I'm looking for a closed solution or an approximation to $$\int x^{-a} \text{erf}\left( b - c x^{-d} \right) dx,$$ where $a, b, c, d > 0$.
-
$\begingroup$ there is no closed-form solution; concerning the approximation, you will want to specify in what limit, which parameters are small? $\endgroup$– Carlo BeenakkerCommented Feb 10, 2022 at 11:17
-
$\begingroup$ Dear @CarloBeenakker, if possible for $x >0$. These are the exact values of a, b, c and d: $a=1.2743$, $b=16.33$, $c=18.7525$, and $d=0.0308$. I hope that helps. $\endgroup$– Felipe Augusto de FigueiredoCommented Feb 10, 2022 at 12:04
-
1$\begingroup$ since $d\ll 1$ you may approximate the integral by $(1-a)^{-1}x^{1-a} \text{erf}(b-c)$, I compared the plots and the agreement is quite good. $\endgroup$– Carlo BeenakkerCommented Feb 10, 2022 at 12:14
-
$\begingroup$ @CarloBeenakker, I see, but I'm checking the limits of $x$ and it is $250 < x < 7e5$. $\endgroup$– Felipe Augusto de FigueiredoCommented Feb 10, 2022 at 12:34
-
$\begingroup$ @CarloBeenakker , it seems the sign is changed. See this matlab script: a = 1.2743; b = 16.33; c = 18.7525; d = 0.0308; gamma_min = 284.9588; gamma_max = 7.1209e+05; f = @(x) (x.^(-a)).*(erf(b - c.*(x.^(-d)))); res1 = integral(f, gamma_min, gamma_max); res2 = ((1./(1 - a)).*(gamma_max.^(1-a)).*erf(b-c)) - ((1./(1 - a)).*(gamma_min.^(1-a)).*erf(b-c)); fprintf(1, 'res 1: %e\n', res1); fprintf(1, 'res 2: %e\n', res2); res 1: 6.210717e-01 res 2: -6.825698e-01 $\endgroup$– Felipe Augusto de FigueiredoCommented Feb 10, 2022 at 13:18
1 Answer
From the comments I understand that the OP seeks an approximation of $$I=\int_{x_1}^{x_2} x^{-a} \text{erf}\left( b - c x^{-d} \right)\, dx$$ for $x_2\gg x_1\gg 1$. A complication which will limit the accuracy of the approximation is that $d\ll 1$. If I ignore that for a moment, and assume all coefficients $a,b,c,d$ are of order unity, then a large-$x$ expansion of the integrand gives the approximation $$I_{\text{appr}}=\int_{x_1}^{x_2}x^{-a}\left(\text{erf}(b)-\frac{2c e^{-b^2} }{x^{d}\sqrt{\pi }}\right)\,dx$$ $$\qquad={x_2}^{-a} {x_1}^{-a} \left(\frac{2 e^{-b^2} c {x_2}^{-d} {x_1}^{-d} \left({x_2} {x_1}^{a+d}-{x_1} {x_2}^{a+d}\right)}{\sqrt{\pi } (a+d-1)}+\frac{\text{erf}(b) \left({x_1} {x_2}^a-{x_2} {x_1}^a\right)}{a-1}\right).$$ The values of interest to the OP are $\{a,b,c,d,x_1,x_2\}=\{1.2743, 16.33, 18.7525, 0.0308, 284.959, 712090\}$ In this case $I=0.621072$ while $I_{\text{appr}}=0.682988$, an error of 10%. If the parameter $d$ is increased slightly to $0.05$ the agreement improves to four decimal places.
-
$\begingroup$ Dear, I've just checked, and as you said, if d is increased, the approximation gets closer to the true value. Would it be possible to add the steps you took in order to find such a aproximation, please? $\endgroup$ Commented Feb 10, 2022 at 16:23
-
1$\begingroup$ added the expansion step $\endgroup$ Commented Feb 10, 2022 at 16:36