Actual source code: pounders.c
1: #include <../src/tao/leastsquares/impls/pounders/pounders.h>
3: static PetscErrorCode pounders_h(Tao subtao, Vec v, Mat H, Mat Hpre, void *ctx)
4: {
5: PetscFunctionBegin;
6: PetscFunctionReturn(PETSC_SUCCESS);
7: }
9: static PetscErrorCode pounders_fg(Tao subtao, Vec x, PetscReal *f, Vec g, void *ctx)
10: {
11: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)ctx;
12: PetscReal d1, d2;
14: PetscFunctionBegin;
15: /* g = A*x (add b later)*/
16: PetscCall(MatMult(mfqP->subH, x, g));
18: /* f = 1/2 * x'*(Ax) + b'*x */
19: PetscCall(VecDot(x, g, &d1));
20: PetscCall(VecDot(mfqP->subb, x, &d2));
21: *f = 0.5 * d1 + d2;
23: /* now g = g + b */
24: PetscCall(VecAXPY(g, 1.0, mfqP->subb));
25: PetscFunctionReturn(PETSC_SUCCESS);
26: }
28: static PetscErrorCode pounders_feval(Tao tao, Vec x, Vec F, PetscReal *fsum)
29: {
30: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
31: PetscInt i, row, col;
32: PetscReal fr, fc;
34: PetscFunctionBegin;
35: PetscCall(TaoComputeResidual(tao, x, F));
36: if (tao->res_weights_v) {
37: PetscCall(VecPointwiseMult(mfqP->workfvec, tao->res_weights_v, F));
38: PetscCall(VecDot(mfqP->workfvec, mfqP->workfvec, fsum));
39: } else if (tao->res_weights_w) {
40: *fsum = 0;
41: for (i = 0; i < tao->res_weights_n; i++) {
42: row = tao->res_weights_rows[i];
43: col = tao->res_weights_cols[i];
44: PetscCall(VecGetValues(F, 1, &row, &fr));
45: PetscCall(VecGetValues(F, 1, &col, &fc));
46: *fsum += tao->res_weights_w[i] * fc * fr;
47: }
48: } else {
49: PetscCall(VecDot(F, F, fsum));
50: }
51: PetscCall(PetscInfo(tao, "Least-squares residual norm: %20.19e\n", (double)*fsum));
52: PetscCheck(!PetscIsInfOrNanReal(*fsum), PETSC_COMM_SELF, PETSC_ERR_USER, "User provided compute function generated Inf or NaN");
53: PetscFunctionReturn(PETSC_SUCCESS);
54: }
56: static PetscErrorCode gqtwrap(Tao tao, PetscReal *gnorm, PetscReal *qmin)
57: {
58: #if defined(PETSC_USE_REAL_SINGLE)
59: PetscReal atol = 1.0e-5;
60: #else
61: PetscReal atol = 1.0e-10;
62: #endif
63: PetscInt info, its;
64: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
66: PetscFunctionBegin;
67: if (!mfqP->usegqt) {
68: PetscReal maxval;
69: PetscInt i, j;
71: PetscCall(VecSetValues(mfqP->subb, mfqP->n, mfqP->indices, mfqP->Gres, INSERT_VALUES));
72: PetscCall(VecAssemblyBegin(mfqP->subb));
73: PetscCall(VecAssemblyEnd(mfqP->subb));
75: PetscCall(VecSet(mfqP->subx, 0.0));
77: PetscCall(VecSet(mfqP->subndel, -1.0));
78: PetscCall(VecSet(mfqP->subpdel, +1.0));
80: /* Complete the lower triangle of the Hessian matrix */
81: for (i = 0; i < mfqP->n; i++) {
82: for (j = i + 1; j < mfqP->n; j++) mfqP->Hres[j + mfqP->n * i] = mfqP->Hres[mfqP->n * j + i];
83: }
84: PetscCall(MatSetValues(mfqP->subH, mfqP->n, mfqP->indices, mfqP->n, mfqP->indices, mfqP->Hres, INSERT_VALUES));
85: PetscCall(MatAssemblyBegin(mfqP->subH, MAT_FINAL_ASSEMBLY));
86: PetscCall(MatAssemblyEnd(mfqP->subH, MAT_FINAL_ASSEMBLY));
88: PetscCall(TaoResetStatistics(mfqP->subtao));
89: /* PetscCall(TaoSetTolerances(mfqP->subtao,*gnorm,*gnorm,PETSC_DEFAULT)); */
90: /* enforce bound constraints -- experimental */
91: if (tao->XU && tao->XL) {
92: PetscCall(VecCopy(tao->XU, mfqP->subxu));
93: PetscCall(VecAXPY(mfqP->subxu, -1.0, tao->solution));
94: PetscCall(VecScale(mfqP->subxu, 1.0 / mfqP->delta));
95: PetscCall(VecCopy(tao->XL, mfqP->subxl));
96: PetscCall(VecAXPY(mfqP->subxl, -1.0, tao->solution));
97: PetscCall(VecScale(mfqP->subxl, 1.0 / mfqP->delta));
99: PetscCall(VecPointwiseMin(mfqP->subxu, mfqP->subxu, mfqP->subpdel));
100: PetscCall(VecPointwiseMax(mfqP->subxl, mfqP->subxl, mfqP->subndel));
101: } else {
102: PetscCall(VecCopy(mfqP->subpdel, mfqP->subxu));
103: PetscCall(VecCopy(mfqP->subndel, mfqP->subxl));
104: }
105: /* Make sure xu > xl */
106: PetscCall(VecCopy(mfqP->subxl, mfqP->subpdel));
107: PetscCall(VecAXPY(mfqP->subpdel, -1.0, mfqP->subxu));
108: PetscCall(VecMax(mfqP->subpdel, NULL, &maxval));
109: PetscCheck(maxval <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "upper bound < lower bound in subproblem");
110: /* Make sure xu > tao->solution > xl */
111: PetscCall(VecCopy(mfqP->subxl, mfqP->subpdel));
112: PetscCall(VecAXPY(mfqP->subpdel, -1.0, mfqP->subx));
113: PetscCall(VecMax(mfqP->subpdel, NULL, &maxval));
114: PetscCheck(maxval <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "initial guess < lower bound in subproblem");
116: PetscCall(VecCopy(mfqP->subx, mfqP->subpdel));
117: PetscCall(VecAXPY(mfqP->subpdel, -1.0, mfqP->subxu));
118: PetscCall(VecMax(mfqP->subpdel, NULL, &maxval));
119: PetscCheck(maxval <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "initial guess > upper bound in subproblem");
121: PetscCall(TaoSolve(mfqP->subtao));
122: PetscCall(TaoGetSolutionStatus(mfqP->subtao, NULL, qmin, NULL, NULL, NULL, NULL));
124: /* test bounds post-solution*/
125: PetscCall(VecCopy(mfqP->subxl, mfqP->subpdel));
126: PetscCall(VecAXPY(mfqP->subpdel, -1.0, mfqP->subx));
127: PetscCall(VecMax(mfqP->subpdel, NULL, &maxval));
128: if (maxval > 1e-5) {
129: PetscCall(PetscInfo(tao, "subproblem solution < lower bound\n"));
130: tao->reason = TAO_DIVERGED_TR_REDUCTION;
131: }
133: PetscCall(VecCopy(mfqP->subx, mfqP->subpdel));
134: PetscCall(VecAXPY(mfqP->subpdel, -1.0, mfqP->subxu));
135: PetscCall(VecMax(mfqP->subpdel, NULL, &maxval));
136: if (maxval > 1e-5) {
137: PetscCall(PetscInfo(tao, "subproblem solution > upper bound\n"));
138: tao->reason = TAO_DIVERGED_TR_REDUCTION;
139: }
140: } else {
141: PetscCall(gqt(mfqP->n, mfqP->Hres, mfqP->n, mfqP->Gres, 1.0, mfqP->gqt_rtol, atol, mfqP->gqt_maxits, gnorm, qmin, mfqP->Xsubproblem, &info, &its, mfqP->work, mfqP->work2, mfqP->work3));
142: }
143: *qmin *= -1;
144: PetscFunctionReturn(PETSC_SUCCESS);
145: }
147: static PetscErrorCode pounders_update_res(Tao tao)
148: {
149: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
150: PetscInt i, row, col;
151: PetscBLASInt blasn = mfqP->n, blasn2 = blasn * blasn, blasm = mfqP->m, ione = 1;
152: PetscReal zero = 0.0, one = 1.0, wii, factor;
154: PetscFunctionBegin;
155: for (i = 0; i < mfqP->n; i++) mfqP->Gres[i] = 0;
156: for (i = 0; i < mfqP->n * mfqP->n; i++) mfqP->Hres[i] = 0;
158: /* Compute Gres= sum_ij[wij * (cjgi + cigj)] */
159: if (tao->res_weights_v) {
160: /* Vector(diagonal) weights: gres = sum_i(wii*ci*gi) */
161: for (i = 0; i < mfqP->m; i++) {
162: PetscCall(VecGetValues(tao->res_weights_v, 1, &i, &factor));
163: factor = factor * mfqP->C[i];
164: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn, &factor, &mfqP->Fdiff[blasn * i], &ione, mfqP->Gres, &ione));
165: }
