void ML_multi_DownhillSimplex::start_amoeba(int& nfunk)
{
nfunk = 0;
amoeba(simplex,
funk_vals,
get_function_tolerance(),
&ML_multi::bounded_eval_neg_log_likelihood_at,
// &ML_multi::eval_neg_log_likelihood_at,
nfunk);
// Now set the ratepvector and Likelihood parameters to reflect
// the minimum found at vertex 0 of the simplex
v_ratep_type parameters(simplex[0]);
assert(parameters.size() == branch_rate_manager.get_ratepvector().size());
branch_rate_manager.get_ratepvector().assign(parameters);
compute();
if (DEBUG_START_AMOEBA) {
std::cout << "end of start_amoeba(): ";
std::cout << " ftol=" << get_gradient_tolerance();
std::cout << " nfunk=" << nfunk;
std::cout << " minimum at simplex[0]=" << get_neg_log_likelihood();
std::cout << std::endl << std::endl << "parm\tval" << std::endl;
for (size_t i = 0; i < branch_rate_manager.get_ratepvector().size(); ++i) {
std::cout << branch_rate_manager.get_ratepvector()[i].get_name()
<< "\t"
<< branch_rate_manager.get_ratepvector()[i].get_ratep()
<< std::endl;
}
std::cout << std::endl;
if (DEBUG_PRINT_DATA_STRUCTURES)
print_data_structures();
}
}
void zspace_minimization(char *fname)
{
int n,niter,i,j;
double *a,**pp,*yy,FTOL=1.0E-3,chi2min,qromo(),midpnt(),s1;
fprintf(stderr,"\n\nCHI2 MINIMIZATION OF W_P(R_P)+xi(s,p) DATA..........\n");
fprintf(stderr, "-----------------------------------------------------\n\n");
OUTPUT=0;
wp_input();
Work.imodel=2;
/* Find the number of free parameters in the minimization
* for the real-space correlation function.
*/
for(n=0,i=1;i<=100;++i)
{
n+=HOD.free[i];
if(OUTPUT)
printf("zspace_min> free[%i] = %d\n",i,HOD.free[i]);
}
wp.ncf=n;
a=dvector(1,n);
if(POWELL)
pp=dmatrix(1,n,1,n);
else
pp=dmatrix(1,n+1,1,n);
yy=dvector(1,n+1);
initial_zspace_values(a,pp,yy);
if(POWELL)
{
powell(a,pp,n,FTOL,&niter,&chi2min,func_chi2);
chi2min = func_chi2(a);
}
else
amoeba(pp,yy,n,FTOL,func_chi2,&niter);
s1=qromo(func_galaxy_bias,log(HOD.M_low),log(HOD.M_max),midpnt);
GALAXY_BIAS=s1/GALAXY_DENSITY;
if(!ThisTask) {
printf("ZSPACEMIN %e %e ",chi2min,HOD.M_min);
for(i=1;i<=n;++i)printf("%e ",a[i]);
printf(" %f\n",GALAXY_BIAS);
}
output_parameter_file(fname);
}
void fit_color_samples()
{
int n,niter,i,j;
double *a,**pp,*yy,FTOL=1.0E-3,chi2min,s1,dlogm,m;
FILE *fp;
char aa[1000];
mcmc_color_minimization();
fprintf(stderr,"\n\nCHI2 MINIMIZATION OF W_P(R_P) COLOR DATA..........\n");
fprintf(stderr, "--------------------------------------------\n\n");
HOD.blue_fraction = 0.5857; /* <- Millenium fraction (sloan);; SDSS fraction -> 0.565; */
HOD.blue_fraction = 0.6555; /* <- Millenium fraction (B-V>0.8) */
HOD.blue_fraction = 0.565; /* SDSS -19,-20 */
HOD.blue_fraction = 0.492; /* SDSS -20,-21 */
HOD.blue_fraction = 0.379; /* SDSS -21 */
wp_color.ON = 1;
if(POWELL)
FTOL=1.0E-3;
else
FTOL=1.0E-5;
for(n=0,i=1;i<=7;++i)
{
n+=HOD.free[i];
if(!OUTPUT)continue;
printf("wp_min> free[%i] = %d\n",i,HOD.free[i]);
}
/* The parameters that govern the blue fraction aren't
* listed in the HOD.free array, so add them in.
