MSc Applied Mathematics Project 2021-22

Uncertainty Quantification in Topological Data Analysis

Supervisors: Barbara Bravi andAnthea Monod
Topological data analysis (TDA) is a 21st century branch of data science that leverages ideas from pure
mathematics for feature extraction in data. Persistent homology1,5 is a tool from TDA that adapts the
classical concept of homology from algebraic topology to data settings as point clouds or finite metric
spaces. Persistent homology has been widely applied with great success in many applications including
viral evolution8 , medical imaging3 , sensor networks6 , neuroscience7 , finance2 , and many more. Essentially,
persistent homology captures the “shape” and “size” of simplicial complexes as skeletal representations of
data and summarizes this information in a persistence diagram. Persistence diagrams are algebraic objects
by construction that live in a well-defined metric space. They are also random objects, by nature of the
underlying point cloud from which they arise.
Despite their widespread use in data analysis, there remains much to be understood about the interplay
between the algebraicity and randomness of persistence diagrams. This project will explore this intersection
and seek to better understand the random nature of points on a persistence diagram as well as the random
nature within persistence diagram space using theory from uncertainty quantification—such as polynomial
chaos9 and spline-based methods4 —to perform moment and density estimation on persistence diagrams.
Time permitting, there may be the opportunity to explore an application to biochemical reaction networks.

Prerequisites

The following skills are required for this project.

• Interest in geometry, algebra and topology: Prior experience with algebraic topology is not required,
but undergraduate experience (with good results) in both algebra and point-set topology is strongly
recommended.
• Experience with probability, statistics, and stochasticity/randomness via appropriate choice of mod-
ules offered by the programme.
• Independence and creativity: The intersection of the fields of topological data analysis and uncer-
tainty quantification is novel and though there is some relevant work, the area is wide open for
exploration and potential for several different aspects of study. The ideal candidate will play an
active role in directing the project in a direction commensurate with their skills and interests.
• Strong programming skills: Ability to write code in your choice of language. Note that the most
widely used TDA libraries are written in Python and C++.

References
1. G. Carlsson, A.J. Zomorodian (2005). Computing Persistent Homology. Discrete Computational Geom-
etry. 33:249–274.
2. J. Charlier (2019). From persistent homology to reinforcement learning with applications for retail bank-
ing. arXiv. 1911.11573.
3. L. Crawford, A. Monod, A.X. Chen, S. Mukherjee, R. Rabadán (2020). Predicting Clinical Outcomes
in Glioblastoma: An Application of Topological and Functional Data Analysis. Journal of the American
Statistical Association. 115(531):1139–1150.
4. A. Ditkowski, G. Fibich, A. Sagiv (2020). Density Estimation in Uncertainty Propagation Problems Using
a Surrogate Model. SIAM Journal on Uncertainty Quantification 8(1):261–300.
5. H. Edelsbrunner, D. Letscher, A. Zomorodian (2000). Topological persistence and simplification. Pro-
ceedings of the 41st Annual Symposium on Foundations of Computer Science. 454–463.
6. R. Ghrist, V. de Silva (2006). Coordinate-free coverage in sensor networks with controlled boundaries via
homology. International Journal of Robotics Research.. 25:1205–1222.
7. C. Giusti, E. Pastalkova, C. Curto, V. Itskov (2015), Clique topology reveals intrinsic geometric structure
in neural correlations. Proceedings of the National Academy of Sciences.. 112(44):13455–13460.
8. A. Monod, S. Kališnik, J.Á. Patiño-Galindo, L. Crawford (2019). Tropical sufficient statistics for persis-
tent homology. SIAM Journal on Applied Algebra and Geometry. 3(2): 337–371.
9. N. Wiener (1938). The Homogeneous Chaos. American Journal of Mathematics 60(4):897–936.
Statistical learning of the generation probability of immune receptors
Supervisor: Barbara Bravi (Department of Mathematics, Imperial College London)
Co-supervisor: Becca Asquith (Department of Infectious Disease, Imperial College London)
Collaborators: Ben Willcox (Birmingham), Carrie Willcox (Birmingham) and Derek Macallan (SGUL)

T cells are cells of the immune system which play a major role in protecting the organism against
infections and malignancies. Their action is based on the molecular recognition of infected or cancer
cells, which relies on the binding between the T cell membrane receptors and protein targets displayed
on the surface of infected or cancer cells. Every individual is endowed with an extremely diverse set of
T-cell receptors; such diversity is generated through a stochastic process, the V(D)J recombination [1],
whereby gene segments are stochastically chosen, combined and modied (through random insertion or
removal of nucleotides) to form the gene from which the receptor is translated. After this generation step,
the receptors undergo a process of selection based on their binding properties, which essentially discards
the receptors that are likely to be non-functional.
The so-called γδ T cells [2] are a subset of T cells whose functional characterization is far from
complete, particularly in comparison to a much better known type of T cells called αβ T cells. For
instance, the statistics of receptor generation and selection have been characterized quantitatively for αβ T
cells, using techniques of statistical learning that allow one to build probabilistic models of generation [3,4]
and selection [5] from large datasets of receptor sequences (i.e., the receptors' constitutive chains of
nucleotides or amino acids). On the other hand, the same characterization for γδ T cells is lacking, and
even whether or not γδ T cells are subject to selection during their development is still an open question.
We aim at tackling these questions in a quantitative way by detecting the statistical signatures of
generation and selection mechanisms in newly produced datasets of γδ T-cell receptor sequences. Starting
from this dataset, the main objective of this project is to construct a probabilistic model of receptor
generation for γδ T cells, using the statistical learning approach proposed for αβ T cells [3]. Such an
approach is based on the Expectation-Maximization algorithm to infer the parameters of a probabilistic
model describing step by step the process underlying generation, i.e. V(D)J recombination.
The inferred model will inform us of the diversity and the statistical properties of the γδ receptor
repertoires resulting from generation, setting a baseline expectation in terms of receptor statistical distri-
bution; modelling the dierences with respect to this expectation will eventually allow us to detect and
quantify signatures of selection. Time permitting, we will start to explore also this second question by
learning, using dierent information in the same dataset, a probabilistic model of γδ receptor selection.
This work will be performed in collaboration with our experimental colleagues, Ben Willcox, Carrie
Willcox and Derek Macallan who have generated the experimental data.

References
1. Hozumi N, Tonegawa S. Evidence for Somatic Rearrangement of Immunoglobulin Genes Coding for Variable and
Constant Regions. PNAS. 1976;73(10):36283632.
2. Ribot JC, Lopes N, Silva-Santos B. Γδ T Cells in Tissue Physiology and Surveillance. Nat Rev Immunol. 2021;21(4):221
232.
3. Marcou Q, Mora T, Walczak AM. High-Throughput Immune Repertoire Analysis with IGoR. Nature Communications.
2018;9(1):110.
4. Sethna Z, Elhanati Y, Callan CG, Walczak AM, Mora T. OLGA: Fast Computation of Generation Probabilities of B-
and T-Cell Receptor Amino Acid Sequences and Motifs. Bioinformatics. 2019;35(17):29742981.
5. Sethna Z, Isacchini G, Dupic T, Mora T, Walczak AM, Elhanati Y. Population Variability in the Generation and
Selection of T-Cell Repertoires. PLOS Computational Biology. 2020;16(12):e1008394.
A probabilistic modelling approach to the detection of immune
escape in cancer evolution
Supervisor: Barbara Bravi (Department of Mathematics, Imperial College London)
Co-supervisor: Benjamin D. Greenbaum (Department of Epidemiology and Biostatistics, Memorial Sloan
Kettering Cancer Center, New York)

Cancer progression is characterized by the appearance of genetic mutations, and what factors shape
the observed distribution of mutations is an outstanding open question. In particular, it is not clear
what is relative interplay between immune escape, i.e., the occurrence of mutations disrupting immune
recognition of cancer cells, and the `mutational signatures' of dierent cancer types [1], i.e., the char-
acteristic patterns in the distribution of mutations reecting the intrinsic or extrinsic factors (such as
environmental carcinogens, UV radiation, smoking) from which they might originate.
The objective of this project is to design a quantitative framework to address this question, based on
tness models and probabilistic machine learning tools that are able to measure the cost of mutations
in terms of their potential to be recognized by the immune system [24]. The design of such framework
requires two main steps: (i) sampling by Monte Carlo the background distribution purely induced by
mutational signatures, starting from genes that are known to harbour oncogenic mutations (such as p53
and KRAS) and using mutational rates that reect mutational signatures (from the COSMIC database
cancer.sanger.ac.uk/signatures); (ii) learning the probabilistic models that describe the potential of muta-
tions to be recognized constrained to dierent genetic make-ups (since dierences in the genetic make-up
lead to dierences in the immune response across individuals). The models' predictions will be then
deployed to map the distribution of mutational costs in terms of immune recognition. The MSc project
will focus on step (ii), starting from the background distribution for p53 obtained in the work [5].
The result will be a reference model for cancer mutational landscapes providing a baseline expectation
of the immune escape potential of typical mutations in cancer, which can be used as term of comparison
to detect deviations with respect to this baseline indicating immune escape.

Possible directions that could be developed as a PhD project include: (i) a renement of the back-
ground distribution simulation step, by taking into account higher-order dependencies among nucleotides
of mutational signatures; (ii) comparisons to cancer evolution datasets to dene a quantitative measure
of immune escape based on the deviation with respect to the reference model's baseline expectation;
(iii) a systematic application to mouse datasets, where it is possible to obtain a prole of early-stage vs
late-stage mutations and hence to assess, through the designed framework, the temporal progression of
immune escape.