167: /* compute Hres = sum_ij [wij * (*ci*Hj + cj*Hi + gi gj' + gj gi') ] */
168: /* vector(diagonal weights) Hres = sum_i(wii*(ci*Hi + gi * gi')*/
169: for (i = 0; i < mfqP->m; i++) {
170: PetscCall(VecGetValues(tao->res_weights_v, 1, &i, &wii));
171: if (tao->niter > 1) {
172: factor = wii * mfqP->C[i];
173: /* add wii * ci * Hi */
174: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn2, &factor, &mfqP->H[i], &blasm, mfqP->Hres, &ione));
175: }
176: /* add wii * gi * gi' */
177: PetscCallBLAS("BLASgemm", BLASgemm_("N", "T", &blasn, &blasn, &ione, &wii, &mfqP->Fdiff[blasn * i], &blasn, &mfqP->Fdiff[blasn * i], &blasn, &one, mfqP->Hres, &blasn));
178: }
179: } else if (tao->res_weights_w) {
180: /* General case: .5 * Gres= sum_ij[wij * (cjgi + cigj)] */
181: for (i = 0; i < tao->res_weights_n; i++) {
182: row = tao->res_weights_rows[i];
183: col = tao->res_weights_cols[i];
185: factor = tao->res_weights_w[i] * mfqP->C[col] / 2.0;
186: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn, &factor, &mfqP->Fdiff[blasn * row], &ione, mfqP->Gres, &ione));
187: factor = tao->res_weights_w[i] * mfqP->C[row] / 2.0;
188: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn, &factor, &mfqP->Fdiff[blasn * col], &ione, mfqP->Gres, &ione));
189: }
191: /* compute Hres = sum_ij [wij * (*ci*Hj + cj*Hi + gi gj' + gj gi') ] */
192: /* .5 * sum_ij [wij * (*ci*Hj + cj*Hi + gi gj' + gj gi') ] */
193: for (i = 0; i < tao->res_weights_n; i++) {
194: row = tao->res_weights_rows[i];
195: col = tao->res_weights_cols[i];
196: factor = tao->res_weights_w[i] / 2.0;
197: /* add wij * gi gj' + wij * gj gi' */
198: PetscCallBLAS("BLASgemm", BLASgemm_("N", "T", &blasn, &blasn, &ione, &factor, &mfqP->Fdiff[blasn * row], &blasn, &mfqP->Fdiff[blasn * col], &blasn, &one, mfqP->Hres, &blasn));
199: PetscCallBLAS("BLASgemm", BLASgemm_("N", "T", &blasn, &blasn, &ione, &factor, &mfqP->Fdiff[blasn * col], &blasn, &mfqP->Fdiff[blasn * row], &blasn, &one, mfqP->Hres, &blasn));
200: }
201: if (tao->niter > 1) {
202: for (i = 0; i < tao->res_weights_n; i++) {
203: row = tao->res_weights_rows[i];
204: col = tao->res_weights_cols[i];
206: /* add wij*cj*Hi */
207: factor = tao->res_weights_w[i] * mfqP->C[col] / 2.0;
208: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn2, &factor, &mfqP->H[row], &blasm, mfqP->Hres, &ione));
210: /* add wij*ci*Hj */
211: factor = tao->res_weights_w[i] * mfqP->C[row] / 2.0;
212: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn2, &factor, &mfqP->H[col], &blasm, mfqP->Hres, &ione));
213: }
214: }
215: } else {
216: /* Default: Gres= sum_i[cigi] = G*c' */
217: PetscCall(PetscInfo(tao, "Identity weights\n"));
218: PetscCallBLAS("BLASgemv", BLASgemv_("N", &blasn, &blasm, &one, mfqP->Fdiff, &blasn, mfqP->C, &ione, &zero, mfqP->Gres, &ione));
220: /* compute Hres = sum_ij [wij * (*ci*Hj + cj*Hi + gi gj' + gj gi') ] */
221: /* Hres = G*G' + 0.5 sum {F(xkin,i)*H(:,:,i)} */
222: PetscCallBLAS("BLASgemm", BLASgemm_("N", "T", &blasn, &blasn, &blasm, &one, mfqP->Fdiff, &blasn, mfqP->Fdiff, &blasn, &zero, mfqP->Hres, &blasn));
224: /* sum(F(xkin,i)*H(:,:,i)) */
225: if (tao->niter > 1) {
226: for (i = 0; i < mfqP->m; i++) {
227: factor = mfqP->C[i];
228: PetscCallBLAS("BLASaxpy", BLASaxpy_(&blasn2, &factor, &mfqP->H[i], &blasm, mfqP->Hres, &ione));
229: }
230: }
231: }
232: PetscFunctionReturn(PETSC_SUCCESS);
233: }
235: static PetscErrorCode phi2eval(PetscReal *x, PetscInt n, PetscReal *phi)
236: {
237: /* Phi = .5*[x(1)^2 sqrt(2)*x(1)*x(2) ... sqrt(2)*x(1)*x(n) ... x(2)^2 sqrt(2)*x(2)*x(3) .. x(n)^2] */
238: PetscInt i, j, k;
239: PetscReal sqrt2 = PetscSqrtReal(2.0);
241: PetscFunctionBegin;
242: j = 0;
243: for (i = 0; i < n; i++) {
244: phi[j] = 0.5 * x[i] * x[i];
245: j++;
246: for (k = i + 1; k < n; k++) {
247: phi[j] = x[i] * x[k] / sqrt2;
248: j++;
249: }
250: }
251: PetscFunctionReturn(PETSC_SUCCESS);
252: }
254: static PetscErrorCode getquadpounders(TAO_POUNDERS *mfqP)
255: {
256: /* Computes the parameters of the quadratic Q(x) = c + g'*x + 0.5*x*G*x'
257: that satisfies the interpolation conditions Q(X[:,j]) = f(j)
258: for j=1,...,m and with a Hessian matrix of least Frobenius norm */
260: /* NB --we are ignoring c */
261: PetscInt i, j, k, num, np = mfqP->nmodelpoints;
262: PetscReal one = 1.0, zero = 0.0, negone = -1.0;
263: PetscBLASInt blasnpmax = mfqP->npmax;
264: PetscBLASInt blasnplus1 = mfqP->n + 1;
265: PetscBLASInt blasnp = np;
266: PetscBLASInt blasint = mfqP->n * (mfqP->n + 1) / 2;
267: PetscBLASInt blasint2 = np - mfqP->n - 1;
268: PetscBLASInt info, ione = 1;
269: PetscReal sqrt2 = PetscSqrtReal(2.0);
271: PetscFunctionBegin;
272: for (i = 0; i < mfqP->n * mfqP->m; i++) mfqP->Gdel[i] = 0;
273: for (i = 0; i < mfqP->n * mfqP->n * mfqP->m; i++) mfqP->Hdel[i] = 0;
275: /* factor M */
276: PetscCallBLAS("LAPACKgetrf", LAPACKgetrf_(&blasnplus1, &blasnp, mfqP->M, &blasnplus1, mfqP->npmaxiwork, &info));
277: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine getrf returned with value %" PetscBLASInt_FMT, info);
279: if (np == mfqP->n + 1) {
280: for (i = 0; i < mfqP->npmax - mfqP->n - 1; i++) mfqP->omega[i] = 0.0;
281: for (i = 0; i < mfqP->n * (mfqP->n + 1) / 2; i++) mfqP->beta[i] = 0.0;
282: } else {
283: /* Let Ltmp = (L'*L) */
284: PetscCallBLAS("BLASgemm", BLASgemm_("T", "N", &blasint2, &blasint2, &blasint, &one, &mfqP->L[(mfqP->n + 1) * blasint], &blasint, &mfqP->L[(mfqP->n + 1) * blasint], &blasint, &zero, mfqP->L_tmp, &blasint));
286: /* factor Ltmp */
287: PetscCallBLAS("LAPACKpotrf", LAPACKpotrf_("L", &blasint2, mfqP->L_tmp, &blasint, &info));
288: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine potrf returned with value %" PetscBLASInt_FMT, info);
289: }
291: for (k = 0; k < mfqP->m; k++) {
292: if (np != mfqP->n + 1) {
293: /* Solve L'*L*Omega = Z' * RESk*/
294: PetscCallBLAS("BLASgemv", BLASgemv_("T", &blasnp, &blasint2, &one, mfqP->Z, &blasnpmax, &mfqP->RES[mfqP->npmax * k], &ione, &zero, mfqP->omega, &ione));
295: PetscCallBLAS("LAPACKpotrs", LAPACKpotrs_("L", &blasint2, &ione, mfqP->L_tmp, &blasint, mfqP->omega, &blasint2, &info));
296: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine potrs returned with value %" PetscBLASInt_FMT, info);
298: /* Beta = L*Omega */
299: PetscCallBLAS("BLASgemv", BLASgemv_("N", &blasint, &blasint2, &one, &mfqP->L[(mfqP->n + 1) * blasint], &blasint, mfqP->omega, &ione, &zero, mfqP->beta, &ione));
300: }
302: /* solve M'*Alpha = RESk - N'*Beta */
303: PetscCallBLAS("BLASgemv", BLASgemv_("T", &blasint, &blasnp, &negone, mfqP->N, &blasint, mfqP->beta, &ione, &one, &mfqP->RES[mfqP->npmax * k], &ione));
