Cross-Domain Scientific Discovery · Evolutionary Engine

inZOR-ND / inZORi

inZOR-ND is now the strongest public face of the project: a validated engine for hard scientific discovery problems, from active space selection and quantum hardware to nonlinear solvers, plasma disruption, and grid stability. inZORi is the broader evolutionary engine behind these cross-domain results.
20 published studies  ·  First public test: 14 February 2026  ·  Independent Research  ·  United Kingdom

What a new visitor should know

This project is not presented as a black-box AI claim. It is a published body of cross-domain evidence showing that the same evolutionary search logic can produce strong results on scientifically hard problems.

20
published studies currently listed
8/8
benchmarks won in active space validation
7/7
IBM hardware QEC runs won
1.59×
BAWS-NR speedup across 6 domains

What this project is claiming

Current flagship message

inZOR-ND now anchors the public scientific story: active space selection validated on 8 benchmarks, supported by equally strong public results in IBM quantum hardware, universal Newton-Raphson acceleration, plasma disruption law discovery, and critical-threshold grid stability.

1. What the project is today

inZORi is the general evolutionary engine. inZOR-ND is one of its strongest public scientific validations: automatic active space selection for CASSCF / SA-CASSCF, benchmarked across dissociation curves, transition-metal chemistry, excited states, and large combinatorial spaces.

More broadly, the same engine has already been used in published studies across:

2. Why this matters

Hard search spaces

The strongest public results are not toy examples. They involve large combinatorial spaces, multi-geometry consistency, difficult convergence regimes, or real hardware constraints.

Useful where heuristics weaken

The engine is especially relevant where fixed heuristics, handcrafted priors, or standard solver setups become brittle under the same protocol.

One engine, many domains

The public evidence is not confined to one niche. The same underlying search logic appears in chemistry, quantum hardware, solvers, plasma physics, and energy systems.

3. What has already been demonstrated publicly

Flagship study Published evidence Why it is important
inZOR-ND active space validation 8/8 benchmarks, 6 molecular systems, up to 430 kcal/mol advantage, only method converged on all systems under the tested workflow Shows the engine can solve a hard scientific selection problem where orbital choice is usually fragile and manually tuned
IBM hardware-native QEC 7/7 IBM hardware runs won, shallower circuits than Steane on Heron hardware Demonstrates hardware-aware discovery rather than only simulation success
BAWS-NR universal 1.59× mean speedup across 6 domains, 142,056 converged solves, convergence preserved Shows the same engine can improve nonlinear numerical workflows, not only scientific ranking problems
Fusion disruption law Cross-machine law derived from plasma current dynamics, validated on MAST, C-Mod, and HL-2A Suggests interpretable discovery of compact scientific laws, not just black-box prediction
PFΔ grid studies Critical threshold result at 99.9% vs 0% on a 1354-bus Pan-European grid benchmark Shows relevance in operationally meaningful regimes, not just under easy conditions

4. How it works at brochure level

The public explanation is intentionally simple:

  1. A domain defines the environment and evaluator. The problem can be a molecule, a circuit family, a nonlinear solve, a plasma indicator, or a grid scenario.
  2. A population-based evolutionary search explores candidate solutions. Candidates are evaluated under the domain-specific protocol rather than by a one-size-fits-all reward formula.
  3. Selection pressure keeps what remains viable. Better-performing candidates survive, recombine, and continue exploring the space.
  4. The same engine can be reused across domains. What changes is the evaluator and the problem framing, not the core search idea.

This page deliberately stays at brochure level. It explains enough to be credible while leaving implementation details to the published studies and reproducible test pages.

5. Where this engine is strongest

✓ Strong at

  • Combinatorial discovery problems where the useful structure is hard to write down in advance
  • Implicit or evaluator-defined objectives where success is best measured by simulation, convergence, or physical validity
  • Cross-domain reuse of the same search engine with different evaluators
  • Scientific problems where fixed heuristics degrade at the hard edge of the workflow

✗ Not the right fit for

  • Simple one-shot inference tasks where a direct model lookup is the real need
  • Problems with a single trivial optimum where evolutionary search adds no value
  • Ultra-low-latency settings where there is no room for iterative search
  • Claims of universal superiority on every benchmark class; that is not the project message

Research Tests & Empirical Validation

The project currently lists 20 published studies. The strongest public evidence is concentrated in flagship results that already answer the most important external questions: does it work, on what kinds of problems, and where is it better than conventional baselines?

View all research tests →
inZOR-ND Active Space8/8 benchmarks · 6 molecular systems · only method converged on all tested systems
IBM Hardware QEC7/7 hardware runs won · hardware-native circuits on Heron
BAWS-NR Universal1.59× speedup · 6 domains · 142,056 converged solves
Fusion Disruption LawCross-machine discovery from plasma current dynamics alone
PFΔ Phase 699.9% vs 0% at the critical threshold on a 1354-bus grid
ENTSO-E Real LoadValidation on Romania, Germany, and France real consumption profiles

6. Current stage

7. Contact

For feedback, collaboration, or technical questions:

Email: dumitru.novic@gmail.com

Researchers, technical teams, and partners can reuse or adapt this high-level presentation while keeping the published study pages as the evidence layer.

"inZORi doesn't simulate evolution — it implements it. inZOR-ND shows what that can already do on hard scientific problems."