Crate russell_lab

source ·
Expand description

Russell - Rust Scientific Library

russell_lab: Scientific laboratory for linear algebra and numerical mathematics

Important: This crate depends on external libraries (non-Rust). Thus, please check the Installation Instructions on the GitHub Repository.

§Introduction

This crate extends Rust with scientific computing capabilities, such as special mathematical functions (e.g., Bessel, beta, elliptic integral, erf, erfc, gamma), auxiliary mathematical functions (e.g., ramp, Heaviside, boxcar, logistic), numeric constants (e.g., SQRT_PI, EULER, NAPIER), numerical differentiation (e.g., to check analytical derivatives), functions to validate calculations, code performance measurement, least-square fitting, reading/writing table-formatted data, generation of grid coordinates for plotting, interpolation, and others. This crate also implements several functions to perform linear algebra computations (e.g., Matrix, Vector, Matrix-Vector, Eigen-decomposition, SVD, Inverse, and more).

The code shall be implemented in native Rust code as much as possible. However, thin interfaces (“wrappers”) are implemented for some of the best tools available in numerical mathematics, including OpenBLAS and Intel MKL.

The code is organized in modules:

  • algo – Structs and algorithms that roughly depend on the other modules
  • base – “Base” functionality to help other modules
  • check – Functions to assist in unit and integration testing
  • math (not re-exported) – Mathematical “special” functions and constants
  • matrix – Matrix struct and associated functions
  • matvec – Functions operating on matrices and vectors
  • vector – Vector struct and associated functions

§Linear algebra

For linear algebra, the main structures are NumVector and NumMatrix, that are generic Vector and Matrix structures. The Matrix data is stored as column-major. The Vector and Matrix are f64 and Complex64 aliases of NumVector and NumMatrix, respectively.

The figure below illustrates the base structures (struct) for linear algebra computations. The vector and matrix structures are generic but require the type parameter to implement a set of traits. The essential trait is the Num trait provided by the num-traits crate. This trait allows vectors and matrices in Russell to hold any numerical type. The Copy trait enables the possibility of cloning vectors and matrices directly; however, for efficiency, the BLAS functions can be called instead. DeserializedOwned and Serialize allow vectors and matrices to be serialized/deserialized, a feature that propagates for all other structures depending on NumVector and NumMatrix. This feature allows, for instance, writing and reading JSON files with simulation results. NumCast is required by NumVector because the linspace member function needs to cast the number of grid points to the final type (e.g., f64). NumMatrix requires the AddAssign and MulAssign traits to implement the convenience member functions named add and mul.

Vector-Matrix

NumVector has a single data member named data (a Vec<T>) holding all elements. Thus, NumVector wraps the native Rust vector struct. NumVector implements various methods, including functions for constructing and transforming vectors. One helpful method is linspace to generate a sequence of equally spaced numbers. NumVector implements the Index trait, allowing the access of elements using the [i] notation where i is the index.

NumMatrix stores all entries in the column-major representation. The main reason for using the column-major representation is to make the code work directly with Fortran routines, such as the BLAS/LAPACK functions. It is worth noting that BLAS/LAPACK have functions that accept row-major matrices. Nonetheless, these functions add an overhead due to temporary memory allocation and copies, including transposing matrices.

Thus, column-major data storage is essential. However, this requirement yields an undesirable side effect. Unfortunately, it is no longer possible to use the Index trait to access matrix elements using a notation such as a[i][j] as in other languages. Alternatively, russell_lab implements member functions get(i, j) and set(i, j, value) to access and update matrix elements.

NumMatrix also implements the AsArray2D trait, allowing matrices to be constructed from fixed-size nested lists of elements (stack-allocated), growable nested arrays of type Vec<Vec<T>> (heap-allocated) and slices that are views into an array. For example, matrices can be created as follows:

use russell_lab::{NumMatrix, StrError};

fn main() -> Result<(), StrError> {
    let a = NumMatrix::<i32>::from(&[
        [1, 2, 3], //
        [4, 5, 6], //
        [7, 8, 9], //
    ]);
    println!("{}", a);
    assert_eq!(
        format!("{}", a),
        "┌       ┐\n\
         │ 1 2 3 │\n\
         │ 4 5 6 │\n\
         │ 7 8 9 │\n\
         └       ┘"
    );
    assert_eq!(a.as_data(), &[1, 4, 7, 2, 5, 8, 3, 6, 9]);
    Ok(())
}

Finally, NumVector and NumMatrix implement the Display trait, allowing them to be pretty printed.