References
1. Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin AV, et al. Signatures of Mutational
Processes in Human Cancer. Nature. 2013;500(7463):415421.
2. Šuksza M, Riaz N, Makarov V, Balachandran VP, Hellmann MD, Solovyov A, et al. A Neoantigen Fitness Model Predicts
Tumour Response to Checkpoint Blockade Immunotherapy. Nature. 2017;551(7681):517520.
3. Bravi B, Tubiana J, Cocco S, Monasson R, Mora T, Walczak AM. RBM-MHC: A Semi-Supervised Machine-Learning
Method for Sample-Specic Prediction of Antigen Presentation by HLA-I Alleles. cels. 2021;12(2):195202.e9.
4. Bravi B, Balachandran VP, Greenbaum BD, Walczak AM, Mora T, Monasson R, et al. Probing T-Cell Response by
Sequence-Based Probabilistic Modeling. PLOS Computational Biology. 2021;17(9):e1009297.
5. Hoyos D, et al. Under revision. 2021.
Professor Colin Cotter

Modified CR1-DG0 discretisation for atmosphere and ocean models have worked a lot on compatible finite element
methods for geophysical fluid dynamics (see this book for example:
https://link.springer.com/book/10.1007/978-3-030-23957-2) which have a number of special properties that make
them great for this application area. One disadvantage is that they are not very efficient when using explicit time
integration methods because you still have to solve a globally-coupled matrix-vector system at each Runge-Kutta
stage. On the other hand, another finite element discretisation, CR1-DG0, only requires the solution of a diagonal
matrix, meaning that it is good
for explicit time-integration methods. However, CR1-DG0 does not share the special properties of compatible finite
elements. I have
discovered a modification of CR1-DG0 that recovers these properties. I would like to publish this work but it
requires the design and implementation of numerical experiments to demonstrate and benchmark this in practice,
hence this project. The project will develop code
using the Firedrake library: https://www.firedrakeproject.org/

Efficient solution of energy-conserving methods for geophysical flows:

My recently graduated PhD student Golo Wimmer developed space discretisations for geophysical flows that
conserve energy whilst controlling numerical wiggles through upwind discretisations. These have wonderful
properties but are yet to impact practical use because
of the lack of efficient algorithms to solve the systems of nonlinear equations that arise from them (Golo
demonstrated them using methods that are 10-40 times slower than the algorithms that can be used on standard
discretisations). I have designed a new approach to
efficiently solving these equations based on an "augmented Lagrangian" technique, and this project would
implement and explore this approach. The project will develop code using the Firedrake library:
https://www.firedrakeproject.org/

Finite element neural operators:

Neural operators are neural net architectures that are designed to

learn operators mapping between infinite dimensional function spaces. They can be used to e.g. learn the time-T
solution map for
time-dependent PDEs, solve PDE constrained inverse problems, etc. This project will investigate a new type of
neural operator that is implementable on finite element spaces defined on unstructured meshes, which can be applied
to e.g. fluid dynamics problems in complicated domains such as flow past objects etc. The method is based upon
composition of fast approximate solvers for elliptic PDEs with coefficients that play the role of the neural network
parameters. In
this project we will explore this finite element neural operator
approach by developing an implementation using the Firedrake library: https://www.firedrakeproject.org/
 Michele Coti Zelati: Enhanced diffusion for radial flows

The goal of this project is to study the enhanced diffusion properties of a passive scalar f that satisfies the
advection diffusion equation
∂t f + u · ∇f = ν∆f, (1)
where ν > 0 is a small diffusion parameter and u is a given, time-independent and divergence-free velocity vector
field. In fairly general settings, it is not hard to show that the energy of the solution (namely, the L2 -norm) decays
q
exponentially as e−νt . However, it is expected that the presence of the flow u speeds up the rate to e−ν t , for some
q ∈ (0, 1) depending on u. This is a manifestation of a phenomenon called enhanced diffusion.
In this project, we will take a look at the case in which u describes a circular flow, which in polar coordinates
look like  
− sin θ
u(r, θ) = v (r) , r ≥ 0, θ ∈ [0, 2π), (2)
cos θ
for some smooth decreasing function v : [0, ∞) → R, with v 0 (0) = 0. The goal is to describe in mathematically
rigorous terms the features of Figure 1.

F IGURE 1. The evolution of a drop of slightly diffusive “cream” radial stirred into a “cup of coffee”
with impermeable walls. Initially, pure advection is the dominant effect. As time progresses, the
solution becomes radially symmetric. After this time, the cream simply diffuses across the (circular)
streamlines.

The problem can be approached in various ways: it can be rephrased into studying the spectral properties of the
operator u · ∇ − ν∆, or it can be studied using an energy method called hypocoercivity, see for instance [1].
Prerequisites. Students taking on this project are required to have some basic understanding of differential
equations, multivariable calculus, Fourier series and Hilbert spaces.

References
[1] M. Coti Zelati and M. Dolce, Separation of time-scales in drift-diffusion equations on R2 , J. Math. Pures Appl. (9) 142 (2020), 58–75.

1
 Michele Coti Zelati: Landau damping for active Brownian particles

Active matter systems consisting of many interacting self-propelled particles occur in many applications ranging
from synthetic self-propelled colloids, microtubules, and bacterial suspensions, to large-scale systems such as fish
schools, bird flocks and collective robotics.
If one considers identical Brownian particles moving in a periodic box, a so-called mean-field limit procedure
allows to derive a partial differential equation for the macroscopic density f = f (t, x, θ) of particles with direction
angle θ, at time t and position x, of the form
∂t f + ∇ · (v(ρ)f e(θ)) = De ∆f + DR ∂θθ f. (1)
Here, De is an effective diffusion depending on how crowded the system is, DR is a free rotational diffusion with
diffusion coefficient, e(θ) = (cos θ, sin θ) and v is a nonlinear effective speed, which depends on the the spatial
macroscopic density
Z 2π
ρ(t, x) = f (t, x, θ)dθ. (2)
0
The goal of this project is to study the stability of the homogeneous solution f? = φ/2π, of mass φ ∈ [0, 1],
for various models of the form (1). Even when no diffusion is present (i.e. De = DR = 0), the problem can
be understood by means of a mysterious stability mechanism in plasma physics, known as Landau damping. It
originates by an oscillatory behaviour known as phase mixing, and was only recently understood in an important
article by C. Mouhot and fields medalist C. Villani [1]. The project will require the understanding of a powerful
theory of oscillatory integrals, know as stationary phase method.
Prerequisites. Students taking on this project are required to have some basic understanding of differential
equations, multivariable calculus, Fourier series and Hilbert spaces.

References
[1] C. Mouhot and C. Villani, On Landau damping, Acta Math. 207 (2011), no. 1, 29–201. MR2863910

1
 Wave localization in random media
Dr Bryn Davies and Prof Richard Craster
Key words: differential equations, wave physics, numerical simulations

Figure 1: When a defect is introduced to a periodic structure, there can exist

eigenmodes that are localized in a region of the defect.

This project will be an opportunity to become familiar with cutting-edge ideas in wave physics and
contribute to active research. A general principle in wave physics is that if a perturbation is made to a
periodic structure then waves with specific frequencies will be localized (that is, guided or trapped) in a
region of that perturbation (see Figure 1). This has been studied in many different settings and exploited
for many different applications. One perspective on this, which has been widely studied, is that we
can assign topological indices to periodic structures and use these to predict the occurrence of localized
modes (which are often known as topologically protected modes to describe their special properties) [3].
Conversely, the behaviour when random perturbations are made to the structure is less well understood.
In this case, localized modes have been shown to occur and landscape functions have been used to predict
where they may occur [2].
In this project, we are interested in studying the occurrence of localization in randomly perturbed
media and investigating the extent to which this can be understood using the established theory of topo-
logical edge modes. Initially, we will explore a one-dimensional discrete lattice, modelled using difference
equations. A similar system was studied in [4] for the case of deterministic, localized perturbations.
The student will develop numerical codes to model the problem and perform experiments to understand
the behaviour. They will also need to familiarise themselves with the background literature on An-
derson and weak localization [2] and topological insulators [3]. One-dimensional systems based on the
Su-Schrieffer–Heeger (SSH) model are of particular interest [1]. The fundamental question is how this
theory can be translated to our setting and used to predict the observed behaviour.
If good progress is made, these results can be extended to mass-string systems (modelled by ordinary
differential equations) and, subsequently, to multidimensional partial differential systems, similar to those
studied in [1].
Prerequisites:
• basic understanding of differential equations,
• some working knowledge of simulations in MATLAB or Python,
• interest in wave physics.

References
[1] H. Ammari, B. Davies, E. O. Hiltunen, and S. Yu. Topologically protected edge modes in one-
dimensional chains of subwavelength resonators. J. Math. Pure. Appl., 144:17–49, 2020.
[2] M. Filoche and S. Mayboroda. Universal mechanism for anderson and weak localization. Proc. Natl.
Acad. Sci. USA, 109(37):14761–14766, 2012.

[3] A. B. Khanikaev, S. H. Mousavi, W.-K. Tse, M. Kargarian, A. H. MacDonald, and G. Shvets. Photonic
topological insulators. Nat. Mater., 12(3):233–239, 2013.
[4] M. Makwana and R. Craster. Localised point defect states in asymptotic models of discrete lattices.
Q. J. Mech. Appl. Math., 66(3):289–316, 2013.

1 of 1
Benchmarking implementations of quantum stochastic simulators
Dr Thomas Elliott
The field of quantum information theory combines information theory with quantum
physics [1], opening a new paradigm of information processing. Quantum computers –
devices that can perform computations with quantum information – promise to provide
significant advantages for fields that rely on immense computational power, such as ma-
chine learning, financial modelling, and computational biology. While the realisation of
quantum computers poses an extreme engineering challenge, in recent years the threshold
where quantum computers can outperform our best classical computers (at specific tasks)
has been surpassed [2], heralding the beginning of practical quantum computational ad-
vantages.
One such area of application is the simulation of complex stochastic processes. To repli-
cate the behaviour of a stochastic process, it is in general necessary to store information
about the past of the process, requiring the simulator to have a memory. It has been
shown that simulators capable of storing and processing quantum information can do
this with a lower memory overhead than possible with any classical counterpart [3]. In
some circumstances, this quantum advantage can scale without bound [4]. Given the
ubiquity of stochastic simulation, these results are expected to find application across the
quantitative sciences.
This project will explore the proof-of-principle implementation of such quantum stochas-
tic simulators on cloud-accessible quantum processors, such as the IBM Quantum Ex-
perience [5]. Focussing on toy processes that require only a few quantum bits (qubits)
to realise, the student will first trial the implementation on classical simulators of noisy
quantum processors, and then later on real quantum processors. We will then inves-
tigate how the performance of these implementations can be quantified and evaluated.
Subsequently, we will explore methods for improving this performance, such as incor-
porating quantum error correction techniques and recompiling computations to require
fewer quantum operations.
The project will be at least in part numerical, with scope for a significant analytical
component. By the end of the project, the student will have gained experience in state-of-
the-art research in quantum computation and information theory. Some prior knowledge
of the field would be useful, but not essential. Familiarity with basic quantum mechanics
is required.
[1] Nielsen, M. A. and Chuang, I. L. Quantum Computation and Quantum Information Cambridge University Press (2010).
[2] Arute, F. et al. Quantum supremacy using a programmable superconducting processor Nature 574, 505 (2019).
[3] Gu, M. et al. Quantum mechanics can reduce the complexity of classical models Nat. Commun. 3, 762 (2012).
[4] Elliott, T. J. et al. Extreme Dimensionality Reduction with Quantum Modeling Phys. Rev. Lett. 125, 260501 (2020).
[5] IBM Quantum quantum-computing.ibm.com.