304: PetscCallBLAS("LAPACKgetrs", LAPACKgetrs_("T", &blasnplus1, &ione, mfqP->M, &blasnplus1, mfqP->npmaxiwork, &mfqP->RES[mfqP->npmax * k], &blasnplus1, &info));
305: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine getrs returned with value %" PetscBLASInt_FMT, info);
307: /* Gdel(:,k) = Alpha(2:n+1) */
308: for (i = 0; i < mfqP->n; i++) mfqP->Gdel[i + mfqP->n * k] = mfqP->RES[mfqP->npmax * k + i + 1];
310: /* Set Hdels */
311: num = 0;
312: for (i = 0; i < mfqP->n; i++) {
313: /* H[i,i,k] = Beta(num) */
314: mfqP->Hdel[(i * mfqP->n + i) * mfqP->m + k] = mfqP->beta[num];
315: num++;
316: for (j = i + 1; j < mfqP->n; j++) {
317: /* H[i,j,k] = H[j,i,k] = Beta(num)/sqrt(2) */
318: mfqP->Hdel[(j * mfqP->n + i) * mfqP->m + k] = mfqP->beta[num] / sqrt2;
319: mfqP->Hdel[(i * mfqP->n + j) * mfqP->m + k] = mfqP->beta[num] / sqrt2;
320: num++;
321: }
322: }
323: }
324: PetscFunctionReturn(PETSC_SUCCESS);
325: }
327: static PetscErrorCode morepoints(TAO_POUNDERS *mfqP)
328: {
329: /* Assumes mfqP->model_indices[0] is minimum index
330: Finishes adding points to mfqP->model_indices (up to npmax)
331: Computes L,Z,M,N
332: np is actual number of points in model (should equal npmax?) */
333: PetscInt point, i, j, offset;
334: PetscInt reject;
335: PetscBLASInt blasn = mfqP->n, blasnpmax = mfqP->npmax, blasnplus1 = mfqP->n + 1, info, blasnmax = mfqP->nmax, blasint, blasint2, blasnp, blasmaxmn;
336: const PetscReal *x;
337: PetscReal normd;
339: PetscFunctionBegin;
340: /* Initialize M,N */
341: for (i = 0; i < mfqP->n + 1; i++) {
342: PetscCall(VecGetArrayRead(mfqP->Xhist[mfqP->model_indices[i]], &x));
343: mfqP->M[(mfqP->n + 1) * i] = 1.0;
344: for (j = 0; j < mfqP->n; j++) mfqP->M[j + 1 + ((mfqP->n + 1) * i)] = (x[j] - mfqP->xmin[j]) / mfqP->delta;
345: PetscCall(VecRestoreArrayRead(mfqP->Xhist[mfqP->model_indices[i]], &x));
346: PetscCall(phi2eval(&mfqP->M[1 + ((mfqP->n + 1) * i)], mfqP->n, &mfqP->N[mfqP->n * (mfqP->n + 1) / 2 * i]));
347: }
349: /* Now we add points until we have npmax starting with the most recent ones */
350: point = mfqP->nHist - 1;
351: mfqP->nmodelpoints = mfqP->n + 1;
352: while (mfqP->nmodelpoints < mfqP->npmax && point >= 0) {
353: /* Reject any points already in the model */
354: reject = 0;
355: for (j = 0; j < mfqP->n + 1; j++) {
356: if (point == mfqP->model_indices[j]) {
357: reject = 1;
358: break;
359: }
360: }
362: /* Reject if norm(d) >c2 */
363: if (!reject) {
364: PetscCall(VecCopy(mfqP->Xhist[point], mfqP->workxvec));
365: PetscCall(VecAXPY(mfqP->workxvec, -1.0, mfqP->Xhist[mfqP->minindex]));
366: PetscCall(VecNorm(mfqP->workxvec, NORM_2, &normd));
367: normd /= mfqP->delta;
368: if (normd > mfqP->c2) reject = 1;
369: }
370: if (reject) {
371: point--;
372: continue;
373: }
375: PetscCall(VecGetArrayRead(mfqP->Xhist[point], &x));
376: mfqP->M[(mfqP->n + 1) * mfqP->nmodelpoints] = 1.0;
377: for (j = 0; j < mfqP->n; j++) mfqP->M[j + 1 + ((mfqP->n + 1) * mfqP->nmodelpoints)] = (x[j] - mfqP->xmin[j]) / mfqP->delta;
378: PetscCall(VecRestoreArrayRead(mfqP->Xhist[point], &x));
379: PetscCall(phi2eval(&mfqP->M[1 + (mfqP->n + 1) * mfqP->nmodelpoints], mfqP->n, &mfqP->N[mfqP->n * (mfqP->n + 1) / 2 * (mfqP->nmodelpoints)]));
381: /* Update QR factorization */
382: /* Copy M' to Q_tmp */
383: for (i = 0; i < mfqP->n + 1; i++) {
384: for (j = 0; j < mfqP->npmax; j++) mfqP->Q_tmp[j + mfqP->npmax * i] = mfqP->M[i + (mfqP->n + 1) * j];
385: }
386: blasnp = mfqP->nmodelpoints + 1;
387: /* Q_tmp,R = qr(M') */
388: blasmaxmn = PetscMax(mfqP->m, mfqP->n + 1);
389: PetscCallBLAS("LAPACKgeqrf", LAPACKgeqrf_(&blasnp, &blasnplus1, mfqP->Q_tmp, &blasnpmax, mfqP->tau_tmp, mfqP->mwork, &blasmaxmn, &info));
390: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine geqrf returned with value %" PetscBLASInt_FMT, info);
392: /* Reject if min(svd(N*Q(:,n+2:np+1)) <= theta2 */
393: /* L = N*Qtmp */
394: blasint2 = mfqP->n * (mfqP->n + 1) / 2;
395: /* Copy N to L_tmp */
396: for (i = 0; i < mfqP->n * (mfqP->n + 1) / 2 * mfqP->npmax; i++) mfqP->L_tmp[i] = mfqP->N[i];
397: /* Copy L_save to L_tmp */
399: /* L_tmp = N*Qtmp' */
400: PetscCallBLAS("LAPACKormqr", LAPACKormqr_("R", "N", &blasint2, &blasnp, &blasnplus1, mfqP->Q_tmp, &blasnpmax, mfqP->tau_tmp, mfqP->L_tmp, &blasint2, mfqP->npmaxwork, &blasnmax, &info));
401: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine ormqr returned with value %" PetscBLASInt_FMT, info);
403: /* Copy L_tmp to L_save */
404: for (i = 0; i < mfqP->npmax * mfqP->n * (mfqP->n + 1) / 2; i++) mfqP->L_save[i] = mfqP->L_tmp[i];
406: /* Get svd for L_tmp(:,n+2:np+1) (L_tmp is modified in process) */
407: blasint = mfqP->nmodelpoints - mfqP->n;
408: PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF));
409: PetscCallBLAS("LAPACKgesvd", LAPACKgesvd_("N", "N", &blasint2, &blasint, &mfqP->L_tmp[(mfqP->n + 1) * blasint2], &blasint2, mfqP->beta, mfqP->work, &blasn, mfqP->work, &blasn, mfqP->npmaxwork, &blasnmax, &info));
410: PetscCall(PetscFPTrapPop());
411: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine gesvd returned with value %" PetscBLASInt_FMT, info);
413: if (mfqP->beta[PetscMin(blasint, blasint2) - 1] > mfqP->theta2) {
414: /* accept point */
415: mfqP->model_indices[mfqP->nmodelpoints] = point;
416: /* Copy Q_tmp to Q */
417: for (i = 0; i < mfqP->npmax * mfqP->npmax; i++) mfqP->Q[i] = mfqP->Q_tmp[i];
418: for (i = 0; i < mfqP->npmax; i++) mfqP->tau[i] = mfqP->tau_tmp[i];
419: mfqP->nmodelpoints++;
420: blasnp = mfqP->nmodelpoints;
422: /* Copy L_save to L */
423: for (i = 0; i < mfqP->npmax * mfqP->n * (mfqP->n + 1) / 2; i++) mfqP->L[i] = mfqP->L_save[i];
424: }
425: point--;
426: }
428: blasnp = mfqP->nmodelpoints;
429: /* Copy Q(:,n+2:np) to Z */
430: /* First set Q_tmp to I */
431: for (i = 0; i < mfqP->npmax * mfqP->npmax; i++) mfqP->Q_tmp[i] = 0.0;
432: for (i = 0; i < mfqP->npmax; i++) mfqP->Q_tmp[i + mfqP->npmax * i] = 1.0;
434: /* Q_tmp = I * Q */
435: PetscCallBLAS("LAPACKormqr", LAPACKormqr_("R", "N", &blasnp, &blasnp, &blasnplus1, mfqP->Q, &blasnpmax, mfqP->tau, mfqP->Q_tmp, &blasnpmax, mfqP->npmaxwork, &blasnmax, &info));
436: PetscCheck(info == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "LAPACK routine ormqr returned with value %" PetscBLASInt_FMT, info);
438: /* Copy Q_tmp(:,n+2:np) to Z) */
439: offset = mfqP->npmax * (mfqP->n + 1);
440: for (i = offset; i < mfqP->npmax * mfqP->npmax; i++) mfqP->Z[i - offset] = mfqP->Q_tmp[i];
442: if (mfqP->nmodelpoints == mfqP->n + 1) {
443: /* Set L to I_{n+1} */
444: for (i = 0; i < mfqP->npmax * mfqP->n * (mfqP->n + 1) / 2; i++) mfqP->L[i] = 0.0;
445: for (i = 0; i < mfqP->n; i++) mfqP->L[(mfqP->n * (mfqP->n + 1) / 2) * i + i] = 1.0;
446: }
447: PetscFunctionReturn(PETSC_SUCCESS);
