2
$\begingroup$

I did a few examples using the CRT to solve congruences where everything was in terms of integers. I'm trying to use the same technique for polynomials over $\mathbb{Q}$, but I'm getting stuck.


Here's an example with integers:

$\begin{cases}x \equiv 1 \, (\mathrm{mod} \, 5) \\ x \equiv 2 \, (\mathrm{mod} \, 7) \\ x \equiv 3 \, (\mathrm{mod} \, 9) \\ x \equiv 4 \, (\mathrm{mod} \, 11). \end{cases}$

Since all the moduli are pairwise relatively prime, we can use the CRT. Here's some notation I'm using:

$\bullet \, M$ denotes the product of the moduli (in this case, $M = 5 \cdot7 \cdot 9 \cdot 11$)

$\bullet \, m_i $ denotes the modulus in the $i^{\mathrm{th}}$ congruence

$\bullet \, M_i$ denotes $\dfrac{M}{m_i}$

$\bullet \, y_i$ denotes the inverse of $M_i$ (mod $m_i$), i.e. $y_i$ satisfies $y_i M_i \equiv 1$ (mod $m_i$).

Then $x = \displaystyle \sum_{i = 1}^n a_iM_iy_i$, and this solution is unique (mod $M$).


Now I want to apply the same technique to the following:

$\begin{cases} f(x) \equiv 1 \, (\mathrm{ mod } \, x^2 + 1) \\ f(x) \equiv x \, (\mathrm{mod} \, x^4), \end{cases}$

where $f(x) \in \mathbb{Q}(x)$. Having checked that the moduli are relatively prime, we should be able to use the CRT. Using the notation above, I have the following:

$M = (x^4)(x^2 + 1)$

$M_1 = x^4$

$M_2 = x^2 + 1$

Here's where I run into a problem. I need to find $y_1, y_2$ such that

$\begin{cases} y_1 (x^4) \equiv 1 \, (\mathrm{mod} \, x^2 + 1) \\ y_2 (x^2+1) \equiv 1 \, (\mathrm{mod} \, x^4). \end{cases}$

But how does one find $y_1, y_2$?

$\endgroup$
  • 1
    $\begingroup$ "Having checked that the moduli are relatively prime..." but that means precisely that there exist $p_1(x)$ and $p_2(x)$ such that $p_1(x)x^4 + p_2(x)(x^2+1)=1$. $\endgroup$ – kccu Apr 21 at 19:23
  • $\begingroup$ Are you saying that I can find $p_1(x)$ and $p_2(x)$ in general by using the extended Euclidean algorithm, and that $p_1(x)$ and $p_2(x)$ are precisely my $y_1$ and $y_2$? $\endgroup$ – Junglemath Apr 21 at 19:56
3
$\begingroup$

To find $y_1$ and $y_2$ consider solving the problem $$y_1x^4+y_2(x^2+1)=1.$$ This is not always easy to solve, but in this case a solution comes to mind. Note that by difference of squares $$(x^2-1)(x^2+1)=x^4-1,$$ hence $$x^4+[(-1)(x^2-1)](x^2+1)=1.$$ This tells us that we can choose $$y_1=1,$$ $$y_2=(1-x^2).$$

$\endgroup$
  • $\begingroup$ Is there an algorithmic way of solving these, rather than relying on intuition? $\endgroup$ – Junglemath Apr 21 at 19:59
  • $\begingroup$ @Junglemath By Euclidean algorithm you can find two polynomials $p(x), q(x) \in \mathbb{Q}[x]$ such that $p(x) x^4 + q(x) (x^2 + 1) = 1$. $\endgroup$ – Paolo Apr 21 at 20:09
  • 2
    $\begingroup$ @Junglemath The polynomials over a field form a Euclidean Domain, so yes, there is. If the gcd of $f_1,f_2\in\mathbb{Q}[x]$ is a unit, then you can perform the Euclidean Algorithm to find their gcd. Now, you can reverse the algorithm to write them as a linear combination of their gcd the same way you would for integers. I said not easy, because the process can be time consuming and very tedious. $\endgroup$ – Melody Apr 21 at 20:10
  • 2
    $\begingroup$ @Junglemath I didn't really convert it. I knew in advanced the single equation had a solution. This is because $x^4$ and $x^2+1$ have no common roots, hence no common irreducible factors. This means they are relatively prime, in which case we can write $1$ as a linear combination. Knowing that, I knew solving the single equation would give rise to a solution to the congruence equations. This is completely analogous to how you can solve everything over the integers. $\endgroup$ – Melody Apr 21 at 20:22
  • 2
    $\begingroup$ @Junglemath I describe here at length this method of scaling the Bezout equation into a CRT solution. $\endgroup$ – Bill Dubuque Apr 21 at 21:28
2
$\begingroup$

Bu applying $\ ab\bmod ac\, =\, a(b\bmod c)\ $ [Mod Distributive Law] $ $ it is a bit simpler:

$ f-x\,\bmod\, {x^{\large 4}(x^{\large 2}\!+\!1)}\, =\, x^{\large 4}\underbrace{{\left[\dfrac{\color{#c00}f-x}{\color{#0a0}{x^{\large 4}}}\bmod {x^{\large 2}\!+\!1}\right]}}_{\large \color{#0a0}{x^{\Large 4}} \ \equiv\ 1\ \ \ {\rm by}\ \ \ x^{\Large 2}\ \equiv\ \ -1 } =\, x^{\large 4}[1-x],\ $ by $\,\color{#c00}f\equiv 1\pmod{\!x^{\large 2}\!+\!1}$

Remark $ $ Here are further examples done using MDL (an operational form of CRT).

You can find further details here on transforming the Bezout equation into a CRT solution (the method sketched in Melody's answer).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.