Work Description

Title: Predicting Crystal Structures using Digital Alchemy Inverse Materials Design and the Random Forest Technique of Machine Learning Open Access Deposited
Attribute Value
  • The data were generated as a part of a model in predicting crystal structures solely from knowledge about the colloidal particles without the need for simulations or experiments. We used the Digital Alchemy inverse materials design approach to find optimal and near-optimal hard, convex, colloidal, polyhedral shapes for 13 target structures. To construct the predictive model we performed Alchemical Monte Carlo (Alch-MC) simulations on the target structures. We placed a minimum of N = 100 particles in a periodic simulation box, with the exact number chosen to be a multiple of the number of particles in the unit cell of one of the 13 target structures. Particle shapes were initialized with as many as 64 vertices randomly generated to create a convex shape. Monte Carlo (MC) sweeps were performed to allow particle translations, rotations, and shape moves via vertex re-location.
  • The data are the 13 target structures used in developing our model for predicting colloidal crystal structures from the geometries of particular shapes. The target structures are: simple cubic (SC), body-centered cubic (BCC), face-centered cubic (FCC), simple chiral cubic (SCC), hexagonal (HEX-1-0.6), diamond (D), graphite (G), honeycomb (H), body-centered tetragonal (BCT-1-1-2.4), high-pressure Lithium (Li), Manganese (beta-Mn), Uranium (beta-U), Tungsten (beta-W). At least nine simulations were run on each of the target structures. All of the data are formatted as .pos files.
Contact information
Funding agency
  • Other Funding Agency
Citations to related material
  • " Yina Geng, Greg van Anders, Sharon C. Glotzer, ""Predicting colloidal crystals from shapes via inverse design and machine learning [pre-print]"""
Resource type
Last modified
  • 10/04/2019
  • 01/28/2018
To Cite this Work:
Geng, Y., Van Anders, G., Glotzer, S. (2018). Predicting Crystal Structures using Digital Alchemy Inverse Materials Design and the Random Forest Technique of Machine Learning [Data set]. University of Michigan - Deep Blue.


Files (Count: 3; Size: 1.58 GB)

#given a pos file name, return the particle vertices array in all frames
import numpy as np

def getShapeVerticesFromPos(filename):
#particle shape vertices
allVerts = []

with open(filename,'r') as f: #With syntax for using the open command deletes it when scope ends
for line in f:
#line with 'poly3d' has shape vertices information
if 'poly3d' in line:
temp = line.replace('"','');
temp = temp.split()
#number of vertices
N_vertices = int(temp[3]);
#one shape vertices
verts = np.zeros(N_vertices*3, dtype=float);
for i in range(N_vertices*3):
verts[i] = float(temp[i+4]);
#append the shape to result

allVerts = np.array(allVerts, float)
return allVerts

#filename = "diamond_seed1_k0.00_phi0.60_N128_V32.pos"
#allVerts = getShapeVerticesFromPos(filename)

Download All Files (To download individual files, select them in the “Files” panel above)

Total work file size of 1.58 GB may be too large to download directly. Consider using Globus (see below).

Files are ready   Download Data from Globus
Best for data sets > 3 GB. Globus is the platform Deep Blue Data uses to make large data sets available.   More about Globus