448: }
450: /* Only call from modelimprove, addpoint() needs ->Q_tmp and ->work to be set */
451: static PetscErrorCode addpoint(Tao tao, TAO_POUNDERS *mfqP, PetscInt index)
452: {
453: PetscFunctionBegin;
454: /* Create new vector in history: X[newidx] = X[mfqP->index] + delta*X[index]*/
455: PetscCall(VecDuplicate(mfqP->Xhist[0], &mfqP->Xhist[mfqP->nHist]));
456: PetscCall(VecSetValues(mfqP->Xhist[mfqP->nHist], mfqP->n, mfqP->indices, &mfqP->Q_tmp[index * mfqP->npmax], INSERT_VALUES));
457: PetscCall(VecAssemblyBegin(mfqP->Xhist[mfqP->nHist]));
458: PetscCall(VecAssemblyEnd(mfqP->Xhist[mfqP->nHist]));
459: PetscCall(VecAYPX(mfqP->Xhist[mfqP->nHist], mfqP->delta, mfqP->Xhist[mfqP->minindex]));
461: /* Project into feasible region */
462: if (tao->XU && tao->XL) PetscCall(VecMedian(mfqP->Xhist[mfqP->nHist], tao->XL, tao->XU, mfqP->Xhist[mfqP->nHist]));
464: /* Compute value of new vector */
465: PetscCall(VecDuplicate(mfqP->Fhist[0], &mfqP->Fhist[mfqP->nHist]));
466: CHKMEMQ;
467: PetscCall(pounders_feval(tao, mfqP->Xhist[mfqP->nHist], mfqP->Fhist[mfqP->nHist], &mfqP->Fres[mfqP->nHist]));
469: /* Add new vector to model */
470: mfqP->model_indices[mfqP->nmodelpoints] = mfqP->nHist;
471: mfqP->nmodelpoints++;
472: mfqP->nHist++;
473: PetscFunctionReturn(PETSC_SUCCESS);
474: }
476: static PetscErrorCode modelimprove(Tao tao, TAO_POUNDERS *mfqP, PetscInt addallpoints)
477: {
478: /* modeld = Q(:,np+1:n)' */
479: PetscInt i, j, minindex = 0;
480: PetscReal dp, half = 0.5, one = 1.0, minvalue = PETSC_INFINITY;
481: PetscBLASInt blasn = mfqP->n, blasnpmax = mfqP->npmax, blask, info;
482: PetscBLASInt blas1 = 1, blasnmax = mfqP->nmax;
484: PetscFunctionBegin;
485: blask = mfqP->nmodelpoints;
486: /* Qtmp = I(n x n) */
487: for (i = 0; i < mfqP->n; i++) {
488: for (j = 0; j < mfqP->n; j++) mfqP->Q_tmp[i + mfqP->npmax * j] = 0.0;
489: }
490: for (j = 0; j < mfqP->n; j++) mfqP->Q_tmp[j + mfqP->npmax * j] = 1.0;
492: /* Qtmp = Q * I */
493: PetscCallBLAS("LAPACKormqr", LAPACKormqr_("R", "N", &blasn, &blasn, &blask, mfqP->Q, &blasnpmax, mfqP->tau, mfqP->Q_tmp, &blasnpmax, mfqP->npmaxwork, &blasnmax, &info));
495: for (i = mfqP->nmodelpoints; i < mfqP->n; i++) {
496: PetscCallBLAS("BLASdot", dp = BLASdot_(&blasn, &mfqP->Q_tmp[i * mfqP->npmax], &blas1, mfqP->Gres, &blas1));
497: if (dp > 0.0) { /* Model says use the other direction! */
498: for (j = 0; j < mfqP->n; j++) mfqP->Q_tmp[i * mfqP->npmax + j] *= -1;
499: }
500: /* mfqP->work[i] = Cres+Modeld(i,:)*(Gres+.5*Hres*Modeld(i,:)') */
501: for (j = 0; j < mfqP->n; j++) mfqP->work2[j] = mfqP->Gres[j];
502: PetscCallBLAS("BLASgemv", BLASgemv_("N", &blasn, &blasn, &half, mfqP->Hres, &blasn, &mfqP->Q_tmp[i * mfqP->npmax], &blas1, &one, mfqP->work2, &blas1));
503: PetscCallBLAS("BLASdot", mfqP->work[i] = BLASdot_(&blasn, &mfqP->Q_tmp[i * mfqP->npmax], &blas1, mfqP->work2, &blas1));
504: if (i == mfqP->nmodelpoints || mfqP->work[i] < minvalue) {
505: minindex = i;
506: minvalue = mfqP->work[i];
507: }
508: if (addallpoints != 0) PetscCall(addpoint(tao, mfqP, i));
509: }
510: if (!addallpoints) PetscCall(addpoint(tao, mfqP, minindex));
511: PetscFunctionReturn(PETSC_SUCCESS);
512: }
514: static PetscErrorCode affpoints(TAO_POUNDERS *mfqP, PetscReal *xmin, PetscReal c)
515: {
516: PetscInt i, j;
517: PetscBLASInt blasm = mfqP->m, blasj, blask, blasn = mfqP->n, ione = 1, info;
518: PetscBLASInt blasnpmax = mfqP->npmax, blasmaxmn;
519: PetscReal proj, normd;
520: const PetscReal *x;
522: PetscFunctionBegin;
523: for (i = mfqP->nHist - 1; i >= 0; i--) {
524: PetscCall(VecGetArrayRead(mfqP->Xhist[i], &x));
525: for (j = 0; j < mfqP->n; j++) mfqP->work[j] = (x[j] - xmin[j]) / mfqP->delta;
526: PetscCall(VecRestoreArrayRead(mfqP->Xhist[i], &x));
527: PetscCallBLAS("BLAScopy", BLAScopy_(&blasn, mfqP->work, &ione, mfqP->work2, &ione));
528: PetscCallBLAS("BLASnrm2", normd = BLASnrm2_(&blasn, mfqP->work, &ione));
529: if (normd <= c) {
530: blasj = PetscMax((mfqP->n - mfqP->nmodelpoints), 0);
531: if (!mfqP->q_is_I) {
532: /* project D onto null */
533: blask = (mfqP->nmodelpoints);
534: PetscCallBLAS("LAPACKormqr", LAPACKormqr_("R", "N", &ione, &blasn, &blask, mfqP->Q, &blasnpmax, mfqP->tau, mfqP->work2, &ione, mfqP->mwork, &blasm, &info));
535: PetscCheck(info >= 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "ormqr returned value %" PetscBLASInt_FMT, info);
536: }
537: PetscCallBLAS("BLASnrm2", proj = BLASnrm2_(&blasj, &mfqP->work2[mfqP->nmodelpoints], &ione));
539: if (proj >= mfqP->theta1) { /* add this index to model */
540: mfqP->model_indices[mfqP->nmodelpoints] = i;
541: mfqP->nmodelpoints++;
542: PetscCallBLAS("BLAScopy", BLAScopy_(&blasn, mfqP->work, &ione, &mfqP->Q_tmp[mfqP->npmax * (mfqP->nmodelpoints - 1)], &ione));
543: blask = mfqP->npmax * (mfqP->nmodelpoints);
544: PetscCallBLAS("BLAScopy", BLAScopy_(&blask, mfqP->Q_tmp, &ione, mfqP->Q, &ione));
545: blask = mfqP->nmodelpoints;
546: blasmaxmn = PetscMax(mfqP->m, mfqP->n);
547: PetscCallBLAS("LAPACKgeqrf", LAPACKgeqrf_(&blasn, &blask, mfqP->Q, &blasnpmax, mfqP->tau, mfqP->mwork, &blasmaxmn, &info));
548: PetscCheck(info >= 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "geqrf returned value %" PetscBLASInt_FMT, info);
549: mfqP->q_is_I = 0;
550: }
551: if (mfqP->nmodelpoints == mfqP->n) break;
552: }
553: }
554: PetscFunctionReturn(PETSC_SUCCESS);
555: }
557: static PetscErrorCode TaoSolve_POUNDERS(Tao tao)
558: {
559: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
560: PetscInt i, ii, j, k, l;
561: PetscReal step = 1.0;
562: PetscInt low, high;
563: PetscReal minnorm;
564: PetscReal *x, *f;
565: const PetscReal *xmint, *fmin;
566: PetscReal deltaold;
567: PetscReal gnorm;
568: PetscBLASInt info, ione = 1, iblas;
569: PetscBool valid, same;
570: PetscReal mdec, rho, normxsp;
571: PetscReal one = 1.0, zero = 0.0, ratio;
572: PetscBLASInt blasm, blasn, blasncopy, blasnpmax;
573: static PetscBool set = PETSC_FALSE;
575: /* n = # of parameters
576: m = dimension (components) of function */
577: PetscFunctionBegin;
578: PetscCall(PetscCitationsRegister("@article{UNEDF0,\n"
579: "title = {Nuclear energy density optimization},\n"
580: "author = {Kortelainen, M. and Lesinski, T. and Mor\'e, J. and Nazarewicz, W.\n"
581: " and Sarich, J. and Schunck, N. and Stoitsov, M. V. and Wild, S. },\n"
582: "journal = {Phys. Rev. C},\n"
583: "volume = {82},\n"
584: "number = {2},\n"
585: "pages = {024313},\n"
586: "numpages = {18},\n"
587: "year = {2010},\n"
588: "month = {Aug},\n"
589: "doi = {10.1103/PhysRevC.82.024313}\n}\n",
590: &set));
591: tao->niter = 0;
592: if (tao->XL && tao->XU) {
593: /* Check x0 <= XU */
