isspecular parameter ignored in MMC at air-interface

[lvdtweel]

Dear @fangq and colleagues,

I have encountered discrepancies for how specular reflection at air-interface (escaping background into volume) is handled between MCX and MMC. Based on the documentation, this behavior seems to be unintentional.

The simulated volume contains a cube (label 1) with 30 mm edges sandwiched between two 3 mm thick layers of air (label 0) with properties ((0, 0, 1, 1), (1e-3, 1e-1, 0.5, 1.35)). A pencil source is placed off-center in the background with an angle of incidence 45° relative to the normal of the surface.

Expected behavior:

• isspecular = True: specular reflection into the background and refraction at the interface, resulting in the ray propagating into the medium at an angle determined from Snell's law.
• isspecular = False: no consideration of the refractive index mismatch at the air interface. The ray propagates at 0° relative to angle of incidence.

This behavior is observed in MCX: the ray refracts at the interface by default, and is propagating in a straight line when disabling isspecular.

The behavior is not expected when using MMC, as the beam is always refracted regardless of the value of the isspecular parameter. The parameter is set correctly, as it's value is reflected by the Dgcfgisspecular build argument used to build the kernel.

MCX (rev. aa62f2$v2025.10, pmcx 0.6.1) and MMC (rev. ae42a8$v2025.10, pmmc 0.3.6) were both installed from the wheels provided via pypi. The used system is based on Ubuntu 24.04 running on the Intel Cascade Lake platform with a NVIDIA L4.

The used Python environment and minimal working example in a Jupyter notebook (including outputs) can be found in this repository.

[fangq]

@lvdtweel, thanks for reporting this, but this is expected.

isspecular only handles one specific type of event - a launched photon moving from the background medium (in mcx - 0-valued voxels or outside of the bounding box; in mmc - 0-labeled element) to enter a non-background medium (label > 0) for the first time.

in your example, two layers are both assigned with non-zero labels (top being 1, bottom being 2). MMC considers this as internal reflection/transmission, and always applies the Fresnel's law.

if you want to trigger isspecular, you should set the top-layer's element ID to 0 instead of 1.

please give it a try and let me know if you can see expected result.

[lvdtweel]

Dear @fangq ,

Thank you for your prompt reply and for sharing your insight regarding the role of the background in how initial specular reflections are handled. I am however aware of the role of the background in specular reflections and believe I understand how these are handled by MCX and MMC.

In the example in the previously linked notebook, the last column of the element connectivity matrix is updated before assigning it to the elemprop parameter in the MMC configuration dictionary. Specifically:

• The outer layers, which were originally labeled 1 and 3, are reassigned to 0.
• The inner layer, which was originally labelled 2, is reassigned to 1.

elemprop = (el2[:, 4] == 2).astype(np.uint8)
mmc_config["elem"] = el2[:, :4]
mmc_config["elemprop"] = elemprop

The visualized mesh in the figure produced by cell 7 confirms this assignment. If this is indeed expected behavior, any guidance to adjust my input to reproduce the results obtained from MCX would be much appreciated.

Kind regards,

Lennard

[fangq]

thanks, yes, I noticed that you had set region labels.

I also noticed that isspecular flag was not really implemented in either CPU or GPU code in mmc.

check out this comment I made in this old post

https://groups.google.com/g/mmc-users/c/sN01hWcGUKQ/m/acIqtxc9BwAJ

I believe I had implemented a special flag in mmc to handle external boundary reflection, but I will need time to go over the code.