Logarithmic Factors at the Ratio-Test Boundary

Why Logarithmic Factors Create a Boundary Case

The ratio test and root test are especially useful when terms contain factorials, exponentials, or powers whose behavior changes rapidly with the index. But some positive series decay more slowly. A typical example has terms \(1/[n(\ln n)^q]\), where the logarithmic factor modifies the familiar \(1/n\) scale only gradually.

For this family, both tests reach the same boundary value, regardless of the real exponent \(q\). That does not mean all these series behave alike. It means the ratio and root tests do not distinguish their behavior, so another test is needed.

The Ratio Limit

The ratio test examines the limit of the quotient of consecutive terms, when that limit exists. For a series of positive terms, a limit less than one implies convergence, while a limit greater than one implies divergence. A limit equal to one gives no conclusion.

For the logarithmic terms, calculate the quotient before taking the limit:

The first factor tends to one. Also, \(\ln(n+1)/\ln n\to1\), so its reciprocal tends to one. Raising positive quantities tending to one to the fixed real power \(q\) still gives a limit of one. Therefore

This result holds for every fixed real \(q\), including negative values. The ratio test therefore does not decide convergence anywhere in this family.

The Root Limit

The root test examines \(\lim_{n\to\infty}\sqrt[n]{a_n}\), when this limit exists. A limit less than one implies convergence, a limit greater than one implies divergence, and a limit equal to one is inconclusive.

Taking a logarithm makes the root calculation manageable. Since \(a_n>0\),

Both \(\ln n/n\) and \(\ln(\ln n)/n\) tend to zero. Thus the displayed logarithm tends to zero. Exponentiating gives

The root test, like the ratio test, reaches its inconclusive case for every fixed \(q\). The logarithmic factor does affect the series, but its effect is too gradual to appear in either of these limits.

Use the Integral Test to Resolve the Boundary

To distinguish the values of \(q\), use the integral test. In the form needed here, if \(f\) is positive, continuous, and decreasing for all \(x\) beyond some point, then the series \(\sum f(n)\) and the improper integral \(\int f(x)\,dx\) over the corresponding tail either both converge or both diverge. A finite number of initial terms does not change convergence.

Set \(f(x)=1/[x(\ln x)^q]\) for \(x>1\). This function is positive and continuous on its domain. Differentiating gives

For sufficiently large \(x\), \(\ln x+q>0\). The denominator is positive, so \(f'(x)<0\) eventually. Thus the required eventual decrease holds for every fixed real \(q\), including negative \(q\).

For completeness, the comparison behind the integral test follows from the fact that a decreasing positive function satisfies \(f(n+1)\le f(x)\le f(n)\) whenever \(n\le x\le n+1\). Integrating over that interval gives

Summing these inequalities over consecutive intervals shows that the integral and the series tail bound one another, apart from at most a finite initial term. Consequently, one is finite exactly when the other is finite.

Now evaluate the improper integral. The substitution \(u=\ln x\), so \(du=dx/x\), gives

When \(q\ne1\), an antiderivative of \(u^{-q}\) is \(u^{1-q}/(1-q)\). If \(q>1\), then \(1-q<0\), so \(u^{1-q}\to0\) as \(u\to\infty\); the improper integral is finite. If \(q<1\), then \(1-q>0\), so \(u^{1-q}\to\infty\); the integral diverges. When \(q=1\), the antiderivative is \(\ln u\), which also tends to infinity. Therefore

This threshold comes from the integral, not from the ratio or root limit. Those two limits remain one on both sides of the threshold.

Worked Examples

A Common Mistake

A frequent error is to treat a ratio or root limit equal to one as if it proved divergence. For example, after finding

one might incorrectly say, “The limit is at least one, so the series diverges.” The ratio test’s divergence condition is a limit strictly greater than one; equality is not included. In this example the series actually converges, as the integral calculation shows.

The correction is to record the test’s conclusion accurately: the ratio limit is one, so the ratio test gives no decision. The same is true of the root test when its limit is one. A separate argument, such as the integral test, is needed to classify the series.

Why the Logarithmic Threshold Matters

The exponent \(q\) controls a difference that the ratio and root limits cannot detect. A power of \(\ln n\) changes more slowly than a fixed exponential factor, so taking consecutive-term ratios or nth roots removes the distinction between \(q=2\), \(q=1\), and \(q=\frac12\). Yet the accumulated contribution of the terms is sensitive to that difference: the series converges only when the logarithmic power is strictly greater than one.

The practical lesson is to recognize the form before repeatedly applying a test that cannot decide it. For terms proportional to \(1/[n(\ln n)^q]\), compute the ratio and root limits to see the shared boundary value, then use an appropriate test that can resolve the logarithmic threshold.