594: PetscReal val;
596: PetscCall(VecCopy(tao->solution, mfqP->Xhist[0]));
597: PetscCall(VecAXPY(mfqP->Xhist[0], -1.0, tao->XU));
598: PetscCall(VecMax(mfqP->Xhist[0], NULL, &val));
599: PetscCheck(val <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "X0 > upper bound");
601: /* Check x0 >= xl */
602: PetscCall(VecCopy(tao->XL, mfqP->Xhist[0]));
603: PetscCall(VecAXPY(mfqP->Xhist[0], -1.0, tao->solution));
604: PetscCall(VecMax(mfqP->Xhist[0], NULL, &val));
605: PetscCheck(val <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "X0 < lower bound");
607: /* Check x0 + delta < XU -- should be able to get around this eventually */
609: PetscCall(VecSet(mfqP->Xhist[0], mfqP->delta));
610: PetscCall(VecAXPY(mfqP->Xhist[0], 1.0, tao->solution));
611: PetscCall(VecAXPY(mfqP->Xhist[0], -1.0, tao->XU));
612: PetscCall(VecMax(mfqP->Xhist[0], NULL, &val));
613: PetscCheck(val <= 1e-10, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_OUTOFRANGE, "X0 + delta > upper bound");
614: }
616: blasm = mfqP->m;
617: blasn = mfqP->n;
618: blasnpmax = mfqP->npmax;
619: for (i = 0; i < mfqP->n * mfqP->n * mfqP->m; ++i) mfqP->H[i] = 0;
621: PetscCall(VecCopy(tao->solution, mfqP->Xhist[0]));
623: /* This provides enough information to approximate the gradient of the objective */
624: /* using a forward difference scheme. */
626: PetscCall(PetscInfo(tao, "Initialize simplex; delta = %10.9e\n", (double)mfqP->delta));
627: PetscCall(pounders_feval(tao, mfqP->Xhist[0], mfqP->Fhist[0], &mfqP->Fres[0]));
628: mfqP->minindex = 0;
629: minnorm = mfqP->Fres[0];
631: PetscCall(VecGetOwnershipRange(mfqP->Xhist[0], &low, &high));
632: for (i = 1; i < mfqP->n + 1; ++i) {
633: PetscCall(VecCopy(mfqP->Xhist[0], mfqP->Xhist[i]));
635: if (i - 1 >= low && i - 1 < high) {
636: PetscCall(VecGetArray(mfqP->Xhist[i], &x));
637: x[i - 1 - low] += mfqP->delta;
638: PetscCall(VecRestoreArray(mfqP->Xhist[i], &x));
639: }
640: CHKMEMQ;
641: PetscCall(pounders_feval(tao, mfqP->Xhist[i], mfqP->Fhist[i], &mfqP->Fres[i]));
642: if (mfqP->Fres[i] < minnorm) {
643: mfqP->minindex = i;
644: minnorm = mfqP->Fres[i];
645: }
646: }
647: PetscCall(VecCopy(mfqP->Xhist[mfqP->minindex], tao->solution));
648: PetscCall(VecCopy(mfqP->Fhist[mfqP->minindex], tao->ls_res));
649: PetscCall(PetscInfo(tao, "Finalize simplex; minnorm = %10.9e\n", (double)minnorm));
651: /* Gather mpi vecs to one big local vec */
653: /* Begin serial code */
655: /* Disp[i] = Xi-xmin, i=1,..,mfqP->minindex-1,mfqP->minindex+1,..,n */
656: /* Fdiff[i] = (Fi-Fmin)', i=1,..,mfqP->minindex-1,mfqP->minindex+1,..,n */
657: /* (Column oriented for blas calls) */
658: ii = 0;
660: PetscCall(PetscInfo(tao, "Build matrix: %" PetscInt_FMT "\n", (PetscInt)mfqP->size));
661: if (1 == mfqP->size) {
662: PetscCall(VecGetArrayRead(mfqP->Xhist[mfqP->minindex], &xmint));
663: for (i = 0; i < mfqP->n; i++) mfqP->xmin[i] = xmint[i];
664: PetscCall(VecRestoreArrayRead(mfqP->Xhist[mfqP->minindex], &xmint));
665: PetscCall(VecGetArrayRead(mfqP->Fhist[mfqP->minindex], &fmin));
666: for (i = 0; i < mfqP->n + 1; i++) {
667: if (i == mfqP->minindex) continue;
669: PetscCall(VecGetArray(mfqP->Xhist[i], &x));
670: for (j = 0; j < mfqP->n; j++) mfqP->Disp[ii + mfqP->npmax * j] = (x[j] - mfqP->xmin[j]) / mfqP->delta;
671: PetscCall(VecRestoreArray(mfqP->Xhist[i], &x));
673: PetscCall(VecGetArray(mfqP->Fhist[i], &f));
674: for (j = 0; j < mfqP->m; j++) mfqP->Fdiff[ii + mfqP->n * j] = f[j] - fmin[j];
675: PetscCall(VecRestoreArray(mfqP->Fhist[i], &f));
677: mfqP->model_indices[ii++] = i;
678: }
679: for (j = 0; j < mfqP->m; j++) mfqP->C[j] = fmin[j];
680: PetscCall(VecRestoreArrayRead(mfqP->Fhist[mfqP->minindex], &fmin));
681: } else {
682: PetscCall(VecSet(mfqP->localxmin, 0));
683: PetscCall(VecScatterBegin(mfqP->scatterx, mfqP->Xhist[mfqP->minindex], mfqP->localxmin, INSERT_VALUES, SCATTER_FORWARD));
684: PetscCall(VecScatterEnd(mfqP->scatterx, mfqP->Xhist[mfqP->minindex], mfqP->localxmin, INSERT_VALUES, SCATTER_FORWARD));
686: PetscCall(VecGetArrayRead(mfqP->localxmin, &xmint));
687: for (i = 0; i < mfqP->n; i++) mfqP->xmin[i] = xmint[i];
688: PetscCall(VecRestoreArrayRead(mfqP->localxmin, &xmint));
690: PetscCall(VecScatterBegin(mfqP->scatterf, mfqP->Fhist[mfqP->minindex], mfqP->localfmin, INSERT_VALUES, SCATTER_FORWARD));
691: PetscCall(VecScatterEnd(mfqP->scatterf, mfqP->Fhist[mfqP->minindex], mfqP->localfmin, INSERT_VALUES, SCATTER_FORWARD));
692: PetscCall(VecGetArrayRead(mfqP->localfmin, &fmin));
693: for (i = 0; i < mfqP->n + 1; i++) {
694: if (i == mfqP->minindex) continue;
696: PetscCall(VecScatterBegin(mfqP->scatterx, mfqP->Xhist[ii], mfqP->localx, INSERT_VALUES, SCATTER_FORWARD));
697: PetscCall(VecScatterEnd(mfqP->scatterx, mfqP->Xhist[ii], mfqP->localx, INSERT_VALUES, SCATTER_FORWARD));
698: PetscCall(VecGetArray(mfqP->localx, &x));
699: for (j = 0; j < mfqP->n; j++) mfqP->Disp[ii + mfqP->npmax * j] = (x[j] - mfqP->xmin[j]) / mfqP->delta;
700: PetscCall(VecRestoreArray(mfqP->localx, &x));
702: PetscCall(VecScatterBegin(mfqP->scatterf, mfqP->Fhist[ii], mfqP->localf, INSERT_VALUES, SCATTER_FORWARD));
703: PetscCall(VecScatterEnd(mfqP->scatterf, mfqP->Fhist[ii], mfqP->localf, INSERT_VALUES, SCATTER_FORWARD));
704: PetscCall(VecGetArray(mfqP->localf, &f));
705: for (j = 0; j < mfqP->m; j++) mfqP->Fdiff[ii + mfqP->n * j] = f[j] - fmin[j];
706: PetscCall(VecRestoreArray(mfqP->localf, &f));
708: mfqP->model_indices[ii++] = i;
709: }
710: for (j = 0; j < mfqP->m; j++) mfqP->C[j] = fmin[j];
711: PetscCall(VecRestoreArrayRead(mfqP->localfmin, &fmin));
712: }
714: /* Determine the initial quadratic models */
715: /* G = D(ModelIn,:) \ (F(ModelIn,1:m)-repmat(F(xkin,1:m),n,1)); */
716: /* D (nxn) Fdiff (nxm) => G (nxm) */
717: blasncopy = blasn;
718: PetscCallBLAS("LAPACKgesv", LAPACKgesv_(&blasn, &blasm, mfqP->Disp, &blasnpmax, mfqP->iwork, mfqP->Fdiff, &blasncopy, &info));
719: PetscCall(PetscInfo(tao, "Linear solve return: %" PetscInt_FMT "\n", (PetscInt)info));
721: PetscCall(pounders_update_res(tao));
723: valid = PETSC_TRUE;
725: PetscCall(VecSetValues(tao->gradient, mfqP->n, mfqP->indices, mfqP->Gres, INSERT_VALUES));
726: PetscCall(VecAssemblyBegin(tao->gradient));
727: PetscCall(VecAssemblyEnd(tao->gradient));
728: PetscCall(VecNorm(tao->gradient, NORM_2, &gnorm));
729: gnorm *= mfqP->delta;
730: PetscCall(VecCopy(mfqP->Xhist[mfqP->minindex], tao->solution));
732: tao->reason = TAO_CONTINUE_ITERATING;
733: PetscCall(TaoLogConvergenceHistory(tao, minnorm, gnorm, 0.0, tao->ksp_its));
734: PetscCall(TaoMonitor(tao, tao->niter, minnorm, gnorm, 0.0, step));
735: PetscUseTypeMethod(tao, convergencetest, tao->cnvP);
