Hi everyone,
I’m working on fitting data from nanosized materials, where accurately reproducing the lineshape is crucial. I’ve noticed that my fittings often misses smaller features. For example, it tends to smooth over subtle bumps in the experimental data and “cut through” them, rather than following the detailed structure of the pattern.
I’d like to adjust the objective function so that it gives more importance to these small deviations in shape, rather than focusing primarily on perfectly matching peak heights. My suspicion is that the optimizer is being driven too much by high-intensity regions, while small deviations elsewhere are underweighted.
I’ve heard that sometimes using the square root of intensities can help balance this. Is there a practical way to implement such an approach? Would defining custom weightings be a good way to achieve this?