Date | Day | Time | Week no. | Presenter | Affil. | Dept. Contact | Title |
|---|---|---|---|---|---|---|---|
2024-01-26 | Friday | 12pm | 0 | Marcos Prates | Universidade Federal de Minas Gerais | James Sweeney | Spatial Confounding Beyond Generalized Linear Mixed Models: Extension to Shared Components and Spatial Frailty Models |
2024-02-01 | Thursday | 12pm | 1 | Emmanuil H. Georgoulis | Heriot-Watt University | Natalia Kopteva | Hypocoercivity-preserving Galerkin discretisations of kinetic equations |
2024-02-02 | Friday | 12pm | 1 | William Lee | University of Huddersfield | Kevin Moroney | Nonuniform flow in coffee extraction |
2024-02-09 | Friday | 12pm | 2 | Cameron Hall | University of Bristol | Kevin Moroney | Opportunities and challenges in optimising power grid stability |
2024-02-16 | Friday | 12pm | 3 | Srikanth Toppaladoddi | University of Leeds | Anthony Bonfils | Brownian Motion, Polar Oceans, and the Statistical Physics of Climate |
2024-02-23 | Friday | 12pm | 4 | Maeve Upton | University of Limerick | James Sweeney | Bayesian generalised additive models for quantifying sea-level change. |
2024-03-08 | Friday | 12pm | 6 | Ben Taylor | University College Cork | James Sweeney | Modelling and inference for Spatial Processes Under Aggregation and Change of Support |
2024-03-15 | Friday | 12pm | 7 | Michael Fop | University College Dublin | Shirin Moghaddam | Model-based clustering of networks with compositional edges |
2024-03-28 | Thursday | 12pm | 9 | Graham Benham | University College Dublin | Doireann O'Kiely | Wave-driven propulsion |
2024-04-04 | Thursday | 12pm | 10 | Dan Giles | UCL Centre for Artificial Intelligence | James Gleeson | Embedding sub-grid variability into hybrid climate simulations to improve convective modelling |
2024-04-05 | Friday | 12pm | 10 | Kieran Mulchrone | University College Cork | David O'Sullivan | Rate-Induced Tipping of the Compost Bomb: Sizzling Summers, Heteroclinic Canards and Metastable Zombie Fires |
2024-04-12 | Friday | 12pm | 11 | Ann Smith | University of Huddersfield | Kevin Moroney | Predictive Maintenance in the Digital Era |
2024-04-16 | Tuesday | 12pm | 12 | Allan Greenleaf | University of Rochester | Cliff Nolan | Partition Optimization: Multilinear Estimates from Linear Bounds |
2024-04-19 | Friday | 12pm | 12 | Mel Devine | University College Dublin | David O'Sullivan | Using Complementarity Problems for Game Theory Optimisation: an application for electricity market modelling. |
2024-05-09 | Thursday | 12pm | 15 | Natalya Pya Arnqvis | University College Dublin | Kevin Burke | Extended generalized additive modelling with shape constraints |
2024-07-01 | Monday | 12pm | 16 | Graeme Hocking | Murdoch University, Perth, Australia | James Gleeson | Splashes, waves and bores - unsteady, free-surface flows |
2024-08-01 | Thursday | 12pm | 19 | Laura Keane | York University | Doireann O'Kiely | Simulation and analysis of double charge layers in electrolyte models for Lithium-ion batteries |
AY 2023-2024 Semester 2 Seminar Schedule
Timetable
Abstracts
As seminar abstracts become available from the speakers this page will be updated.
Seminar week 0 by Marcos Prates
Date: 2024-01-26 at 12pm
Speaker: Marcos Prates (Universidade Federal de Minas Gerais)
Host: James Sweeney
Title: Spatial Confounding Beyond Generalized Linear Mixed Models: Extension to Shared Components and Spatial Frailty Models
Abstract: Spatial confounding is defined as the confounding between the fixed and spatial random effects in generalized linear mixed models (GLMMs). It gained attention in the past years, as it may generate unexpected results in modeling. We introduce solutions to alleviate the spatial confounding beyond GLMMs for two families of statistical models. In the shared component models, multiple count responses are recorded at each spatial location, which may exhibit similar spatial patterns. Therefore, the spatial effect terms may be shared between the outcomes in addition to specific spatial patterns. Our proposal relies on the use of modified spatial structures for each shared component and specific effects. Spatial frailty models can incorporate spatially structured effects and it is common to observe more than one sample unit per area which means that the support of fixed and spatial effects differs. Thus, we introduce a projection-based approach for reducing the dimension of the data. An R package named “RASCO: An R package to Alleviate Spatial Confounding” is provided. Cases of lung and bronchus cancer in the state of California are investigated under both methodologies and the results prove the efficiency of the proposed methodology.
