stephenbrooks.orgForumMuon1GeneralPhaseRotB going quite fast
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2004-05-22 15:06:24
I know the purple graphs are a bit messed up at present, but if you compare how far PhaseRotB's muon percentage has come in the last 2 days with this graph http://stephenbrooks.org/muon1/graphs/phaserota.png , you'll see it's got to where the first optimisation took just under a week to get to.  This is probably because of the client speed increase over v4.4, though could also be manual seeding with PhaseRotA results, if someone got clever.
Pollock[Romulus2]
2004-05-22 18:31:02
Clever?
I won't admit to that.  Big Grin
Pollock[Romulus2]
2004-05-22 18:35:12
quote:
Originally posted by Pollock[Romulus2]:
_Clever?_
I won't admit to that.  Big Grin


BTW: Nice work on v4.41!  It's much faster and the tweaking possibilities are plentiful.  Smile What more could we ask, except for a more cooperative bending chicane.  That thing has ruined everything that I have tried.  Mad
excaliber[Free-DC.org]
2004-05-22 19:02:02
I've alwasys wished I knew how to tweak stuff.  People have tried explaining, I really just still dont understand.  Alas, it is not my cup of tea, apparently.

Smile
kitsura
2004-05-22 20:11:41
Its too troublesome to seed with so many variables, or do you happen to have a utility, Pollock that I could use?
AySz88
2004-05-22 22:27:57
Can someone link me to a thread where tweaking is explained?
Pollock[Romulus2]
2004-05-22 22:43:22
quote:
Originally posted by kitsura:
Its too troublesome to seed with so many variables, or do you happen to have a utility, Pollock that I could use? 


I wish there were an easy way to do it.  If I had any knowledge of programming, it would not be difficult to write an app to handle the job.  The biggest problem is that the recent versions of the client seems to scatter the parameters in no particular order.  They look completely different than they actually run in the client.  The simplest way to get them in the proper order is to remove your 'results.dat' file for one run.  The first result should have the parameters in the exact order it will run.  After that, it is just as simple as changing the parameters to run a test queue. 
The v4.41 makes it more challenging with two extra types of simulations. 

[SolenoidsTo15cm] are the same as every other version. 

[PhaseRotB] is easy to work with.  You can start by taking your highest result from the [PhaseRotA] version and running it as a test queue in the B version.  Be sure to change the A to B and it will run as a normal queue.

[ChicaneLinacB] is not as complicated as it looks.  If the parameters are in the proper order, it has two sections.  The first half is just a typical [SolenoidsTo15cm] run.  Then a bending chicane and a linear accelerator are added to it.  The easiest way that I have found is to take a good [SolenoidsTo15cm] run and add the rest of the parameters.  It will usually produce a low percentage with a huge number of mpts.  I have not had much luck with it.  The bending chicane in the middle keeps leaking all of my muons.  Mad
kitsura
2004-05-22 23:11:43
quote:
Originally posted by AySz88:
Can someone link me to a thread where tweaking is explained? 

Here you go!
kitsura
2004-05-23 00:21:50
quote:
Originally posted by Pollock[Romulus2]:
quote:
Originally posted by kitsura:
Its too troublesome to seed with so many variables, or do you happen to have a utility, Pollock that I could use? 


I wish there were an easy way to do it.  If I had any knowledge of programming, it would not be difficult to write an app to handle the job.  The biggest problem is that the recent versions of the client seems to scatter the parameters in no particular order.  They look completely different than they actually run in the client.  The simplest way to get them in the proper order is to remove your 'results.dat' file for one run.  The first result should have the parameters in the exact order it will run.  After that, it is just as simple as changing the parameters to run a test queue. 
The v4.41 makes it more challenging with two extra types of simulations. 

[SolenoidsTo15cm] are the same as every other version. 

[PhaseRotB] is easy to work with.  You can start by taking your highest result from the [PhaseRotA] version and running it as a test queue in the B version.  Be sure to change the A to B and it will run as a normal queue.

[ChicaneLinacB] is not as complicated as it looks.  If the parameters are in the proper order, it has two sections.  The first half is just a typical [SolenoidsTo15cm] run.  Then a bending chicane and a linear accelerator are added to it.  The easiest way that I have found is to take a good [SolenoidsTo15cm] run and add the rest of the parameters.  It will usually produce a low percentage with a huge number of mpts.  I have not had much luck with it.  The bending chicane in the middle keeps leaking all of my muons.  Mad

I kinda have the impression that parameter arrangements don't make a difference, correct me if i'm wrong Stephen, and if I could get my hands on a copy of a compiler (apart from Java which is really slow) I should be able to write all the parameters keyed into the program into a simple text file.  But I don't have one and nope I don't have the money to buy one too since I don't primarily do software development, so maybe Stephen you can create such a utility if its not to much trouble.
AySz88
2004-05-23 00:41:44
That's what the bending chicane is supposed to do - it 'filters' out all the bad off-energy particles.
kitsura
2004-05-23 01:36:18
I just re-read the RAL report and went through the bending chicane part very slowly and realised that manually seeding it with the best solenoid result might not be the best way to optimise it.  In fact if I understand it correctly all the soleniods simulations are infact useless when it comes together with the bending chicane:
quote:
The second optimisation presented here varies the solenoid channel in the same way as before but also tracks the particles through the second segment of the design.  The score for each configuration is calculated from the number of particles retained at the end of this second part, meaning that although the chicane itself cannot be modified due to beam optics considerations, the decay channel can be refined to be optimally compatible with it. 
Stephen Brooks
2004-05-23 06:19:14
Just to clear up some questions before I go away:

1. Seeding manual designs is done by writing the 1st line (the 'genome'/design line) yourself and then putting "TEST <LatticeNameHere>" as the second line.  You then put that into queue.txt before running.

2. The order of the parameters does not matter - that's why each one has a name associated with it like 'ls29f'. You can convert out of this format using the Results2CSV utility on the main page, though converting back might require some programming to do it in a short space of time!  Of course just copying the first line of a result and putting it into queue.txt will work to recheck that result (possibly under a different lattice, if they are at all compatible).
Pollock[Romulus2]
2004-05-23 15:50:20
quote:
Originally posted by kitsura:
I just re-read the RAL report and went through the bending chicane part very slowly and realised that manually seeding it with the best solenoid result might not be the best way to optimise it.  In fact if I understand it correctly all the soleniods simulations are infact useless when it comes together with the bending chicane:
quote:
The second optimisation presented here varies the solenoid channel in the same way as before but also tracks the particles through the second segment of the design.  The score for each configuration is calculated from the number of particles retained at the end of this second part, meaning that although the chicane itself cannot be modified due to beam optics considerations, the decay channel can be refined to be optimally compatible with it. 




Using the solenoids result as a front-end for the chicane is not intended to be a perfect solution, just an easy solution.  Most of the results to this point have a score below 50 mpts which indicates that very few (if any) muons are actually surviving long enough to make it through the linear accelerator before the run is dead.  Using the solenoids result as a front-end simply assures that some muons will survive through the second half, due to the large volume of muons (a known quantity) rather than depending on a random (very poor) front-end generated by the client.

NOTE: The same principle works with phaserotB.  The first half of the simulation is simply a standard solenoids-type channel. 
0.182958 (204.1 Mpts) [v4.41]- my result
0.100712 (56.9 Mpts) [v4.41] - by Mr._O_[US-Distributed]

These are both single runs.  The major difference is the point total.  A higher total indicates that the front-end is producing a large number of muons.  That allows a greater chnace to develop the back-end (phaserot) channel because a large number of muons actually get that far.
kitsura
2004-05-23 20:01:06
Yes I've tried using the solenoid result to seed chicanelinac and got a slightly better result but the drawback is that the mpt count is very high which means that for so much muons the channel is not efficient enough to get through the bending chicane.  This means that in the long run a more efficient method is going to produce a better score much faster.
Pollock[Romulus2]
2004-05-24 18:30:35
quote:
Originally posted by kitsura:
Yes I've tried using the solenoid result to seed chicanelinac and got a slightly better result but the drawback is that the mpt count is very high which means that for so much muons the channel is not efficient enough to get through the bending chicane.  This means that in the long run a more efficient method is going to produce a better score much faster. 