737: mfqP->nHist = mfqP->n + 1;
738: mfqP->nmodelpoints = mfqP->n + 1;
739: PetscCall(PetscInfo(tao, "Initial gradient: %20.19e\n", (double)gnorm));
741: while (tao->reason == TAO_CONTINUE_ITERATING) {
742: PetscReal gnm = 1e-4;
743: /* Call general purpose update function */
744: PetscTryTypeMethod(tao, update, tao->niter, tao->user_update);
745: tao->niter++;
746: /* Solve the subproblem min{Q(s): ||s|| <= 1.0} */
747: PetscCall(gqtwrap(tao, &gnm, &mdec));
748: /* Evaluate the function at the new point */
750: for (i = 0; i < mfqP->n; i++) mfqP->work[i] = mfqP->Xsubproblem[i] * mfqP->delta + mfqP->xmin[i];
751: PetscCall(VecDuplicate(tao->solution, &mfqP->Xhist[mfqP->nHist]));
752: PetscCall(VecDuplicate(tao->ls_res, &mfqP->Fhist[mfqP->nHist]));
753: PetscCall(VecSetValues(mfqP->Xhist[mfqP->nHist], mfqP->n, mfqP->indices, mfqP->work, INSERT_VALUES));
754: PetscCall(VecAssemblyBegin(mfqP->Xhist[mfqP->nHist]));
755: PetscCall(VecAssemblyEnd(mfqP->Xhist[mfqP->nHist]));
757: PetscCall(pounders_feval(tao, mfqP->Xhist[mfqP->nHist], mfqP->Fhist[mfqP->nHist], &mfqP->Fres[mfqP->nHist]));
759: rho = (mfqP->Fres[mfqP->minindex] - mfqP->Fres[mfqP->nHist]) / mdec;
760: mfqP->nHist++;
762: /* Update the center */
763: if ((rho >= mfqP->eta1) || (rho > mfqP->eta0 && valid == PETSC_TRUE)) {
764: /* Update model to reflect new base point */
765: for (i = 0; i < mfqP->n; i++) mfqP->work[i] = (mfqP->work[i] - mfqP->xmin[i]) / mfqP->delta;
766: for (j = 0; j < mfqP->m; j++) {
767: /* C(j) = C(j) + work*G(:,j) + .5*work*H(:,:,j)*work';
768: G(:,j) = G(:,j) + H(:,:,j)*work' */
769: for (k = 0; k < mfqP->n; k++) {
770: mfqP->work2[k] = 0.0;
771: for (l = 0; l < mfqP->n; l++) mfqP->work2[k] += mfqP->H[j + mfqP->m * (k + l * mfqP->n)] * mfqP->work[l];
772: }
773: for (i = 0; i < mfqP->n; i++) {
774: mfqP->C[j] += mfqP->work[i] * (mfqP->Fdiff[i + mfqP->n * j] + 0.5 * mfqP->work2[i]);
775: mfqP->Fdiff[i + mfqP->n * j] += mfqP->work2[i];
776: }
777: }
778: /* Cres += work*Gres + .5*work*Hres*work';
779: Gres += Hres*work'; */
781: PetscCallBLAS("BLASgemv", BLASgemv_("N", &blasn, &blasn, &one, mfqP->Hres, &blasn, mfqP->work, &ione, &zero, mfqP->work2, &ione));
782: for (i = 0; i < mfqP->n; i++) mfqP->Gres[i] += mfqP->work2[i];
783: mfqP->minindex = mfqP->nHist - 1;
784: minnorm = mfqP->Fres[mfqP->minindex];
785: PetscCall(VecCopy(mfqP->Fhist[mfqP->minindex], tao->ls_res));
786: /* Change current center */
787: PetscCall(VecGetArrayRead(mfqP->Xhist[mfqP->minindex], &xmint));
788: for (i = 0; i < mfqP->n; i++) mfqP->xmin[i] = xmint[i];
789: PetscCall(VecRestoreArrayRead(mfqP->Xhist[mfqP->minindex], &xmint));
790: }
792: /* Evaluate at a model-improving point if necessary */
793: if (valid == PETSC_FALSE) {
794: mfqP->q_is_I = 1;
795: mfqP->nmodelpoints = 0;
796: PetscCall(affpoints(mfqP, mfqP->xmin, mfqP->c1));
797: if (mfqP->nmodelpoints < mfqP->n) {
798: PetscCall(PetscInfo(tao, "Model not valid -- model-improving\n"));
799: PetscCall(modelimprove(tao, mfqP, 1));
800: }
801: }
803: /* Update the trust region radius */
804: deltaold = mfqP->delta;
805: normxsp = 0;
806: for (i = 0; i < mfqP->n; i++) normxsp += mfqP->Xsubproblem[i] * mfqP->Xsubproblem[i];
807: normxsp = PetscSqrtReal(normxsp);
808: if (rho >= mfqP->eta1 && normxsp > 0.5 * mfqP->delta) {
809: mfqP->delta = PetscMin(mfqP->delta * mfqP->gamma1, mfqP->deltamax);
810: } else if (valid == PETSC_TRUE) {
811: mfqP->delta = PetscMax(mfqP->delta * mfqP->gamma0, mfqP->deltamin);
812: }
814: /* Compute the next interpolation set */
815: mfqP->q_is_I = 1;
816: mfqP->nmodelpoints = 0;
817: PetscCall(PetscInfo(tao, "Affine Points: xmin = %20.19e, c1 = %20.19e\n", (double)*mfqP->xmin, (double)mfqP->c1));
818: PetscCall(affpoints(mfqP, mfqP->xmin, mfqP->c1));
819: if (mfqP->nmodelpoints == mfqP->n) {
820: valid = PETSC_TRUE;
821: } else {
822: valid = PETSC_FALSE;
823: PetscCall(PetscInfo(tao, "Affine Points: xmin = %20.19e, c2 = %20.19e\n", (double)*mfqP->xmin, (double)mfqP->c2));
824: PetscCall(affpoints(mfqP, mfqP->xmin, mfqP->c2));
825: if (mfqP->n > mfqP->nmodelpoints) {
826: PetscCall(PetscInfo(tao, "Model not valid -- adding geometry points\n"));
827: PetscCall(modelimprove(tao, mfqP, mfqP->n - mfqP->nmodelpoints));
828: }
829: }
830: for (i = mfqP->nmodelpoints; i > 0; i--) mfqP->model_indices[i] = mfqP->model_indices[i - 1];
831: mfqP->nmodelpoints++;
832: mfqP->model_indices[0] = mfqP->minindex;
833: PetscCall(morepoints(mfqP));
834: for (i = 0; i < mfqP->nmodelpoints; i++) {
835: PetscCall(VecGetArray(mfqP->Xhist[mfqP->model_indices[i]], &x));
836: for (j = 0; j < mfqP->n; j++) mfqP->Disp[i + mfqP->npmax * j] = (x[j] - mfqP->xmin[j]) / deltaold;
837: PetscCall(VecRestoreArray(mfqP->Xhist[mfqP->model_indices[i]], &x));
838: PetscCall(VecGetArray(mfqP->Fhist[mfqP->model_indices[i]], &f));
839: for (j = 0; j < mfqP->m; j++) {
840: for (k = 0; k < mfqP->n; k++) {
841: mfqP->work[k] = 0.0;
842: for (l = 0; l < mfqP->n; l++) mfqP->work[k] += mfqP->H[j + mfqP->m * (k + mfqP->n * l)] * mfqP->Disp[i + mfqP->npmax * l];
843: }
844: PetscCallBLAS("BLASdot", mfqP->RES[j * mfqP->npmax + i] = -mfqP->C[j] - BLASdot_(&blasn, &mfqP->Fdiff[j * mfqP->n], &ione, &mfqP->Disp[i], &blasnpmax) - 0.5 * BLASdot_(&blasn, mfqP->work, &ione, &mfqP->Disp[i], &blasnpmax) + f[j]);
845: }
846: PetscCall(VecRestoreArray(mfqP->Fhist[mfqP->model_indices[i]], &f));
847: }
849: /* Update the quadratic model */
850: PetscCall(PetscInfo(tao, "Get Quad, size: %" PetscInt_FMT ", points: %" PetscInt_FMT "\n", mfqP->n, mfqP->nmodelpoints));
851: PetscCall(getquadpounders(mfqP));
852: PetscCall(VecGetArrayRead(mfqP->Fhist[mfqP->minindex], &fmin));
853: PetscCallBLAS("BLAScopy", BLAScopy_(&blasm, fmin, &ione, mfqP->C, &ione));
854: /* G = G*(delta/deltaold) + Gdel */
855: ratio = mfqP->delta / deltaold;
856: iblas = blasm * blasn;
857: PetscCallBLAS("BLASscal", BLASscal_(&iblas, &ratio, mfqP->Fdiff, &ione));
858: PetscCallBLAS("BLASaxpy", BLASaxpy_(&iblas, &one, mfqP->Gdel, &ione, mfqP->Fdiff, &ione));
859: /* H = H*(delta/deltaold)^2 + Hdel */
860: iblas = blasm * blasn * blasn;
861: ratio *= ratio;
862: PetscCallBLAS("BLASscal", BLASscal_(&iblas, &ratio, mfqP->H, &ione));
863: PetscCallBLAS("BLASaxpy", BLASaxpy_(&iblas, &one, mfqP->Hdel, &ione, mfqP->H, &ione));
865: /* Get residuals */
866: PetscCall(pounders_update_res(tao));
868: /* Export solution and gradient residual to TAO */
869: PetscCall(VecCopy(mfqP->Xhist[mfqP->minindex], tao->solution));
870: PetscCall(VecSetValues(tao->gradient, mfqP->n, mfqP->indices, mfqP->Gres, INSERT_VALUES));
871: PetscCall(VecAssemblyBegin(tao->gradient));
872: PetscCall(VecAssemblyEnd(tao->gradient));
