Suppose I have a function $f(s,t) = [(1-t^2)(1-s^2t^2)]^{-1/2}$.

Is there a way to get the general coefficient in this power series of the form $s^{2k} t^{2n}$?


As Artes shows, Series can produce selected coefficients. We can go further: FindSequenceFunction finds closed forms for lists. We have to commit some small sins:

  • We assert that odd coefficients are 0 with opportune [[;; ;;2]]s, not too bad
  • We assert that there are closed forms for each coefficient list $(a_n(k)s^{2k}t^{2n})_n$, and
  • The general coefficient $a(n,k)=a_n(k)$ will admit an obvious closed form.

So we try

FindSequenceFunction@#[[;; ;;2]]&/@CoefficientList[Series[
],{s,t}][[;; ;;2]]

which yields


where there are a few garbage elements towards the end of the list. This is promising. Type


to obtain Pochhammer[1/2,#1-1]/Pochhammer[1,#1-1]&. It would seem


is the series expansion we are looking for. Obviously this method is highly manual. Still, Mathematica was of assistance.

Expanding some definitions and applying some identities, we obtain $$f(s,t)=\sum_{n=0}^\infty\sum_{i=0}^\infty\frac{(-1)^n\prod_{j=1}^i\left(j-n-\frac12\right)}{\Gamma(n+1)\Gamma(1+i-n)}s^{2n}t^{2i}.$$ I think those two $\Gamma$'s can be eliminated, but I'm not very good at complex analysis. I don't know where this is convergent.

Here's a coinciding plot of the two functions

  ((-1)^n Product[j-n-1/2,{j,i}])/(Gamma[n+1]Gamma[1+i-n])s^(2n)t^(2i),{n,0,10},{i,0,10}];

Assuming that one deals with series expansion around $(s,t)=(0,0)$ we can define

cf[k_, n_] := 
  Series[1/Sqrt[(1 - t^2) (1 - s^2 t^2)], {s, 0, 2 k}, {t, 0, 2 n}] // 
  Normal // Coefficient[#, s^(2 k) t^(2 n)] &

We have defined cf to get only even coefficients (2k,2n), since the others do vanish.

Table[cf[k, n], {k, 6}, {n, 6}] // MatrixForm

enter image description here

One can get the same with Array as well

Array[ cf, {6, 6}] == Table[ cf[k, n], {k, 6}, {n, 6}]

This appraoch would be satisfactory for small {n, k}. For bigger ones there is SeriesCoefficient (as another answer recalled).

sf[k_, n_]:= SeriesCoefficient[ 1/Sqrt[(1 - t^2) (1 - s^2 t^2)], {s, 0, 2k}, {t, 0, 2n}]

Array[ cf, {6, 6}] == Array[ sf, {6, 6}]

And sf is much better than cf:

AbsoluteTiming[sf[350, 400];]
{0.0015182, Null}
AbsoluteTiming[cf[350, 400];]
{10.2873, Null}

The standard way to do this would be to simply use SeriesCoefficient, e.g.

SeriesCoefficient[1/Sqrt[1 + t^2], {t, 0, n}]

produces the general form

(I^n (1/2 (-1 + n))!)/(Sqrt[\[Pi]] (n/2)!)

for Mod[n, 2] == 0 && n >= 0. However, it seems to have trouble with your particular function, so you may need to resort to guessing as @Adam has shown...


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.