void
QOP_asqtad_force_multi_asvec_qdp(QOP_info_t *info, QDP_ColorMatrix *links[],
QDP_ColorMatrix *force[], QOP_asqtad_coeffs_t *coef,
REAL eps[], QDP_ColorVector *xin[], int nsrc)
{
#define NC QDP_get_nc(xin[0])
REAL coeff[nsrc];
REAL OneLink[nsrc], Lepage[nsrc], Naik[nsrc], FiveSt[nsrc], ThreeSt[nsrc], SevenSt[nsrc];
REAL mNaik[nsrc], mLepage[nsrc], mFiveSt[nsrc], mThreeSt[nsrc], mSevenSt[nsrc];
QDP_ColorVector *P3[8][nsrc];
QDP_ColorVector *P5[8][nsrc];
QDP_ColorVector *P5tmp[8][8][nsrc];
QDP_ColorVector *P5s[4][nsrc];
QDP_ColorVector *P5tmps[4][8][nsrc];
//QDP_ColorVector *xin[nsrc];
QDP_ColorVector *xintmp[8][nsrc];
QDP_ColorVector *Pmu[nsrc];
QDP_ColorVector *Pmutmp[8][nsrc];
QDP_ColorVector *Pnumu[nsrc];
QDP_ColorVector *Pnumutmp[8][nsrc];
QDP_ColorVector *Prhonumu[nsrc];
QDP_ColorVector *Prhonumutmp[8][nsrc];
QDP_ColorVector *P7[nsrc];
QDP_ColorVector *P7tmp[8][nsrc];
QDP_ColorVector *P7rho[nsrc];
QDP_ColorVector *ttv[nsrc];
int i, dir;
int mu, nu, rho, sig;
double nflop1 = 253935;
double nflop2 = 433968;
double nflop = nflop1 + (nflop2-nflop1)*(nsrc-1);
double dtime;
dtime = -QOP_time();
ASQTAD_FORCE_BEGIN;
QOP_trace("test 1\n");
/* setup parallel transport */
QDP_ColorMatrix *tmpmat = QDP_create_M();
for(i=0; i<QOP_common.ndim; i++) {
fbshift[i] = QDP_neighbor[i];
fbshiftdir[i] = QDP_forward;
fblink[i] = links[i];
fbshift[OPP_DIR(i)] = QDP_neighbor[i];
fbshiftdir[OPP_DIR(i)] = QDP_backward;
fblink[OPP_DIR(i)] = QDP_create_M();
QDP_M_eq_sM(tmpmat, fblink[i], QDP_neighbor[i], QDP_backward, QDP_all);
QDP_M_eq_Ma(fblink[OPP_DIR(i)], tmpmat, QDP_all);
}
tv = ttv;
for(i=0; i<nsrc; i++) {
tv[i] = QDP_create_V();
}
QOP_trace("test 2\n");
/* Allocate temporary vectors */
for(i=0; i<nsrc; i++) {
Pmu[i] = QDP_create_V();
Pnumu[i] = QDP_create_V();
Prhonumu[i] = QDP_create_V();
P7[i] = QDP_create_V();
P7rho[i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
xintmp[dir][i] = QDP_create_V();
Pmutmp[dir][i] = QDP_create_V();
Pnumutmp[dir][i] = QDP_create_V();
Prhonumutmp[dir][i] = QDP_create_V();
P7tmp[dir][i] = QDP_create_V();
}
#if 1
for(mu=0; mu<4; mu++) {
P5s[mu][i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
P5tmps[mu][dir][i] = QDP_create_V();
}
}
#else
for(mu=0; mu<8; mu++) {
P5[mu][i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
P5tmp[mu][dir][i] = QDP_create_V();
//printf("%p %p\n", P5tmp[mu][dir][i], &(P5tmp[mu][dir][i])); fflush(stdout);
if(P5tmp[mu][dir][i]==NULL) {
fprintf(stderr, "error: can't create V\n");
QDP_abort();
}
}
}
#endif
}
//printf("%p\n", P5tmp[0][4][0]); fflush(stdout);
for(mu=0; mu<8; mu++) {
for(i=0; i<nsrc; i++) {
P3[mu][i] = QDP_create_V();
//P5[mu][i] = QDP_create_V();
}
}
for(mu=0; mu<4; mu++) {
tempmom_qdp[mu] = force[mu];
QDP_M_eqm_M(tempmom_qdp[mu], tempmom_qdp[mu], QDP_odd);
}
/* Path coefficients times fermion epsilon */
