5

In calculating the integral

$$\int_{-\infty}^{\infty} \frac{\sin x}{x}dx$$

by contour integration, we use

$$\int_{-\infty}^{\infty} \frac{\sin x}{x}dx = \int_{-\infty}^{\infty} \frac{\operatorname{Im}(e^{ix})}{x}dx =\operatorname{Im}\left(\int_{-\infty}^{\infty} \frac{e^{ix}}{x}dx \right)$$

but in the process, we have gone from an integral which is well-defined with no real singularities to one with a real singularity which in fact is just undefined as an improper integral. Therefore, in the source I am reading, we take the cauchy principal value (CPV) of the integral on the RHS instead of treating it as an improper integral. This principal value is calculated by use of the Residue Theorem.

My question: There are different ways to treat singularities in integrals. How do we know that this one (the CPV) will give us the correct result for $\int_{-\infty}^{\infty} \frac{\sin x}{x}dx$? Of course, knowing the answer by other methods, we can compare and see it was correct, but I'm looking to understand why the reasoning is valid.

Response to 1st answer: Simply saying that the integral converges is not enough. We need some way to know that in particular the CPV is the correct notion of integration for the exponential integral. Clearly, not any notion of integration which converges must give the correct result.

Response to 2nd answer: The question I ask is: by what reasoning is the notion of CPV in $\int_{-\infty}^{\infty} \frac{\sin x}{x}dx =\operatorname{Im}\left(CPV\int_{-\infty}^{\infty} \frac{e^{ix}}{x}dx \right)$ justified. Of course the first integral is the same as an improper integral or CPV, this just doesn't answer the question.

  • The cauchy principle value is the same thing really as evaluating the indefinite integral and taking the limit, which is really the same as when we have bounds of $\infty$ anyway – Henry Lee Apr 27 '21 at 19:10
  • I'm not sure what you mean. Both are defined in terms of a limit. But the CPV has one limit, and the improper integral has two independent limits which approach the singularity from both sides. They don't give equivalent results for an integral like $\int_{-1}^1 \frac{1}{x} dx$ and many others. – doublefelix Apr 28 '21 at 04:37
  • That is a different case since you are integrating over a discontinuity, whilst in the integral we are talking about the limit of the function does exist, we are just trying to justify the limit of its integral – Henry Lee Apr 28 '21 at 11:44
  • The CPV in this question is for the integral of $e^{ix}/x$, which does have a singularity at 0. The bounds at $\pm \infty$ make no difference because the tails of integral are convergent. I think you did not thoroughly read the question – doublefelix May 10 '21 at 15:00
  • The integral is discussed here https://math.mit.edu/~jorloff/18.04/notes/topic9.pdf – Gribouillis May 23 '21 at 08:05
  • @Gribouillis Thanks for the link, it is a nice one. They also only say that since the CPV is convergent it must be correct, which is Learn'd Astonomer's answer. But clearly not any convergent notion of integration must give the correct answer. So more is needed. – doublefelix May 23 '21 at 10:22

5 Answers5

2

In order to say that the Cauchy principal value agrees with the ordinary value of an improper integral you only need to know that the integral coverges, which you may show using a convergence test. This is because the Cauchy principal value is simply a particularly easy way of taking the limit. The Cauchy principal value takes the limit symmetrically. If the limit exists, it means you can take it in any manner you like, for example symmetrically.

