xref: /libCEED/examples/fluids/qfunctions/newtonian.h (revision e71202f6ed1f7658a951fb96c00ca70137d558f9) !
1 // Copyright (c) 2017-2022, 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 /// Operator for Navier-Stokes example using PETSc
10 
11 
12 #ifndef newtonian_h
13 #define newtonian_h
14 
15 #include <math.h>
16 #include <ceed.h>
17 #include "newtonian_types.h"
18 
19 #ifndef M_PI
20 #define M_PI    3.14159265358979323846
21 #endif
22 
23 typedef struct {
24   CeedScalar pressure;
25   CeedScalar velocity[3];
26   CeedScalar temperature;
27 } StatePrimitive;
28 
29 typedef struct {
30   CeedScalar density;
31   CeedScalar momentum[3];
32   CeedScalar E_total;
33 } StateConservative;
34 
35 typedef struct {
36   StateConservative U;
37   StatePrimitive Y;
38 } State;
39 
40 CEED_QFUNCTION_HELPER CeedScalar Dot3(const CeedScalar u[3],
41                                       const CeedScalar v[3]) {
42   return u[0]*v[0] + u[1]*v[1] + u[2]*v[2];
43 }
44 
45 CEED_QFUNCTION_HELPER StatePrimitive StatePrimitiveFromConservative(
46   NewtonianIdealGasContext gas, StateConservative U, const CeedScalar x[3]) {
47   StatePrimitive Y;
48   for (int i=0; i<3; i++) Y.velocity[i] = U.momentum[i] / U.density;
49   CeedScalar e_kinetic = .5 * Dot3(Y.velocity, Y.velocity);
50   CeedScalar e_potential = -Dot3(gas->g, x);
51   CeedScalar e_total = U.E_total / U.density;
52   CeedScalar e_internal = e_total - e_kinetic - e_potential;
53   Y.temperature = e_internal / gas->cv;
54   Y.pressure = (gas->cp / gas->cv - 1) * U.density * e_internal;
55   return Y;
56 }
57 
58 CEED_QFUNCTION_HELPER StatePrimitive StatePrimitiveFromConservative_fwd(
59   NewtonianIdealGasContext gas, State s, StateConservative dU,
60   const CeedScalar x[3], const CeedScalar dx[3]) {
61   StatePrimitive dY;
62   for (int i=0; i<3; i++) {
63     dY.velocity[i] = (dU.momentum[i] - s.Y.velocity[i] * dU.density) / s.U.density;
64   }
65   CeedScalar e_kinetic = .5 * Dot3(s.Y.velocity, s.Y.velocity);
66   CeedScalar de_kinetic = Dot3(dY.velocity, s.Y.velocity);
67   CeedScalar e_potential = -Dot3(gas->g, x);
68   CeedScalar de_potential = -Dot3(gas->g, dx);
69   CeedScalar e_total = s.U.E_total / s.U.density;
70   CeedScalar de_total = (dU.E_total - e_total * dU.density) / s.U.density;
71   CeedScalar e_internal = e_total - e_kinetic - e_potential;
72   CeedScalar de_internal = de_total - de_kinetic - de_potential;
73   dY.temperature = de_internal / gas->cv;
74   dY.pressure = (gas->cp / gas->cv - 1)
75                 * (dU.density * e_internal + s.U.density * de_internal);
76   return dY;
77 }
78 
79 CEED_QFUNCTION_HELPER State StateFromU(NewtonianIdealGasContext gas,
80                                        const CeedScalar U[5], const CeedScalar x[3]) {
81   State s;
82   s.U.density = U[0];
83   s.U.momentum[0] = U[1];
84   s.U.momentum[1] = U[2];
85   s.U.momentum[2] = U[3];
86   s.U.E_total = U[4];
87   s.Y = StatePrimitiveFromConservative(gas, s.U, x);
88   return s;
89 }
90 
91 CEED_QFUNCTION_HELPER State StateFromU_fwd(NewtonianIdealGasContext gas,
92     State s, const CeedScalar dU[5],
93     const CeedScalar x[3], const CeedScalar dx[3]) {
94   State ds;
95   ds.U.density = dU[0];
96   ds.U.momentum[0] = dU[1];
97   ds.U.momentum[1] = dU[2];
98   ds.U.momentum[2] = dU[3];
99   ds.U.E_total = dU[4];
100   ds.Y = StatePrimitiveFromConservative_fwd(gas, s, ds.U, x, dx);
101   return ds;
102 }
103 
104 CEED_QFUNCTION_HELPER void FluxInviscid(NewtonianIdealGasContext gas, State s,
105                                         StateConservative Flux[3]) {
106   for (int i=0; i<3; i++) {
107     Flux[i].density = s.U.momentum[i];
108     for (int j=0; j<3; j++)
109       Flux[i].momentum[j] = s.U.momentum[i] * s.Y.velocity[j]
110                             + s.Y.pressure * (i == j);
111     Flux[i].E_total = (s.U.E_total + s.Y.pressure) * s.Y.velocity[i];
