I have a question regarding energy deposition in a 15-crystal setup. When I simulate energy deposition using, for example, 40 keV photons, I notice that the energy deposited in the later crystals is sometimes higher than in the earlier ones. For instance, here is a sample output:
Ideally, the energy deposition should decrease as we move from the first to the last crystal, but that’s not happening consistently.
I’m generating an energy deposition response matrix from 20 to 200 keV in 10 keV steps and using it to unfold experimental spectra. However, due to these fluctuations, the chi-square function doesn’t perform well.
Is there a known reason for these non-monotonic energy deposition patterns? And is there a recommended method to smooth or correct these fluctuations to improve the response matrix?
Perhaps this is due to statistical fluctuation. There are no errors, only Energy deposition fluctuations. But I would like to avoid it as it is crucial for me. I am taking 1e7 primaries with 5 cycles.
Please find attached the screenshot of the LIS file showing the energy deposition with errors. Since I’m having trouble understanding this, I am also including my inp and flair files for your reference.
From this file, I notice that some of the values you scored have errors of up to 60%. As a general guideline, keep in mind that scored values with errors larger than 50% are completely unreliable. You should aim for errors smaller than 10% ideally and sometimes even smaller for anything serious.
I advise you find a strategy to reduce these errors. This could be run more primaries, use biasing or use a more targeted source.
I tried implementing biasing (see image)
but it didn’t help, could you please see if I applied it correctly. Then I tried to increase primaries to 1e8 which reduces errors significantly compare to 1e7 primaries now the errors are 1e-2 to 32 percent.
But I would like to reduce it more as you mentioned less than 10 percent, what are the other steps should I take?
Regarding the source configuration, I am currently using a point-like, collimated source. If there are any adjustments you would recommend for a more “targeted” source setup, I would appreciate your guidance.
Increasing the primaries by a factor 10 will reduce the error by a factor 1/sqrt(10)=0.3 in the best case which is consistent with what you observe here. To reduce them by a factor 6 you would need 36 times more primaries. See the relevant lecture on Monte Carlo techniques, also in the latest FLUKA course.
As for the source, you may want to consult the relevant page in the manual. In general you just want to make sure that all primaries you simulate actually contribute in a meaningful way to the result you want to score. I do not see an obvious problem in your simulation but simply be aware of that.
I’ve been working on the biasing example from the FLUKA course and encountered a few challenges, particularly with determining the correct importance values. I wasn’t entirely sure if I was applying them correctly.
Other than this
Without using biasing, and to reduce the statistical errors, I increased the number of primaries at lower energies as you had suggested:
40 keV → 1e10 primaries
80 keV → 2e9 primaries
Rest → 1e9 primaries
This significantly improved the results—now the maximum error is around 2–3% (please see the attached screenshots).
Looking at the energy deposition results:
For 180 and 240 keV, the deposition appears quite linear.
At low energies like 40 keV, I observe energy deposition in the first two layers, after which the values drop significantly but are not strictly zero.
For 80 keV, there’s deposition in three layers, and similarly, the subsequent values are small but non-zero.
The issue is that these small values beyond the expected layers interfere with interpolation and introduce inconsistencies or gaps.
My main concern now is understanding and defining the noise level. Ideally, I’d like to consider the very small values after the expected deposition layers as zero. For example:
40 keV → deposition in 1st and 2nd layers → rest should be zero
80 keV → deposition in 1st to 3rd layers → rest should be zero
However, while 80 and 100 keV seem to follow a consistent noise level, 40 keV behaves differently—the values beyond the second layer are higher than I would expect for pure noise. Ideally, the noise threshold should be the same across 40, 80, and 100 keV, but this doesn’t seem to be the case at the moment.
To illustrate what I mean, here is an example of the kind of ideal deposition matrix I expect:
40 keV: 1 2 0 0
80 keV: 1 2 3 0
120keV: 1 2 3 4
Sorry for the long message, but I hope you’ll understand what I mean.
It is difficult to further answer because we are beyond the point of software support. I am not in a good position to answer as it heavily depends on how you intend to use these results so I cannot provide a straightforward answer. Let me just comment the following:
Besides statistical errors, there remains uncertainties on the theory, the materials and geometry. In your case they might have become dominant.
Measurements always have an error.
If photons are produced towards a fiducial volume, there’s always a non-zero average energy deposition in that volume. Defining a “zero” is highly dependent of what you want to achieve.
If you have specific questions about biasing please create a separate post.
