Dispersion interactions in machine learning potentials for large-scale atomistic simulations

Heikki Muhli

Dispersion interactions in machine learning potentials for large-scale atomistic simulations

Julkaisun otsikon käännös: Dispersiovuorovaikutukset koneoppivissa potentiaaleissa suuren skaalan atomistisia simulaatioita varten

Heikki Muhli

Teknillisen fysiikan laitos

Tutkimustuotos: Doctoral Thesis › Collection of Articles

Abstrakti

Dispersiovuorovaikutukset ovat molekyylien välisiä vuorovaikutuksia, joita on kaikissa materiaaleissa. Sen lisäksi, että nämä vuorovaikutukset saavat alkunsa elektronien korrelaatiosta, niiden kumulatiivisuuden ja pitkän kantaman luonteen vuoksi suurin osa tiheysfunktionaaliteoriassa (DFT) käytetyistä vaihto- ja korrelaatiofunktionaaleista ei pysty mallintamaan näitä vuorovaikutuksia tarkasti. Tämän vuoksi laskuihin lisätään usein jonkinlainen dispersiokorjaus, jotta saadaan huomioitua tämä puuttuva vuorovaikutus. Monissa systeemeissä dispersionvuorovaikutusten tarkka mallintaminen vaatii monen kappaleen dispersion (MBD) huomioonottamista, mutta nämä vuorovaikutukset nostavat huomattavasti simulaatioiden vaatimia laskennallisia resursseja verrattuna usein käytettyihin pareittain additiviisiin malleihin. Tämä tekee niistä mahdottomia käyttää suuren skaalan atomistisissa simulaatioissa. Hiljattain koneoppiminen (ML) on tehnyt tuloaan materiaalitieteeseen mahdollistaen simulaatioiden vaatimien resurssien vähentämisen kehittämällä ML-potentiaaleja, jotka on kehitetty DFT-laskujen tuloksia tietokantana käyttäen. Nämä potentiaalit pystyvät interpoloimaan uusia tuloksia systeemien potentiaalienergiapinnoilla ilman uusia DFT-laskuja. Koska nämä potentiaalit käyttävät lokaaleihin atomiympäristöihin perustuvia deskriptoreja tekemään ennustuksia laskuista, jotka on alun perin laskettu DFT:llä, niistä puuttuu samalla tavalla dispersionvuorovaikutusten kontribuutio. Näiden vuorovaikutusten epälokaalius ja ML-deskriptorien lokaalius on merkittävä haste dispersiovuorvaikutusten sisällyttämisessä ML-potentiaaleihin. Tässä väitöskirjassa näytetään kuinka dispersiovuorovaikutusten lisääminen ML-potentiaaleihin on mahdollista käyttämällä atomiympäristöjen lokaalia parametrisaatiota. Ensin pareittain additiivinen dispersiomalli muutetaan ML-potentiaaleille sopivaksi siten, että se tuottaa atomikeskisiä dispersioenergioita lokaaleista parametreista. Menetelmä yleistetään MBD:hen johtamalla atomikeskinen dispersioenergia globaalista MBD-energiasta siten, että lopullisen mallin laskennallinen skaalaus on lineaarinen suhteessa atomien lukumäärään. Koska MBD-mallista aiheutuu väistämättä laskennallisia lisäkustannuksia tehokkaille ML-potentiaaleille, kehitetään lisäksi menetelmä pareittain additiivisen dispersiomallin uudelleen parametrisointiin molekyylidynamiikkasimulaatioiden (MD) aikana käyttäen MBD-mallin antamia periodisia korjauksia. Tällä tavalla saavutetaan menetelmä, joka nopeuttaa simulaatioita huomattavasti ja aiheuttaa hyväksyttävän virheen lopullisiin tuloksiin. Koska menetelmä on lineaarisesti skaalaava ja helposti rinnakkaisohjelmoitava, voidaan sillä ajaa suuren skaalan atomistisia MD-simulaatioita murto-osalla siitä ajasta, joka DFT:llä menisi samoihin laskuihin, samalla tuottaen kvanttimekaanisesti tarkkoja MBD-vuorovaikutukset sisältäviä tuloksia.