Contact: [email protected]
 DIFFUSE INTERFACE METHODS FOR TWO-PHASE COMPLEX FLUIDS

Dr Andrea Giorgini

The motion of two-phase fluids and their interaction at the interface is a fascinating and chal-
lenging problem in fluid mechanics. Complex phenomena already occur in simple experiments
when the spatial regions occupied by a single flow is deformed by moving fluid structures, and
the interface area decreases its characteristic length scale, develops singularities and changes its
topology. These phenomena are ubiquitous in materials science, engineering and biology: from
coalescence of drops to phase separation, from topological defects of liquid crystals to electrowet-
ting, and from non-isothermal flows to tumor growth dynamics. The Diffuse Interface (DI) theory,
also called Phase Field theory, represents nowadays a successful method to simulate fluid mix-
tures, being able to capture the main features of the mutual interplay in the motion of two fluids
and large interface deformations. The key concept of DI theory is to represent the interface (or
phase boundary) as region with finite thickness among which physical quantities have a rapid but
smooth variation. The interfaces are described as the level set of a order parameter, which is usu-
ally the fluid concentration or rescaled volume fraction. The dynamics of the order parameter is
deduced from the principles of fluid mechanics, thermodynamics and statistical mechanics. In
contrast to other methods in which the interface is a time-dependent surface, the main advantage
of the DI formulation is the Eulerian description of the interface. In addition, DI models can be
regarded as a regularization of Sharp Interface models (free boundary problems) for simple fluid
mixtures driven by surface tension or surface evolution problems driven by mean curvature. On the
other hand, DI models can be used to represent the dynamics of complex fluids (polymers, liquid
crystals and biofluids), which are those mixtures whose molecular interaction at the microscopic
scale affects the macroscopic dynamics.

Project Description. This project is devoted to the mathematical study of systems of partial
differential equations originating from the DI theory for fluid mixtures. These models consist of
the Navier-Stokes equations for the velocity of the mixture coupled with the phase field equations
for the order parameter. The student’s task will consist in studying the well-posedness theory and
longtime behavior of a DI system for complex fluids.
 CAHN-HILLIARD DYNAMICS IN MATERIALS SCIENCE

Dr Andrea Giorgini

The Cahn-Hilliard equation is a famous model in Materials Science. It was proposed in 1958 by
J.W. Cahn and J.E. Hilliard to provide a mesoscale description of the evolution of microstructures
during the phase separation in a binary alloys system. Such phenomenon is characterized by an
early stage where the so-called spatial spinodal decomposition takes place, followed by the coars-
ening process. In the latter, the average size of these spatial domains with same phase increases
over time at the expense of the smaller ones (see the figure below). This occurs when a homoge-
neous mixture undergoes a rapid cooling below a certain critical temperature. The Cahn-Hilliard
equation and its variants are particular Diffuse Interface (Phase Field) models and have been em-
ployed for different phenomena which are characterized by pattern formation, segregation-like
processes and interface motion, such as grain boundary, nucleation, liquid-liquid phase transition
and crystallization. In a bounded domain Ω ⊂ Rd, d = 2, 3, we consider the Gindzburg-Landau
free energy Z
ε 1
E(φ) = |∇φ|2 + Ψ(φ) dx,
Ω 2 ε
where φ is the phase function, ε is a parameter related to the thickness of the interface, and the free
energy density Ψ is the Flory-Huggins potential
 
θ θ0
Ψ(s) = (1 + s) log(1 + s) + (1 − s) log(1 − s) − s2 , s ∈ [−1, 1],
2 2
where the constant parameters θ and θ0 fulfill the conditions 0 < θ < θ0 . The Cahn-Hilliard system
is the gradient flow with respect to the (H 1 (Ω))0 metric of the total free energy E(φ), namely
1
∂t φ = ∆ − ε∆φ + Ψ0 (φ) , in Ω × (0, ∞),


ε
subject to physical boundary conditions.
Project Description. This project is devoted to the study of the recent theoretical results con-
cerning local, nonlocal and fractional Cahn-Hilliard equations with logarithmic (singular) poten-
tials. More precisely, the student’s task will consists in understanding the techniques for the exis-
tence and uniqueness of weak solutions, regularity theory and validity of the separation property
for Cahn-Hilliard type equations. The study of numerical approximations can be also discussed.

F IGURE 1. Different stages of the phase separation in the Cahn-Hilliard equation

 Optimal vibration control of slender structures

Supervisor: Dante Kalise

Description

In this project we will study active vibration control strategies for slender struc-
tures such as bridges, trusses, and aircraft wings. We will study this problem in the
framework of PDE-constrained optimization, where the control is designed by op-
timizing a performance index (vibration mitigation plus control effort) constrained
to a PDE governing the structural vibration dynamics. Since standard structural
vibration models are linear, after a suitable discretization the active vibration con-
trol problem can be cast as a Linear-quadratic Regulator (LQR) problem, and can
be solved trough a large-scale Algebraic Riccati Equation. As an extension, we will
study a mathematical formulation for the optimal location and design of sensors
and actuators in the framework of bi-level optimization.

Plan

The project is split into three stages:

(1) Modelling and approximation of structural vibration phenomena: beam and

plate models, time/frequency-dependent formulations, FEM and spectral
approximation.
(2) PDE-constrained optimization. The LQR problem for structural dynamics,
optimality conditions for the stationary problem. Numerical approximation
of large-scale control problems.
(3) Actuator/sensor aspects: piezoelectric actuators, optimal actuator/sensor
location and design.

A background/interest on computational mathematics and mathematical modelling

with PDEs is essential.

References
(1) E. Hernández, D. Kalise and E. Otárola. Numerical approximation of the
LQR problem in a strongly damped wave equation, Computational Opti-
mization and Applications, 47:161–178 (2010).
(2) E. Hernández, D. Kalise and E. Otárola. A locking-free scheme for the
LQR control of a Timoshenko beam, J. Comput. Appl. Math. 235(5):
1383-1393 (2011).
(3) M. S. Edalatzadeh, D. Kalise, K. A. Morris and K. Sturm. Optimal Actu-
ator Design for the Euler-Bernoulli Vibration Model Based on LQR Per-
formance and Shape Calculus, IEEE Control. Syst. Lett. 6: 1334-1339
(2022).
 Deep Neural Networks for Real-time Trajectory Planning

Supervisor: Dante Kalise

Description

Different problems in robotic locomotion can be studied in the framework of optimal

control theory. For example, we can look for a set of actions (controls) which take
a drone from point A to B by minimizing the amount of time or energy that
is required to complete the task. These control signals are often expressed as
feedback laws, that is, real-time actions which can be implemented solely based
on the current state (position, velocity) of the robot. In this project, we will
develop a deep learning formulation of the optimal trajectory planning problem
outlined above. y combining methods from dynamical systems, control theory, and
polynomial systems, we will generate a synthetic dataset of optimal input-output
pairs. This synthetic dataset will be used to cast a supervised learning problem to
approximate an optimal feedback law to be applied in real time after a measurement
of the current state of the system.

Plan

The project is split into three stages:

(1) A brief introduction to optimal control: minimum time formulation, bang-

bang controls, switching structures.
(2) Parametrizing optimal control problems as nonlinear and/or polynomial
systems. Solution strategies via iterative methods or computer algebra.
(3) Synthetic data generation and supervised learning approaches for synthe-
sizing feedback laws. Applications in robotic locomotion.

Prerequisites: basic knowledge of dynamical systems, computational simulation

tools, and machine learning (DNNs, supervised learning). Programming skills in
Matlab and/or Python. Willingness to learn about polynomial systems and/or
computer algebra is desirable.

References
(1) U. Walther, T. T. Georgiou and A. Tannenbaum. On the computation of
switching surfaces in optimal control: a Grobner basis approach, in IEEE
Transactions on Automatic Control, vol. 46, no. 4, pp. 534-540, April
2001, doi: 10.1109/9.917655.
(2) B. Karg and S. Lucia. Efficient Representation and Approximation of Model
Predictive Control Laws via Deep Learning, in IEEE Transactions on Cyber-
netics, vol. 50, no. 9, pp. 3866–3878, Sept. 2020, doi: 10.1109/TCYB.2020.2999556.
(3) B. Azmi, D. Kalise and K. Kunisch. Optimal Feedback Law Recovery by
Gradient-Augmented Sparse Polynomial Regression, in Journal of Machine
Learning Research, 22(48):1–32, 2021.
 Time-optimal control of agent-based dynamics

Supervisors: Dante Kalise and Grigorios A. Pavliotis

Description

In this project we will study the synthesis of control laws for agent-based models
arising in large animal populations (bird flocks, fish shoal, and sheep) and swarm
robotics. The objective is to determine optimal control signals which are able to
steer the population towards a desired state, e.g. a certain spatial configuration or
flocking state. For this, we will follow a dynamic optimization approach, deriving
optimality conditions which characterize an optimal intervention. In particular, we
will focus on the synthesis of optimal controls which are able to achieve the objective
in a minimum amount of time. The derivation of optimality conditions guides the
construction of numerical methods for an effective control synthesis. Moreover, we
will explore the control synthesis as the number of agents grows, paving the way
for a mean-field modelling of the time-optimal control problem.
A background on dynamical systems, numerical analysis, and optimisation/control
is desirable.

Plan

The project is split into three stages:

(1) Mathematical modelling of collective behaviour phenomena: animal be-

haviour, agent-based models, and swarm robotics.
(2) Time-optimal control, optimality conditions for agent-based dynamics.
(3) Computational synthesis of control laws through dynamic optimization.
Mean-field scaling of the control problem.

References
(1) Y. Chen and T. Kolokolnikov. A minimal model of predator-swarm inter-
actions, J. R. Soc. Interface 11(94):20131208 (2014).
(2) D. Strömbom, R.P. Mann, A.M. Wilson, S. Hailes, J. Morton, D.J. Sumpter,
and A. J. King. Solving the shepherding problem: heuristics for herding au-
tonomous, interacting agents, J. R. Soc. Interface 11(100):20140719 (2014).
(3) M. Bongini, M. Fornasier, M. Hansen, and M. Maggioni . Inferring in-
teraction rules from observations of evolutive systems I: The variational
approach, Math. Models Methods Appl. Sci. 27(5):909–951 (2017).
(4) R. Bailo, M. Bongini, J.A. Carrillo and D. Kalise. Optimal consensus con-
trol of the Cucker-Smale model, IFAC-PapersOnLine 51(13)(2018):1–6.
 Self-sustaining processes and coherent structures in
two-dimensional flows

Supervisor: Andrew Walton

Description
The topic of transition to turbulence in wall-bounded flows is a classical problem. Reynolds
pioneering experiment (1) demonstrates how pipe flow undergoes this transition from an orderly
or laminar state to a disordered or turbulent one. One well-established route from laminar to
turbulent flow occurs via a sequence of steps, beginning with the linear instability of infinitesimal
disturbances (e.g. see (2)). This description, however, is problematical in view of the fact that
many wall-bounded flows such as the pipe flow mentioned above can be shown to be stable to
small disturbances. Evidently, any theory which claims to explain the route to transition in such
flows must therefore be inherently nonlinear.
Within a turbulent flow it is often possible to discern the existence of coherent structures de-
spite the apparent randomness of the overall motion. For example, streaky structures possessing
cross-stream variation can be seen in many experimental visualizations of shear flows (e.g. (3)).
In addition these streaky structures often appear to be accompanied by a longitudinal vortex
structure in the cross-stream plane.
Numerical simulations of turbulent flow (e.g. (4)) have made it possible to observe an inter-
action between three distinct structures in the flow: a flow in the cross-stream plane with no
downstream dependence, known as a roll flow, a streamwise component possessing some cross-
stream variation (a streak), and finally a three-dimensional travelling wave propagating in the
streamwise direction.
The aim of this project is to explore mathematically the interaction between these three
components in a three-dimensional flow and to examine whether a similar mechanism can operate
in two dimensions.

