All matrices and vectors in this post have entries in the field $\mathbb{F}_2$.

Fix some $n \geq 1$. For an $n \times n$ matrix $X$, write $X_0$ for the column vector whose entries are the diagonal entries in $X$. The following curious fact arose in a paper I am writing:

Fact: Let $X$ be a symmetric $n \times n$ matrix and let $A$ be an arbitrary $n \times n$ matrix. Then $(AXA^t)_0 = A (X_0)$. Here $A (X_0)$ means that we are multiplying the column vector $X_0$ by $A$.

This is easy enough to prove in a grungy way, but to me it basically comes out of nowhere. Does anyone know a conceptual reason for it? Or perhaps a bigger picture in which it sits?

  • 4
    $\begingroup$ This likely can be seen as a statement about the representation theory of $G = SL_n(\mathbb{F}_2)$ over $\mathbb{F}_2$. If $\rho$ is the obvious representation, then this says that the symmetric component of $\rho \otimes \rho^t$ further decomposes to a diagonal-only component and a zero-diagonal component. $\endgroup$ – user44191 Dec 12 '18 at 23:27
  • 2
    $\begingroup$ Perhaps it is better to express this in terms of vector spaces: $X$ corresponds to a symmetric bilinear form $b$ on a vector space $V$ over $\mathbb{F}_2$, and $X_0$ corresponds to the linear map $v \mapsto b(v,v)$. Further $A^tXA$ corresponds to the bilinear form $b'$ defined by $b'(v,w) = b(Av, Aw)$, with the associated linear map $v \mapsto b'(v,v)$ etc. Looking at it this way, the equality seems clear. $\endgroup$ – spin Dec 12 '18 at 23:59
  • 5
    $\begingroup$ @user44191 It's a nontrivial extension, not a direct sum. The diagonal zero symmetric matrices are a subspace and the "diagonal only" are a quotient. $\endgroup$ – David E Speyer Dec 13 '18 at 1:35

I think one way of explaining this is via quadratic forms. The usual correspondence sends $X$ to the quadratic form $X(v)=vXv^t,$ $v$ a row vector. But in characteristic $2$, this formula simplifies to


Now examine what happens to $AXA^t$. For any characteristic, we get $AXA^t(v)=vAXA^tv^t=(vA)X(vA)^t=X(vA).$ In characteristic $2$, this implies that $$(v(AXA^t)_0)^2=(vA(X_0))^2,$$



as desired.

You can also phrase this in a representation-theoretic language (following the suggestion of user44191). Take the algebraic group $\operatorname{GL}(n)$ over $\mathbb{Z}.$ It has representations $U$ and $V$ (again over $\mathbb{Z}$) corresponding to $n\times n$ symmetric matrices and quadratic forms on $\mathbb{Z}^n.$ The natural map $U\rightarrow V$ is an isomorphism once you localize away from $2$, but let us examine the map $U/2U\rightarrow V/2V.$ This map is our map $X\mapsto(vX_0)^2,$ which shows that $U/2U$ has the desired quotient.

  • $\begingroup$ This is very helpful. Thanks! $\endgroup$ – Alice Dec 13 '18 at 20:38

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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