Chemical questions. Computational thinking.
I explore and design computational methods to connect chemical reaction mechanisms, catalysis, and process behaviour across chemistry and chemical engineering.
I investigate how reacting systems evolve and use computation to test hypotheses against evidence—from competing pathways that shape product distributions to operating conditions that change reaction kinetics.
My aim is to identify what the evidence supports, where uncertainty remains, and which calculation or experiment could resolve it.
- 01Computer scienceRepresent & computeIoT engineering · algorithms & data structures
- 02Control science & engineeringResolve the dynamicsPhD · microscopic polymer modelling
- 03Chemical engineering & chemistryConnect scalesPolymerisation methods · multiscale catalysis
Methods find their way into chemistry.
Chemical systems
Research purposes
Computational methods
Research purpose
Industrial processes
Predict process behaviour and polymer distributions under changing conditions and at steady state.
Research purpose
Scientific exploration
Develop models to investigate how reaction mechanisms shape kinetics and product formation.
Further questions · Crossings ↗Plastic upcycling
Catalysis
Connecting atomistic chemistry with kinetic modelling.
Explore this researchIndustrial process prediction
Conventional polymerisation
Predicting polymer distributions in dynamic and steady-state processes.
- KMC & hybrid models
- Poisson sampling
- Steady-state Monte Carlo
- Parallel computing & chain data
- PINNs: batch & grade transitions
New process & product design
Photopolymerisation
Predicting how reaction conditions shape polymer microstructure.
Explore this researchBeyond chemical applications
Deep learning for 0–1 linear programmingTime-dependent KMCComputational methods & algorithms
How can we preserve the chemistry that matters,
while making its computation practical?
Keep the detail that explains the chemistry.
I represent individual chains or whole distributions, depending on what the question needs.
Hierarchical polymer-chain representations
- Challenge
- Microscopic chain features must remain accessible as reaction events change the population.
- Approach
- Organise chain information in hierarchical data structures, so simulation can retain relevant detail.
- In practice
- Polymer microstructure ↗
Continuous distribution models
- Challenge
- Tracking many discrete chain lengths leads to large population-balance systems.
- Approach
- Represent distributions as functions of chain length and time, reformulating the balance equations for neural-network solution.
- In practice
- Dynamic batch MWD ↗ · Polyolefin grade transitions ↗
Let that description evolve.
Reaction events and governing equations connect microscopic change with process behaviour.
Event-based & deterministic–stochastic models
- Challenge
- Molecular events and macroscopic balances describe different scales of the same process.
- Approach
- Use KMC for microscopic events and couple stochastic simulation with deterministic balances where appropriate.
- In practice
- Dynamic & photoinduced polymerisation ↗
Time-dependent kinetic Monte Carlo
- Challenge
- Reaction propensities can vary between events in a changing environment.
- Approach
- Develop algorithms that account for this time dependence in non-stationary reacting systems.
- In practice
- Non-stationary reacting systems
Learning with kinetic & conservation equations
- Challenge
- A flexible function approximator still needs to satisfy the process model.
- Approach
- With Shenhua Jiao and collaborators, constrain neural-network solutions using the governing equations.
- In practice
- Dynamic batch MWD ↗ · Polyolefin grade transitions ↗
Make that detail computationally practical.
I reduce repeated work, separate timescales, and organise computation without discarding the information of interest.
Superbasin acceleration
- Challenge
- Fast reaction cycles can dominate simulation while slower changes determine progress.
- Approach
- Treat fast cycles efficiently to reduce repeated event simulation.
- In practice
- Photoiniferter RAFT ↗
Poisson-based fast kinetic Monte Carlo
- Challenge
- Detailed batch-reactor simulation can require very many reaction events.
- Approach
- Use Poisson sampling to accelerate the simulation of polymerisation.
- In practice
- Dynamic batch processes ↗
Multi-step steady-state Monte Carlo
- Challenge
- Reaching the steady state through the full start-up transient can be expensive.
- Approach
- Combine deterministic and stochastic modelling to obtain steady-state polymer distributions efficiently.
- In practice
- Steady-state processes ↗
Adaptive sub-box parallelisation
- Challenge
- Detailed dynamic Monte Carlo calculations need an efficient division of computational work.
- Approach
- Partition simulations into computational sub-boxes to make use of parallel execution.
- In practice
- Dynamic polymerisation ↗
Curriculum learning
- Challenge
- Coupled physical constraints can make network training difficult.
- Approach
- Introduce constraints in stages, beginning with more tractable objectives before tackling strongly coupled ones.
- In practice
- Polyolefin grade transitions ↗
Applications
Turning plastic waste into valuable fuels with scalable green catalysts
Another life for polymers.
With Prof. Michail Stamatakis at the University of Oxford, I work on multiscale modelling of catalytic plastic upcycling. MAP4PWV connects molecular-level chemistry with kinetic modelling to investigate polymer hydrocracking on bifunctional catalysts.
Project on CORDIS ↗Project support
MAP4PWV is funded by the European Union under the Marie Skłodowska-Curie Actions, grant agreement No. 101271961.
Industrial process prediction
Conventional polymerisation
Predict polymer distributions as a process evolves—and at steady state.
Batch polymerisation · Industrial slurry polyethylene production
Dynamic processes
How does the polymer population change over time?
KMC method development
Faster microscopic simulation
I developed adaptive sub-box parallel acceleration and Poisson-based fast KMC to predict evolving polymer populations with less computational work.
AI method development · PINNs
Learning with the laws of the process.
With Shenhua Jiao and collaborators, I use physics-informed neural networks (PINNs) to solve population balances and predict evolving molecular-weight distributions (MWDs).
Dynamic batch MWDsPolyolefin grade transitions
Steady-state processes
What distribution does sustained operation produce?
Reach the distribution without the full start-up
My multi-step steady-state Monte Carlo method combines deterministic and stochastic modelling to predict microscopic polymer distributions without simulating the full transient.
Related collaborative work derives analytical MWDs for steady-state radical polymerisation.
New process & product design
Photoinduced polymerisation
Connect reaction conditions with the polymer microstructure they produce.
Two light-driven reaction systems
Photo-ATRP
Compare process conditions through their effects on the polymer.
Predict microscopic properties as the reaction proceeds
For photoinduced atom-transfer radical polymerisation, I combine Monte Carlo simulation with deterministic–stochastic hybrid methods. Accelerated dynamic modelling makes comparisons of reaction conditions more practical.
Photoiniferter RAFT
Explore how chain-transfer agents shape polymer formation.
Resolve competing timescales
I model reversible addition–fragmentation chain-transfer polymerisation with multiple chain-transfer agents. Superbasin acceleration reduces repeated simulation of fast reaction cycles, enabling more efficient exploration of process and product choices.
Scientific software
PolymInsight
I am a co-author of PolymInsight, an open process simulation software for polymerisation with microstructural quality indices. It brings model definition, model reduction and adaptive solution methods into a Python-based platform.
The software supports dynamic batch and steady-state continuous stirred-tank reactor studies, connecting process simulation with molecular-weight-distribution prediction.
Read the software paper ↗Back to research areas ↑