The hypothesis is written down before the test
Including the conditions that would falsify it. We have killed our own headline claim this way more than once, and the replacement was better each time.
The Lab
Research discipline, engineering clock
Not a software vendor, not a generalist consultancy. A small group doing frontier work in compressible and reactive flow, and applying it to industrial problems while it is still frontier work.
The problem we exist to solve
Industry pays for the delay.
Trace how a genuinely better method reaches the engineer who needs it. It is conceived inside a large institution. It waits for a funding cycle. It is written up, reviewed and published two or three years later, implemented in a code assuming a national-laboratory machine. A decade on, some fraction appears in a commercial solver behind a per-core licence, with the assumptions that made it work stripped from the documentation.
Meanwhile the engineer with a hydrogen containment problem or a transonic buffet margin has a decision to make this quarter. They get the decade-old method, or a correlation.
None of that delay is scientific. It is structural. An independent lab can keep every part of the academic method that makes a result trustworthy, and drop the parts that only serve the institution.
How we work
Six practices we do not negotiate on
Speed without discipline is confident guessing. These are what let us move quickly and still stand behind the number.
Attribution, not anecdote
When something works, we find out which component is responsible — comparisons where exactly one thing changes. Slower than claiming the win, and the difference between a method and a coincidence.
Canonical verification gates every change
Order-of-accuracy studies, Riemann problems, steady detonation structure, ignition against an independent chemistry library, cross-checks against a reference code. Nothing lands on a case only we run.
Byte-identical reproducibility
Refactors must produce bit-for-bit identical results on adaptive reacting cases. It pays for itself the first time a number moves and you need to know why.
We never feed the answer to the solver
No force-refining a band where we expect the shock. No clipping a sensor where we have decided the flow is smooth. Those make the picture better and the prediction worthless.
Limitations are part of the deliverable
Every result states where it stops being trustworthy. A result that only survives selective quotation is not one — and you will find the edge eventually, better from us than from a design review.
The team
An unusual combination
Normally the person who derives a scheme from kinetic theory and the person who makes it saturate a GPU sit in different groups, trading artefacts across a wall. The model gets simplified to fit the code, the code gets slow to fit the model, and nobody owns the gap.
Here the same people who derive the flux decide the memory layout it reads from. It is the single biggest reason a small lab can compete with a large one.
First-principles modelling
- Kinetic theory and lattice-Boltzmann formulations for compressible flow
- Detonation structure, stability and nonlinear dynamics
- Stiff chemical kinetics and operator coupling
- Non-equilibrium indicators and shock detection theory
- Conservative discretisation on unstructured polyhedra
Robust, efficient implementation
- Data-parallel algorithm design and memory-layout engineering
- GPU offload through standard C++, with no vendor kernel language
- Parallel tree algorithms and space-filling-curve partitioning
- Mesh generation, geometry queries and domain decomposition
- Reproducible build systems and verification infrastructure
Engagement
Three ways in
We are looking for industry collaborators, research sponsors and academic partners. Not software customers.
Industry collaboration
Bring us a flow that is not working.
A shock your solver smears, a detonation that will not stay stable, a case that takes a fortnight per run. We take it as a research problem with an engineering deadline, and you get the method, the evidence and the limitations — not a black box.
- Fixed-scope studies against your own reference data
- Method development where off-the-shelf closures fail
- Verification evidence and stated limitations, every time
Research sponsorship
Fund a frontier method, and get it first.
Sponsoring a line of work buys results as they happen, not a conference paper in two years. Sponsors shape problem selection, see the negative results too, and get the capability applied to their own configurations while it is being built.
- Named research lines with citable outputs
- Early access as capability matures
- Influence over what gets attacked first
Academic partnership
Rigour is the part we refuse to trade.
Hypotheses stated before testing, falsification conditions written down first, cross-code comparison, reproducible artefacts. We collaborate with university groups on benchmarks, shared validation and co-authored work.
- Co-authored publications and shared benchmarks
- Independent cross-validation against your solver
- Reproducible artefacts on every published result
Who you would be working with
A small lab, on purpose
Deliberately small. It keeps the person who answers your question the same person who derived the method, with no account layer in between.