Actual source code: dt.c

  1: /* Discretization tools */

  3: #include <petscdt.h>
  4: #include <petscblaslapack.h>
  5: #include <petsc/private/petscimpl.h>
  6: #include <petsc/private/dtimpl.h>
  7: #include <petsc/private/petscfeimpl.h>
  8: #include <petscviewer.h>
  9: #include <petscdmplex.h>
 10: #include <petscdmshell.h>

 12: #if defined(PETSC_HAVE_MPFR)
 13:   #include <mpfr.h>
 14: #endif

 16: const char *const        PetscDTNodeTypes_shifted[] = {"default", "gaussjacobi", "equispaced", "tanhsinh", "PetscDTNodeType", "PETSCDTNODES_", NULL};
 17: const char *const *const PetscDTNodeTypes           = PetscDTNodeTypes_shifted + 1;

 19: const char *const        PetscDTSimplexQuadratureTypes_shifted[] = {"default", "conic", "minsym", "diagsym", "PetscDTSimplexQuadratureType", "PETSCDTSIMPLEXQUAD_", NULL};
 20: const char *const *const PetscDTSimplexQuadratureTypes           = PetscDTSimplexQuadratureTypes_shifted + 1;

 22: static PetscBool GolubWelschCite       = PETSC_FALSE;
 23: const char       GolubWelschCitation[] = "@article{GolubWelsch1969,\n"
 24:                                          "  author  = {Golub and Welsch},\n"
 25:                                          "  title   = {Calculation of Quadrature Rules},\n"
 26:                                          "  journal = {Math. Comp.},\n"
 27:                                          "  volume  = {23},\n"
 28:                                          "  number  = {106},\n"
 29:                                          "  pages   = {221--230},\n"
 30:                                          "  year    = {1969}\n}\n";

 32: /* Numerical tests in src/dm/dt/tests/ex1.c show that when computing the nodes and weights of Gauss-Jacobi
 33:    quadrature rules:

 35:    - in double precision, Newton's method and Golub & Welsch both work for moderate degrees (< 100),
 36:    - in single precision, Newton's method starts producing incorrect roots around n = 15, but
 37:      the weights from Golub & Welsch become a problem before then: they produces errors
 38:      in computing the Jacobi-polynomial Gram matrix around n = 6.

 40:    So we default to Newton's method (required fewer dependencies) */
 41: PetscBool PetscDTGaussQuadratureNewton_Internal = PETSC_TRUE;

 43: PetscClassId PETSCQUADRATURE_CLASSID = 0;

 45: /*@
 46:   PetscQuadratureCreate - Create a `PetscQuadrature` object

 48:   Collective

 50:   Input Parameter:
 51: . comm - The communicator for the `PetscQuadrature` object

 53:   Output Parameter:
 54: . q - The `PetscQuadrature` object

 56:   Level: beginner

 58: .seealso: `PetscQuadrature`, `Petscquadraturedestroy()`, `PetscQuadratureGetData()`
 59: @*/
 60: PetscErrorCode PetscQuadratureCreate(MPI_Comm comm, PetscQuadrature *q)
 61: {
 62:   PetscFunctionBegin;
 63:   PetscAssertPointer(q, 2);
 64:   PetscCall(DMInitializePackage());
 65:   PetscCall(PetscHeaderCreate(*q, PETSCQUADRATURE_CLASSID, "PetscQuadrature", "Quadrature", "DT", comm, PetscQuadratureDestroy, PetscQuadratureView));
 66:   (*q)->ct        = DM_POLYTOPE_UNKNOWN;
 67:   (*q)->dim       = -1;
 68:   (*q)->Nc        = 1;
 69:   (*q)->order     = -1;
 70:   (*q)->numPoints = 0;
 71:   (*q)->points    = NULL;
 72:   (*q)->weights   = NULL;
 73:   PetscFunctionReturn(PETSC_SUCCESS);
 74: }

 76: /*@
 77:   PetscQuadratureDuplicate - Create a deep copy of the `PetscQuadrature` object

 79:   Collective

 81:   Input Parameter:
 82: . q - The `PetscQuadrature` object

 84:   Output Parameter:
 85: . r - The new `PetscQuadrature` object

 87:   Level: beginner

 89: .seealso: `PetscQuadrature`, `PetscQuadratureCreate()`, `PetscQuadratureDestroy()`, `PetscQuadratureGetData()`
 90: @*/
 91: PetscErrorCode PetscQuadratureDuplicate(PetscQuadrature q, PetscQuadrature *r)
 92: {
 93:   DMPolytopeType   ct;
 94:   PetscInt         order, dim, Nc, Nq;
 95:   const PetscReal *points, *weights;
 96:   PetscReal       *p, *w;

 98:   PetscFunctionBegin;
 99:   PetscAssertPointer(q, 1);
100:   PetscCall(PetscQuadratureCreate(PetscObjectComm((PetscObject)q), r));
101:   PetscCall(PetscQuadratureGetCellType(q, &ct));
102:   PetscCall(PetscQuadratureSetCellType(*r, ct));
103:   PetscCall(PetscQuadratureGetOrder(q, &order));
104:   PetscCall(PetscQuadratureSetOrder(*r, order));
105:   PetscCall(PetscQuadratureGetData(q, &dim, &Nc, &Nq, &points, &weights));
106:   PetscCall(PetscMalloc1(Nq * dim, &p));
107:   PetscCall(PetscMalloc1(Nq * Nc, &w));
108:   PetscCall(PetscArraycpy(p, points, Nq * dim));
109:   PetscCall(PetscArraycpy(w, weights, Nc * Nq));
110:   PetscCall(PetscQuadratureSetData(*r, dim, Nc, Nq, p, w));
111:   PetscFunctionReturn(PETSC_SUCCESS);
112: }

114: /*@
115:   PetscQuadratureDestroy - Destroys a `PetscQuadrature` object

117:   Collective

119:   Input Parameter:
120: . q - The `PetscQuadrature` object

122:   Level: beginner

124: .seealso: `PetscQuadrature`, `PetscQuadratureCreate()`, `PetscQuadratureGetData()`
125: @*/
126: PetscErrorCode PetscQuadratureDestroy(PetscQuadrature *q)
127: {
128:   PetscFunctionBegin;
129:   if (!*q) PetscFunctionReturn(PETSC_SUCCESS);
131:   if (--((PetscObject)*q)->refct > 0) {
132:     *q = NULL;
133:     PetscFunctionReturn(PETSC_SUCCESS);
134:   }
135:   PetscCall(PetscFree((*q)->points));
136:   PetscCall(PetscFree((*q)->weights));
137:   PetscCall(PetscHeaderDestroy(q));
138:   PetscFunctionReturn(PETSC_SUCCESS);
139: }

141: /*@
142:   PetscQuadratureGetCellType - Return the cell type of the integration domain

144:   Not Collective

146:   Input Parameter:
147: . q - The `PetscQuadrature` object

149:   Output Parameter:
150: . ct - The cell type of the integration domain

152:   Level: intermediate

154: .seealso: `PetscQuadrature`, `PetscQuadratureSetCellType()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
155: @*/
156: PetscErrorCode PetscQuadratureGetCellType(PetscQuadrature q, DMPolytopeType *ct)
157: {
158:   PetscFunctionBegin;
160:   PetscAssertPointer(ct, 2);
161:   *ct = q->ct;
162:   PetscFunctionReturn(PETSC_SUCCESS);
163: }

165: /*@
166:   PetscQuadratureSetCellType - Set the cell type of the integration domain

168:   Not Collective

170:   Input Parameters:
171: + q  - The `PetscQuadrature` object
172: - ct - The cell type of the integration domain

174:   Level: intermediate

176: .seealso: `PetscQuadrature`, `PetscQuadratureGetCellType()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
177: @*/
178: PetscErrorCode PetscQuadratureSetCellType(PetscQuadrature q, DMPolytopeType ct)
179: {
180:   PetscFunctionBegin;
182:   q->ct = ct;
183:   PetscFunctionReturn(PETSC_SUCCESS);
184: }

186: /*@
187:   PetscQuadratureGetOrder - Return the order of the method in the `PetscQuadrature`

189:   Not Collective

191:   Input Parameter:
192: . q - The `PetscQuadrature` object

194:   Output Parameter:
195: . order - The order of the quadrature, i.e. the highest degree polynomial that is exactly integrated

197:   Level: intermediate

199: .seealso: `PetscQuadrature`, `PetscQuadratureSetOrder()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
200: @*/
201: PetscErrorCode PetscQuadratureGetOrder(PetscQuadrature q, PetscInt *order)
202: {
203:   PetscFunctionBegin;
205:   PetscAssertPointer(order, 2);
206:   *order = q->order;
207:   PetscFunctionReturn(PETSC_SUCCESS);
208: }

210: /*@
211:   PetscQuadratureSetOrder - Set the order of the method in the `PetscQuadrature`

213:   Not Collective

215:   Input Parameters:
216: + q     - The `PetscQuadrature` object
217: - order - The order of the quadrature, i.e. the highest degree polynomial that is exactly integrated

219:   Level: intermediate

221: .seealso: `PetscQuadrature`, `PetscQuadratureGetOrder()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
222: @*/
223: PetscErrorCode PetscQuadratureSetOrder(PetscQuadrature q, PetscInt order)
224: {
225:   PetscFunctionBegin;
227:   q->order = order;
228:   PetscFunctionReturn(PETSC_SUCCESS);
229: }

231: /*@
232:   PetscQuadratureGetNumComponents - Return the number of components for functions to be integrated

234:   Not Collective

236:   Input Parameter:
237: . q - The `PetscQuadrature` object

239:   Output Parameter:
240: . Nc - The number of components

242:   Level: intermediate

244:   Note:
245:   We are performing an integral $\int f(x) w(x) dx$, where both $f$ and $w$ (the weight) have `Nc` components.

247: .seealso: `PetscQuadrature`, `PetscQuadratureSetNumComponents()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
248: @*/
249: PetscErrorCode PetscQuadratureGetNumComponents(PetscQuadrature q, PetscInt *Nc)
250: {
251:   PetscFunctionBegin;
253:   PetscAssertPointer(Nc, 2);
254:   *Nc = q->Nc;
255:   PetscFunctionReturn(PETSC_SUCCESS);
256: }

258: /*@
259:   PetscQuadratureSetNumComponents - Sets the number of components for functions to be integrated

261:   Not Collective

263:   Input Parameters:
264: + q  - The `PetscQuadrature` object
265: - Nc - The number of components

267:   Level: intermediate

269:   Note:
270:   We are performing an integral $\int f(x) w(x) dx$, where both $f$ and $w$ (the weight) have `Nc` components.

272: .seealso: `PetscQuadrature`, `PetscQuadratureGetNumComponents()`, `PetscQuadratureGetData()`, `PetscQuadratureSetData()`
273: @*/
274: PetscErrorCode PetscQuadratureSetNumComponents(PetscQuadrature q, PetscInt Nc)
275: {
276:   PetscFunctionBegin;
278:   q->Nc = Nc;
279:   PetscFunctionReturn(PETSC_SUCCESS);
280: }

282: /*@C
283:   PetscQuadratureGetData - Returns the data defining the `PetscQuadrature`

285:   Not Collective

287:   Input Parameter:
288: . q - The `PetscQuadrature` object

290:   Output Parameters:
291: + dim     - The spatial dimension
292: . Nc      - The number of components
293: . npoints - The number of quadrature points
294: . points  - The coordinates of each quadrature point
295: - weights - The weight of each quadrature point

297:   Level: intermediate

299:   Note:
300:   All output arguments are optional, pass `NULL` for any argument not required

302:   Fortran Note:
303:   Call `PetscQuadratureRestoreData()` when you are done with the data

305: .seealso: `PetscQuadrature`, `PetscQuadratureCreate()`, `PetscQuadratureSetData()`
306: @*/
307: PetscErrorCode PetscQuadratureGetData(PetscQuadrature q, PeOp PetscInt *dim, PeOp PetscInt *Nc, PeOp PetscInt *npoints, PeOp const PetscReal *points[], PeOp const PetscReal *weights[])
308: {
309:   PetscFunctionBegin;
311:   if (dim) {
312:     PetscAssertPointer(dim, 2);
313:     *dim = q->dim;
314:   }
315:   if (Nc) {
316:     PetscAssertPointer(Nc, 3);
317:     *Nc = q->Nc;
318:   }
319:   if (npoints) {
320:     PetscAssertPointer(npoints, 4);
321:     *npoints = q->numPoints;
322:   }
323:   if (points) {
324:     PetscAssertPointer(points, 5);
325:     *points = q->points;
326:   }
327:   if (weights) {
328:     PetscAssertPointer(weights, 6);
329:     *weights = q->weights;
330:   }
331:   PetscFunctionReturn(PETSC_SUCCESS);
332: }

334: /*@
335:   PetscQuadratureEqual - determine whether two quadratures are equivalent

337:   Input Parameters:
338: + A - A `PetscQuadrature` object
339: - B - Another `PetscQuadrature` object

341:   Output Parameter:
342: . equal - `PETSC_TRUE` if the quadratures are the same

344:   Level: intermediate

346: .seealso: `PetscQuadrature`, `PetscQuadratureCreate()`
347: @*/
348: PetscErrorCode PetscQuadratureEqual(PetscQuadrature A, PetscQuadrature B, PetscBool *equal)
349: {
350:   PetscFunctionBegin;
353:   PetscAssertPointer(equal, 3);
354:   *equal = PETSC_FALSE;
355:   if (A->ct != B->ct || A->dim != B->dim || A->Nc != B->Nc || A->order != B->order || A->numPoints != B->numPoints) PetscFunctionReturn(PETSC_SUCCESS);
356:   for (PetscInt i = 0; i < A->numPoints * A->dim; i++) {
357:     if (PetscAbsReal(A->points[i] - B->points[i]) > PETSC_SMALL) PetscFunctionReturn(PETSC_SUCCESS);
358:   }
359:   if (!A->weights && !B->weights) {
360:     *equal = PETSC_TRUE;
361:     PetscFunctionReturn(PETSC_SUCCESS);
362:   }
363:   if (A->weights && B->weights) {
364:     for (PetscInt i = 0; i < A->numPoints; i++) {
365:       if (PetscAbsReal(A->weights[i] - B->weights[i]) > PETSC_SMALL) PetscFunctionReturn(PETSC_SUCCESS);
366:     }
367:     *equal = PETSC_TRUE;
368:   }
369:   PetscFunctionReturn(PETSC_SUCCESS);
370: }

372: static PetscErrorCode PetscDTJacobianInverse_Internal(PetscInt m, PetscInt n, const PetscReal J[], PetscReal Jinv[])
373: {
374:   PetscScalar *Js, *Jinvs;
375:   PetscInt     i, j, k;
376:   PetscBLASInt bm, bn;

378:   PetscFunctionBegin;
379:   if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS);
380:   PetscCall(PetscBLASIntCast(m, &bm));
381:   PetscCall(PetscBLASIntCast(n, &bn));
382: #if defined(PETSC_USE_COMPLEX)
383:   PetscCall(PetscMalloc2(m * n, &Js, m * n, &Jinvs));
384:   for (i = 0; i < m * n; i++) Js[i] = J[i];
385: #else
386:   Js    = (PetscReal *)J;
387:   Jinvs = Jinv;
388: #endif
389:   if (m == n) {
390:     PetscBLASInt *pivots;
391:     PetscScalar  *W;

393:     PetscCall(PetscMalloc2(m, &pivots, m, &W));

395:     PetscCall(PetscArraycpy(Jinvs, Js, m * m));
396:     PetscCallLAPACKInfo("LAPACKgetrf", LAPACKgetrf_(&bm, &bm, Jinvs, &bm, pivots, &info));
397:     PetscCallLAPACKInfo("LAPACKgetri", LAPACKgetri_(&bm, Jinvs, &bm, pivots, W, &bm, &info));
398:     PetscCall(PetscFree2(pivots, W));
399:   } else if (m < n) {
400:     PetscScalar  *JJT;
401:     PetscBLASInt *pivots;
402:     PetscScalar  *W;

404:     PetscCall(PetscMalloc1(m * m, &JJT));
405:     PetscCall(PetscMalloc2(m, &pivots, m, &W));
406:     for (i = 0; i < m; i++) {
407:       for (j = 0; j < m; j++) {
408:         PetscScalar val = 0.;

410:         for (k = 0; k < n; k++) val += Js[i * n + k] * Js[j * n + k];
411:         JJT[i * m + j] = val;
412:       }
413:     }

415:     PetscCallLAPACKInfo("LAPACKgetrf", LAPACKgetrf_(&bm, &bm, JJT, &bm, pivots, &info));
416:     PetscCallLAPACKInfo("LAPACKgetri", LAPACKgetri_(&bm, JJT, &bm, pivots, W, &bm, &info));
417:     for (i = 0; i < n; i++) {
418:       for (j = 0; j < m; j++) {
419:         PetscScalar val = 0.;

421:         for (k = 0; k < m; k++) val += Js[k * n + i] * JJT[k * m + j];
422:         Jinvs[i * m + j] = val;
423:       }
424:     }
425:     PetscCall(PetscFree2(pivots, W));
426:     PetscCall(PetscFree(JJT));
427:   } else {
428:     PetscScalar  *JTJ;
429:     PetscBLASInt *pivots;
430:     PetscScalar  *W;

432:     PetscCall(PetscMalloc1(n * n, &JTJ));
433:     PetscCall(PetscMalloc2(n, &pivots, n, &W));
434:     for (i = 0; i < n; i++) {
435:       for (j = 0; j < n; j++) {
436:         PetscScalar val = 0.;

438:         for (k = 0; k < m; k++) val += Js[k * n + i] * Js[k * n + j];
439:         JTJ[i * n + j] = val;
440:       }
441:     }

443:     PetscCallLAPACKInfo("LAPACKgetrf", LAPACKgetrf_(&bn, &bn, JTJ, &bn, pivots, &info));
444:     PetscCallLAPACKInfo("LAPACKgetri", LAPACKgetri_(&bn, JTJ, &bn, pivots, W, &bn, &info));
445:     for (i = 0; i < n; i++) {
446:       for (j = 0; j < m; j++) {
447:         PetscScalar val = 0.;

449:         for (k = 0; k < n; k++) val += JTJ[i * n + k] * Js[j * n + k];
450:         Jinvs[i * m + j] = val;
451:       }
452:     }
453:     PetscCall(PetscFree2(pivots, W));
454:     PetscCall(PetscFree(JTJ));
455:   }
456: #if defined(PETSC_USE_COMPLEX)
457:   for (i = 0; i < m * n; i++) Jinv[i] = PetscRealPart(Jinvs[i]);
458:   PetscCall(PetscFree2(Js, Jinvs));
459: #endif
460:   PetscFunctionReturn(PETSC_SUCCESS);
461: }

463: /*@
464:   PetscQuadraturePushForward - Push forward a quadrature functional under an affine transformation.

466:   Collective

468:   Input Parameters:
469: + q           - the quadrature functional
470: . imageDim    - the dimension of the image of the transformation
471: . origin      - a point in the original space
472: . originImage - the image of the origin under the transformation
473: . J           - the Jacobian of the image: an [imageDim x dim] matrix in row major order
474: - formDegree  - transform the quadrature weights as k-forms of this form degree (if the number of components is a multiple of (dim choose `formDegree`),
475:                 it is assumed that they represent multiple k-forms) [see `PetscDTAltVPullback()` for interpretation of `formDegree`]

477:   Output Parameter:
478: . Jinvstarq - a quadrature rule where each point is the image of a point in the original quadrature rule, and where the k-form weights have
479:   been pulled-back by the pseudoinverse of `J` to the k-form weights in the image space.

481:   Level: intermediate

483:   Note:
484:   The new quadrature rule will have a different number of components if spaces have different dimensions.
485:   For example, pushing a 2-form forward from a two dimensional space to a three dimensional space changes the number of components from 1 to 3.

487: .seealso: `PetscQuadrature`, `PetscDTAltVPullback()`, `PetscDTAltVPullbackMatrix()`
488: @*/
489: PetscErrorCode PetscQuadraturePushForward(PetscQuadrature q, PetscInt imageDim, const PetscReal origin[], const PetscReal originImage[], const PetscReal J[], PetscInt formDegree, PetscQuadrature *Jinvstarq)
490: {
491:   PetscInt         dim, Nc, imageNc, formSize, Ncopies, imageFormSize, Npoints, pt, i, j, c;
492:   const PetscReal *points;
493:   const PetscReal *weights;
494:   PetscReal       *imagePoints, *imageWeights;
495:   PetscReal       *Jinv;
496:   PetscReal       *Jinvstar;

498:   PetscFunctionBegin;
500:   PetscCheck(imageDim >= PetscAbsInt(formDegree), PetscObjectComm((PetscObject)q), PETSC_ERR_ARG_INCOMP, "Cannot represent a %" PetscInt_FMT "-form in %" PetscInt_FMT " dimensions", PetscAbsInt(formDegree), imageDim);
501:   PetscCall(PetscQuadratureGetData(q, &dim, &Nc, &Npoints, &points, &weights));
502:   PetscCall(PetscDTBinomialInt(dim, PetscAbsInt(formDegree), &formSize));
503:   PetscCheck(Nc % formSize == 0, PetscObjectComm((PetscObject)q), PETSC_ERR_ARG_INCOMP, "Number of components %" PetscInt_FMT " is not a multiple of formSize %" PetscInt_FMT, Nc, formSize);
504:   Ncopies = Nc / formSize;
505:   PetscCall(PetscDTBinomialInt(imageDim, PetscAbsInt(formDegree), &imageFormSize));
506:   imageNc = Ncopies * imageFormSize;
507:   PetscCall(PetscMalloc1(Npoints * imageDim, &imagePoints));
508:   PetscCall(PetscMalloc1(Npoints * imageNc, &imageWeights));
509:   PetscCall(PetscMalloc2(imageDim * dim, &Jinv, formSize * imageFormSize, &Jinvstar));
510:   PetscCall(PetscDTJacobianInverse_Internal(imageDim, dim, J, Jinv));
511:   PetscCall(PetscDTAltVPullbackMatrix(imageDim, dim, Jinv, formDegree, Jinvstar));
512:   for (pt = 0; pt < Npoints; pt++) {
513:     const PetscReal *point      = PetscSafePointerPlusOffset(points, pt * dim);
514:     PetscReal       *imagePoint = &imagePoints[pt * imageDim];

516:     for (i = 0; i < imageDim; i++) {
517:       PetscReal val = originImage[i];

519:       for (j = 0; j < dim; j++) val += J[i * dim + j] * (point[j] - origin[j]);
520:       imagePoint[i] = val;
521:     }
522:     for (c = 0; c < Ncopies; c++) {
523:       const PetscReal *form      = &weights[pt * Nc + c * formSize];
524:       PetscReal       *imageForm = &imageWeights[pt * imageNc + c * imageFormSize];

526:       for (i = 0; i < imageFormSize; i++) {
527:         PetscReal val = 0.;

529:         for (j = 0; j < formSize; j++) val += Jinvstar[i * formSize + j] * form[j];
530:         imageForm[i] = val;
531:       }
532:     }
533:   }
534:   PetscCall(PetscQuadratureCreate(PetscObjectComm((PetscObject)q), Jinvstarq));
535:   PetscCall(PetscQuadratureSetData(*Jinvstarq, imageDim, imageNc, Npoints, imagePoints, imageWeights));
536:   PetscCall(PetscFree2(Jinv, Jinvstar));
537:   PetscFunctionReturn(PETSC_SUCCESS);
538: }

540: /*@C
541:   PetscQuadratureSetData - Sets the data defining the quadrature

543:   Not Collective

545:   Input Parameters:
546: + q       - The `PetscQuadrature` object
547: . dim     - The spatial dimension
548: . Nc      - The number of components
549: . npoints - The number of quadrature points
550: . points  - The coordinates of each quadrature point
551: - weights - The weight of each quadrature point

553:   Level: intermediate

555:   Note:
556:   `q` owns the references to points and weights, so they must be allocated using `PetscMalloc()` and the user should not free them.

