[lecture notes in mathematics] proceedings of liverpool singularities — symposium i volume 192 ||...
TRANSCRIPT
A PROOF OF THE ~K~LGRAN,GE, ~ P ~ T I O N THEOREM
L. Nirenb erg
Malgrange's preparation theorem [3], [4] is an extension to real C ~ functions
of the Weirstrass preparation theorem. In this paper we give a short, rather
direct proof of the theorem which applies also to complex functions: t
Theorem 1. Let f(t, x) be a C ~ complex-valued function defined in a
nei~hourhood of the ori6in in R n+l, here t E R I, x E R n, and let p > 0 be the
first integer such that
(~)P f(O, 0) / 0 . (i)
Then in a nei~hbourhood of the ori6in one has the fact orization
(2)
where
(3)
and q
If f
f(t,x) : Q(t, x)P ,
P P = t p + ~ kj(x)t p-j
i
and ~j are ~ complex-valued functions with Q(0, 0) / 0.
is real there is such a factorization with Q and P real .
This paper is a slightly expanded version of a letter written to B. Ma!grange in
November 1969. The work was supported by the U.S. Air Force Contract AF-49(638), 1719. Reproduction in whole or in part is permitted for any purpose of the U.S. government.
~Nalgrange informed me that the complex case can be derived from the results in [4]. In addition, G.I. Eskin, using the results of [4], observed independently that for each positive integer k the complex function f admits the factori~ation (2) with Q and ~i of class C k . See a&so the papers in this volume by S. Lojasiewicz ~and G. Glaeser.
- 98 -
af (0, O) # 0 then one For p = i the theorem states that if f(O, O) = 0 ,
has the C °o factorization
(4) f(t, x) = Q(t, x)(t + kl(X)) , Q(0, 0) # 0 .
If f is real this follows easily from the implicit function theorem, and the
factorization is then unique. However for complex f this simple looking result
seems to be nontrivial. Furthermore, the factorization is not unique, for the
function f t + i(x + e -l/x2 = ) , here n = 1 , admits the factorization f = 1 .f
and f = Q(t, x)(t + ix) - one easily sees that Q belongs to C ~ , and
Q(o, o) : i .
The preparation theorem is based on the Maigrange division theorem which we
state for complex functions :
C ~ , ,,,,,, ...... ,,,,,,, ,, Theorem 2. ~ Let f(t, x) be a complex-valued function in a nei~hbour-
hood of the orisin in R n+l and let
tP-J x : (h ..... ~) (5) P(t, A) = t p + Aj ,
1
be a generic monic polynomial in t with constant complex coefficients
in a nei~hbourhood of the origin one has division by P(t, %) :
(6) f(t, x) : q(t, x, ~) P(t, %) + r(t, x, %) ,
%. . Then
where q and r are complex-vaiued C ~ functions in a n eighbourhood of the origin
in R n+l × ~P , an d r is a polynomial in t of desTee less than p . A~ain if
f and the ~. are real one can obtain this with q and r real. J
tThe functions q and r are not unique. In the real case a stronger form of Theorem 2 was proved by J. Mather [5]. He showed that for every f one can find q and r in (6) in such a way as to depend linearly on f . In addition he established a global form of the theorem. Our proof does not yield q and r depending linearly on f . In a paper in this volume Mather shows how to modify
our proof (in particular he presents an improved version of our Lemma l) so as to obtain q and r depending linearly on f .
- 9 9 -
Proof of Theorem ! : This follows [3] • Let
consider the division (6) by the generic polynomial
remainder r has the form
r = rj(x, A) t p-j
1
We shall solve the p complex equations
(7) rj(x,x) = o, j =l ..... p
for k = k(x) , regarding this system as 2p real equations for
Re lj , Im Aj , j = 1 ,..., p .
f(t, x) =
f be as in Theorem I and
P( t , ~) of (5) . ~he
2p real unknown
We then obtain the desired factorization
q(t , x, X(x)) P( t , X(x)) .
From the definition of p we see that rj(O, 0) = O, j = 1,..., p and
q(O,0,0) / 0 hence to solve (7) it suffices to show that the Jacobian matrix
a(Re r j , Im rj) is nonsingular at x = O , ~ = 0 . In fact we shall show that
a(~e ~k' ~m ~k) Or. Or.
the (complex) p xp matrix --/ (0, O) is nonsingular while 0 (0,0) = 0 ,
from which the desired result follows easily.
a 1 (__a a a~ 2 O~k+ i~vk ) , for ~ = ~k + irk .
