alglib 0.1.1
Sign up to get free protection for your applications and to get access to all the features.
- data/History.txt +7 -0
- data/Manifest.txt +253 -0
- data/README.txt +33 -0
- data/Rakefile +27 -0
- data/ext/Rakefile +24 -0
- data/ext/alglib.i +24 -0
- data/ext/alglib/Makefile +157 -0
- data/ext/alglib/airyf.cpp +372 -0
- data/ext/alglib/airyf.h +81 -0
- data/ext/alglib/alglib.cpp +8558 -0
- data/ext/alglib/alglib_util.cpp +19 -0
- data/ext/alglib/alglib_util.h +14 -0
- data/ext/alglib/ap.cpp +877 -0
- data/ext/alglib/ap.english.html +364 -0
- data/ext/alglib/ap.h +666 -0
- data/ext/alglib/ap.russian.html +442 -0
- data/ext/alglib/apvt.h +754 -0
- data/ext/alglib/bdss.cpp +1500 -0
- data/ext/alglib/bdss.h +251 -0
- data/ext/alglib/bdsvd.cpp +1339 -0
- data/ext/alglib/bdsvd.h +164 -0
- data/ext/alglib/bessel.cpp +1226 -0
- data/ext/alglib/bessel.h +331 -0
- data/ext/alglib/betaf.cpp +105 -0
- data/ext/alglib/betaf.h +74 -0
- data/ext/alglib/bidiagonal.cpp +1328 -0
- data/ext/alglib/bidiagonal.h +350 -0
- data/ext/alglib/binomialdistr.cpp +247 -0
- data/ext/alglib/binomialdistr.h +153 -0
- data/ext/alglib/blas.cpp +576 -0
- data/ext/alglib/blas.h +132 -0
- data/ext/alglib/cblas.cpp +226 -0
- data/ext/alglib/cblas.h +57 -0
- data/ext/alglib/cdet.cpp +138 -0
- data/ext/alglib/cdet.h +92 -0
- data/ext/alglib/chebyshev.cpp +216 -0
- data/ext/alglib/chebyshev.h +76 -0
- data/ext/alglib/chisquaredistr.cpp +157 -0
- data/ext/alglib/chisquaredistr.h +144 -0
- data/ext/alglib/cholesky.cpp +285 -0
- data/ext/alglib/cholesky.h +86 -0
- data/ext/alglib/cinverse.cpp +298 -0
- data/ext/alglib/cinverse.h +111 -0
- data/ext/alglib/clu.cpp +337 -0
- data/ext/alglib/clu.h +120 -0
- data/ext/alglib/correlation.cpp +280 -0
- data/ext/alglib/correlation.h +77 -0
- data/ext/alglib/correlationtests.cpp +726 -0
- data/ext/alglib/correlationtests.h +134 -0
- data/ext/alglib/crcond.cpp +826 -0
- data/ext/alglib/crcond.h +148 -0
- data/ext/alglib/creflections.cpp +310 -0
- data/ext/alglib/creflections.h +165 -0
- data/ext/alglib/csolve.cpp +312 -0
- data/ext/alglib/csolve.h +99 -0
- data/ext/alglib/ctrinverse.cpp +387 -0
- data/ext/alglib/ctrinverse.h +98 -0
- data/ext/alglib/ctrlinsolve.cpp +297 -0
- data/ext/alglib/ctrlinsolve.h +81 -0
- data/ext/alglib/dawson.cpp +234 -0
- data/ext/alglib/dawson.h +74 -0
- data/ext/alglib/descriptivestatistics.cpp +436 -0
- data/ext/alglib/descriptivestatistics.h +112 -0
- data/ext/alglib/det.cpp +140 -0
- data/ext/alglib/det.h +94 -0
- data/ext/alglib/dforest.cpp +1819 -0
- data/ext/alglib/dforest.h +316 -0
- data/ext/alglib/elliptic.cpp +497 -0
- data/ext/alglib/elliptic.h +217 -0
- data/ext/alglib/estnorm.cpp +429 -0
- data/ext/alglib/estnorm.h +107 -0
- data/ext/alglib/expintegrals.cpp +422 -0
- data/ext/alglib/expintegrals.h +108 -0
- data/ext/alglib/faq.english.html +258 -0
- data/ext/alglib/faq.russian.html +272 -0
- data/ext/alglib/fdistr.cpp +202 -0
- data/ext/alglib/fdistr.h +163 -0
- data/ext/alglib/fresnel.cpp +211 -0
- data/ext/alglib/fresnel.h +91 -0
- data/ext/alglib/gammaf.cpp +338 -0
- data/ext/alglib/gammaf.h +104 -0
- data/ext/alglib/gqgengauss.cpp +235 -0
- data/ext/alglib/gqgengauss.h +92 -0
- data/ext/alglib/gqgenhermite.cpp +268 -0
- data/ext/alglib/gqgenhermite.h +63 -0
- data/ext/alglib/gqgenjacobi.cpp +297 -0
- data/ext/alglib/gqgenjacobi.h +72 -0
- data/ext/alglib/gqgenlaguerre.cpp +265 -0
- data/ext/alglib/gqgenlaguerre.h +69 -0
- data/ext/alglib/gqgenlegendre.cpp +300 -0
- data/ext/alglib/gqgenlegendre.h +62 -0
- data/ext/alglib/gqgenlobatto.cpp +305 -0
- data/ext/alglib/gqgenlobatto.h +97 -0
- data/ext/alglib/gqgenradau.cpp +232 -0
- data/ext/alglib/gqgenradau.h +95 -0
- data/ext/alglib/hbisinv.cpp +480 -0
- data/ext/alglib/hbisinv.h +183 -0
- data/ext/alglib/hblas.cpp +228 -0
- data/ext/alglib/hblas.h +64 -0
- data/ext/alglib/hcholesky.cpp +339 -0
- data/ext/alglib/hcholesky.h +91 -0
- data/ext/alglib/hermite.cpp +114 -0
- data/ext/alglib/hermite.h +49 -0
- data/ext/alglib/hessenberg.cpp +370 -0
- data/ext/alglib/hessenberg.h +152 -0
- data/ext/alglib/hevd.cpp +247 -0
- data/ext/alglib/hevd.h +107 -0
- data/ext/alglib/hsschur.cpp +1316 -0
- data/ext/alglib/hsschur.h +108 -0
- data/ext/alglib/htridiagonal.cpp +734 -0
- data/ext/alglib/htridiagonal.h +180 -0
- data/ext/alglib/ialglib.cpp +6 -0
- data/ext/alglib/ialglib.h +9 -0
- data/ext/alglib/ibetaf.cpp +960 -0
- data/ext/alglib/ibetaf.h +125 -0
- data/ext/alglib/igammaf.cpp +430 -0
- data/ext/alglib/igammaf.h +157 -0
- data/ext/alglib/inv.cpp +274 -0
- data/ext/alglib/inv.h +115 -0
- data/ext/alglib/inverseupdate.cpp +480 -0
- data/ext/alglib/inverseupdate.h +185 -0
- data/ext/alglib/jacobianelliptic.cpp +164 -0
- data/ext/alglib/jacobianelliptic.h +94 -0
- data/ext/alglib/jarquebera.cpp +2271 -0
- data/ext/alglib/jarquebera.h +80 -0
- data/ext/alglib/kmeans.cpp +356 -0
- data/ext/alglib/kmeans.h +76 -0
- data/ext/alglib/laguerre.cpp +94 -0
- data/ext/alglib/laguerre.h +48 -0
- data/ext/alglib/lbfgs.cpp +1167 -0
- data/ext/alglib/lbfgs.h +218 -0
- data/ext/alglib/lda.cpp +434 -0
- data/ext/alglib/lda.h +133 -0
- data/ext/alglib/ldlt.cpp +1130 -0
- data/ext/alglib/ldlt.h +124 -0
- data/ext/alglib/leastsquares.cpp +1252 -0
- data/ext/alglib/leastsquares.h +290 -0
- data/ext/alglib/legendre.cpp +107 -0
- data/ext/alglib/legendre.h +49 -0
- data/ext/alglib/linreg.cpp +1185 -0
- data/ext/alglib/linreg.h +380 -0
- data/ext/alglib/logit.cpp +1523 -0
- data/ext/alglib/logit.h +333 -0
- data/ext/alglib/lq.cpp +399 -0
- data/ext/alglib/lq.h +160 -0
- data/ext/alglib/lu.cpp +462 -0
- data/ext/alglib/lu.h +119 -0
- data/ext/alglib/mannwhitneyu.cpp +4490 -0
- data/ext/alglib/mannwhitneyu.h +115 -0
- data/ext/alglib/minlm.cpp +918 -0
- data/ext/alglib/minlm.h +312 -0
- data/ext/alglib/mlpbase.cpp +3375 -0
- data/ext/alglib/mlpbase.h +589 -0
- data/ext/alglib/mlpe.cpp +1369 -0
- data/ext/alglib/mlpe.h +552 -0
- data/ext/alglib/mlptrain.cpp +1056 -0
- data/ext/alglib/mlptrain.h +283 -0
- data/ext/alglib/nearunityunit.cpp +91 -0
- data/ext/alglib/nearunityunit.h +17 -0
- data/ext/alglib/normaldistr.cpp +377 -0
- data/ext/alglib/normaldistr.h +175 -0
- data/ext/alglib/nsevd.cpp +1869 -0
- data/ext/alglib/nsevd.h +140 -0
- data/ext/alglib/pca.cpp +168 -0
- data/ext/alglib/pca.h +87 -0
- data/ext/alglib/poissondistr.cpp +143 -0
- data/ext/alglib/poissondistr.h +130 -0
- data/ext/alglib/polinterpolation.cpp +685 -0
- data/ext/alglib/polinterpolation.h +206 -0
- data/ext/alglib/psif.cpp +173 -0
- data/ext/alglib/psif.h +88 -0
- data/ext/alglib/qr.cpp +414 -0
- data/ext/alglib/qr.h +168 -0
- data/ext/alglib/ratinterpolation.cpp +134 -0
- data/ext/alglib/ratinterpolation.h +72 -0
- data/ext/alglib/rcond.cpp +705 -0
- data/ext/alglib/rcond.h +140 -0
- data/ext/alglib/reflections.cpp +504 -0
- data/ext/alglib/reflections.h +165 -0
- data/ext/alglib/rotations.cpp +473 -0
- data/ext/alglib/rotations.h +128 -0
- data/ext/alglib/rsolve.cpp +221 -0
- data/ext/alglib/rsolve.h +99 -0
- data/ext/alglib/sbisinv.cpp +217 -0
- data/ext/alglib/sbisinv.h +171 -0
- data/ext/alglib/sblas.cpp +185 -0
- data/ext/alglib/sblas.h +64 -0
- data/ext/alglib/schur.cpp +156 -0
- data/ext/alglib/schur.h +102 -0
- data/ext/alglib/sdet.cpp +193 -0
- data/ext/alglib/sdet.h +101 -0
- data/ext/alglib/sevd.cpp +116 -0
- data/ext/alglib/sevd.h +99 -0
- data/ext/alglib/sinverse.cpp +672 -0
- data/ext/alglib/sinverse.h +138 -0
- data/ext/alglib/spddet.cpp +138 -0
- data/ext/alglib/spddet.h +96 -0
- data/ext/alglib/spdgevd.cpp +842 -0
- data/ext/alglib/spdgevd.h +200 -0
- data/ext/alglib/spdinverse.cpp +509 -0
- data/ext/alglib/spdinverse.h +122 -0
- data/ext/alglib/spdrcond.cpp +421 -0
- data/ext/alglib/spdrcond.h +118 -0
- data/ext/alglib/spdsolve.cpp +275 -0
- data/ext/alglib/spdsolve.h +105 -0
- data/ext/alglib/spline2d.cpp +1192 -0
- data/ext/alglib/spline2d.h +301 -0
- data/ext/alglib/spline3.cpp +1264 -0
- data/ext/alglib/spline3.h +290 -0
- data/ext/alglib/srcond.cpp +595 -0
- data/ext/alglib/srcond.h +127 -0
- data/ext/alglib/ssolve.cpp +895 -0
- data/ext/alglib/ssolve.h +139 -0
- data/ext/alglib/stdafx.h +0 -0
- data/ext/alglib/stest.cpp +131 -0
- data/ext/alglib/stest.h +94 -0
- data/ext/alglib/studenttdistr.cpp +222 -0
- data/ext/alglib/studenttdistr.h +115 -0
- data/ext/alglib/studentttests.cpp +377 -0
- data/ext/alglib/studentttests.h +178 -0
- data/ext/alglib/svd.cpp +620 -0
- data/ext/alglib/svd.h +126 -0
- data/ext/alglib/tdbisinv.cpp +2608 -0
- data/ext/alglib/tdbisinv.h +228 -0
- data/ext/alglib/tdevd.cpp +1229 -0
- data/ext/alglib/tdevd.h +115 -0
- data/ext/alglib/tridiagonal.cpp +594 -0
- data/ext/alglib/tridiagonal.h +171 -0
- data/ext/alglib/trigintegrals.cpp +490 -0
- data/ext/alglib/trigintegrals.h +131 -0
- data/ext/alglib/trinverse.cpp +345 -0
- data/ext/alglib/trinverse.h +98 -0
- data/ext/alglib/trlinsolve.cpp +926 -0
- data/ext/alglib/trlinsolve.h +73 -0
- data/ext/alglib/tsort.cpp +405 -0
- data/ext/alglib/tsort.h +54 -0
- data/ext/alglib/variancetests.cpp +245 -0
- data/ext/alglib/variancetests.h +134 -0
- data/ext/alglib/wsr.cpp +6285 -0
- data/ext/alglib/wsr.h +96 -0
- data/ext/ap.i +97 -0
- data/ext/correlation.i +24 -0
- data/ext/extconf.rb +6 -0
- data/ext/logit.i +89 -0
- data/lib/alglib.rb +71 -0
- data/lib/alglib/correlation.rb +26 -0
- data/lib/alglib/linearregression.rb +63 -0
- data/lib/alglib/logit.rb +42 -0
- data/test/test_alglib.rb +52 -0
- data/test/test_correlation.rb +44 -0
- data/test/test_correlationtest.rb +45 -0
- data/test/test_linreg.rb +35 -0
- data/test/test_logit.rb +43 -0
- data/test/test_pca.rb +27 -0
- metadata +326 -0
@@ -0,0 +1,175 @@
|
|
1
|
+
/*************************************************************************
|
2
|
+
Cephes Math Library Release 2.8: June, 2000
|
3
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
4
|
+
|
5
|
+
Contributors:
|
6
|
+
* Sergey Bochkanov (ALGLIB project). Translation from C to
|
7
|
+
pseudocode.