558: .seealso: `PetscQuadrature`, `PetscQuadratureCreate()`, `PetscQuadratureGetData()`
559: @*/
560: PetscErrorCode PetscQuadratureSetData(PetscQuadrature q, PetscInt dim, PetscInt Nc, PetscInt npoints, const PetscReal points[], const PetscReal weights[]) PeNSS
561: {
562:   PetscFunctionBegin;
564:   if (dim >= 0) q->dim = dim;
565:   if (Nc >= 0) q->Nc = Nc;
566:   if (npoints >= 0) q->numPoints = npoints;
567:   if (points) {
568:     PetscAssertPointer(points, 5);
569:     q->points = points;
570:   }
571:   if (weights) {
572:     PetscAssertPointer(weights, 6);
573:     q->weights = weights;
574:   }
575:   PetscFunctionReturn(PETSC_SUCCESS);
576: }

578: static PetscErrorCode PetscQuadratureView_Ascii(PetscQuadrature quad, PetscViewer v)
579: {
580:   PetscInt          q, d, c;
581:   PetscViewerFormat format;

583:   PetscFunctionBegin;
584:   if (quad->Nc > 1)
585:     PetscCall(PetscViewerASCIIPrintf(v, "Quadrature on a %s of order %" PetscInt_FMT " on %" PetscInt_FMT " points (dim %" PetscInt_FMT ") with %" PetscInt_FMT " components\n", DMPolytopeTypes[quad->ct], quad->order, quad->numPoints, quad->dim, quad->Nc));
586:   else PetscCall(PetscViewerASCIIPrintf(v, "Quadrature on a %s of order %" PetscInt_FMT " on %" PetscInt_FMT " points (dim %" PetscInt_FMT ")\n", DMPolytopeTypes[quad->ct], quad->order, quad->numPoints, quad->dim));
587:   PetscCall(PetscViewerGetFormat(v, &format));
588:   if (format != PETSC_VIEWER_ASCII_INFO_DETAIL) PetscFunctionReturn(PETSC_SUCCESS);
589:   for (q = 0; q < quad->numPoints; ++q) {
590:     PetscCall(PetscViewerASCIIPrintf(v, "p%" PetscInt_FMT " (", q));
591:     PetscCall(PetscViewerASCIIUseTabs(v, PETSC_FALSE));
592:     for (d = 0; d < quad->dim; ++d) {
593:       if (d) PetscCall(PetscViewerASCIIPrintf(v, ", "));
594:       PetscCall(PetscViewerASCIIPrintf(v, "%+g", (double)quad->points[q * quad->dim + d]));
595:     }
596:     PetscCall(PetscViewerASCIIPrintf(v, ") "));
597:     if (quad->Nc > 1) PetscCall(PetscViewerASCIIPrintf(v, "w%" PetscInt_FMT " (", q));
598:     for (c = 0; c < quad->Nc; ++c) {
599:       if (c) PetscCall(PetscViewerASCIIPrintf(v, ", "));
600:       PetscCall(PetscViewerASCIIPrintf(v, "%+g", (double)quad->weights[q * quad->Nc + c]));
601:     }
602:     if (quad->Nc > 1) PetscCall(PetscViewerASCIIPrintf(v, ")"));
603:     PetscCall(PetscViewerASCIIPrintf(v, "\n"));
604:     PetscCall(PetscViewerASCIIUseTabs(v, PETSC_TRUE));
605:   }
606:   PetscFunctionReturn(PETSC_SUCCESS);
607: }

609: /*@
610:   PetscQuadratureView - View a `PetscQuadrature` object

612:   Collective

614:   Input Parameters:
615: + quad   - The `PetscQuadrature` object
616: - viewer - The `PetscViewer` object

618:   Level: beginner

620: .seealso: `PetscQuadrature`, `PetscViewer`, `PetscQuadratureCreate()`, `PetscQuadratureGetData()`
621: @*/
622: PetscErrorCode PetscQuadratureView(PetscQuadrature quad, PetscViewer viewer)
623: {
624:   PetscBool isascii;

626:   PetscFunctionBegin;
629:   if (!viewer) PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)quad), &viewer));
630:   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
631:   PetscCall(PetscViewerASCIIPushTab(viewer));
632:   if (isascii) PetscCall(PetscQuadratureView_Ascii(quad, viewer));
633:   PetscCall(PetscViewerASCIIPopTab(viewer));
634:   PetscFunctionReturn(PETSC_SUCCESS);
635: }

637: /*@C
638:   PetscQuadratureExpandComposite - Return a quadrature over the composite element, which has the original quadrature in each subelement

640:   Not Collective; No Fortran Support

642:   Input Parameters:
643: + q              - The original `PetscQuadrature`
644: . numSubelements - The number of subelements the original element is divided into
645: . v0             - An array of the initial points for each subelement
646: - jac            - An array of the Jacobian mappings from the reference to each subelement

648:   Output Parameter:
649: . qref - The dimension

651:   Level: intermediate

653:   Note:
654:   Together `v0` and `jac` define an affine mapping from the original reference element to each subelement

656: .seealso: `PetscQuadrature`, `PetscFECreate()`, `PetscSpaceGetDimension()`, `PetscDualSpaceGetDimension()`
657: @*/
658: PetscErrorCode PetscQuadratureExpandComposite(PetscQuadrature q, PetscInt numSubelements, const PetscReal v0[], const PetscReal jac[], PetscQuadrature *qref)
659: {
660:   DMPolytopeType   ct;
661:   const PetscReal *points, *weights;
662:   PetscReal       *pointsRef, *weightsRef;
663:   PetscInt         dim, Nc, order, npoints, npointsRef, c, p, cp, d, e;

665:   PetscFunctionBegin;
667:   PetscAssertPointer(v0, 3);
668:   PetscAssertPointer(jac, 4);
669:   PetscAssertPointer(qref, 5);
670:   PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, qref));
671:   PetscCall(PetscQuadratureGetCellType(q, &ct));
672:   PetscCall(PetscQuadratureGetOrder(q, &order));
673:   PetscCall(PetscQuadratureGetData(q, &dim, &Nc, &npoints, &points, &weights));
674:   npointsRef = npoints * numSubelements;
675:   PetscCall(PetscMalloc1(npointsRef * dim, &pointsRef));
676:   PetscCall(PetscMalloc1(npointsRef * Nc, &weightsRef));
677:   for (c = 0; c < numSubelements; ++c) {
678:     for (p = 0; p < npoints; ++p) {
679:       for (d = 0; d < dim; ++d) {
680:         pointsRef[(c * npoints + p) * dim + d] = v0[c * dim + d];
681:         for (e = 0; e < dim; ++e) pointsRef[(c * npoints + p) * dim + d] += jac[(c * dim + d) * dim + e] * (points[p * dim + e] + 1.0);
682:       }
683:       /* Could also use detJ here */
684:       for (cp = 0; cp < Nc; ++cp) weightsRef[(c * npoints + p) * Nc + cp] = weights[p * Nc + cp] / numSubelements;
685:     }
686:   }
687:   PetscCall(PetscQuadratureSetCellType(*qref, ct));
688:   PetscCall(PetscQuadratureSetOrder(*qref, order));
689:   PetscCall(PetscQuadratureSetData(*qref, dim, Nc, npointsRef, pointsRef, weightsRef));
690:   PetscFunctionReturn(PETSC_SUCCESS);
691: }

693: /* Compute the coefficients for the Jacobi polynomial recurrence,

695:    J^{a,b}_n(x) = (cnm1 + cnm1x * x) * J^{a,b}_{n-1}(x) - cnm2 * J^{a,b}_{n-2}(x).
696:  */
697: #define PetscDTJacobiRecurrence_Internal(n, a, b, cnm1, cnm1x, cnm2) \
698:   do { \
699:     PetscReal _a = (a); \
700:     PetscReal _b = (b); \
701:     PetscReal _n = (n); \
702:     if (n == 1) { \
703:       (cnm1)  = (_a - _b) * 0.5; \
704:       (cnm1x) = (_a + _b + 2.) * 0.5; \
705:       (cnm2)  = 0.; \
706:     } else { \
707:       PetscReal _2n  = _n + _n; \
708:       PetscReal _d   = (_2n * (_n + _a + _b) * (_2n + _a + _b - 2)); \
709:       PetscReal _n1  = (_2n + _a + _b - 1.) * (_a * _a - _b * _b); \
710:       PetscReal _n1x = (_2n + _a + _b - 1.) * (_2n + _a + _b) * (_2n + _a + _b - 2); \
711:       PetscReal _n2  = 2. * ((_n + _a - 1.) * (_n + _b - 1.) * (_2n + _a + _b)); \
712:       (cnm1)         = _n1 / _d; \
713:       (cnm1x)        = _n1x / _d; \
714:       (cnm2)         = _n2 / _d; \
715:     } \
716:   } while (0)

718: /*@
719:   PetscDTJacobiNorm - Compute the weighted L2 norm of a Jacobi polynomial.

721:   $\| P^{\alpha,\beta}_n \|_{\alpha,\beta}^2 = \int_{-1}^1 (1 + x)^{\alpha} (1 - x)^{\beta} P^{\alpha,\beta}_n (x)^2 dx.$

723:   Input Parameters:
724: + alpha - the left exponent > -1
725: . beta  - the right exponent > -1
726: - n     - the polynomial degree

728:   Output Parameter:
729: . norm - the weighted L2 norm

731:   Level: beginner

733: .seealso: `PetscQuadrature`, `PetscDTJacobiEval()`
734: @*/
735: PetscErrorCode PetscDTJacobiNorm(PetscReal alpha, PetscReal beta, PetscInt n, PetscReal *norm)
736: {
737:   PetscReal twoab1;
738:   PetscReal gr;

740:   PetscFunctionBegin;
741:   PetscCheck(alpha > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Exponent alpha %g <= -1. invalid", (double)alpha);
742:   PetscCheck(beta > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Exponent beta %g <= -1. invalid", (double)beta);
743:   PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "n %" PetscInt_FMT " < 0 invalid", n);
744:   twoab1 = PetscPowReal(2., alpha + beta + 1.);
745: #if defined(PETSC_HAVE_LGAMMA)
746:   if (!n) {
747:     gr = PetscExpReal(PetscLGamma(alpha + 1.) + PetscLGamma(beta + 1.) - PetscLGamma(alpha + beta + 2.));
748:   } else {
749:     gr = PetscExpReal(PetscLGamma(n + alpha + 1.) + PetscLGamma(n + beta + 1.) - (PetscLGamma(n + 1.) + PetscLGamma(n + alpha + beta + 1.))) / (n + n + alpha + beta + 1.);
750:   }
751: #else
752:   {
753:     PetscInt alphai = (PetscInt)alpha;
754:     PetscInt betai  = (PetscInt)beta;
755:     PetscInt i;

757:     gr = n ? (1. / (n + n + alpha + beta + 1.)) : 1.;
758:     if ((PetscReal)alphai == alpha) {
759:       if (!n) {
760:         for (i = 0; i < alphai; i++) gr *= (i + 1.) / (beta + i + 1.);
761:         gr /= (alpha + beta + 1.);
762:       } else {
763:         for (i = 0; i < alphai; i++) gr *= (n + i + 1.) / (n + beta + i + 1.);
764:       }
765:     } else if ((PetscReal)betai == beta) {
766:       if (!n) {
767:         for (i = 0; i < betai; i++) gr *= (i + 1.) / (alpha + i + 2.);
768:         gr /= (alpha + beta + 1.);
769:       } else {
770:         for (i = 0; i < betai; i++) gr *= (n + i + 1.) / (n + alpha + i + 1.);
771:       }
772:     } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "lgamma() - math routine is unavailable.");
773:   }
774: #endif
775:   *norm = PetscSqrtReal(twoab1 * gr);
776:   PetscFunctionReturn(PETSC_SUCCESS);
777: }

779: static PetscErrorCode PetscDTJacobiEval_Internal(PetscInt npoints, PetscReal a, PetscReal b, PetscInt k, const PetscReal *points, PetscInt ndegree, const PetscInt *degrees, PetscReal *p)
780: {
781:   PetscReal ak, bk;
782:   PetscReal abk1;
783:   PetscInt  i, l, maxdegree;

785:   PetscFunctionBegin;
786:   maxdegree = degrees[ndegree - 1] - k;
787:   ak        = a + k;
788:   bk        = b + k;
789:   abk1      = a + b + k + 1.;
790:   if (maxdegree < 0) {
791:     for (i = 0; i < npoints; i++)
792:       for (l = 0; l < ndegree; l++) p[i * ndegree + l] = 0.;
793:     PetscFunctionReturn(PETSC_SUCCESS);
794:   }
795:   for (i = 0; i < npoints; i++) {
796:     PetscReal pm1, pm2, x;
797:     PetscReal cnm1, cnm1x, cnm2;
798:     PetscInt  j;

800:     x   = points[i];
801:     pm2 = 1.;
802:     PetscDTJacobiRecurrence_Internal(1, ak, bk, cnm1, cnm1x, cnm2);
803:     pm1 = (cnm1 + cnm1x * x);
804:     l   = 0;
805:     while (l < ndegree && degrees[l] - k < 0) p[l++] = 0.;
806:     while (l < ndegree && degrees[l] - k == 0) {
807:       p[l] = pm2;
808:       for (PetscInt m = 0; m < k; m++) p[l] *= (abk1 + m) * 0.5;
809:       l++;
810:     }
811:     while (l < ndegree && degrees[l] - k == 1) {
812:       p[l] = pm1;
813:       for (PetscInt m = 0; m < k; m++) p[l] *= (abk1 + 1 + m) * 0.5;
814:       l++;
815:     }
816:     for (j = 2; j <= maxdegree; j++) {
817:       PetscReal pp;

819:       PetscDTJacobiRecurrence_Internal(j, ak, bk, cnm1, cnm1x, cnm2);
820:       pp  = (cnm1 + cnm1x * x) * pm1 - cnm2 * pm2;
821:       pm2 = pm1;
822:       pm1 = pp;
823:       while (l < ndegree && degrees[l] - k == j) {
824:         p[l] = pp;
825:         for (PetscInt m = 0; m < k; m++) p[l] *= (abk1 + j + m) * 0.5;
826:         l++;
827:       }
828:     }
829:     p += ndegree;
830:   }
831:   PetscFunctionReturn(PETSC_SUCCESS);
832: }

834: /*@
835:   PetscDTJacobiEvalJet - Evaluate the jet (function and derivatives) of the Jacobi polynomials basis up to a given degree.

837:   Input Parameters:
838: + alpha   - the left exponent of the weight
839: . beta    - the right exponetn of the weight
840: . npoints - the number of points to evaluate the polynomials at
841: . points  - [npoints] array of point coordinates
842: . degree  - the maximm degree polynomial space to evaluate, (degree + 1) will be evaluated total.
843: - k       - the maximum derivative to evaluate in the jet, (k + 1) will be evaluated total.

845:   Output Parameter:
846: . p - an array containing the evaluations of the Jacobi polynomials's jets on the points.  the size is (degree + 1) x
847:       (k + 1) x npoints, which also describes the order of the dimensions of this three-dimensional array: the first
848:       (slowest varying) dimension is polynomial degree; the second dimension is derivative order; the third (fastest
849:       varying) dimension is the index of the evaluation point.

851:   Level: advanced

853:   Notes:
854:   The Jacobi polynomials with indices $\alpha$ and $\beta$ are orthogonal with respect to the
855:   weighted inner product $\langle f, g \rangle = \int_{-1}^1 (1+x)^{\alpha} (1-x)^{\beta} f(x)
856:   g(x) dx$.

858: .seealso: `PetscDTJacobiEval()`, `PetscDTPKDEvalJet()`
859: @*/
860: PetscErrorCode PetscDTJacobiEvalJet(PetscReal alpha, PetscReal beta, PetscInt npoints, const PetscReal points[], PetscInt degree, PetscInt k, PetscReal p[])
861: {
862:   PetscInt   i, j, l;
863:   PetscInt  *degrees;
864:   PetscReal *psingle;

866:   PetscFunctionBegin;
867:   if (degree == 0) {
868:     PetscInt zero = 0;

870:     for (i = 0; i <= k; i++) PetscCall(PetscDTJacobiEval_Internal(npoints, alpha, beta, i, points, 1, &zero, &p[i * npoints]));
871:     PetscFunctionReturn(PETSC_SUCCESS);
872:   }
873:   PetscCall(PetscMalloc1(degree + 1, &degrees));
874:   PetscCall(PetscMalloc1((degree + 1) * npoints, &psingle));
875:   for (i = 0; i <= degree; i++) degrees[i] = i;
876:   for (i = 0; i <= k; i++) {
877:     PetscCall(PetscDTJacobiEval_Internal(npoints, alpha, beta, i, points, degree + 1, degrees, psingle));
878:     for (j = 0; j <= degree; j++) {
879:       for (l = 0; l < npoints; l++) p[(j * (k + 1) + i) * npoints + l] = psingle[l * (degree + 1) + j];
880:     }
881:   }
882:   PetscCall(PetscFree(psingle));
883:   PetscCall(PetscFree(degrees));
884:   PetscFunctionReturn(PETSC_SUCCESS);
885: }

887: /*@
888:   PetscDTJacobiEval - evaluate Jacobi polynomials for the weight function $(1.+x)^{\alpha} (1.-x)^{\beta}$ at a set of points
889:   at points

891:   Not Collective

893:   Input Parameters:
894: + npoints - number of spatial points to evaluate at
895: . alpha   - the left exponent > -1
896: . beta    - the right exponent > -1
897: . points  - array of locations to evaluate at
898: . ndegree - number of basis degrees to evaluate
899: - degrees - sorted array of degrees to evaluate

901:   Output Parameters:
902: + B  - row-oriented basis evaluation matrix B[point*ndegree + degree] (dimension npoints*ndegrees, allocated by caller) (or `NULL`)
903: . D  - row-oriented derivative evaluation matrix (or `NULL`)
904: - D2 - row-oriented second derivative evaluation matrix (or `NULL`)

906:   Level: intermediate

908: .seealso: `PetscDTGaussQuadrature()`, `PetscDTLegendreEval()`
909: @*/
910: PetscErrorCode PetscDTJacobiEval(PetscInt npoints, PetscReal alpha, PetscReal beta, const PetscReal *points, PetscInt ndegree, const PetscInt *degrees, PeOp PetscReal B[], PeOp PetscReal D[], PeOp PetscReal D2[])
911: {
912:   PetscFunctionBegin;
913:   PetscCheck(alpha > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "alpha must be > -1.");
914:   PetscCheck(beta > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "beta must be > -1.");
915:   if (!npoints || !ndegree) PetscFunctionReturn(PETSC_SUCCESS);
916:   if (B) PetscCall(PetscDTJacobiEval_Internal(npoints, alpha, beta, 0, points, ndegree, degrees, B));
917:   if (D) PetscCall(PetscDTJacobiEval_Internal(npoints, alpha, beta, 1, points, ndegree, degrees, D));
918:   if (D2) PetscCall(PetscDTJacobiEval_Internal(npoints, alpha, beta, 2, points, ndegree, degrees, D2));
919:   PetscFunctionReturn(PETSC_SUCCESS);
920: }

922: /*@
923:   PetscDTLegendreEval - evaluate Legendre polynomials at points

925:   Not Collective

927:   Input Parameters:
928: + npoints - number of spatial points to evaluate at
929: . points  - array of locations to evaluate at
930: . ndegree - number of basis degrees to evaluate
931: - degrees - sorted array of degrees to evaluate

933:   Output Parameters:
934: + B  - row-oriented basis evaluation matrix B[point*ndegree + degree] (dimension `npoints`*`ndegrees`, allocated by caller) (or `NULL`)
935: . D  - row-oriented derivative evaluation matrix (or `NULL`)
936: - D2 - row-oriented second derivative evaluation matrix (or `NULL`)

938:   Level: intermediate

940: .seealso: `PetscDTGaussQuadrature()`
941: @*/
942: PetscErrorCode PetscDTLegendreEval(PetscInt npoints, const PetscReal *points, PetscInt ndegree, const PetscInt *degrees, PeOp PetscReal B[], PeOp PetscReal D[], PeOp PetscReal D2[])
943: {
944:   PetscFunctionBegin;
945:   PetscCall(PetscDTJacobiEval(npoints, 0., 0., points, ndegree, degrees, B, D, D2));
946:   PetscFunctionReturn(PETSC_SUCCESS);
947: }

949: /*@
950:   PetscDTIndexToGradedOrder - convert an index into a tuple of monomial degrees in a graded order (that is, if the degree sum of tuple x is less than the degree sum of tuple y,
951:   then the index of x is smaller than the index of y)

953:   Input Parameters:
954: + len   - the desired length of the degree tuple
955: - index - the index to convert: should be >= 0

957:   Output Parameter:
958: . degtup - filled with a tuple of degrees

960:   Level: beginner

962:   Note:
963:   For two tuples x and y with the same degree sum, partial degree sums over the final elements of the tuples
964:   acts as a tiebreaker.  For example, (2, 1, 1) and (1, 2, 1) have the same degree sum, but the degree sum over the
965:   last two elements is smaller for the former, so (2, 1, 1) < (1, 2, 1).

967: .seealso: `PetscDTGradedOrderToIndex()`
968: @*/
969: PetscErrorCode PetscDTIndexToGradedOrder(PetscInt len, PetscInt index, PetscInt degtup[])
970: {
971:   PetscInt i, total;
972:   PetscInt sum;

974:   PetscFunctionBeginHot;
975:   PetscCheck(len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "length must be non-negative");
976:   PetscCheck(index >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "index must be non-negative");
977:   total = 1;
978:   sum   = 0;
979:   while (index >= total) {
980:     index -= total;
981:     total = (total * (len + sum)) / (sum + 1);
982:     sum++;
983:   }
984:   for (i = 0; i < len; i++) {
985:     PetscInt c;

987:     degtup[i] = sum;
988:     for (c = 0, total = 1; c < sum; c++) {
989:       /* going into the loop, total is the number of way to have a tuple of sum exactly c with length len - 1 - i */
990:       if (index < total) break;
991:       index -= total;
992:       total = (total * (len - 1 - i + c)) / (c + 1);
993:       degtup[i]--;
994:     }
995:     sum -= degtup[i];
996:   }
997:   PetscFunctionReturn(PETSC_SUCCESS);
998: }

1000: /*@
1001:   PetscDTGradedOrderToIndex - convert a tuple into an index in a graded order, the inverse of `PetscDTIndexToGradedOrder()`.

