Let $K$ be an inifinite field of characteristic different from 2. The well-known Amitsur-Levitzki theorem states that the algebra $M_n(K)$ satisfy the standard polynomial identity of degree $2n$, $$s_{2n}(x_1,\dots,x_{2n})=\sum_{\sigma\in S_{2n}}(-1)^{\sigma}x_{\sigma(1)}\cdots x_{\sigma(2n)}$$ Moreover, it does not satisfy any other identity of degree less than $2n$. In particular, if $m < n$, $s_{2m}$ is an identity for $M_m(K)$ and is not an identity for $M_n(K)$.

My question is the following:

If $R$ is a unitary associative noncommutative $K$-algebra that satisfy a polynomial identity, is it true that if $m < n$ then there is an identity of $M_m(R)$ which is not an identity for $M_n(R)$?

In the language of T-ideals, is the inclusion $T(M_n(R))\subset T(M_m(R))$ a proper one?

Equivalently, the PI-equivalence of $M_n(R)$ and $M_m(R)$ imply that $m = n$?

Of course, if the condition that $R$ is a unitary algebra is removed, nilpotent algebras can give counter-examples.

  • 3
    $\begingroup$ The A-L theorem actually works for any commutative ring $K$. The very best theorems do not have hypotheses :-) $\endgroup$ – Mariano Suárez-Álvarez Dec 18 '12 at 15:31
  • 1
    $\begingroup$ A central polynomial for $M_n(R)$ would fit... $\endgroup$ – Ilya Bogdanov Dec 18 '12 at 16:31
  • $\begingroup$ Ilya Bogdanov, the problem is that we do not know if $M_n(R)$ has a central polynomial. If $R$ is commutative, it is ok, but this is not the case here. $\endgroup$ – Thiago Dec 18 '12 at 23:45
  • $\begingroup$ Related question: Does the truth of any statement of real matrix algebra stabilize in sufficiently high dimension? mathoverflow.net/questions/34186/… $\endgroup$ – Joel David Hamkins May 13 '13 at 1:48

This is not true in general. For example, if $R$ is a free associative algebra of rank $>1$, then $M_n(R)$ does not have nontrivial identities for any $n$.

Moreover, I believe there are examples of algebras $R$ such that much stronger condition holds: $M_n(R)$ is isomorphic to $M_m(R)$ for some $m \ne n$. This reminds cancellation problems in the commutative setting, though I cannot provide examples of such algebras.

This is true if, for example, if $R$ is finite-dimensional and prime (even not necessary associative). Then we can note that $M_n(R)$ is prime too, pass to the algebraic closure of the base field, and invoke theorem of Razmyslov that finite-dimensional prime algebras over algebraically closed fields are determined by their identities (Yu.P. Razmyslov, Identities of Algebras and Their Representations, AMS, 1994 (translation from Russian), around p. 30).


I was not careful enough when reading the question, sorry. The question explicitly asks for situation when $R$ is a PI algebra, so the example with a free algebra obviously does not qualify.

I still think this is not true in the whole generality: probably just the condition of being unitary is to weak, one should demand something like (semi)primeness.

I can think of two approaches. First, there is a lot of works about identities of tensor product of algebras, and of $M_n(R)$ in particular (typical results: if $R$ satisfies the standard identity of degree $k$, then $M_n(R)$ satisfies the standard identity of some given degree in terms of $n$ and $k$, see e.g. M. Domokos, Eulerian polynomial identities and algebras satisfying a standard identity, J. Algebra 169 (1994), N3, 913-928 DOI: 10.1006/jabr.1994.1317). Second, perhaps one can do something along the lines of Sections 4 and 5 of arXiv:0911.5414.

| cite | improve this answer | |
  • 1
    $\begingroup$ If $R$ is a ring of row-finite column finite infinite matrices, then $R\cong M_2(R)$. $\endgroup$ – Mariano Suárez-Álvarez May 11 '13 at 10:00
  • 1
    $\begingroup$ Thiago has the assumption that $R$ satisfies a nontrivial polynomial identity. Associativity is, of course, an identity, but, I suspect, he just forgot to assume that $R$ is an associative ring. $\endgroup$ – Misha May 12 '13 at 23:26
  • $\begingroup$ Right, Misha! I assume R to be an associative ring. In particular, Pasha, the free associative algebra of rank >1 is not PI, as well. I will fix this. Thanks for the answers. $\endgroup$ – Thiago May 13 '13 at 0:53
  • $\begingroup$ Mariano, also in this case, $R$ is not PI. $\endgroup$ – Thiago May 13 '13 at 0:56
  • 2
    $\begingroup$ @Thiago, I was just providing an example to Pasha's second paragraph, quite independently of what the question asked for. $\endgroup$ – Mariano Suárez-Álvarez May 13 '13 at 5:08

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.