112   }
113 }
114 
115 CEED_QFUNCTION_HELPER void FluxInviscid_fwd(NewtonianIdealGasContext gas,
116     State s, State ds, StateConservative dFlux[3]) {
117   for (int i=0; i<3; i++) {
118     dFlux[i].density = ds.U.momentum[i];
119     for (int j=0; j<3; j++)
120       dFlux[i].momentum[j] = ds.U.momentum[i] * s.Y.velocity[j] +
121                              s.U.momentum[i] * ds.Y.velocity[j] + ds.Y.pressure * (i == j);
122     dFlux[i].E_total = (ds.U.E_total + ds.Y.pressure) * s.Y.velocity[i] +
123                        (s.U.E_total + s.Y.pressure) * ds.Y.velocity[i];
124   }
125 }
126 
127 // Kelvin-Mandel notation
128 CEED_QFUNCTION_HELPER void KMStrainRate(const State grad_s[3],
129                                         CeedScalar strain_rate[6]) {
130   const CeedScalar weight = 1 / sqrt(2.);
131   strain_rate[0] = grad_s[0].Y.velocity[0];
132   strain_rate[1] = grad_s[1].Y.velocity[1];
133   strain_rate[2] = grad_s[2].Y.velocity[2];
134   strain_rate[3] = weight * (grad_s[2].Y.velocity[1] + grad_s[1].Y.velocity[2]);
135   strain_rate[4] = weight * (grad_s[2].Y.velocity[0] + grad_s[0].Y.velocity[2]);
136   strain_rate[5] = weight * (grad_s[1].Y.velocity[0] + grad_s[0].Y.velocity[1]);
137 }
138 
139 CEED_QFUNCTION_HELPER void KMUnpack(const CeedScalar v[6], CeedScalar A[3][3]) {
140   const CeedScalar weight = 1 / sqrt(2.);
141   A[0][0] = v[0];
142   A[1][1] = v[1];
143   A[2][2] = v[2];
144   A[2][1] = A[1][2] = weight * v[3];
145   A[2][0] = A[0][2] = weight * v[4];
146   A[1][0] = A[0][1] = weight * v[5];
147 }
148 
149 CEED_QFUNCTION_HELPER void NewtonianStress(NewtonianIdealGasContext gas,
150     const CeedScalar strain_rate[6], CeedScalar stress[6]) {
151   CeedScalar div_u = strain_rate[0] + strain_rate[1] + strain_rate[2];
152   for (int i=0; i<6; i++) {
153     stress[i] = gas->mu * (2 * strain_rate[i] + gas->lambda * div_u * (i < 3));
154   }
155 }
156 
157 CEED_QFUNCTION_HELPER void ViscousEnergyFlux(NewtonianIdealGasContext gas,
158     StatePrimitive Y, const State grad_s[3], const CeedScalar stress[3][3],
159     CeedScalar Fe[3]) {
160   for (int i=0; i<3; i++) {
161     Fe[i] = - Y.velocity[0] * stress[0][i]
162             - Y.velocity[1] * stress[1][i]
163             - Y.velocity[2] * stress[2][i]
164             - gas->k * grad_s[i].Y.temperature;
165   }
166 }
167 
168 // *****************************************************************************
169 // Helper function for computing flux Jacobian
170 // *****************************************************************************
171 CEED_QFUNCTION_HELPER void computeFluxJacobian_NS(CeedScalar dF[3][5][5],
172     const CeedScalar rho, const CeedScalar u[3], const CeedScalar E,
173     const CeedScalar gamma, const CeedScalar g[3], const CeedScalar x[3]) {
174   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2]; // Velocity square
175   CeedScalar e_potential = -(g[0]*x[0] + g[1]*x[1] + g[2]*x[2]);
176   for (CeedInt i=0; i<3; i++) { // Jacobian matrices for 3 directions
177     for (CeedInt j=0; j<3; j++) { // Rows of each Jacobian matrix
178       dF[i][j+1][0] = ((i==j) ? ((gamma-1.)*(u_sq/2. - e_potential)) : 0.) -
179                       u[i]*u[j];
180       for (CeedInt k=0; k<3; k++) { // Columns of each Jacobian matrix
181         dF[i][0][k+1]   = ((i==k) ? 1. : 0.);
182         dF[i][j+1][k+1] = ((j==k) ? u[i] : 0.) +
183                           ((i==k) ? u[j] : 0.) -
184                           ((i==j) ? u[k] : 0.) * (gamma-1.);
185         dF[i][4][k+1]   = ((i==k) ? (E*gamma/rho - (gamma-1.)*u_sq/2.) : 0.) -
186                           (gamma-1.)*u[i]*u[k];
187       }
188       dF[i][j+1][4] = ((i==j) ? (gamma-1.) : 0.);
189     }
190     dF[i][4][0] = u[i] * ((gamma-1.)*u_sq - E*gamma/rho);
191     dF[i][4][4] = u[i] * gamma;
192   }
193 }
194 
195 // *****************************************************************************