1003:   Input Parameters:
1004: + len    - the length of the degree tuple
1005: - degtup - tuple with this length

1007:   Output Parameter:
1008: . index - index in graded order: >= 0

1010:   Level: beginner

1012:   Note:
1013:   For two tuples x and y with the same degree sum, partial degree sums over the final elements of the tuples
1014:   acts as a tiebreaker.  For example, (2, 1, 1) and (1, 2, 1) have the same degree sum, but the degree sum over the
1015:   last two elements is smaller for the former, so (2, 1, 1) < (1, 2, 1).

1017: .seealso: `PetscDTIndexToGradedOrder()`
1018: @*/
1019: PetscErrorCode PetscDTGradedOrderToIndex(PetscInt len, const PetscInt degtup[], PetscInt *index)
1020: {
1021:   PetscInt i, idx, sum, total;

1023:   PetscFunctionBeginHot;
1024:   PetscCheck(len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "length must be non-negative");
1025:   for (i = 0, sum = 0; i < len; i++) sum += degtup[i];
1026:   idx   = 0;
1027:   total = 1;
1028:   for (i = 0; i < sum; i++) {
1029:     idx += total;
1030:     total = (total * (len + i)) / (i + 1);
1031:   }
1032:   for (i = 0; i < len - 1; i++) {
1033:     total = 1;
1034:     sum -= degtup[i];
1035:     for (PetscInt c = 0; c < sum; c++) {
1036:       idx += total;
1037:       total = (total * (len - 1 - i + c)) / (c + 1);
1038:     }
1039:   }
1040:   *index = idx;
1041:   PetscFunctionReturn(PETSC_SUCCESS);
1042: }

1044: static PetscBool PKDCite       = PETSC_FALSE;
1045: const char       PKDCitation[] = "@article{Kirby2010,\n"
1046:                                  "  title={Singularity-free evaluation of collapsed-coordinate orthogonal polynomials},\n"
1047:                                  "  author={Kirby, Robert C},\n"
1048:                                  "  journal={ACM Transactions on Mathematical Software (TOMS)},\n"
1049:                                  "  volume={37},\n"
1050:                                  "  number={1},\n"
1051:                                  "  pages={1--16},\n"
1052:                                  "  year={2010},\n"
1053:                                  "  publisher={ACM New York, NY, USA}\n}\n";

1055: /*@
1056:   PetscDTPKDEvalJet - Evaluate the jet (function and derivatives) of the Proriol-Koornwinder-Dubiner (PKD) basis for
1057:   the space of polynomials up to a given degree.

1059:   Input Parameters:
1060: + dim     - the number of variables in the multivariate polynomials
1061: . npoints - the number of points to evaluate the polynomials at
1062: . points  - [npoints x dim] array of point coordinates
1063: . degree  - the degree (sum of degrees on the variables in a monomial) of the polynomial space to evaluate.  There are ((dim + degree) choose dim) polynomials in this space.
1064: - k       - the maximum order partial derivative to evaluate in the jet.  There are (dim + k choose dim) partial derivatives
1065:             in the jet.  Choosing k = 0 means to evaluate just the function and no derivatives

1067:   Output Parameter:
1068: . p - an array containing the evaluations of the PKD polynomials' jets on the points.  The size is ((dim + degree)
1069:       choose dim) x ((dim + k) choose dim) x npoints, which also describes the order of the dimensions of this
1070:       three-dimensional array: the first (slowest varying) dimension is basis function index; the second dimension is jet
1071:       index; the third (fastest varying) dimension is the index of the evaluation point.

1073:   Level: advanced

1075:   Notes:
1076:   The PKD basis is L2-orthonormal on the biunit simplex (which is used as the reference element
1077:   for finite elements in PETSc), which makes it a stable basis to use for evaluating
1078:   polynomials in that domain.

1080:   The ordering of the basis functions, and the ordering of the derivatives in the jet, both follow the graded
1081:   ordering of `PetscDTIndexToGradedOrder()` and `PetscDTGradedOrderToIndex()`.  For example, in 3D, the polynomial with
1082:   leading monomial x^2,y^0,z^1, which has degree tuple (2,0,1), which by `PetscDTGradedOrderToIndex()` has index 12 (it is the 13th basis function in the space);
1083:   the partial derivative $\partial_x \partial_z$ has order tuple (1,0,1), appears at index 6 in the jet (it is the 7th partial derivative in the jet).

1085:   The implementation uses Kirby's singularity-free evaluation algorithm, <https://doi.org/10.1145/1644001.1644006>.

1087: .seealso: `PetscDTGradedOrderToIndex()`, `PetscDTIndexToGradedOrder()`, `PetscDTJacobiEvalJet()`
1088: @*/
1089: PetscErrorCode PetscDTPKDEvalJet(PetscInt dim, PetscInt npoints, const PetscReal points[], PetscInt degree, PetscInt k, PetscReal p[])
1090: {
1091:   PetscInt   degidx, kidx, d, pt;
1092:   PetscInt   Nk, Ndeg;
1093:   PetscInt  *ktup, *degtup;
1094:   PetscReal *scales, initscale, scaleexp;

1096:   PetscFunctionBegin;
1097:   PetscCall(PetscCitationsRegister(PKDCitation, &PKDCite));
1098:   PetscCall(PetscDTBinomialInt(dim + k, k, &Nk));
1099:   PetscCall(PetscDTBinomialInt(degree + dim, degree, &Ndeg));
1100:   PetscCall(PetscMalloc2(dim, &degtup, dim, &ktup));
1101:   PetscCall(PetscMalloc1(Ndeg, &scales));
1102:   initscale = 1.;
1103:   if (dim > 1) {
1104:     PetscCall(PetscDTBinomial(dim, 2, &scaleexp));
1105:     initscale = PetscPowReal(2., scaleexp * 0.5);
1106:   }
1107:   for (degidx = 0; degidx < Ndeg; degidx++) {
1108:     PetscInt  e;
1109:     PetscInt  m1idx = -1, m2idx = -1;
1110:     PetscInt  n;
1111:     PetscInt  degsum;
1112:     PetscReal alpha;
1113:     PetscReal cnm1, cnm1x, cnm2;
1114:     PetscReal norm;

1116:     PetscCall(PetscDTIndexToGradedOrder(dim, degidx, degtup));
1117:     for (d = dim - 1; d >= 0; d--)
1118:       if (degtup[d]) break;
1119:     if (d < 0) { /* constant is 1 everywhere, all derivatives are zero */
1120:       scales[degidx] = initscale;
1121:       for (e = 0; e < dim; e++) {
1122:         PetscCall(PetscDTJacobiNorm(e, 0., 0, &norm));
1123:         scales[degidx] /= norm;
1124:       }
1125:       for (PetscInt i = 0; i < npoints; i++) p[degidx * Nk * npoints + i] = 1.;
1126:       for (PetscInt i = 0; i < (Nk - 1) * npoints; i++) p[(degidx * Nk + 1) * npoints + i] = 0.;
1127:       continue;
1128:     }
1129:     n = degtup[d];
1130:     degtup[d]--;
1131:     PetscCall(PetscDTGradedOrderToIndex(dim, degtup, &m1idx));
1132:     if (degtup[d] > 0) {
1133:       degtup[d]--;
1134:       PetscCall(PetscDTGradedOrderToIndex(dim, degtup, &m2idx));
1135:       degtup[d]++;
1136:     }
1137:     degtup[d]++;
1138:     for (e = 0, degsum = 0; e < d; e++) degsum += degtup[e];
1139:     alpha = 2 * degsum + d;
1140:     PetscDTJacobiRecurrence_Internal(n, alpha, 0., cnm1, cnm1x, cnm2);

1142:     scales[degidx] = initscale;
1143:     for (e = 0, degsum = 0; e < dim; e++) {
1144:       PetscReal ealpha;
1145:       PetscReal enorm;

1147:       ealpha = 2 * degsum + e;
1148:       for (PetscInt f = 0; f < degsum; f++) scales[degidx] *= 2.;
1149:       PetscCall(PetscDTJacobiNorm(ealpha, 0., degtup[e], &enorm));
1150:       scales[degidx] /= enorm;
1151:       degsum += degtup[e];
1152:     }

1154:     for (pt = 0; pt < npoints; pt++) {
1155:       /* compute the multipliers */
1156:       PetscReal thetanm1, thetanm1x, thetanm2;

1158:       thetanm1x = dim - (d + 1) + 2. * points[pt * dim + d];
1159:       for (e = d + 1; e < dim; e++) thetanm1x += points[pt * dim + e];
1160:       thetanm1x *= 0.5;
1161:       thetanm1 = (2. - (dim - (d + 1)));
1162:       for (e = d + 1; e < dim; e++) thetanm1 -= points[pt * dim + e];
1163:       thetanm1 *= 0.5;
1164:       thetanm2 = thetanm1 * thetanm1;

1166:       for (kidx = 0; kidx < Nk; kidx++) {
1167:         PetscCall(PetscDTIndexToGradedOrder(dim, kidx, ktup));
1168:         /* first sum in the same derivative terms */
1169:         p[(degidx * Nk + kidx) * npoints + pt] = (cnm1 * thetanm1 + cnm1x * thetanm1x) * p[(m1idx * Nk + kidx) * npoints + pt];
1170:         if (m2idx >= 0) p[(degidx * Nk + kidx) * npoints + pt] -= cnm2 * thetanm2 * p[(m2idx * Nk + kidx) * npoints + pt];

1172:         for (PetscInt f = d; f < dim; f++) {
1173:           PetscInt km1idx, mplty = ktup[f];

1175:           if (!mplty) continue;
1176:           ktup[f]--;
1177:           PetscCall(PetscDTGradedOrderToIndex(dim, ktup, &km1idx));

1179:           /* the derivative of  cnm1x * thetanm1x  wrt x variable f is 0.5 * cnm1x if f > d otherwise it is cnm1x */
1180:           /* the derivative of  cnm1  * thetanm1   wrt x variable f is 0 if f == d, otherwise it is -0.5 * cnm1 */
1181:           /* the derivative of -cnm2  * thetanm2   wrt x variable f is 0 if f == d, otherwise it is cnm2 * thetanm1 */
1182:           if (f > d) {
1183:             p[(degidx * Nk + kidx) * npoints + pt] += mplty * 0.5 * (cnm1x - cnm1) * p[(m1idx * Nk + km1idx) * npoints + pt];
1184:             if (m2idx >= 0) {
1185:               p[(degidx * Nk + kidx) * npoints + pt] += mplty * cnm2 * thetanm1 * p[(m2idx * Nk + km1idx) * npoints + pt];
1186:               /* second derivatives of -cnm2  * thetanm2   wrt x variable f,f2 is like - 0.5 * cnm2 */
1187:               for (PetscInt f2 = f; f2 < dim; f2++) {
1188:                 PetscInt km2idx, mplty2 = ktup[f2];
1189:                 PetscInt factor;

1191:                 if (!mplty2) continue;
1192:                 ktup[f2]--;
1193:                 PetscCall(PetscDTGradedOrderToIndex(dim, ktup, &km2idx));

1195:                 factor = mplty * mplty2;
1196:                 if (f == f2) factor /= 2;
1197:                 p[(degidx * Nk + kidx) * npoints + pt] -= 0.5 * factor * cnm2 * p[(m2idx * Nk + km2idx) * npoints + pt];
1198:                 ktup[f2]++;
1199:               }
1200:             }
1201:           } else {
1202:             p[(degidx * Nk + kidx) * npoints + pt] += mplty * cnm1x * p[(m1idx * Nk + km1idx) * npoints + pt];
1203:           }
1204:           ktup[f]++;
1205:         }
1206:       }
1207:     }
1208:   }
1209:   for (degidx = 0; degidx < Ndeg; degidx++) {
1210:     PetscReal scale = scales[degidx];

1212:     for (PetscInt i = 0; i < Nk * npoints; i++) p[degidx * Nk * npoints + i] *= scale;
1213:   }
1214:   PetscCall(PetscFree(scales));
1215:   PetscCall(PetscFree2(degtup, ktup));
1216:   PetscFunctionReturn(PETSC_SUCCESS);
1217: }

1219: /*@
1220:   PetscDTPTrimmedSize - The size of the trimmed polynomial space of k-forms with a given degree and form degree,
1221:   which can be evaluated in `PetscDTPTrimmedEvalJet()`.

1223:   Input Parameters:
1224: + dim        - the number of variables in the multivariate polynomials
1225: . degree     - the degree (sum of degrees on the variables in a monomial) of the trimmed polynomial space.
1226: - formDegree - the degree of the form

1228:   Output Parameter:
1229: . size - The number ((`dim` + `degree`) choose (`dim` + `formDegree`)) x ((`degree` + `formDegree` - 1) choose (`formDegree`))

1231:   Level: advanced

1233: .seealso: `PetscDTPTrimmedEvalJet()`
1234: @*/
1235: PetscErrorCode PetscDTPTrimmedSize(PetscInt dim, PetscInt degree, PetscInt formDegree, PetscInt *size)
1236: {
1237:   PetscInt Nrk, Nbpt; // number of trimmed polynomials

1239:   PetscFunctionBegin;
1240:   formDegree = PetscAbsInt(formDegree);
1241:   PetscCall(PetscDTBinomialInt(degree + dim, degree + formDegree, &Nbpt));
1242:   PetscCall(PetscDTBinomialInt(degree + formDegree - 1, formDegree, &Nrk));
1243:   Nbpt *= Nrk;
1244:   *size = Nbpt;
1245:   PetscFunctionReturn(PETSC_SUCCESS);
1246: }

1248: /* there was a reference implementation based on section 4.4 of Arnold, Falk & Winther (acta numerica, 2006), but it
1249:  * was inferior to this implementation */
1250: static PetscErrorCode PetscDTPTrimmedEvalJet_Internal(PetscInt dim, PetscInt npoints, const PetscReal points[], PetscInt degree, PetscInt formDegree, PetscInt jetDegree, PetscReal p[])
1251: {
1252:   PetscInt  formDegreeOrig = formDegree;
1253:   PetscBool formNegative   = (formDegreeOrig < 0) ? PETSC_TRUE : PETSC_FALSE;

1255:   PetscFunctionBegin;
1256:   formDegree = PetscAbsInt(formDegreeOrig);
1257:   if (formDegree == 0) {
1258:     PetscCall(PetscDTPKDEvalJet(dim, npoints, points, degree, jetDegree, p));
1259:     PetscFunctionReturn(PETSC_SUCCESS);
1260:   }
1261:   if (formDegree == dim) {
1262:     PetscCall(PetscDTPKDEvalJet(dim, npoints, points, degree - 1, jetDegree, p));
1263:     PetscFunctionReturn(PETSC_SUCCESS);
1264:   }
1265:   PetscInt Nbpt;
1266:   PetscCall(PetscDTPTrimmedSize(dim, degree, formDegree, &Nbpt));
1267:   PetscInt Nf;
1268:   PetscCall(PetscDTBinomialInt(dim, formDegree, &Nf));
1269:   PetscInt Nk;
1270:   PetscCall(PetscDTBinomialInt(dim + jetDegree, dim, &Nk));
1271:   PetscCall(PetscArrayzero(p, Nbpt * Nf * Nk * npoints));

1273:   PetscInt Nbpm1; // number of scalar polynomials up to degree - 1;
1274:   PetscCall(PetscDTBinomialInt(dim + degree - 1, dim, &Nbpm1));
1275:   PetscReal *p_scalar;
1276:   PetscCall(PetscMalloc1(Nbpm1 * Nk * npoints, &p_scalar));
1277:   PetscCall(PetscDTPKDEvalJet(dim, npoints, points, degree - 1, jetDegree, p_scalar));
1278:   PetscInt total = 0;
1279:   // First add the full polynomials up to degree - 1 into the basis: take the scalar
1280:   // and copy one for each form component
1281:   for (PetscInt i = 0; i < Nbpm1; i++) {
1282:     const PetscReal *src = &p_scalar[i * Nk * npoints];
1283:     for (PetscInt f = 0; f < Nf; f++) {
1284:       PetscReal *dest = &p[(total++ * Nf + f) * Nk * npoints];
1285:       PetscCall(PetscArraycpy(dest, src, Nk * npoints));
1286:     }
1287:   }
1288:   PetscInt *form_atoms;
1289:   PetscCall(PetscMalloc1(formDegree + 1, &form_atoms));
1290:   // construct the interior product pattern
1291:   PetscInt (*pattern)[3];
1292:   PetscInt Nf1; // number of formDegree + 1 forms
1293:   PetscCall(PetscDTBinomialInt(dim, formDegree + 1, &Nf1));
1294:   PetscInt nnz = Nf1 * (formDegree + 1);
1295:   PetscCall(PetscMalloc1(Nf1 * (formDegree + 1), &pattern));
1296:   PetscCall(PetscDTAltVInteriorPattern(dim, formDegree + 1, pattern));
1297:   PetscReal centroid = (1. - dim) / (dim + 1.);
1298:   PetscInt *deriv;
1299:   PetscCall(PetscMalloc1(dim, &deriv));
1300:   for (PetscInt d = dim; d >= formDegree + 1; d--) {
1301:     PetscInt Nfd1; // number of formDegree + 1 forms in dimension d that include dx_0
1302:                    // (equal to the number of formDegree forms in dimension d-1)
1303:     PetscCall(PetscDTBinomialInt(d - 1, formDegree, &Nfd1));
1304:     // The number of homogeneous (degree-1) scalar polynomials in d variables
1305:     PetscInt Nh;
1306:     PetscCall(PetscDTBinomialInt(d - 1 + degree - 1, d - 1, &Nh));
1307:     const PetscReal *h_scalar = &p_scalar[(Nbpm1 - Nh) * Nk * npoints];
1308:     for (PetscInt b = 0; b < Nh; b++) {
1309:       const PetscReal *h_s = &h_scalar[b * Nk * npoints];
1310:       for (PetscInt f = 0; f < Nfd1; f++) {
1311:         // construct all formDegree+1 forms that start with dx_(dim - d) /\ ...
1312:         form_atoms[0] = dim - d;
1313:         PetscCall(PetscDTEnumSubset(d - 1, formDegree, f, &form_atoms[1]));
1314:         for (PetscInt i = 0; i < formDegree; i++) form_atoms[1 + i] += form_atoms[0] + 1;
1315:         PetscInt f_ind; // index of the resulting form
1316:         PetscCall(PetscDTSubsetIndex(dim, formDegree + 1, form_atoms, &f_ind));
1317:         PetscReal *p_f = &p[total++ * Nf * Nk * npoints];
1318:         for (PetscInt nz = 0; nz < nnz; nz++) {
1319:           PetscInt  i     = pattern[nz][0]; // formDegree component
1320:           PetscInt  j     = pattern[nz][1]; // (formDegree + 1) component
1321:           PetscInt  v     = pattern[nz][2]; // coordinate component
1322:           PetscReal scale = v < 0 ? -1. : 1.;

1324:           i     = formNegative ? (Nf - 1 - i) : i;
1325:           scale = (formNegative && (i & 1)) ? -scale : scale;
1326:           v     = v < 0 ? -(v + 1) : v;
1327:           if (j != f_ind) continue;
1328:           PetscReal *p_i = &p_f[i * Nk * npoints];
1329:           for (PetscInt jet = 0; jet < Nk; jet++) {
1330:             const PetscReal *h_jet = &h_s[jet * npoints];
1331:             PetscReal       *p_jet = &p_i[jet * npoints];

1333:             for (PetscInt pt = 0; pt < npoints; pt++) p_jet[pt] += scale * h_jet[pt] * (points[pt * dim + v] - centroid);
1334:             PetscCall(PetscDTIndexToGradedOrder(dim, jet, deriv));
1335:             deriv[v]++;
1336:             PetscReal mult = deriv[v];
1337:             PetscInt  l;
1338:             PetscCall(PetscDTGradedOrderToIndex(dim, deriv, &l));
1339:             if (l >= Nk) continue;
1340:             p_jet = &p_i[l * npoints];
1341:             for (PetscInt pt = 0; pt < npoints; pt++) p_jet[pt] += scale * mult * h_jet[pt];
1342:             deriv[v]--;
1343:           }
1344:         }
1345:       }
1346:     }
1347:   }
1348:   PetscCheck(total == Nbpt, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Incorrectly counted P trimmed polynomials");
1349:   PetscCall(PetscFree(deriv));
1350:   PetscCall(PetscFree(pattern));
1351:   PetscCall(PetscFree(form_atoms));
1352:   PetscCall(PetscFree(p_scalar));
1353:   PetscFunctionReturn(PETSC_SUCCESS);
1354: }

1356: /*@
1357:   PetscDTPTrimmedEvalJet - Evaluate the jet (function and derivatives) of a basis of the trimmed polynomial k-forms up to
1358:   a given degree.

1360:   Input Parameters:
1361: + dim        - the number of variables in the multivariate polynomials
1362: . npoints    - the number of points to evaluate the polynomials at
1363: . points     - [npoints x dim] array of point coordinates
1364: . degree     - the degree (sum of degrees on the variables in a monomial) of the trimmed polynomial space to evaluate.
1365:            There are ((dim + degree) choose (dim + formDegree)) x ((degree + formDegree - 1) choose (formDegree)) polynomials in this space.
1366:            (You can use `PetscDTPTrimmedSize()` to compute this size.)
1367: . formDegree - the degree of the form
1368: - jetDegree  - the maximum order partial derivative to evaluate in the jet.  There are ((dim + jetDegree) choose dim) partial derivatives
1369:               in the jet.  Choosing jetDegree = 0 means to evaluate just the function and no derivatives

1371:   Output Parameter:
1372: . p - an array containing the evaluations of the PKD polynomials' jets on the points.

1374:   Level: advanced

1376:   Notes:
1377:   The size of `p` is `PetscDTPTrimmedSize()` x ((dim + formDegree) choose dim) x ((dim + k)
1378:   choose dim) x npoints,which also describes the order of the dimensions of this
1379:   four-dimensional array\:

1381:   the first (slowest varying) dimension is basis function index;
1382:   the second dimension is component of the form;
1383:   the third dimension is jet index;
1384:   the fourth (fastest varying) dimension is the index of the evaluation point.

1386:   The ordering of the basis functions is not graded, so the basis functions are not nested by degree like `PetscDTPKDEvalJet()`.
1387:   The basis functions are not an L2-orthonormal basis on any particular domain.

1389:   The implementation is based on the description of the trimmed polynomials up to degree r as
1390:   the direct sum of polynomials up to degree (r-1) and the Koszul differential applied to
1391:   homogeneous polynomials of degree (r-1).

