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
#include "Matrix.h"
#include "DenseMatrix.h"
#include "PolynomialSolver.h"
namespace ATC_matrix {
//-----------------------------------------------------------------------------
//* performs a matrix-matrix multiply with optional transposes BLAS version
// C = b*C + a*A*B
//-----------------------------------------------------------------------------
void MultAB(const MATRIX &A, const MATRIX &B, DENS_MAT &C,
const bool At, const bool Bt, double a, double b)
{
static char t[2] = {'N','T'};
char *ta=t+At, *tb=t+Bt;
int sA[2] = {A.nRows(), A.nCols()}; // sizes of A
int sB[2] = {B.nRows(), B.nCols()}; // sizes of B
GCK(A, B, sA[!At]!=sB[Bt], "MultAB<double>: matrix-matrix multiply");
if (!C.is_size(sA[At],sB[!Bt]))
{
C.resize(sA[At],sB[!Bt]);
if (b != 0.0) C.zero();
}
// get pointers to the matrix sizes needed by BLAS
int *M = sA+At; // # of rows in op[A] (op[A] = A' if At='T' else A)
int *N = sB+!Bt; // # of cols in op[B]
int *K = sA+!At; // # of cols in op[A] or # of rows in op[B]
double *pa=A.ptr(), *pb=B.ptr(), *pc=C.ptr();
#ifdef COL_STORAGE
dgemm_(ta, tb, M, N, K, &a, pa, sA, pb, sB, &b, pc, M);
#else
dgemm_(tb, ta, N, M, K, &a, pb, sB+1, pa, sA+1, &b, pc, N);
#endif
}
//-----------------------------------------------------------------------------
//* returns the inverse of a double precision matrix
//-----------------------------------------------------------------------------
DenseMatrix<double> inv(const MATRIX& A)
{
SQCK(A, "DenseMatrix::inv(), matrix not square"); // check matrix is square
DENS_MAT invA(A); // Make copy of A to invert
// setup for call to BLAS
int m, info, lwork=-1;
m = invA.nRows();
int *ipiv = new int[m<<1]; // need 2m storage
int *iwork=ipiv+m;
dgetrf_(&m,&m,invA.ptr(),&m,ipiv,&info); // compute LU factorization
GCK(A,A,info<0,"DenseMatrix::inv() dgetrf error: Argument had bad value.");
GCK(A,A,info>0,"DenseMatrix::inv() dgetrf error: Matrix not invertible.");
if (info > 0) {
delete [] ipiv;
invA = 0;
return invA;
}
// LU factorization succeeded
// Compute 1-norm of original matrix for use with dgecon
char norm = '1'; // Use 1-norm
double rcond, anorm, *workc = new double[4*m];
anorm = dlange_(&norm,&m,&m,A.ptr(),&m,workc);
// Condition number estimation (warn if bad)
dgecon_(&norm,&m,invA.ptr(),&m,&anorm,&rcond,workc,iwork,&info);
GCK(A,A,info<0, "DenseMatrix::inv(): dgecon error: Argument had bad value.");
GCK(A,A,rcond<1e-14,"DenseMatrix::inv(): Matrix nearly singular, RCOND<e-14");
// Now determine optimal work size for computation of matrix inverse
double work_dummy[2] = {0.0,0.0};
dgetri_(&m, invA.ptr(), &m, ipiv, work_dummy, &lwork, &info);
GCK(A,A,info<0,"DenseMatrix::inv() dgetri error: Argument had bad value.");
GCHK(info>0,"DenseMatrix::inv() dgetri error: Matrix not invertible.");
// Work size query succeded
lwork = (int)work_dummy[0];
