In the current metacalibration/ngmix_fit_superbit3.py implementation, we are computing both the full R_gamma matrix and it's inverse for the g{i}_MC = R_gamma^-1 * g{i}_noshear column here. Importantly this does not include the selection component and is unused as far as I can tell.
Just a few lines later, the actual mcal response-weighted shape estimate g_Rinv is computed using the simpler g{i}_Rinv = g{i}_noshear / (R{ii}_gamma + R{ii}_S). If we're already spending the cycles to compute the actual R_inv for the gamma component, why not just do this for the full responsivity matrix and store that instead?
In the current
metacalibration/ngmix_fit_superbit3.pyimplementation, we are computing both the full R_gamma matrix and it's inverse for theg{i}_MC = R_gamma^-1 * g{i}_noshearcolumn here. Importantly this does not include the selection component and is unused as far as I can tell.Just a few lines later, the actual mcal response-weighted shape estimate
g_Rinvis computed using the simplerg{i}_Rinv = g{i}_noshear / (R{ii}_gamma + R{ii}_S). If we're already spending the cycles to compute the actualR_invfor the gamma component, why not just do this for the full responsivity matrix and store that instead?