1393: .seealso: `PetscDTPKDEvalJet()`, `PetscDTPTrimmedSize()`
1394: @*/
1395: PetscErrorCode PetscDTPTrimmedEvalJet(PetscInt dim, PetscInt npoints, const PetscReal points[], PetscInt degree, PetscInt formDegree, PetscInt jetDegree, PetscReal p[])
1396: {
1397:   PetscFunctionBegin;
1398:   PetscCall(PetscDTPTrimmedEvalJet_Internal(dim, npoints, points, degree, formDegree, jetDegree, p));
1399:   PetscFunctionReturn(PETSC_SUCCESS);
1400: }

1402: /* solve the symmetric tridiagonal eigenvalue system, writing the eigenvalues into eigs and the eigenvectors into V
1403:  * with lds n; diag and subdiag are overwritten */
1404: static PetscErrorCode PetscDTSymmetricTridiagonalEigensolve(PetscInt n, PetscReal diag[], PetscReal subdiag[], PetscReal eigs[], PetscScalar V[])
1405: {
1406:   char          jobz   = 'V'; /* eigenvalues and eigenvectors */
1407:   char          range  = 'A'; /* all eigenvalues will be found */
1408:   PetscReal     VL     = 0.;  /* ignored because range is 'A' */
1409:   PetscReal     VU     = 0.;  /* ignored because range is 'A' */
1410:   PetscBLASInt  IL     = 0;   /* ignored because range is 'A' */
1411:   PetscBLASInt  IU     = 0;   /* ignored because range is 'A' */
1412:   PetscReal     abstol = 0.;  /* unused */
1413:   PetscBLASInt  bn, bm, ldz;  /* bm will equal bn on exit */
1414:   PetscBLASInt *isuppz;
1415:   PetscBLASInt  lwork, liwork;
1416:   PetscReal     workquery;
1417:   PetscBLASInt  iworkquery;
1418:   PetscBLASInt *iwork;
1419:   PetscReal    *work = NULL;

1421:   PetscFunctionBegin;
1422: #if !defined(PETSCDTGAUSSIANQUADRATURE_EIG)
1423:   SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP_SYS, "A LAPACK symmetric tridiagonal eigensolver could not be found");
1424: #endif
1425:   PetscCall(PetscBLASIntCast(n, &bn));
1426:   PetscCall(PetscBLASIntCast(n, &ldz));
1427: #if !defined(PETSC_MISSING_LAPACK_STEGR)
1428:   PetscCall(PetscMalloc1(2 * n, &isuppz));
1429:   lwork  = -1;
1430:   liwork = -1;
1431:   PetscCallLAPACKInfo("LAPACKstegr", LAPACKstegr_(&jobz, &range, &bn, diag, subdiag, &VL, &VU, &IL, &IU, &abstol, &bm, eigs, V, &ldz, isuppz, &workquery, &lwork, &iworkquery, &liwork, &info));
1432:   lwork  = (PetscBLASInt)workquery;
1433:   liwork = iworkquery;
1434:   PetscCall(PetscMalloc2(lwork, &work, liwork, &iwork));
1435:   PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF));
1436:   PetscCallLAPACKInfo("LAPACKstegr", LAPACKstegr_(&jobz, &range, &bn, diag, subdiag, &VL, &VU, &IL, &IU, &abstol, &bm, eigs, V, &ldz, isuppz, work, &lwork, iwork, &liwork, &info));
1437:   PetscCall(PetscFPTrapPop());
1438:   PetscCall(PetscFree2(work, iwork));
1439:   PetscCall(PetscFree(isuppz));
1440: #elif !defined(PETSC_MISSING_LAPACK_STEQR)
1441:   jobz = 'I'; /* Compute eigenvalues and eigenvectors of the
1442:                  tridiagonal matrix.  Z is initialized to the identity
1443:                  matrix. */
1444:   PetscCall(PetscMalloc1(PetscMax(1, 2 * n - 2), &work));
1445:   PetscCallLAPACKInfo("LAPACKsteqr", LAPACKsteqr_("I", &bn, diag, subdiag, V, &ldz, work, &info));
1446:   PetscCall(PetscFPTrapPop());
1447:   PetscCall(PetscFree(work));
1448:   PetscCall(PetscArraycpy(eigs, diag, n));
1449: #endif
1450:   PetscFunctionReturn(PETSC_SUCCESS);
1451: }

1453: /* Formula for the weights at the endpoints (-1 and 1) of Gauss-Lobatto-Jacobi
1454:  * quadrature rules on the interval [-1, 1] */
1455: static PetscErrorCode PetscDTGaussLobattoJacobiEndweights_Internal(PetscInt n, PetscReal alpha, PetscReal beta, PetscReal *leftw, PetscReal *rightw)
1456: {
1457:   PetscReal twoab1;
1458:   PetscInt  m = n - 2;
1459:   PetscReal a = alpha + 1.;
1460:   PetscReal b = beta + 1.;
1461:   PetscReal gra, grb;

1463:   PetscFunctionBegin;
1464:   twoab1 = PetscPowReal(2., a + b - 1.);
1465: #if defined(PETSC_HAVE_LGAMMA)
1466:   grb = PetscExpReal(2. * PetscLGamma(b + 1.) + PetscLGamma(m + 1.) + PetscLGamma(m + a + 1.) - (PetscLGamma(m + b + 1) + PetscLGamma(m + a + b + 1.)));
1467:   gra = PetscExpReal(2. * PetscLGamma(a + 1.) + PetscLGamma(m + 1.) + PetscLGamma(m + b + 1.) - (PetscLGamma(m + a + 1) + PetscLGamma(m + a + b + 1.)));
1468: #else
1469:   {
1470:     PetscReal binom1, binom2;
1471:     PetscInt  alphai = (PetscInt)alpha;
1472:     PetscInt  betai  = (PetscInt)beta;

1474:     PetscCheck((PetscReal)alphai == alpha && (PetscReal)betai == beta, PETSC_COMM_SELF, PETSC_ERR_SUP, "lgamma() - math routine is unavailable.");
1475:     PetscCall(PetscDTBinomial(m + b, b, &binom1));
1476:     PetscCall(PetscDTBinomial(m + a + b, b, &binom2));
1477:     grb = 1. / (binom1 * binom2);
1478:     PetscCall(PetscDTBinomial(m + a, a, &binom1));
1479:     PetscCall(PetscDTBinomial(m + a + b, a, &binom2));
1480:     gra = 1. / (binom1 * binom2);
1481:   }
1482: #endif
1483:   *leftw  = twoab1 * grb / b;
1484:   *rightw = twoab1 * gra / a;
1485:   PetscFunctionReturn(PETSC_SUCCESS);
1486: }

1488: /* Evaluates the nth jacobi polynomial with weight parameters a,b at a point x.
1489:    Recurrence relations implemented from the pseudocode given in Karniadakis and Sherwin, Appendix B */
1490: static inline PetscErrorCode PetscDTComputeJacobi(PetscReal a, PetscReal b, PetscInt n, PetscReal x, PetscReal *P)
1491: {
1492:   PetscReal pn1, pn2;
1493:   PetscReal cnm1, cnm1x, cnm2;

1495:   PetscFunctionBegin;
1496:   if (!n) {
1497:     *P = 1.0;
1498:     PetscFunctionReturn(PETSC_SUCCESS);
1499:   }
1500:   PetscDTJacobiRecurrence_Internal(1, a, b, cnm1, cnm1x, cnm2);
1501:   pn2 = 1.;
1502:   pn1 = cnm1 + cnm1x * x;
1503:   if (n == 1) {
1504:     *P = pn1;
1505:     PetscFunctionReturn(PETSC_SUCCESS);
1506:   }
1507:   *P = 0.0;
1508:   for (PetscInt k = 2; k < n + 1; ++k) {
1509:     PetscDTJacobiRecurrence_Internal(k, a, b, cnm1, cnm1x, cnm2);

1511:     *P  = (cnm1 + cnm1x * x) * pn1 - cnm2 * pn2;
1512:     pn2 = pn1;
1513:     pn1 = *P;
1514:   }
1515:   PetscFunctionReturn(PETSC_SUCCESS);
1516: }

1518: /* Evaluates the first derivative of P_{n}^{a,b} at a point x. */
1519: static inline PetscErrorCode PetscDTComputeJacobiDerivative(PetscReal a, PetscReal b, PetscInt n, PetscReal x, PetscInt k, PetscReal *P)
1520: {
1521:   PetscReal nP;
1522:   PetscInt  i;

1524:   PetscFunctionBegin;
1525:   *P = 0.0;
1526:   if (k > n) PetscFunctionReturn(PETSC_SUCCESS);
1527:   PetscCall(PetscDTComputeJacobi(a + k, b + k, n - k, x, &nP));
1528:   for (i = 0; i < k; i++) nP *= (a + b + n + 1. + i) * 0.5;
1529:   *P = nP;
1530:   PetscFunctionReturn(PETSC_SUCCESS);
1531: }

1533: static PetscErrorCode PetscDTGaussJacobiQuadrature_Newton_Internal(PetscInt npoints, PetscReal a, PetscReal b, PetscReal x[], PetscReal w[])
1534: {
1535:   PetscInt  maxIter = 100;
1536:   PetscReal eps     = PetscExpReal(0.75 * PetscLogReal(PETSC_MACHINE_EPSILON));
1537:   PetscReal a1, a6, gf;

1539:   PetscFunctionBegin;
1540:   a1 = PetscPowReal(2.0, a + b + 1);
1541: #if defined(PETSC_HAVE_LGAMMA)
1542:   {
1543:     PetscReal a2, a3, a4, a5;
1544:     a2 = PetscLGamma(a + npoints + 1);
1545:     a3 = PetscLGamma(b + npoints + 1);
1546:     a4 = PetscLGamma(a + b + npoints + 1);
1547:     a5 = PetscLGamma(npoints + 1);
1548:     gf = PetscExpReal(a2 + a3 - (a4 + a5));
1549:   }
1550: #else
1551:   {
1552:     PetscInt ia, ib;

1554:     ia = (PetscInt)a;
1555:     ib = (PetscInt)b;
1556:     gf = 1.;
1557:     if (ia == a && ia >= 0) { /* compute ratio of rising factorals wrt a */
1558:       for (PetscInt k = 0; k < ia; k++) gf *= (npoints + 1. + k) / (npoints + b + 1. + k);
1559:     } else if (b == b && ib >= 0) { /* compute ratio of rising factorials wrt b */
1560:       for (PetscInt k = 0; k < ib; k++) gf *= (npoints + 1. + k) / (npoints + a + 1. + k);
1561:     } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "lgamma() - math routine is unavailable.");
1562:   }
1563: #endif

1565:   a6 = a1 * gf;
1566:   /* Computes the m roots of P_{m}^{a,b} on [-1,1] by Newton's method with Chebyshev points as initial guesses.
1567:    Algorithm implemented from the pseudocode given by Karniadakis and Sherwin and Python in FIAT */
1568:   for (PetscInt k = 0; k < npoints; ++k) {
1569:     PetscReal r = PetscCosReal(PETSC_PI * (1. - (4. * k + 3. + 2. * b) / (4. * npoints + 2. * (a + b + 1.)))), dP;

1571:     if (k > 0) r = 0.5 * (r + x[k - 1]);
1572:     for (PetscInt j = 0; j < maxIter; ++j) {
1573:       PetscReal s = 0.0, delta, f, fp;
1574:       PetscInt  i;

1576:       for (i = 0; i < k; ++i) s = s + 1.0 / (r - x[i]);
1577:       PetscCall(PetscDTComputeJacobi(a, b, npoints, r, &f));
1578:       PetscCall(PetscDTComputeJacobiDerivative(a, b, npoints, r, 1, &fp));
1579:       delta = f / (fp - f * s);
1580:       r     = r - delta;
1581:       if (PetscAbsReal(delta) < eps) break;
1582:     }
1583:     x[k] = r;
1584:     PetscCall(PetscDTComputeJacobiDerivative(a, b, npoints, x[k], 1, &dP));
1585:     w[k] = a6 / (1.0 - PetscSqr(x[k])) / PetscSqr(dP);
1586:   }
1587:   PetscFunctionReturn(PETSC_SUCCESS);
1588: }

1590: /* Compute the diagonals of the Jacobi matrix used in Golub & Welsch algorithms for Gauss-Jacobi
1591:  * quadrature weight calculations on [-1,1] for exponents (1. + x)^a (1.-x)^b */
1592: static PetscErrorCode PetscDTJacobiMatrix_Internal(PetscInt nPoints, PetscReal a, PetscReal b, PetscReal *d, PetscReal *s)
1593: {
1594:   PetscInt i;

1596:   PetscFunctionBegin;
1597:   for (i = 0; i < nPoints; i++) {
1598:     PetscReal A, B, C;

1600:     PetscDTJacobiRecurrence_Internal(i + 1, a, b, A, B, C);
1601:     d[i] = -A / B;
1602:     if (i) s[i - 1] *= C / B;
1603:     if (i < nPoints - 1) s[i] = 1. / B;
1604:   }
1605:   PetscFunctionReturn(PETSC_SUCCESS);
1606: }

1608: static PetscErrorCode PetscDTGaussJacobiQuadrature_GolubWelsch_Internal(PetscInt npoints, PetscReal a, PetscReal b, PetscReal *x, PetscReal *w)
1609: {
1610:   PetscReal mu0;
1611:   PetscReal ga, gb, gab;
1612:   PetscInt  i;

1614:   PetscFunctionBegin;
1615:   PetscCall(PetscCitationsRegister(GolubWelschCitation, &GolubWelschCite));

1617: #if defined(PETSC_HAVE_TGAMMA)
1618:   ga  = PetscTGamma(a + 1);
1619:   gb  = PetscTGamma(b + 1);
1620:   gab = PetscTGamma(a + b + 2);
1621: #else
1622:   {
1623:     PetscInt ia = (PetscInt)a, ib = (PetscInt)b;

1625:     PetscCheck(ia == a && ib == b && ia + 1 > 0 && ib + 1 > 0 && ia + ib + 2 > 0, PETSC_COMM_SELF, PETSC_ERR_SUP, "tgamma() - math routine is unavailable.");
1626:     /* All gamma(x) terms are (x-1)! terms */
1627:     PetscCall(PetscDTFactorial(ia, &ga));
1628:     PetscCall(PetscDTFactorial(ib, &gb));
1629:     PetscCall(PetscDTFactorial(ia + ib + 1, &gab));
1630:   }
1631: #endif
1632:   mu0 = PetscPowReal(2., a + b + 1.) * ga * gb / gab;

1634: #if defined(PETSCDTGAUSSIANQUADRATURE_EIG)
1635:   {
1636:     PetscReal   *diag, *subdiag;
1637:     PetscScalar *V;

1639:     PetscCall(PetscMalloc2(npoints, &diag, npoints, &subdiag));
1640:     PetscCall(PetscMalloc1(npoints * npoints, &V));
1641:     PetscCall(PetscDTJacobiMatrix_Internal(npoints, a, b, diag, subdiag));
1642:     for (i = 0; i < npoints - 1; i++) subdiag[i] = PetscSqrtReal(subdiag[i]);
1643:     PetscCall(PetscDTSymmetricTridiagonalEigensolve(npoints, diag, subdiag, x, V));
1644:     for (i = 0; i < npoints; i++) w[i] = PetscSqr(PetscRealPart(V[i * npoints])) * mu0;
1645:     PetscCall(PetscFree(V));
1646:     PetscCall(PetscFree2(diag, subdiag));
1647:   }
1648: #else
1649:   SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP_SYS, "A LAPACK symmetric tridiagonal eigensolver could not be found");
1650: #endif
1651:   { /* As of March 2, 2020, The Sun Performance Library breaks the LAPACK contract for xstegr and xsteqr: the
1652:        eigenvalues are not guaranteed to be in ascending order.  So we heave a passive aggressive sigh and check that
1653:        the eigenvalues are sorted */
1654:     PetscBool sorted;

1656:     PetscCall(PetscSortedReal(npoints, x, &sorted));
1657:     if (!sorted) {
1658:       PetscInt  *order, i;
1659:       PetscReal *tmp;

1661:       PetscCall(PetscMalloc2(npoints, &order, npoints, &tmp));
1662:       for (i = 0; i < npoints; i++) order[i] = i;
1663:       PetscCall(PetscSortRealWithPermutation(npoints, x, order));
1664:       PetscCall(PetscArraycpy(tmp, x, npoints));
1665:       for (i = 0; i < npoints; i++) x[i] = tmp[order[i]];
1666:       PetscCall(PetscArraycpy(tmp, w, npoints));
1667:       for (i = 0; i < npoints; i++) w[i] = tmp[order[i]];
1668:       PetscCall(PetscFree2(order, tmp));
1669:     }
1670:   }
1671:   PetscFunctionReturn(PETSC_SUCCESS);
1672: }

1674: static PetscErrorCode PetscDTGaussJacobiQuadrature_Internal(PetscInt npoints, PetscReal alpha, PetscReal beta, PetscReal x[], PetscReal w[], PetscBool newton)
1675: {
1676:   PetscFunctionBegin;
1677:   PetscCheck(npoints >= 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of points must be positive");
1678:   /* If asking for a 1D Lobatto point, just return the non-Lobatto 1D point */
1679:   PetscCheck(alpha > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "alpha must be > -1.");
1680:   PetscCheck(beta > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "beta must be > -1.");

1682:   if (newton) PetscCall(PetscDTGaussJacobiQuadrature_Newton_Internal(npoints, alpha, beta, x, w));
1683:   else PetscCall(PetscDTGaussJacobiQuadrature_GolubWelsch_Internal(npoints, alpha, beta, x, w));
1684:   if (alpha == beta) { /* symmetrize */
1685:     PetscInt i;
1686:     for (i = 0; i < (npoints + 1) / 2; i++) {
1687:       PetscInt  j  = npoints - 1 - i;
1688:       PetscReal xi = x[i];
1689:       PetscReal xj = x[j];
1690:       PetscReal wi = w[i];
1691:       PetscReal wj = w[j];

1693:       x[i] = (xi - xj) / 2.;
1694:       x[j] = (xj - xi) / 2.;
1695:       w[i] = w[j] = (wi + wj) / 2.;
1696:     }
1697:   }
1698:   PetscFunctionReturn(PETSC_SUCCESS);
1699: }

1701: /*@
1702:   PetscDTGaussJacobiQuadrature - quadrature for the interval $[a, b]$ with the weight function
1703:   $(x-a)^\alpha (x-b)^\beta$.

1705:   Not Collective

1707:   Input Parameters:
1708: + npoints - the number of points in the quadrature rule
1709: . a       - the left endpoint of the interval
1710: . b       - the right endpoint of the interval
1711: . alpha   - the left exponent
1712: - beta    - the right exponent

1714:   Output Parameters:
1715: + x - array of length `npoints`, the locations of the quadrature points
1716: - w - array of length `npoints`, the weights of the quadrature points

1718:   Level: intermediate

1720:   Note:
1721:   This quadrature rule is exact for polynomials up to degree 2*`npoints` - 1.

1723: .seealso: `PetscDTGaussQuadrature()`
1724: @*/
1725: PetscErrorCode PetscDTGaussJacobiQuadrature(PetscInt npoints, PetscReal a, PetscReal b, PetscReal alpha, PetscReal beta, PetscReal x[], PetscReal w[])
1726: {
1727:   PetscInt i;

1729:   PetscFunctionBegin;
1730:   PetscCall(PetscDTGaussJacobiQuadrature_Internal(npoints, alpha, beta, x, w, PetscDTGaussQuadratureNewton_Internal));
1731:   if (a != -1. || b != 1.) { /* shift */
1732:     for (i = 0; i < npoints; i++) {
1733:       x[i] = (x[i] + 1.) * ((b - a) / 2.) + a;
1734:       w[i] *= (b - a) / 2.;
1735:     }
1736:   }
1737:   PetscFunctionReturn(PETSC_SUCCESS);
1738: }

1740: static PetscErrorCode PetscDTGaussLobattoJacobiQuadrature_Internal(PetscInt npoints, PetscReal alpha, PetscReal beta, PetscReal x[], PetscReal w[], PetscBool newton)
1741: {
1742:   PetscFunctionBegin;
1743:   PetscCheck(npoints >= 2, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of points must be positive");
1744:   /* If asking for a 1D Lobatto point, just return the non-Lobatto 1D point */
1745:   PetscCheck(alpha > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "alpha must be > -1.");
1746:   PetscCheck(beta > -1., PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "beta must be > -1.");

1748:   x[0]           = -1.;
1749:   x[npoints - 1] = 1.;
1750:   if (npoints > 2) PetscCall(PetscDTGaussJacobiQuadrature_Internal(npoints - 2, alpha + 1., beta + 1., &x[1], &w[1], newton));
1751:   for (PetscInt i = 1; i < npoints - 1; i++) w[i] /= (1. - x[i] * x[i]);
1752:   PetscCall(PetscDTGaussLobattoJacobiEndweights_Internal(npoints, alpha, beta, &w[0], &w[npoints - 1]));
1753:   PetscFunctionReturn(PETSC_SUCCESS);
1754: }

1756: /*@
1757:   PetscDTGaussLobattoJacobiQuadrature - quadrature for the interval $[a, b]$ with the weight function
1758:   $(x-a)^\alpha (x-b)^\beta$, with endpoints `a` and `b` included as quadrature points.

1760:   Not Collective

1762:   Input Parameters:
1763: + npoints - the number of points in the quadrature rule
1764: . a       - the left endpoint of the interval
1765: . b       - the right endpoint of the interval
1766: . alpha   - the left exponent
1767: - beta    - the right exponent

1769:   Output Parameters:
1770: + x - array of length `npoints`, the locations of the quadrature points
1771: - w - array of length `npoints`, the weights of the quadrature points

1773:   Level: intermediate

1775:   Note:
1776:   This quadrature rule is exact for polynomials up to degree 2*`npoints` - 3.