196 // Helper function for computing flux Jacobian of Primitive variables
197 // *****************************************************************************
198 CEED_QFUNCTION_HELPER void computeFluxJacobian_NSp(CeedScalar dF[3][5][5],
199     const CeedScalar rho, const CeedScalar u[3], const CeedScalar E,
200     const CeedScalar Rd, const CeedScalar cv) {
201   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2]; // Velocity square
202   // TODO Add in gravity's contribution
203 
204   CeedScalar T    = ( E / rho - u_sq / 2. ) / cv;
205   CeedScalar drdT = -rho / T;
206   CeedScalar drdP = 1. / ( Rd * T);
207   CeedScalar etot =  E / rho ;
208   CeedScalar e2p  = drdP * etot + 1. ;
209   CeedScalar e3p  = ( E  + rho * Rd * T );
210   CeedScalar e4p  = drdT * etot + rho * cv ;
211 
212   for (CeedInt i=0; i<3; i++) { // Jacobian matrices for 3 directions
213     for (CeedInt j=0; j<3; j++) { // j counts F^{m_j}
214 //        [row][col] of A_i
215       dF[i][j+1][0] = drdP * u[i] * u[j] + ((i==j) ? 1. : 0.); // F^{{m_j} wrt p
216       for (CeedInt k=0; k<3; k++) { // k counts the wrt vel_k
217         dF[i][0][k+1]   =  ((i==k) ? rho  : 0.);   // F^c wrt u_k
218         dF[i][j+1][k+1] = (((j==k) ? u[i] : 0.) +  // F^m_j wrt u_k
219                            ((i==k) ? u[j] : 0.) ) * rho;
220         dF[i][4][k+1]   = rho * u[i] * u[k]
221                           + ((i==k) ? e3p  : 0.) ; // F^e wrt u_k
222       }
223       dF[i][j+1][4] = drdT * u[i] * u[j]; // F^{m_j} wrt T
224     }
225     dF[i][4][0] = u[i] * e2p; // F^e wrt p
226     dF[i][4][4] = u[i] * e4p; // F^e wrt T
227     dF[i][0][0] = u[i] * drdP; // F^c wrt p
228     dF[i][0][4] = u[i] * drdT; // F^c wrt T
229   }
230 }
231 
232 CEED_QFUNCTION_HELPER void PrimitiveToConservative_fwd(const CeedScalar rho,
233     const CeedScalar u[3], const CeedScalar E, const CeedScalar Rd,
234     const CeedScalar cv, const CeedScalar dY[5], CeedScalar dU[5]) {
235   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2];
236   CeedScalar T    = ( E / rho - u_sq / 2. ) / cv;
237   CeedScalar drdT = -rho / T;
238   CeedScalar drdP = 1. / ( Rd * T);
239   dU[0] = drdP * dY[0] + drdT * dY[4];
240   CeedScalar de_kinetic = 0;
241   for (int i=0; i<3; i++) {
242     dU[1+i] = dU[0] * u[i] + rho * dY[1+i];
243     de_kinetic += u[i] * dY[1+i];
244   }
245   dU[4] = rho * cv * dY[4] + dU[0] * cv * T // internal energy: rho * e
246           + rho * de_kinetic + .5 * dU[0] * u_sq; // kinetic energy: .5 * rho * |u|^2
247 }
248 
249 // *****************************************************************************
250 // Helper function for computing Tau elements (stabilization constant)
251 //   Model from:
252 //     PHASTA
253 //
254 //   Tau[i] = itau=0 which is diagonal-Shakib (3 values still but not spatial)
255 //
256 // Where NOT UPDATED YET
257 // *****************************************************************************
258 CEED_QFUNCTION_HELPER void Tau_diagPrim(CeedScalar Tau_d[3],
259                                         const CeedScalar dXdx[3][3], const CeedScalar u[3],
260                                         const CeedScalar cv, const NewtonianIdealGasContext newt_ctx,
261                                         const CeedScalar mu, const CeedScalar dt,
262                                         const CeedScalar rho) {
263   // Context
264   const CeedScalar Ctau_t = newt_ctx->Ctau_t;
265   const CeedScalar Ctau_v = newt_ctx->Ctau_v;
266   const CeedScalar Ctau_C = newt_ctx->Ctau_C;
267   const CeedScalar Ctau_M = newt_ctx->Ctau_M;
268   const CeedScalar Ctau_E = newt_ctx->Ctau_E;
269   CeedScalar gijd[6];
270   CeedScalar tau;
271   CeedScalar dts;
272   CeedScalar fact;
273 
274   //*INDENT-OFF*
275   gijd[0] =   dXdx[0][0] * dXdx[0][0]
276             + dXdx[1][0] * dXdx[1][0]
277             + dXdx[2][0] * dXdx[2][0];
