Research · AI · Quantum Physics

Ideogenesis AI

Acquiring new physical insights via AI algorithms on ensembles
of data from quantum lattice simulations

  • Symmetry-Exact Tensors Abelian and non-Abelian quantum numbers conserved exactly across all tensor operations
  • Tensor Network Methods Ground state, finite temperature, and dynamical properties for quantum many-body systems
  • Quantum Lattice Models Spin chains, Hubbard models, and frustrated geometries spanning 1D to quasi-2D
  • AI-Driven Discovery Emergent physical correlations uncovered via data ensembles from quantum simulations

A New Paradigm for Physical Research

Ideogenesis AI is a research organization dedicated to bridging the gap between traditional theoretical physics and modern AI capabilities. We believe that breakthrough discoveries emerge not only from derivations, but also from observations — direct patterns and correlations extracted from numerical or experimental simulations.

Our work centers on strongly correlated quantum many-body systems. We build the computational infrastructure — tensor network libraries, symmetry engines, and simulation algorithms — that produces the high-quality data our AI framework learns from.

From Simulation to Discovery

The technology stack assembles from low-level recoupling primitives up to full tensor network algorithms — each component supports the next, culminating in data that drives the Ideogenesis AI framework.

  1. Yuzuha CG Engine

    SU(2) Clebsch–Gordan recoupling engine. Computes angular momentum coupling coefficients for symmetry-aware tensor operations.

    Python
  2. Nicole Tensor Library

    Symmetry-aware tensor library for many-body systems. Provides block-sparse tensor objects that preserve quantum numbers during tensor operations.

    Python
  3. Alice TN Algorithms

    1D tensor network methods built upon Nicole. Implements infrastructure and algorithms for ground state, finite-T and dynamical properties.

    Python
  4. Ideogenesis AI Framework

    Transformer-based architectures trained on quantum lattice simulation data. Designed to acquire new physical insights by uncovering emergent correlations and latent structure beyond the reach of conventional analysis.

    Private · Python

The Ideogenesis Framework

Built on top of the technology stack, the Ideogenesis Framework provides a comprehensive suite of transformer-based architectures and analysis tools for discovering emergent physics from data.

Transformer Core Architecture

State-of-the-art transformer architecture optimized for lattice system analysis, featuring novel attention mechanisms with locality biases and modular Processor/Propagator/Attention building blocks.

Analysis Diagnostic Tools

Advanced model diagnostics implementing attention propagation analysis, carrier transitions, Markov spectrum computation, and Omnimetry — a statistical framework for measuring physical observables.

Visualize Visualization Suite

Comprehensive visualization that transforms complex model outputs into intuitive representations, with TensorBoard integration and specialized attention visualization tools.

Strata CLI Resource management

A Homebrew-inspired resource management system that streamlines handling of datasets, models, and experimental artifacts — with centralized registry management, automatic versioning, and dependency resolution.

View org ↗

Pre-trained models and curated datasets available on Hugging Face — lattice systems, quantum simulations, and complex dynamical data.

Tensor Network Simulations

Each component embodies living, collaborative open-source efforts. Together, our projects cover the ground from symmetry bookkeeping to quantum simulation algorithms.

  • Nicole Tensor Library
    Python

    A symmetry-aware tensor library for many-body quantum systems. Implements block-sparse tensor objects that preserve quantum numbers during operations.

    Symmetry-AwareGPU / TPU / NPUAutogradML / AI Ecosystem
  • Alice TN Algorithms
    Python

    A collection of 1D tensor network algorithms built upon Nicole. Includes DMRG, (upcoming XTRG, tanTRG, TDVP, TaSK etc.) and MPS / MPO infrastructure.

    Tensor Networks1D AlgorithmsAutoMPSAutoMPO
  • Yuzuha CG Engine
    Python

    SU(2) recoupling engine for tensor network algorithms. Computes Clebsch–Gordan coefficients and Wigner symbols needed for symmetric tensor operations.

    SU(2)Clebsch–GordanWignerYuzuha Protocol

More repositories — including the private Ideogenesis Framework — live on the organization page.

Browse all repositories