1
$\begingroup$

Let $A$, $B$ be $d$ by $d$ matrices where $A$, $A+B$ are symmetric positive semi-definite matrices and $\text{det}B=0$. $f(t)=(\text{det}(A+tB))^{\frac{1}{d}}$ is my function where $t\in[0,1]$.

If $f(t)\leq f(0)+tf'(0)$, then $f$ is concave over $[0,1]$.

I just asked To be concave on $[0,1]$, $f(t)\leq f(0)+tf'(0)$ is enough? for general function, but turns out there is a counterexample.

$f$ is special here, which is a composition of exponent function and determinent, it could be true. Thanks in advance.

My trial : $f(t)=(\text{det}(A+tB))^{\frac{1}{d}}$ looks like a exponent $\frac{1}{d}$ of a polynomial function of order $d$. then $f(t)=(a_{1}t^{d}+\cdots)^{\frac{1}{d}}$.


Second trial : Let $f(t)=(a_{d}t^{d}+\cdots+a_{0})^{1/d}$.

$tf(1/t)=(\det(tA+B))^{1/d}$, LHS$=(a_{d}+\cdots+a_{0}t^{d})^{1/d}$ take $t=0$, then $a_{d}=\text{det}B=0$.

So $f(t)=(a_{d-1}t^{d-1}+\cdots+a_{0})^{1/d}$

We can see $f$ is smooth. Let's claim $f$ doesn't have zero on $(0,1)$.

Suppose not, i.e. $f(z)=0$ for some $z\in(0,1)$. We know $f(1)=\left(\text{det}(A+B)\right)^{1/d}\geq0$.

Then, we can change $A$ into $A+zB$ then the condition $\tilde{f}(t)\leq\tilde{f}(0)+t\tilde{f}'(0)$ doesn't make sence. So $f$ doesn't have any zero on $(0,1)$.

Here, I can't go farther.

$\endgroup$
  • $\begingroup$ you need special matrices $A,B$ if you want the determinant to be non-negative or positive $\endgroup$ – Calvin Khor Sep 6 '18 at 8:45
  • $\begingroup$ @CalvinKhor Thanks for comment, I edited my question. $\endgroup$ – kayak Sep 6 '18 at 8:48
  • $\begingroup$ Can you provide more context, and the textbook? also either the claim that $f(0) \in \mathbb R^+$ is false, or in fact A is more than pos. semi-def(or I don't like your notation $\mathbb R^+$). $\endgroup$ – Calvin Khor Sep 7 '18 at 18:29
  • $\begingroup$ What do you mean by changing $A$ into $A+zB$, and how did you use the inequality? also, can you tell us what textbook you are using? $\endgroup$ – Calvin Khor Sep 9 '18 at 13:31
  • 1
    $\begingroup$ found it on the arxiv arxiv.org/abs/1705.00331 $\endgroup$ – Calvin Khor Sep 9 '18 at 13:40
1
$\begingroup$

Some headway... Let $C(t) = A+tB$ for $t\in[0,1]$. Note that $C(t)$ is a convex combination of positive semi-definite matrices $A$ and $A+B$, so it is positive semi-definite. Thus, $\det C(t) \ge 0$ and thus $f(t)$ is well defined for every $t\in[0,1]$.

One can attempt to employ the highschool second derivative test. Define $s := 1/d$, and $g(t):= \det C(t)$ for notational brevity. Then we have

$$ f'(t) = \frac{d}{dt} (g(t)^s ) =sg(t)^{s-1} g'(t)$$ and $$ f''(t) = s(s-1) g(t)^{s-2} g'(t)^2 + sg(t)^{s-1} g''(t)$$

Now observe that

so \begin{align}f''(t) &= s(s-1)g(t)^s [\operatorname{tr}(C^{-1} B)]^2 + s g(t)^s\left([\operatorname{tr}(C^{-1} B)]^2 +0-\operatorname{tr}(C^{-1}BC^{-1}B) \right) \\ &= sg(t)^s \left( s [\operatorname{tr}(C^{-1} B)]^2 -\operatorname{tr}((C^{-1}B)^2) \right) \end{align}

Similarly to the linked question, it seems that the given inequality $ f(t) \le f(0) + tf'(0)$, i.e., $$g(t)^s \le g(0)^s + tsg(0)^{s}\operatorname{tr}(A^{-1}B)$$ is not helpful. It seems like we need to know an inequality between traces of $D:=C^{-1}B$ and $D^2$. As the trace is a sum of eigenvalues, it may be useful to note that
$$ s(\operatorname{tr} D)^2 - \operatorname{tr}(D^2) = s(\sum_i \lambda_i )^2 - (\sum_i \lambda_i^2) $$

If one assumes that $A,A+B$ are better behaved, namely positive definite, then $ f > 0$ is smooth on $(0,1)$, so the above has a chance of working. If in addition the dimension is 2, then there is the equality $$ \det M = \frac12((\operatorname{tr}M)^2 - \operatorname{tr}(M^2) ) $$ so we can notice that $$ f''(t) = \frac12g(t)^{1/2} \left( -\frac12\operatorname{tr}((C^{-1}B)^2)\right) < 0$$ which proves the result true in this special case, without using the inequality $f(t) \le f(0) + tf'(0)$. There are also similar inequalities in dimensions $>2$ but I don't know how to use them.

$\endgroup$

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.