Waring’s problem for matrices

Probably a well-know question, but I haven't solved it, so I'll ask.

I can show that every matrix in $M_2(\mathbb{R})$ is the sum of two squares of matrices in $M_2(\mathbb{R})$. If $n>2$, I can also show that every matrix in $M_n(\mathbb{R})$ is the sum of three squares of matrices in $M_n(\mathbb{R})$.

So my question is : Is every matrix in $M_n(\mathbb{R})$ is the sum of two squares of matrices in $M_n(\mathbb{R})$ (n>2)?

-
According to Wikipedia, every matrix is the average of two unitaries (en.wikipedia.org/wiki/Unitary_matrix#Properties). – Steve Huntsman Apr 9 2010 at 23:56
That sounds wrong to me. The set of unitary n by n matrices is compact, so the set of averages of two unitary matrices is compact too. So the latter set cannot equal the space of all n by n matrices. – Bjorn Poonen Apr 10 2010 at 0:28
Wikipedia missed the requirement that the norm of the matrix should be ≤ 1. – François G. Dorais Apr 10 2010 at 0:46
It's fixed now. – Steve Huntsman Apr 10 2010 at 1:22
One remark: Waring's problem usually takes place in the context of integers, not real numbers. So one might ask: what can one say about sums of squares of nxn matrices? Here there is a nice story,having to do with Siegel modular forms and powers of the Siegel theta function; see e.g. A.N.Andrianov's book "Quadratic forms and Hecke operators." – JSE Apr 10 2010 at 2:29
show 2 more comments

1 Answer

The answer is YES if $n$ is even. But if $n$ is odd, then the answer is NO since $-I$ is not a sum of two squares. See

Griffin and Krusemeyer, Matrices as sums of squares, Linear and Multilinear Algebra 5 (1977/78), no. 1, 33-44

for the proofs of these facts and generalizations.

-
 Thanks Bjorn, this is useful. – Portland Apr 10 2010 at 1:17