Absorbance coefficient mu_a's impact on the simulation result.

[Wenxin Fang]

Hi Docter Fang,

I was trying to use mcxcl on my computer and found out that the change of mu_a in cfg.prop did not change the result partial pathlength. From figure 1 in your paper "Monte Carlo simulation of photon migration in 3D turbid media accelerated by graphics processing units" , it has a step "Computer attenuation based on absorption", which I assume will be changed with mu_a. Also when I rerun the simulation and no seed was set, the results were exactly the same. Could you please help me with both problems? 

Thank you.

Wenxin

[Fang, Qianqian]

Hi Wenxin,

There are several flavors of the Monte Carlo simulations. MCX/MMC uses something called the microscopic Beer-Lambert law (mBLL) approach (as shown in this paper https://pubmed.ncbi.nlm.nih.gov/23201658/ by Angelo Sassaroli and Fabrizio Martelli).

In this flavor of MC, mus/g/n determines the photon path/trajectories, and mua determines the weight.

When changing mua, path does not change, but the weight does. Therefore, when you perform your analysis, you must consider the updated weights to reflect mua changes.

On your second question, if you do not set a seed, mcx uses a default seed of 1648335518, see

https://github.com/fangq/mcx/blob/v2025/src/mcx_utils.c#L5556

I also want to highlight that exact reproducibility of any output quantity that are accumulated over multiple threads, such as fluence, is not guaranteed in mcx. Even using the same seed, running on the same GPU may yield nearly identical results, it can produce different results when running at a different time, OS, or over different GPUs. More discussions on this can be found in several previous discussion threads

https://groups.google.com/g/mcx-users/search?q=reproducibility

Qianqian

---

From: 'Wenxin Fang' via mcx-users <mcx-...@googlegroups.com>
Sent: Tuesday, March 4, 2025 1:23 AM
To: mcx-users <mcx-...@googlegroups.com>
Subject: [mcx-users] Absorbance coefficient mu_a's impact on the simulation result.

 

--
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 visit https://groups.google.com/d/msgid/mcx-users/80bd5c52-29e4-4fd3-8cca-86acf405087dn%40googlegroups.com.

[Wenxin Fang]

Hi Professor Fang,

Thank you for getting back to me. I am currently building a two-layer model with the same properties, I was wondering if the mean path length should be the same for both layers if the photon number launched is large enough. I am running it on my PC; thus, the photon number is relatively small, making it difficult to validate my thoughts.

Thank you,

Wenxin

[Qianqian Fang]

if they have the same mus, then their (scattering) mean-free-paths (1/mus) should the same.

To view this discussion visit https://groups.google.com/d/msgid/mcx-users/bb635311-9ea1-4520-8d5b-c2dfa686d1can%40googlegroups.com.