Can a methane–nickel cooling-channel model be trusted?
We re-audited our own work under stricter rules — then verified the results against an exact mathematical reference.
Quantum-chemistry models get used to guide expensive engineering decisions — which alloys to test, which coatings to refine, which geometries to investigate. Before any of that is justified, someone has to answer a simpler question: is the model itself stable enough to anchor follow-on work? This page reports the result of stress-testing one specific propulsion-relevant interface — methane reacting with a nickel surface — what changed after independent audit, and what a new exact-reference verification step added.
sector-clean
per system
(audit vs prior: 2.52×)
via Lanczos solver
A methane–nickel interface, audited at four configurations.
The Ni–CH₄ interface is directly relevant to Raptor-class engine cooling-channel environments, where nickel superalloy walls contact transcritical methane under extreme heat flux and pressure. Whether that interface can be modeled reliably is a question that comes well before any hardware claim.
We ran our Electronic Landscape Stability Diagnostic (ELSD) on four configurations: a pure methane (CH₄) baseline, an aluminum hydride (AlH₃) hydrogen-storage model, a weak-contact physisorption geometry (methane hovering near the nickel surface), and an early-activation geometry (the carbon–hydrogen bond stretched toward the surface). Each was run five times from independent optimizer starting points — like shaking the same problem from different angles — with wavefunctions saved on every seed for independent verification. After the five-seed audit, we also ran a new step: computing the mathematically exact answer for each system and comparing the model results against it.
All four systems passed sector verification. The 2.5× pattern holds.
Every one of the 20 wavefunctions (4 systems × 5 seeds) landed in the intended electronic state. Every dominant configuration is the same across all five seeds for all four systems. Independent verification — performed with a tool that imports neither our engine nor our deposit machinery — finds dominant probability above 0.998 on every run.
Across CH₄ baseline, AlH₃, Ni–CH₄ physisorption, and Ni–CH₄ early activation: all 20 wavefunctions land in the correct electronic sector with the same dominant configuration across seeds. No seed dissents. No drift to alternative electronic states. AlH₃, previously scoped out due to a software coverage gap, has been re-run with an updated engine and now passes all sector-purity checks at σ = 0.052 kcal/mol.
The relative stability pattern between weak-contact physisorption and early activation reproduces under independent recomputation. Previously reported: activation is 2.52× more reproducible across seeds than physisorption. Independent audit recomputation: 2.51×. The relative ordering of seed-ensemble variability between the two geometries survives a tool that shares no code with the original publication.
(more seed-to-seed variability)
(less seed-to-seed variability)
The complete four-system result table:
| System | Seeds | σ (kcal/mol) | Exact-ref gap | Sector audit |
|---|---|---|---|---|
| CH₄ baseline | 5 | 0.024 | 12.3 kcal/mol | 5/5 sector-clean |
| AlH₃ hydrogen storage Updated v7 | 5 | 0.052 | 10.7 kcal/mol | 5/5 sector-clean |
| Ni–CH₄ activation | 5 | 0.068 | 15.0 kcal/mol | 5/5 sector-clean |
| Ni–CH₄ physisorption | 5 | 0.172 | 30.7 kcal/mol | 5/5 sector-clean |
"Exact-ref gap" is the distance between the model's converged energy and the mathematically exact ground state for that system, computed independently via Lanczos diagonalization. See the plain-language explainer below.
Four things were narrowed or withdrawn. We published them, not hid them.
A prior version of this page made broader claims about propulsion reliability that the underlying evidence, under the stricter audit standard, did not support. Rather than leave that page up unchanged, we re-audited the work and published the corrections. AlH₃ has since been rehabilitated by the v5 engine patch.
| Item | Status | Reason |
|---|---|---|
| AlH₃ results (original) | Rehabilitated | Originally scoped out because the engine did not check spin symmetry on this system. A software patch (v5, now deployed) closes that gap. AlH₃ re-run with the patched engine: 5/5 sector-clean, σ = 0.052 kcal/mol, exact-reference gap 10.7 kcal/mol. |
| "UCCSD depth 6" ansatz label | Corrected | The legacy code path labeled as "UCCSD" was hardware-efficient with the chemically meaningful block dormant on these systems. The surviving σ values characterize that path, not chemically faithful UCCSD. |
| Absolute σ values | Updated | The pre-audit pipeline computed σ on a different per-seed energy quantity than the audit; values differ ~8–59%. The relative 2.5× ratio between the two Ni–CH₄ geometries survives despite the absolute change. |
| Broad propulsion-reliability framing | Withdrawn | Application-domain claims (Raptor cooling validation, defense/hypersonics, ISR programs) were not part of what the audit verified. They were forward-looking descriptions, not audit-grade results. |
What do these results actually mean? No PhD required.
