To this point we have only considered nicely behaved integrals \(\int_a^b f(x)\dee{x}\text{.}\) Though the algebra involved in some of our examples was quite difficult, all the integrals had
finite limits of integration \(a\) and \(b\text{,}\) and
The first has an infinite domain of integration and the integrand of the second tends to \(\infty\) as \(x\) approaches the left end of the domain of integration. We’ll start with an example that illustrates the traps that you can fall into if you treat such integrals sloppily. Then we’ll see how to treat them carefully.
Very wrong. But it is not an example of “not even wrong” — which is a phrase attributed to the physicist Wolfgang Pauli who was known for his harsh critiques of sloppy arguments. The phrase is typically used to describe arguments that are so incoherent that not only can one not prove they are true, but they lack enough coherence to be able to show they are false. The interested reader should do a little searchengineing and look at the concept of falisfyability.
. In fact, the answer is ridiculous. The integrand \(\frac{1}{x^2} \gt 0\text{,}\) so the integral has to be positive.
is applicable only when \(F'(x)\) exists and equals \(f(x)\) for all\(a\le x\le b\text{.}\) In this case \(F'(x)=\frac{1}{x^2}\) does not exist for \(x=0\text{.}\) The given integral is improper. We’ll see later that the correct answer is \(+\infty\text{.}\)
Let us put this example to one side for a moment and turn to the integral \(\int_a^\infty\frac{\dee{x}}{1+x^2}\text{.}\) In this case, the integrand is bounded but the domain of integration extends to \(+\infty\text{.}\) We can evaluate this integral by sneaking up on it. We compute it on a bounded domain of integration, like \(\int_a^R\frac{\dee{x}}{1+x^2}\text{,}\) and then take the limit \(R\rightarrow\infty\text{.}\)
To be more precise, we actually formally define an integral with an infinite domain as the limit of the integral with a finite domain as we take one or more of the limits of integration to infinity.
We must also be able to deal with an integral like \(\int_0^1\frac{\dee{x}}{x}\) that has a finite domain of integration, but whose integrand is unbounded near one limit of integration 2
This will, in turn, allow us to deal with integrals whose integrand is unbounded somewhere inside the domain of integration.
. Our approach is similar — we sneak up on the problem. We compute the integral on a smaller domain, such as \(\int_t^1\frac{\dee{x}}{x}\text{,}\) with \(t \gt 0\text{,}\) and then take the limit \(t\rightarrow 0+\text{.}\)
Let \(a \lt c \lt b\text{.}\) If the integrals \(\int_a^T f(x)\dee{x}\) and \(\int_t^b f(x)\dee{x}\) exist for all \(a \lt T \lt c\) and \(c \lt t \lt b\text{,}\) then
Notice that (c) is used when the integrand is unbounded at some point in the middle of the domain of integration, such as was the case in our original example
More generally, if an integral has more than one “source of impropriety” (for example an infinite domain of integration and an integrand with an unbounded integrand or multiple infinite discontinuities) then you split it up into a sum of integrals with a single “source of impropriety” in each. For the integral, as a whole, to converge every term in that sum has to converge.
The domain of the integral \(\int_1^\infty\frac{\dee{x}}{x^p}\) extends to \(+\infty\) and the integrand \(\frac{1}{x^p}\) is continuous and bounded on the whole domain.
The antiderivative of \(1/x^p\) changes when \(p=1\text{,}\) so we will split the problem into three cases, \(p \gt 1\text{,}\)\(p=1\) and \(p \lt 1\text{.}\)
The domain of integration of the integral \(\int_0^1\frac{\dee{x}}{x^p}\) is finite, but the integrand \(\frac{1}{x^p}\) becomes unbounded as \(x\) approaches the left end, \(0\text{,}\) of the domain of integration.
This time the domain of integration of the integral \(\int_0^\infty\frac{\dee{x}}{x^p}\) extends to \(+\infty\text{,}\) and in addition the integrand \(\frac{1}{x^p}\) becomes unbounded as \(x\) approaches the left end, \(0\text{,}\) of the domain of integration.
We saw, in Example 1.12.9, that the first integral diverged whenever \(p\ge 1\text{,}\) and we also saw, in Example 1.12.8, that the second integral diverged whenever \(p\le 1\text{.}\)
This suggests that the signed area to the left of the \(y\)-axis should exactly cancel the area to the right of the \(y\)-axis making the value of the integral \(\int_{-1}^1\frac{\dee{x}}{x}\) exactly zero.
diverge so \(\int_{-1}^1\frac{\dee{x}}{x}\) diverges. Don’t make the mistake of thinking that \(\infty-\infty=0\text{.}\)It is undefined. And it is undefined for good reason.
