N ) is an equivalent volume form. If f is a diffeomorphism and 11M and v M are equivalent volume forms on M, then f*(IlM) and f*(v M) are equivalent volume forms on N. Proof This follows from the fact that r*(gm) = (g
0
f)r*m, which implies r*(gm) = (g 0 [-I)r*m
when f is a diffeomorphism. _
6.5.6 Definition Let M be an orientable manifold with orientation [Ill. A chart (U,
452
Chapter 6 DifferrmtioJ Farms
form
dxl" ... " dxD E nn(u'). From 6.5.5 we see mat mis definition does not depend on me choice of me representativc
from
[~].
If M is orientable. we can find an atlas in which every chart has positive orientation by choosing an atlas of connected charts and. if a chart has negative orientation. by composing it with a reflection. Thus. in 6.S.2(ii). me atlas consists of positively oriented charts. If M is not orientable. mere is an orientable manifold M and a 2-to-l COO surjective local diffeomorphism It: M -+ M. The manifold M is called me orientable double covering and is useful for reducing certain facts to the orientable case. The double covering is constructed as follows. Let M= {(m. [~m]) I m E M. [~m] an orientation of TmM}. Define a chart at (m. [~m]) in me following way. Fix an orientation [co] of RD and an orientation reversing isomorphism A of RD. e.g .• the isomorphism given by A(e l ) =-el' A(e) = e j • i = 2 •...• n. where {el' ...• eo} is me standard basis of JRD. If
453
Section 6.5 Orientation, Volume Elements, and Codifjerentiol
codimension k with trivial normal bundle. That is. there are C"" maps Ni : V ~ TM. i
= 1•...•
k such that Ni(v) E Tv(M). and Ni(v) span a subspace Wv such that TvM = Tv V EB Wv for all v E V. Then V is orientable. Proof Let Il be a volume fonn on M. Consider the restriction III V : V ~ An(M). Let us first note that Il I V is a smooth mapping of manifolds. This follows by using charts with the submanifold property. where the local representation is a restriction to a subspace. Now define Ilo : V ~ An-k(v) as follows. For Xi' ... , Xn-k E ~(V), put
It is clear that Ilo(v) cF- 0 for all v. It remains only to show that Ilo is smooth, but this follows from the fact that III V is smooth. _ If g is a Riemannian metric, then
we write g# = (
. .. af (grad t)1 = gIJ_.. axJ
(1)
6.5.9 Corollary Suppose M is an orientable paracompact manifold. H E .'RM) and c E IR is a regular value of H. Then V = H-I(c) is an orientable submanifold of M of codimension one. if it is nonempty. Proof Suppose c is a regular value of H and H-I(c) = V cF- 0. Then V is a submanifold of codimension one. Let g be a Riemannian metric of M and N = grad(H) I V. Then N(v) e Tv V for v E V, because Tv V is the kernel of dH(v), and dH(v)[N(v») =g(N, N)(v) > 0 as dH(v) cFo by hypothesis. Then 6.5.8 applies, and so V is orientable. _ Thus if we interpret V as the "energy surface," we see that it is an oriented submanifold for "almost all" energy values by Sard's theorem. 6.5.10 Definition Let M and N be two orientable n-manifolds with volume forms IlM and IlN' respectively. Then we call a C~ map f: M ~ N volume preserving (with respect to IlM and IlN) if r*IlN E
=Ilw
orientation preserving if r*{IlN) E [IlM], and orientation reversing if r* (IlN)
[-IlM)' An orientable manifold admitting (not admitting) an orientation reversing
diffeomorphism is called reversible (respectively non-reversible).
From 6.5.5, [r*IlN) depends only on [IlN)' Thus the first part of the definition explicitly depends on IlM and IlN' while the last four parts depend only on the orientations [IlMl and [IlN)'
Chapter 6 DifferentUd Forms
454
Furthermore, we see from 6.5.5 that if f is volume preserving with respect to 11M and I1N' then f is volume preserving with respect to hl1M and gl1N iff h
=g
f. It is also clear that if f is
0
volume preserving with respect to 11M and 11N' then f is orientation preserving with respect to [11M) and [I1N)' 6.5.11 Proposition Let M and N be n-manifolds with volwne forms 11M and I1N' respectively.
Suppose f: M ~ N is of class C~. Then r* (I1N) is a volume form iff f is a local diffeomorphism; that is,for each m
E
M, there is a neighborhood V of m such that f I V : V
~
feY) is a diffeomorphism. If M is connected, then f is a local diffeomorphism iff f is
orientation preserving or orientation reversing. Proof If f is a local diffeomorphism, then clearly r*(I1N)(m)
Tc
0, by 6.2.3(ii). Conversely, if
r*(I1N) is a volume form, then the determinant of the derivative of the local representative is not zero, and hence the derivative is an isomorphism. The result then follows by the inverse function theorem. The second statement follows at once from the first and 6.5.4. • Next we consider the global analog of the determinant. 6.5.12 Definition Suppose M and N are n-manifolds with volume forms 11M and I1N'
respectively. If f: M
~
N is of class
C~,
the unique
C~
function J(I1M' I1N)f E .'r(M) SItch
that r*I1N = (J(I1M' I1N)f)I1M is called the Jacobian determinant of f (with respect to 11M and I1N)' If f: M ~ M we write J/ =J(I1, l1)f. Note that J(I1M' I1N)f(m) = det(Tmf), the determinant being taken with respect to the volume forms I1 M(m) on TmM and I1N(f(m» on Tf(m)N. The basic properties of determinants that were developed in §6.2 also hold in the global case, as follows.
First, we have the following
consequences of 6.5.11. 6.5.13 Proposition f is a local diffeomorphism iff J(I1M' I1N)f(m)
Tc
0 for all mE M.
Second, we have consequences of the definition and properties of pull-back. 6.5.14 Proposition Let (M,I1) be a volume manifold. (i) If f: M
~
M and g: M
~
Mare of class
(ii) If h : M ~ M is the identity, then Jf1h = 1. (iii) If f: M ~ M is a diffeomorphism, then
C~,
then
Section 6.5 Omntation, Volume Elements, and CodijJerentiaJ
455
Proof For (i),
Part (ii) follows since h* is the identity. For (iii) we have
6.5.15 Proposition Let (M, [11M]) and (N, [11:--;]) be oriented manifolds and f: M ---t N be of
class C"'. Then f is orientation preserving iff J(llw IlN)f(m) > 0 for all m E M, and orientation reversing if J(IlM' IlN)f(m) < 0 for all mE M. Also, f is volwne preserving with respect to 11\1 and IlN iff J(llw IlN)f = 1. This proposition follows from the definitions. Note that the first two assertions depend only on the orientations [11M) and [IlN) since
which the reader can easily check. Here g E .'T(N) , h E .'T(M) , g(n)
* 0, and
hem)
*0
for all n
EN, mE M.
We have seen that in !R 3 the divergence of a vector field is expressible in terms of the standard volume clement 11 =dx /\ dy /\ dz by the use of the metric in !R3 (see Example 6.4.3D). There is, however, a second characterization of the divergence that does not require a metric but only a volume form 11, namely
as a simple computation shows. This can now be generalized. 6.5.16 Definition Let (M, 11) be a volume manifold; i.e., M is an orientable manifold with a
volwne fonn 11. Let X be a vector field on M. The unique function div).LX E X(M), such that Lxll = (div).LX)1l is called the divergence of X. We say X is incompressible (with respect to 11)
if div).LX = O. 6.5.17 Proposition Let (M,Il) be a volwne manifold and X a vector field on M. (i) If f
E
.'T(M) and f(m)
* 0 for all
m E M, then
Chapter 6 Differential Forms
456
Proof Since
Lx
is a derivation.
As fl1 is a volume fonn. (div f j!X)(fl1) = (Lx f)11 + f(divj!X)I1. Then (i) follows. For (ii). Lgxl1 = gLxl1 + dg /\ i x l1. and from the antiderivation property of ix ' dg /\ ixl1 =- ix(dg /\ 11) + i xdg /\ 11. But dg /\ 11 E nn+l(M). and hence dg /\ 11 = O. Also. ixdg = Lxg and so Lgxl1 = gLxl1 + (L x g)l1. The result follows from this. • 6.5.18 Proposition Let (M. 11) be a volume manifold and X a vector field on M. Then X is incompressible (with respect to 11) if the flow of X is volume preserving: that is, the local diffeomorphism Ft : U --+ V is volume preserving with respect to 111 U and 111 V. Proof Since X is incompressible, Lxl1 = 0, and so 11 is constant along the flow of X; l1(m) = (Ft*l1)(m). Thus Ft is volume preserving. Conversely, if (Ft*I1)(m) = l1(m), then Lxl1 = O.• 6.5.19 CoroJlary Let (M,I1) be a volume manifold and X a vector field with flow Ft on M. Then X is incompressible iff Jj!Ft = 1 for all t E JR. The considerations regarding the Jacobian and the divergence can also be carried out for one-densities. If 111M I, II1N I are one-densities and f: M --+ N is C-, we shall write r*II1N I = J( 111M I, II1N I, f) 111M I, where the pull back is defined as for fonns. Then 6.5.13 and 6.5.14 go through for one-densities. The Lie derivative of a one-density is defined by
and one defines the divergence of X with respect to 1111 as in 6.5.16. Then it is easy to check that 6.5.17-6.15.19 have analogues for one-densities. We shall now globalize the concepts pertaining to Riemannian volume fonns and densilies. as well as the Hodge star operator discussed in §6.2. 6.5.20 Proposition Let (M, g) be a pseudo-Riemannian manifold of signature s; i.e., g(m) has signature s for all m E M. (i) If Mis orientable, then there exists a unique volume element 11 = l1(g) on M, called the g-volume (or pseudo-Riemannian volume of g), such that 11 equals one on all positively
Section 6.5 OrientoJion, Volume Elements, and Codifferential
457
oriented orthofUJrmal bases on the tangent spaces to M. If XI' .... Xn is such a basis in an open set U of M and if ~I ..... ~n is the dual basis, then Il vI' .... vn
E
=~I /\ ... /\ ~n.
More generally. if
TxM are positively oriented. then ll(x)(vl' .... vn) = I det[g(x)(v i • vj )lll!2.
a E JR. there exists a unique a-density I ilia. called the g-a-density (or the
(ii) For every
pseudo-Riemannian a-density of g). such that Illl a equals 1 on all orthonormal bases of the tangent spaces to M. If Xl' .... Xn is such a basis in an open set U of M with dual basis ~ I. .... ~n. then Illl a = I ~I/\ ... /\ ~n la. More generally, if vI' .... vn E TxM. then
The proof is straightforward from 6.2.10 and the fact that Il and Illl a are smooth. Also note that in an oriented chart (x I.... , xn) on M, we have Il = I det[gij lll!2 dx l /\ ... /\ dxn. As in the vector space situation. g induces a pseudo-Riemannian metric on Ak(M) by
(2) where a. ~
E
Ak(M)x and ~il"'ik are the components of the associated contravariant tensor to ~
g(x). As in 6.2.11. if XI' .... ~ is an orthonormal basis in U eM with dual basis ~I •...• ~n. then the elements ~il
/\ ... /\
~\ where i l < ... < ik form an orthonormal basis of A\U).
The Hodge operator is defined pointwise on an orientable pseudo-Riemannian manifold with pseudo-Riemannian volume form Il by *: d«M) ~ nn-k(M). (*a)(x) = *a(x). i.e .• a /\ *~ =
E
nk(M). The properties in Propositions 6.2.12 and 6.2.13 carry over since
they hold pointwise. The exterior derivative and the Hodge star operator enable us to introduce the following linear operator. (The reason for the strange-looking factor (-1) in the definition is so a later integration by parts formula. proved in 7.2.13. will come out simple.)
6.5.21 Definition The codifferential 0: nk-I(M) ~ nk(M) is defined by o(no(M)) = 0 and on k+ 1 forms ~ by o~ = (_I)nk+l+lnd(g) *d*~ Notice that since d 2 =0 and ** is a multiple of the identity. 02 =O. For example. in JR.3. let a = ady /\ dz - bdz /\ dz + cdx /\ dy. Then *a =adx + bdy + cdz. so and
Chapter 6 DifferentWl Forms
458
Thus. as nk + I + Ind(g) = 4 is even.
The formula for
I)~
in coordinates is given by
1
k+1 _I_I det[g.111/2g . ... g. ~ [ "" (-OPg'd, ... g'~,jp-lgtirg'p1>" ... g'kik,'A. . 1 det[g .. 11 112 k+ I 1) 1,', Ik'k t L..t f'lI···h" I)
ax
p=1
or as a contravariant tensor
where ~ = ~, ... , ' 1
k hl
dx" /\ ... /\ dx'k~.
(sum over r l < ... < rk + l ) is the usual coordinate expression for
~.
Formula (3) is messy to
prove directly. However it follows fairly readily from integration by parts in local coordinates and the fact that I) is the adjoint of d. a fact that will be proved in Chapter 7 (see 7.2.13 and Exercise
7.5G). We now shall express the divergence of a vector field X E X(M) in terms of
I).
We define
the divergence div g(X) of X with respect to a pseudo-Riemannian metric g to be the divergence of X with respect to the pseudo-Riemannian volume 11 = Il(g) of g; i.e .• Lx ll = divg(X)Il. To compute the divergence. we prepare a lemma.
6.5.22 Lemma ixll =
* xU.
Proof At points where X vanishes this relation is trivial. So let X(x) *" 0 and choose v2' ... , Vo E T"M such that {X(x)/g(X, X)(x). v2..... vol is a positively oriented orthonormal basis of T"M. Then by 6.2.14,
This may also be seen in coordinates using formula (12) of §6.2.
6.5.23 Proposition Let g be a pseudo-Riemannian metric on the orienta table n-manifold M. Then div /X) = -oxU. (4)
Section 6.5 Orientation, Volume Elements, and Codifferential
459
In (positively oriented) local coordinates (5)
Proof Let Lx ll = dixll = d * xf' by the lemma. But then (div g X)1l the definition of I) and the fonnula for **. Since *1
=11,
=- * I)xf' =_(I)Xb) * 1
by
we get (4). To prove fonnula (5),
write 11 = 1det[~j111/2dxl /\ ... /\ dx n and compute Lxll = dixll in these coordinates. We have A
ixll = 1det[gi!2Xk(-I)k dx l /\ ... /\ dxk /\ ... /\ dxn and so
k) 1
. a 1det [gij 111/2X dx /\ ... /\ dx n dlxll = (ai a
1
----:-I!2~ --k (I det[gij11
1det [gij11
1/2
k
X )11·
•
ax
6.5.24 Definition The Laplace-Beltrami operator onfunctions on a orientable pseudo-Riemannian
manifold is defined by V2 =div
0
grad.
Thus in a positively oriented chart, equation (5) gives af) V2 f=ldet[g·ll -1/2 - a ( g 'k Idet[gll 112 . 1J axk 1J ax'
(6)
Exercises 6.5A Let f: JR.n
~
JR.n be a diffeomorphism with positive Jacobian and f(O) = O. Prove that there
is a continuous curve ft of diffcomorphismsjoning f to the identity. (Hint: First join f to Df(O) by gt(x) = f(tx)/t.) 6.58 If t is a tensor density of M, that is, t = t' ® 11, where 11 is a volume fonn, show that
6.5C A map A: E ~ E is said to be derived from a variational principle if there is a function L: E
~
JR. such that dL(x)·v = (A(x), v),
where (,) is an inner product on E. Prove Vainberg's theorem: A comes from a variational principle if and only if DA(x) is a symmetric linear operator. Do this by applying the Poincare lemma to the one-fonn a(x)· v = (A(x) ,v) (see Hughes and
460
Chapter 6 Differential Forms
Marsden [1977]). 6.SD Show in three different ways that the sphere
sn
is orientable by using 6.5.2 and the two
charts given in 3.1.2, by constructing an explicit n-form, and by using 6.5.9. 6.SE Use formula (6) to show that in polar coordinates (r, e) in ]R2,
and that in spherical coordinates (p,
where Il
e,
in ]R3,
=cos <po
6.SF Let (M, Il) be a volume manifold. Prove the identity
6.SG Let f: M
~
N be a diffeomorphism of connected oriented manifolds WIth boundary.
Assuming that T mf: T mM
~
Tf(m)N is orientation preserving for some m
E
Int(M), show
that J(f) > 0 on M; i.e., f is orientation preserving. 6.SH Let g be a pseudo-Riemannian metric on M and define g.. = Ag for A> O. Let *.. be the Hodge-star operator defined by g.. and set *1 A(n/2)-k
= *.
Show that if a
E
Qk(M), *.. a
=
* a.
6.51 In]R3 equipped with the standard Euclidean metric show that for any vector field F and any function f we have: div F
= -liP>,
curl F = (1i*P»#, and grad f = -(*Ii*f)#.
6.5.1 Show that if M and N are orientab1e, then so is M x N. 6.SK (i) Let 0: M
~
M be an involution of M, Le., 000 = identity, and assume that the
equivalence relation defined by 0 is regular, i.e., there exists a surjective submersion 1t: M ~ N such that 1t- 1(n)
= (m, o(m)},
where 1t(m)
= n.
Let Q±M)
= {a E
Q(M) I
0* a = ±a} be the ± 1 eigenspaces of 0*. Show that 1t* : Q(N) ~ Q /M) is an isomorphism. (Hint: To show that range 1t* = Q +(M), note that 1t 0 0 = 1t implies range 1t* C Q +(M). For the converse inclusion, note that Tm1t is an isomorphism, so a form at 1t(m) uniquely determines a form at m. Show that this resulting form is
Section 6.5 Orientation, Volume Elements, and Codifferential
461
smooth by working in a chart on M diffeomorphic to a chart on N.) (ii) Show that lRP' is orientable if n is odd and is not orientable if n is even. (Hint: In (i) take M = So C ]R0+I, cr(x) = x, and N =Un. Let CO be a volume element on So induced by a volume element of Ro+l. Show cr*co =(_1)0+I CO• Now apply (i) to orient RlP'° for n odd. If n is even, let v be an n-fonn on JRlP'0; then 1t*v = fco. Show that f(x) =-fe-x) so f must vanish at a point of So.) 6.5L In Example 3.4.8C, the Mobius band M was defined as the quotient of ]R 2 by the equivalence relation (x, y) - (x+k, (_1)ky) for any k E Z. (i) Show that this equivalence relation is regular. Show that M is a non-compact. connected. two-manifold. (ii) Define cr:]R2 ~ R2 by cr(x, y) = (x + 1, -y). Show 1t 0 cr = cr, where cr: R2 ~ M, is the canonical projection. If v E n2(M) define f E .1(]R2) by 1t*v = fco. where co is an area fonn on R2. Show that f(x + 1, -y) = -f(x, y). (iii) Conclude that f must vanish at a point of ]R2, and that this implies M is not orientable. 6.5M The Klein bottle K is defined as the quotient of R2 by the equivalence relation defined by (x, y) - (x + n, (-l)0y + m) for any n, m E Z. (i) Show that this equivalence realtion is regular. Show that K is a compact, connected.
smooth, two-manifold. (ii) Use 6.5L(ii), (iii) to show that K is non-orientable. 6.5N Orientation in vector bundles. Let 1t : E ~ B be a vector bundle with finite dimensional fiber modeled on a vector space E, and assume B is connected. The vector bundle is said to be orientable if the line bundle L = E* 1\ .•• 1\ E* (dim (E) times) has a global nowhere vanishing section. An orientation of E is an equivalence class of global nowhere vanishing sections of L under the equivalence relation: cr I - cr 2 iff there exists f E .1'(B), f> 0 such that cr2 = fcrl' (i) Prove that E is orientable iff L is a trivial line bundle. Show that E admits exactly two orientations. Showthat an orientation [crJ of E induces an orientation in each fiber of E. (ii) Show that a manifold M is orient able iff its tangent bundle is an oriented vector bundle. (iii) Let E, F be vector bundles over the same base. Show that if two of E. F and E EEl F are orientable, so is the third. (iv) Let E, F be vector bundles (over possibly different bases) Show that Ex F is orientable if and only if E and F are both orientable. Conclude that if M. N are finite dimensional manifolds, then M x N is orientable if and only if M and N are both orientable. (v) Show that E EEl E* is an orientable vector bundle if E is any vector bundle. (Hint:
Chapter 6 DigerenJiol Forms
462
Consider the section Q(x)«el' a. 1). (e2 • a. 2 » = (a.2 • e l )
-
(a. p e 2 ) of (E (B E*)
/\ (E (B E*).)
(vi)
Choose an orientation of the vector space E and assume B admits partitions of unity. Show that the vector bundle atlas all of whose change of coordinate maps have positive determinant relative to the orientation of E. when restricted to the fiber. (Hint: If E is oriented and '!f: rr-I(V) ~ V' x E is a vector bundle chart with V' open in the model space of B. and V is connected in B. define
1t- I(V) ~
orientation preserving and
00
on F and define on a vector bundle chart (V.
with V connected in B. rr-I(V) = V. w(V) : V ~ L I V by w(V)(b)(el' ...• eT ) = w(
».
then w(VI)(b)=det",(~1
o
(
charts. ,Vj connected. i = I. 2. Next. glue the w(V)'s together using a partition of unity.) (vii) Vse (iv) to show that if E and F are oriented. then there exists a vector bundle atlas on E such that all
1t- I (b) ~
F are orientation preserving
isomorphisms. Such an atlas is called positively oriented. (viii)
Let E. F. B be finite dimensional. E oriented by 0" and B oriented by 00. Show that 1t*w /\ 0" is a volume form on E. Conclude that an orientation of B and an orientation of F uniquely determine an orientation of E as a manifold. This orientation is called the local product orientation.
(ix)
Show that any vector bundle oriented double cover 1t: orientation of 1t- I(b)}. p:
1t:
E
E ~ B. B
~
B with finite dimensional fiber has an
where
B=
(b. [Ilb]) I b E B. [llb1 is an
~ B is the map p(b.lllb]) = b. and
E = p*E.
Find the vector bundle charts of E and show that the fiber at (b. Illb]) is oriented by Ip *(~)l. where p: E ~ E is the mapping induced by p on the pull-bal:k bundle E. If E = TB. what is E? 6.50 Let M be a compact manifold and :M the space of Riemannian metrics on M. Let 'T be a space of tensor fields on M of a fixed type. A mapping cI> : :M ~ 'T is called covariant if for every diffeomorphism ((g). (i)
Show that covariant maps satisfy the identity
for every vector field X. (Assume cI> is differentiable and :M, 'T are given suitable Banach space topologies.)
Section 6.5 Orientation, Volume Elements, and Codifferential
463
(ii) Show that if M is oriented, then the map g>-+ l1(g), the volume element of g, is covariant. Is the identity in (i) anything interesting? 6.5P Let X be a vector field density on the oriented n-manifold M; i.e., X = F ® 11, where F is a vector field and 11 is a density. Use Exercise 6.58 to define div X and to show it makes intrinsic sense. 6.5Q Show that an orientable line bundle over a base admitting partitions of unity is trivial. (Hint: Since the bundle is orientable there exist local charts which when restricted to each fiber give positive functions. Regard these as local sections and then glue.) 6.5R Let 1t: E
~
SI be a vector bundle with n-dimensional fiber. If E is orientable show that
it is isomorphic to a trivial bundle over SI. Show that if p : F
~
SI is a non-orientable
vector bundle with n-dimensional fiber and if E is non-orientable, then E and F are isomorphic. Conclude that there are exactly two isomorphism classes of vector bundles with n-dimensional fiber over SI. Construct a representative for the class corresponding to the non-orientable case. (Hint: Construct a non-orientable vector bundle like the Mobius band: the equivalence relation has a factor (-I)2n-1 if the dimension of the fiber is 2n-1 or 2n. To prove non-orientability, proceed as in 6.5L.) 6.5S Let {el' ... , en+tl be the standard basis of ~n+1 and 0n+1 = el induced volume form. On Sn define (On E on(sn) by
1\ ... 1\
en+1 be the
for s E sn, vI' ... , vn E Tssn. (i) Use Proposition 6.5.8 to show that (On is a volume form on Sn; (On is called the (ii)
standard volume form on sn. ~ IRn+1 \ (OJ be given by f(t, s) = ts, where IR + is defined to be the set {t E IR It> O}. Show that if IR + is oriented by dt, Sn by (On' and ~n+1
Let f: IR + x Sn
by 0,,+1' then (JI)(t. s) = tn. Conclude that f is orientation preserving.
Chapter 7 Integration on Manifolds The integral of an n-form on an n-manifold is defined by piecing together integrals over sets in ]Rn using a partition of unity subordinate to an atlas. The change-of-variables theorem guarantees that the integral is well defined. independent of the choice of atlas and partition of unity. Two basic theorems of integral calculus. the change-of-variables theorem and Stokes' theorem. arc discussed in detail along with some applications.
~7.1
The Definition of the Integral
The aim of this section is to define the integral of an n-form on an oriented n-manifold M and prove a few of its basic properties. We begin with a summary of the basic results in ~n. Suppose f:]Rn ~]R is continuous and has compact support. Then f dx \ ... dxn is defined to
f
be the Riemann integral over any rectangle containing the support of f.
7.1.1 Definition Let U C]Rn be open and
0) E
nn(u) have compact support. If, relative to the
standard basis of ]Rn.
ro(x) =
I
, 0 ) ; ... ;
n.
where the components of
1
0)
; ' (x) dx I 1\ ... 1\ dx'u =
n
0)\ ...
n(x) dx
\
n
1\'"
1\
dx •
are given by 0);1 ... i,,(X) =
O)(x)(e;I' ...• e;).
then we define
Recall that if ~ is any integrable function and f:]Rn change o/variables theorem states that ~ 0 f is integrable and
~]Rn
is any diffeomorphism. thc
Section 7.1 The Definition ofthe Integral
465
where n = dx l /\ ... /\ dxn is Ihe standard volume form on JRn, and Jnf is Ihe Jacobian determinant of f relative to n. This charge of variables Iheorem can be rephrased in terms of pull backs in Ihe following form. 7.1.2 Change of Variables in Rn Let U and V be open subsets of JRn and suppose f: U
-+ V is an orientation-preserving diffeomorphism. If co E nn(v} has compact support, then r*co E
nn(u} has compact support as well and
(2) Proof If co = COl ... ndxl /\ ... /\ dxn, Ihen r*co = (COl ... n 0 f)(Jnf)n, where the n-form n = dx l /\ ... /\ dxn is Ihe standard volume form on JRn. As discussed in §6.5, Jnf> O. Since f is a diffeomorphism, Ihe support of r* co is r-I(supp co), which is compact. Then from (I),
Suppose Ihat (U, cp) is a chart on a manifold M and CO E nn(M}. If supp CO
C
U, we
may form co I U, which has Ihe same support. Then CP.(co I U} has compact support, so we may state Ihe following. 7.3.13 Definition Let M be an orientable n-manifold with orientation [n]. Suppose co E nn(M} has compact support C C U, where (U, cp) is a positively oriented chart. Then we define
1
co = cp.( coIU}.
(ep)
7.1.4 Proposition Suppose co E nn(M} has compact support C
C
U
n V, where
(U, cp), (V,
"') are two positively oriented charts on the oriented manifold M. Then
r J(ep)
Proof By 7.1.2
I CP.(co I U) = I ('"
[Recall that for diffeomorphisms f* Thus we may define
0
co-
r
co
J('I')
cp-I}.cp.(CO I U). Hence
= (11)*
and (f 0 g)*
Iv co = I(ep) co, where
= f*
0
I CP.(co I U) = I 'I'*(co I U).
g*.] •
(U, cp) is any positively oriented chart
containing the compact support of CO (if one exists). More generally, we can define co has compact support as follows.
IM co
where
Chapter 7 Integration on Manifolds
466
7.1.5 Definition Let M be an oriented manifold and 5'l an atlas of positively oriented charts. Let P = {(V u'
~ro
(so rou has
(3)
7.1.6I'roposilion (i) Tlte slim (3) contains only afinite number of nonzero terms. (ii) For any other atLllS ofpositillc/yoriented charts and subordinate partition of unity Q we
fQro.
have fpro= The common vaLue is denoted f~ ro, and is called the integral of ro E nn(M). Proof For any m E M, there is a neighborhood V such that only a finite number of gu are nonzero on U. By compactness of supp ro, a finite number of such neighborhoods cover the support of ro. Hence only a finite number of gu are nonzero on the union of these V. For (ii), let P = {(Vu'
and ~u~l3guhl3(m) = 1, for all mE M. Since ~l3hl3 = 1, we get
7.1.7 Change of Variables Theorem Suppose M and N are oriented n-manifolds and f: M
~
N is an orientation-preserving diffeomorphism. If ro E nn(N) has compact support, then
r*ro has compact support and
I I ro=
N
~
(4)
f*ro.
Proof First, supp r* ro = f-I(supp ro). which is compact. To prove (2). let {(V j•
= LflRn(
0
II) *(gj 0 II)ro =
L
ro. •
1
This result is summarized by the following commutative diagram:
Section 7.1 The Definition ofthe Integral
.:-
nn(M) .::;;::::=======~
7.1.8 Definition Let (M,Il) be a volume manifold. Suppose f support. Then we call fll the integral of f with respect to 11.
467
nn(N)
E
.'r(M) and f has compact
fM
The reader ean check that since the Riemann integral is IR-linear, so is the integral just defined. The next theorem will show that the integral defined by (4) ean be obtained in a unique way from a measure on M. (The reader unfamiliar with measure theory ean find the necessary background in Royden [1968]; this result will not be essential for future sections.) The integral we have described can clearly be extended to all continuous functions with compact support. Then we have the following.
7.1.9 Riesz Representation Theorem Let (M, 11) be a volume manifold. Let '13 denote the Borel sets of M, the o-agebra generated by the open (or closed, or compact) subsets of M. Then there is a unique measure mJ.L on '13 such that for every continuous function of compact support
(5) Proof Existence of such a measure mJ.L is proved in Royden [1968]. For uniqueness, it is enough to consider bounded open sets (by the Hahn extension theorem). Thus, let U be open in M, and let C u be its characteristic function. We construct a sequence of C'" functions of compact support CPn such that CPn .J.. C u ' pointwise. Hence from the monotone convergence theorem, fCPnll = fCPn
dmJ.L ~
fCu dmJ.L = mJ.L(U). Thus, mJ.L is unique. •
The space LP(M, 11), p
E
integrable. For p ~ 1, the norm
IR, consists of all measurable functions f such that I f IP is
II flip
=Cf I f 1PdmJ.L)l/p
makes U(M, 11) into a Banach space
(functions that differ only on a set of measure zero are identified). The use of these spaces in studying objects on M itself is discussed in §7.4. The next propositions give an indication of some of the ideas. If F: M ~ N is a measurable mapping and m M is a measure on M, then F*m M is the measure on N defined by F*mM(A) = mM(r-I(A». If F is bijective, we set F*(m N ) = (r-I)*m N . If f: M ~ lR is an integrable function, then fmM defined by
is the measure on M
468
Chapter 7 Integration on Manifolds
for every measurable set A in M. 7.1.10 Proposition Suppose M and N are orientable n-manifolds with volwneforms
~M
and Let F be an orientation preserving C1
~N and corresponding measures mM and mN . diffeomorphism of M onto N. Then
(6) Proof Let f be any
C~
function with compact support on M. By 7.1.7,
J/dmN = =
!/~N
=
LF*(f~N)
=
L(f O
F)(J("'M>!lNl)~M
L(f O F)(J("'M''''N)F)dmM'
As in the proof of 7.1.9, this relation holds for f the characteristic function of F(A). That is,
In preparation for the next result, we notice that on a volume manifold (M, of any vector field X is orientation preserving for each t
E
~),
the flow F t
lR (regard this as a statement on the
domain of the flow, if the vector field is not complete). Indeed, since Ft is a diffeomorphism,
J,.,(Ft) is nowhere zero; since it is continuous in t and equals one at t = 0, it is positive for all t. 7.1.11 Proposition Let M be an orientable manifold with volume form ~ and corresponding measure mw Let X be a (possibly time-dependent) C 1 vector field on M withflow Ft. The following are equivalent (if the flow of X is not complete. the statements involving it are understood to /wld on its domain):
=0;
(i)
div,.,X
(ii)
J,.,Ft
(iii)
(iv)
Ft*ml-' = m,., for all t E lR; F t *~ = ~ for all t E lR;
(v)
fMfdml-'= fM(foFt)dml-' for all fE
= 1 for all
t
E
lR;
Ll(M,~) andall tE lR.
Proof Statement (i) is equivalent to (ii) by 6.5.19. Statement (ii) is equivalent to (iii) by (6) and to (iv) by definition. We shall prove that (ii) is equivalent to (v). If JI-'Ft = 1 for all t E lR and f is continuous with compact support, then
Section 7.1 The Definition ofthe Integral
469
Hence, by uniqueness in 7.1.9, we have fMf dmll = fM(f 0 Ft)dm ll for all integrable f, and so (ii) implies (v). Conversely, if
then taking f to be continuous with compact support, we see that
Thus, for every integrable f, fM(f implies (ii). •
0
Ft)dm ll = fM(f
0
Ft)(IIlFt)dmw Hence JIlFt = I, and so (v)
The following result is central to continuum mechanics (see below and §8.2 for applications). 7.1.12 Transport Theorem Let (M,I1) be a volume manifold and X a vector field on M with flow Ft. For f E .1"(M x IR) and letting ft(m) = f(m, t), we have
(7) for any open set U in M. Proof By the flow characterization of Lie derivatives and 6.S.17(ii), we have
= F;(:
= F;[
11) + F;[(Lx ft)11 + ft(divIlX)I1]
(~~ + divll(ft X»)11J.
Thus by the change-of-variables formula,
7.1.13 Example Let p(x, t) be the density of an ideal fluid moving in a compact region of 1R3 with smooth boundary. One of the basic assumptions of fluid dynamics is conservation of mass: the mass of the fluid in the open set U remains unchanged during the motion described by a flow Ft' This means that
:J
F,{Ul
p(x, t)dx = 0
(8)
Chapter 7 Integration on Manifolds
470
for aH open sets U. By the transport theorem, (8) is equivalent to the equotion of continuity
a;
+ div(pu)
=
(9)
0,
where u represents the velocity of the fluid particles. We shaH return to this example in §8.2. As another application of 7.1.11, we prove the foHowing.
Poincare Recurrence Theorem Let (M, Il) be a volume manifold, mil the corresponding measure, and X a time-independent divergence-free vector field with flow Ft' Suppose A is a measurable set in M such that mll(A) < 00, Ft(x) exists for all t E R if x E A, and Ft(A) C A. Thenfor each measurable subset B of A and T ~ 0, there exists S ~ T such that B n Fs(B) T- 0. Therefore, a trajectory starting in B returns infinitely often to B. 7.1.14
Proof By 7.1.11, the sets B, FT(B), F2T(B), ... aH have the same finite measure. Since mll(A)
< 00, they cannot aH be disjoint, so there exist integers k > ~ > 0 satisfying FkT(B) n Fn(B) 0. Since FkT =(FTt (as X is time-independent), we get F(k_t)T(B)
T-
n B T- 0 . •
The Poincare recurrence theorem is one of the forerunners of ergodic thoery, a topic that wiH be discussed briefly in §7.4. A related result is the foHowing. 7.1.15 Schwarzschild Capture Theorem Let (M, Il) be a volume manifold, X a time -
independent divergence-free vector field with flow F t , and A a measurable subset of M with finite measure. Assume that for every x E A, the trajectory t ...... Ft(x) exists for all t E R Then for almost all x E A (relative to mil) the following are equivalent: (i) Ft(x) E A for all t ~ 0; (ii)
Ft(x)
E
A for all t:5; O.
Proof Let AI = nt~Ft(A) be the set of points in A which have their future trajectory completely in A. Similarly, consider A2 = nt~cft(A). By 7.1.11, for any
which shows, by letting
1: ~
00,
1:
~ 0,
that
Reasoning similarly for A 2, we get Il(A I) = Il(A I
n A2) =Il(A 2 ), so that
Section 7.1 The Definition ofthe Integral Let S Al \ S
= (AI \(AI n A 2» U (A2 \(AI n A2»; then ml!(S) = 0 and = Al n A2 = A2 \ S which proves the desired equivalence. •
471
seA. Moreover, we have
So far only integration on orientable manifolds has been discussed. A similar procedure can be carried out to define the integral of a one-density (see §6.S) on any manifold, orientable or not.
The only changes needed in the foregoing definitions and propositions are to replace the Iacobians with their absolute values and to use the definition of divergence with respect to a given density as discussed in §6.S. All definitions and propositions go through with these modifications. F-valued one-forms and one-densities can also be integrated in the following way. If co = Eti=ICOifi' where fp ... , fn is an ordered basis of F, then we set fM co = Eti=l( fM coi)fi
E
F. It
is easy to sce that this definition is independent of the chosen basis of F and that all the basic properties of the integral remain unchanged.
On the other hand, the integraL of
vector-bundLe-vaLued n-forms on M is not defined unless additional special structures (such as
triviality of the bundle) are used. In particular, integration of vector or general tensor fields is not defined.
Exercises 7.1A Let M be an n-manifold and J.I. a volume form on M. If X is a vector field on M with flow Ft show that
(Hint: Compute (d/dt)Ft*J.I. using the Lie derivative formula.)
7.1 B Prove the following generalization of the transport theorem
where co t is a time-dependent k-form on M and V is a k-dimensional submanifold of M. 7.1C (i) Let q> : SI ~ SI be the map defined by q>(e i9 ) = e2i9 , where e
E
[0,21t]. Let, by
abuse of notion, de denote the standard volume of SI. Show that the following identity holds:
ISl q>*(d9) = 2ISl de.
(ii) Let q> : M ~ N be a smooth surjective map. Then q> called a k-Jold covering map if
every n E N has an open neighborhood V such that q>-I(V) = U I U ... U Uk' are disjoint open sets each of which is diffeomorphic by q> to V. Generalize (i) in the q>*co = co. following way. If co E nn(N) is a volume form, show that
IM
kIN
Chapter 7 Integration on Manifolds
472
7.ID Define the integration of Banach space valued n·forms on an n·manifold M. Show that if the Banach space is lR '. you recover the coordinate definition given at the end of this section. If E. F are Banach spaces and A E L(E. F). define A* E L(Q(M. E). Q(M. F» by (A*a)(m)
= A(a(m».
Show that
(fM)
A*
0
= A (fM) on 0
QD(M. E).
7.IE Let M and N be oriented manifolds and endow M x N with the product orientation. Let PM : M x N -+ M and PN: M x N -+ N be the projections. If a E Qdirn M(M) and ~ E Qdirn N(N) have compact support show that
has compact support and is a (dim M + dim N)-form on M x N. Prove Fubini's Theorem
7.IF Fiber Integral Let cp: M -+ N be a surjective submersion. dim M = m and dim N = n. The map cp is said to be orientable if there exists 11 E QP(M). where p = m - n. such that for each YEN. jy *11 is a volume form on cp-l(y). where jy: cp-l(y) -+ M is the inclusion. An orientation of cp is an equivalence class of p-forms under the relation: 111 Tt2 iff there exists f E !F(M). f> 0 such that Tt2 = f Tt I. (i) If cp : M -+ N is a vector bundle. show that orientability of cp is equivalent to
orientability of the vector bundle as defined in Exercise 6.SM. (ii) If cp is oriented by Tt and N by ro. show that cp*ro 1\ 11 is a volume form on M. The orientation on M defined by this volume is called the local product orientation of M (compare with Exercise 6.SM(vi». (iii) Let
Q
n supp a is compact. for any compact set KeN}.
supported k-forms on M.
Show that
Q
is an
!F(M)-submodule of Qk(M). and is invariant under the interior product. exterior differential. and Lie derivative. (iv) If a E Q
1\ ••• 1\
T/N (k times) by
[~(x)(ui' ...• up)] (Txcp(v l ) •...• Txcp(vk»)
=a(x)(v l •...• vk• up ...• up)'
where cp(x) = y. x E M. vi' ...• vk E TxM. and ui' ...• up E ker(Txcp) Tx(cp-l(y». Assume cp is oriented. Definethejiberintegral
J. :Q~+P(M) -+ Qk(N) fib
by
(J.
fib
a)(x) =
J
ny.
if cp(x) = y;
=
Section 7.1 The Definition ofthe Integral
473
the right-hand side is understood as the integral of a vector-valued p-form on thc oriented p-manifold q>-l(y). Prove that ffiba is a smooth k-form on N. (Hint: Use charts in which q> is a projection and apply the theorem of smoothness of the integral with respect to parameters.) Show that if q> : M
~
N is a locally trivial fiber bundle
with N paracompact. ffib is surjective. (v) Let ~ e n'(N) have compact support and let a e n./+p(M). Show that q>*~ 1\ a e n cpk+t+P(M) and that
1.
fib
(Hint: Let E = Ty *N
~ 1\ 1.
(q> *~ 1\ a) =
1\ ... 1\
a.
fib
Ty *N (k times) and let F be the wedge product ~ + k
times. Define A e L(E. F) by A(y) = ~(y) 1\ Y and show that (q>*~ 1\ a)y = A*(~) using the notation of Exercise 7.1D. Then apply ffib to this identity and use 7.1D). (vi) Assume N is paracompact and oriented. q> is oriented. and endow M with the local product orientation. Prove the following iteraJed integration (Fubini-type)formula
by following the three steps below. Step 1: Using a partition of unity. reduce to the case M = N x P where q> : M
~
N is
the projection and M. N. P are Euclidean spaces. Step2: Use (v) and Exercise 7.1E to show that for ~ E nn(N) and ye np(p) with compact support.
r r (~x y)
JNJfib
=
r (~x y)
JM
Step 3: Since M. N. and P are ranges of coordinate patches. show that any ro e nm(M) with compact support is of the form ~ x y. (vii) Let q>: M ~ N and q>': M' ~ N' be oriented surjective submersions and let f: M ~ M'. and fo: N ~ N' be smooth maps satisfying fo 0 q> = q>'
0
f. Show that
Jfib r* = 0
f'
fo * 0 f'fib' where fib denotes the fiber integral of q>'. (viii) Let q>: M ~ N be an oriented surjective submersion and assume X e X(M) and Y X(N) are q>-related. Prove that
E
(For more information on the fiber integral see Bourbaki [1971) and Greub. Halperin. and Vanstone [1973). Vol. I.)
474
Chapter 7 IntegratiDn on Manifolds
7.1G Let
~
N be a smooth orientation preserving map. where M and N are volume
manifolds of dimension m and n respectively. For a define the linear functional
for all ~
E
E
nk(M) with compact support.
nm-k(N); i.e .•
If there is a 'YE nn-m+k(N) satisfying (
=fM ~"'Y.
identify
(i) If
a is a volume form. this definition corresponds to that for the push-forward of
measures. (iii) If
Exercise 7.1F(iv). (Hint: Prove the identity Exercise 7.1F(v) and (vi).) (M.~)
7.1H Let (i)
fM
a
Jfib a.
as defined in
=IN(~ " ffib a)
be a paracompact n-dimensional volume manifold.
If (N. v) is another paracompact n-dimensional volume manifold and f: M
fN
-fM f"'ro
an orientation reversing diffeomorphism. show that ro = nn(N) with compact support. (Hint: Use the proof of 7.1.12.) If"
(ii)
E
using
~
N is
for any ro E
nn(M) has compact support and -M denotes the manifold M endowed with
f-M"
fM'"
=the orientation [-~l. show that oriented atlas for (M. [~]). then -JI = {(Ui'
0
(Hint: If JI = {(Ui •
7.11 Let ron be the standard voiume form on sn. Show that
• if n = 2m+ 1. m ~ 0
and
by the following steps below. (i) Let M C ]Rn+1 be the annulus {x sn E
E
]Rn+1 10 < a < II x II < b < oo} and let f: lao b[ x
~
A bl! the diffeomorphism f(t. s) = ts. Use Exercise 6.SS(ii) to show that for x ]Rn+l.
where nn+1 = e l " ... " en+1 for Ie\, .... en+I } the standard basis of ]Rn+l. and where dt x ron denotes the product volume form on lao b[ x sn.
Section 7.1 The Definition of the Integral (ii) Deduce the equality
(iii) Let a
J. 0
and b i
00
to deduce the equality
Prove that ~
2
l_~e-u du = {it,
to deduce the required formula for
r ron'
Js•
475
§7.2 Stokes' Theorem Stokes' theorem states that if IX. is an (n-I)-form on an orientable n-manifold M, 111en the integral of dIX. over M equals the integral of IX. over aM, me boundary of M. As we shall see in me next section, the classical theorems of Gauss, Green, and Stokes are special cases of mis result. Before stating Stokes' theorem formally, we need to discuss manifolds wim boundary and meir orientations. 7.2.1 Definition Let E be a Ba/Ulch space and AE E*. Let
E".=
(XE EIA(x)~O}. cal/eda
hal/-space of E, and let veE". be an open set (in the topology induced on E". from E). Call Int V = V n (x EEl A(X) > O} the interior of V and av = V n ker A the boundary of U. If E
=Rn
and A is the projection on the j-th factor, then
E".
is denoted by Rnj and is called positive
j-th half-space. Rnn is also denoted by Rn+. We have V = Int V U av, Int V is open in V, Int V
=0.
av
The situation is shown in Fig. 7.2.1. Note that
is closed in V (not in E), and aU
av
n
is not the topological boundary of
V in E, but it is me topological boundary of V intersected with mat of of the notation au is temporary.
E"..
This inconsistent use
intU
au Figure 7.2.1 A manifold with boundary will be obtained by piecing togemer sets of me type shown in the figure. To carry this out, we need a notion of local smoothness to be used for overlap maps of charts. 7.2.2 Definition Let E and F be Banach spaces, A E E*, Il E F*, V be an open set in EA, and V be an open set in Fw A map f:V~V iscalledsmoothifforeachpoint XE V there are
Section 7.2 Stokes' Theorem
477
open neighborfwods V I of x in E and V I of f(x) in F and a smooth map fl : V I
n VI = fll V n VI'
that fl V
is a smooth map g: V
~
V I such
We define Df(x) = Dfl(x). The map f is a difjeomorphism
if there
V which is an inverse of f. (In this case Df(x) is an isomorphism of E
---7
with F.) We must prove that this definition of Of is independent of the choice of fl' that is, we have to show that if
=
D
°
for all x
n £,.. If x E Int(W n E,..), this fact is obvious.
W
E
choose a sequence xn
Int(W n E)
E
such that xn
---7
If
n £,. = 0, then x E a(W n E),
x; but then 0 = D
~
D
hence D
Int V,
E
£,. be open,
C
aFw Then D
Proof The quotient F/ker E
F/ker
~
~
C
---7
FI! be a smooth map, and assume thatfor some Xo
aFI! = ker~.
is isomorphic to JR, so that fixing f with
E
Sf: ker
~
EEl JR
---7
diffeomorphisms (in the sense of 7.2.2) of ker F I ~(y) ~
OJ.
~
lR given by
F given by Sf(y, t) ~
=y
+ tf which induces
x [0, oc[ with FI! and of ker
~
x )-.00, 0] wilh
Denote by p: F ~ lR the linear map given by Sf- I followed by the
projection kerJ.l EEl JR ~ JR, so that y if p(y)
~
JR is the unique number for which t[f] = [y). This isomorphism in tum
defines the isomorphism E
> 0, the element [f]
forms a basis. Therefore [f] determines the isomrophism T f : F/ker ~
Tf([y]) = t, where t
{y
~(O
0 (respectively
= 0,
~
E
FI! (respectively ker~, (y
E
F I ~(y) ~ O}) if and only
0).
Notice that the rclation
(p
0
° Letting t ~ 0, we get (p we get (p
0
D
~
°
for all x
E
V, implies that 0
O(tX») ~(poD
D
° 0
0
D 0
0
for all x
E
E
E. Similarly, for t <
°
and letting t ~ 0,
E. The conclusion is (D
C
ker
•
~.
Intuitively, this says that if
~
°
and maps an interior point
t,
the boundary, then the derivative must be zero in the normal direction. The reader may also wish to prove 7.2.3 from the implicit mapping theorem. Now we carry this idea one step further. 7.2.3 Lemma Let V be open in EA.' V be open in FW and f: V ~ V be a diffeomorphism.
