Integrals of the following form show up frequently in Physics, especially in electrodynamics. I have the following integral that I have no idea how to approach solving it:


I can solve each of them separately, or look them up in table integrals, but, the caveat is that separately, the integrals diverge, but as a whole, the integral converges. The only way I have been able to solve this is via Walfram Alpha and only with numbers--that is, replacing $z^2$ by 4 say and trying to figure out what goes where in the final answer.

I would love for someone to guide me as to how I can solve this integral exactly on paper. Additionally, could someone explain why the integral diverges when analyzed separately compared to when analyzed as a whole? This seems to break the general rule:


[Perhaps some condition such as smoothness or continuity of functions is being broken in the presented case, but I am not sure.]

  • $\begingroup$ $dx'$ is the integration variable. $x$, $y$, $z$, and $a$ are constants with respect to this integration. $\endgroup$ – Ptheguy Feb 15 '18 at 18:35
  • $\begingroup$ @Moo that means integrating with respect to $x'$. $\endgroup$ – Botond Feb 15 '18 at 18:36
  • $\begingroup$ You can find the indefinite integral by trigonmetric substitution, but I don't know if that will help you find your improper integral. $\endgroup$ – Jonathan Feb 15 '18 at 18:37
  • $\begingroup$ The splitting rule is not broken, it simply holds iff the two functions are both singularly integrable, that is if the two separated integrals both converge. $\endgroup$ – Turing Feb 15 '18 at 18:37
  • 1
    $\begingroup$ @VonNeumann, Do you mind explaining singularly integrable. Does it mean they each individually converge? $\endgroup$ – Ptheguy Feb 15 '18 at 18:39

To find $$\int \frac{1}{\sqrt{u^2+a^2}} du$$ (a > 0), let $x = a \tan \theta$, Then $dx = a \sec^2 \theta d \theta$, and $\sqrt{u^2+a^2} = a \sec \theta$ (here we are using the identity $\sec^2 \theta = 1 + \tan^2 \theta$). This transforms the integral to $$\int \frac{a \sec^2 \theta}{a \sec \theta} d \theta = \int \sec \theta d \theta = \ln | \sec \theta + \tan \theta| = \ln | \frac{\sqrt{u^2+a^2}}{a}+\frac{u}{a}| + C.$$ Furthermore, properties of logs allow as to slurp the $a$'s in the denominator into the $+C$, so $$\int \frac{1}{\sqrt{u^2+a^2}} du = \ln | \sqrt{u^2+a^2}+u|+C$$

Now if we let $b = \sqrt{(y-a)^2+z^2}$, this and a simple substitution gives that your indefinite integral is equal to $$\ln | \sqrt{(x'-x)^2+b^2}+x'-x|-\ln|\sqrt{x'^2+a^2}+x'| + C = \ln|\frac{\sqrt{(x'-x)^2+b^2}+x'-x}{\sqrt{x'^2+a^2}+x'}| +C.$$ So if $f(x')$ is this expression, we want $\lim_{x' \to \infty} f(x') - \lim_{x' \to -\infty} f(x')$. It's a standard Calc I exercise to show that $\lim_{x' \to \infty} f(x') = \ln(1)=0$. To find $\lim_{x' \to -\infty} f(x')$ multiply the top and bottom by the conjugates of both the top and bottom to get $$f(x') = \ln|\frac{b^2}{a^2} \frac{\sqrt{x'^2+a^2} - x'}{\sqrt{(x'-x)^2+b^2}-(x'-x)}|.$$ Again the Calc I technique shows that $\lim_{x' \to -\infty} f(x') = \ln\frac{b^2}{a^2}$.

So your over answer is $-\ln(\frac{b^2}{a^2}) = 2 \ln(\frac{a}{b})= 2 \ln \frac{a}{\sqrt{(y-a)^2+z^2}}$.

To answer your second question, improper integrals are defined in terms of limits, and the limit law $$\lim_{x \to a} (f(x)+g(x))=\lim_{x\to a} f(x) + \lim_{x \to a} g(x)$$ only holds when both $\lim_{x\to a} f(x)$ and $\lim_{x\to a} g(x)$ exist.

  • $\begingroup$ Thank you for such a detailed explanation $\endgroup$ – Ptheguy Feb 16 '18 at 0:45

FOREOWRD: this answer doesn't represent a proper answer to the above question because instead of $x'^2 + a^2$ I wrote $x'^2 - a^2$. Yet I find it "cool and useful" and I will leave it here the same. I will however provide for another answer about the second terms of the integral.

Before a numerical analysis, it's better to rewrite it in another way.

Let's define:

$$(x-x') = p$$

$$(y-a)^2 + z^2 = k^2$$

$$x'^2 - a^2 = (x'+a)(x'-a) = (x-p+a)(x-p-a) = -(p-a+x)(-p-a-x) = p^2 - c^2$$

Where $c = a+x$


$$-\int_{-\infty}^{+\infty} \left[\frac{1}{\sqrt{p^2 + k^2}} - \frac{1}{\sqrt{p^2-c^2}}\right]dp$$

The indefinite integrals, separately, are trivial:

$$\int \frac{1}{\sqrt{p^2 + k^2}}\ dp = \log \left(\sqrt{k^2+p^2}+k\right)$$

$$\int \frac{1}{\sqrt{p^2-c^2}}\ dp = \log \left(\sqrt{k^2+p^2}+p\right)$$

Which shows you why they are separately divergent.

The the indefinite integral of them both is:

$$-\int \left[\frac{1}{\sqrt{p^2 + k^2}} - \frac{1}{\sqrt{p^2-c^2}}\right]dp = \log \left(\sqrt{p^2-c^2}+p\right) - \log \left(\sqrt{k^2-c^2}+p\right)$$

The problem now is: What are $c$ and $k$?

From how we defined it, $k$ is surely always positive, and it also doesn't cause problems.

But $c$, well $c = a+x$, and we don't know if it's always positive. Also it appears as $-c^2$ hence $p^2 - c^2 \geq 0$ is required to use log unifying property.

We also need $p$ such that the two logarithm arguments will be strictly positive.

Why am I telling you all this? Because log of negative argument "exist" in the comple field, but we couldn't use the usual properties. This is why we cannot say anything a propri, unless we make strong assumptions.

Anyway, assuming it is all good, we can write the result as

$$\log\left(\frac{\sqrt{p^2 - c^2} + p}{\sqrt{p^2 + k^2} + p}\right)$$

Evaluating it as $p\to \pm \infty$ we easily find

$$\to 0 ~~~~~~~ \text{for} ~~ p\to +\infty$$ $$\to \log \left(-\frac{c^2}{k^2}\right) ~~~~~~~ \text{for} ~~ p\to -\infty$$

The final result is anyway complex.

  • $\begingroup$ Don't you want $x'^2+a^2$ instead of $x'^2-a^2$? $\endgroup$ – Jonathan Feb 15 '18 at 19:08
  • 2
    $\begingroup$ @JonathanBrown Damn. I messed all up. $\endgroup$ – Turing Feb 15 '18 at 19:10
  • $\begingroup$ @VonNeumann, so does that mean this solution is not correct? A complex final answer is a bit weird since this integral represents the potential of two infinitely long wires :) $\endgroup$ – Ptheguy Feb 15 '18 at 21:03
  • $\begingroup$ @Ptheguy My final answer is not correct because at the beginning I messed up a sign. That sign is what will make the final result real, as Jonatan Brown showed you here down. I left the answer because for someone it may come helpful (for other purposes), but please: up vote and accept Jonatan's one because it's complete and correct! $\endgroup$ – Turing Feb 15 '18 at 22:27

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.