Unexpected photon trajectories for very low scattering coefficients about mcx HOT 4 CLOSED

thanks, I am able to reproduce this issue in both python and mcxlab. I agree with you both the scattering behavior and the out-of-bbx trajectories are incorrect. I will look into this and will post my updates here.

mcxlab test code is attached

%%octave

cfg.nphoton=1e4;
cfg.seed=42;

cfg.vol=permute(uint8(ones(32,32)), [3,1,2]); % from 2d to 3d
cfg.vol(1,10:22, 10:22)=2;
cfg.issrcfrom0=1;
cfg.srctype='pencil';

cfg.srcpos=[0,0,16];   % src position must be located in the 2D plane
cfg.srcdir=[0 1 0];    % src dir must align in the plan (y-z in this case)

cfg.gpuid=1;
cfg.autopilot=1;

cfg.prop=[0, 0, 1, 1,
    0.02, 1e-11, 0.9, 1,
    0.30, 80.25, 0.9, 1];
cfg.unitinmm=0.1;

cfg.tstart=0;
cfg.tend=5e-9;
cfg.tstep=5e-9;
cfg.outputtype='energy';

[flux,detp,vol,seeds,traj]=mcxlab(cfg);
sortedlines=mcxplotphotons(traj);

from mcx.

@dkalsan, again, thanks for reporting this issue.

I looked into it further, and was able to identify the cause - the handling of near-zero mus causes issues when combined with the new DDA ray-tracing approach implemented last year. The photon voxel indices (flipdir(1:3)) no longer follow the floating-point positions (p.{x,y,z}). this was not a problem before using the DDA based photon advancing.

after a one-line change, I believe this behavior has been fixed. see my above commit.

the above matlab demo code now returns correct trajectories, see below.

feel free to reopen the ticket if you see anything else is not fixed.

from mcx.

Interestingly, mcxcl/mcxlabcl does not have this behavior - it uses native_divide for the affected line, which appears to be able to handle divided by 0.

from mcx.

Perfect, I will give it a test. Thanks a lot for taking the time for this!

from mcx.