CSE 8803 / ME 8883 - Materials Informatics Course - Fall 2016

Join Policy
Invite Only
20 Aug 2016

High-dimensional PES metamodel via DFT - Blog 5: Battling against the curse of dimensionality using permutation in material science (Group:WangTran)

CSE8803 / ME 8883 Materials Informatics

Fall 2016 Georgia Tech

Anh Tran and Zhiyu Wang

While the curse of dimensionality has casted on the meta-modeling approach, one may reduce its impact by many other physics-based insights. With respect to our static ab-initio calculations, we have mentioned that permutation in has helped

(i) to reduce the distance between the input in Kriging, our surrogate model

(ii) to create a ghost image, containing both inputs and output, by re-enumerating the Fe atoms.

Our particle searching algorithm has returned some results, promising, but also have much space for improving. We are currently searching for the minimum energy path (MEP) by modelling it using multiple Bezier curves. The Bezier curve endpoints are the local minima of the MEP. As the searching algorithm advances, these points can be split within the algorithm to create a new Bezier curve. Our research attention focuses on how well Kriging can help in term of surrogate model.

Potential approaches include

(i) performing 2-points statistics on each image, with input and output, then performing PCA later on the 2-points statistics results

(ii) voxelize the image first, then performing 2-points statistics.

The issue is currently discussed by Prof. Kalidindi and Joshua Gomberg. As the PACE cluster is up again, after being maintained for 3 days, we are now looking forward running the algorithm again. It may take 5-6 days to receive the results back.

Created on , Last modified on