AlphaML: machine learning of molecular polarizabilities

About AlphaML

AlphaML was developed by the Ceriotti group at EPFL and the DiStasio group at Cornell. AlphaML uses a machine-learning framework based on comparisons of local atomic environments to predict the polarizabilities of molecules. This model was trained on the 7211 compounds in the QM7b dataset, containing up to 7 heavy atoms (C,N,O,S,Cl) as well as hydrogen. For these molecules, the polarizability was calculated using B3LYP-DFT, SCAN0-DFT and CCSD. A model trained on the 5400 most diverse members of the QM7b set predicts the CCSD polarizability of the remaining molecules with an error of 0.04 a.u., an improvement of an order of magnitude over the prediction of DFT, and the polarizability of a showcase set of more complex molecules with an error of 0.24 a.u., comparable in accuracy to the DFT prediction.

Disclaimer: The model used in this web-app has been optimized for execution speed rather than accuracy, and so have been found to deviate by about 0.02 a.u. from those of the published manuscript (0.002 a.u. for the CCSD polarizabilities).

Please read and cite the following: David M. Wilkins, Andrea Grisafi, Yang Yang, Ka Un Lao, Robert A. DiStasio Jr. and Michele Ceriotti. Accurate molecular polarizabilities by coupled-cluster theory and machine learning , Proc. Natl. Acad. Sci. 116, 3401 (2019).

Version 1.1 Beta; 02.03.2019

This is the first release of AlphaML, with basic functionality. The polarizabilities of molecules containing H,C,N,O,S,Cl can be predicted. It has only been tested for input structures in which all atomic coordinates are fully relaxed with DFT. Version 1.1 Beta introduces radial scaling to improve the accuracy of predictions.

What AlphaML does

AlphaML takes as an input the atomic coordinates of a molecule, and

predicts the dipole polarizability of this molecule, as well as a decomposition of this polarizability into atom-centred environments.
provides a JSmol interactive output and copy-paste content in the extended xyz format.

AlphaML has so far been tested for DFT-optimized molecular configurations of molecules containing H, C, N, O, S and Cl atoms. See Wilkins et al..

Version history

02.03.2019: Version 1.1 Beta

Inclusion of radial scaling to improve the accuracy of predictions.

13.08.2018: Version 1.0 Beta

Implementation of AlphaML framework as described in Wilkins et al..

Upload your molecule

Version 1.1 Beta of AlphaML predicts polarizabilities of structures containing only H,C,N,O,S,Cl.

Upload a molecule:

Select the file format:

Select the AlphaML model:

By continuing, you agree with the terms of use of this service.

Please be patient.
Depending on the workload of the server the calculation might only start in a few minutes.

Otherwise, pick an example

How to cite

If you use this tool, please cite the following work:

David M. Wilkins, Andrea Grisafi, Yang Yang, Ka Un Lao,Robert A. DiStasio Jr. and Michele Ceriotti: Accurate molecular polarizabilities by coupled-cluster theory and machine learning Proc. Natl. Acad. Sci. 116, 3401 (2019).
The input parsers use a number of libraries (see name in the dropdown list) from ASE, pymatgen.
The web interface used in this tool is based on that of ShiftML.

Acknowledgment

We are grateful for support from:

Funding by the European Research Council under the European Union's Horizon 2020 research and innovation program
(grant agreement no. 677013-HBMAP)
Partial support from Cornell University through startup funding
Computational resources of the Argonne Leadership Computing Facility at Argonne National Laboratory, which is supported by Office of Science of the US Department of Energy Contract DE-AC02-06CH11357. Additional computing time was provided by the Scientific IT and Application Support Center of EPFL and by the Swiss National Supercomputing Centre (CSCS) under project ID s843.