Exact conservation laws for neural network integrators of dynamical systems

Speaker: Dr Eike Mueller

Abstract: The solution of time dependent differential equations with neural networks has attracted a lot of attention recently. The central idea is to learn the laws that govern the evolution of the solution from data, which might be polluted with random noise. However, in contrast to other machine learning applications, usually a lot is known about the system at hand. For example, for many dynamical systems physical quantities such as energy or momentum are exactly conserved. Hence, the network has to learn these conservation laws from data and they will only be satisfied approximately due to finite training time and random noise. In this talk we present an alternative approach which uses Noether's Theorem to inherently incorporate conservation laws into the architecture of the neural network. We demonstrate that this leads to better predictions for the motion of a massive relativistic particle in the Schwarzschild metric.

Applications of multireference methods in computational chemistry

Speaker: Dr Elizaveta Suturina

Abstract: Computational chemistry methods are subject to continuous development. Yet, just a few methods tend to dominate the field e.g. density functional theory with one of the most widely used functionals like B3LYP.

In my talk, I will present a less popular ab initio method called complete active space self-consistent field (CASSCF) and a range of applications where CASSCF is the most appropriate and efficient method. I will share our latest findings regarding a series of transition metal complexes and how input from computational chemistry helps us understand their magnetic properties and underlying magneto-structural correlations.

Survival analysis of a wave energy convertor

Speaker: Dr Haoyu Ding

Abstract: The research on wave energy converters (WECs) has attracted great interest, however, there are still many challenges remaining in the development of effective, reliable and economically viable WECs. Because WECs are always exposed to harsh marine environments, survivability under extreme marine environments that cause extreme loads and large responses is one of the challenges that are needed to be addressed. OpenFOAM® has been proved to be an accurate and mature viscous flow model for simulating wave-structure interactions even under violent waves.

This research aims to apply OpenFOAM® with a RANS-based turbulence model to investigate the survivability of a CorPower WEC system as part of a joint EPSRC project. The focused waves generated by New Wave theory and based on the JONSWAP spectrum are applied for the short-term survival analysis. Both translational and rotational motion responses of the WEC buoy are presented and discussed. The motion responses in time histories have a more obviously nonlinear form after the main crest of focused waves passed through the buoy. The viscous flow around the WEC buoy is also investigated in this research. Violent vorticities are observed downstream of the buoy over the following two wave periods after the main crest passed.

Electronic structure of two dimensional materials

Speaker: Professor Daniel Wolverson

Abstract: First-principles DFT calculations are a powerful tool for the calculation of the electronic and lattice properties of materials. They are applied here to the study of the electronic states in new two dimensional materials including metals and semiconductors, as well as strongly correlated electronic systems, where complex behaviour emerges. Our calculations support experimental scanning probe and synchrotron-based investigations of these materials and are vital to interpret the large volumes of experimental data that we obtain. I will give one or two examples of this process, from experiment to calculation, at a level suitable for a non-specialist.

Using in-plane anisotropy to engineer Janus monolayers of rhenium dichalcogenides

Speaker: Dr Marcin Mucha-Kruczynski

Abstract: The new class of Janus two-dimensional (2D) transition metal dichalcogenides with two different interfaces is currently gaining increasing attention due to the possibility to access properties different from the typical 2D materials. We show that in-plane anisotropy of a 2D atomic crystal, like ReS2 or ReSe2, allows the formation of a large number of inequivalent Janus monolayers. We use first-principles calculations to investigate the structural stability of 29 distinct ReX(2−x)Yx (X, Y being S or Se) structures, which can be obtained by selective exchange of exposed chalcogens in a ReX2 monolayer. We also examine the electronic properties and work function of the most stable Janus monolayers and show that a large number of inequivalent structures provides a way to engineer spin-orbit splitting of the electronic bands.

We find that the breaking of inversion symmetry leads to sizable spin splittings and spontaneous dipole moments that are larger than those in other Janus dichalcogenides. Moreover, our calculations suggest that the work function of the Janus monolayers can be tuned by varying the content of the substituting chalcogen. Our work demonstrates that in-plane anisotropy provides additional flexibility in sublayer engineering of 2D atomic crystals.

Anion-polarisation: Induced short-range-order in the heterocationic lithium-ion cathode material Li2FeSO

Speaker: Dr Samuel Coles

Abstract: Short-range order in cation-disordered cathodes can have a significant effect on their electrochemical properties. Here, we report a combined computational and experimental study of short-range order in the cation-disordered antiperovskite Li2FeSO, using density functional theory, Monte-Carlo simulations, and synchrotron X-ray pair-distribution-function experiments. We predict partial short-range cation-ordering characterised by preferential OLi4Fe2 oxygen coordination, with a weak energetic preference for cis-OLi4Fe2 over trans-OLi4Fe2 oxygen coordination geometries.

The preference for cis-OLi4Fe2 oxygen coordination and the presence of other non-OLi4Fe2 oxygen coordination geometries results in long-range disorder, in agreement with previous experimental data. This contrasts with the predictions obtained from a simple ionic “point-charge” model, which instead predicts almost exclusive trans-OLi4Fe2 oxygen coordination and corresponding long-range crystallographic order. The absence of long-range order in Li2FeSO can therefore be attributed to the relative stability of cis-OLi4Fe2, and other non-OLi4Fe2 oxygen-coordination motifs, compared to the trans-OLi4Fe2 coordination that is favoured by point-charge electrostatics. We show that this effect can be attributed to the polarisation of oxygen and sulfur anions in polar coordination environments, which stabilises these coordination geometries relative to non-polar anion coordination.

