xref: /libCEED/examples/fluids/qfunctions/newtonian.h (revision 5c677226cd04abd1d571aee943392fe51960a7dc)
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 void FluxInviscid(NewtonianIdealGasContext gas, State s,
92                                         StateConservative Flux[3]) {
93   for (int i=0; i<3; i++) {
94     Flux[i].density = s.U.momentum[i];
95     for (int j=0; j<3; j++)
96       Flux[i].momentum[j] = s.U.momentum[i] * s.Y.velocity[j]
97                             + s.Y.pressure * (i == j);
98     Flux[i].E_total = (s.U.E_total + s.Y.pressure) * s.Y.velocity[i];
99   }
100 }
101 
102 CEED_QFUNCTION_HELPER void FluxInviscid_fwd(NewtonianIdealGasContext gas,
103     State s, State ds, StateConservative dFlux[3]) {
104   for (int i=0; i<3; i++) {
105     dFlux[i].density = ds.U.momentum[i];
106     for (int j=0; j<3; j++)
107       dFlux[i].momentum[j] = ds.U.momentum[i] * s.Y.velocity[j] +
108                              s.U.momentum[i] * ds.Y.velocity[j] + ds.Y.pressure * (i == j);
109     dFlux[i].E_total = (ds.U.E_total + ds.Y.pressure) * s.Y.velocity[i] +
110                        (s.U.E_total + s.Y.pressure) * ds.Y.velocity[i];
111   }
112 }
113 
114 // Kelvin-Mandel notation
115 CEED_QFUNCTION_HELPER void KMStrainRate(const State grad_s[3],
116                                         CeedScalar strain_rate[6]) {
117   const CeedScalar weight = 1 / sqrt(2.);
118   strain_rate[0] = grad_s[0].Y.velocity[0];
119   strain_rate[1] = grad_s[1].Y.velocity[1];
120   strain_rate[2] = grad_s[2].Y.velocity[2];
121   strain_rate[3] = weight * (grad_s[2].Y.velocity[1] + grad_s[1].Y.velocity[2]);
122   strain_rate[4] = weight * (grad_s[2].Y.velocity[0] + grad_s[0].Y.velocity[2]);
123   strain_rate[5] = weight * (grad_s[1].Y.velocity[0] + grad_s[0].Y.velocity[1]);
124 }
125 
126 CEED_QFUNCTION_HELPER void KMUnpack(const CeedScalar v[6], CeedScalar A[3][3]) {
127   const CeedScalar weight = 1 / sqrt(2.);
128   A[0][0] = v[0];
129   A[1][1] = v[1];
130   A[2][2] = v[2];
131   A[2][1] = A[1][2] = weight * v[3];
132   A[2][0] = A[0][2] = weight * v[4];
133   A[1][0] = A[0][1] = weight * v[5];
134 }
135 
136 CEED_QFUNCTION_HELPER void NewtonianStress(NewtonianIdealGasContext gas,
137     const CeedScalar strain_rate[6], CeedScalar stress[6]) {
138   CeedScalar div_u = strain_rate[0] + strain_rate[1] + strain_rate[2];
139   for (int i=0; i<6; i++) {
140     stress[i] = gas->mu * (2 * strain_rate[i] + gas->lambda * div_u * (i < 3));
141   }
142 }
143 
144 CEED_QFUNCTION_HELPER void ViscousEnergyFlux(NewtonianIdealGasContext gas,
145     StatePrimitive Y, const State grad_s[3], const CeedScalar stress[3][3],
146     CeedScalar Fe[3]) {
147   for (int i=0; i<3; i++) {
148     Fe[i] = - Y.velocity[0] * stress[0][i]
149             - Y.velocity[1] * stress[1][i]
150             - Y.velocity[2] * stress[2][i]
151             - gas->k * grad_s[i].Y.temperature;
152   }
153 }
154 
155 // *****************************************************************************
156 // Helper function for computing flux Jacobian
157 // *****************************************************************************
158 CEED_QFUNCTION_HELPER void computeFluxJacobian_NS(CeedScalar dF[3][5][5],
159     const CeedScalar rho, const CeedScalar u[3], const CeedScalar E,
160     const CeedScalar gamma, const CeedScalar g[3], const CeedScalar x[3]) {
161   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2]; // Velocity square
162   CeedScalar e_potential = -(g[0]*x[0] + g[1]*x[1] + g[2]*x[2]);
163   for (CeedInt i=0; i<3; i++) { // Jacobian matrices for 3 directions
