For background on these issues related to B-field and tracking, this is a good read.
My thoughts on what should be done going forward:
-Make a detector with a map that has fixed B-field values, similar to our current geometry, and replicate current results using it. (as Pelle suggests)
-Determine the acceptable miss distance parameter, based upon what we consider reasonable level of error for our experiment. The example in the G4 doc sets this to 10 microns. Currently, we are operating with much larger acceptable values here. By default, there is no maximum miss distance, so in certain circumstances tracking could be very inaccurate, as we saw with the issue Pelle identified.
-Determine the best setting for delta intersection, which determines the accuracy of tracking along the chord to geometric boundaries. I am not sure what this should be exactly, but the doc suggests it should not differ greatly from the miss distance. Likely, these values could be the same to achieve reasonable results.
-Investigate which mag field propagator is most appropriate for our field. This is based upon the smoothness of the field, or lack thereof, etc. The above link has more information. Unfortunately, I do not know of a good comprehensive doc page that covers all of them, so we may be left with a combination of simply trying them out and reading the code documentation (if there is any!).
-Characterize the performance penalty of using a 3D mag field map in the simulation and reconstruction. This is likely to be at least 10x slower in simulation. We should determine exactly how much we lose from introducing the more complex B-field. I am actually quite worried here that if there is a significant amount of MC samples that need to be produced using the field map, this could be a real show stopper unless we have sufficient throughput or can somehow greatly increase the performance. This concern is based on some limited prior experience using field maps for ILC.
-Additionally, determine the added performance effect of using the different interpolators with parameters for miss distance and delta intersection. I have a feeling there is going be a wide range of simulation performance effects depending on those three settings (propagator, miss distance, delta intersection).
The geometric "solution" introduced by Pelle, which forces a new step at the edge of the mag field, should work fine for our basically fixed B-field, or at least Pelle claims it greatly reduces the errors he was seeing. But I do not think this strategy works once we introduce varying B-field values with the map. The tracking accuracy will need to be roughly on the level of the B-field grid or below. Actually, I'm guessing the parameters will need to be well below the grid resolution, but again this is something that needs investigation.
From: Hansson Adrian, Per Ola
Sent: Tuesday, September 24, 2013 10:52 AM
To: Graf, Norman A.
Cc: Uemura, Sho; McCormick, Jeremy I.
Subject: Re: latest HPS detector
one thing that would help is to have someone diff into the 3D field map and the related tools we are gonna have to understand better (or at least I'd like to).
For example, it would be good to run the 3D field map and check speed and then understand what the best algorithms to use are; Jeremy told me that there are many available implementations on the propagators in G4. After that I guess tuning the parameters for that algorithm is probably needed. I'd suggest to make a 3D field map that is uniform field over a certain region (like our ideal field in HPS) so that we can solve for the true trajectory manually and then cross-check with the truth residuals that everything is working great.
I would like to work on this but realistically I have so many other commitments in place that it will be hard to get to it in a reasonable time. If you can work on that it would be great. I guess one can estimate that from just looking at the existing field map and I would be interested in understanding how large effect we expect from the 3D field map. Or put it the other way, how non-uniform does it have to be in order for our existing approximation to work. The answer is not easy as I guess it depends on the exact form of the field but anyway.
Let me know what you think.
On Sep 22, 2013, at 8:27 PM, Graf, Norman A. wrote:
> Hello Pelle,
> Thanks. Let me know if I can be of any assistance.
> From: Hansson Adrian, Per Ola
> Sent: Sunday, September 22, 2013 10:15 AM
> To: Graf, Norman A.
> Cc: Uemura, Sho; McCormick, Jeremy I.
> Subject: Re: latest HPS detector
> Hi Norman,
> I have not had time to check all the details that I wanted to. I will do it during Monday and Tuesday and commit before Tuesday evening if I don't find any issues.
> On Sep 20, 2013, at 4:18 PM, Graf, Norman A. wrote:
>> Hello Pelle,
>> Have you released the version of the latest HPS geometry with the
>> thin vacuum volumes at the edges of the box dipole field?
>> Also, has anyone tried running the detector which has the field map?
Use REPLY-ALL to reply to list
To unsubscribe from the HPS-SOFTWARE list, click the following link: