 |
| Stephen Brooks 2007-11-16 17:12:34 | There's a new lattice out just this evening. I thought I'd write a bit about it instead of just letting it turn up in your client leaving you to wonder what it is. After a diversion into the D lattices (D in this case standing for 'dual-harmonic'), which included a much longer phase rotation section with a doubled-frequency section at the end, I've gone back to trying to develop the current best phase rotation results based on the successful PhaseRotC. If you want to know how many muons is "successful", you can now see on the optimisation dashboard at the bottom of the main page I've translated our traditional % figure into muons per second generated at the end of the section of accelerator we are studying. The standard neutrino factory goal defined in Paris in 1999 is 10^21 muon decays per "year", where "year" means 10^7 seconds partly due to physicists liking powers of ten but more importantly such a complex machine needs a lot of maintenance, so if you are operating 24/7 for just 4 months a year you're feeling quite pleased with yourself. That makes 10^14 μ/sec a good yardstick to judge designs by and the dual-harmonics, whether due to intrinsic inflexibility or the very large number of parameters they generated swamping the optimiser, weren't getting there. PhaseRotEby5 is three things: a verification exercise, an speed exercise and an extension of parameter space. Verification because its space basically contains the PhaseRotC optimum, so you would hope it will eventually match this value (used as the 'baseline' on the stats) and if it misses by a mile I'll have found some aberrant optimiser behaviour to try and fix. Speed because I've parameterised it with a lot fewer parameters, blocking the cells in sections of the accelerator that are usually quite uniform in groups of 5. I personally don't like doing this all the time (and you might notice I've kept S1-S9 near the target separate) but it may be the way to get around the slow stepping behaviour seen in the later PhaseRots. Since each RF cavity contributes rather incrementally to the rotation, blocking them in fives will not lose a lot of the useful options (the rapidly-varying ones cancel out!). Finally, the extension of parameter space is pretty conservative given I'm trying to do the other two things, but I've let the lengths of the decay channel and phase rotation to be most sensible values. The decay channel was fixed at 30 metres previously (except in the DecayRots where it was 5m). This way it can vary all the way from 7m to 95m, getting near the DecayRot range at the low end, though the DecayRots didn't produce all that high yield so I'm expecting an optimum somewhere near the 30m mark, but perhaps not exactly that. |
| Stephen Brooks 2007-11-16 20:37:35 | First person other than me to submit results on this new lattice was [ARS] flyingass, whether he knows it or not. |
| Stephen Brooks 2007-11-16 21:45:33 | Now [TA]amd.borg and PCZ[Free-DC.org] have joined in, the latter beating my score. For those of you who have forgotten, the -1.xxxxxx scores are before a sufficient amount of beam gets to the end, so they are scored on a range of -2 to -1 proportional to what distance down the accelerator the beam gets. |
| Stephen Brooks 2007-11-16 22:14:20 | Er, that stage didn't last long. OK, the -0.xxxxxx results are ones where a minimal amount of beam gets through for yield scoring to be meaningful, but that beam arrives outside the required energy window at the end of the accelerator. The score is rated from -1 to 0 on the basis of how far the median energy is from the acceptible band. |
| RGtx 2007-11-16 23:24:45 | Next update, we go positive! |
| Stephen Brooks 2007-11-17 00:19:55 | Well! It's all going quicker than expected.  You've just generated 3.5765 × 10^12 muons per second. |
| RGtx 2007-11-17 00:21:01 | Er, Stephen, Perhaps you shouldn't have removed the PhaseRotDD lattice file until all results have been returned, Members are now getting the old "Record rejected due to bad checksum" error when returning results. |
| Stephen Brooks 2007-11-17 00:29:25 | I had a few of those from people. They should just say "Yes" in some way and I think the results will actually go through and get sent anyway. If you look at the v4.43d changelog, you'll find the clients running in "invisible" mode shouldn't halt at this error. But I'll look to see if I can do anything else in later client versions to make it go smoother. I'll remember the "overlap" trick if I do another lattice while we're still using v4.43x. |
| Stephen Brooks 2007-11-17 22:22:06 | First over 10^13 is akmazero with 1.0357 × 10^13 muons/sec (0.367415%). We're currently an order of magnitude away from beating the PhaseRotC best result. |
| Stephen Brooks 2007-11-19 17:55:42 | Bah. Now it's gone flat around 0.55% and I think I know why (I just made a program that views the evolution of parameters as an animation and in this case it proved useful). Don't be surprised if there's a replacement for this lattice very soon! |
| Stephen Brooks 2007-11-19 18:46:51 | --[I'll remember the "overlap" trick if I do another lattice while we're still using v4.43x.]-- Actually, there's no point in overlapping because at some point the lattice will be deleted anyway. So I've not overlapped this new one. |
| Stephen Brooks 2007-11-19 22:05:56 | So then, as if to prove me wrong, PhaseRotEby5 jumps over 0.6% (in fact one of my results). I'll put them both on and "race" them for a few days... better to let reality decide which is best than my assumption of which is better. |
| Stephen Brooks 2007-11-24 00:37:02 | Well done to [TA]amd.borg who has submitted the first results over 1%: a 1.033627. |