  • I think you may have misinterpreted, I'm not asking if CPV(integral) = improper(integral). The question is whether improper(integral1) = Im(CPV(integral2)). – doublefelix Apr 26 '21 at 17:52
  • The answer is yes. That follows from what I have written and the fact that $\Im(\int f , dz) = \int \Im(f) , dz$. – Thusle Gadelankz Apr 26 '21 at 18:06
  • But could we not use the same logic to say that $Improper \int \frac{sinx}{x}dx=Im[Improper \int \frac{e^x}{x}dx]=undefined$? It seems only one of these lines of reasoning should be valid, and not the other. – doublefelix Apr 26 '21 at 18:28
  • The reason that the integral you are referring to is not defined is that its real part does not converge. This is because $\frac{\cos(x)}{x}$ is very badly behaved near 0, whereas $\frac{\sin(x)}{x}$ had a limit. That is exactly why we have to take the Cauchy principal value. However, since you know that the real part converges (by some convergence test of your choosing), you know that the CVP gives the correct result of the reason above, also explaines by saulspatz. – Thusle Gadelankz Apr 26 '21 at 19:03
  • I'm not convinced by this reasoning; just because you can show that an integral is convergent to some value doesn't mean that the correct value is the CPV. To clarify my question, I added two contradictory lines of reasoning in an edit. One or both of them must be wrong, but it is not clear to me which one and why. – doublefelix Apr 28 '21 at 04:40
  • You are wrong in writing that the convergence of the integral does not guarantee that the CPV is the correct value. You should think about my and saulspatz’ arguments and the definition of convergence for improper integrals. As to your lines, line 1 is right and line 2 is clearly wrong since you have equated a convergent integral with a divergent one. You need to the CPV. – Thusle Gadelankz Apr 28 '21 at 05:20
  • It is incorrect to say that any convergent integral must be a cauchy principal value. There are other notions of regularizing integrals which give convergent results. I know that line 2 is wrong, because by other methods the value of this integral can be calculated and it agrees with the CPV. My question is, what is wrong with the reasoning in line 2, and why does it work in line 1? – doublefelix Apr 28 '21 at 23:07
  • I have explained this. This follows from the additive property and tha fact when a limit converges, you are free to choose the way in which you evaluate yiur limit (symmetrically in the case of CPV). You should really stop and think about this. This discussion has now become rather too lengthy for a comment section. Maybe your analysis professor will explain this better. – Thusle Gadelankz Apr 29 '21 at 04:37
  • Yes but the exponential integral does not converge, so there we would not be free to choose which way to assign a value to the integral. For example here another way of getting a finite value from a divergent improper integral is discussed by introducing a parameter $\epsilon$. In general there can be more than one way. How do we know that, in this case, the CPV of the exponential integral is the correct notion of integration to give the proper value for the sinx/x integral, when there may be more than one regularization possible? – doublefelix Apr 29 '21 at 05:09
  • Either my question was misunderstood, or I'm just being very thick-headed, but so far I still feel it was misunderstood. – doublefelix Apr 29 '21 at 05:10
  • I think the issue I have with your answer is, in brief, that not every convergent notion of integration is equal to the CPV. So just showing that the result should be finite does not prove it should be given by the imaginary part of specifically a CPV integral. I appreciate the help nonetheless, but I think the question is not yet satisfactorily answered. – doublefelix May 10 '21 at 15:03
2

Since this is a question about rigor we should state what integral we are using. From (1) This integral is not Lebesgue integrable as it is not absolutely convergent, it is an improper Riemann integral though so we will use the Riemann integral.

Let us have a visual of the function being integrated

enter image description here


When we state the purely real integral $$I_1 = \int_{-\infty}^{\infty} \frac{\sin x}{x}dx$$ this is short-hand for $$I_1 = \lim_{a \to -\infty} \lim_{b \to \infty} \int_{a}^{b} f(x) dx$$ where $$f(x) = \begin{cases} \frac{\sin(x)}{x} & x \not = 0 \\ 1 & x = 0 \\ \end{cases}$$

The convergence of this integral must be addressed. It does not converge absolutely so we should attempt to cancel parts of the integral with itself. We will show that the right side (from $2\pi$ to $\infty$) converges by chopping it up and subtracting the negative part of each sine wave from the positive part. For each segment we have

$$\begin{align} &\, \left|\int_{\pi 2 k}^{\pi (2 k + 2)} \frac{\sin(x)}{x} dx\right| \\ \le&\, \left|\int_{\pi 2 k}^{\pi (2 k + 1)} \frac{\sin(x)}{x} - \frac{\sin(x)}{x + \pi} dx\right| \\ \le&\, \left|\int_{\pi 2 k}^{\pi (2 k + 1)} \frac{\sin(x)}{\pi (2 k + 1)} - \frac{\sin(x)}{\pi (2 k + 2)} dx\right| \\ \le&\, \left|\int_{\pi 2 k}^{\pi (2 k + 1)} \frac{\sin(x)}{\pi} \left[\frac{1}{2 k + 1} - \frac{1}{2 k + 2}\right] dx\right| \\ \end{align}$$

and the alternating harmonic series converges. Note that each segment is positive and their sum converges absolutely. Note that we showed the integral from $0$ to $\infty$ converges independently of the integral from $-\infty$ to $0$.