Then f restricts to diffeomorphisms Int
r: Int V
~
Int V and
ar: av ~ avo
Chapter 7 Integration on Manifolds
478
Proof Assume first that
au = 0, that is, that V n ker A. = 0. We shall show that av = 0 and *" 0, there exists x V such that f(x) aV and hence by
hence we take Int f = f. If av
E
E
definition of smoothness there are open neighborhoods V) and f(x)
E
V\, and smooth maps f) : VI
=f- I I V n V).
Let xn
E
V\, xn ~ x, Yn
C
~
V\, g) : V)
E
V) \aV, and Yn
V and V)
C
F, such that x
E
such that f I V) = fl' gl I V
~ V)
= f(x n).
V)
n V)
We have
and similarly
so that Df(x)-) exists and equals Dg)(f(x». But by 7.2.3, Df(x)(E)
C
ker 11 , which is
impossible, Df(x) being an isomorphism. Assume that aV
*" 0.
If we assume av
=0,
then, working with f-) instead of f, the
above argument leads to a contradiction. Hence av neighborhood V)
C
V, V)
af(V) = 0, and f(V)
n av = 0,
=av
*" 0.
av) = 0.
Let x
:J
and f I av : av
E
Int V so that x has a
Thus, by the preceding argument,
is open in V\av. This shows that feint V)
with f- I , we conclude that f(lnt V) But then f(aV)
and hence
C
Int V. Similarly, working
Int V and hence f: Int V ~ Int V is a diffeomorphism.
~
av is a diffeomorphism as well. _
Now we are ready to define a manifold with boundary.
7.2.5 Definition A manifold with boundary is a set M together with an atlas of charts with boundary on M; charts with boundary are pairs (V,
Int M is a manifold (with atlas obtained from (U,
2 aM is a manifold (with atlas obtained from (V,
C
C
EA. by the set
Eo". by a
C
aE = ker A.); 3 aM is the topological boundary of Int M in M (although Int M is not the topological interior of M). Summarizing, we have proved the following. 7.2.6 Proposition If M is a manifold with boundary, then its interior Int M and its boundary
479
Section 7.2 Stokes Theorem
aM are smooth manifolds without boundary. Moreover, if f: M
~
N is a diffeomorphism, N
being another manifold with boundary, then f induces, by restriction, two diffeomorphisms Int f: Int M ~ Int Nand af: aM ~ aN. If n = dim M, then dim(lnt M) = n and dim(aM) = n - 1.
[R"
I
Figure 7.2.2 To integrate a differential n-form over an n-manifold M, M must be oriented. If Int M is oriented, we want to choose an orientation on aM compatible with it. In the classical Stokes theorem for surfaces, it is crucial that the boundary curve be oriented, as in Fig. 7.2.3.
M
Figure 7.2.3 The tangent bundle to a manifold with boundary is defined in the same way as for manifolds without boundary. Recall that any tangent vector in TxM has the form [dc(t)/dtll t =~, where c : [a, b 1 ~ M is a Cl curve, a < b, and 1: E [a, b 1. If x E aM, we consider curves c : [a, b] ~ M such that c(b) = x. If q> : U ~ U' C E" is a chart at m, then [d(q> 0 c)(t)/dt]I t = b in general
480
Chapter 7 IntegratiJJn on Manifolds
points out of U /• as in Fig. 7.2.4. Therefore. TxM is isomorphic to the model space E of M even if x E aM (see Fig. 7.2.5). It is because of this result that tangent vectors are derivatives of Cl-curves defined on closed intervals. Had we defined tangent vectors as derivatives of C1-curves defined on open intervals. TyM for y E aM would be isomorphic to ker A. and not to E. In Fig. 7.2.4. E = JRD. A. is the projection onto the n-th factor. and c(t) is defined on a closed interval whereas the Cl-curve d(t) is defined on an open interval.
.....
",
T,M
T,8M
Figure 7.2.4 Having defined the tangent bundle. all of our previous constructions including tensor fields and exterior forms as well as operations on them such as the Lie derivative. interior product. and exterior derivative carry over directly to manifolds with boundary. One word of caution though: the fundamental relation between Lie derivatives and flows still holds if one is careful to take into account that a vector field on M has integral curves which could run into the boundary in finite time and with fmite velocity. (If the vector field is tangent to aM. this will not happen.)
Figure 7.2.S
481
Section 7.2 Stokes' Theorem
Next. we turn to the problem of orientation. As-for manifolds without boundary a volume
form on an n-manifold with boundary M is a nowhere vanishing n-form on M. Fix an orientation on JRn. Then a chart (U.
E
~
JRn is
U. If M is paracompact this latter condition is equivalent to
orientabilityof M (the proof is as in 6.5.2.) Therefore. for paracompact manifolds. an orientation on M is just a smooth choice of orientations of all the tangent spaces. "smooth" meaning that for all the charts of a certain atlas. called the oriented charts. the maps D(
~
0 to define the charts at the boundary. Had we done that. the very definition
of an oriented chart would be in jeopardy. For example. consider M
= {x E
charts must have range in JR+
= [0. 1]
and agree that all
JR I x ~ O}. Then an example of an orientation reversing
chart at x = 1 is O}. then a positively oriented chart at 1 is
r=x..JI-L+-[--_. t [ o
]]
1
R+
x t--l-x R_-·~-----------------+~~O
Figure 7.2.6 Once oriented charts and atlases are defined. the theory of integration for oriented paracompact manifolds with boundary proceeds as in §7.1. Finally we define the boundary orientation of aM. At every x E aM. the linear space T x(aM) has codimension one in T xM so that there are (in a chart on M intersecting aM) exactly two kinds of vectors not in ker A.: those for which their representatives v satisfy A.(v) > 0 or A.(v) < O. i.e .• the inward and outward pointing vectors. By Lemma 7.2.4. a change of chart does not affect the property of a vector being outward or inward (see Figure 7.2.4). If dim M = n. these considerations enable us to define the induced orientation of aM in the following way. 7.2.7 Definition Let M be an oriented n-manifold with boundary. x E aM and
482
Chapter 7 Integration on Manifolds
For example, we could choose for n the outward pointing vector to lR\ and perpendicular to ker A. If A: lRn ~ lR is the projection on the n-th factor, then (Tx<Jl)-I(n) = - a/ax n and the situation is illustrated in Figure 7.2.5.
7.2.8 Stokes' Theorem Let M be an oriented smooth paracompact n-manifold with boundary and (X E nn-I(M) have compact support. Let i : aM ~ M be the inclusion map so that i*(X E nn-l(aM). Then
r
JaM
r d(X
i*(X=
JM
(I)
or for short, (I ')
If aM = 0, the left hand side of(1) or (1') is set equal to zero. Proof Since integration was constructed with partitions of unity subordinate to an atlas and both sides of the equation to be proved are linear in (X, we may assume without loss of generality that (X is a form on U C lRn+ with compact support. Write n
(X =
L
(-l)i-l(Xidx l" ... " (dxit " ... "dxn,
(2)
i=1 where
above a term means that it is deleted. Then n
.
~ a(X' 1 n d(X = LJ - i dx " ... " dx i=1 ax
(3)
and thus
Jlr d(X U
~l a(Xi 1 n = LJ n - i dx ... dx . i=1 R ax
(4)
There are two cases: aU = 0 and au", 0. If au = 0, we have fau(X = O. The integration of the i-th term in the sum occurring in (4) is
1.
(1
Rn - 1
R
a(Xi . dx iJ dx 1 ... (dx i)A ... dxn (no sum) ax'
(5)
and f~(a(Xi/axi)dx = 0 since (Xi has a compact support. Thus (4) is zero as desired. If au '" 0, then we can do the same trick for each term except the last, which is, by the fundamental theorem of calculus,
483
Section 7.2 Stokes' Theorem
1 (i ~ Rn- 1
0
aan dXnJ dx l ... dx n- I = ax n
1
lan(x 1, ... , xn-I , 0 )dx 1 ... dx n-I .
Rn -
(6)
since an has compact support. Thus
1 I U
da = -
R n .- 1
a n(x 1, ... , xn-I , O)dx 1 ... dx n-I
(7)
On the other hand,
f f a =
ilU
a =
ilR:
foR:
( -I) n-I a n(x 1, ... , xn-I ,
0 )d Id X A'" A Xn-I .
(8)
But lR,n-1 = a:Rn+ and the usual orientation on lR,n-1 is oot the boundary orientation. The outward unit normal is - en = (0, ... , 0, -1) and hence the boundary orientation has the sign of the ordered basis {-en,el, ... ,en_I},whichis (_I)n. Thus(8)becomes =
f
(-1)
n-I n 1 ex (x , ... , xn-I ,O)dx 1 A'"
A
dx
n-I
ilR:
= (_1)2n-1
r
JR,n.-t
exn(X, 1
... ,
x
n-I , O)d x 1 ... d x n-I.
(9)
Since (_1)2n-1 = -1, combining (7) and (9), we get the desired result. _ This basic theorem reduces to the usual theorems of Green, Stokes, and Gauss in lR,2 and lR,3,
as we shall see in the next section. For forms with less smoothness or without compact support, the best results are somewhat subtle. See Gaffney [1954], Morrey [1966], Yau [1976], Karp [1981] and the remarks at the end of Supplement 7.2B. Next we draw some important consequences from Stokes' theorem. 7.2.9 Gauss'Theorem Let M be an oriented paracompact n-manifold with boundary and X a vector field on M with compact support. Let ~ be a volume form on M. Then
r (div X)~ = filM ix~'
JM
(10)
Proof Recall that
The result is thus a consequence of Stokes' theorem. _ If M carries a Riemannian metric, there is a unique outward-pointing unit normal nilM along
484
Chapter 7 Integration on Manifolds
aM. and M and aM carry corresponding uniquely determined volume forms IlM and llaM and associated measures mM and maM' Then Gauss' theorem reads as follows. 7.2.10 Corollary
where (X. naM) is the inner product of X and naM"
Proof Let llaM denote the volume element on aM induced by the Riemannian volume element IlM E nn(M); i.e .• for any positively oriented basis vI' ...• vn_1 E Tx(aM). and charts chosen so that naM = - a/ax n at the point x.
Since (iXIlM)(X)(VI' ...• vn_l) = IlM(X)(Xi (X)V i + Xn(x)~. vI' ...• Vn_l) axn
= -Xn(X)llaM(X)(VI' ...• vn-I) and xn =- (X. naM ). the corollary follows by Gauss' theorem. _ 7.2.11 Corollary If X is divergence-free on a compact boundaryless manifold with a volume element Il. then X as an operator is skew-symmetric; that is,for f and g E j{M).
L
X[f]gll = -
t
fX[glll·
Proof Since X is divergence free. Lx(hll) = (Lxh)1l for any h E j"(M). Thus X[flgll + fX[glll = L x (fg)1l = Lx(fgll)· Integration and the use of Stokes' theorem gives the result. _ 7.2.12 Corollary If M is compact without boundary X E X(M). a then
L
Lx a /\
~
= -
t
E
nk(M) and ~
E
nn-k(M).
a /\ Lx~·
Proof Since a /\ ~ E nn(M). the formula follows by integrating both sides of the relation dix(a /\~) = Lx(a /\ ~) = Lxa /\ ~ + a /\ Lx~ and using Stokes' theorem .•
485
Section 7.2 Stokes' Theorem
7.2.13 Corollary If M is a compact orientable. boundaryless n-dimensional pseudo-Riemannian
manifold with a matric g of index Ind(g). then d and 8 are adjoints. i.e .•
for a
E
d«M) and ~
E
nk+1(M).
Proof Recall from 6.5.21 that 8~
=(_l)nk+l+lnd(g) *d*~.
so that
da/\ *~-a/\*8~ = da"*~+(_l)nk+lnd(g)a,,**d*~ da /\
*~
+ (_l)nk+lnd(g)+k(n-k)+Ind(g)a /\
d*~
da /\ *~ + (-Ita" d*~ d(a /\ *~) since k2 + k is an even number for any integer k. Integrating both sides of the equation and using Stokes' theorem gives the result. _ The same identity
(da.~)
= (a. 8~) holds for noncompact manifolds. possibly with
boundary. provided either a or ~ has compact support in Int M.
~ SUPPLEMENT 7.2A Stokes' Theorem/or Nonorientable Manifolds
Let M be a nonorientable paracompact n-manifold with a smooth boundary aM and inclusion map i : aM
~
M. We would like to give meaning to the formula
in Stokes' theorem. Clearly. both sides makes sense if dp and i*(p) are defined in such a way that they are densities on M and aM. respectively. Here d should be some operator analogous to the exterior differential. and p should be a section of some bundle over M analogous to An-l(M). Denote the as yet unknown bundle analogous to Ak(M) by A\(M) and its space of sections n\(M). Then we desire an operator d: n k,(M) ~ n k +1, (M). k = O..... n. and desire A\(M) to be isomorphic to I A I(M). To guess what A\(M) might be. let us first discuss A\(M). The key difference between an n-form (j) and a density p is their transformation property under a linear map A: Tm M ~ TmM as follows:
486
Chapter 7 Integration on Manifolds
for mE M and vI' ...• vn E TmM. If vI' .... vn is a basis. then det(A) > 0 if A preserves the orientation given by vI' .... vn and det(A) < 0 if A reverses this orientation. Thus p can be thought of as an object behaving likc an n-form at every m E M once an orientation of T mM is given; i.e .. p should be thought of as an n-form with values in some line bundle (a bundle with one-dimensional fibers) associated with the concept of orientation. This definition would then generalize to any k; Ak~(M) will be line-bundle-valued k-forms on M. We shall now construct thi s line bundle. At every point of M there are two orientations. Using them. we construct the oriented
M -t M (see 6.5.7). Since M is not a line bundle. some other construction is in
double covering
order. At every m E M. a line is desired such that the positive half-line should correspond to one orientation of T mM and the negative half-line to the other. The fact that must be taken into account is that multiplication by a negative number switches these two half-lines. To incorporate this idea. identify (m.
[~].
a) with (m.
[-~].
-a) where mE M. a E JR. and
[~]
is an orientation of
TmM. Thus. define the orientation line bundle oeM) = {(m. [~]. a) I m E M. a E JR. and [~l] is an orientation of TmM) )/- where - is the equivalence relation (m. [~]. a) - (m. [-~]. -a). Denote by (m. [~]. a) the elements of oeM).
It can be checked that the map 1t : oeM) -t M
defined by 1t«m. [~]. a») = m is a line bundle with bundle charts given by'll: rr-l(U) -t
C
JRn x JR 1-+ «
0
0
u' x R
If M is paracompact.
then oeM) is an orientable vector bundle (see exercise 6.SN). If in addition M is also connected. then M is orientable if and only if oeM) is trivial line bundle; the proof is similar to that of Proposition 6.5.7. 7.2.14 Definition A twisted k-form on M is a o(M)-valued k-form on M. The bundle of twisted k-forms is denoted by A\(M) and sections of this bundle are denoted nk ~(M) or r-(AkiM)). Locally. a section p E nkiM) can be written as p =
ru;
where
a E nk(U) and
~ is an
orientation of U regarded as a locally constant section of oeM) over U. The operators d : Qk~(M) -t Qk+liM) and ix : n k~(M) -t n k- 1~(M). where X E X(M). are defined to be the unique operators such that if p = a~ in the neighborhood U. then dp = (da)~ and ixp
= (ixa)~.
One has Lx
= ix
0
d + d 0 i x ' Note that if M is orientable. A\(M)
coincides with Ak(M). Next we show that the line bundles
I A(M) I and AniM) are isomorphic. If A E I A(M) 1m
and vi' .... vn E TmM. define (I..)(v l ..... v n) = (m. [o(vl' ...• v n)]. I..(vl' .... v n). if {vI' .... vnl is a basis of TmM. and setting it equal to
487
Section 7.2 Stokes' Theorem
O. if (vI' ...• vnl are linearly dependent. where [cr(v i •...• vn)] denotes the orientation of TmM given by the ordered basis (vI' ...• vnl. (A) is skew symmetric and homogeneous with respect to scalar multiplication since if (vI' ...• vnl is a basis and a
E
JR. we have
(A)(V 2 • vi' v 3 •...• v n) = (m. [cr(v 2 • vi' v 3 •...• v o)]' A(V 2 • vi' v 3 •...• vn»
= (m. [- cr(vi' v 2 ••..• vn)]. A(V I •···• vn» = (m. [cr(v i •...• vn)]. -A(V I •...• Yo»~ =- (A)(V I' ...• Vn) and
»
(A)(avi' v2 •.••• vn) = (m. [cr(avi' v 2 •...• Yo)], A(av i •...• v o
= (m.
[(sign a)cr(vi' ...• vn)],
I a I A(V I •...•
Yo»~
»
= (m. [cr(v i •...• vn)], aA(v l •...• v n
= a(A)(v l •...• Yo)· The proof of additivity is more complicated. Let vi'vI" v 2 •...• vn E TmM. Ifboth {vi' ...• vnl and
{vI'. v2 •...• vnl are linearly dependent. then so are (vI + vI" v 2 •...• vol and the additivity
property of (A) is trivially verified. So assume that {vI' ...• vn} is a basis of TmM and write vI' aiv i + ... + anv n. Therefore
=
A.(v l '. v 2•••.• v n) A(V I + vI" v 2 •..• v n)
= I a l I A(Vi' ...• vn)' and =
11 + all A.(v i •...• vn).
Moreovcr. if (i) a l > O. then [cr(v i •...• v n)] = [cr(v l '. v 2 •...• v n)] = [cr(v i + vI" v 2 •...• v n)]; (ii) a l = O. then [cr(vi' ...• vn)] = [cr(v i + vI" v 2 •...• v n)] and (A)(V I'. v 2 •...• v n) = 0; (iii) -1 < a l < O. then [cr(v i •...• v o)] = [-<J(v I'. v 2 •...• v n)] = [cr(v i + vI" v 2 •.••• v n)]; (iv) a l = -1. then [cr(v i •...• v n)] = [- cr(v l '. v 2 •...• v n)] and (A)(V I + VI" v 2 •...• v n) = 0; (v) a l < -1. then [cr(v i •...• v n)] = [- cr(v l '. v 2 •...• v n)] = [- cr(v i + VI" v 2 •...• vn)]. Additivity is now checked in all five cases separately. For example. in case (iii) we have
»
(A)(V I + VI" v 2 •...• v n) = (m. [cr(v i + VI" v 2 •...• vn]' A(V I + VI" ...• v n = (m. [cr(v i •...• v n)]. (1 + al)A(v l •...• vn»
= (m.
»
[cr(vi' ...• v n)]. A(Vi' ...• v n
»
+ (m. [- cr(v l '. V2 ••.•• v n»). -I a l I A(Vi' .... v n = (A)(Vi' ...• vn) + (A)(V I'. V2 •...• vn)·
Thus has values in A n~(M). The map is clearly linear and injective and thus is an isomorphism of
I A(M) I with
An~(M). Denote also by the induced isomorphism of
I Q(M) I
488
Chapter 7 Integration on Manifolds
with nn~(M). The integral of p E nniM) is defined to be the integral of the density ell-I(p) over M. In local coordinates the expression for ell is
where I;no is the basis element of the space sections of o(U) given by I;no(u)(vl' ...• vn) =(u. [o(vl' ...• vn)l. sign(det(vim. where (vi) are the components of the vector Vj in the coordinates (x I•...• xn) of U. Therefore
and
for any smooth functions a. b : U -+ JR. Finally. for the formulation of Stokes' Theorem. if i : dM -+ M is the inclusion and p E nn-I~(M). the induced twisted (n-1)-form i*p on dM is defined by setting (i*p)(m)(vl' ...• vn_l ) = (m. [sign[llnl 0(- d/dXn• v\, ...• vn_I)]. p'(m)(vl' ...• vn_I if v\, ...• vn_1 are linearly independent and setting it equal to zero. if vi' ...• vn-I are linearly dependent. where (x I •...• xn) is a coordinate system at m with dM described by xn = 0 and p(m)(vl' ...• vn_l ) = (m.
».
sign[llml. p'(m)(vl' ...• vn_ I » with p'(m) skew symmetric; moreover sign[llml = +1 (respectively -1) if [Ilml and [O(-d/dX n• v\, ...• vn_l)l define the same (respectively opposite) orientation of TmM. If M =U. where U is open in JR\ and p =a a~ E nn-\(U). then i*p = (-l)ni*(aa)l;n-l o. In particular. if n
~= L~idxl
/\ ... /\ (dx i)" /\ ... /\ dxn.
i=1
we have
With this observation. the proof of Stokes' Theorem 7.2.8 gives the following. 7.2.15 Nonorientable Stokes' Theorem Let M be a paracompact rwrwrientable n-manifold with smooth boundary dM and pEnn-I ~(M). a twisted (n-1 )-form with compact support. Then
r i*p. J.Mdp = JaM
489
Section 7.2 Stokes' Theorem
The same statement holds for vector-valued twisted (n-I)-forms and all corollaries go through replacing everywhere (n-I)-forms with twisted (n-I)-forms. For example, we have the following. 7.2.16 Non-Orientable Gauss Theorem Let M be a nonorientable Riemannian n-manifold
with associated density ~M' Thenfor X
E
X(M) with compact support
f. div(X)~M f M
=
aM
(X· n)!laM
where n is the outward unit normal of aM, !laM is the induced Riemannian density of aM and LX!lM = (div X)!lM' For a concrete situation in lR3 involving these ideas, sec Exercise 7.31
$:n
~ SUPPLEMENT 7.2B Stokes' Theorem on Manifolds with Piecewise Smooth Boundary
The statement of Stokes' theorem we have given does not apply when M is, say a cube or a cone, sinee these sets do not have a smooth boundary. If the singular portion of the boundary (the four vertices and 12 edges in case of the cube, the vertex and the base circle in case of the cone), is of Lebesgue measure zero (within the boundary) it should not eontribue to the boundary integral and we can hope that Stokes' theorem still holds. This supplement discusses such a version of Stokes' theorem inspired by Holmann and Rummier [1972]. (See Lang [1972] for an alternative approach.) First we shall give the definition of a manifold with piecewise smooth boundary. A glance at the definition of a manifold with boundary makes it clear that one could define a manifold with
corners, by choosing charts that make regions near the boundary diffeomorphic to open subsets of a finite intersection of positive closed half-spaces. Unfortunately, singular points on the boundary-osuch as the vertex of a cone--need not be of this type. Thus, instead of trying to classify the singular points up to diffeomorphism and then make a formal intrinsic definition, it is simpler to consider manifolds already embedded in a bigger manifold. Then we can impose a condition on the boundary to insure the validity of Stokes' theorem. 7.2.17 Definition Let U C lR n- 1 be open and f: U ~ lR be continuous. A point p on the graph of f, r f = {(x, f(x» I x E U}, is called regular if there is an open neighborhood V of P such that V () r f is an (n - 1) dimensional smooth submanifold of V. Let Pf denote the set of regular points. Any point in <Jf = r f \ Pf is called singular. The mapping f is called piecewise smooth if Pf is Lebesgue measurable, 1t(<Jf ) has measure zero in U (where 1t: U x lR ~ U is the projection) and f !1t(<Jf ) is locally Holder; i.e.Jor each compact set K C 1t(<Jf ) there are
490
Chapter 7 Integration on Manifolds
constants c(K) > 0, 0 < a(K) ~ 1 such that I f(x) - f(y) I ~ c(K) II x - Y 1I'1(K) for all x, y E K. Note that Pr is open in rr and that Int(rn U Pr' where rr-
= {(x, y) E
U x JR.I y ~
f(x)}, is a manifold with boundary Pr' Thus Pf has positive orientation induced from the standard orientation of JR.n. This will be called the positive orientation of r f. We are now ready to define manifolds with piecewise smooth boundary. 7.2.18 Definition Let M be an n-manifold. A closed subset N of M is said to be a manifold
with piecewise smooth boundary iffor every pEN there exists a chart (U,
=U' x
n U) = r f- n
n U) = ri n
See Fig. 7.2.7.
Pn = regular points
Figure 7.2.7 It is readily verified that the condition on N is chart independent, using the fact that the composition of a piecewise smooth map with a diffeomorphism is still piecewise smooth. Thus, regular and singular points of bd(N) make intrinsic sense and are defined in terms of an arbitrary chart satisfying the conditions of the preceding definition. Let PN and aN denote the regular and singular part of the boundary bd(N) of N in M.
IN'"
To formulate Stokes' theorem, we define for" an n-form (respectively, density) on M with compact support. This is done as usual via a partition of unity; PN and aN play no role since they have Lebesgue measure zero in every chart: PN because it is an (n - I)-manifold and aN by definition. It is not so simple to define Ibd(N)~ for ~E nn-l(M) lemma is needed.
(respectively, a density). First a
Section 7.2 Stokes' Theorem
Let ~ E nn-I(u x JR). where U is open in JRn-l. supp(~) is compact and f:
7.2.19 Lemma U
~
491
~ R. where i: Pr ~ U x JR is the inclusion and I.E nn-I(Pr) is the boundary
JR is a piecewise smooth mapping. Then there is a smooth bounded function a: Pr
such that i*~
= aA
volume form induced by the canonical volume form of U x JR
C
JRn on Int(rn U Pr'
Proof The existence of the function a on Pf follows since nn-I(pf) is one-dimensional with a basis element A. We prove that a is bounded. Let p E Pf and v!' ...• v n_1 E Tp(pf) be an orthonormal basis with respect to the Riemannian metric on Pf induced from the standard metric of JRn • and denote by n the outward unit normal. Then a(p)
= a(p)(dxl
Let Vi
=
/\ ... /\ dxn)(p)(n. v!' ...• v n_l )
I
I
.
=
a(p)A(p)(vl' ...• vn_l )
=
~(p)(vl' ...• vn_I ).
I
a Vij -a j . Smce -aI •...• -n and n. VI' .... Vn_1 are orthonormal bases of ax p ax p ax p
Tp(U x JR). we must have I vi I::;: 1 for all i. j. Hence if n
~ = L~idx I /\
... /\ (dx i) 1\ / \
... / \
dxn.
i=1 then
I a(p) I
I
~(P)(vI"'"
I t~i(P) L
vn_l) I =
1=1
(-l)i(sign Ci)vf(1) ...
v~~~-I) I
(JES n_ 1
n
=
LI~i(p)l(n-I)! i=1
which is bounded. since
~
has compact support. T
In view of this lemma and the fact that Cif has measure zero. we can define
1~ f i*~ =
Ir
=
Pr
faA. Pr
Now we can define. via a partition of unity. the integral of 11
E
nn-I(M) (or a twisted (n -
I)-form) by
J.bd(N) 11=f11 PN 7.2.20 Piecewise Smooth Stokes Theorem Let M be a paracompact n-manifold and N a closed submanifold of M with piecewise smooth boundary. If (i) M is orientable and 0> E nn-I(M) has compact support. or (ii) M is nonorientable and 0> E nn-liM) is a twisted (n - I)-form (see the preceding
492
Chapter 7 Integration on Manifolds
supplement) which has compact support, then
IrN dro
=
L
(N)
roo
The proof of this theorem reduces via a partition of unity to the local case. Thus it suffices to prove that if U is open in lItn- 1 , ro
nn-l(u x TIt) has compact support, and f: U ~ lit is a
E
piecewise smooth mapping, then (1)
The left -hand side of (I) is to be understood as the integral over the compact measurable set
r f- n
supp(ro). For the proof of (I) we use three lemmas.
7.2.21 Lemma Equation (I) holds if ro vanishes in a neighborhood of Proof Let V be an open neighborhood of another open neighborhood of
Of
Of
in U x lit.
Of
in U x lit on which ro vanishes and let W be
(which is closed in V) such that c1(W) n (U x lit) c V. The
r f-
C
0 is an n-dimensional
submanifold of 0 with bd(rf- n 0) = r f no. Since supp(dro) n r f
C
r f- n 0 and supp(ro) n
set 0 = (U x lit) \ cl(W) is open and since it is disjoint from
rf
C
r f n 0,
Of'
by the usual Stokes theorem, we have
The purpose of the next two lemmas is to construct approximations to dro and ro if ro does not vanish near
Of.
For this we need translates of bump functions with control on their derivatives.
7.2.22 Lemma Let C be a box (rectangular paral/elipiped) in lItn of edge lengths 2~i and let 0 be the box with the same center as C but of edge lengths 4~/3. There exists a C~ function
= 1.
Then 'V(x 1, •.• , xn)
=
U3
and choose an integer
E
=
N such that 2IN < E. Let h : lit ~ [0,1] be a bump function that is equal to 1 for I t I < 1/2 and that vanishes for I t I > 1. Then f: lit
~
[0, 1 J. defined by f(t) = 1 - h(t) is a
C~
function
vanishing for I t I < 1/2 and equal to 1 for I t I > 1. Let fn(t) = f(nt) for all positive integers n and note that
I fn'(t) I = n I f'(nt) I::; Cn.
Section 7.2 Stokes' Theorem
493
Define the COO function
where the product is taken over integers z such that I z I < 2Na + 1. Note that if I t I < a + l/4N
Z is chosen such that I t - z/2N I < l/4N. then fN(t - z{2N} =0 and I z I .:; 2N I t I + 1/2 <2Na+l. so that cp(t} =0. Similarly if Itl>a+2IN and Izl<2Na+l.then It-zJ2NI~ltl and
Z E
- I z I/2N > lIN so that cp(t} = 1. Finally. let Ito - a 1< 2IN and let za
E
Z be such that Ito - zrJ2N I <
lIN. All factors
fN(ta - z/2N} are one in a neighborhood of to. unless Ita - z/2N I.:; lIN. In that case we have the inequality I Z - za I .:; I z - 2Nto I + 12Nta -
Zo I .:; 3. Hence at most seven factors in the product ar.
not identically 1 in a neighborhood of ta' Hence I cp'(to} I':; 7CN =Ne .
...
7.2.23 Lemma Let K be a compact subset of CJp the singular set of f. For every e> 0 there is
a neighborhood U E of K in U x JR and a C~ function CPE: U x JR -+ [0. Il. which vanishes on a neighborhood of K in UE• is one on the complement of UE' and is such that i = 1•...• n. and
(ii) vol(UE}':; e and q(U E n Pf}':; e. where q is the measure on Pf associated with the volume form A. E nn-I(pf}' and vol(UE} is the Lebesgue measure of UE in JRn. Proof Partition JRn-1 by closed cubes D of edge length 42/3. 2.:; 1. At most 2n such cubes can meet at a vertex. The set 7t(K}. where 7t: U x JR -+ U is the projection. can be covered by finitely many open cubes C of edge length 22. each one of these cubes containing a cube D and having the same center as C. Since 7t(K} and K have measure zero. choose 2 so small that for given 15 > O. (i) the (n - I}-dimensional volume of Uj=l ..... LCj is smaller than or equal to 0; and (ii) q(7t- 1(U j=l ..... LCj) n Pf}':; 15. Since f is locally Holder and 7t(K} is compact. there exist constants 0 < u.:; 1 and k > 0
II x - y lin for x. y E 7t(K}. We can assume k
~ 1 without loss of I generality. In each of the sets 7t- (C j} = C j x JR. choose a box Pj with base Cj and height
such that I f(x} - f(y} I':; k
(2k2}l/n such that 7t(K) is covered by parallelipeds Pj' with the same center as Pj and edge lengths equal to two-thirds of the edge lengths of Pj' Let V = U j=l ..... LPj' Then 7t(V} = U j=l ..... LCj and since at most 2n of the Pj intersect vol(V} and
= 2k22n vol(7t(V}} .:; 2n+1k20':; 2n+1ko
494
Chapter 7 Integration on Manifolds
q(V
n Pr) ~ o.
By the previous lemma. for each Pi there is a COO function
~
[0. 1) that vanishes on
Pt. is equal to 1 on the complement of Pi' and sUPxeRnlla
xR
~
C~.
[0. 1) is
L
vanishes in a neighborhood of K and equals one in the
complement of V. But at most 2n of the Pi can intersect. so that
I I II, a IT I a
_
~
L
-
ax
j
i=1 ax
< n ._ - 2 A/~. J - 1•...• n.
k"i
Hence
Now let 0 = min!E. £/22n +1kA}.
Proof of Equation (1) Let n
00=
I, Wi dx l A ... A (dxl)" A ... A dx
n • dw=
bdx l A ... A dxn.
i=l and i*w = aA.. Then oi. b. and a are continuous and bounded on U x R and Pr respectively; i.e .• I oi(x) I ~ M. I b(x) I ~ N for x E U x R and I a(y) I ~ N for y E PC' where M. N > 0 are constants. Let U£ and
(2)
We have
and
I r_dOl Jrr
r_d(
Jrr
~ I r _bO -
~ N vol(U£) + M
I
A··· Adxnl +
is, J i=1 rr
[i
sup i=1 xeR
A
00
I
r
r
Wi
II a
Ia
(4)
495
Section 7.2 Stokes' Theorem
From (2)-(4) we get
for all
E > O.
which proves the equality . •
In analysis it can be useful to have hypotheses on the smoothness of strategy for sharper results is to approximate ~
00.
(J)
as well as on the
(J)
need only be C I . An effeclive
by smooth forms
0\ so that both sides of Stokes'
boundary that are as weak as possible. Our proofs show that theorem converge as k
(J)
A useful class of forms for which this works are those in Sobolev
spaces. function spaces encountered in the study of partial differential equations. The HOlder nature of the boundary of N in Stokes' theorem is exactly what is needed to make this approximation process work. The key ingredients are approximation properties in M (which arc obtained from those in JR.n) and the Calderon extension theorem to reduce approximations in N to those in JR.n. (Proofs of thcse facts may be found in Stein [1970]. Marsden [1973]. and Adams [1975].) ~
SUPPLEMENT 7.2C Stokes'Theorem on Chains
I@"
In algebraic topology it is of interest to integrate forms over images of simp1cxes. This box adapts Stokes' theorem to this case. The result could be obtained as a corollary of the piecewise smooth Stokcs Theorem. but we shall give a self-contained and independent proof. 7.2.24 Definition The standard p-simplex is the closed set
The vertices of ~p are the p+1 points Opposite to each Vi there is the i-thface
Vo
= (0 ..... 0)
VI
p--IJ: ~p--I ~ ~p
= O. O..... 0) ..... vp = (0 ..... O.
1).
given by (see Fig. 7.2.8):
and "'"
~p-I.i
Y ..... Yp--I) -_ ( YI ..... Yi-I • 0 • yi ..... yp-I);<. • IJ 1'" 0 .
(I
A Ck-singular p-simplex on a C'-manifold M. 1:5 k :5 r. is a Ck_map s: U ~ M. where U is an open neighborhood of ~p in JR.p. The points s(vo)' .... s(vp) are the vertices of the singular p-simplex s and the map SO P--IJ : V ~ M. for V an open neighborhood of ~p--I in
Chapter 7 Integration on Manifolds
496
and c;I>p-i,i extended by the same formula from ~i to V. is called the i-thface of the singular p-simplex s. A c!'-singular p-chain on M is afiniteformallinear combination with real coefficients of Ck-singular p-simp/exes. The boundary of a singular p-simp/ex s is the singular
lRP-i
(p-I)-chain
as
defined by
as
P
~ (_I)i S £..
=
0
c;I>
p-
i'
,I
i=O
and that of a singular p-chain is obtained by extending It is straightforward to verify that for j < i. If s: U ~ M. ~p
a a= 0 0
a from the simplexes by linearity to chains.
using the relation
c;I>p_ij
0
c;I>p-2j
=
c;I>p_2,i_i
0
c;I> p-2,i-l
C
U. is a singular p-simplex, ro
E
QP(M). and s"'ro = a dx i
/\ ... /\
dx P E QP(U). the integral of ro over s is defined by
f.
ro =
JL\ a dx i ... dxP•
where the integral on the right is the usual integral in
lRP.
The integral of ro over a p-chain is
obtained by linear extension.
~
__________________. .Vl
Figure 7.2.8 7.2.25 Stokes' Theorem on Chains If c is any singular p-chain and ro E Q?-i(M). then
i
dro
=
c
r roo
Jac
Proof By linearity it suffices to prove the formula if c = s, a singular p-simplex. If
*ro
S
=
L P
j=i
then
. i'
(_1)'1-
i
.1\
cff dx /\ ... /\ (dxJ )
/\ ... /\
dxP•
Section 7.2 Stokes' Theorem
d(s '" co) =
1·
aoY i dx I 1\'" i=1 ax
1\
497
dxP
and denoting the coordinates in a neighborhood V of ~I by (yl, ... , yp-I) = y, we get
and
Thus the formula in the statement becomes
~5,\
t 5 [(I
aoY(X) I P_ axi dx ... dx - i=1
p.
.
_
iip-I
'\.-1 oY 1 i=1 Y , Y , ... , Y
J
1
y ,
. I, ... , . I 0, y1,. ... , y p-I ) dy I ... dy p-I . - oY(y
(1)
By Fubini's theorem, each summand on the left hand side of (1) equals
J'\.-1[11-~/ =J
J
acd(x) . dx I ... dx p -'\ ax]
acd dxiJ dx 1 ... (i)" dx ... dx p ax]
"'i.
0
[cd(x l , ... , xi-I, 1-
'\.-1
I
xk, xi+l, ... , xp)
bi
- cd(x l , ... , xi-I, 0, xi+\ ... , xP)] dx l ... (dxi)" ... dxP. Break up this integral as a difference of two terms. In the first term perform the change of variables (y l , ... , Yp-I) 1-+ ( x 2, ... , xi-I , 1 - ""'k"i ~ xk,xi+1 , ... , xP) W h'ICh h as J aco b'Ian equaI t0 (1)1' , use the change of variables formula from calculus in the multiple integral involving the absolute value of the Jacobian, and note that x I = 1 - Li=l ..... p-I yi. In the second term perform lhe change of variables (yl, ... , yp-I) 1-+ (xl, ... , xH, xi+l, ... , xP) which has Jacobian equal to one. Then we get
J '\
acd(x) i dx I ... dxP = ax -
J [-j( UT
'\.-1
1-
~ L.- yi, y , ... , yP-IJ 1
i=1
coi(yl, ... , yi-I, 0, yi, ... , yP-I)] dyl ... dyp-I
and formula (1) is thus proved for each corresponding summand. _
498
Chapter 7 Integration on Manifolds
Instead of singular p-chains one can consider infinite singular p-chains defined as infinite fonnal sums with real coefficients
L j E I ajS j
aj ~ o} is locally finite, i.e., each m
E
such that for each i E I the family of sets (Sj('\)
I
M has a neighborhood intersecting only finitely many (or
no) sets of this family. On compact manifolds only finitely many coefficients in an infinite singular p-chain are non-zero and thus infinite singular p-chains are singular p-chains. The statement and the proof of Stokes' theorem on chains remain unchanged if c is an infinite p-chain and nlt-I(M) has compact support. fl:n
W E
Exercises 7.2A Let M and N be oriented n-manifolds with boundary and f: M
~
N an oricntation-
preserving diffeomorphism. Show that the change-of-variables fonnula and Stokes' theorem imply that
r*
0
d
=d
0
f*.
7.2D Let M be a compact orientable boundaryless n-manifold and a
E
nn-I(M). Show that da
vanishes at some point. 7.2C Let M be a compact (n + I)-dimensional manifold with boundary, f: aM map and WE nn(N) where
dw = O.
Show that if f extends to M, then
7.2D Let (M.Il) be a volume manifold with aM
~
N a smooth
faMr*w = O.
=0.
(i) Show that the divergence of a vector field X is uniquely detennined by the condition
J.Mf(div X)1l = - J,M(LxDIl
(*)
for any f with compact support. (ii) What docs (*) become if M is compact with boundary? (iii) X(x. y, z) =(y, -x. 0) defines a vector field on S2. Calculate div X. 7.2E Let M be a paracompact manifold with boundary. Show that there is a positive smooth function f: M ~ [0, oo[ with 0 a regular value. such that aM = r-I(O). (Hint: First do it locally and then patch the local functions together with a partition of unity.) 7.2F Let M be a boundaryless manifold and f: M ~ 1R a C'" mapping having a regular value a. Show that r-1([a,ooD is a manifold with boundary r-I(a). 7.2G Let f: M ~ N be a C'" mapping, aM ~ 0, aN ~ 0, and let peN be a submanifold of N. Assume that f hold.
mP, (fl aM) mP
and that in addition one of the following conditions
499
Section 7.2 Stokes' Theorem
(i) P is boundaryless and P
C
Int N; or
(ii) ap *- 0 and ap c aN; or (iii) ap *- 0, f map, and (f I aM) map. Show that f-I(P) is a submanifold of M whose boundary equals af-I(p) = f-I(p)
n aM,
in case (i), and af-I(p) = f-I(ap) in cases (ii) and (iii). If all manifolds are finite dimensional, show that dim M - dim r-I(P)
=dim N -
dim P. Formulate and prove the
statement replacing this equality between dimensions for infinite dimensional manifolds. (Hint: At the boundary, work with a boundary chart using the technique in the proof of
3.5.11.) 7.2H Without some kind of transversality conditions on f I aM for f: M ~ N a smooth map, even if f m P, where P is a submanifold of N (like the ones in the previous exercise), f-I(P) is in gencral not a submanifold. For example, let M = JR2 +' N = JR, P = {OJ and f(x, y) = y + X(x) for a smooth function X : JR ~ R Show that f is a smooth surjective submersion. Find the conditions under which f I aM has 0 as a regular value. Construct a smooth function X for which these conditions are violated and f-I(O) is not a manifold.
(Hint: Take for X a smooth function which has infinitely many zeros converging to zero.) 7.21
(i)
Show that if M is a boundary less manifold, there is a connected manifold N with M = aN. (Hint: Think of semi-infinite cylinders.)
(ii) Construct an example for (i) in which M is compact but N cannot be chosen to be compact. (Hint: Assume dim M = 0.) 7.2J Let M be a manifold, X a smooth vector field on M with flow FI and a. E nk(M). We call a. an invariant k-form of X when Lxa. = O. Prove the following. Poincare-Cartan Theorem a. is an invariant k-form of X ifffor all oriented compact k-manifolds with boundary (V, aV) and CO mappings
Iv
of FI contains
Chapter 7 Integration on Manifolds
500
7.2L Let X be a vector field on a manifold M with flow Ft and a.
E
Qk(M). Then a. is called
a relatively invariant k-form of X if Lxa. is closed. Prove the following
Poincare-Cartan Theorem
a. is a relatively invariant (k-l)-form of X ifffor all
oriented compact k-manifolds with boundary (V. oV) and COO maps q>: V the domain of F t contains q>(V) for 0
~
~
M such that
t ~ T. we have
jav (FtOq>Oi)*a.=jav (q>oi)*a.. where i : oV
~
V is the inclusion map.
E X(M). let 5l. x be the set of all invariant forms of X. 'R.x the set of all relatively invariant forms of X. C the set of all closed forms in Q(M). and 'E the set of all exact
7.2 M If X
forms in Q(M). Show that (i) 5l.x c ~. 'E c C c real vector subspace.
'R.x.
5l. x is a differential sub algebra of Q(M). but
'R.x is only a
i LX 1t (ii) 0 -75l. x -7 Q(M) -7 Q(M) -7 Q(M)/lm(L x ) -7 0 is exact. i
d
1t
(iii) 0 -7 C --7 ~ -7 5l.x -75l. xI'E n 5l.x -7 0 is exact. (iv) d(5l.x ) c 5l.x and i x (5l. x ) c 5l. x .
Theorem for manifolds with boundary: Let M and N be Ck manifolds, where M is Lindelof, having a boundary oM, and N is boundaryless. Let f:
7.2N Smale-Sard
M ~ N be a Ck Fredholm map and let of = f 10M. If k > max(O. index(T xf) for every
x E M. show that '1(r n 'l\.ar is residual in N. 7.20 The Boundaryless Double Let M be a manifold with boundary. Show that the topological space obtained by identifying the points of oM in the disjoint union of M with itself is a boundary less manifold in which M embeds. called the boundaryless double of M. (Hint: Glue together the two boundary charts.) 7.2P Let M be a manifold with oM
oF
0. Assume M admits partitions of unity. Show that oM
is orientable and hence by Exercise 6.5Q. the algebraic normal bundle v(oM) = (TM I oM)!f(oM) is trivial. (Hint: Use Proposition 6.5.8. Locally n(m) = %x" for m
E
oM;
glue these together.) 7.2Q Collars Let M be a manifold with boundary. A collar for M is a diffeomorphism of oM x [0. 1[ onto an open neighborhood of aM in M that is the identity on aM. (i)
Show that a manifold with boundary and admitting partitions of unity has a collar.
(Hint: Via a partition of unity. construct a vector field on M that points inward when restricted to aM. Then look at the integral curves starting on aM to define the collar.)
Section 7.2 Stokes's Theorem
(ii)
501
Let
isotopic, i.e., there is a smooth map H:] - 10, 1 + E[ x aM x [0, 1[ that H(O, m, t) = <po(m, t), H(l, m, t) =
E
~
M such
aM x [0, 1[ and
that H(s,', -) is an embedding for all s E ]-10, 1 + 10[. (Hint: Let Ui be the image of
=N n aM
and TnN is not a subset of
Tn(aM) for all n E aN. Show that aM has a collar which restricts to a collar of aN in N.
7.2R Let M and N be manifolds with boundary and let
Use collars to construct a homeomorphism of a neighborhood U of V with the space ]- I, 1[ x V which maps V pointwise to V x {O} and which maps V n M and V n N diffeomorphically onto V x ]0, 1 [ and V x ] - I, O[ respectively. Construct a differentiable structure out of those on M, N, and U.
The "uniqueness theorem of glueing" states that the differentiable structures on the space M U", N obtained in (i) by making various choices are all diffeomorphic. The rest of this exercise uses this facl. (ii)
Two compact boundary less manifolds M I' M2 are called cobordant if there is a compact manifold with boundary N, called the cobordism, such that
aN
equals
the disjoint union of MI with M2. Show that "cobordism" is an equivalence relation. (Hint: For transitivity, glue the manifolds along one common component of their boundaries.) (iii) Show that the operation of disjoint union induces the structure of an abelian group on the set of cobordism classes in which each clement has order two. (Hint: The zero element is the class of any compact manifold which is the boundary of another compact manifold.) (iv) Show that the operation of taking products of manifolds induces a multiplicative law on the set of cobordism classes, thus making this set }} a ring. (v) Repeat parts (ii) and (iii) for oriented manifolds obtaining a gradcd ring (that is, [M, ~l·[N, v] = (_I)dim M + dim I\'[N, v]·[M, /.1]), the graded ring of oriented cobordism classes ®. Are the clements of ® still of order two relative to addition? = (vi) Denote by }}n, ®n, the cobordism classes of a given dimension. Show that
»0
Chapter 7 Integration on Manifolds
S02
7J./27J., ®o = Z, (vii) Assume M
:n l = ®I = O.
and
N are boundary less manifolds,
M
compact,
and
P
a
boundary less submanifold of N. Assume f, g : M -t N are smoothly homotopic maps such that f
mP
and g
mP.
Show that II(P) and g-I(p) are cobordant.
(Hint: Choose a smooth homotopy H transverse to P. What is (JH-I(p)?) 7.2S
(i) Let X be a vector field density on a finite dimensional manifold M, i.e., X = X ® P for X E X(M) and pEl n(M)
I. Recall from exercise 6.SP that the density
div X, defined to be (div pX)p, is independent of the representation of X as X ® p. Show that for any f
E
.'F(M), Y
E
X(M), and any diffeomorphism
M wehave
nk-I(M) and 1 be a tensor density of type (k, 0) which is completely
E
anti symmetric. Let a· 1 denote the contraction of a with the first k-\ indices of the tensor part of 1 producing a vector field density. Defme the contravariat exterior
derivative
(ft by requiring the following relation for all a
E
Qk-I(M):
div(a . 1) = da· 1 + a· (ft, where da· 1 and a· (ft means contraction on all indices. Show that if 1 = t ® p, where locally
then the local expression of (ft is a j .. j ) ' a a1 = - . (t 1 k-1) - . ax) axIl Show Ihal
a
/\ ... /\ - . -
I n ® 1 dx /\ ... /\ dx
1
ax'k-l
iJ2 = O.