/* Load path coefficients from table */
for(i=0; i<nsrc; i++) {
OneLink[i] = coef->one_link * eps[i];
Naik[i] = coef->naik * eps[i]; mNaik[i] = -Naik[i];
ThreeSt[i] = coef->three_staple * eps[i]; mThreeSt[i] = -ThreeSt[i];
FiveSt[i] = coef->five_staple * eps[i]; mFiveSt[i] = -FiveSt[i];
SevenSt[i] = coef->seven_staple * eps[i]; mSevenSt[i] = -SevenSt[i];
Lepage[i] = coef->lepage * eps[i]; mLepage[i] = -Lepage[i];
}
#if 0
printf("nsrc = %i\n", nsrc);
printf("coeffs = %g %g %g %g %g %g\n", OneLink[0], ThreeSt[0], FiveSt[0],
SevenSt[0], Lepage[0], Naik[0]);
#endif
/* *************************************** */
QOP_trace("start force loop\n");
for(mu=0; mu<8; mu++) {
//u_shift_hw_fermion(temp_x_qdp, Pmu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)]);
u_shift_color_vecs(xin, Pmu, OPP_DIR(mu), nsrc, xintmp[OPP_DIR(mu)]);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) ) {
//u_shift_hw_fermion(Pmu, P3[sig], sig, temp_hw[sig]);
u_shift_color_vecs(Pmu, P3[sig], sig, nsrc, Pmutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu]: x--+ *
* | | *
* o o *
* the 1 link in the path: - (numbering starts form 0) */
add_forces_to_mom(P3[sig], Pmu, sig, mThreeSt, nsrc);
}
}
for(nu=0; nu<8; nu++) if( (nu!=mu)&&(nu!=OPP_DIR(mu)) ) {
int nP5 = 0;
//Pnumu = hw_qdp[OPP_DIR(nu)];
//u_shift_hw_fermion(Pmu, Pnumu, OPP_DIR(nu), temp_hw[OPP_DIR(nu)]);
u_shift_color_vecs(Pmu, Pnumu, OPP_DIR(nu), nsrc, Pmutmp[OPP_DIR(nu)]);
//QDP_V_veq_V(Pnumu, P3[OPP_DIR(nu)], QDP_all, nsrc);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) &&
(sig!=nu)&&(sig!=OPP_DIR(nu)) ) {
#if 1
for(i=0; i<nsrc; i++) {
P5[sig][i] = P5s[nP5][i];
for(dir=0; dir<8; dir++) P5tmp[sig][dir][i] = P5tmps[nP5][dir][i];
}
#endif
nP5++;
//u_shift_hw_fermion(Pnumu, P5[sig], sig, temp_hw[sig]);
u_shift_color_vecs(Pnumu, P5[sig], sig, nsrc, Pnumutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu]: x--+ *
* | | *
* o o *
* the 2 link in the path: + (numbering starts form 0) */
add_forces_to_mom(P5[sig], Pnumu, sig, FiveSt, nsrc);
}
}
QOP_trace("test 4\n");
for(rho=0; rho<8; rho++) if( (rho!=mu)&&(rho!=OPP_DIR(mu)) &&
(rho!=nu)&&(rho!=OPP_DIR(nu)) ) {
//Prhonumu = hw_qdp[OPP_DIR(rho)];
//u_shift_hw_fermion(Pnumu, Prhonumu, OPP_DIR(rho),
// temp_hw[OPP_DIR(rho)] );
u_shift_color_vecs(Pnumu, Prhonumu, OPP_DIR(rho), nsrc,
Pnumutmp[OPP_DIR(rho)]);
//QDP_V_veq_V(Prhonumu, P5[OPP_DIR(rho)], QDP_all, nsrc);
for(sig=0; sig<8; sig++) if( (sig!=mu )&&(sig!=OPP_DIR(mu )) &&
(sig!=nu )&&(sig!=OPP_DIR(nu )) &&
(sig!=rho)&&(sig!=OPP_DIR(rho)) ) {
/* Length 7 paths */
//P7 = hw_qdp[sig];
//u_shift_hw_fermion(Prhonumu, P7, sig, temp_hw[sig] );
QOP_trace("test 43\n");
u_shift_color_vecs(Prhonumu, P7, sig, nsrc, Prhonumutmp[sig]);
QOP_trace("test 44\n");
//QDP_V_eq_r_times_V(P7[0], &SevenSt[0], P7[0], QDP_all);
