Bug #2860
Understand discrepancy between unbinned and binned/stacked results for MSH
| Status: | Closed | Start date: | 03/26/2019 | |
|---|---|---|---|---|
| Priority: | Immediate | Due date: | ||
| Assigned To: |  Knödlseder Jürgen | % Done: | 100% | |
| Category: | - | |||
| Target version: | 1.6.0 | |||
| Duration: |
Description
The minor axis of the elliptical Gaussian fitted to MSH is smaller for unbinned than for binned/stacked analysis. This discrepancy seems to be very robust.
Recurrence
No recurrence.
History
#1
Updated by Knödlseder Jürgen about 5 years ago
- Status changed from New to In Progress
- % Done changed from 0 to 10
It looks like if the integration precision in the GCTAResponseIrf::irf_elliptical method is not sufficient. The number of integration iterations is fixed in this method to 5, and increasing the number brings the fit results closer to the values that were simulated. Below a case where only source events were simulated. With iter=6 or 7 the unbinned results are comparable to the stacked analysis result.
| Run | Nobs-Npred | RA | Dec | PA | Minor | Major | Prefactor | Index |
| Simulation | 228.547 | -59.174 | 61.961 | 0.04965 | 0.11648 | 6.93 | 2.45 | |
| Unbinned, iter=5 | 0.050 | 228.553 | -59.177 | 63.001 | 0.049 | 0.120 | 7.185 | 2.435 |
| Unbinned, Iter=6 | 0.002 | 228.553 | -59.177 | 62.658 | 0.056 | 0.121 | 7.186 | 2.434 |
| Unbinned, Iter=7 | 0.000 | 228.553 | -59.177 | 62.653 | 0.055 | 0.121 | 7.186 | 2.435 |
| Stacked, iter=5 | 0.032 | 228.552 | -59.176 | 63.379 | 0.056 | 0.122 | 7.331 | 2.424 |
#2
Updated by Knödlseder Jürgen about 5 years ago
And here now the analysis of the MSH data. For iter=6 or 7 the size of the 2D Gaussian is comparable to the binned and stacked analyses.
| Run | Nobs-Npred | TS | RA | Dec | PA | Minor | Major | Prefactor | Index |
| Unbinned, iter=5 | 0.543 | 877.3 | 228.547 | -59.173 | 62.352 | 0.050 | 0.117 | 6.943 | 2.433 |
| Unbinned, iter=6 | 0.015 | 905.3 | 228.547 | -59.173 | 59.001 | 0.071 | 0.118 | 7.479 | 2.437 |
| Unbinned, iter=7 | 0.026 | 905.1 | 228.544 | -59.173 | 58.286 | 0.076 | 0.119 | 7.600 | 2.438 |
| Binned, iter=5 | 132.025 | 910.0 | 228.547 | -59.173 | 59.810 | 0.082 | 0.123 | 8.044 | 2.414 |
| Stacked, iter=5 | 872.8 | 228.545 | -59.170 | 58.164 | 0.071 | 0.119 | 7.460 | 2.403 |
#3
Updated by Knödlseder Jürgen about 5 years ago
- Status changed from In Progress to Closed
- % Done changed from 10 to 100
And here the results for a simulation where the source model was amplified by a factor of 10 over the nominal flux value. Also here it was necessary to move to iter=6 to get morphological results consistent with the simulations and consistent with the stacked analysis results.
| Run | Nobs-Npred | TS | RA | Dec | PA | Minor | Major | Prefactor | Index |
| Simulation | 228.547 | -59.174 | 61.961 | 0.04965 | 0.11648 | 69.3 | 2.45 | ||
| Unbinned, iter=5 | 0.234 | 32417.3 | 228.548 | -59.175 | 62.038 | 0.044 | 0.117 | 69.4 | 2.473 |
| Unbinned, iter=6 | 0.017 | 32525.3 | 228.548 | -59.175 | 62.070 | 0.051 | 0.118 | 69.9 | 2.474 |
| Unbinned, iter=7 | 0.017 | 32513.1 | 228.548 | -59.174 | 62.061 | 0.050 | 0.118 | 69.9 | 2.473 |
| Stacked, iter=5 | 53.003 | 33884.7 | 228.548 | -59.174 | 61.830 | 0.051 | 0.119 | 70.7 | 2.454 |
So the precision was changed to 6, and the code was merged into devel.