University of Crete (UOC)
Multi-scale modeling of macromolecular systems via physics-based data-driven models
Abstract
dc:descriptionMolecular dynamics (MD) simulations study the physical movement of atoms and molecules. They do so by treating the underlying physical system as an N-body problem. After defining themolecular model, that is, the interaction potentials between atoms of the system, the equations of motion can be numerically integrated to acquire the evolution of the system through time.Molecular dynamics suffers from the huge computational demands due to the complexity of the calculations involved, and is thus limited regarding the size of the simulated systems andthe duration of the runs (simulation experiments). The continuous growth of available computing power, most notably the common availability of clusters of parallel processing units,multiprocessors, and programmable graphics processing units, has enabled the simulation of much larger systems. Largely parallel computing is mandatory in molecular dynamics simu-lations when large physical systems are to be examined, and thankfully it has been available during the past several years and is becoming all the more available.Still one often needs to study the model systems for times much longer, or for systems much larger than the typical spatiotemporal scales that are accessible by atomistic MD simulations. Examples of such systems are usually biological systems or large molecular weight polymer systems, which are characterized by an enormous range of time and length scales. Multiscale MD tackles this problem by removing degrees of freedom from the system, through approximations, that hopefully capture the important physical phenomena. The process of aggregating several atoms in order to represent them by a single point particle is called Coarse Graining (CG) and the aggregate particle is called the CG particle, ”bead” or ”superatom”. The opposite process of reintroducing the constituent atoms from a CG superatom is called ”back-insertion”. Note that this is a mathematically ill-posed problem as several atomistic configurations might correspond to the same CG one. Through back-insertion we are able to revert to studying atom-level phenomena from a system state point reached using coarse graining. Some studies of materials, such as the mechanical studies, must be performed in the macrosopic-continuum scale and being able to extract macroscopic material properties and model parameters from MD simulations is of paramount importance.Overall the spatiotemporal scales of material modeling and simulation cannot be tackled by a single approach. Full detail atomistic simulations are limited to time lengths of the order of microseconds, since the integration time step must be small enough to capture the fastest bond vibrations of the system. At the same time, the length scale of the simulated systems is typically of the order of tenths of nanometers, as the large number of atoms becomes overwhelming for standard available allocations in most supercomputers. CG simulations extend these limits by far. The spatial scale may be increased by an order of magnitude or so, since more than several atoms are aggregated to a superatom. The temporal scale may be increased by more than an order of magnitude, as the heavier CG superatoms vibrate slower than the atomistic ones, which allows a larger integration time step to be used. Also, the CG friction coefficient is far less than the atomistic one, and thus the faster diffusion of the CG superatoms allows one to favorably scale the CG modeled time. At the continuum level, times of the order of milliseconds and lengths of the order of millimeters must be modeled. This is totally out of scale for both CG and all-atom MD simulations. An alternative approach is to calculate material properties from MD simulations and feed the model parameters to established continuum models, by also potentially improving these models by insights obtained from the atomistic perspective.The main goal of this thesis is twofold. First, to develop a set of methodologies that cover the whole spectrum of multiscale modeling of molecular systems. These cover both the coarsegraining and the back-insertion processes, as well as the bridging of the atomistic and the continuum scales. Second, to investigate the accuracy of these methods by applying them toseveral molecular systems, focusing on macromolecular (polymeric) materials.We first test a few well-established methods of coarse graining using small toy molecular systems. We then proceed to coarse graining industrial rubber and studying the model’s trans-ferability against the proportions of the rubber’s constituents. Afterwards, we complement the coarse graining process with a density-dependent potential term, while studying a polyisoprene melt, in order to capture the proper pressure and isothermal compressibility. Moreover, we devise an algorithm for reintroducing the atomistic details from a coarse-grained (rubber) polymer system. We proceed to study the origin of mechanical properties of polymer-nanoparticle composite systems and so illustrate how the atomistic and continuum levels can be bridged. Finally, we experiment with some data-driven machine learning methods of performing the coarse graining process.
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chazirakis, Antonios
- Χαζηράκης, Αντώνιος
Subjects
dc:subject × 16- Προσομοιώσεις μοριακής δυναμικής
- Αδροποίηση μοριακών συστημάτων
- Επανεισαγωγή ατομιστικών λεπτομερειών
- Μοντελοποίηση πολλαπλών κλιμάκων
- Molecular dynamics simulations
- Coarse graining of molecular systems
- Atomistic detail back insertion
- Multi scale modeling
- Μαθηματικά
- Φυσικές Επιστήμες
- Εφαρμοσμένα μαθηματικά
- Μοντελοποίηση και Προσομοίωση
- Mathematics
- Natural Sciences
- Applied Mathematics
- Modeling and Simulation
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62530