Introduction
For many improper integrals, a power of \(x\) determines convergence. At a critical power, however, that power alone no longer decides the result. Factors involving \(\ln x\), and then \(\ln\ln x\), can settle the question. Each new logarithmic layer matters precisely when the previous layer is itself at its threshold.
This tutorial studies tails that have the form of a power of \(x\) multiplied by powers of \(\ln x\) and \(\ln\ln x\). The key is to compare in stages: first inspect the power of \(x\); if it is critical, inspect the power of \(\ln x\); if that too is critical, inspect the power of \(\ln\ln x\).
The Family of Tails and Its Thresholds
For real numbers \(p,q,r\), define the positive function
Negative values of \(q\) or \(r\) are allowed. For example, a negative \(r\) means that \((\ln\ln x)^{-r}\) appears in the numerator. The convergence classification is:
- \(p>1\);
- \(p=1\) and \(q>1\);
- \(p=1,\ q=1,\) and \(r>1\).
The theorem gives a hierarchy. If \(p\ne1\), neither logarithmic factor changes the outcome. If \(p=1\), the power \(q\) decides unless \(q=1\). Only when both \(p=1\) and \(q=1\) does the value of \(r\) determine convergence.
Why Logarithms Cannot Overcome a Noncritical Power
We first justify why \(p\) settles the result whenever it differs from \(1\). We use the power-integral test: \(\int_A^\infty x^{-s}\,dx\) converges for \(s>1\) and diverges for \(s\le1\), where \(A>0\).
The comparison works because any fixed powers of logarithms grow more slowly than any positive power of \(x\). More precisely, for \(a>0\) and real \(b,c\),
To see this, put \(t=\ln x\). The expression becomes \(e^{-at}t^b(\ln t)^c\). For sufficiently large \(t\), \(\ln t\le t\), so the absolute size of the factors \(t^b(\ln t)^c\) is bounded above by some fixed power \(t^M\). Since \(e^{-at}t^M\to0\), the displayed limit follows.
If \(p>1\), choose \(s\) with \(1<s<p\). The ratio of \(f(x)\) to the convergent comparison function \(x^{-s}\) is
The limit comparison theorem, including its zero-limit form, then gives convergence: for sufficiently large \(x\), \(f(x)\le x^{-s}\), and the comparison integral converges.
If \(p<1\), choose \(s\) with \(p<s<1\). Now the same ratio satisfies
The comparison integral \(\int_{e^e}^{\infty}x^{-s}\,dx\) diverges because \(s<1\). Since \(f(x)\ge x^{-s}\) for all sufficiently large \(x\), the integral of \(f\) diverges as well. These comparisons show that fixed logarithmic powers cannot change the outcome when \(p\ne1\).
When the Power of \(x\) Is Critical
Now let \(p=1\). The integral is
Use the substitution \(u=\ln x\), so \(du=dx/x\). The lower endpoint \(x=e^e\) becomes \(u=e\), giving
This substitution reveals the next threshold. When \(q>1\), choose \(s\) with \(1<s<q\). The ratio of \(u^{-q}(\ln u)^{-r}\) to \(u^{-s}\) is \(u^{s-q}(\ln u)^{-r}\), which tends to zero because \(q-s>0\). Since \(\int_e^\infty u^{-s}\,du\) converges, the transformed integral converges.
When \(q<1\), choose \(s\) with \(q<s<1\). The same ratio tends to infinity. The comparison integral \(\int_e^\infty u^{-s}\,du\) diverges, so the transformed integral diverges. Thus, unless \(q=1\), the exponent of \(\ln x\) decides the result.
At \(q=1\), a second substitution is required. Set \(v=\ln u\), so \(dv=du/u\). The remaining integral becomes
The power-integral test gives convergence exactly when \(r>1\). If \(r\le1\), this last integral diverges. This proves the full classification: each substitution exposes a new power-integral threshold only when the preceding exponent is exactly critical.
