The process of fitting the background in a diffraction pattern is potentially complex, in that GSAS-II provides four different computational mechanisms for background fitting and these are commonly used in combination. These will be described further, below. GSAS-II provides some unique tools can assist in background fitting. They are not needed for the simple cases, but may greatly simplify the problems associated with background modeling for the problem cases, where background is not a smooth slowly increasing or decreasing function.
The process of fitting a background is conceptually simple. Any scattering that is not Bragg scattering from the sample needs to be defined as part of the model used for fitting. The background model is added to the Bragg scattering to produce a simulation that reproduces the entire observed diffraction pattern. If the background is not well fit, it will degrade the overall fit and may cause parameters in the Bragg fitting to deviate from the correct values as the refinement attempts to minimize the differences between the observed and computed pattern. Most commonly, the parameters that would be affected by a poor background fit are the atomic displacement parameters (\(\rm U_{iso}\)) and/or peak shapes, but on occasion structural parameters can be biased. I should note that I am a firm believer that the final fits the background model should be optimized against the data rather than be fit by what the scientist thinks is correct. The best model predicted by the data is what is needed to minimize the overall fit while also obtaining chemically plausible results. While in the early stages of a fit, it may be best to fix the background to be in the locations that look right to our eye, my preference is that in the final stages, the profile be refined. There is one exception to this, which is when all the \(\rm U_{iso}\) values refine to much lower than expected values. In this case it may be necessary to use a hybrid approach, as will be discussed in §11.5 of this chapter.
Before discussing background fitting, I will define one term that is useful for these discussions, the baseline. I will speak of the baseline as function that describes the background as a function of TOF, \(2\theta \), Q, etc. This is the average intensity well away from any Bragg peaks, but will be well below the data in the regions where peaks are found. Note that where noise levels are high, there may well be data points that are below the baseline, since the noise will scatter the data points around this average. Date may not return to the baseline for quite some distance from peaks, if the broadening has appreciable Lorentzian character (from sample broadening), since the Lorentzian line shape has very long peak tails.
It should be noted that the same background terms are used both for peak fitting (see Chapter 17) and for Rietveld fitting and can be refined in either mode.