 |
| Stephen Brooks 2011-07-12 15:24:52 | The two lattices we had (Linac900Ext-6Xc2 for version 4.45 and -10td2 for everyone) seem to have found their maximum, having not changed for a week. Of those, the 6Xc2 running on the newer client exhibited a smoother-looking optimisation curve but apparently got stuck in a local maximum in the end. I've released two new lattices that change the RF phase ranges: they are actually reparameterisations of the same problems, but could optimise differently. 10tc2 is like 10td2 but shifts the phase ranges by 180 degrees so the boundary is somewhere else (this 'c2' is actually the original setup that I only changed with the 'd2' lattices). 6X is like 6Xc2 except the RF phase only goes around one cycle rather than two. Having two cycles allows the phases to move around more easily but produces 2^{number of RF cavities} copies of every maximum, which might confuse the optimiser. So I'm going back to the original formulation on the improved v4.45 client. |
