Consider the $n\times n$ matrices given by $A_n=1/n*I_{n\times n}$. For every finite $n$ each $A_n$ has rank $n$ but the limit is the zero matrix, and has rank zero. So the limit of a sequence of rank $n$ matrices need not be of rank $n$. Now I am considering the reverse question.

Consider a sequence $A_k \in \mathbb{R}^{m \times n}$ (or $\mathbb{C}^{m \times n}$). Is it possible to construct a sequence such that each $A_k$ has some rank $d \le \min(m,n)$ but: $\operatorname{rank} \left( \lim \limits _{k \to \infty} A_n \right) >d$?

If not how would one go about proving that matrices cannot jump rank?

  • $\begingroup$ Indeed the rank cannot jump to higher values. To show this for square matrices, one can start with their characteristic polynomials. $\endgroup$ – Did Apr 18 '18 at 14:21
  • $\begingroup$ The rank is lower semicontinuous on the set of all matrices. It can only jump down. $\endgroup$ – Hans Engler Apr 18 '18 at 14:31

The set of matrices of rank (at most) $d$ is closed in the set of $m\times n$ matrices.

You can generalize the proof that the set of singular matrices is closed, by considering the set of matrices of rank less than $d$ as the set $f^{-1}({0})$, for $f\colon \mathbb{R}^{m\times n} \to \mathbb{R}$ defined by $$ f(A) = \sum_{A'} \lvert\det(A')\rvert $$ where the sum if over all $(d+1)\times (d+1)$ sub-matrices. Then $f$ is continuous, and $f(A)$ is non-zero if and only if $A$ has rank at least $d+1$.


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.