Industry

Simulations for industry

I build molecular models that help R&D teams decide what to make and what to test. In my doctorate, I worked with Kemira on barrier coatings for paper and board, and the same methods work in many other industries. For each industry below, you will find the questions I can help with, how I would model them, my closest past work, and a small model that you can run in the browser.

Contact me on LinkedIn
Doctorate with industry
At Aalto University (2021 to 2026), I worked on the simulations in coating projects with Kemira.
From atoms to films
All-atom MD, coarse-grained models, DPD, Monte Carlo, machine learning potentials, and continuum models. I pick the method that fits the scale of the problem.
Checked against experiments
I first test every model against data that already exists. My work is in eight papers, seven published and one under review.
Large simulation campaigns
I script my runs so that they restart on their own, and I run them on CSC supercomputers and university clusters with LAMMPS, GROMACS, HOOMD-blue, and OpenMM.

Industries where I can help

nine examples

  1. 01Paper, board, and packagingcoatings that keep water out
  2. 02Paints, coatings, and adhesiveshow films form and how well they stick
  3. 03Plastics, composites, and recyclingfillers and polymer interfaces
  4. 04Pharmaceuticals and drug deliveryhow drugs cross membranes
  5. 05Batteries and electrolyteshow ions move
  6. 06Semiconductors and electronicsself-assembly on wafers
  7. 07Carbon capture and gas separationsorbents for CO2
  8. 08Home and personal caresurfactants and micelles
  9. 09Metals and alloysdefects and machine learning potentials

The same methods also work for food emulsions, cosmetics, lubricants, construction chemicals, textiles, and membranes for water treatment.

01Paper, board, and packaging

Barrier coatings for paper and board

Packaging is moving from plastic films to thin bio-based coatings on paper and board. A good coating keeps water and grease out, covers the fibers without holes, and does not crack when the board is folded. This was the topic of my doctoral work.

Questions I can help with
  • Which polymer and block ratio give a closed, ordered film on cellulose or starch?
  • How much coating is enough, and when does the film crack or leave bare spots?
  • Why does one chemistry let water in after stretching while another does not?
How I would model it
I use DPD to follow how the coating forms, from the polymer solution to the dry film. Then I use all-atom MD to see how water gets into the hydrophobic layer, before and after stretching.
Closest to my work
Paper 1 (block length ratio), paper 2 (cracking and the composition window), paper 3 (water under strain), and my doctoral thesis. The two figures are small versions of the models in papers 1 and 2.

A live 2D version of the model in paper 1. A mixed layer of copolymers that has settled on the surface sorts itself into layers, with the hydrophilic beads on the substrate. Change the number of hydrophilic beads per chain and see how the film changes.

The composition window from paper 2. Click the map to set the amounts of copolymer and hydrophobic polymer. With too little copolymer, the coating forms islands, and with too much, the film cracks.

02Paints, coatings, and adhesives

How films form and how well they stick

Paints, primers, and adhesives have to wet the surface, form a film as they dry, and stay attached under load. A small change in chemistry can decide whether a joint breaks at the interface or inside the material.

Questions I can help with
  • Does the binder spread on this surface, or does it pull back and leave holes?
  • Does the joint break at the interface or inside the adhesive, and how can we move the weak point?
  • Which surface treatment or primer gives the strongest bond, and how much of it is needed?
How I would model it
All-atom MD gives adhesion energies and simulated pull-off tests at the interface. DPD shows how the film forms while it dries. I compare both with peel or pull-off measurements.
Closest to my work
In paper 6, we found how many grafted chains give the best adhesion between graphene and PLA, and that too many chains make the interface itself the weak point. Paper 4 is about silane monolayers, which are used as primers and adhesion promoters.

Based on paper 6. A graphene sheet is pulled off a polymer. With few grafted chains, the polymer tears and stays on the sheet. With many, the chains cover the sheet, and it comes off cleanly with less work. The chart shows the published energies, and the sketch is a drawing.

