3
$\begingroup$

Let $E$ be a finite dimensional vector space over some field $k$. Let $u$ be an endomorphism of $E$. Let $P$ and $Q$ be two polynomials in $k[X]$. Then the following identities are satisfied:
$$\operatorname{Im}\gcd(P,Q)(u)=\operatorname{Im}P(u)+\operatorname{Im}Q(u)$$ $$\operatorname{Ker}\gcd(P,Q)(u)=\operatorname{Ker}P(u)\cap \operatorname{Ker}Q(u)$$ $$\operatorname{Im}\operatorname{lcm}(P,Q)(u)=\operatorname{Im}P(u)\cap\operatorname{Im}Q(u)$$ $$\operatorname{Ker}\operatorname{lcm}(P,Q)(u)=\operatorname{Ker}P(u)+ \operatorname{Ker}Q(u)$$

I was able to show the two first identities concerning the $\gcd$ quite easily : one inclusion comes from the fact that $\gcd(P,Q)$ divides both $P$ and $Q$, the other is given by a Bézout identity.

However surprisingly, I could not find a way to prove the two last identities concerning the $\operatorname{lcm}$. Namely, one implication always come easily from the fact that the $\operatorname{lcm}$ is a common multiple of $P$ and $Q$ (that is, we have $\subset$ for the identity with the images, and $\supset$ for the identity with the kernels).

However, as we have no relation such as the Bézout identity for the $\operatorname{lcm}$, I am having a hard time to prove the reverse inclusion. I tried to use the fact that $\operatorname{lcm}(P,Q)|PQ$ but that was not really conclusive.

Would you know a nice argument to prove the reverse inclusions?

NB: It is possible that the finite dimensional hypothesis is unneeded.

$\endgroup$
2
$\begingroup$

Actually, I just found a way to answer the problem. Let us write down $$P':=\frac{P}{\gcd(P,Q)} \quad ; \quad Q':=\frac{Q}{\gcd(P,Q)}$$ so that $P'$ and $Q'$ are coprime. We have the Bézout identity $AP' + BQ' = 1$ for some polynomials $A$ and $B$.

The kernels:
Let $x\in \operatorname{Ker}\operatorname{lcm}(P,Q)(u)$. By the Bézout identity given above, we know that $x=AP'(u)(x) + BQ'(u)(x)$. I claim that $AP'(u)(x)\in \operatorname{Ker}Q(u)$ and that $BQ'(u)(x)\in \operatorname{Ker}P(u)$.

Indeed, we have $$Q(u)[AP'(u)(x)]=AQP'(u)(x)=\lambda A\operatorname{lcm}(P,Q)(u)(x)=0$$ by the usual properties of the $k$-algebra $k[u]$ and because $QP=\lambda\operatorname{lcm}(P,Q)\gcd(P,Q)$, where $\lambda$ is the product of the dominant coefficients of $P$ and $Q$. The case of $BQ'(u)(x)$ is exactly similar to this.

Hence, we showed that $x\in \operatorname{Ker}P(u)+ \operatorname{Ker}Q(u)$.

The images:
Let $z\in \operatorname{Im}P(u)\cap\operatorname{Im}Q(u)$. We may find $x, y\in E$ such that $z=P(u)(x)=Q(u)(y)$. By the Bézout identity, we have

$$z=AP'(u)(z) + BQ'(u)(z)=AP'Q(u)(y)+BQ'P(u)(x)$$

expressing $z$ in terms either of $x$ either of $y$. Now, as before, we have $P'Q=Q'P=\lambda\operatorname{lcm}(P,Q)$. Hence,

$$z=\operatorname{lcm}(P,Q)(u)[\lambda A(u)(y)+\lambda B(u)(x)]\in \operatorname{Im}\operatorname{lcm}(P,Q)(u)$$

This concludes the proof.

$\endgroup$

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.