Tensory

Tensory is a fast, minimal, and extensible tensor network computation library written in Rust.

Features

• High performance: Built with Rust's zero-cost abstractions and efficient memory management.
• Modular design: Easy to extend with custom tensor structures and backends.
• Rust idioms: Follows Rust best practices for safety and concurrency.
• Cross-platform: Supports Linux, macOS, and Windows.

Example

Add Tensory to your Cargo.toml:

[dependencies]
tensory-core = { git = "https://github.com/todo-group/tensory.git" }
tensory-basic = { git = "https://github.com/todo-group/tensory.git" }
tensory-linalg = { git = "https://github.com/todo-group/tensory.git" }
tensory-ndarray = { git = "https://github.com/todo-group/tensory.git" }
anyhow = "1.0.100"
openblas-src = { version = "0.10.13", features = ["system"] }
blas-src = { version = "0.14.0", features = ["openblas"] }

Then, use it in your code:

use tensory_basic::mapper::VecMapper;
use tensory_core::prelude::*;
use tensory_linalg::prelude::*;
use tensory_ndarray::{NdDenseTensor, NdDenseTensorExt};

extern crate blas_src;

type Tensor<'a, E> = NdDenseTensor<E, VecMapper<&'a str>>;

fn main() -> anyhow::Result<()> {
    let t = Tensor::<f64>::random(lm!["a"=>10, "b"=>15, "c"=>20, "d"=>25])?;
    let (u, s, v) = (&t).svd(ls![&"a", &"b"], "us", "sv")?.with(((),))?;
    Ok(())
}

Getting Help

Read Examples for sample codes, or see API Docs to know details.

Roadmap

• High-Dimentional-Diagonal Tensor
• Normalized tensor (for TRG)
• Sparse Tensor
• Lazy Contraction Tensor
• Allocation-aware interface
• GPU backend (CUDA, ROCm, ...)
• Grassmann Tensor
• Symbolic Function Tensor

License

This project is licensed under the UNDEFIND! License.

Acknowledgements

Inspired by:

• ITensor (C++)
• ndarray (Rust)
• grassmanntn (python)