float rayMarch( vec3 ray0, vec3 hRay ){
float dt = dtmax;
float t = tmin;
while( t<tmax ){
float val = evalFunc( ray0 + t*hRay );
if( abs(val) > iso ) break;
t += dt;
}
//return t;
return bisec( ray0, hRay, t-dt, dt );
}
static const kd_node_t *kd_index_hlp (kd_tree_t *tree, unsigned *idx, unsigned n,
bnds_t *bnds // bounds of the subtree
) {
kd_node_t *node;
unsigned n_lo, n_hi, cut_dim;
float cut_val;
float lo, hi; // save value for the cutted dimension
assert(n > 1);
node = kd_alloc_node(tree);
bisec(tree, idx, n, bnds, &cut_dim, &cut_val, &n_lo);
node->cut_dim = cut_dim;
node->cut_val = cut_val;
lo = node->lower = bnds->lo[cut_dim];
hi = node->upper = bnds->hi[cut_dim];
// construct the left subtree
bnds->hi[cut_dim] = cut_val;
assert(n_lo > 0);
if (n_lo == 1) {
node->left.leaf = idx[0];
}
else {
node->left.node = kd_index_hlp(tree, idx, n_lo, bnds);
}
bnds->hi[cut_dim] = hi;
// construct the right subtree
bnds->lo[cut_dim] = cut_val;
n_hi = n - n_lo;
if (n_hi == 1) {
node->right.leaf = idx[n-1];
}
else {
node->right.node = kd_index_hlp(tree, idx + n_lo, n_hi, bnds);
}
bnds->lo[cut_dim] = lo;
return node;
}
int
ripper(
double range_min,
double range_max,
double delta,
unsigned int opt,
int results_max,
double * results ,
double val,
double (*fevol)(double, int, char *, void *),
void * param)
{
double cursor = range_min;
double cursor_last = cursor;
double b=range_max,a=range_min;
double value;
double value_last;
double epsilon = 1.e-15;
double t=0.,s,dt; /* t in (0,1] */
int results_collected = 0;
value_last = fevol( cursor, 0, "", param ) - val;
switch( opt )
{
case RIPPER_LINEAR:
dt = delta/(b-a);
break;
case RIPPER_DENSE1:
dt = sqrt(1.-pow(1.-delta/(b-a),2));
break;
default:
dt = delta/(b-a);
}
/* printf("cursor_last = %.5E, cursor = %.5E, dt = %.5E, s = %.5E\n", cursor_last, cursor, dt, s); */
while( cursor < range_max )
{
t+=dt;
switch( opt ) {
case RIPPER_LINEAR:
s=t;
break;
case RIPPER_DENSE1:
s=1.-sqrt(1.-t*t);
break;
default:
s=t;
}
cursor_last = cursor;
cursor = a+s*(b-a);
value = fevol( cursor, 0, "", param ) - val;
/* printf("cursor_last = %.5E, cursor = %.5E, dt = %.5E, s = %.5E\n", cursor_last, cursor, dt, s); */
/* if signs are opposite */
if( value_last*value < 0 )
{
results[results_collected] = bisec( cursor_last, cursor, epsilon, val, fevol, param );
results_collected++;
printf("collected result in (%.15f,%.15f), %i results in total\n",
cursor_last, cursor, results_collected);
if( results_collected >= results_max )
return results_collected;
value_last = value;
}
}
return results_collected;
}
/**
*
*
* @param X
*
* @author Takahiro Misawa (The University of Tokyo)
* @author Kazuyoshi Yoshimi (The University of Tokyo)
* @return
*/
int Lanczos_EigenValue(struct BindStruct *X)
{
fprintf(stdoutMPI, "%s", cLogLanczos_EigenValueStart);
FILE *fp;
char sdt[D_FileNameMax],sdt_2[D_FileNameMax];
int stp, iproc;
long int i,iv,i_max;
unsigned long int i_max_tmp, sum_i_max;
int k_exct,Target;
int iconv=-1;
double beta1,alpha1; //beta,alpha1 should be real
double complex temp1,temp2;
