Gaussian API

Service for fitting repulsive potentials in density-functional tight-binding with Gaussian process regression.


Gaussian regression (GPR), along with other emerging machine learning techniques, has become more and more popular in computational chemistry, physics, biology and life sciences. In conjunction with the molecular dynamics simulations (MD), these approaches have been shown to be rather useful for prediction of a wide variety of molecular and materials’ properties and functionalities. However, due to the novelty of techniques, the procedures for their application as well as their validation are far from being standardized. 

At the same time, GPR approach can be used to tune the parametrization in approximate computational techniques in a bias-free manner. A typical example of a popular and rather useful approximate technique is the Density functional tight binding method (DFTB). This method aims to approximate the Density functional theory (DFT) approach, providing comparable accuracy at only a fraction of its computational expense. With DFTB, simulations on time- and length-scales which are absolutely unfeasible with DFT have become reality. 

Aside from the above-mentioned advantage, DFTB also allows a direct access to electronic properties, unlike other empirical approximate methods. This is a great advantage of DFTB, compared to other computational approaches with similar computational cost. Still, all these advantages come into play at the expense of the complication that empirical parameters need to be introduced. Due to this drawback, the method transferability (as compared to other more exact quantum chemical approaches) is significantly reduced. The DFTB segment related to its electronic structure part allows for construction of workflows for transferable parametrization. However, the repulsive part of the potential is rather tricky to treat in this context. This service provides methods for fitting the repulsive part of the potential on the basis of GPR approach. The training data that can be used are DFT-DFTB energy or force residues. The methods can be applied to many elements at once, i.e. for computing the repulsive potentials in the case of molecules that contain multiple “organic elements”, such as C, H, O etc. 

Resource organization

Resource providers

ON-BOARDED 
...
University Ss. Cyril and Methodius, Faculty of Computer Science and Engineering

Rudjer Boskovic 16, 1000 Skopje, North Macedonia (MK)

ON-BOARDED 
...
University Ss. Cyril and Methodius, Faculty of Natural Science and Mathematics

Arhimedova bb, 1000 Skopje, North Macedonia (MK)

ON-BOARDED 
...
University Ss. Cyril and Methodius, Faculty of Computer Science and Engineering

Rudjer Boskovic 16, 1000 Skopje, North Macedonia (MK)

 THEMATIC
ON-BOARDED 

...

Website

https://gaussian.chem-api.finki.ukim.mk/

Readiness

TRL8

Domain

Natural Sciences

Subdomain

Chemical Sciences Physical Sciences Biological Sciences Other Natural Sciences Computer & Information Sciences Earth & Related Environmental Sciences

Category

Compute

Subcategory

Job Execution APIs Repository/Gateway

Target users

Research Groups Students Research Projects Researchers Research Organisations Research Networks Research Communities

Access type

Virtual

Access mode

Free

Tags

computational chemistry API gaussian

Contact

Bojana Koteska, Assistant professor
email: bojana.koteska@finki.ukim.mk
phone: 0038923070377

Helpdesk e-mail

bojana.koteska@finki.ukim.mk

Security contact e-mail

cirt@finki.ukim.mk

The NI4OS-Europe project is funded by the European Commission under the Horizon 2020 European research infrastructures grant agreement no. 857645.