A mathematical model can describe how CO2 spreads through a geological formation, how a battery cell ages, how stormwater finds its way through a city, or how turbulence behaves on the approach to an airport. But models do more than describe and predict. They are used to choose and to steer: which route the delivery van should take, which trains to prioritise when something has gone wrong, how to size an energy system, how to regulate a process while it is running. In practice, the difference between understanding a system and acting inside it comes down to which algorithms you have.
Computational science is also the foundation beneath artificial intelligence. Behind every model that learns lie numerical linear algebra, optimisation and programming close to the hardware – the same methods that make a simulation fast enough to be worth running. And the influence runs both ways: machine learning now gives us faster and more flexible ways to compute, in models that combine physics with data.
Our users are in industry, public administration and research. We build subsurface simulation tools used around the world to plan and assess CO2 storage, geothermal energy and petroleum production. We forecast turbulence at Norwegian airports, plan train services, optimise warehouse operations and distribution, and cut the time it takes to test a new battery. Some results are visible as research, others as software in daily use: every day, around 12,000 hauliers across the Nordics run routes computed with technology developed here.
Computational science is also a discipline that carries other disciplines. Acoustics, sensor technology, robotics, health technology and quantum technology all rest on models, discretisation and algorithms. Much of our work is therefore done alongside groups who know the physics, the biology or the logistics better than we do, and where our contribution is what makes the problem computable in the first place.
Our research includes
- Physics-based simulation of processes and systems, to predict behaviour, reduce risk and optimise performance
- Discretisation in space and time, from finite volume and finite element to isogeometric analysis, with the grids and time integration that go with them
- Coupled problems and multiphysics, where flow, mechanics, heat, chemistry and electrochemistry act on each other at once
- Linear and nonlinear solvers that keep up as models grow
- Multiscale and reduced-order methods, which make large models computable without losing what matters
- Discrete optimisation for planning, routing, sequencing, staffing and timetabling
- Continuous and PDE-constrained optimisation, where the physical model itself is part of the optimisation problem
- Differentiable simulators, adjoint methods and automatic differentiation, which give not just an answer but also how that answer depends on each individual input
- Uncertainty quantification and data assimilation, so you know how much to trust an answer
- Geometry and spline technology for design, manufacturing and digital twins
- AI-driven models that combine machine learning with established computational methods
- Agentic AI that lets you talk to your simulator, connecting language models both to the simulation tools and to the work of building them
- Hardware-accelerated algorithms and high-performance computing, including GPU-based methods we have been developing since 2004
A method is not finished when it is published. It is finished when someone else can run it on their own problem. That is why so much of our work ends up as software: open source such as MRST, OPM Flow, Jutul, BattMo, IFEM, GoTools, GPU Ocean and Swim, and commercial tools delivered through industrial partners.
What can we help with?
Plenty of computational work is done perfectly well with tools that already exist. We come in when it is not – when the model is too large, the physics too tangled, the geometry too unusual, or the speed requirement too tight for a general-purpose tool to cope with. We can tell you whether a problem is computable at all, build a prototype that settles the question quickly, develop the method that is missing, or put it inside the software you already use. We are also glad to look at models other people have built. And what is bespoke here today is often standard equipment ten years from now. Computational science is the bridge between theory and practice. Software is what the bridge is made of.