(iv) Prove the following properties of a: a(1/\ <J)
= aT /\ <J + (-lf1 /\ a<J,
and Lxift
= aLxT,
where 1 is a (k , D)-tensor density, <J is a (~, D)-tensor density, X
E
X(M), and
ax =
Section 7.2 Stokes's Theorem
503
div X. (v) Let
h' =
X /\, for X
nk+I(M), X
E
E
X(M). Show that ixa·, = a . h' for any a
E
X(M) and, is a completely anti symmetric (k, 0) tensor density.
Prove the analog of Cartan's formula:
Lx = h
0
a +a
0
h.
(Hint: Integrate the
defining relation in (iii) for a any form with support in Int M and extend the formula by continuity to aM.) (vi) Formulate and prove a global formula for aL(al' ... , ak-I)' ,
a (k, OJ-tensor
density, analogous to Palais' formula 6.4.11(ii). 7.2T Prufer Manifold Let lRd denote the set lR with the discrete topology; it is thus a zero dimensional manifold. Let P=(lR2+xlR d)!R, where lR2+= {(x,Y)IY~O) and R is the equivalence relation: (x, y, a) R (x', y', a') iff [(y = y' > 0 and a + xy = a' + x' y') or (y = y' = 0 and a = a', x = x')]. Show that P is a Hausdorff two dimensional manifold, ap oF
0, and ap is a disjoint union of uncountably many copies of lR. Show that P is not paracompact. 7.2U In the notation of Supplement 7.2C, verify that a
0
a=
o.
~7.3
The Classical Theorems of Green, Gauss, and Stokes
This section obtains these three classical theorems as a consequence of Stokes' theorem for differential forms. We begin with Green's theorem, which relates a line integral along a closed piecewise smooth cW"Ve C in the plane 1R2 to a double integral over the region D enclosed by C. (Piecewise smooth means that the cW"Ve C has only finitely many comers.) Recall from advanced calculus that the line integral of a one-form [a, b] ~ 1R2 is defined by
0)
=P dx + Q dy
along a cW"Ve C parametrized by y:
7.3.1 Green's Theorem Let D be a closed bounded region in 1R2 bounded by a closed positively oriented piecewise smooth curve C. (Positively oriented means the region D is on your left as you traverse the curve in the positive direction.) Suppose P: D CI. Then
~ IR
and Q : D
~ IR
are
Proof We assume the boundary C = dD is smooth. (The piecewise smooth case follows from the generalization of Stokes' theorem outlined in Supplement 7.28). Let 0) = P(x, y)dx + Q(x, y)dy E nl(D). Since dO) = (dQ/dx - dP/dy) dx 1\ dy and the measure associated with the volume dx 1\ dy on 1R2 is the usual Lebesgue measure dx dy, the formula of the theorem is a restatement of Stokes' theorem for this case. _ This theorem may be phrased in terms of the divergence and the outward unit normal. If C is given parametrically by t 1-+ (x(t), y(t», then the outward unit normal is n = (y'(t), - x'(t))/J x'(t)2 + y'(t)2
and the infinitesimal arc-length (the volume element of C) is ds =
(1)
Jx'(t)2 + y'(t)2 dt.
(See Fig.
7.3.1.) If X = Pd/dx + Qd/oy E 3t(D), recall that div X = *d*)(f> = dP/dx + dQ/dy. 7.3.2 Corollary Let D be a region in 1R2 bounded by a closed piecewise smooth curve C. If X 3t(D), then
E
Ie (X· n) ds = I ID (div X) dx dy.
Section 7.3 The Classical Theorems o/Green, Gauss, and Stokes
where
505
Ie f ds denotes the line integral of the function f over the positively oriented curve C and
X • n is the dot product.
/)
Figure 7.3.1 Proof Using formula (I) for n we have
fe
(X • n) ds =
f
[P(x(t)y(t»y'(t) - Q(x(t), y(t»x'(t)]dt =
f/
dy - Q dx
(2)
by the definition of the line integral. By 7.3.1, (2) equals
Taking P(x, y) = x and Q(x, y) = y in Green's theorem, we get the following. 7.3.3 Corollary Let D be a region in
JR.2
area of D is given by area(D) =
bounded by a closed piecewise smooth curve C. The
±Ie
x dy - y dx.
The classical Stokes theorem for surfaces relates the line integral of a vector field around a simple closed curve C in
JR.3
to an integral over a surface S for which C =
advanced calculus that the line integral of a vector field X in is defined by
LX. ds
=
f
JR.3
X(o(t» • o'(t)dt.
over the curve
as. 0:
Recall from [a, b] ~ JR.3
(3)
The surface integral of a compactly supported two-form co in JR.3 is defined to be the integral of the pull-back of co to the oriented surface. If S is an oriented surface, n is called the outward unit normal at XES if n is perpendicular to TxS and (n, el' e 2 ) is a positively oriented basis of JR.3 whenever (e l , e 2 ) is a positively oriented basis of TxS. Thus S is
Chapter 7 Integration on Manifolds
506
orientable iff the normal bundle to S, which has one-dimensional fiber, is trivial. In particular, the area element v of S is given by 6.5.8. That is (4)
and ~
=0
IsW in a form familiar from vector calculus.
dx /\ dy /\ dz. We want to express
P dy /\ dz + Q dy /\ dz + R dx /\ dy so that w =0
x Recall that a /\ *P
=0
=0
* Xb,
Let
W =0
where
a a a Pax +Qay +R az '
(a, P> ~ so that letting a
=0
n b, and P =0 Xb, we get n b /\ *Xb
=0
(X • n)~.
Applying both sides to (n, vi' v2 ) and using (2) gives
(5) (the base point x is suppressed). The left side of (4) is *Xb( vi' v2 ) since n b is one on n and zero on
VI
and v 2 . Thus (4) becomes
* Xb Therefore
=0
(X • n)v .
(6)
Is w Is (X • n) dS Is X • dS, =
=
where dS, the measure on S defined by v, is identified with a surface integral familiar from vector calculus. A physical interpretation of
Is (X· n)dS
may be useful. Think of X as the velocity field
of a fluid, so X is pointing in the direction in which the fluid is moving across the surface S and X· n measures the volume of fluid passing through a unit square of the tangent plane to S in unit
Is (X· n)dS is the net quantity offluidflowing across the sUrface per unit time, i.e., the rate offluidflow. Accordingly, this integral is also called the flux of X across the time. Hence the integral
surface. 7.3.4 Classical Stokes Theorem Let S be an oriented compact surface in IR3 and X a C l
vector field defined on S and its boundary. Then
r (curl X) • n dS
Js
=0
J as
X· ds.
where n is the outward unit normal to S (Fig. 7.3.2). Proof First extend X via a bump function to all of 1R3 so that the extended X still has compact
Section 7.3 The Classical Theorems of Green, Gauss, and Stokes
support. By definition,
Ias X • ds = Ias Xb
by the standard metric in
]R3.
where b denotes the index lowering action defined
= *(curl
But dXb
507
X)b (sec 6.4.3(C» so that by (3), (6), and
Stokes' theorem,
f
as
X· ds
x
Figure 7.3.2 7.3.5 Examples A The historical origins of Stokes' formula (6) arc connected with Faraday's law, which is
discussed in Chapter 8 and examble B below. In fluid dynamics, (5) is related to Kelvin's circulation theorem, to be discussed in §8.2. Here we concentrate on a physical interpretation of the curl operator. Suppose X represents the velocity vector field of a fluid. Let us apply Stokes' theorem to a disk Dr of radius r at a point P
f
aD,
X· ds
=
f
aD,
E ]R3
(Fig. 7.3.3). We get
(curl X) • n ds
= (curl X • n)(Q)1tf2,
the last equality coming from the mean value theorem for integrals; here Q E Dr is some point given by the mean value theorem and 1tf2 is the area of Dr' Thus ((curl X) • n)(P)
=
lim _1_ r-->O 1tr2
The number
Ie X • ds
f
aD,
X· ds.
(7)
is called the circulation of X around the closed curve C. It
represents the net amount of turning of the fluid in a counterclockwise direction around C.
,~" r
Figure 7.3.3
Chapter 7 Integration on Manifolds
508
Formula (7) gives the following physical interpretation for curl X, namely: (curl X) • n is the circulation of X per unit area on a sUrface perpendicular to n. The magnitude of (curl X) • n
is clearly maximized when n = (curl X)/II curl X II. Curl X is called the vorticity vector. B One of Maxwell's equations of electromagnetic theory states that if E(x, y, z, t) and H(x, y, z) represent the electric and magntic fields at time t, then V x E
=- aH/at,
where V x
E is computed by holding t fixed, and aH/at is computed by holding x, y, and z constant. Let us use Stokes' theorem to determine what this means physically. Assume S is a surface to which Stokes' theorem applies. Then
(The last equality may be justified if H is ct.) Thus we obtain (8)
Equality (8) is known as Faraday's law. The quantity
fas E • ds
represents the "voltage" around
as, and if as were a wire, a current would flow in proportion to this voltage. Also
fs H • dS
is
called theflux of H, or the magnetic flux. Thus, Faraday's law says that the voltage around a loop equals the negative of the rate of change of magnetic flux through the loop.
C Let X E X(1R 3 ). Since
is contractible, the Poincare lemma shows that curl X = 0
]R3
iff X = grad f for some function f
E
.'F(]R3).
This in tum is equivalent (by Stokes' theorem) to
fe
either of the following: (i) for any oriented simple closed curve C, X • ds = 0, or (ii) for any oriented simple curves C t , C2 with the same end points Ie 1X • ds = Ie 2 X • ds. The function f can be found in the following way:
f(x, y, z) =
I
x
0
X t (t, 0, O)dt +
lY X (x, t, O)dt + lZX (x, y, t)dt. 0
2
0
3
Thus, for example, if
X
then curl X =
°
a a a y ax + (z cos(yz) + x) ay + y cos(yz) az'
and so X = grad f, for some f. Using the formula (9), one finds f(x, y, z) = xy + sin yz.
D The same arguments apply in
X
]R2
by the use of Green's theorem. Namely if
(9)
Section 7.3 The Classical Theorems o/Green, Gauss, and Stokes
509
then X = grad f, for some f E .'J(JR.2) and conversely. E The following statement is again a reformulation of the Poincare lemma: let X then div X = 0 iff X = curl Y for some Y 7.3.6 Classical Gauss Theorem
E
E
X(JR.3) ,
X(lRh •
Let 0 be a compact set with nonempty interior in JR.3
bouruied by a surface S that is piecewise smooth. Thenfor X a C 1 vector field on 0 US,
fn (div
fs (X
X) dV =
0
(10)
n)dS,
where dV denotes the staruiard volume element (Lebesgue measure) in JR.3 (Fig. 7.3.4). Proof Either use 7.2.10 or argue as in 7.3.4. By (6),
By Stokes' theorem, this equals
since d*Xb = (div X) dx /\ dy /\ dz. •
X
n
Figure 7.3.4 7.3.7 Example
We shall use the preceding theorem to prove Gauss' law
1n
ron -dS r3
-
j
41t, if 0 E 0
0,
if 0
where 0 is a compact set in JR.3 with nonempty interior.
~0
ao
(11)
is the surface bounding O. which
Chapter 7 Integration on Manifolds
510
is assumed to be piecewise smooth. n is the outward unit normal. 0 i"
n.
If 0 i"
o inside n
(In.
and where
apply 7.3.6 and the fact that div(r/r3) = 0 to get the result. If 0 E
n.
surround
by a ball DE of radius £ (Fig. 7.3.5). Since the orientation of (l0E induced from
\ DE is the opposite of that induced from DE (namely it is given by the
n
inward unit normal). Gauss'
theorem gives (12)
since 0 i"
n \ DE
and thus on
JaD since
J
n \ DE'
div(r/r3) = O. But on (l0E' r = £ and n = -r/£. so that
J
E
2 1 2 P3ndS=£4dS =--41t£ =-41t r aDE £ £2
dS = 41t£2. the area of the sphere of radius £.
aDE
aD,.--....-.---.•
~
•
. . . . . . t.:. . .. .•....
Figure 7.3.5 In electrostatics Gauss' law (II) is used in the following way. The potential due to a point E JR.3 is given by q/(4m). r = (x 2 + y2 + z2)l!2. The corresponding electric field is defined to be minus the gradient of this potential; i.e .. E = qr/(4m3). Thus Gauss' law states that charge q at 0
fanE. n
the total electricjlux dS equals q if 0 E n and equals zero, if 0 i" n. A continuous charge distribution in n described by a charge density p is related to E by P
fan
fn
= div E. By Gauss' theorem the electric flux E • n dS = p dV. which represents the total charge inside n. Thus. the relationship div E = P may be phrased as follows. The flux out of II swface of an electric field equals the total charge inside the swface.
Section 7.3 The Classical Theorems of Green, Gauss, and Stokes
511
Exercises 7.3A Use Green's theorem to show that (i) The area of the ellipse x 2/a2 + y2jb2 == 1 is 1tab. (ii) The area of the hypocycloid x == a cos 3 9, y == b sin 3 9 is 31ta2/8. (iii) The area of one loop of the four-leaved rose r == 3 sin 29 is 91t/8. 7.3B Why does Green's theorem fail in the unit disk for -y dx/(x 2 + y2) + X dy/(x 2 + y2)? 7.3C For an oriented surface S and a fixed vector a, show that
2
r a. n dS == Jras(a x r). dS.
Js
7.3D Let the components of the vector field X
E
3\(]R3) be homogeneous of degree one; i.e., X
satisfies Xi(tx, ty, tz) == tXi(x, y, z), i == 1,2, 3. grad f, where f == (xXi + yX 2 + zX3)/2. 7.3E Let S be the surface of a region
n
Show that if curl X == 0, then X ==
in ]R3. Show that
volume(n) = (1/3)
Lr . n dS.
Give an intuitive argument why this should be so. (Hint: Think of cones.) 7.3F Let S be a closed (i.e., compact boundaryless) oriented surface in ]R3. (i) Show in two ways that fs(curl X) • n dS == O. (ii) Let X = 3\(S) and f E :J(S). Make sense of and show that
Is (grad DX • dS -Is(f curl X)dS =
where grad, curl, and div are taken in ]R3. 7.3G Let X and Y be smooth vector fields on an open set DC ]R3, with aD smoolh, and c1(D) compact. Show that
r Y • curl X dV Jr X • curl Y dV + Jr (X x Y) • n
JD
=
D
dS.
aD
where n is the outward unit normal to dD and dS the induced surface measure on dD. (Hint: Show that y. curl X - X • curl Y == div(X x Y).)
512
Chapter 7 Integration on Manifolds
7.3H If C is a closed curve bounding a surface S show that f/(grad g) • ds =
Is (grad f x grad g) • n dS
= -
fcg(grad 0 • ds
where f and g are C2 functions. 7.31 (A. Lenard) Faraday's law relates the line integral of the electric field around a loop C to the surface integral of the rate of change of the magnetic field over a surface S with boundary C. Regarding the equation V x E = - dH/dt as the basic equation, Faraday's law is a
consequence of Stokes' theorem, as we have seen in Example 7.3.5B. Suppose we are given electric and magnetic fields in space that satisfy the equation V x E = - dH/dt. Suppose C is the boundary of the Mobius band shown in Fig. 6.5.1. Since the Mobius band cannot be oriented, Stokes' theorem does not apply. What becomes of Faraday's law? Resolve the issue in two ways: (i) by using the results of Supplement 7.2A or a direct reformulation of Stokes' theorem for nonorientable surfaces, and (ii) realizing C as the boundary of an orientable surface. If dH/dt is arbitrary, in general does a current flow around C or not?
~7.4
Induced Flows on Function Spaces and Ergodicity
This section requires some results from functional analysis. Specifically we shall require a
knowledge of Stone's theorem and self-adjoint operators. The required results may be found in Supplement 7.4A and 7.4B at the end of this section. Flows on manifolds induce flows on tangent bundles, tensor bundles, and spaces of tensor fields by means of push-forward. In this section we shall be concerned mainly with the induced flow on the space of functions. This induced flow is sometimes called the Liouville flow. Let M be a manifold and Jl a volume element on M; i.e., (M, Jl) is a volume manifold. If F t is a (volume-preserving) flow on M, then F t induces a linear one-parameter group (of isometrics) on the Hilbert space H = L2(M Jl) by
The association of V t with F t replaces a nonlinear finite-dimensional problem with a linem
infinite-dimensional one. There have been several theorems that relate properties of Ft and V t . The best known of these is the result of Koopman [1931], which shows that V t has one as a simple eigenvalue for all t if and only if F t is ergodic. (If there are no other eigenvalues, then F t is called weakly mixing.) A few basic results on ergodic theory are given below. We refer the reader to the excellent texts of Halmos [1956], Arnold and Avez [1967], and Bowen [1975) for more information. We shall first present a result of Povzner [1966], which relates the completeness of the flow of a divergence-frce vector field X to the skew-adjointness of X as an operator. (The hypothesis of divergence free is removed in Exercises 7.4A-C.) We begin with a lemma due to Ed Nelson. 7.4.1 Lemma Let A be an (unbounded) self-adjoint operator on a complex Hilbert space '}{. Let Do
C
D(A) (the domain of A) be a dense linear subspace of '}{ and suppose V t = e itA (the
unitary one-parameter group generated by A) leaves Do invariant. Then Ao:= (A restricted to Do) is essentially self-adjoint; that is, the closure of
Ao
is A.
Proof Let A denote the closure of Ao. Since A is closed and extends A o' A extends A. We
need to prove that A extends A. For A> 0, A - iA is surjective with a bounded inverse. First of all, we prove that A - iAo has dense range. If not, there is a v
E
'}{
such that
(v, AX - iAox) = 0 for all x E Do.
Chapter 7 IntegratWn on Manifolds
514
In particular, since Do is VI-invariant,
so
Since Do is dense, this holds for all x E J{. Thus II VI II = 1 and A. > 0 imply v = O. Therefore (A. - iAu)-' makes sense and (A. - iAt' is its closure. It follows from Supplement 7.4A that A is the closure of Au (see 7.4.15 and the remarks following it) .• 7.4.2 Proposition
Let X be a
C~
divergence-free vector field on (M, 11) with a complete flow ~c
Fl' Then iX is an essentially self-adjoint operator on in the complex Hilbert space L 2(M, 11).
= ~ functions with compact support
Proof Let Vtf = f 0 F-I be the unitary one-parameter group induced from Fl' A straightforward convergence argument shows that V/ is continuous in t in L2(M,I1). In Lemma 7.4.1, choose Do = C~ functions with compact support. This is clearly invariant under VI' If f E Do' then
.!V dt If I
1=0
=.!f dt 0 F-I
I 1=0
= -
df· X '
so the generator of VI is an extension of -X (as a differential operator) on Do. The corresponding essentially self-adjoint operator is therefore iX . • Now we prove the converse of 7.4.2. That is, if iX is essentially self-adjoint, then X has an almost everywhere complete flow. This then gives a functional-analytic characterization of completeness. 7.4.3 Theorem Let M be a manifold with a volume element 11 and X be a c~ divergence-free vector field on M. Suppose that, as an operator on L2(M, 11), iX is essentially self-adjoint on the c~ functions with compact support. Then, except possibly for a set of points x of measure zero, the flow FI(x) of X is defined for all t E R
We shall actually prove that, if the defect index of iX is zero in the upper half-plane (i.e., if (iX + i)(C~c) is dense in L2), then the flow is defined, except for a set of measure zero, for all t> O. Similarly, if the defect index of iX is zero in the lower half-plane, the flow is essentially complete for t < O. The converses of these more general results can be established along the lines of the proof of 7.4.1. Proof (E. Nelson-private communication) Suppose that there is a set E of finite positive measure such that if x E E, FI(x) fails to be defined for t sufficiently large. Let Er be the set of x E E
Section 7.4 Induced Flows on Function Spaces and Ergodicity
for which Ft(x) is undefined for t ~ T. Since E Replacing E by
~,
515
=
UT:1:i Ep some ET has positive measure. we may assume that all points of E "move to infinity" in a time '5T.
If f is any function on M, we adopt the convention that f(Ft(x» = 0 if Ft(x) is undefined. For any x
M, and t < -T, Ft(x) must be either in the complement of E or
E
undefined; otherwise it would be a point of E that did not move to infinity in time T. Hence XE(Ft(x» = 0 for t < - T, where XE is the characteristic function of E. We now define a function on M by
Note that the integral converges because the integrand vanishes for t < - T. In fact, we have
Moreover, g is in L2. Indeed, because F t is measure-preserving, where defined, denoting by II 112 the L 2 norm, we have II XE ° F, 112 '5 II XE 11 2, so that
IIglb '5
i~
e--'tIiXEoF,lbd't '5 IIXElbeT .
-T
The function g is nonzero because E has positive measure. Fix a point x this case F/Ft(x»
E
M. Then Ft(x) is defined for t sufficiently small. It is easy to see that in
and Fnt(x) are defined or undefined together, and in the former case they are
equal. Hence we have XE(F,(Ft(x» = XE(Fnt(x»
for t sufficiently small. Therefore, for t
sufficiently small
Now if
f
C~
with compact support, we have
g(x)X[
=
lim
t->O
= lim
t->O
f f~ g(x)
1
-t
t
dll
=
g(x)
f
t
lim
f
g(F_t(x») - g(x)
t->O
t
g(x)
These equalities are justi tied because on the support of
Chapter 7 Integration on Manifolds
516
Methods of functional analysis applied to L 2(M, Il) can, as we have seen, be used to obtain theorems relevant to flows on M. Related to this is a measure-theoretic analogue of the fact that any automorphism of the algebra .1{M) is induced by a diffeomorphism of M (see Supplement 4.2C). This result, due to Mackey [1962]. states that if V t is a linear isometry on L2(M, Il), which is multiplicative (i.e., Vt(fg) = (V/)(Vtg), where defined), then V t is induced by some measure preserving flow Ft on M. This may be used to give another proof of theorem 7.4.3. A central notion in statistical mechanics is that of ergodicity; this is intended to capture the idea that a flow may be random or chaotic. In dealing with the motion of molecules, the founders of statistical mechanics, particularly Boltzmann and Gibbs, made such hypotheses at the outset. One of the earliest precise defintions of randomness of a dynamical system was minimality: the orbit of almost every point is dense. In order to prove useful theorems, von Neumann and Birkhoff in the early 1930s required the strong assumption of ergodicity, defined as follows. 7.4.4 Definition Let S be a measure space and Ft a (measurable) flow on S. We call Ft
ergodic if the only invariant measurable sets are 0 and all of S. Here, invariant means Ft(A) = A for all t E IR and we agree to write A = B if A and B differ by a set of measure zero. (It is not difficult to see that ergodicity implies minimality if we are on a second countable Borel space.) A function f: S
~
IR will be called a constant of the motion if f
0
Ft = f a.e. (almost
everywhere) for each t E R 7.4.5 Proposition Aflow Ft on S is ergodic iff the only constants of the motion are constant a.e. Proof If Ft is ergodic and f is a constant of the motion, the two sets {x E S I f(x) ~ a} and {x S I f(x) .:; a} are invariant, so f must be constant a.e. The converse follows by taking f to be a
E
characteristic function. • The first major step in ergodic theory was taken by J. von Neumann [1932], who proved the mean ergodic theorem which remains as one of the most important basic theorems. The setting is in Hilbert space, but we shall see how it applies to flows of vector fields in 7.4.7. 7.4.6 Mean Ergodic Theorem Let H be a real or complex Hilbert space and V t : H
~
H a
strongly continuous one-parameter unitary group (i.e., V t is unitary for each t, is aflow on H andfor each x E H, t 1-+ Vtx is continuous). Let the closed subspace flo be defined by
flo = {x E
HI Vtx
=~
for all t E 1R}
Section 7.4 Induced Flows on FunctWn Spaces and Ergodicity
517
and let lP' be the orthogonal projection onto Ho. Thenfor any x E H,
tl~ I +J~ Usx ds -lP' x II
=
o.
The point av(x) so defined is called the time average of x.
Proof of 7.4.6 (F. Riesz [1944]) We must show that
If lP'x = x, this means x
E Ho' so U,(x) = x; the result is clearly true in this case. We can therefore suppose that lP'x = 0 by decomposing x = lP'x + (x -lP'x). Note that
where.l denotes the orthogonal complement. This is an easy verification using unitarity of U t and Ut- I = U-t. It follows that ker lP' is the closure of the space spanned by elements of the form Us y - y. Indeed ker lP' = Hl, and if A is any set in H, and B = A.l, then B.l is the closure of the span of A. Therefore, for any
E
> 0, there exists t l ,
... ,
tn and xI' ... , xn such that
It follows from this, again using unitarity of U t' that it is enough to prove our assertion for x of
the form U,y - y. Thus we must establish 1
lim -
t .... ~ t
J,t Us(U,y 0
y) ds = O.
For t > 't we may estimate this integral as follows:
11+ E(UsU.Y - Usy)ds II = 11- +fa U,(y)ds + +fH U,(y)ds "
To apply 7.4.6 to a measure-preserving flow Ft on S, we consider the unitary one-parameter group Ut(f) = f 0 F t on L2(S, ~). We only require a minimal amount of continuity on F t here, namely, we assume that if s ~ t, F,(x) ~ Ft(x) for a.e. XES. We shall also
518
a<;sume
Chapter 7 IntegratWn on Manifolds
~(s)
<
00
for convenience. Under these hypotheses, U( is a strongly continuous unitary
onc-parameter group. The verification is an exercise in the use of the dominated convergence theorem. 7.4.7 Corollary In the hypotheses above F( is ergodic average av(f)(x)
=
11( (f
lim -
t----..±oo t
0
0
if and only iffor each
f
E
L2(S) its time
Fs)(x) ds
(the limit being in the L2-mean) is constant a.e. In this case the time average av(f) necessarily equals the space average
Is fdj.l/~(S)
a.e.
Proof Ergodicityof F( is equivalent by Proposition 7.4.5 to dim Ho = 1, where Ho is the closed subspace of L2(S) given in theorem 7.4.6. If dim Ho = 1, P(f) =
Is fd~/~(S)
so the
equality of av(f) with P(f) a.e. is a consequence of Theorem 7.4.6. Conversely if any f E L 2(S) has a.e. constant time average av(f) then taking f to be a constant of motion, it follows that f = av(f) is constant a.e. Therefore dim Ho = 1. • Thus, if F( is ergodic, the time average of a function is constant a.e. and equals its space average. A refinement of this is the individual ergodic theorem of G. D. Birkoff [1931], in which one obtains convergence almost everywhere. Also, if ~(S) =
00
but f E LI(S) n L2(S), one still
concludes a.e. convergence of the time average. (If f is only L2, mean convergence to zero is still assured by 7.4.5.) Modem work in dynamical systems, following the ideas in §4.3, has shown that for many interesting flows arising in the physical sciences, the motion can be "chaotic" on large regions of phase space without being ergodic. Much current research is focused on trying to prove analogues of the ergodic theorems for such cases. (See, for instance, Guckenhcimer and Holmes [1985] and Eckmann and Ruelle [1985] and references therein.) A particularly important example of an ergodic flow is the irrational flow on the torus.
'It"" given by F(([
~
7.4.9 Proposition The linear flow F( on 1I'" determined by v
E
IRn is ergodic
if and only if it is
irrational. Proof Assume the flow is irrational and let f E L2(;.n) be a constant of the motion. Expand f([
Section 7.4 Induced Flows on Function Spaces and Ergodicity
f([q>))
=
L akeik ·
and
(f 0 F,)([q>))
=
~~
L bk(t)eik .
519
~~
where the convergence is in L2 and the Fourier coefficients are given by
(The measure dq>
is chosen such that the total volume of 1r' equals one.) Since f is a constant
of the motion, ak = bk(t) for all k E zn and all t E lit which implies that e kv ' = 1 for all k E T, t E lit. Thus k·Y = 0 which by hypothesis forces k = O. Consequently all ak := 0 with the exception of 30 and thus f:= 30 a.e. Conversely, assume F, is ergodic and that k· A:= {[\jIl A
*0
E
:= 0 for some k
Y
E
zn\ {OJ.
Then the set
1r' I k . \jI := 0) is closed and hence measurable and invariant under F,. But clearly
and A
* 1r'
which shows that F, is not ergodic. _
7.4.10 Corollary Let F, be an irrational flow on 1r' determined by
Y.
Then every trajectory of
F, is uniformly distributed on 1r', i.e.,for any measurable set A in 1r'.
lim (measure A(t)/t)
=
measure A
l---7±CCI
where A(t):= {s
E
[0. tll Fs([\jI))
E
A) and the measure of 1r' is assumed to be equal to one.
Proof Let XA be the characteristic function of A. Then
by 7.4.7 and 7.4.9. _
7.4.11 Corollary Every trajectory of a quasiperiodic flow F, on 1r' is dense if and only if the /low is irrational. Proof By translation of the initial condition it is easily seen that every trajectory is dense on 1r' if and only if the trajectory through [0] is dense. Assume first that the flow is irratil, 1al. If {F,([O))
It E
lit) is not dense in 1r' then there is an open set U in 1r' not containing any !'JOint of this trajectory. Thus, in the notation of corollary 7.4.10, U(t):= 0. This contradicts 7.4.10 since the
measure of U is strictly positive.
Chapter 7 Integration on Manifolds
520
Conversely, assume that the trajectory through [OJ is dense and let f be a continuous constant of the motion for Ft. This implies that f is a constant. Since continuous functions are dense in L 2, this in tum implies that any L 2-constant of the motion is constant a.e. By Proposition 7.4.5, Ft is ergodic and by Proposition 7.4.9, F t is irrational. _
u:w
SUPPLEMENT 7.4A
Unbounded and Self Adjoint Operators. (Written in collaboration with P. Chernoff.) In many applications involving differential equations, the operators one meets are not defined
on the whole Banach space E and are not continuous. Thus we are led to consider a linear transformation A: DACE
~
E where DAis a linear subspace of E (the domain of A). If DA
is dense in E, we say A is densely defined. We speak of A as an operator and this shall mean linear operator unless otherwise specified. Even though A is not usually continuous, it might have the important property of being closed. We say A is closed if its graph
rA
is a closed subset of E x E. This is equivalent to (XnE
DA,xn~XE
E and
Axn~YE
E) implies (XE DA and Ax=y).
An operator A (with domain DA) is called closable if c1(rA)' the closure of the graph of A, is the graph of an operator, say, A. We call A the closure of A. It is easy to see that A is closable iff (X n E D A' Xn ~ 0 and AXn ~ y) implies y = O}. Clearly A is a closed operator that is an extension of A; i.e., DA ~DA and A =A on DA. One writes this as A~A. The closed graph theorem asserts that an everywhcre defined closed operator is bounded. (See §2.2.) However, if an operator is only densely defined, "closed" is weaker than "bounded." If A is a closed operator, the map x ...... (x, Ax) is an isomorphism between D A and the closed subspace
r A'
Hence if we set
III x 1112 = II x 112 + II Ax 11 2, DA becomes a Banach space. We call the norm III III on DA the graph norm. Let A be an operator on a real or complex Hilbert space H with dense domain DA' The adjoint of A is the operator A* with domain DA" defined as follows:
LJ~.= (YE Hlthereisa ZE H such that (Ax,y) =(x,z) for all XE DA} and
Section 7.4 Induced Flows on Function Spaces and Ergodicity
521
From the fact that DAis dense, we see that A * is indeed well defined (there is at most one such z for any y E H). It is easy to see that if A::::> B then B*::::> A *. If A is everywhere defined and bounded, it follows from the Riesz representation theorem
(Supplement 2.2A) that A* is everywhere defined; moreover it is not hard to see that, in this case, IIA*II=IIAII. An operator A is symmetric (Hermitian in the complex case) if A * ::::> A; i.e., (Ax, y) = (x, Ay) for all x, y E DA' If A* = A (this includes the condition DA' = D A)' then A is called
self-adjoint. An everywhere defined symmetric operator is bounded (from the closed graph theorem) and so is self-adjoint. It is also easy to see that a self-adjoint operator is closed. One must be aware that, for technical reasons, it is the notion of self-adjoint rather than symmetric, which is important in applications. Correspondingly, verifying self-adjointness is often difficult while verifying symmetry is usually trivial. Sometimes it is useful to have another concept at hand, that of essential self-adjointness. First, it is easy to check that any symmetric operator A is closable. The closure A is easily seen to be symmetric. One says that A is essentially self-adjoint when its closure A is self-adjoint. Let A be a self-adjoint operator. A dense subspace C CHis said to be a core of A if C C D A and the closure of A restricted to C is again A. Thus if C is a core of A one can recover A just be knowing A on C. We shall now give a number of propositions concerning the foregoing concepts, which arc useful in applications. Most of this is classical work of von Neumann. We begin with the following. 7.4.12 Proposition Let A be a closed symmetric operator of a complex Hilbert space H. If A is self-adjoint then A + AI is surjective for every complex number A with 1m A of- 0 (I is the
identity operator). Conversely, if A is symmetric and A - iI and A + iI are both surjective then A is self-adjoint. Proof Let A be self-adjoint and A = a + i~,
~ of-
O. For xED A we have
II(A + A)X 112 = II(A + a)x 112 + i~(x, Ax) - i~(Ax, x) + ~2 II X 112 = II(A + a)x 112 + ~211 x If ~ ~2 II
X
11 2,
where A + A means A + AI. Thus we have the inequality II(A + A)X II
~
I 1m A III x II.
(1)
Since A is closed, it follows from (1) that the range of A + A is a closed set for 1m A of- O. Indeed, let Yn = (A + A)X n ~ y. By the inequality (1), II xn - xm 11:0; II Yn - Ym II II 1m AI so xn converges to, say x. Also AXn converges to y - Ax; thus x E DA and y - Ax = Ax as A is
522
Chapter 7 Integration on Manifolds
closed. Now suppose y is orthogonal to the range of A + AI. Thus (Ax + Ax, y) = 0 for all xEDA' or (Ax, y) = - (x, Ay). By definition, yE DA* and A*y=- T y; since A=A*, yE DA and Ay=- Ty,weobtain (A + AI)y = O. Thus the range of A + AI is all of H. Conversely, suppose A+i and A-i are onto. Let yE DA*. Thus for all XE DN «A + i)x, y) = (x, (A * - i)y) = (x, (A - i)z) for some zED A since A - i is onto. Thus «A + i)x, y) = «A + i) x, z) and it follows that y = z. This proves that DA* C D A and so DA = D A*. The result follows . • If A is self-adjoint then for 1m A ~ 0, AI - A is onto and from (1) is one-to-one. Thus (AI - A)-I: H ~ H exists, is bounded, and we have
II (AI-Af 1 11$_1_. IImAI
(2)
This operator (AI - A)-l is called the resolvent of A. Notice that even though A is an unbounded operator, the resolvent is bounded. The same argument used to prove 7.4.12 shows the following. 7.4.13 Proposition A symmetric operator A is essentially self-adjoint iff the ranges of A + iI and A - iI are dense. If A is a (closed) symmetric operator then the ranges of A + iI and A - iI are (closed)
subspaces. The dimensions of their orthogonal complements are called the deficiency indices of A. Thus 7.4.12 and 7.4.13 can be restated as: a closed symmetric operator (resp., a symmetric operator) is self-adjoint (resp., essentially self-adjoint) iff it has deficiency indices (0,0). If A is a closed symmetric operator then from (I), A + iI is one-to-one and we can consider the inverse (A + il)-l, defined on the range of A + iI. One calls (A - iI)(A + il)-l the Cayley transform of A. It is always isometric, as is easy to check. Thus A is self-adjoint iff its Cayley transform is unitary. Let us return to the graph of an operator A for a moment. The adjoint can be described entirely in terms of its graph and this is often convenient. Define an isometry l: H (fJ H ~ H (fJ H by lex, y) = (-y, x); note that j2 =-1.
Section 7.4 Induced Flows on Function Spaces and Ergodicity
523
7.4.14 Proposition Let A be densely defined. Then (r A).l = J(r A+) and -r A+ = J(r A).l. In
particular, A * is closed, and if A is closed, then
where H EB H carries the usual inner product: «xI' x), (Yl' Y2» = (xl' YI) + (x 2' Y2)' Proof Let (Z,y)E J(rA+), so yE D A+ and z=-A*y. Let XE DN Wehave «x, Ax), (-A *y, y» = (x, -A *y) + (Ax, y) = 0, and so J(rA+)crA.l. Conversely if (z, y) E (rA).l, then (x, z) + (Ax, y) = 0 for all XED A' Thus by definition, y E D A+ and z = -A *y. This proves the opposite inclusion. _ Thus if A is a closed operator, the statement H EB H = r A EB J(r A+) means that given e, f E H, the equations x - A *y = e and Ax + y = f have exactly one solution (x, y). If A is densely defined and symmetric, then its closure satisfies A
C
A * since A * is closed. There are other important consequences of 7.4.14 as well.
7.4.15 Corollary For A densely defined and closeable, we have (i)
A =A**,
Suppose A: DA
and
C
(ii)
A*=A*.
H ~ H is one-to-one. Then we get an operator A-I defined on the
range of A. In terms of graphs:
where K(x, y) = (y, x); note that K2 = I, K is an isometry and KJ = -JK. It follows for example that if A is self-adjoint, so is A-I, since
Chapter 7 Integration on Manifolds
524
Next we consider possible self-adjoint extensions of a symmetric operator. 7.4.16 Proposition Let A be a symmetric densely defined operator on H and A its closure. The following are equivalent: (i) A is essentially self-adjoint. (ii) A* is self-adjoint. (iii) A** ::J A*. (iv) A has exactly one self-adjoint extension. (v) A=A*.
Proof By definition, (i) means A* = A. But we know A* = A * and A = A ** by 7.4.15. Thus (i), (ii), (v) are equivalent. These imply (iii). Also (iii) implies (ii) since A cAe A * c A ** = A and so A * = A **. To prove (iv) is implied let Y be any self-adjoint extention of A. Since Y is closed, Y::J A. But A = A * so Y extends the self-adjoint operator A *; i.e., Y::J A. Taking adjoints, A * = A::J Y* = Y so Y = A. To prove that (iv) implies the others is a bit more complicated. We shall in fact give a more general result in 7.4.18 below. First we need some notation. Let 0+ = range(A + iI)-L CHand 0_ = range(A - ilf C H called the positive and negative defect spaces. Using the argument in 7.4.12 it is easy to check that
7.4.17 Lemma Using the graph norm on DA*' we have the orthogonal direct sum
°
° °
Proof Since 0+, 0_ are closed in H they are closed in A *. Also A c A * is closed since A * is an extention of A and hence of A. It is ea~y to see that the indicated spaces are orthogollal. Forexamplelct XE DA and yE 0_. Then using the inner product «x, y» = (x, y) + (A *x, A *y) gives «x, y» = (x, y) + (A *x, -iy) = (x, y) - i( A *x, y). Since x E DA = D A *, by 7.4.16(v), we get «x, y» = (x, y) - i(x, A *y) = (x, y) - (x, y) = o.
Section 7.4 Induced Flows on FunctUm Spaces and Ergodicity
525
To see that D A " = DA EEl D+ EEl D_ it suffices to show that the orthogonal complement of DA EEl D+ EEl D_ is zero. Let UE (DAEElD+EElD_).L, so «u, x» = «u, y» = «u, z» = 0 for all XE DA, yE D+, ZE D_. From «u,x»=O we get (u,x)+(A*u,A*x)=O or A*UE D A" and A * A *u = - u. It follows that (I - iA *)u ED +. But from «u, y» = 0 we have «(1iA *)u, y) = 0 and so (1- iA *)u = O. Hence u
E
D_. Taking Z = u gives u = O. ,.
7.4.18 Proposition The self·adJoint extensions of a symmetric densely defined operator A (if any) are obtained as follows. Let T : D+ ~ D_ be an isometry mapping D+ onto D_ and let
r TeD+
EEl D_ be its graph. Then the restriction of A * to D A EEl r T is a self·adJoint extension of A. Thus A has self-adjoint extensions iff its defect indices (dim D+' dim D_) are equal and these extensions are in one-to-one correspondence with all isometries of D+ onto D_. Assuming this result for a moment, we give the following. Completion of Proof of 7.4.16 If there is only one self-adjoint extension it follows from 7.4.18 that D+=D_= {OJ soby7.4.13 A is essentially self-adjoint. _ Proof of 7.4.18 Let B be a self-adjoint extension of A. Then A* = A*::::> B so B is the restriction of A * to some subspace containing D A' We want to show that these subspaccs are of the form D A EEl r T as stated. Suppose first that T: D+ ~ D_ is an isometry onto and let DA EEl
r T"
First of all, one proves that
A
A
be the restriction of A * to
is symmetric: i.e., for u, xED A and v, y E D+ that
(Ax + A *y + A *Ty, u + v + Tv) = (x + y + Ty, Au + A *v + A *Tv). This is a straightforward computation using the definitions. To show that
A
is self-adjoint, we show that DA"
a nonzero ZEDA" such that either
A
*z = iz or
C
DA' If this does not hold there exists
A *z = -iz.
This follows from Lemma 7.4.17
applied to the operator A. (Observe that A is a closed operator--this easily follows.) Now A ::::>A so A*::::>A*. Thus ZE D+ or ZE D_. Suppose ZE D+. Then z+TzE DA so as «DA' z» = 0 «(,» denotes the inner product relative to A),
o=
«z + Tz, z» =
«;~,
z» + «Tz, z» = 2(z, z),
since Tz E D_. Hence z = O. In a similar way one sees that if Z E D_ then z = O. Hence
A
is self-adjoint. We will leave the details of the converse to the reader (they are similar to the foregoing). The idea is this: if A is restriction of A* to a subspace DA ED V for V C D+ ED D_ and A is
526
Chapter 7 Integration on Manifolds
symmetric, then V is the graph of a map T: W subspace
we
D+. Then self-adjointness of
C
D+
~
D_ and (Tu, Tv) = (u, v), for a
A. implies that in fact W = D+ and T is onto. _
A convenient test for establishing the equality of the deficiency indices is to show that T commutes with a conjugation U; i.e., an anti linear isometry U: H ~ H satisfying U 2 = I; anti linear means U(ux) = aUx for complex scalars u and U(x + y) = Ux + Uy for x, Y E H. It is easy to see that U is the isometry required from D+ to D_ (use D+ = range (A + iI)l.). As a corollary, we obtain an important classical result of von Neumann: Let H be L2 of a
measure space and let A be a (closed) symmetric operator that is real in the sense that it commutes with complex conjugation. Then A admits self-adjoint extensions. (Another sufficient condition of a different nature, due to Friedrichs, is given below.) This result applies to many quantum mechanical operators. However, one is also intcrested in essential self-adjointness, so that thc self-adjoint extension will be unique. Methods for proving this for specific operators in quantum mechanics are given in Kato [1966] and Reed and Simon [1974]. For corresponding questions in elasticity, see Marsden and Hughes [1983]. We now give some additional results that illustrate methods for handling self-adjoint operators. 7.4.19 Proposition Let A be a self-adjoint and B a bounded self-adjoint operator. Then A + B
(with donwin D A) is self-adjoint. If A is essentially self-adjoint on D A then so is A + B. Proof A + B is certainly symmetric on DA' Let y E D(A+B)' so that for all xEDA' «A + B)x, y) = (x, (A + B)*y). The left side is (Ax, y) + (Bx, y) =(Ax, y) + (x, By) since B is everywhere defined. Thus (Ax, y) = (x, (A + B)*y - By). Hence y E D A• = DA and Ay = A*y = (A + B)*y - By. Hence y E DA+B = DA. Let A be the closure of A. For the second part, it suffices to show that the closure of A + B equals A + B. But if x E DA there is a sequence xn E D A such that xn
~
x, and AXn
~
Ax. Then Bxn·-t Bx as B is bounded so x belongs to the domain of the closure of A + B. _ In general, the sum of two self-adjoint operators need not be self-adjoint. (See Nelson [1959] and Chernoff [1974] for this and related examples.)
Section 7.4 Induced Flows on Function Spaces and Ergodicity
527
7.4.20 Proposition Let A be a symmetric operator. If the range of A is all of H then A is self-adjoint.
Proof We first observe that A is one-to-one. Indeed Jet Ax = O. Then for any y E DA' 0 = (Ax, y) = (x, Ay). But A is onto and so x = O. Thus A admits an everywhere defined inverse A-I, which is therefore self-adjoint. Hence A is self-adjoint (we proved earlier than the inverse of a self-adjoint operator is self-adjoint). _ We shall use these results to prove a theorem that typifies the kind of techniques one uses. 7.4.21 Proposition Let A be a symmetric operator on H and suppose A 5 0; that is (Ax, x) 5
o for
xED A' Suppose I - A has dense range. Then A is essentially self-adjoint.
Proof Note that ((1- A)u, u) = (u, u) - (Au, u) ~
and so by the Schwarz inequality we have 11(1- A)u II
~
II U 112
II u II. It follows that the closure of I - A
which equals I - A has closed range, which by hypothesis must be all of H. By 7.4.20 I - A is self-adjoint and so by 7.4.21 A is self-adjoint. _ 7.4.22 Corollary If A is self-adjoint and A 5 0, then for any A. > 0, A. - A is onto, (A. - A)-l exists and
II (A. - A)-l II 5 IfA.
(3)
Proof As before we have «A. - A)u, u) ~ A. II U 112 which yields II(A. - A)u II ~ A. II u II. As A is closed, this implies that the range of A. - A is closed. If we can show it is dense, the result will follow. Suppose Y is orthogonal to the range: «A.-A)u,y)=O for all
UE
DA .
This means that (A. - A)*y = 0, or since A is sC\f-adjoint, y gives 0 = «A. - A)y, y) ~
Ail y 112
E
D A' Making the choice u = y
so Y = O. _
Note that an operator A has dense range iff A* is one-to-one; i.e., A*w = 0 implies w= O. For a given symmetric operator A, we considered the general problem of self-adjoint extensions of A and classified these in terms of the defect spaces. Now, under different
528
Chapter 7 Integration on Manifolds
hypotheses. we construct a special self-adjoint extension (even though A need not he essentially self-adjoint). This result is useful in many applications. including quantum mechanics. A symmetric operator A on H is called lower semi-bounded if there is a constant such that (Ax.x)~cllxIl2 forall
XE
CElt
DA" Upper semi-bounded is defined similarly. If A is
either upper or lower semi-bounded then A is called semi-bounded. Observe that if A is positive or negative then A is semi-bounded. As an example. let A = _V2 + V where V2 is the Laplacian and let V be a real valued continuous function and bounded below. say V(x) ~ a. Let H = L2(IRn. C) and DA the C~ functions with compact support. Then - V2 is positive so (Af. f)
=(- V2 f. f) + (Vf. f) ~ a(f. f).
and thus A is semi-bounded. We already know that this operator is real so has self-adjoint extensions by von Neumann's theorem. However. the self-adjoint extension constructed below (called the Friedrichs extension) is "natural." Thus the actual construction is as important as the statement. 7.4.23 Theorem A semi-bounded symmetric (densely defined) operator admits a self-adjoint
extension. Proof After multiplying by -1 if necessary and replacing A by A + (1 - a)I we can suppose (Ax. x) ~ II X 112. Consider the inner product on DA given by «x. y» = (Ax. y). (Using symmetry of A and the preceding inequality one easily checks that this is an inner product.) Let HI be the completion of DA in this inner product. Since the HI-norm is stronger than the H-norm. we have HI C H (i.e .• the injection DA C H extends uniquely to the completion). Now let H-I be the dual of HI. We have an injection of H into H- I defined as follows: if y is fixed and x 1-+ (x, y) is a linear functional on H, it is also continuous on HI since I (x, y) I ::; II x 1111 y II::; III x 11111 y II,
where III· III is the norm of HI. Thus HI
C
H
C
Wi.