|
8
|
+
|
9
|
+
See subroutines comments for additional copyrights.
|
10
|
+
|
11
|
+
Redistribution and use in source and binary forms, with or without
|
12
|
+
modification, are permitted provided that the following conditions are
|
13
|
+
met:
|
14
|
+
|
15
|
+
- Redistributions of source code must retain the above copyright
|
16
|
+
notice, this list of conditions and the following disclaimer.
|
17
|
+
|
18
|
+
- Redistributions in binary form must reproduce the above copyright
|
19
|
+
notice, this list of conditions and the following disclaimer listed
|
20
|
+
in this license in the documentation and/or other materials
|
21
|
+
provided with the distribution.
|
22
|
+
|
23
|
+
- Neither the name of the copyright holders nor the names of its
|
24
|
+
contributors may be used to endorse or promote products derived from
|
25
|
+
this software without specific prior written permission.
|
26
|
+
|
27
|
+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
28
|
+
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
29
|
+
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
30
|
+
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
31
|
+
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
32
|
+
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
33
|
+
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
34
|
+
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
35
|
+
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
36
|
+
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
37
|
+
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
38
|
+
*************************************************************************/
|
39
|
+
|
40
|
+
#ifndef _normaldistr_h
|
41
|
+
#define _normaldistr_h
|
42
|
+
|
43
|
+
#include "ap.h"
|
44
|
+
#include "ialglib.h"
|
45
|
+
|
46
|
+
/*************************************************************************
|
47
|
+
Error function
|
48
|
+
|
49
|
+
The integral is
|
50
|
+
|
51
|
+
x
|
52
|
+
-
|
53
|
+
2 | | 2
|
54
|
+
erf(x) = -------- | exp( - t ) dt.
|
55
|
+
sqrt(pi) | |
|
56
|
+
-
|
57
|
+
0
|
58
|
+
|
59
|
+
For 0 <= |x| < 1, erf(x) = x * P4(x**2)/Q5(x**2); otherwise
|
60
|
+
erf(x) = 1 - erfc(x).
|
61
|
+
|
62
|
+
|
63
|
+
ACCURACY:
|
64
|
+
|
65
|
+
Relative error:
|
66
|
+
arithmetic domain # trials peak rms
|
67
|
+
IEEE 0,1 30000 3.7e-16 1.0e-16
|
68
|
+
|
69
|
+
Cephes Math Library Release 2.8: June, 2000
|
70
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
71
|
+
*************************************************************************/
|
72
|
+
double erf(double x);
|
73
|
+
|
74
|
+
|
75
|
+
/*************************************************************************
|
76
|
+
Complementary error function
|
77
|
+
|
78
|
+
1 - erf(x) =
|
79
|
+
|
80
|
+
inf.
|
81
|
+
-
|
82
|
+
2 | | 2
|
83
|
+
erfc(x) = -------- | exp( - t ) dt
|
84
|
+
sqrt(pi) | |
|
85
|
+
-
|
86
|
+
x
|
87
|
+
|
88
|
+
|
89
|
+
For small x, erfc(x) = 1 - erf(x); otherwise rational
|
90
|
+
approximations are computed.
|
91
|
+
|
92
|
+
|
93
|
+
ACCURACY:
|
94
|
+
|
95
|
+
Relative error:
|
96
|
+
arithmetic domain # trials peak rms
|
97
|
+
IEEE 0,26.6417 30000 5.7e-14 1.5e-14
|
98
|
+
|
99
|
+
Cephes Math Library Release 2.8: June, 2000
|
100
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
101
|
+
*************************************************************************/
|
102
|
+
double erfc(double x);
|
103
|
+
|
104
|
+
|
105
|
+
/*************************************************************************
|
106
|
+
Normal distribution function
|
107
|
+
|
108
|
+
Returns the area under the Gaussian probability density
|
109
|
+
function, integrated from minus infinity to x:
|
110
|
+
|
111
|
+
x
|
112
|
+
-
|
113
|
+
1 | | 2
|
114
|
+
ndtr(x) = --------- | exp( - t /2 ) dt
|
115
|
+
sqrt(2pi) | |
|
116
|
+
-
|
117
|
+
-inf.
|
118
|
+
|
119
|
+
= ( 1 + erf(z) ) / 2
|
120
|
+
= erfc(z) / 2
|
121
|
+
|
122
|
+
where z = x/sqrt(2). Computation is via the functions
|
123
|
+
erf and erfc.
|
124
|
+
|
125
|
+
|
126
|
+
ACCURACY:
|
127
|
+
|
128
|
+
Relative error:
|
129
|
+
arithmetic domain # trials peak rms
|
130
|
+
IEEE -13,0 30000 3.4e-14 6.7e-15
|
131
|
+
|
132
|
+
Cephes Math Library Release 2.8: June, 2000
|
133
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
134
|
+
*************************************************************************/
|
135
|
+
double normaldistribution(double x);
|
136
|
+
|
137
|
+
|
138
|
+
/*************************************************************************
|
139
|
+
Inverse of the error function
|
140
|
+
|
141
|
+
Cephes Math Library Release 2.8: June, 2000
|
142
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
143
|
+
*************************************************************************/
|
144
|
+
double inverf(double e);
|
145
|
+
|
146
|
+
|
147
|
+
/*************************************************************************
|
148
|
+
Inverse of Normal distribution function
|
149
|
+
|
150
|
+
Returns the argument, x, for which the area under the
|
151
|
+
Gaussian probability density function (integrated from
|
152
|
+
minus infinity to x) is equal to y.
|
153
|
+
|
154
|
+
|
155
|
+
For small arguments 0 < y < exp(-2), the program computes
|
156
|
+
z = sqrt( -2.0 * log(y) ); then the approximation is
|
157
|
+
x = z - log(z)/z - (1/z) P(1/z) / Q(1/z).
|
158
|
+
There are two rational functions P/Q, one for 0 < y < exp(-32)
|
159
|
+
and the other for y up to exp(-2). For larger arguments,
|
160
|
+
w = y - 0.5, and x/sqrt(2pi) = w + w**3 R(w**2)/S(w**2)).
|
161
|
+
|
162
|
+
ACCURACY:
|
163
|
+
|
164
|
+
Relative error:
|
165
|
+
arithmetic domain # trials peak rms
|
166
|
+
IEEE 0.125, 1 20000 7.2e-16 1.3e-16
|
167
|
+
IEEE 3e-308, 0.135 50000 4.6e-16 9.8e-17
|
168
|
+
|
169
|
+
Cephes Math Library Release 2.8: June, 2000
|
170
|
+
Copyright 1984, 1987, 1988, 1992, 2000 by Stephen L. Moshier
|
171
|
+
*************************************************************************/
|
172
|
+
double invnormaldistribution(double y0);
|
173
|
+
|
174
|
+
|
175
|
+
#endif
|
@@ -0,0 +1,1869 @@
|
|
1
|
+
/*************************************************************************
|
2
|
+
Copyright (c) 1992-2007 The University of Tennessee. All rights reserved.
|
3
|
+
|
4
|
+
Contributors:
|
5
|
+
* Sergey Bochkanov (ALGLIB project). Translation from FORTRAN to
|
6
|
+
pseudocode.
|
7
|
+
|
8
|
+
See subroutines comments for additional copyrights.
|
9
|
+
|
10
|
+
Redistribution and use in source and binary forms, with or without
|
11
|
+
modification, are permitted provided that the following conditions are
|
12
|
+
met:
|
13
|
+
|
14
|
+
- Redistributions of source code must retain the above copyright
|
15
|
+
notice, this list of conditions and the following disclaimer.
|
16
|
+
|
17
|
+
- Redistributions in binary form must reproduce the above copyright
|
18
|
+
notice, this list of conditions and the following disclaimer listed
|
19
|
+
in this license in the documentation and/or other materials
|
20
|
+
provided with the distribution.
|
21
|
+
|
22
|
+
- Neither the name of the copyright holders nor the names of its
|
23
|
+
contributors may be used to endorse or promote products derived from
|
24
|
+
this software without specific prior written permission.
|
25
|
+
|
26
|
+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
27
|
+
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
28
|
+
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
29
|
+
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
30
|
+
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
31
|
+
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
32
|
+
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
33
|
+
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
34
|
+
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
35
|
+
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
36
|
+
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
37
|
+
*************************************************************************/
|
38
|
+
|
39
|
+
#include <stdafx.h>
|
40
|
+
#include "nsevd.h"
|
41
|
+
|
42
|
+
static void internaltrevc(const ap::real_2d_array& t,
|
43
|
+
int n,
|
44
|
+
int side,
|
45
|
+
int howmny,
|
46
|
+
ap::boolean_1d_array vselect,
|
47
|
+
ap::real_2d_array& vl,
|
48
|
+
ap::real_2d_array& vr,
|
49
|
+
int& m,
|
50
|
+
int& info);
|
51
|
+
static void internalhsevdlaln2(const bool& ltrans,
|
52
|
+
const int& na,
|
53
|
+
const int& nw,
|
54
|
+
const double& smin,
|
55
|
+
const double& ca,
|
56
|
+
const ap::real_2d_array& a,
|
57
|
+
const double& d1,
|
58
|
+
const double& d2,
|
59
|
+
const ap::real_2d_array& b,
|
60
|
+
const double& wr,
|
61
|
+
const double& wi,
|
62
|
+
ap::boolean_1d_array& rswap4,
|
63
|
+
ap::boolean_1d_array& zswap4,
|
64
|
+
ap::integer_2d_array& ipivot44,
|
65
|
+
ap::real_1d_array& civ4,
|
66
|
+
ap::real_1d_array& crv4,
|
67
|
+
ap::real_2d_array& x,
|
68
|
+
double& scl,
|
69
|
+
double& xnorm,
|
70
|
+
int& info);
|
71
|
+
static void internalhsevdladiv(const double& a,
|
72
|
+
const double& b,
|
73
|
+
const double& c,
|
74
|
+
const double& d,
|
75
|
+
double& p,
|
76
|
+
double& q);
|
77
|
+
|
78
|
+
/*************************************************************************
|
79
|
+
Finding eigenvalues and eigenvectors of a general matrix
|
80
|
+
|
81
|
+
The algorithm finds eigenvalues and eigenvectors of a general matrix by
|
82
|
+
using the QR algorithm with multiple shifts. The algorithm can find
|
83
|
+
eigenvalues and both left and right eigenvectors.
|
84
|
+
|
85
|
+
The right eigenvector is a vector x such that A*x = w*x, and the left
|
86
|
+
eigenvector is a vector y such that y'*A = w*y' (here y' implies a complex
|
87
|
+
conjugate transposition of vector y).
|
88
|
+
|
89
|
+
Input parameters:
|
90
|
+
A - matrix. Array whose indexes range within [0..N-1, 0..N-1].
|
91
|
+
N - size of matrix A.
|
92
|
+
VNeeded - flag controlling whether eigenvectors are needed or not.
|
93
|
+
If VNeeded is equal to:
|
94
|
+
* 0, eigenvectors are not returned;
|
95
|
+
* 1, right eigenvectors are returned;
|
96
|
+
* 2, left eigenvectors are returned;
|
97
|
+
* 3, both left and right eigenvectors are returned.
|
98
|
+
|
99
|
+
Output parameters:
|
100
|
+
WR - real parts of eigenvalues.
|
101
|
+
Array whose index ranges within [0..N-1].
|
102
|
+
WR - imaginary parts of eigenvalues.
|
103
|
+
Array whose index ranges within [0..N-1].
|
104
|
+
VL, VR - arrays of left and right eigenvectors (if they are needed).
|
105
|
+
If WI[i]=0, the respective eigenvalue is a real number,
|
106
|
+
and it corresponds to the column number I of matrices VL/VR.
|
107
|
+
If WI[i]>0, we have a pair of complex conjugate numbers with
|
108
|
+
positive and negative imaginary parts:
|
109
|
+
the first eigenvalue WR[i] + sqrt(-1)*WI[i];
|
110
|
+
the second eigenvalue WR[i+1] + sqrt(-1)*WI[i+1];
|
111
|
+
WI[i]>0
|
112
|
+
WI[i+1] = -WI[i] < 0
|
113
|
+
In that case, the eigenvector corresponding to the first
|
114
|
+
eigenvalue is located in i and i+1 columns of matrices
|
115
|
+
VL/VR (the column number i contains the real part, and the
|
116
|
+
column number i+1 contains the imaginary part), and the vector
|
117
|
+
corresponding to the second eigenvalue is a complex conjugate to
|
118
|
+
the first vector.
