Topo.bgy, Vol. 8, pp. 167-171. Pergamon Press, 1969.

Printed in Great Britain

ON CO-SUFFICIENCY OF JETS OF POTENTIAL FUNCTIONS

TZEE-CHAR Kuo

(Received 17June 1968)

0 1. THE RESULTS

LET V : R” + R be a polynomial of degree r with V(0) = 0. We say V is a Co-sufficient r-jet

(of a potential function) if the following holds. For any C*-function P : R” + R, where
q 2 r + 1, whose partial derivatives of order 6 r and P(0) vanish at 0, there exists a local
homeomorphism g with g(0) = 0 such that the diagram

R”

I
\;

R" -'R
Vi-P
ll

is (locally) commutative.
In this paper, we shall give a criterion for CO-sufficiency of r-jets. We shall also need an
improvement of the existence theorem of ordinary differential equations. This is our
Theorem 2 in $4. Mather ([2]) has solved a related problem about C “-sufficiency of map-
pings f: R” -+ RP. Let us write V(X) = H,(X) + * * * + H,.(x), where Hi(X) is a homogeneous

form of degree i. Recall that grad Hi(x) = 2, . . . ,z .

1 n
THEOREM1. I’ there exists E > 0 and k 6 r such that lgrad H,(x) + * . . + grad W,(x)] 3
lxlk-‘s for all x in a neighbourhood of 0, then V = HI + * . - + H, is a CO-suficient r-jet.
A proof is given in $2.
COROLLARY 1. If Hi = 0 for i 6 k - 1 and Hk = 0 is a non-singular projective variety,
i.e. grad Hk(x) = 0 only when x = 0, then V = Hk is a Co-sufJicient k-jet.
When k = 2, this is the well-known Morse Lemma ([3]).
One might conjecture that if H, = 0 is non-singular, then V = HI + . . . + H, is always
CO-sufficient. This is false. An example (due to Thorn?) is as follows. Let V(x, y) =
x2 + 2x(x2 - y’). Then V + (x2 - y2)2 = [x + (x2 - y2)12 is a complete square. Hence
V is not sufficient.
It turns out that if the varieties Hi = 0, i = 1, . . . , r intersect transversally in some
sense, then V is CO-sufficient. This is our Corollary 2 below.
167
168 TZEE-CHAR KU0

Let each projective variety Hi = 0 be stratified regularly with respect to tangent

planes in the sense of [6]. A stratification of a given variety W is a decomposition of
W=M,v a** u M, into a disjoint union of manifolds M, satisfying certain conditions
([4]), [S], [6]). Each Mt is called a stratum.
A family {S,} of submanifolds of a manifold M is said to intersect transversally if at
any x E M the sum of normal spaces N,(x) of S, is actually a direct sum. It is understood
that N,(x) = 0 if x $ S, .
COROLLARY 2. Suppose the projective varieties Hi = 0, i 6 r have no common singularity.
Also, suppose the following holds. For any point ii of the real projective space RP”-I, let
Hj = 0 be the first variety such that either Hj(ii) 4 0 or U is a regular point of Hj = 0,
then the stratum, tf any, of Hj = 0 containing ii together with all strata of Hi = 0 (containing
ii), i < j, form a family of transversally intersecting mantfolds. Then the hypothesis of Theorem
1 is satisfied and V is C o-su#icient.

We give the proof in $3.

$2. THE PROOF OF THEOREM 1

If HI $0,
then V(x) = x1 for a suitable basis. Then V is obviously sufficient. In the
following, we assume HI = 0.
Let P(x) be a given Cq-function (43 r + 1) with lim,,, lP(x)l/lxl” = 0. Let F(x, a) =
V(x) + aP(x) where a E R. For all a, grad F = (aF/ax,, . . . , aF/aa) = 0 when x = 0. Hence
F = 0 has (0, a) as singularities.

LEMMA1. There exists 6 > 0 such that Igrad FI 2 z Ixlk-’ for 0 6 1x1 < 6, a E [0, 11,

where E was given in the hypothesis of Theorem 1.

Proof. For a E [0, 11, x =+ 0,
[grad FI 3 lgrad V(x) + a grad w>I
3 lgrad VI - Igrad’Pl

3 Ixlk-’ &--
IgradPI
i Ixlk-’

2 1x1”-’ i for 1x1small. q.e.d.