Now the inner product on HI defines an isomorphism B: HI ~ H-I. Let C be the operator with domain Dc = {x E HI I B(x) E H}, and Cx = Bx for x E Dc- Thus C is an extension of A. This will be the extension we sought. We shall prove that C is self-adjoint. By definition, C is surjective; in fact C: Dc ~ H is a linear isomorphism. Thus by 7.4.20 it suffices to show that C is symmetric. Indeed for x. y E Dc we have, by definition, (Cx, y)
=
«x, y»
=
«y, x»
=
(Cy, x)
=
(x, Cy). •
The self-adjoint extension C can be alternatively described as follows. Let HI be as before and let C be the restriction of A* to DA. n HI. We leave the verification as an exercise. b
Section 7.4 Induced Flows on Function Spaces and Ergodicity
~
529
SUPPLEMENT 7.4B Stone's Theorem
(written in collaboration with P. Chernoff) Here we give a self-contained proof of Stone's theorem for unbounded self-adjoint operators
A on a complex Hilbert space H. This guarantees that the one-parameter group eitA of unitary operators exists. In fact, there is a one-to-one correspondence between self-adjoint operators and continuous one-parameter unitary groups. A continuous one-parameter unitary group is a homomorphism t 1-+ VI from lR to the group of unitary operators on H, such that for each x
E
H the map t 1-+ Vlx is continuous. The infinitesimal generator A of VI is defined by
its domain D consisting of those x for which the indicated limit exists. We insert the factor i for convenience; iA is often called the generator.
7.4.24 Stone's Theorem. Let VI be a continuous one-parameter unitary group. Then the generator A of VI is self-adjoint. (In particular, by Supplement 7.4A, it is closed and densely defined.) Conversely, let A be a given self-adjoint operator. Then there exists a unique one-parameter unitary group VI whose generator is A. Before we begin the proof, let us note that if A is a bounded self-adjoint operator then one can form the series I + itA
+
;!
(itA)2
+
;!
(itA)3
+ ...
which converges in the operator norm. It is straightforward to verify that VI is a continuous one-parameter unitary group and that A is its generator. Because of this, one often writes eitA for the unitary group whose generator is A even if A is unbounded. (In the context of the so-called "operational calculus" for self-adjoint operators, one can show that eilA really is the result of applying the function e ilO to A; however, we shall not go into these matters here.) Proof of Stone's Theorem (first half) Let VI be a given continuous unitary group. In a series of lemmas, we shall show that the generator A of VI is self-adjoint.
7.4.25 Lemma The domain D of A is invariant under each VI' and moreover AVtx = VIAx for each XED. Proof Suppose xED. Then
Chapter 7 Integration on Manifolds
530
which converges to Ut(iAx)
= iUtAx
as h
~
O. The lemma follows by the definition of A. _
7.4.26 Corollary A is closed. Proof If XED then, by 7.4.25
(1)
Hence
Now suppose that xn E D, xn
~
x, and AXn
~
y. Then we have, by (I),
(2)
Thus
(Here we have taken the limit under the integral sign because the convergence is unifonn; indccd IIU~Axn
-
U~YII
= IIAxn -
yll
~
0 independent of "t E [0, tJ.) Then, by (2),
dd UtX t
Hence xED and y
=Ax.
I
iy.
t=O
Thus A is closed. _
7.4.27 Lemma A is densely defined. Proof Let x E H, and let
C~
function with compact support on IR. Define
x
i~
We shall show that x~x 0 is given, one can find N so large that IIUtx - xii < e if I t I < lIN. Suppose that n > N. Then
Section 7.4 Induced Flows on Function Spaces and Ergodicity
531
Finally, we show that xq> E 0; moreover, we shall show that iAxq> = - xq>" Indeed,
= -
f ~ do'
it
UHcrx d'! = - f~ do'
0
~
l'crHtU,X d'!.
Integrating by parts and using the fact that
That is, - f:u,Xq>' d'! = Utxq> - xq> , from which the assertion follows. • Thus far we have made no significant use of the unitarity of U t . We shall now do so. 7.4.28 Lemma A is symmetric. Proof Take x, yEO. Then we have
Tlddt
(Ax, y)
=
It=O =
(Utx, y)
+
1t (x, U_ty)
= - .;.. (x, 1
iAy)
=
It=O
Tlddt
= -
*
I
(x, U t y) t=O
+
I
1t (x, UtY) t=O
(x, Ay) . •
To complete the proof that A is self-adjoint, let y E 0* and x E O. By 7.4.25.28,
(UtY, x) = (y. U_tx) = (y. x) + (y, i Itu,AX d'!)
= (y.
x) - i I t (y. U,Ax)d'! = (y, x) - i I t (y, AU,x) d1:
Chapter 7 Integration on Manifolds
532
1,-1 (U_, A * y, x) d't
= (y, x) - i 0
1,1 (0, A *y, x) d1:
= (y, x) + i 0
Because D is dense, it follows that
Hence, differentiating, we see that y E D and A *y = Ay. Thus A = A *. Proof of Stone's theorem (second half) We are now given a self-adjoint operator A. We shall construct a continuous unitary group UI whose generator is A.
7.4.29 Lemma If A> 0, then 1+ AA2: DA2 ~ H is bijective, (I + AA2)-1 : H ~ DA2 is bounded by I, and DA 2, the domain of A2 , is dense. Proof If A is self-adjoint, so is
...;T:A.
It is therefore enough to establish the lemma for A = 1.
First we establish surjectivity. By 7.4.14 and 7.4.26, if
E H is given there exists a unique solution (x, y) to the
Z
equations x - Ay = 0, Ax + y = z. From the first equation, x
= Ay.
The second equation then yields A2y + y = z, so 1+ A2 is
surjective. For xED A2, note that «I + A2)x, x) ~ II
X
11 2, so II (I + A2)x II ~ II x II. Thus 1+ A2 is
one-to-one and II (I + A2)-1 11$ 1. Now suppose that u is orthogonal to D A2. We can find a v such that u
=v + A2v.
Then 0= (u, v) = (v + A2v, v) = II
V
112 + II Av 11 2,
whence v = 0 and therefore u = O. Consequently D A2 is dense in H . • For A> 0, define an operator AI.. by AI.. = A( I + A A2)-I. Note that AI.. is defined on all of H because if x E H then (I + J...A2)-lx E D A2 C D, so A (I + J...A2)-lx makes sense.
7.4.30 Lemma AI.. is a bounded self-adjoint operator. Also AI.. and Proof Pick x E H. Then by 7.4.29
A~
commute for all A,
~
> o.
Section 7.4 Induced Flows on Function Spaces and Ergodicity
533
Ail AA.x 112 =(AA(I + AA 2)-lx, A(I + AA2)-lx) = (AA2(1 + AA2)-lx, (I + AA2)-lx)
:0; «I + AA2)(I + AA2)-lx, (I + AA 2)-lx) :0; 11(1 + AA2)-lx 112:0; so
II AA.II :0; 1/ v-:;:'
II x 11 2,
and thus AA. is bounded.
We now show that AA. is self-adjoint. First we shall show that if xED, then
Indeed, if xED we have AA.xED A2 by 7.4.29 and so
Now suppose xED and y is arbitrary. Then «I + AA2)-IAx, y) = «I + AA2)-IAx, (I + AA2)(I + AA2)-ly)
(Ax, (I + }.A2t 1y)
= (x, A.y).
Because D is dense and AI.. bounded, this relation must hold for all x E H. Hence AA. is self-adjoint. The proof that AA.~ = AJ.LAA. is a calculation that we leave to the reader. _ Since AA. is bounded, we can form the continuous one-parameter unitary groups UtA. = eitAA., A> 0 using power series or the results of §4.1. Since AA. and AJ.L commute, it follows that UsA. and
ut
commute for every s and t.
7.4.31 Lemma If XED then limA. -> OAA.x Proof If xED we have AA.x - Ax
=Ax.
= (I + AA 2t
1Ax - Ax
=- AA2(1 + AA2)-1 Ax.
It is therefore
enough to show that for every y E H, AA2(1 + AA2)-ly ~ O. From the inequality 11(1 + AA2)y 112 ~ II AAZy 11 2, valid for A ~ 0, we see that II }.A2(1 + }.A2)-1 II :0; I. Thus it is even enough to show the preceding equality for all y in some dense subspace of H. Suppose y E D A2, which is dense by 7.4.29. Then
which indeed goes to zero with A. _ 7.4.32 Lemma For each x E H, limA.-> oUtAx exists. If we call the limit Utx, then {Utl is a
continlWus one-parameter unitary group. Proof We have
534
Chapter 7 Integration on Manifolds
whence (3)
II AAx - Af1x II ~ 0 as A,
Now suppose that xED. Then by 7.4.31, AAx ~ Ax, so that Il ~ O. Because of (3) it follows that {U,A~
h>0
is uniformly Cauchy as A ~ 0 on every
compact t-interval. It follows that lim A-> oU,Ax = U,X exists and is a continuous function of t. Moreover, since D is dense and all of the UtA have norm I, an easy approximation argument shows that the preceding conclusion holds even if x E' D.
It is obvious that each U t is a linear opeator. Furthermore,
so Ut is isometric. Trivially, U o = I. Finally, (UsU,x, y) = lim (UsAUtx, y) 1..->0
lim (U/x, U~sY)
1..->0
=
=
lim (U,x, U~sY)
1..->0
lim (U~+tx, y)
1..->0
=
(Us+tx, y),
so UsU, = Us+!' Thus, Us has an inverse, namely U_ s' and so Us is unitary. _
7.4.33 Lemma If xED, then , . Ux-x Iu n---
,->0
t
iAx.
Proof We have
(4) Now
uniformly for
1:
E [0, tl as A ~ O. Thus letting A ~ 0 in (4), we get
(5)
Section 7.4 Induced Flows on FunctWn Spaces and Ergodicily
535
for all xED. The lemma follows directly from (5) . • 7.4.34 Lemma If
Ux-x lim - ' - -
1-->0
iw
t
exists, then xED.
Proof It suffices to show that XE D*, the domain of A*, since D=D*. Let yE D*. Then by 7.4.33,
(x, iAy)
= lim
1-->0
<x, U-,y-y) -t
= - lim
,-->0
<-Ux-x ) '- - , y t
= - (iw, y).
So (x, Ay) = (w, y). Thus x E D* and so as A is self-adjoint, XED . • Let us finally prove uniqueness. Let c(t) be a differentiable curve in H such that c(t) E D and c'(t) = iA(c(t». We claim that c(t) = U1c(0). Indeed consider, h(t) = U-tc(t). Then
lI(c(t + 't) - c(t» - (U,c(t) - c(t» II. Hence h(t + 't) - h(t)
~O
't
as
't ~
0, so h is constant. But h(t) = h(O) means c(t) = U,c(O) . •
From the proof of Stone's theorem, one can deduce the following Laplace transform expression for the resolvent, which we give for the sake of completeness. 7.4.35 Corollary Let Re A. > O. Thenfor all x E H,
Proof The foregoing is formally an identity if one thinks of U, as e itA . Indeed, if A is bounded then it follows just by manipulation of the power series: One has e-Ald tA = e-t(A. - iA), as one can
see by expanding both sides; next
l
R
o
e-'(A-iN x dt = ('11.-1·A)-l [ x-e-R(A-iA) x1
(integrate the series term by term). Letting R
~
00,
one has the result.
Chapter 7 lnUgration on Manifolds
536
Now for arbitrary A we know that U,x = lim!, ~ 0 U,!'x, uniformly on bounded intervals. It follows that
It remains to show that this limit is (A - iA)-'x. Now
But (A - iAt'x
E
0 (see Proposition 7.4.12) and so by 7.4.31,
Because II(A - iA!,)-' II ~ IRe AI-' it follows that
In closing, we mention that many of the results proved have generalizations to continuous one-parameter groups or semi-groups of linear operators in Banach spaces (or on locally convex spaces). The central result, due to Hille and Yosida, characterizes generators of semi-groups. Our proof of Stone's theorem is based on methods that can be used in the more general context. Expositions of this more general context are found in, for example, Kato [1966] and Marsden and Hughes [1983, ch 6].
Il:JJ Exercises
(Exercises 7.4A-C form a unit.) 7.4A Given a manifold M, show that the space of half-densities on M carries a natural inner product. Let its completion be denoted J)(M), which is called the intrinsic Hilbert space of M. If IJ. is a density on M, define a bijection of L2(M, IJ.) with J)(M) by ff-+ f..J1l. Show that it is an isometry. 7.48 If F, is the (local) flow of a smooth vector field X, show that F, induces a flow of isometrics on J)(M). (Make no assumption that X is divergence-free.) Show that the generator iX' = Lx of the induced flow on J)(M) is
and check directly that X' is a symmetric operator on the space of half-densities with
Section 7.4 Induced Flows on Function Spaces and Ergodicity
537
compact support. 7.4C Prove that Ft is complete a.e. if and only if X' is essentially self-adjoint. 7.4D Consider the flow in
]R2
associated with a reflecting particle: for t > 0, set
Ft(q, p) = q + tp if q > 0, q + tp > 0 and Ft(q, p) = - q - tp if q > 0, q + tp < 0 and set
What is the generator of the induced unitary flow? Is it essentially self-adjoint on the C'" functions with compact support away from the line q =O?
7.4E Let M be an oriented Riemannian manifold and L2(Ak(M» the space of L2 k-forms with
f
inner product
Wi/.S Introduction to Hodge-DeRham Theory and Topological Applications of Differential Forms Recall that a k-fonn a is called closed if do. = 0 and exact if 0.= dP for some k - 1 fonn p. Since d 2 =0, every exact fonn is closed, but the converse need not hold. Let ker dk range d
k-l'
(where dk denotes the exterior derivative on k-fonns), and call it the k-th deRham cohomology group of M. (The group structure here is a real vector space.) The celebrated deRham theorem states that for a finite-dimensional compact manifold, these groups are isomorphic to the singular cohomology groups (with real coefficients) defmed in algebraic topology; the isomorphism is given by integration. For "modem" proofs, see Singer and Thorpe [1967] or Warner [1971]. Thc original books of Hodge [1952] and deRham [1955] remain excellent sources of infonnation as well. A special but important case of the deRham theorem is proved in Supplement 7.58. The scope of this section is to infonnally discuss the Hodge decomposition theory based on differential operators and to explain how it is related to the deRham cohomology groups. In addition, some topological applications of the theory are given, such as the Brouwer fixed-point theorem, and the degree of a map is defined. In the sequel, M will denote a compact oriented Riemannian manifold, and I) the codifferential operator. At first we assume M has no boundary. Later we will discuss the case in which M has a boundary.
7.5.1 Definition The Laplace-deRham operator ~ : nt«M) ~ nt«M) is defined by ~ =dl) + I)d. A form for which ~o. = 0 is called harmonic. Let :J{k = {a E Qk(M) I ~o. = O} denote the vector space of harmonic k-forms. If f E J{o(M), then M = d8f + /idf = I)df = - div grad f. So M = - '\7 2f, where '\7 2 is the Laplace-Beltrami operator. This minus sign can be a source of confusion and one has to be careful. Recall that the L2-inner product in Qk(M) is defined by (a,
P)
=
L
0./\
*P
and that d and I) are adjoints with respect to this inner product. That is, (do., P) = (a, I)P) for all a E Qk-l(M), P E Qk(M). Thus is follows that for a, P E Qk(M), we have
Section 7.5 Introduction to Hodge-DeRham Theory
539
(!la, P) = (doa, P) + ( oda , P) = ( oa, oP) + (da, dP)
= (a, dOP) + (a, odP) = (a, !lP), and thus ~ is symmetric. This computation also shows that (~a, a) ~ 0 for all a
E
nk(M).
7.5.2 Proposition Let M be a compact boundaryless oriented Riemannian manifold and a nk(M). Then
~a
=0
iff oa = 0 and
E
da = O.
Proof It is obvious from the expression ~a = Oda + Oda that if da = 0 and oa = 0, then ~a = O. Conversely, the previous computation shows that 0 = (~a, a) = (da, da) + (oa, oa), so the result follows. •
7.5.3 The Hodge Decomposition Theorem Let M be a compact, boundaryless, oriented, Riemannian manifold and let ak(M) such that
0)
0) E
nk(M). Then there is an a
E
nH(M), P E nk+1(M) and "y E
= da + op +"y and ~("y) = O. Furthermore, da, op, and "y ar~ rnutUlllly L 2
orthogonal and so are uniquely determined. That is, (1)
We can easily check that the spaces in the Hodge decomposition are orthogonal. For example,
dnk-1(M) and oak+1(M) are orthogonal since (da, oP) = (dda, P) = 0,
o being the adjoint of d
and d 2 = O.
The basic idea behind the proof of the Hodge theorem can be abstracted as follows. We consider a linear operator T on a Hilbert space E and assume that T2 = O. In our case T = d and E is the space of L2 forms. (We ignore the fact that T is only densely defmed.) Let T* be the adjoint of T. Let J{ = {x EEl Tx = 0 and T*x = O}. We assert that E = c1(range T) EB c1(range T*) EB J{
(2)
which, apart from technical points on understanding the closures, is the essential content of the Hodge decomposition. To prove (2), note that the ranges of T and T* arc orthogonal because (Tx, T*y) = (Tx 2, y) = O. If C denotes the orthogonal complement of cl(range T) EB c1(range T*), then J{ c C. If x E C then (Ty, x) = 0 for all y implies T*x = O. Similarly, Tx = 0, so C c J{ and hence C= J{ The complete proof of the Hodge theorem requires elliptic estimates and may be found in Money [1966]. For more elementary expositions, consult Fladers [1963] and Warner [1971].
540
Chapter 7 Integration on Manifolds
7.5.4 Corollary Let:J-(k derwte the space of harmonic k-forms. Then the vector spaces '}{k and Hk (= ker dk/range d k- 1) are isomorphic. Proof Map :J-(k -+ ker d k by inclusion and then to Hk by projection. We need to show that this map is an isomorphism. Suppose "I E :J-(k and ["I] = 0 where ["I]
E
Hk is the class of "I. But
["I] = 0 means that "I is exact; "I = d~. But since '6"1 = O. "I is orthogonal to d~; i.e .. "I is
orthogonal to itself. so "1= O. Thus the map "I
f-+
["I] is one-to-one. Next let [m]
E
Hk. We
can. by the Hodge theorem. decompose m = da + '6~ + "I. where "I E :J-(k. Since dm =0. d'6~ =
o so f-+
0 = (~. d'6~) = ('6~. '6~). so '6~ = O. Thus m = da + "I and hence [m] = ["I], so the map "I ["I] is onto. _ The space '}{k == Hk is finite dimensional. Again the proof relies on elliptic theory (the
kernel of an elliptic operator on a compact manifold is finite dimensional). The Hodge theorem plays a fundamental role in incompressible hydrodynamics. as we shall see in §8.2. It allows the introduction of the pressure for a given fluid state. It has applications to many other areas of mathematical physics and engineering as well; see for example. Fischer and Marsden [1979] and Wyatt et aI. [1978]. Below we shall state a generalization of the Hodge theorem for some decomposition theorems for general elliptic operators (rather than the special case of the Laplacian). However. we first pause to discuss what happens if a boundary is present. This theory was worked out by Kodaira [1949]. Duff and Spencer [1952], and Morrey [1966. Ch. 7]. Differentiability across the boundary is very delicate. but important. Some of the best results in this regard are due to Morrey. Note that d and '6 may not be adjoints in this case. because boundary terms arise when we intregrate by parts (see Exercise 7.5E). Hence we must impose certain boundary conditions. Let a
E
Qk(M). Then a is called parallel or tangent to aM if the normal part. defined by na =
i*(*a) is zero where i : aM -+ M is the inclusion map. Analogously. a is perpendicular to aM
if its tangent part. defmed by ta = i*(a) is zero. Let X be a vector field on M. Using the metric. we know when X is tangent or perpendicular to aM. Now X corresponds to the one-form xt' and also to the (n-1)-form ixJ.! =
*xt' (J.! denotes the Riemarmian volume form). One checks that X is tangent to aM if and only if xt' is tangent to aM iff ixJ.! is rwrmal to aM. Similarly X is normal to aM iff ixJ.! is tangent to aM. Set
Q\(M) = {a E Qk(M) I a is tangent to aM} Qk n(M) = {a E Qk(M) I a is perpdendicular to aM}. and :J-(k(M) = {a E Qk(M) I da = O. 8a = O}. The condition that da = 0 and 8a = 0 is now stronger than t.a = O. One calls elements of '}{k harmonic fields. after Kodaira [1949].
Section 7.5 Introduction to Hodge-DeRham Theory
541
7.5.5 The Hodge Theorem for Manifolds with Boundary Let M be a compact oriented Riemannian manifold with boundary. The following decomposition holds:
One can easily check from the following fonnula (obtained from Stokes' theorem):
(da,~)
= (a,
o~) +
1 (1M
a
I\*~
(see Exercise 7.5E), that the summands in this decomposition are orthogonal. Two other closely related decompositions of interest are (i)
where
are the co-closed k-forms tangent to aM and, dually (ii) where Ck n are the closed k-forms normal to aM. To put the Hodge theorem in a general context, we give a brief discussion of differential operators and their symbols. (See Palais [1965a], Wells [1980), and Marsden and Hughes [1983) for more infonnation and additional details on proofs.) Let E and F be vector bundles of M and let C"'(E) denote the C'" sections of E. Assume M is Riemannian and that the fibers of E and F have inner products. A k-th order differential operator is a linear map D:
C~(E) ~
C"'(F) such
that, if f E C~(E) and f vanishes to k-th order at x E M, then D(f)(x) =O. It is not difficult to see that vanishing to k-th order makes intrinsic sense independent of charts and that D is a k-th order differential operator iff in local charts D has the fonn
Where j
= (j I' ... , js)
is a multi-index and a j is a C~ matrix-valued function of x (the matrix
Corresponding to linear maps of E to F). The operator D has an adjoint operator D* given in charts (with the standard Euclidean inner product on fibers) by
542
Chapter 7 Integratinn on Manifolds
O*(h)
"
1 aljl LJ (_I)ljl .
=
. (pajh), p a"xli ... a"xl,
O:;ljl~
where p dx 1 /\
... /\
dx" is the volume element on M and a/ is the transpose of a r The crucial
property of 0* is (g, O*h) = (Og, h). where (.) denotes the L2 inncr product. g
E
C~C<E). and h
C~C
E
C'" sections with compact support. For example. we have the operators ~ C~(Ak+l)
(first order)
C~(Ak) ~ C~(Ak-l)
(first order)
d: C"'(Ak) 8:
t.: C~(Ak) ~ C~(Ak) where d*
= 8.
8*
=d
and t.*
= t..
(second ordcr)
The symbol of 0 assigns to each /;
E
T* xM. a linear map
0(1;) : Ex ~ Fx defined by
where g
E
C~(M.
JR). dg(x) = /; and f E
C~(E).
f(x) = e. By writing this out in coordinates one
sees that 0(/;) so defined is independent of g and f and is a homogeneous polynomial expression in /; of degree k obtained by substituting each /;j in place of a/ax j in the highest order terms. For example. if
(aij is for each i. j a map of Ex to F x), For real-valued functions. the classical definition of an elliptic operator is that the foregoing quadratic form be definite. This can be generalized
o is called elliptic if
a~
follows:
0(/;) is an isomorphism for each /; = O. To see that t. : C~(Ak) ~ C~(Ak)
is elliptic one uses the following facts: 1 The symbol of d is 0(/;) = /; /\.
2 The symbol of 8 is 0(/;) = i s#. 3 The symbol is multiplicative: 0(/;)(0 1 0 O2 )
= 0(/;)(0 1)
0
0(/;)(0 2 ),
From these. it follows by a straightforward calculation that the symbol of t. is given by 0(/;)a = II
S 11 2 a.
so t. is elliptic. (Compute
all but one ternl cancel.)
S /\ (is#u) + is#(S /\ a)
applied to (v I ..... vk). noting that
Section 7.5 Introduction to Hodge·DeRham Theory
543
Elliptic Splitting Theorem or Fredholm Alternative Let 0 be an elliptic operator as above. Then
7.5.6
CO"(F)
=O(C~(E»
Ef)
ker 0*
Indeed this Iwlds ifit is merely assumed that either 0 or 0* has injective symbol. The proof of this leans on elliptic estimates that are not discussed here. As in the Hodge theorem, the idea is that the L2 orthogonal complement of range 0 is ker 0*. This yields an L2 splitting and we get a CO" splitting via elliptic estimates. The splitting in case 0 (resp. 0*) has injective symbol relies on the fact that then 0*0 (resp. DO *) is elliptic. For example, the equation
0u
=f
is soluble iff f is orthogonal to ker 0 *. More
~u = f is soluble if f is orthogonal to the constants; i.e.,
specifically,
f fd~ = O.
The Hodge Theorem is derived from the elliptic splitting theorem as follows.
Since
~
is
elliptic and symmetric C~(Ak(M» = range ~ Ef) ker ~ = range ~ Ef) :;(
Now write a k-form w as w = ~p + 'Y = dOp + Sdp + 'Y, so to get 7.5.3, we can choose a.
= op
and ~ = dp. ~ SUPPLEMENT 7.SA Introduction to Degree Theory
One of the purposes of degree theory is to providc algebraic measures of the number of solutions of nonlinear equations. Its development rests on Stokes' theorem. It links beautifully caleulus on manifolds with ideas on differential and algebraic topology.
All manifolds in this section are assumed to be finite dimensional, paracompact and LindelOf We begin with an extandability result. 7.5.7 Proposition Let V and N be orientable manifolds, dim(V) = n + I and dim(N) = n.
If f: dV E
~
N is a smooth proper map that extends to a smooth map of V to N, then/or every
Qn(N) with compact support,
Proof
W
favr*w = O.
Let F: V ~ N be a smooth extension of f. Then by Stokes' theorem
r lw= r
kv
kv
F*w=
r dF*W= r F*dw=O,
~
~
since dw = O. • This proposition will be applied to the case V = [0, 1] x M. For this purpose let us recall the
544
Chapter 7 Integration on Manifolds
product orientation (see Exercise 6.SN). If N and M are orientable manifolds (at most one of which has a boundary). then N x M is a manifold (with boundary). which is orientable in the following way. Let
1t1 :
Nx M
~
N and
~:
NxM
~
M be the canonical projections and [w],
[ll J orientations on Nand M respectively. Then the orientation of N x M is defined to be [1t 1*w /\ 1t1 *ll]. Alternatively. if v!' ...• vn E TxN and wI' ...• wm E TyM are positively oriented bases in the respective tangent spaces. then (v!' 0) •...• (vn • 0). (0. wI)' .... (0. wm) E Tex,y/N x M) is defined to be a positively oriented basis in their product. Thus. for [0. 1] x M. a natural
orientation will be given at every point (t. x) E [0. 1] x M by (1, 0). (0. vI)' .... (0, vm)' where vI' .... vm E TxM is a positively oriented basis. The boundary orientation of [0. 1] x M is determined according to definition 7.2.7. Since
a([o.
1] x M) = ({ O) x M) U ({ I) x M). every element of this union is oriented by the orientation
of M. On the other hand. this union is oriented by the boundary orientation of [0. 1] x M. Since the outward normal at (1. x) is (1. 0). we see that a positively oriented basis of T(1, x)( {l} x M) is given by (0. vI)' .... (0. v m) for vI' .... vm E TxM a positvely oriented basis. However. since the outward normal at (0. x) is (-1. 0). a positively oriented basis of Teo, x)( {O} x M) must consist of elements
(0. WI)' .... (0. wm) such that (-1. 0). (0. WI)' .... (0. wm) is positively oriented in [0. 1) x M. i.e .• defines the same orientation as (1.0). (0. VI)' .... (0. vm),
for VI' .... vm E TxM a positively oriented basis. This means that wI' .... wm E TxM is negatively oriented (see Fig. 7.5.1.). Thus the oriented manifold a([o. 1) x M) is the disjoint
union of {O} x M. where M is negatively oriented. with {I} x M. where M is positively oriented. ( 1.0)
Figure 7.5.1
7.5.8 Definition Two smooth mappings f. g : M ~ N. are called Cr-homotopic if there is a C-map F:[O.I]xM~N suchthat F(O.m)=f(m) and F(1,m)=g(m).forall mE M. The homotopy F is called proper if F is a proper map; in this case f and g are said to be properly C-homotopic maps.
Section 7.5 IntroductWn to Hodge-DeRham Theory
545
Note that if f and g are properly homotopic then necessarily f and g are proper as restrictions of a proper map to the closed sets (o) x M and (I) x M, respectively. 7.5.9 Proposition Let M and N be orientable n-manifolds, with M boundaryless,
with compact support, and f, g : M
~
0) E
nn(N)
N be properly smooth homotopic maps. Then
Proof There are two ways to do this. Method 1. Let F : [0, 1] x M
~
N be the proper homotopy between f and g. By the remarks
preceding definition 7.5.8, we have
I
f* 0) = -
{O}xM
r f*
JM
and
0)
I
g *0) =
{I}xM
r g*
JM
0),
so that
r g*
JM
0) -
r f*
JM
0)
=
r
Ja([O, II x M)
(F I a([O, 1] x M)t 0) = 0
by proposition 7.5.7. Method 2. By 6.4.16,
[*0) -
g*O) = d11 for some 11
E
nn-I(M) which has compact support
since the homotopy between f and g is a proper map. Then by Stokes' theorem
Remark Properness of f and g does not suffice in the hypothesis of Proposition 7.5.9. For example, if M = N = JR., then f(x)
=x
and g(x)
0)
= x2
= adx with a
~
0 a
C~
function satisfying supp(a) C ]- 00,0[,
are smoothly but not properly homotopic via F(t, x)
= (l -
t) x + tx 2
and {<><>f* 0) > 0, while {<><> g* 0) = 0 since g *0) = O.
Degree Theorem Let M and N be oriented manifolds, N connected, M boundaryless, and f: M ~ N a smooth proper map. Then there is an integer dcg(f) constant on the proper homotopy class of f, called the degree of f such that for any 11 E nn(N) with
7.5.10
compact support, (1)
If x E M is a regular point of f, let signeT xf) be 1 or -1 depending on whether the isomorphism TJ: TxM ~ Tf(x)N preserves or reverses orientation. The integer deg(f) is given by
Chapter 7 Integration on Manifolds
546
deg(f)
L
=
(2)
sign(Txf).
XEC1(y)
where y is an arbitrary regular value of f;
if Y e f(M) the right hand side is by convention equal
to zero.
JM
Proof By proposition 7.5.9. £*11 depends only on the proper homotropy class of f (and on 11). By Sard's Theorem. there is a regular value y of f. There are two possibilities: either 1{.f = N \ f(M) or not. If 1{.f = N \ f(M). then ~'!'! is never onto for all x
E
M. For any v I' ...• V n E
TxM •
since Ti(v l ) •...• Txf(vn) are linearly dcpendent. Thus deg(f) exists and equals zero. Assume 1{.r n f(M)
~
12' and let y
E
'1l r n
f(M). Since M and N have the same
dimension. r-I(y) is a zero - dimensional submanifold of M. hence discrete. Properness of f implies that {I(y) is also compact. i.e .• {I(y) = {XI' ...• xk+ll. where Txl is orientation preserving for i = 1•...• k and orientation reversing for i = k + 1•...• k + ~. The inverse function theorem implies that there are open neighborhoods Vi of xi and V of y such that r-I(V) = VI U ... U VHI' Vi n Vj =12' and if f I Vi : Vi the change of variables formula
~
V is a diffeomorphism. If sUPP(l1) c V. then by
(k -
and so the theorem is proved for sUPP(11) c
~/J 11 = (k -~) Ir 11 v N
(3)
v.
To deal with a general 11 proceed in the following way. For the open neighborhood V of 11. consider the collection of open subsets of N. S = {<jl(V) I <jl is a diffeomorphism propcrly homotopic to the identity}. We shall prove that S covers N. Let n
E
N; we will show that there
is a diffeomorphism <jl properly homotopic to the identity such that <jl(n) = y. Let c: [0. 1] ~ N be a smooth curve with c(O) = n and c(1) = y. As in Proposition 5.5.9. use a partition of unity to extend c'(t) to a smooth vector field X
E
X(N) such that X
vanishes outside a compact
neighborhood of c([O. 1]). The flow F( of X is complete by Proposition 4.1.20 and is the identity outside the above compact neighborhood of c([O. 1D. Thus the restriction F: [0. 1] x N ~ N is proper. Then <jl = FI is a proper diffeomorphism properly homotopic to the identity on N and <jl(n) = FI(n) = c(1) = y. Since S covers N. choose a partition of unity {(Va' h a ») subordinate to S and let lla = hall; thus sUPP(l1) eVa C <jla(V) for some <jla' Since all <jla are orientation preserving. the change of variables formula and (3) give
Section 7.5 Introduction to Hodge-DeRham Theory
Since
547
0
f is properly
homotopic to f. Thus by Proposition 7.5.9. fM(
Notice that by construction. if deg(f)
oF-
O. then f is onto. so f(x) = Y is solvable for x
given y. 7.5.11 Corollary Let V and N be orientable manifolds with dim(V) = n + I. and dim(N) = n.
If f:
av ~ N
extends to V. then deg(f)
=O.
This is a reformulation of7.5.7. Similarly. 7.5.9 is equivalent to the following. 7.5.12 Corollary Let M. N be orientable n-manifolds. N connected, M boundaryless, and let f. g : M
~
N be snwoth properly honwtopic maps. Then deg(f) = deg(g).
This corollary is useful in three important applications. The first concerns vector fields on spheres. 7.5.13 Hairy Ball Theorem Every vector field on an even dimensional sphere has a critical point.
Proof Let s2n be embedded as the unit sphere in ]R2n+1 and X E X(s2n). Then X defines a map f: s2n ~ ]R2n+1 with components f(x) = (fl(x) •...• f 2n +l (x)) satisfying fl(x)x l + ... + f2n+l(x)x2n+1 = 0; rex) are the components of X in ]R2n+l. Assume that X has no critical point. Replacing f by fill f II. we can assume that f: s2n ~ s2n. The map F: [0. IJ
X
s2n ~ s2n.
F(t. x) = (cos 7tt)x + (sin 7tt)f(x)
is a smooth proper homotopy between F(O. x) = x and F(l. x) = - x. That is. the identity fd is homotopic to the antipodal map A: s2n ~ s2n. A(x) = - x.
Thus by corollary 7.5.12. deg A =
1. However. since the lacobain of A is -I (this is the place where we use evenness of the dimension of the sphere). A is orientation reversing and thus by the Degree Theorem 7.5.10. deg A = -I. which is a contradiction. _ The second application is to prove the existence of fixed points for maps of the unit ball to itself.
Chapter 7 Integration on Manifolds
548
7.5.14 Brouwer's Fixed-Point Theorem Any smooth mapping of the closed unit ball of JRn into itself has a fixed point.
Proof Let B denote the closed unit ball in JRn and let sn-I sphere. If f: B -+ B has no fixed point, define g(x)
E
= aB
be its boundary, the unit
sn-I to be the intersection of the line
starting at f(x) and going through x with sn-I. The map g: B -+ sn-I so defined is smooth and for x E sn-I, g(x) = x. If n = 1 this already gives a contradiction, since g must map B = [-I, 1] onto {-I, I} = So, which is disconnected. For n ~ 2, defme a smooth proper homotopy F: [0, 1] x sn-I -+ sn-I by F(t, x) = g(tx). Thus F is a homotopy between the constant map c: sn-I -+ sn-I, c(x) = g(O) and the identity of sn-I. But c*co = 0 for any CO E nn-I(sn-I), so that by 7.5.10, deg c = O. On the other hand, by 7.5.12, deg c = I, which is false. _ The Brouwer fixed point theorem is valid for continuous mappings and is proved in thc following way. If f has no fixed points, then by compactness there exists a positive constant K>
o such that
1/
f(x) - x II > K for all x E B.
Let e < minCK, 2) and choose 0 > 0 such that
20/0 + 0) < e; i.e., 0 < d(2 - e). By the Weierstrass approximation theorem (see, for example, Marsden [1974a]) there exists a polynomial mapping q : JRn -+ JRn such that 1/ f(x) - q(x) 1/ < 0 for all x E B. The image q(B) lies inside the closed ball centered at 0 of radius 1 + 0, so that p '" q/(l + 0) : B -+ B and
1/
f(x) - p(x)
1/
~
I
f(x) -
:~)o
"
+"
-
:~)o {~~ II ~
for all x E B. Since p is smooth by 7.5.14, it has a fixed point, say
o< K < 1/ f(x o) -
Xo 1/ ~ 1/
f(x o) - p(xO>
1/
+
1/
12: 0 < e
Xo E
p(x o) - Xo 1/
B. Then ~
e,
which contradicts the choice e < K. Brouwer's fixed point theorem isfaise in an open ball, for the open ball is diffeomorphic to JRn and translation provides a counterexample. The proof we have given is not "constructive." For example, it is not clear how to base a numerical search on this proof, nor is it obvious that the fixed point we have found varies continuously with f. For these aspects, see Chow, Mallet-Paret and Yorke [1978]. A third application of 7.5.12 is a topological proof of the fundamental theorem of algebra. 7.5.15 The Fundamental Theorem of Algebra Any polynomial p : C -+ C of degree n > 0 has a root. Proof Assume without loss of generality that p(z) = zn + an_1z n- 1 + ... + 30, where ai E C, and regard p as a smooth map from JR2 to JR2. If P has no root, then we can define the smooth map
Section 7.5 Introduction to Hodge-DeRham Theory
fez) = p(z)/1 p(z)
I whose restriction to S' we denote by g: S I
549
~ S' .
Let R > 0 and define for t E [0, 1] and z E 5',
5ince p,(z)/(Rz)n R~
00,
= 1 + t[an_,(Rz) + ... + ad(Rz)n]
and the coefficient of t converges to zero as
we conclude that for sufficiently large R, none of the p, has zeros on 5 '. Thus
F: [0, 1] x
s' ~ 5'
defined by F(t, z) =
is a smooth proper homotopy of dn(z) = zn with g(Rz) which in tum is properly homotopic to g(z). On the other hand, G : [0, 1] x 5' ~ S' defined by G(t, z) = f(tz) is a proper homotopy of the constant mapping c: 5' ~ 5', c(z)
= f(O)
with g. Thus dn is properly homotopic to a constant map and hence deg dn = 0 by corollary 7.5.12. However, if 5' is parameterized by arc length e, 0 ~ e ~ 21t, then dn maps the segment 0 ~ e ~ 21t/n onto the segment 0 ~ e ~ 21t since dn has the effect e ie ...... eine . If (() denotes the corresponding volume form on 5', the change of variables formula thus gives
1SI which for n
*0
~ (() =
n1 (() SI
=
21tn, i.e., deg
~ = n,
is a contradiction. _
The fundamental theorem of algebra shows that any polynomial p: C ~ C of degree n can be written as
where z" ... , zm are the distinct roots of p, k" ... , k m are their multiplicities, k, + ... + km = n. and c
E
C is the coefficient of zn in p(z). The fundamental theorem of algebra can be refmed to
take into account multiplicities of roots in the following way.
7.5.16 Proposition Let D be a compact subset of C with open interior and smooth boundary aD. Asswne !hat the polynomial p : C ~ C has no zeros on aD. Then the total number of zeros of p which lie in the interior of D, counting multiplicities, equals the degree of the map p/l pi: aD~ S'. Proof Let z" ... , zm be the roots of p in the interior of D with multiplicities k(1), .... k(m). Around each zi construct an open disk Di centered at zi' Di C D, such that aD n aDi = 0, and aDi n aDj = 0, for all i j. Then V = D \ (D I U ... U Dm) is a smooth compact two-dimensional
*
Chapter 7 Integration on Manifolds
550
u aO I u ... aO m . The boundary orientation of aO j induced by V is opposite to the usual boundary orientation of aOj as the boundary of the disk OJ; see Figure
manifold whose boundary is aD
7.5.2. Since pI! p I is defined on all of V. corollary 7.5.11 implies that the degree of pI! pi: aV
--> Sl is zero. But the degree of a map defined on a disjoint union of manifolds is the sum of the individual degrees and thus the degree of pI! p I on aD equals the sum of the degrees of pI! p I on all aDj' The proposition is therefore proved if we show that the degree of pI! p I on aOj is the multiplicity k(i) of the root Zj'
0,
D
Figure 7.5.2
Let m
r(z) = c
II (z - Zj)
kG)
so p(z)
j=l.j;tj and the only zero of p(z) in the disk OJ is Zj' Then Zj + Rjz
aDj, where R j is the radius of OJ. is a diffeomorphism and therefore the degree of pI! pi: aO j --> SI equals the degree of (p 0 Sl. The homotopy H: [0. 1] x SI --> Sl of zk(j)arg r(z) with (p
0
E
I r(zj + tRjz) I
is proper and smooth. since Zj + tRjZ E OJ for all Z E SI. t E [0. IJ. Thus in aO j we have deg(p/I p j) = deg«p 0
Let U be an open subset of ]Rn and f: U -->]Rn be a C l proper map. Assume there is a closed subset K C U such that for all x E U \ K. the Jacobian J(f)(x) does 7.5.17 Proposition
Section 7.S Introduction to Hodge-DeRham Theory
SSI
fUJt change sign and is not identically equal to zero. Then f is surjective. Proof The map f cannot be constant since the Jacobian J(f)(x) is not identically zero for all x E U\K. For the same reason. f has a regular value yE f(U\K). for if all values in f(V\K) are singular. J(f) will vanish on V \ K. If Y E f(V \ K) is a regular value of f then sign(Txf) does not change for all x E r-1(y) so by the degree theorem 7.S.10. deg f"* O. which then implies that fis onto. The orientation preserving character of proper diffeomorphisms is characterized in terms of the degre as follows. 7.S.18 Proposition Let M and N be oriented boundaryless connected manifolds and f: M ~ N
a proper local diffeomorphism. Then deg f = I. if and only if f is an orientation preserving diffeomorphism. Proof If f is an orientation preserving diffeomorphism then deg f = 1 by theorem 7.S.10. Conversely. let f be a proper local diffeomorphism with deg f= 1. Define V± = {m E M I sign Tmf = il}. Since f is a local diffeomorphism. V± are open in M. Connectedness of M and M = V+ U V_. V+ n V_ = 0 imply that M = V+ or M = V_. Let us show that V_ = 0. Since deg f = I. f is onto and hence if n E N. f- 1(n)"* 0 is a discrete submanifold of M. Properness of f implies that f- 1(n) = {m(I) ....• m(k)}. Since f is a local diffeomorphism of a neighborhood Vi of m(i) onto a neighborhood V of n. sign Tm(J is the same for all i = I • ...• k (for otherwise J(f) must vanish somewhere). Thus deg f = ±k according to whether Tm(i/ preserves or reverses orientation. Since deg f = I. this implies V_ = 0 and k = I. i.e .. f is injective. Thus f is a bijective local diffeomorphism. i.e .. a diffeomorphism. -
fl:n
~ SUPPLEMENT 7.SB Zero and n-Dimensional Cohomology
Here we compute HO(M) and HO(M) for a connected n-manifold M. Rt:call LhaL the cohomology groups are defined by Hk(M) = ker dkjrange d k- t • where d k : nk(M) ~ n k+1(M) is the exterior differential. If n ke(M) denotes the k-forms with compact support. then d k : nkC<M) ~ n k+ 1e(M) and one forms in the same manner Hke(M). the compactly supported cohomology groups of M. Thus HO(M) = (f E .'F(M) I d f = O} == IR since any locally constant function on a connected space is constant. If M were not connected. then ~(M) = IRe. where c is the number of connected components of M. By the Poincare \emma. if M is contractible. then Hq(M) = 0 for q"* O. The rest of this supplement is devoted to the proof and applications of the following special case of deRham's theorem. (We will provide a supplementary chapter with further development of these ideas on request.)
Chapter 7 InJegratWn on Manifolds
SS2
7.S.19 Theorem Let M be a boundary less connected n-manifold. (i) If M is orientable, then H"c(M) == JR, the isomorphism being given by integration: [00]1-+
fMOO. In particular OOE n"/M) is exact iff fMOO=O.
(ii) If M is nonorientable, then Hnc(M) = O. (iii) If M is non-compact, orientable or not, then W(M) =
o.
Before starting the actual proof, let us discuss (i). The integration mapping
fM : n"(M) ~ JR
is linear and onto. To see that it is onto, let 00 be an n-form with support in a chart in which the local expression is 00 = f dx I /\ ... /\ dx" with f a bump function. Then
fM 00 = fRn f(x) dx > O.
Since we can multiply 00 by any scalar, the integration map is onto. Any 00 with nonzero integral cannot be exact by Stokes' theorem. This last remark also shows that integration induces a mapping, which we shall still call integration,
fM : Hnc(M) ~ JR,
which is linear and onto. Thus,
in order to show that it is an isomorphism as (i) states, it is necessary and sufficient to prove it is injective, i.e., to show that if
fM 00 = 0
for 00 E n"(M), then 00 is exact. The proof of this will be
done in the following lemmas. 7.S.20 Lemma Theorem 7.S.19 holds for M = s!. Proof Let p : JR ~ SI be given by pet) = eit and 00 E nl(s!). Then p*oo = f dt for f E .'T(JR) a 21t-periodic function. Let F be an anti derivative of f. Since
o=
1 = f. 00
Sl
t+21t
f(s) ds = F(t + 21t) - F(t)
t
for all t E JR, we conclude that F is also 21t-periodic, so it induces a unique map G E 1(SI), determined by p*G = F. Hence p*oo = dF = p*dG implies 00 = dG since p is a surjective submersion. T 7.S.21 Lemma Theorem 7.S.19 holds for M = SO, s> 1. Proof This will be done by induction on n, the case n = 1 being the previous lemma. Write sn = NUS,
where N = {x E S" I x"+! ~ O} is the closed northern hemisphere and S = {x E S" I
x"+1 ~ O} the closed southern hemisphere. Then N n S = sn-I is oriented in two different ways as the boundary of N and S, respectively. Let
ON = {x E sn I x"+1 > - £},
Os = {x E S"
I
x"+1 < t} be open contractible neighborhoods of N and S, respectively. Thus by the Poincare lemma, there exist aN E nn-I(ON)' as E nn-I(Os) such that daN = 00 on ON' das = 00 on Os. Hence by hypothesis and Slokes' theorem,
Section 7.5 Introduction to Hodge-DeRham Theory
553
where i : sn-I ~ sn is the inclusion of sn-I as the equator of sn; the minus sign appears on the second integral because the orientations of Sn-I and as are opposite. By induction, i*(aN - as) E on-I(sn-I) is exact.
°
°
= ON n Os and note that the map r: ~ sn-I, sending each XES to rex) E Let sn-I, the intersection of the meridian through x with the equator sn-I, is smooth. Then r 0 i is
the identity on sn-I. Also, i 0 r is homotopic to the identity of 0, the homotopy being given by
°
sliding x E along the meridian to rex). Since deaN - as) = ffi - ffi = 0 on 0, by Theorem 6.4.16 we conclude that (aN - as) - r*i*(a N - ag) is exact on 0. But we just showed that i*(~ - ag) E on-I(sn-I) is exact, and hence r*i*(~ - as) E on-I(o) is also exact. Hence aN - as
E
on-I(o) is exact. Thus, there exists ~ E on-2(0) such that aN - as = d~ on 0. Now
use a bump function to extend ~ to a form y E on-2(sn) so that on 0, ~
V, where V is an open set such that cl(U)
C
=y
and y =0 on sn \
V. Then if x EN if
XES
is by construction COO and dA = ffi. " 7.5.22 Lemma A compactly supported n-form ffi
compactly supported (n - I)-form on ]Rn Proof Let
0:
sn
~]Rn
E
on(]Rn) is the exterior derivative of a
iff IRn ffi = O.
be the stereographic projection from the north pole (0, ... , 1) E sn onto
]Rn and assume without loss of generality that (0, ... , 1) eo- I (supp ffi). By the previous lemma, O*ffi
= do.,
for some
a
E
on-I(sn) since 0
= IRn ffi = Isn O*ffi
by the change-of-variables
formula. But 0* ffi = do. is zero in a contractible neighborhood U of the north pole, so that by the Poincare lemma, a
=d~
on-2(sn) such that ~
=y
on U, where ~
E
on-2(u). Now extend ~ to an (n - 2)-form
yE
on U and y = 0 outside a neighborhood of cl(U). But then 0*(0.-
dy) is compactly supported in ]Rn and do*(a - dy) = o*da = ffi. " 7.2.23 Lemma Let M be a boundaryless connected n-manifold. Then Wc(M) is at most
one-dimensional. Proof Let (UO'
fR"
;::: 0 and f(x) dx = 1. To prove the lemma, it is sufficient to show that for every 11 there exists a number c E ]R such that 11 - Cffi = d~ for some ~ E Q"-I c(M).
