• No results found

Reducing Problem 1 to Problem 2

In this section, we show how instances of Problem 1 can be reduced to instances of Problem 2. The main technical ingredient, stated in Lemma 8 below, generalizes to any s > 1 the one given for s = 1 by Zeh, Gentner, and Augot in [39, Proposition 3]. To prove it, we use the same steps as in [39]; we rely on the notion of Hasse derivatives, which allow us to write Taylor expansions in positive characteristic (see for instance Hasse [19] or Roth [31, pp. 87, 276]).

In what follows, comparison and addition of s-tuples of integers are defined com- ponentwise. For example, writing i 6 j is equivalent to ik 6 jk for k = 1, . . . , s,

and ij denotes (i1−j1, . . . , is−js). Similarly, ify= (y1, . . . , ys) is in K[X]s then

Y y= Y1−y1, . . . , Ys−ys. Finally, for products of binomial coefficients, we shall

write j i = j1 i1 · · · js is .

Note that this coefficient is zero when i 66 j; note also that the monomial Yi has

total degree degY(Yi) =|i|=i1+· · ·+is.

IfAis any commutative ring with unity andA[Y] denotes the ring of polynomials in Y1, . . . , Ys over A, then for a polynomial P(Y) = PjPjY

j in A[Y] and a multi-

by P[i]=X j>i j i PjYj−i.

The Hasse derivative satisfies the following property (Taylor expansion): for all a in As,

P(Y) =X

i

P[i](a)(Y a)i.

The following lemma shows how Hasse derivatives can help rephrase the vanishing condition (iv) of Problem 1. Below, the polynomials G and R are as defined in the introduction.

Lemma 8. For any polynomial Q in K[X,Y], Q satisfies the vanishing condition (iv) of Problem 1 if and only if for all i in Ns such that |i|< m,

Q[i](X,R) = 0 modGm−|i|.

Proof. Since the xr’s defining G =Qnr=1(X−xr) are pairwise distinct, it suffices to

prove, for 1 6 r 6 n, the following equivalence for the point (xr,yr): Q(xr,yr) = 0

with order at least m if and only if for all i in Ns such that |i| < m, Q[i](X,R) = 0 mod (Xxr)m−|i|. Now, up to a shift one can assume that this point is 0∈Ks+1;

in other words, it suffices to show that for R(0) = 0 ∈ Ks, we have Q(0,0) = 0 with order at least m if and only if, for all i in Ns such that |i|< m, Xm−|i| divides

Q[i](X,R).

Assume first that 0 Ks+1 is a root of Q of order at least m. Then, Q(X,Y) = P

jQjY

j has only monomials of total degree at least m, so that for j > i, each

nonzero QjYj−i has only monomials of total degree at least m− |i|. Now, R(0) =

0 Ks implies that X divides each component of R. Consequently, Xm−|i| divides

QjRj−i for each j >i, and thus Q[i](X,R) as well.

Conversely, let us assume that for all i in Ns such that |i| < m, Xm−|i| divides

Q[i](X,R), and show that Q has no monomial of total degree less than m. Writing the Taylor expansion of Q with A=K[X] and a=R, we obtain

Q(X,Y) = X

i

Q[i](X,R)(Y R)i.

Each component of R being a multiple of X, we deduce that for the multi-indices i

at least m. Using our assumption, the same conclusion follows for the multi-indices such that |i|< m.

For i inNs, with |i|< m, define the polynomials

Pi=Gm−|i| as well as Fi= (Fi,j)j∈Γ, with Fi,j = j i Rj−i mod Pi = j1 i1 Rj1−i1 1 · · · js is Rjs−is s modPi.

Then, the previous lemma implies that Qsatisfies properties (ii)-(iv) if and only if it can be written as Q=PjΓQjYj, with deg(Qj)< Nj for all j and, for all i inNs

such that |i|< m,

X

j∈Γ

QjFi,j = 0 modPi.

The latter conditions express the problem of finding such a Q as an instance of Problem 2. In order to make the reduction completely explicit, define further

µ= s+m1 s , ν =|Γ|,

and choose arbitrary orders on the set of indices {i Ns | |i| < m} and Γ, that is, bijections

φ:{0, . . . , µ1} → {iNs | |i|< m} and ψ :{0, . . . , ν1} →Γ. To i in {0, . . . , µ− 1}, we can then associate M′

i = Mφ(i), and similarly to j in

{0, . . . , ν1}we associate N′

j =Nψ(j); we also set Pi′ =Pφ(i) and Fi,j′ =Fφ(i),ψ(j).

Proposition 9. Let (s, ℓ, m, n, k, b) be parameters for Problem 1, and let the param- eters µ, ν,M′ = (M′

0, . . . , Mµ′−1),N′ = (N0′, . . . , Nν′−1) be as above.

Then, one can reduce an instance of Problem 1 with parameters (s, ℓ, m, n, k, b)

to an instance of Problem 2 with parameters (µ, ν,M′,N′) and input polyno-

mials (P′

i,Fi′) using O(rM(M) log(M)) operations in K, where we write r =

max( s+ms−1,|Γ|) = max(µ, ν).

Proof. The only thing left to do is the complexity analysis. First, we need to compute

namely G, . . . , Gm, so it can be done using O(mM(mn)) operations; this will be dominated by the cost of the third step below.

Then, we have to compute the interpolation polynomials R= (R1, . . . , Rs)(using

Lagrange Interpolation). Each of them can be computed inO(M(n) log(n)) operations inK, for a total ofO(sM(n) log(n)), which isO(sM(M) log(M)). We have|Γ|>s, so that this cost isO(rM(M) log(M)), except in the extreme case when Γ = {(0, . . . ,0)}, so that |Γ|= 1; in that case, however, we need not compute R and we can omit the cost of interpolation.

Finally, we compute Fi,j modPi for every i,j. This is done by fixing i and

computing all products Fi,j modPi incrementally, starting from R1, . . . , Rs. Each

product takes O(M(Mi)) operations in K. Summing over all j leads to a cost of

O(|Γ|M(Mi)) per index i. Summing over all i and using the super-linearity of M

leads to a total cost of O(|Γ|M(M)), which is O(rM(M)).

Thus, in order to prove Theorem 4, it is enough to prove Theorem 3: in view of the cost reported in Theorem 3 for solving the linear system, the cost of reduction given above will always be negligible.

Related documents