Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Suppose we have fields $\mathbb{K}_1, \mathbb{K}_2$ s.t. $\mathbb{K}_1 \subset \mathbb{K}_2$. On a paper I'm reading there is a flashy reference to some algebraic results concerning similitude of matrices over $\mathbb{K}_1$ when viewed as matrices over $\mathbb{K}_2$. The paper tells nothing more, so I'm left wondering what exactly those results are. The simplest thing I can think of is the following.

Question Let $A, B$ be square $n\times n$ matrices over $\mathbb{K}_1$. If there exists $P_2 \in \mathrm{GL}(n, \mathbb{K}_2)$ s.t.

$$ P_2^{-1}AP_2=B, $$

then there exists $P_1 \in \mathrm{GL}(n, \mathbb{K}_1)$ s.t. $$ P_1^{-1}AP_1=B. $$ Is it true?

share|cite|improve this question
up vote 4 down vote accepted

Here is a somewhat more elementary proof that works if the field $\mathbb{K}_1$ is sufficiently big (for example it works with infinite fields). Assume $A$ and $B$ are similar over $\mathbb{K}_2$, so you have $P \in \textrm{GL}_n(\mathbb{K}_2)$ such that

$$P A = B P$$

Now write $P = (p_{i,j})$, and pick a basis $(e_1, \ldots, e_r)$ of $\textrm{Vect}_{\mathbb{K}_1}(p_{i,j})$ as a $\mathbb{K}_1$-vector space. We can thus write

$$P = \sum_{i = 1}^r e_i P_i$$

with $P_i \in M_n(\mathbb{K}_1)$ for $i \in [\!|1,r|\!]$ (note that the $P_i$ need not be invertible in general). And because $(e_1, \ldots, e_r)$ is free, we get $P_i A = B P_i$ for all $i \in [\!|1,r|\!]$. Now consider the polynomial

$$f(X_1, \ldots, X_r) = \det \left(\sum_{i = 1}^r X_i P_i \right) \in \mathbb{K}_1[X_1, \ldots, X_r]$$

Since $P$ is invertible, we have $f(e_1, \ldots, e_r) \neq 0$, so $f$ is non zero. And if $|\mathbb{K}_1| > n$, then there exist $(\lambda_1, \ldots, \lambda_r) \in \mathbb{K}_1^r$ such that $f(\lambda_1, \ldots, \lambda_r) \neq 0$ (see lemma below for an explanation). So the matrix $P' = \sum_{i = 1}^r \lambda_i P_i$ is in $\textrm{GL}_n(\mathbb{K}_1)$ and satisfies $A = P'^{-1} B P'$.

Lemma : Let $f \in \mathbb{K}[X_1, \ldots, X_r]$ be a non zero polynomial. Assume

$$|\mathbb{K}| > \deg(f) = \max_{1 \le i \le r} (\deg_{i}(f))$$

Then there exist a point $(\lambda_1, \ldots, \lambda_r) \in \mathbb{K}^r$ such that $f(\lambda_1, \ldots, \lambda_r) \neq 0$.

Proof : We will prove this by induction on the number $r \ge 1$ of variables. The case $r = 1$ follows from the fact that a polynomial of degree $d$ has at most $d$ roots. If $r \ge 2$, write $d = \deg_{r}(f)$ and

$$f = \sum_{k = 0}^{d} a_k X_r^k$$

where the $a_i$ are polynomials in $r-1$ variables with $\deg(a_i) \le \deg(f)$. Now $a_d$ is non zero, so by induction, there is a point $(\lambda_1, \ldots, \lambda_{r-1}) \in \mathbb{K}_1^{r-1}$ such that $a_{d}(\lambda_1, \ldots, \lambda_{r-1}) \neq 0$. Finally, since $g(X) = f(\lambda_1, \ldots, \lambda_{r-1}, X)$ is a non zero polynomial with $\deg(g) \le \deg(f)$, you can find $\lambda_n$ such that $g(\lambda_n) \neq 0$. Which concludes.

share|cite|improve this answer
Sorry, I don't understand what $\text{Vect}_{\mathbb{K}_1}(p_{i,j})$ is supposed to be. – Qiaochu Yuan Jun 7 '11 at 16:28
@Qiaochu: I understand $\mathrm{Vect}_{\mathbb{K}_1}$ as the vector subspace of $\mathbb{K}_2$ (regarded as a $\mathbb{K}_1$-vector space) spanned by $(p_{1, 1}, p_{1, 2} \ldots p_{n, n})$. $$$$ @Joel: I like very much this proof. It couldn't be simpler than that. The only thing I can't understand is that "if $\mathbb{K}_1$ is sufficiently big...". Why? We have a polynomial $f$ and know that it is not null, because in an appropriate extension we have $f(e_1 \ldots e_r)\ne 0$. So we must have $f(\mathrm{something})\ne 0$ for $\mathrm{something} \in \mathbb{K}_1$... Where am I wrong? – Giuseppe Negro Jun 7 '11 at 16:50
@dissonance: take $\mathbb{K}_1 = \mathbb{F}_p$ and, for example, $f(x_1) = x_1^p - x_1$. For a homogeneous example take $f(x_1, x_2) = x_1^p x_2 - x_2^p x_1$. – Qiaochu Yuan Jun 7 '11 at 16:55
@Qiaochu: Ok, now I see. Thank you. Can we say that Joel's proof works only if $\mathrm{char}(\mathbb{K}_1)$ is greater than some lower bound? And, if affirmative, how big is this lower bound? – Giuseppe Negro Jun 7 '11 at 17:05
@dissonance : The assumption is not on the characteristic of the field, but rather on its cardinality. By sufficiently large, I mean that if $|\mathbb{K}_1| > n$ (where $n$ is the size of the matrices), the proof should work. I included an explanation in the answer above. – Joel Cohen Jun 7 '11 at 22:45

Yes. It is true. $A$ and $B$ are similar iff $\lambda I-A$ and $\lambda I-B$ are equivalent iff they have the same invariant factors or the same determinate factors. These are only depend on matrix itself.

see Similar matrix

share|cite|improve this answer
See also rational canonical form – Aaron Jun 7 '11 at 16:57

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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