Further, while sometimes all that is needed is to pick the right function and refine its coefficients, that does not always work, so GSAS-II provides both a manual and a semi-automated way to “bootstrap” the background.
One key step in background fitting is to use the right data limits for the refinement. It is common that the low Q portion of a diffraction pattern has a complex background. If there are no diffraction peaks in that region, one is only making the fitting process more difficult by including that portion of the pattern. Make sure to set the low Q data limit (low angle or long TOF time) so that the lowest-order observed diffraction peaks are included in the computation but do not include data at much lower Q than where the first peak reaches the baseline. While typically not a problem with respect to background fitting, including data at too high a Q value is also not a good idea. Look at the data at high Q and see if there are observable changes in intensity or if the intensities are too small to see relative to the noise level. Fitting noisy data can be helpful, but do not include data at high Q where there is only background to fit. The most important reason to not include this “flat” data at high Q is not for background considerations, but due to the time needed to perform computations. As the range of the pattern increases in Q, the number of reflections increases by \(\approx \rm Q^3\) and the time needed to compute the pattern also increases on that order. This is most noticeable with TOF data, where a small change in the minimum TOF time can add a very large number of reflections to the pattern. I have seen least-squares refinement cycle times drop from minutes to seconds with the choice of sensible diffraction limits.
Limits for the range of data to be used in fitting are stored on the Limits data tree entry and can be changed there, but can also be changed by dragging the green and red vertical dashed lines from the plot for the background tree entry.