03Plastics, composites, and recycling

Fillers and polymer interfaces

Fillers like graphene, talc, glass fibers, or carbon black only improve a plastic when they are well spread out and the polymer sticks to them. In recycling, the same questions come up for blends of plastics that do not mix.

Questions I can help with
  • Will this filler spread out in our polymer, or will it clump?
  • Which surface treatment or coupling agent improves load transfer?
  • Which compatibilizer helps a recycled blend, and how much is needed?
How I would model it
All-atom MD at the filler surface gives adhesion, pull-out, and load transfer. DPD, with parameters from MD, shows dispersion and blend structure at larger scales.
Closest to my work
My M.Sc. work, papers 5 to 8. We designed surface treatments for graphene in PLA with simulations, and we studied shape memory composites with MD set up next to lab experiments.

04Pharmaceuticals and drug delivery

How drugs cross membranes

Before a molecule is made, it helps to know if it will cross cell membranes, dissolve well, and stay in its carrier. Free energy simulations can estimate this from the molecular structure.

Questions I can help with
  • Will this molecule cross a cell membrane, and where does it sit on the way?
  • How well does it dissolve in water compared with the formulation?
  • Will a micelle or polymer carrier hold the drug and release it at the right rate?
How I would model it
Umbrella sampling gives the free energy profile across a membrane, and the permeability follows from it. Alchemical free energy methods give solubility, and coarse-grained models describe the carriers.
Closest to my work
In project 31, I pulled ibuprofen through a membrane in GROMACS with 27 umbrella windows. In project 18, my hydration free energies matched FreeSolv within 0.17 kcal/mol. Project 1 predicts solubility and lipophilicity from structure.

05Batteries and electrolytes

How ions move in electrolytes

A good electrolyte lets ions move fast, is not too viscous, and stays stable at the electrodes. MD gives these transport properties and shows what the ions are doing.

Questions I can help with
  • How do conductivity and viscosity change with salt, solvent, and additives?
  • Do the ions form pairs, and how much current does that cost?
  • How do solvent and ions arrange at the electrode surface?
How I would model it
All-atom MD with scaled-charge or polarizable force fields, or with machine learning potentials. Conductivity and viscosity come from the Green-Kubo and Einstein relations.
Closest to my work
I have not worked on batteries yet, but these are the methods I use every day. In project 26, my viscosity from MD was within 6 % of NIST data. Project 27 automates diffusion runs on clusters, and project 10 predicts viscosity with calibrated error bars.

Ions in an electric field. Cations and anions drift in opposite directions. When the Coulomb attraction is stronger, they form pairs, and fewer free charges carry the current. This is one reason why conductivity stops rising at high salt concentration.

06Semiconductors and electronics

Self-assembly on wafers

Block copolymers can draw lines only a few nanometers wide on a wafer. Molecular monolayers are used as adhesion promoters, release layers, and masks for selective deposition. Both are self-assembly at a surface, which is what my research is about.

Questions I can help with
  • Which block copolymer and guide pattern give straight lines with few defects?
  • How does a monolayer pack on an oxide surface, and what does water do to it?
  • Will a resist or underfill stick to the surface after processing?
How I would model it
Field models like the one in the figure cover patterns over hundreds of nanometers. DPD and MD show the polymer near the guide and monolayers on oxide surfaces.
Closest to my work
Papers 1 and 2 are about block copolymer films that assemble on a surface. Paper 4 is about alkylsilane monolayers on silica, dry and in water, which I built with my own tool apadana-v4.

07Carbon capture and gas separation

Sorbents for CO2 capture

A good sorbent takes CO2 from flue gas or air, ignores most of the nitrogen and water, and gives the CO2 back without much energy. Simulations can screen thousands of porous materials before any of them is made.