164     for (CeedInt j=0; j<3; j++) { // Rows of each Jacobian matrix
165       dF[i][j+1][0] = ((i==j) ? ((gamma-1.)*(u_sq/2. - e_potential)) : 0.) -
166                       u[i]*u[j];
167       for (CeedInt k=0; k<3; k++) { // Columns of each Jacobian matrix
168         dF[i][0][k+1]   = ((i==k) ? 1. : 0.);
169         dF[i][j+1][k+1] = ((j==k) ? u[i] : 0.) +
170                           ((i==k) ? u[j] : 0.) -
171                           ((i==j) ? u[k] : 0.) * (gamma-1.);
172         dF[i][4][k+1]   = ((i==k) ? (E*gamma/rho - (gamma-1.)*u_sq/2.) : 0.) -
173                           (gamma-1.)*u[i]*u[k];
174       }
175       dF[i][j+1][4] = ((i==j) ? (gamma-1.) : 0.);
176     }
177     dF[i][4][0] = u[i] * ((gamma-1.)*u_sq - E*gamma/rho);
178     dF[i][4][4] = u[i] * gamma;
179   }
180 }
181 
182 // *****************************************************************************
183 // Helper function for computing flux Jacobian of Primitive variables
184 // *****************************************************************************
185 CEED_QFUNCTION_HELPER void computeFluxJacobian_NSp(CeedScalar dF[3][5][5],
186     const CeedScalar rho, const CeedScalar u[3], const CeedScalar E,
187     const CeedScalar Rd, const CeedScalar cv) {
188   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2]; // Velocity square
189   // TODO Add in gravity's contribution
190 
191   CeedScalar T    = ( E / rho - u_sq / 2. ) / cv;
192   CeedScalar drdT = -rho / T;
193   CeedScalar drdP = 1. / ( Rd * T);
194   CeedScalar etot =  E / rho ;
195   CeedScalar e2p  = drdP * etot + 1. ;
196   CeedScalar e3p  = ( E  + rho * Rd * T );
197   CeedScalar e4p  = drdT * etot + rho * cv ;
198 
199   for (CeedInt i=0; i<3; i++) { // Jacobian matrices for 3 directions
200     for (CeedInt j=0; j<3; j++) { // j counts F^{m_j}
201 //        [row][col] of A_i
202       dF[i][j+1][0] = drdP * u[i] * u[j] + ((i==j) ? 1. : 0.); // F^{{m_j} wrt p
203       for (CeedInt k=0; k<3; k++) { // k counts the wrt vel_k
204         dF[i][0][k+1]   =  ((i==k) ? rho  : 0.);   // F^c wrt u_k
205         dF[i][j+1][k+1] = (((j==k) ? u[i] : 0.) +  // F^m_j wrt u_k
206                            ((i==k) ? u[j] : 0.) ) * rho;
207         dF[i][4][k+1]   = rho * u[i] * u[k]
208                           + ((i==k) ? e3p  : 0.) ; // F^e wrt u_k
209       }
210       dF[i][j+1][4] = drdT * u[i] * u[j]; // F^{m_j} wrt T
211     }
212     dF[i][4][0] = u[i] * e2p; // F^e wrt p
213     dF[i][4][4] = u[i] * e4p; // F^e wrt T
214     dF[i][0][0] = u[i] * drdP; // F^c wrt p
215     dF[i][0][4] = u[i] * drdT; // F^c wrt T
216   }
217 }
218 
219 CEED_QFUNCTION_HELPER void PrimitiveToConservative_fwd(const CeedScalar rho,
220     const CeedScalar u[3], const CeedScalar E, const CeedScalar Rd,
221     const CeedScalar cv, const CeedScalar dY[5], CeedScalar dU[5]) {
222   CeedScalar u_sq = u[0]*u[0] + u[1]*u[1] + u[2]*u[2];
223   CeedScalar T    = ( E / rho - u_sq / 2. ) / cv;
224   CeedScalar drdT = -rho / T;
225   CeedScalar drdP = 1. / ( Rd * T);
226   dU[0] = drdP * dY[0] + drdT * dY[4];
227   CeedScalar de_kinetic = 0;
228   for (int i=0; i<3; i++) {
229     dU[1+i] = dU[0] * u[i] + rho * dY[1+i];
230     de_kinetic += u[i] * dY[1+i];
231   }
232   dU[4] = rho * cv * dY[4] + dU[0] * cv * T // internal energy: rho * e
233           + rho * de_kinetic + .5 * dU[0] * u_sq; // kinetic energy: .5 * rho * |u|^2
234 }
235 
236 // *****************************************************************************
237 // Helper function for computing Tau elements (stabilization constant)
238 //   Model from:
239 //     PHASTA
240 //
241 //   Tau[i] = itau=0 which is diagonal-Shakib (3 values still but not spatial)
