NTOT vs SUMDES?

[alan (Alan Van Drie)]

I’m I correct that variables: An, Dn, Hn SUMABS, SUMDES and SUMHITS and the green lines you see in the texture plot all represent virtual particles, not the equivalent real particles? While NTOT and QCONST_N represents the real particles scaled up from the virtual ones? I see the ratio NTOT/SUMDES gets smaller as the simulation runs longer.

I was calculating the probability to desorbing from a source and adsorbing onto another so I was using a formula like…A313/SUMDES as apposed to A313/NTOT. However, if I work with the real quantities then the formula 0.97* AR313* 1E-4* Z313/QCONST_Q should give the same answer. The 0.97 is the sticking factor, since the flux Z313 also includes the non-absorbing hits. However, I get an answer that is off by about 0.05%. I would expect it to be exactly the same if the internal calculations going from the virtual particles to the real ones is done correctly. This makes think there is a mistake somewhere in this process.

Cheers,
Alan

[alan (Alan Van Drie)]

A follow up to this, I played around with a simpler geometry and found the same problem. Only if the sticking factor = 1 did the two give the same answer. So I’m guessing the problem lies in the reflected flux.

SF=1 everywhere

SF=0.01 everywhere

[rkerseva (Roberto Kersevan)]

Hello Alan:
During the simulation there are N_cores particles which are being simulated at any times, and ADDED taken into account by the variables you mention. If you e.g. calculate your transmission probability using Ax/sumdes (x=adsorbing facet number) then you’ll divide a number which is Ncores too big. Clearly Ncores is typically 4-8-16, depending on the machine you use (I’m talking single computers, on clusters could be much bigger I guess), so the ratio SUMDES/(SUMDES+Ncores) gets smaller and smaller very rapidly (unless you run a simulation which generates very few particles, that is).
I’m not sure this answers your comment, I leave it to Marton to elaborate further.
Cheers.

[alan (Alan Van Drie)]

Finally figured out the difference. The sticking factor is just the probability of sticking which is a random event with that average. So using 0.01 in the formula is not the current value based on the H6 & A6. SF=1-(H6-A6)/H6 = A6/H6, whose long term average should converge to 0.01. Using this expression for the sticking value the two formulas match…so in the end the error was in my assumptions.

Cheers,
Alan

[alan (Alan Van Drie)]

I’m not sure either lol!

[maarton (Marton Ady)]

There are many questions asked, let’s start from the beginning:

SUMABS

all absorbed Monte Carlo particles. Should increase as simulation runs. It is the sum of the indivdual numbers in the ABS column in the facet list.

While SUMABS is usually equal to the sum of absorptions on facets, there are 3 special cases when not:

• Time dependent mode with “calculate const. flow” OFF: in this case particles are eliminated mid-flight when their flight time is over the last moment of interest
• Radioactive decay enabled and the particle decayed before sticking
• Collisions with background gas (coming soon!) caused the particle to thermalize and reach Brownian motion

SUMDES

Same, for desorption, also displayed under the Begin/Pause button:

Roberto is right that SUMDES=SUMABS + N_core, or in plain simple English, particles that already absorbed on a facet plus particles currently traced:

See that my laptop has 8 cores.

SUMHIT

All Monte Carlo hits. Usually the sum of hits on facets, except when background gas collisions are ON, in which gas volume scattering is also a hit.

NTOT

Physical molecules until “latest moment”. The latest moment is the last recorded time plus half the time windows. By default it’s 5E-11s (last moment = 0s and time window is 1E-10s)

It is also displayed in Global settings:

As you see, this doesn’t make much sense in steady-state simulations.

Instead, you can use:

QCONST and QCONST_N

Also displayed in Global Settings:

Which is the gas entering the system every second, in either mbar.l/s or molecules/sec.

All of the above is true if the “low flux mode”, a rarely used physics mode that allows that particles represent less and less physical molecules, is turned OFF. I assume it is always the case for Alan.

[maarton (Marton Ady)]

As for the discrepancy: yes, the 0.01 means that on average 1% of particles will get stuck. In reality, there is some discrepancy:

The lower the sticking, the lower this discrepancy becomes, because the “positive events” (sticking) get more rare, and Monte Carlo statistics lower.

In your extreme case of sticking=1, there is a 100% probability of sticking so even a single event provides perfect outcome:

(note that particles can’t bounce so nothing comes back to facet 1, the source).

In case of a low sticking factor 1E-5, the error is large, around 0.67%:

[alan (Alan Van Drie)]

Thanks for the all the gory details…it is helpful. I’m glad you are doing collisions, it will help with some of my work with transitional flow. Collisional-less molflow will over estimate transitional flow by around 25%, at least in my models, which isn’t too bad to begin with. I came up with a 7 layer stack of different types of facets that can simulate collisions with a probability that can be set based on the density of neighboring facets…3 layers if there is azimuthal symmetry. When I had time I was going to simulate transitional flow down a pipe using the CLI and making successive simulations runs until it converged to steady state…but since you building collisions into the software I can save my time.

Thanks,
Alan