Plan
There are possibilities for the project to be more or less computationally-oriented depending on
the interests of the student but there will be some numerical computation (in Python or Matlab)
required.
The project will commence with some reading on Self-Sustaining Processes and Coherent
Structures which will be made available by me. The student will be encouraged to fill in some
of the details within these notes and to reproduce some of the computations.
Next, a self-sustaining process for two-dimensional flows involving wave-mean flow interaction
will be studied theoretically for a general parallel (or nearly parallel) base flow.
Following this, the effect of this mechanism upon specific flows such as pipe flow and plane
and annular Poiseuille and Couette flows will be studied.
If time allows the instability structures derived will be compared with exact solutions of the
two-dimensional Navier-Stokes equations.
Prerequisites: Some prior exposure to fluid mechanics is desirable together with
a knowledge of Matlab or Python.

1
References
1. An experimental investigation of the circumstances which determine whether the motion of
water shall be direct or sinuous, and of the law of resistance in parallel channels. Reynolds,
O. (1883) Phil. Trans. R. Soc., 174, 935-982.

2. Introduction to hydrodynamic stability, Drazin, P. G. (2002) Cambridge Texts in Applied

Mathematics.

3. The structure of turbulent boundary layers, Kline S. J., Reynolds, W. C., Schraub, F. A.&
Runstadler, P. W. (1967) J. Fluid Mech., 30, 741–773.

4. Regeneration mechanisms of near-wall turbulence structures. Hamilton, J. M., Kim, J. &

Waleffe, F. (1995) J. Fluid Mech., 287, 317–348.

2
 The nonlinear stability of plane Couette-Poiseuille flow with
zero mean advection velocity

Supervisor: Andrew Walton

Description
The topic of transition to turbulence in wall-bounded flows is a classical problem. Reynolds
pioneering experiment (1) demonstrates how pipe flow undergoes this transition from an orderly
or laminar state to a disordered or turbulent one. One well-established route from laminar to
turbulent flow occurs via a sequence of steps, beginning with the linear instability of infinitesimal
disturbances (e.g. see (2)). This description, however, is problematical in view of the fact that
many wall-bounded flows such as the pipe flow mentioned above can be shown to be stable to
small disturbances. Evidently, any theory which claims to explain the route to transition in such
flows must therefore be inherently nonlinear.
A particular flow which can be shown to be linearly stable is that between parallel moving
walls subject to a constant streamwise pressure gradient such that the resulting velocity profile,
when averaged across the channel, is zero. This flow can be set up relatively easily in a laboratory
and, despite the property of linear stability referred to above, exhibits transition to turbulence
as the relative wall speed is increased. An interesting feature of this process is that before the
flow becomes fully turbulent we observe the formation of a self-sustained turbulent spot in an
otherwise laminar flow (3).
Although the linear stability of this flow has been well-studied there are very few theoretical
studies of the nonlinear stability properties of this flow. This project aims to use high Reynolds
number asymptotic methods including the use of nonlinear critical layer theory to understand
the interactions occurring between the boundary layers and the core flow that self-sustain this
sophisticated nonlinear structure.

Plan
The project will commence with a derivation of the basic flow and an investigation of its linear
stability properties via the solution of an Orr-Sommerfeld eigenvalue problem using Chebyshev
collocation (4).
Once the linear stability is established we can take a nonlinear asymptotic approach at high
Reynolds number where the disturbance field is practically free from viscosity except within thin
layers adjoining each wall and a region where the basic flow velocity is close to the phase speed of
the propagating perturbation: this region is known as a critical layer. By linking the properties
of the various layers we can establish a self-sustaining interaction in which the amplitude of the
disturbance can be related to its wavelength and phase speed. Although a substantial part of
this analysis is analytical, some numerical calculations are necessary.
With the self-sustaining mechanism established, we can calculate the distortion to the mean
flow induced by the disturbance and compare it to that found in the experiments referred to
earlier.
Prerequisites: Some prior exposure to fluid mechanics is desirable together with
a knowledge of Matlab or Python.

1
References
1. An experimental investigation of the circumstances which determine whether the motion of
water shall be direct or sinuous, and of the law of resistance in parallel channels. Reynolds,
O. (1883) Phil. Trans. R. Soc., 174, 935-982.

2. Introduction to hydrodynamic stability. Drazin, P. G. (2002) Cambridge Texts in Applied

Mathematics.

3. Couette-Poiseuille flow experiment with zero mean advection velocity: subcritical transi-
tion to turbulence. Klotz, L., Lemoult, G., Frontczak, I., Tuckerman, L. S. & Wesfreid, J.
E. (2017) Phys. Rev. Fluids, 2, 043904.

4. Stability and transition in shear flows. Schmid, P. J. & Henningson, D. S. (2001) Applied
Mathematical Sciences, Springer.

2
 Investigation of the stability properties of shear flows over
compliant surfaces

Supervisor: Andrew Walton

Description
The behaviour of fluid flows over flexible surfaces is a challenging one for theoretical fluid dy-
namicists. The traditional instabilities one encounters over a rigid wall are supplemented by
elastic modes arising from the response of the boundary. In this project we will consider flows
through channels with either one or both walls possessing compliant properties. The flow can
be generated by applying a pressure gradient, moving the walls or a combination of both effects.
A simple spring-backed plate model will be adopted to describe the motion of the boundary
and we will assume that any perturbations of the surface are small, so that their effect on the
basic flow will be a linear one. In the case of plane Couette flow, where the flow over rigid
walls is linearly stable, it is possible to show that the surface flexibility destabilizes the flow.
The aim of the project is to investigate the linear stability properties of the ensuing flow both
at asymptotically large Reynolds number and at finite Reynolds number. The former analysis
requires the use of matched asymptotic expansions, while the latter involves computation and
requires the student to write some of their own code.

Plan
There are possibilities for the project to be more or less computationally-oriented depending on
the interests of the student but there will be some numerical computation (in Python or Matlab)
required.
The project will start by considering plane Couette flow subject to rigid boundaries. The
linear stability eigenvalue problem will be posed and solved numerically (using either existing
code or that developed by the student) and the flow will be visualised.
Following this, there will be some modelling of the flexible boundaries resulting in a modified
stability problem that incorporates the wall compliance via a boundary condition. This new
problem will be solved for a range of parameters, and the effects of wall stiffness upon stability
will be documented quantitatively.
Next, the instabilities induced will be analyzed asymptotically at large Reynolds number,
leading to multi-zone interactive structures that elucidate the key physical mechanisms un-
derlying the instability process. The solutions of the corresponding high Reynolds number
eigenrelations will be compared with the computations.
If time allows, the same ideas will be applied to plane Poiseuille flow, in which the rigid state
is already unstable and also to a linear combination of Poiseuille and Couette flows in which the
wall sliding speed provides an extra parameter.
Prerequisites: Some prior exposure to fluid mechanics is desirable together with
a knowledge of Matlab or Python. Some experience of perturbation methods and
matched asymptotic expansions would also be helpful, but not essential.

1
References
1. Introduction to hydrodynamic stability, Drazin, P. G. (2002) Cambridge Texts in Applied
Mathematics.

2. On the stability of plane Poiseuille flow between compliant boundaries. Nagata, M. &
Cole, T. R. (1999) Transactions on Modelling and Simulation, vol. 21
https://www.witpress.com/Secure/elibrary/papers/CMEM99/CMEM99023FU.pdf

3. Instabilities in a plane channel flow between compliant walls, Davies, C. & Carpenter, P.
W. (1997) J. Fluid Mech., 352, 205–243.

2
Pavel Berloff - Estimating Eulerian Transport Tensors from the Lagrangian Trajectories

Stirring of ocean water by mesoscale currents (``eddies'') leads to large-scale transport of

many important
oceanic properties (``tracers'').
These eddy-induced transports can be related to the large-scale tracer gradients, using the
concept of turbulent
diffusion.
The concept is widely used to describe these transports in the real ocean and to represent
them in climate models.
This study focuses on the inherent complexity of the corresponding coefficient tensor (``K-
tensor'') and its
components, defined here in all its spatio-temporal complexity.
Results so far demonstrate that this comprehensive K-tensor is space-, time-, direction- and
even tracer-dependent.
Using numerical simulations with idealized intermediate-complexity models of the North
Atlantic circulation,
it was shown that these properties lead to many discoveries about the involved eddy
effects.

This Project will address important question: to what degree components of the K-tensor
can be approximated from
trajectories of infinitesimal passive (i.e., Lagrangian) particles released in the flow?
This is motivated by the fact that most of the relevant oceanic observations are made in
terms of the Lagrangian
particle approach.
The leading hypothesis is that the real differences between the actual K-tensor and its
Lagrangian approximation
will be huge, and this knowledge will shake up the whole research field.
Knowing these differences quantitatively, as well as qualitatively, will be practically
important new knowledge.
The student will be expected to work with model output data and master theory of
turbulent transport.

Prerequisites: Interest in mathematical modelling, coding skills, ODEs and PDEs, numerical
methods.
 Collective Behavior of interacting particle systems
MSc project 2021 − 2022
Professor G.A. Pavliotis

Interacting particle systems arise in many applications raging from plasma physics and stellar dynamics
to biology and even to algorithms for sampling and optimization and to mathematical models in the social
sciences.
Quite often, the interactions between particles (agents) at the microscale lead to the emergence of col-
lective behavior at the macroscale. This collective behavior can manifest itself in terms of the formation
of clusters (in stellar dynamics models), the emergence of consensus (in models for opinion formation), of
synchronization (in systems of interacting nonlinear oscillators), the swarming of animal populations etc.
The goal of this project is the numerical and analytical study of systems of interacting particles that
exhibit collective behavior and, in particular, the formation of clusters. in particular, we will study the
Langevin dynamics √
q̈i = −∇ΦN (qN ) − γ ṗi + 2γβ −1 Ḃi , (1)
where qN = (q1 , . . . qN ) denotes the position of N particles in R2 , pN = (p1 , . . . pN ) the momentum
vector, γ the friction coefficient, β the inverse temperature and B1 , . . . BN a collection of N independent
two-dimensional standard Brownian motions. The potential ΦN consists of a confining potential and of
pairwise interactions,
∑N
κ ∑
N
ΦN (qN ) = V (xj ) − W (xi , xj ), (2)
2
j=1 i,j=1

where W is a symmetric function. We will consider, in particular the case of Gaussian interactions
|x−y|2
W (x, y) = e− 2σ 2 (3)
−(x−xe )
Other choices, such as the Morse potential W (x) = λ2 (e−2(x−xe )−e , will also be considered. The
parameter (interaction strength) κ can be either positive or negative and it can depend on the number of
particles, e.g. κ ∼ N1 . We will consider the dynamics (1) both in R2N × R2N as well as in ΛN × R2N , where
Λ denotes a box of size L in R2 with periodic boundary conditions. The goal will be to study the problem by
means of extensive numerical simulations as well as by studying analytically the mean field/thermodynamic
limit N → +∞.
Prerequisites for this project are statistical mechanics, stochastic differential equations and numerical
methods for ODEs, SDEs, and PDEs. Useful references are [4, 1, 5, 2, 3].

