8
$\begingroup$

For $F(x)={}_2F_1 (a,b;c;x)$, with $c=a+b$, $a>0$, $b>0$, it has been proved in [1] that $\log F(x)$ is convex on $(0,1)$.

I numerically checked that with a variety of $a,\ b$ values, $\log F(x)$ is not only convex, but also has a Taylor series in x consisting of strictly positive coefficients. Can this be proved?

[1] Generalized convexity and inequalities, Anderson, Vamanamurthy, Vuorinen, Journal of Mathematical Analysis and Applications, Volume 335, Issue 2, http://www.sciencedirect.com/science/article/pii/S0022247X07001825#

$\endgroup$

3 Answers 3

5
$\begingroup$

Here's a sketch of a proof of a stronger statement: the coefficients of the Taylor series for $\log{}_2F_1(a,b;a+b+c;x)$ are rational functions of $a$, $b$, and $c$ with positive coefficients.

To see this we first note that $$\begin{aligned} \frac{d\ }{dx} \log {}_2F_1(a,b;a+b+c;x) &= \frac{\displaystyle \frac{d\ }{dx}\,{}_2F_1(a,b;a+b+c;x)}{{}_2F_1(a,b;a+b+c;x)}\\[3pt] &=\frac{ab}{a+b+c}\frac{{}_2F_1(a+1,b+1;a+b+c+1;x)}{{}_2F_1(a,b;a+b+c;x)}. \end{aligned} $$ Then $$ \begin{gathered} \frac{{}_2F_1(a+1,b+1;a+b+c+1;x)}{{}_2F_1(a,b;a+b+c;x)} = \frac{{}_2F_1(a+1,b+1;a+b+c+1;x)}{{}_2F_1(a,b+1;a+b+c;x)} \\ \hfill\times \frac{{}_2F_1(a,b+1;a+b+c;x)}{{}_2F_1(a,b;a+b+c;x)}.\quad \end{gathered} $$ We have continued fractions for the two quotients on the right. Let $S(x; a_1, a_2, a_3, \dots)$ denote the continued fraction $$\cfrac{1}{1-\cfrac{a_1x} {1-\cfrac{a_2x} {1-\cfrac{a_3x} {1-\ddots} }}} $$ Then $$\begin{gathered}\frac{{}_2F_1(a+1,b+1;a+b+c+1;x)}{{}_2F_1(a,b+1;a+b+c;x)} = S \left( x;{\frac { \left( b+1 \right) \left( b+c \right) }{ \left( a +b+c+1 \right) \left( a+b+c \right) }}, \right.\hfill\\ \left. {\frac { \left( a+1 \right) \left( a+c \right) }{ \left( a+b+c+2 \right) \left( a+b+c+1 \right) }}, {\frac { \left( b+2 \right) \left( b+c+1 \right) }{ \left( a+b+c+3 \right) \left( a+b+c+2 \right) }}, \right.\\ \hfill \left. {\frac { \left( a+2 \right) \left( a+c+1 \right) }{ \left( a+b+c+4 \right) \left( a+b+c+3 \right) }},\dots \right) \end{gathered} $$ and $$\begin{gathered} \frac{{}_2F_1(a,b+1;a+b+c;x)}{{}_2F_1(a,b;a+b+c;x)} =S \left( x,{\frac {a}{a+b+c}}, {\frac { \left( b+1 \right) \left( b+c \right) }{ \left( a+b+c+1 \right) \left( a+b+c \right) }}, \right.\hfill\\ \left. {\frac { \left( a+1 \right) \left( a+c \right) }{ \left( a+b+c+2 \right) \left( a+b+c+1 \right) }}, {\frac { \left( b+2 \right) \left( b+c+1 \right) }{ \left( a+b+c+3 \right) \left( a+b+c+2 \right) }}, \right.\\ \hfill \left. {\frac { \left( a+2 \right) \left( a+c+1 \right) }{ \left( a+b+c+4 \right) \left( a+b+c+3 \right) }}, \dots\right) \end{gathered} $$ The first of these continued fractions is Gauss's well-known continued fraction, and the second can easily be derived from the first. It follows from these formulas that the coefficients of the Taylor series for $\log{}_2F_1(a,b;a+b+c;x)$ are rational functions of $a$, $b$, and $c$ with positive coefficients.

$\endgroup$
3
$\begingroup$

Please note the paper, may be it will be useful:

D. Karp, S.M. Sitnik, Log-convexity and log-concavity of hypergeometric-like functions, Journal of Mathematical Analysis and Applications, Volume 364, Issue 2, P. 384-394.

There is some general result in this paper on positive Taylor coefficients.

$\endgroup$
0
$\begingroup$

Just like to point out that, using the same partial fraction expansions as in the answer posted, I think you can see that for $a=b>0$, $c\ge a$, the coefficients are also positive. So it is not necessary that $c \ge a+b$ if $a=b$, just that $c \ge a$.

Log convexity is true under more general conditions: If $a,b > 0$ and $c\ge\min(a,b)$, then $\log F(x)$ is convex.

The statement is equivalent to showing that $\log F(1-x)$ is convex on $(0,\infty)$.

When $c=\min(a,b)$, $F(x)$ reduces to an elementary function. When $a,b>0$ and $c > \min(a,b)$, using the integral representation of the hypergeometric function, we see that it is a convex combination of completely monotonic functions (a composition of CM function and Bernstein function which is CM). Thus $F(1-x)$ is completely monotonic. Since completely monotonic functions are log convex, the result follows.

$\endgroup$
2
  • $\begingroup$ Does this tell us whether the coefficients of its Taylor series are positive? This is what the OP was hoping to show. Or are you just pointing out a strengthening of the result in [1] mentioned in the initial post? $\endgroup$ Aug 10, 2023 at 2:52
  • $\begingroup$ Yes, sorry I meant the log convexity, not the positive Taylor series. $\endgroup$
    – japalmer
    Aug 10, 2023 at 2:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.