On average, how long does it take to compute elastic properties for a compound?

Materials Project
,
1

The reference for the elastic properties is:

de Jong M, Chen W, Angsten T, Jain A, Notestine R, Gamst A, Sluiter M, Ande CK, van der Zwaag S, Plata JJ, Toher C, Curtarolo S, Ceder G, Persson KA, Asta M. Charting the complete elastic properties of inorganic crystalline compounds. Scientific Data 2: 150009 (2015) | doi:10.1038/sdata.2015.9.

There is also methodology information given in Elastic Constants | Materials Project Documentation.

I had trouble finding any discussion of the runtime required for the elasticity properties.

On a compound’s page (e.g. mp-1025377: Cd(AgI2)2 (Tetragonal, I-42m, 121)), the following information is present:

Calculation Summary
Elasticity
Methodology
Structure Optimization
Run Type
GGA	
Energy Cutoff
700 eV
# of K-points
None	
U Values
--
Pseudopotentials
VASP PAW: Cd Ag I	
Final Energy/Atom
-2.2365 eV
Corrected Energy
-17.1716 eV
How do we arrive at this value?
Detailed input parameters and outputs for all calculations
GGA NSCF Uniform  GGA DFPT Dielectric  GGA Static  
GGA Structure Optimization
GGA Structure Optimization
GGA Structure Optimization
GGA NSCF Line

image
image1009Ă—732 48.1 KB

Clicking into each of the links for “detailed input parameters and outputs for all calculations”, I’m having trouble seeing which one would correspond to calculation of the elastic properties. I might be missing something obvious.

2

The runtime will vary significantly depending on the computer it’s run on. In general, each individual calculation is not that onerous. The only issue with elastic tensors is that it requires a fairly large number of individual calculations since we do a separate calculation for each deformation, and then fit the tensor afterwards.

1 Like
3

Thanks! That’s good to know. Do you know if the full tensor is something that takes on the order of minutes, hours, or days? (i.e. in terms of CPU time). If not, no worries. Thanks for the patience as I’ve bombarded discourse with questions the last week or so.

4

Depends on size of the supercell but ~hours across ~5-10 compute nodes is roughly the order of magnitude I’ve experienced, but there may be ways to speed it up significantly

1 Like
5

Whew! It sounds like roughly 100-1000 CPU hrs per compound. I think I can now appreciate why there’s a voting system and a fraction of compounds that have a computed elastic tensor. The full MP database might take something like 50 million CPU hours… Thank you!