I read (page 8 here) that if $A$ and $B$ are rectangular matrices so that the product $AB$ is defined, then $$(1)\quad||AB||_F^2\leq ||A||_F^2||B||_F^2$$

Does that mean that the inequality above also holds when the number of rows of $A$ is larger than the number of columns of $B$? The justification (Cauchy Swartz):

$$||AB||_F^2=\sum_{i=1}^n\sum_{j=1}^k(a_i^\top b_j)^2\leq \sum_{i=1}^n\sum_{j=1}^k||a_i||_2^2||b_j||^2_2=||A||_F^2||B||_F^2$$

does not require $k$ (the number of columns of $B$) to equal $n$ (the number of rows of $A$). Intuitively, you could also add imaginary columns of 0's to $B$, so I can believe the claim. On the other hand, in other places I only see $(1)$ claimed for matrix of the same size and have had a hard time finding it claimed for the more general case (where $A$ $B$ are merely multiplicative) online.


Indeed, the only requirement to have the inequality that you wrote for the Frobenius norm and for arbitrary matrices is that the product $AB$ is defined.

If you are looking for a reference see for example the book Numerical Linear Algebra, by Trefethen and Bau, page 23.

  • $\begingroup$ Thanks! Exactly what I needed $\endgroup$ – user1963 Feb 6 '16 at 12:21

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.