Tensor — Multi-Physics Modeling for the Cislunar Mission

A cislunar mission runs on models nobody can afford to get wrong: the lattice that takes a lander’s shock, the power network that keeps a surface base alive through every outage, the deck whose modes must stay clear of the reaction wheels. Models that size usually go to a cluster, far from where the decision is made. Tensor runs them on one machine.

The unlock is a compact exact representation: a large grid or mesh problem replaced by a small exact object fixed at its boundary, with its decisions made by characters, divisors, cut-cycle coordinates, and prime-coded keys rather than by floating tolerances. One mathematics carries three offerings: lattice structures under shock, heat, and contact; power networks certified against every single-line outage; and modal clearance for flexible structures.

Lattice OS · Tensor Mesh

84.9 seconds vs 47.9 minutes

Same million-element lattice impact. Same equations, same mesh, same answers.

Time

MUMPS 5.8.2 with METIS2,872.1 s

Tensor84.9 s

Peak memory

MUMPS 5.8.2 with METIS53.4 GB

Tensor1.7 GB

34× faster · 31× less memory measured, one desktop

A periodic slab of body-centred-cubic lattice cells, 1,064,448 quadratic tetrahedra (6,438,528 unknowns), struck by a striker and followed for 512 implicit time steps. MUMPS with METIS ordering is the open-source standard among direct solvers. Both ran on the same desktop computer, one run at a time, 31 October to 3 November 2026; their answers agree to 4.1×10⁻¹⁰. Exact readings.

Three offerings — engagements are direct

Mesh

IMPACT · contact · heat · deform

Finite-element modeling of lattice structures — the cells that absorb a lander’s shock, isolate a payload, and ride out the swing between lunar day and night — under impact, contact, coupled heat, and large deformation, every cell resolved.

34× faster and 31× less memory than MUMPS with METIS on a million-element lattice impact, same equations and mesh. 1,089,994,752 elements in 1.8 GB on the same desktop.

[ engagement-direct ] View data sheet →

Grid

SIZE · certify · condition · survive every outage

Certified headroom for a new data center, generating project, or lunar base load: how many megawatts the network can reliably take across your demand range and every single-line outage, and the exact demand and outage that would take them away. You get the number with its proof, instead of re-solving your network once per outage and per demand case.

115 MW certified for a new load across all 1,998,000 single-line outages of a million-bus network, in a 794-byte certificate; each question answered 29,925× faster, in 99.1% less memory, than HiGHS re-solving warm.

[ engagement-direct ] View data sheet →

Spectral

COUNT · isolate · clear · certify

Modal analysis for flexible structures: which modes of a finite-element model fall inside the frequency bands that matter, counted exactly rather than estimated — for a payload deck that must stay clear of its reaction wheels and antenna mechanisms across every payload and mounting change.

545× faster and 22.5× less memory than MUMPS given the whole matrix, at 750,141 degrees of freedom, one core each, with every mode count certified and the cold time under a third of a second at every size.

[ engagement-direct ] View data sheet →

The common machine

The three offerings share one engine, the Holonomic Boundary Compiler. A model is answered from the small exact object its boundary fixes, not from its whole interior. On the lattice slab, memory held at 1.7 to 1.8 GB from two cells wide to 384 — a 36,864-fold range in elements — and a million-bus outage atlas fits in a certificate under a kilobyte.

[ direct engagement — vetted partners ] Engage by email →

Tell us the model that is too large, too slow, or too far from a cluster to run where you need the answer. A scoped, direct conversation follows — no tiers, no funnel.

tony@fancyland.net

— Tony Matos