873: PetscCall(VecNorm(tao->gradient, NORM_2, &gnorm));
874: gnorm *= mfqP->delta;
875: /* final criticality test */
876: PetscCall(TaoLogConvergenceHistory(tao, minnorm, gnorm, 0.0, tao->ksp_its));
877: PetscCall(TaoMonitor(tao, tao->niter, minnorm, gnorm, 0.0, step));
878: PetscUseTypeMethod(tao, convergencetest, tao->cnvP);
879: /* test for repeated model */
880: if (mfqP->nmodelpoints == mfqP->last_nmodelpoints) {
881: same = PETSC_TRUE;
882: } else {
883: same = PETSC_FALSE;
884: }
885: for (i = 0; i < mfqP->nmodelpoints; i++) {
886: if (same) {
887: if (mfqP->model_indices[i] == mfqP->last_model_indices[i]) {
888: same = PETSC_TRUE;
889: } else {
890: same = PETSC_FALSE;
891: }
892: }
893: mfqP->last_model_indices[i] = mfqP->model_indices[i];
894: }
895: mfqP->last_nmodelpoints = mfqP->nmodelpoints;
896: if (same && mfqP->delta == deltaold) {
897: PetscCall(PetscInfo(tao, "Identical model used in successive iterations\n"));
898: tao->reason = TAO_CONVERGED_STEPTOL;
899: }
900: }
901: PetscFunctionReturn(PETSC_SUCCESS);
902: }
904: static PetscErrorCode TaoSetUp_POUNDERS(Tao tao)
905: {
906: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
907: PetscInt i, j;
908: IS isfloc, isfglob, isxloc, isxglob;
910: PetscFunctionBegin;
911: if (!tao->gradient) PetscCall(VecDuplicate(tao->solution, &tao->gradient));
912: if (!tao->stepdirection) PetscCall(VecDuplicate(tao->solution, &tao->stepdirection));
913: PetscCall(VecGetSize(tao->solution, &mfqP->n));
914: PetscCall(VecGetSize(tao->ls_res, &mfqP->m));
915: mfqP->c1 = PetscSqrtReal((PetscReal)mfqP->n);
916: if (mfqP->npmax == PETSC_DEFAULT) mfqP->npmax = 2 * mfqP->n + 1;
917: mfqP->npmax = PetscMin((mfqP->n + 1) * (mfqP->n + 2) / 2, mfqP->npmax);
918: mfqP->npmax = PetscMax(mfqP->npmax, mfqP->n + 2);
920: PetscCall(PetscMalloc1(tao->max_funcs + 100, &mfqP->Xhist));
921: PetscCall(PetscMalloc1(tao->max_funcs + 100, &mfqP->Fhist));
922: for (i = 0; i < mfqP->n + 1; i++) {
923: PetscCall(VecDuplicate(tao->solution, &mfqP->Xhist[i]));
924: PetscCall(VecDuplicate(tao->ls_res, &mfqP->Fhist[i]));
925: }
926: PetscCall(VecDuplicate(tao->solution, &mfqP->workxvec));
927: PetscCall(VecDuplicate(tao->ls_res, &mfqP->workfvec));
928: mfqP->nHist = 0;
930: PetscCall(PetscMalloc1(tao->max_funcs + 100, &mfqP->Fres));
931: PetscCall(PetscMalloc1(mfqP->npmax * mfqP->m, &mfqP->RES));
932: PetscCall(PetscMalloc1(mfqP->n, &mfqP->work));
933: PetscCall(PetscMalloc1(mfqP->n, &mfqP->work2));
934: PetscCall(PetscMalloc1(mfqP->n, &mfqP->work3));
935: PetscCall(PetscMalloc1(PetscMax(mfqP->m, mfqP->n + 1), &mfqP->mwork));
936: PetscCall(PetscMalloc1(mfqP->npmax - mfqP->n - 1, &mfqP->omega));
937: PetscCall(PetscMalloc1(mfqP->n * (mfqP->n + 1) / 2, &mfqP->beta));
938: PetscCall(PetscMalloc1(mfqP->n + 1, &mfqP->alpha));
940: PetscCall(PetscMalloc1(mfqP->n * mfqP->n * mfqP->m, &mfqP->H));
941: PetscCall(PetscMalloc1(mfqP->npmax * mfqP->npmax, &mfqP->Q));
942: PetscCall(PetscMalloc1(mfqP->npmax * mfqP->npmax, &mfqP->Q_tmp));
943: PetscCall(PetscMalloc1(mfqP->n * (mfqP->n + 1) / 2 * (mfqP->npmax), &mfqP->L));
944: PetscCall(PetscMalloc1(mfqP->n * (mfqP->n + 1) / 2 * (mfqP->npmax), &mfqP->L_tmp));
945: PetscCall(PetscMalloc1(mfqP->n * (mfqP->n + 1) / 2 * (mfqP->npmax), &mfqP->L_save));
946: PetscCall(PetscMalloc1(mfqP->n * (mfqP->n + 1) / 2 * (mfqP->npmax), &mfqP->N));
947: PetscCall(PetscMalloc1(mfqP->npmax * (mfqP->n + 1), &mfqP->M));
948: PetscCall(PetscMalloc1(mfqP->npmax * (mfqP->npmax - mfqP->n - 1), &mfqP->Z));
949: PetscCall(PetscMalloc1(mfqP->npmax, &mfqP->tau));
950: PetscCall(PetscMalloc1(mfqP->npmax, &mfqP->tau_tmp));
951: mfqP->nmax = PetscMax(5 * mfqP->npmax, mfqP->n * (mfqP->n + 1) / 2);
952: PetscCall(PetscMalloc1(mfqP->nmax, &mfqP->npmaxwork));
953: PetscCall(PetscMalloc1(mfqP->nmax, &mfqP->npmaxiwork));
954: PetscCall(PetscMalloc1(mfqP->n, &mfqP->xmin));
955: PetscCall(PetscMalloc1(mfqP->m, &mfqP->C));
956: PetscCall(PetscMalloc1(mfqP->m * mfqP->n, &mfqP->Fdiff));
957: PetscCall(PetscMalloc1(mfqP->npmax * mfqP->n, &mfqP->Disp));
958: PetscCall(PetscMalloc1(mfqP->n, &mfqP->Gres));
959: PetscCall(PetscMalloc1(mfqP->n * mfqP->n, &mfqP->Hres));
960: PetscCall(PetscMalloc1(mfqP->n * mfqP->n, &mfqP->Gpoints));
961: PetscCall(PetscMalloc1(mfqP->npmax, &mfqP->model_indices));
962: PetscCall(PetscMalloc1(mfqP->npmax, &mfqP->last_model_indices));
963: PetscCall(PetscMalloc1(mfqP->n, &mfqP->Xsubproblem));
964: PetscCall(PetscMalloc1(mfqP->m * mfqP->n, &mfqP->Gdel));
965: PetscCall(PetscMalloc1(mfqP->n * mfqP->n * mfqP->m, &mfqP->Hdel));
966: PetscCall(PetscMalloc1(PetscMax(mfqP->m, mfqP->n), &mfqP->indices));
967: PetscCall(PetscMalloc1(mfqP->n, &mfqP->iwork));
968: PetscCall(PetscMalloc1(mfqP->m * mfqP->m, &mfqP->w));
969: for (i = 0; i < mfqP->m; i++) {
970: for (j = 0; j < mfqP->m; j++) {
971: if (i == j) {
972: mfqP->w[i + mfqP->m * j] = 1.0;
973: } else {
974: mfqP->w[i + mfqP->m * j] = 0.0;
975: }
976: }
977: }
978: for (i = 0; i < PetscMax(mfqP->m, mfqP->n); i++) mfqP->indices[i] = i;
979: PetscCallMPI(MPI_Comm_size(((PetscObject)tao)->comm, &mfqP->size));
980: if (mfqP->size > 1) {
981: PetscCall(VecCreateSeq(PETSC_COMM_SELF, mfqP->n, &mfqP->localx));
982: PetscCall(VecCreateSeq(PETSC_COMM_SELF, mfqP->n, &mfqP->localxmin));
983: PetscCall(VecCreateSeq(PETSC_COMM_SELF, mfqP->m, &mfqP->localf));
984: PetscCall(VecCreateSeq(PETSC_COMM_SELF, mfqP->m, &mfqP->localfmin));
985: PetscCall(ISCreateStride(MPI_COMM_SELF, mfqP->n, 0, 1, &isxloc));
986: PetscCall(ISCreateStride(MPI_COMM_SELF, mfqP->n, 0, 1, &isxglob));
987: PetscCall(ISCreateStride(MPI_COMM_SELF, mfqP->m, 0, 1, &isfloc));
988: PetscCall(ISCreateStride(MPI_COMM_SELF, mfqP->m, 0, 1, &isfglob));
990: PetscCall(VecScatterCreate(tao->solution, isxglob, mfqP->localx, isxloc, &mfqP->scatterx));
991: PetscCall(VecScatterCreate(tao->ls_res, isfglob, mfqP->localf, isfloc, &mfqP->scatterf));
993: PetscCall(ISDestroy(&isxloc));
994: PetscCall(ISDestroy(&isxglob));
995: PetscCall(ISDestroy(&isfloc));
996: PetscCall(ISDestroy(&isfglob));
997: }
999: if (!mfqP->usegqt) {
1000: KSP ksp;
1001: PC pc;
1002: PetscCall(VecCreateSeqWithArray(PETSC_COMM_SELF, mfqP->n, mfqP->n, mfqP->Xsubproblem, &mfqP->subx));
1003: PetscCall(VecCreateSeq(PETSC_COMM_SELF, mfqP->n, &mfqP->subxl));
1004: PetscCall(VecDuplicate(mfqP->subxl, &mfqP->subb));
1005: PetscCall(VecDuplicate(mfqP->subxl, &mfqP->subxu));
1006: PetscCall(VecDuplicate(mfqP->subxl, &mfqP->subpdel));
1007: PetscCall(VecDuplicate(mfqP->subxl, &mfqP->subndel));
1008: PetscCall(TaoCreate(PETSC_COMM_SELF, &mfqP->subtao));
1009: PetscCall(PetscObjectIncrementTabLevel((PetscObject)mfqP->subtao, (PetscObject)tao, 1));
