Let $\gamma>0$. I would like a nice way to prove that $$\int_{\begin{array}{c} 0\leq s,t\leq1\\ s+t\leq1 \end{array}}t^{\gamma-1}s^{\gamma-1}(1-t-s)^{\gamma-1}dtds=\frac{\Gamma(\gamma)^3}{\Gamma(3\gamma)}.$$

Edit/Remark: zhoraster's answer along with induction allows one to prove the generalized identity when integrating over the simplex $$\int_{\begin{array}{c} 0\leq t_{i}\leq1\\ t_{1}+\cdots+t_{k}\leq1 \end{array}}t_{1}^{x_{1}-1}\cdots t_{k}^{x_{k}-1}\left(1-\sum_{i=1}^{k}t_{i}\right)^{x_{k+1}-1}dt_{1}\cdots dt_{k}=\frac{\Gamma(x_{1})\Gamma(x_{2})\cdots\Gamma(x_{k+1})}{\Gamma(x_{1}+\cdots+x_{k+1})} $$ for any $0<x_i<1$

This integral came about while I was reading through Balog and Friedlander's paper A Hybrid Theorem of Vinogradov and Piatetski-Shapiro. Let $0<\gamma<1$, and let$$P_{\gamma}=\left\{ p\ :\ p=\lfloor n^{1/\gamma}\rfloor\ \text{for some }n\right\} $$ be the set of Piatetski-Shapiro primes. While this set of primes is very sparse, it has many nice properties and it is known that a version of the Prime number theorem holds when $\gamma$ is not too small (see this paper of Heath-Brown) $$\sum_{\begin{array}{c} p\leq x\\ p\in P_{\gamma} \end{array}}1\sim\frac{x^{\gamma}}{\log x}. $$ In the paper of Balog and Friedlander, they prove that for $\gamma$ in some range close to $1$, we have $$\frac{1}{\gamma^{3}}\sum_{\begin{array}{c} p_{1}+p_{2}+p_{3}=N\\ p_{i}\in P_{\gamma} \end{array}}p_{1}^{1-\gamma}p_{2}^{1-\gamma}p_{3}^{1-\gamma}\log p_{1}\log p_{2}\log p_{3}\sim\frac{1}{2}\mathfrak{S}(N)N^{2}, $$ where $\mathfrak{S}(N)$ is a particular singular series. Note that this implies that every sufficiently large integer can be written as the sum of three Piatetski-Shapiro primes, which proves a variant of Goldbach's Ternary conjecture for all sufficient large integers. From this they deduce without proof that $$\sum_{\begin{array}{c} p_{1}+p_{2}+p_{3}=N\\ p_{i}\in P_{\gamma} \end{array}}1\sim\frac{1}{2}\frac{\mathfrak{S}(N)N^{3\gamma-1}}{\left(\log N\right)^{3}}\frac{\gamma^{3}\Gamma(\gamma)^{3}}{\Gamma(3\gamma)}.$$ When I applied partial summation, this deduction reduced to evaluating the integral written above.

  • 2
    $\begingroup$ Is the substitution $t =(1-s)u$ counted as a nice way? Anyway, I got $$\frac{\Gamma(2-\gamma)^3}{\Gamma(6-3\gamma)} $$ for your integral. Letting $\gamma \nearrow +\infty$ correctly reflects the fact that your integral diverges when $\gamma = 2$, which makes me doubt your claim... $\endgroup$ – Sangchul Lee Oct 22 '15 at 13:15
  • $\begingroup$ When you compare zhoraster's integral with your one, you will find that the exponents are all negated. That's why my answer is different. $\endgroup$ – Sangchul Lee Oct 22 '15 at 13:33
  • 1
    $\begingroup$ @SangchulLee: I just noticed that exactly as you posted the comment. (my mistake) I have edited things now. $\endgroup$ – Eric Naslund Oct 22 '15 at 13:34

Check the signs. $$ \int_{\substack{ 0\leq s,t\leq1\\ s+t\leq1}}t^{\gamma-1}s^{\gamma-1}(1-t-s)^{\gamma-1}dt\,ds= \int_0^1 t^{\gamma-1}\int_0^{1-t}s^{\gamma-1}(1-t-s)^{\gamma-1}ds\,dt \\ = \int_0^1 t^{\gamma-1}(1-t)^{2\gamma-1}\int_0^{1}u^{\gamma-1}(1-u)^{\gamma-1}du\,dt = B(\gamma,2\gamma)B(\gamma,\gamma) = \frac{\Gamma(2\gamma)\Gamma(\gamma)^3}{\Gamma(2\gamma)\Gamma(3\gamma)} = \frac{\Gamma(\gamma)^3}{\Gamma(3\gamma)}. $$

  • $\begingroup$ Great, thanks! I thought I was missing simple substitution like this. $\endgroup$ – Eric Naslund Oct 22 '15 at 13:30
  • $\begingroup$ This substitution proves what looks like a $2$-dimensional beta function identity: $$\int_{\begin{array}{c} 0\leq s,t\leq1\\ s+t\leq1 \end{array}}t^{a-1}s^{b-1}(1-t-s)^{c-1}dtds=\frac{\Gamma(a)\Gamma(b)\Gamma(c)}{ \Gamma(a+b+c)} $$ $\endgroup$ – Eric Naslund Oct 22 '15 at 13:37
  • $\begingroup$ @EricNaslund, yes, it can be proved exactly as the usual beta function identity. $\endgroup$ – zhoraster Oct 22 '15 at 13:45
  • $\begingroup$ What is the usual way to prove the beta function identity? Are you suggesting writing $\Gamma(a)\Gamma(b)\Gamma(c)$ as an integral over $\mathbb{R}^3$? $\endgroup$ – Eric Naslund Oct 22 '15 at 13:49
  • 1
    $\begingroup$ @EricNaslund, yes, exactly. Integrate over $x+y+z = t$, getting $B(a,b,c) t^{a+b+c-1} e^{-t}$. Then integrate over $t$, which gives $B(a,b,c)\Gamma(a+b+c)$. $\endgroup$ – zhoraster Oct 22 '15 at 14:17

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.