Canonical form of symmetric integer matrix M - MathOverflow most recent 30 from http://mathoverflow.net2013-05-22T03:48:22Zhttp://mathoverflow.net/feeds/question/97448http://www.creativecommons.org/licenses/by-nc/2.5/rdfhttp://mathoverflow.net/questions/97448/canonical-form-of-symmetric-integer-matrix-mCanonical form of symmetric integer matrix MXiao-Gang Wen2012-05-20T01:09:45Z2012-05-20T22:29:42Z
<p>Let $M$, $N$ be a symmetric matrix over a ring $R$.
$M$ and $N$ are said to be equivalent if there exist an invertible
matrix $U$ (over the same ring $R$) such that $N=U M U^T$ ($U^T$ is the transpose of $U$).
A question is that what is the simple canonical form of $M$
under such an equivalent relation.</p>
<p>We know that when $R$ is the ring of real numbers, every
real symmetric matrix is equivalent to
an diagonal matrix with diagonal entries being 1, -1, or 0.</p>
<p>When $R$ is the ring of integers, do we have a similar result?</p>
<p>If there is no nice results, we may assume $M$
to satisfy additional conditions:</p>
<p>(a) $|\det(M)|=1$</p>
<p>(b) There exist a $J$ such that $J^2=1$ and $JMJ^T=-M$.</p>
<p>Thanks!</p>
<p>Edit: I am also interested in finding the simple canonical form of integer symmetric matrices $M$, that satisfy</p>
<p>(a) $|\det(M)|=1$</p>
<p>(b) There exist a $J$ such that $J^2=-1$ and $JMJ^T=-M$.</p>
http://mathoverflow.net/questions/97448/canonical-form-of-symmetric-integer-matrix-m/97451#97451Answer by Will Jagy for Canonical form of symmetric integer matrix MWill Jagy2012-05-20T02:06:15Z2012-05-20T02:06:15Z<p>I don't see how to diagonalize the quadratic form $2xy.$ As far as your conditions, we have
$$ M = \left( \begin{array}{cc}
0 & 1 \\
1 & 0<br>
\end{array}
\right) $$ and
$$ J = \left( \begin{array}{cc}
1 & 0 \\
0 & -1<br>
\end{array}
\right), $$ with
$$ \left( \begin{array}{cc}
1 & 0 \\
0 & -1<br>
\end{array}
\right)
\left( \begin{array}{cc}
0 & 1 \\
1 & 0<br>
\end{array}
\right)
\left( \begin{array}{cc}
1 & 0 \\
0 & -1<br>
\end{array}
\right) =
\left( \begin{array}{cc}
0 & -1 \\
-1 & 0<br>
\end{array}
\right)
$$
However, the only diagonal matrices with determinant $-1$ are $\pm J,$ which does not work as $x^2 - y^2$ does not represent any numbers congruent to $2 \pmod 4.$ </p>
<p>Let's see, Conway and Sloane refer to Watson for his 2-adic canonical form for their work on the Mass Formula, so I can recommend the book Integral Quadratic Forms by George Leo Watson. More recent, SPLAG, which is Sphere Packings, Lattices, and Groups by Conway and Sloane. </p>
http://mathoverflow.net/questions/97448/canonical-form-of-symmetric-integer-matrix-m/97457#97457Answer by Igor Rivin for Canonical form of symmetric integer matrix MIgor Rivin2012-05-20T03:37:23Z2012-05-20T03:37:23Z<p>For symplectic unimodular symmetric (or skew) matrices, such a result is <a href="http://dl.dropbox.com/u/5188175/zarrow.pdf" rel="nofollow">shown in</a>
Zarrow, Robert
A canonical form for symmetric and skew-symmetric extended symplectic modular matrices with applications to Riemann surface theory.
Trans. Amer. Math. Soc. 204 (1975), 207–227. </p>
<p>You might be able to extend it to the nonsymplectic case (though I am a bit skeptical).</p>
http://mathoverflow.net/questions/97448/canonical-form-of-symmetric-integer-matrix-m/97458#97458Answer by Will Jagy for Canonical form of symmetric integer matrix MWill Jagy2012-05-20T05:00:04Z2012-05-20T22:29:42Z<p>It's all in the correct reference. Cassels, <em>Rational Quadratic Forms</em>, chapter 9 "Integral Forms over the Rational Integers," pages 163-164, Examples 9-11. Example 11(i) says that, for "odd" matrices, we can cut down the dimension by 2 and write $y_1^2 - y_2^2 + g(z_1, \ldots , z_{n-2}).$ The determinant of $g$ is still $\pm 1,$ so the only problem is that $g$ may be "even." </p>
<p>Next, if $f$ is "even" the quadratic form can, in fact, be written $ 2y_1 y_2 + g(z_1, \ldots , z_{n-2}).$</p>
<p>So, all we really need is to show, as in Sylvester's Law of Inertia, that the resulting form $g$ continues to be indefinite. Presumably your condition with $J M J^T = -M$ can do this. </p>
<p>Otherwise, without your $J$ condition, Example 11(vi) says that either $f$ or $-f,$ if "even," is equivalent to a sum of some $2x_j y_j$ terms along with a single $\mathbb E_8$ lattice. <a href="http://store.doverpublications.com/0486466701.html" rel="nofollow">CASSELS</a> </p>
<p>I was uneasy about the possible need to mix 2 by 2 blocks of both types, despite Hahn's statement, but
$$<br>
\left( \begin{array}{cccc}
3 & 4 & 2 & 2 \\
2 & 3 & 1 & 2 \\
0 & 1 & 1 & 1 \\
2 & 3 & 2 & 1<br>
\end{array}
\right)
\left( \begin{array}{cccc}
1 & 0 & 0 & 0 \\
0 & -1 & 0 & 0 \\
0 & 0 & 0 & 1 \\
0 & 0 & 1 & 0<br>
\end{array}
\right)
\left( \begin{array}{cccc}
3 & 2 & 0 & 2 \\
4 & 3 & 1 & 3 \\
2 & 1 & 1 & 2 \\
2 & 2 & 1 & 1<br>
\end{array}
\right) =
\left( \begin{array}{cccc}
1 & 0 & 0 & 0 \\
0 & -1 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 0 & 0 & -1<br>
\end{array}
\right)
$$</p>