278 
279   gijd[1] =   dXdx[0][0] * dXdx[0][1]
280             + dXdx[1][0] * dXdx[1][1]
281             + dXdx[2][0] * dXdx[2][1];
282 
283   gijd[2] =   dXdx[0][1] * dXdx[0][1]
284             + dXdx[1][1] * dXdx[1][1]
285             + dXdx[2][1] * dXdx[2][1];
286 
287   gijd[3] =   dXdx[0][0] * dXdx[0][2]
288             + dXdx[1][0] * dXdx[1][2]
289             + dXdx[2][0] * dXdx[2][2];
290 
291   gijd[4] =   dXdx[0][1] * dXdx[0][2]
292             + dXdx[1][1] * dXdx[1][2]
293             + dXdx[2][1] * dXdx[2][2];
294 
295   gijd[5] =   dXdx[0][2] * dXdx[0][2]
296             + dXdx[1][2] * dXdx[1][2]
297             + dXdx[2][2] * dXdx[2][2];
298   //*INDENT-ON*
299 
300   dts = Ctau_t / dt ;
301 
302   tau = rho*rho*((4. * dts * dts)
303                  + u[0] * ( u[0] * gijd[0] + 2. * ( u[1] * gijd[1] + u[2] * gijd[3]))
304                  + u[1] * ( u[1] * gijd[2] + 2. *   u[2] * gijd[4])
305                  + u[2] *   u[2] * gijd[5])
306         + Ctau_v* mu * mu *
307         (gijd[0]*gijd[0] + gijd[2]*gijd[2] + gijd[5]*gijd[5] +
308          + 2. * (gijd[1]*gijd[1] + gijd[3]*gijd[3] + gijd[4]*gijd[4]));
309 
310   fact=sqrt(tau);
311 
312   Tau_d[0] = Ctau_C * fact / (rho*(gijd[0] + gijd[2] + gijd[5]))*0.125;
313 
314   Tau_d[1] = Ctau_M / fact;
315   Tau_d[2] = Ctau_E / ( fact * cv );
316 
317 // consider putting back the way I initially had it  Ctau_E * Tau_d[1] /cv
318 //  to avoid a division if the compiler is smart enough to see that cv IS
319 // a constant that it could invert once for all elements
320 // but in that case energy tau is scaled by the product of Ctau_E * Ctau_M
321 // OR we could absorb cv into Ctau_E but this puts more burden on user to
322 // know how to change constants with a change of fluid or units.  Same for
323 // Ctau_v * mu * mu IF AND ONLY IF we don't add viscosity law =f(T)
324 }
325 
326 // *****************************************************************************
327 // This QFunction sets a "still" initial condition for generic Newtonian IG problems
328 // *****************************************************************************
329 CEED_QFUNCTION(ICsNewtonianIG)(void *ctx, CeedInt Q,
330                                const CeedScalar *const *in, CeedScalar *const *out) {
331   // Inputs
332   const CeedScalar (*X)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0];
333 
334   // Outputs
335   CeedScalar (*q0)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0];
336 
337   // Context
338   const SetupContext context = (SetupContext)ctx;
339   const CeedScalar theta0    = context->theta0;
340   const CeedScalar P0        = context->P0;
341   const CeedScalar cv        = context->cv;
342   const CeedScalar cp        = context->cp;
343   const CeedScalar *g        = context->g;
344   const CeedScalar Rd        = cp - cv;
345 
346   // Quadrature Point Loop
347   CeedPragmaSIMD
348   for (CeedInt i=0; i<Q; i++) {
349     CeedScalar q[5] = {0.};
350 
351     // Setup
352     // -- Coordinates
353     const CeedScalar x[3] = {X[0][i], X[1][i], X[2][i]};
354     const CeedScalar e_potential = -(g[0]*x[0] + g[1]*x[1] + g[2]*x[2]);
355 
356     // -- Density
357     const CeedScalar rho = P0 / (Rd*theta0);
358 
359     // Initial Conditions
360     q[0] = rho;
361     q[1] = 0.0;
362     q[2] = 0.0;
363     q[3] = 0.0;
364     q[4] = rho * (cv*theta0 + e_potential);
365 
366     for (CeedInt j=0; j<5; j++)
367       q0[j][i] = q[j];
368   } // End of Quadrature Point Loop
369   return 0;
370 }
371 
372 // *****************************************************************************
373 // This QFunction implements the following formulation of Navier-Stokes with
374 //   explicit time stepping method
375 //
376 // This is 3D compressible Navier-Stokes in conservation form with state
377 //   variables of density, momentum density, and total energy density.
