olymmath

Putnam 2015 B6

  • Session B

For each positive integer kk, let A(k)A(k) be the number of odd divisors of kk in the interval [1,2k)[1, \sqrt{2k}). Evaluate ∑k=1∞(−1)k−1A(k)k.\sum_{k=1}^\infty (-1)^{k-1} \frac{A(k)}{k}.

Curated text Kedlaya Putnam archive. Quoted with source.

Solutions

Human solution Solution 1 by Manjul Bhargava, Kiran Kedlaya and Lenny Ng

The terms of the source let olymmath link to this solution but not copy it. Read the solution (Kedlaya Putnam archive).

Human solution Solution 2 by Eric Naslund

We have that ∑k=1∞(−1)k−1A(k)k=(∑k=1∞(−1)k−12k−1)2=π216.\sum_{k=1}^\infty (-1)^{k-1}\frac{A(k)}{k}=\left(\sum_{k=1}^\infty \frac{(-1)^{k-1}}{2k-1}\right)^2=\frac{\pi^2}{16}.

Firstly, we would like to rearrange the terms in this series, but we have to be very careful in doing so as this is a conditionally convergent series. To do things rigorously, we have to truncate and take limits, and so define T(x)=∑k≤x(−1)k−1A(k)kT(x)=\sum_{k\leq x}(-1)^{k-1}\frac{A(k)}{k} so that T=lim⁡x→∞T(x)T=\lim_{x\rightarrow\infty}T(x) is the series in question. Writing A(k)=∑d∣k, d oddd≤2k1A(k)=\sum_{\begin{array}{c} d|k,\ d\text{ odd}\\ d\leq\sqrt{2k} \end{array}}1 and switching the order of summation we obtain T(x)=∑k≤x(−1)k−1k∑d∣k, d oddd<2k1=∑d<2xd odd∑d2/2<k≤xd∣k(−1)k−1k=∑d<2xd odd1d∑d2<k≤xd(−1)k−1k.T(x)=\sum_{k\leq x}\frac{(-1)^{k-1}}{k}\sum_{\begin{array}{c} d|k,\ d\text{ odd}\\ d<\sqrt{2k} \end{array}}1=\sum_{\begin{array}{c} d<\sqrt{2x}\\ d\text{ odd} \end{array}}\sum_{\begin{array}{c} d^{2}/2<k\leq x\\ d|k \end{array}}\frac{(-1)^{k-1}}{k}=\sum_{\begin{array}{c} d<\sqrt{2x}\\ d\text{ odd} \end{array}}\frac{1}{d}\sum_{\frac{d}{2}<k\leq\frac{x}{d}}\frac{(-1)^{k-1}}{k}. Let S(x)=∑k≤x(−1)k−1kS(x)=\sum_{k\leq x}\frac{(-1)^{k-1}}{k} and note that S(xd)=log⁡2+O(dx)S\left(\frac{x}{d}\right)=\log2+O\left(\frac{d}{x}\right). Hence T(x)=∑d≤2xd odd1d[S(xd)−S(d2)]=∑d≤2xd odd1d[log⁡2−S(d2)]+O(1x)T(x)=\sum_{\begin{array}{c} d\leq\sqrt{2x}\\ d\text{ odd} \end{array}}\frac{1}{d}\left[S\left(\frac{x}{d}\right)-S\left(\frac{d}{2}\right)\right]=\sum_{\begin{array}{c} d\leq\sqrt{2x}\\ d\text{ odd} \end{array}}\frac{1}{d}\left[\log2-S\left(\frac{d}{2}\right)\right]+O\left(\frac{1}{\sqrt{x}}\right) and so upon taking x→∞x\rightarrow \infty we have that T=∑d odd1d∑k>d2(−1)k−1k=∑k=1∞(−1)k−1k∑d<2kd odd1d.T=\sum_{d\text{ odd}}\frac{1}{d}\sum_{k>\frac{d}{2}}\frac{(-1)^{k-1}}{k}=\sum_{k=1}^{\infty}\frac{(-1)^{k-1}}{k}\sum_{\begin{array}{c} d<2k\\ d\text{ odd} \end{array}}\frac{1}{d}. Now, we can write this as 12∑k=1∞(−1)k−1k∑j=1k(12j−1+12k−(2j−1)).\frac{1}{2}\sum_{k=1}^{\infty}\frac{(-1)^{k-1}}{k}\sum_{j=1}^{k}\left(\frac{1}{2j-1}+\frac{1}{2k-(2j-1)}\right). Truncating the first sum at k=Nk=N, we have T=lim⁡N→∞∑j=1N∑k=jN(−1)k−1(2j−1)(2k−(2j−1)).T=\lim_{N\rightarrow \infty}\sum_{j=1}^{N}\sum_{k=j}^{N}\frac{(-1)^{k-1}}{(2j-1)(2k-(2j-1))}. Shifting the second sum to start at k=1k=1, this becomes T=lim⁡N→∞∑j=1N(−1)j−1(2j−1)∑k=1N−j+1(−1)k−1(2k−1)=lim⁡N→∞(∑j=1N(−1)j−1(2j−1))2−lim⁡N→∞∑j=1N(−1)j−1(2j−1)∑N−j+2N(−1)k−1(2k−1).T=\lim_{N\rightarrow\infty}\sum_{j=1}^{N}\frac{(-1)^{j-1}}{(2j-1)}\sum_{k=1}^{N-j+1}\frac{(-1)^{k-1}}{(2k-1)}= \lim_{N\rightarrow\infty}\left(\sum_{j=1}^{N}\frac{(-1)^{j-1}}{(2j-1)}\right)^2-\lim_{N\rightarrow \infty}\sum_{j=1}^{N}\frac{(-1)^{j-1}}{(2j-1)}\sum_{N-j+2}^{N}\frac{(-1)^{k-1}}{(2k-1)}. Now, the second term goes to zero, and the first term is the square of the well known Leibniz series for π/4\pi/4. Thus ∑k=1∞(−1)k−1A(k)k=π216.\sum_{k=1}^\infty (-1)^{k-1}\frac{A(k)}{k}=\frac{\pi^2}{16}.

Mathematics Stack Exchange. Open licence.

Locator putnam/2015/b6 · Tag 9271H6QRQY · Edit this page on GitHub · Keys: j next, k previous