I have accrossed this integral when I run some of my computation in Wolfram alpha with many values of $t$ , Really seems to conjecture that :

$$\int_{0}^{t}\operatorname{erf}(x+\sqrt{1-\log (x)} )dx \sim t$$ for every $t$ , The key I dea i have used is the series expansion of that function $x+\sqrt{1-\log (x)}$ at $x=e$ as shown in the below image , by substitution in the titled function that's make the integral is very complicated and as a reason we do not have a closed form for it this make obstacle to me to get what i have conjuctred , Now my justification if the conjecture is true is that :$\lim_{x\to \infty}\operatorname{erf}(x+\sqrt{1-\log (x)} ) =\lim_{x\to e}\operatorname{erf}(x+\sqrt{1-\log (x)} )=1 $ , But this is not enough and is not clear at all , Now my question how do i show the titled equality if it is true ?

enter image description here

  • $\begingroup$ What's the meaning of this integral for $t \geq e$? $\endgroup$ – Ian Jul 24 '19 at 22:28
  • $\begingroup$ it's not defined for x greater than e $\endgroup$ – zeraoulia rafik Jul 24 '19 at 22:29
  • 3
    $\begingroup$ You say "for every $t$", that's the objection I am raising. Besides that, what you are really doing is taking an estimate valid for small $x$ which is that for such small $x$ you have $\sqrt{1-\log(x)} \approx +\infty$ so $\mathrm{erf}(x+\sqrt{1-\log(x)}) \approx \lim_{x \to \infty} \mathrm{erf}(x) = 1$ and you get an estimate of $t$ for the integration. As $t \to 0$ this estimate is correct to leading order; for finite $t$ you need to say something about the error in the approximation more carefully (which will be a bit tricky because the inner function is singular). $\endgroup$ – Ian Jul 24 '19 at 22:30
  • $\begingroup$ For t greater than e Wolfram alpha assumed complex closed values to 0 in imaginary part , I said for every to since the imaginary part go to be 0 , and i havn't took any condition where should i compute that integral over R or C $\endgroup$ – zeraoulia rafik Jul 24 '19 at 22:34
  • $\begingroup$ $\operatorname{erf}(x+\sqrt{1-\ln(x)})$ will be very close to $1$. The minimum is $\approx 0.988$. This is why the integral is approximately equal to $t$. $\endgroup$ – Varun Vejalla Jul 24 '19 at 22:42

Too long for a comment.

automaticallyGenerated gave the correct explanation. Just let me elaborate considering $$f(x)=\operatorname{erf}(x+\sqrt{1-\ln(x)})$$ for which $$f'(x)=\frac{2}{\sqrt{\pi }}e^{-\left(x+\sqrt{1-\log (x)}\right)^2} \left(1-\frac{1}{2 x \sqrt{1-\log(x)}}\right)$$ So, the derivative cancels when $$2 x \sqrt{1-\log(x)}=1$$ that is to say when $$1+\frac 12 \log\left(\frac 1 {x^2} \right)=\frac 1 {4x^2}$$ the solution of which being $$x_*=\sqrt{2 W\left(\frac{e^2}{2}\right)}\approx 1.52260$$ where $W(.)$ is Lambert function. $$f(x_*)=0.998762$$


Interesting would be the Taylor expansion around $x=1$; this would be $$f(x)=\text{erf}(2)+\frac{x-1}{e^4 \sqrt{\pi }}-\frac{3 (x-1)^2}{4 e^4 \sqrt{\pi }}-\frac{(x-1)^3}{8 e^4 \sqrt{\pi }}+\frac{137 (x-1)^4}{192 e^4 \sqrt{\pi }}+O\left((x-1)^5\right)$$ making $$\int_0^t f(x)\,dx\approx\left(\text{erf}(2)-\frac{175}{192 e^4 \sqrt{\pi }}\right) t-\frac{35 t^2}{96 e^4 \sqrt{\pi }}+\frac{125 t^3}{96 e^4 \sqrt{\pi }}-\frac{143 t^4}{192 e^4 \sqrt{\pi }}+\frac{137 t^5}{960 e^4 \sqrt{\pi }}$$ which, numerically is $$0.985904 t-0.00376742 t^2+0.0134551 t^3-0.0076963 t^4+0.00147468 t^5$$

| cite | improve this answer | |

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.