Graduate theses
Current Record: 21 of 1665
|
Title |
Development and application of Machine Learning approaches to investigate the configurational space of binary alloys |
Alternative Title |
Ανάπτυξη και εφαρμογή μεθόδων μηχανικής μάθησης για την μελέτη του χώρου διατάξεων (configurational space) δυαδικών κραμάτων |
Author
|
Γκάνας, Μάριος
|
Thesis advisor
|
Λυμπεράκης, Λιβέριος
|
Abstract |
In this thesis, a novel methodology has been developed that allows for the
identification and analysis of configurational patterns in alloys. This meth-
odology explicitly incorporates the symmetry properties of the parent lattice
and is based on unsupervised machine learning approaches. More specifically,
unit cells of various sizes and symmetries are used to describe the configura-
tions. These configurations are represented by vectors with lengths equal to
the number of atoms enclosed by the unit cells.
To search for patterns and dominant configurations within the thousands of
vectors, a self-consistent clustering algorithm has been developed. By applying
this approach, cluster centers are constructed, and the representation vectors
are assigned to these cluster centers. Moreover, a degree of order parameter
is defined, allowing the assignment of the highest symmetry cluster center to
each lattice site.
The aforementioned methodology is applied to pseudobinary InGaN alloys.
As input, large alloy structures consisting of more than 105 atoms, produced
by Monte Carlo calculations, are used. The results confirm the tendency of
In atoms to align as second nearest neighbors in InGaN, leading to √3 ×
√3 translational symmetry. The outcome of the developed methodology, i.e.,
cluster centers, can be directly used in density functional theory calculations
or to produce special quasirandom structures with modified probabilities for
the occupation of the lattice sites.
|
Language |
English |
Issue date |
2024-07-11 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Physics--Graduate theses
|
|
Type of Work--Graduate theses
|
Permanent Link |
https://elocus.lib.uoc.gr//dlib/5/7/b/metadata-dlib-1718778117-140509-29927.tkl
|
Views |
548 |