Seminar week 1 by Emmanuil H. Georgoulis
Date: 2024-02-01 at 12pm
Speaker: Emmanuil H. Georgoulis (Heriot-Watt University)
Host: Natalia Kopteva
Title: Hypocoercivity-preserving Galerkin discretisations of kinetic equations
Abstract: Numerous physical, chemical, biological, and social dynamic processes are characterised by convergence to long-time equilibria. These are often described as PDEs of kinetic type, whereby position'' andvelocity’’ are independent variables; well known examples of such are Kolmogorov and Fokker-Planck equations. These may also arise when modelling multi-agent interacting processes of particles, individuals, etc. In many important cases the diffusion/dissipation required to arrive to such equilibria is explicitly present in some of the spatial directions only, that is there exist evolution PDEs with degenerate diffusion yet converging to equilibrium states as time goes to infinity. This, somewhat counter-intuitive at first, state of affairs suggests that decay to equilibrium is due to finer hidden structure, which allows for the transport terms to also “propagate dissipation” to the spatial directions in which no dissipation appears explicitly in the PDE model. This property has been studied extensively by Villani who coined the term “hypocoercivity” to describe it in his celebrated 2009 AMS Memoir.
In the talk I will present recent results on how to design and analyse hypocoercivity-preserving Galerkin discretisations, in an effort to port the concept of hypocoercivity in the design and analysis of cutting-edge numerical methods. To that end I plan to start by presenting the first provably hypocoercivity-preserving Galerkin (non-conforming) finite element method for the model problem of Kolmogorov equation and discuss some very recent results on a different approach for the same equation. I plan to conclude with some first results of Galerkin methods for the inhomogeneous Fokker-Plack equation on exponentially weighted function spaces. Some of the results are joint work with Zhaonan Dong (INRIA, Paris) and Philip Herbert (Sussex).
Seminar week 1 by William Lee
Date: 2024-02-02 at 12pm
Speaker: William Lee (University of Huddersfield)
Host: Kevin Moroney
Title: Nonuniform flow in coffee extraction
Abstract: Coffee is a drink made by dissolving the soluble parts of ground, roast coffee beans in water. In espresso coffee this is done at high temperatures and pressures. It seems obvious that grinding coffee more finely will lead to more coffee being extracted. However, a recent experiment showed that, beyond a cutoff point grinding coffee more finely results in lower extraction. One possible explanation for this is that fine grinding promotes uneven extraction in the coffee bed. To explore this a low dimensional model in which there are two possible pathways for flow and coffee extraction is derived and analysed. This model shows that, below a critical grind size, there is decreasing extraction with decreasing grind size as is seen experimentally. In the model this is due to a complicated interplay between an initial imbalance in the porosities and permeabilities of the two pathways which is increased by flow and extraction, leading to the complete extraction of all soluble coffee from one pathway.
Seminar week 2 by Cameron Hall
Date: 2024-02-09 at 12pm
Speaker: Cameron Hall (University of Bristol)
Host: Kevin Moroney
Title: Opportunities and challenges in optimising power grid stability
Abstract: Decarbonising electricity generation is essential to any attempt to achieve net zero carbon dioxide emissions. However, the move to variable renewable energy sources such as wind turbines and solar photovoltaics creates new challenges for ensuring the stability of power grids. A major reason for this is that variable renewable energy sources often have very low inherent inertia. Inertia in this context is a measure of the tendency of the grid to maintain its frequency, even when imbalances between power supply and power demand create a push for the frequency to change. Since power grids can only be stable when power frequency is maintained within a very narrow range, the low inertia of variable renewable energy sources tends to destabilise the power grid, leading to possible grid failures.