378 //
379 // State Variables: q = ( rho, U1, U2, U3, E )
380 //   rho - Mass Density
381 //   Ui  - Momentum Density,      Ui = rho ui
382 //   E   - Total Energy Density,  E  = rho (cv T + (u u)/2 + g z)
383 //
384 // Navier-Stokes Equations:
385 //   drho/dt + div( U )                               = 0
386 //   dU/dt   + div( rho (u x u) + P I3 ) + rho g khat = div( Fu )
387 //   dE/dt   + div( (E + P) u )                       = div( Fe )
388 //
389 // Viscous Stress:
390 //   Fu = mu (grad( u ) + grad( u )^T + lambda div ( u ) I3)
391 //
392 // Thermal Stress:
393 //   Fe = u Fu + k grad( T )
394 // Equation of State
395 //   P = (gamma - 1) (E - rho (u u) / 2 - rho g z)
396 //
397 // Stabilization:
398 //   Tau = diag(TauC, TauM, TauM, TauM, TauE)
399 //     f1 = rho  sqrt(ui uj gij)
400 //     gij = dXi/dX * dXi/dX
401 //     TauC = Cc f1 / (8 gii)
402 //     TauM = min( 1 , 1 / f1 )
403 //     TauE = TauM / (Ce cv)
404 //
405 //  SU   = Galerkin + grad(v) . ( Ai^T * Tau * (Aj q,j) )
406 //
407 // Constants:
408 //   lambda = - 2 / 3,  From Stokes hypothesis
409 //   mu              ,  Dynamic viscosity
410 //   k               ,  Thermal conductivity
411 //   cv              ,  Specific heat, constant volume
412 //   cp              ,  Specific heat, constant pressure
413 //   g               ,  Gravity
414 //   gamma  = cp / cv,  Specific heat ratio
415 //
416 // We require the product of the inverse of the Jacobian (dXdx_j,k) and
417 // its transpose (dXdx_k,j) to properly compute integrals of the form:
418 // int( gradv gradu )
419 //
420 // *****************************************************************************
421 CEED_QFUNCTION(RHSFunction_Newtonian)(void *ctx, CeedInt Q,
422                                       const CeedScalar *const *in, CeedScalar *const *out) {
423   // *INDENT-OFF*
424   // Inputs
425   const CeedScalar (*q)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0],
426                    (*Grad_q)[5][CEED_Q_VLA] = (const CeedScalar(*)[5][CEED_Q_VLA])in[1],
427                    (*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2],
428                    (*x)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[3];
429   // Outputs
430   CeedScalar (*v)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0],
431              (*Grad_v)[5][CEED_Q_VLA] = (CeedScalar(*)[5][CEED_Q_VLA])out[1];
432   // *INDENT-ON*
433 
434   // Context
435   NewtonianIdealGasContext context = (NewtonianIdealGasContext)ctx;
436   const CeedScalar mu     = context->mu;
437   const CeedScalar cv     = context->cv;
438   const CeedScalar cp     = context->cp;
439   const CeedScalar *g     = context->g;
440   const CeedScalar dt     = context->dt;
441   const CeedScalar gamma  = cp / cv;
442   const CeedScalar Rd     = cp - cv;
443 
444   CeedPragmaSIMD
445   // Quadrature Point Loop
446   for (CeedInt i=0; i<Q; i++) {
447     CeedScalar U[5];
448     for (int j=0; j<5; j++) U[j] = q[j][i];
449     const CeedScalar x_i[3] = {x[0][i], x[1][i], x[2][i]};
450     State s = StateFromU(context, U, x_i);
451 
452     // -- Interp-to-Interp q_data
453     const CeedScalar wdetJ      =   q_data[0][i];
454     // -- Interp-to-Grad q_data
455     // ---- Inverse of change of coordinate matrix: X_i,j
456     // *INDENT-OFF*
457     const CeedScalar dXdx[3][3] = {{q_data[1][i],
458                                     q_data[2][i],
459                                     q_data[3][i]},
460                                    {q_data[4][i],
461                                     q_data[5][i],
462                                     q_data[6][i]},
463                                    {q_data[7][i],
464                                     q_data[8][i],
465                                     q_data[9][i]}
466                                   };
467     // *INDENT-ON*
468 
469     State grad_s[3];
470     for (int j=0; j<3; j++) {
471       CeedScalar dx_i[3] = {0}, dU[5];
472       for (int k=0; k<5; k++) dU[k] = Grad_q[0][k][i] * dXdx[0][j]
473                                         + Grad_q[1][k][i] * dXdx[1][j]
474                                         + Grad_q[2][k][i] * dXdx[2][j];