Julkaisun otsikon käännös	Dispersiovuorovaikutukset koneoppivissa potentiaaleissa suuren skaalan atomistisia simulaatioita varten
Alkuperäiskieli	Englanti
Pätevyys	Tohtorintutkinto
Myöntävä instituutio	Aalto-yliopisto
Valvoja/neuvonantaja	Ala-Nissilä, Tapio, Vastuuprofessori Caro, Miguel, Ohjaaja
Kustantaja	Aalto University
Painoksen ISBN	978-952-64-2357-9
Sähköinen ISBN	978-952-64-2358-6
Tila	Julkaistu - 2025
OKM-julkaisutyyppi	G5 Artikkeliväitöskirja

Tutkimusalat

dispersiovuorovaikutukset
koneoppiminen
tiheysfunktionaaliteoria
monen kappaleen dispersio
molekyylidynamiikka

Pääsy asiakirjaan

http://urn.fi/URN:ISBN:978-952-64-2358-6

Siteeraa tätä

@phdthesis{e74aba648ce74203bbd37d987685819c,

title = "Dispersion interactions in machine learning potentials for large-scale atomistic simulations",

abstract = "Dispersion interactions are intermolecular interactions that are present in all materials. Due to their cumulative and long-range nature, and the fact that these interactions originate from electron correlation, they are not properly captured by most exchange-correlation functionals used in density-functional theory (DFT). Instead, some kind of dispersion correction is often added to these calculations to account for the missing dispersion interactions. In many systems, accurate modeling of dispersion interactions requires many-body dispersion (MBD) to be taken into account, but these interactions significantly increase the computational resources required for simulations compared to the often used pairwise additive models, making them intractable for large-scale atomistic simulations. With the recent advent of machine learning (ML) in materials science, the problem with the computational cost of the DFT calculations can be mitigated by training ML potentials for various materials using DFT data as the reference database. These potentials are able to interpolate new results on the potential energy surfaces of the systems without further DFT calculations. However, because these potentials are trained on the DFT data based on descriptors of local atomic environments, they also lack proper dispersion interactions. The non-local nature of these interactions and the locality requirement of the ML potentials pose a challenge to the inclusion of these interactions in ML frameworks. In this thesis, we show how dispersion interactions can be included in ML potentials using local parametrization of the atomic environment. We first modify a pairwise-additive dispersion model for the ML framework such that it produces atom-centered contributions to the dispersion energies from local parameters. Next, the method is generalized to MBD by deriving the atom-centered dispersion from the global MBD energy such that the computational scaling of the final model is linear with respect to the number of atoms. Since the MBD implementation inevitably incurs a significant additional computational cost to the computationally efficient ML potentials, we also design a method to reparametrize the pairwise-additive model with periodic corrections from the MBD model on the fly during molecular dynamics simulations. With this method we achieve a significant speed-up for the simulations with an acceptable error in the results. As linear-scaling and readily parallelizable by design, our methodology allows one to run extremely large-scale molecular dynamics in a fraction of the computational time required for the corresponding DFT calculations, while producing quantum-mechanically accurate results that include MBD interactions.",

keywords = "dispersion interactions, machine learning, density-functional theory, many-body, dispersion, molecular dynamics, dispersiovuorovaikutukset, koneoppiminen, tiheysfunktionaaliteoria, monen kappaleen dispersio, molekyylidynamiikka, dispersion interactions, machine learning, density-functional theory, many-body, dispersion, molecular dynamics",

author = "Heikki Muhli",

year = "2025",

language = "English",

isbn = "978-952-64-2357-9",

series = "Aalto University publication series Doctoral Theses",

publisher = "Aalto University",

number = "11",

address = "Finland",

school = "Aalto University",

}

TY - BOOK

T1 - Dispersion interactions in machine learning potentials for large-scale atomistic simulations

