I asked this question on math.stackexchange:
Does this integral converge when $\frac{1}{p}+\frac{1}{q}\ge1$?
No answers or very useful comments there. May be it is more appropraite for mathoverflow.
Fix a small $\delta>0$ and let $p,q>1$. Consider the integral
$$I(p,q):=\int\limits_{1-\delta}^{1+\delta} \int\limits_{y/2}^{2y}\frac{1}{|y-x|^{\frac{1}{p}}|1-x|^{\frac{1}{q}}} \,\mathrm{d}x\,\mathrm{d}y. $$
I am trying to show that $I(p,q)$ diverges if $\frac{1}{p}+\frac{1}{q}\geq 1$. I am not sure this is even the case ? Any hints on how to handle this?
Remark: This seems to be related to the failure of the Hardy-Littlewood-Sobolev inequality (HLS) at the endpoint $p=1$. HLS reads:
If $1<p,q<\infty$, $f\in L^p$ and
$$Tf(x):=\int_{\mathbb{R}^n} \frac{f(y)}{|x-y|^{\gamma}}dy$$
Then $$\|Tf\|_q\leq \|f\|_p$$ if and only if $$\frac{1}{p}-\frac{1}{q}=1-\frac{\gamma}{n}.$$
Many thanks.
NIntegrate[ 1/RealAbs[y - x]^(1/2)/RealAbs[1 - x]^(2/3), {y, 3/4, 5/4}, {x, y/2, 2 y}, Exclusions -> {y == x}, AccuracyGoal -> 4, PrecisionGoal -> 4]
results in $8.66016$, confirming the convergence. $\endgroup$