Function: asympnum
Section: sums
C-Name: asympnum0
Prototype: GD0,L,DGp
Help: asympnum(expr,{k=20},{alpha = 1}): asymptotic expansion of expr assuming
 it has rational coefficients with reasonable height; k and alpha are as
 in extnum.
Doc: Asymptotic expansion of \var{expr}, corresponding to a sequence $u(n)$,
 assuming it has the shape
 $$u(n) \approx \sum_{i \geq 0} a_i n^{-i\alpha}$$
 with rational coefficients $a_i$ with reasonable height; the algorithm
 is heuristic and performs repeated calls to limitnum, with
 \kbd{k} and \kbd{alpha} are as in \kbd{limitnum}
 \bprog
 ? f(n) = n! / (n^n*exp(-n)*sqrt(n));
 ? asympnum(f)
 %2 = []   \\ failure !
 ? l = limitnum(f)
 %3 = 2.5066282746310005024157652848110452530
 ? asympnum(n->f(n)/l) \\ normalize
 %4 = [1, 1/12, 1/288, -139/51840]
 @eprog\noindent and we indeed get a few terms of Stirling's expansion. Note
 that it helps to normalize with a limit computed to higher accuracy:
 \bprog
 ? \p100
 ? L = limitnum(f)
 ? \p38
 ? asympnum(n->f(n)/L) \\ we get more terms!
 %6 = [1, 1/12, 1/288, -139/51840, -571/2488320, 163879/209018880,\
       5246819/75246796800, -534703531/902961561600]
 @eprog\noindent If \kbd{alpha} is not an integer, loss of accuracy is
 expected, so it should be precomputed to double accuracy, say:
 \bprog
 ? \p38
 ? asympnum(n->-log(1-1/n^Pi),,Pi)
 %1 = [0, 1, 1/2, 1/3]
 ? asympnum(n->-log(1-1/sqrt(n)),,1/2)
 %2 = [0, 1, 1/2, 1/3, 1/4, 1/5, 1/6, 1/7, 1/8, 1/9, 1/10, 1/11, 1/12, \
   1/13, 1/14, 1/15, 1/16, 1/17, 1/18, 1/19, 1/20, 1/21, 1/22]

 ? localprec(100); a = Pi;
 ? asympnum(n->-log(1-1/n^a),,a) \\ better !
 %4 = [0, 1, 1/2, 1/3, 1/4, 1/5, 1/6, 1/7, 1/8, 1/9, 1/10, 1/11, 1/12]
 @eprog

 \synt{asympnum}{void *E, GEN (*u)(void *,GEN,long), long muli, GEN alpha, long prec}, where \kbd{u(E, n, prec)} must return $u(n)$ in precision \kbd{prec}.
 Also available is
 \fun{GEN}{asympnum0}{GEN u, long muli, GEN alpha, long prec}, where $u$
 must be a vector of sufficient length as above.