This appears to give \(\infty-\infty=\log 7\text{.}\) Of course the number \(7\) was picked at random. You can make \(\infty-\infty\) be any number at all, by making a suitable replacement for \(7\text{.}\)
For \(x\ge e\text{,}\) the denominator \(x(\log x)^p\) is never zero. So the integrand is bounded on the entire domain of integration and this integral is improper only because the domain of integration extends to \(+\infty\) and we proceed as usual.
Now we move on to general \(n\text{,}\) using the same type of computation as we just used to evaluate \(\Gamma(2)\text{.}\) For any natural number \(n\text{,}\)
The Gamma function is far more important than just a generalisation of the factorial. It appears all over mathematics, physics, statistics and beyond. It has all sorts of interesting properties and its definition can be extended from natural numbers \(n\) to all numbers excluding \(0,-1,-2,-3,\cdots\text{.}\) For example, one can show that \(\Gamma(1-z)\Gamma(z) = \frac{\pi}{\sin \pi z}.\)
Subsection1.12.3Convergence Tests for Improper Integrals
It is very common to encounter integrals that are too complicated to evaluate explicitly. Numerical approximation schemes, evaluated by computer, are often used instead (see Section 1.11). You want to be sure that at least the integral converges before feeding it into a computer 4
Applying numerical integration methods to a divergent integral may result in perfectly reasonably looking but very wrong answers.
. Fortunately it is usually possible to determine whether or not an improper integral converges even when you cannot evaluate it explicitly.
For pedagogical purposes, we are going to concentrate on the problem of determining whether or not an integral \(\int_a^\infty f(x)\dee{x}\) converges, when \(f(x)\) has no singularities for \(x\ge a\text{.}\) Recall that the first step in analyzing any improper integral is to write it as a sum of integrals each of has only a single “source of impropriety” — either a domain of integration that extends to \(+\infty\text{,}\) or a domain of integration that extends to \(-\infty\text{,}\) or an integrand which is singular at one end of the domain of integration. So we are now going to consider only the first of these three possibilities. But the techniques that we are about to see have obvious analogues for the other two possibilities. And, at the end of this section, we’ll give the analogous tools that are appropriate for use with an integrand which is singular at one end of a finite domain of integration.
Now let’s start. Imagine that we have an improper integral \(\int_a^\infty f(x)\dee{x}\text{,}\) that \(f(x)\) has no singularities for \(x\ge a\) and that \(f(x)\) is complicated enough that we cannot evaluate the integral explicitly 5
You could, for example, think of something like our running example \(\int_a^\infty e^{-t^2} \dee{t}\text{.}\)
. The idea is find another improper integral \(\int_a^\infty g(x)\dee{x}\)
with \(g(x)\) simple enough that we can evaluate the integral \(\int_a^\infty g(x)\dee{x}\) explicitly, or at least determine easily whether or not \(\int_a^\infty g(x)\dee{x}\) converges, and
with \(g(x)\) behaving enough like \(f(x)\) for large \(x\) that the integral \(\int_a^\infty f(x)\dee{x}\) converges if and only if \(\int_a^\infty g(x)\dee{x}\) converges.
Let \(a\) be a real number. Let \(f\) and \(g\) be functions that are defined and continuous for all \(x\ge a\) and assume that \(g(x)\ge 0\) for all \(x\ge a\text{.}\)
If \(|f(x)|\le g(x)\) for all \(x\ge a\) and if \(\int_a^\infty g(x)\dee{x}\) converges then \(\int_a^\infty f(x)\dee{x}\) also converges.
We will not prove this theorem, but, hopefully, the following supporting arguments should at least appear reasonable to you. Consider the figure below:
We have separated the regions in which \(f(x)\) is positive and negative, because the integral \(\int_a^\infty f(x)\dee{x} \) represents the signed area of the union of \(\big\{\ (x,y)\ \big|\ x\ge a,\ 0\le y\le f(x)\ \big\}\) and \(\big\{\ (x,y)\ \big|\ x\ge a,\ f(x)\le y\le 0\ \big\}\text{.}\)
Solution: We will use Theorem 1.12.17 to answer the question.
So we want to find another integral that we can compute and that we can compare to \(\int_1^\infty e^{-x^2}\dee{x}\text{.}\) To do so we pick an integrand that looks like \(e^{-x^2}\text{,}\) but whose indefinite integral we know — such as \(e^{-x}\text{.}\)
So, by Theorem 1.12.17, with \(a=1\text{,}\)\(f(x)=e^{-x^2}\) and \(g(x)=e^{-x}\text{,}\) the integral \(\int_1^\infty e^{-x^2}\dee{x}\) converges too (it is approximately equal to \(0.1394\)).