One key tool for ensuring grid stability is mathematical modelling. Models of power grids can be used to estimate the inherent stability of the system, and models can be used to optimise the controllable features of generators to help ensure grid stability. Recent sophisticated optimisation methods have been proposed for finding optimal generator parameters for power grids. In this talk, I will describe some of these optimisation methods and explore some of their strengths and weaknesses. In particular, I will look at how robust the optimisation methods are when certain model parameters are not known exactly, or when multiple generators are combined in a model and treated as a single generator.
Seminar week 3 by Srikanth Toppaladoddi
Date: 2024-02-16 at 12pm
Speaker: Srikanth Toppaladoddi (University of Leeds)
Host: Anthony Bonfils
Title: Brownian Motion, Polar Oceans, and the Statistical Physics of Climate
Abstract: In this talk, I will show how tools from statistical physics can be used to study the Earth’s climate. The specific problem addressed is the geophysical-scale evolution of Arctic sea ice. Using an analogy with Brownian motion, the original evolution equation for the sea ice thickness distribution function, g(h), by Thorndike et al. (J. Geophys. Res. 80(33), 4501(1975)) is transformed to a Fokker-Planck-like equation. The steady solution for wintertime is g(h) = N(q, H) * h^q * exp(-h/H), where q and H are expressible in terms of moments over the transition probabilities between thickness categories. This solution exhibits the functional form used in observational fits and shows that for h << 1, g(h) is controlled by both thermodynamics and mechanics, whereas for h >> 1 only mechanics controls g(h). Furthermore, seasonality is introduced by using the Eisenman-Wettlaufer (Proc. Natl. Acad. Sci. USA 106, 28 (2009)) and Semtner (J. Phys. Oceanogr. 6, 379 (1976)) models for the thermal growth of sea ice. The time-dependent problem is studied by numerically integrating the Fokker-Planck equation. The results obtained from these numerical integrations and their comparison with submarine and satellite observations of ice thickness will also be discussed.
Seminar week 4 by Maeve Upton
Date: 2024-02-23 at 12pm
Speaker: Maeve Upton (University of Limerick)
Host: James Sweeney
Title: Bayesian generalised additive models for quantifying sea-level change.
Abstract: The 2021 Intergovernmental Panel on Climate Change report highlighted how rates of sea level rise are the fastest in at least the last 3,000 years. As a result, understanding historical sea level trends globally and locally is important to comprehend the dynamics and impacts of sea level change. The influence of different sea level drivers, for example thermal expansion, ocean dynamics and glacial – isostatic adjustment (GIA), has changed throughout time and space. Therefore, a useful statistical model requires both flexibility in time and space and have the capability to examine these separate drivers, whilst taking account of uncertainty.
In this talk, I will discuss the statistical models we developed to examine historic relative sea level changes, employing sea-level proxy and tide gauge data and the noisy input uncertainty method to account for uncertainty. Our approach uses Generalised Additive Models (GAMs) within a Bayesian framework which enables separate modelling of sea level components, smooth calculation of rates and the ability to incorporate external prior information guiding the evolution of sea level change over time and space. Our findings reveal that current sea levels along North America’s Atlantic coast are the highest in at least the past 15 centuries. GAMs demonstrate the different drivers of relative sea level change, indicating that GIA dominated until the 20th century when a sharp rise in sea level change rates occurred.
Seminar week 6 by Ben Taylor
Date: 2024-03-08 at 12pm
Speaker: Ben Taylor (University College Cork)
Host: James Sweeney
Title: Modelling and inference for Spatial Processes Under Aggregation and Change of Support
Abstract: The issues of aggregation and change of support are common in practical applications involving space-time data. They arise when the true process is continuous in space-time, but only data from different aggregation units, e.g. point-locations, or administrative regions, are available. The challenges posed by such data are often ignored, or substantially simplified in practice. In this talk, I will introduce two strands of research concerning the modelling of spatial/spatiotemporal data measured across multiple types of support, one concerning the modelling of spatiotemporal point process data on malaria in Zambia and a second concerning geostatistical modelling of land suitability in Wales.