//QDP_V_eq_r_times_V(P7[1], &SevenSt[1], P7[1], QDP_all);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu+rho]: x--+ *
* | | *
* o o *
* the 3 link in the path: - (numbering starts form 0) */
QOP_trace("test 45\n");
add_forces_to_mom(P7, Prhonumu, sig, mSevenSt, nsrc);
QOP_trace("test 46\n");
//mom_meq_force(P7, Prhonumu, sig);
}
/* Add the force F_rho the 2(4) link in the path: + */
//P7rho = hw_qdp[rho];
//u_shift_hw_fermion(P7, P7rho, rho, temp_hw[rho]);
QOP_trace("test 47\n");
u_shift_color_vecs(P7, P7rho, rho, nsrc, P7tmp[rho]);
QOP_trace("test 48\n");
side_link_forces(rho,sig,SevenSt,Pnumu,P7,Prhonumu,P7rho, nsrc);
QOP_trace("test 49\n");
//side_link_3f_force2(rho,sig,Pnumu,P7,Prhonumu,P7rho);
/* Add the P7rho vector to P5 */
for(i=0; i<nsrc; i++) {
if(FiveSt[i]!=0) coeff[i] = SevenSt[i]/FiveSt[i];
else coeff[i] = 0;
QOP_trace("test 410\n");
QDP_V_peq_r_times_V(P5[sig][i], &coeff[i], P7rho[i], QDP_all);
QOP_trace("test 411\n");
}
} /* sig */
} /* rho */
QOP_trace("test 5\n");
#define P5nu P7
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) &&
(sig!=nu)&&(sig!=OPP_DIR(nu)) ) {
/* Length 5 paths */
/* Add the force F_nu the 1(3) link in the path: - */
//P5nu = hw_qdp[nu];
//u_shift_hw_fermion(P5[sig], P5nu, nu, temp_hw[nu]);
u_shift_color_vecs(P5[sig], P5nu, nu, nsrc, P5tmp[sig][nu]);
side_link_forces(nu, sig, mFiveSt, Pmu, P5[sig], Pnumu, P5nu, nsrc);
/* Add the P5nu vector to P3 */
for(i=0; i<nsrc; i++) {
if(ThreeSt[i]!=0) coeff[i] = FiveSt[i]/ThreeSt[i];
else coeff[i] = 0;
QDP_V_peq_r_times_V(P3[sig][i], &coeff[i], P5nu[i], QDP_all);
}
} /* sig */
} /* nu */
#define Pmumu Pnumu
#define Pmumutmp Pnumutmp
#define P5sig Prhonumu
#define P5sigtmp Prhonumutmp
#define P3mu P7
#define Popmu P7
#define Pmumumu P7
/* Now the Lepage term... It is the same as 5-link paths with
nu=mu and FiveSt=Lepage. */
//u_shift_hw_fermion(Pmu, Pmumu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)] );
u_shift_color_vecs(Pmu, Pmumu, OPP_DIR(mu), nsrc, Pmutmp[OPP_DIR(mu)]);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) ) {
//P5sig = hw_qdp[sig];
//u_shift_hw_fermion(Pmumu, P5sig, sig, temp_hw[sig]);
u_shift_color_vecs(Pmumu, P5sig, sig, nsrc, Pmumutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu]: x--+ *
* | | *
* o o *
* the 2 link in the path: + (numbering starts form 0) */
add_forces_to_mom(P5sig, Pmumu, sig, Lepage, nsrc);
}
/* Add the force F_nu the 1(3) link in the path: - */
//P5nu = hw_qdp[mu];
//u_shift_hw_fermion(P5sig, P5nu, mu, temp_hw[mu]);
u_shift_color_vecs(P5sig, P5nu, mu, nsrc, P5sigtmp[mu]);
side_link_forces(mu, sig, mLepage, Pmu, P5sig, Pmumu, P5nu, nsrc);
/* Add the P5nu vector to P3 */
for(i=0; i<nsrc; i++) {
if(ThreeSt[i]!=0) coeff[i] = Lepage[i]/ThreeSt[i];
else coeff[i] = 0;
QDP_V_peq_r_times_V(P3[sig][i], &coeff[i], P5nu[i], QDP_all);