* NB: There are four parameters for these two functions,
* one of which is fit by the number densities.
*/
n+=3;
if(OUTPUT)printf("wp_min> Number of free parameters: %d\n",n);
wp_color_input();
wp.ncf=n;
a=dvector(1,n);
if(POWELL)
pp=dmatrix(1,n,1,n);
else
pp=dmatrix(1,n+1,1,n);
yy=dvector(1,n+1);
initial_color_values(a,pp,yy);
if(POWELL)
{
if(OUTPUT)printf("wp_min> starting powell.\n");
powell(a,pp,n,FTOL,&niter,&chi2min,chi2_wp_color);
chi2min = chi2_wp_color(a);
}
else
{
if(OUTPUT)printf("wp_min> starting amoeba.\n");
amoeba(pp,yy,n,FTOL,chi2_wp_color,&niter);
for(i=1;i<=n;++i)a[i]=pp[1][i];
chi2min = chi2_wp_color(a);
}
s1=qromo(func_galaxy_bias,log(HOD.M_low),log(HOD.M_max),midpnt);
GALAXY_BIAS=s1/GALAXY_DENSITY;
printf("POWELL %e %e %e ",chi2min,HOD.M_min,HOD.fblue0_cen);
if(HOD.pdfc==7)
printf("%e ",HOD.M_cen_max);
for(i=1;i<=n;++i)printf("%e ",a[i]);
printf(" %f\n",GALAXY_BIAS);
/* Output the fit and the HOD curve.
*/
sprintf(aa,"%s.fit",Task.root_filename);
fp=fopen(aa,"w");
fprintf(fp,"%e %e %e ",chi2min,HOD.M_min,HOD.fblue0_cen);
if(HOD.pdfc==7)
fprintf(fp,"%e ",HOD.M_cen_max);
for(i=1;i<=n;++i)fprintf(fp,"%e ",a[i]);
fprintf(fp," %f\n",GALAXY_BIAS);
fclose(fp);
sprintf(aa,"%s.HOD",Task.root_filename);
fp=fopen(aa,"w");
dlogm=(log(HOD.M_max)-log(HOD.M_low))/99;
for(i=1;i<=100;++i)
{
m=exp((i-1)*dlogm)*HOD.M_low;
fprintf(fp,"%e %e %e %e\n",m,N_cen(m),N_sat(m),N_avg(m));
}
fclose(fp);
}
/*
* Fit to a gaussian
*/
int SimplexFitting::fitGaussian(double* fitting, double &err, int
howToInitParams)
// mIndex1 and mIndex2 are the
//starting index and endind index of the fitting range.
{
float pp[VARNUM+1][VARNUM+1], yy[VARNUM+1];
float da[VARNUM]={2.0, 2.0, 2.0, 2.0, 2.0};
float a[VARNUM]={0.7, 0.0, 0.2, 0.0, 0.0};
float min_a[VARNUM];
float ptol[VARNUM];
int nvar=4;
int iter, jmin, i, iter_counter;
float errmin;
float delfac=2.0;
float ftol2=5.0e-4;
float ptol2=5.0e-4;
float ftol1=1.0e-5;
float ptol1=1.0e-5;
//init params
if (howToInitParams == 0) {
a[0]=mRaw[mIndex1];
a[2]=1.414*(mIndex2-mIndex1)/(mDim-1);
} else {
a[0]=mRaw[(mIndex1+mIndex2)/2-1];
a[1]=0.5*(mIndex2-mIndex1)/(mDim-1); //sigma=0.3*(mIndex2-mIndex1)/(mDim-1);
a[2]=1.414*0.3*(mIndex2-mIndex1)/(mDim-1);
}
//find the range of fitting data;
double min, max;
double range;
if( (mIndex2-mIndex1)>0 ){
if( mRaw[mIndex1]> mRaw[mIndex1+1] ){
max=mRaw[mIndex1];
min=mRaw[mIndex1+1];
}else{
min=mRaw[mIndex1];
max=mRaw[mIndex1+1];
}
for(i=mIndex1+1;i<mIndex2;i++){
if( mRaw[i]>max )
max=mRaw[i];
else if( mRaw[i]<min)
min=mRaw[i];
}
range=max-min;
} else
range=mRaw[mIndex1];
iter_counter=0;
err=100000.0;
double scaling;
while(iter_counter<MAX_ITER){
amoebaInit(&pp[0][0], yy, VARNUM+1, nvar, delfac, ptol2, a, da,
&SimplexFitting::funk, ptol);
amoeba(&pp[0][0], yy, VARNUM+1, nvar, ftol2, &SimplexFitting::funk,
