Anda di halaman 1dari 2

22-04-2015

Affine Invariance of Shifted Partials

Introduction

It is said that the dimension of Span of Shifted Partials is invariant under affine substitution. That is, it
will not increase under an affine substitution (using the same set of variables). In fact it remains the same
if the substitution is invertible. (A substitution is invertible if the original expression can be unambiguously
recovered from the substituted expression.) We try to prove/disprove these statements here.

1.1

Definition

Let F be a field. With nonnegative integer parameters k, `, the shifted partials measure is a function
SPDk,` : F[x] {0, 1, 2, . . . } defined as follows. Let f F[x]. Furthermore for any xk we use the
k
f
f . (Of course, for any permutation of we have
notation 1 2 ...k f , or sometimes f , for 1...
k
=k
k
f = f .) Let f denote the set { f : x }. Then
def

SPDk,` (f ) = dim(spanF (x` =k f )).

1.2

(1)

Notations.

Let |x| = n and f F[x]. For g F[x]n , by f (g) we mean the polynomial obtained by substituting xi with
symbol gi in f (x). f (g(x)) denotes the polynomial obtained by substituting xi with the polynomial gi (x)
in f (x). If an operator op is first applied on f and then the coordinates are changed to g then we denote
this by (op f )(g). The other way round is denoted simply by op f (g).

1.3

Restricting to linear substitutions.

An affine form is a linear polynomial. A linear form is a homogeneous linear polynomial. Partly for warmup
and partly for I am skeptical of the truth of the claim with affine substitutions, I will focus first on substitution
with linear forms. So let gi (x)s be linear forms.

1.4

Invariance under linear substitutions.

First we show the invariance of partial derivatives without shift.


Theorem 1 Let f F[x]. If g F[x]n are linear forms then spanF ( =k f (g(x))) spanF (( =k f )(g(x))).
Moreover if g1 , . . . , gn are linearly independent then the two spans become equal.
Proof: To prove the first part it suffices to show that every element of =k f (g(x)) is a linear combination of elements in ( =k f )(g(x)). Particularly, we show, by inducting on k, that the derivative (w.l.o.g.)
1

Lecture Affine Invariance of Shifted Partials:

x1 ,...,xk f (g(x)) equals the sum

(xj1 ,xj2 ,...,xjk f )(g(x))

j1 ,...,jk [n]

xt gjt . Since g1 , . . . , gn are affine forms

t[k]

it follows that xt gjt is a constant for every t [k] and that the sum is our required linear combination. It
remains to present the induction. For k = 0 the claim trivially holds. Assume the inductive hypothesis for
k 1. Then
x1 ,...,xk f (g(x)) = x1 x2 ,...,xk f (g(x))

X
= x1
(xj2 ,...,xjk f )(g(x))
j2 ,...,jk [n]



x1 xj2 ,...,xjk f )(g(x))

xt gjt

j2 ,...,jk [n]

xt gjt

j2 ,...,jk [n]

(from the linearity of derivatives)

t=2,...,k

(from the inductive hypothesis)

t=2,...,k

xj1 ,xj2 ,...,xjk f )(g(x)) x1 gj1

j1 [n]

(xj1 ,xj2 ,...,xjk f )(g(x))

j1 ,...,jk [n]

xt gjt

(from chain rule)

t=2,...,k

xt gjt .

t[k]

(k)

Proving
the moreover part then reduces to showing that the nk nk matrix A(k) , defined by Ai,j =
Q
t[k] xit gjt , is full rank. This we show by inducting on k but we first make an observation: Linear
independence of g1 , . . . , gn implies that the n n matrix B, where Bi,j is given by the coefficient of xi in
gj , is full rank. Incidentally, Bi,j = xi gj as well. Proceeding, the k = 0 case is trivial. From the inductive
hypothesis we have that A(k1) is full rank. We observe that A(k) is the Kronecker product B A(k1) . That
(k)
(k1)
is, A(i1 ,...,ik ),(j1 ,...,jk ) = Bi1 ,j1 A(i2 ,...,ik ),(j2 ,...,jk ) . This implies that rank(A(k) ) = rank(B) rank(A(k1) ) =
n nk1 = nk . That is, A(k) is full rank.

Anda mungkin juga menyukai