5
$\begingroup$

The answer to another question (Upper bound of the fraction of Gamma functions) gave an asymptotic upper bound for an expression with Gamma functions: $$\left(\frac{\Gamma(a+b)}{a\Gamma(a)\Gamma(b)}\right)^{1/a}\!\leq \,C\,\frac{a+b}a, \forall a,b\geq\frac12$$

What is the best possible value for the constant $C$ in that statement?

$\endgroup$

3 Answers 3

2
$\begingroup$

The optimal $C$ is $\mathrm{e}$.

Proof:

We have $$\ln C \ge \ln a - \ln(a + b) + \frac{\ln \Gamma(a + b) -\ln a - \ln\Gamma(a) - \ln\Gamma(b)}{a}.$$

Let $$F(a, b) := \ln a - \ln(a + b) + \frac{\ln \Gamma(a + b) -\ln a - \ln\Gamma(a) - \ln\Gamma(b)}{a}.$$ We have $$\frac{\partial F}{\partial b} = - \frac{1}{a+b} + \frac{\psi(a + b) - \psi(b)}{a} \ge 0 \tag{1}$$ where $\psi(\cdot)$ is the digamma function defined by $\psi(u) = \frac{\mathrm{d} \ln \Gamma(u)}{\mathrm{d} u} = \frac{\Gamma'(u)}{\Gamma(u)}$. The proof of (1) is given at the end.

Fixed $a\ge 1/2$, we have $$G(a) := \lim_{b\to \infty} F(a, b) = \ln a + \frac{ -\ln a - \ln\Gamma(a)}{a} $$ where we have used $$\lim_{b\to \infty} -\ln(a + b) + \frac{\ln \Gamma(a + b) - \ln\Gamma(b)}{a} = 0.$$ (Note: Use $\sqrt{2\pi}\, x^{x-1/2}\mathrm{e}^{-x} \le \Gamma(x) \le \sqrt{2\pi}\, x^{x-1/2}\mathrm{e}^{-x}\mathrm{e}^{\frac{1}{12x}}$ for all $x > 0$.)

We have $$ a^2 G'(a) = a - 1 - a\psi(a) + \ln a + \ln \Gamma(a) \ge 0.\tag{2} $$ The proof of (2) is given at the end.

We have $$\lim_{a\to \infty} G(a) = 1.$$ (Note: Use $\sqrt{2\pi}\, x^{x-1/2}\mathrm{e}^{-x} \le \Gamma(x) \le \sqrt{2\pi}\, x^{x-1/2}\mathrm{e}^{-x}\mathrm{e}^{\frac{1}{12x}}$ for all $x > 0$.)

Thus, the optimal $C$ is $\mathrm{e}$.


Proof of (1):

Using Theorem 5 in [1]: for all $u > 0$, $$\ln u - \frac{1}{2u} - \frac{1}{12u^2} < \psi(u) < \ln u - \frac{1}{2u} - \frac{1}{12(u+1/14)^2},$$ we have \begin{align*} &- \frac{1}{a+b} + \frac{\psi(a + b) - \psi(b)}{a}\\ \ge{}&- \frac{1}{a+b} + \frac{1}{a} \left(\ln (a+b) - \frac{1}{2(a+b)} - \frac{1}{12(a+b)^2}\right)\\ &\qquad - \frac{1}{a}\left(\ln b - \frac{1}{2b} - \frac{1}{12(b+1/14)^2}\right)\\ ={}& \frac{1}{a}\ln(1 + a/b) - \frac{1}{a+b} - \frac{1}{2a(a+b)} - \frac{1}{12a(a+b)^2} + \frac{1}{2ab} + \frac{1}{12a(b+1/14)^2}\\ \ge{}& \frac{1}{a}\left(\ln(1 + a/b) - \frac{a/b}{1 + a/b}\right) + \frac{1}{2ab} - \frac{1}{2a(a+b)} + \frac{1}{12a(b+1/14)^2} - \frac{1}{12a(a+b)^2}\\ \ge{}&0 \end{align*} where we use $\ln(1+x) \ge \frac{x}{1+x}$ for all $x \ge 0$.

We are done.

$\phantom{2}$

Proof of (2):

Using $\Gamma(x) \ge \sqrt{2\pi}\, x^{x-1/2}\mathrm{e}^{-x}$ and $\psi(u) < \ln u - \frac{1}{2u} - \frac{1}{12(u+1/14)^2}$ for all $u > 0$ (Theorem 5 in [1]), we have \begin{align*} &a - 1 - a\psi(a) + \ln a + \ln \Gamma(a)\\ \ge{}& a - 1 - a \left(\ln a - \frac{1}{2a} - \frac{1}{12(a+1/14)^2}\right) + \ln a + \frac12\ln(2\pi) + (a-1/2)\ln a - a\\ ={}& \frac12\ln(2\pi a) - \frac12 + \frac{a}{12(a+1/14)^2}\\ \ge{}& 0. \end{align*}

We are done.


Reference

[1] L. Gordon, “A stochastic approach to the gamma function”, Amer. Math. Monthly, 9(101), 1994, 858-865.

$\endgroup$
0
$\begingroup$

Too long for a comment. You know certainly that the so-called Beta function $B$ ( a classical special function) is defined by $$ B(a,b)=\frac{\Gamma (a)\Gamma (b)}{\Gamma (a+b)}=\int_0^1 t^{a-1}(1-t)^{b-1} dt \quad\text{ for $\Re a>0, \Re b> 0$.} $$ You are thus looking for a lowerbound for the $B$ function. When $a, b$ are real-valued and large the Beta function is equivalent to $$ \sqrt{2π}\frac{a^{a-\frac 12}b^{b-\frac 12}}{(a+b)^{a+b-\frac 12}}. $$ You can also fix $b$ and consider $a$ large to get the equivalent $$ \Gamma (b) a^{-b}. $$ Both asymptotics are obtained from Stirling's formula.

$\endgroup$
0
$\begingroup$

Taking $f(a,b)=\frac{a+b}{a}$ as in the other question, I conjecture that the optimal $C$ is $e$.

One bit of evidence is that the limit of the ratio for $a=cb$ and $b\to\infty$ is $(c+1)^{1/c}$ which converges to $e$ as $c\to\infty$. Also the limit for $a=b^{1/2}$ as $b\to\infty$ is exactly $e$ and the ratio seems to be increasing.

$\endgroup$

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.