Let X(x, a) be the projection of the vector (0, 1) E R” x R to the direction of grad Fat
(x, a), 0 < 1x1< 6. A simple computation yields X(x, a) = lgrad PI-‘P(x) grad F.
LEMMA 2. Let v(x, a) = (0, 1) - X(x, a) for 0 < 1x1< 6, and let ~(0, a) = (0, 1). Then
(i) v(x, a) is Co for 0 6 1x1< 6, Cq-’ for 0 < 1x1< 6.
(ii) lj2 Iv(x, a) - v(0, a)l/lxl = 0 uniformly for a E [0, 11.

(iii) for x =I=0, v(x, a) is tangent to the level surface F = C passing through (x, a).
(iv) there exists an n > 0 so that for 0 6 1x1 < n, the inner products (v(x, a), (0, 1)) > 0.
 ON C”-sm~c~~~cY 0~ JETS OF POTENTIAL FUNCTIONS 169

Proof. By Lemma 1, grad F + 0 for 0 < 1x1< 6, hence ZJ(X,a) is C4- ’ for 0 c 1x1< 6.
Now,
limoix/-‘10(x, a) - u(0, a)1 =,“nJ 1x/- ‘1X(x, a)1

=~~~x~-l~gradF(-ljP(x)($~~o~-‘~~~-k~P(x),=O.
--t
This proves (i) and (ii). The construction of v(x, a) implies (iii). To prove (iv), we observe
that
(0(x, a), (0, 1)) = 1 - [grad PI-‘P(x)” 3 1 - 4s-2)xl-2k+2P(x)2 > 0 for 1x1 small. q.e.d.
We shall find that properties (i) (ii) of v(x, a) are stronger than the hypothesis of Theor-
em 2 of $4. The conclusions are that local solutions of 2)(x,a) (exist and) are unique, and
that the terminal value is a continuous function of the initial value and time. Granting this,
let cp(t; x0, ao) denote the solution with ~(0; x0, ao) = (x0, ao). The obvious solution
~(t; 0,O) = (0, t) satisfies the initial condition q(O; 0,O) = (0,O). For all x in the neighbour-
hood of 0, the a-component of any solution &t; x, 0) increases monotonicahy with t, by(iv)
of Lemma 2. Hence rp(t; x, 0) meets the hyperplane a = 1 at a unique point g(x). The map-
ping x --t g(x) is a local homeomorphism. By (iii) of Lemma 2, F is constant along each 40.
Hence
I’(x) = t;(x, 0) = P@(x), 1) = J’@(x)) + P(g(x)),
proving Theorem 1.
Corollary 1 follows immediately. If Sk = 0 is non-singular, /grad H&)1 is bounded
away from 0 for 1241
= 1.

!j 3. THE PROOF OF COROLLARY 2

Let u E R”, lul = 1, be a representative of U. It suffices to show that for all u in a neigh-
bourhood of U, and all p > 0, there exists E > 0 such that

g+-‘lgrad H,(u) 2 E

Suppose this were not the case. There would exist a sequence (p,, u,>, ju,,l = 1 such that
(i) lim (P,, un) = (0, 4,
(ii) lim / j$lp;l+i grad Hi(un) = 0, and
n+ul
(iii) for each i gj, either grad Hi&,) = 0 for all n or limn+, Igrad Hi(U, gradIY&)
exists.
Let C be a regular point of Hj = 0. For each i g’j, grad Hi(UJ approaches a (possibIy
zero) normal vector to the stratum of Hi = 0 at U.This is by the regularity of the stratifica-
tion. These normal vectors lie in different summands of a direct sum, because the family of
strata intersect transversally. It follows that grad Hi(U) + 0 and ID= 1p- j+'grad H,(u,Jl + 0
would yield a contradiction.
Finally, if Hi(~) + 0, then grad Hi(U) is not perpendicular to the radius vector u in
R” (Euler’s Theorem). A similar argument completes the proof.
170 TZEE-CHAR KU0

0 4. THE EXISTENCE THEOREM

Let X(x) be a vector field defined and continuous in an open subset U of R”. It is well-
known that at least one local solution exists through each point of U. Let cp(t, uO) be a
solution defined for 1t 1< y with ~(0, uO) = u,, . We say X(x) satisfies the Lipschitz condition
along cp with Lipschitz constant K = K(q) if for all x in a neighbourhood of the point set
{q(t, ~0) : It I < ~1, I%4 - JJdt, u,Nl < K Ix - dt> GJI.
This condition is certainly weaker than the usual Lipschitz condition.
THEOREM 2. Let X(x) be a vector jield dejined and continuous in an open subset U of
R”. Suppose for each u0 E U, there is a solution cp(t, IA,,) with ~(0, u,,) = u0 along which X
satisjies the Lipschitz condition. Then X admits a unique solution through each point of U,
i.e. two solutions satisfying the same initial condition are identical on their common domain of
dejinition.
Moreover, the solution q(t, u,,), defined on its maximal interval of existence, is a con-
tinuous function oft and uO.
We might exploit the differentiability of cp in another paper.
Proof. Uniqueness. It suffices to show the following. Let q(t) be a solution along which
X satisfies the Lipschitz condition with a constant K. And let t(t) be any solution with
t(O) =I=q(O). Then t(t) 4 q(t) for all t $0.
Define a “ Liapounov function” V by
V(x, r) = ezKt Ix - (p(t)l’.
For all x, t, V(x, t) 3 0. And V(x, t) = 0 if and only if x = q(t). Moreover, letting a =
5(t) - V(t)>