475       dx_i[j] = 1.;
476       grad_s[j] = StateFromU_fwd(context, s, dU, x_i, dx_i);
477     }
478 
479     CeedScalar strain_rate[6], kmstress[6], stress[3][3], Fe[3];
480     KMStrainRate(grad_s, strain_rate);
481     NewtonianStress(context, strain_rate, kmstress);
482     KMUnpack(kmstress, stress);
483     ViscousEnergyFlux(context, s.Y, grad_s, stress, Fe);
484 
485     StateConservative F_inviscid[3];
486     FluxInviscid(context, s, F_inviscid);
487 
488     // Total flux
489     CeedScalar Flux[5][3];
490     for (int j=0; j<3; j++) {
491       Flux[0][j] = F_inviscid[j].density;
492       for (int k=0; k<3; k++)
493         Flux[k+1][j] = F_inviscid[j].momentum[k] - stress[k][j];
494       Flux[4][j] = F_inviscid[j].E_total + Fe[j];
495     }
496 
497     for (int j=0; j<3; j++) {
498       for (int k=0; k<5; k++) {
499         Grad_v[j][k][i] = wdetJ * (dXdx[j][0] * Flux[k][0] +
500                                    dXdx[j][1] * Flux[k][1] +
501                                    dXdx[j][2] * Flux[k][2]);
502       }
503     }
504 
505     const CeedScalar body_force[5] = {0, s.U.density *g[0], s.U.density *g[1], s.U.density *g[2], 0};
506     for (int j=0; j<5; j++)
507       v[j][i] = wdetJ * body_force[j];
508 
509     // jacob_F_conv[3][5][5] = dF(convective)/dq at each direction
510     CeedScalar jacob_F_conv[3][5][5] = {0};
511     computeFluxJacobian_NS(jacob_F_conv, s.U.density, s.Y.velocity, s.U.E_total,
512                            gamma, g, x_i);
513     CeedScalar grad_U[5][3];
514     for (int j=0; j<3; j++) {
515       grad_U[0][j] = grad_s[j].U.density;
516       for (int k=0; k<3; k++) grad_U[k+1][j] = grad_s[j].U.momentum[k];
517       grad_U[4][j] = grad_s[j].U.E_total;
518     }
519 
520     // strong_conv = dF/dq * dq/dx    (Strong convection)
521     CeedScalar strong_conv[5] = {0};
522     for (int j=0; j<3; j++)
523       for (int k=0; k<5; k++)
524         for (int l=0; l<5; l++)
525           strong_conv[k] += jacob_F_conv[j][k][l] * grad_U[l][j];
526 
527     // -- Stabilization method: none, SU, or SUPG
528     CeedScalar stab[5][3] = {{0.}};
529     CeedScalar tau_strong_conv[5] = {0.}, tau_strong_conv_conservative[5] = {0};
530     CeedScalar Tau_d[3] = {0.};
531     switch (context->stabilization) {
532     case STAB_NONE:        // Galerkin
533       break;
534     case STAB_SU:        // SU
535       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
536       tau_strong_conv[0] = Tau_d[0] * strong_conv[0];
537       tau_strong_conv[1] = Tau_d[1] * strong_conv[1];
538       tau_strong_conv[2] = Tau_d[1] * strong_conv[2];
539       tau_strong_conv[3] = Tau_d[1] * strong_conv[3];
540       tau_strong_conv[4] = Tau_d[2] * strong_conv[4];
541       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
542                                   tau_strong_conv,
543                                   tau_strong_conv_conservative);
544       for (int j=0; j<3; j++)
545         for (int k=0; k<5; k++)
546           for (int l=0; l<5; l++)
547             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_conv_conservative[l];
548 
549       for (int j=0; j<5; j++)
550         for (int k=0; k<3; k++)
551           Grad_v[k][j][i] -= wdetJ*(stab[j][0] * dXdx[k][0] +
552                                     stab[j][1] * dXdx[k][1] +
553                                     stab[j][2] * dXdx[k][2]);
554       break;
555     case STAB_SUPG:        // SUPG is not implemented for explicit scheme
556       break;
557     }
558 
559   } // End Quadrature Point Loop
560 
561   // Return
562   return 0;
563 }
564 
565 // *****************************************************************************
566 // This QFunction implements the Navier-Stokes equations (mentioned above) with
567 //   implicit time stepping method
568 //
569 //  SU   = Galerkin + grad(v) . ( Ai^T * Tau * (Aj q,j) )
570 //  SUPG = Galerkin + grad(v) . ( Ai^T * Tau * (q_dot + Aj q,j - body force) )
571 //                                       (diffussive terms will be added later)
572 //
573 // *****************************************************************************
