Molekularna dinamika na strojno naučenoj plohi potencijalne energije

master's thesis

• Title (english): Molecular dynamics on machine-learned potential energy surface
• Author: Juraj Ovčar
• Mentor: Ivor Lončarić (mentor)
• Mentor: Ivo Batistić (mentor)
• Committee member: Ivo Batistić (predsjednik povjerenstva)
• Committee member: Ivor Lončarić (član povjerenstva)
• Committee member: Željko Skoko (član povjerenstva)
• Committee member: Davor Horvatić (član povjerenstva)
• Committee member: Damir Pajić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Physics); Zagreb
• Defense date and country: 2019-07-16, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Physics

Abstract

Za modeliranje materijala iz prvih principa danas se predominantno koristi teorija funkcionala gustoće, koja pruža dobar omjer preciznosti i računalne efikasnosti. Međutim, zbog trenutnih računalnih ograničenja, obično je prezahtjevno modelirati dinamiku sustava s desecima atoma. Jedna od obećavajućih metoda za približavanje simulacija realnim sustavima i povećanje skala na kojima je moguće provoditi dinamiku je interpolacija potencijala dobivenog s teorijom funkcionala gustoće pomoću neuronskih mreža. U ovom diplomskom radu smo simulirali sustav molekula ugljičnog monoksida adsorbiranih na površini paladija ozračenoj femtosekundnim laserskim pulsom. Prikazali smo proces strojnog učenja te implementirali molekularnodinamičke simulacije u modelu Langevinove dinamike s vremenski ovisnim temperaturama.

Abstract (english)

Nowadays, density functional theory is the predominant tool used in first-principles material modeling, providing an adequate ratio between precision and computational efficiency. However, because of current computational limitations, it is usually too demanding to simulate dynamics of systems consisting of tens of atoms. A promising method for reducing the gap between simulations and realistic systems and extending the scales across which it is possible to simulate dynamics is interpolation of potentials calculated with density functional theory via neural networks. In this master’s thesis we have simulated a system of carbon monoxide molecules adsorbed on palladium irradiated with a femtosecond laser pulse. We have shown the process of machine-learning and implemented molecular-dynamics simulations in the Langevin dynamics model with time dependent temperatures.

• Keywords: teorija funkcionala gustoće; strojno učenje; neuronske mreže; ploha potencijalne energije; heterogena kataliza; laserski inducirana dinamika; Langevinova dinamika; model dviju temperatura; molekularno-dinamičke simulacije
• Keywords (english): density functional theory; machine learning; neural networks; potential energy surface; heterogeneous catalysis; laser-induced dynamics; Langevin dynamics; two temperature model; molecular dynamics simulations
• Language: croatian
• URN:NBN: urn:nbn:hr:217:000337
• Study programme: Title: Physics; specializations in: Research; Course: Research; Study programme type: university; Study level: integrated undergraduate and graduate; Academic / professional title: magistar/magistra fizike (magistar/magistra fizike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-09-04 11:42:10