242 //
243 // Where NOT UPDATED YET
244 // *****************************************************************************
245 CEED_QFUNCTION_HELPER void Tau_diagPrim(CeedScalar Tau_d[3],
246                                         const CeedScalar dXdx[3][3], const CeedScalar u[3],
247                                         const CeedScalar cv, const NewtonianIdealGasContext newt_ctx,
248                                         const CeedScalar mu, const CeedScalar dt,
249                                         const CeedScalar rho) {
250   // Context
251   const CeedScalar Ctau_t = newt_ctx->Ctau_t;
252   const CeedScalar Ctau_v = newt_ctx->Ctau_v;
253   const CeedScalar Ctau_C = newt_ctx->Ctau_C;
254   const CeedScalar Ctau_M = newt_ctx->Ctau_M;
255   const CeedScalar Ctau_E = newt_ctx->Ctau_E;
256   CeedScalar gijd[6];
257   CeedScalar tau;
258   CeedScalar dts;
259   CeedScalar fact;
260 
261   //*INDENT-OFF*
262   gijd[0] =   dXdx[0][0] * dXdx[0][0]
263             + dXdx[1][0] * dXdx[1][0]
264             + dXdx[2][0] * dXdx[2][0];
265 
266   gijd[1] =   dXdx[0][0] * dXdx[0][1]
267             + dXdx[1][0] * dXdx[1][1]
268             + dXdx[2][0] * dXdx[2][1];
269 
270   gijd[2] =   dXdx[0][1] * dXdx[0][1]
271             + dXdx[1][1] * dXdx[1][1]
272             + dXdx[2][1] * dXdx[2][1];
273 
274   gijd[3] =   dXdx[0][0] * dXdx[0][2]
275             + dXdx[1][0] * dXdx[1][2]
276             + dXdx[2][0] * dXdx[2][2];
277 
278   gijd[4] =   dXdx[0][1] * dXdx[0][2]
279             + dXdx[1][1] * dXdx[1][2]
280             + dXdx[2][1] * dXdx[2][2];
281 
282   gijd[5] =   dXdx[0][2] * dXdx[0][2]
283             + dXdx[1][2] * dXdx[1][2]
284             + dXdx[2][2] * dXdx[2][2];
285   //*INDENT-ON*
286 
287   dts = Ctau_t / dt ;
288 
289   tau = rho*rho*((4. * dts * dts)
290                  + u[0] * ( u[0] * gijd[0] + 2. * ( u[1] * gijd[1] + u[2] * gijd[3]))
291                  + u[1] * ( u[1] * gijd[2] + 2. *   u[2] * gijd[4])
292                  + u[2] *   u[2] * gijd[5])
293         + Ctau_v* mu * mu *
294         (gijd[0]*gijd[0] + gijd[2]*gijd[2] + gijd[5]*gijd[5] +
295          + 2. * (gijd[1]*gijd[1] + gijd[3]*gijd[3] + gijd[4]*gijd[4]));
296 
297   fact=sqrt(tau);
298 
299   Tau_d[0] = Ctau_C * fact / (rho*(gijd[0] + gijd[2] + gijd[5]))*0.125;
300 
301   Tau_d[1] = Ctau_M / fact;
302   Tau_d[2] = Ctau_E / ( fact * cv );
303 
304 // consider putting back the way I initially had it  Ctau_E * Tau_d[1] /cv
305 //  to avoid a division if the compiler is smart enough to see that cv IS
306 // a constant that it could invert once for all elements
307 // but in that case energy tau is scaled by the product of Ctau_E * Ctau_M
308 // OR we could absorb cv into Ctau_E but this puts more burden on user to
309 // know how to change constants with a change of fluid or units.  Same for
310 // Ctau_v * mu * mu IF AND ONLY IF we don't add viscosity law =f(T)
311 }
312 
313 // *****************************************************************************
314 // This QFunction sets a "still" initial condition for generic Newtonian IG problems
315 // *****************************************************************************
316 CEED_QFUNCTION(ICsNewtonianIG)(void *ctx, CeedInt Q,
317                                const CeedScalar *const *in, CeedScalar *const *out) {
318   // Inputs
319   const CeedScalar (*X)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0];
320 
321   // Outputs
322   CeedScalar (*q0)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0];
323 
324   // Context
325   const SetupContext context = (SetupContext)ctx;
