next up previous
Next: Symbolic Newton Iteration Up: Advanced Computer Algebra: From Newton Previous: Division with remainder using Newton

P-adic Expansions and Approximations

The goal of this section is to formalize in an algebraic point of view the notions of expansions and approximations well known in numerical analysis.

Definition 6   Let R be an Euclidean domain and let p $ \in$ R be a prime element. Let a $ \in$ E be an element. We say that a has degree zero w.r.t. p and we write deg(a, p) = 0 if for every positive integer k $ \in$ $ \mbox{${\mathbb N}$}$ the quotient of a w.r.t. pk is zero. Assume from now on that a does not have degree zero w.r.t. p. If for every positive integer $ \ell$ there exists a positive integer k > $ \ell$ such that the quotient of a w.r.t. pk is not zero, then we say a has infinite degree w.r.t. p and we write deg(a, p) = + $ \infty$, otherwise we say that a has finite degree w.r.t. p and we write deg(a, p) < + $ \infty$. Assume furthermore that deg(a, p) < + $ \infty$ holds. The largest positive integer k $ \in$ $ \mbox{${\mathbb N}$}$ such that the quotient of a w.r.t. pk is not zero is called the degree of a w.r.t. p and is denoted by deg(a, p).

Definition 7   The Euclidean size $ \delta$ of an Euclidean domain R is said regular if for every a, b $ \in$ R with b $ \neq$ 0 there exists a couple (q, r) $ \in$ E×E such that
  1. a = bq + r,
  2. r $ \neq$$ \Rightarrow$   $ \delta$(r) < $ \delta$(b),
  3. q $ \neq$$ \Rightarrow$   $ \delta$(q) < $ \delta$(aand  $ \delta$(b) $ \leq$ $ \delta$(a).

Proposition 8   Let R be an Euclidean domain with a regular Euclidean size $ \delta$. Let p be a prime element. Then for every non-zero a $ \in$ E there exists an integer d $ \in$ N and elements a0, a1,..., ad $ \in$ E such that
  1. a = adpd + ... + a1p + a0,
  2. ad $ \neq$ 0,
  3. for every i = 0 ... d we have deg(ai, p) = 0.
The sequence (a0, a1,..., ad) is called a p-adic expansion of a w.r.t. p.

Proof. Let (q0, a0) be a couple quotient-remainder given as in Definition 7 when dividing a w.r.t. p.

First we prove that a0 must have degree 0 w.r.t. p. This is obvious if a0 = 0. So let us assume that a0 $ \neq$ 0 holds. This implies $ \delta$(p) > $ \delta$(a0). If a0 would have positive degree k w.r.t. p, then, by the properties of a regular Euclidean size, we would have $ \delta$(a0) $ \geq$ $ \delta$(pk). Since $ \delta$(pk) $ \geq$ $ \delta$(p) holds (by definition of an Euclidean domain) we would have a contradiction with $ \delta$(p) > $ \delta$(a0). Therefore in any case we have deg(a0, p) = 0.

Next, let us prove the proposition for those elements a which have quotient 0 w.r.t. p. Observe that if one quotient of a w.r.t. p is zero then we have $ \delta$(a) < $ \delta$(p). Hence every quotient of a w.r.t. p is null. (Otherwise we would have $ \delta$(p) $ \leq$ $ \delta$(a) by the properties of a regular Euclidean size.) Since a = a0 and since a0 has degree 0 w.r.t. p, the element a has degree 0 w.r.t. p and (a0) is a p-adic expansion of a w.r.t. p. Therefore the proposition is proved in the case of the elements a which have quotient 0 w.r.t. p.

We can assume now that a does not have quotient 0 w.r.t. p and thus q0 $ \neq$ 0 and $ \delta$(q0) < $ \delta$(a) hold. Then let (q1, a1) be a couple quotient-remainder given as in Definition 7 when dividing q0 w.r.t. p. If q1 $ \neq$ 0 then we have $ \delta$(q1) < $ \delta$(q0) and let (q2, a2) be a couple quotient-remainder given as in Definition 7 when dividing q1 w.r.t. p. Continuing in this manner we obtain a finite sequence a0, a1,..., ad $ \in$ R and a finite sequence q0, q1,..., qd $ \in$ R such that we have

a = q0p + a0 with q0 $\displaystyle \neq$ 0
q0 = q1p + a1 with q1 $\displaystyle \neq$ 0
q1 = q2p + a2 with q2 $\displaystyle \neq$ 0
$\displaystyle \vdots$ $\displaystyle \vdots$ $\displaystyle \vdots$ $\displaystyle \vdots$ $\displaystyle \vdots$ $\displaystyle \vdots$ $\displaystyle \vdots$
qd2 = qd-1p + ad-1 with qd-1 $\displaystyle \neq$ 0
qd-1 = qdp + ad with qd = 0
   

