Solving large systems of linear equations using sparse matrix methods.
let mut m = sparse21::Matrix::from_entries(vec![
(0, 0, 1.0),
(0, 1, 1.0),
(0, 2, 1.0),
(1, 1, 2.0),
(1, 2, 5.0),
(2, 0, 2.0),
(2, 1, 5.0),
(2, 2, -1.0),
]);
let soln = m.solve(vec![6.0, -4.0, 27.0]);
// => vec![5.0, 3.0, -2.0]
Sparse methods are primarily valuable for systems in which the number of non-zero entries is substantially less than the overall size of the matrix. Such situations are common in physical systems, including electronic circuit simulation. All elements of a sparse matrix are assumed to be zero-valued unless indicated otherwise.
Sparse21 exposes two primary data structures:
Matrix
represents anf64
-valued sparse matrixSystem
represents a system of linear equations of the formAx=b
, including aMatrix
(A) and right-hand-sideVec
(b).
Once matrices and systems have been created, their primary public method is solve
, which returns a (dense) Vec
solution-vector.
Sparse21 matrices can be constructed from a handful of data-sources
Matrix::new
creates an empty matrix, to which elements can be added via the add_element
and add_elements
methods.
let mut m = Matrix::new();
m.add_element(0, 0, 11.0);
m.add_element(7, 0, 22.0);
m.add_element(0, 7, 33.0);
m.add_element(7, 7, 44.0);
let mut m = Matrix::new();
m.add_elements(vec![(0, 0, 11.0),
(7, 0, 22.0),
(0, 7, 33.0),
(7, 7, 44.0)
]);
The arguments to add_element
are a row (usize
), column (usize
), and value (f64
).
Adding elements (plural) via add_elements
takes a vector of (usize, usize, f64)
tuples, representing the row, col, and val.
Unlike common mathematical notation, all locations in sparse21
matrices and vectors are zero-indexed.
Adding a non-zero at the "first" matrix element therefore implies calling add_element(0, 0, val)
.
Creating a Matrix
from data entries with Matrix::from_entries
:
let mut m = Matrix::from_entries(vec![
(2, 2, -1.0),
(2, 1, 5.0),
(2, 0, 2.0),
(1, 2, 5.0),
(1, 1, 2.0),
(0, 2, 1.0),
(0, 1, 1.0),
(0, 0, 1.0),
]);
The Matrix::identity
method returns a new identity matrix of size (n x n):
let mut m = Matrix::identity(3);
let soln = m.solve(vec![11.1, 30.3, 99.9]);
assert_eq!(soln, vec![11.1, 30.3, 99.9]);
Sparse21 matrices are built for solving equation-systems. The primary public method of a Matrix
is solve()
, which accepts a Vec
right-hand-side as its sole argument, and returns a solution Vec
of the same size.
You may have noticed all examples to date declare matrices as mut
, perhaps unnecessarily. This is on purpose. The Matrix::solve
method (un-rustily) modifies the matrix in-place. For larger matrices, the in-place modification saves orders of magnitude of memory, as well as time creating and destroying elements. While in-place self-modification falls out of line with the Rust ethos, it follows a long lineage of scientific computing tools for this and similar tasks.
So: in order to be solved, matrices must be declared mut
.
Sparse21 equation-systems can be loaded directly from a (custom-format) file including the elements and RHS.
let s = sparse21::System::from_file(Path::new("my_matrix.mat"))?;
let soln = s.solve();
It is often more convenient to split a sparse21::System
into its constituent matrix and RHS. The split
method returns a tuple of the two:
let s = sparse21::System::from_file(&p)?;
// Split into parts, and solve
// More or less equivalent to `s.solve()`
let (mut mat, rhs) = s.split();
let res = mat.solve(rhs);