double complex cbeta1;
double E[5],ebefor;
int mythread;
// for GC
double dnorm;
double complex cdnorm;
long unsigned int u_long_i;
dsfmt_t dsfmt;
#ifdef lapack
double **tmp_mat;
double *tmp_E;
int int_i,int_j,mfint[7];
#endif
sprintf(sdt_2, cFileNameLanczosStep, X->Def.CDataFileHead);
i_max=X->Check.idim_max;
k_exct = X->Def.k_exct;
if(initial_mode == 0){
sum_i_max = SumMPI_li(X->Check.idim_max);
X->Large.iv = (sum_i_max / 2 + X->Def.initial_iv) % sum_i_max + 1;
iv=X->Large.iv;
fprintf(stdoutMPI, " initial_mode=%d normal: iv = %ld i_max=%ld k_exct =%d \n\n",initial_mode,iv,i_max,k_exct);
#pragma omp parallel for default(none) private(i) shared(v0, v1) firstprivate(i_max)
for(i = 1; i <= i_max; i++){
v0[i]=0.0;
v1[i]=0.0;
}
sum_i_max = 0;
for (iproc = 0; iproc < nproc; iproc++) {
i_max_tmp = BcastMPI_li(iproc, i_max);
if (sum_i_max <= iv && iv < sum_i_max + i_max_tmp) {
if (myrank == iproc) {
v1[iv - sum_i_max+1] = 1.0;
if (X->Def.iInitialVecType == 0) {
v1[iv - sum_i_max+1] += 1.0*I;
v1[iv - sum_i_max+1] /= sqrt(2.0);
}
}/*if (myrank == iproc)*/
}/*if (sum_i_max <= iv && iv < sum_i_max + i_max_tmp)*/
sum_i_max += i_max_tmp;
}/*for (iproc = 0; iproc < nproc; iproc++)*/
}/*if(initial_mode == 0)*/
else if(initial_mode==1){
iv = X->Def.initial_iv;
fprintf(stdoutMPI, " initial_mode=%d (random): iv = %ld i_max=%ld k_exct =%d \n\n",initial_mode,iv,i_max,k_exct);
#pragma omp parallel default(none) private(i, u_long_i, mythread, dsfmt) \
shared(v0, v1, iv, X, nthreads, myrank) firstprivate(i_max)
{
#pragma omp for
for (i = 1; i <= i_max; i++) {
v0[i] = 0.0;
}
/*
Initialise MT
*/
#ifdef _OPENMP
mythread = omp_get_thread_num();
#else
mythread = 0;
#endif
u_long_i = 123432 + labs(iv) + mythread + nthreads * myrank;
dsfmt_init_gen_rand(&dsfmt, u_long_i);
if (X->Def.iInitialVecType == 0) {
#pragma omp for
for (i = 1; i <= i_max; i++)
v1[i] = 2.0*(dsfmt_genrand_close_open(&dsfmt) - 0.5) + 2.0*(dsfmt_genrand_close_open(&dsfmt) - 0.5)*I;
}
else {
#pragma omp for
for (i = 1; i <= i_max; i++)
v1[i] = 2.0*(dsfmt_genrand_close_open(&dsfmt) - 0.5);
}
}/*#pragma omp parallel*/
cdnorm=0.0;
#pragma omp parallel for default(none) private(i) shared(v1, i_max) reduction(+: cdnorm)
for(i=1;i<=i_max;i++){
cdnorm += conj(v1[i])*v1[i];
}
cdnorm = SumMPI_dc(cdnorm);
dnorm=creal(cdnorm);
dnorm=sqrt(dnorm);
#pragma omp parallel for default(none) private(i) shared(v1) firstprivate(i_max, dnorm)
for(i=1;i<=i_max;i++){
v1[i] = v1[i]/dnorm;
}
}/*else if(initial_mode==1)*/
//Eigenvalues by Lanczos method
TimeKeeper(X, cFileNameTimeKeep, cLanczos_EigenValueStart, "a");
mltply(X, v0, v1);
stp=1;
TimeKeeperWithStep(X, cFileNameTimeKeep, cLanczos_EigenValueStep, "a", stp);
alpha1=creal(X->Large.prdct) ;// alpha = v^{\dag}*H*v
alpha[1]=alpha1;
cbeta1=0.0;
#pragma omp parallel for reduction(+:cbeta1) default(none) private(i) shared(v0, v1) firstprivate(i_max, alpha1)
for(i = 1; i <= i_max; i++){
cbeta1+=conj(v0[i]-alpha1*v1[i])*(v0[i]-alpha1*v1[i]);
}
cbeta1 = SumMPI_dc(cbeta1);
beta1=creal(cbeta1);
beta1=sqrt(beta1);
beta[1]=beta1;