1
References
[1] F. Bavaud. Equilibrium properties of the Vlasov functional: the generalized Poisson-Boltzmann-Emden
equation. Rev. Modern Phys., 63(1):129–148, 1991.

[2] A. Campa, T. Dauxois, D. Fanelli, and S. Ruffo. Physics of long-range interacting systems. Oxford
University Press, Oxford, 2014. With a foreword by David Mukamel.

[3] N. Martzel and C. Aslangul. Mean-field treatment of the many-body Fokker-Planck equation. J. Phys.
A, 34(50):11225–11240, 2001.

[4] J. Messer and H. Spohn. Statistical mechanics of the isothermal lane-emden equation. Journal of
Statistical Physics, 29(3):561–578, 1982.

[5] H. A. Posch, H. Narnhofer, and W. Thirring. Dynamics of unstable systems. Phys. Rev. A (3),
42(4):1880–1890, 1990.

2
 Stochastic Gradient Descent in Continuous Time

Professor G.A. Pavliotis

The methods and models of machine learning are rapidly becoming de

facto tools for the analysis and interpretation of large data sets. The
ability to synthesise and simplify high-dimensional data raises the
possibility that neural networks may also find applications as efficient
representations of known high-dimensional functions. Training a given
neural network remains one of the central challenges in applications due
to the slow dynamics of training and the complexity of the objective
function. Parameter optimisation in machine learning typically relies on
the stochastic gradient descent algorithm (SGD), which makes an
empirical estimate of the gradient of the objective function over a small
number of sample points. In order to study the properties of stochastic
gradient descent for neural network optimisation, it is possible to recast
the standard training procedure in terms of a system of interacting
particles. It is then possible to use tools from stochastic differential
equations and statistical mechanics to study the improve methodologies
for the training of neural networks. The goal of this project will be to
study recent works on the dynamical study of the training of neural
networks and to implement some of the proposed methodologies for
improving the performance of the SGD algorithm.

Sirignano, J.; Spiliopoulos, K. Stochastic gradient descent in continuous

time. SIAM J. Financial Math. 8 (2017), no. 1, 933–961.

Sirignano, J.; Spiliopoulos, K. Stochastic gradient descent in continuous

time: a central limit theorem. Stoch. Syst. 10 (2020), no. 2, 124–151.

Parameters as interacting particles: long time convergence and

asymptotic error scaling of neural networks
G Rotskoff, E Vanden-Eijnden
Advances in neural information processing systems, 7146-7155 (2018)

Neural networks as interacting particle systems: Asymptotic convexity of

the loss landscape and universal scaling of the approximation error
GM Rotskoff, E Vanden-Eijnden stat 1050, 22 (2018)

Parameter estimation for the McKean-Vlasov stochastic differential

equation
L Sharrock, N Kantas, P Parpas, GA Pavliotis
arXiv preprint arXiv:2106.13751
MSc Applied project proposals – Pruessner, Gunnar
Project 1 - Interacting particles far from non-equilibrium

The study of non-equilibrium systems is often concerned with entropy and entropy production, that
is the rate with which entropy is being produced by a process. While this is normally studied as an
ensemble average, some authors are using methods based on single, stochastic trajectories.
We are interested in studying the entropy production for a non-trivial Markov process. We consider
the motion of a particle that is constrained to move on a ring of states. The dynamics of the particle
are governed by a Poisson process, the particle can hop to the nearest right or left site
independently after an exponentially distributed waiting time, that depends on the position of
another particle that is moving randomly as well.
The entropy production for this process can be calculated for different levels of description: The two
particles may be considered distinguishable or indistinguishable and we may consider an ensemble
of trajectories or a single one. We want to use traditional as well as field-theoretic methods. The aim
of the project is to survey the different methods and to devise a procedure to efficiently change
from one description to another.

Project 2 - Entropy production of random walk on evolving graphs

Similar to the project above, we want to study the entropy production for a particular type of
process. In the present project we consider the entropy production when the very Markov matrix of
the underlying process changes.
For many systems the state space changes with time. If we model the dynamics on the state space as
a random walk on a graph, there will be a competition of timescales between the "mixing time” of
the process on the graph and the characteristic time of the change of the graph. What features does
the entropy have as a consequence? A related problem is the characterisation of the entropy
production of the evolving graph itself (e.g. comparing different remodelling dynamics). This project
will require serious initial numerical input.

Project 3 - Field theory of defect pair creation and annihilation under coarse-graining

Active nematics can be characterised by the creation and annihilation of defect pairs with opposite
topological charge. Unlike passive nematics, these defects behave like “active” quasi-particles. Can
we use Doi-Peliti field theories to characterise these effective dynamics and study the entropy
production of the active vs passive case?

Project 4 - Entropy production determines the material properties of logic active matter

This is in the spirit of Marchetti’s recent papers on logic active matter, where active particles have
some internal state that is modified by interaction/measurement. The setup is a 1d box with active
particles where each particle is either self-propelling to the left or to the right (the internal state is 1
bit), maybe with a bit of diffusion. With some rate m each particle “measures” it’s surrounding area
and detects whether it’s nearest neighbour is to its left or its right (1 bit of information). It then
adjusts the direction of its self-propulsion so that it moves away from its nearest neighbour. If the
measurement rate is sufficiently fast (compared to the self-propulsion), the particles will organise in
what resembles a 1d lattice (they “crystallise”), while for low measurement rates this should looks
like a gas of run-and-tumble particles. The idea is to explore these two phases and the transition. For
example, what happens if I compress the box in the “active solid” phase? Do I get pressure waves
and how do they propagate? What kind of material is this? Also, the entropy production of the
measurement itself is an interesting question, which can also be explored at the level of a single
particle, if we fix the position of its neighbours to simplify things.
 Transition to chaos in model spatially-developing flows

Supervisors: Dr. P. K. Ray and Prof. D. T. Papageorgiou

Project Description

Consider the figure below. The image on the left is a bifurcation diagram for the logistic map, xn+1 =
rxn (1 − xn ); this map was analyzed by Feigenbaum 40 years ago in his foundational study on the period-
doubling route to chaos [1]. The image on the right illustrates the spatial development of a fluid boundary layer
as it transitions from a steady, laminar flow to fully-developed turbulence. The basic question motivating this
project is, can the bifurcation diagram on the left at all describe the dynamics depicted in the boundary layer
on the right? Boundary layers have great practical significance – consider water flowing through pipes in your
home, air flow over an aircraft wing, or hurricanes after landfall. While these flows are undoubtedly important,
they are also enormously complicated and require simulation and analysis of the 3-D Navier-Stokes equations. In
this project, we make a pragmatic compromise and focus on the spatially-developing 2-D Kuramoto-Sivashinsky
(K-S) equation, ut + (u + c)ux + ∇2 u + ∇4 u = 0, which retains many important features of the Navier-Stokes
equations but is simpler to analyze and simulate. Transition to chaos in the K-S system has already been
investigated for confined dynamics on a periodic domain, x ∈ [0, L) [2], and in this project, we will analyze
open flows developing along the half line, x ∈ [0, ∞). Numerical simulations will be used as a ‘laboratory’
for investigating transition scenarios. Statistical methods and nonlinear time series analysis will be applied to
simulation results and connections to insights gained from both chaos theory and linear stability analysis will
be explored and explained.

Learning Outcomes

There will be several learning outcomes emerging from this project:

- You will learn about numerical methods for nonlinear PDEs and acquire proficiency in scientific computing
- You will also learn about nonlinear time series analysis and statistical analysis of complex spatio-temporal
data
- You will learn about linear stability analysis of spatially developing flows and asymptotic methods used to
connect linear theory to observed nonlinear dynamics.

Background: The following courses (or equivalent) from the Applied Mathematics program could prove
useful, however not all are essential: Fluid Dynamics I/II, Hydrodynamic Stability, Asymptotic Analysis,
Numerical Solution of ODEs, Computational PDEs. Some programming experience is essential.

References

[1] S.H. Strogatz, Nonlinear dynamics and chaos, 2000.

[2] Y.-S. Smyrlis and D.T. Papageorgiou. Predicting chaos for infinite-dimensional systems: The
Kuramoto-Sivashinsky equation, a case study. Proc. Natl. Acad. Sci. USA, 88:11129-11132, 1991.
 Waves and turbulence in pedestrian flows

Supervisor: Dr. Prasun Ray

Project Description

Analysis and optimization of pedestrian flows is fundamen-

tally important for both urban design and public health and
safety. These flows can exhibit a range of complex phenomena,
and the aim of this project is to use simulations of pedestrian
flows to investigate these dynamics. You will simulate pedestrian
flows in canonical configurations such as uni- and bi-directional
traffic in a hallway using the descision-based model developed
in [1], and analyze the waves and turbulence that are known to
develop under certain conditions. You will also carry out com-
parisons with experimental studies of pedestrian behavior and
critically assess the strengths and weaknesses of the model and
develop improvements as appropriate.
Illustration of a pedstrian (p1 ) analyzing her
Learning Outcomes path (taken from [1])

There will be several learning outcomes emerging from this project:

· You will learn about numerical methods for large systems of ODEs and optimzation; you will also acquire
proficiency in scientific computing
· You will learn about mathematical modeling of pedestrian behavior
· You will also learn about the dynamics of linear and nonlinear waves in the context of pedestrian traffic

Background: The following courses (or equivalent) may be useful, however not all are essential:
Hydrodynamic Stability, Numerical Solution of ODEs, Computational PDEs, Scientific Computation. Python
programming experience will be helpful.

Reference

[1] M. Moussaid, D. Helbing, & G. Theraulaz. How simple rules determine pedestrian behavior and crowd
disasters. Proc. Natl. Acad. Sci. USA, 108::6884-6888 2011.
MSc projects with Dr Ory Schnitzer and Dr Gunnar Peng

The projects involve mathematical modelling (fluids, waves,…), asymptotic analysis and for some projects
also numerical calculations.

A. Vertical oscillations of levitated “Leidenfrost” drops. Drops are able to levitate for
relatively long times above a sufficiently hot substrate, supported by a cushion of their own vapour. This is
the “Leidenfrost effect,” which you can observe in your kitchen by drizzling water on a very hot pan. There
is currently a lot of interest in studying the rich dynamics of Leidenfrost drops, including their high mobility,
symmetry breaking and spontaneous motion, self-induced oscillations and more; experiments are far ahead
of theory. In this project, we will try to rationalise recent observations of Leidenfrost drops exhibiting
vertical oscillations which can grow into limit-cycle-like bouncing or trampolining. We will start by modelling
the vertical dynamics of a levitated Leidenfrost solid (a chunk of “dry ice”).