§Functionality

The linear algebra functions currently handle only (f64, i32) pairs, i.e., accessing the (double, int) C functions. We also consider (Complex64, i32) pairs.

There are many functions for linear algebra, such as (for Real and Complex types):

The russell_lab functions are higher-level than the BLAS/LAPACK counterparts, thus losing some of the generality of BLAS/LAPACK. Each BLAS/LAPACK function wrapped by russell_lab is carefully documented and thoroughly tested.

russell_lab implements functions organized in the vector, matvec, and matrix directories. These directories correspond to the BLAS terminology as Level 1, Level 2, and Level 3.

All vector functions are prefixed with vec_ and complex_vec_, whereas all matrix functions are prefixed with mat_ and complex_mat_. The matvec functions have varied names, albeit descriptive.

russell_lab implements several other interfaces to BLAS/LAPACK. However, some functions are complemented with native Rust code to provide a higher-level interface or improve the functionality. For instance, after calling the DGEEV function, mat_eigen post-processes the results via the dgeev_data function to extract the results from LAPACK’s compact representation, generating a more convenient interface to the user.

An example of added functionality is the mat_pseudo_inverse function, which computes the pseudo-inverse of a rectangular matrix using the singular-value decomposition. This function is based on the singular-value decomposition routine provided by LAPACK.

§Complex numbers

For convenience, this library re-exports:

  • num_complex::Complex64 – Needed for ComplexMatrix, ComplexVector and complex functions
  • num_complex::ComplexFloat – Needed for the intrinsic Complex64 operators such as abs() an others

When using the crate::cpx! macro, the Complex64 type must be imported as well. For example:

use russell_lab::{cpx, Complex64};

println!("{}", cpx!(1.0, 2.0));

§Examples

§(matrix) Eigen-decomposition of a small matrix

use russell_lab::{mat_eigen, Matrix, Vector};
use russell_lab::StrError;

fn main() -> Result<(), StrError> {
    let data = [
        [2.0, 0.0, 0.0],
        [0.0, 3.0, 4.0],
        [0.0, 4.0, 9.0],
    ];
    let mut a = Matrix::from(&data);
    let m = a.nrow();
    let mut l_real = Vector::new(m);
    let mut l_imag = Vector::new(m);
    let mut v_real = Matrix::new(m, m);
    let mut v_imag = Matrix::new(m, m);
    mat_eigen(&mut l_real, &mut l_imag, &mut v_real, &mut v_imag, &mut a)?;
    println!("eigenvalues =\n{}", l_real);
    println!("eigenvectors =\n{}", v_real);
    Ok(())
}

§(math) Bessel functions

use russell_lab::math;
use russell_lab::{Vector, StrError};

fn main() -> Result<(), StrError> {
    let xx = Vector::linspace(0.0, 15.0, 101)?;
    let j0 = xx.get_mapped(|x| math::bessel_j0(x));
    let j1 = xx.get_mapped(|x| math::bessel_j1(x));
    let j2 = xx.get_mapped(|x| math::bessel_jn(2, x));
    Ok(())
}

Re-exports§

Modules§

  • algo
    This module implements algorithms built from base, math, and vector-matrix routines
  • base
    This module implements a “base” functionality to help other modules
  • check
    This module contains functions to compare float numbers and arrays for unit testing
  • math
    This module implements mathematical (specialized) functions and constants
  • matrix
    This module contains functions for calculations with matrices
  • matvec
    This module contains functions for calculations with matrices and vectors
  • vector
    This module contains functions for calculations with vectors

Macros§

  • cpx
    Allocates a new Complex64 number

Traits§

  • ComplexFloat
    Generic trait for floating point complex numbers.

Type Aliases§

  • Complex64
    Defines complex numbers with f64-real and f64-imaginary parts
  • StrError
    Defines the error output as a static string