next up previous
Next: Quadratic Hensel Lifting Up: Hensel Lifting Previous: Hensel Lifting

Linear Hensel Lifting

Theorem 15   Let R be a commutative ring with identity element. Let f, g0, h0 be univariate polynomials in R[x] and let m $ \in$ R. We assume that the following relation holds

f  $\displaystyle \equiv$  g0 h0 modm (121)

We assume also that g0 and h0 are relatively prime modulo m, that is there exists s, t $ \in$ R such that

sg0 + th0  $\displaystyle \equiv$  1 modm (122)

Then, for every integer $ \ell$ there exist g($\scriptstyle \ell$), h($\scriptstyle \ell$) $ \in$ R[x] such that we have
  1. f  $ \equiv$  g($\scriptstyle \ell$)h($\scriptstyle \ell$)modm$\scriptstyle \ell$,
  2. g0  $ \equiv$  g($\scriptstyle \ell$)modm.

Proof. We apply Theorem 13 with

$\displaystyle \cal {I}$ = $\displaystyle \langle$m$\displaystyle \rangle$n = 1, r = 2, f1(x1, x2) = x1x2 - f. (123)

Thus we have

U  =  $\displaystyle \left(\vphantom{ u_{11} \ u_{12} }\right.$u11 u12$\displaystyle \left.\vphantom{ u_{11} \ u_{12} }\right)$  =  $\displaystyle \left(\vphantom{g_0 \ h_0 }\right.$g0 h0$\displaystyle \left.\vphantom{g_0 \ h_0 }\right)$ (124)

Observe that this matrix is left-invertible modulo m if and only if g0, h0 are relatively prime modulo m. Indeed, the matrix U is left-invertible modulo m iff there exists a matrix

U-1 = $\displaystyle \left(\vphantom{ \begin{array}{c} s \\  t \end{array} }\right.$$\displaystyle \begin{array}{c} s \\  t \end{array}$$\displaystyle \left.\vphantom{ \begin{array}{c} s \\  t \end{array} }\right)$    

such that U U-1  $ \equiv$  $ \left(\vphantom{ 1 }\right.$1$ \left.\vphantom{ 1 }\right)$modm, that is

sg0 + th0  $\displaystyle \equiv$  1 modm (125)

We prove now the claim of the theorem. Assume we have computed g($\scriptstyle \ell$), h($\scriptstyle \ell$) $ \in$ R such that f  $ \equiv$  g($\scriptstyle \ell$)h($\scriptstyle \ell$)modm$\scriptstyle \ell$ and g  $ \equiv$  g($\scriptstyle \ell$)modm hold. We want to compute g($\scriptstyle \ell$+1), h($\scriptstyle \ell$+1) $ \in$ R. Let v be such that

f1(g($\scriptstyle \ell$), h($\scriptstyle \ell$)) = g($\scriptstyle \ell$)h($\scriptstyle \ell$) - f = vm$\scriptstyle \ell$ (126)

We look for g$\scriptstyle \ell$, h$\scriptstyle \ell$ $ \in$ R such that

g($\scriptstyle \ell$+1) = g($\scriptstyle \ell$) + g$\scriptstyle \ell$m$\scriptstyle \ell$  and  h($\scriptstyle \ell$+1) = h($\scriptstyle \ell$) + h$\scriptstyle \ell$m$\scriptstyle \ell$ (127)

Following the proof of Theorem 13 we are led to solve the equation

v + g$\scriptstyle \ell$h0 + h$\scriptstyle \ell$g0 $\displaystyle \equiv$  0 modm (128)

A solution of this equation is

$\displaystyle \left(\vphantom{\begin{array}{c} g_{\ell} \\  h_{\ell} \end{array} }\right.$$\displaystyle \begin{array}{c} g_{\ell} \\  h_{\ell} \end{array}$$\displaystyle \left.\vphantom{\begin{array}{c} g_{\ell} \\  h_{\ell} \end{array} }\right)$ = $\displaystyle \left(\vphantom{ \begin{array}{c} - s v\\  - t v \end{array} }\right.$$\displaystyle \begin{array}{c} - s v\\  - t v \end{array}$$\displaystyle \left.\vphantom{ \begin{array}{c} - s v\\  - t v \end{array} }\right)$ (129)

This proves the claim of the theorem. $ \qedsymbol$

Remark 23   The proof of Theorem 15 provides a solution (g$\scriptstyle \ell$, h$\scriptstyle \ell$) to Equation (128). It is desirable to add constraints such that Equation (128) has a unique soulution. In particular, one would like to guarantee that for every integer $ \ell$ the polynomials g($\scriptstyle \ell$) and g0 have the same degree. This problem is solved in the following sub-section where we study Linear Diophantine Equations.

