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
|
#!/usr/bin/env python
"""Tests for the linalg.isolve.lgmres module
"""
from __future__ import division, print_function, absolute_import
from numpy.testing import TestCase, assert_, assert_allclose, assert_equal, run_module_suite
import numpy as np
from numpy import zeros, array, allclose
from scipy.linalg import norm
from scipy.sparse import csr_matrix, eye, rand
from scipy.sparse.linalg.interface import LinearOperator
from scipy.sparse.linalg import splu
from scipy.sparse.linalg.isolve import lgmres, gmres
Am = csr_matrix(array([[-2,1,0,0,0,9],
[1,-2,1,0,5,0],
[0,1,-2,1,0,0],
[0,0,1,-2,1,0],
[0,3,0,1,-2,1],
[1,0,0,0,1,-2]]))
b = array([1,2,3,4,5,6])
count = [0]
def matvec(v):
count[0] += 1
return Am*v
A = LinearOperator(matvec=matvec, shape=Am.shape, dtype=Am.dtype)
def do_solve(**kw):
count[0] = 0
x0, flag = lgmres(A, b, x0=zeros(A.shape[0]), inner_m=6, tol=1e-14, **kw)
count_0 = count[0]
assert_(allclose(A*x0, b, rtol=1e-12, atol=1e-12), norm(A*x0-b))
return x0, count_0
class TestLGMRES(TestCase):
def test_preconditioner(self):
# Check that preconditioning works
pc = splu(Am.tocsc())
M = LinearOperator(matvec=pc.solve, shape=A.shape, dtype=A.dtype)
x0, count_0 = do_solve()
x1, count_1 = do_solve(M=M)
assert_(count_1 == 3)
assert_(count_1 < count_0/2)
assert_(allclose(x1, x0, rtol=1e-14))
def test_outer_v(self):
# Check that the augmentation vectors behave as expected
outer_v = []
x0, count_0 = do_solve(outer_k=6, outer_v=outer_v)
assert_(len(outer_v) > 0)
assert_(len(outer_v) <= 6)
x1, count_1 = do_solve(outer_k=6, outer_v=outer_v)
assert_(count_1 == 2, count_1)
assert_(count_1 < count_0/2)
assert_(allclose(x1, x0, rtol=1e-14))
# ---
outer_v = []
x0, count_0 = do_solve(outer_k=6, outer_v=outer_v, store_outer_Av=False)
assert_(array([v[1] is None for v in outer_v]).all())
assert_(len(outer_v) > 0)
assert_(len(outer_v) <= 6)
x1, count_1 = do_solve(outer_k=6, outer_v=outer_v)
assert_(count_1 == 3, count_1)
assert_(count_1 < count_0/2)
assert_(allclose(x1, x0, rtol=1e-14))
def test_arnoldi(self):
np.random.rand(1234)
A = eye(10000) + rand(10000,10000,density=1e-4)
b = np.random.rand(10000)
# The inner arnoldi should be equivalent to gmres
x0, flag0 = lgmres(A, b, x0=zeros(A.shape[0]), inner_m=15, maxiter=1)
x1, flag1 = gmres(A, b, x0=zeros(A.shape[0]), restart=15, maxiter=1)
assert_equal(flag0, 1)
assert_equal(flag1, 1)
assert_(np.linalg.norm(A.dot(x0) - b) > 1e-3)
assert_allclose(x0, x1)
def test_cornercase(self):
np.random.seed(1234)
# Rounding error may prevent convergence with tol=0 --- ensure
# that the return values in this case are correct, and no
# exceptions are raised
for n in [3, 5, 10, 100]:
A = 2*eye(n)
b = np.ones(n)
x, info = lgmres(A, b, maxiter=10)
assert_equal(info, 0)
assert_allclose(A.dot(x) - b, 0, atol=1e-14)
x, info = lgmres(A, b, tol=0, maxiter=10)
if info == 0:
assert_allclose(A.dot(x) - b, 0, atol=1e-14)
b = np.random.rand(n)
x, info = lgmres(A, b, maxiter=10)
assert_equal(info, 0)
assert_allclose(A.dot(x) - b, 0, atol=1e-14)
x, info = lgmres(A, b, tol=0, maxiter=10)
if info == 0:
assert_allclose(A.dot(x) - b, 0, atol=1e-14)
def test_nans(self):
A = eye(3, format='lil')
A[1,1] = np.nan
b = np.ones(3)
x, info = lgmres(A, b, tol=0, maxiter=10)
assert_equal(info, 1)
def test_breakdown_with_outer_v(self):
A = np.array([[1, 2], [3, 4]], dtype=float)
b = np.array([1, 2])
x = np.linalg.solve(A, b)
v0 = np.array([1, 0])
# The inner iteration should converge to the correct solution,
# since it's in the outer vector list
xp, info = lgmres(A, b, outer_v=[(v0, None), (x, None)], maxiter=1)
assert_allclose(xp, x, atol=1e-12)
def test_breakdown_underdetermined(self):
# Should find LSQ solution in the Krylov span in one inner
# iteration, despite solver breakdown from nilpotent A.
A = np.array([[0, 1, 1, 1],
[0, 0, 1, 1],
[0, 0, 0, 1],
[0, 0, 0, 0]], dtype=float)
bs = [
np.array([1, 1, 1, 1]),
np.array([1, 1, 1, 0]),
np.array([1, 1, 0, 0]),
np.array([1, 0, 0, 0]),
]
for b in bs:
xp, info = lgmres(A, b, maxiter=1)
resp = np.linalg.norm(A.dot(xp) - b)
K = np.c_[b, A.dot(b), A.dot(A.dot(b)), A.dot(A.dot(A.dot(b)))]
y, _, _, _ = np.linalg.lstsq(A.dot(K), b)
x = K.dot(y)
res = np.linalg.norm(A.dot(x) - b)
assert_allclose(resp, res, err_msg=repr(b))
def test_denormals(self):
# Check that no warnings are emitted if the matrix contains
# numbers for which 1/x has no float representation, and that
# the solver behaves properly.
A = np.array([[1, 2], [3, 4]], dtype=float)
A *= 100 * np.nextafter(0, 1)
b = np.array([1, 1])
xp, info = lgmres(A, b)
if info == 0:
assert_allclose(A.dot(xp), b)
if __name__ == "__main__":
run_module_suite()
|