double *work = new double[lwork]; // Allocate vector of appropriate size
// Compute and store matrix inverse
dgetri_(&m,invA.ptr(),&m,ipiv,work,&lwork,&info);
GCK(A,A,info<0,"DenseMatrix::inv() dgetri error: Argument had bad value.");
GCHK(info>0,"DenseMatrix::inv() dgetri error: Matrix not invertible.");
// Clean-up
delete [] ipiv;
delete [] workc;
delete [] work;
return invA;
}
//-----------------------------------------------------------------------------
//* returns all eigenvalues & e-vectors of a pair of double precision matrices
//-----------------------------------------------------------------------------
DenseMatrix<double> eigensystem(const MATRIX& AA, const MATRIX & BB,
DenseMatrix<double> & eVals, bool normalize)
{
DENS_MAT A(AA); // Make copy of A
DENS_MAT B(BB);
int m = A.nRows(); // size
eVals.resize(m,1); // eigenvectors
//A.print("A");
//B.print("B");
SQCK(A, "DenseMatrix::eigensystem(), matrix not square"); // check matrix is square
SQCK(B, "DenseMatrix::eigensystem(), matrix not square"); // check matrix is square
SSCK(A, B, "DenseMatrix::eigensystem(), not same size");// check same size
// workspace
int lwork=-1; //1+(NB+6+2*NMAX)*NMAX)
double tmp[1];
double *work = tmp;
int liwork = -1; // 3+5*NMAX
int itmp[1];
int *iwork = itmp;
// Solve the generalized symmetric eigenvalue problem
// A*x = lambda*B*x (ITYPE = 1)
// only accesses upper triangle
char vectors[] = "Vectors", upper[] = "Upper";
int type = 1, info;
// query optimal sizes
dsygvd_(&type,vectors,upper,&m,A.ptr(),&m,B.ptr(),&m,
eVals.ptr(),work,&lwork,iwork,&liwork,&info);
// returns optimal sizes LWOPT = WORK(1), LIWOPT = IWORK(1)
lwork = int(work[0]);
liwork = iwork[0];
work = new double[lwork];
iwork = new int[liwork];
dsygvd_(&type,vectors,upper,&m,A.ptr(),&m,B.ptr(),&m,
eVals.ptr(),work,&lwork,iwork,&liwork,&info);
GCK(A,B,info!=0,"DenseMatrix::eigensystem(), error");
//eVals.print("e-values");
//(A.transpose()).print("e-vectors");
// normalize
if (normalize) {
for (int j = 0; j < A.nCols(); j++) {
double scale = 0.0;
for (int i = 0; i < A.nRows(); i++) {
scale += A(i,j)*A(i,j);
}
scale = 1.0/sqrt(scale);
for (int i = 0; i < A.nRows(); i++) {
A(i,j) *= scale;
}
}
//(A.transpose()).print("normalized e-vectors");
}
delete [] work;
delete [] iwork;
//return A.transpose();
return A; // column storage
}
//-----------------------------------------------------------------------------
//* returns (1-norm) condition number
//-----------------------------------------------------------------------------
double condition_number(const MATRIX& AA)
{
DenseMatrix<double> eVals, I;
I.identity(AA.nRows());
eigensystem(AA, I, eVals);
// [1] William W. Hager, "Condition Estimates," SIAM J. Sci. Stat. Comput. 5, 1984, 311-316, 1984.
// [2] Nicholas J. Higham and Françoise Tisseur, "A Block Algorithm for Matrix 1-Norm Estimation with an Application to 1-Norm Pseudospectra, "SIAM J. Matrix Anal. Appl., Vol. 21, 1185-1201, 2000.