Definition 10   Let R be a commutative ring with identity element. A linear Diophantine equation over R is an equation of the form

f1s1 + ... + fnsn = a (130)

where f1,..., fn, a are given in R and where s1,..., sn are unknowns in R.

Theorem 16   Let R be an Euclidean domain and let f, g, h, a $ \in$ R such that h = gcd(f, g). We consider the linear Diophantine equation

sf + tg = a. (131)

Then the following hold
(i)
Equation (131) has a solution if and only if h divides a.
(ii)
If h $ \neq$ 0 and if (s0, t0) $ \in$ R2 is a solution of Equation (131) then every other solution is of the form

(s0 + r$\displaystyle {\frac{{g}}{{h}}}$, t0 - r$\displaystyle {\frac{{f}}{{h}}}$where  r $\displaystyle \in$ R. (132)

(iii)
If R = F[x] where F is a field, h $ \neq$ 0, and if Equation (131) is solvable, where

degf + degg - degh > dega (133)

then there is a unique solution (s0, t0) $ \in$ R2 of Equation (131) such that

degs < degg - degh  and  degt < degf - degh. (134)

Proof. First we prove (i). If (s0, t0) $ \in$ R2 is a solution of Equation (131), then h which divides s0f + t0g, divides also a. Conversly, we assume that h = gcd(f, g) divides a. The claim is trivial if h = 0. (Indeed, this implies f = g = 0 and also a = 0.) Otherwise, let (s0, t0) $ \in$ R2 be computed by the Extended Euclidean Algorithm applied to (f, g), such that we have s0f + t0g = h. Then (s, t) = (s0a/h, t0a/h) is a solution of Equation (131).

Now we prove (ii). Assume h $ \neq$ 0 and let (s0, t0) $ \in$ R2 is a solution of Equation (131). Since h $ \neq$ 0, then f /h and g/h are coprime. Let (s, t) be in R2. Then we have

sf + tg = a $\displaystyle \iff$ (s - s0)f + (t - t0)g = 0
  $\displaystyle \iff$ (s - s0)(f /h) = - (t - t0)(g/h)
  $\displaystyle \iff$ (s - s0)(f /h) = (t0 - t)(g/h)
  $\displaystyle \iff$ (g/h) | (s - s0)  and  (f /h) | (t0 - t)
  $\displaystyle \iff$ ($\displaystyle \exists$r $\displaystyle \in$ R) (s - s0) = r(g/h)  and  (t0 - t) = r(f /h)
  $\displaystyle \iff$ ($\displaystyle \exists$r $\displaystyle \in$ Rs = s0 + r(g/h)  and  t = t0 - r(f /h).
   

Finally, we prove (iii). Let (s1, t1) $ \in$ R2 be a solution of Equation (131). Let q and t0 be the quotient and the remainder of t1 w.r.t. f /h. Hence we have

t1 = f /hq + t0  and  degt0 < degf - degh. (135)

We define

s0 = s1 + qg/h. (136)

Observe that

(s0, t0) = (s1, t1) + q(g/h, - f /h)    

solves Equation (131) too. By Relation (135) and by hypothesis we have

deg(t0g) < degf + degg - degh  and  dega < degf + degg - degh    

leading to

degs0 + degf = degs0f = deg(a - t0g) < degf + degg - degh.    

Therefore we have

degt0 < degf - degh  and  degs0 < degg - degh.    