|
119
|
+
Arrays whose indexes range within [0..N-1, 0..N-1].
|
120
|
+
|
121
|
+
Result:
|
122
|
+
True, if the algorithm has converged.
|
123
|
+
False, if the algorithm has not converged.
|
124
|
+
|
125
|
+
Note 1:
|
126
|
+
Some users may ask the following question: what if WI[N-1]>0?
|
127
|
+
WI[N] must contain an eigenvalue which is complex conjugate to the
|
128
|
+
N-th eigenvalue, but the array has only size N?
|
129
|
+
The answer is as follows: such a situation cannot occur because the
|
130
|
+
algorithm finds a pairs of eigenvalues, therefore, if WI[i]>0, I is
|
131
|
+
strictly less than N-1.
|
132
|
+
|
133
|
+
Note 2:
|
134
|
+
The algorithm performance depends on the value of the internal parameter
|
135
|
+
NS of the InternalSchurDecomposition subroutine which defines the number
|
136
|
+
of shifts in the QR algorithm (similarly to the block width in block-matrix
|
137
|
+
algorithms of linear algebra). If you require maximum performance
|
138
|
+
on your machine, it is recommended to adjust this parameter manually.
|
139
|
+
|
140
|
+
|
141
|
+
See also the InternalTREVC subroutine.
|
142
|
+
|
143
|
+
The algorithm is based on the LAPACK 3.0 library.
|
144
|
+
*************************************************************************/
|
145
|
+
bool rmatrixevd(ap::real_2d_array a,
|
146
|
+
int n,
|
147
|
+
int vneeded,
|
148
|
+
ap::real_1d_array& wr,
|
149
|
+
ap::real_1d_array& wi,
|
150
|
+
ap::real_2d_array& vl,
|
151
|
+
ap::real_2d_array& vr)
|
152
|
+
{
|
153
|
+
bool result;
|
154
|
+
ap::real_2d_array a1;
|
155
|
+
ap::real_2d_array vl1;
|
156
|
+
ap::real_2d_array vr1;
|
157
|
+
ap::real_1d_array wr1;
|
158
|
+
ap::real_1d_array wi1;
|
159
|
+
int i;
|
160
|
+
double mx;
|
161
|
+
|
162
|
+
ap::ap_error::make_assertion(vneeded>=0&&vneeded<=3, "RMatrixEVD: incorrect VNeeded!");
|
163
|
+
a1.setbounds(1, n, 1, n);
|
164
|
+
for(i = 1; i <= n; i++)
|
165
|
+
{
|
166
|
+
ap::vmove(&a1(i, 1), &a(i-1, 0), ap::vlen(1,n));
|
167
|
+
}
|
168
|
+
result = nonsymmetricevd(a1, n, vneeded, wr1, wi1, vl1, vr1);
|
169
|
+
if( result )
|
170
|
+
{
|
171
|
+
wr.setbounds(0, n-1);
|
172
|
+
wi.setbounds(0, n-1);
|
173
|
+
ap::vmove(&wr(0), &wr1(1), ap::vlen(0,n-1));
|
174
|
+
ap::vmove(&wi(0), &wi1(1), ap::vlen(0,n-1));
|
175
|
+
if( vneeded==2||vneeded==3 )
|
176
|
+
{
|
177
|
+
vl.setbounds(0, n-1, 0, n-1);
|
178
|
+
for(i = 0; i <= n-1; i++)
|
179
|
+
{
|
180
|
+
ap::vmove(&vl(i, 0), &vl1(i+1, 1), ap::vlen(0,n-1));
|
181
|
+
}
|
182
|
+
}
|
183
|
+
if( vneeded==1||vneeded==3 )
|
184
|
+
{
|
185
|
+
vr.setbounds(0, n-1, 0, n-1);
|
186
|
+
for(i = 0; i <= n-1; i++)
|
187
|
+
{
|
188
|
+
ap::vmove(&vr(i, 0), &vr1(i+1, 1), ap::vlen(0,n-1));
|
189
|
+
}
|
190
|
+
}
|
191
|
+
}
|
192
|
+
return result;
|
193
|
+
}
|
194
|
+
|
195
|
+
|
196
|
+
/*************************************************************************
|
197
|
+
Obsolete 1-based subroutine
|
198
|
+
*************************************************************************/
|
199
|
+
bool nonsymmetricevd(ap::real_2d_array a,
|
200
|
+
int n,
|
201
|
+
int vneeded,
|
202
|
+
ap::real_1d_array& wr,
|
203
|
+
ap::real_1d_array& wi,
|
204
|
+
ap::real_2d_array& vl,
|
205
|
+
ap::real_2d_array& vr)
|
206
|
+
{
|
207
|
+
bool result;
|
208
|
+
ap::real_2d_array s;
|
209
|
+
ap::real_1d_array tau;
|
210
|
+
ap::boolean_1d_array sel;
|
211
|
+
int i;
|
212
|
+
int info;
|
213
|
+
int m;
|
214
|
+
|
215
|
+
ap::ap_error::make_assertion(vneeded>=0&&vneeded<=3, "NonSymmetricEVD: incorrect VNeeded!");
|
216
|
+
if( vneeded==0 )
|
217
|
+
{
|
218
|
+
|
219
|
+
//
|
220
|
+
// Eigen values only
|
221
|
+
//
|
222
|
+
toupperhessenberg(a, n, tau);
|
223
|
+
internalschurdecomposition(a, n, 0, 0, wr, wi, s, info);
|
224
|
+
result = info==0;
|
225
|
+
return result;
|
226
|
+
}
|
227
|
+
|
228
|
+
//
|
229
|
+
// Eigen values and vectors
|
230
|
+
//
|
231
|
+
toupperhessenberg(a, n, tau);
|
232
|
+
unpackqfromupperhessenberg(a, n, tau, s);
|
233
|
+
internalschurdecomposition(a, n, 1, 1, wr, wi, s, info);
|
234
|
+
result = info==0;
|
235
|
+
if( !result )
|
236
|
+
{
|
237
|
+
return result;
|
238
|
+
}
|
239
|
+
if( vneeded==1||vneeded==3 )
|
240
|
+
{
|
241
|
+
vr.setbounds(1, n, 1, n);
|
242
|
+
for(i = 1; i <= n; i++)
|
243
|
+
{
|
244
|
+
ap::vmove(&vr(i, 1), &s(i, 1), ap::vlen(1,n));
|
245
|
+
}
|
246
|
+
}
|
247
|
+
if( vneeded==2||vneeded==3 )
|
248
|
+
{
|
249
|
+
vl.setbounds(1, n, 1, n);
|
250
|
+
for(i = 1; i <= n; i++)
|
251
|
+
{
|
252
|
+
ap::vmove(&vl(i, 1), &s(i, 1), ap::vlen(1,n));
|
253
|
+
}
|
254
|
+
}
|
255
|
+
internaltrevc(a, n, vneeded, 1, sel, vl, vr, m, info);
|
256
|
+
result = info==0;
|
257
|
+
return result;
|
258
|
+
}
|
259
|
+
|
260
|
+
|
261
|
+
static void internaltrevc(const ap::real_2d_array& t,
|
262
|
+
int n,
|
263
|
+
int side,
|
264
|
+
int howmny,
|
265
|
+
ap::boolean_1d_array vselect,
|
266
|
+
ap::real_2d_array& vl,
|
267
|
+
ap::real_2d_array& vr,
|
268
|
+
int& m,
|
269
|
+
int& info)
|
270
|
+
{
|
271
|
+
bool allv;
|
272
|
+
bool bothv;
|
273
|
+
bool leftv;
|
274
|
+
bool over;
|
275
|
+
bool pair;
|
276
|
+
bool rightv;
|
277
|
+
bool somev;
|
278
|
+
int i;
|
279
|
+
int ierr;
|
280
|
+
int ii;
|
281
|
+
int ip;
|
282
|
+
int iis;
|
283
|
+
int j;
|
284
|
+
int j1;
|
285
|
+
int j2;
|
286
|
+
int jnxt;
|
287
|
+
int k;
|
288
|
+
int ki;
|
289
|
+
int n2;
|
290
|
+
double beta;
|
291
|
+
double bignum;
|
292
|
+
double emax;
|
293
|
+
double ovfl;
|
294
|
+
double rec;
|
295
|
+
double remax;
|
296
|
+
double scl;
|
297
|
+
double smin;
|
298
|
+
double smlnum;
|
299
|
+
double ulp;
|
300
|
+
double unfl;
|
301
|
+
double vcrit;
|
302
|
+
double vmax;
|
303
|
+
double wi;
|
304
|
+
double wr;
|
305
|
+
double xnorm;
|
306
|
+
ap::real_2d_array x;
|
307
|
+
ap::real_1d_array work;
|
308
|
+
ap::real_1d_array temp;
|
309
|
+
ap::real_2d_array temp11;
|
310
|
+
ap::real_2d_array temp22;
|
311
|
+
ap::real_2d_array temp11b;
|
312
|
+
ap::real_2d_array temp21b;
|
313
|
+
ap::real_2d_array temp12b;
|
314
|
+
ap::real_2d_array temp22b;
|
315
|
+
bool skipflag;
|
316
|
+
int k1;
|
317
|
+
int k2;
|
318
|
+
int k3;
|
319
|
+
int k4;
|
320
|
+
double vt;
|
321
|
+
ap::boolean_1d_array rswap4;
|
322
|
+
ap::boolean_1d_array zswap4;
|
323
|
+
ap::integer_2d_array ipivot44;
|
324
|
+
ap::real_1d_array civ4;
|
325
|
+
ap::real_1d_array crv4;
|
326
|
+
|
327
|
+
x.setbounds(1, 2, 1, 2);
|
328
|
+
temp11.setbounds(1, 1, 1, 1);
|
329
|
+
temp11b.setbounds(1, 1, 1, 1);
|
330
|
+
temp21b.setbounds(1, 2, 1, 1);
|
331
|
+
temp12b.setbounds(1, 1, 1, 2);
|
332
|
+
temp22b.setbounds(1, 2, 1, 2);
|
333
|
+
temp22.setbounds(1, 2, 1, 2);
|
334
|
+
work.setbounds(1, 3*n);
|
335
|
+
temp.setbounds(1, n);
|
336
|
+
rswap4.setbounds(1, 4);
|
337
|
+
zswap4.setbounds(1, 4);
|
338
|
+
ipivot44.setbounds(1, 4, 1, 4);
|
339
|
+
civ4.setbounds(1, 4);
|
340
|
+
crv4.setbounds(1, 4);
|
341
|
+
if( howmny!=1 )
|
342
|
+
{
|
343
|
+
if( side==1||side==3 )
|
344
|
+
{
|
345
|
+
vr.setbounds(1, n, 1, n);
|
346
|
+
}
|
347
|
+
if( side==2||side==3 )
|
348
|
+
{
|
349
|
+
vl.setbounds(1, n, 1, n);
|
350
|
+
}
|
351
|
+
}
|
352
|
+
|
353
|
+
//
|
354
|
+
// Decode and test the input parameters
|
355
|
+
//
|
356
|
+
bothv = side==3;
|
357
|
+
rightv = side==1||bothv;
|
358
|
+
leftv = side==2||bothv;
|
359
|
+
allv = howmny==2;
|
360
|
+
over = howmny==1;
|
361
|
+
somev = howmny==3;
|
362
|
+
info = 0;
|
363
|
+
if( n<0 )
|
364
|
+
{
|
365
|
+
info = -2;
|
366
|
+
return;
|
367
|
+
}
|
368
|
+
if( !rightv&&!leftv )
|
369
|
+
{
|
370
|
+
info = -3;
|
371
|
+
return;
|
372
|
+
}
|
373
|
+
if( !allv&&!over&&!somev )
|
374
|
+
{
|
375
|
+
info = -4;
|
376
|
+
return;
|
377
|
+
}
|
378
|
+
|
379
|
+
//
|
380
|
+
// Set M to the number of columns required to store the selected
|
381
|
+
// eigenvectors, standardize the array SELECT if necessary, and
|
382
|
+
// test MM.
|
383
|
+
//
|
384
|
+
if( somev )
|
385
|
+
{
|
386
|
+
m = 0;
|
387
|
+
pair = false;
|
388
|
+
for(j = 1; j <= n; j++)
|
389
|
+
{
|
390
|
+
if( pair )
|
391
|
+
{
|
392
|
+
pair = false;
|
393
|
+
vselect(j) = false;
|
394
|
+
}
|
395
|
+
else
|
396
|
+
{
|
397
|
+
if( j<n )
|
398
|
+
{
|
399
|
+
if( t(j+1,j)==0 )
|
400
|
+
{
|
401
|
+
if( vselect(j) )
|
402
|
+
{
|
403
|
+
m = m+1;
|
404
|
+
}
|
405
|
+
}
|
406
|
+
else
|
407
|
+
{
|
408
|
+
pair = true;
|
409
|
+
if( vselect(j)||vselect(j+1) )
|
410
|
+
{
|
411
|
+
vselect(j) = true;
|
412
|
+
m = m+2;
|
413
|
+
}
|
414
|
+
}
|
415
|
+
}
|
416
|
+
else
|
417
|
+
{
|
418
|
+
if( vselect(n) )
|
419
|
+
{
|
420
|
+
m = m+1;
|
421
|
+
}
|
422
|
+
}
|
423
|
+
}
|
424
|
+
}
|
425
|
+
}
|
426
|
+
else
|
427
|
+
{
|
428
|
+
m = n;
|
429
|
+
}
|
430
|
+
|
431
|
+
//
|
432
|
+
// Quick return if possible.