}
/* Length 3 paths (Not the Naik term) */
/* Add the force F_mu the 0(2) link in the path: + */
if(GOES_FORWARDS(mu)) {
//P3mu = hw_qdp[mu]; /* OK to clobber P5nu */
//u_shift_hw_fermion(P3[sig], P3mu, mu, temp_hw[mu]);
//u_shift_color_vecs(P3[sig], P3mu, mu, 2, temp_hw[mu]);
for(i=0; i<nsrc; i++) {
QDP_V_eq_V(P5sig[i], P3[sig][i], QDP_all);
}
u_shift_color_vecs(P5sig, P3mu, mu, nsrc, P5sigtmp[mu]);
}
/* The above shift is not needed if mu is backwards */
side_link_forces(mu, sig, ThreeSt, xin, P3[sig], Pmu, P3mu, nsrc);
}
/* Finally the OneLink and the Naik term */
if(GOES_BACKWARDS(mu)) {
/* Do only the forward terms in the Dslash */
/* Because I have shifted with OPP_DIR(mu) Pmu is a forward *
* shift. */
/* The one link */
add_forces_to_mom(Pmu, xin, OPP_DIR(mu), OneLink, nsrc);
/* For the same reason Pmumu is the forward double link */
/* Popmu is a backward shift */
//Popmu = hw_qdp[mu]; /* OK to clobber P3mu */
//u_shift_hw_fermion(xin, Popmu, mu, temp_hw[mu]);
u_shift_color_vecs(xin, Popmu, mu, nsrc, xintmp[mu]);
/* The Naik */
/* link no 1: - */
add_forces_to_mom(Pmumu, Popmu, OPP_DIR(mu), mNaik, nsrc);
/* Pmumumu can overwrite Popmu which is no longer needed */
//Pmumumu = hw_qdp[OPP_DIR(mu)];
//u_shift_hw_fermion(Pmumu, Pmumumu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)]);
u_shift_color_vecs(Pmumu, Pmumumu, OPP_DIR(mu), nsrc, Pmumutmp[OPP_DIR(mu)]);
/* link no 0: + */
add_forces_to_mom(Pmumumu, xin, OPP_DIR(mu), Naik, nsrc);
} else {
/* The rest of the Naik terms */
//Popmu = hw_qdp[mu]; /* OK to clobber P3mu */
//u_shift_hw_fermion(xin, Popmu, mu, temp_hw[mu]);
u_shift_color_vecs(xin, Popmu, mu, nsrc, xintmp[mu]);
/* link no 2: + */
/* Pmumu is double backward shift */
add_forces_to_mom(Popmu, Pmumu, mu, Naik, nsrc);
}
/* Here we have to do together the Naik term and the one link term */
}/* mu */
QOP_trace("test 6\n");
QOP_trace("test 7\n");
for(mu=0; mu<4; mu++) {
QDP_M_eq_M(tmpmat, tempmom_qdp[mu], QDP_even);
QDP_M_eqm_M(tmpmat, tempmom_qdp[mu], QDP_odd);
QDP_M_eq_antiherm_M(tempmom_qdp[mu], tmpmat, QDP_all);
}
QDP_destroy_M(tmpmat);
//printf("%p\n", P5tmp[0][4][0]); fflush(stdout);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
/* Free temporary vectors */
for(i=0; i<nsrc; i++) {
QDP_destroy_V(Pmu[i]);
QDP_destroy_V(Pnumu[i]);
QDP_destroy_V(Prhonumu[i]);
QDP_destroy_V(P7[i]);
QDP_destroy_V(P7rho[i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(dir=0; dir<8; dir++) {
QDP_destroy_V(xintmp[dir][i]);
QDP_destroy_V(Pmutmp[dir][i]);
QDP_destroy_V(Pnumutmp[dir][i]);
QDP_destroy_V(Prhonumutmp[dir][i]);
QDP_destroy_V(P7tmp[dir][i]);
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(mu=0; mu<4; mu++) {
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
QDP_destroy_V(P5s[mu][i]);