1778: .seealso: `PetscDTGaussJacobiQuadrature()`
1779: @*/
1780: PetscErrorCode PetscDTGaussLobattoJacobiQuadrature(PetscInt npoints, PetscReal a, PetscReal b, PetscReal alpha, PetscReal beta, PetscReal x[], PetscReal w[])
1781: {
1782:   PetscFunctionBegin;
1783:   PetscCall(PetscDTGaussLobattoJacobiQuadrature_Internal(npoints, alpha, beta, x, w, PetscDTGaussQuadratureNewton_Internal));
1784:   if (a != -1. || b != 1.) { /* shift */
1785:     for (PetscInt i = 0; i < npoints; i++) {
1786:       x[i] = (x[i] + 1.) * ((b - a) / 2.) + a;
1787:       w[i] *= (b - a) / 2.;
1788:     }
1789:   }
1790:   PetscFunctionReturn(PETSC_SUCCESS);
1791: }

1793: /*@
1794:   PetscDTGaussQuadrature - create Gauss-Legendre quadrature

1796:   Not Collective

1798:   Input Parameters:
1799: + npoints - number of points
1800: . a       - left end of interval (often-1)
1801: - b       - right end of interval (often +1)

1803:   Output Parameters:
1804: + x - quadrature points
1805: - w - quadrature weights

1807:   Level: intermediate

1809:   Note:
1810:   See {cite}`golub1969calculation`

1812: .seealso: `PetscDTLegendreEval()`, `PetscDTGaussJacobiQuadrature()`
1813: @*/
1814: PetscErrorCode PetscDTGaussQuadrature(PetscInt npoints, PetscReal a, PetscReal b, PetscReal x[], PetscReal w[])
1815: {
1816:   PetscFunctionBegin;
1817:   PetscCall(PetscDTGaussJacobiQuadrature_Internal(npoints, 0., 0., x, w, PetscDTGaussQuadratureNewton_Internal));
1818:   if (a != -1. || b != 1.) { /* shift */
1819:     for (PetscInt i = 0; i < npoints; i++) {
1820:       x[i] = (x[i] + 1.) * ((b - a) / 2.) + a;
1821:       w[i] *= (b - a) / 2.;
1822:     }
1823:   }
1824:   PetscFunctionReturn(PETSC_SUCCESS);
1825: }

1827: /*@
1828:   PetscDTGaussLobattoLegendreQuadrature - creates a set of the locations and weights of the Gauss-Lobatto-Legendre
1829:   nodes of a given size on the domain $[-1,1]$

1831:   Not Collective

1833:   Input Parameters:
1834: + npoints - number of grid nodes
1835: - type    - `PETSCGAUSSLOBATTOLEGENDRE_VIA_LINEAR_ALGEBRA` or `PETSCGAUSSLOBATTOLEGENDRE_VIA_NEWTON`

1837:   Output Parameters:
1838: + x - quadrature points, pass in an array of length `npoints`
1839: - w - quadrature weights, pass in an array of length `npoints`

1841:   Level: intermediate

1843:   Notes:
1844:   For n > 30  the Newton approach computes duplicate (incorrect) values for some nodes because the initial guess is apparently not
1845:   close enough to the desired solution

1847:   These are useful for implementing spectral methods based on Gauss-Lobatto-Legendre (GLL) nodes

1849:   See <https://epubs.siam.org/doi/abs/10.1137/110855442>  <https://epubs.siam.org/doi/abs/10.1137/120889873> for better ways to compute GLL nodes

1851: .seealso: `PetscDTGaussQuadrature()`, `PetscGaussLobattoLegendreCreateType`
1852: @*/
1853: PetscErrorCode PetscDTGaussLobattoLegendreQuadrature(PetscInt npoints, PetscGaussLobattoLegendreCreateType type, PetscReal x[], PetscReal w[])
1854: {
1855:   PetscBool newton;

1857:   PetscFunctionBegin;
1858:   PetscCheck(npoints >= 2, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must provide at least 2 grid points per element");
1859:   newton = (PetscBool)(type == PETSCGAUSSLOBATTOLEGENDRE_VIA_NEWTON);
1860:   PetscCall(PetscDTGaussLobattoJacobiQuadrature_Internal(npoints, 0., 0., x, w, newton));
1861:   PetscFunctionReturn(PETSC_SUCCESS);
1862: }

1864: /*@
1865:   PetscDTGaussTensorQuadrature - creates a tensor-product Gauss quadrature

1867:   Not Collective

1869:   Input Parameters:
1870: + dim     - The spatial dimension
1871: . Nc      - The number of components
1872: . npoints - number of points in one dimension
1873: . a       - left end of interval (often-1)
1874: - b       - right end of interval (often +1)

1876:   Output Parameter:
1877: . q - A `PetscQuadrature` object

1879:   Level: intermediate

1881: .seealso: `PetscDTGaussQuadrature()`, `PetscDTLegendreEval()`
1882: @*/
1883: PetscErrorCode PetscDTGaussTensorQuadrature(PetscInt dim, PetscInt Nc, PetscInt npoints, PetscReal a, PetscReal b, PetscQuadrature *q)
1884: {
1885:   DMPolytopeType ct;
1886:   PetscInt       totpoints = dim > 1 ? dim > 2 ? npoints * PetscSqr(npoints) : PetscSqr(npoints) : npoints;
1887:   PetscReal     *x, *w, *xw, *ww;

1889:   PetscFunctionBegin;
1890:   PetscCall(PetscMalloc1(totpoints * dim, &x));
1891:   PetscCall(PetscMalloc1(totpoints * Nc, &w));
1892:   /* Set up the Golub-Welsch system */
1893:   switch (dim) {
1894:   case 0:
1895:     ct = DM_POLYTOPE_POINT;
1896:     PetscCall(PetscFree(x));
1897:     PetscCall(PetscFree(w));
1898:     PetscCall(PetscMalloc1(1, &x));
1899:     PetscCall(PetscMalloc1(Nc, &w));
1900:     totpoints = 1;
1901:     x[0]      = 0.0;
1902:     for (PetscInt c = 0; c < Nc; ++c) w[c] = 1.0;
1903:     break;
1904:   case 1:
1905:     ct = DM_POLYTOPE_SEGMENT;
1906:     PetscCall(PetscMalloc1(npoints, &ww));
1907:     PetscCall(PetscDTGaussQuadrature(npoints, a, b, x, ww));
1908:     for (PetscInt i = 0; i < npoints; ++i)
1909:       for (PetscInt c = 0; c < Nc; ++c) w[i * Nc + c] = ww[i];
1910:     PetscCall(PetscFree(ww));
1911:     break;
1912:   case 2:
1913:     ct = DM_POLYTOPE_QUADRILATERAL;
1914:     PetscCall(PetscMalloc2(npoints, &xw, npoints, &ww));
1915:     PetscCall(PetscDTGaussQuadrature(npoints, a, b, xw, ww));
1916:     for (PetscInt i = 0; i < npoints; ++i) {
1917:       for (PetscInt j = 0; j < npoints; ++j) {
1918:         x[(i * npoints + j) * dim + 0] = xw[i];
1919:         x[(i * npoints + j) * dim + 1] = xw[j];
1920:         for (PetscInt c = 0; c < Nc; ++c) w[(i * npoints + j) * Nc + c] = ww[i] * ww[j];
1921:       }
1922:     }
1923:     PetscCall(PetscFree2(xw, ww));
1924:     break;
1925:   case 3:
1926:     ct = DM_POLYTOPE_HEXAHEDRON;
1927:     PetscCall(PetscMalloc2(npoints, &xw, npoints, &ww));
1928:     PetscCall(PetscDTGaussQuadrature(npoints, a, b, xw, ww));
1929:     for (PetscInt i = 0; i < npoints; ++i) {
1930:       for (PetscInt j = 0; j < npoints; ++j) {
1931:         for (PetscInt k = 0; k < npoints; ++k) {
1932:           x[((i * npoints + j) * npoints + k) * dim + 0] = xw[i];
1933:           x[((i * npoints + j) * npoints + k) * dim + 1] = xw[j];
1934:           x[((i * npoints + j) * npoints + k) * dim + 2] = xw[k];
1935:           for (PetscInt c = 0; c < Nc; ++c) w[((i * npoints + j) * npoints + k) * Nc + c] = ww[i] * ww[j] * ww[k];
1936:         }
1937:       }
1938:     }
1939:     PetscCall(PetscFree2(xw, ww));
1940:     break;
1941:   default:
1942:     SETERRQ(PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Cannot construct quadrature rule for dimension %" PetscInt_FMT, dim);
1943:   }
1944:   PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, q));
1945:   PetscCall(PetscQuadratureSetCellType(*q, ct));
1946:   PetscCall(PetscQuadratureSetOrder(*q, 2 * npoints - 1));
1947:   PetscCall(PetscQuadratureSetData(*q, dim, Nc, totpoints, x, w));
1948:   PetscCall(PetscObjectChangeTypeName((PetscObject)*q, "GaussTensor"));
1949:   PetscFunctionReturn(PETSC_SUCCESS);
1950: }

1952: /*@
1953:   PetscDTStroudConicalQuadrature - create Stroud conical quadrature for a simplex {cite}`karniadakis2005spectral`

1955:   Not Collective

1957:   Input Parameters:
1958: + dim     - The simplex dimension
1959: . Nc      - The number of components
1960: . npoints - The number of points in one dimension
1961: . a       - left end of interval (often-1)
1962: - b       - right end of interval (often +1)

1964:   Output Parameter:
1965: . q - A `PetscQuadrature` object

1967:   Level: intermediate

1969:   Note:
1970:   For `dim` == 1, this is Gauss-Legendre quadrature

1972: .seealso: `PetscDTGaussTensorQuadrature()`, `PetscDTGaussQuadrature()`
1973: @*/
1974: PetscErrorCode PetscDTStroudConicalQuadrature(PetscInt dim, PetscInt Nc, PetscInt npoints, PetscReal a, PetscReal b, PetscQuadrature *q)
1975: {
1976:   DMPolytopeType ct;
1977:   PetscInt       totpoints;
1978:   PetscReal     *p1, *w1;
1979:   PetscReal     *x, *w;

1981:   PetscFunctionBegin;
1982:   PetscCheck(!(a != -1.0) && !(b != 1.0), PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must use default internal right now");
1983:   switch (dim) {
1984:   case 0:
1985:     ct = DM_POLYTOPE_POINT;
1986:     break;
1987:   case 1:
1988:     ct = DM_POLYTOPE_SEGMENT;
1989:     break;
1990:   case 2:
1991:     ct = DM_POLYTOPE_TRIANGLE;
1992:     break;
1993:   case 3:
1994:     ct = DM_POLYTOPE_TETRAHEDRON;
1995:     break;
1996:   default:
1997:     ct = DM_POLYTOPE_UNKNOWN;
1998:   }
1999:   totpoints = 1;
2000:   for (PetscInt i = 0; i < dim; ++i) totpoints *= npoints;
2001:   PetscCall(PetscMalloc1(totpoints * dim, &x));
2002:   PetscCall(PetscMalloc1(totpoints * Nc, &w));
2003:   PetscCall(PetscMalloc2(npoints, &p1, npoints, &w1));
2004:   for (PetscInt i = 0; i < totpoints * Nc; ++i) w[i] = 1.;
2005:   for (PetscInt i = 0, totprev = 1, totrem = totpoints / npoints; i < dim; ++i) {
2006:     PetscReal mul;

2008:     mul = PetscPowReal(2., -i);
2009:     PetscCall(PetscDTGaussJacobiQuadrature(npoints, -1., 1., i, 0.0, p1, w1));
2010:     for (PetscInt pt = 0, l = 0; l < totprev; l++) {
2011:       for (PetscInt j = 0; j < npoints; j++) {
2012:         for (PetscInt m = 0; m < totrem; m++, pt++) {
2013:           for (PetscInt k = 0; k < i; k++) x[pt * dim + k] = (x[pt * dim + k] + 1.) * (1. - p1[j]) * 0.5 - 1.;
2014:           x[pt * dim + i] = p1[j];
2015:           for (PetscInt c = 0; c < Nc; c++) w[pt * Nc + c] *= mul * w1[j];
2016:         }
2017:       }
2018:     }
2019:     totprev *= npoints;
2020:     totrem /= npoints;
2021:   }
2022:   PetscCall(PetscFree2(p1, w1));
2023:   PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, q));
2024:   PetscCall(PetscQuadratureSetCellType(*q, ct));
2025:   PetscCall(PetscQuadratureSetOrder(*q, 2 * npoints - 1));
2026:   PetscCall(PetscQuadratureSetData(*q, dim, Nc, totpoints, x, w));
2027:   PetscCall(PetscObjectChangeTypeName((PetscObject)*q, "StroudConical"));
2028:   PetscFunctionReturn(PETSC_SUCCESS);
2029: }

2031: static PetscBool MinSymTriQuadCite       = PETSC_FALSE;
2032: const char       MinSymTriQuadCitation[] = "@article{WitherdenVincent2015,\n"
2033:                                            "  title = {On the identification of symmetric quadrature rules for finite element methods},\n"
2034:                                            "  journal = {Computers & Mathematics with Applications},\n"
2035:                                            "  volume = {69},\n"
2036:                                            "  number = {10},\n"
2037:                                            "  pages = {1232-1241},\n"
2038:                                            "  year = {2015},\n"
2039:                                            "  issn = {0898-1221},\n"
2040:                                            "  doi = {10.1016/j.camwa.2015.03.017},\n"
2041:                                            "  url = {https://www.sciencedirect.com/science/article/pii/S0898122115001224},\n"
2042:                                            "  author = {F.D. Witherden and P.E. Vincent},\n"
2043:                                            "}\n";

2045: #include "petscdttriquadrules.h"

2047: static PetscBool MinSymTetQuadCite       = PETSC_FALSE;
2048: const char       MinSymTetQuadCitation[] = "@article{JaskowiecSukumar2021\n"
2049:                                            "  author = {Jaskowiec, Jan and Sukumar, N.},\n"
2050:                                            "  title = {High-order symmetric cubature rules for tetrahedra and pyramids},\n"
2051:                                            "  journal = {International Journal for Numerical Methods in Engineering},\n"
2052:                                            "  volume = {122},\n"
2053:                                            "  number = {1},\n"
2054:                                            "  pages = {148-171},\n"
2055:                                            "  doi = {10.1002/nme.6528},\n"
2056:                                            "  url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/nme.6528},\n"
2057:                                            "  eprint = {https://onlinelibrary.wiley.com/doi/pdf/10.1002/nme.6528},\n"
2058:                                            "  year = {2021}\n"
2059:                                            "}\n";

2061: #include "petscdttetquadrules.h"

2063: static PetscBool DiagSymTriQuadCite       = PETSC_FALSE;
2064: const char       DiagSymTriQuadCitation[] = "@article{KongMulderVeldhuizen1999,\n"
2065:                                             "  title = {Higher-order triangular and tetrahedral finite elements with mass lumping for solving the wave equation},\n"
2066:                                             "  journal = {Journal of Engineering Mathematics},\n"
2067:                                             "  volume = {35},\n"
2068:                                             "  number = {4},\n"
2069:                                             "  pages = {405--426},\n"
2070:                                             "  year = {1999},\n"
2071:                                             "  doi = {10.1023/A:1004420829610},\n"
2072:                                             "  url = {https://link.springer.com/article/10.1023/A:1004420829610},\n"
2073:                                             "  author = {MJS Chin-Joe-Kong and Wim A Mulder and Marinus Van Veldhuizen},\n"
2074:                                             "}\n";

2076: #include "petscdttridiagquadrules.h"

2078: // https://en.wikipedia.org/wiki/Partition_(number_theory)
2079: static PetscErrorCode PetscDTPartitionNumber(PetscInt n, PetscInt *p)
2080: {
2081:   // sequence A000041 in the OEIS
2082:   const PetscInt partition[]   = {1, 1, 2, 3, 5, 7, 11, 15, 22, 30, 42, 56, 77, 101, 135, 176, 231, 297, 385, 490, 627, 792, 1002, 1255, 1575, 1958, 2436, 3010, 3718, 4565, 5604};
2083:   PetscInt       tabulated_max = PETSC_STATIC_ARRAY_LENGTH(partition) - 1;

2085:   PetscFunctionBegin;
2086:   PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Partition number not defined for negative number %" PetscInt_FMT, n);
2087:   // not implementing the pentagonal number recurrence, we don't need partition numbers for n that high
2088:   PetscCheck(n <= tabulated_max, PETSC_COMM_SELF, PETSC_ERR_SUP, "Partition numbers only tabulated up to %" PetscInt_FMT ", not computed for %" PetscInt_FMT, tabulated_max, n);
2089:   *p = partition[n];
2090:   PetscFunctionReturn(PETSC_SUCCESS);
2091: }

2093: /*@
2094:   PetscDTSimplexQuadrature - Create a quadrature rule for a simplex that exactly integrates polynomials up to a given degree.

2096:   Not Collective

2098:   Input Parameters:
2099: + dim    - The spatial dimension of the simplex (1 = segment, 2 = triangle, 3 = tetrahedron)
2100: . degree - The largest polynomial degree that is required to be integrated exactly
2101: - type   - `PetscDTSimplexQuadratureType` indicating the type of quadrature rule

2103:   Output Parameter:
2104: . quad - A `PetscQuadrature` object for integration over the biunit simplex
2105:             (defined by the bounds $x_i >= -1$ and $\sum_i x_i <= 2 - d$) that is exact for
2106:             polynomials up to the given degree

2108:   Level: intermediate

2110: .seealso: `PetscDTSimplexQuadratureType`, `PetscDTGaussQuadrature()`, `PetscDTStroudCononicalQuadrature()`, `PetscQuadrature`
2111: @*/
2112: PetscErrorCode PetscDTSimplexQuadrature(PetscInt dim, PetscInt degree, PetscDTSimplexQuadratureType type, PetscQuadrature *quad)
2113: {
2114:   PetscDTSimplexQuadratureType orig_type = type;

2116:   PetscFunctionBegin;
2117:   PetscCheck(dim >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative dimension %" PetscInt_FMT, dim);
2118:   PetscCheck(degree >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative degree %" PetscInt_FMT, degree);
2119:   if (type == PETSCDTSIMPLEXQUAD_DEFAULT) type = PETSCDTSIMPLEXQUAD_MINSYM;
2120:   if (type == PETSCDTSIMPLEXQUAD_CONIC || dim < 2) {
2121:     PetscInt points_per_dim = (degree + 2) / 2; // ceil((degree + 1) / 2);
2122:     PetscCall(PetscDTStroudConicalQuadrature(dim, 1, points_per_dim, -1, 1, quad));
2123:   } else {
2124:     DMPolytopeType    ct;
2125:     PetscInt          n    = dim + 1;
2126:     PetscInt          fact = 1;
2127:     PetscInt         *part, *perm;
2128:     PetscInt          p = 0;
2129:     PetscInt          max_degree;
2130:     const PetscInt   *nodes_per_type     = NULL;
2131:     const PetscInt   *all_num_full_nodes = NULL;
2132:     const PetscReal **weights_list       = NULL;
2133:     const PetscReal **compact_nodes_list = NULL;
2134:     const char       *citation           = NULL;
2135:     PetscBool        *cited              = NULL;

2137:     switch (dim) {
2138:     case 0:
2139:       ct = DM_POLYTOPE_POINT;
2140:       break;
2141:     case 1:
2142:       ct = DM_POLYTOPE_SEGMENT;
2143:       break;
2144:     case 2:
2145:       ct = DM_POLYTOPE_TRIANGLE;
2146:       break;
2147:     case 3:
2148:       ct = DM_POLYTOPE_TETRAHEDRON;
2149:       break;
2150:     default:
2151:       ct = DM_POLYTOPE_UNKNOWN;
2152:     }
2153:     if (type == PETSCDTSIMPLEXQUAD_MINSYM) {
2154:       switch (dim) {
2155:       case 2:
2156:         cited              = &MinSymTriQuadCite;
2157:         citation           = MinSymTriQuadCitation;
2158:         max_degree         = PetscDTWVTriQuad_max_degree;
2159:         nodes_per_type     = PetscDTWVTriQuad_num_orbits;
2160:         all_num_full_nodes = PetscDTWVTriQuad_num_nodes;
2161:         weights_list       = PetscDTWVTriQuad_weights;
2162:         compact_nodes_list = PetscDTWVTriQuad_orbits;
2163:         break;
2164:       case 3:
2165:         cited              = &MinSymTetQuadCite;
2166:         citation           = MinSymTetQuadCitation;
2167:         max_degree         = PetscDTJSTetQuad_max_degree;
2168:         nodes_per_type     = PetscDTJSTetQuad_num_orbits;
2169:         all_num_full_nodes = PetscDTJSTetQuad_num_nodes;
2170:         weights_list       = PetscDTJSTetQuad_weights;
2171:         compact_nodes_list = PetscDTJSTetQuad_orbits;
2172:         break;
2173:       default:
2174:         max_degree = -1;
2175:         break;
2176:       }
2177:     } else {
2178:       switch (dim) {
2179:       case 2:
2180:         cited              = &DiagSymTriQuadCite;
2181:         citation           = DiagSymTriQuadCitation;
2182:         max_degree         = PetscDTKMVTriQuad_max_degree;
2183:         nodes_per_type     = PetscDTKMVTriQuad_num_orbits;
2184:         all_num_full_nodes = PetscDTKMVTriQuad_num_nodes;
2185:         weights_list       = PetscDTKMVTriQuad_weights;
2186:         compact_nodes_list = PetscDTKMVTriQuad_orbits;
2187:         break;
2188:       default:
2189:         max_degree = -1;
2190:         break;
2191:       }
2192:     }

2194:     if (degree > max_degree) {
2195:       PetscCheck(orig_type == PETSCDTSIMPLEXQUAD_DEFAULT, PETSC_COMM_SELF, PETSC_ERR_SUP, "%s symmetric quadrature for dim %" PetscInt_FMT ", degree %" PetscInt_FMT " unsupported", orig_type == PETSCDTSIMPLEXQUAD_MINSYM ? "Minimal" : "Diagonal", dim, degree);
2196:       // fall back to conic
2197:       PetscCall(PetscDTSimplexQuadrature(dim, degree, PETSCDTSIMPLEXQUAD_CONIC, quad));
2198:       PetscFunctionReturn(PETSC_SUCCESS);
2199:     }

2201:     PetscCall(PetscCitationsRegister(citation, cited));

2203:     PetscCall(PetscDTPartitionNumber(n, &p));
2204:     for (PetscInt d = 2; d <= n; d++) fact *= d;

2206:     PetscInt         num_full_nodes      = all_num_full_nodes[degree];
2207:     const PetscReal *all_compact_nodes   = compact_nodes_list[degree];
2208:     const PetscReal *all_compact_weights = weights_list[degree];
2209:     nodes_per_type                       = &nodes_per_type[p * degree];

2211:     PetscReal      *points;
2212:     PetscReal      *counts;
2213:     PetscReal      *weights;
2214:     PetscReal      *bary_to_biunit; // row-major transformation of barycentric coordinate to biunit
2215:     PetscQuadrature q;

2217:     // compute the transformation
2218:     PetscCall(PetscMalloc1(n * dim, &bary_to_biunit));
2219:     for (PetscInt d = 0; d < dim; d++) {
2220:       for (PetscInt b = 0; b < n; b++) bary_to_biunit[d * n + b] = (d == b) ? 1.0 : -1.0;
2221:     }

2223:     PetscCall(PetscMalloc3(n, &part, n, &perm, n, &counts));
2224:     PetscCall(PetscCalloc1(num_full_nodes * dim, &points));
2225:     PetscCall(PetscMalloc1(num_full_nodes, &weights));

2227:     // (0, 0, ...) is the first partition lexicographically
2228:     PetscCall(PetscArrayzero(part, n));
2229:     PetscCall(PetscArrayzero(counts, n));
2230:     counts[0] = n;

2232:     // for each partition
2233:     for (PetscInt s = 0, node_offset = 0; s < p; s++) {
2234:       PetscInt num_compact_coords = part[n - 1] + 1;

2236:       const PetscReal *compact_nodes   = all_compact_nodes;
2237:       const PetscReal *compact_weights = all_compact_weights;
2238:       all_compact_nodes += num_compact_coords * nodes_per_type[s];
2239:       all_compact_weights += nodes_per_type[s];

2241:       // for every permutation of the vertices
2242:       for (PetscInt f = 0; f < fact; f++) {
2243:         PetscCall(PetscDTEnumPerm(n, f, perm, NULL));