Here a~ = 1 (10 . i a 2 % )'
Applying a/a~ to (6) we find
@r~ (0,0) t p-j = - t p 8
j = l aXk aik
for all t , from which it follows that
(6) we find
q(t, 0, O) , k = i,..., p ,
~r. (0, O)= 0 . Applying a//O~ to
q(t, O, 0)t p-k + t p 0 ~ ar.
OZ]< q(t , O, O) + a2~ (0,0) t p-j = 0 j = l
k = i,..., p .
Suppose that for some complex vector ~ = (~i'''" ~p) '
- i00 -
at. c00) k
k
= 0 , j = i,..., p ;
then we find
q(t,o, o) ~t p-k ~k:°(tP)' as t*O. 1
Since q(0,O,O) ~ 0 this implies that ~ tP-k ~k ~ 0 for all
1
~ 0 . Hence arj/0~ k (O, O) is nonsingular. Q.e.d.
t which means
Our proof of Theorem 2 is an adaptation of one of the well known proofs of
the Weierstrass Division Theorem (Theorem 2 in the analytic case), which we repeat
here. If f is analytic we have, by Cauchy's integral theorem
(8) f ( t , x ) = 2~il ~ f(~,x),~_t d~ aD
where D is a disc about the origin in the complex z plane containing t , and
all the roots of P(z, ~t) for [%[ small, in its interior. The function
P(~, x) - P(t, x) R(t, z, ~) =
z - t
is analytic, and a polynomial in t and z of degree less than p . Replacing
1/(z - t) in the proceding integral by
(9) l P( t , X) R(t, z, X) z-t = (z-t) P(z,A) + P(z,k) '
we find
f(t,x) = P(t,~) i f(z,x) l f (s- t )P(z ,X) ds + 2~i f R(t,
aD aD
z, ~) dz .
This is the desired form (6) with
- I01 -
q (, - t) P(~, ~)
aD aD
This proof was also used by Mather in [5] •
We shall make use of the following extension lemma.
Lemma i. t Let f(t, x) beta C °o complex-valued function in a nei~hbourhood
• C °° , of the origin in R n+l There exists a complex function F(z, x, %) defined
in a nei~hbourhood of the origi n for z ~ ~I, x e R n ~P . , ~ E , satisfying
(i) F(t, x, %) = f(t, x) for t real
(ii) F z vanishes of infinite order on Im z = 0 and on P(z,2~) = 0.
Proof of Theorem 2 : We wish to apply the analogue of (8) to our extension
F . This is the well known generalization of the Cauchy integral theorem for a
smooth , but non-holomorphic function of a complexvariable z (see for instance
Theorem 1.2.1. in [i] ):
f(t,x):F(t,x,~) : 1 ~ F<z,x,~l dz+ 1 ~j dz dz ( ) ^ 2~i z-t ~ F z z,x,% z - t
aD D
where the area of integration is a disc D about the origin containing in its
interior the point t and all roots z of P(z, ~) for IA I small. Substituting
(9) for 1/(z- t) in these integrals we obtain (6) with :
1 r F(z,x,%) 1 r r F~(z,x,A) dz + dz dz q 2~i J (~-t~P(z,~) ~ JJ (z-t~P(z,~ ^
~D, D
r - 2~i f F(z,x,%) P(z,k) dz + ~'~ If Fz (z,x,%) R(t.z,2~)p(z,;t) dz ^ d~
~D D
In his paper in this issue Mather proves a global version of this and obtains F depending linearly on f ; the F constructed here depends on f in a nonlinear way.
- 102 -
The denominators in the double integrals are zero when z - t or P(z, ~) vanishes
- but the numerators vanish of infinite order there, by Lemma 1. Hence those
integrals converge absolutely. Furthermore one may differentiate under the integral
signs with respect to t, x, and k ; the resulting integrals are all absolutely
convergent. Thus q and r are of class C ~ and Theorem 2 is proved for the
complex case. In case f is real we may use ~F(z,x,X) + F(~,x,k)) in place of
F in the preceding; it is then readily verified that q and r are real for
real.