This proves the existence. Let us prove the unicity. Let (s, t) be a solution of Equation (131). We know that there exists r $ \in$ R such that s = s0 + rg/h. Since degs0 < degg - degh holds, if r $ \neq$ 0, we have

degs $\displaystyle \geq$ deg(g/h) = degg - degh.    

This implies the uniquemess. $ \qedsymbol$

Theorem 17   Let R be a commutative ring with identity element. Let f, g0, h0 be univariate monic polynomials in R[x]. Let m $ \in$ R be such that R/m is a field. We assume that the following relation holds

f  $\displaystyle \equiv$  g0 h0 modm (137)

and that g0 and h0 are relatively prime modulo m. Then, for every integer $ \ell$ there exist unique monic polynomials g($\scriptstyle \ell$), h($\scriptstyle \ell$) $ \in$ R[x] such that we have
  1. f  $ \equiv$  g($\scriptstyle \ell$)h($\scriptstyle \ell$)modm$\scriptstyle \ell$,
  2. g0  $ \equiv$  g($\scriptstyle \ell$)modm,
  3. h0  $ \equiv$  h($\scriptstyle \ell$)modm.

Proof. By induction on $ \ell$ $ \geq$ 1. The clain is clear for $ \ell$ = 1. So let us assume it is true for $ \ell$ $ \geq$ 1 and consider monic polynomials g($\scriptstyle \ell$+1), h($\scriptstyle \ell$+1) $ \in$ R[x] satisfying

f  $\displaystyle \equiv$  g($\scriptstyle \ell$+1)h($\scriptstyle \ell$+1)modm$\scriptstyle \ell$+1 (138)

and

g0  $\displaystyle \equiv$  g($\scriptstyle \ell$+1)modm  and  h0  $\displaystyle \equiv$  h($\scriptstyle \ell$+1)modm. (139)

Such polynomials g($\scriptstyle \ell$+1), h($\scriptstyle \ell$+1) exist by Theorem 15. (The fact that they can be chosen monic is left to the reader as an exercise.) Observe that we have

f  $\displaystyle \equiv$  g($\scriptstyle \ell$+1)h($\scriptstyle \ell$+1)modm$\scriptstyle \ell$ (140)

So the induction hypothesis leads to

g($\scriptstyle \ell$)  $\displaystyle \equiv$  g($\scriptstyle \ell$+1)modm$\scriptstyle \ell$  and  h($\scriptstyle \ell$)  $\displaystyle \equiv$  h($\scriptstyle \ell$+1)modm$\scriptstyle \ell$ (141)

Hence there exist polynomials qg, qh $ \in$ (R/m)[x] such that

g($\scriptstyle \ell$+1) = g($\scriptstyle \ell$) + m$\scriptstyle \ell$qg  and  h($\scriptstyle \ell$+1) = h($\scriptstyle \ell$) + m$\scriptstyle \ell$qh. (142)

Since f, g0, h0 are given to be monic, it is easy to prove that for k $ \geq$ 1 we have

degqg < degg0  and  degqh < degh0 (143)

In addition, observe that combining Equation (138) and Equation (142) we obtain

f $\displaystyle \equiv$ g($\scriptstyle \ell$)h($\scriptstyle \ell$) + $\displaystyle \left(\vphantom{g^{({\ell})} q_h + h^{({\ell})} q_g }\right.$g($\scriptstyle \ell$)qh + h($\scriptstyle \ell$)qg$\displaystyle \left.\vphantom{g^{({\ell})} q_h + h^{({\ell})} q_g }\right)$m$\scriptstyle \ell$modm$\scriptstyle \ell$+1
(144)

The induction hypothesis shows that f - g($\scriptstyle \ell$)h($\scriptstyle \ell$) is a multiple of m$\scriptstyle \ell$. Then we obtain

g0qh + h0qg  $\displaystyle \equiv$  $\displaystyle {\frac{{f - g^{({\ell})} h^{({\ell})} }}{{m^{\ell}}}}$modm. (145)

By Theorem 16 and Equation (143), the Equation (145) has a unique solution. $ \qedsymbol$


next up previous
Next: Quadratic Hensel Lifting Up: Hensel Lifting Previous: Hensel Lifting
Marc Moreno Maza
2004-04-27