B. Trapped modes. Waveguides are devices that are designed to convey propagating waves (acoustic,
electromagnetic, etc.). Nonetheless, waves can also be perfectly trapped within a waveguide by suitably
designed obstacles, namely without interacting with waves propagating towards or outwards from the
obstacle. Such waves are described by solutions to the Helmholtz equation known as trapped modes, or
bound states in the continuum. In this project, we will combine asymptotic methods and conformal
mappings to identify and calculate approximations for trapped modes in slit channels. This approach will
allow us to identify and describe new trapped modes as well as simplify existing results in the literature.

C. Active particles interacting with a boundary. Chemically active particles self-propel in a

viscous fluid by generating around themselves an asymmetric solute distribution, which in turn animates a
flow around the particle. This project is concerned with chemically active particles that are isotropic in their
geometry and surface properties — they self-propel, in a random direction, as a consequence of a
symmetry breaking instability associated with a positive feedback loop between advective-diffusive solute
transport and Stokes flow. We will use asymptotic methods and Stokes-flow theory to study the
trajectories of isotropically active particles interacting with a distant boundary.

D. Vortex shedding in strong-field electrophoresis. Electrophoresis refers to the field-driven

motion of a charged particle that is immersed in an electrolyte. Direct numerical simulations in the strong-
field regime have recently revealed the formation of vortices on one side of the particle; above a critical
field strength, the vortices periodically detach resulting in an unsteady flow reminiscent of the Karman
vortex street (only at zero Reynolds number!). We will use asymptotic tools to study the Poisson–Nernst–
Planck–Stokes set of equations governing this phenomenon, and investigate the origin of these vortices and
their stability, building on existing strong-field theories in the literature.
MSc projects with Dr Igor Shevchenko
• Large scale low-frequecny variability of the midlatitude ocean circulation
Understanding origins of the large-scale low-frequency variability (LFV) of the ocean is not only
one of the central questions in the Earth system modelling and geophysical fluid dynamics, but
also one of the serious challenges in predictive understanding of climate change. The midlat-
itude atmosphere and ocean possess significant interannual variability and several large-scale
variability modes on decadal and interdecadal timescales. Physical origins of the LFV modes
remain unclear, and it is not even known to what extent these origins are intrinsic atmospheric,
intrinsic oceanic, or coupled oceanic-atmospheric. This project focuses on studying the intrinsic
oceanic LFV.

• Absorbing boundary conditions for nonlinear wave equations

Many problems in science and engineering are naturally formulated in unbounded domains; typ-
ical examples originate from fluid dynamics, solid mechanics, aerodynamics, electrodynamics,
acoustics, etc. However, numerical simulations of such problems require a finite computational
region. This project is aimed to design absorbing boundary conditions for efficient and robust
numerical simulations of nonlinear wave equations in unbounded domains.

• Stochastic parameterisations for ocean models

Stochastic parameterisations of oceanic eddies play an important role in geophysical fluid
dynamics because of their ability to represent complex physical processes with relatively simple
models. In this project we develop parameterisations for the quasi-geostrophic model of wind-
driven ocean gyres and analyse their efficiency in modelling unresolved scales.

• Multiscale oceanic energetics

The goal of this project is to study inter-scale energy transfers in the ocean, examine the
multi-scale nature of the forward and backward energy cascade, and how the energy transfers
depend on viscosity.

• Modelling the ocean with primitive equations

Modelling the ocean with primitive equations is a vast and active area of research in geophysics.
The goal of this project is to simulate and study ocean currents in the North Atlantic with
using the Regional Ocean Modelling System (ROMS).

• Bifurcation analysis of dynamical systems with degenerative solutions

In this project we consider convection in a porous material saturated with fluid and heated from
below. This problem belongs to the class of dynamical systems with nontrivial cosymmetry,
which gives rise to a hidden parameter in the system and continuous families of infinitely many
equilibria, and leads to non-trivial bifurcations. It is planned to study nonlinear phenomena
resulting from the existence of cosymmetry, describe different non-classical bifurcations, and
the selection scenarios (namely, which of infinitely many equilibria can be realized in physical
experiments).

1
Dr Philipp Thomas - Stochastic processes in biology

Web: https://www.ma.imperial.ac.uk/~pthomas

PROJECT 1: Hidden Markov models on trees to understand cancer cell proliferation

Supervisors: Fern Hughes ([email protected]), Dr Alexis Barr ([email protected]),

Dr Philipp Thomas ([email protected])

Modern microscopy experiments allow us to observe dividing cells and their offspring over time.
Such technologies promise to reveal crucial insights into diseases like cancer - a disease of
abnormal cell growth. To understand the biological mechanisms of cell proliferation and division,
we require stochastic models that can account for cell-to-cell variation observed in the data.

A simple model for this stochastic process is a Markov chain where each cell divides based on a
stochastic internal state. In this project, you will extract the hidden dynamics of these states from
single-cell experimental data of healthy and cancer cells acquired by our experimental
collaborators (Barr Lab @ Hammersmith Hospital campus) using Hidden Markov Models
(HMM). To this end, we will develop the common sequential HMM to a tree-structured HMM,
which accounts for the branching process nature of cell division. In particular, we aim to extend
the expectation maximisation algorithm to fit tree-structured data and the Viterbi algorithm to
extract the hidden dynamics. The developed methodology is expected to advance our
understanding of cancer cell proliferation and heterogeneity.

Prior knowledge of biology is not required. Experience with HMMs is beneficial but not essential.

Literature:
Murphy - Machine Learning: A Probabilistic Perspective (2012) - Book
(http://noiselab.ucsd.edu/ECE228/Murphy_Machine_Learning.pdf)

Bishop - Pattern Recognition and Machine Learning (2006) - Book

(https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-a
nd-Machine-Learning-2006.pdf)

Nakashima et al. 2020 - Lineage EM algorithm for inferring latent states from cellular lineage
trees (https://doi.org/10.1093/bioinformatics/btaa040)

Mohammadi et al. 2021 (preprint) - A lineage tree-based hidden Markov model to quantify
cellular heterogeneity and plasticity (https://doi.org/10.1101/2021.06.25.449922)
PROJECT 2: Age-structured models of cancer cell populations

Supervisors: Fern Hughes ([email protected]), Dr Alexis Barr ([email protected]),

Dr Philipp Thomas ([email protected])

Cells can enter a resting state called ‘quiescence’, where they temporarily stop cycling and
dividing. In cancer, we believe quiescence can allow cells to escape drug treatment, resulting in
tumour recurrence once they re-enter the cell cycle. In this project, you will combine stochastic
modelling with data from Dr. Alexis Barr’s lab at the LMS to understand how quiescence gives
cancer cells a proliferative advantage over healthy cells.

You will combine agent-based stochastic models with data from a variety of lung cancer cell
lines. For each cell line, you will use the growth curve data along with measured quiescent
fraction to inform an age structured model of a cell population. The aim is to understand how the
emergence of a quiescent subpopulation alters population growth and measured phenotypic
distributions and how they are altered in cancer. The theoretical predictions will be fitted to the
data to provide a mechanistic understanding of cancer cell proliferation and guide future
experiments.

Literature:
Thomas, Philipp - Making sense of snapshot data: ergodic principle for clonal cell populations.
(https://doi.org/10.1098/rsif.2017.0467)

Barr, Alexis R., et al. - DNA damage during S-phase mediates the proliferation-quiescence
decision in the subsequent G1 via p21 expression. (https://doi.org/10.1038/ncomms14728)

PROJECT 3: Stochastic population dynamics in time-dependent environments

Supervisor: Dr Paul Piho ([email protected]), Dr Philipp Thomas

([email protected])

Controlling the growth of cell populations is an important problem in biomedical applications. In

this project, we will investigate solutions to age-structured population dynamics with noisy
time-dependent inputs corresponding to effects of the environment.

We study and compare the effects of noise in age-dependent and age-independent population
growth dynamics both via analytical and simulation-based methods. In particular, we develop
methods for simple populations to quantify their asymptotic growth. Of particular interest will be
to distinguish features of individuals in growing from dying populations (super-/subcritical
processes).

Some basic knowledge of stochastic processes is advantageous but not necessary.

Literature:
Inaba, H (2017). Age-structured population dynamics in demography and epidemiology
(Chapter 1), Springer Book.

PROJECT 4: Approximate inference methods for live cell microscopy

Supervisor: Dr Barbara Bravi ([email protected]), Dr Philipp Thomas

([email protected])

Cellular reaction networks are inherently noisy. Understanding the sources of this variation is
important since it affects many essential cell functions. Modern time-lapse microscopy allows us
to answer this question but the wealth data it generates requires sophisticated statistical
inference tools to connect data with mechanistic models.

The chemical master equation models intracellular reactions as a continuous-time Markov

chain. While inference tools for Markov chains are well developed, they do not immediately
apply to stochastic reaction networks because the number of accessible states in these
networks is, at least in principle, unbounded. The idea of this project is to combine the
Baum-Welch algorithm (expectation maximization) for inference of Markov chains from time
series data with the Finite State Projection Algorithm (FSP) to find the unknown parameters of
the network. The FSP algorithm is a finite-dimensional approximation of the master equation
which also provides error estimates and convergence guarantees. You will probe the efficacy of
the methodology using cutting-edge live cell microscopy data of bacteria.

Literature:
Wikipedia - Baum-Welch Algorithm
(https://en.wikipedia.org/wiki/Baum%E2%80%93Welch_algorithm)

Munsky et al (2006) The finite state projection algorithm for the solution of the chemical master
equation J. Chem. Phys. 124, 04410 (https://aip.scitation.org/doi/10.1063/1.2145882)

PROJECT 5: Bayesian inference of stochastic reaction networks

Supervisor: Dr Philipp Thomas ([email protected])

Biochemical reactions occur at random times in living cells. To reliably estimate the rates of
these reactions, we need to take into account this stochasticity observed in vivo. Bayesian
inference is often the preferred choice for this purpose, but it is usually expensive because it
requires sampling from the posterior distribution of parameters via MCMC. In this project, we will
explore an alternative route to the inverse problem. The goal will be to use an asymptotic
expansion of the stochastic process to approximate the posterior distribution directly. The results
are expected to provide analytical insights into how informative stochasticity is about the
parameters underlying the system under study.

Literature:
Fröhlich, F., Thomas, P., Kazeroonian, A., Theis, F. J., Grima, R., & Hasenauer, J. (2016).
Inference for stochastic chemical kinetics using moment equations and system size expansion.
PLoS Computational Biology, 12(7), e1005030.

