Message: Re: Replicas vs Navigation for large voxel numbers Not Logged In (login)
 Next-in-Thread Next-in-Thread
 Next-in-Forum Next-in-Forum

None Re: Replicas vs Navigation for large voxel numbers 

Forum: Geometry
Re: None Replicas vs Navigation for large voxel numbers (Ben Morgan)
Re: Feedback Re: Replicas vs Navigation for large voxel numbers (Pedro Arce)
Date: 10 Dec, 2009
From: Ben Morgan <Ben Morgan>

Hi Pedro,
On Mon, 07 Dec 2009 13:47:55 GMT, Pedro Arce wrote:
> If you use G4PhantomParameterisation there is an option to avoid filling
> the material indices set (sorry, I will document it): if you do not
> define the material indices, it will always return index = 0 (the first
> material in your list).

Thanks! I tried that and it will run with 1000x1000x1000 voxels! It's about a factor of 3 slower than my 'virtual' navigation method, though there seems a significant start up time (e.g. running 100 events doesn't take a factor of 100 longer than 1 event) - so for large jobs it may not be so bad.

I did wonder about adding the step length correction to my 'virtual' navigator to enable a full comparison (as in G4PSDoseDeposit_RegNav). However, do I understand correctly that G4RegularNavigationHelper::theStepLengths records the actual step lengths made in each voxel - rather than fractional lengths of the geometric step passed to G4PSDoseDeposit_RegNav::ProcessHits? If that's the case I guess my virtual navigator won't be able to accurately score energy deposits...

> In any case, I think the memory problem you will have will not be
> related to the material indices, but to the map that is used for keeping
> the deposited energy in each voxel. Geant4 scorers use a
> std::map, and I found that this occupies 50 Mb of
> memory for each 1 million entries. You may write your own scorers using
> two std::vector, that occupy "only" 10 Mb for 2X1 million entries.

So far this seems to be o.k. - I'm using my own scorer with a std::map for now, although only for my 'virtual' navigator at the moment.

> The regular parameterisation is of course thought for box-like
> geometries, but if your volume is a tube you may use a parallel geometry
> for scoring.

I'll have a look at using parallel worlds for scoring - I'd rather avoid this if possible though it'll be useful to compare with regular navigation.

Thanks again for the help!

Ben.

 Add Message Add Message
to: "Re: Replicas vs Navigation for large voxel numbers"

 Subscribe Subscribe

This site runs SLAC HyperNews version 1.11-slac-98, derived from the original HyperNews


[ Geant 4 Home | Geant 4 HyperNews | Search | Request New Forum | Feedback ]