This process stops since we have

$\displaystyle \delta$(a) > $\displaystyle \delta$(q0) > $\displaystyle \delta$(q1) > $\displaystyle \delta$(q2) > ... > $\displaystyle \delta$(qd-1) (83)

and since the Euclidean size of an element has value in $ \mbox{${\mathbb N}$}$  $ \cup$  { - $ \infty$}. Clearly we have

a = adpd + ... + a1p + a0.    

Moreover ad = qd-1 $ \neq$ 0 holds. In addition, for every i = 0 ... d we have deg(ai, p) = 0 by the first part of this proof. $ \qedsymbol$

Remark 15   Proposition 8 applies in the ring $ \mathbb {Z}$ of integers and the ring $ \bf k$[x] of univariate polynomials over a field $ \bf k$. It extends also for the ring R[y] of univariate polynomials over an Euclidean domain R with a regular Euclidean size $ \delta$ and a prime element p $ \in$ R. Indeed, one can apply Proposition 8 to each coefficient of a polynomial $ \phi$ $ \in$ R[y].

As we shall see now, it extends to a ring R[y] of univariate polynomials over a commutative ring with identity element and to the element p = y - g where g is an element of R.

Proposition 9   Let R be a commutative ring with identity element. Let $ \phi$ $ \in$ R[y] be a polynomial of degree n and let g $ \in$ R be an element. We define p = y - g. There exists a unique sequence $ \phi_{{0}}^{{}}$,$ \phi_{{1}}^{{}}$,...,$ \phi_{{n}}^{{}}$ $ \in$ R such that we have

$\displaystyle \phi$ = $\displaystyle \phi_{{0}}^{{}}$ + $\displaystyle \phi_{{1}}^{{}}$p + ... + $\displaystyle \phi_{{n}}^{{}}$pn. (84)

The above expression is called the Taylor expansion of $ \phi$ at g,

Proof. By induction on n. If n = 0, the result is clear. Otherwise let q and r be the quotient and the remainder in the division of $ \phi$ by p. We have

$\displaystyle \phi$ = qp + r    and    r $\displaystyle \in$ R.    

Since r = $ \phi$(g) the conclusion (existence and unicity) follows by the induction hypothesis. $ \qedsymbol$

Remark 16   We can characterize the Taylor expansion of $ \phi$ at g by means of the notion of a formal derivative.

Definition 8   Let R be a commutative ring with identity element. For

$\displaystyle \phi$ = $\displaystyle \phi_{{0}}^{{}}$ + $\displaystyle \phi_{{1}}^{{}}$y + $\displaystyle \phi_{{2}}^{{}}$y2 + ... + $\displaystyle \phi_{{n}}^{{}}$yn    

we define the formal derivative of $ \phi$ by

$\displaystyle \phi$$\scriptstyle \prime$ = $\displaystyle \phi_{{1}}^{{}}$ + 2$\displaystyle \phi_{{2}}^{{}}$y + ... + n$\displaystyle \phi_{{n}}^{{}}$yn-1. (85)

Remark 17   The terminology formal derivative comes from the fact that we do not make use of the notion of a limit to define this derivative. Indeed, if R is a finite ring, then this would not make clear sense. However we retrieve the usual properties of derivatives.

Proposition 10   Let R be a commutative ring with identity element. Let $ \phi$,$ \psi$ $ \in$ R[y] and let a, b $ \in$ R. Then the following properties hold
  1. (a$ \phi$ + b$ \psi$)' = a$ \phi$$\scriptstyle \prime$ + b$ \phi$$\scriptstyle \prime$,
  2. ($ \phi$$ \psi$)' = $ \phi$$\scriptstyle \prime$$ \psi$ + $ \phi$$ \psi$$\scriptstyle \prime$,
  3. ($ \phi$($ \psi$))' = $ \psi$$\scriptstyle \prime$$ \phi$$\scriptstyle \prime$($ \psi$).

