MCNP ignoring particle histories

[oksuz_]

Hi,

I have been doing some simulation with MCNP. They take a long time. I think there is a way to ignore some particle histories, which reduces the time for simulations. In my model, there is a point neutron source, an object, and a detector. Most of the neutrons do not reach the detector. So, I want to ignore them. Can anybody give me a resource from which I can get some information?

Thank you in advance!

 

[Astronuc]

What version of MCNP? Does one have access to manuals?

Can one find a discussion of how one addresses leakage in a volume?

 

[oksuz_]

It is MCNP 6. I have been referring to the user manual version6.2, however, I could not locate the leakage in volume part in it. Could you elaborate this part a little bit more?

Thank you in advance.

 

[rpp]

In general, the methods you are looking for are called "variance reduction methods".
They are described in the MCNP manuals and after doing a quick google search, I found the LANL report "MCNP Variance Reduction Examples".
https://mcnp.lanl.gov/pdf_files/la-ur-12-25907.pdf

There are a lot of variance reduction methods available, and they can be applied to both energy and space.
To recommend a specific method to use, we need to know more about the problem
* What is the source energy?
* What is the geometry?
* What is the detector reaction?

In addition, this isn't really a variance reduction technique, but if you have a high energy neutron source, and a target with a high energy threshold,
you can also set a cutoff energy to instruct the code to ignore any particles that fall below the cutoff. Refer to the manual for more information.

 

[PSRB191921]

Hi,

I have been doing some simulation with MCNP. They take a long time. I think there is a way to ignore some particle histories, which reduces the time for simulations. In my model, there is a point neutron source, an object, and a detector. Most of the neutrons do not reach the detector. So, I want to ignore them. Can anybody give me a resource from which I can get some information?

Thank you in advance!

Hi,
Perhaps you can try to reduce the dimension of your geometry.
You can also try to put importance to 0, so you will have no particles transport in some cell (you must be sure that the result will not be change).
you can try, if possible a wwg,
If you have your input file we can try to look at it.

 

[oksuz_]

In general, the methods you are looking for are called "variance reduction methods".
They are described in the MCNP manuals and after doing a quick google search, I found the LANL report "MCNP Variance Reduction Examples".
https://mcnp.lanl.gov/pdf_files/la-ur-12-25907.pdf

There are a lot of variance reduction methods available, and they can be applied to both energy and space.
To recommend a specific method to use, we need to know more about the problem
* What is the source energy?
* What is the geometry?
* What is the detector reaction?

In addition, this isn't really a variance reduction technique, but if you have a high energy neutron source, and a target with a high energy threshold,
you can also set a cutoff energy to instruct the code to ignore any particles that fall below the cutoff. Refer to the manual for more information.

I am using 14-MeV neutrons. You can see the simplified set-up below.

Both the object and the detector contain Hydrogen and Carbon. As for the reactions, I am assuming all possible reactions inside the object. However, radiography tally is tallying neutron flux. So, I am not sure if every possible reaction occurs inside the detector.

(The object and the detector are not scaled. Assuming the detector is big enough)

 

[oksuz_]

Hi,
Perhaps you can try to reduce the dimension of your geometry.
You can also try to put importance to 0, so you will have no particles transport in some cell (you must be sure that the result will not be change).
you can try, if possible a wwg,
If you have your input file we can try to look at it.

It is kind of a real-world scenario. So, I do not want to change the dimension of the geometry so much. There are three cells whose importance must be 1. Otherwise, MCNP gives an error.

Please see the input file below.

 

[PSRB191921]

Hi,
the geometry "11 So 450" is very big.Why don't you try a
parallelepiped 420x12x20 something like that
for you fatal error try to copy/paste your file (or put it in ASCII)

 

[oksuz_]

Hi,
the geometry "11 So 450" is very big.Why don't you try a
parallelepiped 420x12x20 something like that
for you fatal error try to copy/paste your file (or put it in ASCII)

I tried rectangular parallelepiped, however, the time that it takes for the simulation to complete was pretty much the same with when a sphere was used. Even it was a little bit longer (I do not know why).

I tried cone beam instead of a point source. The number of neutrons reaching the detector was higher, however, it took a day for 1.5e4 particles. I am not sure if there is a way to reduce computer time while not compromising statistic.

 

[PSRB191921]

and a parallel beam ?

 

[oksuz_]

I did not try it since a big plane source would not be suitable for a real-world scenario (at least mine). The source will eventually be a neutron generator. It is going to provide a cone beam with proper shielding. But I can give it a try.

 

[Tom.G]

More Core(s)! CPU cores that is.

All recent laptops & office computers use multicore processors. By default, MCNP6 will only use 1 of the processor cores (to avoid hogging all of the system resources). If you want to use more than 1 cpu core, just add "tasks n" to the MCNP execution line command, where "n" is the number of cpu-cores you want MCNP6 to use. For example,

bash> mcnp6 i=myinp.txt tasks 4​

would direct MCNP6 to use 4 cpu-cores when following histories. "n" should be chosen considering your specific processor: For my laptop, with an Intel i7 with 2 cpu-cores & 2 hyperthreads per core, I usually use "tasks 4".

The default version of MCNP6 from the installation includes the threading capability. It is not necessary to copy other versions, rename executables, etc., just run the code with the appropriate "tasks n" supplied.
It will also run on computer clusters up to at least 128.

above from: https://laws.lanl.gov/vhosts/mcnp.lanl.gov/mcnp_faq.shtml
(under Parallel Processing at the bottom of the page)

Cheers,
Tom

 

[PSRB191921]

After reflexion a "DXTRAN" could be a good solution,
of course you can use the Tom.g's solution with tasks (I should have thought about it !) If you have a 16 cores you can try with "tasks 13" (no more )

 

[oksuz_]

More Core(s)! CPU cores that is.

All recent laptops & office computers use multicore processors. By default, MCNP6 will only use 1 of the processor cores (to avoid hogging all of the system resources). If you want to use more than 1 cpu core, just add "tasks n" to the MCNP execution line command, where "n" is the number of cpu-cores you want MCNP6 to use. For example,

bash> mcnp6 i=myinp.txt tasks 4​

would direct MCNP6 to use 4 cpu-cores when following histories. "n" should be chosen considering your specific processor: For my laptop, with an Intel i7 with 2 cpu-cores & 2 hyperthreads per core, I usually use "tasks 4".

The default version of MCNP6 from the installation includes the threading capability. It is not necessary to copy other versions, rename executables, etc., just run the code with the appropriate "tasks n" supplied.
It will also run on computer clusters up to at least 128.

above from: https://laws.lanl.gov/vhosts/mcnp.lanl.gov/mcnp_faq.shtml
(under Parallel Processing at the bottom of the page)

Cheers,
Tom

I am already using this feature. Still, it takes quite a long time.

 

[PSRB191921]

and dxtran ?

 

[oksuz_]

and dxtran ?

Not yet. I am working on it. Thanks.