&iter, ptol, &jmin);
for (i = 0; i < nvar; i++)
a[i] = pp[i][jmin];
amoebaInit(&pp[0][0], yy, VARNUM+1, nvar, delfac, ptol1, a, da,
&SimplexFitting::funk, ptol);
amoeba(&pp[0][0], yy, VARNUM+1, nvar, ftol1, &SimplexFitting::funk,
&iter, ptol, &jmin);
for (i = 0; i < nvar; i++)
a[i] = pp[i][jmin];
funk(a, &errmin);
if( (errmin<err || errmin<MIN_ERROR*range) && a[0]>0.0){
err=errmin;
for(i=0;i<nvar;i++)
min_a[i]=a[i];
}
iter_counter++;
if(errmin<MIN_ERROR*range && a[0]>0.0)
break;
//re-initialize parameters
if(iter_counter%2)
scaling=0.5*iter_counter/(MAX_ITER-1);
else
scaling=-0.5*iter_counter/(MAX_ITER-1);
a[0]=(1+scaling)*mRaw[mIndex1];
if(howToInitParams==0)
a[1]=0.0;
else
a[1]=0.5*(mIndex2-mIndex1)/(mDim-1);
a[2]=(1+scaling)*1.414*(mIndex2-mIndex1)/(mDim-1);
a[3]=0.0;
}// iteration
for (i=0; i < mDim; i++) {
fitting[i]=min_a[0]*exp( -((float)(i-mIndex1)/(mDim-1)-min_a[1])*
((float)(i-mIndex1)/(mDim-1)-min_a[1])/(min_a[2]*min_a[2])) +min_a[3];
}
if( debugLevel>=3){
printf("Iteration Num=%d threshold=%f Simplex fitting parameters for \
range %d to %d are:\n", iter_counter, MIN_ERROR*range, mIndex1, mIndex2);
printf("Fitting error=%f\t a[0]=%f\t a[1]=%f\t a[2]=%f\t a[3]=%f\n",err,
min_a[0], min_a[1], min_a[2], min_a[3]);
}
// Main programm
int main (int argc, char *argv[]) {
Search_settings sett;
Command_line_opts opts;
Search_range s_range;
Aux_arrays aux_arr;
double *F; // F-statistic array
int i, j, r, c, a, b, g;
int d, o, m, k;
int bins = 2, ROW, dim = 4; // neighbourhood of point will be divide into defined number of bins
double pc[4]; // % define neighbourhood around each parameter for initial grid
double pc2[4]; // % define neighbourhood around each parameter for direct maximum search (MADS & Simplex)
double tol = 1e-10;
// double delta = 1e-5; // initial step in MADS function
// double *results; // Vector with results from Fstatnet function
// double *maximum; // True maximum of Fstat
// double results_max[11];
double s1, s2, s3, s4;
double sgnlo[4];
double **arr; // arr[ROW][COL], arrg[ROW][COL];
double nSource[3];
double sinalt, cosalt, sindelt, cosdelt;
double F_min;
char path[512];
double x, y;
ROW = pow((bins+1),4);
#ifdef YEPPP
yepLibrary_Init();
Yep64f *results_max = (Yep64f*)malloc(sizeof(Yep64f)*11);
Yep64f *results_first = (Yep64f*)malloc(sizeof(Yep64f)*11);
Yep64f *results = (Yep64f*)malloc(sizeof(Yep64f)*11);
Yep64f *maximum = (Yep64f*)malloc(sizeof(Yep64f)*11);
// Yep64f *sgnlo = (Yep64f*)malloc(sizeof(Yep64f)*4);
// Yep64f *nSource = (Yep64f*)malloc(sizeof(Yep64f)*3);
Yep64f *mean = (Yep64f*)malloc(sizeof(Yep64f)*4);
enum YepStatus status;
#endif
pc[0] = 0.015;
pc[1] = 0.015;
pc[2] = 0.015;
pc[3] = 0.015;
for (i = 0; i < 4; i++){
pc2[i] = 2*pc[i]/bins;
}
// Time tests
double tdiff;
clock_t tstart, tend;
// Command line options