Eonc(M)
Chapter 7 Integration on Manifolds
554
First assume'll
E
nnc(M) has supp('ll) entirely contained in a chart (V,
Vi' ... , Vk be a fmite covering of a curve starting in Vo and ending on V k = V such that Vi n V i+1 =0. Let u i E nn/v), i = 1, ... , k - I be non-negative n-forms such that supp(u) c Vi' supp(u)
n Vi+1 '* 0, and
fRn
= 1. Let Uo = co, and
~ = 'll. But then
by the change-of-variables formula, so that with c i = -1/ fRn
i = 1, ... , k.
Put c = ck ... c 1 and
Then 'll- cco = Uk - cu o = Uk - ckUk_1 + ck(Uk_1 - ck_1Uk_ 2) + ... + (ck ... C2)(U 1 - c 1uo)
= d~k + ckd~k_1 + ... + (ck ... c2)d~1 = d~. Let 'll
E
nnc(M) be arbitrary and {Xi I i = 1, ... , k} a partition of unity subordinate to the
given atlas {(Vi'
C=
L.
k
Ci
and
U =
i=1
L.
Ui
E
nn-I c(M),
i=1
then k
1'\ - cco =
L.
k
(Xi'll - CiCO)
i=1
=
L.
dUi
= du.
T
i=1
Proof of theorem 7.5.19 (i) By the preceding lemma, II"c(M) is zero- or one-dimensional. We have seen that
f
M :
one-dimensional; i.e.,
II" c (M) -t IR
is linear and onto so that necessarily
Hn c(M)
is
fMCO = 0 iff co is exact.
(ii) Let M be the oriented double covering of M and It: M -t M the canonical projection. Define It#: Hn(M) -t Hn(M) by It#[u) = [It*u). We shall first prove that It# is the zero map. Let {VJ be an open covering of M by chart domains and
{xJ
a subordinate partition of unity.
Let n-I(V) = u/ u ut Then (V~ Ij = 1, 2} is an open covering of M by chart domains and the maps 'l'ii =Xi 0 It/2 : u~ -t JR, j = 1,2, form a subordinate partition of unity on M. Let U E nn c(M). Then
Section 7.5 Introduction to Hodge-DeRham Theory
555
0,
each term vanishing since their push-forwards by the coordinate maps coincide on lR. n and Vj I and
V?
have opposite orientations. By (i), we conclude that 1t*a = d~ for some ~ E nn-I(M);
i.e., 1t#[a] = [1t*a] = [0] for all [a] E H"c(M). We shall now prove that n# is injective, which will show that Hnc(M) = O. Let a E nn C<M) be such that 1t* a = d ~ for some ~ E nn-I e( tvt) and let r: M ~ M be the diffeomorphism associating to (m, [ro]) E M the point (m, [-ro]) E M. Then clearly 1t 0 r = 1t so that d(r*~) = r*(d~) = r*1t*a = (1t
r)*a = 1t*a = d~. Define YE nn-I e (M) by setting y= (l/2)(~ +r*~) and note that r*y= y and dy= (d~ + dr*~)/2 = d~ = 1t*a. But y projects to a 0
well-defined form yE nn-Ie(M) such that 1t*Y= y, since r*y=
i
Thus 1t*a = dy= d1t*y=
1t*dy, which implies that a = dy, since 1t is a surjective submersion. (iii) Assume first that ro E nn e(M) has its support contained in a relatively compact chart domain V I of M. Then out of a finite open relatively compact covering of cl(V I) by chart domains, pick a relatively compact chart domain V 2 which does not intersect supp(ro). Working with cl(U 2) \ V I' find a relatively compact chart domain V3 such that V I n V3 = 0, V 2 n V3 'F 0, V3 n (M \ (VI U V 2 » 'F 0. Proceed inductively to find a sequence {Vn} of relatively compact chart domains such that Vn n-1, n, n+1, and such that supp(ro)
n Vn+1 'F 0, C
Vn n Vn_1 'F 0, Vn n Vrn = 0 for all m 'F Vi' V 2 n supp(ro) = 0. Since M is not compact, this
sequence can be chosen to be infinite; see Figure 7.5.3.
Figure 7.5.3 Now choose in each Vn n Vn+1 an n-form ron with compact support such that
Since Hne(V n) = lR. by (i), Vn being orientable, ron-I and ron define the same cohomology
556
Chapter 7 Integration on Manifolds
class, i.e., there is lln E nn-I C
(On-l
= (On
recursively (0 = dll1 + (01 = d(111 + 112) + Wz = ... = d
+ dlln. If we let (00
[t
= (0,
we get
lliJ + (On = ....
,=1
We claim that (0 = d(Ln.;,ll1n)' where the sum is finite since any point of the manifold belongs to at most two U;s. Thus, if pE Un~IUn' let pE Un sothat d(Ln~ll1n)(p)=dlln_l(p)+dlln(P)+ dll n+I(P) = dll n-I(P) + (On_I(P) - (On+I(P) = dll n_I(P) + (On_I(P) - (On+2(P) - dll n+2(P) with the convention 110 = O. Since Un n U n+2 =0 and supp (On+2 ' supp lln+2 C U n+2, it follows that the last two terms vanish. Thus d(Ln;o,ll1n)(P) = dll n_I(P) + (On_I(P). If n = I, this proves the desired equality. If n ~ 2, then d(Ln~ll1n)(P) = dll n_I(P) + (Onjp) = (On_2(P) and Un n Un_2 =0
= O.
implies that d(Ln;o,ll1n)
Since also (O(p)
=0
in this case, the desired equality holds
again. Finally, if p Ii" U n~l Un' then both sides of the equality are zero and we showed that (0 is exact, (0 = d(Ln~ll1n)' with SUPP(Ln~ll1n) C U n~l Un. Now if (0 E nn(M), let ((Ui' g)} be a partition of unity subordinate to a locally finite atla~ of M whose chart domains are relatively compact. Thus supp(g,(O)
C U i and by what we just proved, is;(O = dlli' with sUPP(l1i) contained in the union of the chain of open sets (Uni), U 1i =
Ui' as described above. Refine each such chain, such that all its elements are one of the Uj's. Since at most two of the U ni intersect for each fixed i, it follows that the sum Lilli is locally finite and therefore 11
=Lilli
E nn-I(M). Finally, (0 = Ligi(O
=Lidlli = dll,
thus showing that
(0 is exact and hence Hn(M) = O. • One can use this result -t
a~
an alternative method to introduce the degree of a proper map f: M
N between oriented manifolds; i.e., that integer deg(f) such that
r f* 11 =
JM
deg(f)
r 11
IN
for any 11 E nnc(N). Indeed, since the isomorphism Hnc(N):; lit is given by [11] ......
fNll,
the
linear map [11] ...... fMr*l1 of H" c(N) to lit must be some multiple of this isomorphism:
J, f* 11 = deg(f) J,11 for all 11 E nn c(N). M
N
To prove that deg(f) is an integer in this context and that the formula (ii) for deg(f) is independent on the regular value y, note that if y is any regular value of f and x E f-l(y), then there exist compact neighborhoods V of y and U of x such that flU: U -t V is a diffeomorphism. Since f-l(y) is compact and discrete, it must be finite, say f-l(y) = (Xl' ... , xk ). This shows that f-I(V) = U I U ... U Uk with all U i disjoint and the sum in the degree formula is finite. Shrink V if necessary to lie in a chart domain. Now choose 11 E nn c(N) satisfying sUPP(l1)
C
V. Then
Section 7.51ntroduction to Hodge-DeRham Theory
by the change-of-variables formula in
jRo,
557
so the claim follows.
Degree theory can be extended to infinite dimensions as well and has important applications to
partial differential equations and bifurcations. This theory is similar in spirit to the above and
was developed by Leray and Schauder in the 1930s. See Chow and Hale [1982], Choquet-Bruhat et aI. [1977], Nirenberg [1974], and Elworthy and Tromba [1970b] for modem accounts. flJJ
Exercises 7.5A Poincare duality
Show that * induces an isomorphism *: Hk ~
w-l<
and H\ ~
Ho.k e ·
7.5B (For students knowing some algebraic topology) Develop some basic properties of deRham cohomology groups such as homotopy invariance, exact sequences, Mayer-Vietoris sequences and excision. Use this to compute the cohomology of some standard simple spaces (tori, spheres, projective spaces). 7.5C (i) Show that any smooth vector field X on a compact Riemannian manifold (M, g) can be written uniquely as X=Y+gradp where Y has zero divergence (and is parallel to aM if M has boundary). (ii) Show directly that the equation L\p = - div X, (grad p)-n = X· n is formally soluble using the ideas of the Fredholm alternative. 7.5D Show that any symmetric two-tensor h on a compact Riemannian manifold (M. g) can be uniquely decomposed in the form
where 15k = 0, 15 being the divergence of g, defined by 15k = (Log)*k, where (Log) * is the adjoint of the operator X 1-+ Lxg. (See Berger and Ebin [1969] and Cantor [1981] for more information.)
Chapter 7 Integration on Manifolds
SS8
7.SE Let a
E
(lk-t(M), P E nk(M), where M is a compact oriented Riemannian manifold with
boundary. Show that (i)
(da, P) - (a, dP) =
(Hint: Show that * ap
=(-l)kd*P
(ii)
faM a
/\ *p.
and use Stokes theorem or 7.2.13.)
(daa, P) - (da, dP) =
faM aa /\ *P
(da, dP) - (a, adp) =
faM a
/\ *dP
(iii) (Green's formula)
(~a, P) - (a, ~P) =
faMe aa /\ *P -
dP /\ *a + a /\ *dP - P /\ *da).
(Hint: Show first that
(~a, P) - (da, dP) - ( aa, ap) =
faMe aa /\ *P -
P /\ *da).)
7.SF (For students knowing algebraic topology) Define relative cohomology groups and relate them to the Hodge decomposition for manifolds with boundary. 7.SG Prove the local formulas
(aa)
r
... r k 1
1
=
k+ 1 I det[gi) I
where it < ... < ik and a
E
-1/2
nk+I(M) according to the following guidelines. Prove first the
second formula. Work in a chart (U,
= 1. Then extend x
]Rn,
denote it by a' and consider the set
B 4 (0).
(i) Show from 7.SE(i) that (dP, a') = (P, aa') for any P E n k+I(B 4 (0». (ii) In the explicit expression for (dP, a'), perform an integration by parts and justify it.
Section 7.5 Introduction to Hodge-DeRham Theory
(iii) Find the expression for ba' by comparing
(~,
559
ba') with the expression found in (iii)
and argue that it must hold on
~
M be a diffeomorphism of an oriented Riemannian manifold (M, g) and let 15g
denote the codifferential corresponding to the metric g and
(')g the inner product on
(lk(M) corresponding to the metric g. Show that (i) (a, ~)g = (
E
E
(lk(M). (Hint: Use the fact that d and 15
are adjoints.) 7.51
(i) Let cI and c2 be two differentiably homotopic curves and
0) E
(l1(M) a closed
one-form. Show that
(ii)
Let M be simply connected. Show that Hl(M) = O. (Hint: For rna E M, let c be
(iii)
Show that HI(SI) cf. 0 by exhibiting a closed one-form that is not exact.
a curve from Il10 to m
E
M. Then f(m)
=Ie 0)
7.5 J The H opj degree theorem states that f and g: Mn
is well defined by (i) and df = 0).)
~
Sn
are homotopic iff they have
the same degree. By consulting references if necessary, prove this theorem in the context of Supplements 7.5A ,B. (Hint: Consult GuiIIemin and Pollack [1974] and Hirsch [1976].) 7.5 K What does the degree of a map have to do with Exercise 7.2D on integration over the fiber? Give some examples and a discussion. 7.5 L Show that the equations Zl3
+ sin(1 z 12 )z7 + 3z4 + 2 =0
z8 + cosCl z j2)z5 +510g(1 z 12 )z4 + 53
=0
have a root. 7.5M Let f: M ~ N where M and N are compact orientable boundary less manifolds and N is contractible. Show that deg(f)
=O.
Conclude that the only contractible compact manifold
(orientable or not) is the one-point space. (Hint: Show that the oriented double covering of a contractible non-orientable manifold is contractible.) 7.5 N Show that every smooth map f: Sn ~ Tn, n> 1 has degree zero. (Hint: Show that f is homotopic to a constant map.) Conclude that sn and Tn are not diffeomorphic if n> 1.
Chapter 8 Applications This chapter presents some applications of manifold theory and tensor analysis to physics and engineering. Our selection is a of limited scope and depth, with the intention of providing an introduction to the techniques. There are many other applications of the ideas of this book as well. We list below a few selected references for further reading in the same spirt. l. Arnold [1978], Abraham and Marsden [1978], Chernoff and Marsden [1974], Weinstein
[1977], and Marsden [1981] for Hamiltonian mechanics. 2. Marsden and Hughes [1983] for elasticity theory. 3. Flanders [1963], von Westenholtz [1981], and Schutz [1980] for applications to control theory. 4. Hermann [1980], Knowles [1981], and Schutz [1980] for diverse applications. 5. Bleecker [1981] for Yang-Mills theory. 6. Misner, Thome, and Wheeler [1973] and Hawking and Ellis [1973] for general relativity.
§8.1 Hamiltonian Mechanics Our starting point is Newton's second law in ]R3, which states that a particle which has mass m > 0, and is moving in a given potential field Vex) where x E ]R3, moves along a curve
x
x
x(t) satisfying the equation of motion m =- grad Vex). If we introduce the momentum p =m and the energy H(x, p) = (l!2m)1I p 112 + Vex) then Newton's law becomes Hamilton's equations: ·i aH x = api'
. aH Pi = - ax i
i = 1,2,3.
To study this system of first-order equations for given H, we introduce the matrix
where I is the 3 x 3 identity; note that the equations become ~ =.If grad H(~) where ~ = (x, p).
Section 8.1 Hamiltoniiln Mechanics
561
In complex notation, setting z =x + ip, they may be written as .
z=
2' aH 1
az
Suppose we make a change of coordinates, w = f(~), where f: lR6 ~ lR6 ~(t) satisfies Hamilton's equations, the equations satisfied by w(t) are
is smooth. If
w= A~ = All grad~H(~) =
AlTA *gradwH(~(w», where Aij = (awi/a~j) is the Jacobian matrix of f, A * is the transpose of
A and
~(w)
denotes the inverse function of f. The equations for w will be Hamiltonian with
energy K(w) = H(~(w»
if AlTA * = IT. A transformation satisfying this condition is called
canonical or symplectic. One of the things we do in this chapter is to give a coordinate free treatment of this and related concepts. The space lR3 x lR3 of the ~'s is called the phase space. For a system of N particles we would use lR3 :-1 x
lR,3N.
However, many fundamental physical systems have a phase space that is
a manifold rather than Euclidean space, so doing mechanics soley in the context of Euclidean space is to constraining. For example, the phase space for the motion of a rigid body about a fixed point is the tangent bundle of the group 80(3) of 3 x 3 orthogoanl matrices with determinant +1. This manifold is diffeomorphic to rup3 and is topologically nontrivial. To generalize the notion of a Hamiltonian system to the context of manifolds, we first need to geometrize the symplectic matrix IT. In infinite dimensions a few technical points need attention before proceeding.
Let E be a Banach space and B : E x E ~ lR a continuous bilinear mapping. Then B induces a continuous map Bb: E ~ E*, e 1-+ Bb(e) defined by Bb(e)· f = B(e, f). We call B
weakly nondengerate if Bb is injective, i.e., B(e, f) = 0 for all fEE implies e = O. We call B nondengerate or strongly nondegenerate if Bb is an isomorphism. By the open mapping theorm it follows that B is nondegenrate iff B is weakly nondegenerate and Bb is onto. If E is finite dimensional there is no difference between strong and weak nondegeneracy. However, if infinite dimensions the distinction is important to bear in mind, and the issue does
come up in basic examples, as we shall see in Supplement 8.1A. Let M be a Banach manifold. By a weak Riemannian structure we mean a smooth assignment g: x 1-+ < , >x = g(x) of a weakly nondegenerate inner product (not necessarily complete) to each tangent space TxM. Here smooth means that in a local chart U C E, the mapping g: x 1-+ <, >x E L2(E, E; lR) is smooth, where L2(E, E; lR) denotes the Banach space of bilinear maps of Ex E to R Equivalently, smooth means g is smooth as a section of the vector bundle L2(TM, TM; lR) whose fiber at x E M is L2(TxM, TxM; lR). By a Riemanniiln
manifold we mean a weak Ricmannian manifold in which the topology of Hilbert space.
<, >x
<, >x
is nondegenerate. Equivalently,
is complete on TxM, so that the model space E must be isomorphic to a
For example the L 2 inner product r1
0
(f, g ) = Jo f(x)g(x)dx on E = C ([0,
n, lR)
Chapter 8 Applications
562
is a weak Riemannian metric on E but is not a Riemannian metric. 8.1.1 Definition Let P be a manifold modeled on a Banach space E. By a symplecticform we
mean a two-form (J) on P such that (i) (J) is closed, i.e., d(J)
(ii) for each
Z E
=0;
P, (J)x: TzP x TzP
~
IR is weakly nondegenerate.
If (J)z in (ii) is nondegenerate, we speak of a strong symplectic form. If(ii) is dropped we
refer to (J) as a presymplectic form. (For the moment the reader may wish to assume P is finite dimensional, in which case the weak - strong distinction vanishes.) The first result is referred to as Darboux's theorem. Our proof follows Moser [1965] and Weinstein [1969]. 8.1.2 The Darboux Theorem Let (J) be a strong symplectic form on the Banach manifold P.
Then for each x E P there is a local coordinate chart about x in which (J) is constant. Proof The proof proceeds by the Lie transform method 5.4.7. We can assume P = E and x = 0 E E. Let (J)I be the constant form equaling (J)o = ro(O). Let n = (J)I - (J) and (J)I = (J) + tn, for
o$
t $ 1. For each t, (J)I (0) = (J)(O) is nondegenerate. Hence by openness of the set of linear
isomorphisms of E to E*, there is a neighborhood of 0 on which (J)I is nondegenerate for all 0$ t $ 1. We can assume that this neighborhood is a ball. Thus by the Poincare lemma, n = da for some one-form a. We can suppose a(O) = O. Define a smooth vector field i",
ro. = -
~
by
a,
which is possible since (J)t is strongly non-degenerate. Since Xt(O) there is a sufficiently small ball on which the integral curves of
~
= 0, by corollary 4.1.25,
will be defined for time at least
one. Let F t be the flow of Xt starting at Fo = identity. By the Lie derivative formula for time-dependent vector fields (Theorem 5.4.4) we have
Therefore,
Ft (J)I = F0 *(J)o
= (J), so F I provides the chart transforming (J) to the constant form
(J)I' • We note without proof that such a result is not true for Riemannian structures unless they are flat. Also, the analogue of Darboux's theorem is known to be not valid for weak symplectic forms. (For the example, see Abraham and Marsden [1978], Exercise 3.2H and for conditions under which it is valid, see Marsden [1981].)
Section 8.1 Hamiltonum Mechanics
563
8.1.3 Corollary If P is finite dimensional and OJ is a symplectic form, then (i) P is even dimensional, say dim P = 2n; (ii) locally about each point there are coordinates x I•...• xn. yl ..... yn such that n
OJ=
L
dx i
/\
d/.
i=1
Such coordinates are called canonical..
Proof By elementary linear algebra. any skew symmetrie bilinear form that is nondegenerate has the canonical form
where I is the n x n identity. (This is proved by the same method as 6.2.9.) This is the matrix version of (ii) pointwise on P. The result now follows from Darboux's theorem. _ As a bilinear form.
is given in canonical coordinates by w«x l • YI)' (x 2• Y2» = (Y2' x l )(y I' x2)· In complex notation with z = x + iy it reads w(zi' z2) =- 1m (zl' z2)' This form for canonical coordinates extends to infinite dimensions (see Cook [1966), Chernoff and Marsden [1974] and Abraham and Marsden [1978]. section 3.1 for details). OJ
Now we are ready to consider canonical symplectic forms. 8.1.4 Definition Let Q be a manifold modeled on a Banach space E. Let T*Q be its contangent bundle, and 1t : T*Q ~ Q the projection. Define the canonical one-form e on T*Q I:Jy e(a)w = a . T1t(w). where a
E
Tq *Q and
In a chart U
C
WE
Tu(T*Q). The canonical two-form is defined by
E
de.
E. the formula for e becomes e(x. a) . (e.
where (x. a) written
OJ = -
U x E* and (e.~)
E
~) =
a(e).
Ex E*. If Q is finite dimensional. this formula may be
where ql ..... qn. Pl' .... Pn are coordinates for T*Q and the summation convention is enforced. Using the local formula for d from formula (6) in Table 6.4.1.
564
Chapter 8 Applications
or. in the finite-dimensional case.
In the infinite-dimensional case one can check that co is weakly non degenerate and is strongly nondegenerate iff E is reflexive (Marsden [1968b]). If ( •
>x
is a weak Riemannian (or pseudo-Riemannian) metric on Q. the smooth vector
t':
bundle map x' X E Q. is injective on fibers. If (. > is a strong Riemannian metric. then
=- d<3
where <3 =
Q
become
and
where ql •...• qn. til •...• tin are coordinates for TQ. This follows by substituting Pi = gij tij into CO =dqi A dPi' In the infinite-dimensional case. if (. >x is a weak metric. then co is a weak symplectic form locally given by and
where Dx denotes the derivative with respect to x. One can also check that if (. >x is a strong metric and Q is modeled on a reflexive space. then Q is a strong symplectic form. 8.1.5 Definition Let (P. co) be a symplectic manifold. A (smooth) map canonical or symplectic when [*co = co. If follows that [*(co A
..• A
co) = co
A'"
A
f: P
~
P is called
co (k times). If P is 2n-dimensional. then 11 =
co A ... A co (n times) is nowhere vanishing. so is a volume form; for instance by a computation one finds 11 to be a multiple of the standard Euclidean volume in canonical coordinates. In paticular. note symplectic manifolds are orientable. We call 11 the phase volume or the Liouville form. Thus a symplectic map preserves the phase volume. and so is necessarily a local diffeomorphism. A map f: PI ~ P2 between symplectic manifolds (PI' COl) and (P2 • COz) is called symplectic if f" co2 = COl' As above. if P I and P2 have the same dimension. then f is a local diffeomoorphism and preserves the phase volume.
565
Section 8.1 HamiIJonUln Mechanics
We now discuss symplectic maps induced by maps on the base space of a cotangent bundle. 8.1.6 Proposition Let f: Q 1 ~ Q2 be a diffeonwrphism; define the cotangent lift of f I:Jy
where q E Q2' u q E T q*Q2 and VETr'(q) QI; i.e .• T*f is the pointwise adjoint of Tf. Then T*f is symplectic and infact (T*f)*9 1 = 9 2 where 9i is the canonical one-form on Qi' i = 1.2. Proof Let 1ti : T*Q i ~ Qi be the cotangent bundle projection. i = 1.2. For w in the tangent space to T*Q2 at u q • we have (T*f)*9 1(uq)(w)
=9 1(T*f(Uq))(TT*f. w) =T*f(Uq) . (T1t 1 . TT*f· w) = T*f (uq) . (T( ltl =
since. by construction. f
0
ltl
0
u q • (T1t 2
.
0
T*f) . w) = u q . (T(f 0
ltl
0
T*f) . w
w) = 9i U q) . w
T*f = 1t 2 · •
In coordinates. if we write f(q I..... qn)
= (QI ..... Qn).
then T*f has the effect
where aQi p=-p. J
a~
,
(evaluated at the corresponding points). That this transformation is always canonical and in fact preserves the canonical one-form may be verified directly:
PidQ
i
aQi
k
k
= Pi - k dq = Pk dq . aq
Sometimes one refers to canonical transformations of this type as "point transformations" since they arise from general diffeomorphisms of Q 1 to Q2' Notice that lifts of diffeomorphisms satisfy
that is. the following diagram commutes:
Chapter 8 Applications
566
T*Q 2
T*f
~l Q2 f
..
T*Q,
..
Q,
I"
Notice also that T*(f 0 g) = T*g 0 T*f and compare with T(f 0 g) = Tf 0 Tg. 8.1.7 Corollary If Q, and Q2 are Riemannian (or pseudo-Riemannian) manifolds and f: Q, ~ Q2 is an isometry, then Tf: TQ, ~ TQ 2 is symplectic, and infact (Tf)*0 2 = 0,. Proof This follows from the identity Tf= g2#
0
(T*f)-' og/'. All maps in this composition are
symplectic and thus Tf is as well. • So far no mention has been made of Hamilton's equations. Now we are ready to consider them. 8.1.8 Definition Let (P. w) be a symplectic manifold. A vector field X: P
~
TP is called
Hamiltonio.n if there is a C' function H: P ~ IR such that
We say X is locally Hamiltonio.n if ixw is closed. We write X = X H because usually in examples one is given H and then one constmc's the Hamiltonian vector field XII" If w is only weakly nondegenerate. then given a smooth function H:P
~
R XII need not exist on all of P. Rather than being a pathology. this is quite essential in
infinite dimensions, for the vector fields then correspond to partial differential equations and are only densely defined. The condition ix"w = dH is equivalent to wz(XH(z).v) = dH(z) . v. for z E P and v E TzP. Let us express this condition in canonical coordinates (q' •...• qO. p" ...• Po) on a 2n-dimensional symplectic manifold P, i.e., when W = dqi /\ dPi. If X = Aia/aqi
+ Bia/api' then
Section 8.1 Hamiltonian Mechanics
567
This equals aH i aH dH=-.dq +-a dPi aq' Pi iff
i.e.,
If
where I is the n x n identity matrix, the formula for XH can be expressed as
More intrinsically, one can write XH = m# dH, so one sometimes says that XH is the symplectic
gradunt of H. Note that the formula XH = Jf grad H is a little misleading in this respect, since no metric structure is actually needed and it is really the differential and not the gradient that is essential. From the local expression for XH we see that (qi(t), Pi(t» is an integral curve of XH iff
Hamilton's equations hold; .i aH aH q = api' Pi = - aqi .
We now give a couple of simple properties of Hamiltonian systems. The proofs are a bit more technical for densely defined vector fields, so for purposes of these theorems, we work with C"" vector fields. 8.1.9 Theorem Let X H be a Hamiltonian vector field on the (weak) symplectic manifold (P, m)
and let Ft be the flow of X H . Then
(i) Ft is symplectic, i.e. Ft*m = m, and (ii) energy is conserved, i.e. H 0 Ft = Ft' Proof Since Fa = identity, it suffices to show that (d/dt)F/m = O. But by the basic connection between Lie derivatives and flows (§5.4 and §6.4):
Chapter 8 Applications
568 d
*
*
* •
*.
-dt Fl ffi(X) = Fl (Lx Hro)(x)= Fl (dl XHro)(x) + Fl (lydroXx). .... The first term is zero because it is d 2 H = 0 and the second is zero because ro is closed. (ii) By the chain rule,
But this is zero
~
view of the skew symmetry of roo _
A corollary of (i) in finite dimensions is Liouville's theorem: Fl preserves the phase
volume. This is seen directly in canonical coordinates by observing that XH is divergence-free.
We shall now generalize (ii). Define for any functions f, g: U -t JR, U open in P, their Poisson bracket by
Since we see that {f,g}= Lx/=-L~. If
and
Thus
iff
preserves the Poisson bracket iff it preserves Hamilton's equations. We have . • IX ro = d (*0
so that by the (weak) nondcgeneracy of ro and the fact that any v E TzP equals some X h (z) for a C~ function h defined in a neighborhood of z, we conclude that
Section 8.1 Hamiltonian Mechanics
569
8.1.10 Proposition Let (PI' 0)1) and (P 2 '0)2) be symplectic manifolds and
~
P2 a
(i)
(ii) .c for any local f: U ~ lR, where U is open in P 2 (i.e.,
Hamilton's equations). Conservation of energy is generalized in the following way.
8.1.11 Corollary (i) Let X H be a Hamiltonian vector field on the (weak) symplectic manifold (P,O) with (local) flow Fe Thenfor any C~ function f: U ~ JR., U open in P, we have
(ii) The curve c(t) satisfies Hamilton's equations defined by H if and only
if
d dt f(c(t» = {f, H} (c(t»
for any
C~
function f: U
~
JR., where U is open in P.
Proof (i)
by the fonnula for Lie derivatives and the previous proposition. (ii) Sinee df(c(t»/dt =df(c(t» . (dc/dt) and (C, H}(c(t» = (Lx/)(c(t» = df(c(t» . XH(c(t»,
the equation in the statement of the proposition holds iff e'(t) = XH(c(t»
by the Hahn-Banach
theorem and 4.2.14 . • One writes i = {f, H} to stand for the equation in (ii). This equation is called the equation of motion in Poisson bracketformulation. Two functions f, g : P ~ JR. are said to be in involution or to Poisson commute if if, g) = O. Any function Poisson commuting with the Hamiltonian of a mechanical system is, by 8.1.11, necessarily constant along on the flow of the Hamiltonian vector field. This is why such functions are called constants of the motion. A classical theorem of Liouville states that in a mechanical system with a 2n-dimensional phase space admitting k constants of the motion in involution and independent almost everywhere (that is, the differentials are independent on an open dense set) one
570
Chapter 8 Applications
can reduce the dimension of the phase space to 2(n - k). In particular, if k = n, the equations of
motion can be "explicitly" integrated. In fact, under certain additional hypotheses, the trajectories of the mechanical system are straight lines on hlgh-dimensional cylinders or tori. If the motion takes place on tori, the explicit integration of the equations of motion goes under the name of finding action-angle variables. See Arnold [1978], and Abraham and Marsden [1978, pp. 392-400) for details and Exercise 8.10 for an example. In infinite-dimensional systems the situation is considerably more complicated. A famous example is the Korteweg-deVries (KdV) equation; for this example we also refer to Abraham and Marsden [1978, pp. 462-72) and references therein. The following supplement gives some elementary but still interesting examples of infinite-dimensional Hamiltonian systems.
~ Supplement S.IA Two Infinite-Dimensional Examples
8.1.12 The Wave Equation as a Hamiltonian System The wave equation for a function u(x, t), where x E lRn and t E lR is given by d 2u
-
=
2 2 V u + m u, (where m 2': 0 is a constant),
dt 2 with u and ~
=i1u/dt
given at t = O. The energy is H(u, u) =
We define H on pairs (u,
u)
~(J 1U 12 dx +
f
II Vu 11 2dx ).
of finite energy by setting
where HI consists of functions in L2 whose first (distributional) derivatives are also in L2. (The Sobolev spaces H' defmed this way are Hilbert spaces that arise in many problems involving partial differential equations. We only treat them informally here.) Let 0 = H2 X H : D~P by •
•
X
HI and define
2
XH(u, u) = (u, V u + m2u).
Let the symplectic form be associated with the L2 metric as in the discussion following 8.1.4, namely • ro«(u, ~), (v, ~)) = vu dx uv dx.
f
-f
It is now an easy verification using integration by parts, to show that XH' co and H are in the
Section 8.1 Hamiltonum Mechanics
571
proper relation, so in this sense the wave equation is Hamiltonian. That the wave equation has a flow on P follows from (the real form of) Stone's theorem (see Supplement 7.4A and Yosida
[1980]). 8.1.13 The Schrodinger Equation Let P = '}{ a complex Hilbert space with
(0
=-21m (,
).
Let Hop be a self-adjoint operator with domain D and let
and H(<jl) Again it is easy to check that
(0,
XH
=(Hop<jl, <jl),
<jl
E
D.
and H are in the correct relation. Thus, X H is
Hamiltonian. Note that \jf(t) is an integral curve of XH if
which is the abstract Schrodinger eqUfltion of quantum mechanics. That X H has a flow is a special case of Stone's theorem. We know from general principles that the flow e itHop will be symplectic. flJJ The additional structure needed for unitarity is exactly complex linearity. Turning our attention to geodesics and to Lagrangian systems, let M be a (weak) Riemannian manifold with metric (, the metric (, (x, v)
E
>x
>x
on the tangent space TxM. The spray S: TM ..... T2M of
is the vector field on TM defined locally by S(x, v) = «x, v), (v, y(x, v»), for
TxM, where y is defined by
and Dx(v, V>X . w means the derivative of (v, v)x with respect to x in the direction of w. If M is finite dimensional, the Christoffel symbols arc defined by putting yi(x, v)
= - rijk(x)vi vk
Equation (1) is equivalent to
The verification that S is well-defined independent of the charts is not too difficult. Notice that y is quadratic in v. We will show below that S is the Hamiltonian vector field on TM associated with the kinetic energy (v, v)/2. The projection of the integral curves of S to M are called geodesics. Their local equations are thus
Chapter 8 Applications
572
'x = y(x.
~).
which in the finite-dimensional case becomes
Xi + rjk
~j ~k = O. i = 1..... n.
The definition of y in (1) makes sense in the infinite as well as the finite-dimensional case, whereas the coordinate definition of
r i jk
makes sense only in finite dimensions. This then
provides a way to deal with geodesics in infinite-dimensional spaces. Let t>-+ (x(t).
vet»~
be an integral curve of S. That is. x(t) = vet)
and
vet)
=y(x(t). v(t».
(2)
As we remarked. these will shortly be shown to be Hamilton's equations of motion in the absence of a potential. To include a potential. let V: M
~
lR be given. At each x. we have the
differential of V. dV(x) E Tx *M. and we define grad Vex) by (grad Vex). w >x = dV(x) . w.
(3)
(In infinite dimensions, it is an extra assumption that grad V exists. since the map T xM ~ T x*M induced by the metric is not necessarily bijective.) The equations of motion in the potential field V are given by x(t) = vet);
vet) =y(x(t). v(t» - grad V(x(t»).
(4)
The total energy. kinetic plus potential. is given by H(v x) = (1/2) IIvxW + Vex). The vector field XH determined by H relative to the symplectic structure on TM induced by the metric. is given by (4). This will be part of a more general derivation of Lagrange's equations given below.
I@'"
Supplement 8.1B Geodesics
Readers familiar with Riemannian geometry can reconcile the present approach to geodesics based on Hamiltonian mechanics to the standard one in the following way. Define the covariant derivative V: 3t(M) x 3t(M) ~ 3t(M) locally by (V x Y)(x) = yx(X(x). Y(x» + DY(x) . X(x). where X(x) and Y(x) are the local representatives of X and Y in the model space E of M
Section 8.1 Hamiltonian Mechanics
and Yx : E x E
-t
573
E denotes the symmetric bilinear continuous mapping defined by polarization of
the quadratic form y(x. v). In finite dimensions. if E
= ]Rn.
quadratic form on ]Rn determined by the Christoffel symbols
then y(x. v) is an JRn-valued
r jk'
The defining relation for Vx Y
becomes n Vx
y _ xiyk....i -
1
where locally i
a
a Xi ayk a + --. - k ax' ax] ax
jk - .
k
•
a
X=X - . and y=y - . ax' ai It is a straightforward exercise to show that the foregoing definition of V x Y is chart independent and that V satisfies the following conditions defining an qffine connection: (i) V is lR-bilinear.
= f Vx Y and VxfY = f Vx Y + X[f]Y. = DY(x) . X(x) - DX(x) . Y(x) = [X. Y](x).
(ii) for f: M -tlR smooth. VIX Y (iii) (V x Y - Vy X)(x)
by the local formula for the Jacobi-Lie bracket of two vector fields. (The equivalence of sprays and affine connections was introduced by Palais and Singer [1960].) If c(t) is a curve in M and X E X(M). the covariant derivative of X along c is defined by
where c is a vector field coinciding with c(t) at the points c(t). Locally. using the chain rule. this becomes DX d (it(c(t» = - Yc(t)(X(c{t)). X(c{t))) + dt X(c(t).
which also shows that the definition of DX/dt depends only on c(t) and not on how c is extended to a vector field. In finite dimensions. the coordinate form of the preceding equation is
where c(t) denotes the tangent vector to the curve at c(t). The vector field X is called autoparallel or is parallel-transported along c if DX/dt = O. Thus c is autoparallel along c iff in any coordinate system we have c(t) - Yc(t)( c(t). c(t» = 0
or. in finite dimensions
574
Chapter 8 Applications
That is, c is autoparallel along c
iff c
is a geodesic.
There is feedback between Hamiltonian systems and Riemannian geometry. For example, conservation of energy for geodesics is a direct consequence of their Hamiltonian character but can also be checked directly. Moreover, the fact that the flow of the geodesic spray on TM consists of canonical transformations is also useful in geometry, for example, in the study of closed geodesics (cf. Klingenberg [1978]). On the other hand, Riemannian geometry raises questions (such as parallel transport and curvature) that are useful in studying Hamiltonian systems.
~
We now generalize the idea of motion in a potential to that of a Lagrangian system; these are, however, still special types of Hamiltonian systems. We begin with a manifold M and a given function L: TM
~
lR called the LagrangWn. In case of motion in a potential, take
which differs from the energy in that -V is used rather than +V. The Lagranian L defines a map called thefiber derivative, IFL: TM ~ T*M as follows: let v, w
E
TxM, and set lFL(V)'W""ddt L(v+ tw)I
",0
.
That is, lFL(v)· w is the derivative of L along the fiber in direction w. In the case of L(v x) = (112) (v x' vX>x - V(x), we see that lFL(v x)' Wx = (vx' wX>x' so we recover the usual map
r/': TM
~ T*M associated with the bilinear form (, >x'
Since T*M carries a canonical symplectic form
(J),
we can use lFL to obtain a closed
two-form 0\. on TM:
0\. = (lFL)* (J) . A local coordinate computation yields the following local formula for COr.: if M is modeled on a linear space E, so locally TM looks like U x E where U E
C
E is open, then O\.(u, e) for (u, e)
U x E is the skew symmetric bilinear form on E x E given by (J)L(U,
e)· ((el' ez)' (fl' fz»
= D 1(DzL(u, e)· e 1)· fl - D 1(D zL(u, e)· f1)· e 1 + Dz(DzL(u, e) . e 1) • fz - DzCDzL(u, e) . f 1) • ez '
(5)
where Dl and Dz denote the indicated partial derivatives of L. In finite dimensions this reads
where (qi, qi) are the standard local coordinates on TQ. It is easy to see that
0\. is (weakly) nondegenerate if DzDzL(u, e) is (weakly) nondegenerate;
Section 8.1 Hamiltonian Mechanics
in this case L is called (weakly) nondegenerate.
In the case of motion in a potential,
nondegeneracy of ffiL amounts to nondegeneracy of the metric (, defined by A: TM
-7
575
>x.
The action of L is
lR, A(v) = JFL(v) . v, and the energy of L is E = A - L. In charts, E(u, e) = D2L(u, e) . e - L(u, e),
and in finite dimensions, E is given by the expression .) dL·i . E(q, q = - . q - L(q, q).
dei' Given L, we say that a vector field Z on TM is a Lagrangian vector field or a Lagrangian system for L if the Lagrangian condition holds: ffiL(V)(Z(V), w)
=dE(v) . w
(6)
for all v E TqM, and WE Tv(TM). Here dE denotes the differential of E. We shall see that for motion in a potential, this leads to the same equations of motion as we found before. If
~
were a weak. symplectic form there would exist at most one such Z, which would be
the Hamiltonian vector field for the Hamiltonian E. The dynamics is obtained by find the integral curves of Z; that is, the curves t 1-+ vet) ETM satisfying (dv/dt)(t) = Z(v(t». From the Lagranian condition it is easy to check that energy is conserved (even though L may be degenerate).
8.1.14 Proposition Let Z be a Lagrangian vector field for L and let vet) E TM be an integral curve of Z. Then E(v(t» is constant in t. Proof By the chain rule, d dt E(v(t)) = dE(v(t) . v'(t) = dE(v(t» . Z(v(t) = ffiL(V(t»(Z(v(t», Z(v(t» = 0
by skew symmetry of ffiL' • We now generalize our previous local expression for the spray of a metric, and the equations of motion in the presence of a potential. In the general case the equations are called Lagrange's equations. 8.1.15 Proposition. Let Z be a Lagrangian system for L and suppose Z is a second-order equation (that is, in a chart U x E for TM, Z(u, e) = (u, e, e, Z2(u, e» for some map Z2: U x E -7 E). Then in the chart U x E, an integral curve (u(t), vet)) E U x E of Z satisfies Lagrange's equations: that is,
Chapter 8 Applications
576
(7)
for all
wEE.
If D2 D2L, or equivalently~, is weakly nondegenerate, then Z is automatically
second order. In case of motion in a potential, (7) reduces to the equations (4).
Proof From the definition of the energy E we have locally
(a term D2 L(u, e) . f2 has cancelled). Locally we may write Z(u, e) Using formula (5) for
(OL'
=(u, e, Y1(u, e), Y2(u, e».
the condition on Z may be written
D 1(D 2L(u, e)· Y1(u, e»· fl - D 1(D 2L(u, e)· f l )· Y)(u, e)
+ D 2(D 2L(u, e) . Y)(u, e»· f2 - D 2(D 2L(u, e) . f)· Y2(u, e) = D)(D2L(u, e) . e) . f) - D)L(u, e) . f) + DzeD2L(u, e) . e) . f 2. Thus if
~
o we get
(9)
is a weak symplectic form, then D 2 D 2 L(u, e) is weakly nondenerage, so setting f)
=
Y)(u, e) = e, i.e., Z is a second-order equation. In any case, if we assume that Z is
second order, then condition (9) becomes
for all f)
E
E.
If (u(t),
vet»~
differentiation, then ~ = v and
is an integral curve of Z
ii = Y 2(u, ~),
and using dots to denote time
so
by the chain rule. _ The condition of being second order is intrinsic; Z is second order iff T'tM where 'tM : TM
~
0
Z = identity,
M is the projection. See Exercise 8.1D.
In finite dimensions Lagrange's equations (7) take the form
1t (:~ )=
:>
i
=
I, ... , n .
Proposition Assume q> : Q ~ Q is a diffeomorphism which leaves a weakly nondegenerate Lagrangian L invariant, i.e., L 0 Tq> = L. Then c(t) is an integral curve of the
8.1.16
Section 8.1 Hamiltonian Mechanics
if and only if
Lagrangian vector field Z
T
Proof Invariance of Lunder
T
0
577
0
0
T
0
lFL so that
lFL)*co =(lFL)* (T*
by Proposition 8.1.6. We also have for any v E TQ, A(T O}
so that TQ = Q
]R2. Define the Poincare metric g on Q by
X
g(x, y)«u l , u2), (vI, v2» = (ulv l + u 2v2)/y2 and consider the Lagrangian
L(x, y, vI, v 2) = [(vl)2 + (v2)2]/y2 defined by g. L is
non degenerate and thus by Proposition 8.1.15, the Lagrangian vector field Z defined by L is a second order equation. By local existence and uniqueness of integral curves, for every point (xo, Yo) E Q and every vector (vlo, v2o) E TCx.y) Q, there is a unique geodesic yet) satisfying yeO) = (xo, Yo), y'(O) = (vlo, v2o). We shall detcrmine the geodesics of g by taking advantage of invariance properties of L. Note that the reflection r: (x, y) E Q 1-+ (-x, y) E Q leaves L invariant. Furthermore, consider the homographies h(z) = (az + b)/(cz + d) for a, b, c, d E]R
satisfying ad - bc = 1,
where z = x + iy. Since Im[h(z)] = Im[z/(cz + d)2], it follows that h(Q) = Q and since Tzh(v) = v/(cz + d)2, where v = vI + iv 2, it follows that h leaves L invariant. Therefore, by Proposition
8.1.16, y is a geodesic if and only if roy and hoy are. In particular, if yeO) = (0, Yo), Y '(0) = (0, uo), then (r
0
y)(O) = (0, Yo) and (r 0 y) '(0) = (0, uo), i.e., y = roy and thus y is
the semiaxis y > 0, x = 0. Since for any qo E Q and tangent vector Vo to Q at qo there exists a homography h such that h(iyo) = qo, and the tangent of h at iyo in the direction iuo is Va, it fOllows that the geodesics of g are images by h of thc semiaxis (0, y) I y > OJ. If c =
°
or d
= 0, this image equals the ray (bid, y) I y > OJ or the ray (alc, y) I y > OJ. If both c '" 0, d '" 0, then the image equals the arc of the circle centered at «ad + bc)!2cd, 0) of radius 1/(2cd). Thus
the geodesics of the Poincare upper halfplane are either rays parallel to the y-axis or arcs of circles centered on the x-axis.
(See Fig. 8.1.1). The Poincare upper half-plane is a model of the
Lobatchevski geometry.
Two gcodesics in Q are called parallel if they do not intersect in Q.
Given either a ray parallel to Oy or a semicircle centered on Ox and a point not on this geodesic, there are infinitely many semicircles passing through this point and not intersecting the geodesics, i.e., through a point not on a geodesic there are infinitely many geodesics parallel to it. •
578
Chapter 8 Applications
y
--4---~--~----4-4-----4-~----~--~--~------~------~_x
Figure 8.1.1 Geodesics in the Poincare Upper Half Plane We close with a completeness result for Lagrangian systems generalizing Example 4.1.23B. 8.1.18 Definition. A C2 function Vo: [0, 00] -+ IR is called positively complete if it is
decreasing andfor any
e > sup{Vo(t)}, it satisfies l~O
l~[e X
-
Vo(t)r l12 dt dt
=
+ 00, where x ~ O.
The last condition is independent of e. Examples of positively complete functions are _t a , -t[log(1 + t)]a, - t log(l + t) [log(log(l + t) + 1)]a, etc.
Let Q be a complete weak Riemannian function and I et Z be the Lagrangian vector field for L(v) = : TQ -+ Q is the tangent bundle projection. Suppose there is a
8.1.19 Theorem (Weinstein and Marsden [1970])
manifold, V: Q -+ IR be a II v 112 (2 - V("t(v», where
C2 'I:
positively complete function Vo and a point q' Q. Then Z is complete.
Q such that V(q)
E
~
Vo(d(q, q'» for all q
E
Proof Let c(t) be an integral curve of Z and let q(t) = ('I: 0 c)(t) be its projection in Q. Let qo = q(O) and consider the differential equation on IR
f"(t) =
with initial conditions f(O) = d(q', qo), ['(0)
dV - -;
(10)
(f(t»
=,J2(~ -
Vo(f(O» , where
.
~ = E(c{t)) =
.
E(c(O»
~ V(qo).
We can assume ~ > V(CJo), for if ~ = V(qo)' then q(O) = 0; now if q(t) = 0, the conclusion is trivially satisfied, so we need to work under the assumption that there exists a La for which q(La) '" 0; by time translation we can assume La =
o.
Section 8.1 Hamiltonian Mechanics
We show that the solution f(t) of (10) is defined for all t
~
579 O. Multiplying both sides of
(10) by f'(t) and integrating yields 2 . 1 2['(t) = p - Vo(f(t)). I.e .• t(s)
=
JS dCq'. qo)
[2(P - Vo(u))]
By hypothesis. the integral on the right diverges and hence t(s) that f(t) exists for all t ~
~
+
00
-1/2
as s
du .
~
+
00.
This shows
o.
For t ~ O. conservation of energy and the estimate on the potential V imply
d(q(t). q') :::; d(q(t). qo) + d(qo. q') :::; d(q'. qo) + = d(q'. qo)
r'
f:
ilil(s)[[ ds
r'
+ Jo [2(P - Vo(q(s))] 1/2 ds:::; d(q'. qo) + Jo [2(P - Vo(d(q(s). q'))] 1/2 ds.
