Physics-Constrained Scientific Computing
Compute reality.
Data to AI.
The unified SciML stack. Phaethon bridges the gap between chaotic physical data and advanced AI through native-speed tabular pipelines, dimensional tensor mechanics, and physics-constrained deep learning.
Architecture
Built for the entire
scientific stack.
Five deeply integrated modules executing absolute physical integrity from data ingestion to differential calculus.
phaethon.units
Dimensional Tensor Algebra
Metaclass-driven physics engine across 90+ domains. Enforces mathematical integrity via Isomorphic Firewalls and Domain Locks. Features Isomorphic Firewalls, Semantic Domain Locks, real-time logarithmic scale evaluation, and deep NumPy protocol integration for zero-overhead vectorized tensor mechanics.
phaethon.Schema
Hybrid Data Engineering
Declarative pipelines for Pandas and Polars. Leverages a Rust backend for extreme-speed physical string parsing and C++ RapidFuzz for fuzzy ontologies. Executes vectorized imputation, anomaly clipping, and strict Axiom boundary enforcement.
[dataframe] or [polars]
phaethon.Dataset
Polyglot I/O Storage
A dimension-aware columnar store unifying discrete semantics and continuous physics tensors. Secures data via .phx cryptographic archives, with native PyArrow (.parquet) and HDF5 (.h5) integrations for scientific interoperability.
[h5], [arrow] or [io]
phaethon.ml + phaethon.pinns
Classical & Deep Sci-ML
An end-to-end SciML engine. Wraps Scikit-Learn with physics-aware meta-estimators and automated Buckingham Pi feature synthesis. Deep PyTorch integration unlocks physical autograd (PTensor), native PDE calculus, and Fourier Neural Operators.
[sklearn], [torch] or [sciml]
phaethon.units
Dimensional algebra
that thinks.
Intercepts complex mathematics at runtime. Isolates semantic anomalies, computes matrices, and tracks SI units to prevent silent physics failures.
Isomorphic Firewalls
Actively tracks Phantom Units to isolate mathematically identical but conceptually distinct dimensions (e.g., Frequency vs. Radioactivity).
Semantic Domain Locks
Rejects illegal casting between specialized domains (e.g., Energy vs. Torque) requiring explicit algebraic synthesis to proceed.
Real-Time Logarithmic Scaling
Natively evaluates non-linear arithmetic (e.g., Decibels, pH), implicitly linearizing values to prevent mathematical impossibilities.
phaethon.pinns & phaethon.ml
Neural networks that
obey physics.
Bridge the gap between data-driven Deep Learning and fundamental reality with dimension-aware operators and physics-informed losses.
Physics-Aware Autograd (PTensor)
Tensors retain dimensional identity through complex matrix operations, SVDs, and gradient descent inside PyTorch's computational graph.
Neural Operators & Calculus
Built-in Spectral Convolutions for FNOs and native differential calculus (grad, curl, laplace) to solve coupled Partial Differential Equations.
Axiom-Bounded Predictions
Scikit-Learn meta-estimators and Physics-Informed Loss Tribunals violently penalize networks that mathematically extrapolate into impossible bounds.
phaethon.Schema
Data pipelines
at machine speed.
Normalizes millions of irregular records with exceptional throughput by circumventing Python's standard string handling bottlenecks.
Rust-Powered Extraction
A dedicated Rust engine slices through chaotic, mixed-type physical text, bypassing Python overhead for extreme parsing speeds.
Vectorized Data Healing
Leverages the C/C++ backends of Pandas and Polars for zero-copy imputation, time-series interpolation, and anomaly clipping.
C++ RapidFuzz Ontologies
Utilizes high-speed Levenshtein distance matching to clean typographical errors and map real-world categories into strict ontologies.
Installation
Start with what
you need.
Phaethon is fully modular. Install only the engine components relevant to your scientific workload without bloating your environment.
Physics Engine
Dimensional Algebra
90+ domains, domain locks, native .phx storage, global contexts, and NumPy tensor wrappers. Zero extra dependencies.
pip install phaethonData Engineering
Schema + Tabular
Hybrid Rust/C++ pipeline engine for extreme-scale physical string parsing and validation.
pip install 'phaethon[dataframe]'pip install 'phaethon[polars]'Sci-ML Bridge
Scikit-Learn Integration
Buckingham Pi automated feature synthesis and strictly bounded classical estimators.
pip install 'phaethon[sklearn]'Deep Learning
PINNs + PyTorch
PTensor autograd, native physical calculus, and spectral layers for PDE solvers.
pip install 'phaethon[torch]'I/O Storage
HDF5 & Parquet
Adds external dependencies (h5py, pyarrow) to the columnar store for language interoperability.
pip install 'phaethon[io]'Enterprise Bundle
Full Sci-ML Stack
Installs the complete scientific ecosystem in a single, unified command.
pip install 'phaethon[sciml]'