|
433
|
+
//
|
434
|
+
if( n==0 )
|
435
|
+
{
|
436
|
+
return;
|
437
|
+
}
|
438
|
+
|
439
|
+
//
|
440
|
+
// Set the constants to control overflow.
|
441
|
+
//
|
442
|
+
unfl = ap::minrealnumber;
|
443
|
+
ovfl = 1/unfl;
|
444
|
+
ulp = ap::machineepsilon;
|
445
|
+
smlnum = unfl*(n/ulp);
|
446
|
+
bignum = (1-ulp)/smlnum;
|
447
|
+
|
448
|
+
//
|
449
|
+
// Compute 1-norm of each column of strictly upper triangular
|
450
|
+
// part of T to control overflow in triangular solver.
|
451
|
+
//
|
452
|
+
work(1) = 0;
|
453
|
+
for(j = 2; j <= n; j++)
|
454
|
+
{
|
455
|
+
work(j) = 0;
|
456
|
+
for(i = 1; i <= j-1; i++)
|
457
|
+
{
|
458
|
+
work(j) = work(j)+fabs(t(i,j));
|
459
|
+
}
|
460
|
+
}
|
461
|
+
|
462
|
+
//
|
463
|
+
// Index IP is used to specify the real or complex eigenvalue:
|
464
|
+
// IP = 0, real eigenvalue,
|
465
|
+
// 1, first of conjugate complex pair: (wr,wi)
|
466
|
+
// -1, second of conjugate complex pair: (wr,wi)
|
467
|
+
//
|
468
|
+
n2 = 2*n;
|
469
|
+
if( rightv )
|
470
|
+
{
|
471
|
+
|
472
|
+
//
|
473
|
+
// Compute right eigenvectors.
|
474
|
+
//
|
475
|
+
ip = 0;
|
476
|
+
iis = m;
|
477
|
+
for(ki = n; ki >= 1; ki--)
|
478
|
+
{
|
479
|
+
skipflag = false;
|
480
|
+
if( ip==1 )
|
481
|
+
{
|
482
|
+
skipflag = true;
|
483
|
+
}
|
484
|
+
else
|
485
|
+
{
|
486
|
+
if( ki!=1 )
|
487
|
+
{
|
488
|
+
if( t(ki,ki-1)!=0 )
|
489
|
+
{
|
490
|
+
ip = -1;
|
491
|
+
}
|
492
|
+
}
|
493
|
+
if( somev )
|
494
|
+
{
|
495
|
+
if( ip==0 )
|
496
|
+
{
|
497
|
+
if( !vselect(ki) )
|
498
|
+
{
|
499
|
+
skipflag = true;
|
500
|
+
}
|
501
|
+
}
|
502
|
+
else
|
503
|
+
{
|
504
|
+
if( !vselect(ki-1) )
|
505
|
+
{
|
506
|
+
skipflag = true;
|
507
|
+
}
|
508
|
+
}
|
509
|
+
}
|
510
|
+
}
|
511
|
+
if( !skipflag )
|
512
|
+
{
|
513
|
+
|
514
|
+
//
|
515
|
+
// Compute the KI-th eigenvalue (WR,WI).
|
516
|
+
//
|
517
|
+
wr = t(ki,ki);
|
518
|
+
wi = 0;
|
519
|
+
if( ip!=0 )
|
520
|
+
{
|
521
|
+
wi = sqrt(fabs(t(ki,ki-1)))*sqrt(fabs(t(ki-1,ki)));
|
522
|
+
}
|
523
|
+
smin = ap::maxreal(ulp*(fabs(wr)+fabs(wi)), smlnum);
|
524
|
+
if( ip==0 )
|
525
|
+
{
|
526
|
+
|
527
|
+
//
|
528
|
+
// Real right eigenvector
|
529
|
+
//
|
530
|
+
work(ki+n) = 1;
|
531
|
+
|
532
|
+
//
|
533
|
+
// Form right-hand side
|
534
|
+
//
|
535
|
+
for(k = 1; k <= ki-1; k++)
|
536
|
+
{
|
537
|
+
work(k+n) = -t(k,ki);
|
538
|
+
}
|
539
|
+
|
540
|
+
//
|
541
|
+
// Solve the upper quasi-triangular system:
|
542
|
+
// (T(1:KI-1,1:KI-1) - WR)*X = SCALE*WORK.
|
543
|
+
//
|
544
|
+
jnxt = ki-1;
|
545
|
+
for(j = ki-1; j >= 1; j--)
|
546
|
+
{
|
547
|
+
if( j>jnxt )
|
548
|
+
{
|
549
|
+
continue;
|
550
|
+
}
|
551
|
+
j1 = j;
|
552
|
+
j2 = j;
|
553
|
+
jnxt = j-1;
|
554
|
+
if( j>1 )
|
555
|
+
{
|
556
|
+
if( t(j,j-1)!=0 )
|
557
|
+
{
|
558
|
+
j1 = j-1;
|
559
|
+
jnxt = j-2;
|
560
|
+
}
|
561
|
+
}
|
562
|
+
if( j1==j2 )
|
563
|
+
{
|
564
|
+
|
565
|
+
//
|
566
|
+
// 1-by-1 diagonal block
|
567
|
+
//
|
568
|
+
temp11(1,1) = t(j,j);
|
569
|
+
temp11b(1,1) = work(j+n);
|
570
|
+
internalhsevdlaln2(false, 1, 1, smin, double(1), temp11, 1.0, 1.0, temp11b, wr, 0.0, rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
571
|
+
|
572
|
+
//
|
573
|
+
// Scale X(1,1) to avoid overflow when updating
|
574
|
+
// the right-hand side.
|
575
|
+
//
|
576
|
+
if( xnorm>1 )
|
577
|
+
{
|
578
|
+
if( work(j)>bignum/xnorm )
|
579
|
+
{
|
580
|
+
x(1,1) = x(1,1)/xnorm;
|
581
|
+
scl = scl/xnorm;
|
582
|
+
}
|
583
|
+
}
|
584
|
+
|
585
|
+
//
|
586
|
+
// Scale if necessary
|
587
|
+
//
|
588
|
+
if( scl!=1 )
|
589
|
+
{
|
590
|
+
k1 = n+1;
|
591
|
+
k2 = n+ki;
|
592
|
+
ap::vmul(&work(k1), ap::vlen(k1,k2), scl);
|
593
|
+
}
|
594
|
+
work(j+n) = x(1,1);
|
595
|
+
|
596
|
+
//
|
597
|
+
// Update right-hand side
|
598
|
+
//
|
599
|
+
k1 = 1+n;
|
600
|
+
k2 = j-1+n;
|
601
|
+
k3 = j-1;
|
602
|
+
vt = -x(1,1);
|
603
|
+
ap::vadd(work.getvector(k1, k2), t.getcolumn(j, 1, k3), vt);
|
604
|
+
}
|
605
|
+
else
|
606
|
+
{
|
607
|
+
|
608
|
+
//
|
609
|
+
// 2-by-2 diagonal block
|
610
|
+
//
|
611
|
+
temp22(1,1) = t(j-1,j-1);
|
612
|
+
temp22(1,2) = t(j-1,j);
|
613
|
+
temp22(2,1) = t(j,j-1);
|
614
|
+
temp22(2,2) = t(j,j);
|
615
|
+
temp21b(1,1) = work(j-1+n);
|
616
|
+
temp21b(2,1) = work(j+n);
|
617
|
+
internalhsevdlaln2(false, 2, 1, smin, 1.0, temp22, 1.0, 1.0, temp21b, wr, double(0), rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
618
|
+
|
619
|
+
//
|
620
|
+
// Scale X(1,1) and X(2,1) to avoid overflow when
|
621
|
+
// updating the right-hand side.
|
622
|
+
//
|
623
|
+
if( xnorm>1 )
|
624
|
+
{
|
625
|
+
beta = ap::maxreal(work(j-1), work(j));
|
626
|
+
if( beta>bignum/xnorm )
|
627
|
+
{
|
628
|
+
x(1,1) = x(1,1)/xnorm;
|
629
|
+
x(2,1) = x(2,1)/xnorm;
|
630
|
+
scl = scl/xnorm;
|
631
|
+
}
|
632
|
+
}
|
633
|
+
|
634
|
+
//
|
635
|
+
// Scale if necessary
|
636
|
+
//
|
637
|
+
if( scl!=1 )
|
638
|
+
{
|
639
|
+
k1 = 1+n;
|
640
|
+
k2 = ki+n;
|
641
|
+
ap::vmul(&work(k1), ap::vlen(k1,k2), scl);
|
642
|
+
}
|
643
|
+
work(j-1+n) = x(1,1);
|
644
|
+
work(j+n) = x(2,1);
|
645
|
+
|
646
|
+
//
|
647
|
+
// Update right-hand side
|
648
|
+
//
|
649
|
+
k1 = 1+n;
|
650
|
+
k2 = j-2+n;
|
651
|
+
k3 = j-2;
|
652
|
+
k4 = j-1;
|
653
|
+
vt = -x(1,1);
|
654
|
+
ap::vadd(work.getvector(k1, k2), t.getcolumn(k4, 1, k3), vt);
|
655
|
+
vt = -x(2,1);
|
656
|
+
ap::vadd(work.getvector(k1, k2), t.getcolumn(j, 1, k3), vt);
|
657
|
+
}
|
658
|
+
}
|
659
|
+
|
660
|
+
//
|
661
|
+
// Copy the vector x or Q*x to VR and normalize.
|
662
|
+
//
|
663
|
+
if( !over )
|
664
|
+
{
|
665
|
+
k1 = 1+n;
|
666
|
+
k2 = ki+n;
|
667
|
+
ap::vmove(vr.getcolumn(iis, 1, ki), work.getvector(k1, k2));
|
668
|
+
ii = columnidxabsmax(vr, 1, ki, iis);
|
669
|
+
remax = 1/fabs(vr(ii,iis));
|
670
|
+
ap::vmul(vr.getcolumn(iis, 1, ki), remax);
|
671
|
+
for(k = ki+1; k <= n; k++)
|
672
|
+
{
|
673
|
+
vr(k,iis) = 0;
|
674
|
+
}
|
675
|
+
}
|
676
|
+
else
|
677
|
+
{
|
678
|
+
if( ki>1 )
|
679
|
+
{
|
680
|
+
ap::vmove(temp.getvector(1, n), vr.getcolumn(ki, 1, n));
|
681
|
+
matrixvectormultiply(vr, 1, n, 1, ki-1, false, work, 1+n, ki-1+n, 1.0, temp, 1, n, work(ki+n));
|
682
|
+
ap::vmove(vr.getcolumn(ki, 1, n), temp.getvector(1, n));
|
683
|
+
}
|
684
|
+
ii = columnidxabsmax(vr, 1, n, ki);
|
685
|
+
remax = 1/fabs(vr(ii,ki));
|
686
|
+
ap::vmul(vr.getcolumn(ki, 1, n), remax);
|
687
|
+
}
|
688
|
+
}
|
689
|
+
else
|
690
|
+
{
|
691
|
+
|
692
|
+
//
|
693
|
+
// Complex right eigenvector.
|
694
|
+
//
|
695
|
+
// Initial solve
|
696
|
+
// [ (T(KI-1,KI-1) T(KI-1,KI) ) - (WR + I* WI)]*X = 0.
|
697
|
+
// [ (T(KI,KI-1) T(KI,KI) ) ]
|
698
|
+
//
|
699
|
+
if( fabs(t(ki-1,ki))>=fabs(t(ki,ki-1)) )
|
700
|
+
{
|
701
|
+
work(ki-1+n) = 1;
|
702
|
+
work(ki+n2) = wi/t(ki-1,ki);
|
703
|
+
}
|
704
|
+
else
|
705
|
+
{
|
706
|
+
work(ki-1+n) = -wi/t(ki,ki-1);
|
707
|
+
work(ki+n2) = 1;
|
708
|
+
}
|
709
|
+
work(ki+n) = 0;
|
710
|
+
work(ki-1+n2) = 0;
|
711
|
+
|
712
|
+
//
|
713
|
+
// Form right-hand side
|
714
|
+
//
|
715
|
+
for(k = 1; k <= ki-2; k++)
|
716
|
+
{
|
717
|
+
work(k+n) = -work(ki-1+n)*t(k,ki-1);
|
718
|
+
work(k+n2) = -work(ki+n2)*t(k,ki);
|
719
|
+
}
|
720
|
+
|
721
|
+
//
|
722
|
+
// Solve upper quasi-triangular system:
|
723
|
+
// (T(1:KI-2,1:KI-2) - (WR+i*WI))*X = SCALE*(WORK+i*WORK2)
|
724
|
+
//
|
725
|
+
jnxt = ki-2;
|
726
|
+
for(j = ki-2; j >= 1; j--)
|
727
|
+
{
|
728
|
+
if( j>jnxt )
|
729
|
+
{
|
730
|
+
continue;
|
731
|
+
}
|
732
|
+
j1 = j;
|
733
|
+
j2 = j;
|
734
|
+
jnxt = j-1;
|
735
|
+
if( j>1 )
|
736
|
+
{
|
737
|
+
if( t(j,j-1)!=0 )
|
738
|
+
{
|
739
|
+
j1 = j-1;
|
740
|
+
jnxt = j-2;
|
741
|
+
}
|
742
|
+
}
|
743
|
+
if( j1==j2 )
|
744
|
+
{
|
745
|
+
|
746
|
+
//
|
747
|
+
// 1-by-1 diagonal block
|
748
|
+
//
|
749
|
+
temp11(1,1) = t(j,j);
|
750
|
+
temp12b(1,1) = work(j+n);
|
751
|
+
temp12b(1,2) = work(j+n+n);
|
752
|
+
internalhsevdlaln2(false, 1, 2, smin, 1.0, temp11, 1.0, 1.0, temp12b, wr, wi, rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
753
|
+
|
754
|
+
//
|
755
|
+
// Scale X(1,1) and X(1,2) to avoid overflow when
|
756
|
+
// updating the right-hand side.
