Multiphysics and multi-scale, in one solver

One simulator for the whole problem.

One model carries one representation from electronic structure to bulk material, with heat, stress, current and light on the same mesh. Read energy per mode and the harmonics live, not after the solve. No system-level eigensolver in the loop.

Orbital scale runs today. All four scales are implemented; testing completes by end of 2026. Private alpha follows, design partners first.

Apply to be a design partner See the numbers ↓

Six months, up to 32 GPUs or CPUs. You run the compute.
Free for academia, $5k for startups and industry.

One quantum-dot device resolved at three scalesA fabricated semiconductor device on the right, a quantum dot magnified from its active layer in the middle, and the electronic structure inside that dot on the left. One mesh, refined further at each depth. Electronic structure Quantum dot Fabricated device
Figure 1One quantum-dot device at three scales: the fabricated device, the dot inside its active layer, and the electronic structure inside the dot. One mesh, refined where the physics needs it. Schematic, not a screen capture.

02 · The evidence

Local coupling recovers correlation without a global solve

Running today, orbital scale

Every point advances from its own state and its neighbors, with no system-level eigensolver anywhere in the loop. Correlated electronic structure is the most demanding case for that, so it is the one we measured.

H₂ dissociation, measured against the exact 4.7477 eV

Our mean field lands on the Hartree–Fock limit: deviating by 0.8 meV, which is within the ±2.4 meV margin of error. Spectral correlation then adds back 1.0193 eV of the 1.1118 eV no mean field method can reach, all of it through neighbor coupling.

Table 1Columns marked ours are Bayris runs, 15 September 2026. The Hartree–Fock limit and the expected bond are published values. Spectrally recovered quantities add to our mean field to give our total. All values in electron volts.
Molecule Eigensolver HF (eV)published Computed mean field (eV)ours Spectrally recovered (eV)ours Total bond (eV)ours Expected bond (eV)published
H₂ 3.6367 3.6359 +1.0193 4.6552 4.7477

Why this matters above the orbital scale

Correlated electronic structure is the least forgiving case there is: it is the place a local update rule is most likely to need a global solve. It did not need one. No node's state reaches another except through its neighbors. The same update rule runs the nanoscale and continuum scales, which are already implemented and will be fully tested by end of 2026.

Runtime cost follows how big your system is rather than its square or its cube, so a bigger model may mean a longer run on the same machine, as long as it can hold the additional storage. And because the representation is spectral, energy per mode and the harmonics are read straight off the state as the run proceeds, rather than rebuilt from a converged result afterward.

03 · How it works

The spectrum is the state

Most simulators make you wait for convergence before you can look at anything. Here energy per mode and the harmonics are readable mid-run, with no system-level eigensolver in the loop.

Read it while it runs

Pick any point at any step and the values are already there: energy per mode, the harmonics as they shift, and spectral entropy. You can inspect two-point correlation, such as quantum correlation, between any two points to see what is disordering and what is still coupled. At the orbital scale, you can watch a bond form rather than reconstruct it from a converged snapshot.

Example

One mesh element, the mode families it carries, and the spectrum inside one family One mesh element One tetrahedral element Four nodes. One scale tag. Scale tag · orbital carries The state it carries Mode families Photon Electric and magnetic field Spin-up and spin-down density Exchange–correlation … and the other families this scale tag selects One family, expanded Re Im phase DC frequency

Example

One mesh element, the mode families it carries, and the spectrum inside one family One mesh element One tetrahedral element Four nodes. One scale tag. Scale tag · orbital carries The state it carries Mode families Photon Electric and magnetic field Spin-up and spin-down density Exchange–correlation … and the other families this scale tag selects expanding one of them One family, expanded spin-up and spin-down density Re Im phase DC frequency
Figure 2One example mesh element with an associated scale. The tag can name any scale, and each mesh element selects its own mode families. Each mode family holds a complex spectrum, not a collapsed state. The points here show the shape such a spectrum can take, not a measurement.
Table 2The scale tag on an element decides which of these mode families it carries. The families listed under "Every scale" can exist in any scale tag. Molecular runs on its own topology, tuned for large molecules; a multi-scale model can span orbital → nanoscale → continuum.
Scale Modes the element can carry Status
Every scale Photon; electric and magnetic field; vector and scalar potential; temperature; phonon; torsion. Running today in orbital scale Other scales fully tested by end of 2026
Orbital and molecular Spin-up and spin-down density; current density; exchange–correlation, as the on-top pair density; orbital character (s, p, d); nuclear spin. Running today in orbital scale
Nanoscale Electron populations; plasmon; polariton; magnon; spin current; Berry curvature; strain; stress. Implemented · tested by end of 2026
Continuum Elastic and plastic deformation; fluid velocity; strain rate; vorticity; interface; strain; stress. Implemented · tested by end of 2026

04 · The peer group

Where the methods differ

Among the established methods, linear scaling and all-electron do not come together. Bayris does both at the orbital scale, and reads the spectrum while the run proceeds.

