1 // Copyright (c) 2017-2026, Lawrence Livermore National Security, LLC and other CEED contributors.
2 // All Rights Reserved. See the top-level LICENSE and NOTICE files for details.
3 //
4 // SPDX-License-Identifier: BSD-2-Clause
5 //
6 // This file is part of CEED: http://github.com/ceed
7
8 /// @file
9 /// libCEED QFunctions for diffusion operator example for a scalar field on the sphere using PETSc
10
11 #include <ceed/types.h>
12 #ifndef CEED_RUNNING_JIT_PASS
13 #include <math.h>
14 #endif
15
16 // -----------------------------------------------------------------------------
17 // This QFunction sets up the geometric factors required for integration and coordinate transformations when reference coordinates have a different
18 // dimension than the one of physical coordinates
19 //
20 // Reference (parent) 2D coordinates: X \in [-1, 1]^2
21 //
22 // Global 3D physical coordinates given by the mesh: xx \in [-R, R]^3 with R radius of the sphere
23 //
24 // Local 3D physical coordinates on the 2D manifold: x \in [-l, l]^3 with l half edge of the cube inscribed in the sphere
25 //
26 // Change of coordinates matrix computed by the library:
27 // (physical 3D coords relative to reference 2D coords)
28 // dxx_j/dX_i (indicial notation) [3 * 2]
29 //
30 // Change of coordinates x (on the 2D manifold) relative to xx (phyisical 3D):
31 // dx_i/dxx_j (indicial notation) [3 * 3]
32 //
33 // Change of coordinates x (on the 2D manifold) relative to X (reference 2D):
34 // (by chain rule)
35 // dx_i/dX_j [3 * 2] = dx_i/dxx_k [3 * 3] * dxx_k/dX_j [3 * 2]
36 //
37 // mod_J is given by the magnitude of the cross product of the columns of dx_i/dX_j
38 //
39 // The quadrature data is stored in the array q_data.
40 //
41 // We require the determinant of the Jacobian to properly compute integrals of the form: int( u v )
42 //
43 // q_data[0]: mod_J * w
44 //
45 // We use the Moore–Penrose (left) pseudoinverse of dx_i/dX_j, to compute dX_i/dx_j (and its transpose), needed to properly compute integrals of the
46 // form: int( gradv gradu )
47 //
48 // dX_i/dx_j [2 * 3] = (dx_i/dX_j)+ = (dxdX^T dxdX)^(-1) dxdX
49 //
50 // and the product simplifies to yield the contravariant metric tensor
51 //
52 // g^{ij} = dX_i/dx_k dX_j/dx_k = (dxdX^T dxdX)^{-1}
53 //
54 // Stored: g^{ij} (in Voigt convention) in
55 //
56 // q_data[1:3]: [dXdxdXdxT00 dXdxdXdxT01]
57 // [dXdxdXdxT01 dXdxdXdxT11]
58 // -----------------------------------------------------------------------------
SetupDiffGeo(void * ctx,CeedInt Q,const CeedScalar * const * in,CeedScalar * const * out)59 CEED_QFUNCTION(SetupDiffGeo)(void *ctx, CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) {
60 const CeedScalar *X = in[0], *J = in[1], *w = in[2];
61 CeedScalar *q_data = out[0];
62
63 // Quadrature Point Loop
64 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) {
65 // Read global Cartesian coordinates
66 const CeedScalar xx[3] = {X[i + 0 * Q], X[i + 1 * Q], X[i + 2 * Q]};
67
68 // Read dxxdX Jacobian entries, stored as
69 // 0 3
70 // 1 4
71 // 2 5
72 const CeedScalar dxxdX[3][2] = {
73 {J[i + Q * 0], J[i + Q * 3]},
74 {J[i + Q * 1], J[i + Q * 4]},
75 {J[i + Q * 2], J[i + Q * 5]}
76 };
77
78 // Setup
79 // x = xx (xx^T xx)^{-1/2}
80 // dx/dxx = I (xx^T xx)^{-1/2} - xx xx^T (xx^T xx)^{-3/2}
81 const CeedScalar mod_xx_sq = xx[0] * xx[0] + xx[1] * xx[1] + xx[2] * xx[2];
82 CeedScalar xx_sq[3][3];
83 for (int j = 0; j < 3; j++) {
84 for (int k = 0; k < 3; k++) xx_sq[j][k] = xx[j] * xx[k] / (sqrt(mod_xx_sq) * mod_xx_sq);
85 }
86
87 const CeedScalar dxdxx[3][3] = {
88 {1. / sqrt(mod_xx_sq) - xx_sq[0][0], -xx_sq[0][1], -xx_sq[0][2] },
89 {-xx_sq[1][0], 1. / sqrt(mod_xx_sq) - xx_sq[1][1], -xx_sq[1][2] },
90 {-xx_sq[2][0], -xx_sq[2][1], 1. / sqrt(mod_xx_sq) - xx_sq[2][2]}
91 };
92
93 CeedScalar dxdX[3][2];
94 for (int j = 0; j < 3; j++) {
95 for (int k = 0; k < 2; k++) {
96 dxdX[j][k] = 0;
97 for (int l = 0; l < 3; l++) dxdX[j][k] += dxdxx[j][l] * dxxdX[l][k];
98 }
99 }
100
101 // J is given by the cross product of the columns of dxdX
102 const CeedScalar J[3] = {dxdX[1][0] * dxdX[2][1] - dxdX[2][0] * dxdX[1][1], dxdX[2][0] * dxdX[0][1] - dxdX[0][0] * dxdX[2][1],
