xref: /aosp_15_r20/external/eigen/test/lu.cpp (revision bf2c37156dfe67e5dfebd6d394bad8b2ab5804d4)
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2008-2009 Benoit Jacob <[email protected]>
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 #include "main.h"
11 #include <Eigen/LU>
12 #include "solverbase.h"
13 using namespace std;
14 
15 template<typename MatrixType>
matrix_l1_norm(const MatrixType & m)16 typename MatrixType::RealScalar matrix_l1_norm(const MatrixType& m) {
17   return m.cwiseAbs().colwise().sum().maxCoeff();
18 }
19 
lu_non_invertible()20 template<typename MatrixType> void lu_non_invertible()
21 {
22   STATIC_CHECK(( internal::is_same<typename FullPivLU<MatrixType>::StorageIndex,int>::value ));
23 
24   typedef typename MatrixType::RealScalar RealScalar;
25   /* this test covers the following files:
26      LU.h
27   */
28   Index rows, cols, cols2;
29   if(MatrixType::RowsAtCompileTime==Dynamic)
30   {
31     rows = internal::random<Index>(2,EIGEN_TEST_MAX_SIZE);
32   }
33   else
34   {
35     rows = MatrixType::RowsAtCompileTime;
36   }
37   if(MatrixType::ColsAtCompileTime==Dynamic)
38   {
39     cols = internal::random<Index>(2,EIGEN_TEST_MAX_SIZE);
40     cols2 = internal::random<int>(2,EIGEN_TEST_MAX_SIZE);
41   }
42   else
43   {
44     cols2 = cols = MatrixType::ColsAtCompileTime;
45   }
46 
47   enum {
48     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
49     ColsAtCompileTime = MatrixType::ColsAtCompileTime
50   };
51   typedef typename internal::kernel_retval_base<FullPivLU<MatrixType> >::ReturnType KernelMatrixType;
52   typedef typename internal::image_retval_base<FullPivLU<MatrixType> >::ReturnType ImageMatrixType;
53   typedef Matrix<typename MatrixType::Scalar, ColsAtCompileTime, ColsAtCompileTime>
54           CMatrixType;
55   typedef Matrix<typename MatrixType::Scalar, RowsAtCompileTime, RowsAtCompileTime>
56           RMatrixType;
57 
58   Index rank = internal::random<Index>(1, (std::min)(rows, cols)-1);
59 
60   // The image of the zero matrix should consist of a single (zero) column vector
61   VERIFY((MatrixType::Zero(rows,cols).fullPivLu().image(MatrixType::Zero(rows,cols)).cols() == 1));
62 
63   // The kernel of the zero matrix is the entire space, and thus is an invertible matrix of dimensions cols.
64   KernelMatrixType kernel = MatrixType::Zero(rows,cols).fullPivLu().kernel();
65   VERIFY((kernel.fullPivLu().isInvertible()));
66 
67   MatrixType m1(rows, cols), m3(rows, cols2);
68   CMatrixType m2(cols, cols2);
69   createRandomPIMatrixOfRank(rank, rows, cols, m1);
70 
71   FullPivLU<MatrixType> lu;
72 
73   // The special value 0.01 below works well in tests. Keep in mind that we're only computing the rank
74   // of singular values are either 0 or 1.
75   // So it's not clear at all that the epsilon should play any role there.
