In my research about distribution theory in the topic of probability and statistic, I came across the following integral:
∫∞0erf(1/x)erfc(1/x)xdx I run it using WolframAlpha and as shown here I have got 2Cπ, where C is Catalan's constant. The latter let me to believe that ∫t0erf(1/x)erfc(1/x)xdx have a nice closed form which i didn't Get it using integration by part and series asymptotic of both error function and complementary error function, Any way to get that closed form?
Note: erfc is the complementary error function.
Answer
I'm not sure what definition of the error function you have, but I will use:
erf(x)=2x√π∫10e−x2z2dz,erfc(x)=2x√π∫∞1e−x2y2dy
Now notice that via the substitution x→1x your integral is:
I=∫∞0erf(1/x)erfc(1/x)xdx=∫∞0erf(x)erfc(x)xdx
Using the integral representation of the error function that I mentioned we get:
\require{cancel}I=\left(\frac{2}{\sqrt{\pi}}\right)^2 \int_0^1 \int_1^\infty \int_0^\infty \frac{xe^{-x^2z^2}\cdot \cancel xe^{-x^2y^2}dz}{\cancel x}dx dydz
\overset{\large x^2=t}=\frac{2}{\pi}\int_0^1 \int_1^\infty \int_0^\infty e^{-t(z^2+y^2)}dtdydx=\frac{2}{\pi}\int_0^1 \int_1^\infty \frac{1}{z^2+y^2}dydz
=\frac{2}{\pi}\int_0^1 \frac{\arctan z}{z}dz=\frac{2}{\pi}\operatorname{Ti}_2(1)=\frac{2G}{\pi}
You might try the same approach for your more general integral, although at first sight it won't look that nice.
No comments:
Post a Comment