Why Cancellation Needs Care
For ordinary finite numbers, subtraction is always defined. If two quantities are both large, their difference may nevertheless be small. Improper integrals create a subtlety: a divergent improper integral is not a finite number, even when its cutoff expressions look like quantities that could cancel.
The key is to keep track of the cutoff limits. A cutoff expression is a finite integral, but the improper integral is defined only when its limit exists as a finite real number. If two improper integrals do not converge, there are no two finite integral values to subtract.
Cutoff Limits and Convergence
For a function \(f\) that is integrable on every finite interval \([a,R]\), its improper integral over \([a,\infty)\) is defined by a limit:
provided that this limit exists as a finite real number. If the limit is infinite or does not exist, the improper integral diverges; it does not have a finite value that can be used in ordinary arithmetic.
Suppose \(f\) and \(g\) are each integrable on every finite interval starting at \(a\). For a finite cutoff \(R\), ordinary linearity gives
This equality is valid for every finite \(R\). To use it to subtract the two improper integrals, however, both cutoff integrals on the right must converge to finite values as \(R\to\infty\). In that case, the limit laws for finite real numbers justify taking the limit of the difference. If either improper integral fails to converge, this reasoning does not produce a difference of two improper-integral values.
Worked Example: A Constant Tail
Worked Example: A Constant Tail
Consider \(\int_1^\infty 1\,dx\). First evaluate it with a finite cutoff \(R>1\):
The antiderivative \(x\) is correct because its derivative is \(1\). As \(R\to\infty\), \(R-1\to+\infty\). Therefore \(\int_1^\infty 1\,dx\) diverges to \(+\infty\) and is not a finite real number.
This example also shows why the cutoff expression and the improper integral must not be confused: \(R-1\) is a finite number for each finite \(R\), but the improper integral is the limit, and that limit is not finite.
Matched Cutoffs Do Not Define a Difference of Divergent Integrals
Consider subtracting \(\int_1^\infty 1\,dx\) from itself. Both improper integrals diverge. If we use the same finite cutoff \(R\) for both, their truncated difference is zero:
This calculation is correct for each \(R\). It shows that the integral of the pointwise difference \(1-1=0\), evaluated with a common cutoff, is zero. It does not show that the difference of the two divergent improper integrals is defined. That expression would attempt to subtract two quantities that are not finite integral values.
One way to see the distinction is to let the two cutoffs vary independently. For \(R,S>1\), the truncated difference is
Its behavior depends on how \(R\) and \(S\) tend to infinity. Setting \(S=R\) gives \(0\). Setting \(S=R+1\) gives \(-1\). Setting \(S=2R\) gives \(-R\), which tends to \(-\infty\). Thus there is no single value produced by letting both cutoffs grow independently.
Worked Example: Two Divergent Integrals with Logarithmic Terms
Worked Example: Two Divergent Integrals with Logarithmic Terms
Consider the cutoff integrals for \(f(x)=1+\frac{1}{x}\) and \(g(x)=1\), starting at \(1\). For \(R>1\), an antiderivative of \(f\) is \(x+\ln x\), since its derivative is \(1+\frac{1}{x}\). Therefore
An antiderivative of \(g\) is \(x\), with derivative \(1\), so for \(S>1\),
The first cutoff integral tends to \(+\infty\), as does the second. Subtracting their finite cutoff expressions gives
Different ways of linking the cutoffs produce different results:
- If \(S=R\), the difference is \(\ln R\), which tends to \(+\infty\).
- If \(S=R+\ln R\), the difference is \(0\) for every \(R>1\).
- If \(S=R+2\ln R\), the difference is \(-\ln R\), which tends to \(-\infty\).
Each substitution follows from \(R-S+\ln R\). These outcomes show that the truncated difference has no value independent of the cutoff relationship. With the common cutoff, the truncated difference is \(\int_1^R \frac{1}{x}\,dx=\ln R\), which also confirms that the improper integral of the pointwise difference diverges.
The Difference Integrand Is a Separate Question
For finite cutoffs, linearity always permits combining the integrands:
Taking \(R\to\infty\) on the left asks whether the improper integral of \(f-g\) converges. It is possible for that integral to converge even when the two separate integrals of \(f\) and \(g\) diverge. In that situation, the value belongs to the integral of the difference; it is not obtained by subtracting the two divergent integrals.
For example, if \(f(x)=g(x)=1\), then \(f-g=0\), and \(\int_1^\infty 0\,dx=0\). The separate integrals \(\int_1^\infty 1\,dx\) both diverge. The zero value is justified by integrating the zero function, not by assigning values to the divergent integrals and subtracting them.
This distinction is useful whenever two expressions seem to have large terms that cancel. Before using cancellation, ask what has actually been defined: a limit of the difference of cutoff expressions, an integral of the difference function, or a difference of two convergent improper integrals. These are not interchangeable when separate integrals diverge.
Why Linearity Requires Convergence
For convergent improper integrals, linearity follows directly from the finite-cutoff identity and the limit laws. Suppose
where \(A\) and \(B\) are finite real numbers. Then
The last equality is the limit law for the difference of two convergent real-valued quantities. It requires both limits to be finite. If the cutoff integrals diverge, the hypotheses for this limit law are not met, and the calculation cannot be used to define \(A-B\).
Use ordinary linearity to combine the integrands or subtract the finite cutoff values.
Determine whether each cutoff expression has a finite limit as its cutoff tends to infinity.
If both separate limits are finite, their difference is valid. Otherwise, assess the integral of the difference separately, if that is the quantity of interest.
Common Mistake: Treating Infinity as a Number
A common error is to write “\(\infty-\infty=0\)” because two expressions look identical, or because they grow at the same rate. This is not a valid real-number calculation: \(+\infty\) is not a finite number, and subtraction of two divergent limits is not defined by ordinary limit laws.
For a concrete correction, compare the two cutoff expressions \(R-1\) and \(S-1\). Their difference is
If \(S=2R\), the result is \(R-2R=-R\), which tends to \(-\infty\). It is not \(1-R\). More importantly, other choices of the cutoffs give other outcomes, so this subtraction does not define a value for the two divergent improper integrals.
Check Your Understanding
Use cutoff expressions to justify each answer. Do not treat a divergent integral as a finite value.
- Does \(\int_2^\infty 1\,dx\) converge? Show its cutoff expression.
- For two cutoffs \(R,S>1\), simplify \(\int_1^R 1\,dx-\int_1^S 1\,dx\). What happens for \(S=R+3\)?
- Why does using the same cutoff in \(\int_1^R 1\,dx-\int_1^R 1\,dx\) not define the difference of the two improper integrals?
- For the functions \(1+\frac{1}{x}\) and \(1\) on \([1,\infty)\), what is the integrand difference, and does its improper integral converge?
- State the condition that permits subtracting two improper integrals using linearity.
