Calculus · real student question

Let f(x) = sum from n = 1 to infinity of x^n / n^2, the dilogarithm Li2(x). Is f differentiable at x = -1?

Question

Let

f(x)=n=1xnn2=Li2(x)f(x)=\sum_{n=1}^{\infty}\frac{x^n}{n^2}=\operatorname{Li}_2(x)

Is ff differentiable at x=1x=-1?

Step-by-step solution

  1. Locate x = −1 relative to the interval of convergence. The ratio test gives radius R=1R=1, and at both endpoints (±1)nn2\sum \tfrac{(\pm 1)^n}{n^2} converges absolutely (it is dominated by 1/n2\sum 1/n^2). So ff is defined on the closed interval [1,1][-1,1], and x=1x=-1 is an endpoint — which is exactly why the question is not automatic. Inside (1,1)(-1,1) differentiability is free; at an endpoint it must be argued.

  2. Differentiate term by term inside the interval. For x<1|x|<1,

    f(x)=n=1nxn1n2=n=1xn1n=ln(1x)xf'(x)=\sum_{n=1}^{\infty}\frac{n x^{n-1}}{n^2}=\sum_{n=1}^{\infty}\frac{x^{n-1}}{n}=-\frac{\ln(1-x)}{x}

    using n1tnn=ln(1t)\sum_{n\ge 1}\tfrac{t^n}{n}=-\ln(1-t) with t=xt=x and then dividing by xx.

  3. Test the differentiated series at the endpoint. Putting x=1x=-1 into the derivative series gives

    n=1(1)n1n=112+1314+\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n}=1-\frac12+\frac13-\frac14+\cdots

    The alternating harmonic series converges (by the alternating series test) to ln2\ln 2. This is the decisive difference from the case x=+1x=+1, where the same series becomes the divergent harmonic series and the derivative fails to exist.

  4. Upgrade convergence to differentiability with Abel's theorem. Since the differentiated series converges at x=1x=-1, it converges uniformly on [1,0][-1,0], and a uniformly convergent series of derivatives may be integrated/differentiated term by term up to and including the endpoint. Hence ff is differentiable at x=1x=-1 (one-sidedly from within [1,1][-1,1], and two-sidedly if one uses the analytic continuation of Li2\operatorname{Li}_2 past 1-1) with

    f(1)=n=1(1)n1n=ln2f'(-1)=\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n}=\ln 2

  5. Confirm with the closed form. The formula f(x)=ln(1x)xf'(x)=-\tfrac{\ln(1-x)}{x} is continuous at x=1x=-1 and gives

    f(1)=ln21=ln20.6931f'(-1)=-\frac{\ln 2}{-1}=\ln 2\approx 0.6931

    agreeing with the series value. (For reference, f(1)=(1)nn2=π212f(-1)=\sum \tfrac{(-1)^n}{n^2}=-\tfrac{\pi^2}{12}.)

    Yes; f(1)=ln20.693\boxed{\text{Yes; }f'(-1)=\ln 2\approx 0.693}

  6. Contrast the two endpoints. At x=1x=1 the difference quotient f(1)f(x)1x=n11+x++xn1n2\tfrac{f(1)-f(x)}{1-x}=\sum_{n\ge 1}\tfrac{1+x+\cdots+x^{n-1}}{n^2} tends to 1n=\sum \tfrac{1}{n}=\infty, so ff has a vertical tangent there and is not differentiable. The single extra factor of (1)n(-1)^n at x=1x=-1 turns a divergent series into a convergent one — the whole answer turns on that.

Answer

Yes: f(1)=n=1(1)n1n=ln20.693\text{Yes: }f'(-1)=\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n}=\ln 2\approx 0.693

Need to solve a different problem like this? Open the solver →