But by assumption, - (da/dt I = Kit(t)

K 1~x1 - q(t)1 - 1X(&t)) - X(cp(t))( > 0 whenever
c(t), q(t) are both defined and It(t) - q(t)] is small. Thus V(t(t), t) can never be zero,
proving the uniqueness.
Continuity. A general theorem asserts that continuity follows from uniqueness (Cl].
Theorem 2.1, p. 94). We shall give a simple proof for our case, again based on Liapunov’s
idea.
Let cp(t, uo) be a solution defined at least for 0 6 t 6 T. It suffices to show that for all
u1 near u,, , cp(t, ul) is also defined at least for 0 g t 4 T and that the map (t, q) + q(t, ul) is
continuous at (T, uO).
Assume first that X satisfies the Lipschitz condition along the whole of cp(t, uo) with
Lipschitz constant K.
Let W(X, t) = e-2Kt Ix - q(t, u0)12.
 ON ~~~~~~~~~~~~ OF JETS OF POTENTIAL FUNCTIONS 171

Again, we have W(x, t) 3 0, and W(x, t) = 0 only if x = p(t).

For 6 > 0, let
N, = {(x, t) : t E C-6, T+ S], W(x, t) 6 S}.
The Na’s form a basis for the neighbourhood system of the point set {cp(t, uo) : t E [0, T]}.
Choose z so small that for (x, t) E N,,

/Xx) - X(cp(6 uo)>l < K Ix - V(6 uo)l*

We claim that if t(t) is any solution with (r(O), 0) E N,, then (c(t), t) E N, for all
I E [0, T]. In fact, a simple computation yields

$ WtXO, 0 6 2epZK’
IalP,
where CI = 5(t) - &I, P = Ix(&)) - Xcp(t))l - Kl50) - Cal.
If (t(b), b) E N,, then p < 0 and d Wjdt < 0 at t = b. Hence for some E > 0 and all
t E [b, b + E],
w(W), t) < WC(b), b), (5(t), t) E Nz.
Thus if (c(O), 0) E N,, (t(t), t) E N, for all t E [0, T].
This proves that (t, ur) + &t, ur) is continuous in U, at u. .
That q(t, ui) is continuous in both t and u1 follows from the following observation. If
X(x) is continuous, N a given neighbourhood of a given point x0. Then there exists an
E > 0 and a neighbourhood M of x0 such that for any solution 5(t) with 5(O) E A4, T(t) EN
for all ItI < E.
Now the general case. The time interval [0, T] can be divided into [To = 0, T,,
. ..) T, = T] in such a way that X(x) satisfies the Lipschitz condition along cp(t, uo) for
t E [Ti, Ti+,], for each i. Then cp(t, uo) is continuous at (Ti+1 - Ti, uo). Since X(x) is
independent of t (an autonomous system),
cp(T, uo) = cp(T, - T,, q(T,, ~0)) = .a.
= cp(T, - T,-,, cp(T,-, - K-2 7 . . .,)
Thus cp,(T, uo) is a composition of s continuous mappings. q.e.d.

Acknowledgements-The author wishes to thank M. G. Barratt, D. H. Fowler, J. N. Mather, W. Shih,

R. Thorn and C. Zeeman for many helpful discussions and communications.

REFERENCES
1. P. HARTMAN:Ordinary Differential Equations, Wiley, 1964.
2. J. N. MATHER:Stability of C” mappings III, Publ. Z.H.E.S. (to appear).
3. J. MILNOR: Morse theory, Ann. Math. Stud. No. 51 (1963).
4. R. THOM: Local topological properties of differentiable mappings, Difirential Analysis, Bombay Colloq.
1964.
5. R. THOM: La stabilite topologique des applications polynomiales, L’Enseignement Math. 1962.
6. H.WHITNEY: Local properties of analytic varieties, Differential and Combinatorial Topology, Princeton,
1965.

University of Manchester