We remark that H~rmander [2], in quite a different connection, has used the
device of extending smooth functions of real variables to complex variables as
functions satisfying the Cauchy-Riemann equations to infinite order on the real
subvariety.
In proving Lemaa 1 we shall make use of a special, simple case of the Whitney
extension theorem for which we include a proof :
Lemma 2. Let u(y), v(y) be C ~ complex-valued functions in a nei~hbourhood
of the origin in R N
t~J)
D°(u-v)
Then there is a C ~
D~(F - u) : o
~=(F - v) : o
sat%s.f~in g (here D ~ represents any derivative of order
= 0 o~n Yl = "'" = Yk = 0 for all
function F(y) satisfying
on Yl = "'" = Yj = 0 forall ~ ,
o_n Yj+l .... = Yk = 0 for all
proof : The function
where the sum is over
where ~(s) is a
for s > l, and where
The choice of the t. J
(z .
D~(u-v)(O ..... O,yj+I,---,Y N) • ¢(ti~ I
J
1
: (~i''''' ~j' 0,..., 0) , has the desired property,
function on s ~> 0 which is i on 0 ~ s ~< 1/2 and zero
t..~ co is a rapidly increasing sequence tending to infinity.
depends on the functions u and v , and the resulting F
- I03 -
therefore does not@epend linearly on u, v .
Proof of Lemma 1 :
it has been proved for
variable in (z, ~) space to straighten out the variety P(z, l) =
l' = (h i ..... kp_ I) alone and introduce a new variable
= P(z, ~) .
Then in (z, l' , N) space the variety P = 0 becomes
@/aT now becomes , in the new coordinates,
(io) L a + ~ _t a~ a~
P' represents the derivative of P where
independent of A P
We shall construct two
Induction on p ; Lemma I is true for p = 0 . Assuming
p-i we wish to prove it for p . First we make a change of
0 . Leave z and
to replace ;t : P
= 0 , and the operator
with respect to z - note that P' is
C ~ functions v(z, x, A') and u(z, x, A', ~) such
that
(ii)
(12)
and
(i3)
(l~)
(15)
v(t, X, ~') = f(t, X) for t real ,'
V vanishes of infinite order on Im z = 0 Z
u=v on fl = 0
Lu vanishes of infinite order on ~ = 0
D~(u - v) = 0 on Im z = Re ~ = Im ~ = 0 for all ~ .
Using Lemma 2 we can then put these functions together to obtain the desired
function F(z, x, ~) .
The induction hypothesis is used to construct v , namely v(z, x, ~')
C ~ function satisfying (ll), (12) and a/so satisfying :
(16) v_ vanishes of infinite order on Z
P,(z,~'):p2 -l+(p-l)~s p-2+...÷~p_l = o.
is a
Then we set
- l O ~ -
l a )J v(z,X') ~ ¢(1~12 tj) (- P~ a~
u
0
where
for
Next
t./~ oo is rapidly increasing (depending on the function v) . J
Because v vanishes of infinite order on the set where P' = 0 one sees that Z
tj rapidly increasing the function u belongs to C °O and satisfies (13).
Lu = ~ ( 1 a + !)u
p--r a~ a~
= p' ( 1 a )j+l v(., x,) £ [~(1.t 2 tj÷ l) - ¢(l~12tj)] P-~ a~ J~ 0
_l ~ )J v(z, k') ~j ~'(i~i2tj + P' ( - p-T a~ j'T t j . ) ;
0
and one sees that Lu vanishes of infinite order on
to verify (15) we see from (12) that
u - v (s , x ' ) ~(I.12to )
= 0 , i.e. (14) holds. Final]z,
vanishes of infinite order on Imz : 0 . Since v(z,A')(~(I~I2t0' ' -
infinite order on ~ = O , w@ infer that (15) holds.
!) vanishes of
REFERZNC E S
[1] L. H~rmander,
[2] L. H~rmander,
[3] B. Malgrange,
[4] B. ~algrange,
An Introduction to Complex Analysis in Several Variables.
D. Van Nostrand Co., Inc., Princeton, 1966
On the singularities of solutions of partial differenti-
-al equations. Conf. on Functional Analysis and Related
Topics, Tokyo, 1969, 31-40.
The preparation theorem for differentiable functions.
Differential Analysis, Oxford Univ. Press, 1964, 203-208.
Ideals of differentiable functions, Oxford Univ.Press,
1966.