326   const CeedScalar theta0    = context->theta0;
327   const CeedScalar P0        = context->P0;
328   const CeedScalar cv        = context->cv;
329   const CeedScalar cp        = context->cp;
330   const CeedScalar *g        = context->g;
331   const CeedScalar Rd        = cp - cv;
332 
333   // Quadrature Point Loop
334   CeedPragmaSIMD
335   for (CeedInt i=0; i<Q; i++) {
336     CeedScalar q[5] = {0.};
337 
338     // Setup
339     // -- Coordinates
340     const CeedScalar x[3] = {X[0][i], X[1][i], X[2][i]};
341     const CeedScalar e_potential = -(g[0]*x[0] + g[1]*x[1] + g[2]*x[2]);
342 
343     // -- Density
344     const CeedScalar rho = P0 / (Rd*theta0);
345 
346     // Initial Conditions
347     q[0] = rho;
348     q[1] = 0.0;
349     q[2] = 0.0;
350     q[3] = 0.0;
351     q[4] = rho * (cv*theta0 + e_potential);
352 
353     for (CeedInt j=0; j<5; j++)
354       q0[j][i] = q[j];
355   } // End of Quadrature Point Loop
356   return 0;
357 }
358 
359 // *****************************************************************************
360 // This QFunction implements the following formulation of Navier-Stokes with
361 //   explicit time stepping method
362 //
363 // This is 3D compressible Navier-Stokes in conservation form with state
364 //   variables of density, momentum density, and total energy density.
365 //
366 // State Variables: q = ( rho, U1, U2, U3, E )
367 //   rho - Mass Density
368 //   Ui  - Momentum Density,      Ui = rho ui
369 //   E   - Total Energy Density,  E  = rho (cv T + (u u)/2 + g z)
370 //
371 // Navier-Stokes Equations:
372 //   drho/dt + div( U )                               = 0
373 //   dU/dt   + div( rho (u x u) + P I3 ) + rho g khat = div( Fu )
374 //   dE/dt   + div( (E + P) u )                       = div( Fe )
375 //
376 // Viscous Stress:
377 //   Fu = mu (grad( u ) + grad( u )^T + lambda div ( u ) I3)
378 //
379 // Thermal Stress:
380 //   Fe = u Fu + k grad( T )
381 // Equation of State
382 //   P = (gamma - 1) (E - rho (u u) / 2 - rho g z)
383 //
384 // Stabilization:
385 //   Tau = diag(TauC, TauM, TauM, TauM, TauE)
386 //     f1 = rho  sqrt(ui uj gij)
387 //     gij = dXi/dX * dXi/dX
388 //     TauC = Cc f1 / (8 gii)
389 //     TauM = min( 1 , 1 / f1 )
390 //     TauE = TauM / (Ce cv)
391 //
392 //  SU   = Galerkin + grad(v) . ( Ai^T * Tau * (Aj q,j) )
393 //
394 // Constants:
395 //   lambda = - 2 / 3,  From Stokes hypothesis
396 //   mu              ,  Dynamic viscosity
397 //   k               ,  Thermal conductivity
398 //   cv              ,  Specific heat, constant volume
399 //   cp              ,  Specific heat, constant pressure
400 //   g               ,  Gravity
401 //   gamma  = cp / cv,  Specific heat ratio
402 //
403 // We require the product of the inverse of the Jacobian (dXdx_j,k) and
404 // its transpose (dXdx_k,j) to properly compute integrals of the form:
405 // int( gradv gradu )
406 //
407 // *****************************************************************************
408 CEED_QFUNCTION(RHSFunction_Newtonian)(void *ctx, CeedInt Q,
409                                       const CeedScalar *const *in, CeedScalar *const *out) {
410   // *INDENT-OFF*
411   // Inputs
412   const CeedScalar (*q)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0],
413                    (*dq)[5][CEED_Q_VLA] = (const CeedScalar(*)[5][CEED_Q_VLA])in[1],
414                    (*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2],
415                    (*x)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[3];
416   // Outputs
417   CeedScalar (*v)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0],
