Why Parameter Thresholds Matter
In an improper integral, convergence can depend on how an integrand behaves near a singular endpoint or far out on an infinite interval. When the integrand contains parameters, each endpoint can impose its own condition. A parameter value that makes one endpoint harmless may still make another endpoint divergent.
The key is to identify the power that describes the integrand near each problematic endpoint, apply the threshold appropriate to that endpoint, and require all the resulting conditions to hold. Equality is not usually included: at the threshold, the corresponding power integral diverges logarithmically.
The Exponent Thresholds
For a power \(x^p\), the convergence condition depends on whether the endpoint is zero or infinity. These conditions follow directly from an antiderivative when \(p\ne -1\); when \(p=-1\), the antiderivative is a logarithm.
To see the finite-endpoint condition, first suppose \(p\ne -1\). For \(0<\varepsilon<1\),
If \(p+1>0\), then \(\varepsilon^{p+1}\to 0\) as \(\varepsilon\to 0^+\), so the limit is finite. If \(p+1<0\), then \(\varepsilon^{p+1}\to\infty\), so the integral diverges. At \(p=-1\),
For the infinite endpoint, take \(R>1\). When \(p\ne -1\),
This has a finite limit as \(R\to\infty\) exactly when \(p+1<0\). At \(p=-1\), the integral is \(\ln R\), which tends to infinity. Thus the threshold reverses between zero and infinity.
The same rules apply to a power of the distance from a finite endpoint: \((x-a)^p\) near \(a\) on the right has the zero-endpoint threshold \(p>-1\), and \((b-x)^p\) near \(b\) on the left has that same threshold. At infinity, the relevant power is a power of \(x\), and its exponent must be less than \(-1\).
Using Asymptotic Behavior
Many integrands are not exactly powers, but behave like powers near an endpoint. For positive functions \(f\) and \(g\), if \(f(x)/g(x)\) tends to a positive finite number \(L\) at an endpoint, then \(f\) and \(g\) have the same convergence behavior there. Indeed, sufficiently close to the endpoint, their ratio is bounded above and below by positive constants, so their integrals are bounded by constant multiples of one another. This is the comparison principle for endpoint behavior.
For example, if \(f(x)/x^p\to 1\) as \(x\to 0^+\), then the integral of \(f\) near zero converges exactly when \(p>-1\). If \(f(x)/x^p\to 1\) as \(x\to\infty\), it converges at infinity exactly when \(p<-1\). When the resulting exponent equals \(-1\), the comparison is to \(1/x\), so the integral diverges.
Worked Example: Two Singular Endpoints on a Finite Interval
Worked Example: Finding the Parameter Range on \((0,1)\)
Determine the values of \(t\) for which
converges.
Step 1: Examine the endpoint \(x=0\). As \(x\to 0^+\), the factor \((1-x)^{t+1}\) tends to \(1\), which is positive and finite. Thus the integrand behaves like \(x^{2t-3}\). The zero-endpoint threshold requires
Step 2: Examine the endpoint \(x=1\). As \(x\to 1^-\), the factor \(x^{2t-3}\) tends to \(1\). The integrand behaves like \((1-x)^{t+1}\), so convergence at this endpoint requires
Step 3: Require both conditions. The integral converges only when \(t>1\) and \(t>-2\). Their intersection is \(t>1\).
Check: At \(t=2\), the endpoint exponents are \(2t-3=1\) at zero and \(t+1=3\) at one. Both exceed \(-1\), so the integral converges. At the boundary \(t=1\), the exponent at zero is \(-1\), while the exponent at one is \(2\). The integral diverges at zero despite being integrable at one. This confirms that the strict condition \(t>1\), not \(t\ge 1\), is required.
Worked Example: A Parameter at an Infinite Endpoint
Worked Example: A Rational Power on \([1,\infty)\)
For real parameters \(\alpha\) and \(\beta\), determine when
converges.
