Recent notes
Research notes, prospectuses, and formalization roadmaps.
Environmental series — download the proof pack
A single PDF with all five environmental-series articles: cover pages, printable layout, and source URLs.
Download proof-pack-climate-series.pdf-
What the Cap Was Hiding
A correction gate that refused seven in ten of the cases it could act on turned out to be measuring its own estimator, not the physics — and the failure that exposed it was preserved rather than discarded.
-
Sharing DFT anchors reduces projected evaluations by 72.4%
A frozen, artifact-backed analysis of 29 analyzable diffusion paths compares four separately evaluated sparse-anchor sets with their shared union.
-
The Small Cell Held: A 16-Element MLIP Elastic Benchmark
A preserved 16-element comparison of 1×1×1 and 3×3×3 MatPES elastic calculations, with the correction failure reported alongside the result that survived
-
The Union Verdict: A De-brief on the Z1 Campaign
The full account you asked for before we talk — the verdict, the economics, the failures, the catches, and where I'd push back
-
The Savings Stack: A Booklet on Doing Less, and the Commons That Makes Everyone Faster
Seven savings layers from the literature, one measured sharing economy, and the theorem commons — as a free, designed booklet with a hashable data pack
-
An Order of Effort: Where Correction Laws Hold, and Where the Boundary Begins
The measured ladder from proven corrections to proven impossibility — and the order in which our next rounds climb it
-
The Z1 Barrier Panel: Evaluating Foundation MLIPs on 42 Solid-State Electrolyte Chemistries
One step in the trust-layer program: the locked Li-ion conductor panels behind our barrier campaigns — what they are, why they are the right adversaries, and what a migration barrier actually is
-
Four Gates, Three Honest Failures, One Live Experiment
A plain-language status report on our four preregistered Round-4 campaigns — what we measured, what failed, what is still running, and how you can check every number
-
Beyond Carbon: The Error Geometry of Environmental Materials
The same under-coordination error that corrupts climate-materials predictions also corrupts water, air, methane, refrigerant, mineral, PFAS, and cement discovery.
-
Cement, Concrete, and the Weight of the Built World
How corrected universal machine-learning interatomic potentials can accelerate low-CO₂ binders, alternative clinkers, and CO₂-cured concrete by taming amorphous and metastable phase chemistry.
-
Critical Minerals, PFAS, and the Remediation Imperative
Critical-mineral recovery and PFAS remediation share the same computational bottleneck — accurate binding and activation energies in under-coordinated environments — and the correction layer addresses both.
-
Methane and Refrigerants: Cutting the Non-CO₂ Climate Forcers
How corrected atomistic predictions can accelerate low-temperature methane conversion and the discovery of low-GWP refrigerants.
-
Water and Air: Correcting the Molecules Humans Drink and Breathe
How corrected uMLIPs recover accurate binding and barrier predictions for water desalination, atmospheric harvesting, lithium-selective extraction, and air-quality catalysts.
-
The Correction That Hurt, and the Theorem That Stopped It
A pre-registered, two-round trial of raw vs. Lupine-corrected uMLIPs on two of the five portfolio targets, with a LAMMPS classical leg and kernel-checked correction licenses
-
Lupi Gains HFC Refrigerant Research Payloads
Lupi now streams real hydrofluorocarbon refrigerant trajectories with full research payloads: charges, velocities, forces, thermo, and temperature profiles.
-
Non-CO₂ Climate Forcers, Now Formalized in Lean
Lupine Rhizo now formally accounts for methane, N₂O, HFCs, and SF₆ inside its Lean 4 climate library.
-
A Field, Not a Neural Net
The environment error field as a measured physical correction layer for universal machine-learning interatomic potentials.
-
Five Materials That Could Unlock 5–12 GtCO₂/Year
A portfolio-level view of five material classes targeted for gigatonne-scale climate impact.
-
From Predicted Crystal to Commercial Cell
How Lupine Science turns computational predictions into commercial climate materials through a sequenced partner chain.
-
Investing in the Trust Layer
The Lupine Science correction-and-verification layer as investable climate infrastructure
-
The 0.2% Synthesis Problem
Why most computationally predicted materials never reach a synthesis vessel, and why that matters for climate-critical materials.
-
The Trust Layer
The founding thesis of Lupine Science — verification as the missing infrastructure of AI-driven discovery
-
A smooth, environment-resolved error field underlies the systematic property errors of universal machine-learned interatomic potentials
Universal machine-learned interatomic potentials (uMLIPs) err systematically away from equilibrium; the test is whether those errors share a single, measurable shape.
-
The Order Is Right, the Size Is Wrong
Foundation machine-learned interatomic potentials (CHGNet, MACE-MP, MACE-MPA-0) measured across 21 materials and up to 9 properties against 228 published reference values
-
Why LUPI?
A browser-native molecular viewer for makeability
-
Why Lupine Science?
Trustworthy AI for materials discovery
-
From Fantasy Frameworks to Makeable Materials: A Prospectus for Formalized MOF Discovery
Metal–organic frameworks (MOFs), covalent organic frameworks (COFs), and other high-value reticular / complex molecular structures