Reconstruction errors are generated by Minuit MIGRAD code. It seems there is an "ErrorDef" which needs to be set. It defaults to 1 which is the appropriate value for a chi-square fit, but it should be set to 0.5 for a likelihood fit. Documentation also suggests using MINOS to get better errors. This was in the code but commented out.
Testing:
- version 303: Same as previous
- version 303A: Set ErrorDef to 0.5
- version 303B: Uncommented call to MINOS
Version 303 vs Version 242 (raw data)
X ErrorVersion 303A vs Version 303 (raw data)
X ErrorVersion 303B vs Version 303A (raw data)
X ErrorChecking how the energy errors relate to the energy of the shower:
Version 242Version 303 looks more reasonable.
Going to keep the ErrorDef change but remove the call to MINOS, as it is a performance hit.