The finite part from $0$ to $2 \pi$ easily converges as we always have $\sin(x)/x \le 1$.


When we state a contour integral like

$$I_2 = \mathrm{p.v.} \int_\gamma \frac{e^{i z} - e^{-i z}}{2 i \cdot z} dz$$

($\gamma$ denoting a line from $-\infty$ to $\infty$).

the meaning (2) is that the we delete from the contour $\gamma$ an $\epsilon$ sized ball around the singularity and take the limit of $\epsilon$ to 0.

So $$I_2 = \lim_{\epsilon \to 0} \lim_{a \to -\infty} \lim_{b \to \infty} \left(\int_a^{-\epsilon} + \int_{\epsilon}^b\right) \frac{e^{i z} - e^{-i z}}{2 i \cdot z} dz$$

Regarding convergence of $I_2$, this integral (or half of it, to be precise) drops out algebraically from an application of Cauchy's Theorem (that a contour integral around a closed curve not containing poles gives 0) to the meromorphic function $e^{iz}/z$. The details are here.


Now to show $I_1 = I_2$. The single exceptional point at $x=0$ has no bearing on the value of the integral $I_1$ so we may split $I_1$ around 0 and add a $\lim_{\epsilon \to 0}$ on it. Now $I_1 - I_2 = 0$ can be shown. Rewrite $I_2$ to use the complex $\sin$ function.

$$\begin{align} I_1 - I_2 =& \left[\lim_{a \to -\infty} \lim_{b \to \infty} \int_{a}^{b} f(x) dx \right] - \left[\lim_{\epsilon \to 0} \lim_{a \to -\infty} \lim_{b \to \infty} \left(\int_a^{-\epsilon} + \int_{\epsilon}^b\right) \frac{e^{i z} - e^{-i z}}{2 i \cdot z} dz\right] \\ =& \left[\lim_{\epsilon \to 0} \lim_{a \to -\infty} \lim_{b \to \infty} \left(\int_a^{-\epsilon} + \int_{\epsilon}^b\right) \frac{\sin(x)}{x} dx \right] - \left[\lim_{\epsilon \to 0} \lim_{a \to -\infty} \lim_{b \to \infty} \left(\int_a^{-\epsilon} + \int_{\epsilon}^b\right) \frac{\sin(z)}{z} dz\right] \\ =& \lim_{\epsilon \to 0} \lim_{a \to -\infty} \lim_{b \to \infty} \left(\int_a^{-\epsilon} + \int_{\epsilon}^b\right) \left[\frac{\sin(x)}{x} - \frac{\sin(x)}{x} \right] dx \\ =& 0 \end{align}$$


  • 1
    I spent a fair amount of time reading the answer, but the big picture is: it doesn't seem to address the question. Let me explain: In the method of this answer, the first step is essentially to write $\sin(z) = \frac{e^{iz}-e^{-iz}}{2i}$. Then we calculate the integral (steps are shown in the other post you linked). This circumvents the method in my question by doing it a different way. Nonetheless it is interesting to learn that it can be done in the way you write about. The method you used seems fully legitimate and I don't have any questions about it, unlike the one in the OP. – doublefelix May 24 '21 at 18:33
2