2245:         // check if it is a valid permutation
2246:         PetscInt digit;
2247:         for (digit = 1; digit < n; digit++) {
2248:           // skip permutations that would duplicate a node because it has a smaller symmetry group
2249:           if (part[digit - 1] == part[digit] && perm[digit - 1] > perm[digit]) break;
2250:         }
2251:         if (digit < n) continue;

2253:         // create full nodes from this permutation of the compact nodes
2254:         PetscReal *full_nodes   = &points[node_offset * dim];
2255:         PetscReal *full_weights = &weights[node_offset];

2257:         PetscCall(PetscArraycpy(full_weights, compact_weights, nodes_per_type[s]));
2258:         for (PetscInt b = 0; b < n; b++) {
2259:           for (PetscInt d = 0; d < dim; d++) {
2260:             for (PetscInt node = 0; node < nodes_per_type[s]; node++) full_nodes[node * dim + d] += bary_to_biunit[d * n + perm[b]] * compact_nodes[node * num_compact_coords + part[b]];
2261:           }
2262:         }
2263:         node_offset += nodes_per_type[s];
2264:       }

2266:       if (s < p - 1) { // Generate the next partition
2267:         /* A partition is described by the number of coordinates that are in
2268:          * each set of duplicates (counts) and redundantly by mapping each
2269:          * index to its set of duplicates (part)
2270:          *
2271:          * Counts should always be in nonincreasing order
2272:          *
2273:          * We want to generate the partitions lexically by part, which means
2274:          * finding the last index where count > 1 and reducing by 1.
2275:          *
2276:          * For the new counts beyond that index, we eagerly assign the remaining
2277:          * capacity of the partition to smaller indices (ensures lexical ordering),
2278:          * while respecting the nonincreasing invariant of the counts
2279:          */
2280:         PetscInt last_digit            = part[n - 1];
2281:         PetscInt last_digit_with_extra = last_digit;
2282:         while (counts[last_digit_with_extra] == 1) last_digit_with_extra--;
2283:         PetscInt limit               = --counts[last_digit_with_extra];
2284:         PetscInt total_to_distribute = last_digit - last_digit_with_extra + 1;
2285:         for (PetscInt digit = last_digit_with_extra + 1; digit < n; digit++) {
2286:           counts[digit] = PetscMin(limit, total_to_distribute);
2287:           total_to_distribute -= PetscMin(limit, total_to_distribute);
2288:         }
2289:         for (PetscInt digit = 0, offset = 0; digit < n; digit++) {
2290:           PetscInt count = counts[digit];
2291:           for (PetscInt c = 0; c < count; c++) part[offset++] = digit;
2292:         }
2293:       }
2294:       PetscCheck(node_offset <= num_full_nodes, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Node offset %" PetscInt_FMT " > %" PetscInt_FMT " number of nodes", node_offset, num_full_nodes);
2295:     }
2296:     PetscCall(PetscFree3(part, perm, counts));
2297:     PetscCall(PetscFree(bary_to_biunit));
2298:     PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, &q));
2299:     PetscCall(PetscQuadratureSetCellType(q, ct));
2300:     PetscCall(PetscQuadratureSetOrder(q, degree));
2301:     PetscCall(PetscQuadratureSetData(q, dim, 1, num_full_nodes, points, weights));
2302:     *quad = q;
2303:   }
2304:   PetscFunctionReturn(PETSC_SUCCESS);
2305: }

2307: /*@
2308:   PetscDTTanhSinhTensorQuadrature - create tanh-sinh quadrature for a tensor product cell

2310:   Not Collective

2312:   Input Parameters:
2313: + dim   - The cell dimension
2314: . level - The number of points in one dimension, $2^l$
2315: . a     - left end of interval (often-1)
2316: - b     - right end of interval (often +1)

2318:   Output Parameter:
2319: . q - A `PetscQuadrature` object

2321:   Level: intermediate

2323: .seealso: `PetscDTGaussTensorQuadrature()`, `PetscQuadrature`
2324: @*/
2325: PetscErrorCode PetscDTTanhSinhTensorQuadrature(PetscInt dim, PetscInt level, PetscReal a, PetscReal b, PetscQuadrature *q)
2326: {
2327:   DMPolytopeType  ct;
2328:   const PetscInt  p     = 16;                        /* Digits of precision in the evaluation */
2329:   const PetscReal alpha = (b - a) / 2.;              /* Half-width of the integration interval */
2330:   const PetscReal beta  = (b + a) / 2.;              /* Center of the integration interval */
2331:   const PetscReal h     = PetscPowReal(2.0, -level); /* Step size, length between x_k */
2332:   PetscReal       xk;                                /* Quadrature point x_k on reference domain [-1, 1] */
2333:   PetscReal       wk = 0.5 * PETSC_PI;               /* Quadrature weight at x_k */
2334:   PetscReal      *x, *w;
2335:   PetscInt        K, k, npoints;

2337:   PetscFunctionBegin;
2338:   PetscCheck(dim <= 1, PETSC_COMM_SELF, PETSC_ERR_SUP, "Dimension %" PetscInt_FMT " not yet implemented", dim);
2339:   PetscCheck(level, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must give a number of significant digits");
2340:   switch (dim) {
2341:   case 0:
2342:     ct = DM_POLYTOPE_POINT;
2343:     break;
2344:   case 1:
2345:     ct = DM_POLYTOPE_SEGMENT;
2346:     break;
2347:   case 2:
2348:     ct = DM_POLYTOPE_QUADRILATERAL;
2349:     break;
2350:   case 3:
2351:     ct = DM_POLYTOPE_HEXAHEDRON;
2352:     break;
2353:   default:
2354:     ct = DM_POLYTOPE_UNKNOWN;
2355:   }
2356:   /* Find K such that the weights are < 32 digits of precision */
2357:   for (K = 1; PetscAbsReal(PetscLog10Real(wk)) < 2 * p; ++K) wk = 0.5 * h * PETSC_PI * PetscCoshReal(K * h) / PetscSqr(PetscCoshReal(0.5 * PETSC_PI * PetscSinhReal(K * h)));
2358:   PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, q));
2359:   PetscCall(PetscQuadratureSetCellType(*q, ct));
2360:   PetscCall(PetscQuadratureSetOrder(*q, 2 * K + 1));
2361:   npoints = 2 * K - 1;
2362:   PetscCall(PetscMalloc1(npoints * dim, &x));
2363:   PetscCall(PetscMalloc1(npoints, &w));
2364:   /* Center term */
2365:   x[0] = beta;
2366:   w[0] = 0.5 * alpha * PETSC_PI;
2367:   for (k = 1; k < K; ++k) {
2368:     wk           = 0.5 * alpha * h * PETSC_PI * PetscCoshReal(k * h) / PetscSqr(PetscCoshReal(0.5 * PETSC_PI * PetscSinhReal(k * h)));
2369:     xk           = PetscTanhReal(0.5 * PETSC_PI * PetscSinhReal(k * h));
2370:     x[2 * k - 1] = -alpha * xk + beta;
2371:     w[2 * k - 1] = wk;
2372:     x[2 * k + 0] = alpha * xk + beta;
2373:     w[2 * k + 0] = wk;
2374:   }
2375:   PetscCall(PetscQuadratureSetData(*q, dim, 1, npoints, x, w));
2376:   PetscFunctionReturn(PETSC_SUCCESS);
2377: }

2379: /*@C
2380:   PetscDTTanhSinhIntegrate - Approximate $\int_a^b f(x)\,dx$ to a requested precision using adaptive tanh-sinh (double-exponential) quadrature

2382:   Not Collective; No Fortran Support

2384:   Input Parameters:
2385: + func   - the integrand callback (`func(x, ctx, &value)` evaluates the integrand at point `x`)
2386: . a      - lower limit of integration
2387: . b      - upper limit of integration
2388: . digits - target number of correct decimal digits
2389: - ctx    - optional user context passed to `func`

2391:   Output Parameter:
2392: . sol - the approximate value of the integral

2394:   Level: developer

2396:   Notes:
2397:   Doubles the number of quadrature points at each refinement level until the change in the
2398:   integral falls below the requested tolerance. Suitable for smooth integrands and integrands
2399:   with endpoint singularities.

2401:   For arbitrary-precision arithmetic via MPFR, see `PetscDTTanhSinhIntegrateMPFR()`.

2403: .seealso: `PetscDTTanhSinhIntegrateMPFR()`, `PetscDTGaussQuadrature()`
2404: @*/
2405: PetscErrorCode PetscDTTanhSinhIntegrate(void (*func)(const PetscReal[], PetscCtx, PetscReal *), PetscReal a, PetscReal b, PetscInt digits, PetscCtx ctx, PetscReal *sol)
2406: {
2407:   const PetscInt  p     = 16;           /* Digits of precision in the evaluation */
2408:   const PetscReal alpha = (b - a) / 2.; /* Half-width of the integration interval */
2409:   const PetscReal beta  = (b + a) / 2.; /* Center of the integration interval */
2410:   PetscReal       h     = 1.0;          /* Step size, length between x_k */
2411:   PetscInt        l     = 0;            /* Level of refinement, h = 2^{-l} */
2412:   PetscReal       osum  = 0.0;          /* Integral on last level */
2413:   PetscReal       psum  = 0.0;          /* Integral on the level before the last level */
2414:   PetscReal       sum;                  /* Integral on current level */
2415:   PetscReal       yk;                   /* Quadrature point 1 - x_k on reference domain [-1, 1] */
2416:   PetscReal       lx, rx;               /* Quadrature points to the left and right of 0 on the real domain [a, b] */
2417:   PetscReal       wk;                   /* Quadrature weight at x_k */
2418:   PetscReal       lval, rval;           /* Terms in the quadature sum to the left and right of 0 */
2419:   PetscInt        d;                    /* Digits of precision in the integral */

2421:   PetscFunctionBegin;
2422:   PetscCheck(digits > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must give a positive number of significant digits");
2423:   PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF));
2424:   /* Center term */
2425:   func(&beta, ctx, &lval);
2426:   sum = 0.5 * alpha * PETSC_PI * lval;
2427:   /* */
2428:   do {
2429:     PetscReal lterm, rterm, maxTerm = 0.0, d1, d2, d3, d4;
2430:     PetscInt  k = 1;

2432:     ++l;
2433:     /* PetscPrintf(PETSC_COMM_SELF, "LEVEL %" PetscInt_FMT " sum: %15.15f\n", l, sum); */
2434:     /* At each level of refinement, h --> h/2 and sum --> sum/2 */
2435:     psum = osum;
2436:     osum = sum;
2437:     h *= 0.5;
2438:     sum *= 0.5;
2439:     do {
2440:       wk = 0.5 * h * PETSC_PI * PetscCoshReal(k * h) / PetscSqr(PetscCoshReal(0.5 * PETSC_PI * PetscSinhReal(k * h)));
2441:       yk = 1.0 / (PetscExpReal(0.5 * PETSC_PI * PetscSinhReal(k * h)) * PetscCoshReal(0.5 * PETSC_PI * PetscSinhReal(k * h)));
2442:       lx = -alpha * (1.0 - yk) + beta;
2443:       rx = alpha * (1.0 - yk) + beta;
2444:       func(&lx, ctx, &lval);
2445:       func(&rx, ctx, &rval);
2446:       lterm   = alpha * wk * lval;
2447:       maxTerm = PetscMax(PetscAbsReal(lterm), maxTerm);
2448:       sum += lterm;
2449:       rterm   = alpha * wk * rval;
2450:       maxTerm = PetscMax(PetscAbsReal(rterm), maxTerm);
2451:       sum += rterm;
2452:       ++k;
2453:       /* Only need to evaluate every other point on refined levels */
2454:       if (l != 1) ++k;
2455:     } while (PetscAbsReal(PetscLog10Real(wk)) < p); /* Only need to evaluate sum until weights are < 32 digits of precision */

2457:     d1 = PetscLog10Real(PetscAbsReal(sum - osum));
2458:     d2 = PetscLog10Real(PetscAbsReal(sum - psum));
2459:     d3 = PetscLog10Real(maxTerm) - p;
2460:     if (PetscMax(PetscAbsReal(lterm), PetscAbsReal(rterm)) == 0.0) d4 = 0.0;
2461:     else d4 = PetscLog10Real(PetscMax(PetscAbsReal(lterm), PetscAbsReal(rterm)));
2462:     d = PetscAbsInt(PetscMin(0, PetscMax(PetscMax(PetscMax(PetscSqr(d1) / d2, 2 * d1), d3), d4)));
2463:   } while (d < digits && l < 12);
2464:   *sol = sum;
2465:   PetscCall(PetscFPTrapPop());
2466:   PetscFunctionReturn(PETSC_SUCCESS);
2467: }

2469: /*@C
2470:   PetscDTTanhSinhIntegrateMPFR - High-precision version of `PetscDTTanhSinhIntegrate()` that uses the MPFR arbitrary-precision library to evaluate the quadrature

2472:   Not Collective; No Fortran Support

2474:   Input Parameters:
2475: + func   - the integrand callback (`func(x, ctx, &value)` evaluates the integrand at point `x`)
2476: . a      - lower limit of integration
2477: . b      - upper limit of integration
2478: . digits - target number of correct decimal digits (also drives the working MPFR precision)
2479: - ctx    - optional user context passed to `func`

2481:   Output Parameter:
2482: . sol - the approximate value of the integral

2484:   Level: developer

2486:   Note:
2487:   Requires PETSc to be configured with `--with-mpfr`; otherwise an error is raised.

2489: .seealso: `PetscDTTanhSinhIntegrate()`, `PetscDTGaussQuadrature()`
2490: @*/
2491: #if defined(PETSC_HAVE_MPFR)
2492: PetscErrorCode PetscDTTanhSinhIntegrateMPFR(void (*func)(const PetscReal[], PetscCtx, PetscReal *), PetscReal a, PetscReal b, PetscInt digits, PetscCtx ctx, PetscReal *sol)
2493: {
2494:   const PetscInt safetyFactor = 2; /* Calculate abscissa until 2*p digits */
2495:   PetscInt       l            = 0; /* Level of refinement, h = 2^{-l} */
2496:   mpfr_t         alpha;            /* Half-width of the integration interval */
2497:   mpfr_t         beta;             /* Center of the integration interval */
2498:   mpfr_t         h;                /* Step size, length between x_k */
2499:   mpfr_t         osum;             /* Integral on last level */
2500:   mpfr_t         psum;             /* Integral on the level before the last level */
2501:   mpfr_t         sum;              /* Integral on current level */
2502:   mpfr_t         yk;               /* Quadrature point 1 - x_k on reference domain [-1, 1] */
2503:   mpfr_t         lx, rx;           /* Quadrature points to the left and right of 0 on the real domain [a, b] */
2504:   mpfr_t         wk;               /* Quadrature weight at x_k */
2505:   PetscReal      lval, rval, rtmp; /* Terms in the quadature sum to the left and right of 0 */
2506:   PetscInt       d;                /* Digits of precision in the integral */
2507:   mpfr_t         pi2, kh, msinh, mcosh, maxTerm, curTerm, tmp;

2509:   PetscFunctionBegin;
2510:   PetscCheck(digits > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must give a positive number of significant digits");
2511:   /* Create high precision storage */
2512:   mpfr_inits2(PetscCeilReal(safetyFactor * digits * PetscLogReal(10.) / PetscLogReal(2.)), alpha, beta, h, sum, osum, psum, yk, wk, lx, rx, tmp, maxTerm, curTerm, pi2, kh, msinh, mcosh, NULL);
2513:   /* Initialization */
2514:   mpfr_set_d(alpha, 0.5 * (b - a), MPFR_RNDN);
2515:   mpfr_set_d(beta, 0.5 * (b + a), MPFR_RNDN);
2516:   mpfr_set_d(osum, 0.0, MPFR_RNDN);
2517:   mpfr_set_d(psum, 0.0, MPFR_RNDN);
2518:   mpfr_set_d(h, 1.0, MPFR_RNDN);
2519:   mpfr_const_pi(pi2, MPFR_RNDN);
2520:   mpfr_mul_d(pi2, pi2, 0.5, MPFR_RNDN);
2521:   /* Center term */
2522:   rtmp = 0.5 * (b + a);
2523:   func(&rtmp, ctx, &lval);
2524:   mpfr_set(sum, pi2, MPFR_RNDN);
2525:   mpfr_mul(sum, sum, alpha, MPFR_RNDN);
2526:   mpfr_mul_d(sum, sum, lval, MPFR_RNDN);
2527:   /* */
2528:   do {
2529:     PetscReal d1, d2, d3, d4;
2530:     PetscInt  k = 1;

2532:     ++l;
2533:     mpfr_set_d(maxTerm, 0.0, MPFR_RNDN);
2534:     /* PetscPrintf(PETSC_COMM_SELF, "LEVEL %" PetscInt_FMT " sum: %15.15f\n", l, sum); */
2535:     /* At each level of refinement, h --> h/2 and sum --> sum/2 */
2536:     mpfr_set(psum, osum, MPFR_RNDN);
2537:     mpfr_set(osum, sum, MPFR_RNDN);
2538:     mpfr_mul_d(h, h, 0.5, MPFR_RNDN);
2539:     mpfr_mul_d(sum, sum, 0.5, MPFR_RNDN);
2540:     do {
2541:       mpfr_set_si(kh, k, MPFR_RNDN);
2542:       mpfr_mul(kh, kh, h, MPFR_RNDN);
2543:       /* Weight */
2544:       mpfr_set(wk, h, MPFR_RNDN);
2545:       mpfr_sinh_cosh(msinh, mcosh, kh, MPFR_RNDN);
2546:       mpfr_mul(msinh, msinh, pi2, MPFR_RNDN);
2547:       mpfr_mul(mcosh, mcosh, pi2, MPFR_RNDN);
2548:       mpfr_cosh(tmp, msinh, MPFR_RNDN);
2549:       mpfr_sqr(tmp, tmp, MPFR_RNDN);
2550:       mpfr_mul(wk, wk, mcosh, MPFR_RNDN);
2551:       mpfr_div(wk, wk, tmp, MPFR_RNDN);
2552:       /* Abscissa */
2553:       mpfr_set_d(yk, 1.0, MPFR_RNDZ);
2554:       mpfr_cosh(tmp, msinh, MPFR_RNDN);
2555:       mpfr_div(yk, yk, tmp, MPFR_RNDZ);
2556:       mpfr_exp(tmp, msinh, MPFR_RNDN);
2557:       mpfr_div(yk, yk, tmp, MPFR_RNDZ);
2558:       /* Quadrature points */
2559:       mpfr_sub_d(lx, yk, 1.0, MPFR_RNDZ);
2560:       mpfr_mul(lx, lx, alpha, MPFR_RNDU);
2561:       mpfr_add(lx, lx, beta, MPFR_RNDU);
2562:       mpfr_d_sub(rx, 1.0, yk, MPFR_RNDZ);
2563:       mpfr_mul(rx, rx, alpha, MPFR_RNDD);
2564:       mpfr_add(rx, rx, beta, MPFR_RNDD);
2565:       /* Evaluation */
2566:       rtmp = mpfr_get_d(lx, MPFR_RNDU);
2567:       func(&rtmp, ctx, &lval);
2568:       rtmp = mpfr_get_d(rx, MPFR_RNDD);
2569:       func(&rtmp, ctx, &rval);
2570:       /* Update */
2571:       mpfr_mul(tmp, wk, alpha, MPFR_RNDN);
2572:       mpfr_mul_d(tmp, tmp, lval, MPFR_RNDN);
2573:       mpfr_add(sum, sum, tmp, MPFR_RNDN);
2574:       mpfr_abs(tmp, tmp, MPFR_RNDN);
2575:       mpfr_max(maxTerm, maxTerm, tmp, MPFR_RNDN);
2576:       mpfr_set(curTerm, tmp, MPFR_RNDN);
2577:       mpfr_mul(tmp, wk, alpha, MPFR_RNDN);
2578:       mpfr_mul_d(tmp, tmp, rval, MPFR_RNDN);
2579:       mpfr_add(sum, sum, tmp, MPFR_RNDN);
2580:       mpfr_abs(tmp, tmp, MPFR_RNDN);
2581:       mpfr_max(maxTerm, maxTerm, tmp, MPFR_RNDN);
2582:       mpfr_max(curTerm, curTerm, tmp, MPFR_RNDN);
2583:       ++k;
2584:       /* Only need to evaluate every other point on refined levels */
2585:       if (l != 1) ++k;
2586:       mpfr_log10(tmp, wk, MPFR_RNDN);
2587:       mpfr_abs(tmp, tmp, MPFR_RNDN);
2588:     } while (mpfr_get_d(tmp, MPFR_RNDN) < safetyFactor * digits); /* Only need to evaluate sum until weights are < 32 digits of precision */
2589:     mpfr_sub(tmp, sum, osum, MPFR_RNDN);
2590:     mpfr_abs(tmp, tmp, MPFR_RNDN);
2591:     mpfr_log10(tmp, tmp, MPFR_RNDN);
2592:     d1 = mpfr_get_d(tmp, MPFR_RNDN);
2593:     mpfr_sub(tmp, sum, psum, MPFR_RNDN);
2594:     mpfr_abs(tmp, tmp, MPFR_RNDN);
2595:     mpfr_log10(tmp, tmp, MPFR_RNDN);
2596:     d2 = mpfr_get_d(tmp, MPFR_RNDN);
2597:     mpfr_log10(tmp, maxTerm, MPFR_RNDN);
2598:     d3 = mpfr_get_d(tmp, MPFR_RNDN) - digits;
2599:     mpfr_log10(tmp, curTerm, MPFR_RNDN);
2600:     d4 = mpfr_get_d(tmp, MPFR_RNDN);
2601:     d  = PetscAbsInt(PetscMin(0, PetscMax(PetscMax(PetscMax(PetscSqr(d1) / d2, 2 * d1), d3), d4)));
2602:   } while (d < digits && l < 8);
2603:   *sol = mpfr_get_d(sum, MPFR_RNDN);
2604:   /* Cleanup */
2605:   mpfr_clears(alpha, beta, h, sum, osum, psum, yk, wk, lx, rx, tmp, maxTerm, curTerm, pi2, kh, msinh, mcosh, NULL);
2606:   PetscFunctionReturn(PETSC_SUCCESS);
2607: }
2608: #else

2610: PetscErrorCode PetscDTTanhSinhIntegrateMPFR(void (*func)(const PetscReal[], void *, PetscReal *), PetscReal a, PetscReal b, PetscInt digits, PetscCtx ctx, PetscReal *sol)
2611: {
2612:   SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "This method will not work without MPFR. Reconfigure using --download-mpfr --download-gmp");
2613: }
2614: #endif

2616: /*@
2617:   PetscDTTensorQuadratureCreate - create the tensor product quadrature from two lower-dimensional quadratures

2619:   Not Collective

2621:   Input Parameters:
2622: + q1 - The first quadrature
2623: - q2 - The second quadrature