Thomas, P., & Grima, R. (2015). Approximate probability distributions of the master equation.
Physical Review E, 92(1), 012120.
Phonons in an expanding or contracting BEC – Dr Ryan Barnett
In quantum mechanics, any particle can be classified either as a fermion or a boson. When a
collection of bosons is cooled to a sufficiently low temperature, an exotic state of matter called a
Bose-Einstein condensate (BEC) is formed. Since its realisation in 1995 (for a good pedagogical
review see [1]), BEC has proven to be a fertile ground for applications in quantum simulations. For
example, they can provide clean realisations of toy models used to understand disparate physical
systems.
The low-energy excitations in BECs are often called phonons and are similar (though sometimes
superficially) to vibrational modes of a crystal or sound waves in air. In this project, we will
investigate such collective excitations of a BEC that is expanding or contracting. Experiments along
these lines have recently been carried in [2] and have strong analogies with certain cosmological
models. We will attempt to understand the puzzles found in these experiments using quantum field
theory methods of many-particle systems.
[1] A. J. Leggett, Bose-Einstein condensation in the alkali gases: Some fundamental concepts, Rev.
Mod. Phys. 73, 307
[2] S. Banik, M. Gutierrez Galan, H. Sosa-Martinez, M. Anderson, S. Eckel, I. B. Spielman, G.
K. Campbell, Hubble Attenuation and Amplification in Expanding and Contracting Cold-Atom
Universes arXiv:2107.08097
Exotic states in rotating condensates – Dr Ryan Barnett
Diagonalising the Hamiltonian corresponding to a charged particle confined to a plane with a
constant perpendicular magnetic field is one of the canonical problems in quantum mechanics. The
resulting spectrum has a fascinating feature: there are (typically) large degeneracies associated with
each eigenenergy. Such collections of degenerate states are called Landau levels, and the one with
lowest energy is called the lowest Landau level (LLL). These large degeneracies play an essential
role in (fractional) quantum Hall systems [3].
Systems of bosons in the LLL have been experimentally explored much less, partially due to the
fact that most bosons used the lab are neutral (i.e. they carry no charge). For such systems, a
magnetic field (as experienced by charged particles) can be mimicked by mechanically rotating the
system. Recent experiments have reached the LLL regime using such rotation with the additional
technical trick of geometrical squeezing [4]. In these experiments a bosonic crystalline phase was
observed.
This project will aim to understand this experiment using the method of exact diagonalisation.
[3] R. B. Laughlin, Fractional quantization, Rev. Mod. Phys. 71, 863 (1999)
[4] Biswaroop Mukherjee, Airlia Shaffer, Parth B. Patel, Zhenjie Yan, Cedric C. Wilson, Valentin
Crépel, Richard J. Fletcher, Martin Zwierlein, Crystallization of Bosonic Quantum Hall States,
arXiv:2106.11300

1
Dynamics of Active Fluids: an exploration of the Toner-Tu equation

Supervisor: Thibault Bertrand

Description

Active systems take energy from their environment to transform it into motion. These systems are
driven far from equilibrium [1] and display a wealth of new phenomena forbidden by equilibrium ther-
modynamics, including the emergence of novel collective properties including large scale collective
motion [2], clustering [3], and self-jamming [4]. Studying active matter offers hope to uncover new
physics, shine light on complex biological processes and perspectives to develop functional materials
and smart devices. Complex and robust collective behaviors can be the result of interactions between
very simple constituent agents; finding a general framework to understand how active particles syn-
ergistically interact to perform a task is appealing and has many applications.
While the dynamics of conventional fluids is governed by the famous Navier-Stokes equation [5],
the dynamics of active fluids is well-described by the Toner-Tu equation [6,7,8,9,10]. This equation
was originally derived on the basis of symmetry considerations [6]. For the past two decades, sev-
eral studies have rederived hydrodynamic equations by systematically coarse-graining microscopic
models of active particles to finally end up with a Toner-Tu equation [10].

Plan

Using a combination of analytics and numerical simulations, we will study the emergent phases
stemming from the activity in the Toner-Tu equation. On the numerical side, we will develop various
methods to numerically solve the Toner-Tu equation and compare them [11]. These methods may
include: finite differences, finite elements and/or classical pseudo-spectral and spectral methods
which were hugely successful in classical hydrodynamics.

Prerequisites: Good coding skills are essential. Familiarity with partial differential equations and/or
fluid dynamics would is desirable.

References

[1] M. E. Cates. Reports on Progress in Physics, 75(4):042601, 2012.

[2] A. Bricard, J-B Caussin, N. Desreumaux, O. Dauchot, and D. Bartolo. Nature, 503(7474):9598,
2013.
[3] I. Buttinoni, J. Bialké, F. Kümmel, H. Löwen, C. Bechinger, and T. Speck. Phys. Rev. Lett.,
110:238301, 2013.
[4] S. Henkes, Y. Fily, and M. C. Marchetti. Phys. Rev. E, 84:040301, 2011.
[5] G. K. Batchelor. An Introduction to Fluid Dynamics. Cambridge University Press, 1967.
[6] J. Toner and Y. Tu. Phys. Rev. Lett. 75, 4326, 1995.
[7] J. Toner and Y. Tu. Phys. Rev. E 58, 4828, 1998.
[8] J. Toner, Y. Tu, S. Ramaswamy. Annals of Physics, 318, 170244, 2005.
 [9] S. Ramaswamy. Ann. Rev. Cond. Mat. Phys., 1, 323-345, 2010.
[10] M. C. Marchetti, J. F. Joanny, S. Ramaswamy, T. B. Liverpool, J. Prost, Madan Rao, and R. Aditi
Simha. Rev. Mod. Phys. 85, 1143, 2013.
[11] J. Dunkel, S. Heidenreich, K. Drescher, H. H. Wensink, M. Br, R. E. Goldstein. Phys. Rev. Lett.,
110, 228102 (2013)

2
Inertial Active Matter

Supervisor: Thibault Bertrand

Description

In recent years, a growing effort has been devoted to research in the field of Active Matter [1]. Active
matter generically describes the dynamics of large systems of agents that consume energy from
their surroundings to perform a function. Interactions between agents and with their environment
may have striking consequences, e.g. it can generate novel behaviour including self-organisation
and collective motion that are not observed in passive systems. These systems occur across the
natural world and understanding their structure is a fundamentally interdisciplinary field of research.
A classical example of collective motion is the flocking behaviour observed in birds, fish and bacteria
[2,3].
Agent-based modelling in which one stipulates the microscopic interactions between agents allows us
to investigate the link between agent-agent interactions and the emergence of particular macroscopic
behaviour, both computationally and analytically [4]. Of particular interest are minimal models which
reduce the description of agents to a few key features, providing the foundation for the development
of more specific and complex models. In the Active Brownian Particle (ABP) model, particles are
self-propelled with force Fp in a direction that diffuses in time with diffusivity DR . A system of ABPs is
commonly described by the following equations of motion for individual positions, ri , and orientations,
θi : X p
ζ ṙi (t) = F(ri , rj ) + Fp (cos θi , sin θi ), θ̇i (t) = 2DR ηi (t)
j6=i

where F is the agent-agent interaction and ηi (t) is a noise term. In deriving these equations one
assumes that the friction ζ is large, a fair assumption when modelling microscopic organisms such
as cells or bacteria. This is called the overdamped limit. In models where particles interact only via
purely-repulsive interactions F, agents can crowd and form dense clusters leading to the so-called
Motility Induced Phase Separation [5].
While a lot of effort has been devoted to understanding active matter in the overdamped limit, the
underdamped or inertial limit remains elusive [6]. Mathematically, this implies that inertia is not negli-
gible and may have striking consequences. For instance, it is thought that the presence of inertia can
hinder the mechanism that generates the MIPS phase separation. We often work in the overdamped
limit because it makes analytical progress more tractable. From a modelling standpoint, understand-
ing the role of inertia is necessary to relate these minimal models to larger scale physical systems
like animal flocking or robotic swarms.
Many questions remain unanswered in these underdamped systems: how can we quantify the effect
of inertia on the presence of MIPS? What are the differences in the physical properties of the clusters
formed with and without inertia? Is there a new collective behaviour that is present only when we
have inertia present in the system?

Plan

The student will develop interacting active particle simulations in the underdamped limit including
that of systems exhibiting MIPS that includes the effect of inertia. This will involve solving a 2nd order
stochastic differential equation. From here, the candidate would be expected to investigate the role
that inertia plays in perturbing the well-established results for the overdamped systems. A starting
test for the code would be for the candidate to recreate the results seen in the overdamped systems
before expanding.
While the project would be mainly computational, the candidate would also be expected to develop
an understanding of the analytical study undertaken on the overdamped versions of the model. We
believe that these questions are currently the subject of much attention and that this project provides
the candidate with a great opportunity to make a relevant contribution to the work being done in the
field.

Prerequisites: Solid coding skills in C/C++, matlab or python. A strong interest in mathematical
physics is essential.

References

[1] S. Ramaswamy, Annu. Rev. Condens. Matter Phys. 1:32345, 2010.

[2] T. Vicsek, A. Czirok, E. Ben-Jacob, I. Cohen and O. Shochet, Phys. Rev. Lett., 75:1226-1229,
1995.
[3] T. Vicsek and A. Zafeiris, Phys. Rep. 517, 71-140, 2012.
[4] F. Schweitzer, Brownian Agents and Active Particles, Springer-Verlag, Heidelberg, Germany,
2003.
[5] M. E. Cates and J. Tailleur, Annu. Rev. Condens. Matter Phys. 6, 219, 2015.
[6] H. Löwen, J. Chem. Phys. 152, 040901, 2020

2
Worm blobs and the dynamics of entangled active filaments

Supervisor: Thibault Bertrand

Description

Living systems at all scales aggregate in large numbers for a variety of functions including mating,
predation, and survival. The majority of such systems consist of unconnected individuals that col-
lectively flock, school, or swarm [1,2]. Already striking collective behaviors can emerge from simple
interactions between biological agents, but in some cases, the individuals can even be are physically
connected to each other, forming an additional class of entangled active matter systems with emer-
gent collective properties. In particular, recent experiments showed that aquatic worms such as the
California blackworm (Lumbriculus variegatus) can entangle their bodies into dense blobs to protect
themselves against environmental stresses [2].
From a theoretical point of view, these macroscopic worms can be thought of as active Brownian
polymers [3], whose length and the stiffness is dependent on the species under consideration. In
recent years, there has been huge research interest in the study of active polymers; existing models
usually describe these active polymers a beadstrings in which each bead (or monomer) is subject
to both thermal fluctuations but also an active force (or self-propulsion). Two strategies have been
considered to model this active forcing: (i) the active force on each monomer can be applied in a
random direction subject to rotational diffusion or (ii) the active force can be applied tangentially to
chain. These two different conventions have been shown to lead to very different structural properties.
The first kind of chain undergoes shrinkage followed by swelling of the structure with active forces [4],
whereas the tangentially driven chain has been shown to form spirals and undergo snake-like motion
transiently [5].
While the behavior of active polymer in porous media has been the focus of some recent work
with a recent study showing that stiffness of active filaments promote their efficient transport in two-
dimensional porous media [6,7], studies of dense assemblies of active polymers are crucially lacking
in the literature. We know that while flexible chains curl up and tend to get trapped when they form
spirals, stiff active polymers do not. In this project, we will investigate the collective dynamics of worm
blobs.

