MathGroup Archive 2000

[Date Index] [Thread Index] [Author Index]

Search the Archive

RE Looking for a faster way

In [mg24998] looking for a faster way  Otto Linsuain wrote:

> Hi to all.
> I am looking for a faster way to achieve the following:Have two lists 
> with the formats:
> xlist  = {x1, x2, x3, x4,...}
> flist  = {f1, f2, f3, f4, ....}
> the x's and f's are just numbers.Want the matrix:
> myMatrix[[i,j]]=(fi-fj)/(xi-xj). There are probably a thousand ways 
> to avoid the Indeterminate along the diagonal.Any of them is fine,
> as long as the off-diagonal part is not touched.Speed is the real 
> problem.My solution so far is:
>  myMatrix[xlist_,flist_]:=
>  Outer[Subtract,flist,flist]/(Outer[Subtract, xlist,xlist] + 
>        IdentityMatrix[Length[xlist]])
> the IdentityMatrix makes the diagonal zero,rather than Indeterminate.
> This solution is not very fast because it involves 2 Outers,so that 
> the speed goes roughly as 2 N^2. If I could avoid one of them
> I would have just one N^2,possibly+something-linear-in-N.
> ........
> ........
> For lists in the length range I am interested in (from 200 to 1000 
> elements) the second solution is slower than the first,despite the 
> fact that it uses only one Outer.If anyone can
> think of something faster than the first solution I would really 
> appreciate it.
> Otto Linsuain.

There is no need to work with Outer. I have tried the function
  myMatrix[xlist_,flist_]:= Map[(# - flist)&, flist] /
              ( Map[(# - xlist)&,xlist] + IdentityMatrix[Length[xlist]])
  and it works fine.
 xlist = Table[Random[],500];
 flist = Sin[xlist];
 Using the first method I got
 {37.9 Second,Null}

 Using the second method the result was much better:
 {2.5833 Second,Null}
 (I have a Power Macintosh 7500/100 running under OS 9.04)
 When I changed the number of elements in xlist to 1000,
 the first method the first method could not complete the computations
 complaining that there was not enough memory. The second method
 gave the result {9.46667 Second,Null}
 Ranko Bojanic
 bojanic at

  • Prev by Date: Mathematica Publication Question
  • Next by Date: Re: Simple integral wrong
  • Previous by thread: Re: Re: Mathematica Publication Question
  • Next by thread: discontinuous 3D plot and error messages