Since
r1
f(t) = d(q'. qo) + Jo [2(P - Vo(f(s))]
112
ds
it follows that d(q(t). q ') :::; f(t); see Exercise 4.II(v) or the reasoning in Example 4.1.238 plus an approximation of d(q(t). q ') by Cl
functions. Hence if Q is finite dimensional. q(t)
remains in a compact set for finite t-intervals. t
~
O. Therefore c(t) does as well. V(q(t)) being
bounded below on such a finite t-interval. Proposition 4.1.9 implies that c(t) exists for all t
~
O.
The proof in infinite dimensions is done in Supplement 8.1e. If F, is the local flow of Z. from 'r(F-tv)) = 'r(F,(-v)) (reversibility). it follows that c(t) exists also for all t:::; 0 and so the theorem is proved. _
(kiF Supplement s.l.e Completeness of Lagrangian Vector Fields on Hilbert Manifolds
This supplement provides the proof of theorem 8.1.19 for infinite dimensional Riemannian manifolds. We start with a few facts of general interest. Let (Q. g) be a Riemannian manifold and 'r; TQ ~ Q the tangent bundle projection. For vETqQ. the subspace Vv = ker Tv 'r = T veTqQ)
C
T v(TQ) is called the vertical subspace of
Tv(TQ). The local expression of the covariant derivative V' defined by g in Supplement 8.18 shows that V'yX depends only on the point values of Y and thus it defines a linear map (V'X)(q); v E TqQ ....... (V'yX)(q) E TqQ where Y E X(Q) is any vector field satisfying Y(q) = v. Let jv; TqQ ~ Tv(TqQ) = Vv denote the isomorphism identifying the tangent space to a linear space with the linear space itself and consider the map hv; TqQ
~
Tv(TQ) by
L
0
(V'X)(q) : TqQ ~ Vv. Define the horizontal map
Chapter 8 Applications
580
where vET qQ and X E X(Q) satisfies X(q) = v. Loeally. if E is the model of Q. hy has the expression hy: (x. u) E V x E ..... (x. v. u - y/u. v» E V x E x Ex E. This shows that hy is a linear eontinuous injective map with split image. The image of hy is called the horizontal subspace of T/TQ) and is denoted by l\.. It is straightforward to check that Ty(TQ)=VyEllHy and that Ty'tIl\.:Hy~TqQ. jy:TqQ~Vy are Banach space isomorphisms. Declaring them to be isometries and Hy perpendicular to Vy gives a metric gT on TQ. We have proved that if (Q. q) is a (weak) Riemannian manifold, then g induces a metric gT on TQ. The following result is taken from Ebin [1970]. 8.1.20 Proposition If (Q. g) is a complete (weak) Riemannian manifold then so is (TQ. gT). Proof. Let (v n) be a Cauchy sequence in TQ and let qn = 't(v n)' Since 't is distance decreasing it follows that (qn) is a Cauchy sequence in Q and therefore convergent to q E Q by completeness of Q. If E is the model of Q. E is a Hilbert space. again by completeness of Q. Let (V. cp) be a chart at q and assume that V is a closed ball in the metrie dermed by g of radius 3£. Also. assume that TyT cp : T/TM) ~ Ex E is an isometry for all v E TqQ which implies that for £ small enough there is a C > 0 such that IITTcp(w)II :;; C IIwl! for all WE TTU. This means that all curves in TV are stretched by Tcp by a factor of C. Let V
C
V be the closed
ball ofradius £ centered at q and let n. m be large enough so the distance between vn and vrn is smaller than £ and vn' vrn E TV. If Y is a path from vn to vrn oflength < 2£. then 't 0 Y is a path from qn to qrn of length < 2£ and therefore 't 0 Y C V. which in turn implies that y C TV. Moreover. Tcp 0 Y has length < 2C£ and therefore the distance between Tcp(v n) and Tcp(v rn) in E is at most 2C£. This shows that (Tcp(v n») is a Cauchy sequence in E and hence convergent. Since Tcp is a diffeomorphism. {vn} is convergent. _ In general. completeness of a vector field on M implies completeness of the first variation equation on TM.
Proof of 8.1.19. Let c:]a.b[
~
TQ be a maximal integral curve of Z. We shall prove that
lil1\ibc(t) exists in TQ which implies. by local existence and uniqueness. that c can be continued beyond b. i.e .• that b = + 00. One argues similarly for a. We have shown that q(t) = ('t 0 c)(t) is bounded on finite t-intervals. Since V(q(t» is bounded on such a finite t-interval. it follows that
Section 8.1 Hamiltonian Mechanics
581
Z. Since V(c(t» depends only on q(t) and since q(t) extends continuously to q(b), it follows that IIV(c(t»1I is bounded as t i b. Since IIS(c(t»1I = IIc(t)1I and II ~(t)1I2 = IIS(c(t»1I 2 + IIV(c(t»1I 2 by the definition of the metric gT, it follows that II ~(t)1I Therefore
(c(tn ))
remains bounded on finite t-intervals.
is Cauchy and Proposition 8.1.20 implies that c(t) can be continuously
extended to c(b). • Remark Note that completeness of Q, an estimate on the potential V, and conservation of the ~
enrgy E, replaces "compactness" in Proposition 4.1.19 with "boundedness."
Exercises 8.1A Let (M,O) be a symplectic manifold with
co =de and f: M --t M a local diffeomorphism.
Prove that f is a symplectic iff for every compact oriented two-manifold B with boundary, Be M, wehave
JaB e-I.naB) e
8.1B (1. Moser) Use the method of proof of Darboux's theorem to prove that if M is a compact manifold, 11 and v are two volume forms with the same orientation, and
JIl= Jv, then there is a diffeomorphism f: M --t M such that r*v form method. Since JIl = Jv, 11- v
=da.
=11.
(Hint: Use the Lie trans-
(see Supplement 7.5B); put v t
and define X t by letting the interior product of
~
with v t be a.. Let
=tv
X t and set f =
(i)
Write down Hamilton's equations.
(ii)
Show that fl(q, p) = PI + P2 + P3' f2 = H, and f3(q, p) =
~(pi + pi + p~) + PI(exp(ql -
+ (1 -t)11
be the flow of
q2) + exp(q3 _ ql»
+ P2(exp(ql _ q2) + exp(q2 _ q3» + piexp(ql _ q2) + exp(q2 _ q3»
582
Chapter 8 Applications
are in involution and are independent everywhere. (iii)
Prove the same thing for gl
=
fl'
gz
= exp(ql - q2) + exp(q2 _ q3)+ exp(q3 - q\ and
g3(q, p) = PIP2P3 - PI exp (q2 - q3) - P2 exp(q3 _ ql) - P3 exp(ql _ q2). (iv)
Can you establish (iii) without explicitly computing the Poisson brackets? (Hint: Express gl' g2' g3 as polynomials of fl' f2' f3·)
8.1 D A second-order equaton on a manifold M is a vector field X on TM such that T'tM
0
X ==
IdTM . Show that (i) X is a second-order equation iff for all integral curves c of X in TM wc have ('tM 0 c)' =c. One calls 't M 0 C a base integral curve. (ii) X is a second· order equation iff in every chart the local representative of X has the form (u, e)
1-+
(u, e, e, V(u, e».
(iii) If M is finite dimensional and X is a second-order equation, then the base integral curves satisfy
where (x, x) denotes standard coordinates on TM. 8.1E Noether theorem Prove the following result for Lagrangian systems.
Let Z be a Lagrangian vector field for L: TM -7 lIt and suppose Z is a second· order equation. Let t be a one-parameter group of diffeomorphisms of M generated by the vector field Y: M -7 TM. Suppose that for each real number t, LoT t = L. Then the function P(Y): TM -7 lIt, defined by P(Y)(v) = lFL(v) . Y is constant along integral curves of Z. 8.1F Use Exercise 8.1E to show conservation of linear (resp., angular) momentum for the motion of a particle in lIt3 moving in a potential that has a translation (resp., rotational) symmctry. 8.1G Consider lIt2n+2 with coordinates (ql, ... , qn, E, PI"'" Pn' t) and define the symplectic form 0) = dqi 1\ dPi + dE 1\ dt. Consider a function P(q, p, E, t) = H(q, p, t) - E. Show that the vector field X = qia/aqi, + . il;?/api +, Ea/at + 'ta/aE defined by ixO) = dP reproduces familiar equations for q, p, t and E. 8.1H Show that the wave equation (see supplement 8.IA) may be derived as a Lagrangian system.
Section 8.1 Hamillonum Mechanics
583
8.11 Refer to Example 8.1.13 on the Schrooinger equation. Let A and B be self adjoint operators on J{ and let fA: J{ ~ lR be given by fA('V) = ('V. A'll) (the expectation value of A in the state'll). Show that Poisson brackets and commutators are related by
8.U Show that the geodesic flow of a compact Riemannian manifold is complete. [Warning: Compact pseudo-Riemannian manifolds need not be complete; see Wolf [1982] and Marsden [ 1973]]. 8.1K Show that any isometry of a weak pseudo-Riemannian manifold maps geodesics to geodesics. (A map
(i) If F, is the flow of the spray of g show that t(F,(sv» = t (Fs/v». where t: TQ Q is the projection. (ii) Let U v
E
be any bounded set in TqQ.
Show that there is an E > 0 such that for any
U. the integral curve of the spray with initial condition v exists for a time
~
E.
(Hint: Let V cUbe an open neighborhood of v such that all integral curves starting in V exist for time ~ Ii. Find R > 0 such that R-1U
C
V and use (i).)
8.1M A weak pseudo-Riemannian manifold (Q. g) is called homogeneous if for any x. y
E
Q
there is an isometry
~
E.)
§8.2 Fluid Mechanics We present a few of the basic ideas concerning the motion of an ideal fluid from the point of view of manifolds and differential forms. This is usually done in the context of Euclidean space using vector calculus. For the latter approach and additional details. the reader should consult one of the standard texts on the subject such as Batchelor (1967), Chorin and Marsden [1979], or Gurtin [19811. The use of manifolds and differential forms can give additional geometric insight. The present section is for expository reasons somewhat superficial and is intended only to indicate how to use differential forms and Lie derivatives in fluid mechanics. Once the basics are understood. more sophisticated questions can be asked. such as: in what sense is fluid mechanics an infinite - dimensional Hamiltonian system? For the answer. see Arnold [19781. Abraham and Marsden [19781. Marsden and Weinstein [19831. and Marsden. Ratiu. and Weinstein [19841. For analogous topics in elasticity. see Marsden and Hughes [1983], and for plasmas. see §8.4 and Marsden and Weinstein [19821. Let M be a compact. oriented finite-dimensional Riemannian n-manifold. possibly with boundary. Let the Riemannian volume form be denoted 11
E
nn(M). and the corresponding
volume element dll. Usually M is a bounded region with smooth boundary in two - or three dimensional Euclidean space. oriented by the standard basis. and with the standard Euclidean volume form and inner product. Imagine M to be filled with fluid and the fluid to be in motion. Our object is to describe this motion. Let x E M be a point in M and consider the particle of fluid moving through x at time t = O. For example. we can imagine a particle of dust suspended in the fluid; this particle traverses a
trajectory which we denote
=
Let u(x. t) denote the velocity of the particle of fluid
moving through x at time t. Thus. for each fixed time. u is a vector field on M. See Figure 8.2.1. We call u the velocity field of the fluid. Thus the relationship between u and
that is. u is a time-dependent vector field with evolution operator
Our derivation of the equations is based on three basic principles. which we ~hall treat in turn:
Section 8.2 Fluid Meclulnics
585
1 Mass is neither created nor destroyed. 2 (Newton's second law) The rate of change of momentum of a portion of the fluid equals the force applied to it .
3 Energy is neither created nor destroyed.
trajectory of fluid particle moving through x at time I = o.
x
Mi
t)
Figure 8.2.1
1. Conservation of mass. This principle says that the total mass of the fluid, which at time t = W, remains unchanged after time t; i.e.,
ooccupied a nice region
(We call a region W "nice" when it is an open subset of M with smooth enough boundary to allow us to use Stokes' theorem.) Let us recall how to use the transport theorem 7.1.12 to derive the continuity equation. Using the change-of-variables formula, conservation of mass may be rewritten as
for any nice region W in M, which is equivalent to
where J(
by the Lie derivative formula and 6.5.17. Thus
~+ div (p,u) = 0
586
Chapter 8 Applications
is the differential form of the law of conservation of mass, also known as the continuity equation. Because of shock waves that could be present, p and u may not be smooth enough to justify the steps leading to the differential form of this law; the integral form will then be the one to usc. Also note that the Riemannian metric has as yet played no role; only the volume clement of M was needed. 2. Balance of momentum. Newton's second law asserts that the rate of change of momentum of a portion of the fluid equals the total force applied to it. To see how to apply this principle on a general manifold, let us discuss the situation M C JR.3 first. Here we follow the standard vector calculus conventions and write the velocity fields in boldface type. The momentum of a portion of the fluid at time t that at time t =0 occupied the region W is
f
pudll·
Here and in what follows the integral is JR.3- valued, so we apply all theorems on integration componentwise. For any continuum, forces acting on a piece of material are of two types. First there are forces of stress, whereby the piece of material is acted on by forces across its surface by the rest of the continuum. Second, there are external, or body forces, such as gravity or a magnetic field, which exert a force per unit volume on the continuum. The clear formulation of surface stress forces in a continuum is usually attributed to Cauchy. We shall assume that the body forces are given by a given force density b, i.e., the total body forces acting on Ware fwPb dll. In continuum mechanics the forces of stress are assumed to be of the form fawa(x, t) . n da, where da is the induced volume clement on the boundary, n is the outward unit normal, and a(x, t) is a time-dependent contravariant symmetric two-tensor, called the Cauchy stress tensor. The contraction a(t, x) . n is understood in the following way: if a has components a ij and n has components nk, then a· n is a vector with components (a· n)i = gjk aijnk, where g is the metric (in our case gjk =0jk). The vector a· n, called the Cauchy traction vector, measures the force of contact (per unit area orthogonal to n) between two parts of the continuum. (A theorem of Cauchy states that if one postulates the existence of a continuous Cauchy traction vector field T(x, t, n) satisfying balance of momentum, then it must be of the form a· n, for a two-tensor, a; moreover if balance of moment of momentum holds, a must be symmetric. See Chorin and Marsden [1979], Gurtin [1981], or Marsden and Hughes [1983] for details.) Balance of momentum is said to hold when
for any nice region W in M = JR.3. If diva denotes the vector with components (div(a li ), div(a 2i ), div(a 3i then by Gauss' theorem
»,
Section 8.2 Fluid Mechanics
J
(J . n da =
iJG\(W)
J
587
(div (J) dll·
G\(W)
By the change-of-variables formula and Lie derivative formula, we get
so that the balance of momentum is equivalent to ap i Ttu +
i
au ( ) iii . i.)i PTt + dp' u u + pLuu + pu dlv U = pb + (dlV (J .
But dp . u + p div u = div(pu) and by conservation of mass, ap/at + div(pu) = O. Also, Luu i = (aui/axi)ui = (u .V)u i, so we get
~ ut
+ (u • V)u = b + 2-div (J,
P
which represents the basic equations of motion. Here the quantity au/at + (u ·V) u is usually called the material derivative and is denoted by Du/dt. These equations are for any continuum, be it elastic or fluid. An ideal fluid is by definition a fluid whose Cauchy stress tensor (J is given in terms of a function p(x, t) called the pressure, by (Jii = _pgii. In this case, balance of momentum in differential form becomes the Euler equations for an idea/fluid: au + (u • V)u ut
"T
=
I p
b - -grad p.
The assumption on the stress (J in an ideal fluid means that if S is any fluid surface in M with outward unit normal n, then the force of stress per unit area exerted across a surface element S at x with normal n at time t is -p(x, t)n (see Figure 8.2.2). Let us return to the context of a Riemannian manifold M. First, it is not elear what the veetor-valued integrals should mean. But even if we could make sense out of this, using, say parallel transport, there is a more serious problem with the integral form of balance of momentum as stated. Nanlely, if one changes coordinates, then balance of momentum does not look the same. One says that the integral form of balance of momentum is not covariant. Therefore we shall concentrate on the differential form and from now on we shall deal only with ideal fluids. (For a detailed discussion of how to formulate the basic integral balance laws of continuum mechanics eovariantly, sec Marsden and Hughes 11983]. A genuine difficulty with shock wave theory is that the notion of weak solution is not a coordinate independent eoneept.)
Chapter 8 Applications
588
Figure 8.2.2 Rewrite Euler's equations in
]R3
with indices down; i.e .• take the flat of these equations.
Then the i-th equation. i = 1.2.3 is
We seek an invariant meaning for the sum of the last three terms on the left-hand side. For fixed this expression is aUi aUi aUj aUj u·-=u·-+u·--uJ ax j J ax j J axi axi That is. Euler's equations can be written in the invariant form 1 aub b bIb ---.,-- +L u - - d(u (u))=- -dp+ b at u 2 P
We postulate this equation as the balance of momentum in M for an ideal fluid. The reader familiar with Riemannian connections (see Supplement 8.1B) can prove that this form is equivalent to the form
au
at + V' uu = by showing that
I p-grad P + b
Section 8.2 Fluid Mechanics
589
where Vuu is the covariant derivative of u along itself, with V the Riemannian connection given by g. The boundary conditions that should be imposed come from the physical significance of ideal fluid: namely, no friction should exist between the fluid and aM; i.e., u is tangent to aM at points of aM. Summarizing, the equations of motion of an ideal fluid on a compact Riemannian manifold M with smooth boundary aM and outward unit normal n are au b
bIb
-s:- + L u - -d(u (u» Ol
u
2
= -
1 b ap -dp + b and - + div(pu) =
at
P
o.
We also have the boundary conditions u lIaM, i.e.,
U·
n = 0 on aM;
and initial conditions u(x, 0) = uo(x) given on M. We shall assume b =0 from now on for simplicity. 3. Conservation of energy A basic problem of ideal fluid dynamics is to solve the initial-boundary-value problem. The unknowns are u, p, and p, i.e., n + 2 scalar unknowns. We have, however, only n + 1 equations. Thus one might suspect that to specify the fluid motion, one more equation is needed. This is in fact true and the law of conservation of energy will supply the necessary extra equation in fluid mechanics. (The situation is similar for general continua; see Marsden and Hughes [1983].) For a fluid moving in M with velocity field u, the kinetic energy of the fluid is
Ekinetic =
where II u 112
=(u, u)
11
'2
.vi
p II u II 2 dJ..l
is the square length of the vector function u. We assume that the total
energy of the fluid can be written E total
= ~inetic + E intemal ,
where Eintemal is the energy that relates to energy we cannot "see" on a macroscopic scale and derives from sources such as intermolecular potentials and molecular vibrations. If energy is pumped into the fluid or if we allow the fluid to do work,
Etotal
will change. We describe two
particular examples of energy equations that are useful. A Assume that Eintemal = constant. Then we ought to have ~inetic as a constant of the motion; i.e.,
Chapter 8 Applications
590
To deal with this equation it is convenient to use the following.
8.2.1 Transport Theorem with Mass Density Let f be a time-deperuient smoothfunction on M. Then if W is any (nice) open set in M.
J
J
Of -d pf dll = P-dll. dt'l''(W) 'l'l(W) dt where Df/dt = df/dt + LJ. Proof By the change-of-variables fonnula. the Lie derivative fonnula. div(pu) = u[p) + p div(u), and conservation of mass. we have
Making use of Lu(1I u 112)
= LU
the transport lemma. and Euler's
equations. we get
d (I r I r ( dll u 112 2) r dub I o = di 2' J,Mpll u II2) dll = - J, P -dt- + Lull u II dIJ. = J, Pat' u dIJ. + '2 2M M =
J./ ~; .
u dll + Lp( Luub) . u dll-
=-
f
\1
=
i
\1
dp' u dll =
{(div u)w- LJPIl)} M =
f
\1
b
(Luu)' u dll
~ iPd(ub(u)) . u dIJ.
f
{(div u)PIl- d(iuPll)}
f
(by the Leibniz rule for Lu)
(div u)PIl·
\1
The last equality is obtained by Stokes' theorem and the boundary conditions 0 = (u . n)da = iull'
If we imagine this to hold for the same fluid in all conceivable motions. we are forced to postulate one of the additional equations div u = 0 or p = O. The case div u = 0 is that of an incompressible fluid. Thus in this case the Euler equations are
Section 8.2 Fluid Mechanics
dub
at + Luu
b
a;
591
1 2 1 - 2'd II u II = - pdp
+ div(pu) = 0 div u = 0
with the boundary condition iull = 0 on aM and initial condition u(x, 0)
= Uo(x).
The case p = (
is also possible but is less interesting. For a homogeneous incompressible fluid, with constant density p, Euler's equations can be reformulated in terms of the Hodge decomposition theorem (see Section 7.5). Nonhomogeneous incompressible flow requires a weighted Hodge decomposition (see Marsden [1976]). Recall that anyone-form a can be written in a unique way as a =
d~
+ y, where '6y= O. Define the linear
operator IP': nl(M) ~ lYE n1(M)1 '6y= OJ by lP'(a) = y. We are now in a position to reformulate Euler's equations. Let n
10 =0
be the set of C~
one-forms y with '6y=O and y tangent to aM; i.e., *y1iJM=0. Let T:nlo=o~nlo=o be defined by
Thus Euler's equations can be written as dub/at + T(ub) = 0, which is in the "standard form" for an evolution equation. Note that T is nonlinear. Another important feature of T is that it is nonlocal; this is because lP'(a)(x) depends on the values of a on all of M and not merely those in
the neighborhood of x E M. B We postulate an internal energy over the region W to be of the form
Emtemal =
fw pw dll ,
where the function w is the internal energy density per unit mass. We assume that energy is balanced in the sense that the rate of change of energy in a region equals the work done on it:
By the transport theorem and arguing as in our previous results, this reduces to
Since W is arbitrary, pdivu+p
~
=
O.
592
Chapter 8 Applications
Now assume that w depends on the fluid motion through the density; i.e., the internal energy depends only on how much the fluid is compressed. Such a fluid is called ideal isentropic or
barotropic. The preceding identity then becomes . O= P dlv
U
+ P(dW at + dw· u )
=
dW at dP + P ap dW d p. u = P d·IV U + P ap dW (- P d·IV u) P d·IV U + P ap
using the equation of continuity. Since this is an identity and we are not restricting div u, we get
If P is a given function of p note that w
=- f p del/Pl.
Note that dp/p
= dew + p/p).
This
follows from p = p 2 w' by a straightforward calculation in which p and ware regarded as functions of p. The quantity w + p/p = w + p w' is called the enthalpy and is often denoted h. Thus Euler's equations for compressible ideal isentropic flow are aub
at + Luu
bIb
-
a;
"2 d(u
(u» +
dp
p
=
0
+ div(pu) = 0
u(x, 0) = uo(x) on M and u· n = 0 on dM. where p
= p 2 w '(p)
is a function of p, called an equation of state, which depends on the
particular fluid. It is known that these equations lead to a well-posed initial value problem (i.e., there is a local existence and uniqueness theorem) only if p'(p) > O. This agrees with the common experience that increasing the surrounding pressure on a vlume of fluid causes a decrease in occupied volume and hence an increase in density. Many gases can often be viewed as satisfying our hypotheses, with p = ApY where A and y are constants and y
~
1.
Cases A and B above are rather opposite. For instance, if p = Po is a constant for an incompressible fluid, then clearly p cannot be an invertible function of p. However, the case p = constant may be regarded as a limiting case p'(p)
~
00.
In case B, p is an explicit function of p.
In case A, p is implicitly determined by the condition div u = O. Finally, notice that in neither case A or B is the possibility of a loss of total energy due to friction taken into account. This leads to the subject of viscous fluids, not dealt with here. Given a fluid flow with velocity field u(x, t), a streamline at a fixed time t is an integral curve of u; i.e., if xes) is a streamline parametrized by s at the instant t, then xes) satisfies dx ds
=
u(x(s), t), t fixed.
593
Section 8.2 Fluid Meclulnics
On the other hand, a trajectory is the curve traced out by a particle as time progresses, as explained at the beginning of this section; i.e., is a solution of the differential equation dx
dt = u(x(t), t) with given initial conditions. If u is independent of t (i.e., au/at = 0), then, streamlines and trajectories coincide. In this case, the flow is called stationary or steady. This condition means that the "shape" of the fluid flow is not changing. Even if each particle is moving under the flow, the global configuration of the fluid does not change. The following criteria for steady solutions for homogeneous incompressible flow is a direct consequence of EuIer's equations, written in the form aub/at + lP'(Luub) = 0, where lP' is the Hodge projection to the co-closed I-forms.
8.2.2 Proposition Let u t bea solution to the Euler equations for homogeneous incompressible flow on a compact manifold M and
where per, e) =
s:
f(s)s ds .
Clearly such a U need not be harmonic. We saw that for compressible ideal isentropic flow, the total energy
JM (II U112/2 + pw) dll
is conserved. We can refine this a little for stationary flows as follows.
8.2.3 Bernoulli's Theorem For stationary compressible ideal isentropic flow, with p a function of p,
I
2
Zllull +
f p-=Zllull dp
I
2
p
+w+p
J
dp/p =w + pip denotes a potential for the dp/p. The same holds for stationary homogeneous (p = constant in space = Po) incompressible flow with dp/p replaced by p/Po' If body forces deriving from a potential U is constant along streamlines where the entlullpy
one form
are present, i.e.,
J
If> = -
dU, then the conserved quantity is
594
Chapter 8 Applications
Proof Since Lu(ub).u
= d(ub(u))·u,
for stationary ideal compressible or incompressible
homogeneous flows we have dub bIb dp 0 = - . u = - ( L u )·u + -d(u )) . u - - . u = u 2 p
at
1 - -2 (d
2
1
II u II ). u - -p dp . u ,
so that dub J,, as' u(x(s))ds '2
since x'(s)
=u(x(s)).
The two-form
=
0
• 0)
=dub
is called vorticity, which, in JR.3 can be identified with curl u. Our
assumptions so far have precluded any tangential forces and thus any mechanism for starting or stopping rotation. Hence, intuitively, we might expect rotation to be conserved. Since rotation is intimately related to the vorticity, we can expect the vorticity to be involved. We shall now prove that this is so. Let C be a simple closed contour in the fluid at t = 0 and let C, be the contour carried along the flow. In other words, Ct =
c,
_I--~--I
Figure 8.2.3 8.2.4 Kelvin Circulation Theorem Let M be a manifold and ~ C M a smooth closed loop, i.e., a compact one-manifold. Let u t solve the Euler equations on M for ideal isentropic compressible or homogeneous incompressible flow and ~(t) be the image of ~ at time t when each particle moves under the flow
Section 8.2 Fluid Mechanics
595
in time; i.e.,
Proof Let
However, Luub + dub(dt is exact from the equations of motion and the integral of an exact form over a closed loop is zero. _ We now use Stokes' theorem, which will bring in the vorticity. If L is a surface (a two-dimensional submanifold of M) whose boundary is a closed contour C, then Stokes' theorem yields
See Fig. 8.2.4.
n
Figure 8.2.4 Thus, as a corollary of the circulation theorem, we can conclude: 8.2.5 Helmholtz' Theorem Uruier the hypotheses 0/8.2.4, the flux o/vorticity across a surface moving with the fluid is constant in time. We shall now show that
(0
and 11 = co/p are Lie propagated by the flow.
8.2.6 Proposition For isentropic or homogeneous incompressible flow, we have
0)
596
Chapter 8 Applications
called the vorticity-stream equation and (ii) where 'Ill(x) = 'Il(x, t) and J(
Thus ro ( -+dp'u ap ) =1 - (aro -+Lro) - -
p at
= -
U
p2 at
%( ~ + dp·u + p div u) + 'Il div u
= 'Il div u
by conservation of mass. From aro/at + Luro = 0 it follows that (a/at)(
. Proof ;~,
iq>;~,11 = il;"Il' which gives
=
=
. J() Iq>;~,
~o· •
Notice that the vorticity as a two-form is Lie transported by the flow but as a vector field it is vorticity/p, which is Lie transported. Here is another instance where distinguishing between forms and vector fields makes an important diffcrence. The flow
~
M (volume preserving for incompressible flow) as the configuration
space. The references noted at the beginning of this section carry out this program (see also Exercise 8.21 and §8.4).
597
Section 8.2 Fluid Mechanics
Exercises 8.2A In classical texts on fluid mechanics, the identity (u oV)u=(l/2)V(u ou)+(Vxu)xu is often used. To what identity does this correspond in this section? 8.2B A flow is called potentialflow if ul> = d
f dp/p = constant on a streamline.
8.2C Complex variables texts "show" that the gradient of
0)
has a one-dimensional kernel. (i) Using
Frobenius' theorem, show that this distribution is integrable. one-dimensional leaves with integral curves of
~
(ii)
Identify the
(see 8.2.7)--these are called vortex lines.
(iii) Show that vortex lines are propagated by the flow. 8.2F Assume dim M = 3. A vortex tube T is a closed oriented two-manifold in M that is a union
of vortex lines. The strength of the vortex tube is the flux of vorticity across a
surface E inside T whose boundary lies on T and is transverse to the vortex lines. Show that vortex tubes are propagated by the flow and have a strength that is constant in time. 8.2G Let f: ]R3 ~]R be a linear function and g: S2 ~]R be its restriction to the unit sphere. Show that dg gives a stationary solution of Euler's equations for flow on the two-sphere. 8.2H Stream Functions (i) For incompressible flow in ]R2, show that there is a function 'V such that u l = d'V/dy and u2 = - d\jf/dx. One calls = 0*'1'. (iii) Show that u is a Hamiltonian vector field (see Section 8.1) with energy 'V directly in ]R2 and then for arbitrary two-dimensional Riemannian manifolds M. (iv) Do stream functions exist for arbitrary fluid flow on li2? On S2?
598 (v)
Chapter 8 Applications
Show that the vorticity is
(J)
= M",.
8.21 Clebsch Variables; Clebsch [1859]. Let ~ be the space of functions on a compact manifold M with the dual space ~ *, taken to be densities on M; the pairing between f E ~ and pE ~*is (f,p)=JMfp. (i) On the symplectic manifold ~x ~* x ~x ~* with variables (a, A, 1.1, p), show that Hamilton's equations for a given Hamiltonian H are . I5H a =
sr'
. I5H 1.1 =
8P'
j. _
I5H
- - l5a'
. _ I5H p - - 151.1 '
where 15H/I5A is the junctional derivative of H defined by I5H
.
.
(sr' A) = DH(A)'A . (ii) In the ideal isentropic compressible fluid equations, set M = pub, the momentum
density, where dx denotes the Riemannian volume form on M. Identify the density o(x)dx E ~ * with the function o(x) E ~ and write M = - (pdl.1 + Ada)dx. For momentum densities of this form show that Hamilton's equations written in the variables (a, A, 1.1, p) imply Euler's equation and the equation of continuity.
§8.3 Electromagnetism Classical electromagnetism is governed by Maxwell's field equations. The form of these equations depends on the physical units chosen, and changing these units introduces factors like
41t, c = the speed of light, simplest form; thus c
Eo = the dielectric constant and 110 = the magnetic permeability.
100 , ~o
discussion assumes that
This
are constant; the choice of units is such that the equations take the
= 100 = ~o = 1
and factors 41t disappear. We also do not consider
Maxwell's equations in a material, where one has to distinguish E from D, and B from H. Let E, B, and J be time dependent C1-vector fields on 1R3 and p: JR.3 x JR. ~ JR. a scalar. These are said to satisfy MaxweU's equatWns with charge density p and current density J when the following hold: div E
=
P
curl B -
aB at aE at
(1)
(no magnetic sources)
(2)
=
0
(Faraday's law o/induction)
(3)
=
J
(Ampere's law).
(4)
div B = 0 = 0 curl E +
(Gauss's law)
E is called the electricjield and B the magneticjield.
The quantity
In p dV = Q
is called the charge of the set
n
C JR.3. By Gauss' theorem, (1)
is equivalent to
(5) for any (nice) open set
n C JR.3; i.e., the electric flux out of a closed surface equals the total charge
inside the surface. This generalizes Gauss' law for a point charge discussed in Section 7.3. By the same reasoning, (2) is equivalent to
lan
(6)
B. n dS = O.
That is, the magnetic flUX out of any closed surface is zero. In other words there are no magnetic SOurces inside any closed surface. By Stokes' theorem, (3) is equivalent to
Ias for any closed loop
as
E. ds =
r(curl E) • n dS = - aadrs B· n dS
Js
bounding a surface S. The quanity
JasE' ds is called the voltage around
600
as.
Chapter 8 Applications
Thus, Faraday's law of induction (3) says that the voltage around a loop equals the negative of
the rate of change of the magnetic flux through the loop. Finally, again by the classical Stokes' theorem, (4) is equivalent to
(8)
JsJ· n dS has the physical interpretation of current, this form states that if E is constant in
Since
time, then the magnetic potential difference JasB. ds around a loop equals the current through the loop. In general, if E varies in time, Ampere's law states that the magnetic potential difference around a loop equals the total current in the loop plus the rate of change of electric flUX through the loop. We now show how to express Maxwell's equations in terms of differential forms. Let M = jR4 = {(x, y, z, t)} with the Lorentz metric g on jR4 of diagonal form (1, 1, 1, - 1) in standard coordinates (x, y, z, t).
8.3.1 Proposition There is a unique twojorm F on lR4 , called the Faraday twojorm such that
(9)
Eb = - i a/ilt F ; Bb
(Here the b is Euclidean in jR3 and the
*
=- i a/ilt * F.
(10)
is Lorentzian in jR4.)
Proof If F = F xy dx
1\
dy + F zx dz
+FxI dx
1\
1\
dx + F yz dy 1\ dz
dt+ FYI dy I\dt + Fzt dz
1\
dt,
then (see Example 6.2.14E),
* F = Fxy dz 1\ - Fxl dy
dt + F zx dy 1\
1\
dt + F yz dx 1\ dt
dz - FYI dz 1\ dx - F zt dx
1\
dy
and so
- i a/ilt F = F xl dx + F yt dy + F zt dz and
- i alilt
* F = Fxy dz + F zx dy + Fyz dx.
Thus, F is uniquely determined by (9) and (10), namely F = El dx
1\
dt + E2 dy 1\ dt + E3 dz 1\ dt
+ B3 dx
1\
dy + B2 dz
1\
dx + B 1 dy 1\ dz. •
We started with E and B and used them to construct F, but one can also take F as the primitive object and construct E and B from it using (9) and (10). Both points of view are useful. Similarly, out of p and J we can form the source one-form j = - p dt + J 1 dx + J2 dy +
601
Section 8.3 Electromagnetism
13 dz; i.e., j is uniquely determined by the equations - ia/dtj relation, J is regarded as being defined on
=p
and i a/dt *j
=*.F;
in the last
]R4.
8.3.2 Proposition Maxwell's equations (1)-(4) are equivalent to the equations
of = j
dF = 0 and on the manifold
]R4
endowed with the Lorentz metric.
Proof A straightforward computation shows that
dF = (curl E +
~~)xdy A dz A dt + (curl E + ~~ ) ydz A dx A dt
+ (curIE+
~~)zdXAdYAdt+(diV
B)dxAdYAdz.
Thus dF =0 is equivalent to (2) and (3). Since the index of the Lorentz metric is 1, we have 0 =* d
*.
Thus
of = * d * F = * d( - Eldy A dz - E2dz A dx - E 3dx A dy + Bldx A dt + B2dy A dt + B3dz A dt) =
* [ - (div E)dx A dy A dz + ( curl B - ~ ) xdy A dz A dt + + (curl B(curlB-
Thus
~)ydzAdxAdt+(curIB- ~~)zdXAdYAdt]
~)xdx+(curIB- ~)ydY + (curIB- ~~)zdZ-(diVE)dt
of
= j iff (1) and (4) hold . • As a skew matrix, we can represent F as follows
F=
x
y
0
B3 _B2 EI
_B3
BI
E2
0
E3
0
B2 -BI
z
-EI _E2 _E3
X Y z
0
Recall from Section 6.5 and Exercise 7.SG, the formula
Since Idet [gkt 11 = I, Maxwell's equations can be written in terms of the Faraday two-form F in
602
Chapter 8 Applications
components as Fij,k + Fjk~ + Fkij =
and
pk,k
°
(1)
=- ji,
(12)
where Fij,k = aF/dx k, etc. Since 02 = 0, we obtain
°=
02F = oj
= *[
= * d * j = * d( -
(~ + div J )dx
1\
dy
P dx 1\
1\
dy
1\
dz 1\ dtJ =
dz + (*
P) 1\ dt)
~ + div J;
i.e., ap/ at + div J = 0, which is the continuity equation (see Section 8.2). Its integral form is, by Gauss' theorem -dQ = -d
dt
for any bounded open set
n.
f.
dtQ
P dV =
I
an
J • n dS
Thus the continuity equation says that the flux of the current density
out of a closed surface equals the rate of change of the total charge inside the sUrface.
Next we show that Maxwell's equations are Lorentz invariant, i.e., are special-relativistic. The Lorentz group L is by definition the orthogonal group with respect to the Lorentz metric g,
i.e.,
L = {A E uL(JR.4)
I g(Ax, Ay) = g(x, y) for all
x, y
E
JR.4}.
Lorentz in variance means that F satisfies Maxwell's equations with j iff A *F satisfies them with A *j, for any A
E
L. But due to Proposition 8.3.1 this is clear since pull-back commutes with d
and orthogonal transformations commute with the Hodge operator (sec Exercise 6.2D) and thus they commute with O. As a 4 x 4 matrix, the Lorentz transformation A acts on F by F 1-+ A *F = AFAT. Let us sec that the action of A
E
L mixes up E's and B's. (This is the source of statements like: "A
moving observer sees an electric field partly converted to a magnetic field.") Proposition 8.3.1 defines E and B intrinsically in terms of F. Thus, if one performs a Lorentz transformation A on F, the new resulting electric and magnetic fields E' and B' with respect to the Lorentz unit normal A *(aJdt) to the image A(JR.3 x 0) in JR.4 are given by
(E'f
=-i A-a/at A *F,
(B'f
=- i A-a/at A *F.
For a Lorentz transformation of the form x' -
x-vt
-~'
Y '= y, z
= z,
t ' = ---,t""-""v",,x=-
~
(the special-relativistic analogue of an observer moving uniformly along the x-axis with velocity v) we get
Section 8.3 Electromagnetism
603
and
[
2 3 B3+vE2J .
B'= BI, B -vE,
~~
We leave the verification to the reader. By the way we have set things up, note that Maxwell's equations make sense on any Lorentz
manifold; i.e., a four-dimensional manifold with a pseudo-Riemannian metric of signature (+, +, +, -). Maxwell's vacuum equations (i.e., j = 0) will now be shown to be conformally invariant on any Lorentz manifold (M, g). A diffeomorphism
E
n2(M) where (M, g) is a Lorentz manifold, satisfy dF = 0 and of
= j. Let
Hence Maxwell's vacuum equations (with j = 0) are conformally invariant; i.e.,
if F salisfies
them, so does
I det [gks11-If2 gjr ~(grVhpab I det [gks1 I 1(2)
ax
shows that when one replaces g by f2g, we get
and so
• Let us now discuss the energy equation for the electromagnetic field. Introduce the energy
density of the field 'E= (E E + B • B) 12 and the Poynting energy-flux vector 0
S = E x B.
Poynting's theorem states that
a'E - at =div S+ E oj. This is a straightforward computation using (3) and (4). We shall extend this result to
]R4
the same time, shall rephrase it in the framework of forms. Introduce the stress-energy-momentum tensor (or the Maxwell stress tensor) T by
and, at
604
Chapter 8
ApplicaJions
(13)
(or intrinsically. T = F . F - (1/4) (F. F)g. where F· F denotes a single contraction of F with itself). A straightforward computation shows that the divergence of T equals (19)
where Fpq,i = (ClFpqlClxk)gik. Taking into account /iF =j written in the form (12). it follows that (14) For i = 4. the relation (14) becomes Poynting's theorem.
[Poynting's theorem can also be
understood in terms of a Hamiltonian formulation; see 8.4.2 below. The Poynting energy-flux vector is the Noether conserved quantity for the action of the diffeomorphism group of JR.3 on T* A.. where A. is the space of vector potentials A defined in the following paragraph. and Poynting's theorem is just conservation of momentum (Ncether's theorem). We shall not dwell upon these considerably more advanced aspects and refer the interested reader to Abraham and Marsden (1978).) It is clear that T is a symmetric 2-tensor. As a symmetric matrix,
a Ex B] [ T= (Ex B)T 'E where a is the stress tensor and 'E
is the energy density. The symmetric 3 x 3 matrix a
has the following components all =
~[_ (E? _
0 22 =
~[(EI)2 + (BI)2 _
0 33 =
~[(El)2 + (BI)2 + (E2)2 + (B2)2 _
2
(B 1)2
2
+ (E2)2 + (B2)3 + (E3)2 + (B 3)2] (E2)2 _ (B2)2
+ (E 3)3 + (B3)2)
2
0 12 = EIE2
(E 3 )2 _ (B 3 )2]
+ B IB2
0 13 = EI E 3 + BI B 3 0 23 = E2 E 3 + B2B'.
We close this section with a discussion of Maxwell's equations in terms of vector potentials. We first do this directly in terms of E and B. Since div B
=O.
if B is smooth on all of JR.3,
there exists a vector field A. called the vector potential. such that B = curl A. by the Poincare lemma. This vector field A is not unique and one could use equally well A' = A + grad f for some (possibly time-dependent) function f: JR.3 ~ R This freedom in the choice of A is called
Section 8.3 Electromagnetism
605
gauge freedom. For any such choice of A we have by (3) aB = curl E + at a curl A = curl (Ea+A o= curl E + at at). so that again by the Poincare lemma there exists a (time-dependent) function
E+
aA at = -
grad <po
(8)
Recall that the Laplace-Bc\trami operator on functions is defined by V2f = div(grad f). On vector fields in JR.3 this operator may be defined componentwise. Then it is easy to check that curl (curl A) = grad(div A) - V2A. Using this identity. (8). and B =curl A in (4). we get aE a ( aA ) J = curl B - at = curl(curl A) - at - at - grad
a2A + a (grad
2
= grad(dlv A) - V A + -
C\.
at2
01
and thus (16)
From (1) we obtain as before
p
= div E = diV( -
a: -
grad
that is. V2
(17)
It is apparent that (16) and (17) can be considerably simplified if one could choose. using the gauge frecclom. the vector potential A and the function
divA+
~=
O.
606
Chapter 8 Applications
So, assume one has Ao' o df/dt satisfy div A + dq>/dt = 0. This becomes, in tenns of f,
d(
0= div(Ao + grad f) + dt q>o -
df)
dt
.
= dlv
dq>O 2 An + at +V f-
d2f
dt2 ;
(18)
i.e.,
This equation is the calssical inhomogeneous wave equation. The homogeneous wave equation (right-hand side equals zero) has solutions f(t, x, y, z)
= 'I'(x -
t) for any function 'V. This
solution propagates the graph of 'V like a wave--hence the name wave equation. Now we we can draw some conclusions regarding Maxwell's equations. In tenns of the vector potential A and the function q>, (1) and (4) become
(19)
which again are inhomogeneous wave equations. Conversely, if A and q> satisfy the foregoing equations and div A + dq>/dt = 0, then E = - grad q> - dAldt and B = curl A satisfy Maxwell's equations. Thus in
]R4,
this procedure reduces the study of Maxwell's equations to the wave equation,
and hence solutions of Maxwell's equations can be expected to be wavelike. We now repeat the foregoing constructions on ]R4 using differential fonns. Since dF = 0, on
]R4
we can write F = dG for a one-fonn G. Note that F is unchanged if we replace G by
G + df. This again is the gauge fr'edom Substituting F = dG into of = j gives OdG = j. Since 8
=do + Od
is the Laplace-deRham operator in
]R4,
we get
8G =j - dOG.
°
Suppose we try to choose G so that OG = we can let G = Go + df and demand that
(20)
(a gauge condition). To do this, given an initial Go'
so f must satisfy 8f =- OG o. Thus, if the gauge condition (21)
holds, then Maxwell's equations become
607
Section 8.3 Electromagnetism
L'.G =j.
(22)
Equation (21) is equivalent to (18) and (22) to (19) by choosing G = Ab +
Exercises 8.3A Assume that the Faraday two-form F depends only on t - x.
° of =°
(i) Show that dF =
is then equivalent to B3
=E2, B3 =E2,
B2 =- E3, B 1 =0.
(ii) Show that
is then equivalent to
B2 =- E3, El
=0.
These solutions of Maxwell's equations are called plane electromagnetic waves; they are determined only by E2, E3 or B2, B3, respectively. 8.38 Let u = dldt. Show that the Faraday two-form FE n2(]R4) is given in terms of E and B
E
X (]R4) by F = ub /\ Eb
- *(ub /\ Bb).
9.3C Show that the Poynting vector satisfies
where u = d/dt and E, B E X (lR4). 8.3D Let (M, g) be a Lorentzian four-manifold and u E X (M) a timelike unit vector field on M; i.e., g(u, u) =-1. (i) Show that any a E n2(M) can be written in the form
(ii) Show that if iua = 0, where a E n2(M) ("a is orthogonal to u"), then *a is decomposable, i.e., *a is the wedge product of two one-forms. Prove that a is also locally decomposable. (Hint: Use the Darboux theorem.) 8.3E
The field of a stationary point charge is given by
E=~ 8=0, 47tr 3 '
where r is the vector xi + yj + zk in ]R3 and r is its length. Use this and a Lorentz transformation to show that the electromagnetic field produced by a charge e moving along the x-axis with velocity v is
608
Chapter 8 Applications
and, using spherical coordinates with the x-axis as the polar axis,
Br = 0,
B = 0 B = e
'q>
e(I 2
y2)y 2
sin 9 2
4m (1 - y sin 9)
3/2
(the magnetic field lines are thus circles centered on the polar axis and lying in planes perpendicular to it). 8.3F (Misner, Thorne, and Whecler [1973].) The following is the Faraday two-form for the field of an electric dipole of magnitude PI oscillating up and down parallel to the z-axis.
Verify that dF =0 and
of =0,
except at the origin.
8.3G Let the Lagrangian for electromagnetic theory be
Check that JL/Jg ij is the stress-energy-momentum tensor Tij (see Hawking and Ellis [1973, sec. 3.3]).
~8.4 The Lie-Poisson Bracket in Continuum Mechanics and Plasma Physics
This section studies the equations of motion for some Hamiltonian systems in Poisson bracket formation. As opposed to §8.1. the emphasis is placed here on the Poisson bracket rather than on the underlying symplectic structure. This naturally leads to a generalization of Hamiltonian mechanics to systems whose phase space is a "Poisson manifold". We do not intend to develop here the theory of Poisson manifolds but only to illustrate it with the most important example. the Lie-Poisson bracket. See Marsden and Ratiu [1988] for further details. If (P. ro) is a (weak) symplectic manifold. H: P
~
JR a smooth Hamiltonian with
Hamiltonian vector field XH E X(p) whose flow is denoted by
is equivalent to
(2) for any smooth locally defined function f: U ~ JR. U open in P. In (2). {. } denotes the Poisson bracket defined by ro. i.e .• (3)
Finally. recall that the Poisson bracket is an anti symmetric bilinear operation on .'f(P) which satisfies the Jacobi identity {{F.G} .H} + {{G.H} .F} + {{H.F}.G} =0.
(4)
i.e.. (.'f(P). { • } ) is a Lie algebra. In addition. the multiplicative ring structure and the Lie algebra structure of .1(P) are connected by the Leibniz rule {FG. H} =F{G. H} + G{F. H}.
(5)
i.e.. { . } is a derivation in each argument. These observations naturally lead to the following generalization of the concept of symplectic manifolds. 8.4.1 Definition A smooth manifold P is called a Poisson manifold if .'f(P). the ring of functions on P. admits a Lie algebra structure which is a derivation in each argument. The bracket
Chapter 8
610
Applications
operation on .'J(P) is called a Poisson bracket and is usually derwted by ( • }. From the remarks above. we see that any (weak) symplectic manifold is a Poisson manifold. One of the purposes of this section is to show that there arc physically important Poisson manifolds which are not symplectic. But even in the symplectic context it is sometimes easier to compute the Poisson bracket than the symplectic form. as the following example shows.
