Half a Billion Spins. 21 Seconds. One GPU.
•
Origin Neural AI
Configure a problem and watch the engine solve it in real time.
Live demonstrations proving engine capability across 8 real-world domains.
Classic 3-regular graph with known optimal cut of 12/15 edges. Verify the engine finds the exact solution.
The FMO light-harvesting complex transfers energy at 95%+ efficiency via quantum coherence. Find the optimal energy pathway.
Partition co-occurring cancer mutations to separate driver mutations from passenger noise using graph partitioning.
Color edges of the complete graph on 9 vertices avoiding monochromatic triangles and K4 cliques. A frontier combinatorics problem.
Find the shortest tour visiting 10 cities exactly once. The quintessential NP-hard routing problem solved via Ising encoding.
At the critical clause ratio α=4.267, satisfiability undergoes a sharp phase transition. Navigate this computational boundary.
Sherrington-Kirkpatrick model with dense random couplings. The prototypical NP-hard problem from statistical mechanics.
Markowitz mean-variance portfolio selection encoded as QUBO with cardinality constraints. Find the minimum-risk asset allocation.
Partition a drug-drug interaction graph to identify potentially dangerous combination clusters.
Optimize warehouse-to-store delivery routes across a clustered distribution network.
Partition a network graph to isolate vulnerable nodes from critical assets.
Optimize item placement to minimize picker travel distance using co-order frequency.
Optimize a simplified protein folding energy landscape on a lattice model.
Schedule 16 exams across time slots to minimize student conflicts.
Run a live benchmark from 1M to 500M spins on the RTX 6000 Ada (48GB VRAM). Watch the engine solve at industrial scale in real time.
500M only: single GPU solve — host generates the 22 GB sparse problem (~13s), GPU solves in ~22s. Reproduces the paper's headline claim.
Full sweep: 7 sequential scales (1M, 10M, 25M, 50M, 100M, 200M, 500M) — full Table 4 reproducibility. Only one runs at a time on the GPU.
Performance data from reproducible benchmark suites.
Run a live GPU benchmark from 1M to 500M spins on the RTX 6000 Ada (48GB VRAM). Tests 7 problem scales with timing, throughput, and VRAM measurement.
31 peer-reviewed papers. 500+ computational checks. Zero falsifications.
9 significant figures derived from basin arithmetic. The most precise theoretical prediction from a computational framework.
Exhaustive GPU campaign confirms R(5,5) = 43 via structural obstruction. 865 violations at the barrier.
Universal design limit validated against 59/59 biological systems. Predicts why nature chose quantum coherence.
PCAWG whole-genome sequencing data. 6 independent validations including ENCODE, TCGA, and multi-species.
Nine-step proof chain from Kato-Rellich perturbation theory through GUE statistics to Hadamard product formula.
Half a billion optimization variables solved on a single RTX 6000 Ada. 15.3 GB of 48 GB VRAM used.
Integrate the DSC-3 Isomorphic Engine into your applications. 26 authenticated endpoints across 5 categories.
API requests use X-API-Key header for core endpoints, or Authorization: Bearer for V1 Engine endpoints. Partner keys support both methods with full 500M access.
# Core endpoints (Python API)
curl -X POST https://engine.originneural.ai/solve/ising \
-H "X-API-Key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"n_spins": 100, "couplings": [{"i":0,"j":1,"value":1.0}]}'
# V1 Engine endpoints (Rust API, up to 500M)
curl -X POST https://engine.originneural.ai/v1/solve?sync=true \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"problem_type":"ising","n":10000,"density":0.05,"seed":42,"preset":"fast"}'
https://engine.originneural.ai
/solve/ising
Core
Solve an Ising model. Supports dense matrix or sparse couplings up to 500M spins. Returns optimal spin configuration, energy, and solver metadata.
// Request
{ "n_spins": 50, "couplings": [{"i":0,"j":1,"value":1.0}, ...],
"field": [0.1, -0.3, ...], "n_restarts": 10 }
// Response
{ "spins": [1,-1,1,...], "energy": -47.3, "solver_name": "SBM",
"elapsed_ms": 12.4, "n_restarts": 10,
"all_results": [{"solver_name":"SBM","energy":-47.3}, ...] }
/solve/qubo
Solve a QUBO (Quadratic Unconstrained Binary Optimization) problem. Input upper-triangular Q matrix. Returns binary assignment and objective value.
/solve/tsp
Solve Traveling Salesman Problem via Ising encoding. Input city coordinates, receive optimal tour and total distance.
/solve/ensemble
Recommended
Run all 16 solvers in parallel and return the best result. Includes per-solver timing, energy comparison, and winner selection. Best for production workloads.