ebefor=0;
/*
Set Maximum number of loop to the dimention of the Wavefunction
*/
i_max_tmp = SumMPI_li(i_max);
if(i_max_tmp < X->Def.Lanczos_max){
X->Def.Lanczos_max = i_max_tmp;
}
if(i_max_tmp < X->Def.LanczosTarget){
X->Def.LanczosTarget = i_max_tmp;
}
if(i_max_tmp == 1){
E[1]=alpha[1];
vec12(alpha,beta,stp,E,X);
X->Large.itr=stp;
X->Phys.Target_energy=E[k_exct];
iconv=0;
fprintf(stdoutMPI," LanczosStep E[1] \n");
fprintf(stdoutMPI," stp=%d %.10lf \n",stp,E[1]);
}
else{
#ifdef lapack
fprintf(stdoutMPI, " LanczosStep E[1] E[2] E[3] E[4] E_Max/Nsite\n");
#else
fprintf(stdoutMPI, " LanczosStep E[1] E[2] E[3] E[4] \n");
#endif
for(stp = 2; stp <= X->Def.Lanczos_max; stp++){
#pragma omp parallel for default(none) private(i,temp1, temp2) shared(v0, v1) firstprivate(i_max, alpha1, beta1)
for(i=1;i<=i_max;i++){
temp1 = v1[i];
temp2 = (v0[i]-alpha1*v1[i])/beta1;
v0[i] = -beta1*temp1;
v1[i] = temp2;
}
mltply(X, v0, v1);
TimeKeeperWithStep(X, cFileNameTimeKeep, cLanczos_EigenValueStep, "a", stp);
alpha1=creal(X->Large.prdct);
alpha[stp]=alpha1;
cbeta1=0.0;
#pragma omp parallel for reduction(+:cbeta1) default(none) private(i) shared(v0, v1) firstprivate(i_max, alpha1)
for(i=1;i<=i_max;i++){
cbeta1+=conj(v0[i]-alpha1*v1[i])*(v0[i]-alpha1*v1[i]);
}
cbeta1 = SumMPI_dc(cbeta1);
beta1=creal(cbeta1);
beta1=sqrt(beta1);
beta[stp]=beta1;
Target = X->Def.LanczosTarget;
if(stp==2){
#ifdef lapack
d_malloc2(tmp_mat,stp,stp);
d_malloc1(tmp_E,stp+1);
for(int_i=0;int_i<stp;int_i++){
for(int_j=0;int_j<stp;int_j++){
tmp_mat[int_i][int_j] = 0.0;
}
}
tmp_mat[0][0] = alpha[1];
tmp_mat[0][1] = beta[1];
tmp_mat[1][0] = beta[1];
tmp_mat[1][1] = alpha[2];
DSEVvalue(stp,tmp_mat,tmp_E);
E[1] = tmp_E[0];
E[2] = tmp_E[1];
E[3] = tmp_E[2];
E[4] = tmp_E[3];
d_free1(tmp_E,stp+1);
d_free2(tmp_mat,stp,stp);
#else
bisec(alpha,beta,stp,E,4,eps_Bisec);
#endif
ebefor=E[Target];
childfopenMPI(sdt_2,"w", &fp);
#ifdef lapack
fprintf(stdoutMPI, " stp = %d %.10lf %.10lf xxxxxxxxxx xxxxxxxxx xxxxxxxxx \n",stp,E[1],E[2]);
fprintf(fp, "LanczosStep E[1] E[2] E[3] E[4] E_Max/Nsite\n");
fprintf(fp, "stp = %d %.10lf %.10lf xxxxxxxxxx xxxxxxxxx xxxxxxxxx \n",stp,E[1],E[2]);
#else
fprintf(stdoutMPI, " stp = %d %.10lf %.10lf xxxxxxxxxx xxxxxxxxx \n",stp,E[1],E[2]);
fprintf(fp, "LanczosStep E[1] E[2] E[3] E[4] \n");
fprintf(fp,"stp = %d %.10lf %.10lf xxxxxxxxxx xxxxxxxxx \n",stp,E[1],E[2]);
#endif
fclose(fp);
}
if(stp>2 && stp%2==0){
childfopenMPI(sdt_2,"a", &fp);
#ifdef lapack
d_malloc2(tmp_mat,stp,stp);
d_malloc1(tmp_E,stp+1);
for(int_i=0;int_i<stp;int_i++){
for(int_j=0;int_j<stp;int_j++){
tmp_mat[int_i][int_j] = 0.0;
}
}
tmp_mat[0][0] = alpha[1];
tmp_mat[0][1] = beta[1];
for(int_i=1;int_i<stp-1;int_i++){
tmp_mat[int_i][int_i] = alpha[int_i+1];
tmp_mat[int_i][int_i+1] = beta[int_i+1];
tmp_mat[int_i][int_i-1] = beta[int_i];
}
tmp_mat[int_i][int_i] = alpha[int_i+1];
tmp_mat[int_i][int_i-1] = beta[int_i];
DSEVvalue(stp,tmp_mat,tmp_E);
E[1] = tmp_E[0];
E[2] = tmp_E[1];
E[3] = tmp_E[2];
E[4] = tmp_E[3];
E[0] = tmp_E[stp-1];
d_free1(tmp_E,stp+1);
