These analyses are performed on generated pseudodata using the BKM02 and BKM10 DVCS cross-section formulations at JLab Hall-A and Hall-B kinematic as well as on the corresponding experimental data listed bellow:

There are 2 different GPD models used for generating the pseudodata: a model based on the KM15 model and an out-of-the-hat model (model1).

At twist-2 the pure dvcs cross-section does not depend on the azimuthal angle and therefore, taking the pure dvcs as a constant in the total cross-section,  there will be 4 fit parameters to extract, i.e. ReH, ReE, ReHt and the dvcs constant.

Least Squares extraction

The least squares extraction analysis was performed using the ROOT framework.

  • Least Squares Fit Optimization

Local Fit Tuning Parameters

1- Minimizer type and algorithm

Genetic algorithm was also explored but unless strict limits are set for the fit parameters the extraction does not improve.

The best minimizer type found for both formulations was Minuit2 with Migrad algorithm for the pseudadata generated with KM15 model.

2- Strategy: Used only by Minuit and Minuit2, related to the computation of the Hessian matrix.

BKM02

No significant differences between strategy 0 and 1.

Strategy 0 gives small errors for Minuit2.

Strategy 2 fails to converge.

BKM10

Minuit2 with strategy 0 gives better results.


The strategy parameter configuration will be set to 0 for both formulations for the pseudadata generated with KM15 model.

3- Tolerance: The tolerance controls the criteria to stop the minimization iterations.

BKM02

BKM10

Overall, setting the tolerance at 10 gives better results for both formulations for the pseudadata generated with KM15 model although the errors are unrealistic.

  • Constraining BKM10 with BKM02

BKM02 does not predict the dvcs constant better than preforming the fit with BKM10, therefore the dvcs constant should not be initially estimated with the BKM02 formulation. The CFFs are extracted closer to the true values with BKM02 but they do not improve significantly.

  • Line Fit Method
  • Stepwise regression

ANN extraction

  • No labels