Fingerprinting in STM4 — Getting Started

  1. Start STM4. Obviously it has to be installed and properly licensed (see Installation and prerequisites).
  2. Go to Libraries and select STM4.
STM4 library
  1. Go to the very right of the module list. A column with heading Experimental will appear. Double click on it, then double click on the topmost Crystal Fingerprint folder.
Experimental sublibrary
  1. Drag and drop in the Applications area below the module list (see the first image). Normally you instantiate the module called FpEnergyLandscape. Only if you don't have energies for the set of structures, then use FpCompareApp. Something as next image will appear.
Load Structures for FpEnergyLandscape
  1. Now load the structures you want to analyze. Select the module user interface for Read Structure. Normally USPEX creates structures in POSCARS formats with all structures concatenated in the same file. So select the file format, select the file, select Multiframe from file. You see the End counter will changes form 1 to the actual number of structures contained in the input file. If you want, enter the atomic types actually present in your file in the bottom text field (titled: POSCAR atoms names)
  2. Now switch with the Modules drop-down menu and select Read Energies. Here enter the name of the energy file. Its format is very simple, just an energy value per line in the same order as the structures. You have to set the toggle if the energies are per atom and not per structure.
Load Energies for FpEnergyLandscape
  1. The next step seems is needed to overcome a unwanted side effect of how AVS/Express behave (remember dataflow?). After loading the energy file, switch to the module Structure Similarity. Here in red you see one structure has been loaded. Push the Reset button (the loaded structures goes to zero).
  2. Return to the Read Structure module (see above). Push the Reset toggle. Push the Run toggle and look the Count entry tick. If you want, stop the loading with the Run toggle.
  3. When finished loading, return to the Structure Similarity module. Note the number of structure loaded.
Main fingerprinting page
  1. Optionally filter loaded structure by energy. This is done in the Select Structures section.
  2. Next step is to compute fingerprints. Normally the only change you have to do here is to select Per element diffraction as fingerprinting method. You can force the cutoff distance for the approximation to an infinite crystal (the system show the value it will use). You can change the bin size, that is inversely related to the resulting fingerprint space dimensionality. You can change also the Gaussian smoothing width. There is normally no need to change these values.
  3. Push the green button Compute fingerprints. Beware, could be a time consuming step, so plan a visit to the coffee machine. At the end, the done message is show plus the resulting space dimensionality.
  4. Then push the Compute distances green button. Its another time consuming step. Normally use the Cosine distance method.
  5. Last step is to adjust the grouping step. Adjust the tolerance value and select the grouping method Pseudo SSN. At the bottom look at the groups found and the not grouped structures counts.
Main fingerprinting page, bottom part
  1. Now you can start analyzing the structures in fingerprint space. Switch to module Fingerprints Analysis. Note that two windows (Chart and 2D plot) are already active.
Fingerprint analysis page
  1. The first section changes the 2D chart. The most important entries are: Distance map and Distance map ordered. The Grouping quality entry is a diagnostic tool to check if the grouping parameters are correct (as much blue as possible).
  2. The second section governs the chart show.
  3. Next section contains controls for the selected chart. The toggle Display Gaussian fit superimposes a Gaussian fit to various histograms.
  4. Next section, if enabled, show and manipulate a probe to lookup values from the chart.
  5. In the last section you can write the displayed values in a comma separated values formatted file to look at them inside another tool.
  6. Now select module Fingerprint Scatteplot. It tries to project points from the high dimensional fingerprint space down to 2D space preserving as much as possible interpoints distances. Adjust timestep (reduce it if you have many points). Change the Num. retries value if you want (each retry perturb the points to escape local minima). The final result is the best retry, that is the one with the lowers temperature i.e. stress.
Fingerprint scatterplot page
  1. The coloring of the points could be changed to show: temperature (i.e. stress), group (black mean ungrouped entry), step, energy.
  2. The last two entries are visual diagnostics. They show 2D distances versus original distances between points. The scatterplot quality is best when the points fall around the diagonal. Map efficiency binned is the same, but the distances are binned. That is the lighter colored points are the more represented values, a better view of the distance distribution.
  3. For the next step (energy landscape computation) remember to select coloring by energy.
  4. Select module Energy Landscape and toggle Enable. The landscape appears in the Energy Landscape window. Adjust the Scale value to have a reasonably depth surface.
Fingerprint Energy Landscape page
  1. If the surface has steps or missing pieces, play with the Dimension and Search range sliders.
  2. That's all.
  3. An optional step explores and compare structures.
  4. Select the Graphical Select module.
Structures compare page
  1. The structures could be selected by two methods. Graphical and Textual. With the Graphical method, you have to draw a line around the points you want to analyze on the scatterplot windows. The line is draw pushing the right mouse button. The line is automatically closed. The selected structures appears in two parallel lists. Select one in the first column and one in the second one. The selected structures appear in the Visual compare window. Toggling Atom color by structure colors the atoms of the first column structure red and from the second column green.
  2. The Textual selection build the two lists by presenting the structures in various ways (grouped/non grouped/all to all).
  3. Using the Move Structure module, one of the structures could be moved to try to superimpose to the other for example. The size balancing slides changes the atom radius in opposite way for the two structures to help resolve coincident atoms.
  4. Last step is to write out the selected structures. For this go to the Write Structure module.