xref: /petsc/src/mat/tests/ex118.c (revision 21e3ffae2f3b73c0bd738cf6d0a809700fc04bb0)
1 static char help[] = "Test LAPACK routine DSTEBZ() and DTEIN().  \n\n";
2 
3 #include <petscmat.h>
4 #include <petscblaslapack.h>
5 
6 extern PetscErrorCode CkEigenSolutions(PetscInt, Mat, PetscInt, PetscInt, PetscScalar *, Vec *, PetscReal *);
7 
8 int main(int argc, char **args)
9 {
10 #if defined(PETSC_USE_COMPLEX) || defined(PETSC_MISSING_LAPACK_STEBZ) || defined(PETSC_MISSING_LAPACK_STEIN)
11   PetscFunctionBeginUser;
12   PetscCall(PetscInitialize(&argc, &args, (char *)0, help));
13   SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_SUP_SYS, "This example requires LAPACK routines dstebz and stien and real numbers");
14 #else
15   PetscReal   *work, tols[2];
16   PetscInt     i, j;
17   PetscBLASInt n, il = 1, iu = 5, *iblock, *isplit, *iwork, nevs, *ifail, cklvl = 2;
18   PetscMPIInt  size;
19   PetscBool    flg;
20   Vec         *evecs;
21   PetscScalar *evecs_array, *D, *E, *evals;
22   Mat          T;
23   PetscReal    vl = 0.0, vu = 4.0, tol = 1000 * PETSC_MACHINE_EPSILON;
24   PetscBLASInt nsplit, info;
25 
26   PetscFunctionBeginUser;
27   PetscCall(PetscInitialize(&argc, &args, (char *)0, help));
28   PetscCallMPI(MPI_Comm_size(PETSC_COMM_WORLD, &size));
29   PetscCheck(size == 1, PETSC_COMM_WORLD, PETSC_ERR_WRONG_MPI_SIZE, "This is a uniprocessor example only!");
30 
31   n    = 100;
32   nevs = iu - il;
33   PetscCall(PetscMalloc1(3 * n + 1, &D));
34   E     = D + n;
35   evals = E + n;
36   PetscCall(PetscMalloc1(5 * n + 1, &work));
37   PetscCall(PetscMalloc1(3 * n + 1, &iwork));
38   PetscCall(PetscMalloc1(3 * n + 1, &iblock));
39   isplit = iblock + n;
40 
41   /* Set symmetric tridiagonal matrix */
42   for (i = 0; i < n; i++) {
43     D[i] = 2.0;
44     E[i] = 1.0;
45   }
46 
47   /* Solve eigenvalue problem: A*evec = eval*evec */
48   PetscCall(PetscPrintf(PETSC_COMM_SELF, " LAPACKstebz_: compute %d eigenvalues...\n", nevs));
49   LAPACKstebz_("I", "E", &n, &vl, &vu, &il, &iu, &tol, (PetscReal *)D, (PetscReal *)E, &nevs, &nsplit, (PetscReal *)evals, iblock, isplit, work, iwork, &info);
50   PetscCheck(!info, PETSC_COMM_SELF, PETSC_ERR_USER, "LAPACKstebz_ fails. info %d", info);
51 
52   PetscCall(PetscPrintf(PETSC_COMM_SELF, " LAPACKstein_: compute %d found eigenvectors...\n", nevs));
53   PetscCall(PetscMalloc1(n * nevs, &evecs_array));
54   PetscCall(PetscMalloc1(nevs, &ifail));
55   LAPACKstein_(&n, (PetscReal *)D, (PetscReal *)E, &nevs, (PetscReal *)evals, iblock, isplit, evecs_array, &n, work, iwork, ifail, &info);
56   PetscCheck(!info, PETSC_COMM_SELF, PETSC_ERR_USER, "LAPACKstein_ fails. info %d", info);
57   /* View evals */
58   PetscCall(PetscOptionsHasName(NULL, NULL, "-eig_view", &flg));
59   if (flg) {
60     PetscCall(PetscPrintf(PETSC_COMM_SELF, " %d evals: \n", nevs));
61     for (i = 0; i < nevs; i++) PetscCall(PetscPrintf(PETSC_COMM_SELF, "%" PetscInt_FMT "  %g\n", i, (double)evals[i]));
62   }
63 
64   /* Check residuals and orthogonality */
65   PetscCall(MatCreate(PETSC_COMM_SELF, &T));
66   PetscCall(MatSetSizes(T, PETSC_DECIDE, PETSC_DECIDE, n, n));
67   PetscCall(MatSetType(T, MATSBAIJ));
68   PetscCall(MatSetFromOptions(T));
69   PetscCall(MatSetUp(T));
70   for (i = 0; i < n; i++) {
71     PetscCall(MatSetValues(T, 1, &i, 1, &i, &D[i], INSERT_VALUES));
72     if (i != n - 1) {
73       j = i + 1;
74       PetscCall(MatSetValues(T, 1, &i, 1, &j, &E[i], INSERT_VALUES));
75     }
76   }
77   PetscCall(MatAssemblyBegin(T, MAT_FINAL_ASSEMBLY));
78   PetscCall(MatAssemblyEnd(T, MAT_FINAL_ASSEMBLY));
79 
80   PetscCall(PetscMalloc1(nevs + 1, &evecs));
81   for (i = 0; i < nevs; i++) {
82     PetscCall(VecCreate(PETSC_COMM_SELF, &evecs[i]));
83     PetscCall(VecSetSizes(evecs[i], PETSC_DECIDE, n));