103 dxdX[0][0] * dxdX[1][1] - dxdX[1][0] * dxdX[0][1]};
104
105 // Use the magnitude of J as our detJ (volume scaling factor)
106 const CeedScalar mod_J = sqrt(J[0] * J[0] + J[1] * J[1] + J[2] * J[2]);
107
108 // Interp-to-Interp q_data
109 q_data[i + Q * 0] = mod_J * w[i];
110
111 // dxdX_k,j * dxdX_j,k
112 CeedScalar dxdXTdxdX[2][2];
113 for (int j = 0; j < 2; j++) {
114 for (int k = 0; k < 2; k++) {
115 dxdXTdxdX[j][k] = 0;
116 for (int l = 0; l < 3; l++) dxdXTdxdX[j][k] += dxdX[l][j] * dxdX[l][k];
117 }
118 }
119
120 const CeedScalar detdxdXTdxdX = dxdXTdxdX[0][0] * dxdXTdxdX[1][1] - dxdXTdxdX[1][0] * dxdXTdxdX[0][1];
121
122 // Compute inverse of dxdXTdxdX, which is the 2x2 contravariant metric tensor g^{ij}
123 CeedScalar dxdXTdxdX_inv[2][2];
124 dxdXTdxdX_inv[0][0] = dxdXTdxdX[1][1] / detdxdXTdxdX;
125 dxdXTdxdX_inv[0][1] = -dxdXTdxdX[0][1] / detdxdXTdxdX;
126 dxdXTdxdX_inv[1][0] = -dxdXTdxdX[1][0] / detdxdXTdxdX;
127 dxdXTdxdX_inv[1][1] = dxdXTdxdX[0][0] / detdxdXTdxdX;
128
129 // Stored in Voigt convention
130 q_data[i + Q * 1] = dxdXTdxdX_inv[0][0];
131 q_data[i + Q * 2] = dxdXTdxdX_inv[1][1];
132 q_data[i + Q * 3] = dxdXTdxdX_inv[0][1];
133 } // End of Quadrature Point Loop
134
135 // Return
136 return 0;
137 }
138
139 // -----------------------------------------------------------------------------
140 // This QFunction sets up the rhs and true solution for the problem
141 // -----------------------------------------------------------------------------
SetupDiffRhs(void * ctx,CeedInt Q,const CeedScalar * const * in,CeedScalar * const * out)142 CEED_QFUNCTION(SetupDiffRhs)(void *ctx, CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) {
143 // Inputs
144 const CeedScalar *X = in[0], *q_data = in[1];
145 // Outputs
146 CeedScalar *true_soln = out[0], *rhs = out[1];
147
148 // Context
149 const CeedScalar *context = (const CeedScalar *)ctx;
150 const CeedScalar R = context[0];
151
152 // Quadrature Point Loop
153 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) {
154 // Read global Cartesian coordinates
155 CeedScalar x = X[i + Q * 0], y = X[i + Q * 1], z = X[i + Q * 2];
156 // Normalize quadrature point coordinates to sphere
157 CeedScalar rad = sqrt(x * x + y * y + z * z);
158 x *= R / rad;
159 y *= R / rad;
160 z *= R / rad;
161 // Compute latitude and longitude
162 const CeedScalar theta = asin(z / R); // latitude
163 const CeedScalar lambda = atan2(y, x); // longitude
164
165 true_soln[i + Q * 0] = sin(lambda) * cos(theta);
166
167 rhs[i + Q * 0] = q_data[i + Q * 0] * 2 * sin(lambda) * cos(theta) / (R * R);
168 } // End of Quadrature Point Loop
169
170 return 0;
171 }
172
173 // -----------------------------------------------------------------------------
174 // This QFunction applies the diffusion operator for a scalar field.
175 //
176 // Inputs:
177 // ug - Input vector gradient at quadrature points
178 // q_data - Geometric factors
179 //
180 // Output:
181 // vg - Output vector (test functions) gradient at quadrature points
182 // -----------------------------------------------------------------------------
Diff(void * ctx,CeedInt Q,const CeedScalar * const * in,CeedScalar * const * out)183 CEED_QFUNCTION(Diff)(void *ctx, CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) {
184 // Inputs
185 const CeedScalar *ug = in[0], *q_data = in[1];
186 // Outputs
187 CeedScalar *vg = out[0];
188
189 // Quadrature Point Loop
190 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) {
191 // Read spatial derivatives of u
192 const CeedScalar du[2] = {ug[i + Q * 0], ug[i + Q * 1]};
193 // Read q_data
194 const CeedScalar w_det_J = q_data[i + Q * 0];
195 // -- Grad-to-Grad q_data
196 // ---- dXdx_j,k * dXdx_k,j
197 const CeedScalar dXdxdXdx_T[2][2] = {
198 {q_data[i + Q * 1], q_data[i + Q * 3]},
199 {q_data[i + Q * 3], q_data[i + Q * 2]}
200 };
201
202 for (int j = 0; j < 2; j++) { // j = direction of vg
203 vg[i + j * Q] = w_det_J * (du[0] * dXdxdXdx_T[0][j] + du[1] * dXdxdXdx_T[1][j]);
204 }
205 } // End of Quadrature Point Loop
206
207 return 0;
208 }
209 // -----------------------------------------------------------------------------
210