MathGroup Archive 2002

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

Search the Archive

Re: Question regarding Compile

  • To: mathgroup at smc.vnet.net
  • Subject: [mg33780] Re: Question regarding Compile
  • From: Madhusudan Singh <ar at sc.hc.de>
  • Date: Tue, 16 Apr 2002 03:50:31 -0400 (EDT)
  • Organization: University of Michigan Electrical Engineering and Computer Science
  • References: <a8m33c$oaq$1@smc.vnet.net> <a8tt4m$fmg$1@smc.vnet.net> <a90hj5$kfv$1@smc.vnet.net> <a939s1$pe3$1@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

Thanks for your response.

> > integrand := Compile[{{y, _Real}, {x1, _Real}, {x2, _Real}}, f1[pl[argument[y]] + x1, x2, c]];
> >
> > pl[] is an array, f1 and argument are functions (fairly simple functions).
> > c is a constant.
> >
> > answer1[x1_, x2_] := Exp[-2 Integrate[integrand[y, x1, x2], {y, a, b}]];
> >
> > a=0, b~1e-8
> >
> > I get an error :
> >
> > CompiledFunction::cfsa: Argument y at position should be a machine-size
> > real number.
>
>
> You can't symbolical integrate a compiled function !  and you can't pass
> a
> symbol like y (without a value) to a compiled function where you
> declared
> the argument as Real

So perhaps the solution is to expand integrand directly inside
the definition of answer1 and the resulting expression directly inside
jt3 ? That way, the variable y would not be an issue at all.

>
> >
> > when I try to use answer1 with
> >
> > jt3=q NIntegrate[v[x1, 50, 0.2] g3[0.2, x1, 50] partjt3[x1]
> > answer1[0.0, x1], {x1, 50, maxpl + dummyinfinity}, Method -> MonteCarlo, MaxPoints -> 10000];
> >
> > v, g3, partjt3 are simple functions.
> >
> > As you can see, this is quite complicated. However, since a and b are
> > fixed numbers, and nowhere near machine precision, I do not not understand
> > the error above.
> >
> > This is leading to other errors later.
> >
> > Any ideas ?
>
> Use NIntegrate[] instead of Integrate[]. There is no need to compile the
> argument of NIntegrate[] because Mathematica will do it by default.
>

I wonder why do calculations take so long then (one step of this
calculation takes as long as 40 minutes). Maybe I should try to use some
timing statements to identify the step that takes the longest to run.

Thanks.



  • Prev by Date: Making Tensor operations more efficient
  • Next by Date: RE: RE: How to Return from within a Table[]-Command?
  • Previous by thread: Re: Question regarding Compile
  • Next by thread: Using rules from Solve