d_free2(tmp_mat,stp,stp);
fprintf(stdoutMPI, " stp = %d %.10lf %.10lf %.10lf %.10lf %.10lf\n",stp,E[1],E[2],E[3],E[4],E[0]/(double)X->Def.NsiteMPI);
fprintf(fp,"stp=%d %.10lf %.10lf %.10lf %.10lf %.10lf\n",stp,E[1],E[2],E[3],E[4],E[0]/(double)X->Def.NsiteMPI);
#else
bisec(alpha,beta,stp,E,4,eps_Bisec);
fprintf(stdoutMPI, " stp = %d %.10lf %.10lf %.10lf %.10lf \n",stp,E[1],E[2],E[3],E[4]);
fprintf(fp,"stp=%d %.10lf %.10lf %.10lf %.10lf\n",stp,E[1],E[2],E[3],E[4]);
#endif
fclose(fp);
if(fabs((E[Target]-ebefor)/E[Target])<eps_Lanczos || fabs(beta[stp])<pow(10.0, -14)){
vec12(alpha,beta,stp,E,X);
X->Large.itr=stp;
X->Phys.Target_energy=E[k_exct];
iconv=0;
break;
}
ebefor=E[Target];
}
}
}
sprintf(sdt,cFileNameTimeKeep,X->Def.CDataFileHead);
if(iconv!=0){
sprintf(sdt, cLogLanczos_EigenValueNotConverged);
return -1;
}
TimeKeeper(X, cFileNameTimeKeep, cLanczos_EigenValueFinish, "a");
fprintf(stdoutMPI, "%s", cLogLanczos_EigenValueEnd);
return 0;
}
main(int argc, char *argv[]) {
int i=0, j=0, k=0, N, JMAX=45, M, l, tempnum;
double offsetV[20],offsetR[20],offsetI[20],dm15_templates[20];
double temp3,temp4,temp5,temp6,temp7;
double dm15;
double center_min, center_max;
double f1,f2,fmid1,fmid2, W[MXPTS], rtb, dx, xmid, *w1;
float *w2;
double t[MXPTS], B[MXPTS], V[MXPTS], R[MXPTS], I[MXPTS], tsigma[MXPTS],
Bsigma[MXPTS], Vsigma[MXPTS], Rsigma[MXPTS], Isigma[MXPTS];
double BB[MXPTS], VV[MXPTS], RR[MXPTS], II[MXPTS], BBsig[200],
VVsig[MXPTS], RRsig[MXPTS], IIsig[MXPTS];
double shiftV=0, shiftR=0, shiftI=0;
char *temp1[20], temp2, outfile1[20], outfile2[20];
FILE *arch1, *arch2, *arch3, *arch4, *arch5;
FILE *arch;
if(argc < 2) {
printf("Usage: dm15temp dm15 \n\n");
exit(0);
}
/* dm15 wanted for the output template */
dm15 = atof(argv[1]);
/* Allowed range of dm15 */
if(dm15 > 1.93 || dm15 < 0.83) {
printf("\nAllowed range: 1.93 > dm15 > 0.83\n\n");
exit(0);
}
/* Reading data of the templates */
arch1 = fopen("dm15temps.dat","r");
while(fscanf(arch1,"%s %s %lf %lf %lf %lf", names[i].temp, names[i].sigma, &temp3, &temp4, &temp5, &temp6)!=EOF) {
dm15_templates[i] = temp3;
offsetV[i] = temp4;
offsetR[i] = temp5;
offsetI[i] = temp6;
i++;
}
N=i; i=0; j=0;
fclose(arch1);
if(dm15 < 0.90) {
center_min= dm15 - 0.10;
center_max= dm15 + 0.05;
}
else {
center_min= dm15 - 0.15;
center_max= dm15 + 0.20;
}
/*
Bisection method to find the weights of a given dm15 template
changing the central value of the triangle function
*/
i=0;
w1 = center_dm15(center_min, dm15_templates, N);
f1= w1[N];
w1 = center_dm15(center_max, dm15_templates, N);
f2= w1[N];
fmid1=bisec(f1,dm15);
fmid2=bisec(f2,dm15);
rtb=fmid1 < 0.0 ? (dx=center_max-center_min,center_min): (dx=center_min-center_max,center_max);
for(j=1; j <= JMAX; j++) {
xmid = rtb + (dx *= 0.5);
/* w1 is the vector of weights */
w1=center_dm15(xmid,dm15_templates,N);
fmid1=w1[N];
fmid2=bisec(fmid1,dm15);
if(fmid2 <= 0.0) rtb=xmid;
if((fabs(dx) < 0.0000001) || (fmid2 == 0.0)) {j=JMAX+1;}
}
/*
Finding the difference between the magnitude at maximum and the
magnitude at Bmax for the new template.