AU - Muhli, Heikki

PY - 2025

Y1 - 2025

N2 - Dispersion interactions are intermolecular interactions that are present in all materials. Due to their cumulative and long-range nature, and the fact that these interactions originate from electron correlation, they are not properly captured by most exchange-correlation functionals used in density-functional theory (DFT). Instead, some kind of dispersion correction is often added to these calculations to account for the missing dispersion interactions. In many systems, accurate modeling of dispersion interactions requires many-body dispersion (MBD) to be taken into account, but these interactions significantly increase the computational resources required for simulations compared to the often used pairwise additive models, making them intractable for large-scale atomistic simulations. With the recent advent of machine learning (ML) in materials science, the problem with the computational cost of the DFT calculations can be mitigated by training ML potentials for various materials using DFT data as the reference database. These potentials are able to interpolate new results on the potential energy surfaces of the systems without further DFT calculations. However, because these potentials are trained on the DFT data based on descriptors of local atomic environments, they also lack proper dispersion interactions. The non-local nature of these interactions and the locality requirement of the ML potentials pose a challenge to the inclusion of these interactions in ML frameworks. In this thesis, we show how dispersion interactions can be included in ML potentials using local parametrization of the atomic environment. We first modify a pairwise-additive dispersion model for the ML framework such that it produces atom-centered contributions to the dispersion energies from local parameters. Next, the method is generalized to MBD by deriving the atom-centered dispersion from the global MBD energy such that the computational scaling of the final model is linear with respect to the number of atoms. Since the MBD implementation inevitably incurs a significant additional computational cost to the computationally efficient ML potentials, we also design a method to reparametrize the pairwise-additive model with periodic corrections from the MBD model on the fly during molecular dynamics simulations. With this method we achieve a significant speed-up for the simulations with an acceptable error in the results. As linear-scaling and readily parallelizable by design, our methodology allows one to run extremely large-scale molecular dynamics in a fraction of the computational time required for the corresponding DFT calculations, while producing quantum-mechanically accurate results that include MBD interactions.

AB - Dispersion interactions are intermolecular interactions that are present in all materials. Due to their cumulative and long-range nature, and the fact that these interactions originate from electron correlation, they are not properly captured by most exchange-correlation functionals used in density-functional theory (DFT). Instead, some kind of dispersion correction is often added to these calculations to account for the missing dispersion interactions. In many systems, accurate modeling of dispersion interactions requires many-body dispersion (MBD) to be taken into account, but these interactions significantly increase the computational resources required for simulations compared to the often used pairwise additive models, making them intractable for large-scale atomistic simulations. With the recent advent of machine learning (ML) in materials science, the problem with the computational cost of the DFT calculations can be mitigated by training ML potentials for various materials using DFT data as the reference database. These potentials are able to interpolate new results on the potential energy surfaces of the systems without further DFT calculations. However, because these potentials are trained on the DFT data based on descriptors of local atomic environments, they also lack proper dispersion interactions. The non-local nature of these interactions and the locality requirement of the ML potentials pose a challenge to the inclusion of these interactions in ML frameworks. In this thesis, we show how dispersion interactions can be included in ML potentials using local parametrization of the atomic environment. We first modify a pairwise-additive dispersion model for the ML framework such that it produces atom-centered contributions to the dispersion energies from local parameters. Next, the method is generalized to MBD by deriving the atom-centered dispersion from the global MBD energy such that the computational scaling of the final model is linear with respect to the number of atoms. Since the MBD implementation inevitably incurs a significant additional computational cost to the computationally efficient ML potentials, we also design a method to reparametrize the pairwise-additive model with periodic corrections from the MBD model on the fly during molecular dynamics simulations. With this method we achieve a significant speed-up for the simulations with an acceptable error in the results. As linear-scaling and readily parallelizable by design, our methodology allows one to run extremely large-scale molecular dynamics in a fraction of the computational time required for the corresponding DFT calculations, while producing quantum-mechanically accurate results that include MBD interactions.

KW - dispersion interactions

KW - machine learning

KW - density-functional theory

KW - many-body

KW - dispersion

KW - molecular dynamics

KW - dispersiovuorovaikutukset

KW - koneoppiminen

KW - tiheysfunktionaaliteoria

KW - monen kappaleen dispersio

KW - molekyylidynamiikka

KW - dispersion interactions

KW - machine learning

KW - density-functional theory

KW - many-body

KW - dispersion

KW - molecular dynamics

M3 - Doctoral Thesis

SN - 978-952-64-2357-9

T3 - Aalto University publication series Doctoral Theses

PB - Aalto University

ER -

Dispersion interactions in machine learning potentials for large-scale atomistic simulations

Abstrakti

Tutkimusalat

Pääsy asiakirjaan

Sormenjälki

Siteeraa tätä