I want to solve the following integral $$ \int_{0}^{t} ds \sqrt{\frac{s}{t-s}} \operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right) $$ with $a\in\mathbb{R}$ and $t\in\mathbb{R}^+$. I have tried some substitutions and the most promising are
- $u=\sqrt{t-s} \ \left(du=-\frac{1}{2\sqrt{t-s}} \ ds\right)$ so that the integral becomes \begin{align} \int_{0}^{t} ds \sqrt{\frac{s}{t-s}} \operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right) &=-2\int_{\sqrt{t}}^{0}du\ \sqrt{s}\operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right)\\ &=\color{orange}{2\int^{\sqrt{t}}_{0}du\ \sqrt{t-u^2}\operatorname{erfc}\left(\frac{a}{\sqrt{t-u^2}}\right).} \end{align}
- $v=\frac{1}{\sqrt{s}} \ \left(dv = -\frac{1}{2s^{3/2}} \ \ \ ds \right)$ which yields \begin{align} \int_{0}^{t} ds \sqrt{\frac{s}{t-s}} \operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right) &=2\int_{\tfrac{1}{\sqrt{t}}}^{\infty} dv \ v^{3/2} \frac{1}{v\sqrt{t-\tfrac{1}{v^2}}} \ \operatorname{erfc}\left(av\right)\\ &=\color{orange}{2\int_{\tfrac{1}{\sqrt{t}}}^{\infty} dv \ \frac{v^{3/2}}{\sqrt{v^2t-1}} \ \operatorname{erfc}\left(av\right)} \end{align}
Furthermore, I have looked at the table of integrals [1], [2] and [3]. However, so far, no luck. Is there someone who knows a way forward?
Let $s=\frac {a^2}{x^2}$ to make $$\int_{0}^{t}\sqrt{\frac{s}{t-s}} \operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right)\,ds=2a^3\int_{\frac{a}{\sqrt{t}}}^\infty\frac{\text{erfc}(x)}{x^3 \sqrt{t x^2-a^2}}\,dx$$
I do not think that the antiderivative can be computed but the definite integral is
$$I=\int_{0}^{t}\sqrt{\frac{s}{t-s}} \operatorname{erfc}\left(\frac{a}{\sqrt{s}}\right)\,ds=t \,\,G_{2,3}^{3,0}\left(\frac{a^2}{t}| \begin{array}{c} 1,2 \\ 0,\frac{1}{2},\frac{3}{2} \end{array} \right)$$
where appears the Meijer G function. This has been checked numerically.
Let $y=\frac t{a^2}$ and then $$\frac I{a^2}=y \,\,G_{2,3}^{3,0}\left(\frac{1}{y}| \begin{array}{c} 1,2 \\ 0,\frac{1}{2},\frac{3}{2} \end{array} \right)$$
To allow comparison with numerical integration, I give you below a few values $$\left( \begin{array}{cc} y & \frac I{a^2} \\ 0 & 0.00000 \\ 1 & 0.16389 \\ 2 & 0.74897 \\ 3 & 1.55103 \\ 4 & 2.47510 \\ 5 & 3.47872 \\ 6 & 4.53912 \\ 7 & 5.64249 \\ 8 & 6.77981 \\ 9 & 7.94477 \\ 10 & 9.13278 \end{array} \right)$$
Edit
I forgot to mention that this is valid only for $a>0$. I did not find any solution for $a<0$.
What is interesting is the expansion of $\frac I{a^2}$ for large values of $y$. It write $$\frac I{a^2}=\frac \pi 2 y-\frac{4 }{\sqrt{\pi }}\sqrt{y}\Bigg[1+ \sum_{n=1}^\infty (-1)^n \frac {\alpha_n+\beta_n \big[\log(4y)-\gamma\big]} {\delta_n\,y^n}\Bigg]$$ The first $\alpha_n$ make the sequence $$\{5,9,20,193,921,1075,9839,265715,\cdots\}$$ the first $\beta_n$ $$\{3,10,21,180,770,819,6930,175032,\cdots\}$$ the first $\delta_n$ $$\{18,400,4704,248832,7434240,62300160,4644864000,1145485393920,\cdots\}$$ None of them in $OEIS$.
Using this truncated expansion for $y=1$ leads to an absolute error of $7.13\times 10^{-8}$ and, for $y=10$ to an absolute error of $5.33\times 10^{-15}$.