This question has been answered.


We all know that Cauchy-Schwartz inequality works for vector inner product. Also, it works for the matrix innner product. Specifically, the following is the relation between matrix inner product and matrix norms $$\tag{1} |<A,B>|^2\le\|A\|^2\|B\|^2 $$ when $A=cB$, the equality holds. The definition of inner product for matrices is $$\tag{2} <A,B>=\text{vec}(A)^H\text{vec}(B)=\text{tr}(A^HB) $$

If the LHS of (1) is replaced by induced matrix norms and matrix product, does the Cauchy Schwartz inequality still work, like the following induced $L_2$-norm(also known as, spectral norm) $$\tag{3} \|AB\|_2^2\le \|A\|_2^2 \|B\|_2^2 $$ where $\|A\|_2=\sup_{u\neq0}\frac{\|Au\|_2}{\|u\|_2}$. Since $\|A\|_2=\max\{\text{eig(A)}\}=\lambda_{\text{max}}(A)$, if (3) holds, it would imply the following inequality $$\tag{4} \lambda_{\text{max}}(AB)^2\le \lambda_{\text{max}}(A)^2\lambda_{\text{max}}(B)^2 $$

  • 2
    $\begingroup$ Please provide the definition of that norm, and some thoughts on why it might/might not work $\endgroup$ Commented Oct 20, 2021 at 15:38
  • $\begingroup$ This is not the CS inequality, but the result holds. $\endgroup$ Commented Oct 20, 2021 at 15:46
  • 1
    $\begingroup$ This is known as submultiplicativity. $\endgroup$
    – While I Am
    Commented Oct 20, 2021 at 16:06

1 Answer 1


Those types of norms are called submultiplicative. Not every norm is submultiplicative, but every operator norm is.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .