7
$\begingroup$

I would like to show that the following function \begin{align} f_a(x)=2^{ax}\frac{\Gamma((a+1)x)}{\Gamma(x)} \end{align} is an increasing function in $x$ for $x \ge 0$ for any fixed $a>0$.

I did some simulations but not sure how to show a proof for this. I wanted to point out that, from simulation, it seems that $\frac{\Gamma((a+1)x)}{\Gamma(x)}$ can be decreasing for values $x=0$.

We can attempt this by showing that the derivative of a logarithm of $f_a(x)$ is positive. Let

\begin{align} g_a(x)= \log (f_a(x)) \end{align}

(here log is base e) and the derivative of $g_a(x)$ is given by \begin{align} \frac{d}{dx} g_a(x)&= \frac{d}{dx} \left( ax \log(2)+ \log (\Gamma( (a+1)x))- \log (\Gamma( x)) \right)\\ &=a \log(2) + (a+1)\psi((a+1)x)-\psi(x) \end{align} where $\psi(x)$ is a digamma function.

Now it remains to show the following inequality for the difference of digamma functions \begin{align} (a+1)\psi((a+1)x)-\psi(x) \ge - a \log(2) . \end{align}

$\endgroup$
  • $\begingroup$ Just prove that $\Gamma((a+1)x)/\Gamma(x)$ is increasing. An increasing times an increasing is increasing. $\endgroup$ – Clayton Jun 29 '18 at 0:34
  • $\begingroup$ Decreasing or increasing? $\endgroup$ – Lord Shark the Unknown Jun 29 '18 at 0:49
  • 1
    $\begingroup$ @Clayton I don't think tha $Γ((a+1)x)/Γ(x)$ is always increasing. Around $x=0$ it can decrease. $\endgroup$ – Lisa Jun 29 '18 at 1:09
  • 1
    $\begingroup$ @Lisa: you’re correct. For some reason, I thought it was for $x\geq1$. $\endgroup$ – Clayton Jun 29 '18 at 3:10
  • 1
    $\begingroup$ We have $$\frac{d}{dx}\log f_a(x) = (\log 2-\gamma)a + a\sum_{n=1}^{\infty} \left[ \frac{1}{n} - \frac{n}{(n+x)(n+(a+1)x)}\right] > 0.$$ This also tells that $$e^{rax}\frac{\Gamma((a+1)x)}{\Gamma(x)}$$ is increasing for any $r \geq \gamma$. $\endgroup$ – Sangchul Lee Jul 5 '18 at 0:54
5
+50
$\begingroup$

We have that $$ r_a(x) = {{\Gamma \left( {\left( {a + 1} \right)x} \right)} \over {\Gamma \left( x \right)}} = {{\Gamma \left( {x + a\,x} \right)} \over {\Gamma \left( x \right)}} = x^{\,\overline {\,a\,x\,} } $$

where $x^{\,\overline {\,y\,} } $ denotes the Rising Factorial.

Now, the Rising Factorial is defined (for $x$ and $y$ real and also complex) as $$ h(x,y) = x^{\,\overline {\,y\,\,} } = {{\Gamma \left( {x + y} \right)} \over {\Gamma \left( x \right)}} = \prod\nolimits_{\;k\, = \,\,0}^{\,y} {\left( {x + k} \right)} $$ where the last term denotes the Indefinite Product, computed for $k$ ranging between the indicated bounds.

So we can write $f_a(x)$ as $$ \bbox[lightyellow] { f_{\,a} (x) = 2^{\,a\,x} {{\Gamma \left( {x + ax} \right)} \over {\Gamma \left( x \right)}} = \prod\nolimits_{\;k\, = \,\,0}^{\,a\,x} {2\left( {x + k} \right)} = 2^{\,a\,x} x^{\,a\,x} \prod\nolimits_{\;k\, = \,\,0}^{\,a\,x} {\left( {1 + k/x} \right)} } \tag{1}$$

