Re: [Visualpython-users] vpython and ode solvers

[Eric A. <Ay...@ma...>]

I've been using Python (occasionally with vpython) for an undergrad- 
level computational physics course, and have had excellent results  
with an RK4 routine. Here's a sample program: http:// 
phys.csuchico.edu/ayars/250/code/spring-pendulum.shtml (Ignore the  
"error processing directive" messages on the webpage -- the page is  
not up to date at the moment but the code shown is good.) As you  
suggest, I solve all variables at once: in this example I use the  
array y[] as a container.

I've found that for displaying things on-screen in vpython, a python- 
based RK4 routine is not the speed-limiting factor. If you need to go  
with a c routine, though, you might consider writing your own and  
then using SWIG (http://www.swig.org/) as a way of making a python  
wrapper for it. I'm new to SWIG (as of this week, actually) but it is  
working very well for me so far.

Hope this helps.
-EA

-----------------------------------------------------------------
Dr. Eric Ayars
Assistant Professor of Physics
California State University, Chico
ay...@ma...


On Jun 22, 2005, at 10:53 AM, Gary wrote:

> I've been wondering if there would be a programming, accuracy, or  
> performance benefit to using a fancy ode solver with vpython rather  
> than the canonical v = v + F/m * dt especially as the number of  
> objects grows.  My thinking was that I could call the solver once  
> per "frame" (say, 1/30 sec), solve for all of the variables all at  
> once, and perhaps have better accuracy, and perhaps get more  
> objects into the simulation.
>
> I began to experiment with scipy's odeint, and got some demos  
> working.  I ran across some instabilites that I tracked down, but  
> before going any further I thought I'd ask:
>
> Does this seem worth the effort?
>
> Is there something I haven't considered that dooms this idea?
>
> Has someone already done this?
>
> The instablilty I found is related to in the way odeint (LSODA??)  
> manages the independent variable.  I'd like to look at other  
> solvers.  Any suggestions?  In particular I'd like to look at a 4th  
> order Runge-Kutta solver, but  the only RK4s that I've found are  
> written in python.  Does anyone know where there's a python-wrapped  
> C or Fortran RK solver?
>
> Thanks,
> Gary
>
>