574 CEED_QFUNCTION(IFunction_Newtonian)(void *ctx, CeedInt Q,
575                                     const CeedScalar *const *in,
576                                     CeedScalar *const *out) {
577   // *INDENT-OFF*
578   // Inputs
579   const CeedScalar (*q)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0],
580                    (*Grad_q)[5][CEED_Q_VLA] = (const CeedScalar(*)[5][CEED_Q_VLA])in[1],
581                    (*q_dot)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2],
582                    (*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[3],
583                    (*x)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[4];
584   // Outputs
585   CeedScalar (*v)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0],
586              (*Grad_v)[5][CEED_Q_VLA] = (CeedScalar(*)[5][CEED_Q_VLA])out[1],
587              (*jac_data)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[2];
588   // *INDENT-ON*
589   // Context
590   NewtonianIdealGasContext context = (NewtonianIdealGasContext)ctx;
591   const CeedScalar mu     = context->mu;
592   const CeedScalar cv     = context->cv;
593   const CeedScalar cp     = context->cp;
594   const CeedScalar *g     = context->g;
595   const CeedScalar dt     = context->dt;
596   const CeedScalar gamma  = cp / cv;
597   const CeedScalar Rd     = cp-cv;
598 
599   CeedPragmaSIMD
600   // Quadrature Point Loop
601   for (CeedInt i=0; i<Q; i++) {
602     CeedScalar U[5];
603     for (int j=0; j<5; j++) U[j] = q[j][i];
604     const CeedScalar x_i[3] = {x[0][i], x[1][i], x[2][i]};
605     State s = StateFromU(context, U, x_i);
606 
607     // -- Interp-to-Interp q_data
608     const CeedScalar wdetJ      =   q_data[0][i];
609     // -- Interp-to-Grad q_data
610     // ---- Inverse of change of coordinate matrix: X_i,j
611     // *INDENT-OFF*
612     const CeedScalar dXdx[3][3] = {{q_data[1][i],
613                                     q_data[2][i],
614                                     q_data[3][i]},
615                                    {q_data[4][i],
616                                     q_data[5][i],
617                                     q_data[6][i]},
618                                    {q_data[7][i],
619                                     q_data[8][i],
620                                     q_data[9][i]}
621                                   };
622     // *INDENT-ON*
623     State grad_s[3];
624     for (int j=0; j<3; j++) {
625       CeedScalar dx_i[3] = {0}, dU[5];
626       for (int k=0; k<5; k++) dU[k] = Grad_q[0][k][i] * dXdx[0][j]
627                                         + Grad_q[1][k][i] * dXdx[1][j]
628                                         + Grad_q[2][k][i] * dXdx[2][j];
629       dx_i[j] = 1.;
630       grad_s[j] = StateFromU_fwd(context, s, dU, x_i, dx_i);
631     }
632 
633     CeedScalar strain_rate[6], kmstress[6], stress[3][3], Fe[3];
634     KMStrainRate(grad_s, strain_rate);
635     NewtonianStress(context, strain_rate, kmstress);
636     KMUnpack(kmstress, stress);
637     ViscousEnergyFlux(context, s.Y, grad_s, stress, Fe);
638 
639     StateConservative F_inviscid[3];
640     FluxInviscid(context, s, F_inviscid);
641 
642 
643     // Total flux
644     CeedScalar Flux[5][3];
645     for (int j=0; j<3; j++) {
646       Flux[0][j] = F_inviscid[j].density;
647       for (int k=0; k<3; k++)
648         Flux[k+1][j] = F_inviscid[j].momentum[k] - stress[k][j];
649       Flux[4][j] = F_inviscid[j].E_total + Fe[j];
650     }
651 
652     for (int j=0; j<3; j++) {
653       for (int k=0; k<5; k++) {
654         Grad_v[j][k][i] = -wdetJ * (dXdx[j][0] * Flux[k][0] +
655                                     dXdx[j][1] * Flux[k][1] +
656                                     dXdx[j][2] * Flux[k][2]);
657       }
658     }
659 
660     const CeedScalar body_force[5] = {0, s.U.density *g[0], s.U.density *g[1], s.U.density *g[2], 0};
661     for (int j=0; j<5; j++)
662       v[j][i] = wdetJ * (q_dot[j][i] - body_force[j]);
663 
664     // jacob_F_conv[3][5][5] = dF(convective)/dq at each direction
665     CeedScalar jacob_F_conv[3][5][5] = {0};
