1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
#[cfg(feature = "serde-serialize-no-std")]
use serde::{Deserialize, Serialize};

use crate::allocator::{Allocator, Reallocator};
use crate::base::{DefaultAllocator, Matrix, OMatrix, Scalar};
use crate::constraint::{SameNumberOfRows, ShapeConstraint};
use crate::dimension::{Dim, DimMin, DimMinimum};
use crate::storage::{Storage, StorageMut};
use simba::scalar::{ComplexField, Field};
use std::mem;

use crate::linalg::PermutationSequence;

/// LU decomposition with partial (row) pivoting.
#[cfg_attr(feature = "serde-serialize-no-std", derive(Serialize, Deserialize))]
#[cfg_attr(
    feature = "serde-serialize-no-std",
    serde(bound(serialize = "DefaultAllocator: Allocator<R, C> +
                           Allocator<DimMinimum<R, C>>,
         OMatrix<T, R, C>: Serialize,
         PermutationSequence<DimMinimum<R, C>>: Serialize"))
)]
#[cfg_attr(
    feature = "serde-serialize-no-std",
    serde(bound(deserialize = "DefaultAllocator: Allocator<R, C> +
                           Allocator<DimMinimum<R, C>>,
         OMatrix<T, R, C>: Deserialize<'de>,
         PermutationSequence<DimMinimum<R, C>>: Deserialize<'de>"))
)]
#[derive(Clone, Debug)]
pub struct LU<T: ComplexField, R: DimMin<C>, C: Dim>
where
    DefaultAllocator: Allocator<R, C> + Allocator<DimMinimum<R, C>>,
{
    lu: OMatrix<T, R, C>,
    p: PermutationSequence<DimMinimum<R, C>>,
}

impl<T: ComplexField, R: DimMin<C>, C: Dim> Copy for LU<T, R, C>
where
    DefaultAllocator: Allocator<R, C> + Allocator<DimMinimum<R, C>>,
    OMatrix<T, R, C>: Copy,
    PermutationSequence<DimMinimum<R, C>>: Copy,
{
}

/// Performs a LU decomposition to overwrite `out` with the inverse of `matrix`.
///
/// If `matrix` is not invertible, `false` is returned and `out` may contain invalid data.
pub fn try_invert_to<T: ComplexField, D: Dim, S>(
    mut matrix: OMatrix<T, D, D>,
    out: &mut Matrix<T, D, D, S>,
) -> bool
where
    S: StorageMut<T, D, D>,
    DefaultAllocator: Allocator<D, D>,
{
    assert!(
        matrix.is_square(),
        "LU inversion: unable to invert a rectangular matrix."
    );
    let dim = matrix.nrows();

    out.fill_with_identity();

    for i in 0..dim {
        let piv = matrix.view_range(i.., i).icamax() + i;
        let diag = matrix[(piv, i)].clone();

        if diag.is_zero() {
            return false;
        }

        if piv != i {
            out.swap_rows(i, piv);
            matrix.columns_range_mut(..i).swap_rows(i, piv);
            gauss_step_swap(&mut matrix, diag, i, piv);
        } else {
            gauss_step(&mut matrix, diag, i);
        }
    }

    let _ = matrix.solve_lower_triangular_with_diag_mut(out, T::one());
    matrix.solve_upper_triangular_mut(out)
}

impl<T: ComplexField, R: DimMin<C>, C: Dim> LU<T, R, C>
where
    DefaultAllocator: Allocator<R, C> + Allocator<DimMinimum<R, C>>,
{
    /// Computes the LU decomposition with partial (row) pivoting of `matrix`.
    pub fn new(mut matrix: OMatrix<T, R, C>) -> Self {
        let (nrows, ncols) = matrix.shape_generic();
        let min_nrows_ncols = nrows.min(ncols);

        let mut p = PermutationSequence::identity_generic(min_nrows_ncols);

        if min_nrows_ncols.value() == 0 {
            return LU { lu: matrix, p };
        }

        for i in 0..min_nrows_ncols.value() {
            let piv = matrix.view_range(i.., i).icamax() + i;
            let diag = matrix[(piv, i)].clone();

            if diag.is_zero() {
                // No non-zero entries on this column.
                continue;
            }

            if piv != i {
                p.append_permutation(i, piv);
                matrix.columns_range_mut(..i).swap_rows(i, piv);
                gauss_step_swap(&mut matrix, diag, i, piv);
            } else {
                gauss_step(&mut matrix, diag, i);
            }
        }

        LU { lu: matrix, p }
    }

    #[doc(hidden)]
    pub fn lu_internal(&self) -> &OMatrix<T, R, C> {
        &self.lu
    }