1010: PetscCall(TaoSetType(mfqP->subtao, TAOBNTR));
1011: PetscCall(TaoSetOptionsPrefix(mfqP->subtao, "pounders_subsolver_"));
1012: PetscCall(TaoSetSolution(mfqP->subtao, mfqP->subx));
1013: PetscCall(TaoSetObjectiveAndGradient(mfqP->subtao, NULL, pounders_fg, (void *)mfqP));
1014: PetscCall(TaoSetMaximumIterations(mfqP->subtao, mfqP->gqt_maxits));
1015: PetscCall(TaoSetFromOptions(mfqP->subtao));
1016: PetscCall(TaoGetKSP(mfqP->subtao, &ksp));
1017: if (ksp) {
1018: PetscCall(KSPGetPC(ksp, &pc));
1019: PetscCall(PCSetType(pc, PCNONE));
1020: }
1021: PetscCall(TaoSetVariableBounds(mfqP->subtao, mfqP->subxl, mfqP->subxu));
1022: PetscCall(MatCreateSeqDense(PETSC_COMM_SELF, mfqP->n, mfqP->n, mfqP->Hres, &mfqP->subH));
1023: PetscCall(TaoSetHessian(mfqP->subtao, mfqP->subH, mfqP->subH, pounders_h, (void *)mfqP));
1024: }
1025: PetscFunctionReturn(PETSC_SUCCESS);
1026: }
1028: static PetscErrorCode TaoDestroy_POUNDERS(Tao tao)
1029: {
1030: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
1031: PetscInt i;
1033: PetscFunctionBegin;
1034: if (!mfqP->usegqt) {
1035: PetscCall(TaoDestroy(&mfqP->subtao));
1036: PetscCall(VecDestroy(&mfqP->subx));
1037: PetscCall(VecDestroy(&mfqP->subxl));
1038: PetscCall(VecDestroy(&mfqP->subxu));
1039: PetscCall(VecDestroy(&mfqP->subb));
1040: PetscCall(VecDestroy(&mfqP->subpdel));
1041: PetscCall(VecDestroy(&mfqP->subndel));
1042: PetscCall(MatDestroy(&mfqP->subH));
1043: }
1044: PetscCall(PetscFree(mfqP->Fres));
1045: PetscCall(PetscFree(mfqP->RES));
1046: PetscCall(PetscFree(mfqP->work));
1047: PetscCall(PetscFree(mfqP->work2));
1048: PetscCall(PetscFree(mfqP->work3));
1049: PetscCall(PetscFree(mfqP->mwork));
1050: PetscCall(PetscFree(mfqP->omega));
1051: PetscCall(PetscFree(mfqP->beta));
1052: PetscCall(PetscFree(mfqP->alpha));
1053: PetscCall(PetscFree(mfqP->H));
1054: PetscCall(PetscFree(mfqP->Q));
1055: PetscCall(PetscFree(mfqP->Q_tmp));
1056: PetscCall(PetscFree(mfqP->L));
1057: PetscCall(PetscFree(mfqP->L_tmp));
1058: PetscCall(PetscFree(mfqP->L_save));
1059: PetscCall(PetscFree(mfqP->N));
1060: PetscCall(PetscFree(mfqP->M));
1061: PetscCall(PetscFree(mfqP->Z));
1062: PetscCall(PetscFree(mfqP->tau));
1063: PetscCall(PetscFree(mfqP->tau_tmp));
1064: PetscCall(PetscFree(mfqP->npmaxwork));
1065: PetscCall(PetscFree(mfqP->npmaxiwork));
1066: PetscCall(PetscFree(mfqP->xmin));
1067: PetscCall(PetscFree(mfqP->C));
1068: PetscCall(PetscFree(mfqP->Fdiff));
1069: PetscCall(PetscFree(mfqP->Disp));
1070: PetscCall(PetscFree(mfqP->Gres));
1071: PetscCall(PetscFree(mfqP->Hres));
1072: PetscCall(PetscFree(mfqP->Gpoints));
1073: PetscCall(PetscFree(mfqP->model_indices));
1074: PetscCall(PetscFree(mfqP->last_model_indices));
1075: PetscCall(PetscFree(mfqP->Xsubproblem));
1076: PetscCall(PetscFree(mfqP->Gdel));
1077: PetscCall(PetscFree(mfqP->Hdel));
1078: PetscCall(PetscFree(mfqP->indices));
1079: PetscCall(PetscFree(mfqP->iwork));
1080: PetscCall(PetscFree(mfqP->w));
1081: for (i = 0; i < mfqP->nHist; i++) {
1082: PetscCall(VecDestroy(&mfqP->Xhist[i]));
1083: PetscCall(VecDestroy(&mfqP->Fhist[i]));
1084: }
1085: PetscCall(VecDestroy(&mfqP->workxvec));
1086: PetscCall(VecDestroy(&mfqP->workfvec));
1087: PetscCall(PetscFree(mfqP->Xhist));
1088: PetscCall(PetscFree(mfqP->Fhist));
1090: if (mfqP->size > 1) {
1091: PetscCall(VecDestroy(&mfqP->localx));
1092: PetscCall(VecDestroy(&mfqP->localxmin));
1093: PetscCall(VecDestroy(&mfqP->localf));
1094: PetscCall(VecDestroy(&mfqP->localfmin));
1095: }
1096: PetscCall(PetscFree(tao->data));
1097: PetscFunctionReturn(PETSC_SUCCESS);
1098: }
1100: static PetscErrorCode TaoSetFromOptions_POUNDERS(Tao tao, PetscOptionItems *PetscOptionsObject)
1101: {
1102: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
1104: PetscFunctionBegin;
1105: PetscOptionsHeadBegin(PetscOptionsObject, "POUNDERS method for least-squares optimization");
1106: PetscCall(PetscOptionsReal("-tao_pounders_delta", "initial delta", "", mfqP->delta, &mfqP->delta0, NULL));
1107: mfqP->delta = mfqP->delta0;
1108: PetscCall(PetscOptionsInt("-tao_pounders_npmax", "max number of points in model", "", mfqP->npmax, &mfqP->npmax, NULL));
1109: PetscCall(PetscOptionsBool("-tao_pounders_gqt", "use gqt algorithm for subproblem", "", mfqP->usegqt, &mfqP->usegqt, NULL));
1110: PetscOptionsHeadEnd();
1111: PetscFunctionReturn(PETSC_SUCCESS);
1112: }
1114: static PetscErrorCode TaoView_POUNDERS(Tao tao, PetscViewer viewer)
1115: {
1116: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
1117: PetscBool isascii;
1119: PetscFunctionBegin;
1120: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
1121: if (isascii) {
1122: PetscCall(PetscViewerASCIIPrintf(viewer, "initial delta: %g\n", (double)mfqP->delta0));
1123: PetscCall(PetscViewerASCIIPrintf(viewer, "final delta: %g\n", (double)mfqP->delta));
1124: PetscCall(PetscViewerASCIIPrintf(viewer, "model points: %" PetscInt_FMT "\n", mfqP->nmodelpoints));
1125: if (mfqP->usegqt) {
1126: PetscCall(PetscViewerASCIIPrintf(viewer, "subproblem solver: gqt\n"));
1127: } else {
1128: PetscCall(TaoView(mfqP->subtao, viewer));
1129: }
1130: }
1131: PetscFunctionReturn(PETSC_SUCCESS);
1132: }
1133: /*MC
1134: TAOPOUNDERS - POUNDERS derivate-free model-based algorithm for nonlinear least squares
1136: Options Database Keys:
1137: + -tao_pounders_delta - initial step length
1138: . -tao_pounders_npmax - maximum number of points in model
1139: - -tao_pounders_gqt - use gqt algorithm for subproblem instead of TRON
1141: Level: beginner
1143: M*/
1145: PETSC_EXTERN PetscErrorCode TaoCreate_POUNDERS(Tao tao)
1146: {
1147: TAO_POUNDERS *mfqP = (TAO_POUNDERS *)tao->data;
1149: PetscFunctionBegin;
1150: tao->ops->setup = TaoSetUp_POUNDERS;
1151: tao->ops->solve = TaoSolve_POUNDERS;
1152: tao->ops->view = TaoView_POUNDERS;
1153: tao->ops->setfromoptions = TaoSetFromOptions_POUNDERS;
1154: tao->ops->destroy = TaoDestroy_POUNDERS;
1156: PetscCall(PetscNew(&mfqP));
1157: tao->data = (void *)mfqP;
1158: /* Override default settings (unless already changed) */
1159: if (!tao->max_it_changed) tao->max_it = 2000;
1160: if (!tao->max_funcs_changed) tao->max_funcs = 4000;
1161: mfqP->npmax = PETSC_DEFAULT;
1162: mfqP->delta0 = 0.1;
1163: mfqP->delta = 0.1;
1164: mfqP->deltamax = 1e3;
1165: mfqP->deltamin = 1e-6;
1166: mfqP->c2 = 10.0;
1167: mfqP->theta1 = 1e-5;
1168: mfqP->theta2 = 1e-4;
1169: mfqP->gamma0 = 0.5;
1170: mfqP->gamma1 = 2.0;
1171: mfqP->eta0 = 0.0;
1172: mfqP->eta1 = 0.1;
1173: mfqP->usegqt = PETSC_FALSE;
1174: mfqP->gqt_rtol = 0.001;
1175: mfqP->gqt_maxits = 50;
1176: mfqP->workxvec = NULL;
1177: PetscFunctionReturn(PETSC_SUCCESS);
1178: }