666     computeFluxJacobian_NS(jacob_F_conv, s.U.density, s.Y.velocity, s.U.E_total,
667                            gamma, g, x_i);
668     CeedScalar grad_U[5][3];
669     for (int j=0; j<3; j++) {
670       grad_U[0][j] = grad_s[j].U.density;
671       for (int k=0; k<3; k++) grad_U[k+1][j] = grad_s[j].U.momentum[k];
672       grad_U[4][j] = grad_s[j].U.E_total;
673     }
674 
675     // strong_conv = dF/dq * dq/dx    (Strong convection)
676     CeedScalar strong_conv[5] = {0};
677     for (int j=0; j<3; j++)
678       for (int k=0; k<5; k++)
679         for (int l=0; l<5; l++)
680           strong_conv[k] += jacob_F_conv[j][k][l] * grad_U[l][j];
681 
682     // Strong residual
683     CeedScalar strong_res[5];
684     for (int j=0; j<5; j++)
685       strong_res[j] = q_dot[j][i] + strong_conv[j] - body_force[j];
686 
687     // -- Stabilization method: none, SU, or SUPG
688     CeedScalar stab[5][3] = {{0.}};
689     CeedScalar tau_strong_res[5] = {0.}, tau_strong_res_conservative[5] = {0};
690     CeedScalar tau_strong_conv[5] = {0.}, tau_strong_conv_conservative[5] = {0};
691     CeedScalar Tau_d[3] = {0.};
692     switch (context->stabilization) {
693     case STAB_NONE:        // Galerkin
694       break;
695     case STAB_SU:        // SU
696       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
697       tau_strong_conv[0] = Tau_d[0] * strong_conv[0];
698       tau_strong_conv[1] = Tau_d[1] * strong_conv[1];
699       tau_strong_conv[2] = Tau_d[1] * strong_conv[2];
700       tau_strong_conv[3] = Tau_d[1] * strong_conv[3];
701       tau_strong_conv[4] = Tau_d[2] * strong_conv[4];
702       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
703                                   tau_strong_conv, tau_strong_conv_conservative);
704       for (int j=0; j<3; j++)
705         for (int k=0; k<5; k++)
706           for (int l=0; l<5; l++)
707             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_conv_conservative[l];
708 
709       for (int j=0; j<5; j++)
710         for (int k=0; k<3; k++)
711           Grad_v[k][j][i] += wdetJ*(stab[j][0] * dXdx[k][0] +
712                                     stab[j][1] * dXdx[k][1] +
713                                     stab[j][2] * dXdx[k][2]);
714       break;
715     case STAB_SUPG:        // SUPG
716       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
717       tau_strong_res[0] = Tau_d[0] * strong_res[0];
718       tau_strong_res[1] = Tau_d[1] * strong_res[1];
719       tau_strong_res[2] = Tau_d[1] * strong_res[2];
720       tau_strong_res[3] = Tau_d[1] * strong_res[3];
721       tau_strong_res[4] = Tau_d[2] * strong_res[4];
722 // Alternate route (useful later with primitive variable code)
723 // this function was verified against PHASTA for as IC that was as close as possible
724 //    computeFluxJacobian_NSp(jacob_F_conv_p, rho, u, E, Rd, cv);
725 // it has also been verified to compute a correct through the following
726 //   stab[k][j] += jacob_F_conv_p[j][k][l] * tau_strong_res[l] // flux Jacobian wrt primitive
727 // applied in the triple loop below
728 //  However, it is more flops than using the existing Jacobian wrt q after q_{,Y} viz
729       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
730                                   tau_strong_res, tau_strong_res_conservative);
731       for (int j=0; j<3; j++)
732         for (int k=0; k<5; k++)
733           for (int l=0; l<5; l++)
734             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_res_conservative[l];
735 
736       for (int j=0; j<5; j++)
737         for (int k=0; k<3; k++)
738           Grad_v[k][j][i] += wdetJ*(stab[j][0] * dXdx[k][0] +
739                                     stab[j][1] * dXdx[k][1] +
740                                     stab[j][2] * dXdx[k][2]);
741       break;
742     }
743     for (int j=0; j<5; j++) jac_data[j][i] = U[j];
744     for (int j=0; j<3; j++) jac_data[5+j][i] = Tau_d[j];
745 
746   } // End Quadrature Point Loop
747 
748   // Return
749   return 0;
750 }
751 // *****************************************************************************
752 #endif // newtonian_h
753