The alternating series, $s_n=\sum_{i=2}^{n} (-1)^i \frac{1}{\ln i}$, (with $\ln$ being the natural logarithm) converges for $n\to\infty$ which can be seen e.g. by the Leibniz test. Can the limit be expressed in a finite closed expression of otherwise known constants?
I got interested while thinking about the values of the sine integral at integer multiples of $\pi$ where I suspect some relation.
Doing my usual pairing of even and odd, if $s_n =\sum_{i=2}^{n} (-1)^i \dfrac{1}{\ln i} $ then
$\begin{array}\\ s_{2n+1} &=\sum_{i=2}^{2n+1} (-1)^i \dfrac{1}{\ln i}\\ &=\sum_{i=1}^{n} (\dfrac{1}{\ln (2i)}-\dfrac{1}{\ln (2i+1)})\\ &=\sum_{i=1}^{n} \dfrac{\ln (2i+1)-\ln (2i)}{\ln (2i)\ln (2i+1)}\\ &=\sum_{i=1}^{n} \dfrac{\ln (1+1/2i)}{\ln (2i)\ln (2i+1)}\\ &<\sum_{i=1}^{n} \dfrac{1/2i}{\ln (2i)\ln (2i+1)}\\ &=\sum_{i=1}^{n} \dfrac{1}{2i\ln (2i)\ln (2i+1)}\\ \end{array} $
and this converges by comparison with $\sum \dfrac1{i \ln^2 i} $.
However, this does converge quite slowly. The integral test, using $\int_2^n \dfrac{dt}{t\ln^2(t)} =\dfrac1{\ln(2)}-\dfrac1{\ln(n)} $ shows how slowly. This, of course, agrees with the result that the sum of an alternating series is between any two consecutive partial sums.
For example, if $n = 10^{10}$, the error is about $\dfrac1{2\ln(10^{10})} =\dfrac1{20\ln(10)} \gt .01 $.
If we apply Euler's transform (https://en.wikipedia.org/wiki/Series_acceleration),
$\begin{array}\\ s &=\sum_{n=0}^{\infty} (-1)^n \dfrac{1}{\ln (n+2)}\\ &=\sum_{n=0}^{\infty} \dfrac{(-1)^n}{2^{n+1}}\sum_{k=0}^n(-1)^k\binom{n}{k} \dfrac{1}{\ln (n-k+2)}\\ &\approx\sum_{n=0}^{\infty} \dfrac{(-1)^n}{2^{n+1}}(\dfrac1{\ln(x+2)})^{(n)}|_{x=2}\\ &\approx\sum_{n=0}^{\infty} \dfrac{(-1)^n}{2^{n+1}}((-1)^{n-1}\dfrac{(n-1)!}{(x+2)^n\ln(x+2)})|_{x=2}\\ &=-\sum_{n=0}^{\infty} \dfrac{(n-1)!}{2^{3n+1}\ln(4)}\\ \end{array} $
and this actually diverges! (modulo any errors on my part)
That's enough for now.