418              (*dv)[5][CEED_Q_VLA] = (CeedScalar(*)[5][CEED_Q_VLA])out[1];
419   // *INDENT-ON*
420 
421   // Context
422   NewtonianIdealGasContext context = (NewtonianIdealGasContext)ctx;
423   const CeedScalar mu     = context->mu;
424   const CeedScalar cv     = context->cv;
425   const CeedScalar cp     = context->cp;
426   const CeedScalar *g     = context->g;
427   const CeedScalar dt     = context->dt;
428   const CeedScalar gamma  = cp / cv;
429   const CeedScalar Rd     = cp - cv;
430 
431   CeedPragmaSIMD
432   // Quadrature Point Loop
433   for (CeedInt i=0; i<Q; i++) {
434     CeedScalar U[5];
435     for (int j=0; j<5; j++) U[j] = q[j][i];
436     const CeedScalar x_i[3] = {x[0][i], x[1][i], x[2][i]};
437     State s = StateFromU(context, U, x_i);
438 
439     // -- Interp-to-Interp q_data
440     const CeedScalar wdetJ      =   q_data[0][i];
441     // -- Interp-to-Grad q_data
442     // ---- Inverse of change of coordinate matrix: X_i,j
443     // *INDENT-OFF*
444     const CeedScalar dXdx[3][3] = {{q_data[1][i],
445                                     q_data[2][i],
446                                     q_data[3][i]},
447                                    {q_data[4][i],
448                                     q_data[5][i],
449                                     q_data[6][i]},
450                                    {q_data[7][i],
451                                     q_data[8][i],
452                                     q_data[9][i]}
453                                   };
454     // *INDENT-ON*
455 
456     State grad_s[3];
457     for (int j=0; j<3; j++) {
458       CeedScalar dx_i[3] = {0};
459       grad_s[j].U.density = dq[0][0][i] * dXdx[0][j]
460                             + dq[1][0][i] * dXdx[1][j] + dq[2][0][i] * dXdx[2][j];
461       for (int k=0; k<3; k++) grad_s[j].U.momentum[k] = dq[0][k+1][i] * dXdx[0][j]
462             + dq[1][k+1][i] * dXdx[1][j] + dq[2][k+1][i] * dXdx[2][j];
463       grad_s[j].U.E_total = dq[0][4][i] * dXdx[0][j] + dq[1][4][i] * dXdx[1][j] +
464                             dq[2][4][i] * dXdx[2][j];
465       dx_i[j] = 1.;
466       grad_s[j].Y = StatePrimitiveFromConservative_fwd(context, s, grad_s[j].U,
467                     x_i, dx_i);
468     }
469 
470     CeedScalar strain_rate[6], kmstress[6], stress[3][3], Fe[3];
471     KMStrainRate(grad_s, strain_rate);
472     NewtonianStress(context, strain_rate, kmstress);
473     KMUnpack(kmstress, stress);
474     ViscousEnergyFlux(context, s.Y, grad_s, stress, Fe);
475 
476     StateConservative F_inviscid[3];
477     FluxInviscid(context, s, F_inviscid);
478 
479     // Total flux
480     CeedScalar Flux[5][3];
481     for (int j=0; j<3; j++) {
482       Flux[0][j] = F_inviscid[j].density;
483       for (int k=0; k<3; k++)
484         Flux[k+1][j] = F_inviscid[j].momentum[k] - stress[k][j];
485       Flux[4][j] = F_inviscid[j].E_total + Fe[j];
486     }
487 
488     for (int j=0; j<3; j++) {
489       for (int k=0; k<5; k++) {
490         dv[j][k][i] = wdetJ * (dXdx[j][0] * Flux[k][0] +
491                                dXdx[j][1] * Flux[k][1] +
492                                dXdx[j][2] * Flux[k][2]);
493       }
494     }
495 
496     const CeedScalar body_force[5] = {0, s.U.density *g[0], s.U.density *g[1], s.U.density *g[2], 0};
497     for (int j=0; j<5; j++)
498       v[j][i] = wdetJ * body_force[j];
499 
500     // jacob_F_conv[3][5][5] = dF(convective)/dq at each direction
501     CeedScalar jacob_F_conv[3][5][5] = {0};
502     computeFluxJacobian_NS(jacob_F_conv, s.U.density, s.Y.velocity, s.U.E_total,
503                            gamma, g, x_i);
504     CeedScalar grad_U[5][3];
