This repository contains code which can be used to compute global structural descriptors for arbitrary atomic systems.
Atomic position files are interpreted as matrices cataloging all atom-to-atom connections.
Using spectral graph theory, these matrices are then converted to scalars which quantify global connectivity.
The atomic simulation environment (ASE) is used to read position files, evaluate atomic distances, assign neighbors, and construct matrices.
As a result, any file format that is understood by ASE can be used by gloab.
Algebraic manipulations are performed using numpy.
A minimal example of code use is as follows:
from ase.io import read
from gloab.encode import Laplacian
atoms = read('/some/path/to/position/file')
system = Laplacian(atoms = atoms)
gc = system.global_connectivityIn this example, a global connectivity value that corresponds to atoms is assigned to the gc variable.
atoms is a prototypical ASE atoms object.
system is an instance that can be used to generate various connectivity properties which are representative of atoms.
As shown through the assignment of gc, connectivity properties are called as properties of the system instance.
Laplacian can be called with additional flags that change underlying behavior.
The first relevant flag is basis, which can be used to select either neighbor-based matrices or distance-based matrices as the basis for calculations.
These are respectively selected by using basis = adjacency or basis = distances.
The second flag is cutoff, which is the limit at which inter-atomic distances will be considered in a distance matrix
Above cutoff, atoms will be assigned distances of 0.
This flag has no use for a basis of adjacency.
The third flag is normalize, which takes a Boolean argument.
If True, the generated Laplacian matrix is the symmetric normalized Laplacian.
If False, it is instead an unnormalized Laplacian.
The fourth flag is radii, which is used to control atomic radii assignments.
When using adjacency as a basis, gloab labels atom neighbors based on overlapping atomic radii.
These can be a default set of atomic radii (as shipped with ASE), or can be customized.
A slightly customized set of radii are used as the default in this repository, as described in a recent publication from our group cite.
Alternatively, a single universal radius can be used for all elements by supplying a single float to the radii flag when instantiating a new Laplacian.
If radii is None, then the default radii will be used.
Once instantiated, the Laplacian can be probed for the different properties summarized here:
| Property | Name |
|---|---|
| adjacency matrix | adjacency |
| degree matrix | degree |
| laplacian matrix | laplacian |
| distance matrix | distances |
| eigenvalue spectrum | eig |
| eigenfunctions | vec |
| spectral gap | gap |
| algebraic connectivity | algebraic_connectivity |
| Laplacian energy like invariant | lel |
| global connectivity | global_connectivity |
| unscaled connectivity | us_connectivity |