My textbook defines the characteristic polynomial of an $n \times n$-matrix $A$ as:

$$p_A = \text{det}(A-XI_n) = \sum_{\sigma \in S_n} \text{sign } \sigma. (a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)})$$

(where $S_n$ is the symmetric group). Then it just states that by reordering according to the powers of $X$, we get the form

$$p_A = a_0+a_1X+...+a_{n-1}X^{n-1}+a_nX^n$$ with coefficients $a_i$ from a field, where $a_0=\text{det } A$, $a_{n-1}=(-1)^{n-1}. \text{tr}(A)$ and $a_n=(-1)^n$.

Could someone clarify how to derive this formula from the definition?


Those equations are the consequence of developing the product on the RHS of the equation $$p_A = \text{det}(A-XI_n) = \sum_{\sigma \in S_n} \text{sign } \sigma. (a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)})$$

1. $a_0=\text{det } A$

in the product $$(a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)})$$ the only term without an $X$ factor is $$a_{1\sigma(1)}...a_{n\sigma(n)}.$$ Hence the only term without $X$ in factor in $$p_A = \text{det}(A-XI_n) = \sum_{\sigma \in S_n} \text{sign } \sigma. (a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)})$$ is $$\sum_{\sigma \in S_n} \text{sign } \sigma. a_{1\sigma(1)}...a_{n\sigma(n)} = \det A.$$

2. $a_{n-1}=(-1)^{n-1}. \text{tr}(A)$

Developing the product $$(a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)})$$ the term with power $X^{n-1}$ is $$\sum_{i=1}^n a_{i\sigma(i)} (-1)^{n-1}[\delta_{1 \sigma(1)} \dots \delta_{n \sigma(n)}]_i$$ where $[\delta_{1 \sigma(1)} \dots \delta_{n \sigma(n)}]_i$ is the product of the $\delta_{j \sigma(j)}$ for $1 \le j \le n$ with the term $\delta_{i \sigma(i)}$ missing. But $[\delta_{1 \sigma(1)} \dots \delta_{n \sigma(n)}]_i$ is not vanishing if and only if $\sigma$ is the identity. Hence $$\sum_{\sigma \in S_n} \text{sign } \sigma. \sum_{i=1}^n a_{i\sigma(i)} (-1)^{n-1}[\delta_{1 \sigma(1)} \dots \delta_{n \sigma(n)}]_i = (-1)^{n-1} \sum_{i=1}^n a_{ii}=(-1)^{n-1} \text{tr} A$$


Here, developing the product $$(a_{1\sigma(1)}-X\delta_{1\sigma(1)})...(a_{n\sigma(n)}-X\delta_{n\sigma(n)}),$$ the only term with $X^n$ in factor is $(-1)^n \delta_{1 \sigma(1)} \dots \delta_{n \sigma(n)}$ which is not vanishing only if $\sigma$ is the identity. And in that case, you get in the sum $(-1)^n X^n$ as the unique term with $x^n$ in factor.


The term $a_0$ has no factor of $X$. So, look at each summand $(a_{1\sigma(1)} - X \delta_{a\sigma(a)}) \cdots (a_{n\sigma(n)}-X \delta_{n\sigma(n)})$, in each factor of that product choose the first term $a_{i\sigma(i)}$ of that factor and ignore the other term $X\delta_{i\sigma(i)}$. You get $a_{1\sigma(1)}\cdots a_{n\sigma(n)}$. Sum those up with appropriate sign and you get the determinant.

The term $a_{n-1} X^{n-1}$ has $n-1$ factors $X$. So, look at each summand $(a_{1\sigma(1)} - X \delta_{a\sigma(a)}) \cdots (a_{n\sigma(n)}-X \delta_{n\sigma(n)})$, choose $n-1$ factors of that product, in each of those chosen factors pick the $X \delta_{i\sigma(i)}$ term of that factor, and in each chosen factor make sure that $i=\sigma(i)$ or else you won't get the factor of $X$ that you want. It follows that $\sigma$ must be the identity in order to get something nonzero. So (up to sign) the term $a_{n-1} X^{n-1}$ is equivalent to $a_{11} X^{n-1} + a_{22} X^{n-1} + \cdots + a_{nn} X^{n-1} = \text{tr}(A) X^{n-1}$. The sign must be $(-1)^{n-1}$ because the $n-1$ chosen factors $X \delta_{i\sigma(i)}$ each come with a minus sign in front.

If you follow those, I'll leave you to follow $a_n = (-1)^n$.


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.