Seminar week 7 by Michael Fop
Date: 2024-03-15 at 12pm
Speaker: Michael Fop (University College Dublin)
Host: Shirin Moghaddam
Title: Model-based clustering of networks with compositional edges
Abstract: Networks represent complex systems that capture interactions among entities, often resulting in the formation of communities. Additionally, networks frequently depict flows of quantities between nodes. For instance, in the Erasmus programme exchange network, interactions between European countries correspond to volumes of students being transferred among them. When modeling such networks, the exchanged volumes are influenced by the nodes’ capacities to send and receive quantities, which can vary significantly across the network and may be associated with uninteresting aspects of the data. In the Erasmus programme network, for example, the volume of outgoing students from a country is limited by the size of its student population. Similarly, the volume of incoming students to a country is associated with the number of universities, which is also related to its student population. Clustering nodes in such networks by directly modeling these volumes often results in clustering solutions that merely reflect the capacities of sending and receiving nodes, masking interesting patterns associated with the relative strength of the connecting flows. We propose a model-based clustering approach that utilizes the relative strength of connections, leveraging concepts from compositional data analysis. We introduce a novel Dirichlet stochastic block model for clustering nodes in networks with compositional edge weights. The model relies on a mixture of Dirichlet distributions, whose parameters are determined by the cluster allocations of sender and receiver nodes, allowing for different propensities for retaining or transferring quantities over the network between the clusters. Consequently, nodes are clustered based on flows as parts of a whole, rather than raw volumes. Inference is implemented using a variation of the classification expectation-maximization algorithm, enabling efficient computations. The model is tested in a number of synthetic data experiments, showing good performance. Furthermore, we showcase the model on two datasets: the Erasmus programme exchange network among European countries and a bike-sharing network for the city of London.
Seminar week 9 by Graham Benham
Date: 2024-03-28 at 12pm
Speaker: Graham Benham (University College Dublin)
Host: Doireann O’Kiely
Title: Wave-driven propulsion
Abstract: Wave-driven propulsion occurs when a floating body, driven into oscillations at the fluid interface, is propelled by the waves generated by its own motion. Wave-driven propulsion has been observed in the case of the waves generated by a honeybee trapped on the surface of water, in the case of “SurferBot”, a centimeter-scale interfacial robot that was inspired by the stricken honeybee, and at much larger scales, in the case of the waves generated by jumping up and down on a canoe, also known as “gunwale bobbing”.
In this seminar I will present a new theory for wave-driven propulsion based on coupling the equations of motion of a floating raft to a quasi-potential flow model of the fluid. Using this model, expressions are derived for the drift speed and propulsive thrust of the raft which in turn are shown to be consistent with global momentum conservation. The validity of the model is explored by describing the motion of SurferBot, demonstrating close agreement with the experimentally determined drift speed and oscillatory dynamics. The efficiency of wave-driven propulsion is then computed as a function of driving oscillation frequency and the forcing location, revealing optimal values for both of these parameters which await confirmation in experiments.
Seminar week 10 by Dan Giles
Date: 2024-04-04 at 12pm
Speaker: Dan Giles (UCL Centre for Artificial Intelligence)
Host: James Gleeson
Title: Embedding sub-grid variability into hybrid climate simulations to improve convective modelling
Abstract: Atmospheric General Circulation Models (AGCMs) play a vital role in our understanding of climate dynamics and how the climate is changing. However, carrying out climate projections using the latest AGCMs is a computationally expensive task due to long integration timescales and the need to explore the impact of different forcing pathways. As a result, climate simulations are typically carried out using spatial resolutions on the order of 100km. This coarse spatial resolution leads to biases associated with cloud formation, convection, precipitation and interactions between the water cycle and the large-scale dynamics.
This work aims to tackle these biases associated with convective scale processes by embedding a multi-output Gaussian Process (MOGP), trained to predict high resolution variability of temperature and specific humidity fields, within the AGCM. A proof-of-concept study will be presented where a trained MOGP model is coupled in-situ with a simplified AGCM. The temperature and specific humidity profiles of the AGCM model outputs are perturbed at fixed time intervals according to the predicted high resolution informed variability. Modelling improvements in the precipitation, outgoing longwave and shortwave radiation patterns are observed in a 10-year simulation run and the physical justifications for these changes will be explored. This work showcases a promising approach towards improving the overall representation of sub-grid cell processes in coarse resolution atmospheric simulations.
