Any tips for debugging MCNP geometry errors?

[Kirk Truax]

Hey all,

I was wondering if anyone had any good tips on debugging mcnp geometry? I'm an intermediate user working on better understanding the program. Does anyone have any tips or tricks that go beyond simply reading the manual?

 

[PSRB191921]

Hi
I think there are no easy answers or ready-made solutions. For my part I use a X-server (X-ming for me) and look at the geometry in XY, XZ AND YZ. But it is true that for complex geometry you can not see everything. With x-ming you can also see the cookie-cutter.
PSR

 

[Stephan_doc]

or you can try using MCNPX Visual Editor

 

[quarkle]

If you have a tiny or non-obvious geometry error (caused by rounding off errors, overlaps, skewed lattice etc) that you can't find with the MCNP plotter or Visual Editor, the following technique can help you pinpoint the error:
- Change all cells to having a void material, but a non-zero neutron/photon importance.
- Surround the entire geometry with a void sphere, again non-zero particle importance
- Surround the sphere with a zero importance void as usual.
- Create a generic point source somewhere within the geometry.
Run MCNP for a very high number of histories. If there are any geometry errors, the run will terminate because of lost particles. You can find the coordinates of the lost particles in the output and file and use those to pinpoint the error.

 

[mha29054]

I did what quarkle said and strangely without material the error disappeared, Is that mean losing particles are not caused by geometry error ?

 

[Alex A]

Show us the input file, or a simplified input file that still shows the issue. The trick is to make things fast for large numbers of particles. If it's fixed the issue, you may have an obvious undefined area around your problem. Or you didn't set the importance correctly.

 

[mha29054]

No sorry, I found the geometrical error , maybe number of particles were not large enough in the previous run (without material) to reach the erroneous parts of geometry.