Recommendations for XPS softwares

The excellent paper by G. Leclerc and J.J. Pireaux (J. Electron Spectroscopy and Related Phenomena, 71(1995)141-164, The use of least squares for XPS peak parameters estimation. Part 1. Myths and realities)   formulates some expectations/recommendations for data evaluation software in XPS. (The red text is cited from the reference above. The blue text shows how  EWA realizes the features in question. )
  • The software be flexible enough to permit a wide choice of possibilities

  • No pre-defined (i.e. compile-time) model exists. Instead, a free-choice (runtime) model can be built up by the user. A wide range of components of different kind, with component supplements can be selected and several options enriches the possibilities.
  • In XPS, the model specification can be separated into four distinct entities
    • peak shape
      • By far the most widely used peak shapes in XPS are the semi - empirical Gaussian-Lorentzian combinations, including the sum, product or convolution of Gaussian and Lorentzian peaks. The list of the available peak shapes is shown during peak creation. That list includes the mentioned semi-empirical peak shapes, although they are only considered as (not too accurate) representation of the Voigt shape. Other, more accurate representations are preferred.
        In these models each peak is described by four parameters: the binding energy, the peak area, the peak width and the amount of Gaussian (or Lorentzian) contribution. The minimum number of peak parameters is three (for the shapes like Gaussian and Lorentzian), plus the optional asymmetry parameter; for the Voigt-approximations either the G/L mixing or the Gaussian and Lorentzian width are the additional parameters.
        Although it may seem equivalent to use peak height as a parameter instead of peak area, peak area is much more useful for inference testing and should be preferred. EWA allows its users to chose between these representations during peak creation; although for some peak shapes one or other representation is not allowed. Where possible, the area representation is the default.
        It is also important for the user to have the possibility of specifying the amount of Gaussian contribution of the peaks as an estimable parameter.
        For asymmetric peaks, the software should of course include some asymmetric lineshapes, such as Doniac-Sunjic, Kuchiev or others. There is a generic way of asymmetrizing a distribution, but these asymmetric distributions are also available.
        For reasons of versatility, the user should also be able to enter his own empirical or theoretical peak shape:
        a numerical (given in tabular form) peak shape can be used. There is no way to provide a peak shape with a formula.
        it is essential to use all possible means of getting the appropriate peak shape before performing the regression, so as to make the model physically and chemically meaningful.
    • background
      • Many types of background should be available to the user, including the most common: constant, linear, Shirley, Tougaard, etc. EWA implements a polynomial (with degree -1...3, and with variable transformation), Shirley (allowing to include sloping background and energy gap), Tougaard (2 and 3 parameters 'Universal' loss function, 'REELS' loss function)
        Any background with estimable parameters should optimally be included in the regression model. In EWA, this is the default. However, the command 'Component Subtract' allows to remove any of the components before performing regression.
        In all cases at least a constant background should be included in the regression model. All background types include a constant background as an option and it is on by default.
    • weights
      • The inverse of the predicted values should be used as a weight factor in XPS. The uncertainty (variance) of the experimental data is stored in a separate store. When the uncertainty is present in the loaded data file, then that value, otherwise a percentage error or the corresponding Poisson uncertainty is calculated from the measured data. Optionally, either the measured or the predicted value is used used in this calculation. If an operation changes the measured data, the uncertainty changes correspondingly.
    • constraints
      • A parameter is said to be constrained when it is given a fixed value or expressed as a function of other parameters. All parameters can be independently fixed, replaced by some other parameter or linked in an additive/muliplicative way to some other parameter. At the moment, only one other parameter can be used to calculate the value of a constrained marameter.
        A less strict method of constraining is also implemented: all parameters have a lower and an upper limit, which cannot be exceeded during parameter adjusting or performing regression, even temporarily; but within the allowed range the parameters are free.
        A constrained parameter is no longer an estimable parameter and should not be taken into account when computing the number of degrees of freedom. This simple case is implemented when the parameter is fixed. However, when calculating the value from some other parameter, either the free or the constrained value should be fixed.