|
757
|
+
//
|
758
|
+
if( xnorm>1 )
|
759
|
+
{
|
760
|
+
if( work(j)>bignum/xnorm )
|
761
|
+
{
|
762
|
+
x(1,1) = x(1,1)/xnorm;
|
763
|
+
x(1,2) = x(1,2)/xnorm;
|
764
|
+
scl = scl/xnorm;
|
765
|
+
}
|
766
|
+
}
|
767
|
+
|
768
|
+
//
|
769
|
+
// Scale if necessary
|
770
|
+
//
|
771
|
+
if( scl!=1 )
|
772
|
+
{
|
773
|
+
k1 = 1+n;
|
774
|
+
k2 = ki+n;
|
775
|
+
ap::vmul(&work(k1), ap::vlen(k1,k2), scl);
|
776
|
+
k1 = 1+n2;
|
777
|
+
k2 = ki+n2;
|
778
|
+
ap::vmul(&work(k1), ap::vlen(k1,k2), scl);
|
779
|
+
}
|
780
|
+
work(j+n) = x(1,1);
|
781
|
+
work(j+n2) = x(1,2);
|
782
|
+
|
783
|
+
//
|
784
|
+
// Update the right-hand side
|
785
|
+
//
|
786
|
+
k1 = 1+n;
|
787
|
+
k2 = j-1+n;
|
788
|
+
k3 = 1;
|
789
|
+
k4 = j-1;
|
790
|
+
vt = -x(1,1);
|
791
|
+
ap::vadd(work.getvector(k1, k2), t.getcolumn(j, k3, k4), vt);
|
792
|
+
k1 = 1+n2;
|
793
|
+
k2 = j-1+n2;
|
794
|
+
k3 = 1;
|
795
|
+
k4 = j-1;
|
796
|
+
vt = -x(1,2);
|
797
|
+
ap::vadd(work.getvector(k1, k2), t.getcolumn(j, k3, k4), vt);
|
798
|
+
}
|
799
|
+
else
|
800
|
+
{
|
801
|
+
|
802
|
+
//
|
803
|
+
// 2-by-2 diagonal block
|
804
|
+
//
|
805
|
+
temp22(1,1) = t(j-1,j-1);
|
806
|
+
temp22(1,2) = t(j-1,j);
|
807
|
+
temp22(2,1) = t(j,j-1);
|
808
|
+
temp22(2,2) = t(j,j);
|
809
|
+
temp22b(1,1) = work(j-1+n);
|
810
|
+
temp22b(1,2) = work(j-1+n+n);
|
811
|
+
temp22b(2,1) = work(j+n);
|
812
|
+
temp22b(2,2) = work(j+n+n);
|
813
|
+
internalhsevdlaln2(false, 2, 2, smin, 1.0, temp22, 1.0, 1.0, temp22b, wr, wi, rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
814
|
+
|
815
|
+
//
|
816
|
+
// Scale X to avoid overflow when updating
|
817
|
+
// the right-hand side.
|
818
|
+
//
|
819
|
+
if( xnorm>1 )
|
820
|
+
{
|
821
|
+
beta = ap::maxreal(work(j-1), work(j));
|
822
|
+
if( beta>bignum/xnorm )
|
823
|
+
{
|
824
|
+
rec = 1/xnorm;
|
825
|
+
x(1,1) = x(1,1)*rec;
|
826
|
+
x(1,2) = x(1,2)*rec;
|
827
|
+
x(2,1) = x(2,1)*rec;
|
828
|
+
x(2,2) = x(2,2)*rec;
|
829
|
+
scl = scl*rec;
|
830
|
+
}
|
831
|
+
}
|
832
|
+
|
833
|
+
//
|
834
|
+
// Scale if necessary
|
835
|
+
//
|
836
|
+
if( scl!=1 )
|
837
|
+
{
|
838
|
+
ap::vmul(&work(1+n), ap::vlen(1+n,ki+n), scl);
|
839
|
+
ap::vmul(&work(1+n2), ap::vlen(1+n2,ki+n2), scl);
|
840
|
+
}
|
841
|
+
work(j-1+n) = x(1,1);
|
842
|
+
work(j+n) = x(2,1);
|
843
|
+
work(j-1+n2) = x(1,2);
|
844
|
+
work(j+n2) = x(2,2);
|
845
|
+
|
846
|
+
//
|
847
|
+
// Update the right-hand side
|
848
|
+
//
|
849
|
+
vt = -x(1,1);
|
850
|
+
ap::vadd(work.getvector(n+1, n+j-2), t.getcolumn(j-1, 1, j-2), vt);
|
851
|
+
vt = -x(2,1);
|
852
|
+
ap::vadd(work.getvector(n+1, n+j-2), t.getcolumn(j, 1, j-2), vt);
|
853
|
+
vt = -x(1,2);
|
854
|
+
ap::vadd(work.getvector(n2+1, n2+j-2), t.getcolumn(j-1, 1, j-2), vt);
|
855
|
+
vt = -x(2,2);
|
856
|
+
ap::vadd(work.getvector(n2+1, n2+j-2), t.getcolumn(j, 1, j-2), vt);
|
857
|
+
}
|
858
|
+
}
|
859
|
+
|
860
|
+
//
|
861
|
+
// Copy the vector x or Q*x to VR and normalize.
|
862
|
+
//
|
863
|
+
if( !over )
|
864
|
+
{
|
865
|
+
ap::vmove(vr.getcolumn(iis-1, 1, ki), work.getvector(n+1, n+ki));
|
866
|
+
ap::vmove(vr.getcolumn(iis, 1, ki), work.getvector(n2+1, n2+ki));
|
867
|
+
emax = 0;
|
868
|
+
for(k = 1; k <= ki; k++)
|
869
|
+
{
|
870
|
+
emax = ap::maxreal(emax, fabs(vr(k,iis-1))+fabs(vr(k,iis)));
|
871
|
+
}
|
872
|
+
remax = 1/emax;
|
873
|
+
ap::vmul(vr.getcolumn(iis-1, 1, ki), remax);
|
874
|
+
ap::vmul(vr.getcolumn(iis, 1, ki), remax);
|
875
|
+
for(k = ki+1; k <= n; k++)
|
876
|
+
{
|
877
|
+
vr(k,iis-1) = 0;
|
878
|
+
vr(k,iis) = 0;
|
879
|
+
}
|
880
|
+
}
|
881
|
+
else
|
882
|
+
{
|
883
|
+
if( ki>2 )
|
884
|
+
{
|
885
|
+
ap::vmove(temp.getvector(1, n), vr.getcolumn(ki-1, 1, n));
|
886
|
+
matrixvectormultiply(vr, 1, n, 1, ki-2, false, work, 1+n, ki-2+n, 1.0, temp, 1, n, work(ki-1+n));
|
887
|
+
ap::vmove(vr.getcolumn(ki-1, 1, n), temp.getvector(1, n));
|
888
|
+
ap::vmove(temp.getvector(1, n), vr.getcolumn(ki, 1, n));
|
889
|
+
matrixvectormultiply(vr, 1, n, 1, ki-2, false, work, 1+n2, ki-2+n2, 1.0, temp, 1, n, work(ki+n2));
|
890
|
+
ap::vmove(vr.getcolumn(ki, 1, n), temp.getvector(1, n));
|
891
|
+
}
|
892
|
+
else
|
893
|
+
{
|
894
|
+
vt = work(ki-1+n);
|
895
|
+
ap::vmul(vr.getcolumn(ki-1, 1, n), vt);
|
896
|
+
vt = work(ki+n2);
|
897
|
+
ap::vmul(vr.getcolumn(ki, 1, n), vt);
|
898
|
+
}
|
899
|
+
emax = 0;
|
900
|
+
for(k = 1; k <= n; k++)
|
901
|
+
{
|
902
|
+
emax = ap::maxreal(emax, fabs(vr(k,ki-1))+fabs(vr(k,ki)));
|
903
|
+
}
|
904
|
+
remax = 1/emax;
|
905
|
+
ap::vmul(vr.getcolumn(ki-1, 1, n), remax);
|
906
|
+
ap::vmul(vr.getcolumn(ki, 1, n), remax);
|
907
|
+
}
|
908
|
+
}
|
909
|
+
iis = iis-1;
|
910
|
+
if( ip!=0 )
|
911
|
+
{
|
912
|
+
iis = iis-1;
|
913
|
+
}
|
914
|
+
}
|
915
|
+
if( ip==1 )
|
916
|
+
{
|
917
|
+
ip = 0;
|
918
|
+
}
|
919
|
+
if( ip==-1 )
|
920
|
+
{
|
921
|
+
ip = 1;
|
922
|
+
}
|
923
|
+
}
|
924
|
+
}
|
925
|
+
if( leftv )
|
926
|
+
{
|
927
|
+
|
928
|
+
//
|
929
|
+
// Compute left eigenvectors.
|
930
|
+
//
|
931
|
+
ip = 0;
|
932
|
+
iis = 1;
|
933
|
+
for(ki = 1; ki <= n; ki++)
|
934
|
+
{
|
935
|
+
skipflag = false;
|
936
|
+
if( ip==-1 )
|
937
|
+
{
|
938
|
+
skipflag = true;
|
939
|
+
}
|
940
|
+
else
|
941
|
+
{
|
942
|
+
if( ki!=n )
|
943
|
+
{
|
944
|
+
if( t(ki+1,ki)!=0 )
|
945
|
+
{
|
946
|
+
ip = 1;
|
947
|
+
}
|
948
|
+
}
|
949
|
+
if( somev )
|
950
|
+
{
|
951
|
+
if( !vselect(ki) )
|
952
|
+
{
|
953
|
+
skipflag = true;
|
954
|
+
}
|
955
|
+
}
|
956
|
+
}
|
957
|
+
if( !skipflag )
|
958
|
+
{
|
959
|
+
|
960
|
+
//
|
961
|
+
// Compute the KI-th eigenvalue (WR,WI).
|
962
|
+
//
|
963
|
+
wr = t(ki,ki);
|
964
|
+
wi = 0;
|
965
|
+
if( ip!=0 )
|
966
|
+
{
|
967
|
+
wi = sqrt(fabs(t(ki,ki+1)))*sqrt(fabs(t(ki+1,ki)));
|
968
|
+
}
|
969
|
+
smin = ap::maxreal(ulp*(fabs(wr)+fabs(wi)), smlnum);
|
970
|
+
if( ip==0 )
|
971
|
+
{
|
972
|
+
|
973
|
+
//
|
974
|
+
// Real left eigenvector.
|
975
|
+
//
|
976
|
+
work(ki+n) = 1;
|
977
|
+
|
978
|
+
//
|
979
|
+
// Form right-hand side
|
980
|
+
//
|
981
|
+
for(k = ki+1; k <= n; k++)
|
982
|
+
{
|
983
|
+
work(k+n) = -t(ki,k);
|
984
|
+
}
|
985
|
+
|
986
|
+
//
|
987
|
+
// Solve the quasi-triangular system:
|
988
|
+
// (T(KI+1:N,KI+1:N) - WR)'*X = SCALE*WORK
|
989
|
+
//
|
990
|
+
vmax = 1;
|
991
|
+
vcrit = bignum;
|
992
|
+
jnxt = ki+1;
|
993
|
+
for(j = ki+1; j <= n; j++)
|
994
|
+
{
|
995
|
+
if( j<jnxt )
|
996
|
+
{
|
997
|
+
continue;
|
998
|
+
}
|
999
|
+
j1 = j;
|
1000
|
+
j2 = j;
|
1001
|
+
jnxt = j+1;
|
1002
|
+
if( j<n )
|
1003
|
+
{
|
1004
|
+
if( t(j+1,j)!=0 )
|
1005
|
+
{
|
1006
|
+
j2 = j+1;
|
1007
|
+
jnxt = j+2;
|
1008
|
+
}
|
1009
|
+
}
|
1010
|
+
if( j1==j2 )
|
1011
|
+
{
|
1012
|
+
|
1013
|
+
//
|
1014
|
+
// 1-by-1 diagonal block
|
1015
|
+
//
|
1016
|
+
// Scale if necessary to avoid overflow when forming
|
1017
|
+
// the right-hand side.