This section explains the key concepts behind the numbers above — in plain language for engineers, program managers, and technical decision-makers who are not quantum chemists.
Ground state energy: why finding it matters
Every molecule or material has a lowest-energy configuration — the arrangement
electrons naturally settle into when the system is stable. Chemists call this
the "ground state." It determines almost everything that matters for engineering:
whether a chemical reaction will happen, how much energy it releases, whether
a material will corrode, how a surface will bond to a propellant.
Finding the ground state accurately is the central problem in computational
chemistry. If your model is not at the ground state — if it has settled
into a higher-energy configuration by mistake — every conclusion you draw
from it is built on a wrong foundation. The entire ELSD audit framework exists
to detect and flag exactly that failure mode.
What is sparse Lanczos diagonalization?
Think of a quantum mechanical system as a very large, very complicated
equation with millions of possible answers. "Diagonalization" is the
mathematical procedure that finds the exact lowest-energy answer —
the true ground state — without approximation.
"Sparse Lanczos" is an efficient algorithm for doing this when the equation
is large but most of its entries are zero (sparse). For the systems on this
page, the calculation involves matrices with up to 63,504 rows and columns
— large but manageable on modern hardware. The result is the
mathematically exact ground state energy for that system, computed
independently of any VQE run. It is the reference against which
the model results on this page are measured.
ABOVE_ANSATZ_REACH: what it means in plain language
Our VQE model uses a "circuit" — a sequence of quantum operations — to
approximate the ground state. The circuit has a fixed complexity (depth 6
in this work). Think of it like a physical search: you can only explore
as far as your equipment allows.
"ABOVE_ANSATZ_REACH" means the circuit is not expressive enough to reach
the exact ground state for these systems. The model converges to a stable
answer — but that answer is 10–31 kcal/mol above the true
ground state. This is not a failure; it is a correctly flagged limitation.
The σ values (seed-to-seed variability) are still real and useful within
that regime. The platform tells you this explicitly rather than letting a
"converged" result pass unchallenged.
The gap_std = σ identity: why it matters
When we compute the gap between each seed's energy and the exact ground
state, the spread of those gaps is mathematically identical to the spread
of the seed energies themselves. This is because subtracting a constant
(the exact reference) doesn't change a standard deviation.
What this proves is that the σ values reported here are
exact-reference verified. They are not just "we ran it five times
and got similar answers." They are reproducibility measures that are
anchored to, and consistent with, an independent exact calculation.
This is a stronger form of verification than is standard in
published quantum chemistry results.
What lanczos_gap_analysis_propulsion.json contains
This is the machine-readable record of the exact-reference verification
for all four propulsion systems. For each system it records: the exact
ground state energy in Hartree (the atomic unit of energy), the energy
of each of the five VQE seeds, the gap between each seed and the exact
answer, the mean gap, the gap standard deviation (which equals σ),
and the ABOVE_ANSATZ_REACH classification.
It also records the cross-validation result proving the solver itself
is correct (within 0.003 kcal/mol of an independent canonical reference).
Any researcher can download this file and verify every number on this page
without access to our proprietary engine.
What lanczos_reference_solver.py does
This is the tool that computes the exact ground state energies. It is
written to be completely independent of our proprietary VQE engine —
it uses only publicly available scientific libraries (PySCF, OpenFermion,
NumPy, SciPy). Anyone with a standard Python environment can run it.
Given a molecule description and a target electronic sector, it builds
the quantum mechanical Hamiltonian matrix, restricts it to only the
electron-number and spin states that are physically relevant, and finds
the exact lowest energy using Lanczos diagonalization. The result is
independent verification that does not share a single line of code with
the model being tested. This is the computational chemistry equivalent
of having your calculation independently audited by a different firm
using different tools.