Seminar week 10 by Kieran Mulchrone
Date: 2024-04-05 at 12pm
Speaker: Kieran Mulchrone (University College Cork)
Host: David O’Sullivan
Title: Rate-Induced Tipping of the Compost Bomb: Sizzling Summers, Heteroclinic Canards and Metastable Zombie Fires
Abstract: The Arctic is the fastest warming region on Earth. Understanding how a rapidly changing climate change impacts Arctic systems is therefore an important challenge. This is the basis of the `Compost-Bomb’ instability, a theorized runaway heating of northern latitude peat soils when atmospheric temperature rises faster than some critical rate, first proposed in [Luke & Cox, European Journal of Soil Science (2011), 62.1] and analysed in [Wieczorek et al, Proceedings of the Royal Society A (2011), 467.2129]. The Compost Bomb instability was one of the first examples of what is known as Rate-induced tipping or R-tipping.The key trigger for the compost bomb instability is heat produced by microbial respiration. Here, the original soil carbon and temperature model of Luke & Cox is augmented with a non-monotone microbial respiration function, for a more realistic representation of the process. This gives rise to a meta-stable state, reproducing the results of [Khvorostyanov et al, Tellus (2008), 60B] where a complex PDE model is used. Two non-autonomous climate forcings are examined: (i) a rise in mean air temperature over decades (ii) a short-lived extreme weather event, with the rate-induced compost bomb observed in each. Using techniques of compactification, singular perturbation theory and desingularisation, we reduce the R-tipping problem to one of heteroclinic orbits, uncovering the tipping mechanism for each climate change scenario.
Seminar week 11 by Ann Smith
Date: 2024-04-12 at 12pm
Speaker: Ann Smith (University of Huddersfield)
Host: Kevin Moroney
Title: Predictive Maintenance in the Digital Era
Abstract: In this seminar, “Predictive Maintenance in the Digital Era,” I’ll explore the essentials of modern predictive maintenance. We’ll cover condition monitoring, data acquisition and management, parameter selection strategies, and model potential.
We’ll start by dissecting condition monitoring, focusing on managing sensor data from large-scale engineering systems. I’ll discuss effective parameter selection methods, including both manual inspection and genetic algorithms.
While real-time monitoring is valuable, it’s a retrospective methodology and at best give instantaneous information on process conditions so typically stops at detection and diagnosis, lacking in prognosis. However, I’ll highlight the potential for future advancements in this area.
We’ll also briefly touch on Functional Principal Component Analysis (FPCA) and its possible role in modeling, as well as the potential of the Sparse Identification of Nonlinear Dynamics (SINDy) algorithm for creating digital twins.
If system dynamics can be reliably recovered from data along with insights into the complexities of predictive maintenance this will pave the way for a more reliable future.
Seminar week 12 by Allan Greenleaf
Date: 2024-04-16 at 12pm
Speaker: Allan Greenleaf (University of Rochester)
Host: Cliff Nolan
Title: Partition Optimization: Multilinear Estimates from Linear Bounds
Abstract: Multilinear operators arise in a wide range of settings, from pure to applied maths. Estimates for multilinear operators are generally more difficult to obtain than for linear ones. I will discuss a very general `cheap’ method of leveraging known estimates for linear operators to (possibly) get estimates for multilinear ones, and give some examples, focusing on generalized Radon transforms.
Seminar week 12 by Mel Devine
Date: 2024-04-19 at 12pm
Speaker: Mel Devine (University College Dublin)
Host: David O’Sullivan
Title: Using Complementarity Problems for Game Theory Optimisation: an application for electricity market modelling.
Abstract: Game theory optimisation involves solving the constrained optimisation problem of several competing players in equilibrium. Numerous mathematical approaches can be used to solve such problems. In this talk, the Complementarity Problem approach will be introduced and discussed. Then, an application to an electricity market model will be presented. In this model, we consider what the optimal investment mix in green technologies (wind energy, solar photovoltaic, and battery storage) will be. The players we model include generating firms, different consumer groups, and a battery storage operator. The uncertainty of wind energy and solar photovoltaic brings stochasticity into the model. We apply the model to a case study of the Irish electricity system in 2030, which is envisaged to have a significant presence of renewable sources. We consider the optimal investment mix when market power (strategic behaviour) is both present and absent from the market. Previous similar work either neglected investment decisions or market power. We observe that the presence of market power increases electricity prices which leads to increased profits for generating firms and higher consumer costs. It also leads to increased investment in green technologies but reduced carbon emissions.