|
1018
|
+
//
|
1019
|
+
if( work(j)>vcrit )
|
1020
|
+
{
|
1021
|
+
rec = 1/vmax;
|
1022
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), rec);
|
1023
|
+
vmax = 1;
|
1024
|
+
vcrit = bignum;
|
1025
|
+
}
|
1026
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+1, j-1), work.getvector(ki+1+n, j-1+n));
|
1027
|
+
work(j+n) = work(j+n)-vt;
|
1028
|
+
|
1029
|
+
//
|
1030
|
+
// Solve (T(J,J)-WR)'*X = WORK
|
1031
|
+
//
|
1032
|
+
temp11(1,1) = t(j,j);
|
1033
|
+
temp11b(1,1) = work(j+n);
|
1034
|
+
internalhsevdlaln2(false, 1, 1, smin, 1.0, temp11, 1.0, 1.0, temp11b, wr, double(0), rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
1035
|
+
|
1036
|
+
//
|
1037
|
+
// Scale if necessary
|
1038
|
+
//
|
1039
|
+
if( scl!=1 )
|
1040
|
+
{
|
1041
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), scl);
|
1042
|
+
}
|
1043
|
+
work(j+n) = x(1,1);
|
1044
|
+
vmax = ap::maxreal(fabs(work(j+n)), vmax);
|
1045
|
+
vcrit = bignum/vmax;
|
1046
|
+
}
|
1047
|
+
else
|
1048
|
+
{
|
1049
|
+
|
1050
|
+
//
|
1051
|
+
// 2-by-2 diagonal block
|
1052
|
+
//
|
1053
|
+
// Scale if necessary to avoid overflow when forming
|
1054
|
+
// the right-hand side.
|
1055
|
+
//
|
1056
|
+
beta = ap::maxreal(work(j), work(j+1));
|
1057
|
+
if( beta>vcrit )
|
1058
|
+
{
|
1059
|
+
rec = 1/vmax;
|
1060
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), rec);
|
1061
|
+
vmax = 1;
|
1062
|
+
vcrit = bignum;
|
1063
|
+
}
|
1064
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+1, j-1), work.getvector(ki+1+n, j-1+n));
|
1065
|
+
work(j+n) = work(j+n)-vt;
|
1066
|
+
vt = ap::vdotproduct(t.getcolumn(j+1, ki+1, j-1), work.getvector(ki+1+n, j-1+n));
|
1067
|
+
work(j+1+n) = work(j+1+n)-vt;
|
1068
|
+
|
1069
|
+
//
|
1070
|
+
// Solve
|
1071
|
+
// [T(J,J)-WR T(J,J+1) ]'* X = SCALE*( WORK1 )
|
1072
|
+
// [T(J+1,J) T(J+1,J+1)-WR] ( WORK2 )
|
1073
|
+
//
|
1074
|
+
temp22(1,1) = t(j,j);
|
1075
|
+
temp22(1,2) = t(j,j+1);
|
1076
|
+
temp22(2,1) = t(j+1,j);
|
1077
|
+
temp22(2,2) = t(j+1,j+1);
|
1078
|
+
temp21b(1,1) = work(j+n);
|
1079
|
+
temp21b(2,1) = work(j+1+n);
|
1080
|
+
internalhsevdlaln2(true, 2, 1, smin, 1.0, temp22, 1.0, 1.0, temp21b, wr, double(0), rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
1081
|
+
|
1082
|
+
//
|
1083
|
+
// Scale if necessary
|
1084
|
+
//
|
1085
|
+
if( scl!=1 )
|
1086
|
+
{
|
1087
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), scl);
|
1088
|
+
}
|
1089
|
+
work(j+n) = x(1,1);
|
1090
|
+
work(j+1+n) = x(2,1);
|
1091
|
+
vmax = ap::maxreal(fabs(work(j+n)), ap::maxreal(fabs(work(j+1+n)), vmax));
|
1092
|
+
vcrit = bignum/vmax;
|
1093
|
+
}
|
1094
|
+
}
|
1095
|
+
|
1096
|
+
//
|
1097
|
+
// Copy the vector x or Q*x to VL and normalize.
|
1098
|
+
//
|
1099
|
+
if( !over )
|
1100
|
+
{
|
1101
|
+
ap::vmove(vl.getcolumn(iis, ki, n), work.getvector(ki+n, n+n));
|
1102
|
+
ii = columnidxabsmax(vl, ki, n, iis);
|
1103
|
+
remax = 1/fabs(vl(ii,iis));
|
1104
|
+
ap::vmul(vl.getcolumn(iis, ki, n), remax);
|
1105
|
+
for(k = 1; k <= ki-1; k++)
|
1106
|
+
{
|
1107
|
+
vl(k,iis) = 0;
|
1108
|
+
}
|
1109
|
+
}
|
1110
|
+
else
|
1111
|
+
{
|
1112
|
+
if( ki<n )
|
1113
|
+
{
|
1114
|
+
ap::vmove(temp.getvector(1, n), vl.getcolumn(ki, 1, n));
|
1115
|
+
matrixvectormultiply(vl, 1, n, ki+1, n, false, work, ki+1+n, n+n, 1.0, temp, 1, n, work(ki+n));
|
1116
|
+
ap::vmove(vl.getcolumn(ki, 1, n), temp.getvector(1, n));
|
1117
|
+
}
|
1118
|
+
ii = columnidxabsmax(vl, 1, n, ki);
|
1119
|
+
remax = 1/fabs(vl(ii,ki));
|
1120
|
+
ap::vmul(vl.getcolumn(ki, 1, n), remax);
|
1121
|
+
}
|
1122
|
+
}
|
1123
|
+
else
|
1124
|
+
{
|
1125
|
+
|
1126
|
+
//
|
1127
|
+
// Complex left eigenvector.
|
1128
|
+
//
|
1129
|
+
// Initial solve:
|
1130
|
+
// ((T(KI,KI) T(KI,KI+1) )' - (WR - I* WI))*X = 0.
|
1131
|
+
// ((T(KI+1,KI) T(KI+1,KI+1)) )
|
1132
|
+
//
|
1133
|
+
if( fabs(t(ki,ki+1))>=fabs(t(ki+1,ki)) )
|
1134
|
+
{
|
1135
|
+
work(ki+n) = wi/t(ki,ki+1);
|
1136
|
+
work(ki+1+n2) = 1;
|
1137
|
+
}
|
1138
|
+
else
|
1139
|
+
{
|
1140
|
+
work(ki+n) = 1;
|
1141
|
+
work(ki+1+n2) = -wi/t(ki+1,ki);
|
1142
|
+
}
|
1143
|
+
work(ki+1+n) = 0;
|
1144
|
+
work(ki+n2) = 0;
|
1145
|
+
|
1146
|
+
//
|
1147
|
+
// Form right-hand side
|
1148
|
+
//
|
1149
|
+
for(k = ki+2; k <= n; k++)
|
1150
|
+
{
|
1151
|
+
work(k+n) = -work(ki+n)*t(ki,k);
|
1152
|
+
work(k+n2) = -work(ki+1+n2)*t(ki+1,k);
|
1153
|
+
}
|
1154
|
+
|
1155
|
+
//
|
1156
|
+
// Solve complex quasi-triangular system:
|
1157
|
+
// ( T(KI+2,N:KI+2,N) - (WR-i*WI) )*X = WORK1+i*WORK2
|
1158
|
+
//
|
1159
|
+
vmax = 1;
|
1160
|
+
vcrit = bignum;
|
1161
|
+
jnxt = ki+2;
|
1162
|
+
for(j = ki+2; j <= n; j++)
|
1163
|
+
{
|
1164
|
+
if( j<jnxt )
|
1165
|
+
{
|
1166
|
+
continue;
|
1167
|
+
}
|
1168
|
+
j1 = j;
|
1169
|
+
j2 = j;
|
1170
|
+
jnxt = j+1;
|
1171
|
+
if( j<n )
|
1172
|
+
{
|
1173
|
+
if( t(j+1,j)!=0 )
|
1174
|
+
{
|
1175
|
+
j2 = j+1;
|
1176
|
+
jnxt = j+2;
|
1177
|
+
}
|
1178
|
+
}
|
1179
|
+
if( j1==j2 )
|
1180
|
+
{
|
1181
|
+
|
1182
|
+
//
|
1183
|
+
// 1-by-1 diagonal block
|
1184
|
+
//
|
1185
|
+
// Scale if necessary to avoid overflow when
|
1186
|
+
// forming the right-hand side elements.
|
1187
|
+
//
|
1188
|
+
if( work(j)>vcrit )
|
1189
|
+
{
|
1190
|
+
rec = 1/vmax;
|
1191
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), rec);
|
1192
|
+
ap::vmul(&work(ki+n2), ap::vlen(ki+n2,n+n2), rec);
|
1193
|
+
vmax = 1;
|
1194
|
+
vcrit = bignum;
|
1195
|
+
}
|
1196
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+2, j-1), work.getvector(ki+2+n, j-1+n));
|
1197
|
+
work(j+n) = work(j+n)-vt;
|
1198
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+2, j-1), work.getvector(ki+2+n2, j-1+n2));
|
1199
|
+
work(j+n2) = work(j+n2)-vt;
|
1200
|
+
|
1201
|
+
//
|
1202
|
+
// Solve (T(J,J)-(WR-i*WI))*(X11+i*X12)= WK+I*WK2
|
1203
|
+
//
|
1204
|
+
temp11(1,1) = t(j,j);
|
1205
|
+
temp12b(1,1) = work(j+n);
|
1206
|
+
temp12b(1,2) = work(j+n+n);
|
1207
|
+
internalhsevdlaln2(false, 1, 2, smin, 1.0, temp11, 1.0, 1.0, temp12b, wr, -wi, rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
1208
|
+
|
1209
|
+
//
|
1210
|
+
// Scale if necessary
|
1211
|
+
//
|
1212
|
+
if( scl!=1 )
|
1213
|
+
{
|
1214
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), scl);
|
1215
|
+
ap::vmul(&work(ki+n2), ap::vlen(ki+n2,n+n2), scl);
|
1216
|
+
}
|
1217
|
+
work(j+n) = x(1,1);
|
1218
|
+
work(j+n2) = x(1,2);
|
1219
|
+
vmax = ap::maxreal(fabs(work(j+n)), ap::maxreal(fabs(work(j+n2)), vmax));
|
1220
|
+
vcrit = bignum/vmax;
|
1221
|
+
}
|
1222
|
+
else
|
1223
|
+
{
|
1224
|
+
|
1225
|
+
//
|
1226
|
+
// 2-by-2 diagonal block
|
1227
|
+
//
|
1228
|
+
// Scale if necessary to avoid overflow when forming
|
1229
|
+
// the right-hand side elements.
|
1230
|
+
//
|
1231
|
+
beta = ap::maxreal(work(j), work(j+1));
|
1232
|
+
if( beta>vcrit )
|
1233
|
+
{
|
1234
|
+
rec = 1/vmax;
|
1235
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), rec);
|
1236
|
+
ap::vmul(&work(ki+n2), ap::vlen(ki+n2,n+n2), rec);
|
1237
|
+
vmax = 1;
|
1238
|
+
vcrit = bignum;
|
1239
|
+
}
|
1240
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+2, j-1), work.getvector(ki+2+n, j-1+n));
|
1241
|
+
work(j+n) = work(j+n)-vt;
|
1242
|
+
vt = ap::vdotproduct(t.getcolumn(j, ki+2, j-1), work.getvector(ki+2+n2, j-1+n2));
|
1243
|
+
work(j+n2) = work(j+n2)-vt;
|
1244
|
+
vt = ap::vdotproduct(t.getcolumn(j+1, ki+2, j-1), work.getvector(ki+2+n, j-1+n));
|
1245
|
+
work(j+1+n) = work(j+1+n)-vt;
|
1246
|
+
vt = ap::vdotproduct(t.getcolumn(j+1, ki+2, j-1), work.getvector(ki+2+n2, j-1+n2));
|
1247
|
+
work(j+1+n2) = work(j+1+n2)-vt;
|
1248
|
+
|
1249
|
+
//
|
1250
|
+
// Solve 2-by-2 complex linear equation
|
1251
|
+
// ([T(j,j) T(j,j+1) ]'-(wr-i*wi)*I)*X = SCALE*B
|
1252
|
+
// ([T(j+1,j) T(j+1,j+1)] )
|
1253
|
+
//
|
1254
|
+
temp22(1,1) = t(j,j);
|
1255
|
+
temp22(1,2) = t(j,j+1);
|
1256
|
+
temp22(2,1) = t(j+1,j);
|
1257
|
+
temp22(2,2) = t(j+1,j+1);
|
1258
|
+
temp22b(1,1) = work(j+n);
|
1259
|
+
temp22b(1,2) = work(j+n+n);
|
1260
|
+
temp22b(2,1) = work(j+1+n);
|
1261
|
+
temp22b(2,2) = work(j+1+n+n);
|
1262
|
+
internalhsevdlaln2(true, 2, 2, smin, 1.0, temp22, 1.0, 1.0, temp22b, wr, -wi, rswap4, zswap4, ipivot44, civ4, crv4, x, scl, xnorm, ierr);
|
1263
|
+
|
1264
|
+
//
|
1265
|
+
// Scale if necessary
|
1266
|
+
//
|
1267
|
+
if( scl!=1 )
|
1268
|
+
{
|
1269
|
+
ap::vmul(&work(ki+n), ap::vlen(ki+n,n+n), scl);
|
1270
|
+
ap::vmul(&work(ki+n2), ap::vlen(ki+n2,n+n2), scl);
|
1271
|
+
}
|
1272
|
+
work(j+n) = x(1,1);
|
1273
|
+
work(j+n2) = x(1,2);
|
1274
|
+
work(j+1+n) = x(2,1);
|
1275
|
+
work(j+1+n2) = x(2,2);
|
1276
|
+
vmax = ap::maxreal(fabs(x(1,1)), vmax);
|
1277
|
+
vmax = ap::maxreal(fabs(x(1,2)), vmax);
|
1278
|
+
vmax = ap::maxreal(fabs(x(2,1)), vmax);
|
1279
|
+
vmax = ap::maxreal(fabs(x(2,2)), vmax);
|
1280
|
+
vcrit = bignum/vmax;
|
1281
|
+
}
|
1282
|
+
}
|
1283
|
+
|
1284
|
+
//
|
1285
|
+
// Copy the vector x or Q*x to VL and normalize.
