{"id":{"repo_id":"uoit","oai_identifier":"oai:ontariotechu.scholaris.ca:10155/967"},"canonical_url":"https://search.dev.ndltd.org/etd/uoit/oai:ontariotechu.scholaris.ca:10155/967","repository":{"repo_id":"uoit","name":"Ontario Institute of Technology","base_url":"https://ontariotechu.scholaris.ca/server/oai/request"},"display":{"title":"Characterizing the potential energy surface of two dimensional and bulk materials using high dimensional neural network potentials","abstract":"Computing material properties at the ab-initio level of detail is computationally prohibitive for large systems or long timescales. As a result, such methods cannot be used to efficiently sample configuration space. Force field methods can efficiently sample configuration space, but rely on large parameter sets that are tuned to specific contexts. In this work we will explore the ænet approach and its application to six systems: 2D silica, bulk silica, graphene, diamond, hexagonal boron nitride, and cubic boron nitride. Here, a general mapping from atomic coordinates to the potential energy surface is obtained using a feed-forward artificial neural network. An approximate Density Functional Theory method, Density Functional Tight Binding (DFTB+), is used to compute quantities required for the reference dataset. It is found that a network made up of linear activation functions in ænet is (almost) equivalent to a one-layer radial basis function network, and is sufficient to learn a reference dataset consisting of structures sampled from a canonical ensemble at various temperatures. We look at how sampling outside of these frequently visited energy states, through data augmentation, significantly increases the complexity of the problem.","abstract_html":"Computing material properties at the ab-initio level of detail is computationally prohibitive for large systems or long timescales. As a result, such methods cannot be used to efficiently sample configuration space. Force field methods can efficiently sample configuration space, but rely on large parameter sets that are tuned to specific contexts. In this work we will explore the ænet approach and its application to six systems: 2D silica, bulk silica, graphene, diamond, hexagonal boron nitride, and cubic boron nitride. Here, a general mapping from atomic coordinates to the potential energy surface is obtained using a feed-forward artificial neural network. An approximate Density Functional Theory method, Density Functional Tight Binding (DFTB+), is used to compute quantities required for the reference dataset. It is found that a network made up of linear activation functions in ænet is (almost) equivalent to a one-layer radial basis function network, and is sufficient to learn a reference dataset consisting of structures sampled from a canonical ensemble at various temperatures. We look at how sampling outside of these frequently visited energy states, through data augmentation, significantly increases the complexity of the problem.","abstract_has_math":false,"creators":["Maharaj, Amber"],"institution":"University of Ontario Institute of Technology","degree_name":"Master of Science (MSc)","degree_level":null,"degree_discipline":"Modelling and Computational Science","degree_department":null,"school":null,"contributors":[],"advisors":["Tamblyn, Isaac"],"committee_chairs":[],"committee_members":[],"year":2018,"date_issued":"2018-08-01","date_published":"2018-08-01","updated_at":"2026-07-24T05:35:43Z","subjects":["Machine learning","Neural networks","Electronic structure","Computational materials science"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10155/967","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Tamblyn, Isaac"]},{"key":"dc:creator","label":"Author","values":["Maharaj, Amber"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2018-09-17T16:26:40Z","2022-03-29T17:25:48Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2018-09-17T16:26:40Z","2022-03-29T17:25:48Z"]},{"key":"dc:date.issued","label":"Date","values":["2018-08-01"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Modelling and Computational Science"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science (MSc)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Ontario Institute of Technology"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine learning","Neural networks","Electronic structure","Computational materials science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10155/967"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Computing material properties at the ab-initio level of detail is computationally prohibitive for large systems or long timescales. As a result, such methods cannot be used to efficiently sample configuration space. Force field methods can efficiently sample configuration space, but rely on large parameter sets that are tuned to specific contexts. In this work we will explore the ænet approach and its application to six systems: 2D silica, bulk silica, graphene, diamond, hexagonal boron nitride, and cubic boron nitride. Here, a general mapping from atomic coordinates to the potential energy surface is obtained using a feed-forward artificial neural network. An approximate Density Functional Theory method, Density Functional Tight Binding (DFTB+), is used to compute quantities required for the reference dataset. It is found that a network made up of linear activation functions in ænet is (almost) equivalent to a one-layer radial basis function network, and is sufficient to learn a reference dataset consisting of structures sampled from a canonical ensemble at various temperatures. We look at how sampling outside of these frequently visited energy states, through data augmentation, significantly increases the complexity of the problem."]},{"key":"dc:title","label":"Title","values":["Characterizing the potential energy surface of two dimensional and bulk materials using high dimensional neural network potentials"]}]}],"canonical_facts":{"dc:contributor.advisor":["Tamblyn, Isaac"],"dc:creator":["Maharaj, Amber"],"dc:date.accessioned":["2018-09-17T16:26:40Z","2022-03-29T17:25:48Z"],"dc:date.available":["2018-09-17T16:26:40Z","2022-03-29T17:25:48Z"],"dc:date.issued":["2018-08-01"],"dc:description.abstract":["Computing material properties at the ab-initio level of detail is computationally prohibitive for large systems or long timescales. As a result, such methods cannot be used to efficiently sample configuration space. Force field methods can efficiently sample configuration space, but rely on large parameter sets that are tuned to specific contexts. In this work we will explore the ænet approach and its application to six systems: 2D silica, bulk silica, graphene, diamond, hexagonal boron nitride, and cubic boron nitride. Here, a general mapping from atomic coordinates to the potential energy surface is obtained using a feed-forward artificial neural network. An approximate Density Functional Theory method, Density Functional Tight Binding (DFTB+), is used to compute quantities required for the reference dataset. It is found that a network made up of linear activation functions in ænet is (almost) equivalent to a one-layer radial basis function network, and is sufficient to learn a reference dataset consisting of structures sampled from a canonical ensemble at various temperatures. We look at how sampling outside of these frequently visited energy states, through data augmentation, significantly increases the complexity of the problem."],"dc:identifier.uri":["https://hdl.handle.net/10155/967"],"dc:language.iso":["en"],"dc:subject":["Machine learning","Neural networks","Electronic structure","Computational materials science"],"dc:title":["Characterizing the potential energy surface of two dimensional and bulk materials using high dimensional neural network potentials"],"dc:type":["Thesis"],"thesis:degree_discipline":["Modelling and Computational Science"],"thesis:degree_name":["Master of Science (MSc)"],"thesis:institution_name":["University of Ontario Institute of Technology"]},"updated_at":"2026-07-24T05:35:43Z"}