    /// The lower triangular matrix of this decomposition.
    #[inline]
    #[must_use]
    pub fn l(&self) -> OMatrix<T, R, DimMinimum<R, C>>
    where
        DefaultAllocator: Allocator<R, DimMinimum<R, C>>,
    {
        let (nrows, ncols) = self.lu.shape_generic();
        let mut m = self.lu.columns_generic(0, nrows.min(ncols)).into_owned();
        m.fill_upper_triangle(T::zero(), 1);
        m.fill_diagonal(T::one());
        m
    }

    /// The lower triangular matrix of this decomposition.
    fn l_unpack_with_p(
        self,
    ) -> (
        OMatrix<T, R, DimMinimum<R, C>>,
        PermutationSequence<DimMinimum<R, C>>,
    )
    where
        DefaultAllocator: Reallocator<T, R, C, R, DimMinimum<R, C>>,
    {
        let (nrows, ncols) = self.lu.shape_generic();
        let mut m = self.lu.resize_generic(nrows, nrows.min(ncols), T::zero());
        m.fill_upper_triangle(T::zero(), 1);
        m.fill_diagonal(T::one());
        (m, self.p)
    }

    /// The lower triangular matrix of this decomposition.
    #[inline]
    pub fn l_unpack(self) -> OMatrix<T, R, DimMinimum<R, C>>
    where
        DefaultAllocator: Reallocator<T, R, C, R, DimMinimum<R, C>>,
    {
        let (nrows, ncols) = self.lu.shape_generic();
        let mut m = self.lu.resize_generic(nrows, nrows.min(ncols), T::zero());
        m.fill_upper_triangle(T::zero(), 1);
        m.fill_diagonal(T::one());
        m
    }

    /// The upper triangular matrix of this decomposition.
    #[inline]
    #[must_use]
    pub fn u(&self) -> OMatrix<T, DimMinimum<R, C>, C>
    where
        DefaultAllocator: Allocator<DimMinimum<R, C>, C>,
    {
        let (nrows, ncols) = self.lu.shape_generic();
        self.lu.rows_generic(0, nrows.min(ncols)).upper_triangle()
    }

    /// The row permutations of this decomposition.
    #[inline]
    #[must_use]
    pub fn p(&self) -> &PermutationSequence<DimMinimum<R, C>> {
        &self.p
    }

    /// The row permutations and two triangular matrices of this decomposition: `(P, L, U)`.
    #[inline]
    pub fn unpack(
        self,
    ) -> (
        PermutationSequence<DimMinimum<R, C>>,
        OMatrix<T, R, DimMinimum<R, C>>,
        OMatrix<T, DimMinimum<R, C>, C>,
    )
    where
        DefaultAllocator: Allocator<R, DimMinimum<R, C>>
            + Allocator<DimMinimum<R, C>, C>
            + Reallocator<T, R, C, R, DimMinimum<R, C>>,
    {
        // Use reallocation for either l or u.
        let u = self.u();
        let (l, p) = self.l_unpack_with_p();

        (p, l, u)
    }
}

impl<T: ComplexField, D: DimMin<D, Output = D>> LU<T, D, D>
where
    DefaultAllocator: Allocator<D, D> + Allocator<D>,
{
    /// Solves the linear system `self * x = b`, where `x` is the unknown to be determined.
    ///
    /// Returns `None` if `self` is not invertible.
    #[must_use = "Did you mean to use solve_mut()?"]
    pub fn solve<R2: Dim, C2: Dim, S2>(
        &self,
        b: &Matrix<T, R2, C2, S2>,
    ) -> Option<OMatrix<T, R2, C2>>
    where
        S2: Storage<T, R2, C2>,
        ShapeConstraint: SameNumberOfRows<R2, D>,
        DefaultAllocator: Allocator<R2, C2>,
    {
        let mut res = b.clone_owned();
        if self.solve_mut(&mut res) {
            Some(res)
        } else {
            None
        }
    }

    /// Solves the linear system `self * x = b`, where `x` is the unknown to be determined.
    ///
    /// If the decomposed matrix is not invertible, this returns `false` and its input `b` may
    /// be overwritten with garbage.
    pub fn solve_mut<R2: Dim, C2: Dim, S2>(&self, b: &mut Matrix<T, R2, C2, S2>) -> bool
    where
        S2: StorageMut<T, R2, C2>,
        ShapeConstraint: SameNumberOfRows<R2, D>,
    {
        assert_eq!(
            self.lu.nrows(),
            b.nrows(),
            "LU solve matrix dimension mismatch."
        );
        assert!(
            self.lu.is_square(),
            "LU solve: unable to solve a non-square system."
        );

        self.p.permute_rows(b);
        let _ = self.lu.solve_lower_triangular_with_diag_mut(b, T::one());
        self.lu.solve_upper_triangular_mut(b)
    }