*/
for(j=0; j < N; j++) {
W[j]=w1[j];
shiftV = offsetV[j] * W[j] + shiftV;
shiftR = offsetR[j] * W[j] + shiftR;
shiftI = offsetI[j] * W[j] + shiftI;
}
j=0; i=0; k=0;
for(i=0; i < MXPTS; i++){
BB[i]=0; VV[i]=0; RR[i]=0; II[i]=0;
BBsig[i]=0; VVsig[i]=0; RRsig[i]=0; IIsig[i]=0;
}
i=0;
/*
Weighting the templates in the template set
*/
for(l=0; l < N; l++) {
if(W[l] > 0) {
arch3=fopen(names[l].temp,"r");
/* Read template data */
while(fscanf(arch3,"%lf %lf %lf %lf %lf",&temp3,&temp4,&temp5,&temp6,&temp7)!=EOF){
t[j]=temp3;
B[j]=temp4;
V[j]=temp5;
R[j]=temp6;
I[j]=temp7;
j++;
}
M=j; j=0;
fclose(arch3);
arch4=fopen(names[l].sigma,"r");
/* Read the approx. uncertainties in each template (saved as look-up tables) */
while(fscanf(arch4,"%lf %lf %lf %lf %lf",&temp3, &temp4, &temp5, &temp6, &temp7)!=EOF){
tsigma[j]=temp3;
Bsigma[j]=temp4;
Vsigma[j]=temp5;
Rsigma[j]=temp6;
Isigma[j]=temp7;
j++;
}
j=0;
fclose(arch4);
/* Weighting */
for(i=0; i<M; i++) {
BB[i] = B[i] * W[l] + BB[i];
VV[i] = V[i] * W[l] + VV[i];
RR[i] = R[i] * W[l] + RR[i];
II[i] = I[i] * W[l] + II[i];
BBsig[i] = pow((Bsigma[i] * W[l]),2.) + BBsig[i];
VVsig[i] = pow((Vsigma[i] * W[l]),2.) + VVsig[i];
RRsig[i] = pow((Rsigma[i] * W[l]),2.) + RRsig[i];
IIsig[i] = pow((Isigma[i] * W[l]),2.) + IIsig[i];
}
}
}
/* Printing out results */
printf("# Constructed light curve template for dm15=%.3f mag\n",dm15);
printf("# Columns: \n# 1 t(Bmax) \n# 2-3 B-B(max) sigma[B-B(max)]^2 \n# 4-5 V-V(max) sigma[V-V(max)]^2 \n# 6-7 R-R(max) sigma[R-R(max)]^2 \n# 8-9 I-I(max) sigma[I-I(max)]^2 \n");
fprintf(stderr, "%lf %lf %lf\n", shiftV, shiftR, shiftI);
for(i=0; i<M; i++) {
/* printf("%5.1lf %10.3lf %10.6lf %10.3lf %10.6lf %10.3lf %10.6lf %10.3lf %10.6lf\n",t[i],BB[i],BBsig[i],VV[i]-shiftV,VVsig[i],RR[i]-shiftR,RRsig[i],II[i]-shiftI,IIsig[i]); */
printf("%5.1lf %10.3lf %10.6lf %10.3lf %10.6lf %10.3lf %10.6lf %10.3lf %10.6lf\n",t[i],BB[i],BBsig[i],VV[i],VVsig[i],RR[i],RRsig[i],II[i],IIsig[i]);
}
}