// Request
{ "n_spins": 100, "couplings": [...], "n_restarts": 5 }
// Response
{ "best_energy": -89.7, "best_solver": "CVTR",
"all_results": [{"solver":"SBM","energy":-87.2,"ms":8.1}, ...],
"elapsed_ms": 45.3 }
/solve/upload
Upload a problem file (.json, .npy, .mtx). Accepts multipart form data with optional solver and restart parameters.
/verify/solution
Verify a claimed solution. Recomputes the energy from spins + couplings and confirms correctness.
/spectral/coherenceCompute spectral coherence metrics. Returns eigenvalue distribution, gap analysis, and coherence score.
/spectral/compressionInformation-theoretic compressibility of a coupling matrix. Returns entropy, rank, and compression ratio.
/spectral/cyclotomicCyclotomic field analysis — detect algebraic structure in spectral data via CM discriminants.
/spectral/lehmer-searchSearch for small Lehmer numbers near a target Mahler measure.
/spectral/moonshineMonster Moonshine connections for McKay-Thompson series coefficients.
/stagnation/analyzeAnalyze solver stagnation using Daugherty regime detection and barrier escape theory.
/tbo/analyzeTrace-based observable analysis — topological signatures from coupling spectra.
/quantum-graph/optimizeOptimize quantum graph scattering matrices via CMA-ES. Cayley graphs, Ihara zeta, bispectral analysis, Hilbert-Polya.
/quantum-graph/validateValidate quantum graph properties — unitarity, spectral gap, universality class, GUE statistics.
/topology/h2Persistent homology (H2 Betti numbers) of an Ising energy landscape. Vietoris-Rips persistence diagrams.
/topology/fingerprintTopological fingerprint — compact hash of the persistence diagram for classification and caching.
/crypto/hashSHA-256 hash with optional BSV anchoring. Returns hex digest and optional on-chain tx ID.
/crypto/zkp/keypairGenerate a Zero-Knowledge Proof keypair for authenticated computation verification.
/crypto/zkp/proveGenerate a ZKP proof that a solution satisfies constraints without revealing it.
/crypto/zkp/verifyVerify a ZKP proof. Returns boolean validity and verification metadata.
/crypto/shamir/splitShamir Secret Sharing — split a secret into N shares with threshold K for reconstruction.
/healthNo AuthHealth check. Returns engine status, GPU availability, and uptime.
/versionNo AuthReturns engine version, build info, and supported problem types.
/docsNo AuthInteractive Swagger/OpenAPI documentation. Try endpoints in your browser.
/v1/solve
500M Scale
Solve Ising, QUBO, MaxCut, SAT, TSP, or Ramsey problems at industrial scale. Supports server-side generation via density parameter — no need to upload matrix data for benchmarking. Async by default; add ?sync=true for synchronous results.
// Request: Generate & solve 10,000-spin Ising problem on-server
POST /v1/solve?sync=true
Authorization: Bearer YOUR_API_KEY
{
"problem_type": "ising",
"n": 10000,
"density": 0.05,
"seed": 42,
"preset": "fast"
}
// Response
{
"job_id": "3ed95200-...",
"result": {
"best_energy": -1124.79,
"best_solver": "Axiom-4",
"elapsed_secs": 0.008,
"all_results": [
{"solver_name":"Axiom-4", "energy":-1124.79, "elapsed_secs":0.008},
{"solver_name":"Lattice-1", "energy":-1055.93, "elapsed_secs":0.005},
...
]
}
}
// Submit async job for 100M spins
POST /v1/solve
Authorization: Bearer YOUR_API_KEY
{ "problem_type": "ising", "n": 100000000, "density": 0.001, "seed": 1, "preset": "fast", "gpu": true }
// Response: { "job_id": "da2742dd-...", "status": "pending" }
// Poll for result
GET /v1/jobs/da2742dd-...
Authorization: Bearer YOUR_API_KEY
/v1/info
Engine metadata: version, solver count, GPU availability, supported problem types, presets, and max_n (currently 500,000,000).
/v1/jobs/{id}
Poll async job status. Returns pending, running, or completed with full result payload.
/v1/mega-benchmark
GPU
Run full-scale benchmark from 1M to 500M spins on GPU. Tests all solver backends at industrial scale. Returns per-scale timing, energy, and throughput.
/v1/generate
Generate random problem instances (Ising, QUBO, MaxCut) with specified N, density, and seed. Returns the problem data for offline analysis.
/v1/stats
Runtime statistics: total jobs, active jobs, completed jobs, average solve time, GPU utilization.
/v1/health
Health check endpoint. Returns engine status, uptime, version, and GPU state.
n |
Problem size (1 to 500,000,000) |
density |
Edge density (0.0–1.0). When present with empty matrix, engine generates the problem server-side. |
seed |
RNG seed for reproducible generation (default: random) |
preset |
fast | production | quality — solver time budget |
gpu |
Force GPU solver (auto-enabled for N > 1M) |