Seminar week 15 by Natalya Pya Arnqvis
Date: 2024-05-09 at 12pm
Speaker: Natalya Pya Arnqvis (University College Dublin)
Host: Kevin Burke
Title: Extended generalized additive modelling with shape constraints
Abstract: Regression models that incorporate smooth functions of predictor variables to explain the relationships with a response variable have gained widespread usage and proved successful in various applications. By incorporating smooth functions of predictor variables, these models can capture complex relationships between the response and predictors while still allowing for interpretation of the results. In situations where the relationships between a response variable and predictors are explored, it is not uncommon to assume that these relationships adhere to certain shape constraints. Examples of such constraints include monotonicity and convexity. Shape-constrained additive models (SCAM) offer a general framework for fitting exponential family generalized additive models with shape restrictions on smooths. The main objective of this talk is to provide extensions of the existing framework for SCAM with a mixture of unconstrained terms and various shape-restricted terms to accommodate smooth interaction of covariates, varying coefficient terms, linear functionals with or without shape constraints as model components, and data with short-term temporal or spatial autocorrelation. The practical usage of the suggested extensions will be illustrated in several examples.
Seminar week 16 by Graeme Hocking
Date: 2024-07-01 at 12pm
Speaker: Graeme Hocking (Murdoch University, Perth, Australia)
Host: James Gleeson
Title: Splashes, waves and bores - unsteady, free-surface flows
Abstract: Superficially it should be relatively simple to simulate the movement of an air-water interface, but the development of curvature singularities places strict limits how far one can go. However, a “fundamental singularities” method does not seem to have this problem and simulations can run until the surface breaks up in some way due to splashing or wave breaking. The availability of such simulations allows us to consider the behaviour of a number of flows. Surface evolution due to flow from sources and into sinks will produce a number of interesting surface effects as described in the title.
Seminar week 19 by Laura Keane
Date: 2024-08-01 at 12pm
Speaker: Laura Keane (York University)
Host: Doireann O’Kiely
Title: Simulation and analysis of double charge layers in electrolyte models for Lithium-ion batteries
Abstract: Rechargeable batteries such as Lithium-ion batteries (LIBs) are becoming widespread in our society as a means of powering devices. Mathematical modelling can be a valuable tool for gaining insight into observed behaviours and in aiding battery testing as we increase our dependence on these LIBs. Double charge layers are narrow boundary regions between the two main components of a battery: the electrode and the electrolyte. The behaviour in these regions can differ from that observed in the bulk due to the reactions occurring and the transfer of ions at the interfaces. Historically, there has been considerably more investigation and modelling regarding liquid electrolytes and their double charge layers in comparison with the less established solid electrolyte. Solid state electrolytes are becomingly increasingly attractive due to their improved safety, lower self-discharge, and higher power densities over the more commonly used liquid electrolyte. We use mathematical modelling techniques such as numerical simulation and asymptotic analysis to gain a deeper understanding of what is happening in these layers, with a particular emphasis on solid electrolytes and elucidating the differences between liquid and solid electrolyte double charge layers. We consider a model for a solid electrolyte derived under thermodynamics principles in zero charge flux equilibrium and isothermal conditions. We use an auxiliary variable to transform from a finite domain to an infinite domain to avoid numerical artifacts of near singularities and to facilitate robust numerical simulations. We use asymptotic techniques to characterize the true width of the boundary layer of the electrolyte. We find that the asymptotic matching between the different regions is non-standard, and we therefore implement a pseudo matching technique to complete our asymptotic solution. From the asymptotics, we identify both strong and weak space charge layers (SCL). The weak SCL layer is characterised by a length which is equivalent to the Debye length of a standard liquid electrolyte. The strong SCL layer is characterised by a scaled Debye length. We identify these length scales formally from the asymptotic reduction of the model allowing a quantifiable measure of SCL widths.