Plan

The project will try to answer a number of interesting questions. Our main aim is to understand
the dynamics of active polymers and focus on the competition between knotting/entanglement of the
filaments and the possible emergence of collective behavior in dense assemblies. To do so, we will
extend and analyse computer simulations by extending existing models [8].
After an initial review of the literature on the subject, you will develop numerical simulations which you
will compare to existing results. We will then extend the existing literature by exploring numerically a
variety of conditions. Our numerical results will be interpreted in the framework of existing theories
for active matter and transport in crowded environments.

Prerequisites: Good coding skills in the programming language of your choice is essential (recom-
mended programming languages: C/C++, matlab or python).
References

[1] T. Vicsek and A. Zafeiris. Physics Reports, 517(3?4):71?140, 8 2012.

[2] Y. Ozkan-Aydin, D. I. Goldman, and M. S. Bhamla. Proceedings of the National Academy of
Sciences, 118(6), 2021.
[3] R. G. Winkler and G. Gompper. The Journal of Chemical Physics, 153(4):040901, 2020.
[4] A. Kaiser, S. Babel, B. ten Hagen, C. von Ferber and H. Löwen. J. Chem. Phys. 142, 124905,
2015.
[5] R. E. Isele-Holder, J. Elgeti and G. Gompper. Soft Matter, 11, 7181-7190, 2015.
[6] B. Chakrabarti, C. Gaillard and D. Saintillan. Soft Matter, 16, 5534-5544, 2020.
[7] Z. Mokhtari and A. Zippelius. Phys. Rev. Lett., 123, 028001, 2019.
[8] C. Nguyen, Y. Ozkan-Aydin, H. Tuazon, D. I. Goldman, M. S. Bhamla, and O. Peleg. Frontiers
in Physics, 9:540, 2021.

2
3D Modelling of Cells and Tissues

Supervisor: Thibault Bertrand

Description

Understanding the mechanics and dynamics of cells and tissues is of vital importance to our un-
derstanding of a diverse array of biological processes, from morphogenesis to cancer progression.
Recent experimental advances have allowed for tissues to be imaged with cellular resolution and so
motivate cell-resolution modelling to describe and analyse the behaviour seen [1].
One branch of such modelling are vertex models. Vertex models are a popular, recently developed
class of 2D cell-resolution models that have successfully replicated several experimental findings
[2,3]. However, these models are limited in the types of tissues they can describe as most biological
tissues are inherently three dimensional. Recent work has sought to rectify this by extending a vertex
model to 3D for an inert tissue of non-motile cells [4]. However, cells are inherently motile and
capturing this activity is crucial for properly describing biological tissues. The aim of this project is
to develop a 3D model of a motile tissue and explore the diverse array of behaviour such a tissue
displays.

Plan

The student will first seek to implement a vertex based model in a similar manner to [4]. Following
this, we will develop a method for introducing activity to model a tissue of motile cells. This is more
complicated than in the 2D case as we now have to consider momentum conservation between cells,
as opposed to cells moving on a rigid substrate.
Beyond the implementation of an active vertex model in 3D, this project has a broad scope and the
model developed can be used to explore a variety of questions depending on the interests of the
student. Possible avenues include studying the effects cell motility has on the rigidity transition to
understand how cell aggregates fluidize, a process of vital importance in morphogenesis, or examin-
ing how single motile cells migrate through tissues, which is central to our understanding of cancer
metastasis and the immune response.

Prerequisites: Good coding skills in the programming language of your choice is essential (recom-
mended programming languages: C/C++, matlab or python). An interest in mathematical physics
and biomathematics is essential.

References

[1] M. R. Shaebani, A. Wysocki, R. G. Winkler, G. Gompper and H. Rieger, Nature Reviews Physics
2, 181-199 (2020)
[2] D. Bi, J. H. Lopez, J. M. Schwarz and M. L. Manning, Nature Physics 11, 1074-1079 (2015)
[3] D. Bi, X. Yang, M. C. Marchetti, and M. L. Manning, Physical Review X 6, 1-13 (2016)
[4] M. Merkel and M. L. Manning, New Journal of Physics 20, 022002 (2018)
Swelling and shrinking dynamics in a thermo-responsive hydrogel

Supervisor: Thibault Bertrand

Description

Swelling is a fundamental process in biology, engineering, and the earth sciences: tissues swell af-
ter injury, wooden structures swell with humidity, and dry soils swell after rainfall. Macroscopically,
swelling is the volumetric growth of a porous material due to the spontaneous imbibition of additional
pore fluid. Swelling is distinct from other growth processes because of the fundamental role of hydro-
dynamics: local expansion of the pore structure is coupled to the evolving fluid distribution, making
swelling inherently dynamic and poromechanical.
The mechanics of polymeric gels has attracted great interest over the years [1-5] and more recently
in the context of hydrogels [6,7]. A hydrogel is a cross-linked network of hydrophilic polymers satu-
rated with water. Hydrogels can experience extremely large and reversible changes in volume during
swelling, which can result in complex changes in shape and the development of surface patterns
[3,8,9,10]. Hydrogels have found a wide variety of practical applications; for example, they are widely
used for moisture absorption and in soft contact lenses. In biomedical engineering, they are used
for drug delivery, wound dressing, and as a scaffold for tissue engineering [11,12]. They have also
shown promise for use as sensors, actuators, and flow controllers, and as a model system in soft
granular matter [13].
We previously studied the swelling and drying of a sphere of hydrogel [14]. For that, we developed a
dynamic model based on large-deformation poromechanics and the theory of ideal elastomeric gels.
We used our model to study the complex internal dynamics of swelling and drying, and to highlight
the fundamentally transient nature of these strikingly different processes.
Thermo-responsive hydrogels are gels whose degree of swelling depends on the ambient tempera-
ture. They can become hydrophobic when heated above a certain temperature, which results in the
expulsion of much of the interstitial fluid and significant shrinking of the gel. These offer a route to
programmable material in which the shape may be tuned by control parameters.

Plan

The dynamics of swelling and shrinking due to temperature changes in thermo-responsive hydrogels
remains poorly understood. In this project, we will extend our poromechanical model for the swelling
of a spherical gel to include temperature dependent behavior [14]. First, we will consider the case of
a material in which temperature equilibration is very fast (i.e. the case where swelling and shrinking
is done at constant temperature). We will then explore the case which includes proper thermal
transport. To do so, we will use both analytics and numerics (including finite difference methods).
You will be responsible for developing new codes and extending existing codes.

Prerequisites: Good coding skills in the programming language of your choice is essential (rec-
ommended programming languages: C/C++, matlab or python). An interest in fluid mechanics is
essential.
References

[1] P. J. Flory and J. Rehner, Jr., J. Chem. Phys. 11, 521 (1943).
[2] M. Quesada-Perez, J. Alberto Maroto-Centeno, J. Forcada, and R. Hidalgo-Alvarez, Soft Matter
7, 10536 (2011).
[3] T. Tanaka, S.-T. Sun, Y. Hirokawa, S. Katayama, J. Kucera, Y. Hirose, and T. Amiya, Nature 325,
796 (1987).
[4] T. Tanaka, Phys. Rev. Lett. 40 (12), 820 (1978).
[5] T. Tanaka and D.J. Fillmore, The Journal of Chemical Physics 70 (3), 1214-1218 (1979).
[6] W. Hong, X. Zhao, J. Zhou, and Z. Suo, J. Mech. Phys. Solids 56, 1779 (2008).
[7] S. A. Chester and L. Anand, J. Mech. Phys. Solids 58, 1879 (2010).
[8] W. Barros, Jr., E. N. de Azevedo, and M. Engelsberg, Soft Matter 8, 8511 (2012).
[9] T. Tallinen, J. Y. Chung, F. Rousseau, N. Girard, J. Lefvre, and L. Mahadevan, Nat. Phys. 12,
588 (2016).
[10] R. Takahashi, Y. Ikura, D. R. King, T. Nonoyama, T. Nakajima, T. Kurokawa, H. Kuroda, Y.
Tonegawa, and J. P. Gong, Soft Matter 12, 5081 (2016).
[11] N. A. Peppas, Y. Huang, M. Torres-Lugo, J. H. Ward, and J. Zhang, Annu. Rev. Biomed. Eng.
2, 9 (2000).
[12] R. Langer, Nature 392, 5 (1998).
[13] C.W. MacMinn, E. R. Dufresne, and J. S.Wettlaufer, Phys. Rev. X 5, 011020 (2015).
[14] T. Bertrand, J. Peixinho, S. Mukhopadhyay, and C.W. MacMinn, Phys. Rev. Applied 6, 064010
(2016).

2
 MSc Projects 2021/22
Dr Martin Rasmussen ([email protected])
Project 1. Resilience in dynamical systems. Resilience describes how
much a dynamical system is away from a bifurcation point where the be-
haviour of the system changes fundamentally. Several concepts of resilience
have been described in the literature. This project builds upon the recently
developed notion of intensity, which has been explored for continuous time
in [1] and for discrete time in [2]. The approach studies pertubations of the
dynamics using a set-valued or control system. Intensity has the advantage
that it is a purely dynamical quantity and for this reason more meaningful
than other approaches. This project aims first at understanding how the
continuous-time theory is related to the discrete-time approach. This is of
particular importance when one uses discretisations. Moreover, the project
aims at extending the theory from determistic dynamical systems to random
dynamical systems. While for random systems with bounded noise, such an
extension seems straightforward (at least conceptually), it needs to be ex-
plored how these ideas extend to systems with unbounded noise and also to
certain questions for bounded noise systems.

[1] K.J. Meyer and R.P. McGehee, Intensity – a metric approach to quan-
tifying attractor robustness in ODEs, 2020, https://arxiv.org/pdf/
2012.10786.pdf.
[2] R.P. McGehee, Some metric properties of attractors with applications to
computer simulations of dynamical systems, 1988, http://www-users.
math.umn.edu/~mcgehee/publications/McGehee1988p/index.html.

Project 2. Rate-induced tipping in discrete-time dynamical sys-

tems. Tipping points describe bifurcations where the output of a dynamical
system changes disproportionately compared to the change in the parame-
ter. In [1], several mathematical mechanisms for tipping have been proposed,
including rate-induced tipping, where a system is pushed across a bifurca-
tion point through a nonautonomous change in the parameter that is non-
adiabatic, i.e. fast enough, so that it cannot be regarded as a constant pa-
rameter in the dynamics. This project aims at understanding rate-induced
tipping in the discrete-time context and builds on the results obtained by
PhD student Michael Hartl [2]. In contrast to the continuous-time case, the
discrete-time case is almost unexplored and more interesting, since one can
observed non-intuitive behaviour already in one dimension.

[1] P. Ashwin, S. Wieczorek, R. Vitolo, and P. Cox, Tipping points in open

systems: bifurcation, noise-induced and rate-dependent examples in the
climate system, Philosophical Transactions of the Royal Society A –
Mathematical, Physical, and Engineering Sciences 370(1962), 2012.

[2] M. Hartl, Non-autonomous Random Dynamical Systems: Stochastic

Approximation and Rate-Induced Tipping, 2019, https://spiral.imperial.
ac.uk/bitstream/10044/1/73914/1/Hartl-M-2019-PhD-Thesis.pdf.