The integral \(\int_{1/2}^\infty e^{-x^2}\dee{x}\) is quite similar to the integral \(\int_1^\infty e^{-x^2}\dee{x}\) of Example 1.12.18. But we cannot just repeat the argument of Example 1.12.18 because it is not true that \(e^{-x^2}\le e^{-x}\) when \(0 \lt x \lt 1\text{.}\)
which is clearly a well defined finite number (its actually about \(0.286\)). It is important to note that we are being a little sloppy by taking the difference of two integrals like this — we are assuming that both integrals converge. More on this below.
So we would expect that \(\int_{1/2}^\infty e^{-x^2}\dee{x}\) should be the sum of the proper integral integral \(\int_{1/2}^1 e^{-x^2}\dee{x}\) and the convergent integral \(\int_1^\infty e^{-x^2}\dee{x}\) and so should be a convergent integral. This is indeed the case. The Theorem below provides the justification.
Let \(a\) and \(c\) be real numbers with \(a \lt c\) and let the function \(f(x)\) be continuous for all \(x\ge a\text{.}\) Then the improper integral \(\int_a^\infty f(x)\ \dee{x}\) converges if and only if the improper integral \(\int_c^\infty f(x)\ \dee{x}\) converges.
exists and is finite. (Remember that, in computing the limit, \(\int_a^c f(x)\dee{x}\) is a finite constant independent of \(R\) and so can be pulled out of the limit.) But that is the case if and only if the limit \(\lim_{R\rightarrow\infty}\int_c^R f(x)\dee{x}\) exists and is finite, which in turn is the case if and only if the integral \(\int_c^\infty f(x)\dee{x}\) converges.
The domain of integration extends to \(+\infty\text{,}\) but we must also check to see if the integrand contains any singularities. On the domain of integration \(x\ge 1\) so the denominator is never zero and the integrand is continuous. So the only problem is at \(+\infty\text{.}\)
This takes practice, practice and more practice. At the risk of alliteration — please perform plenty of practice problems .
. As the only problem is that the domain of integration extends to infinity, whether or not the integral converges will be determined by the behavior of the integrand for very large \(x\text{.}\)
When \(x\) is very large, \(x^2\) is much much larger than \(x\) (which we can write as \(x^2\gg x\)) so that the denominator \(x^2+x\approx x^2\) and the integrand
By Example 1.12.8, with \(p=\frac{3}{2}\text{,}\) the integral \(\int_1^\infty \frac{\dee{x}}{x^{3/2}}\) converges. So we would expect that \(\int_1^\infty\frac{\sqrt{x}}{x^2+x}\dee{x}\) converges too.
Our final task is to verify that our intuition is correct. To do so, we want to apply part (a) of Theorem 1.12.17 with \(f(x)= \frac{\sqrt{x}}{x^2+x}\) and \(g(x)\) being \(\frac{1}{x^{3/2}}\text{,}\) or possibly some constant times \(\frac{1}{x^{3/2}}\text{.}\) That is, we need to show that for all \(x \geq 1\) (i.e. on the domain of integration)
So Theorem 1.12.17(a) and Example 1.12.8, with \(p=\frac{3}{2}\) do indeed show that the integral \(\int_1^\infty\frac{\sqrt{x}}{x^2+x}\dee{x}\) converges.
Notice that in this last example we managed to show that the integral exists by finding an integrand that behaved the same way for large \(x\text{.}\) Our intuition then had to be bolstered with some careful inequalities to apply the comparison Theorem 1.12.17. It would be nice to avoid this last step and be able jump from the intuition to the conclusion without messing around with inequalities. Thankfully there is a variant of Theorem 1.12.17 that is often easier to apply and that also fits well with the sort of intuition that we developed to solve Example 1.12.21.
A key phrase in the previous paragraph is “behaves the same way for large \(x\)”. A good way to formalise this expression — “\(f(x)\) behaves like \(g(x)\) for large \(x\)” — is to require that the limit
\begin{align*}
\lim_{x\rightarrow\infty}\frac{f(x)}{g(x)} & \text{ exists and is a finite nonzero number.}
\end{align*}
Suppose that this is the case and call the limit \(L\ne 0\text{.}\) Then
the ratio \(\frac{f(x)}{g(x)}\) must approach \(L\) as \(x\) tends to \(+\infty\text{.}\)
Let \(-\infty \lt a \lt \infty\text{.}\) Let \(f\) and \(g\) be functions that are defined and continuous for all \(x\ge a\) and assume that \(g(x)\ge 0\) for all \(x\ge a\text{.}\)
If \(\int_a^\infty g(x)\,\dee{x}\) converges and the limit
The domain of integration extends to \(+\infty\text{.}\) On the domain of integration the denominator is never zero so the integrand is continuous. Thus the only problem is at \(+\infty\text{.}\)
Our second task is to develop some intuition about the behavior of the integrand for very large \(x\text{.}\) A good way to start is to think about the size of each term when \(x\) becomes big.
