different default behaviour between pmcx and regular mcx w.r.t. focusing

[G Buist]

Dear Qianqian,

I have been using pmcx and I noticed that I need to set the fourth element of 'srcdir'  to zero to prevent photons from being launched in random directions. While in MCX not specifying the last element of srcdir gives me the standard collimated beam and I believe I have to set the fourth element of srcdir to -inf to get this random direction behaviour.

Is this difference in behaviour between pmcx and mcx intended?

Best,

Gijs Buist

[Qianqian Fang]

hi Gijs,

I did not observe this behavior, at least in all the examples I tested in this google-colab notebook

https://colab.research.google.com/github/fangq/mcx/blob/master/pmcx/tutorials/pmcx_getting_started.ipynb

just like mcxlab, pmcx accepts either a 3-element srcdir, or a 4-element srcdir,

https://github.com/fangq/mcx/blob/540931d07dc32bfb6faf6b53f548655aa70499e1/src/pmcx.cpp#L462

if 3 elements were specified, the 4th element stays as the initial value of 0 initialized at

https://github.com/fangq/mcx/blob/540931d07dc32bfb6faf6b53f548655aa70499e1/src/mcx_utils.c#L317

also, I am not quite sure what you meant as "random direction" - what was your srctype? do you have a test code?

when srcdir(4) was accidentally set, the source should behave either like a convergent (srcdir(4)>0) or divergent (srcdir(4)<0), and only becomes isotropic launch when srcdir(4) is set to NaN.

Qianqian

--
You received this message because you are subscribed to the Google Groups "mcx-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mcx-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mcx-users/68993e6a-92cd-4dfb-a00c-3df268fed562n%40googlegroups.com.

[G Buist]

Hi Qianqian,

Thank you for your response.

To explain my problem and answer your questions: 

I did not observe this behavior, at least in all the examples I tested in this google-colab notebook.

Below this message is some code where I simulate a Gaussian beam in pmcx first without specifying the 4th field of srcdir (so srcdir = [0, 0, 1]) and then the same pmcx simulation where the 4th element is set to 0 ( so srcdir = [0, 0, 1, 0]  ). The results of both are not the same with the first simulation resulting in a diverging beam while the second simulation results in the collimated Gaussian that was also expected for the first. 

The pmcx version that was used is 0.0.12.

Also, I am not quite sure what you meant as "random direction" - what was your srctype? do you have a test code?

When I first tested this with a collimated Gaussian beam I expected a collimated beam but the photons propagated in every direction which I assumed was random but based on your explanation it seems to be a diverging but I have not checked how strong the divergence is.

import numpy as np

import pmcx

import jdata as jd

from matplotlib import pyplot as plt

pmcx.__version__   # print imported pmcx version number

### pmcx version is '0.0.12'

cfg = {}

cfg['nphoton'] = 1e7

cfg['vol'] = np.ones([60, 60, 60], dtype='uint8')

cfg['tstart'] = 0

cfg['tend'] = 5e-9

cfg['tstep'] = 5e-9

cfg['prop'] = [[0, 0, 1, 1], [0.00005, 0.0, 1, 1]]

cfg['srctype'] = 'gaussian'

cfg['srcpos'] = [30,30,0]

cfg['srcdir'] = [0, 0, 1]

cfg['srcparam1'] = [3, 0, 0, 0]

res = pmcx.run(cfg)

flu = res['flux'][:, :, :, 0]

tuplemax = (np.max(flu[:, :, :]) == flu).nonzero()

xmax, ymax, zmax = tuplemax[0][0], tuplemax[1][0], tuplemax[2][0]

plt.figure()

plt.title('crossection at x = '+str(xmax))

plt.imshow(np.log10(flu[xmax, :, :].T))

plt.xlabel('latteral position')

plt.ylabel('axial position')

plt.colorbar(label='intensity (a.u.)')

plt.tick_params(left = False, right = False , labelleft = False ,

                labelbottom = False, bottom = False)

plt.show()

### results look like a diverging beam

### now we set the 4th element of srcdir to 0 and try again

cfg['srcdir'] = [0, 0, 1, 0]

res2 = pmcx.run(cfg)

flu = res2['flux'][:, :, :, 0]

tuplemax = (np.max(flu[:, :, :]) == flu).nonzero()

xmax, ymax, zmax = tuplemax[0][0], tuplemax[1][0], tuplemax[2][0]

plt.figure()

plt.title('crossection at x = '+str(xmax))

plt.imshow(flu[xmax, :, :].T)

plt.xlabel('latteral position')

plt.ylabel('axial position')

plt.colorbar(label='intensity (a.u.)')

plt.tick_params(left = False, right = False , labelleft = False ,

                labelbottom = False, bottom = False)

plt.show()

#### reslults in a collimated Gaussian beam

Best,

Gijs Buist

Op maandag 19 juni 2023 om 21:36:50 UTC+2 schreef Qianqian Fang:

[Qianqian Fang]

hi Gijs,

after looking into this, I realized that this is a critical bug that was introduced from the very first version of pmcx - the code incorrectly assumes the default source focal length (4th element of srcdir) to be 1.0, instead of 0 (which ignores the focal length setting).

this bug affects all simulations using area-sources (focus-able sources), such as disk, fourier, planar and others; it does not affect point sources including pencil/isotropic/cone.

This bug has just been fixed with the following:

https://github.com/fangq/mcx/commit/eaf31def1f5c48f1b50f2805c0e238a924c8371b

https://github.com/fangq/mcx/issues/180

a new version of pmcx, v0.1.1, has been just uploaded to pypi

This is a severe bug and I urge everyone who are using pmcx to upgrade your version using the below command

python3 -m pip install --upgrade pmcx

I apologize that this issue has not been detected until now, but I am glad that we captured this before pmcx is picked up by more users.

Matin: please see my commit - I believe the recent digimouse issue you found was related to this bug. Please update pmcx and run again.

Qianqian

To view this discussion on the web visit https://groups.google.com/d/msgid/mcx-users/a0e750c7-506a-4fd9-af87-8e6cfdbc2bd4n%40googlegroups.com.