Problem with TAYLOR / corrected text



(Resend it because of some dumb textual mistakes:)

Am 01.06.2007 21:55 schrieb Stavros Macrakis:
> Taylor expands into a series a0 + a1*x + a2*x^2 + ...., where the
> exponents of x are constant integers.
>
> It can only expand into series like a0 + a1*x^m + a2*x^(2*m) + ...,
> where m remains symbolic, under rather restricted circumstances.
>
> There are two approaches you could take here.  One is to expand
> 1/(q+1)^(1/m) then substitute q=x^m; the other is to use the powerseries
> routine.
>
> powerseries can expand some series with closed-form coefficience.  In
> this case, for example:
>
>         powerseries(1/(x^m+1)^(1/m),x,0) =>
>           ('sum(x^(i1*m)/beta(-1/m-i1+1,i1+1),i1,0,inf))/(1-1/m)
>
> two-dimensional display:
>
>     inf
>     ====              i1 m
>     \                x
>      >     --------------------------
>     /             1
>     ====   beta(- - - i1 + 1, i1 + 1)
>     i1 = 0        m
>     ---------------------------------
>                       1
>                   1 - -
>                       m
>
Hi Stavros,

 yes, this works fine, thanks.

 But now I don't know, whether I have a mathematical or
 still a technical problem...

 If I cancel in the above sum, (taking care not to use m,
 which lead to divergences) I get a simple binomial-formula
 with the main terms of binomial(1/m , i1)

 as
        sum(i1=0,inf,  binomial(1/m, i1)*(x^m)^i1 )

 where the *first* parameter in binomial(1/m,i1) is constant.

 What I was expecting was, that the first parameter would be
 *varying* and the second would be constant...

 Example:

 Assume the pascal-matrix, containing the binomial-coefficients,
P=
  1
  1  1
  1  2  1
  1  3  3  1
  ...
PInv = P^-1 =
  1
 -1  1
  1 -2  1
 -1  3 -3  1
  ...

and the powerseries-vector V(x) = column([1,x,x^2,x^3,... ])

Then the result expresses, with m=1,

   P * V(x) = V(1+x)

 and the first parameters in the binomials, taken from a row of P, are constant,
 which is reflecting the maxima-result.

 What I wanted was (if m=1)

 1/x* V(1/x))~ * PInv = 1/(1+x) * V(1/(1+x))~

 and here the coefficients of PInv are read column-wise, thus the
 first parameter of the binomials is varying and the second is
 constant.

 So, to define the generating-function for the first column of
 PInv I use (with approprate scaling)

          x
 TAYLOR( --- ,x,0,4) = x V(x)~ * column([1,-1,1,-1,...])
         1+x

 for the second column I use

           x
 TAYLOR( (---)^2,x,0,4 ) =  x^2 V(x)~ * column([1,-2,3,-4,...])
          1+x

 for the second column I use

           x
 TAYLOR( (---)^3,x,0,6 ) =  x^3 V(x)~ * column([1,-3,6,-10,...])
          1+x

 and so on, where binomials are produced, which are varying in the first
 parameter and constant in the second.


 The exponent m at x then should provide a variability of the stepwidth,
 when PInv(m) is understood as an operator, which transfers 1/x V(1/x) to 1/y V(y),
 where y = x+1 or varying stepwidths like y^m = x^m+1
 and I thought it would be an appropriate idea to replace x by x^m and revert this
 by powering of the whole expression by 1/m.

 But how can then the form of the resulting formula change to an
 expression, which reflects a P*V(x)-multiplication instead a
 V(1/x)~ * PInv - multiplication???

 Or is this operation ambiguous, are both ways possible and maxima
 just gets the simpler one?

 * scratchhead *

Gottfried Helms