2625:   Output Parameter:
2626: . q - A `PetscQuadrature` object

2628:   Level: intermediate

2630: .seealso: `PetscQuadrature`, `PetscDTGaussTensorQuadrature()`
2631: @*/
2632: PetscErrorCode PetscDTTensorQuadratureCreate(PetscQuadrature q1, PetscQuadrature q2, PetscQuadrature *q)
2633: {
2634:   DMPolytopeType   ct1, ct2, ct;
2635:   const PetscReal *x1, *w1, *x2, *w2;
2636:   PetscReal       *x, *w;
2637:   PetscInt         dim1, Nc1, Np1, order1, qa, d1;
2638:   PetscInt         dim2, Nc2, Np2, order2, qb, d2;
2639:   PetscInt         dim, Nc, Np, order, qc, d;

2641:   PetscFunctionBegin;
2644:   PetscAssertPointer(q, 3);

2646:   PetscCall(PetscQuadratureGetOrder(q1, &order1));
2647:   PetscCall(PetscQuadratureGetOrder(q2, &order2));
2648:   PetscCheck(order1 == order2, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Order1 %" PetscInt_FMT " != %" PetscInt_FMT " Order2", order1, order2);
2649:   PetscCall(PetscQuadratureGetData(q1, &dim1, &Nc1, &Np1, &x1, &w1));
2650:   PetscCall(PetscQuadratureGetCellType(q1, &ct1));
2651:   PetscCall(PetscQuadratureGetData(q2, &dim2, &Nc2, &Np2, &x2, &w2));
2652:   PetscCall(PetscQuadratureGetCellType(q2, &ct2));
2653:   PetscCheck(Nc1 == Nc2, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "NumComp1 %" PetscInt_FMT " != %" PetscInt_FMT " NumComp2", Nc1, Nc2);

2655:   switch (ct1) {
2656:   case DM_POLYTOPE_POINT:
2657:     ct = ct2;
2658:     break;
2659:   case DM_POLYTOPE_SEGMENT:
2660:     switch (ct2) {
2661:     case DM_POLYTOPE_POINT:
2662:       ct = ct1;
2663:       break;
2664:     case DM_POLYTOPE_SEGMENT:
2665:       ct = DM_POLYTOPE_QUADRILATERAL;
2666:       break;
2667:     case DM_POLYTOPE_TRIANGLE:
2668:       ct = DM_POLYTOPE_TRI_PRISM;
2669:       break;
2670:     case DM_POLYTOPE_QUADRILATERAL:
2671:       ct = DM_POLYTOPE_HEXAHEDRON;
2672:       break;
2673:     case DM_POLYTOPE_TETRAHEDRON:
2674:       ct = DM_POLYTOPE_UNKNOWN;
2675:       break;
2676:     case DM_POLYTOPE_HEXAHEDRON:
2677:       ct = DM_POLYTOPE_UNKNOWN;
2678:       break;
2679:     default:
2680:       ct = DM_POLYTOPE_UNKNOWN;
2681:     }
2682:     break;
2683:   case DM_POLYTOPE_TRIANGLE:
2684:     switch (ct2) {
2685:     case DM_POLYTOPE_POINT:
2686:       ct = ct1;
2687:       break;
2688:     case DM_POLYTOPE_SEGMENT:
2689:       ct = DM_POLYTOPE_TRI_PRISM;
2690:       break;
2691:     case DM_POLYTOPE_TRIANGLE:
2692:       ct = DM_POLYTOPE_UNKNOWN;
2693:       break;
2694:     case DM_POLYTOPE_QUADRILATERAL:
2695:       ct = DM_POLYTOPE_UNKNOWN;
2696:       break;
2697:     case DM_POLYTOPE_TETRAHEDRON:
2698:       ct = DM_POLYTOPE_UNKNOWN;
2699:       break;
2700:     case DM_POLYTOPE_HEXAHEDRON:
2701:       ct = DM_POLYTOPE_UNKNOWN;
2702:       break;
2703:     default:
2704:       ct = DM_POLYTOPE_UNKNOWN;
2705:     }
2706:     break;
2707:   case DM_POLYTOPE_QUADRILATERAL:
2708:     switch (ct2) {
2709:     case DM_POLYTOPE_POINT:
2710:       ct = ct1;
2711:       break;
2712:     case DM_POLYTOPE_SEGMENT:
2713:       ct = DM_POLYTOPE_HEXAHEDRON;
2714:       break;
2715:     case DM_POLYTOPE_TRIANGLE:
2716:       ct = DM_POLYTOPE_UNKNOWN;
2717:       break;
2718:     case DM_POLYTOPE_QUADRILATERAL:
2719:       ct = DM_POLYTOPE_UNKNOWN;
2720:       break;
2721:     case DM_POLYTOPE_TETRAHEDRON:
2722:       ct = DM_POLYTOPE_UNKNOWN;
2723:       break;
2724:     case DM_POLYTOPE_HEXAHEDRON:
2725:       ct = DM_POLYTOPE_UNKNOWN;
2726:       break;
2727:     default:
2728:       ct = DM_POLYTOPE_UNKNOWN;
2729:     }
2730:     break;
2731:   case DM_POLYTOPE_TETRAHEDRON:
2732:     switch (ct2) {
2733:     case DM_POLYTOPE_POINT:
2734:       ct = ct1;
2735:       break;
2736:     case DM_POLYTOPE_SEGMENT:
2737:       ct = DM_POLYTOPE_UNKNOWN;
2738:       break;
2739:     case DM_POLYTOPE_TRIANGLE:
2740:       ct = DM_POLYTOPE_UNKNOWN;
2741:       break;
2742:     case DM_POLYTOPE_QUADRILATERAL:
2743:       ct = DM_POLYTOPE_UNKNOWN;
2744:       break;
2745:     case DM_POLYTOPE_TETRAHEDRON:
2746:       ct = DM_POLYTOPE_UNKNOWN;
2747:       break;
2748:     case DM_POLYTOPE_HEXAHEDRON:
2749:       ct = DM_POLYTOPE_UNKNOWN;
2750:       break;
2751:     default:
2752:       ct = DM_POLYTOPE_UNKNOWN;
2753:     }
2754:     break;
2755:   case DM_POLYTOPE_HEXAHEDRON:
2756:     switch (ct2) {
2757:     case DM_POLYTOPE_POINT:
2758:       ct = ct1;
2759:       break;
2760:     case DM_POLYTOPE_SEGMENT:
2761:       ct = DM_POLYTOPE_UNKNOWN;
2762:       break;
2763:     case DM_POLYTOPE_TRIANGLE:
2764:       ct = DM_POLYTOPE_UNKNOWN;
2765:       break;
2766:     case DM_POLYTOPE_QUADRILATERAL:
2767:       ct = DM_POLYTOPE_UNKNOWN;
2768:       break;
2769:     case DM_POLYTOPE_TETRAHEDRON:
2770:       ct = DM_POLYTOPE_UNKNOWN;
2771:       break;
2772:     case DM_POLYTOPE_HEXAHEDRON:
2773:       ct = DM_POLYTOPE_UNKNOWN;
2774:       break;
2775:     default:
2776:       ct = DM_POLYTOPE_UNKNOWN;
2777:     }
2778:     break;
2779:   default:
2780:     ct = DM_POLYTOPE_UNKNOWN;
2781:   }
2782:   dim   = dim1 + dim2;
2783:   Nc    = Nc1;
2784:   Np    = Np1 * Np2;
2785:   order = order1;
2786:   PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, q));
2787:   PetscCall(PetscQuadratureSetCellType(*q, ct));
2788:   PetscCall(PetscQuadratureSetOrder(*q, order));
2789:   PetscCall(PetscMalloc1(Np * dim, &x));
2790:   PetscCall(PetscMalloc1(Np, &w));
2791:   for (qa = 0, qc = 0; qa < Np1; ++qa) {
2792:     for (qb = 0; qb < Np2; ++qb, ++qc) {
2793:       for (d1 = 0, d = 0; d1 < dim1; ++d1, ++d) x[qc * dim + d] = x1[qa * dim1 + d1];
2794:       for (d2 = 0; d2 < dim2; ++d2, ++d) x[qc * dim + d] = x2[qb * dim2 + d2];
2795:       w[qc] = w1[qa] * w2[qb];
2796:     }
2797:   }
2798:   PetscCall(PetscQuadratureSetData(*q, dim, Nc, Np, x, w));
2799:   PetscFunctionReturn(PETSC_SUCCESS);
2800: }

2802: /* Overwrites A. Can only handle full-rank problems with m>=n
2803:    A in column-major format
2804:    Ainv in row-major format
2805:    tau has length m
2806:    worksize must be >= max(1,n)
2807:  */
2808: static PetscErrorCode PetscDTPseudoInverseQR(PetscInt m, PetscInt mstride, PetscInt n, PetscReal *A_in, PetscReal *Ainv_out, PetscScalar *tau, PetscInt worksize, PetscScalar *work)
2809: {
2810:   PetscBLASInt M, N, K, lda, ldb, ldwork;
2811:   PetscScalar *A, *Ainv, *R, *Q, Alpha;

2813:   PetscFunctionBegin;
2814: #if defined(PETSC_USE_COMPLEX)
2815:   {
2816:     PetscInt i, j;
2817:     PetscCall(PetscMalloc2(m * n, &A, m * n, &Ainv));
2818:     for (j = 0; j < n; j++) {
2819:       for (i = 0; i < m; i++) A[i + m * j] = A_in[i + mstride * j];
2820:     }
2821:     mstride = m;
2822:   }
2823: #else
2824:   A    = A_in;
2825:   Ainv = Ainv_out;
2826: #endif

2828:   PetscCall(PetscBLASIntCast(m, &M));
2829:   PetscCall(PetscBLASIntCast(n, &N));
2830:   PetscCall(PetscBLASIntCast(mstride, &lda));
2831:   PetscCall(PetscBLASIntCast(worksize, &ldwork));
2832:   PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF));
2833:   PetscCallLAPACKInfo("LAPACKgeqrf", LAPACKgeqrf_(&M, &N, A, &lda, tau, work, &ldwork, &info));
2834:   PetscCall(PetscFPTrapPop());
2835:   R = A; /* Upper triangular part of A now contains R, the rest contains the elementary reflectors */

2837:   /* Extract an explicit representation of Q */
2838:   Q = Ainv;
2839:   PetscCall(PetscArraycpy(Q, A, mstride * n));
2840:   K = N; /* full rank */
2841:   PetscCallLAPACKInfo("LAPACKorgqr", LAPACKorgqr_(&M, &N, &K, Q, &lda, tau, work, &ldwork, &info));

2843:   /* Compute A^{-T} = (R^{-1} Q^T)^T = Q R^{-T} */
2844:   Alpha = 1.0;
2845:   ldb   = lda;
2846:   PetscCallBLAS("BLAStrsm", BLAStrsm_("Right", "Upper", "ConjugateTranspose", "NotUnitTriangular", &M, &N, &Alpha, R, &lda, Q, &ldb));
2847:   /* Ainv is Q, overwritten with inverse */

2849: #if defined(PETSC_USE_COMPLEX)
2850:   {
2851:     PetscInt i;
2852:     for (i = 0; i < m * n; i++) Ainv_out[i] = PetscRealPart(Ainv[i]);
2853:     PetscCall(PetscFree2(A, Ainv));
2854:   }
2855: #endif
2856:   PetscFunctionReturn(PETSC_SUCCESS);
2857: }

2859: /* Computes integral of L_p' over intervals {(x0,x1),(x1,x2),...} */
2860: static PetscErrorCode PetscDTLegendreIntegrate(PetscInt ninterval, const PetscReal *x, PetscInt ndegree, const PetscInt *degrees, PetscBool Transpose, PetscReal *B)
2861: {
2862:   PetscReal *Bv;
2863:   PetscInt   i, j;

2865:   PetscFunctionBegin;
2866:   PetscCall(PetscMalloc1((ninterval + 1) * ndegree, &Bv));
2867:   /* Point evaluation of L_p on all the source vertices */
2868:   PetscCall(PetscDTLegendreEval(ninterval + 1, x, ndegree, degrees, Bv, NULL, NULL));
2869:   /* Integral over each interval: \int_a^b L_p' = L_p(b)-L_p(a) */
2870:   for (i = 0; i < ninterval; i++) {
2871:     for (j = 0; j < ndegree; j++) {
2872:       if (Transpose) B[i + ninterval * j] = Bv[(i + 1) * ndegree + j] - Bv[i * ndegree + j];
2873:       else B[i * ndegree + j] = Bv[(i + 1) * ndegree + j] - Bv[i * ndegree + j];
2874:     }
2875:   }
2876:   PetscCall(PetscFree(Bv));
2877:   PetscFunctionReturn(PETSC_SUCCESS);
2878: }

2880: /*@
2881:   PetscDTReconstructPoly - create matrix representing polynomial reconstruction using cell intervals and evaluation at target intervals

2883:   Not Collective

2885:   Input Parameters:
2886: + degree  - degree of reconstruction polynomial
2887: . nsource - number of source intervals
2888: . sourcex - sorted coordinates of source cell boundaries (length `nsource`+1)
2889: . ntarget - number of target intervals
2890: - targetx - sorted coordinates of target cell boundaries (length `ntarget`+1)

2892:   Output Parameter:
2893: . R - reconstruction matrix, utarget = sum_s R[t*nsource+s] * usource[s]

2895:   Level: advanced

2897: .seealso: `PetscDTLegendreEval()`
2898: @*/
2899: PetscErrorCode PetscDTReconstructPoly(PetscInt degree, PetscInt nsource, const PetscReal sourcex[], PetscInt ntarget, const PetscReal targetx[], PetscReal R[])
2900: {
2901:   PetscInt     i, j, k, *bdegrees, worksize;
2902:   PetscReal    xmin, xmax, center, hscale, *sourcey, *targety, *Bsource, *Bsinv, *Btarget;
2903:   PetscScalar *tau, *work;

2905:   PetscFunctionBegin;
2906:   PetscAssertPointer(sourcex, 3);
2907:   PetscAssertPointer(targetx, 5);
2908:   PetscAssertPointer(R, 6);
2909:   PetscCheck(degree < nsource, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Reconstruction degree %" PetscInt_FMT " must be less than number of source intervals %" PetscInt_FMT, degree, nsource);
2910:   if (PetscDefined(USE_DEBUG)) {
2911:     for (i = 0; i < nsource; i++) PetscCheck(sourcex[i] < sourcex[i + 1], PETSC_COMM_SELF, PETSC_ERR_ARG_CORRUPT, "Source interval %" PetscInt_FMT " has negative orientation (%g,%g)", i, (double)sourcex[i], (double)sourcex[i + 1]);
2912:     for (i = 0; i < ntarget; i++) PetscCheck(targetx[i] < targetx[i + 1], PETSC_COMM_SELF, PETSC_ERR_ARG_CORRUPT, "Target interval %" PetscInt_FMT " has negative orientation (%g,%g)", i, (double)targetx[i], (double)targetx[i + 1]);
2913:   }
2914:   xmin     = PetscMin(sourcex[0], targetx[0]);
2915:   xmax     = PetscMax(sourcex[nsource], targetx[ntarget]);
2916:   center   = (xmin + xmax) / 2;
2917:   hscale   = (xmax - xmin) / 2;
2918:   worksize = nsource;
2919:   PetscCall(PetscMalloc4(degree + 1, &bdegrees, nsource + 1, &sourcey, nsource * (degree + 1), &Bsource, worksize, &work));
2920:   PetscCall(PetscMalloc4(nsource, &tau, nsource * (degree + 1), &Bsinv, ntarget + 1, &targety, ntarget * (degree + 1), &Btarget));
2921:   for (i = 0; i <= nsource; i++) sourcey[i] = (sourcex[i] - center) / hscale;
2922:   for (i = 0; i <= degree; i++) bdegrees[i] = i + 1;
2923:   PetscCall(PetscDTLegendreIntegrate(nsource, sourcey, degree + 1, bdegrees, PETSC_TRUE, Bsource));
2924:   PetscCall(PetscDTPseudoInverseQR(nsource, nsource, degree + 1, Bsource, Bsinv, tau, nsource, work));
2925:   for (i = 0; i <= ntarget; i++) targety[i] = (targetx[i] - center) / hscale;
2926:   PetscCall(PetscDTLegendreIntegrate(ntarget, targety, degree + 1, bdegrees, PETSC_FALSE, Btarget));
2927:   for (i = 0; i < ntarget; i++) {
2928:     PetscReal rowsum = 0;
2929:     for (j = 0; j < nsource; j++) {
2930:       PetscReal sum = 0;
2931:       for (k = 0; k < degree + 1; k++) sum += Btarget[i * (degree + 1) + k] * Bsinv[k * nsource + j];
2932:       R[i * nsource + j] = sum;
2933:       rowsum += sum;
2934:     }
2935:     for (j = 0; j < nsource; j++) R[i * nsource + j] /= rowsum; /* normalize each row */
2936:   }
2937:   PetscCall(PetscFree4(bdegrees, sourcey, Bsource, work));
2938:   PetscCall(PetscFree4(tau, Bsinv, targety, Btarget));
2939:   PetscFunctionReturn(PETSC_SUCCESS);
2940: }

2942: /*@
2943:   PetscGaussLobattoLegendreIntegrate - Compute the L2 integral of a function on the GLL points

2945:   Not Collective

2947:   Input Parameters:
2948: + n       - the number of GLL nodes
2949: . nodes   - the GLL nodes
2950: . weights - the GLL weights
2951: - f       - the function values at the nodes

2953:   Output Parameter:
2954: . in - the value of the integral

2956:   Level: beginner

2958: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`
2959: @*/
2960: PetscErrorCode PetscGaussLobattoLegendreIntegrate(PetscInt n, PetscReal nodes[], PetscReal weights[], const PetscReal f[], PetscReal *in)
2961: {
2962:   PetscInt i;

2964:   PetscFunctionBegin;
2965:   *in = 0.;
2966:   for (i = 0; i < n; i++) *in += f[i] * f[i] * weights[i];
2967:   PetscFunctionReturn(PETSC_SUCCESS);
2968: }

2970: /*@C
2971:   PetscGaussLobattoLegendreElementLaplacianCreate - computes the Laplacian for a single 1d GLL element

2973:   Not Collective

2975:   Input Parameters:
2976: + n       - the number of GLL nodes
2977: . nodes   - the GLL nodes, of length `n`
2978: - weights - the GLL weights, of length `n`

2980:   Output Parameter:
2981: . AA - the stiffness element, of size `n` by `n`

2983:   Level: beginner

2985:   Notes:
2986:   Destroy this with `PetscGaussLobattoLegendreElementLaplacianDestroy()`

2988:   You can access entries in this array with AA[i][j] but in memory it is stored in contiguous memory, row-oriented (the array is symmetric)

2990: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementLaplacianDestroy()`
2991: @*/
2992: PetscErrorCode PetscGaussLobattoLegendreElementLaplacianCreate(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA)
2993: {
2994:   PetscReal      **A;
2995:   const PetscReal *gllnodes = nodes;
2996:   const PetscInt   p        = n - 1;
2997:   PetscReal        z0, z1, z2 = -1, x, Lpj, Lpr;
2998:   PetscInt         i, j, nn, r;

3000:   PetscFunctionBegin;
3001:   PetscCall(PetscMalloc1(n, &A));
3002:   PetscCall(PetscMalloc1(n * n, &A[0]));
3003:   for (i = 1; i < n; i++) A[i] = A[i - 1] + n;

3005:   for (j = 1; j < p; j++) {
3006:     x  = gllnodes[j];
3007:     z0 = 1.;
3008:     z1 = x;
3009:     for (nn = 1; nn < p; nn++) {
3010:       z2 = x * z1 * (2. * ((PetscReal)nn) + 1.) / (((PetscReal)nn) + 1.) - z0 * (((PetscReal)nn) / (((PetscReal)nn) + 1.));
3011:       z0 = z1;
3012:       z1 = z2;
3013:     }
3014:     Lpj = z2;
3015:     for (r = 1; r < p; r++) {
3016:       if (r == j) {
3017:         A[j][j] = 2. / (3. * (1. - gllnodes[j] * gllnodes[j]) * Lpj * Lpj);
3018:       } else {
3019:         x  = gllnodes[r];
3020:         z0 = 1.;
3021:         z1 = x;
3022:         for (nn = 1; nn < p; nn++) {
3023:           z2 = x * z1 * (2. * ((PetscReal)nn) + 1.) / (((PetscReal)nn) + 1.) - z0 * (((PetscReal)nn) / (((PetscReal)nn) + 1.));
3024:           z0 = z1;
3025:           z1 = z2;
3026:         }
3027:         Lpr     = z2;
3028:         A[r][j] = 4. / (((PetscReal)p) * (((PetscReal)p) + 1.) * Lpj * Lpr * (gllnodes[j] - gllnodes[r]) * (gllnodes[j] - gllnodes[r]));
3029:       }
3030:     }
3031:   }
3032:   for (j = 1; j < p + 1; j++) {
3033:     x  = gllnodes[j];
3034:     z0 = 1.;
3035:     z1 = x;
3036:     for (nn = 1; nn < p; nn++) {
3037:       z2 = x * z1 * (2. * ((PetscReal)nn) + 1.) / (((PetscReal)nn) + 1.) - z0 * (((PetscReal)nn) / (((PetscReal)nn) + 1.));
3038:       z0 = z1;
3039:       z1 = z2;
3040:     }
3041:     Lpj     = z2;
3042:     A[j][0] = 4. * PetscPowRealInt(-1., p) / (((PetscReal)p) * (((PetscReal)p) + 1.) * Lpj * (1. + gllnodes[j]) * (1. + gllnodes[j]));
3043:     A[0][j] = A[j][0];
3044:   }
3045:   for (j = 0; j < p; j++) {
3046:     x  = gllnodes[j];
3047:     z0 = 1.;
3048:     z1 = x;
3049:     for (nn = 1; nn < p; nn++) {
3050:       z2 = x * z1 * (2. * ((PetscReal)nn) + 1.) / (((PetscReal)nn) + 1.) - z0 * (((PetscReal)nn) / (((PetscReal)nn) + 1.));
3051:       z0 = z1;
3052:       z1 = z2;
3053:     }
3054:     Lpj = z2;

3056:     A[p][j] = 4. / (((PetscReal)p) * (((PetscReal)p) + 1.) * Lpj * (1. - gllnodes[j]) * (1. - gllnodes[j]));
3057:     A[j][p] = A[p][j];
3058:   }
3059:   A[0][0] = 0.5 + (((PetscReal)p) * (((PetscReal)p) + 1.) - 2.) / 6.;
3060:   A[p][p] = A[0][0];
3061:   *AA     = A;
3062:   PetscFunctionReturn(PETSC_SUCCESS);
3063: }

3065: /*@C
3066:   PetscGaussLobattoLegendreElementLaplacianDestroy - frees the Laplacian for a single 1d GLL element created with `PetscGaussLobattoLegendreElementLaplacianCreate()`

3068:   Not Collective

3070:   Input Parameters:
3071: + n       - the number of GLL nodes
3072: . nodes   - the GLL nodes, ignored
3073: . weights - the GLL weightss, ignored
3074: - AA      - the stiffness element from `PetscGaussLobattoLegendreElementLaplacianCreate()`

