PhysTensor

PhysTensor

A strictly-typed physical tensor library for deterministic scientific computing.

Overview

Scientific computing libraries excel at numerical operations but generally treat tensors as dimensionless arrays. This allows mathematically valid yet physically incorrect computations—for example, adding pressure to velocity or multiplying incompatible quantities without warning.

PhysTensor closes this gap by embedding physical dimensions and units directly into tensor operations. Every computation preserves dimensional information, enabling automatic validation of physical consistency while maintaining a familiar tensor programming experience.

Rather than replacing existing numerical backends, PhysTensor acts as a physics-aware layer that integrates dimensional analysis with modern tensor computation.

Why PhysTensor?

PhysTensor is designed around one principle:

A computation should be mathematically valid and physically meaningful.

The library combines:

• Strictly typed tensor objects
• Physical units and dimensions
• Automatic dimensional analysis
• Deterministic execution
• Backend-agnostic tensor computation
• Engineering-oriented validation

Features

Physics-Aware Tensors

Every tensor carries physical dimensional information alongside its numerical values.

Example quantities include:

• Length
• Time
• Mass
• Temperature
• Current
• Velocity
• Acceleration
• Force
• Pressure
• Energy
• Power

Automatic Dimensional Analysis

PhysTensor validates every operation.

Examples:

✅ Force × Distance → Energy

✅ Velocity × Time → Distance

❌ Pressure + Velocity

❌ Energy + Temperature

Invalid operations raise informative exceptions before incorrect results propagate through a scientific workflow.

Strict Typing

Designed for modern Python tooling.

• mypy
• pyright
• Ruff
• IDE autocompletion

Deterministic Scientific Computing

PhysTensor prioritizes reproducibility by ensuring computations produce consistent results across scientific and engineering workflows.

Backend Agnostic

PhysTensor is designed to integrate with existing numerical ecosystems rather than replace them.

Potential backends include:

• NumPy
• PyTorch
• JAX

Extensible Unit System

Define custom physical dimensions and derived quantities for specialized engineering domains.

Installation

pip install phystensor

For development:

git clone https://github.com/OWNER/phystensor.git

cd phystensor

pip install -e .[dev]

Quick Start

from phystensor import Tensor
from phystensor.units import Meter, Second

distance = Tensor([10], unit=Meter)
time = Tensor([2], unit=Second)

velocity = distance / time

print(velocity)

Output

5 m/s

Safety Example

pressure + velocity

Output

DimensionMismatchError:
Cannot add Pressure to Velocity.

Applications

PhysTensor is suitable for:

• Scientific computing
• Computational physics
• Physics-informed machine learning
• Digital twins
• Robotics
• Aerospace engineering
• Mechanical engineering
• Electrical engineering
• Renewable energy systems
• Battery modeling
• Industrial simulation

Architecture

                 User Application
                        │
                ┌───────▼────────┐
                │   PhysTensor   │
                ├────────────────┤
                │ Typed Tensors  │
                │ Unit System    │
                │ Validation     │
                │ Dimensional    │
                │ Analysis       │
                └───────┬────────┘
                        │
        ┌───────────────┼────────────────┐
        │               │                │
      NumPy          PyTorch          JAX

Roadmap

Core

• Typed physical tensors
• Automatic dimensional analysis
• Unit propagation

Future

• Sparse tensors
• GPU acceleration
• Automatic differentiation
• Symbolic dimensions
• PDE utilities
• Finite element helpers
• Physics-informed optimization

Testing

pytest

Static analysis:

ruff check .

black --check .

mypy .

Contributing

Contributions are welcome.

Please:

1. Fork the repository.
2. Create a feature branch.
3. Add tests.
4. Ensure all CI checks pass.
5. Submit a pull request.

Citation

If you use PhysTensor in research, please cite:

@software{phystensor,
  title  = {PhysTensor},
  author = {Asif Kazi},
  year   = {2026},
  url    = {https://github.com/ksaad20/phystensor}
}

License

Released under the Apache License 2.0.

See the LICENSE file for details.

Vision

PhysTensor aims to provide a reliable foundation for physics-aware tensor computing, enabling researchers and engineers to write software that respects both mathematical correctness and physical laws.

PhysTensor
Tensor computing with physical meaning.