Re: matlab code: isovalues and isosurface

[Martin Helm]

Am Samstag, 16. Juli 2011, 12:46:13 schrieb Liam Groener:
> On Jul 16, 2011, at 2:43 AM, Martin Helm wrote:
> > Am Freitag, 15. Juli 2011, 23:53:44 schrieb Liam Groener:
> >> (2) What I find very confusing is that the failure occurs almost
> >> instantly when the for loop is run from 1 to n, but the code runs for
> >> more than a minute (without failue) if the loop is run from 1 to n-1.
> >> It seems like it is actually running it from n to 1 in the former case?
> > 
> > I guess you are using gnuplot as backend? You did not mention it. The
> > reason that the error comes almost instantly is that the loop is very
> > fast, it is finished long before gnuplot starts to show the first plot.
> > So it has nothing to do with the loop running in the wrong order as you
> > suspected but seems to be simply a side effect of the asynchronous
> > behaviour of the plots running in a separate process while the
> > calculations are all finished.
> 
> Yes, you are right. I was fooled by the facts that all the plotting
> commands are inside the loop and that most of the cpu time was being used
> by the octave process rather than by gnuplot. I guess that during the
> loop, the gnuplot commands are placed in a que and octave continues
> feeding them to gnuplot over the length of the run?
> 
> I have been running with both the gnuplot and fltk backends. I swear I was
> getting similar plots from both earlier in the day (actually yesterday
> now). However, recently, I have not been able to make fltk work. It just
> plots the axes and finishes with no actual plot or error message. The
> m-file I'm currently using reads:
> 
> [x,y,z]=meshgrid(-20:.5:20);
> function Psi=FdOn320(x,y,z)
>  r=sqrt(x.^2+y.^2+z.^2);
>  Psi=exp(-r./3).*(3*z.^2-r.^2).*2*sqrt(15/(120*pi))/81;
> endfunction
>  Psi=FdOn320(x,y,z);
>  m=min(Psi(:));
>  M=max(Psi(:));
>  color=[1 1 0;1 0 1; 0 1 1;0 1 0;1 0 0; 0 0 1; 0 0 0];
>  n=5;
>  for i=1:n-1
>      isovalue=m+i*(M-m)/n;
>      fv = isosurface(x,y,z,Psi,isovalue);
>      hpatch = patch(fv);
>      isonormals(x,y,z,Psi,hpatch);
>      Alphalevel=0.2+0.8*i/n;
>      
> set(hpatch,'FaceAlpha',Alphalevel,'FaceColor',color(i,:),'EdgeColor','none
> ') set(hpatch,'FaceColor',color(i,:),'EdgeColor','none')
>       daspect([1,1,1])
>       view(3);
>       axis tight
>  %     camlight left;
>  end
>  s = 'hi'
> 
> Earlier, I was trying small variations in this. Note that I uncommented out
> the daspect command since octave 3.4.2 supports it. I just added the last
> line, my crude way of verifying that the loop finishes fast as you said.

I quickly checked that with fltk since I remembered that the transparency did 
not work with fltk when I last tried it.
If you remove the line with the FaceAlpha setting the graphics will be 
rendered. Choosing
set(hpatch,'FaceColor',color(i,:),'EdgeColor','black')
in the loop gives the attached image.

surface.png
Description: PNG image