Suppose $A$ and $B$ are positive definite (symmetric) real matrices.

Is it true that $(AB)^{1/2}+(BA)^{1/2} \geq A^{1/2}B^{1/2}+B^{1/2}A^{1/2}$ ?

EDIT: Shown to be false below. Extension sought: Consider now a square strictly contractive matrix $S$ (such that $I-SS^T>0$) and consider the inequality $(AB)^{1/2}+(BA)^{1/2}\geq A^{1/2}SB^{1/2}+B^{1/2}S^TA^{1/2}$ ? Does this contraction change things, i.e. is the inequality true?

A related question but "in reverse" is asked here: An inequality on the root of matrix products (part 2 - the reverse case)

  • $\begingroup$ What does $ A\le B $ mean for matrices? $\endgroup$ – Braindead Jan 11 '14 at 10:13
  • $\begingroup$ $\geq$ is in the positive-definite sense $\endgroup$ – John U Jan 11 '14 at 10:15
  • $\begingroup$ Sorry, can you clarify? I don't understand what you mean by that. $\endgroup$ – Braindead Jan 11 '14 at 10:19
  • $\begingroup$ $A\geq B$ implies $A-B\geq0$ or $x^T(A-B)x\geq0$ for all non-zero $x$ $\endgroup$ – John U Jan 11 '14 at 10:24
  • $\begingroup$ $AB$ with $A$ and $B$ both symmetric positive definite has positive eigenvalues.. the full result is $AB$ with $A$ positive definite has the same number of positive, negative and zero eigenvalues as $B$ - see Matrix Analysis (Horn et. al.) $\endgroup$ – John U Jan 11 '14 at 14:17

It's not true. If it's true, by continuity, it's also true for positive semidefinite matrices. Yet we have a counterexample in the latter case: \begin{align*} &A=X^2,\ B=Y^2\ \text{ where }\ X=\sqrt{\frac85}\pmatrix{1&2\\ 2&4},\ Y=\pmatrix{1&0\\ 0&\tfrac12},\ \\ &AB=\pmatrix{8&4\\ 16&8}=R^2,\ \text{ where }\ R=\pmatrix{2&1\\ 4&2},\\ &R+R^T=\pmatrix{4&5\\ 5&4},\\ &XY+YX=\sqrt{\frac85}\pmatrix{2&3\\ 3&4}. \end{align*} Neither is $R+R^T\succeq XY+YX$ (the last entry of $R+R^T$ is smaller than the last entry of $XY+YX$) and nor is $XY+YX\succeq0$ (its determinant is negative).

It should be easy to generate by computer a random counterexample for the positive definite case, but obtaining one with nice matrix entries may require a bit more work.

  • $\begingroup$ Thanks.. the positive definite part of my question is countered simply by noting that $x^T(S+S^T)x=2x^TSx$ which need not be $\geq 0$ when $S$ is non-symmetric but with strictly positive eigenvalues $\endgroup$ – John U Jan 11 '14 at 21:19
  • $\begingroup$ I don't know if I should ask another question but I will try here first (also editing the actual question). What if now we have a square strictly contractive matrix $S$ (such that $I-SS^T>0$) and we consider the inequality $(AB)^{1/2}+(BA)^{1/2}\geq A^{1/2}SB^{1/2}+B^{1/2}S^TA^{1/2}$ ? Does this contraction change things? $\endgroup$ – John U Jan 12 '14 at 5:07
  • $\begingroup$ I guess the answer is no by continuity also but I am also seeking an illustration (through algebra not counter-example) if possible.. $\endgroup$ – John U Jan 12 '14 at 5:11
  • $\begingroup$ @JohnU Your guess is correct: the answer is no. Again, if the hypothesis is true, then by a continuity argument, it will also be true for $I-SS^T=0$ and in particular true for $S=I$. So, the counterexample in the above answer still applies. If you are seeking an illustration "through algebra not counter-example" (what does that mean?), please feel free to post a new question. $\endgroup$ – user1551 Jan 12 '14 at 7:08

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.