ADDITIONAL DATA · PHYSICAL ORACLES · EXPLICIT SIMILARITY GROUPS
Explore the response, not just the structure.
Two complementary synthetic corpora, constructed separately from the active HolonForge/Aster and Mora-3 projects. Select an actual record, rotate its geometry, follow time-resolved fields, and inspect the circuit or Hamiltonian behind it.
Loading computed examples…
Drag to orbit; scroll to zoom; click a site to center a graph wavelet. The time-plane rotation is a projection with time as a fourth coordinate. Gaussian envelopes smear declared site values; they are not ab-initio electron-density isosurfaces.
From a Hamiltonian to circuit operations
Free-fermion Givens circuits have checked numerical recomposition and literal two-mode matchgates. Hubbard circuits include the Jordan–Wigner parity strings and a measured product-formula error. Paired BdG evolution requires a separate Bogoliubov compiler; a missing compiler is displayed explicitly.
PHASE · COHERENCE · TOPOLOGICAL TRANSPORT
Watch an actual quantum state evolve.
Color shows relative complex phase, height and sphere size show probability, and arrows show the computed probability current. Compare matched SSH and QWZ controls, explore a domain-wall zero mode, and select actual eigenstates from the spectrum.
ANALYTIC HYDROGEN · TRUE 3D VOLUME
Electron density clouds: motion, nodes and interference.
These are normalized isolated-hydrogen wavefunctions in atomic units. The 1s–2p superposition produces genuine quantum beats; the complex 2p orbital has phase winding +1 around its nodal axis. These are different phenomena from an SSH winding or a material Chern number. Density gained/lost is zero for a stationary eigenstate; choose the 1s–2p superposition to see quantum-beat redistribution.
Drag to orbit the volume. An isosurface is a chosen |ψ|² level; volume opacity is a display transfer function; colors show phase in the stated gauge. Finite display radius 8a₀ and exposure do not change the analytic normalization. No ab-initio molecular-density or relativistic/spin calculation is claimed.
COHERENT REAL-SPACE DENSITY · DIRECT TIME CONTROLS
Follow an electron density cloud through time.
H₂⁺ uses a fixed-nuclei, one-electron Coulomb Hamiltonian projected into six Gaussian orbitals. The Trp-cage example uses the dataset's explicitly uncalibrated contact Hamiltonian in an orthonormal Gaussian realization. These clouds include spatial coherence; they are not independent blobs scaled only by atomic occupations.
Source orbitals, Gaussian overlaps, Löwdin orthonormalization, full evolved occupied orbitals and particle-number checks are retained. The display uses bounded finite-grid/PNG approximation and temporal interpolation. The peptide cloud is a density of the stated effective model, not DFT, a measured protein charge density or a biological electron-transfer prediction.
TRAINED DIFFUSION · FLOW MATCHING · PERIODIC CHEMISTRY
Generate conformations on a circle and a torus.
Eight actual CPU training runs compare classical and two-qubit hybrid score/flow learners on fixed-geometry butane and pentane MMFF torsion grids. Follow their saved particle paths, rotate the corresponding bond-preserving molecular conformations, and compare the learned fields and endpoint distributions.
Flow transports a uniform torus prior toward the softened finite-grid Gibbs target. Diffusion starts from exact target-plus-noise samples and stops at positive heat time τ=.03; its reference includes that remaining noise. These endpoint laws differ. The earlier trained dessin Gibbs-state models provide the distinct quantum-physics benchmark; actual RG integrates out physical spins. A noise schedule alone is not RG. The trajectories show generative interpolation, not physical molecular time or a new nuclear dynamics simulation.
Fixed other internal coordinates, empirical MMFF94 at 300K, no solvent/full equilibrium, binding affinity or ab-initio density. Quantum circuits are classically simulated; no quantum advantage or improvement to either active model is inferred. Checkpoints, seeds, finite differences, independent circuit calculations and exact-score refinement controls are retained.
What is reused, and what is newly computed?
| Configuration | Rows | Origin |
|---|---|---|
| Aster default | 16,384 | New effective-material simulations |
| Aster source anchors | 576 | 48 existing MP20 training-source structures; newly computed responses |
| Mora default | 22,951 | New MMFF conformer/force simulations of 517 chemical parents |
| Mora PubChem anchors | 74 | 32 existing training-source identities; new checked geometries and forces |
| Mora public PDB anchors | 192 | 8 additional public structures; new uncalibrated contact-model variants |
Separate configurations add 174 Aster and 241 Mora electronic trajectories and 12 Mora nuclear trajectories. Frames do not inflate row counts. Every configuration records origin_class, planned_source_dataset, planned_source_record_id and label_origin. A new simulation does not imply a chemically novel identity absent from all previous training data.
What keeps the partitions separate?
Whole parent, conformer, gauge, intervention and task descendants stay together. Chemical and physical similarity edges are closed transitively before component-level stratified allocation. The resulting threshold-relative audit is not a guarantee against all possible leakage or overfitting.
Generation, curation, native reader checks and reproduction · Exact displayed examples and circuit records
These datasets do not replace either model’s existing corpora. They have not been enrolled into active training. Computed labels retain their method and evidence type; a trained performance improvement or the GRH arithmetic bridge has not been established.