Why the Cauchy Principal Value Gives the Proper Value

Your question boils down to the following $$ \begin{align} &\mathrm{Im}\left(\mathrm{PV}\int_{-\infty}^\infty\frac{e^{ix}}x\,\mathrm{d}x\right)\\ &=\mathrm{Im}\left(\mathrm{PV}\int_{-\infty}^\infty\frac{\cos(x)}x\,\mathrm{d}x+i\,\mathrm{PV}\int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x\right)\tag{1a}\\ &=\mathrm{Im}\left(\lim_{\epsilon\to0}\int_{|x|\gt\epsilon}\frac{\cos(x)}x\,\mathrm{d}x+i\int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x\right)\tag{1b}\\ &=\mathrm{Im}\left(0+i\int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x\right)\tag{1c}\\ &=\int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x\tag{1d} \end{align} $$ Explanation:
$\text{(1a)}$: $e^{ix}=\cos(x)+i\sin(x)$ and $\mathrm{PV}$ is linear
$\text{(1b)}$: definition of $\mathrm{PV}$ on the cosine integral
$\phantom{\text{(1b):}}$ $\mathrm{PV}$ of a convergent integral is that integral
$\text{(1c)}$: the integral of an odd function over a domain
$\phantom{\text{(1c):}}$ symmetric about the origin is $0$
$\text{(1d)}$: take the imaginary part

Actually, it doesn't even matter what the Principal Value of the cosine integral is. As long as it exists and is real, it is eliminated by taking the imaginary part.


Avoid Singularities Altogether

There is no singularity whatsoever in $$ \int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x\tag2 $$ The way to compute this integral and never come close to a singularity is to note that $$ \lim_{R\to\infty}\int_{\gamma_R}\frac{\sin(z)}z\,\mathrm{d}z=0\tag3 $$ where $\gamma_R=[-R,R]\cup[R,R-i]\cup[R-i,-R-i]\cup[-R-i,-R]$. This is because there are no singularities inside this contour for any $R$.

Furthermore, the integral vanishes on $[R,R-i]$ and $[-R-i,-R]$ as $R\to\infty$ since $|\sin(z)|\le\cosh(1)$ and $|z|\ge R$ and the length of each path is $1$. Thus, the integral over both paths is no bigger than $\frac{2\cosh(1)}R$.

Discounting the integrals which vanish, $(3)$ becomes $$ \int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x=\int_{-\infty-i}^{\infty-i}\frac{\sin(z)}z\,\mathrm{d}z\tag4 $$ The path of integration on the right side of $(4)$ passes nowhere near a singularity. Use $\sin(z)=\frac{e^{iz}-e^{-iz}}{2i}$ to evaluate the right side of $(4)$: $$ \begin{align} \int_{-\infty-i}^{\infty-i}\frac{\sin(z)}z\,\mathrm{d}z &=\frac1{2i}\int_{-\infty-i}^{\infty-i}\frac{e^{iz}}z\,\mathrm{d}z-\frac1{2i}\int_{-\infty-i}^{\infty-i}\frac{e^{-iz}}z\tag{5a}\\ &=\frac1{2i}\lim_{R\to\infty}\int_{\gamma_R^+}\frac{e^{iz}}z\,\mathrm{d}z-\frac1{2i}\lim_{R\to\infty}\int_{\gamma_R^-}\frac{e^{-iz}}z\tag{5b} \end{align} $$ where $\gamma_R^+=[-R-i,R-i]\cup Re^{i[0,\pi]}-i$ and $\gamma_R^-=[-R-i,R-i]\cup Re^{i[0,-\pi]}-i$. $\text{(5b)}$ follows because the integrals along the huge arcs go to $0$; $e^{iz}$ vanishes exponentially in the upper half-plane and $e^{-iz}$ vanishes exponentially in the lower half-plane. In fact, the integrals along those arcs are bounded by $$ \begin{align} \int_0^\pi\overbrace{e^{-R\sin(\theta)+1}\vphantom{\frac RR}}^{\large e^{\pm iz}}\overbrace{\ \frac{R\,\mathrm{d}\theta}{R-1}\ }^{\mathrm{d}z/z} &\le\frac{2eR}{R-1}\int_0^{\pi/2}e^{-2R\theta/\pi}\,\mathrm{d}\theta\tag{6a}\\ &\le\frac{2eR}{R-1}\frac\pi{2R}\tag{6b}\\[3pt] &=\frac{e\pi}{R-1}\tag{6c} \end{align} $$ Since $\gamma_R^-$ contains no singularities, the integral on the right-hand side of $\text{(5b)}$ is $0$. Since $\gamma_R^+$ contains the singularity at $0$, whose residue is $1$, we get that the integral on the left-hand side of $\text{(5b)}$ is $2\pi i$.