84     PetscCall(VecSetFromOptions(evecs[i]));
85     PetscCall(VecPlaceArray(evecs[i], evecs_array + i * n));
86   }
87 
88   tols[0] = 1.e-8;
89   tols[1] = 1.e-8;
90   PetscCall(CkEigenSolutions(cklvl, T, il - 1, iu - 1, evals, evecs, tols));
91 
92   for (i = 0; i < nevs; i++) PetscCall(VecResetArray(evecs[i]));
93 
94   /* free space */
95 
96   PetscCall(MatDestroy(&T));
97 
98   for (i = 0; i < nevs; i++) PetscCall(VecDestroy(&evecs[i]));
99   PetscCall(PetscFree(evecs));
100   PetscCall(PetscFree(D));
101   PetscCall(PetscFree(work));
102   PetscCall(PetscFree(iwork));
103   PetscCall(PetscFree(iblock));
104   PetscCall(PetscFree(evecs_array));
105   PetscCall(PetscFree(ifail));
106   PetscCall(PetscFinalize());
107   return 0;
108 #endif
109 }
110 /*------------------------------------------------
111   Check the accuracy of the eigen solution
112   ----------------------------------------------- */
113 /*
114   input:
115      cklvl      - check level:
116                     1: check residual
117                     2: 1 and check B-orthogonality locally
118      A          - matrix
119      il,iu      - lower and upper index bound of eigenvalues
120      eval, evec - eigenvalues and eigenvectors stored in this process
121      tols[0]    - reporting tol_res: || A * evec[i] - eval[i]*evec[i] ||
122      tols[1]    - reporting tol_orth: evec[i]^T*evec[j] - delta_ij
123 */
124 #undef DEBUG_CkEigenSolutions
125 PetscErrorCode CkEigenSolutions(PetscInt cklvl, Mat A, PetscInt il, PetscInt iu, PetscScalar *eval, Vec *evec, PetscReal *tols)
126 {
127   PetscInt    ierr, i, j, nev;
128   Vec         vt1, vt2; /* tmp vectors */
129   PetscReal   norm, norm_max;
130   PetscScalar dot, tmp;
131   PetscReal   dot_max;
132 
133   PetscFunctionBegin;
134   nev = iu - il;
135   if (nev <= 0) PetscFunctionReturn(PETSC_SUCCESS);
136 
137   PetscCall(VecDuplicate(evec[0], &vt1));
138   PetscCall(VecDuplicate(evec[0], &vt2));
139 
140   switch (cklvl) {
141   case 2:
142     dot_max = 0.0;
143     for (i = il; i < iu; i++) {
144       PetscCall(VecCopy(evec[i], vt1));
145       for (j = il; j < iu; j++) {
146         PetscCall(VecDot(evec[j], vt1, &dot));
147         if (j == i) {
148           dot = PetscAbsScalar(dot - (PetscScalar)1.0);
149         } else {
150           dot = PetscAbsScalar(dot);
151         }
152         if (PetscAbsScalar(dot) > dot_max) dot_max = PetscAbsScalar(dot);
153 #if defined(DEBUG_CkEigenSolutions)
154         if (dot > tols[1]) {
155           PetscCall(VecNorm(evec[i], NORM_INFINITY, &norm));
156           PetscCall(PetscPrintf(PETSC_COMM_SELF, "|delta(%d,%d)|: %g, norm: %d\n", i, j, (double)dot, (double)norm));
157         }
158 #endif
159       }
160     }
161     PetscCall(PetscPrintf(PETSC_COMM_SELF, "    max|(x_j^T*x_i) - delta_ji|: %g\n", (double)dot_max));
162 
163   case 1:
164     norm_max = 0.0;
165     for (i = il; i < iu; i++) {
166       PetscCall(MatMult(A, evec[i], vt1));
167       PetscCall(VecCopy(evec[i], vt2));
168       tmp = -eval[i];
169       PetscCall(VecAXPY(vt1, tmp, vt2));
170       PetscCall(VecNorm(vt1, NORM_INFINITY, &norm));
171       norm = PetscAbsReal(norm);
172       if (norm > norm_max) norm_max = norm;
173 #if defined(DEBUG_CkEigenSolutions)
174       if (norm > tols[0]) PetscCall(PetscPrintf(PETSC_COMM_SELF, "  residual violation: %d, resi: %g\n", i, norm));
175 #endif
176     }
177     PetscCall(PetscPrintf(PETSC_COMM_SELF, "    max_resi:                    %g\n", (double)norm_max));
178     break;
179   default:
180     PetscCall(PetscPrintf(PETSC_COMM_SELF, "Error: cklvl=%d is not supported \n", cklvl));
181   }
182 
183   PetscCall(VecDestroy(&vt2));
184   PetscCall(VecDestroy(&vt1));
185   PetscFunctionReturn(PETSC_SUCCESS);
186 }
187