We expect this effect to be generally applicable in cation-disordered materials where the cations have different formal charges. This result illustrates how predicting or understanding local structure in cation-disordered materials may necessitate going beyond simple point-charge models; in particlar, we highlight the potential role of anion polarisation in directing short-range order and consequently in explaining the presence or absence of long-range order.

Constraining the nuclear symmetry energy parameters with resonant shattering flares

Speaker: Duncan Neill

Abstract: Neutron stars (NSs) consist of matter at densities close to that of atomic nuclei, making them ideal environments for studying properties of neutron rich nuclear matter, such as the nuclear symmetry energy. A major aim of nuclear astrophysics is therefore to identify NS observables which can be used to constrain their internal structure, and consequently the underlying nuclear physics.

In our work we investigate resonant shattering flares (RSFs): short flares of gamma-rays which may be produced when an asteroseismic mode within a NS is excited by the tidal field of the star's binary partner. Using the Markov-chain Monte-Carlo method, we perform Bayesian inference of the nuclear symmetry energy parameters with data from injected RSFs. We find that RSFs can provide relatively strong constraints these parameters, and that the constraints are complementary to those from nuclear experiment.

A molecular dynamics approach to modelling oxygen diffusion in PLA and PLA clay nanocomposites

Speaker: Dr Jasmine Lightfoot

Abstract: Poly(lactic acid) (PLA) is a commercially produced polymer which has been praised for its non-toxicity, biodegradability, and production from biobased sources. To enhance the barrier properties of PLA, for use in packaging applications, inorganic clays are often incorporated into the polymer. As such, the gas permeability of the plastic is reduced, as penetrant molecules are forced to adopt a more tortuous diffusion pathway through the polymer, avoiding dispersed clay nanoparticles. The reduction in oxygen permeability of clay composites is reported between factors of 1 and 5, depending on the nature of the clay, dispersion, and percentage weight loading. In published studies, difficulties in modelling gas diffusion in polymers prevents realistic characterisation of barrier properties in neat or clay composite PLA systems from simulation. Although mathematical models exist to predict the impact of nanoclays on gas permeability in PLA systems, these require extensive experimental characterisation and do not perform well in complex systems or at high clay loadings.

Using a previously developed method, oxygen diffusion in neat PLA was simulated. Amorphous models were generated, with rigorous force field validation ensuring their behaviour reproduced experimental characterisation. Following grand canonical Monte Carlo simulation to saturate relaxed polymer systems with oxygen, the transport coefficient was calculated as 4.03 ± 0.58 × 10−8 cm2.s−1. This demonstrated close agreement with experiment, measured at 1.37 × 10−8 cm2.s−1,1 and far exceeded the accuracy of previously reported attempts to predict gas diffusion coefficients from MD - in the order of 10−5 cm2.s−1.

In addition to neat polymer systems, PLA was modelled in contact with a layer of pyrophyllite, to investigate in detail the mechanisms by which clay surfaces reduce gas diffusion in polymers. A structural analysis of the composite system indicated that whilst the central region of the polymer phase was bulk-like in nature, primary and secondary coordination shells formed at the interface. These were characterised by a significantly higher than average polymer density of 1.8 g.cm−3 and 1.6 g.cm−3 respectively, compared to the bulk density of 1.19 g.cm-1. Further analysis demonstrated that atoms in these densely packed regions exhibited limited mobility relative to those of polymers in the bulk. These observations are coherent with experimental reports of polymer rigid amorphous fractions (RAFs), which surround filler particles in composite systems. When analysing the trajectories of oxygen penetrant particles within this system, molecules were seen to dwell at or in proximity to the interface. This is likely due to a positive interaction of oxygen with the clay surface, and a higher polymer density and reduced polymer dynamics in the RAF phase impeding lateral penetrant diffusion. Calculated diffusion coefficients of oxygen in PLA/ pyrophyllite systems were three times lower than in neat systems (1.33 ± 0.27 × 10−8 cm2/s compared to 4.03 ± 0.58 × 10−8 cm2/s), reflecting the barrier improvement factors reported experimentally in PLA/ clay composites.

The Electronic Structure and Cooperative Reactivity of a Highly Unusual Magnesium Sodium Complex

Speaker: Dr Samuel Neale

Abstract: Recently, our experimental collaborators here at Bath reported the synthesis of a highly unusual Magnesium structure which features flanking Sodium cations to ultimately form a "Mg2Na2" quaternary centre. Reactivity with carbon monoxide (CO) demonstrates a degree of cooperativity between all four metal centres as CO can effectively 'click' together and dimerize at room temperature to afford an inserted "–OC≡CO–" fragment wedged within the tetrametallic core. This contribution will focus on efforts to probe the inherent electronic structure of this {Mg2Na2} structure using density functional theory (DFT) calculations, and subsequent analytical tools such as Atoms In Molecules (AIM), Natural Bonding Orbital (NBO), and UV/Vis via TD-DFT, as performed using the Bath HPC facilities. Computational mechanistic studies focussing on elucidating how the unusual CO dimerization reaction proceeds, will also be presented.