Notice that we are using \(A \ll B\) to mean that “\(A\) is much much smaller than \(B\)”. Similarly \(A\gg B\) means “\(A\) is much much bigger than \(B\)”. We don’t really need to be too precise about its meaning beyond this in the present context.
Since \(\int_1^\infty g(x)\dee{x} = \int_1^\infty\frac{\dee{x}}{x}\) diverges, by Example 1.12.8 with \(p=1\text{,}\) Theorem 1.12.22(b) now tells us that \(\int_1^\infty f(x)\dee{x} = \int_1^\infty\frac{x+\sin x}{e^{-x}+x^2}\dee{x}\) diverges too.
We finish this section by giving variants of Theorems 1.12.17 and 1.12.22 that are useful in dealing with an integrand that is singular at one end of a (finite) domain of integration. For the rest of this section, let \(a\lt b\) and let \([a,b]=\big\{x\,\big|\,a\le x\le b\big\}\) be the domain of integration. We allow the integrand to be singular at one end, that we’ll call \(x_s\text{,}\) of the domain of integration. So
if the integrand is singular at \(x=a\text{,}\) we set \(x_s=a\) and write
Below are the graphs \(y=f(x)\) and \(y=g(x)\text{.}\) Suppose \(\displaystyle\int_0^\infty f(x) \dee{x}\) converges, and \(\displaystyle\int_0^\infty g(x) \dee{x}\) diverges. Assuming the graphs continue on as shown as \(x \to \infty\text{,}\) which graph is \(f(x)\text{,}\) and which is \(g(x)\text{?}\)
Decide whether the following statement is true or false. If false, provide a counterexample. If true, provide a brief justification. (Assume that \(f(x)\) and \(g(x)\) are continuous functions.)
If \(\displaystyle\int_{1}^{\infty} f(x) \,\dee{x}\) converges and \(g(x)\ge f(x)\ge 0\) for all \(x\text{,}\) then \(\displaystyle\int_{1}^{\infty} g(x) \,\dee{x}\) converges.
For each of the functions \(h(x)\) described below, decide whether \(\int_{0\vphantom{\frac12}}^\infty h(x) \dee{x}\) converges or diverges, or whether there isn’t enough information to decide. Justify your decision.
Determine whether the integral \(\displaystyle\int_{-2}^2\frac{1}{(x+1)^{4/3}}\,\dee{x}\) is convergent or divergent. If it is convergent, find its value.
\(\displaystyle\int_{-\infty}^{+\infty}\frac{x}{x^2+1}\dee{x}\) converges but \(\displaystyle\int_{-\infty}^{+\infty}\left|\frac{x}{x^2+1}\right|\dee{x}\) diverges
\(\displaystyle\int_{-\infty}^{+\infty}\frac{x}{x^2+1}\dee{x}\) converges, as does \(\displaystyle\int_{-\infty}^{+\infty}\left|\frac{x}{x^2+1}\right|\dee{x}\)
Remark: these options, respectively, are that the integral diverges, converges conditionally, and converges absolutely. You’ll see this terminology used for series in Section 3.4.1.
We craft a tall, vuvuzela-shaped solid by rotating the line \(y = \dfrac{1}{x\vphantom{\frac{1}{2}}}\) from \(x=a\) to \(x=1\) about the \(y\)-axis, where \(a\) is some constant between 0 and 1.
Does the integral \(\displaystyle\int_{-5}^5 \left(\frac{1}{\sqrt{|x|}} + \frac{1}{\sqrt{|x-1|}}+\frac{1}{\sqrt{|x-2|}}\right)\dee{x}\) converge or diverge?
Let \(M_{n,t}\) be the Midpoint Rule approximation for \(\displaystyle\int_0^t \frac{e^{-x}}{1+x}\dee{x}\) with \(n\) equal subintervals. Find a value of \(t\) and a value of \(n\) such that \(M_{n,t}\) differs from \(\int_0^\infty \frac{e^{-x}}{1+x}\dee{x}\) by at most \(10^{-4}\text{.}\) Recall that the error \(E_n\) introduced when the Midpoint Rule is used with \(n\) subintervals obeys
where \(M\) is the maximum absolute value of the second derivative of the integrand and \(a\) and \(b\) are the end points of the interval of integration.
Suppose \(f(x)\) is continuous for all real numbers, and \(\displaystyle\int_1^\infty f(x) \dee{x}\) converges.
If \(f(x)\) is odd, does \(\displaystyle\int_{-\infty\vphantom{\frac12}}^{-1} f(x) \dee{x}\) converge or diverge, or is there not enough information to decide?