76   lu.setThreshold(RealScalar(0.01));
77   lu.compute(m1);
78 
79   MatrixType u(rows,cols);
80   u = lu.matrixLU().template triangularView<Upper>();
81   RMatrixType l = RMatrixType::Identity(rows,rows);
82   l.block(0,0,rows,(std::min)(rows,cols)).template triangularView<StrictlyLower>()
83     = lu.matrixLU().block(0,0,rows,(std::min)(rows,cols));
84 
85   VERIFY_IS_APPROX(lu.permutationP() * m1 * lu.permutationQ(), l*u);
86 
87   KernelMatrixType m1kernel = lu.kernel();
88   ImageMatrixType m1image = lu.image(m1);
89 
90   VERIFY_IS_APPROX(m1, lu.reconstructedMatrix());
91   VERIFY(rank == lu.rank());
92   VERIFY(cols - lu.rank() == lu.dimensionOfKernel());
93   VERIFY(!lu.isInjective());
94   VERIFY(!lu.isInvertible());
95   VERIFY(!lu.isSurjective());
96   VERIFY_IS_MUCH_SMALLER_THAN((m1 * m1kernel), m1);
97   VERIFY(m1image.fullPivLu().rank() == rank);
98   VERIFY_IS_APPROX(m1 * m1.adjoint() * m1image, m1image);
99 
100   check_solverbase<CMatrixType, MatrixType>(m1, lu, rows, cols, cols2);
101 
102   m2 = CMatrixType::Random(cols,cols2);
103   m3 = m1*m2;
104   m2 = CMatrixType::Random(cols,cols2);
105   // test that the code, which does resize(), may be applied to an xpr
106   m2.block(0,0,m2.rows(),m2.cols()) = lu.solve(m3);
107   VERIFY_IS_APPROX(m3, m1*m2);
108 }
109 
lu_invertible()110 template<typename MatrixType> void lu_invertible()
111 {
112   /* this test covers the following files:
113      FullPivLU.h
114   */
115   typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
116   Index size = MatrixType::RowsAtCompileTime;
117   if( size==Dynamic)
118     size = internal::random<Index>(1,EIGEN_TEST_MAX_SIZE);
119 
120   MatrixType m1(size, size), m2(size, size), m3(size, size);
121   FullPivLU<MatrixType> lu;
122   lu.setThreshold(RealScalar(0.01));
123   do {
124     m1 = MatrixType::Random(size,size);
125     lu.compute(m1);
126   } while(!lu.isInvertible());
127 
128   VERIFY_IS_APPROX(m1, lu.reconstructedMatrix());
129   VERIFY(0 == lu.dimensionOfKernel());
130   VERIFY(lu.kernel().cols() == 1); // the kernel() should consist of a single (zero) column vector
131   VERIFY(size == lu.rank());
132   VERIFY(lu.isInjective());
133   VERIFY(lu.isSurjective());
134   VERIFY(lu.isInvertible());
135   VERIFY(lu.image(m1).fullPivLu().isInvertible());
136 
137   check_solverbase<MatrixType, MatrixType>(m1, lu, size, size, size);
138 
139   MatrixType m1_inverse = lu.inverse();
140   m3 = MatrixType::Random(size,size);
141   m2 = lu.solve(m3);
142   VERIFY_IS_APPROX(m2, m1_inverse*m3);
143 
144   RealScalar rcond = (RealScalar(1) / matrix_l1_norm(m1)) / matrix_l1_norm(m1_inverse);
145   const RealScalar rcond_est = lu.rcond();
146   // Verify that the estimated condition number is within a factor of 10 of the
147   // truth.
148   VERIFY(rcond_est > rcond / 10 && rcond_est < rcond * 10);
149 
150   // Regression test for Bug 302
151   MatrixType m4 = MatrixType::Random(size,size);
152   VERIFY_IS_APPROX(lu.solve(m3*m4), lu.solve(m3)*m4);
153 }
154 
lu_partial_piv(Index size=MatrixType::ColsAtCompileTime)155 template<typename MatrixType> void lu_partial_piv(Index size = MatrixType::ColsAtCompileTime)
156 {
157   /* this test covers the following files:
158      PartialPivLU.h
159   */
160   typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
161 
162   MatrixType m1(size, size), m2(size, size), m3(size, size);
163   m1.setRandom();
164   PartialPivLU<MatrixType> plu(m1);
165 
166   STATIC_CHECK(( internal::is_same<typename PartialPivLU<MatrixType>::StorageIndex,int>::value ));
167 
168   VERIFY_IS_APPROX(m1, plu.reconstructedMatrix());
169 
170   check_solverbase<MatrixType, MatrixType>(m1, plu, size, size, size);
171 
172   MatrixType m1_inverse = plu.inverse();
173   m3 = MatrixType::Random(size,size);
174   m2 = plu.solve(m3);
175   VERIFY_IS_APPROX(m2, m1_inverse*m3);
176 
177   RealScalar rcond = (RealScalar(1) / matrix_l1_norm(m1)) / matrix_l1_norm(m1_inverse);
178   const RealScalar rcond_est = plu.rcond();
179   // Verify that the estimate is within a factor of 10 of the truth.