Questions I can help with
  • How much CO2 does this material take up at process conditions, and how selectively?
  • How much of it does one pressure or temperature swing recover?
  • What does humidity do to the uptake?
How I would model it
Grand canonical Monte Carlo gives adsorption isotherms, and MD gives diffusion in the pores. Machine learning models trained on these results can screen large materials databases.
Closest to my work
Monte Carlo and free energy methods are part of my daily work (see the Monte Carlo page), and in project 34, I trained machine learning models on Materials Project data. A full adsorption study would be a new system for me, but with familiar methods.

08Home and personal care

Surfactants and micelles

Detergents, shampoos, and creams are crowded mixtures of surfactants, polymers, oils, and active ingredients. Coarse-grained models show which structures form and where the active ingredients end up.

Questions I can help with
  • Which micelles or phases form at this composition and temperature?
  • Will a fragrance or active stay in the micelles, and how fast does it leave?
  • How does a polymer change the structure, the viscosity, or what sticks to hair and skin?
How I would model it
DPD and Martini models with parameters from all-atom MD. From the same runs, I can map phases over composition and estimate release rates.
Closest to my work
In my doctoral work, I used DPD for amphiphilic copolymers in water. Project 13 derives DPD parameters from MD, and my Martini project builds a model for Pluronic, a polymer surfactant used in many formulations.

Micelles at work. Amphiphilic molecules gather into micelles and pick up oily molecules on the way, like a detergent with grease or a carrier with an active ingredient. With a weaker attraction between the tails, the micelles fall apart.

09Metals and alloys

Defects and machine learning potentials

In metals, simulations help to understand defects, alloying, and how a part starts to yield. Machine learning potentials now bring close to DFT accuracy to large systems. My first degree was in metallurgy.

Questions I can help with
  • How do strain, temperature, and alloying change where a crystal starts to yield?
  • Is a machine learning potential good enough for this property, or does it fail quietly?
  • How do coatings and oxide layers behave at the metal surface?
How I would model it
LAMMPS with EAM-type potentials for large systems, and machine learning potentials such as MACE or GAP, checked against DFT for the properties that matter.
Closest to my work
In project 30, I pulled and indented copper with an EAM potential and compared it with the MACE-MP-0 model. MACE-MP-0 gets the elastic constants right, but its stacking fault energy is about four times too low. In project 32, I fitted a GAP potential for silicon. My B.Sc. thesis was on nickel-alumina coatings on steel.

A crystal heated past melting. Atoms with six neighbors are in the crystal (blue), and the rest are defects or liquid (orange). Melting starts at the surface and at defects.

Which method fits your question?

pick two

Choosing the right model is half the work. Pick what you want to know and the size that matters. The box then shows the method I would start with, what it gives you, and my work that used it.

Fewer lab experiments

a toy example

Lab tests are accurate but slow and expensive. Simulations are cheap per recipe but less accurate. When we use both, the simulation scans the whole design space, and the lab tests only the most promising part.

If you set the simulation accuracy to zero, the advantage disappears. This is why I always start by checking the model against existing data.

How a project runs

five steps

  1. The question. What decision should the result support, and what would change it? A clear question saves months of work.
  2. The model. All-atom, coarse-grained, DPD, or continuum, depending on the length and time scales that matter. I check it against measurements that already exist.
  3. The simulations. A scripted and reproducible set of runs on HPC, so every number can be traced back to its input.
  4. The analysis. I turn the trajectories into numbers the lab can compare with, such as permeability, density profiles, adhesion energies, uptake, and order.
  5. The answer. Trends, mechanisms, and recipes worth testing, with the limits of the model stated clearly.

Working together

four ways

Limits of simulations

Simulations are best at trends, mechanisms, and ranking options. Absolute numbers need calibration against experiments, and a new system takes time to set up and check. The best results come from a few well-chosen experiments together with simulations. The models on this page are simple toy versions that show the idea. A real study uses calibrated models and much larger systems.

If you have a question like the ones above, the easiest way to reach me is LinkedIn.