505     for (int j=0; j<3; j++) {
506       grad_U[0][j] = grad_s[j].U.density;
507       for (int k=0; k<3; k++) grad_U[k+1][j] = grad_s[j].U.momentum[k];
508       grad_U[4][j] = grad_s[j].U.E_total;
509     }
510 
511     // strong_conv = dF/dq * dq/dx    (Strong convection)
512     CeedScalar strong_conv[5] = {0};
513     for (int j=0; j<3; j++)
514       for (int k=0; k<5; k++)
515         for (int l=0; l<5; l++)
516           strong_conv[k] += jacob_F_conv[j][k][l] * grad_U[l][j];
517 
518     // -- Stabilization method: none, SU, or SUPG
519     CeedScalar stab[5][3] = {{0.}};
520     CeedScalar tau_strong_conv[5] = {0.}, tau_strong_conv_conservative[5] = {0};
521     CeedScalar Tau_d[3] = {0.};
522     switch (context->stabilization) {
523     case STAB_NONE:        // Galerkin
524       break;
525     case STAB_SU:        // SU
526       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
527       tau_strong_conv[0] = Tau_d[0] * strong_conv[0];
528       tau_strong_conv[1] = Tau_d[1] * strong_conv[1];
529       tau_strong_conv[2] = Tau_d[1] * strong_conv[2];
530       tau_strong_conv[3] = Tau_d[1] * strong_conv[3];
531       tau_strong_conv[4] = Tau_d[2] * strong_conv[4];
532       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
533                                   tau_strong_conv,
534                                   tau_strong_conv_conservative);
535       for (int j=0; j<3; j++)
536         for (int k=0; k<5; k++)
537           for (int l=0; l<5; l++)
538             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_conv_conservative[l];
539 
540       for (int j=0; j<5; j++)
541         for (int k=0; k<3; k++)
542           dv[k][j][i] -= wdetJ*(stab[j][0] * dXdx[k][0] +
543                                 stab[j][1] * dXdx[k][1] +
544                                 stab[j][2] * dXdx[k][2]);
545       break;
546     case STAB_SUPG:        // SUPG is not implemented for explicit scheme
547       break;
548     }
549 
550   } // End Quadrature Point Loop
551 
552   // Return
553   return 0;
554 }
555 
556 // *****************************************************************************
557 // This QFunction implements the Navier-Stokes equations (mentioned above) with
558 //   implicit time stepping method
559 //
560 //  SU   = Galerkin + grad(v) . ( Ai^T * Tau * (Aj q,j) )
561 //  SUPG = Galerkin + grad(v) . ( Ai^T * Tau * (q_dot + Aj q,j - body force) )
562 //                                       (diffussive terms will be added later)
563 //
564 // *****************************************************************************
565 CEED_QFUNCTION(IFunction_Newtonian)(void *ctx, CeedInt Q,
566                                     const CeedScalar *const *in,
567                                     CeedScalar *const *out) {
568   // *INDENT-OFF*
569   // Inputs
570   const CeedScalar (*q)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0],
571                    (*dq)[5][CEED_Q_VLA] = (const CeedScalar(*)[5][CEED_Q_VLA])in[1],
572                    (*q_dot)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2],
573                    (*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[3],
574                    (*x)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[4];
575   // Outputs
576   CeedScalar (*v)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0],
577              (*dv)[5][CEED_Q_VLA] = (CeedScalar(*)[5][CEED_Q_VLA])out[1];
578   // *INDENT-ON*
579   // Context
580   NewtonianIdealGasContext context = (NewtonianIdealGasContext)ctx;
581   const CeedScalar mu     = context->mu;
582   const CeedScalar cv     = context->cv;
