# How can I invert this series?

I have a series which is expanded around $x=0$:

$$S=2 \sum_{n=0}^4 x^n (c_n\ln x+b_n)$$

where $c_n=-\frac{1}{2}\left(\frac{(2n-1)!!}{2n!!}\right)^2$ and $b_n=-c_n\left(4 \ln 2+2\sum_{k=1}^n(\frac{1}{k}-\frac{2}{2k-1})+\frac{2}{2n-1}\right)$
I want to invert this series(i.e. x of series of function of S) using Mathematica, but it doesn't give me the result. How can I get this?

  c[n_] := -((1/2) ((2 n - 1)!!/(2 n)!!)^2/(2*n - 1));
b[n_] := (-c[n])*(4*Log + 2*Sum[1/k - 2/(2*k - 1), {k, 1, n}] +
2/(2*n - 1));
S = Series[Sum[x^n (c[n] Log[x] + b[n]), {n, 0, 4}], {x, 0, 4}]

• – AccidentalFourierTransform Apr 9 '18 at 22:46
• Works for x=1 :InverseSeries[Series[Sum[x^n (c[n] Log[x] + b[n]), {n, 0, 4}], {x, 1, 4}]] – Mariusz Iwaniuk Apr 9 '18 at 22:47
• I already tried with that, it didn't give the result. – m. bubu Apr 9 '18 at 22:49
• but I need for X=0. – m. bubu Apr 9 '18 at 22:52
• I have doubts about the question as the series is not a monotone function in $0<x<\infty$ – José Antonio Díaz Navas Apr 10 '18 at 10:17

You would like the series expansion of $x(S)$ around $x=0$, however $S$ is infinite at $x=0$ due to the presence of $\log(x)$, so you are effectively looking for an expansion at non-analytic point.

You can shift the singular point away from the origin for the expansion by using $\log(x+\epsilon)$ instead of $\log(x)$ and take the leading order term in $\epsilon$. With this, you can indeed directly use InverseSeries command, which gives

  Normal[Series[InverseSeries[Series[Sum[x^n (c[n] Log[x + \[Epsilon]]+ b[n]), {n, 0, 4}], {x, 0, 4}], S], {\[Epsilon], 0, 1}]]


$$\frac{\epsilon (-b(0)-c(0) \log (\epsilon )+S)^4}{24 c(0)^4}+\frac{\epsilon (-b(0)-c(0) \log (\epsilon )+S)^3}{6 c(0)^3}+\frac{\epsilon (-b(0)-c(0) \log (\epsilon )+S)^2}{2 c(0)^2}+\frac{\epsilon (-b(0)-c(0) \log (\epsilon )+S)}{c(0)}$$

At this point, if you take $\epsilon\to 0$, you get the result 0. However, you can think of this as if you are working with the function $x(S,\epsilon)$ instead of $x(S)\equiv x(S,\epsilon=0)$, and then take $\epsilon\to 0$ at the very end of your overall calculations.

By the way, notice that at leading order in $\epsilon$, only $n=0$ terms contribute.

More natural way to do this than the approach above is to consider the series expansion around a non-singular point, say $x=1$ as @Mariusz suggests.

One also observes same problem for similar cases; for example, if we replace $\log(x)$ with $e^{1/x}$, we get the same problem, and same approach, replacing $x$ with $x+\epsilon$ works there too.

Note that I am not taking into account how you deal with the branch cut: Depending on the specific problem at hand, you may prefer $\log(x\pm\epsilon)$ or even $\log(x\pm i \epsilon)$ for positive infinitesimal $\epsilon$.

I think you are interested in the behavior of x when S is near zero. If so, you can use the new in M12 function AsymptoticSolve. Here is your sum:

sum[x_] = Normal[S];
sum[x] //TeXForm


$$\frac{5 x^4 (-210 \log (x)-473+840 \log (2))}{196608}+\frac{x^3 (-15 \log (x)-31+60 \log (2))}{1536}+\frac{1}{128} x^2 (-3 \log (x)-5+12 \log (2))+\frac{1}{8} x (4 \log (2)-\log (x))+\frac{1}{2} (\log (x)+2-4 \log (2))$$

First we need to find the value of x when S is 0 (there are two solutions, so I arbitrarily pick the first):

rt = x /. First @ Solve[sum[x] == 0, x, Reals];
rt //InputForm

Root[{196608 - 393216*Log + 98304*Log*#1 - 7680*#1^2 + 18432*Log*#1^2 -
3968*#1^3 + 7680*Log*#1^3 - 2365*#1^4 + 4200*Log*#1^4 +
Log[#1]*(98304 - 24576*#1 - 4608*#1^2 - 1920*#1^3 - 1050*#1^4) & , 1.0053977489978152654390529458805439671720.60186555560688}]


Now, use AsymptoticSolve:

g[s_] = x /. First @ AsymptoticSolve[s == sum[x], {x, rt}, {s, 0, 5}] //N


1.0054 + 1.31037 s + 0.731207 s^2 + 0.36329 s^3 + 0.421708 s^4 + 0.600283 s^5

I used N because the unnumericized version is quite messy. Let's check:

Plot[{sum[g[s]], s}, {s, -1, 1}]
` It looks like the approximation is pretty good for $$-.5 < s < .5$$.