During a direct search inversion, the number of calculated dispersion curves is so huge that it is impossible to manually control the individual results of each model. That is the reason why an automatic quality control has been developed.
Figure 3.2 shows that modal curves might be located very close to each other at certain frequencies: at high frequency for Love case, or at osculation points for Rayleigh case (e.g. Forbriger (2003a)). At these points, the distance between two modes might be smaller than the default step calculated above. During the search, crossing two modes in one step results in a constant polarity and hence modes are sought at a higher velocity, ignoring two modes. Another kind of mode jumping may occur for models with low velocity zones (LVZ). For those models only, the dispersion curve may have a non-monotonous shape with a least one maximum (figure 3.7. When moving from frequency 
to 
in figure 3.5, the horizontal line may cross several modes only if higher modes have also a non-monotonous shape at the considered frequency.
Two kinds of tests (detailed in the next two paragraphs) are performed during the computations of dispersion curve to detect any mode jumping. In case of error, the computation is always restarted for the current modal curve to the highest frequency of the user range. Meanwhile, the search step and the relative precision are both divided by a factor 10. At this stage, optimization is not of concern and the step is chosen small enough to avoid mode jumping rather than decreasing slowly until finding the maximum needed size. Four restarts are allowed before stating that the dispersion cannot be calculated for the given model. The default precision and step ratio are reset after each acceptance of a modal curve.