Search-progress.pdf

The Search-progress.pdf file is meant to give an overview of the current status of the search, and of the convergence, while the search is running. It contains several different plots, listed in the following.

R-factor / Generations

The total number of generations is a reasonably good measure of the elapsed time, so plotting the \(R\) factors over generations gives an impression on how much progress is being made. The best \(R\) factor (i.e., corresponding to the current best-fit structure) is drawn as a black line. The blue line tracks the average \(R\) factor over all members of the population; the grey area plots the range between the current best and current worst structure. Note that both the range and the average may sometimes increase, for example if the population is re-initialized (partly) at random, or if new population members are generated due to SEARCH_CULL.

At regular intervals, the \(R\)-factor scatter for the entire population is plotted as points. The best-fit population is black, the worst-fit population is red, and the rest is shaded relative to their distance from the two. If multiple population members share the same \(R\) factor, they are drawn as a bigger point.

Whenever the search is stopped to reduce the SEARCH_CONVERGENCE gaussian parameter, a vertical line is drawn and the new width of the new value is indicated at the top of the plot.

Generations since change

The x-axis is shared with the \(R\)-factor plot above. On the y-axis is the number of generations that have passed since the last improvement to any structure. So, every time any structure improves, a point is added to the scatter plot, marking how many generations had passed before this improvement. Color indicates whether the improvement occured in the best structure, in one of the best 10% of structures, or any other structure (reflecting the SEARCH_CONVERGENCE gaussian criteria). If the delta values become very large, then this means that few improvements are being found, so the search is either converged or the current settings are badly suited for finding improvements. These situations may automatically be handled by the SEARCH_CONVERGENCE settings, i.e., by decreasing the SEARCH_CONVERGENCE gaussian parameter whenever the generation delta becomes large.

Parameter scatter plots

For each parameter under variation, a scatter plot is drawn for the most recent state of the entire population. The y-axis of these scatter plots corresponds to the displacements range of the given parameter. Small labels at the top and bottom indicate the limits of the range (as displacements relative to the positions of the reference calculation). Color coding of the points is the same as for the right-most scatter in the \(R\)-factor/Generation plot, i.e., the current best configuration is drawn as black, the worst as red, and the rest are shaded relative to their distance in R. In addition, the current best-fit population is indicated by black arrows in order to identify it unambiguously.

Parameter scatters are drawn as half-transparent if a parameter is coupled to another parameter. See the searchpars.info file if you are unsure how exactly parameters are coupled.

Point size represents the number of population members which have the same value for the parameter. If more than one structure has the same parameter value, the color (i.e., \(R\) factor) is indicated for the best one; for example, if the best and the worst structure both have a given parameter at the center of the displacement range, then the point at the center of the displacement range is nevertheless drawn as black.

Example

Below are examples of a Search-progress.pdf file that was generated during the example calculation discussed here.

Fig. 26 Example of the first page of a Search-progress.pdf file. The upper plot shows the \(R\) factor during the search, the lower plot shows the generations since the last change.