Back to search

University of Crete (UOC)

Computational and machine learning investigiations of nonlinear complex systems

Abstract

dc:description

This thesis investigates the interplay between nonlinearity, topology, and energy redistribution in discrete complex systems. The main focus is the discrete nonlinear Schrödinger (DNLS) equation on small-world networks, complemented by applications to physics-inspired machine learning and targeted electron transfer in chemical reactions. The first part of the thesis examines how long-range connectivity modifies nonlinear lattice dynamics. Small-world networks provide a controlled interpolation between nearest-neighbor lattices and mean-field-like coupling. Within this framework, network topology is shown to affect the spectral density of states, the localization of stationary solutions, and the bifurcation structure generated by nonlinearity. In the negative-temperature sector of the DNLS model, topology also acts as a thermodynamic control parameter. Sparse and fully connected networks sustain long-lived localized excitations (discrete breathers), whereas intermediate small-world connectivity promotes energy redistribution and weakens localization. These physical ideas are then extended to energy-based machine-learning architectures. Boltzmann-type models are modified by introducing structured sparse connectivity, continuous hidden variables, and DNLS-inspired nonlinear terms. The results show that physically motivated structure can improve reconstruction and completion tasks while reducing unnecessary connectivity. This supports the broader idea that concepts from nonlinear dynamics and statistical mechanics can provide useful inductive biases for machine-learning models. Finally, the thesis investigates targeted electron and energy transfer in chemical reactions using a nonlinear dimer model. When ionic and covalent interactions are balanced near conical intersections, the system can support ultrafast coherent transfer without thermal activation, producing non-Arrhenius behavior distinct from conventional Marcus theory. This mechanism can explain exceptional transfer efficiencies observed in biological systems.

Degree

thesis:*
Grantor dc:publisher
University of Crete (UOC)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Almazova, Natalya
  • Αλμάζοβα, Ναταλία

Subjects

dc:subject × 46

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62319

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Almazova, Natalya; Αλμάζοβα, Ναταλία. Computational and machine learning investigiations of nonlinear complex systems. University of Crete (UOC), 2026. http://hdl.handle.net/10442/hedi/62319