//QDP_destroy_V(P5[mu][i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(dir=0; dir<8; dir++) {
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
QDP_destroy_V(P5tmps[mu][dir][i]);
//printf("%p\n", P5tmp[mu][dir][i]); fflush(stdout);
//QDP_destroy_V(P5tmp[mu][dir][i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("here3\n"); fflush(stdout); }
for(mu=0; mu<8; mu++) {
for(i=0; i<nsrc; i++) {
QDP_destroy_V(P3[mu][i]);
}
//QDP_destroy_V(P5[mu][0]);
//QDP_destroy_V(P5[mu][1]);
}
for(i=0; i<nsrc; i++) {
QDP_destroy_V(tv[i]);
}
//if(QDP_this_node==0) { printf("here4\n"); fflush(stdout); }
for(i=4; i<8; i++) {
QDP_destroy_M(fblink[i]);
}
dtime += QOP_time();
info->final_sec = dtime;
info->final_flop = nflop*QDP_sites_on_node;
info->status = QOP_SUCCESS;
ASQTAD_FORCE_END;
#undef NC
}
int
Rayleigh_min_qdp(QDP_ColorVector *vec, QDP_ColorVector **eigVec,
Real Tolerance, Real RelTol, int Nvecs, int MaxIter,
int Restart, QDP_Subset subset)
{
QLA_Complex cc;
QLA_Real beta, cos_theta, sin_theta;
QLA_Real quot, P_norm, theta, real_vecMp, pMp;
QLA_Real g_norm, old_g_norm, start_g_norm;
QDP_ColorVector *Mvec, *grad, *P, *MP;
int iter;
#ifdef DEBUG
if(QDP_this_node==0) printf("begin Rayleigh_min_qdp\n");
#endif
Mvec = QDP_create_V();
grad = QDP_create_V();
//oldgrad = QDP_create_V();
P = QDP_create_V();
MP = QDP_create_V();
project_out_qdp(vec, eigVec, Nvecs, subset);
normalize_qdp(vec, subset);
Matrix_Vec_mult_qdp(vec, Mvec, subset);
project_out_qdp(Mvec, eigVec, Nvecs, subset);
/* Compute the quotient quot=vev*M*vec */
QDP_r_eq_re_V_dot_V(", vec, Mvec, subset);
/* quot is real since M is hermitian. quot = vec*M*vec */
#ifdef DEBUG
if(QDP_this_node==0) printf("Rayleigh_min: Start -- quot=%g\n", quot);
#endif
/* Compute the grad=M*vec - quot*vec */
QDP_V_eq_V(grad, Mvec, QDP_all);
QDP_V_meq_r_times_V(grad, ", vec, subset);
/* set P (the search direction) equal to grad */
QDP_V_eq_V(P, grad, QDP_all);
/* compute the norms of P and grad */
QDP_r_eq_norm2_V(&P_norm, P, subset);
P_norm = sqrt(P_norm);
QDP_r_eq_norm2_V(&g_norm, grad, subset);
g_norm = sqrt(g_norm);
start_g_norm = g_norm;
//QDP_V_eq_V(oldgrad, grad, subset);
#ifdef DEBUG
if(QDP_this_node==0) printf("Rayleigh_min: Start -- g_norm=%g\n", g_norm);
#endif
iter = 0;
while( (g_norm>Tolerance*quot) &&
( ((iter<MaxIter)&&(g_norm/start_g_norm>RelTol)) || (iter<MINITER) )
) {
iter++;
Matrix_Vec_mult_qdp(P, MP, subset);
QDP_r_eq_re_V_dot_V(&real_vecMp, vec, MP, subset);
QDP_r_eq_re_V_dot_V(&pMp, P, MP, subset);
theta = 0.5*atan(2.0*real_vecMp/(quot*P_norm - pMp/P_norm));
sin_theta = sin(theta);
cos_theta = cos(theta);
if(sin_theta*cos_theta*real_vecMp>0) {
theta = theta - 0.5*M_PI; /* chose the minimum not the maximum */
sin_theta = sin(theta); /* the sin,cos calls can be avoided */
cos_theta = cos(theta);
}
sin_theta = sin_theta/P_norm;
/* vec = cos(theta)*vec +sin(theta)*P/p_norm */