8.4.2 Maxwell's Vacuum Equations as an Infinite Dimensional Hamiltonian System We shall indicate how the dynamical pair of Maxwell's vacuum equations (3) and (4) of the previous section with current J = 0 arc a Hamiltonian system. As the configuration space for Maxwell's equations. we take the space A. of vector potentials. (In more general situations. one should replace A. by the set of connections on a principal bundle over configuration space.) The corresponding phase space is then the cotangent bundle T* A. with the canonical symplectic structure. Elements of T* A. may be identified with pairs (A. Y) where Y is a vector field density on JR3. (We do not distinguish Y and Y d3 x.) The pairing between A's and V's is given by integration. so the canonical symplectic structure (0 on T*A. is
(6)
with associated Poisson bracket
r
8G)
(OF oG of 3 {F, G}(A. Y)= JIl? 8A. oY - oY· 8A d x.
.
(7)
where of/oA is the vector field defined by
with the vector field of/oY defined similarly. With the Hamiltonian H(A. Y) =
If
If
'2 II Y II 2 d 3 x + '2 II curl A II 23 d x.
(8)
Hamilton's equations are easily computed to be ay = at
cur I cur I A and
aA at -- Y.
Ifwe write B for curl A and E for - Y. the Hamiltonian becomes the field energy
(9)
Section 8.4 The Lu-Poisson Bracket in Continuum Mechanics and Plasma Physics
611
(10)
Equation (9) implics Maxwell's equations
aE = curl B at
and
aB =- curl E, at
(11)
and the Poisson bracket of two functions F(A, E), G(A, E) is (12)
We can express this Poisson bracket in terms of E and B = curl A. To do this, we consider functions F: 'llx * (JR.3) ~ JR., where 'll= {curl Z I Z E * (JR.3)}. We pair'll with itself relative to the L2-inner product. This is a weakly non-degenerate pairing since by the Hodge Helmholtz decomposition
for all Z2 E * (JR.3) implies that curl Z\ = Vf, whence M = div Vf = div curl Zl = 0, i.e., f = constant by Liouville's theorem. Therefore curl Zl = Vf = 0, as was to be shown. We compute of/oA in terms of the functional derivative of an arbitrary extension F of F to *(JR.3), where F(A, E) = F(B, E), for B = curl A. Let L be the linear map L(A) = curl A so that F = F 0 (L x Identity) = F 0 (L x Identity). By the chain rule, we have for any oA E * (JR.3),
f.
R3
of. M d\ = DF(A)·M = (DF(B ) 0 DL(A» . M oA
since DL(A) = L and
r X· curl Y d x = f.
~3
3
R3
of
Y· curl X d 3x. Therefore
of
OA = curl oB .
(13)
This formula seems to depend on the extension F of F. However, this is not the case. More precisely, let K: * (JR.3) X * (JR.3) ~ JR. be such that K I 'llx * (JR.3) == o. We claim that if BE 0/, then oK/oB is a gradient. Granting this statement, this shows that (13) is independent of the extension, since any two extensions of F coincide on 'llx * (JR.3) and since curIo grad = O. To prove the claim, note that for any Z E * (JR.3),
Applications
Chapter 8
612
0= DK(B) . curl Z =
fit' ~~.
curl Z d\ =
1., curl ~~ • Z d x 3
whence curl oK/1m = 0, i.e., oK/oB is a gradient. Thus (13) implies of
M
=
of curl oB '
(14)
where on the right-hand side of loB is understood as the functional derivative relative to B of an arbitrary extension of F to X(llt3). Since of /oE = of/oE, the Poisson bracket (12) becomes
{F, C} (B, E) =
f. (
oF oC oC OF) 3 oE' curl oB - oE • curl oB d x
R'
(15)
This bracket was found by Born and Infeld [1935] by a different method. Using the Hamiltonian (16)
equations (11) are equivalent to the Poisson bracket equations
F={F,H}. Indeed, since
oH JoE = E,
curl(oH loB)
{F, H}
= curl B,
(17)
we have
=
. J~., (~~ .
=
flit' (~~ .curl B - E curl· ~~ )d x
=
DF(B)·B + DF(E) . E
=
JR.'
curl B - E· curl
~~ )d3 x 3
Moreover,
F
r (O'F oB
. of .) 3 • B + oE • E d x
f. (
= R'
OF . of .) 3 oB • B + oE • E d x,
where o'F /oB denotes the functional derivative of F relative to B in 'l/, i.e.,
DF(B)' oB
=
J.
R'
o'F
8B' oBd3 x.
(18)
The last equality in the formula for F is proved in the following way. Recall that of /oB is the functional derivative of an arbitrary extension of F computed at B, i.e.,
Section 8.4 The Lie-Poisson Brac~t in Continuum Mechanics and Plasma Physics
DF(B) . Z =
613
J
of 3 OB· Z d x
therefore since OB is a curl, this implies O'F loB and of lOB differ by a gradient which is L2-orthogonal to
B.
since
B is divergence free (again by the Helmholtz-Hodge decomposition).
Therefore (II) holds if and only if (17) does .• We next turn to the most important example of a Poisson manifold which is not symplectic. Let g denote a Lie algebra i.e., as on page 609, a vector space with a pairing [~, l1l of elements of g that is bilinear, anti symmetric and satisfies Jacobi's identity. Let g* denote its "dual", i.e., a vector space weakly paired with g via ( , ): g* x g ~ lR. If g is finite dimensional, we take this pairing to be the usual action of forms on vectors. 9.4.3 Definition For F, G : g* ~ JR., define the (±) Lie-Poisson brackets by OF OG] {F, G} ±(Il) = ± (Il, [ Oil' Oil )
(19)
where Il E g* and OF/Oil, OG/Oll E g are the functional derivatives of F and G, that is, DF(Il)' Oil = (Oil, OF/Oil)· If g is finite dimensional with a basis ~, and the structure constants are defined by
the Lie-Poisson bracket is
8.4.4 Lie-Poisson Theorem
g* with the (±)Ue-Poisson bracket is a Poisson manifold.
Proof Clearly {, }± is bilinear and skew symmetric. To show {, }± is a derivation in each argument, we show that (20)
To prove (20), let OIlE g* be arbitrary. Then
(Oil,
O~~»
= D(FG)(IlHIl = F(Il)DG(Il)' Oil + G(Il) DF(IlHIl = ( Oil, F(Il)
oG of 3il + G(Il) 011) .
Chapter 8 Applications
614
Finally. we prove the Jacobi identity. We start by computing the derivative of the map J.! of/oil
E
g. We have for every 'A.. v
E
E
g* H
g*
i.e .. (21)
Therefore. the derivative of J.! H
[
~~ ~] •
is
oF ~ oG ] (J.!)·v = [ D 2 F(J.!)(v •. ). ~ oG ] + [OF D [ ~. ~. D 2G(J.!)(v •. ) ]
where D2F(J.!)(v.,)
E
(22)
L( g*. JR) is assumed to be represented via ( • ) by an element of g.
Therefore by (19) and (22)
o
(v. oJ.! {F. G}) =D{F. G}{J.!)· v oF OG] =(v. [ ~. oJ.! ) oF OG] = (v. [ ~. oJ.! ) + (ad (OG)* ~ J.!. D2F(J.!)(v ..
where
ad(~): 9 ~ 9
..
»- (ad (OF)* oJ.! J.!. D2G(J.!)(v .. ».
is the linear map ad(~)· 11 = [~. 'Ill and ad@* : g* ~ g* is its dual
defined by (ad@*J.!.'Il) = (J.!. [~. v]). 'Il
E
g. J.!
E
g* .
Therefore. o {F. G}) = (v. [OF oJ.!' oG oJ.! ] ) - (v. D 2F{J.!) ( ad ( oG oJ.! ) * J.! •.) ) + (v. D 2G(J.!) (ad( of oJ.! )* J.! .. ) ) (v. ov o {F. G} = [OF oG )* J.!.' ) + D 2G(J.!)(ad( of ) OV oJ.!' OG] oJ.! - D 2F(J.!) ( ad ( ~ oJ.! )* J.!.'.
(23)
which in turn implies 0 {F.G}, OHJ OGJ OHJ {F. G}(J.!) = (J.!. [ oJ.! oJ.! )=(J.!. [ [OF 0J.!·~·~)
+ D 2F(J.!>( ad(
~~ ) *J.!. ad( ~: ) *J.! ) - D2G(J.!>( ad( ~: ) *J.!. ad( ~: ) *J.! ).
Section 8.4 The Lie-Poisson Bracket in Continuum Mechanics and Plasma Physics 615
The two cyclic penuutations in F, G, H added to the above fonuula sum up to zero: all six tenus involving second derivatives cancel and the three first tenus add up to zero by the Jacobi identity for the bracket of g. • 8.4.S The Free Rigid Body
The equations of motion of the free rigid body described by
an observer fixed on the moving body are given by Euler's equation (24)
TI =TIxw,
where TI, WE JR.3, TIi = Ii Wi' i = 1, 2, 3, I = (II' 12, 13) are the principal moments of inertia, the coordinate system in the body is chosen so that the axes are the principal axes, W is the angular velocity in the body, and TI is the angular momentum in the body. It is straightforward to check that the kinetic energy 1
H(n) = -TI· 2
(25)
W
is a conserved quantity for (24). We shall prove below that (24) are Hamilton's equations with Hamiltonian (25) relative to a
(-) Lie-Poisson structure on JR.3. The vector space JR.3 is in fact a Lie algebra with respect to the bracket operation given by the cross product, i.e. [x, y]
=x x
y. (This is the structure that it inherits from the rotation
group.) We pair JR.3 with itself using the usual dot-product, i.e., (x, y ) =x • y. Therefore, if F: JR.3 ~ JR., of/oTI = VF(TI). Thus, the (-) Lie-Poisson bracket is given via (19) by the triple
product {F, G }(n) = -TI • (VF(TI) x VG(TI».
Since oH/oTI = W, we see that for any F: JR.3
(26)
~ JR.,
d . . dt (F(n» = DF(n) • TI = TI • VF(n) = -TI • (VF(n) x W = VF(n) • (TI x w)
so that
F=
{F, H} for any F: JR.3 ~ JR. is equivalent to Euler's equations of motion (24).
8.4.6 Proposition (Pauli [1953], Martin [1959], Arnold [1966J. Sudarshan and Mukunda
[1974]) Euler's equations (24)for afree rigid body are a Hamiltonian system in JR.3 relative to the (-) Lie Poisson bracket (24) and Hamiltonian function (25). 8.4.7 Ideal Incompressible Homogeneous Fluid Flow In §8.2 we have shown that the equations of motion for an ideal incompressible homogeneous fluid in a region with smooth boundary
an
are given by Euler's equations of motion
n
C JR.3
616
Chapter 8 Applications
at + (v av
• v)v = -vp
) (27)
div v=O v(t, x) E Tx(am for x E an
with initial condition v(O, x) = vo(x), a given vector field on n. Here v(t, x) is the Eulerian or spatial velocity, a time dependent vector field on n. The pressure p is a function of v and is uniquely determined by v (up to a constant) by the Neumann problem (take div and the dot product with n of the first equation in (27»
~p
= -div«v • V)V)
~ = Vp. n = --((v· V)V)· n
on an,
)
(28)
where n is the outward unit normal to an. The kinetic energy
f.
1 nllvll 2 dx 3 H(v)=1
(29)
has been shown in §8.2 to be a conserved quantity for (27). We shall prove below that the first equation in (27) is Hamiltonian relative to a (+) Lie-Poisson bracket with Hamiltonian function given by (29). ... Consider the Lie algebra *div(n) of divergence free vector fields on n tangent to an with bracket given by minus the bracket of vector fields, i.e., for u, v E *div(n) define [u, v] = (v • V)u - (u • V)v.
(30)
(The reason for this strange choice comes from the fact that the usual Lie bracket for vector fields is the right Lie algebra bracket of the diffeomorphism group of n.) Now pair *div(n) with itself via the L2-pairing. As in 8.4.2, using the Hodge-Helmholtz decomposition, it follows that this pairing is weakly non-degenerate. In particular
cSH Yv"=v.
(31)
The (+) Lie-Poisson bracket on *div(n) is
5.
{F,G}(v)= n v·
cSG V) [(8V.
Therefore, for any *div
cSF ( cSF ) cSG ] 3 cSv - 8V. V 8V d x.
(32)
Section 8.4 The Lie-Poisson Bracket in Continuum Mechanics and Plasma Physics
. Jo.r of OV
d dt (F(v» = DF(v)
V=
rv Jo.
=
0
0
.
3
v d x=
((v
0
rv Jo.
[(v
0
0
of - ( OF SV OV
V)
3 Jo.r dv dF
OF) V) OV d x -
V
0
( '21 II v II
0
617
)] 3 V v d x
2) d3x.
To handle the first integral, observe that if ~
f, g : n
JR, then div(fgv) = f div(gv) + gv
0
Vf,
so that by Stokes' theorem and von = 0, div v = 0, we get
In gv
0
Vf d\ =
fan fg von dS - In f div(gv) d x
=-
3
In fv
0
Vgd3 x.
Applying the above relation to g = vi, f = OF, and summing over i = 1, 2, 3 we get Ov'
Irn v so that
F=
0
((v
0
V) OF) Ov d 3x = -
r SV of Jo.
0
«v
0
V)v)d 3x
(F, H) reads (33)
for any F: Xdiv(n) ~ R One would like to conclude from here that the coefficients of oFjov on both sides of (33) are equal. This conclusion, however, is incorrect, since (v oV)v + (I!2) VII v 112 is not divergence free. Thus, applying the Hodge-Helmholtz decomposition, write (v
0
V)v +
'21 VII v II 2 = X -
(34)
Vf
where XE xdiv
~(f + ~ II v II 2) = - div«v
0
V)v)
!(f+~IIVIl2)=-«voV) which coincides with (28), i.e.,
ov)on
)
Chapter 8 Applications
618
f+
±II
v II 2 = P + constant
(35)
Moreover, since
f.n of
~ uV
0
V f d3 x =
J
an
aF
f"" ° n dS dV
l' n
f dlv
of d3 x = 0
~ uV
we have from (33), (34), (35) av at =- X=-(v o V)v- Vp
which is the first equation in (27). We have thus proved 8.4.8 Proposition (Arnold [1966], Marsden and Weinstein [1983].) Euler's equations (27) are a Hamiltonian system on Xdiv(O) relative to the (+) Lie-Poisson bracket (32) and Hamiltonian function given by (29). 8.4.9 The Poisson-Vlasov Equation We consider a collisionless plasma consisting (for notational simplicity) of only one species of particles with charge q and mass m moving in Euclidean space ]R3 with positions x and velocities v. Let f(x, v, t) be the plasma density in the plasma space at time t. In the Coulomb or electrostatic case in :hich there is no magnetic field, the motion of the plasma is described by the Poisson-Vlasov equations which are the (collisionless) Boltzmann equations for the density function f and the Poisson equation for the scalar potential
af af q dqlf af at+Voax -maxoav=O
(36)
~
(37)
q
f
f(x, v) d3 v = Pr,
where ajax and ajav denote the gradients in ]R3 relative to the x and v variables, Pc is the charge density in physical space, and ~ is the Laplacian. Equation (36) can be written in "Hamiltonian" form af (38) at + {f, .?{} = 0, where {, } is the canonical Poisson bracket on phase space, (39) p = mv and J4 = m II v 112 + q
Section 8.4 The Lie-Poisson Bracket in Continuum Mechanics and Plasma Physics
1 {J{c,n =m
(i)1{ af a J{ c af) 1 (a
af ° av- mv
619
af)
0'dX
q a
where (,} + is a (+) Lie-Poisson bracket (to be defined) of functionals and H is the total energy. Let us state this more precisely. Let V = (f E Ck (IR6)1 f ~ 0 as II x II ~ 00, II v II ~ oo} with the L2 -pairing ( , ): V x V ~ 1R; (f, g) = f(x, v) g(x, v) d3 x d3v. If F: V ~ IR is differentiable at f E V, the functional derivative
f
~~
~~
is, by defmition, the unique element
OF g)= DF(f)· g= (M'
E
V such that
I of
M(x, v) g(x, v) d3 x d 3v.
The vector space V is a Lie algebra relative to the canonical Poisson bracket (39) on 1R6 . For two functionals F, G : V ~ IR their (+) Lie-Poisson bracket (F, G
{}
F, G + (x, v) =
I
f(x, v)
t: V ~ IR
is then given by
{
OF M OG } (x, v) d 3x d3v, M'
where ( , } is the canonical bracket (39). For any f, g, h E V we have the formula (40)
lndeed by integration by parts, we get
I I ax
1
=-m 1 = m
af ah 3 3 I axogavdxdv+m
av
I
af ah 3 3 gavoaxgdxdv
af - af ° ah) d 3x d 3v = g(ah ° av
ax
I
g{h, f} d3 x d 3v .
8.4.10 Proposition (Iwinski and Turski [1976], Kaufman and Morrison [1980] and Morrison [1980]) Densities f E V evolve according to the Poisson-Vlasov equation (38) if and only if any
620
Chapter 8 Applications
differentiable function F: V
~ JR.
having functional derivative of/Of evolves by the (+)
Lie-Poisson equation (41 )
with the Hamiltonian H: V ~ JR. equal to the total energy
IJ
H(f) = 2"
Proof First we compute
~~
m
II v II 2 f(x,
using the definition DH(O . Of
H(f) is linear in f and the second term is
Therefore
oH
&
I
= 2" m
v) d 3 x d 3v +
2
II v II + q<Jlr =
.
~J <JlrPr d\ = ~
f M{:Hr, of
f}d 3 x d 3 v
f
=
J
2"<Jlr<x) d 3 x.
J~~
of. Note that the flrstterm of
1IY'<Jlr 112 d 3 x since t.<Jlr =
-
Pr. Then
:Hr. We have
. JM
F(f) = DF(f) . f =
and
Jl
of.f d3 x d3v
{OH }d x d v = f f{OF OH} d x d v = {F, H}+(f). M'&
of &' f = M
3
3
3
3
by (40). Thus (41), for any F having functional derivatives, is equivalent to (38) . •
8.4.11 The Maxwell-Vlasov Equations We consider a plasma consisting of particles with charge ql and mass m moving in Euclidean space
JR.3
with positions x and velocities v. For
simplicity we consider only one species ofparticlc; the general case is similar. Let f(x, v, t) be the plasma density at time
t, E(x, t) and B(x, t) be the electric and magnetic fields.
The
Maxwell-Vlasov equations are:
df + v • df + ..9.(E+
at
aB c1 at
dxm
= -
1 at aE = C
V
x B ).
c
~
0
(42) (43a)
curl E,
curl B -
=
av'
cq f vf(x, v, tM v , 3
(43b)
Section 8.4 The Lie-Poisson Braclwt in Continuum Mechanics and Plasma Physics
621
together with the non evolutionary equations div E = Pr' where Pr = q
f
f(x, v, t)d 3v,
(44a)
div B = O. Letting c ----)
00
leads to the Poisson-Vlasov equation (36) af
at + v where
(44b)
af q a
~
H(f, E, B) =
f211 l
v II 2f(x, v, t)dx dv +
fl[ 2 IIE(x, t)1I 2 + I B(x, t)11 2]3 d x.
(45)
Let 0/= (curl Z I Z EX (JR.3)}.
8.4.12 Theorem (Iwinski and Turski [1976), Morrison [1980], Marsden and Weinstein [1982].) (i) The manifold .'T(JR.6) x X (JR.3) X 0/ is a Poisson manifold relative to the bracket
{}
F, G (r, E, B)
+
=
f {U' 8r8G} f
8F
f(
3 3 d xd v+
8G
J(8G 8G 8G 8B' curl 8B - 8E
8F af 8G af 8F) 3 3 8E • av 8f - liE • av 8f d x d v +
f
fB·
(a 8F
Tv U
x
8F \,3 J' x
• curl 88
8G \,3 3 ava 8r J' x d v.
(46)
(ii) The equations of motion (42) and (43) are equivalent to
F= (F, H} where F is any locally defined function with functional derivatives and ( , } is given by (46). Part (a) follows from general considerations on reduction (sec Marsden and Weinstein [1982]). The direct verification is laborious but straightforward, if one recognizes that the first two terms are the Poisson bracket for the Poisson- Vlasov equation and the Bom-Infeld bracket respectively. Proof of (b) Since 8H/8f =
II v 112/2. 8H/8E = E, curl 8H/8B = curl B, we have, by (40) and
integration by parts in the fourth integral,
Chapter 8 Applications
622
f( { } f {Tf' "2 II II f( Tv" "2 II Tv" f (tv ~~ ~ II f ~~ [{ ~ II ~~ fTv" "2 II II f( f oF
F, H =
v 2} d 3x d 3v +
1
oF i)f 1 oE •
+
+ m
v
112
of curl 8 - E • curl 08 OF)d 3x 08·
-
x :v (
x
oE
v
112) )d 3 x d\
- div.(v x m)]d3 X d 3v
OF [ curl 8 + oE·
+
i)f OF) d 3 d 3
•
V
v 112 'f} - F·
=
E
i)f 1
v 2 d3v ] d3x -
f
E· of 08 curl d 3x
where div. denotes the divergence only with respect to the v-variable. Since I { "2
2}
II v II ,f = - v •
fTv" "2 II f i)f 1
i)f. i)f ax' dlv.(v x m) = Tv" 2
3
v II d x =
-
of 3 E • curl 08 d x =
and we get
{ } fTf F, H =
. fTr·
f f
ax - (E + v x 8)
of [ i ) f v •
(v x 8),
3
v f d x, of 3 08· curl E d x ,
•
'
i)f ] 3 3 Tv" d xd v
f8B.
'.3 OF 0'F B a x taking into account that 8if' 8B
differ by a gradient (by (15» which is L2 -orthogonal to
'II (of which both curl E and B are a
and since F =
of
. 3 3 fd xd v+
f
•
of
.
3
OR· Ed x +
o'F
member), it follows from (48) that the equations (42)-(44) (with q = c = m = 1) are equivalent to (47) . •
Exercises 8.4A Find the symplectic fonn equivalent to the Bom-Infeld bracket (16) on
8.4B Show that the Hamiltonian vector field XH E . given . _ (OH)* ~. by X H() ~ = + ad Bil
{ f:
lR
-t
'* (JR.3).
'* (g*) relative to the (±) Lie-Poisson bracket
IS
8.4C (Gardner [1970]). Let V =
'IIx
JR.I f is
C~,
lim f(x) I xl-+~
=
o}
Section 8.4 The ill-Poisson Bracket in Continuum Mechanics and Plasma Physics
(i)
623
Show that the prescription
defines a Poisson bracket on V for appropriate functions F and G (be careful about what hypotheses you put on F and G). (ii) Show that the Hamiltonian vector field of H : V ~]R is given by
(iii) Let H(O =
S:- (f + ~ ~,,)dX.
Show that the differential equation for X H is the
Korteweg-de Vries equation
8.4D (Ratiu [1980)). Let 9 be a Lie algebra and e E g* be fixed. Show that the prescription
defines a Poisson bracket on g*. (Hint: Look at the formulas in the proof of Theorem 8.4.4.) 8.4E
(i) (Pauli [1953]. R. lost [1964)). Let P be a finite dimensional Poisson manifold satisfying the following condition: If (F. G) = 0 for any locally defined F implies G = constant. Show that there exists an open dense set U in P such that the Poisson bracket restricted to U comes from a symplectic form on U. (Hint: Define B : T*P
x T*P ~]R by B(dF. dG) = (F. G). Show first that U = {p E PI Bp (ex. P) = 0 for all ex E Tp *p implies Pp = 0) is open and dense in P. Then show that B can be inverted at points in U.) (ii) Show that. in general. U ~ P by the following example. On {F. G](x. y) = y (
Show that U in (i) is
]R2
]R2
define
oF oG of OG) ax ay - dY dx
\(Ox-axis). Show that on U. the symplectic form
generating the above Poisson bracket is dx
A
dy/y.
§8.5 Constraints and Control The applications in this fmal section all involve the Frobenius theorem. Each example is necessarily treated briefly, but hopefully in enough detail so the interested reader can pursue the subject further by utilizing the given references. We start with the subject of Iwlorwmic constraints in Hamiltonian systems. A Hamiltonian system as discussed in §8.1 can have a condition imposed that limits the available points in phase space. Such a condition is a constraint. For example, a ball tethered to a string of unit length in IR3 may be considered to be constrained only to move on the unit sphere S2 (or possibly interior to the sphere if the string is collapsible). If the phase space is T*Q and the constraints are all derivable from constraints imposed only on the configuration space (the q's), the constraints are
called Iwlorwmic. For example, if there is one constraint f(q) = 0 for f: Q ~ IR, the constraints on T*Q can be simply obtained by differentiation: df= 0 on T*Q. If the phase~ace is TQ, then the constraints are holonomic iff the constraints on the velocities are saying that the velocities are tangent to some constraint manifold of the positions. A constraint then can be thought of in terms of velocities as a subset E C TM. If it is a subbundle, this constraint is thus holonomic integrable in the sense of Frobenius' theorem.
iff it is
Constraints that are not holonomic, called rwnlwlonomic constraints, are usually difficult to handle. Holonomic constraints can be dealt with in the sense that one understands how to modify the equations of motion when the constraints are imposed, by adding forces of constraint, such as centrifugal force. See, for example Goldstein [1980, Ch. 1], and Abraham and marsden [1978, §3.7J. We shall limit ourselves to the discussion of two examples of nonholonomic constraints.
A classical example of a nonholonomie system is a disk rolling without slipping on a plane. The disk of radius a is constrained to move without slipping on the (x, y)-plane. Let us fix a point P on the disk and call e the angle between the radius at P and the contact point Q of the disk with the plane, as in Fig. 8.5.1. Let (x, y, a) denote the coordinates of the center of the disk. Finally, if q> denotes the angle between the tangent line to the disk at Q and the x-axis, the position of the disk in space is completely determined by (x, y, e, q». These variables form elements of our configuration space M = IR2 X S t X st. The condition that there is no slipping at Q means that the velocity at Q is zero; i.e.,
(total velocity = velocity of center plus the velocity due to rotation by angular velocity de/dt).
Section 8.5 Constraints and Control
625
Figure 8.5.1 The expression of these constraints in terms of differential forms is ffi l
ffi l
= 0,
ffi2
= 0 where
= dx + a cos q> dB and ffiz = dy + a sin q> dB. We compute that ffi = ffi l /\ ffiz = dx /\ dy + a cos q> dB /\ dy + a sin q> dx /\ dB. dffi l = - a sin q> dq> /\ dB , dffi2 = a cos q> dq> /\ dB. dffi l /\ ffi = - a sin q> dq> /\ dB /\ dx /\ dy, dffiz /\ ffi = a cos q> dq> /\ dB /\ dx /\ dy.
These do not vanish identically. Thus, according to Theorem 6.4.20, this system is not integrable and hence these constraints are nonholonomic. A second example of constraints is due to E. Nelson [1967). Consider the motion of a car and denote by (x, y) the coordinates of the center of the front axle, q> the angle formed by the moving direction of the car with the horizontal, and B the angle formed by the front wheels with the car (Fig. 8.5.2).
L Figure 8.5.2
Chapter 8 Applications
626
The configuration space of the car is
]R2
x
1r2 ,
parametrized by (x, y,
prove that the constraints imposed on this motion are nonholonomic. Call the vector field X = a/aa
steer. We want to compute a vector field Y corresponding to drive. Let the car be at the configuration point (x, y,
~
+ sin a
!.
A direct computation shows that the vector field wriggle,
W = [X, Y) = - sin(
~
+ cos a
!, '
and slide,
s = [W, Y) = -
sin
~,
satisfy [X, W) and
=-
Y,
[s, X) = 0,
[S, W) = cos
[S, Y)
= sin a cos
a cos
a:
+ cos a sin
+ sin a sin
~,
~.
Define the vector fields Z!' Z2 by
Zl = [S, Y) = -W + (cos a)S + cos a :
!.
A straightforward calculation shows that
i.e., {X, Zl' Z2' s} span a four dimensional Lie algebra g with one dimensional center spanned by S. In addition, its derived Lie algebra [g, g)
C
g, equals span {Zl'~} and is therefore
abelian and two dimensional. Thus g has no nontrivial non-abelian Lie subalgebras. In particular the subbundle of T(lR2 x
1r2)
spanned by X and Y is not involutive and thus
not integrable. By the Frobenius theorem, the field of two-dimensional planes spanned by X and Y is not tangent to a family of two-dimensional integral surfaces. Thus the motion of the car, subjected only to the constraints of "steer" and "drive" is nonholonomic. On the other hand, the motion of the car subjected to the constraints of "steer", "drive" and "wriggle" is holonomic.
627
Section 8.5 Constraints and Control
Moreover. since the Lie algebra generated by these three vector fields is abelian. the motion of the car with these constraints can be described by applying these three vector fields in any order. Next we turn our attention to some elementary aspects of control theory. We shall restrict our attention to a simple version of a local controllability theorem. For extensions and many additional results. we recommend consulting a few of the important papers and notes such as Brocket [1970. 1983]. Sussmann [1977]. Hermann and Krener [1977]. Russell [1979]. Hermann [1980]. and Ball. Marsden. and Slemrod [1982] and references therein. Consider a system of differential equations of the form w(t) = X(w(t» + p(t)Y(w(t» on a time interval [0. T] with initial conditions w(O)
(1)
= Wo
where w takes values in a Banach
manifold M. X and Y are smooth vector fields on M and p: [0. T]
~
JR
is a prescribed
function called a control. The existence theory for differential equations guarantees that (1) has a flow that depends smoothly on Wo and on p lying in a suitable Banach space Z of maps of [0. T] to JR. such as the space of C 1 maps. Let the flow of (1) be denoted
(2) We consider the curve w(t. O. wo) = wo(t); i.e .• an integral curve of the vector field X. We say that (1) is locally controllable (at time T) if there is a neighborhood U of wo(T) such that for any point hE U. there is apE Z such that w(T; p. wo) = h. In other words. we can alter the endpoint of wo(t) in a locally arbitrary way by altering p. (Fig. 8.4.3).
/ /
I
---- ...... "
U
h
\ wo(T)
:
I I
---- '"
/
Figure 8.5.3 To obtain a condition under which local controllability can be guaranteed. we fix T and Wo
628
Chapter 8 Applications
and consider the map P : Z ~ M; p ...... w(T, p, wo)'
(3)
The strategy is to apply the inverse function theorem to P. The derivative of Ft(w o' p) with respect to p in the direction P E Z is denoted
Differentiating d dt w(t, p) = X(w(t, p» + p(t)Y(w(t, p»
with respect to p at p = 0, we find that in T2M (4)
,
To simplify matters, let us assume M = E is a Banach space and that X is a linear operator, so (4) becomes (5)
Equation (5) has the following solution given by the variation of constants formula
(6)
since wo(t) = etXwo for linear equations. 8.5.1 Proposition If the linear map LT : Z
~
E given by (6) is surjective, then the (1) is locally
controllable (at time T).
Proof This follows from the "local onto" form of the implicit function theorem (see 2.5.9) applied to the map P. Solutions exist for time T for small p since they do for p = 0; see corollary 4.1.25 .• 8.5.2 Corollary Suppose E =]Rn and Y is linear as well. If dim span {Y(wo)' [X, Y](w o)' [X,[X,Y]](w o ), ... } = n, then (1) is locally controllable. Proof We have the Baker-Campbell-Hausdorff formula 2
e-sXYe sX = Y + S[X, Y] + ~ [X, [X, Y]] +
Section 8.5 Constraints and Control
629
obtained by expanding eSx = I + sX + (s2/2)X2 + ... and gathering tenns. Substitution into (6) shows that LT is surjective. _ For the case of nonlinear vector fields and the system (1) on finite-dimensional manifolds, controllability hinges on the dimension of the space obtained by replacing the foregoing commutator brackets by Lie brackets of vector fields, n being the dimension of M. This is related to what are usually called Chow's theorem and Hermes' theorem in control theorey (see Chow [1947]). To see that some condition involving brackets is necessary, suppose that the span of X and Y fonns an involutive distribution of TM. Then by the Frobenius theorem, Wo lies in a unique maximal two-dimensional leaf £J,wo) of the corresponding foliation. But then the solution of (1) can never leave 4wo)' no matter how p is chosen. Hence in such a situation, (1) would not be locally controllable; rather, one would only be able to move in a two-dimensional subspace. If repeated bracketing with X increases the dimension of vectors obtained then the attainable staLes increase in dimension accordingly.
Exercises 8.SA Check that the system in Fig. 8.5.1 is nonholonomic by verifying that there are two vector fields X, Y on M spanning the subset E of TM defined by the constraints x=a9cosq>=O
and
y + a 9 sin q> = 0
such that [X, Y] is not in E; i.e., usc Frobenius' theorem directly rather than using Pfaffian systems. 8.SB Justify the names wriggle and slide for the vector fields W and S in the example of Fig. 8.5.2 using the product formula in Exercise 4.2D. Use these fonnulas to explain the following statement of Nc\son [1967, p. 35]: "the Lie product of "steer" and "drive" is equal to "slide" and "rotate" ( = a/aq» on 9 = 0 and generates a flow which is the simultaneous action of sliding and rotating. This motion is just what is needed to get out of a tight parking spot." 8.SC The word holonomy arises not only in mechanical constraints as explained in this section but also in the theory of connections (Kobayashi-Nomizu [1963, II, §7,8]). What is the relation between the two uses, if any? 8.4D In linear control theory (1) is replaced by
630
Chapter 8 Applications N
W(t)
X . W(t) +
I
i~
where X is a linear vector field on
]Rn
pJt)Y j
,
1
and Yj are constant vectors. By using the methods
used 10 prove 8.5.1, rediscover for yourself the Kalman criterion for local controllability, namely, {Xk Yj I k = 0, 1, ... , n - 1, i = 1, ... , N} spans
]Rn.
References R. Abraham [1963] Lectures of Smale on Differential Topology, Notes, Columbia University. R. Abraham and J. Marsden [ 1978] Foundations of Mechanics, Second Edition, Addison-Wesley, Reading Mass. R. Abraham and J. Robbin [1967] Transversal Mappings and Flows, Addison-Wesley, Reading Mass. R.A. Adams [1975] Sobolev Spaces, Academic Press, New York. V. I. AmoI'd [1966] Sur la geometrie differentielle des groupes de Lie de dimenson infinie et ses applications a L'hydrodynamique des fluids parfaits, Ann. Inst. Fourier. Grenoble, 16,319 - 361. V.I. AmoI'd [1978] Mathematical Methods of Classical Mechanics, Springer Graduate Texts in Mathematics, 60, Springer-Veriag. New York. V.I. Amol'd and A. Avez [1967] Theorie ergodique des systemes dynamiques, Gauthier-Villars, Paris (English ed., Addison-Wesley, Reading, Mass., 1968) J. M. Ball and J.E. Marsden [1984] Quasiconvexity at the boundary, positivity of the second variation and elastic stability, ArchRat. Mech. An. 86,251-277. J.M. Ball, J.E. Marsden and M. Slemrod [1982] Controllability for distributed bilinear systems, SIAM f. Control and Optim., 20, 575-597. G.K. Batchelor [1967] An Introduction to Fluid Dynamics, Cambridge Univ. Press, Cambridge, England. M. Berger and D. Ebin [1969] Some decompositions of the space of symmetric tensors on a Riemannian manifold, f. Diff. Geom., 3, 379-392. G.D. Birkhoff [1931] Proof of the ergodic theorem, Proc. Nat A cad. Sci., 17,656-660. D. Bleecker [1981] Gauge Theory and Variational Principles, Global Analysis: Pure and Applied, I, Addison- Wesley, Reading, Mass. R. Bonic and F. Reis [1966] A characterization of Hilbert space, Acad. Bras. de Cien. 38,239-241. R. Bonic and J. Frampton [1966] Smooth functions on Banach manifolds, f. Math. Mech. 16, 877898. J.M. Bony [1969] Principe du maximum, inegalite de Hamack et unicite du probleme de Cauchy pour les operateurs elliptiques dcgeneres, Ann. Inst. Fourier Grenoble 19,277-304. M. Born and L. Infeld [1935] On the quantization of the new field theory, Proc. Roy. Soc. A. 150, 141-162. R. Bott [1970] On a topological obstruction to integrability, Proc. Symp. Pure Math. 16, 127-131. N. Bourbaki [1971] Varietes differentielles et analytiqes, Fascicule de resultats 33, Hermann.
References
632
J.P. Bourguigon [1975] Une stratification de l'espace des structures ricmanniennes. Compo Math. 30, 1-4l. J. P. Bourguignon and H. Brezis [1974] Remarks on the Euler equation. f. Funct. An. 15,341-363. R. Bowen [1975] Equilibrium States and the Ergrodic Theory of Anosov Diffeo - morphisms, Springer Lecture Notes in Mathematics. 470. G. E. Bredon [1972] Introduction to Compact Transformation Groups, Academic Press. New York. H. Brezis [1970] On Characterization of Flow Invariant Sets. Commun. Pure Appl. Math 23, 261263.
R.W. Brockett [1970] Finite Dimensional Linear Systems. Wiley. New York. R. W. Brockett [1983] A Geometrical Framework for Nonlinear Control and Estimation. CBMS Conference Series. SIAM. M. Buchner and J. Marsden and S. Schecter [1983] Applications of the blowing-up construction and algebraic geometry to bifurcation problems. f. Diff. Eqs. 48,404-433.
•
M. Buchner and J. Marsden and S. Schecter [1983] Examples for the infinite dimensional Morse Lemma. SIAM f. Math An. 14, 1045-1055. D. Burghelea. A. Albu. and T. Ratiu [1975] Compact Lie Group Actions (in Romanian). Monografii Matematice 5, Universitatea Timisoara. W.L. Burke [1980] Spacetime. Geometry. Cosmology. University Science Books. Mill Valley. Ca. C. Camacho. A.L. Neto [1985] Geometric Theory of Foliations. Birkhauser. Boston. Basel. Stuttgart. M. Cantor [1981] Elliptic operators and the decomposition of tensor fields. Bull. Am. Math. Soc. 5, 235-262. C. Caratheodory [1909] Untersuchungen tiber die Grundlagen der Thermodynamik. Math. Ann. 67,
355-386. C. Caratheodory [1965] Calculus of Variation and Partial Differential Equations. Holden-Day. San Francisco. E. Cartan [1945] Les systemes differentiels exterieurs et leur applications geometriques, Hermann. Paris.
EE Chen [1974] Introduction to Plasma Physics. Plenum P. Chernoff [1974] Product Formulas, Nonlinear Semigroups and Addition of Unbollnded Operators. Memoirs of Am. Math. Soc. 140. P. Chernoff and J. Marsden [1974] Properties of Infinite Dimensional Hamiltonian Systems, Springer Lecture Notes in Mathematics. 425. C. Chevalley [1946] Theory of Lie groups. Princeton University Press. Princeton. N.J.
G. Choquet [1969] Lectures on Analysis. 3 vols. Addison-Wesley. Reading. Mass.
References
633
Y. Choquet-Bruhat. C. DeWitt-Morette and M. Dillard-Blieck [1982] Analysis. Manifolds, and Physics. Rev. ed .• North-Holland. Amsterdam. A. Chorin and 1. Marsden [1979) A Mathematical Introduction to Fluid Mechanics. Springer
Universitext. A. Chorin. T.l.R. Hughes. M.F. McCracken and 1. Marsden [1978]. Product formulas and
numerical algorithms. Commun. Pure Appl. Math. 31,205-256. S.N. Chow and 1.K. Hale [1982) Methods of Bifucation Theory. Springer. New York. S.N. Chow. 1. Mallet-Paret and 1. Yorke [1978) Finding zeros of maps: hompotopy methods that are constuctive with probability one. Math. Compo 32,887-899. W. L. Chow [1947) Uber Systeme von linearen partiellen Differentialgleichungen. Math. Ann. 117, 89-105. A. Clebsch [1859] Uber die Integration der hydrodynamischen Gleichungen. 1. Reine. Angew. Math.
56, 1-10. Clemmow and Daugherty [1969) Electrodynamics of Particles and Plasmas. Addison Wesley I.M. Cook [1966) Complex Hilbertian structures on stable linear dynamical systems. 1. Math. Mech. 16, 339-349. M. Craioveanu and T. Ratiu [1976] Elements of Local Analysis. Vol. 1.2 (in Romanian). Monografii Mathematice 6,7. Universitatea Timisoara. R.C. Davidson [1972] Methods in Nonlinear Plasma Theory. Academic Press G. deRham [1984] Differentiable Manifolds: Forms, Currents, Harmonic Fonns. Grundlehren der Math. Wissenschaften. 266. Springer-Verlag. New York. P.A.M. Dirac [1964] Lectures on Quantum Mechanics. Belfer Graduate School of Sci., Monograph Series. 2. Yeshiva University G. Duff and D. Spencer [1952) Harmonic tensors on Riemannian manifolds with boundary. Ann. Math. 56, 128-156. G. Duffing [1918] Erzwungene Schwingungen bei veriinderlichen Eigenfrequenz, Vieweg u. Sohn. Braunschweig. I.E. Dzyaloshinskii and G.E. Volovick [1980) Poisson Brackets in Condensed Matter Physics. Ann. of Phys .. 125. 67-97 D. Ebin [1970] On completeness of Hamiltonian vector fields. Proc. Am. Math. Soc. 26. 632-634. D. Ebin and 1. Marsden [1970) Groups of diffeomorphisms and the motion of an incompressible fluid. Ann Math. 92, 102-163. 1. Eells [1958) On the geometry of function spaces. in Symposium de Topologia Algebrica, Mexico
UNAM. Mexico City. pp. 303-307. H. Elliasson [1967] Geometry of manifolds of maps. 1. Diff. Geom. 1, 169-194.
D. Elworthy and A. Tromba [1970a] Differential structures and Fredholm maps on Banach manifolds. Proc. Symp. Pure Math. 15,45-94.
References
634
D. Elworthy and A. Tromba [1970b] Degree theory on Banach manifolds, Proc. Symp. Pure Math. 18, 86-94. A. Fischer [1970] A theory of superspace, in Relativity, M. Carmelli et al. (Eds). Plenum, New
York. A. Fischer and 1. Marsden [1975] Deformations of the scalar curvature, Duke Math. 1.42, 519-547. A. Fischer and 1. Marsden [1979] Topics in the dynamics of general relativity, in Isolated Gravitating
Systems in General Relativity, 1. Ehlers (ed.) Italian Physical Society. North-Holland, Amsterdam. pp. 322-395. H. Flanders [1963] Differential Forms. Academic Press. New York.
L.E. Fraenkel [1978] Formulae for high derivatives of composite functions. Math. Proc. Camb. Phil. Soc. 83, 159-165. 1. Frampton and A. Tromba [1972] On the classification of spaces of Holder continuous functions, 1. Funct. An. 10, 336-345. T. Fulton. F. Rohrlich and L. Witten [1962] Conformal invariance in physics. Rev. Mod. Phys. 34, 442-457. M.P. Gaffney [1954] A special Stokes's theorem for complete Riemannian manifolds. Ann. Math. 60, 140-145. C.S. Gardener. I.M. Greene. M.D. Kruskal and R.M. Muira [1974] Korteweg-deVries Equation and Generalizations. VI. Methods for Exact Solution. Comm. PUre Appl. Math., 27. 97-133 G. Glaeser [1958] Etude de quelques algebres Tayloriennes. 1. Anal. Math. 11, 1-118. H. Goldstein [1980] Classical Mechanics. 2nd ed. Addison-Wesley, Reading. Mass.
M. Golubitsky and 1. Marsden [1983] The Morse lemma in infinite dimensions via singularity theory. SIAM. 1. Math. An. 14,1037-1044. M. Golubitsky and V. Guillemin [1974] Stable Mappings and their Singularities, Graduate Texts in Mathematics, Vol 14, Springer - Verlag. New York. M. Golubitsky and D. Schaeffer [1985] Singularities and Groups in Bifurcation Theory. I. Springer. Verlag. New York. L. Graves [1950] Some mapping theorems. Duke Math. 1. 17, 111-114. H. Grauert [1958] On Levi's problem and the imbedding of real-analytic manifolds. Am. 1. of. Math. 68, (2). 460-472. 1. Guckenheimer and P. Holmes [1983] Nonlinear Oscillations, Dynamical Systems and Bifurcations 01 Vector Fields. Springer Applied Math. Sciences, Vol. 43. V. Guillemin and A Pollack [1974] Differential Topology, Prentice-Hall. Englewood Cliffs. N.J. V. Guillemin and S. Sternberg [1977] Geometric Asymptotics, American Mathematical Society Surveys. Vol 14.
References
635
V. Guillemin and S. Sternberg [1984] Symplectic Techniques in Physics, Cambridge University Press. M. Gurtin [1981] An Introduction to Continuum Mechanics, Academic Press, New York. 1.K. Hale [1969] Ordinary Differential Equations, Wiley-Interscience, New York. P. R. Halmos [1956] Lectures on Ergodic Theory, Chelsea, New York. R. Hamilton [1982] The inverse function theorem of Nash and Moser, Bull. Am. Math. Soc. 7,65222. P. Hartman [1972] On invariant sets and on a theorem ofWazewski, Proc. Am. Math. Soc. 32,511520. P. Hartman [1973] Ordinary Differential Equations, 2nd ed. Reprinted by Birkhauser, Boston. S. Hawking and G.F.R. Ellis [1973] The Large Scale Structure Univ.Press, Cambridge, England.
0/ Space-Time,
Cambridge
C. Hayashi [1964] Nonlinear Oscillations in Physical Systems, McGraw-Hill, New York.
R. Hennann [1973] Geometry, Physics and Systems, Marcel Dekker, New York. R. Hennann [1977] Differential Geometry and the Calculus o/Variations, 2nd ed. Math. Sci. Press, Brookline, Mass. R. Hennann [1980] Cartanian Geometry, Nonlinear Waves and Control Theory, Part B. Math. Sci. Press, Brookline, Mass. R. Hermann and A.l. Krener [1977] Nonlinear conrollability and observability, IEEE Trans. on Auto. Control. 22,728-740. T.H. Hildebrandt and L.M.Graves [1927] Implicit functions and their differentials in general analysis, Trans. Am. Math. Soc. 29, 127-53. M.W. Hirsch [1976] Differential Topology. Graduate Texts in Mathematics, Vol 33, Springer Verlag. New York. M. W. Hirsch and S. Smale [1974] Differential Equations, Dynamical Systems and Linear Algebra, Academic Press, New York. V.W.D. Hodge[1952] Theory and Applications University Press. Cambridge. England.
0/ Harmonic Integrals, 2nd ed. Cambridge
D.D. Holm, J.E. Marsden, T. Ratiu and A. Weinstein [1985] Nonlinear Stability of Fluid and Plasma Equilibria, Physics Reports, 123, 1-116 H. Holmann and H. RummIer [1972] Alternierende Differentialformen. BI- Wissen - schaftsverlag, ZUrich. P. Holmes [1979a] A nonlinear oscillator with a strange attractor, Phil. Trans. Roy. Soc. London. P. Holmes [1979b] Averaging and chaotic motions in forced oscillations, SIAM 1. on Appl. Math. 38,68-80, and 40, 167-168. V.H. Hopf [1931] Ober die Abbildungen der Dreidimensionalen Sphare auf die KugeWiche, Math. Annalen. 104,637-665.
636
References
W. Huebsch [1955] On the covering homotopy theorem, Annals. Math. 61,555-563. T.1.R. Hughes and J. Marsden [1977] Some Applications of Geometry in Continuum Mechanics, Reports on Math. Phys. 12,35-44. D. Husemoller [1975] Fibre Bundles, 2nd ed. Graduate Texts in Mathematics Vol. 20, SpringerVerlag, New York. M.C. Irwin [1980] Smooth Dynamical Systems, Academic Press. Z.R. Iwi 'nski and K.A. Turski [1976] Canonical Theories of systems Interacting Electromagnetically, Letters in Applied and Engineering Sciences, 4,179-191 F. John [1975] Partial Differential Equations, 2nd ed. Applied Mathematical Sciences, Vol 1. Springer-Verlag, New York. L. Karp [1981] On Stokes' theorem for non-compact manifolds, Proc. Am. Math. Soc. 82.487490.