All-electron means the core states are solved on the mesh instead of being replaced by a pseudopotential, so the tightly bound electrons that set chemistry are computed rather than fitted. Linear-scaling methods give that up to reach O(N). Other all-electron methods keep it and pay with a cubic-scaling eigenproblem. In FEM and CFD the solver never sees an electron, their material properties arrive from a handbook.

Bayris keeps the cores at the orbital scale and still scales linearly, because nothing in the loop diagonalizes the whole system. The same property is why energy per mode and the harmonics are readable at any mesh point while the run proceeds, rather than reconstructed once it has converged.

Table 3Bayris against four established methods, across four capabilities. The Bayris row is scoped to what runs today.

Linear-scaling DFT here means ONETEP, CONQUEST and BigDFT. Orbital scale runs today; molecular, nanoscale and continuum are already implemented, tested by end of 2026.
Method Linear scaling All-electron One representation across scales Spatially resolved spectra
Bayris YesO(N) by construction;
neighbor coupling only		 Yesorbital scale resolves atomic cores on the mesh, no pseudopotential Yesone unified mesh, one coupling rule Nativeread live at any mesh point, at constant cost per point; orbital scale today
Plane-wave DFT Nocubic-scaling diagonalization, O(N³) Nofrozen core; PAW reconstructs valence against a fixed core Noelectronic structure only; a continuum region needs a second tool Reconstructedprojected densities of states, in a pass after the solve
Linear-scaling DFT Yesgapped systems Nopseudopotential valence basis; cores are not solved Noelectronic structure only Reconstructeda post-processing step, after the density kernel converges
LAPW all-electron Nocubic-scaling eigenproblem, O(N³) Yesmuffin-tin spheres plus interstitial Nomuffin-tin and interstitial partition only; no continuum region Reconstructedeigenstates projected after convergence
FEM / CFD Not applicableno electronic structure to scale Not applicablecontinuum fields only Nomaterial properties are supplied from outside the solver Reconstructedmodal analysis of an already-converged state

05 · The problem

When a question spans more than one scale

Plenty of questions sit inside one scale, and one tool answers them. When a question does not, you need a tool for each scale it touches, and you have to move the model between them yourself.

Take a metal contact that degrades in service. Answering that involves the chemistry at the interface, how the material itself behaves, the transport through the device, and the stress in the bulk around it. There is a mature, well-tested tool for each of those, and they do not exchange models directly. What one produces has to be converted before the next can read it.

Each of those conversions is manual: a hand-written translation, re-checked against every new output; the geometry re-meshed from scratch; parameters re-fitted to the new operating point. All three are redone whenever the geometry, the material or the theory changes.

Bayris does not automate those conversions. It removes the need for them: one representation carries the same state from electronic structure to bulk material, so neighboring scales couple through the spectrum at their mesh elements.

The toolchain today, and the one representation that replaces it The chain today is four tool boxes in sequence (quantum chemistry, molecular dynamics, device or nanoscale, FEM or CFD). The arrow between them is cut at three handoffs, labeled hand-written conversion, re-meshed geometry and fitted parameters, and each surviving length of it is thinner than the one before. In the replacement, orbital, nanoscale and continuum form one continuous bar of three touching segments, while molecular sits apart on its own row below and an arrow carries the material properties it generates up into nanoscale and continuum. Molecular is not a segment of that bar, because it runs on its own topology, tuned for large molecules; orbital runs today, nanoscale and continuum are already implemented; fully tested by end of 2026. A multi-tool chain Quantum chemistry for electronic structure Molecular dynamics for conformations Device / nanoscale for transport FEM / CFD for the bulk around them hand-written conversion re-meshed geometry fitted parameters Bayris · one representation Orbital Nanoscale Continuum Molecular supplies material properties generated before the run
The toolchain today, and the one representation that replaces it The chain today is four tool boxes in sequence (quantum chemistry, molecular dynamics, device or nanoscale, FEM or CFD). The arrow between them is cut at three handoffs, labeled hand-written conversion, re-meshed geometry and fitted parameters, and each surviving length of it is thinner than the one before. In the replacement, orbital, nanoscale and continuum form one continuous bar of three touching segments, while molecular sits apart below the end of that bar and an arrow carries the material properties it generates back into nanoscale and continuum. Molecular is not a segment of that bar, because it runs on its own topology, tuned for large molecules; orbital runs today, nanoscale and continuum are already implemented; fully tested by end of 2026. A multi-tool chain Quantum chemistry for electronic structure hand-written conversion Molecular dynamics for conformations re-meshed geometry Device / nanoscale for transport fitted parameters FEM / CFD for the bulk around them Bayris · one representation Orbital Nanoscale Continuum Molecular supplies material properties generated before the run
Figure 3Four tools, three crossings; the arrow thins at each one. One representation replaces them: orbital, nanoscale and continuum span one bar; molecular runs on its own topology, tuned for large molecules. Orbital runs today, nanoscale and continuum by end of 2026.