|
1286
|
+
//
|
1287
|
+
if( !over )
|
1288
|
+
{
|
1289
|
+
ap::vmove(vl.getcolumn(iis, ki, n), work.getvector(ki+n, n+n));
|
1290
|
+
ap::vmove(vl.getcolumn(iis+1, ki, n), work.getvector(ki+n2, n+n2));
|
1291
|
+
emax = 0;
|
1292
|
+
for(k = ki; k <= n; k++)
|
1293
|
+
{
|
1294
|
+
emax = ap::maxreal(emax, fabs(vl(k,iis))+fabs(vl(k,iis+1)));
|
1295
|
+
}
|
1296
|
+
remax = 1/emax;
|
1297
|
+
ap::vmul(vl.getcolumn(iis, ki, n), remax);
|
1298
|
+
ap::vmul(vl.getcolumn(iis+1, ki, n), remax);
|
1299
|
+
for(k = 1; k <= ki-1; k++)
|
1300
|
+
{
|
1301
|
+
vl(k,iis) = 0;
|
1302
|
+
vl(k,iis+1) = 0;
|
1303
|
+
}
|
1304
|
+
}
|
1305
|
+
else
|
1306
|
+
{
|
1307
|
+
if( ki<n-1 )
|
1308
|
+
{
|
1309
|
+
ap::vmove(temp.getvector(1, n), vl.getcolumn(ki, 1, n));
|
1310
|
+
matrixvectormultiply(vl, 1, n, ki+2, n, false, work, ki+2+n, n+n, 1.0, temp, 1, n, work(ki+n));
|
1311
|
+
ap::vmove(vl.getcolumn(ki, 1, n), temp.getvector(1, n));
|
1312
|
+
ap::vmove(temp.getvector(1, n), vl.getcolumn(ki+1, 1, n));
|
1313
|
+
matrixvectormultiply(vl, 1, n, ki+2, n, false, work, ki+2+n2, n+n2, 1.0, temp, 1, n, work(ki+1+n2));
|
1314
|
+
ap::vmove(vl.getcolumn(ki+1, 1, n), temp.getvector(1, n));
|
1315
|
+
}
|
1316
|
+
else
|
1317
|
+
{
|
1318
|
+
vt = work(ki+n);
|
1319
|
+
ap::vmul(vl.getcolumn(ki, 1, n), vt);
|
1320
|
+
vt = work(ki+1+n2);
|
1321
|
+
ap::vmul(vl.getcolumn(ki+1, 1, n), vt);
|
1322
|
+
}
|
1323
|
+
emax = 0;
|
1324
|
+
for(k = 1; k <= n; k++)
|
1325
|
+
{
|
1326
|
+
emax = ap::maxreal(emax, fabs(vl(k,ki))+fabs(vl(k,ki+1)));
|
1327
|
+
}
|
1328
|
+
remax = 1/emax;
|
1329
|
+
ap::vmul(vl.getcolumn(ki, 1, n), remax);
|
1330
|
+
ap::vmul(vl.getcolumn(ki+1, 1, n), remax);
|
1331
|
+
}
|
1332
|
+
}
|
1333
|
+
iis = iis+1;
|
1334
|
+
if( ip!=0 )
|
1335
|
+
{
|
1336
|
+
iis = iis+1;
|
1337
|
+
}
|
1338
|
+
}
|
1339
|
+
if( ip==-1 )
|
1340
|
+
{
|
1341
|
+
ip = 0;
|
1342
|
+
}
|
1343
|
+
if( ip==1 )
|
1344
|
+
{
|
1345
|
+
ip = -1;
|
1346
|
+
}
|
1347
|
+
}
|
1348
|
+
}
|
1349
|
+
}
|
1350
|
+
|
1351
|
+
|
1352
|
+
static void internalhsevdlaln2(const bool& ltrans,
|
1353
|
+
const int& na,
|
1354
|
+
const int& nw,
|
1355
|
+
const double& smin,
|
1356
|
+
const double& ca,
|
1357
|
+
const ap::real_2d_array& a,
|
1358
|
+
const double& d1,
|
1359
|
+
const double& d2,
|
1360
|
+
const ap::real_2d_array& b,
|
1361
|
+
const double& wr,
|
1362
|
+
const double& wi,
|
1363
|
+
ap::boolean_1d_array& rswap4,
|
1364
|
+
ap::boolean_1d_array& zswap4,
|
1365
|
+
ap::integer_2d_array& ipivot44,
|
1366
|
+
ap::real_1d_array& civ4,
|
1367
|
+
ap::real_1d_array& crv4,
|
1368
|
+
ap::real_2d_array& x,
|
1369
|
+
double& scl,
|
1370
|
+
double& xnorm,
|
1371
|
+
int& info)
|
1372
|
+
{
|
1373
|
+
int icmax;
|
1374
|
+
int j;
|
1375
|
+
double bbnd;
|
1376
|
+
double bi1;
|
1377
|
+
double bi2;
|
1378
|
+
double bignum;
|
1379
|
+
double bnorm;
|
1380
|
+
double br1;
|
1381
|
+
double br2;
|
1382
|
+
double ci21;
|
1383
|
+
double ci22;
|
1384
|
+
double cmax;
|
1385
|
+
double cnorm;
|
1386
|
+
double cr21;
|
1387
|
+
double cr22;
|
1388
|
+
double csi;
|
1389
|
+
double csr;
|
1390
|
+
double li21;
|
1391
|
+
double lr21;
|
1392
|
+
double smini;
|
1393
|
+
double smlnum;
|
1394
|
+
double temp;
|
1395
|
+
double u22abs;
|
1396
|
+
double ui11;
|
1397
|
+
double ui11r;
|
1398
|
+
double ui12;
|
1399
|
+
double ui12s;
|
1400
|
+
double ui22;
|
1401
|
+
double ur11;
|
1402
|
+
double ur11r;
|
1403
|
+
double ur12;
|
1404
|
+
double ur12s;
|
1405
|
+
double ur22;
|
1406
|
+
double xi1;
|
1407
|
+
double xi2;
|
1408
|
+
double xr1;
|
1409
|
+
double xr2;
|
1410
|
+
double tmp1;
|
1411
|
+
double tmp2;
|
1412
|
+
|
1413
|
+
zswap4(1) = false;
|
1414
|
+
zswap4(2) = false;
|
1415
|
+
zswap4(3) = true;
|
1416
|
+
zswap4(4) = true;
|
1417
|
+
rswap4(1) = false;
|
1418
|
+
rswap4(2) = true;
|
1419
|
+
rswap4(3) = false;
|
1420
|
+
rswap4(4) = true;
|
1421
|
+
ipivot44(1,1) = 1;
|
1422
|
+
ipivot44(2,1) = 2;
|
1423
|
+
ipivot44(3,1) = 3;
|
1424
|
+
ipivot44(4,1) = 4;
|
1425
|
+
ipivot44(1,2) = 2;
|
1426
|
+
ipivot44(2,2) = 1;
|
1427
|
+
ipivot44(3,2) = 4;
|
1428
|
+
ipivot44(4,2) = 3;
|
1429
|
+
ipivot44(1,3) = 3;
|
1430
|
+
ipivot44(2,3) = 4;
|
1431
|
+
ipivot44(3,3) = 1;
|
1432
|
+
ipivot44(4,3) = 2;
|
1433
|
+
ipivot44(1,4) = 4;
|
1434
|
+
ipivot44(2,4) = 3;
|
1435
|
+
ipivot44(3,4) = 2;
|
1436
|
+
ipivot44(4,4) = 1;
|
1437
|
+
smlnum = 2*ap::minrealnumber;
|
1438
|
+
bignum = 1/smlnum;
|
1439
|
+
smini = ap::maxreal(smin, smlnum);
|
1440
|
+
|
1441
|
+
//
|
1442
|
+
// Don't check for input errors
|
1443
|
+
//
|
1444
|
+
info = 0;
|
1445
|
+
|
1446
|
+
//
|
1447
|
+
// Standard Initializations
|
1448
|
+
//
|
1449
|
+
scl = 1;
|
1450
|
+
if( na==1 )
|
1451
|
+
{
|
1452
|
+
|
1453
|
+
//
|
1454
|
+
// 1 x 1 (i.e., scalar) system C X = B
|
1455
|
+
//
|
1456
|
+
if( nw==1 )
|
1457
|
+
{
|
1458
|
+
|
1459
|
+
//
|
1460
|
+
// Real 1x1 system.
|
1461
|
+
//
|
1462
|
+
// C = ca A - w D
|
1463
|
+
//
|
1464
|
+
csr = ca*a(1,1)-wr*d1;
|
1465
|
+
cnorm = fabs(csr);
|
1466
|
+
|
1467
|
+
//
|
1468
|
+
// If | C | < SMINI, use C = SMINI
|
1469
|
+
//
|
1470
|
+
if( cnorm<smini )
|
1471
|
+
{
|
1472
|
+
csr = smini;
|
1473
|
+
cnorm = smini;
|
1474
|
+
info = 1;
|
1475
|
+
}
|
1476
|
+
|
1477
|
+
//
|
1478
|
+
// Check scaling for X = B / C
|
1479
|
+
//
|
1480
|
+
bnorm = fabs(b(1,1));
|
1481
|
+
if( cnorm<1&&bnorm>1 )
|
1482
|
+
{
|
1483
|
+
if( bnorm>bignum*cnorm )
|
1484
|
+
{
|
1485
|
+
scl = 1/bnorm;
|
1486
|
+
}
|
1487
|
+
}
|
1488
|
+
|
1489
|
+
//
|
1490
|
+
// Compute X
|
1491
|
+
//
|
1492
|
+
x(1,1) = b(1,1)*scl/csr;
|
1493
|
+
xnorm = fabs(x(1,1));
|
1494
|
+
}
|
1495
|
+
else
|
1496
|
+
{
|
1497
|
+
|
1498
|
+
//
|
1499
|
+
// Complex 1x1 system (w is complex)
|
1500
|
+
//
|
1501
|
+
// C = ca A - w D
|
1502
|
+
//
|
1503
|
+
csr = ca*a(1,1)-wr*d1;
|
1504
|
+
csi = -wi*d1;
|
1505
|
+
cnorm = fabs(csr)+fabs(csi);
|
1506
|
+
|
1507
|
+
//
|
1508
|
+
// If | C | < SMINI, use C = SMINI
|
1509
|
+
//
|
1510
|
+
if( cnorm<smini )
|
1511
|
+
{
|
1512
|
+
csr = smini;
|
1513
|
+
csi = 0;
|
1514
|
+
cnorm = smini;
|
1515
|
+
info = 1;
|
1516
|
+
}
|
1517
|
+
|
1518
|
+
//
|
1519
|
+
// Check scaling for X = B / C
|
1520
|
+
//
|
1521
|
+
bnorm = fabs(b(1,1))+fabs(b(1,2));
|
1522
|
+
if( cnorm<1&&bnorm>1 )
|
1523
|
+
{
|
1524
|
+
if( bnorm>bignum*cnorm )
|
1525
|
+
{
|
1526
|
+
scl = 1/bnorm;
|
1527
|
+
}
|
1528
|
+
}
|
1529
|
+
|
1530
|
+
//
|
1531
|
+
// Compute X
|
1532
|
+
//
|
1533
|
+
internalhsevdladiv(scl*b(1,1), scl*b(1,2), csr, csi, tmp1, tmp2);
|
1534
|
+
x(1,1) = tmp1;
|
1535
|
+
x(1,2) = tmp2;
|
1536
|
+
xnorm = fabs(x(1,1))+fabs(x(1,2));
|
1537
|
+
}
|
1538
|
+
}
|
1539
|
+
else
|
1540
|
+
{
|
1541
|
+
|
1542
|
+
//
|
1543
|
+
// 2x2 System
|
1544
|
+
//
|
1545
|
+
// Compute the real part of C = ca A - w D (or ca A' - w D )
|
1546
|
+
//
|
1547
|
+
crv4(1+0) = ca*a(1,1)-wr*d1;
|
1548
|
+
crv4(2+2) = ca*a(2,2)-wr*d2;
|
1549
|
+
if( ltrans )
|
1550
|
+
{
|
1551
|
+
crv4(1+2) = ca*a(2,1);
|
1552
|
+
crv4(2+0) = ca*a(1,2);
|
1553
|
+
}
|
1554
|
+
else
|
1555
|
+
{
|
1556
|
+
crv4(2+0) = ca*a(2,1);
|
1557
|
+
crv4(1+2) = ca*a(1,2);
|
1558
|
+
}
|
1559
|
+
if( nw==1 )
|
1560
|
+
{
|
1561
|
+
|
1562
|
+
//
|
1563
|
+
// Real 2x2 system (w is real)
|
1564
|
+
//
|
1565
|
+
// Find the largest element in C
|
1566
|
+
//
|
1567
|
+
cmax = 0;
|
1568
|
+
icmax = 0;
|
1569
|
+
for(j = 1; j <= 4; j++)
|
1570
|
+
{
|
1571
|
+
if( fabs(crv4(j))>cmax )
|
1572
|
+
{
|
1573
|
+
cmax = fabs(crv4(j));
|
1574
|
+
icmax = j;
|
1575
|
+
}
|
1576
|
+
}
|
1577
|
+
|
1578
|
+
//
|
1579
|
+
// If norm(C) < SMINI, use SMINI*identity.