double max = eVals.maxabs();
double min = eVals.minabs();
return max/min;
}
//-----------------------------------------------------------------------------
//* returns polar decomposition of a square double precision matrix via SVD
//-----------------------------------------------------------------------------
DenseMatrix<double> polar_decomposition(const MATRIX& AA,
DenseMatrix<double> & rotation,
DenseMatrix<double> & stretch,
bool leftRotation)
{
DENS_MAT A(AA); // Make copy of A
SQCK(A, "DenseMatrix::polar_decomposition(), matrix not square");
int m = A.nRows(); // size
DENS_MAT D(m,1);
DENS_MAT U(m,m), VT(m,m); // left and right SVD rotations
// workspace
int lwork=-1;
double tmp[1];
double *work = tmp;
// calculate singular value decomposition A = U D V^T
char type[] = "A"; // all columns are returned
int info;
// query optimal sizes
dgesvd_(type,type,&m,&m,A.ptr(),&m,D.ptr(),
U.ptr(),&m,VT.ptr(),&m,
work,&lwork,&info); // simple: svd, div&conq: sdd
lwork = int(work[0]); // returns optimal size LWOPT = WORK(1)
work = new double[lwork];
// compute SVD
dgesvd_(type,type,&m,&m,A.ptr(),&m,D.ptr(),
U.ptr(),&m,VT.ptr(),&m,
work,&lwork,&info);
//GCK(A,B,info!=0,"DenseMatrix::polar_decomposition(), error");
GCK(A,A,info!=0,"DenseMatrix::polar_decomposition(), error");
delete [] work;
//rotation.resize(m,m);
rotation = U*VT;
// A = R' U' = (U V^T) (V D V^T)
stretch.resize(m,m);
//if (leftRotation) { stretch = (VT.transpose())*D*VT; }
if (leftRotation) {
DENS_MAT V = VT.transpose();
for (int i = 0; i < m; ++i) {
for (int j = 0; j < m; ++j) {
stretch(i,j) = V(i,j)*D(j,0);
}
}
stretch = stretch*VT;
}
// A = V' R' = (U D U^T) (U V^T)
//else { stretch = U*D*(U.transpose()); }
else {
for (int i = 0; i < m; ++i) {
for (int j = 0; j < m; ++j) {
stretch(i,j) = U(i,j)*D(j,0);
}
}
stretch = stretch*(U.transpose());
}
return D;
}
//-----------------------------------------------------------------------------
//* computes the determinant of a square matrix by LU decomposition (if n>3)
//-----------------------------------------------------------------------------
double det(const MATRIX& A)
{
static const double sign[2] = {1.0, -1.0};
SQCK(A, "Matrix::det(), matrix not square"); // check matrix is square
int m = A.nRows();
switch (m) // explicit determinant for small matrix sizes
{
case 1: return A(0,0);
case 2: return A(0,0)*A(1,1)-A(0,1)*A(1,0);
case 3:
return A(0,0)*(A(1,1)*A(2,2)-A(1,2)*A(2,1))
+ A(0,1)*(A(1,2)*A(2,0)-A(1,0)*A(2,2))
+ A(0,2)*(A(1,0)*A(2,1)-A(1,1)*A(2,0));
default: break;
}
// First compute LU factorization
int info, *ipiv = new int[m];
double det = 1.0;
DENS_MAT PLUA(A);
dgetrf_(&m,&m,PLUA.ptr(),&m,ipiv,&info);
GCK(A,A,info>0,"Matrix::det() dgetrf error: Bad argument value.");
if (!info) // matrix is non-singular
{
// Compute det(A) = det(P)*det(L)*det(U) = +/-1 * det(U)
int i, OddNumPivots;
det = PLUA(0,0);
OddNumPivots = ipiv[0]!=(1);
for(i=1; i<m; i++)
{
det *= PLUA(i,i);
OddNumPivots += (ipiv[i]!=(i+1)); // # pivots even/odd
}
det *= sign[OddNumPivots&1];
}
delete [] ipiv; // Clean-up
return det;
}
//-----------------------------------------------------------------------------
//* Returns the maximum eigenvalue of a matrix.
//-----------------------------------------------------------------------------
double max_eigenvalue(const Matrix<double>& A)
{
GCK(A,A,!A.is_size(3,3), "max_eigenvalue only implimented for 3x3");
const double c0 = det(A);
const double c1 = A(1,0)*A(0,1) + A(2,0)*A(0,2) + A(1,2)*A(2,1)
- A(0,0)*A(1,1) - A(0,0)*A(2,2) - A(1,1)*A(2,2);
const double c2 = trace(A);
double c[4] = {c0, c1, c2, -1.0}, x[3];
int num_roots = ATC::solve_cubic(c, x);
double max_root = 0.0;
for (int i=0; i<num_roots; i++) max_root = std::max(x[i], max_root);
return max_root;
}
} // end namescape