3076:   Level: beginner

3078: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementLaplacianCreate()`
3079: @*/
3080: PetscErrorCode PetscGaussLobattoLegendreElementLaplacianDestroy(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA)
3081: {
3082:   PetscFunctionBegin;
3083:   PetscCall(PetscFree((*AA)[0]));
3084:   PetscCall(PetscFree(*AA));
3085:   *AA = NULL;
3086:   PetscFunctionReturn(PETSC_SUCCESS);
3087: }

3089: /*@C
3090:   PetscGaussLobattoLegendreElementGradientCreate - computes the gradient for a single 1d GLL element

3092:   Not Collective

3094:   Input Parameters:
3095: + n       - the number of GLL nodes
3096: . nodes   - the GLL nodes, of length `n`
3097: - weights - the GLL weights, of length `n`

3099:   Output Parameters:
3100: + AA  - the stiffness element, of dimension `n` by `n`
3101: - AAT - the transpose of AA (pass in `NULL` if you do not need this array), of dimension `n` by `n`

3103:   Level: beginner

3105:   Notes:
3106:   Destroy this with `PetscGaussLobattoLegendreElementGradientDestroy()`

3108:   You can access entries in these arrays with AA[i][j] but in memory it is stored in contiguous memory, row-oriented

3110: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementLaplacianDestroy()`, `PetscGaussLobattoLegendreElementGradientDestroy()`
3111: @*/
3112: PetscErrorCode PetscGaussLobattoLegendreElementGradientCreate(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA, PetscReal ***AAT)
3113: {
3114:   PetscReal      **A, **AT = NULL;
3115:   const PetscReal *gllnodes = nodes;
3116:   const PetscInt   p        = n - 1;
3117:   PetscReal        Li, Lj, d0;
3118:   PetscInt         i, j;

3120:   PetscFunctionBegin;
3121:   PetscCall(PetscMalloc1(n, &A));
3122:   PetscCall(PetscMalloc1(n * n, &A[0]));
3123:   for (i = 1; i < n; i++) A[i] = A[i - 1] + n;

3125:   if (AAT) {
3126:     PetscCall(PetscMalloc1(n, &AT));
3127:     PetscCall(PetscMalloc1(n * n, &AT[0]));
3128:     for (i = 1; i < n; i++) AT[i] = AT[i - 1] + n;
3129:   }

3131:   if (n == 1) A[0][0] = 0.;
3132:   d0 = (PetscReal)p * ((PetscReal)p + 1.) / 4.;
3133:   for (i = 0; i < n; i++) {
3134:     for (j = 0; j < n; j++) {
3135:       A[i][j] = 0.;
3136:       PetscCall(PetscDTComputeJacobi(0., 0., p, gllnodes[i], &Li));
3137:       PetscCall(PetscDTComputeJacobi(0., 0., p, gllnodes[j], &Lj));
3138:       if (i != j) A[i][j] = Li / (Lj * (gllnodes[i] - gllnodes[j]));
3139:       if ((j == i) && (i == 0)) A[i][j] = -d0;
3140:       if (j == i && i == p) A[i][j] = d0;
3141:       if (AT) AT[j][i] = A[i][j];
3142:     }
3143:   }
3144:   if (AAT) *AAT = AT;
3145:   *AA = A;
3146:   PetscFunctionReturn(PETSC_SUCCESS);
3147: }

3149: /*@C
3150:   PetscGaussLobattoLegendreElementGradientDestroy - frees the gradient for a single 1d GLL element obtained with `PetscGaussLobattoLegendreElementGradientCreate()`

3152:   Not Collective

3154:   Input Parameters:
3155: + n       - the number of GLL nodes
3156: . nodes   - the GLL nodes, ignored
3157: . weights - the GLL weights, ignored
3158: . AA      - the stiffness element obtained with `PetscGaussLobattoLegendreElementGradientCreate()`
3159: - AAT     - the transpose of the element obtained with `PetscGaussLobattoLegendreElementGradientCreate()`

3161:   Level: beginner

3163: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementLaplacianCreate()`, `PetscGaussLobattoLegendreElementAdvectionCreate()`
3164: @*/
3165: PetscErrorCode PetscGaussLobattoLegendreElementGradientDestroy(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA, PetscReal ***AAT)
3166: {
3167:   PetscFunctionBegin;
3168:   PetscCall(PetscFree((*AA)[0]));
3169:   PetscCall(PetscFree(*AA));
3170:   *AA = NULL;
3171:   if (AAT) {
3172:     PetscCall(PetscFree((*AAT)[0]));
3173:     PetscCall(PetscFree(*AAT));
3174:     *AAT = NULL;
3175:   }
3176:   PetscFunctionReturn(PETSC_SUCCESS);
3177: }

3179: /*@C
3180:   PetscGaussLobattoLegendreElementAdvectionCreate - computes the advection operator for a single 1d GLL element

3182:   Not Collective

3184:   Input Parameters:
3185: + n       - the number of GLL nodes
3186: . nodes   - the GLL nodes, of length `n`
3187: - weights - the GLL weights, of length `n`

3189:   Output Parameter:
3190: . AA - the stiffness element, of dimension `n` by `n`

3192:   Level: beginner

3194:   Notes:
3195:   Destroy this with `PetscGaussLobattoLegendreElementAdvectionDestroy()`

3197:   This is the same as the Gradient operator multiplied by the diagonal mass matrix

3199:   You can access entries in this array with AA[i][j] but in memory it is stored in contiguous memory, row-oriented

3201: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementLaplacianCreate()`, `PetscGaussLobattoLegendreElementAdvectionDestroy()`
3202: @*/
3203: PetscErrorCode PetscGaussLobattoLegendreElementAdvectionCreate(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA)
3204: {
3205:   PetscReal      **D;
3206:   const PetscReal *gllweights = weights;
3207:   const PetscInt   glln       = n;
3208:   PetscInt         j;

3210:   PetscFunctionBegin;
3211:   PetscCall(PetscGaussLobattoLegendreElementGradientCreate(n, nodes, weights, &D, NULL));
3212:   for (PetscInt i = 0; i < glln; i++) {
3213:     for (j = 0; j < glln; j++) D[i][j] = gllweights[i] * D[i][j];
3214:   }
3215:   *AA = D;
3216:   PetscFunctionReturn(PETSC_SUCCESS);
3217: }

3219: /*@C
3220:   PetscGaussLobattoLegendreElementAdvectionDestroy - frees the advection stiffness for a single 1d GLL element created with `PetscGaussLobattoLegendreElementAdvectionCreate()`

3222:   Not Collective

3224:   Input Parameters:
3225: + n       - the number of GLL nodes
3226: . nodes   - the GLL nodes, ignored
3227: . weights - the GLL weights, ignored
3228: - AA      - advection obtained with `PetscGaussLobattoLegendreElementAdvectionCreate()`

3230:   Level: beginner

3232: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementAdvectionCreate()`
3233: @*/
3234: PetscErrorCode PetscGaussLobattoLegendreElementAdvectionDestroy(PetscInt n, PetscReal nodes[], PetscReal weights[], PetscReal ***AA)
3235: {
3236:   PetscFunctionBegin;
3237:   PetscCall(PetscFree((*AA)[0]));
3238:   PetscCall(PetscFree(*AA));
3239:   *AA = NULL;
3240:   PetscFunctionReturn(PETSC_SUCCESS);
3241: }

3243: /*@C
3244:   PetscGaussLobattoLegendreElementMassCreate - Build the elemental mass matrix for a single 1D Gauss-Lobatto-Legendre (GLL) spectral element

3246:   Not Collective; No Fortran Support

3248:   Input Parameters:
3249: + n       - number of GLL nodes
3250: . nodes   - the GLL quadrature nodes
3251: - weights - the GLL quadrature weights

3253:   Output Parameter:
3254: . AA - newly allocated `n` x `n` mass matrix as `PetscReal **`

3256:   Level: beginner

3258:   Note:
3259:   Free with `PetscGaussLobattoLegendreElementMassDestroy()`.

3261: .seealso: `PetscDTGaussLobattoLegendreQuadrature()`, `PetscGaussLobattoLegendreElementMassDestroy()`, `PetscGaussLobattoLegendreElementLaplacianCreate()`, `PetscGaussLobattoLegendreElementAdvectionCreate()`
3262: @*/
3263: PetscErrorCode PetscGaussLobattoLegendreElementMassCreate(PetscInt n, PetscReal *nodes, PetscReal *weights, PetscReal ***AA)
3264: {
3265:   PetscReal      **A;
3266:   const PetscReal *gllweights = weights;
3267:   const PetscInt   glln       = n;

3269:   PetscFunctionBegin;
3270:   PetscCall(PetscMalloc1(glln, &A));
3271:   PetscCall(PetscMalloc1(glln * glln, &A[0]));
3272:   for (PetscInt i = 1; i < glln; i++) A[i] = A[i - 1] + glln;
3273:   if (glln == 1) A[0][0] = 0.;
3274:   for (PetscInt i = 0; i < glln; i++) {
3275:     for (PetscInt j = 0; j < glln; j++) {
3276:       A[i][j] = 0.;
3277:       if (j == i) A[i][j] = gllweights[i];
3278:     }
3279:   }
3280:   *AA = A;
3281:   PetscFunctionReturn(PETSC_SUCCESS);
3282: }

3284: /*@C
3285:   PetscGaussLobattoLegendreElementMassDestroy - Free a 1D GLL elemental mass matrix created with `PetscGaussLobattoLegendreElementMassCreate()`

3287:   Not Collective; No Fortran Support

3289:   Input Parameters:
3290: + n       - number of GLL nodes (ignored)
3291: . nodes   - the GLL quadrature nodes (ignored)
3292: . weights - the GLL quadrature weights (ignored)
3293: - AA      - the mass matrix to free; `*AA` is set to `NULL` on return

3295:   Level: beginner

3297: .seealso: `PetscGaussLobattoLegendreElementMassCreate()`, `PetscDTGaussLobattoLegendreQuadrature()`
3298: @*/
3299: PetscErrorCode PetscGaussLobattoLegendreElementMassDestroy(PetscInt n, PetscReal *nodes, PetscReal *weights, PetscReal ***AA)
3300: {
3301:   PetscFunctionBegin;
3302:   PetscCall(PetscFree((*AA)[0]));
3303:   PetscCall(PetscFree(*AA));
3304:   *AA = NULL;
3305:   PetscFunctionReturn(PETSC_SUCCESS);
3306: }

3308: /*@
3309:   PetscDTIndexToBary - convert an index into a barycentric coordinate.

3311:   Input Parameters:
3312: + len   - the desired length of the barycentric tuple (usually 1 more than the dimension it represents, so a barycentric coordinate in a triangle has length 3)
3313: . sum   - the value that the sum of the barycentric coordinates (which will be non-negative integers) should sum to
3314: - index - the index to convert: should be >= 0 and < Binomial(len - 1 + sum, sum)

3316:   Output Parameter:
3317: . coord - will be filled with the barycentric coordinate, of length `n`

3319:   Level: beginner

3321:   Note:
3322:   The indices map to barycentric coordinates in lexicographic order, where the first index is the
3323:   least significant and the last index is the most significant.

3325: .seealso: `PetscDTBaryToIndex()`
3326: @*/
3327: PetscErrorCode PetscDTIndexToBary(PetscInt len, PetscInt sum, PetscInt index, PetscInt coord[])
3328: {
3329:   PetscInt c, d, s, total, subtotal, nexttotal;

3331:   PetscFunctionBeginHot;
3332:   PetscCheck(len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "length must be non-negative");
3333:   PetscCheck(index >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "index must be non-negative");
3334:   if (!len) {
3335:     PetscCheck(!sum && !index, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Invalid index or sum for length 0 barycentric coordinate");
3336:     PetscFunctionReturn(PETSC_SUCCESS);
3337:   }
3338:   for (c = 1, total = 1; c <= len; c++) {
3339:     /* total is the number of ways to have a tuple of length c with sum */
3340:     if (index < total) break;
3341:     total = (total * (sum + c)) / c;
3342:   }
3343:   PetscCheck(c <= len, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "index out of range");
3344:   for (d = c; d < len; d++) coord[d] = 0;
3345:   for (s = 0, subtotal = 1, nexttotal = 1; c > 0;) {
3346:     /* subtotal is the number of ways to have a tuple of length c with sum s */
3347:     /* nexttotal is the number of ways to have a tuple of length c-1 with sum s */
3348:     if ((index + subtotal) >= total) {
3349:       coord[--c] = sum - s;
3350:       index -= (total - subtotal);
3351:       sum       = s;
3352:       total     = nexttotal;
3353:       subtotal  = 1;
3354:       nexttotal = 1;
3355:       s         = 0;
3356:     } else {
3357:       subtotal  = (subtotal * (c + s)) / (s + 1);
3358:       nexttotal = (nexttotal * (c - 1 + s)) / (s + 1);
3359:       s++;
3360:     }
3361:   }
3362:   PetscFunctionReturn(PETSC_SUCCESS);
3363: }

3365: /*@
3366:   PetscDTBaryToIndex - convert a barycentric coordinate to an index

3368:   Input Parameters:
3369: + len   - the desired length of the barycentric tuple (usually 1 more than the dimension it represents, so a barycentric coordinate in a triangle has length 3)
3370: . sum   - the value that the sum of the barycentric coordinates (which will be non-negative integers) should sum to
3371: - coord - a barycentric coordinate with the given length `len` and `sum`

3373:   Output Parameter:
3374: . index - the unique index for the coordinate, >= 0 and < Binomial(len - 1 + sum, sum)

3376:   Level: beginner

3378:   Note:
3379:   The indices map to barycentric coordinates in lexicographic order, where the first index is the
3380:   least significant and the last index is the most significant.

3382: .seealso: `PetscDTIndexToBary`
3383: @*/
3384: PetscErrorCode PetscDTBaryToIndex(PetscInt len, PetscInt sum, const PetscInt coord[], PetscInt *index)
3385: {
3386:   PetscInt c;
3387:   PetscInt i;
3388:   PetscInt total;

3390:   PetscFunctionBeginHot;
3391:   PetscCheck(len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "length must be non-negative");
3392:   if (!len) {
3393:     PetscCheck(!sum, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Invalid index or sum for length 0 barycentric coordinate");
3394:     *index = 0;
3395:     PetscFunctionReturn(PETSC_SUCCESS);
3396:   }
3397:   for (c = 1, total = 1; c < len; c++) total = (total * (sum + c)) / c;
3398:   i = total - 1;
3399:   c = len - 1;
3400:   sum -= coord[c];
3401:   while (sum > 0) {
3402:     PetscInt subtotal;
3403:     PetscInt s;

3405:     for (s = 1, subtotal = 1; s < sum; s++) subtotal = (subtotal * (c + s)) / s;
3406:     i -= subtotal;
3407:     sum -= coord[--c];
3408:   }
3409:   *index = i;
3410:   PetscFunctionReturn(PETSC_SUCCESS);
3411: }

3413: /*@
3414:   PetscQuadratureComputePermutations - Compute permutations of quadrature points corresponding to domain orientations

3416:   Input Parameter:
3417: . quad - The `PetscQuadrature`

3419:   Output Parameters:
3420: + Np   - The number of domain orientations
3421: - perm - An array of `IS` permutations, one for ech orientation,

3423:   Level: developer

3425: .seealso: `PetscQuadratureSetCellType()`, `PetscQuadrature`
3426: @*/
3427: PetscErrorCode PetscQuadratureComputePermutations(PetscQuadrature quad, PeOp PetscInt *Np, IS *perm[])
3428: {
3429:   DMPolytopeType   ct;
3430:   const PetscReal *xq, *wq;
3431:   PetscInt         dim, qdim, d, Na, o, Nq, q, qp;

3433:   PetscFunctionBegin;
3434:   PetscCall(PetscQuadratureGetData(quad, &qdim, NULL, &Nq, &xq, &wq));
3435:   PetscCall(PetscQuadratureGetCellType(quad, &ct));
3436:   dim = DMPolytopeTypeGetDim(ct);
3437:   Na  = DMPolytopeTypeGetNumArrangements(ct);
3438:   PetscCall(PetscMalloc1(Na, perm));
3439:   if (Np) *Np = Na;
3440:   Na /= 2;
3441:   for (o = -Na; o < Na; ++o) {
3442:     DM        refdm;
3443:     PetscInt *idx;
3444:     PetscReal xi0[3] = {-1., -1., -1.}, v0[3], J[9], detJ, txq[3];
3445:     PetscBool flg;

3447:     PetscCall(DMPlexCreateReferenceCell(PETSC_COMM_SELF, ct, &refdm));
3448:     PetscCall(DMPlexOrientPoint(refdm, 0, o));
3449:     PetscCall(DMPlexComputeCellGeometryFEM(refdm, 0, NULL, v0, J, NULL, &detJ));
3450:     PetscCall(PetscMalloc1(Nq, &idx));
3451:     for (q = 0; q < Nq; ++q) {
3452:       CoordinatesRefToReal(dim, dim, xi0, v0, J, &xq[q * dim], txq);
3453:       for (qp = 0; qp < Nq; ++qp) {
3454:         PetscReal diff = 0.;

3456:         for (d = 0; d < dim; ++d) diff += PetscAbsReal(txq[d] - xq[qp * dim + d]);
3457:         if (diff < PETSC_SMALL) break;
3458:       }
3459:       PetscCheck(qp < Nq, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Transformed quad point %" PetscInt_FMT " does not match another quad point", q);
3460:       idx[q] = qp;
3461:     }
3462:     PetscCall(DMDestroy(&refdm));
3463:     PetscCall(ISCreateGeneral(PETSC_COMM_SELF, Nq, idx, PETSC_OWN_POINTER, &(*perm)[o + Na]));
3464:     PetscCall(ISGetInfo((*perm)[o + Na], IS_PERMUTATION, IS_LOCAL, PETSC_TRUE, &flg));
3465:     PetscCheck(flg, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Ordering for orientation %" PetscInt_FMT " was not a permutation", o);
3466:     PetscCall(ISSetPermutation((*perm)[o + Na]));
3467:   }
3468:   if (!Na) (*perm)[0] = NULL;
3469:   PetscFunctionReturn(PETSC_SUCCESS);
3470: }

3472: /*@
3473:   PetscDTCreateQuadratureByCell - Create default quadrature for a given cell

3475:   Not collective

3477:   Input Parameters:
3478: + ct     - The integration domain
3479: . qorder - The desired quadrature order
3480: - qtype  - The type of simplex quadrature, or PETSCDTSIMPLEXQUAD_DEFAULT

3482:   Output Parameters:
3483: + q  - The cell quadrature
3484: - fq - The face quadrature

3486:   Level: developer

3488: .seealso: `PetscDTCreateDefaultQuadrature()`, `PetscFECreateDefault()`, `PetscDTGaussTensorQuadrature()`, `PetscDTSimplexQuadrature()`, `PetscDTTensorQuadratureCreate()`
3489: @*/
3490: PetscErrorCode PetscDTCreateQuadratureByCell(DMPolytopeType ct, PetscInt qorder, PetscDTSimplexQuadratureType qtype, PetscQuadrature *q, PetscQuadrature *fq)
3491: {
3492:   const PetscInt quadPointsPerEdge = PetscMax(qorder + 1, 1);
3493:   const PetscInt dim               = DMPolytopeTypeGetDim(ct);

3495:   PetscFunctionBegin;
3496:   switch (ct) {
3497:   case DM_POLYTOPE_POINT: {
3498:     PetscReal *x, *w;

3500:     PetscCall(PetscQuadratureCreate(PETSC_COMM_SELF, q));
3501:     PetscCall(PetscQuadratureSetCellType(*q, ct));
3502:     PetscCall(PetscQuadratureSetOrder(*q, 0));
3503:     PetscCall(PetscMalloc1(1, &x));
3504:     PetscCall(PetscMalloc1(1, &w));
3505:     x[0] = 0.;
3506:     w[0] = 1.;
3507:     PetscCall(PetscQuadratureSetData(*q, dim, 1, 1, x, w));
3508:     *fq = NULL;
3509:   } break;
3510:   case DM_POLYTOPE_SEGMENT:
3511:   case DM_POLYTOPE_POINT_PRISM_TENSOR:
3512:   case DM_POLYTOPE_QUADRILATERAL:
3513:   case DM_POLYTOPE_SEG_PRISM_TENSOR:
3514:   case DM_POLYTOPE_HEXAHEDRON:
3515:   case DM_POLYTOPE_QUAD_PRISM_TENSOR:
3516:     PetscCall(PetscDTGaussTensorQuadrature(dim, 1, quadPointsPerEdge, -1.0, 1.0, q));
3517:     PetscCall(PetscDTGaussTensorQuadrature(dim - 1, 1, quadPointsPerEdge, -1.0, 1.0, fq));
3518:     break;
3519:   case DM_POLYTOPE_TRIANGLE:
3520:   case DM_POLYTOPE_TETRAHEDRON:
3521:     PetscCall(PetscDTSimplexQuadrature(dim, 2 * qorder, qtype, q));
3522:     PetscCall(PetscDTSimplexQuadrature(dim - 1, 2 * qorder, qtype, fq));
3523:     break;
3524:   case DM_POLYTOPE_TRI_PRISM:
3525:   case DM_POLYTOPE_TRI_PRISM_TENSOR: {
3526:     PetscQuadrature q1, q2;

3528:     // TODO: this should be able to use symmetric rules, but doing so causes tests to fail
3529:     PetscCall(PetscDTSimplexQuadrature(2, 2 * qorder, PETSCDTSIMPLEXQUAD_CONIC, &q1));
3530:     PetscCall(PetscDTGaussTensorQuadrature(1, 1, quadPointsPerEdge, -1.0, 1.0, &q2));
3531:     PetscCall(PetscDTTensorQuadratureCreate(q1, q2, q));
3532:     PetscCall(PetscQuadratureDestroy(&q2));
3533:     *fq = q1;
3534:     /* TODO Need separate quadratures for each face */
3535:   } break;
3536:   default:
3537:     SETERRQ(PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "No quadrature for celltype %s", DMPolytopeTypes[PetscMin(ct, DM_POLYTOPE_UNKNOWN)]);
3538:   }
3539:   PetscFunctionReturn(PETSC_SUCCESS);
3540: }

3542: /*@
3543:   PetscDTCreateDefaultQuadrature - Create default quadrature for a given cell

3545:   Not collective

3547:   Input Parameters:
3548: + ct     - The integration domain
3549: - qorder - The desired quadrature order

3551:   Output Parameters:
3552: + q  - The cell quadrature
3553: - fq - The face quadrature

3555:   Level: developer

3557: .seealso: `PetscDTCreateQuadratureByCell()`, `PetscFECreateDefault()`, `PetscDTGaussTensorQuadrature()`, `PetscDTSimplexQuadrature()`, `PetscDTTensorQuadratureCreate()`
3558: @*/
3559: PetscErrorCode PetscDTCreateDefaultQuadrature(DMPolytopeType ct, PetscInt qorder, PetscQuadrature *q, PetscQuadrature *fq)
3560: {
3561:   PetscFunctionBegin;
3562:   PetscCall(PetscDTCreateQuadratureByCell(ct, qorder, PETSCDTSIMPLEXQUAD_DEFAULT, q, fq));
3563:   PetscFunctionReturn(PETSC_SUCCESS);
3564: }