    /// Computes the inverse of the decomposed matrix.
    ///
    /// Returns `None` if the matrix is not invertible.
    #[must_use]
    pub fn try_inverse(&self) -> Option<OMatrix<T, D, D>> {
        assert!(
            self.lu.is_square(),
            "LU inverse: unable to compute the inverse of a non-square matrix."
        );

        let (nrows, ncols) = self.lu.shape_generic();
        let mut res = OMatrix::identity_generic(nrows, ncols);
        if self.try_inverse_to(&mut res) {
            Some(res)
        } else {
            None
        }
    }

    /// Computes the inverse of the decomposed matrix and outputs the result to `out`.
    ///
    /// If the decomposed matrix is not invertible, this returns `false` and `out` may be
    /// overwritten with garbage.
    pub fn try_inverse_to<S2: StorageMut<T, D, D>>(&self, out: &mut Matrix<T, D, D, S2>) -> bool {
        assert!(
            self.lu.is_square(),
            "LU inverse: unable to compute the inverse of a non-square matrix."
        );
        assert!(
            self.lu.shape() == out.shape(),
            "LU inverse: mismatched output shape."
        );

        out.fill_with_identity();
        self.solve_mut(out)
    }

    /// Computes the determinant of the decomposed matrix.
    #[must_use]
    pub fn determinant(&self) -> T {
        let dim = self.lu.nrows();
        assert!(
            self.lu.is_square(),
            "LU determinant: unable to compute the determinant of a non-square matrix."
        );

        let mut res = T::one();
        for i in 0..dim {
            res *= unsafe { self.lu.get_unchecked((i, i)).clone() };
        }

        res * self.p.determinant()
    }

    /// Indicates if the decomposed matrix is invertible.
    #[must_use]
    pub fn is_invertible(&self) -> bool {
        assert!(
            self.lu.is_square(),
            "LU: unable to test the invertibility of a non-square matrix."
        );

        for i in 0..self.lu.nrows() {
            if self.lu[(i, i)].is_zero() {
                return false;
            }
        }

        true
    }
}

#[doc(hidden)]
/// Executes one step of gaussian elimination on the i-th row and column of `matrix`. The diagonal
/// element `matrix[(i, i)]` is provided as argument.
pub fn gauss_step<T, R: Dim, C: Dim, S>(matrix: &mut Matrix<T, R, C, S>, diag: T, i: usize)
where
    T: Scalar + Field,
    S: StorageMut<T, R, C>,
{
    let mut submat = matrix.view_range_mut(i.., i..);

    let inv_diag = T::one() / diag;

    let (mut coeffs, mut submat) = submat.columns_range_pair_mut(0, 1..);

    let mut coeffs = coeffs.rows_range_mut(1..);
    coeffs *= inv_diag;

    let (pivot_row, mut down) = submat.rows_range_pair_mut(0, 1..);

    for k in 0..pivot_row.ncols() {
        down.column_mut(k)
            .axpy(-pivot_row[k].clone(), &coeffs, T::one());
    }
}

#[doc(hidden)]
/// Swaps the rows `i` with the row `piv` and executes one step of gaussian elimination on the i-th
/// row and column of `matrix`. The diagonal element `matrix[(i, i)]` is provided as argument.
pub fn gauss_step_swap<T, R: Dim, C: Dim, S>(
    matrix: &mut Matrix<T, R, C, S>,
    diag: T,
    i: usize,
    piv: usize,
) where
    T: Scalar + Field,
    S: StorageMut<T, R, C>,
{
    let piv = piv - i;
    let mut submat = matrix.view_range_mut(i.., i..);

    let inv_diag = T::one() / diag;

    let (mut coeffs, mut submat) = submat.columns_range_pair_mut(0, 1..);

    coeffs.swap((0, 0), (piv, 0));
    let mut coeffs = coeffs.rows_range_mut(1..);
    coeffs *= inv_diag;

    let (mut pivot_row, mut down) = submat.rows_range_pair_mut(0, 1..);

    for k in 0..pivot_row.ncols() {
        mem::swap(&mut pivot_row[k], &mut down[(piv - 1, k)]);
        down.column_mut(k)
            .axpy(-pivot_row[k].clone(), &coeffs, T::one());
    }
}