Concerning the derivative of $r_a(x)$, since $$ \left\{ \matrix{ {\partial \over {\partial x}}h(x,y) = {{\Gamma \left( {x + y} \right)} \over {\Gamma \left( x \right)}}\left( {\psi \left( {x + y} \right) - \psi \left( x \right)} \right) \hfill \cr {\partial \over {\partial y}}h(x,y) = {{\Gamma \left( {x + y} \right)} \over {\Gamma \left( x \right)}}\psi \left( {x + y} \right) \hfill \cr} \right. $$ then, as you already found $$ \eqalign{ & {d \over {dx}}r_a(x) = {\partial \over {\partial x}}h(x,y) + {\partial \over {\partial y}}h(x,y){d \over {dx}}y = \cr & = {{\Gamma \left( {x + ax} \right)} \over {\Gamma \left( x \right)}}\left( {\left( {a + 1} \right)\psi \left( {x + ax} \right) - \psi \left( x \right)} \right) = \cr & = {{\Gamma \left( {x + ax} \right)} \over {\Gamma \left( x \right)}}\left( {a\psi \left( {x + ax} \right) + \left( {\psi \left( {x + ax} \right) - \psi \left( x \right)} \right)} \right) \cr} $$

where, for $0<x$ (and $0<a$) $\Gamma(x+ax)/\Gamma(x)$ is clearly positive.
However, while $\psi(x+ax)-\psi(x)$ is also positive since $\psi(x)$ is increasing in that range, $a\psi(a+ax)$ introduces a negative term for lower $x$.

To determine the limit of $r_a'(x)$ as $x \to 0^+$, let's consider the series development of $$ \bbox[lightyellow] { \left\{ \matrix{ \ln \Gamma (cx) = \ln \left( {{1 \over {cx}}} \right) - \gamma cx + O\left( {x^{\,2} } \right) \hfill \cr \psi (cx) = - {1 \over {cx}} - \gamma + {{\pi ^{\,2} } \over 6}cx + O\left( {x^{\,2} } \right) = \hfill \cr = - {1 \over {cx}} - \gamma + \sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {1 \over {k + 1 + cx}}} \right)} \hfill \cr} \right. } \tag{2}$$

Therefore $$ \eqalign{ & \mathop {\lim }\limits_{x\; \to \;0^{\, + } } {d \over {dx}}r_a(x) = \mathop {\lim }\limits_{x\; \to \;0^{\, + } } {{\Gamma \left( {x + ax} \right)} \over {\Gamma \left( x \right)}}\mathop {\lim }\limits_{x\; \to \;0^{\, + } } \left( {\left( {a + 1} \right)\psi \left( {x + ax} \right) - \psi \left( x \right)} \right) = \cr & = {1 \over {a + 1}}\left( { - \gamma a} \right) \cr} $$

which is negative for $0<a$.

Concerning $f_a(x)$ instead $$ \eqalign{ & {d \over {dx}}f_{\,a} (x) = \;2^{\,a\,x} a\ln 2r_{\,a} (x) + 2^{\,a\,x} {d \over {dx}}r_{\,a} (x) = \cr & = 2^{\,a\,x} r_{\,a} (x)\left( {a\ln 2 + {d \over {dx}}\ln \left( {r_{\,a} (x)} \right)} \right) = \cr & = 2^{\,a\,x} r_{\,a} (x)\left( {a\ln 2 + {d \over {dx}}\ln \Gamma (x + ax) - {d \over {dx}}\ln \Gamma (x)} \right) = \cr & = 2^{\,a\,x} r_{\,a} (x)\left( {a\ln 2 + \left( {a + 1} \right)\psi (x + ax) - \psi (x)} \right) \cr} $$ (which is the equation you already found)

and $$ \bbox[lightyellow] { \eqalign{ & \mathop {\lim }\limits_{x\; \to \;0^{\, + } } f_{\,a} '(x) = 1\left( {\left( {a\ln 2} \right){1 \over {a + 1}} - {{\gamma a} \over {a + 1}}} \right) = \cr & = {a \over {a + 1}}\left( {\ln 2 - \gamma } \right) = {a \over {a + 1}}0.1159 \cdots \cr} } \tag{3}$$