|
1580
|
+
//
|
1581
|
+
if( cmax<smini )
|
1582
|
+
{
|
1583
|
+
bnorm = ap::maxreal(fabs(b(1,1)), fabs(b(2,1)));
|
1584
|
+
if( smini<1&&bnorm>1 )
|
1585
|
+
{
|
1586
|
+
if( bnorm>bignum*smini )
|
1587
|
+
{
|
1588
|
+
scl = 1/bnorm;
|
1589
|
+
}
|
1590
|
+
}
|
1591
|
+
temp = scl/smini;
|
1592
|
+
x(1,1) = temp*b(1,1);
|
1593
|
+
x(2,1) = temp*b(2,1);
|
1594
|
+
xnorm = temp*bnorm;
|
1595
|
+
info = 1;
|
1596
|
+
return;
|
1597
|
+
}
|
1598
|
+
|
1599
|
+
//
|
1600
|
+
// Gaussian elimination with complete pivoting.
|
1601
|
+
//
|
1602
|
+
ur11 = crv4(icmax);
|
1603
|
+
cr21 = crv4(ipivot44(2,icmax));
|
1604
|
+
ur12 = crv4(ipivot44(3,icmax));
|
1605
|
+
cr22 = crv4(ipivot44(4,icmax));
|
1606
|
+
ur11r = 1/ur11;
|
1607
|
+
lr21 = ur11r*cr21;
|
1608
|
+
ur22 = cr22-ur12*lr21;
|
1609
|
+
|
1610
|
+
//
|
1611
|
+
// If smaller pivot < SMINI, use SMINI
|
1612
|
+
//
|
1613
|
+
if( fabs(ur22)<smini )
|
1614
|
+
{
|
1615
|
+
ur22 = smini;
|
1616
|
+
info = 1;
|
1617
|
+
}
|
1618
|
+
if( rswap4(icmax) )
|
1619
|
+
{
|
1620
|
+
br1 = b(2,1);
|
1621
|
+
br2 = b(1,1);
|
1622
|
+
}
|
1623
|
+
else
|
1624
|
+
{
|
1625
|
+
br1 = b(1,1);
|
1626
|
+
br2 = b(2,1);
|
1627
|
+
}
|
1628
|
+
br2 = br2-lr21*br1;
|
1629
|
+
bbnd = ap::maxreal(fabs(br1*(ur22*ur11r)), fabs(br2));
|
1630
|
+
if( bbnd>1&&fabs(ur22)<1 )
|
1631
|
+
{
|
1632
|
+
if( bbnd>=bignum*fabs(ur22) )
|
1633
|
+
{
|
1634
|
+
scl = 1/bbnd;
|
1635
|
+
}
|
1636
|
+
}
|
1637
|
+
xr2 = br2*scl/ur22;
|
1638
|
+
xr1 = scl*br1*ur11r-xr2*(ur11r*ur12);
|
1639
|
+
if( zswap4(icmax) )
|
1640
|
+
{
|
1641
|
+
x(1,1) = xr2;
|
1642
|
+
x(2,1) = xr1;
|
1643
|
+
}
|
1644
|
+
else
|
1645
|
+
{
|
1646
|
+
x(1,1) = xr1;
|
1647
|
+
x(2,1) = xr2;
|
1648
|
+
}
|
1649
|
+
xnorm = ap::maxreal(fabs(xr1), fabs(xr2));
|
1650
|
+
|
1651
|
+
//
|
1652
|
+
// Further scaling if norm(A) norm(X) > overflow
|
1653
|
+
//
|
1654
|
+
if( xnorm>1&&cmax>1 )
|
1655
|
+
{
|
1656
|
+
if( xnorm>bignum/cmax )
|
1657
|
+
{
|
1658
|
+
temp = cmax/bignum;
|
1659
|
+
x(1,1) = temp*x(1,1);
|
1660
|
+
x(2,1) = temp*x(2,1);
|
1661
|
+
xnorm = temp*xnorm;
|
1662
|
+
scl = temp*scl;
|
1663
|
+
}
|
1664
|
+
}
|
1665
|
+
}
|
1666
|
+
else
|
1667
|
+
{
|
1668
|
+
|
1669
|
+
//
|
1670
|
+
// Complex 2x2 system (w is complex)
|
1671
|
+
//
|
1672
|
+
// Find the largest element in C
|
1673
|
+
//
|
1674
|
+
civ4(1+0) = -wi*d1;
|
1675
|
+
civ4(2+0) = 0;
|
1676
|
+
civ4(1+2) = 0;
|
1677
|
+
civ4(2+2) = -wi*d2;
|
1678
|
+
cmax = 0;
|
1679
|
+
icmax = 0;
|
1680
|
+
for(j = 1; j <= 4; j++)
|
1681
|
+
{
|
1682
|
+
if( fabs(crv4(j))+fabs(civ4(j))>cmax )
|
1683
|
+
{
|
1684
|
+
cmax = fabs(crv4(j))+fabs(civ4(j));
|
1685
|
+
icmax = j;
|
1686
|
+
}
|
1687
|
+
}
|
1688
|
+
|
1689
|
+
//
|
1690
|
+
// If norm(C) < SMINI, use SMINI*identity.
|
1691
|
+
//
|
1692
|
+
if( cmax<smini )
|
1693
|
+
{
|
1694
|
+
bnorm = ap::maxreal(fabs(b(1,1))+fabs(b(1,2)), fabs(b(2,1))+fabs(b(2,2)));
|
1695
|
+
if( smini<1&&bnorm>1 )
|
1696
|
+
{
|
1697
|
+
if( bnorm>bignum*smini )
|
1698
|
+
{
|
1699
|
+
scl = 1/bnorm;
|
1700
|
+
}
|
1701
|
+
}
|
1702
|
+
temp = scl/smini;
|
1703
|
+
x(1,1) = temp*b(1,1);
|
1704
|
+
x(2,1) = temp*b(2,1);
|
1705
|
+
x(1,2) = temp*b(1,2);
|
1706
|
+
x(2,2) = temp*b(2,2);
|
1707
|
+
xnorm = temp*bnorm;
|
1708
|
+
info = 1;
|
1709
|
+
return;
|
1710
|
+
}
|
1711
|
+
|
1712
|
+
//
|
1713
|
+
// Gaussian elimination with complete pivoting.
|
1714
|
+
//
|
1715
|
+
ur11 = crv4(icmax);
|
1716
|
+
ui11 = civ4(icmax);
|
1717
|
+
cr21 = crv4(ipivot44(2,icmax));
|
1718
|
+
ci21 = civ4(ipivot44(2,icmax));
|
1719
|
+
ur12 = crv4(ipivot44(3,icmax));
|
1720
|
+
ui12 = civ4(ipivot44(3,icmax));
|
1721
|
+
cr22 = crv4(ipivot44(4,icmax));
|
1722
|
+
ci22 = civ4(ipivot44(4,icmax));
|
1723
|
+
if( icmax==1||icmax==4 )
|
1724
|
+
{
|
1725
|
+
|
1726
|
+
//
|
1727
|
+
// Code when off-diagonals of pivoted C are real
|
1728
|
+
//
|
1729
|
+
if( fabs(ur11)>fabs(ui11) )
|
1730
|
+
{
|
1731
|
+
temp = ui11/ur11;
|
1732
|
+
ur11r = 1/(ur11*(1+ap::sqr(temp)));
|
1733
|
+
ui11r = -temp*ur11r;
|
1734
|
+
}
|
1735
|
+
else
|
1736
|
+
{
|
1737
|
+
temp = ur11/ui11;
|
1738
|
+
ui11r = -1/(ui11*(1+ap::sqr(temp)));
|
1739
|
+
ur11r = -temp*ui11r;
|
1740
|
+
}
|
1741
|
+
lr21 = cr21*ur11r;
|
1742
|
+
li21 = cr21*ui11r;
|
1743
|
+
ur12s = ur12*ur11r;
|
1744
|
+
ui12s = ur12*ui11r;
|
1745
|
+
ur22 = cr22-ur12*lr21;
|
1746
|
+
ui22 = ci22-ur12*li21;
|
1747
|
+
}
|
1748
|
+
else
|
1749
|
+
{
|
1750
|
+
|
1751
|
+
//
|
1752
|
+
// Code when diagonals of pivoted C are real
|
1753
|
+
//
|
1754
|
+
ur11r = 1/ur11;
|
1755
|
+
ui11r = 0;
|
1756
|
+
lr21 = cr21*ur11r;
|
1757
|
+
li21 = ci21*ur11r;
|
1758
|
+
ur12s = ur12*ur11r;
|
1759
|
+
ui12s = ui12*ur11r;
|
1760
|
+
ur22 = cr22-ur12*lr21+ui12*li21;
|
1761
|
+
ui22 = -ur12*li21-ui12*lr21;
|
1762
|
+
}
|
1763
|
+
u22abs = fabs(ur22)+fabs(ui22);
|
1764
|
+
|
1765
|
+
//
|
1766
|
+
// If smaller pivot < SMINI, use SMINI
|
1767
|
+
//
|
1768
|
+
if( u22abs<smini )
|
1769
|
+
{
|
1770
|
+
ur22 = smini;
|
1771
|
+
ui22 = 0;
|
1772
|
+
info = 1;
|
1773
|
+
}
|
1774
|
+
if( rswap4(icmax) )
|
1775
|
+
{
|
1776
|
+
br2 = b(1,1);
|
1777
|
+
br1 = b(2,1);
|
1778
|
+
bi2 = b(1,2);
|
1779
|
+
bi1 = b(2,2);
|
1780
|
+
}
|
1781
|
+
else
|
1782
|
+
{
|
1783
|
+
br1 = b(1,1);
|
1784
|
+
br2 = b(2,1);
|
1785
|
+
bi1 = b(1,2);
|
1786
|
+
bi2 = b(2,2);
|
1787
|
+
}
|
1788
|
+
br2 = br2-lr21*br1+li21*bi1;
|
1789
|
+
bi2 = bi2-li21*br1-lr21*bi1;
|
1790
|
+
bbnd = ap::maxreal((fabs(br1)+fabs(bi1))*(u22abs*(fabs(ur11r)+fabs(ui11r))), fabs(br2)+fabs(bi2));
|
1791
|
+
if( bbnd>1&&u22abs<1 )
|
1792
|
+
{
|
1793
|
+
if( bbnd>=bignum*u22abs )
|
1794
|
+
{
|
1795
|
+
scl = 1/bbnd;
|
1796
|
+
br1 = scl*br1;
|
1797
|
+
bi1 = scl*bi1;
|
1798
|
+
br2 = scl*br2;
|
1799
|
+
bi2 = scl*bi2;
|
1800
|
+
}
|
1801
|
+
}
|
1802
|
+
internalhsevdladiv(br2, bi2, ur22, ui22, xr2, xi2);
|
1803
|
+
xr1 = ur11r*br1-ui11r*bi1-ur12s*xr2+ui12s*xi2;
|
1804
|
+
xi1 = ui11r*br1+ur11r*bi1-ui12s*xr2-ur12s*xi2;
|
1805
|
+
if( zswap4(icmax) )
|
1806
|
+
{
|
1807
|
+
x(1,1) = xr2;
|
1808
|
+
x(2,1) = xr1;
|
1809
|
+
x(1,2) = xi2;
|
1810
|
+
x(2,2) = xi1;
|
1811
|
+
}
|
1812
|
+
else
|
1813
|
+
{
|
1814
|
+
x(1,1) = xr1;
|
1815
|
+
x(2,1) = xr2;
|
1816
|
+
x(1,2) = xi1;
|
1817
|
+
x(2,2) = xi2;
|
1818
|
+
}
|
1819
|
+
xnorm = ap::maxreal(fabs(xr1)+fabs(xi1), fabs(xr2)+fabs(xi2));
|
1820
|
+
|
1821
|
+
//
|
1822
|
+
// Further scaling if norm(A) norm(X) > overflow
|
1823
|
+
//
|
1824
|
+
if( xnorm>1&&cmax>1 )
|
1825
|
+
{
|
1826
|
+
if( xnorm>bignum/cmax )
|
1827
|
+
{
|
1828
|
+
temp = cmax/bignum;
|
1829
|
+
x(1,1) = temp*x(1,1);
|
1830
|
+
x(2,1) = temp*x(2,1);
|
1831
|
+
x(1,2) = temp*x(1,2);
|
1832
|
+
x(2,2) = temp*x(2,2);
|
1833
|
+
xnorm = temp*xnorm;
|
1834
|
+
scl = temp*scl;
|
1835
|
+
}
|
1836
|
+
}
|
1837
|
+
}
|
1838
|
+
}
|
1839
|
+
}
|
1840
|
+
|
1841
|
+
|
1842
|
+
static void internalhsevdladiv(const double& a,
|
1843
|
+
const double& b,
|
1844
|
+
const double& c,
|
1845
|
+
const double& d,
|
1846
|
+
double& p,
|
1847
|
+
double& q)
|
1848
|
+
{
|
1849
|
+
double e;
|
1850
|
+
double f;
|
1851
|
+
|
1852
|
+
if( fabs(d)<fabs(c) )
|
1853
|
+
{
|
1854
|
+
e = d/c;
|
1855
|
+
f = c+d*e;
|
1856
|
+
p = (a+b*e)/f;
|
1857
|
+
q = (b-a*e)/f;
|
1858
|
+
}
|
1859
|
+
else
|
1860
|
+
{
|
1861
|
+
e = c/d;
|
1862
|
+
f = d+c*e;
|
1863
|
+
p = (b+a*e)/f;
|
1864
|
+
q = (-a+b*e)/f;
|
1865
|
+
}
|
1866
|
+
}
|
1867
|
+
|
1868
|
+
|
1869
|
+
|