06 · Fit by field

Where this lands in your field

Orbital runs today. The larger scales consume what orbital/molecular scales generate.

Drug discovery and chemistry

Scales this field draws on

  • Orbital drawn on
  • Molecular drawn on
  • Nanoscale not drawn on
  • Continuum not drawn on

The blocker

  • To rank two novel ligands, you have to resolve about a kcal/mol. A molecular-dynamics force field covers only the chemistry someone already fitted it to, and DFT does not scale to a solvated protein.

What changes

  • The orbital and molecular scales carry dynamic charge transfer throughout the simulation, not charges fixed at setup.
  • Use the orbital scale where the cores matter: all-electron on one adaptive mesh. Cores resolved, not replaced.
  • Use the molecular scale for proteins too large for all-electron. If the mesh cannot resolve an atom’s charge shape, the run refuses rather than return a quietly wrong polarity.

Materials science

Scales this field draws on

  • Orbital drawn on
  • Molecular drawn on
  • Nanoscale drawn on
  • Continuum drawn on

The blocker

  • Your stiffness and conductivity came from someone else’s sample, someone else’s conditions.

What changes

  • Give it an SDF or XYZ. Out comes the material properties the larger scales need.
  • Under your conditions, saved and reused by your CAD.
  • Molecular scale handles systems past orbital’s reach, generating the material properties the larger scales consume.

Quantum devices and semiconductors

Scales this field draws on

  • Orbital sometimes drawn on
  • Molecular sometimes drawn on
  • Nanoscale drawn on
  • Continuum drawn on

The blocker

  • Decoherence limits you, not qubit count. Phonons set T1, unmodeled in device geometry.

What changes

  • One mesh from quantum region to substrate. No interface conditions.
  • Read which modes carry heat and loss, element by element.
  • Orbital scale where the device is most quantum, a dot or a junction. Molecular scale to derive material properties you do not have.

Aerospace and automotive

Scales this field draws on

  • Orbital not drawn on
  • Molecular sometimes drawn on
  • Nanoscale sometimes drawn on
  • Continuum drawn on

The blocker

  • A new alloy has no handbook entry, and the conditions you design for are the ones you cannot easily test.

What changes

  • The material properties your run consumes come from a molecular-scale run in the same system, not a handbook, and each carries the configuration that produced it.
  • Use the continuum scale on your CAD geometry, with thermal, structural and fluid on the same mesh.
  • Use the nanoscale where EM or MEMS behavior matters, on that same mesh.

UnifiedChemAI

In training

Graph transformer for molecular properties, binding sites and affinity. Treat its predictions as hypotheses and simulate the candidates they point to at the orbital or molecular scale.

No accuracy claim until training finishes.

07 · Material properties

Generate the material properties yourself

Implemented · tested by end of 2026

A continuum model needs stiffness, mobility, thermal conductivity. Today you hunt them down in handbooks, papers and databases: measured on someone else's sample, at someone else's temperature. You inherit conditions you did not pick.

Hand it a structure, get the data

An SDF or XYZ file is enough. The framework returns the full set of material properties the nanoscale and continuum scales need, at the conditions you set, and again whenever those conditions change. It is saved, so CAD models reference it and later runs reuse it. Nanoscale and continuum runs consume material properties that already exists; the simulator never invents it mid-run.

Regenerate any number you did not like

Every value carries the run, the input file and the conditions that produced it, and whether a solver computed it or a person typed it in. Re-run it under different conditions, or show a reviewer exactly where it came from.

08 · The deal

Get it before the alpha opens

We are taking a small number of design partners ahead of the private alpha. You bring one real problem and the reference data you would judge an answer by. You get the simulator first, six months of our time on your problem, results we co-publish, and design IP that stays yours.

Six months. Free for academia, $5k for startups and industry. Up to 32 GPUs or CPUs. You run the compute, at cost. One-year market-segment exclusivity is available from $2M.

Read the terms Apply now →

Bring us a system you care about

Tell us the problem and the reference data you would judge an answer by. We reply with a scoped proposal, or with the reason it is not a fit yet.

Apply to be a design partner