Proceeding with the development of the derivative above $$ \bbox[lightyellow] { \eqalign{ & {d \over {dx}}f_{\,a} (x)\;\mathop /\limits_{} \;\left( {2^{\,a\,x} r_{\,a} (x)} \right) = \cr & = a\ln 2 + \left( {a + 1} \right)\psi (x + ax) - \psi (x) = \cr & = a\ln 2 + \left( {a + 1} \right)\left( { - {1 \over {\left( {a + 1} \right)x}} - \gamma + \sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {1 \over {k + 1 + \left( {a + 1} \right)x}}} \right)} } \right) - \left( { - {1 \over x} - \gamma + \sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {1 \over {k + 1 + x}}} \right)} } \right) = \cr & = \left( {a\ln 2 - {1 \over x} - \left( {a + 1} \right)\gamma + {1 \over x} + \gamma } \right) + \left( {a + 1} \right)\sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {1 \over {k + 1 + \left( {a + 1} \right)x}}} \right)} - \sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {1 \over {k + 1 + x}}} \right)} = \cr & = a\left( {\ln 2 - \gamma } \right) + \sum\limits_{0\, \le \,k} {\left( {{a \over {k + 1}} + {1 \over {k + 1 + x}} - {{\left( {a + 1} \right)} \over {k + 1 + \left( {a + 1} \right)x}}} \right)} = \cr & = a\left( {\ln 2 - \gamma } \right) + a\sum\limits_{0\, \le \,k} {\left( {{1 \over {k + 1}} - {{k + 1} \over {\left( {k + 1 + x} \right)\left( {k + 1 + \left( {a + 1} \right)x} \right)}}} \right)} \cr} } \tag{4}$$

and $$ \bbox[lightyellow] { \eqalign{ & 0 \le {1 \over {k + 1}} - {{k + 1} \over {\left( {k + 1 + x} \right)\left( {k + 1 + \left( {a + 1} \right)x} \right)}} = \cr & = {1 \over {k + 1}} - {1 \over {\left( {1 + x/\left( {k + 1} \right)} \right)\left( {k + 1 + \left( {a + 1} \right)x} \right)}}\quad \left| {\;0 \le x,k} \right. \cr} } \tag{5}$$ therefore your thesis is demonstrated.

$\endgroup$
4
$\begingroup$

At least for large $x$, considering $$f_a(x)=2^{ax}\frac{\Gamma((a+1)x)}{\Gamma(x)}$$ take logarithms $$\log(f_a(x) )=a x\log(2)+\log (\Gamma ((a+1) x))-\log (\Gamma ( x))$$ Now, use Stirling approximation for $\log(\Gamma(p))$ and continue with Taylor expansion for large $x$ and get $$\log(f_a(x) )=x \left(a \log \left(\frac{2x}{e}\right)+(a+1) \log (a+1)\right)+\frac{1}{2} \log \left(\frac{1}{a+1}\right)-\frac{a}{12 (a+1) x}+O\left(\frac{1}{x^4}\right)$$

$\endgroup$
1
$\begingroup$

Sketch of proof:

The ratio $G^a(x):=\Gamma(x+ax)/\Gamma(x)$ for $a>0$ is a function that has a single minimum for $x>0$, and the location of that minimum in the limit that $a \to 0$ (but not exactly zero) is the root of the equation $x\,\psi^{\,'}(x) + \psi(x)=0.$ Call this value $x_0$ which has the numerical value of approximately 0.216099. For $a>>0$, the minimum of $G^a(x)$ occurs for some $0<x^*<x_0.$ Since we want an inequality valid for $x \ge 0$ it is natural to look at $G_a(x)$ for $x \sim 0.$ A linear approximation is $G^a_1(x) = 1/(1+a) - a/(1+a)\gamma \, x,$ where $\gamma$ is Euler's constant. The linear approximation will undershoot the value $G^a(x^*)$ which is what will ultimately result in an inequality. To counteract the falling trend from 0 to $x^*$ we seek to find $y$ such that $$\frac{d}{dx} y^x\, G^a(x) |_{x=0} = 0 \text{ or its approx } \frac{d}{dx} y^x\, G^a_1(x) |_{x=0} = 0.$$ This is posed as a question involving a derivative because we want to find the best $y.$ Doing the last calculation it is found that $\log{y}=a\, \gamma.$ However $2^{a\,x}> (e^\gamma)^{a\,x}$ since $e^\gamma \sim 1.78$. Thus the proposed inequality is true and can be improved with 2 replaced by $e^\gamma$. To make this argument rigorous will take some work, but it may only need an appeal to Bohr-Mollerup.

$\endgroup$

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.