180   VERIFY(rcond_est > rcond / 10 && rcond_est < rcond * 10);
181 }
182 
lu_verify_assert()183 template<typename MatrixType> void lu_verify_assert()
184 {
185   MatrixType tmp;
186 
187   FullPivLU<MatrixType> lu;
188   VERIFY_RAISES_ASSERT(lu.matrixLU())
189   VERIFY_RAISES_ASSERT(lu.permutationP())
190   VERIFY_RAISES_ASSERT(lu.permutationQ())
191   VERIFY_RAISES_ASSERT(lu.kernel())
192   VERIFY_RAISES_ASSERT(lu.image(tmp))
193   VERIFY_RAISES_ASSERT(lu.solve(tmp))
194   VERIFY_RAISES_ASSERT(lu.transpose().solve(tmp))
195   VERIFY_RAISES_ASSERT(lu.adjoint().solve(tmp))
196   VERIFY_RAISES_ASSERT(lu.determinant())
197   VERIFY_RAISES_ASSERT(lu.rank())
198   VERIFY_RAISES_ASSERT(lu.dimensionOfKernel())
199   VERIFY_RAISES_ASSERT(lu.isInjective())
200   VERIFY_RAISES_ASSERT(lu.isSurjective())
201   VERIFY_RAISES_ASSERT(lu.isInvertible())
202   VERIFY_RAISES_ASSERT(lu.inverse())
203 
204   PartialPivLU<MatrixType> plu;
205   VERIFY_RAISES_ASSERT(plu.matrixLU())
206   VERIFY_RAISES_ASSERT(plu.permutationP())
207   VERIFY_RAISES_ASSERT(plu.solve(tmp))
208   VERIFY_RAISES_ASSERT(plu.transpose().solve(tmp))
209   VERIFY_RAISES_ASSERT(plu.adjoint().solve(tmp))
210   VERIFY_RAISES_ASSERT(plu.determinant())
211   VERIFY_RAISES_ASSERT(plu.inverse())
212 }
213 
EIGEN_DECLARE_TEST(lu)214 EIGEN_DECLARE_TEST(lu)
215 {
216   for(int i = 0; i < g_repeat; i++) {
217     CALL_SUBTEST_1( lu_non_invertible<Matrix3f>() );
218     CALL_SUBTEST_1( lu_invertible<Matrix3f>() );
219     CALL_SUBTEST_1( lu_verify_assert<Matrix3f>() );
220     CALL_SUBTEST_1( lu_partial_piv<Matrix3f>() );
221 
222     CALL_SUBTEST_2( (lu_non_invertible<Matrix<double, 4, 6> >()) );
223     CALL_SUBTEST_2( (lu_verify_assert<Matrix<double, 4, 6> >()) );
224     CALL_SUBTEST_2( lu_partial_piv<Matrix2d>() );
225     CALL_SUBTEST_2( lu_partial_piv<Matrix4d>() );
226     CALL_SUBTEST_2( (lu_partial_piv<Matrix<double,6,6> >()) );
227 
228     CALL_SUBTEST_3( lu_non_invertible<MatrixXf>() );
229     CALL_SUBTEST_3( lu_invertible<MatrixXf>() );
230     CALL_SUBTEST_3( lu_verify_assert<MatrixXf>() );
231 
232     CALL_SUBTEST_4( lu_non_invertible<MatrixXd>() );
233     CALL_SUBTEST_4( lu_invertible<MatrixXd>() );
234     CALL_SUBTEST_4( lu_partial_piv<MatrixXd>(internal::random<int>(1,EIGEN_TEST_MAX_SIZE)) );
235     CALL_SUBTEST_4( lu_verify_assert<MatrixXd>() );
236 
237     CALL_SUBTEST_5( lu_non_invertible<MatrixXcf>() );
238     CALL_SUBTEST_5( lu_invertible<MatrixXcf>() );
239     CALL_SUBTEST_5( lu_verify_assert<MatrixXcf>() );
240 
241     CALL_SUBTEST_6( lu_non_invertible<MatrixXcd>() );
242     CALL_SUBTEST_6( lu_invertible<MatrixXcd>() );
243     CALL_SUBTEST_6( lu_partial_piv<MatrixXcd>(internal::random<int>(1,EIGEN_TEST_MAX_SIZE)) );
244     CALL_SUBTEST_6( lu_verify_assert<MatrixXcd>() );
245 
246     CALL_SUBTEST_7(( lu_non_invertible<Matrix<float,Dynamic,16> >() ));
247 
248     // Test problem size constructors
249     CALL_SUBTEST_9( PartialPivLU<MatrixXf>(10) );
250     CALL_SUBTEST_9( FullPivLU<MatrixXf>(10, 20); );
251   }
252 }
253