583   const CeedScalar cp     = context->cp;
584   const CeedScalar *g     = context->g;
585   const CeedScalar dt     = context->dt;
586   const CeedScalar gamma  = cp / cv;
587   const CeedScalar Rd     = cp-cv;
588 
589   CeedPragmaSIMD
590   // Quadrature Point Loop
591   for (CeedInt i=0; i<Q; i++) {
592     CeedScalar U[5];
593     for (int j=0; j<5; j++) U[j] = q[j][i];
594     const CeedScalar x_i[3] = {x[0][i], x[1][i], x[2][i]};
595     State s = StateFromU(context, U, x_i);
596 
597     // -- Interp-to-Interp q_data
598     const CeedScalar wdetJ      =   q_data[0][i];
599     // -- Interp-to-Grad q_data
600     // ---- Inverse of change of coordinate matrix: X_i,j
601     // *INDENT-OFF*
602     const CeedScalar dXdx[3][3] = {{q_data[1][i],
603                                     q_data[2][i],
604                                     q_data[3][i]},
605                                    {q_data[4][i],
606                                     q_data[5][i],
607                                     q_data[6][i]},
608                                    {q_data[7][i],
609                                     q_data[8][i],
610                                     q_data[9][i]}
611                                   };
612     // *INDENT-ON*
613     State grad_s[3];
614     for (int j=0; j<3; j++) {
615       CeedScalar dx_i[3];
616       grad_s[j].U.density = dq[0][0][i] * dXdx[0][j]
617                             + dq[1][0][i] * dXdx[1][j] + dq[2][0][i] * dXdx[2][j];
618       for (int k=0; k<3; k++) grad_s[j].U.momentum[k] = dq[0][k+1][i] * dXdx[0][j]
619             + dq[1][k+1][i] * dXdx[1][j] + dq[2][k+1][i] * dXdx[2][j];
620       grad_s[j].U.E_total = dq[0][4][i] * dXdx[0][j] + dq[1][4][i] * dXdx[1][j] +
621                             dq[2][4][i] * dXdx[2][j];
622       dx_i[j] = 1.;
623       grad_s[j].Y = StatePrimitiveFromConservative_fwd(context, s, grad_s[j].U,
624                     x_i, dx_i);
625     }
626 
627     CeedScalar strain_rate[6], kmstress[6], stress[3][3], Fe[3];
628     KMStrainRate(grad_s, strain_rate);
629     NewtonianStress(context, strain_rate, kmstress);
630     KMUnpack(kmstress, stress);
631     ViscousEnergyFlux(context, s.Y, grad_s, stress, Fe);
632 
633     StateConservative F_inviscid[3];
634     FluxInviscid(context, s, F_inviscid);
635 
636 
637     // Total flux
638     CeedScalar Flux[5][3];
639     for (int j=0; j<3; j++) {
640       Flux[0][j] = F_inviscid[j].density;
641       for (int k=0; k<3; k++)
642         Flux[k+1][j] = F_inviscid[j].momentum[k] - stress[k][j];
643       Flux[4][j] = F_inviscid[j].E_total + Fe[j];
644     }
645 
646     for (int j=0; j<3; j++) {
647       for (int k=0; k<5; k++) {
648         dv[j][k][i] = -wdetJ * (dXdx[j][0] * Flux[k][0] +
649                                 dXdx[j][1] * Flux[k][1] +
650                                 dXdx[j][2] * Flux[k][2]);
651       }
652     }
653 
654     const CeedScalar body_force[5] = {0, s.U.density *g[0], s.U.density *g[1], s.U.density *g[2], 0};
655     for (int j=0; j<5; j++)
656       v[j][i] = wdetJ * (q_dot[j][i] - body_force[j]);
657 
658     // jacob_F_conv[3][5][5] = dF(convective)/dq at each direction
659     CeedScalar jacob_F_conv[3][5][5] = {0};
660     computeFluxJacobian_NS(jacob_F_conv, s.U.density, s.Y.velocity, s.U.E_total,
661                            gamma, g, x_i);
662     CeedScalar grad_U[5][3];
663     for (int j=0; j<3; j++) {
664       grad_U[0][j] = grad_s[j].U.density;
665       for (int k=0; k<3; k++) grad_U[k+1][j] = grad_s[j].U.momentum[k];
666       grad_U[4][j] = grad_s[j].U.E_total;
667     }
668 