Putting together $(2)-(5)$, we conclude $$ \int_{-\infty}^\infty\frac{\sin(x)}x\,\mathrm{d}x=\pi\tag7 $$

robjohn
  • 345,667
  • How did you get the 0 in (1c)? I thought that was going to be infinity –  May 25 '21 at 10:33
  • 1
    As the explanation says, it is the integral of an odd function over a domain that is symmetric about the origin. Because $f(x)=\frac{\cos(x)}x$ is odd, $$\int_\epsilon^\infty f(x),\mathrm{d}x=-\int_{-\infty}^{-\epsilon}f(x),\mathrm{d}x$$ the integral over $|x|\gt\epsilon$ is $0$. – robjohn May 25 '21 at 11:48
  • Very nice, this directly answers the question. So we can see that the PV is the (or a) correct notion of integration for the exponential integral by splitting the exponential and then evaluating the cosine integral as a PV. I think the logic is now fully clear. Thank you. – doublefelix May 25 '21 at 14:40
0

The value of the improper integral is $$\lim\limits_{\substack{z\to&\,\,\,\infty \\ y\to&-\infty}}\int_y^z\frac{\sin x}{x}\,\mathrm{d}x$$ Clearly, if this limit exists, then so does the Cauchy principal value, and it has the same value as the integral.

I would appreciate it if someone would show me how to properly format the limit, so that both lines are the same size.

dezdichado
  • 13,888
saulspatz
  • 53,131
  • 1
    @MatthewTowers Thank you. I hope I can remember. – saulspatz Apr 26 '21 at 17:13
  • So we have established that $PV\int \frac{\sin x}{x} dx = Improper(\int \frac{\sin x}{x} dx)$. How do we proceed from there to see that $PV\int \frac{\sin x}{x} dx = Im(PV\int \frac{e^{ix}}{x})$? – doublefelix Apr 26 '21 at 17:56
  • When $x$ is real, $\Im(e^{ix})=\sin x$ so I don't see the problem. Doesn't this just follow from linearity of the integral, and the limit operator? – saulspatz Apr 26 '21 at 18:03
  • But if this logic works, couldn't we just as well say that $Improper \int \frac{\sin x}{x}dx = Im[Improper \int \frac{e^x}{x} dx] = undefined$ and show erraneously that the sinx/x integral is undefined as well. – doublefelix Apr 26 '21 at 18:22
  • @doublefelix Do you mean $\frac{e^{ix}}x$ in the integral? This doesn't make sense to me. Just because $\lim_{x\to\infty}(x+1)$ doesn't exist, we can't say $\lim_{x\to\infty}1$ doesn't exist. – saulspatz Apr 26 '21 at 19:10
  • Thanks for the help. Maybe I am crazy but it's still not clear to me. I have updated the question with an attempt to more precisely hit the source of the issue. – doublefelix Apr 26 '21 at 20:21
0

I don't think there is a way in which we can rigorously use the expression $$\operatorname{Im}\left(\int_{-\infty}^{\infty} \frac{e^{ix}}{x}dx \right)$$

If you test the following in wolfram alpha for example

  • integral from 0.0000001 to inf of e^(ix)/x dx
  • integral from 0.0000000001 to inf of e^(ix)/x dx
  • integral from 0.00000000000001 to inf of e^(ix)/x dx

you can see that the 'value' of this integral (we are using the cauchy principal value of course) would be "$\infty + i \tfrac{\pi}{2}$". Not a valid complex number, it's just a divergent integral so we can't really take the imaginary part to just throw away the infinity.

You may be be able to do manipulation with the divergent integral and still get right answers but I wouldn't trust it.

Note that the convergence of a Reimann integral is stronger than the convergence of the C.P.V.

  • 1
    I don't see any issue with $\text{Im} \left( CPV \int_{-\infty}^{\infty} \frac{e^{ix}}{x}dx \right)$. Of course if you work with the half-line it is divergent, but the CPV of the whole integral is legitimately convergent to $i \pi / 2$. – doublefelix May 25 '21 at 14:42