Worked Examples
Worked Example: The Triple Critical Case
Classify
Here \(p=1\), \(q=1\), and \(r=1\). Both earlier layers are critical, so we must check the final logarithmic exponent. First set \(u=\ln x\), with \(du=dx/x\):
Now set \(v=\ln u\), with \(dv=du/u\). The lower endpoint \(u=e\) becomes \(v=1\):
Therefore the original integral diverges. The check is the transformed power integral: its exponent is \(r=1\), exactly the divergent boundary case.
Worked Example: Convergence Just Beyond the Iterated-Log Threshold
Classify
Here \(p=1\), \(q=1\), and \(r=2\). The first substitution \(u=\ln x\) gives
Set \(v=\ln u\). Since \(dv=du/u\) and \(v=1\) at \(u=e\),
The integral converges, with value \(1\). Differentiating the antiderivative \(-1/v\) gives \(1/v^2\), confirming the evaluation. This is the threshold result: the final exponent \(r=2\) is greater than \(1\).
Worked Example: A Logarithmic Factor Cannot Rescue a Subcritical Power
Classify
The power exponent is \(p=0.8<1\), so we compare with \(x^{-0.9}\), whose integral diverges. The ratio is
This ratio tends to infinity: its reciprocal \(x^{-0.1}(\ln x)(\ln\ln x)^2\) tends to zero by the logarithmic-growth limit above. Thus the original integrand is at least \(x^{-0.9}\) for all sufficiently large \(x\). Since
the original integral diverges by direct comparison. This verifies the classification even though the logarithmic factors appear in the denominator: they cannot overcome the subcritical power.
Worked Example: Convergence When the Logarithmic Exponent Exceeds Its Threshold
Classify
Here \(p=1\), \(q=3/2>1\), and \(r=0\). Set \(u=\ln x\), so \(du=dx/x\), and the lower limit becomes \(e\):
The value is finite, so the integral converges. Differentiating \(-2u^{-1/2}\) gives \(u^{-3/2}\), confirming the transformed evaluation. The classification also checks directly: after \(p=1\), the exponent \(q=3/2\) exceeds \(1\).
A Common Mistake at the Final Threshold
A frequent error is to stop after recognizing that both \(p\) and \(q\) are \(1\), then assume the integral converges because the denominator contains two logarithmic factors. The presence of logarithms alone does not decide the result; their powers matter.
For instance, the calculation
omits a factor. With \(u=\ln x\), we have \(du=dx/x\), but \(\ln\ln x=\ln u\) remains in the denominator. The correct calculation is
The final integral diverges. Keeping the remaining logarithmic factor at every substitution is essential: each layer is tested only after the preceding substitution has exposed it.
Using the Thresholds Efficiently
For this family, a reliable classification procedure is to inspect the exponents in order. Do not compare every factor at once. If \(p\ne1\), stop: the power of \(x\) decides. If \(p=1\), inspect \(q\). If \(q\ne1\), stop: the power of \(\ln x\) decides. Only if \(p=q=1\) should you inspect \(r\).
If \(p>1\), the tail converges; if \(p<1\), it diverges.
If \(q>1\), the tail converges; if \(q<1\), it diverges.
The tail converges exactly when \(r>1\).
This method is useful because it makes clear which factor is doing the decisive work. It also prevents a common overgeneralization: a factor tending to zero does not automatically make an integral converge. Its rate must be compared with the critical scale. The successive substitutions and comparisons identify that scale precisely.
Check Your Understanding
For each integral, decide whether it converges or diverges. Identify the exponent that determines the result, and use substitutions or comparisons to justify your classification.
- \(\int_{e^e}^{\infty}\frac{dx}{x(\ln x)^{1/2}(\ln\ln x)^3}\)
- \(\int_{e^e}^{\infty}\frac{dx}{x\ln x(\ln\ln x)^{3/2}}\)
- \(\int_{e^e}^{\infty}\frac{dx}{x^{3/2}(\ln x)^{-2}}\)
- \(\int_{e^e}^{\infty}\frac{dx}{x(\ln x)(\ln\ln x)}\)
- State which exponent must equal \(1\) before the value of \(r\) can affect convergence.