I'm having trouble following a proof of this generalization of the Cayley-Hamilton theorem:

Suppose that $M$ is an $A$-module generated by $n$ elements, and that $\varphi \in \operatorname{Hom}_A(M,M)$; let $I$ be an ideal of $A$ such that $\varphi(M) \subset IM$. Then there is a relation of the form $$(**)\quad \varphi^n + a_1 \varphi^{n-1} + \cdots + a_{n-1}\varphi + a_n = 0,$$ with $a_i \in I^i$ for $1 \leqslant i \leqslant n$ (where both sides are considered as endomorphisms of $M$).

This is theorem 2.1 in Matsumura's Commutative Ring Theory. The proof there goes as follows:

Let $M = A\omega_1 + \dots + A\omega_n$; by the assumption $\varphi(M) \subset IM$ there exist $a_{ij} \in I$ such that $\varphi(\omega_i) = \sum_{j=1}^n a_{ij}\omega_j$. This can be rewritten $$\sum_{j=1}^n(\varphi\delta_{ij} - a_{ij})\omega_j = 0 \quad (\text{for }\ 1\leqslant i \leqslant n),$$ where $\delta_{ij}$ is the Kronecker symbol. The coefficients of this system of linear equations can be viewed as a square matrix $(\varphi\delta_{ij} - a_{ij})$ of elements of $A'[\varphi]$, the commutative subring of the endomorphism ring $E$ of $M$ generated by the image $A'$ of $A$ together with $\varphi$; let $b_{ij}$ denote its $(i,j)$th cofactor, and $d$ its determinant. By multiplying the above equation through by $b_{ik}$ and summing over $i$, we get $d\omega_k = 0$ for $1 \leqslant k \leqslant n$. Hence $d\cdot M = 0$, so that $d = 0$ as an element of $E$. Expanding the determinant $d$ gives a relation of the form $(**)$.

I would like to understand this particular proof, but I'm having trouble concluding $d\omega_k = 0$ for $1 \leqslant k \leqslant n$ from the multiplication of the above equation by $b_{ik}$ and summing over $i$. It's clear that the right hand side remains zero, but I'm having trouble expanding the left hand side into a useful form. I'm pretty sure I have to use the fact that $d = \sum_{i=1}^n b_{ik} (\varphi\delta_{ik} - a_{ik})$ by the cofactor expansion of the determinant along the $k$th column, but I can only seem conclude $d\omega_k - d\omega_k = 0$, which is not very enlightening.

Thanks in advance.

  • $\begingroup$ this is not a generalization of Caylay-Hamilton. It's called the determinant trick, and is equivalent to Caylay-Hamilton. $\endgroup$
    – adrido
    Jan 4 '14 at 6:51

I can see clearly now that I have thought about it longer. One thing that confused me is that the matrix for $\varphi$ is the transpose of $(a_{ij})$, rather than $(a_{ij})$ itself. This is evident from the identity $\varphi(\omega_i) = \sum_{j=1}^n a_{ij}\omega_j$.

Let $c_{ij} = \varphi\delta_{ij} - a_{ij}$ and put $v_i = \sum_{j=1}^n c_{ij}\omega_j$. Proposition: $u_k = \sum_{i=1}^n b_{ik}v_i = 0$ implies $d\omega_k = 0$.

Proof: The coefficient of $\omega_j$ in $u_k$ is $\sum_{i=1}^n b_{ik} c_{ij}$, which equals $d$ when $j = k$ and zero otherwise. When $j = k$ it's the cofactor expansion of the determinant along the $k$th column. When $j \neq k$ it's a cofactor expansion of the determinant of a matrix where the $j$th and $k$th columns are equal, which is zero.

In terms of the adjugate matrix this can be stated as $X\operatorname{adj}(X) = \operatorname{adj}(X)X = (\det X)I$.


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.