clag-agn/notes
2017-07-05 13:35:33 -04:00

76 lines
3.2 KiB
Plaintext

Ok, a quick summary of what we talked about doing for next week:
1) Make the 18 panel figures with PSD and lag for the 4 frequency bin case
2) Make an 18 panel figure that overlays the 4-bin and 7-bin cases.
3) Try 5 bins and 6 bins: how many wavelengths fully coverage for each? Make 18 panel figures for these too, so we can compare everything
4) Write out ascii files with: low frequency, high frequency, PSD and Lag for each wavelength. This will be what I need to try and fit the lags and PSDs…..
Ed
4bin status
─────────────
bin bounds:
0.008
0.02011893
0.05059644
0.12724332
0.32
all converged but 17
1158A is suspect
4368A is suspect
4392A is suspect
7bin status
─────────────
bin bounds:
0.008
0.01444256
0.02607345
0.04707094
0.08497812
0.15341274
0.27695915
0.5
converged:
2600A
3471A -- suspect
4368A
4775A
6175A
6439A -- suspect
7647A
9157A
stalled:
1746A
1928A
5404A
5468A
broken:
1158A
1479A
2246A
3465A
4392A
8560A
Hi Ed,
I could not find any combination that allowed all models to converge. In all the combinations I've tried over the last 2 weeks, I never found 5 or 6 bins to be an advantage over 7 bins. I think it's reasonable for us to expect we can find 7-bin binnings that will work for some subsets with size about 1/3 of all lightcurves. We can tailor the bins to what we think are more important. That said, I was at least able to find one 4-bin binning that gave convergence for 17 of the 18 lightcurves. Some results are suspect because clag didn't report output for the fitting attempts for some parameters at those wavelengths. I could not get 1764A to converge under any binnings.
We'll want to adjust the window spacing for the graphs, but for now I wanted to be sure the low values showed up. Again, these all have fully-developed errors for the lag measurements as reported by Abdu's program. The 7-bin overlayed result is just a best attempt to find a comprehensive setting. The boundaries and a suspect and convergence report are given below this message.
I had to squeeze the frequency domain to get this to work well. I listed the boundaries below. Pushing up to 0.4 or down to 0.005 breaks many of these. I chose the outside boundaries based on two things: 0.008 is the nyquist frequency for something that occurs every ~60 days, right? So we wouldn't expect there to be much consistently observable lag on that time scale, since the smallest lightcurves aren't much longer than 100 days. At the high end, the lightcurves have a sampling rate of about 1 per day, which suggests the frequency scale should include up to 0.5. As I stated, that was a bit high, but bringing it down to 0.32 at that stage in my troubleshooting left me with the (considerable) result we see now.
I will include a tar.gz archive with tables that are used to create this graph. The tables beginning with "cackett_" are tables in the format you requested corresponding to all of the results shown in the plots.