The integrand is continuous at \(x=1\), so only the infinite endpoint needs a convergence test. Compare it with \(x^{\alpha-\beta}\):
Therefore the integrand has the same convergence behavior at infinity as \(x^{\alpha-\beta}\). The infinite-endpoint threshold gives
Check with a convergent choice: Let \(\alpha=2\) and \(\beta=4\). Then \(\alpha-\beta=2-4=-2<-1\), so the integral converges.
Check the boundary: If \(\alpha=2\) and \(\beta=3\), then \(\alpha-\beta=2-3=-1\). The ratio to \(x^{-1}\) tends to \(1\), so the integral diverges logarithmically. Thus equality \(\beta=\alpha+1\) is excluded, and the condition is strictly \(\beta>\alpha+1\).
Worked Example: Combining Conditions at Zero and Infinity
Worked Example: A Power with a Parameterized Denominator
Suppose \(b>0\) and \(c>0\). Determine the values of \(a\) for which
converges.
Step 1: Apply the threshold at zero. As \(x\to 0^+\), \(x^b\to 0\), so \((1+x^b)^c\to 1\). The integrand behaves like \(x^a\). Therefore convergence at zero requires
Step 2: Apply the threshold at infinity. For large \(x\), factor \(x^b\) from \(1+x^b\). The ratio to \(x^{a-bc}\) is
Thus convergence at infinity requires
Step 3: Combine the conditions. Both endpoints are acceptable exactly when
Check with specific parameter values: Let \(b=2\) and \(c=3\). Then \(bc=6\), so the range becomes \(-1<a<5\). Choose \(a=2\): at zero the power is \(x^2\), with exponent \(2>-1\); at infinity the power is \(x^{2-6}=x^{-4}\), with exponent \(-4<-1\). The integral converges. At \(a=-1\), the zero-endpoint power is \(x^{-1}\), so it diverges there. At \(a=5\), the infinite-endpoint power is \(x^{5-6}=x^{-1}\), so it diverges at infinity. The two strict boundaries are therefore both necessary.
Why Endpoint Conditions Must Be Kept Separate
An improper integral over an interval with multiple problematic endpoints converges only if its contribution at every such endpoint is finite. One endpoint cannot compensate for divergence at another. For an integral over \((0,\infty)\), it is useful to split at any convenient positive number, such as \(1\), and test the parts near zero and infinity separately.
This endpoint-by-endpoint method also prevents a common error: applying one threshold everywhere. The condition \(p>-1\) belongs to a finite zero-type singularity, while \(p<-1\) belongs to an infinite endpoint. The exponent alone is not enough; the location of the endpoint matters.
A Practical Decision Process
For a parameterized power-type improper integral, identify where it is improper before combining any algebraic exponents. Then find the local power at each such endpoint. Factors that tend to a nonzero finite constant do not change that local power. Apply the threshold for that endpoint, solve the resulting inequality, and intersect all parameter conditions. Finally, substitute each boundary value into the local exponent: equality at \(-1\) gives the logarithmic divergence that excludes the boundary.
These thresholds are useful well beyond integrals written as a single power. Rational expressions, products of powers, and parameterized denominators often reduce to a power near each endpoint. The important work is to obtain the correct local exponent at every endpoint rather than treating the entire integrand as if it had one global behavior.
Check Your Understanding
For each question, identify the relevant endpoint or endpoints and state whether equality is included.
- For which values of \(p\) does \(\int_0^1 x^p\,dx\) converge?
- For which values of \(q\) does \(\int_1^\infty x^q\,dx\) converge?
- Determine the values of \(t\) for which \(\int_0^1 x^{t-2}(1-x)^{2t}\,dx\) converges.
- For \(b,c>0\), find the condition on \(a\) that makes \(\int_0^\infty x^a/(1+x^b)^c\,dx\) converge.
- Why does an exponent of \(-1\) at either a zero endpoint or infinity fail the convergence test?
