Common mistakes
From GOSIA
| Line 1: | Line 1: | ||
| - | |||
==Too many fit parameters (forgetting the overall normalization parameter)== | ==Too many fit parameters (forgetting the overall normalization parameter)== | ||
| + | |||
| + | This problem will not cause an error to be reported by Gosia. | ||
If the system is not overdetermined, then fits will not represent a true minimum. With more than a few fit parameters, the system must be very overdetermined using many experiments or scattering angle ranges, etc. | If the system is not overdetermined, then fits will not represent a true minimum. With more than a few fit parameters, the system must be very overdetermined using many experiments or scattering angle ranges, etc. | ||
| Line 7: | Line 8: | ||
Using Gosia 2 in two-state systems with measured yields for both the projectile and the target is designed to overcome this problem. | Using Gosia 2 in two-state systems with measured yields for both the projectile and the target is designed to overcome this problem. | ||
| + | |||
| + | ==Insufficient integration accuracy== | ||
| + | |||
| + | This problem will not cause an error to be reported by Gosia. | ||
| + | |||
| + | Gosia does not test that the [[integrated_yields | integrated yields]] (the accurate yields or cross sections integrated over the energy loss in the target and the solid angle of the scattered particle) are converged. Usually, the most sensitive set is the angular meshpoints, but this is not always the case. | ||
| + | |||
| + | ===Integration meshpoints=== | ||
| + | |||
| + | Gosia does not test convergence for the user's chosen integration meshpoints. | ||
| + | |||
| + | If the integrated yields are not converged, then the predicted cross sections will be inaccurate, and this will also cause an inaccurate fit via the [[corrected_yields | corrected yields]]. | ||
| + | |||
| + | There are two ways to test for convergence: | ||
| + | |||
| + | # The Rachel interface chooses sufficient meshpoints in energy and scattering angle for convergence, but possibly more than necessary. Generating an input with Rachel, or running all calculations in Rachel 'should' always give converged calculations. Rachel also has a built-in test, option "ti" in the Tools button menu, that will test the convergence for a choice of meshpoints. However, sometimes Gosia does not have sufficient memory for this test for very large level schemes, and it must be performed with a subset of the bands. | ||
| + | |||
| + | # The only certain test that there are sufficient meshpoints for convergence is to look at the integrated yields (OP,INTI) with an increasing number of meshpoints 'for each experiment'. Once the required meshpoint spacing is found, calculation time can be minimized by choosing the fewest required. | ||
| + | |||
| + | Once the meshpoint requirement has been found, it does not need to be tested again for OP,CORR [[corrected_yields | corrected yields]] or any other function. | ||
| + | |||
| + | ===Integration meshpoint subdivisions=== | ||
| + | |||
| + | The meshpoint [[integration_subdivisions | subdivisions]] can be tested to see if increasing the number of subdivisions will allow for fewer required meshpoints. Increasing the number of meshpoints significantly increases the calculation time, but increasing the number of subdivisions does not. (It is probably wisest to pick 50 or 100 subdivisions and then test for the required number of meshpoints for convergence.) | ||
