Is the following statement true?

Given a skew symmetric matrix M, among all of its largest rank sub-matrix, there must be one that is the principal submatrix of M.

  • $\begingroup$ It's true, for both symmetric and skew-symmetric matrices, over an arbitrary field (and also hermitian...), and known as "Principal minor theorem". For a generalization, see Theorem 2.9 here (V. Kodiyalam, T. Lam, R. Swan, Determinantal ideals, Pfaffian ideals, and the principal minor theorem. Noncommutative rings, group rings, diagram algebras and their applications, 35-60, Contemp. Math., 456, Amer. Math. Soc., Providence, RI, 2008). I haven't found a more classical reference. $\endgroup$ – YCor Jul 16 at 11:31

Yes. Suppose that you have a general $n \times n$ matrix M. The coefficient of $(-x)^k$ in the characteristic polynomial is the sum of all the principal minors of the matrix of size n-k. From this it follows that the "rank computed by the principal minors" can differ from the actual rank of the matrix only if the matrix is not diagonalizable. Since skew-symmetric matrices are diagonalizable, the result follows.

  • $\begingroup$ Nice, I think this will works for any matrix $M$ whose $0$-eigenspace is the same as the generalized $0$-eigenspace, that is the set of all vectors killed by a power of $M$. $\endgroup$ – Robin Chapman Apr 22 '10 at 14:22
  • 1
    $\begingroup$ Indeed, the "rank by principal minors" computes the multiplicity of 0 in the characteristic polynomial. $\endgroup$ – damiano Apr 22 '10 at 14:30

Another proof: For any $I$ and $J$ two subsets of $\{1,2,\ldots,n\}$ of the same cardinality, let $D(I,J)$ be the minor in rows $I$ and columns $J$. Let $Pf(I)$ be the Pfaffian $\sqrt{D(I,I)}$. We set $D(\emptyset, \emptyset) = Pf(\emptyset) =1$.

Lemma: Every $D(I,J)$ can be written as a quadratic polynomial in the $Pf(K)$'s.

Proof of Lemma: I claim that $$D(I,J) = \sum_{S \subset J \ |J| \ \mbox{even}} (-1)^{|S|/2} \ Pf(I \cup (J \setminus S)) \ Pf(S).$$ (There may be some sign errors.) Every term on the right can be visualized as corresponding to a perfect matching on the indices $J \sqcup I$. Focus on one particular term. Let $(j_1, j_1')$, $(j_2, j_2')$, ..., $(j_r, j_r')$ be the matchings which lie within $J$. This term will appear only when $S$ is a union of some set of these matchings. The total contribution of this term is thus $\sum_{k=0}^r (-1)^k \binom{r}{k}$. This is $1$ for $r=0$ and $0$ for $r>0$.

Application of lemma: Suppose that $Pf(K)=0$ for all $K$ with $|K| > r$. Let $|I|=|J|>r$. Write $D(I,J)$ as a quadratic polynomial in the $Pf(K)$'s. Then, by degree considerations, for every monomial $Pf(K) Pf(K')$ in that quadratic polynomial, one of $Pf(K)$ and $Pf(K')$ is $0$. So imposing that all the principal minors of size $>r$ be $0$ implies that the same is true for non-principal minors.


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.