//dax_p_by_qdp(cos_theta, vec, sin_theta, P, subset);
QDP_V_eq_r_times_V(vec, &cos_theta, vec, subset);
QDP_V_peq_r_times_V(vec, &sin_theta, P, subset);
/* Mvec = cos(theta)*Mvec +sin(theta)*MP/p_norm */
//dax_p_by_qdp(cos_theta, Mvec, sin_theta, MP, subset);
QDP_V_eq_r_times_V(Mvec, &cos_theta, Mvec, subset);
QDP_V_peq_r_times_V(Mvec, &sin_theta, MP, subset);
/* renormalize vec ... */
if( iter%Restart == 0 ) {
#ifdef DEBUG
{
QLA_Real vec_norm;
if(QDP_this_node==0) printf("Renormalizing...");
QDP_r_eq_norm2_V(&vec_norm, vec, subset);
if(QDP_this_node==0) printf(" norm: %g\n", sqrt(vec_norm));
}
#endif
/* Project vec on the orthogonal complement of eigVec */
project_out_qdp(vec, eigVec, Nvecs, subset);
normalize_qdp(vec, subset);
Matrix_Vec_mult_qdp(vec, Mvec, subset);
/* Recompute the quotient */
QDP_r_eq_re_V_dot_V(", vec, Mvec, subset);
/* Recompute the grad */
QDP_V_eq_V(grad, Mvec, QDP_all);
QDP_V_meq_r_times_V(grad, ", vec, subset);
//QDP_r_eq_norm2_V(&g_norm, grad, subset);
//printf("g_norm = %g\n", g_norm);
/* Project P on the orthogonal complement of eigVec */
//QDP_r_eq_norm2_V(&P_norm, P, subset);
//printf("P_norm = %g\n", P_norm);
project_out_qdp(P, eigVec, Nvecs, subset);
//QDP_r_eq_norm2_V(&P_norm, P, subset);
//printf("P_norm = %g\n", P_norm);
/* make P orthogonal to vec */
QDP_c_eq_V_dot_V(&cc, vec, P, subset);
//printf("cc = %g\n", QLA_real(cc));
QDP_V_meq_c_times_V(P, &cc, vec, subset);
//QDP_r_eq_norm2_V(&P_norm, P, subset);
//printf("P_norm = %g\n", P_norm);
/* make P orthogonal to grad */
QDP_c_eq_V_dot_V(&cc, grad, P, subset);
//printf("cc = %g\n", QLA_real(cc));
QDP_V_meq_c_times_V(P, &cc, grad, subset);
QDP_r_eq_norm2_V(&P_norm, P, subset);
P_norm = sqrt(P_norm);
}
QDP_r_eq_re_V_dot_V(", vec, Mvec, subset);
#ifdef DEBUG
node0_printf("Rayleigh_min: %i, quot=%8g g=%8g b=%6g P:%6g\n",
iter, quot, g_norm, beta, P_norm);
#endif
old_g_norm = g_norm;
QDP_V_eq_V(grad, Mvec, QDP_all);
QDP_V_meq_r_times_V(grad, ", vec, subset);
//QDP_V_meq_V(oldgrad, grad, subset);
//QDP_r_eq_re_V_dot_V(&g_norm, oldgrad, grad, subset);
//QDP_V_eq_V(oldgrad, grad, subset);
QDP_r_eq_norm2_V(&g_norm, grad, subset);
g_norm = sqrt(g_norm);
beta = cos_theta*g_norm*g_norm/(old_g_norm*old_g_norm);
if( beta>2.0 ) beta = 2.0; /* Cut off beta */
QDP_c_eq_V_dot_V(&cc, vec, P, subset);
QLA_real(cc) *= beta;
QLA_imag(cc) *= beta;
QDP_V_eq_r_times_V_plus_V(P, &beta, P, grad, subset);
QDP_V_meq_c_times_V(P, &cc, vec, subset);
QDP_r_eq_norm2_V(&P_norm, P, subset);
P_norm = sqrt(P_norm);
}
project_out_qdp(vec, eigVec, Nvecs, subset);
normalize_qdp(vec, subset);
QDP_destroy_V(MP);
QDP_destroy_V(P);
//QDP_destroy_V(oldgrad);
QDP_destroy_V(grad);
QDP_destroy_V(Mvec);
iter++;
#ifdef DEBUG
if(QDP_this_node==0) printf("end Rayleigh_min_qdp\n");
#endif
return iter;
}