T. Kato [1966] Perturbation Theory for Linear Operators, Springer. (Second edition, 1977). A.N. Kaufman and P.J. Morrison [1982] Algebraic Structure of the Plasma Quasilinear Equations, Phys. Lett., 88, 405-406 R. Kaufmann [1979] A singular map of a cube onto a square, 1. Diff Equations 14,593-594. J. Kelley [1975] General Topology. Graduate Texts in Mathematics 27, Springer-Verlag, New York. W.Klingenberg [1978] Lectures on Closed Geodesics. Grundlehren der Math. Wissenschaften 230, Springer-Verlag, New York. G. Knowles [1981] An Introduction to Applied Optimal Control, Academic Press, New York. S. Kobayashi and K Nomizu [1963] Foundations of Differential Geometry, Wiley, New York. K. Kodaira [1949] Harmonic fields in Riemannian manifolds, Ann. of Math. 50,587-665. B.O. Koopman [1931] Hamiltonian systems and transformations in Hilbert space, Proc. Nat. Acad. Sci. 17,315-318. S. Lang [1972] Differential Manifolds, Addison-Wesley, Reading, Mass. H.B. Lawson [1977] The Qualitative Theory of Foliations, American Mathematical Society CBMS Series, Vol. 27. P.O. Lax [1973] Hyperbolic Systems of Conservative Laws and the Mathematical Theory of Shock Waves, SIAM, CBMS Series, Vol. 11. E. Leonard and K.Sunderesan [1973] A note on smooth Banach spaces, 1. Math. Anal. Appl. 43, 450-454. D. Lewis, J.E. Marsden and T. Ratiu [1986b] The Hamiltonian Structure for Dynamic Free Boundary Problems, Physica, 180, 391-404
References
637
J. Lindenstrauss and L. Tzafriri [1971] On the complemented subspace problem. Israel 1. Math. 9, 263-269. L. Loomis and S. Sternberg [1968] Advanced Calculus, Addison-Wesley. Reading Mass.
D. G. Luenberger [1969] Optimization by Vector Space Methods, John Wiley. New York. G.W. Mackey [1963] Mathematical Foundations of Quantum Mechanics. Addison-Wesley. Reading. Mass. G.W. Mackey [1962] Point realizations of transformation groups. Illinois 1. Math. 6, 327-335. J. Marcinkiewicz and A. Zygrnund [1936] On the differentiability of functions and summability of trigonometric series. Fund. Math. 26, 1-43. J.E. Marsden [1968a] Genralized Hamiltonian mechanics. Arch. Rat. Mech. An. 28,326-362. J.E. Marsden [1968b] Hamiltonian one parameter groups. Arch. Rat. Mech. An. 28,362-396. J.E. Marsden [1972] Darboux's theorem fails for weak symplectic forms. Proc. Am. Math. Soc., 32, 590-592. J.E. Marsden [1973] A proof of the Calder6n extension theorem. Can. Math. Bull. 16, 133-136.
I.E. Marsden [1974a] Elementary Classical Analysis. W.H. Freeman. San Francisco. J.E. Marsden [1974b] Applications of Global Analysis in Mathematical Physics, Publish or Perish. Waltham. Mass. J. E. Marsden [1976] Well-posedness of equations of non-homogeneous perfect fluid. Comm. PDE 1,215-230. J.E. Marsden [1981] Lectures on Geometric Methods in Mathematical Physics, CBMS Vol. 37, SIAM. Philadelphia. J.E. Marsden and T. Hughes [1976] A Short Course in Fluid Mechanics. Publish or Perish J.E. Marsden and T.J.R. Hughes [1983] Mathmematical Foundations of Elasticity. Prentice-Hall. Redwood City. Calif. J.E. Marsden. T. Ratiu and A. Weinstein [1982] Semi-direct products and reduction in mechanics. Trans. Am. Math. Soc. 281, 147-177. J.E. Marsden and A. Tromba [1981] Vector Calculus, W.H. Freeman. San Francisco. J.E. Marsden and A. Weinstein [1974] Reduction of symplectic manifolds with symmetry. Rep. Math. Phys .• 5.121-130 J.E. Marsden and A. Weinstein [1982] The Hamiltonian structure of the Maxwell-Vlasov equations. Physica D. 4,394-406. J.E. Marsden and A. Weinstein [1983] Coadjoint orbits. vortices and Clebsch variables for incompressible fluids. Physica D. 7, 305-323 J.E. Marsden and J.Scheurle [1987] The construction and smoothness of invariant manifolds by the deformation method. SIAM 1. Math. An. (to appear).
638
References
J.L. Martin [1959] Generalized Classical Dynamics and teh 'Classical Analogue' of a Fermi Oscillation, Proc. Roy. Soc., A251, 536. R.H. Martin [1973] Differential equations on closed subsets of a Banch space. Trans. Am. Math. Soc. 179, 339-414. W.S. Massey [1977] Algebraic Topology. An Introduction, Graduate Texts in Math. 56, SpringerVerlag, New York. A. Mayer [1872J Ober unbeschrlinkt integrable Systeme von linearen Differential - gleichungen, Math. Ann. 5,448-470.
J. Milnor [1965] Topology from the Differential Viewpoint, University of Virginia Press, Charlottesville, Va. C. Misner, K. Thome and J.A. Wheeler [1973] Gravitation, W.H. Freeman, San Francisco. C.B. Morrey [1966] Multiple Integrals in the Calculus of Variations, Springer-Verlag, New York. PJ. Morrison [1980] The Maxwell-Vlasov Equations as a Continuous Hamiltonian System, Phys. Lett., 80A, 383-386 P.J. Morrison and J.M. Greene [1980] Noncanonical Hamiltonian Density formulation of Hydrodynamics and Ideal Magnetohydrodyanics, Phys. Rev. Lett., 45, 790-794 J. Moser [1965] On the volume elements on a manifold, Trans. Am. Math. Soc. 120,286-294. M. Nagumo [1942] Ober die Lage der Integralkurven gewohnlicher Differential - gleichungen, Proc. Phys. - Math. Soc. lap. 24,551-559. E. Nelson [1959] Analytic vectors, Ann. Math. 70,572-615. E. Nelson [1967] Tensor Analysis, Princeton University Press, Princeton, N.J. E. Nelson [1969] Topics in Dynamics I: Flows, Priceton University Press, Princeton, N.J. L. Nirenberg [1974] Topics in Nonlinear Analysis. Courant Institute Lecture Notes.
S.P. Novikov [1965] Topology of Foliations, Trans. Moscow Math. Soc. 268-304. G.F. Oster and A.S. Perelson [1973] Systems, circuits and thermodynamics, Israel l. Chem. 11, 445-478, and Arch. Rat. Mech. An. 55,230-274, and 57, 31-98. R. Palais [1954] Definition of the exterior drivative in terms of the Lie derivative, Proc. Am. Math. Soc. 5, 902-908. R. Palais [1963] Morse theory on Hilbert manifolds, Topology 2, 299-340. R. Palais [1965a] Seminar on the the Atiyah - Singer Index Theorem. Princeton University Press, Princeton, N.J. R. Palais [1965b] Lectures on the Differential Topology of Infinite Dimensional Manifolds, Notes by S. Greenfield, Brandeis University. R. Palais [1968] Foundations of Global Nonlinear Analysis. Addison-Wesley, Reading, Mass.
References
639
R. Palais [1969] The Morse lemma on Banach spaces, Bull .Am. Math. Soc. 75,968-971. W. Pauli [1953] On the Hamiltonian Structure of Non-local Field Theories, II Nuovo Cimento, X, 648-667 N.H. Pavel [1984] Differential Equations, Flow Invariance and Applications, Research Notes in Mathematics, Pitman, Boston-London. J-P. Penot [1970] Sur Ie thCoreme de Frobenius, Bull. Math. Soc. France, 98,47-80. A. Pfluger [1957] Theorie der Riemannschen Fliichen, Grundlehren der Math. Wissenshaften 89, Springer-Verlag, New York. A. Povzner [1966] A global existence theorem for a nonlinear system and the defect index of a linear
operator, Trans/. Am. Math. Soc. 51, 189-199. M.M. Rao [1972] Notes on characterizing Hilbert space by smoothness and smoothness of Orlicz spaces, J. Math. Anal. Appl. 37, 228-234. B. Rayleigh [1887] The Theory of Sound, 2 vols., (1945 ed.).Dover, New York. G. Rceb [1952] Sur certains proprietes topologiques des varietes feuilletees, Actual. Sci. Ind. No. 1183. Hermann, Paris. R.M. Redheffer [1972] The theorems of Bony and Brezis on flow-invariant sets, Am. Math. Monthly. 79,740-747. M. Reed, and B. Simon [1974] Methods on Modern Mathematical Physics, Vol. 1: Functional Analysis, Vol. 2: Se/f-adjointness and Fourier Analysis, Academic Press, New York. G. Restrepo [1964] Differentiable norms in Banach spaces, Bull. Am. Math. Soc. 70,413-414. F. Riesz [1944] Sur la theorie ergodique, Comm. Math. Helv. 17,221-239.
J. Robbin [1968] On the existence theorem for differential equations, Proc. Am. Math. Soc. 19, 1005-1006. E. H. Roth [1986] Various Aspects of Degree Theory in Banach Spaces, AMS Surveys and Monographs, 23. H. Royden [1968] Real Analysis, 2nd ed., Macmillan, New York. W.Rudin [1966] Real and Complex Analysis, McGraw - Hill, New York. W.Rudin [1973] Functional Analysis, McGraw - Hill, New York. W.Rudin [1976] Principles of Mathematical Analysis, 3rd ed. McGraw-Hill, New York. D. Russell [1979] Mathematics of Finite Dimensional Control Systems, Theory and Design, Marcel Dekker, New York. A. Sard [1942] The measure of the critical values ofdifferentiabIe maps, Bull. Am. Math. Soc. 48, 883-890. B. Schutz [1980] Geometrical Methods of Mathematical Physics, Cambridge University Press, Cambridge, England.
640
References
J.T. Schwartz [1967] Nonlinear Functional Analysis, Gordon and Breach, New York. J. P. Serre [1965] Lie Algebras and Lie Groups, W. A. Benjamin, Inc. Reading. Mass. I. Singer and J Thorpe [1976] Lecture Notes on Elementary Topology and Geometry, Undergraduate
Texts in Mathematics. Springer-Verlag. New York. S. Smale [1964] Morse theory and a nonlinear generalization of the Dirichlet problem. Ann. Math. 80, 382-396. S. Smale [1967] Differentiable dynamical systems. Bull. Am. Math. Soc. 73,747-817. S. Smale [1965] An infinite-dimensional version of Sard's theorem. Amer. 1. Math. 87, 861-866. J. Smoller [1983] Mathematical Theory of Shock Waves and Reaction Diffusion Equations, Grundlehren der Math. Wissenschaffen 258. Springer-Verlag. New York. S.S. Sobolev [1939] On the theory of hyperbolic partial differential equations, Mat. Sb. 5,71-99. A. Sommerfeld [1964] Thermodynamics and Statistical Mechanics, Lectures on Theoretical Physics. Vol. 5. Academic Press. New York.
J.M. Souriau [1970] Structure des Systemes Dynamiques. Dunod. Paris M. Spivak [1979] Differential Geometry, Vols 1-5. Publish or Perish. Waltham. Mass. E. Stein [1970] Singular Integrals and Differentiability Properties of Functions, Princeton Univ. Press. Princeton. N.J. S. Sternberg [1983] Lectures on Differential Geometry, 2nd ed., Chelsea. New York. S. Sternberg [1969] Celestial Mechanics, Vols. 1.2. Addison-Wesley. Reading, Mass. J.J. Stoker [1950] Nonlinear Vibrations, Wiley. New York.
M. Stone [1932a] Linear transformations in Hilbert space. Am. Math. Soc. Colloq. Publ. 15. M. Stone [1932b] On one-parameter unitary groups in Hilbert space. Ann. of Math. 33, 643-648. K. Sundaresan [1967] Smooth Banach spaces. Math. Ann. 173, 191-199. H. J. Sussmann [1975] A generalization of the closed subgroup theorem to quotients of arbitrary manifolds. 1. Diff Geom. 10, 151-166.
H.J. Sussmann [1977] Existence and uniqueness of minimal realizations of nonlinear systems. Math. Systems Theory 10,263-284. F. Takens [1974] Singularities of vector fields. Publ. Math. IHES. 43,47-100. H.F. Trotter [1958] Approximation of semi-groups of operators. Pacific. 1. Math. 8,887-919. V.T. Tuan and D.O. Ang [1979] A representation theorem for differentiable functions, Proc. Am. Math. Soc. 75, 343-350. Y. Ueda [1980] Explosion of strange attractors exhibited by Duffing's equation. Ann. N.Y. Acad. Sci. 357, 422-434.
References
641
V. S. Varadarajan [1974] Lie Groups, Lie Algebras and Their Representations, Graduate Texts in Math. 102. Springer-Verlag. New York. J. von Neumann [1932] Zur Operatorenmethode in der klassischen Mechanik. Ann. Math. 33.587648.789. C. von Westenholz [1981] Differential Forms in Mathematical Physics. North-Holland. Amsterdam. F. Warner [1983] Foundations of Differentiable Manifolds and Lie Groups. Graduate Texts in Math. 94. Springer-Verlag. New York. A. Weinstein [1969] Symplectic structures on Banach manifolds, Bull. Am. Math. Soc. 75. 804-
807. A. Weinstein [1977] Lectures on Symplectic Manifolds, CBMS Conference Series Vol. 29, American Mathematical Society. J.C. Wells [1971] CLpartitions of unity on non-separable Hilbert space. Bull. Am. Math. Soc. 77, 804-807. J.e. Wells [1973] Differentiable functions on Banach spaces with Lipschitz derivatives. J. Diff. Geometry 8, 135-152. R. Wells [1980] Differential Analysis on Complex Manifolds, 2nd ed .• Graduate Texts in Math. 65. Springer-Verlag. New York. H. Whitney [1935] A function not constant on a connected set of critical points. Duke Math. J. 1, 514-517. H. Whitney [1943a] Differentiability of the remainder term in Taylor's formula. Duke Math. J. 10, 153-158. H. Whitney [1943b] Differentiable even functions. Duke Math.1. 10, 159-160. H. Whitney [1944] The self intersections of a smooth n-manifold in 2n-space. Ann. of Math. 45, 220-246. F. Wu and C.A. Desoer [1972] Global inverse function theorem, IEEE Trans. CT. 19, 199-201.
J.L. Wyatt. L.O. Chua. and G.F. Oster [1978] Nonlinear n-port decomposition via the Laplace operator. IEEE Trans. Circuits Systems 25,741-754. S. Yamamuro [1974] Differential Calculus in Topological Linear Spaces. Springer Lecture Notes, 374. S.T. Yau [1976] Some function theoretic properties of complete Riemannian manifolds and thier applications to geometry, Indiana Math J. 25,659-670. J.A. Yorke [1967] Invariance for ordinary differential equations. Math. Syst. Theory. 1, 353-372.
Index
Accumulation point". 4 Action-anglc variablcs. 570. 575 Adjoint". 109.520-525.541.565 Admissiblc charts. 170 Affinc connectiom. 573 Alexandroff theorcm. 34 Algorithms. 254 Altcration mapping. 393 Amperc's law. 600 Analytic functions. 94 Angular variablcs. 447 Annihilators. 439 Antiderivation. 423. 429. 434 Arzcla-Ascoli theorcm. 27-28. 54 Associatcd tcnsor. 342 Asymptotic stability. 299. 304 Atlas. 151. 170 boundary of. 478 equivalencc of. 142. 143 natural. 162. 173.350 orientation of. 449. 452 See a/so Charts At!ractor". 311. 318 Automorphisms. 289. 516 Autoparallel ficlds. 574
Backward cquations. 284 Bairc theorcm. 37-38. 69. 224 Bakcr-Campbcll-Hausdorff formula. 287. 628 Banach-Schaudcr theorcm. 67 Banach spaces. 40-54. 172. 277 Hilbert spaccs and. 52. 387 isomorphism and. 68. 168. 180.228 locally modelcd. 148 manifolds on. 143 ..226. 364 normed.57 paracompact. 378 partitions of unity. 3H7
scparable. 386 split cxact. 69 Banach-Stcinhaus principlc. 69 Bascs. 172. 199.276.339.405 Base integral curve. 582 Basins. 311. 318 Bernoulli theorcm. 593-594. 597 Bifurcation thcory, 329, 557 Bijection. dcfincd, 14 B irkoff thcorcm, 5 18 Bochner intcgral. 381 Boltzmann equations, 516, 618 Bolzano-Wcicrstrass thcorcm, 25. 35 Borcl space, 516 Born-Infcld brackct. 621 Bot! thcorcm, 333 Boundary. 10,26,541 atlas and, 478 boundary less double, 500 charts and. 478 corncrs and, 489 defined, 476 differentiability across, 540 manifold with, 478 oricntation of. 481, 489 piecewise smooth, 489-495 singular points, 489 volume form. 4 Boundary conditions, 589 Bounded opcrators. 300. 30 I. 316 Brouwer fixed-point theorcm. 538, 548 Buckling column. 309 Bump functions, 273. 281, 290, 492. 506
Calderon extension theorcm, 495 Canonical coordinates, 563 Canonical involution. 166 Canonical projection, 19, 50 Cantor sct. 223
644
Cartan formula, 430, 433, 43~\, 503 Cartesian product, 77 Category theorem, 37-3~L 69, 224 Cauchy-Bochner integral. 61 Cauchy-Green tensor. 356, 367 Cauchy-Riemann equations, X2, 137 Cauchy-Schwartz inequality, 42, 43 Cauchy sequence, 10 Cauchy stress tensor. 5X6 Cayley-Hamilton theorem, 73, 140 Cayley numbers, 192, 195 Cayley transform, 522 Chain(s), 2, 496-49X Chain rule, 84,95, 109,267 Change of variables diffeomorphisms and, 464 formula for, 553, 557, 5X5, 5X7 Lie derivative and, 5X7, 590 partition of unity and, 466 pull-backs and, 465 Chaotic systems, 31X, 319, 516 Chapman-Kolmogorov law, 239 Characteristic exponents, 298-299, 305 Characteristic multipliers, 312, 315, 316 Characteristics, method of. 272, 287 Charge density, 510 Charts, 142, 150 boundary of. 478 bundle structure and, 170 covering by, 143 diffeomorphisms and, 153 non-degeneracy and, 235 oriented, 4X I overlap maps of. 171 smoothness and, 476 topology from, 143 See a/so Atlas Choice, axiom of. I Chow's theorem, 629 Christoffel symbols, 571 Circle bundle, 164, 166 Circulation theorem, 507, 594, 595 Clebsch variables, 598 Closed graph theorem, 520 Closed orbits, 311. 316 Closure, 4, 15,227,437 Cobordism, 50 I Coeycle identity, 40 I Codifferentials, 449-463
Index
Codimension, 51, 151 Cohomology groups, 114,437, 53X, 551 Collars, 500 Collation procedure, 210 Columns, buckling of. 309 Compactness, 24-27, 149, 37X boundedness and, 26 compact support, 377,465,483,491 integrals and, 465 of maps, 235 of solution sets, 303 of sphere, 302 Stokes theorem and, 403 See a/so Paraeompactness Comparison lemmas, 261 Compatibility, 142 Complete flow, 514 Completeness convergence and, 303 dynamics and, 249 of manifolds, 381 of metric space, 16 of vector fields, 579-581 Complex dual space, 57 Complex inner product, 41 Composite mapping theorem, 84, 159, 160, 183, 205 Composition, 14--15,36,344,346 Conjugacy, local. 314 Conjugation map, I X6 Connectedness, 31-36, 234 arewise, 32 fiber bundles and, I XX, 388 manifolds and, 218 paracompaetness and, 144 Conservation of energy, 569, 579, 5X9 Conservation of mass, 469, 585, 5X6 Constant of motion, 516, 569 Constraints, forces of. 624 Continuity, 17, 27 definition of. 15 differentiability and, 83 equation of. 470, 585-5X6, 602 of mappings, 14, 16, 21 uniform, 16, 27 Continuum mechanics, 469, 609-623 Contractions, 13X, 341. 429 class tensors, 342 commutation and, 359
Index
contractible spaces, 189 definition of. II derivation property for, 359 fiber bundles and, 190 homotopy and, 437 loops and, 33 manifolds and, 438 mappings and, II, 139,430 Contravariance, 339, 502 Control theory, 444, 627, 629 Co vector field, 268 Convergence, 5 absolute, 54 Cauchy sequences and, 10 completeness and, 303 countability and, 6 of flows, 263 norms and. 5 strong, 58 of subsequences, 25 uniform, 16 weak. 54 Convergence theorem, 467 Coordinate systems, 142, 156 Cotangent bundle, .,) I. 565 Covariance, 339, 4t,2, 572 Covector field, 351, 359 Covering maps, 155, 215, 452 Critical points non-degenerate, 139, 235 Sard's theorem, on, 220 stability of. 299, 305 Cross product, 427 Curl. 427, 435, 508 Curvature of surface. 107 Curvature tensor, 341
Damping, linear, 309 Darboux theorem. 329. 436, 562 Decomposition, 539 Defect spaces. 524 Deficiency indiccs, 522 Deformation lcmma, 437-438 Degree theory, 309. 543. 545. 556 Density. 105, 409. 456 DeRham's theorem. 437.538.551 Derivations, 156, 271, 274, 292
645
Derivative defined, 75-76 directional. 86, 269 functional, 103 linearity of. 76 partial. 89. 107 propcrties of. 83-1 12 symmetry of. III tangcnt bundles and. 156 time-dependent. 372 total. 78. 108 Derived set. 4 Determinants. 404. 454 Devil's staircasc. 223 Diagonal. definition of. 19 Diameter. of sets. 10 Dielectric constant. 599 Diffeomorphisms. 144. 148. 161 algebra and. 289 of automorphism. 516 change of variablcs and. 464 charts and. 153 defined. 116 differential operators. 360 fiber bundle and. 184 flow and, 267 global. 130-131 glueing and, 50 I homotopy and. 131 interior products. 430 inverse function theorem. 119. 135 lifts of. 565 local. 196 manifolds and, 153 one-parameter group of. 249 orientation preserving. 551 pull-back and. 266. 354 push-forward. 354. 360. 361. 428 rank theorem. 128 singular points. 489 smoothness and, 477 of tangent bundles. 163 on tensor fields. 354 vector bundles and, 173 Diffeotopies. 501 Differential forms. 78. 351. 392-400. 417-422 continuity and. 76. 83 contractions of. 429
646 Differential form.', (cOllI.) defined, 26X differential calculus, 40-54 exterior, 41 X geometry and, 156 ideals, 439-444 manifolds and, 143 mappings and, 156, 354 nondi ffeomorphic, 144 operators and, 426 product rule, 352 topological applications, 53X-55Y vector calculus, 427 Dirac delta function, 59 Directional derivative, X6, 269 Direct sum, 46, 524 Dirichlet integral, lOX Dirichlet problem, 125 Displacement vector field, 367 Distance function, 9 Divergence, 2%, 426, 435, 455, 471, 504, 513 Dot product, 341,427 Double covering, 452 Duality, 5X-59, 66 basis and, 33X bundles and, I X3 generalization of. 103 strong, 103 transformations and, 344 Duffing equation, 319 Dyads, 341
Ehresmann theorem, 206, 3X3 Eigenspace, 307, 317 Eigenvalues, 73, 13Y, 29X, 306, 322, 513 Elasticity, 123,341, 373, 526, 584 Electromagnetic theory, 50X, 5 10, 59960X Elliptic operators, 123, 540 Elliptic splitting theorem, 543 Elliptic theory, 231 Embedding, 20 I, 216 Energy equations, 453, 575, 591,603 Enthalpy, 592 Equilibrium point, 2Y8, 304, 305 Equivalence relations, 19,21, 207, 293
Index
Ergodicity, 470, 513-51X Euler equations, 5X7-591, 596, 615 Euler-Lagrange equation, 106 Euler-Poincare characteristic, 30Y Euler symmetry theorem, 91 Evaluation map, 99 Evolution operator, 239, 2X3 Exponential formulas, 2X7, 29Y Exterior algebra, 392-400 Exterior derivative, 423-44X Exterior forms, 393, 439 Extrema, theory of. 110-112, I3X, 213
Faraday two-form, 600 Faraday law, 349, 507, 50X, 512 Fiber bundles, 167, 172, 176, 191 codimension of. 175 connectivity and, I XX, 3XX contractible space and, IX9, 190 coverings and, 189 derivative and, 574 diffeomorphisms and, 184 fibration and, 149, 206, 383 integral and, 472 isomorphism and, I X5, 349 locally trivial, 185-1 X6 manifold products and, 216 maps and, I X5 morphisms and, I X5 paracompact base and, 3X8 path lifting theorem and, 18X preserved, 175.350 pull-backs and. 189 split fiber exact. 69. 179, I XI Stiefel manifolds. 198 tangent bundles. 206 vector bundles and. 173, 174, IX4-IX5 Final topology, 23 Fixed point theorems. 116,301, 53X, 54X Flag manifolds, 149 Floquet normal form. 31 X Flow(s), 429-430 chaotic, 51 X convergence of. 263 diffeomorphism and. 267 flow boxes, 245, 25X-260, 314 flow fields. 370 fluid mechanics, 5X4-598
Index
gradient. 323 hyperbolic, 322 ideal fluids, 587, 592 incompressible, 540, 590 induced, 513-537 irrotational, 519 Lie bracket and, 286 Lie derivatives and, 371, 480, 567, 584 quasiperiodic, 518 stationary, 593 velocity field of. 584 Foliations, 329-332, 335 Force fields, 238 Forms, 394 basic, 448 compactly supported, 472 distributional. 474 F-valued, 268, 421 identities for, 444-446 integral of. 464-475 symplectic, 562 twisted, 486 volume form, 449 Fourier transforms, 125, 170, 519 Frames, 194 Frcchet structure, 86, 137, 144 Fredholm alternative, 123, 236, 543 Fredholm operator, 226, 231 Friedrich extension, 528 Frobenius theorem, 329-337, 373, 436443, 624, 629 Fubini's theorem, 220,472 Functional dependence, 128 Functional derivatives, 103, 107 Fundamental group, 34 Fundamental theorem of algebra, 548, 550 Fundamcntal theorem of calculus, 80, 124
Giieaux differential, 86 Gauge freedom, 605, 606 Gauss map, 166 Gauss theorem, 392-400, 489, 509, 599 Generators, 529, 536 Genericity, 37 Geodesics, 571, 572
647 Geometies, 149, 156 Germs, 293, 294 Gibbs rule, 516 Global basis, 276 Global section, 177 Glueing theorem, 501 Godement theorem, 209 Gradient, 79, 261, 323, 353, 397,427, 453 Gram-Schmidt procedure, 199, 375, 408 Graph norm, 520 Grassmann algebra, 145, 175,209, 397 Gravitational force, 238, 240, 586 Green's theorem, 392-400, 504 Gronwall's inequality, 243, 250, 257, 296,302 Group property, 15, 34, 239
Hadamard-Levy theorem, 130-131, 134 Hahn-Banach theorem, 66, 81, 91, 103, 274,281,360,433,569 Hahn extension theorem, 467 Hairy ball theorem, 547 Hamiltonian systems, 392-400, 618 equations for, 241, 560, 610 holonomic constraints, 624 infinite dimensional, 570, 610 self-consistent, 618 variational principle, 108 vector fields, 566 Harmonic forms, 538, 540 Harmonic oscillator, 299 Harmonic vector field, 593 Hartman-Grobman theorem, 305, 308 Hausdorff manifolds, 6, 24, 208, 305, 382 Heine-Borel theorem, 26-28 Helmholtz-Hodge decomposition, 613 Helmholtz theorem, 595 Hermes theorem, 629 Hermitian operators, 521 Hermitian product, 41 Hessian matrix, 213 Hilbert space, 513, 514 Banach space and, 52, 387 basis in, 55 defined, 45 intrinsic, 536
648 Hilbert space (conI.) manifolds in, 144,579-581 partitions of unity and, 388 Riemannian manifold in, 353 splits in, 52 Hille- Yosida theorem, 536 Hodge-deRham theory, 538-559 Hodge-Helmholtz decomposition, 611, 616 Hodge star operator, 400, 411, 427, 456, 602 Hodge theorem, 231, 538, 539, 591, 611 Hi)lder class, 245, 260, 387, 495 Holomorphic mappings, 65 Holonomy, 624, 629 Homeomorphism, 14, 15 Homoclinic orbit, 320 Homogeneous polynomials, 63 Homotopies, 34, 549 axiom for, 437 contraction and, 437 defined, 36 invariants, 236, 557 lifting lemma, 131, 133, 155,388 Hopf degree theorem, 559 Hopf fibration, 149, 186, 187 Hurewicz fibration, 388 Hyperbolic flows, 305, 307, 322
Ideals, 156, 439-440 Immersion, 199, 200, 228 Implicit function theorem, 40, 121-129, 136, 164,205,329 Incompressible fluid, 540, 590 Indexing, 308, 383 defined, 408 invariant, 236 lowering, 453 raising, 342 Induced flows, 513-537 Infinite-dimensional theory, 40, 291, 433-440, 570 Infinite products, 18 Initial conditions, 589 Inner products, 41,44, 340,414 Insets, 308, 311, 318 Integral(s) compact support and, 465 defined, 249, 464
Index
di fferentiating under, 94 integrating factors, 442 integration by parts, 93, 457 local complete system, 260 subbundles and, 441 Integral curves, 241, 248, 249, 263 Integral manifold, 326 Interior product, 341, 423-448 Intermediate value theorem, 33 Intersection, transversality and, 203 Invariance, 256, 499 Inverse function theorem, 124-130, 136, 196, 454, 546 Inverse images, 15 Inverse mapping theorem, 116, 118, 119 Inversion maps, 135 Involution, 166, 569 Involutive subbundle, 326, 439, 441 Irrational flow, 520 Isometry, 414 Isomorphisms, 62, 65, 68, 126, 134 Isotopies, 50 I
Jacobian matrix, 78, 130,355,402,449, 454, 471 Jacobi identity, 279,444,609,613 Jacobi-Lie bracket, 277, 278, 328, 573 Jordan canonical form, 299, 307 Jordan curve theorem, 34
Kalman criterion, 630 KdV equation, 570 Kelvin circulation theorem, 507, 594 Kernels, 178, 540 K-forms, 418-421, 433-437, 474, 486 K-multilinear maps, 62 Kneser-Glaeser theorem, 223 Koopman theorem, 513 Kronecker delta, 55, 342, 352
Lagrange multipliers, 108, 138,211-213, 305 Lagrangian systems, 252, 571, 574, 575 Landau symbol, 76 Laplace-Beltrami operator, 459, 538, 605 Laplace-deRham operator, 535, 538, 606 Laplacian, defined, 109
649
Index
Lax-Milgram lemma, 134 Leaves, 329 Lebesque measure, 504 Leibniz rule, 87,95,269,271. 609 Liapunov stability, 299, 302, 305, 316 Lie algebra, 278, 444 Lie bracket, 278, 286, 297, 328 Lie derivative, 278, 359-369, 423-448, 456 change of variables and, 587, 590 cynamic systems and, 370-376, 480, 567, 584 formula for, 269-271, 286, 302, 562 Lie groups, 149, 199,327,336 Lie-Poisson bracket, 609-623 Lie transform method, 329, 373, 436 Lifting lemma, 131 Limit cycle, 311 Limit points, 5, 6, 318 Lindelof space, 224, 232, 233, 386* Linear algebra, 338-350 Linear analysis, 65-69 Linear damping, 309 Linear extension theorem, 59, 61 Linear isometry, 516 Linearization, 124,298,308,317 Linear operators, 231. 300, 301. 520, 536 Linear transversality, 70 Line bundle, 191. 486 Line integral. 504 Liouville theorem, 513, 564, 568, 569 Lipschitz condition, 83, 242-244 Lipschitz constant, 138, 255, 256, 263 Lipschitz map, 138 Lipschitz partitions, 387 Lipschitz theorem, 119, 124, 138 Lobatchevskian geometry, 577 Local basis, 276 Local immersion theorem, 125 Local injectivity theorem, 123 Local onto theorem, 196 Local operators, 274, 423 Local representation theorem, 127, 128 Local representatives, 152, 227, 241 Local slice, 210 Local subjectivity theorem, 122, 126 Local submersion theorem, 126 Localizable algebra, 357 Localizable maps, 357
Locally closed mapping, 227 Loops, 33 Lorentz group, 409, 413, 600, 603
MacLaurin expansion, 319 Magnetic field, 508, 586, 599 Manifolds basic idea of. 40, 141 boundary of. 478 charting of. 143 classification of. 144 connectedness of, 144 covering of. 155 differentiable, 143 diffeomorphism and, 153 flag, 149 flows on, 513 Hodge theorem, 541 integral. 326 integration on, 464-475 locally finite, 377 map of. 152 orientable, 451 partition of unity on, 377 product of. 150-154, 163 pseUdo-Riemannian, 459 quotients and, 207, 209, 437 simply connected, 234 types of. 143-144 topology from, 143 See also Submanifolds; Vector bundles Mappings, 150-154 class of. 158 closure of. 227 continuous, 16 coverings of. 155 degree of. 538, 556 derivations, 271, 292 diffeomorphisms and, 354 differentiability, 156, 158,354 Fredholm types, 227 iterates of. 254 linearization of. 317 locally closed, 167-168,227 on manifolds, 152 overlap, 147 Poincare type, 312 proper, 250 quotient manifolds and, 207
Indcx
650
Mappings «('0111.) skcw symmctric. 393 spcctral. 300 symplcctic. 564 tangcnt. 162 transvcrsal. 337 vcctor bundlc. 175 Matcrial dcrivativc. 5R7 Maxima. See Extrcma Maxwcll c4uations. 392-400. 50R. 599603.610 Maxwcll- Vlasov c4uations. 620 Maycr system. 444 Mayer- Vietoris se4uenccs. 557 Mazur-Ulam theorcm. 72 Mean crgodic thcorem. 516 Mean value incquality. 87-RR Measure theory. 49.220.225.467 Metrics. 9-16. 341. 352. 3R2 Milnor manifolds. 144.216 Minima. See Extrcma. Minkowski metric. 353 Mobius band. 174. 177. 191. 209. 451. 461. 512* Model space. 143 Molecular motion. 516. 589 Momentum balancc. 5R6. 59R Monodromy operator. 323 Monotonicity. 134 Morsc functions. 235 Morse Icmma. 329. 373. 375 Morsc-Palais-Tromba lemma. 374-376 Motion. c4uations of. lOR. 239. 5R7. 615 Multilincar algebra. 338 Multilinear maps. 183. 357 Multipliers. Lagrangc. 211-213
Navicr-Stokcs c4uations. 231 Ncighborhood. of points. 3 Nclson lemma. 513 Neumann problcm. 231. 616 Ncumann scrics. 117 Ncwton's laws. 560 N ijcnhuis tcnsor. 368 Nocthcr thcorcm. 5R2, 604 Nonlincar c4uations. 177, 319. 543 Norm(s) defined. 40
differentiability of. 273. 281 c4uivalcnt. 45.301 fibcrs and. 172 graph norm, 520 inner product and. 44 normal forms. 191.373.504 normal space, 41. 43 spectral radius and. 300 Novikov thcorem, 335
Octavcs. algcbra of. 192 Omcga Icmma. 99, 259 Onc-forms. 268.418.440.444 Opcn mapping theorem. 15,67.228.561 Opcrators. 520 closable. 521. 523 dcnsc range of. 527 elliptic, 524 graph of. 522 monodromy and, 324 norm of. 56 opcrational calculus. 529 self-adjoint. 520-525 symmetric, 524 Orbit analysis. 314-317, 320 Ordercd bases. 406 Orientation. 449-463 atlas and. 449. 452 boundary and. 481 criterion for, 452 infinite dimcnsion and. 449 line bundlcs. 486 on manifolds. 451, 453 prescrving, 414. 453 vector spaccs and. 406 Orthogonality. 42, 199 Orthonormal bases. 55, 199.409.414 Outsct. 307 Ovcrdctcrmined systcms. 443 Ovcrlap maps, 147, 476
Pairing, 103 Palais formula, 434 Paracompactness, 144, 289, 290. 378. 382-385, 453 Parametric transversality theorem, 232 Partial derivatives. 89, 163
Index
Partial differential equations. 271. 443. 495.557 nonlinear. 123-125 systems of. 2X2. 2X3 Partitions of unity. 377-3X9. 450. 466. 492. 554 Patching construction. 377. 382 Path lifting theorem. 155. I X8 Pcndulum equation. 309. 319 Penr.eability. 599 Permutation group. 63. 392-400 Pfaffian systems. 439-444 Phase space. 319. SIX. 561. 570. 609 Plasma physics. 5X4. 609-623 Poincare~Cartan theorem. 499. 500 Poincare duality. 557 Poincare~Hopf theorcm. 30X Poincare lemma. 373. 435-436. 445. 447. SOX. 551. 562. 605 Poincare maps. 312-313 Poincare metric. 575 Poincare recurrence theorem. 470 Poincare section technique. 31 X Point transformations. 565 Pointwise addition. 15 Poisson bracket. 56X. 609. 610. 613 Poisson~ Vlasov equation. 61 X. 619. 621 Polar coordinates. 460 Polarization. 63 Polynomials. 63. XO. 139. 54X Potential function. 302. 305. 510. 572. 597 Power series. 65. 299 Poynting's theorem. 603-604 Product axiom. 2 Product formula. 254. 297. 352. 371 Product space. 18. 24 Projection. 19. 167. 172.292 Projective spaces.' 21. 148 Proper maps. 30. 250 Priifer manifolds. 3X2. 503 Pseudometrics. 10.372. 3X4. 45X. 4X5 Pull~back. 454 change of variables and. 465 covariance and. 354 defined. 265. 345-346 diffeomorphism. 354--355 Hodge operator. 602 integral of. 602
651
invertibility and. 371 symplectic manifolds and. 56X vector bundles and. 193. 194.217. 266 Push~forward. 265. 355. 360. 42X
Quantum mechanics. 526 Quasiperiodic flow. 518 Quaternions. 186. 195 Quotient bundles. In. 209 Quotient groups. 437 Quotient manifolds. 18-22.207
Radial projection. 292 Rank theorem. 127. 129.205 Reeb foliation. 331. 335 Refinement. of covering. 378 Reflexivity. 19 Regularity. 196--197. 207. 220. 225. 233 Relativity theory. 123. 249 Representation theorem. 58-59. 80. 103. 467. 517 Residual subsets. 224 Resolvent. 522 Resonance. 319 Restricted bundles. 177 Rhombics. symmetric. 166 Riemannian connections. 5XX Riemannian geometry. 34. 341 Riemannian manifolds. 252. 353. 561 Riemannian metrics. 352-353. 3XO-382. 391. 453. 491 Riemann integral. 61. 464. 467 Riemann~Lebesque lemma. 170 Riesz theorem. 5X-59. 80. 103,467.517 Ring structure. 351
Sard theorem, 197. 220-22 I. 225. 23 I. 453. 546 Schauder fixed~point theorem. 116 Schriidinger equation. 571 Schwarzchild theorem. 470 Schwarz inequality. 134 Second category set. 37 Second countability. 3. 34. 378-385 Segre embedding. 216
652
Self-adjointness, 232, 514, 520-525 Self-intersection, 200 Separability, 4, 6, 386 Separatrix, 31 I, 318 Sequences, 6, 16, 54, 113 Sesquilinearity, 42 Shock waves, 249, 289, 586 Short exact sequence, 181, 194 Shrinking lemma, 383, 385 Shuffle permutations, 394 Signature, defined, 408 Simplexes, images of, 495 Singularities, 249, 293, 298, 329, 489 Sinks, defined, 322 Skew-adjointness, 513 Skew symmetry, 393 Slices, 334, 336 Smale-Sard theorem, 227, 233, 500 Smale theorem, 221-234 Smoothness assumption of, 223, 495 charts and, 476 diffeomorphisms and, 477 piecewise, 489, 504 Stokes' theorem and, 4S3 theorems on, 241, 473 Sobczyk theorem, 52 Sobolev spaces, 102, 125,231, 260,495 Space average, 518 Spectral mapping theorem, 300 Spectral theory, 298, 300, 301, 306 Sphere, 41, 144, 165 Split exact sequence, 69, lSI, 183 Splitting, 51-52, 123,201, 226, 543 Sprays, 571, 573 Stability of critical points, 299 equilibrium and, 305 of fixed points, 298 index of, 30S Liapunov function, 302 of orbits, 315 of subspaces, 323 Stable manifold theorem, 30S, 317 Standard inner product, 42 State, equation of, 592 Statistical mechanics, 516 Steiner's surface, 215 Step function, 60
Index Stiefel manifold, 198, 199 Stokes' theorem, 392-400, 476-506, 595 Stone's theorem, 513, 529, 571 Straightening-out theorem, 247, 26(), 317 Strain tensor, 367 Stratifications, 336 Stream functions, 592, 597-598 Stress tensor, 34 I, 586, 604 String, model for, 109,311 Subbundles, 178, 192, 194,326,439 Subimmersions, 205, 228, 229 Submanifolds, 196 bases, 172 defined, 150 foliation and, 329-330 immersed, 201 open, 150-154 transversals, 20 I vector bundles and, 177 See a/so Manifolds Submersions, 196-197,203,205,229, 383 Subsequences, 25 Subspaces, 18-22, 323 Summation convention, 105,338 Support, of vector fields, 249 Surfaces integral on, 505 Stokes theorem for, 479 tangent vector, 156 theory of, 107 See a/so Manifolds Suspensions, 285 Symmetric operators, 522, 524, 527-528 Symmetric rhombics, 166 Symmetric tensors, 340 Symmetrization operator, 71, 96-97, 135 Symmetry theorems, 19,91,99 Symplectic forms, 149,214,561-567
Tangent, to curves, 83, 158 Tangent bundles, 156-165, 173,332, 352,513 Taylor's theorem, 93, 97, 99, 293 Tensor algebra, 357 Tensor bundles, 59, 349-358, 513 Tensor derivation, 359, 364, 371
653
Index
Tensor fields. 351 carrier of. 377 contractions of. 429 coordinate expression of. 352 differential k-forms, 392-400 mappings, 354 partitions of unity, 377-389 symmetrie. 340, 352 tensor bundles. 349-358 time-dependent. 372 Te.1sorial construction, 182 Tensoriality criterion, 352 Tensor product, 339, 344 Thickening, of sets, 92 Thorn's theorem, 50 I Tietze extension theorem, 379-380 Time-dependent flow, 239, 283 Time-ordered products, 297 Toda lattice, 5S1 Topological index. 30S Topological spaces, 2-7, 9, 18,23 Torus, 20-21. 152, 165,518 Trajectory analysis, 593 Transitivity, 19 Transport theorem, 469. 471. 585, 590-
593 Transposition. 63. 346, 392-400 Transversals, 70, 201. 203. 232.312,
337 Triangle inequality. 41 Trivial bundle. 453 Trivial topology, 2, 10 Trotter product formulas. 287
Unbounded operators, 287, 529 Uncountable spaces, 38 Uniform boundedness principlc, 6'1, 16'1 Uniform continuity. 16, 27 Uniform contraction principle, 138 Uniform convergence, 16. 28 Uniqueness theorem, 246 Unitary group. 214, 529, 533 Unity. partitions of. 377-38'1.450.466.
492. 554 Universal bundles. See Grassmann bundles. Universal covering manifold. 155 Upper bound, defined, 2 Urysohn's lemma, 37'1
Vandermonde determinant. 98 Van der Pol equation. 312 Vanishing theorem, 333 Variational calculus, 103 Variation equation, 580 Variation of constants formula, 261,301.
628 Vector bundles atlas for. 170, 182,417 base of. 172 basic properties of. 171-172 bundle maps, 175-177, 179, 183 constructions with, ISO-181 diffeomorphisms and. 173 fiber bundles and, 172, 173, 175, 184-
185 Grassman bundles, 175 isomorphism and, 17 5, 191 local. 167. 173 orientation in, 461 product bundles, 181 pull-back bundles, 193 quotient and, 178 short exact sequenees, 183 stable, 191 submanifolds and, 177 tangent bundle and, 156-165, 173 tensor algebra and, 182, 338, 349 trivial. 174 Vector calculus, 392-400, 426. 427,
435, 584 Vector fields. 40, 167. 266, 351 autoparallel. 573 basis for, 276 characteristic for. 287 commutation of. 282 completeness of. 248. 579-581 convexity in, 72 defined, 238 derivations and, 293 differential operators, 265-297 divergence-free, 513 identities for, 444-446 Killing fields, 366 Lagrangian for, 575 Lie derivative and, 272, 277-278 line integral and, 505 push-forward of. 268 support for, 249
654
Vector fields (COni.) time-dependent. 2X3, 2X5, 297, 371 vertical, 44S Vector potentiaL 392-400. 405, 604 Vector space. norm in, 46 Velocity vector. 156 Viscous fluids, 592 Voltage equations, SOX Volume clements, 405-409. 449-463, 4XI Von Neumann theorem, 516, 526, 52X Vorticity equation, SOX. 594--596
Index
Wave equation, 109,570.606 Wedge products. 394, 397-39X Weierstrass theorem, 54X Well-ordered sets, 2, 3X3 Whitney embedding theorem, 177 Whitney sum, IX3. 191 Winding number. 309
Zermelo theorem, 2 Zero measure, 225 Zero section, 167, 171. 172 Zorn theorem, 2
Applied Mathematical Sciences co11l. {roln pa~e ii
55. 'Ihida: Operational Calculu>: A Theory of Hyperfunctiom. 56. Chan!!,/Howes: Nonlinear Singular Perturbation Phenomena: Theory and Applications. 57. Reinhardt: Analysis of Approximation Methods Itlf Differential and Integral Equations. 5X. Dwo\'crlHussainilVoigl (ed\.): Theoretical Approaches to Turbulence. 59. SmuienIVerhu/s/: Averaging Mcthod~ in Nonlinear Dynamical Sy-"'tcm~ 60. GhitlChildress: Topics in Geophysical Dynamics: Atmospheric Dynamics. Dynamo Theory and Climate Dynamics. 61. 5,'attingerIWeaw'r: Lie Groups and Algcbra~ with Application\ to Phy~ic~. Geometry. and Mechanics. LaSalle: The Stability and Control of Discrete Processes. Grasman: Asymptotic Methods of Relaxation Oscillations and Applications. Hsu: Cell-to-Cell Mapping: A Method of Glohal Analysis It)r Nonlinear Systems. RalldIArmhrU.Her: Perturbation Methods. Bifurcation Theory and Computer Algebra.
62. 63. 64. 65. 66. 67. 68, 69. 70.
71. 72. 73. 74. 75. 76.
Hla\'(ilekIHllslin?erINelasILO\'f.~ek: Solution of Variationallnc4uali~ics in Mechanics.
Cl>rci!!,llani: The Boltzmann Equation and It~ Applications.
Temam: Infinite Dimensional Dynamical Sy~tcm in Mechanic~ and Phy~ics. GoluhitskvIStewartISchae/jer: Singularities and Groups in Bifurcation Theory. Vol. II. ConstantinIFoiasINico/aenkoITemam: Integral Manifolds and Inertial Manifolds for Dissipative Partial Differential Equations. Carlin: Estimation. Control. and the Discrete Kalman Filter. LochakIMeunil>r: Multipha~e Averaging for Classical Systems. WiRRim: Global Bifurcations and Chaos. MawhinIWil/em: Critical Point Theory and Hamiltonian System,. AhrahamlMarsdenlRmiu: Manifolds. Tensor Analysis. and Applications. 2nd cd. Lagerstrom: Matched Asymptotic Expansions: Ideas and Techniques.