669     // strong_conv = dF/dq * dq/dx    (Strong convection)
670     CeedScalar strong_conv[5] = {0};
671     for (int j=0; j<3; j++)
672       for (int k=0; k<5; k++)
673         for (int l=0; l<5; l++)
674           strong_conv[k] += jacob_F_conv[j][k][l] * grad_U[l][j];
675 
676     // Strong residual
677     CeedScalar strong_res[5];
678     for (int j=0; j<5; j++)
679       strong_res[j] = q_dot[j][i] + strong_conv[j] - body_force[j];
680 
681     // -- Stabilization method: none, SU, or SUPG
682     CeedScalar stab[5][3] = {{0.}};
683     CeedScalar tau_strong_res[5] = {0.}, tau_strong_res_conservative[5] = {0};
684     CeedScalar tau_strong_conv[5] = {0.}, tau_strong_conv_conservative[5] = {0};
685     CeedScalar Tau_d[3] = {0.};
686     switch (context->stabilization) {
687     case STAB_NONE:        // Galerkin
688       break;
689     case STAB_SU:        // SU
690       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
691       tau_strong_conv[0] = Tau_d[0] * strong_conv[0];
692       tau_strong_conv[1] = Tau_d[1] * strong_conv[1];
693       tau_strong_conv[2] = Tau_d[1] * strong_conv[2];
694       tau_strong_conv[3] = Tau_d[1] * strong_conv[3];
695       tau_strong_conv[4] = Tau_d[2] * strong_conv[4];
696       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
697                                   tau_strong_conv, tau_strong_conv_conservative);
698       for (int j=0; j<3; j++)
699         for (int k=0; k<5; k++)
700           for (int l=0; l<5; l++)
701             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_conv_conservative[l];
702 
703       for (int j=0; j<5; j++)
704         for (int k=0; k<3; k++)
705           dv[k][j][i] += wdetJ*(stab[j][0] * dXdx[k][0] +
706                                 stab[j][1] * dXdx[k][1] +
707                                 stab[j][2] * dXdx[k][2]);
708       break;
709     case STAB_SUPG:        // SUPG
710       Tau_diagPrim(Tau_d, dXdx, s.Y.velocity, cv, context, mu, dt, s.U.density);
711       tau_strong_res[0] = Tau_d[0] * strong_res[0];
712       tau_strong_res[1] = Tau_d[1] * strong_res[1];
713       tau_strong_res[2] = Tau_d[1] * strong_res[2];
714       tau_strong_res[3] = Tau_d[1] * strong_res[3];
715       tau_strong_res[4] = Tau_d[2] * strong_res[4];
716 // Alternate route (useful later with primitive variable code)
717 // this function was verified against PHASTA for as IC that was as close as possible
718 //    computeFluxJacobian_NSp(jacob_F_conv_p, rho, u, E, Rd, cv);
719 // it has also been verified to compute a correct through the following
720 //   stab[k][j] += jacob_F_conv_p[j][k][l] * tau_strong_res[l] // flux Jacobian wrt primitive
721 // applied in the triple loop below
722 //  However, it is more flops than using the existing Jacobian wrt q after q_{,Y} viz
723       PrimitiveToConservative_fwd(s.U.density, s.Y.velocity, s.U.E_total, Rd, cv,
724                                   tau_strong_res, tau_strong_res_conservative);
725       for (int j=0; j<3; j++)
726         for (int k=0; k<5; k++)
727           for (int l=0; l<5; l++)
728             stab[k][j] += jacob_F_conv[j][k][l] * tau_strong_res_conservative[l];
729 
730       for (int j=0; j<5; j++)
731         for (int k=0; k<3; k++)
732           dv[k][j][i] += wdetJ*(stab[j][0] * dXdx[k][0] +
733                                 stab[j][1] * dXdx[k][1] +
734                                 stab[j][2] * dXdx[k][2]);
735       break;
736     }
737 
738   } // End Quadrature Point Loop
739 
740   // Return
741   return 0;
742 }
743 // *****************************************************************************
744 #endif // newtonian_h
745