Let $M_n^{sa}$ be the space of $n\times n$ complex Hermitian matrices, let $r < n$, and let $E$ and $F$ be (real) linear subspaces of $M_n$ with ${\rm codim}(E) < r^2$ and ${\rm codim}(F) = 1$. Let $V$ be the set of matrices in $E$ whose rank is at most $r$ and suppose $V$ is not contained in $F$. Is $V \cap (M_n\setminus F)$ dense in $V$?

For this to fail, there would have to be some open subset of $V$ which is contained in $F$, while not all of $V$ is contained in $F$.

I assume the answer is yes, but I think this is an algebraic geometry question and I know next to nothing about the subject $\ldots$

  • $\begingroup$ What if $F=E$ ? $\endgroup$ Dec 8, 2015 at 7:35
  • $\begingroup$ Wouldn't then $V$ be contained in $F$? $\endgroup$
    – Dirk
    Dec 8, 2015 at 7:44
  • $\begingroup$ Right, I missed that bit... $\endgroup$ Dec 8, 2015 at 19:14
  • 1
    $\begingroup$ @Nik You might want to add the tag "real algebraic geometry", since the subspace of Hermitian matrices is not a complex subspace (its defining equations involve conjugates). If everything was "complex" then I think $V$ would be irreducible and what you're asking would be true, but since the situation becomes "real", $V$ might be disconnected, and then I'm not sure. $\endgroup$ Dec 8, 2015 at 20:00
  • $\begingroup$ @MattiaTalpo: did it. Yes, I was hoping I could just solve this by quoting some standard algebraic geometry, but since $V$ isn't a complex variety that seemed unlikely ... $\endgroup$
    – Nik Weaver
    Dec 8, 2015 at 20:07

1 Answer 1


Let $n=4$, $r=3$. Let $E = \mathbb{R} \oplus M_3^{sa}(\mathbb{C})$ (viewed as $4\times 4$ matrices which are $0$ in the first row and column except for the $(1,1)$ entry). Then the (real) codimension of $E$ in $M_4^{sa}(\mathbb C)$ is $6<9=r^2$. Let $F\subset M_4^{sa}(\mathbb{C})$ be the space of all matrices with $0$ in the $(1,1)$ entry. Clearly $V$ is not contained in $F$.

Now consider the matrix $X= 0\oplus I_3$. We have $X\in V\cap F$, and small perturbations $Y$ of $X$ within $V$ must of course also lie in $E$, and thus have the form $Y=c\oplus (I_3+A)$ where $c\in \mathbb R$ and $A\in M_3^{sa}(\mathbb C)$, and we have $\|Y-X\|=\text{max}(|c|, \|A\|)$. But now if $\|Y-X\|<1$, then $\|A\|<1$, so $I+A$ has rank 3 which forces $c=0$ and thus $Y\in F$.

  • $\begingroup$ Yes, that is correct. But I have a condition on dimensions that I didn't think I would need which I would now like to add. Can your counterexample be adapted to give $E$ smaller codimension? $\endgroup$
    – Nik Weaver
    Dec 8, 2015 at 21:27
  • $\begingroup$ Yes, I think it's possible to get the codimension condition on $E$; I've edited the answer accordingly. $\endgroup$
    – Mike Jury
    Dec 8, 2015 at 21:47
  • $\begingroup$ Argh, you are right but that's not what I meant ... I'll have to edit again. $\endgroup$
    – Nik Weaver
    Dec 8, 2015 at 22:47
  • 1
    $\begingroup$ Yes, that's what I meant ... I really hope you can't do this ... $\endgroup$
    – Nik Weaver
    Dec 9, 2015 at 1:19
  • 2
    $\begingroup$ I accept defeat. $\endgroup$
    – Nik Weaver
    Dec 9, 2015 at 1:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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