Proof. See Lemma 9.19 in [vzGG99]. $ \qedsymbol$

Proposition 11   Let R be a commutative ring with identity element. Let $ \phi$ $ \in$ R[y] and let g $ \in$ R. There exists a polynomial $ \psi$ $ \in$ R[y] such that we have

$\displaystyle \phi$(y) = $\displaystyle \phi$(g) + $\displaystyle \phi$$\scriptstyle \prime$(g)(y - g) + $\displaystyle \psi$(y)(y - g)2. (86)

Proof. Follows easily from Proposition 9. $ \qedsymbol$

Remark 18   Proposition 9 admits several generalizations. First one can use a monic polynomial p of degree d > 1. In this case the coefficients $ \phi_{{0}}^{{}}$,$ \phi_{{1}}^{{}}$,...,$ \phi_{{n}}^{{}}$ are polynomials of degree less than d. Another generalization, given by Proposition 12, is for multivariate polynomials. To understand this generalization, it is important to make the following three observations.

Proposition 12   Let R be a commutative ring with identity element. We consider the ring of multivariate polynomials P = R[x1,..., xr][y1,..., yr]. We define $ \bf x$ = (x1,..., xr) and $ \bf y$ = (y1,..., yr). For every polynomial $ \phi$ $ \in$ R[$ \bf x$] there exists a polynomial $ \psi$ $ \in$ P lying in the ideal $ \langle$y1,..., yr$ \rangle^{{2}}_{{}}$ such that

$\displaystyle \phi$(x1 + y1,..., xr + yr) = $\displaystyle \phi$($\displaystyle \bf x$) + $\displaystyle \Sigma_{{{j=1}}}^{{{j=r}}}$$\displaystyle {\frac{{{\partial}{\phi}}}{{{\partial} x_j}}}$($\displaystyle \bf x$)yj + $\displaystyle \psi$. (87)

Remark 19   Using the vocabulary of numerical analysis, Propositions 11 and 12 provides a Taylor expansion of $ \phi$ of order 2. This leads to the following definition.

Definition 9   Let R be a commutative ring with identity element and let $ \cal {I}$ be an ideal of R. Let a, b $ \in$ R be elements and k be a positive integer. We say that a is an $ \cal {I}$-adic approximation of b at order k if we have

a $\displaystyle \equiv$ b mod$\displaystyle \cal {I}$k. (88)

Remark 20   We conclude by two useful properties of ideal-adic approximations.

Proposition 13   Let R be an Euclidean domain with a regular Euclidean size $ \delta$. Let p $ \in$ R be a prime element. Let a $ \in$ R be an element and let (a0, a1,..., ad) be a p-adic expansion of a w.r.t. p. For every positive integer k $ \leq$ d + 1 we define

a(k) = a0 + a1p + ... ak-1pk-1 (89)

Then a(k) is a p-adic approximation of a at order k, that is

a(k) $\displaystyle \equiv$ a modpk. (90)

Proof. Indeed, we have

a - a(k) = akpk + ak+1pk+1 + ... + adpd = (ak + ak+1p + ... + adpd-k)pk    

$ \qedsymbol$

Proposition 14   Let R be a commutative ring with identity element and let $ \cal {I}$ be an ideal of R. Let a, b $ \in$ R be elements, let k be a positive integer and let $ \phi$ $ \in$ R[y] be a polynomial. Then we have

a $\displaystyle \equiv$ b mod$\displaystyle \cal {I}$k  $\displaystyle \Rightarrow$   $\displaystyle \phi$(a) $\displaystyle \equiv$ $\displaystyle \phi$(b)mod$\displaystyle \cal {I}$k (91)

That is, if a is an $ \cal {I}$-adic approximation of b at order k, then $ \phi$(a) is an $ \cal {I}$-adic approximation of $ \phi$(b) at order k.

Proof. This is simply because of the ring-homomorphism between R and R/$ \cal {I}$. $ \qedsymbol$


next up previous
Next: Symbolic Newton Iteration Up: Advanced Computer Algebra: From Newton Previous: Division with remainder using Newton
Marc Moreno Maza
2004-04-27