I agree with what you are saying, but using the solenoids result was not intended to produce a high percentage result.  The reason that I use it is to be sure that some muons (even a small percentage) actually survive through the chicane.  In my simulations, typically at least 3,000 muons (+2%) survive the chicane.  Using the client-generated solenoid channel, very few(or none) ever make it that far.  That produces a negative result before the simulation enters the accelerator.  The only way a positive result will ever be reached is to modify the accelerator at the end of the simulation so that some muons will survive to the end of it. 
We have all had plenty of practice modifying solenoid channels (the front-end), so that is the least problem at this point.  My goal is to modify the accelerator to produce a positive result, not to produce a slightly higher negative result.

The high mpt count means that the muons are surviving longer into the simulation, but it will need to be much higher before the muons will survive long enough to produce a positive result.
Stephen Brooks
2004-05-26 11:29:32
I've just noticed that the muon transfer didn't increase a lot while the stats were down, but has begun increasing a bit faster now (although I'm only looking at a few points).  Maybe since the stats were down, the good results weren't being downloaded and put into the samplefiles, and perhaps the lack of up-to-date samplefiles affects the speed of optimisation more than I thought?

[edit] Actually, no... it doesn't look like PhaseRotB has increased a lot, though the scale is not really proportional.  0.2% is twice as good as 0.1%, but then you've got -1.6% etc. so it's harder to interpret.  Someone just this hour submitted a -0.xx% chicane.
[TA]z
2004-05-26 12:01:40
Maybe you missed my ChicaneLinacB rock the house Wink
Pollock[Romulus2]
2004-05-26 17:52:38
Nice hit, Z

With your computing power and a few 'extreme' runs, you will likely catch me within a day or two.  Big Grin

Phaserot should rise quickly.  The 0.186 result jumped to 0.23 within a few hours on my one machine before it was switched to chicanes.
Maniacken [US-Distributed]
2004-05-26 19:01:30
Hey Pollock and Z, i am comming for you in ChicaneB.  Not implying anything but was just wondering how you made the large jump from -1.7 to -.04. Any secrects you can let us in on.
[TA]z
2004-05-26 22:36:27
quote:
Originally posted by Maniacken:
Hey Pollock and Z, i am comming for you in ChicaneB.  Not implying anything but was just wondering how you made the large jump from -1.7 to -.04. Any secrects you can let us in on. 


No secrets on this end... Honestly I'd be interested in finding out which one of my clients made the leap and with which trial type... Big Grin
kitsura
2004-05-27 03:50:56
Yes I realised it too at first I thought I was looking at PhaseRotB but then I rubbed my eyes and confirmed its Chicanelinac.  Anyway some of my clients have been simulating wholly Chicane and seem to be stuck on 1.7x for several hours.  Looks like time I got myself a dual Opteron machine.
Pollock[Romulus2]
2004-05-27 06:27:13
0.233301 (2128.2 Mpts) [v4.41] [ChicaneLinacB]

This one still needs some adjusting and should grow at least a little bit higher. 
The hardest part was finding the 'linaccells=' setting.  I am still working on the voltage and field parameters for the accelerator.  It is using a 'Solenoids Only' front section that was found in an old file.  It should be sent in the next day or two simply because it takes a long time to run a queue and I want to get a few more trials finished before sending.
[DPC]TeamNWW - Huub
2004-05-27 07:39:08
wait!.. am i correct in assuming some of you guys run JUST PhaseRot OR ChicaneLinac??  how??  I assume something like removing the other lattice files and disabling auto updating of the lattice files?
Maniacken [US-Distributed]
2004-05-27 10:19:46
Yes you can remove the other lattice files, and then not autoupdate the lattice files.

i personally only run the chicanelinacB latttice file.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25162245 accesses.