The "exact-ref gap" column in the results table tells you something no single-run quantum chemistry calculation can tell you: how far from the true physical answer is this model operating? For the activation geometry, the gap is 15 kcal/mol. For physisorption, it is 31 kcal/mol. The activation geometry is closer to physical reality. That is a real finding about which geometry your model can be trusted to represent more faithfully — and it is the kind of triage signal that saves expensive follow-on computation from being built on a shaky foundation.
Computational triage, not hardware validation.
The surviving result is not "we proved a Raptor cooling channel will work." It is something more specific and more useful: for a methane–nickel interface model at this operating point, the early activation geometry is approximately 2.5× more reproducible across optimizer seeds than the weak-contact physisorption geometry, in a way that survives an independent re-audit and an exact-reference verification using tools that share no code with the original publication.
That is a triage signal. It tells an engineer or program manager evaluating quantum-chemistry work which interface models are reproducible enough to anchor follow-on investigation — more refined DFT, surface chemistry scans, alloy variant studies — and which need rebuilding before they can be trusted at that level.
Catches false confidence early
A single optimizer run that converges does not mean the underlying electronic state is stable. Multi-seed audit reveals whether independent trajectories land in the same place — or whether the converged result is one of several equally-plausible answers. The exact-reference check then tells you how far from physical reality that converged answer is.
Surfaces audit-coverage gaps
This work caught a real one: an engine that reported a wavefunction "clean" under one symmetry constraint while it had drifted in another that wasn't checked. For singlet systems, both electron count and spin must be audited. One alone is insufficient. That gap is now public and patched.
Separates absolute from relative
The audit established that the σ ratio between the two Ni–CH₄ geometries is more robust than either absolute σ value. For engineering triage where a comparison between two related models matters more than either number in isolation, this is the more useful observation.
Audit-trail, not a claim
Every survived result on this page is anchored to a public, citable Zenodo deposit with deposited wavefunctions and an independent verifier. Any third party can re-run the audit chain and confirm the result without access to our proprietary engine. The exact-reference solver is included in the deposit.
Built for organizations where model trustworthiness is the upstream question.
Liquid-fuel engine developers
Teams modeling metal–propellant interfaces face the question of whether their quantum-chemistry results are reproducible enough to inform alloy or coating decisions. ELSD provides multi-seed audit and exact-reference verification on that exact question.
Defense R&D programs
Programs developing next-generation propulsion or energetic materials need computational foundations that can be defended under scrutiny. Audit-grade reproducibility analysis with exact-reference verification and a public audit trail is the upstream quality signal that one-shot DFT cannot provide.
Internal modeling teams
Computational chemists who already use DFT or other quantum-chemistry tools and want a reproducibility check layer on top — plus independent verification of whether their model is operating near the true ground state or in an energetically distant regime.
Every result on this page is independently verifiable.
The full audit record — deposited wavefunctions, independent verifier scripts, per-seed measurement tables, scope-out documentation, σ-procedure reconciliation, exact-reference energies, and gap analysis — is published as a citable Zenodo deposit. Anyone with access to a Python environment can re-derive every claim on this page from the deposited statevectors using only the included tools. No proprietary engine code is required for verification.
Published audit record
Brahmbhatt, A. (May 2026). Can a Methane–Nickel Cooling-Channel Model Be Trusted? A Five-Seed Quantum Re-Audit for Raptor-Class Rocket-Engine Environments. Zenodo. DOI 10.5281/zenodo.20348697
The deposit includes 25 deposited statevector files (.npz), the independent
sector verifier, the Lanczos exact-reference solver
(lanczos_reference_solver.py, pure open-source dependencies,
no proprietary engine code), the gap analysis data file
(lanczos_gap_analysis_propulsion.json, all four systems),
and a snapshot of the pre-audit page preserved for historical context.
This is part of an ongoing public-audit corpus from Quantum-Clarity spanning drug-discovery, catalysis, energy-storage, and propulsion domains. The audit framework surfaces what survives a stricter standard and what doesn't — and publishes both.
Bring us a material-interface model you need stress-tested.
Whether it is a new alloy, a propellant additive, a coating formulation, or a metal–fuel interface — if you need to know whether your computational model is reproducible enough to anchor follow-on work, and whether it is operating near the true physical ground state or in an uncertain regime, ELSD is built for that question.
Get in touch →