handle_opts(&sett, &opts, argc, argv);
// Output data handling
/* struct stat buffer;
if (stat(opts.prefix, &buffer) == -1) {
if (errno == ENOENT) {
// Output directory apparently does not exist, try to create one
if(mkdir(opts.prefix, S_IRWXU | S_IRGRP | S_IXGRP
| S_IROTH | S_IXOTH) == -1) {
perror (opts.prefix);
return 1;
}
} else { // can't access output directory
perror (opts.prefix);
return 1;
}
}
*/
sprintf(path, "%s/candidates.coi", opts.dtaprefix);
//Glue function
if(strlen(opts.glue)) {
glue(&opts);
sprintf(opts.dtaprefix, "./data_total");
sprintf(opts.dtaprefix, "%s/followup_total_data", opts.prefix);
opts.ident = 000;
}
FILE *coi;
int z;
if ((coi = fopen(path, "r")) != NULL) {
// while(!feof(coi)) {
/* if(!fread(&w, sizeof(unsigned short int), 1, coi)) { break; }
fread(&mean, sizeof(float), 5, coi);
fread(&fra, sizeof(unsigned short int), w, coi);
fread(&ops, sizeof(int), w, coi);
if((fread(&mean, sizeof(float), 4, coi)) == 4){
*/
while(fscanf(coi, "%le %le %le %le", &mean[0], &mean[1], &mean[2], &mean[3]) == 4){
//Time test
// tstart = clock();
arr = matrix(ROW, 4);
//Function neighbourhood - generating grid around point
arr = neigh(mean, pc, bins);
// Output data handling
/* struct stat buffer;
if (stat(opts.prefix, &buffer) == -1) {
if (errno == ENOENT) {
// Output directory apparently does not exist, try to create one
if(mkdir(opts.prefix, S_IRWXU | S_IRGRP | S_IXGRP
| S_IROTH | S_IXOTH) == -1) {
perror (opts.prefix);
return 1;
}
}
else { // can't access output directory
perror (opts.prefix);
return 1;
}
}
*/
// Grid data
if(strlen(opts.addsig)) {
read_grid(&sett, &opts);
}
// Search settings
search_settings(&sett);
// Detector network settings
detectors_settings(&sett, &opts);
// Array initialization
init_arrays(&sett, &opts, &aux_arr, &F);
// Amplitude modulation functions for each detector
for(i=0; i<sett.nifo; i++) rogcvir(&ifo[i]);
// Adding signal from file
if(strlen(opts.addsig)) {
add_signal(&sett, &opts, &aux_arr, &s_range);
}
// Setting number of using threads (not required)
omp_set_num_threads(1);
results_max[5] = 0.;
// ifo zostaje shared
// ifo....shft i ifo....xdatm{a,b] prerobić na lokalne tablice w fstatnet
// w regionie parallel wprowadzić tablice private aa i bb ; alokować i przekazywać je jako argumenty do fstatnet i amoeba
// Main loop - over all parameters + parallelisation
#pragma omp parallel default(shared) private(d, i, sgnlo, sinalt, cosalt, sindelt, cosdelt, nSource, results, maximum)
{
double **sigaa, **sigbb; // aa[nifo][N]
sigaa = matrix(sett.nifo, sett.N);
sigbb = matrix(sett.nifo, sett.N);
#pragma omp for
for (d = 0; d < ROW; ++d){
for (i = 0; i < 4; i++){
sgnlo[i] = arr[d][i];
// sgnlo[i] = mean[i];
}
sinalt = sin(sgnlo[3]);
cosalt = cos(sgnlo[3]);
sindelt = sin(sgnlo[2]);
cosdelt = cos(sgnlo[2]);
nSource[0] = cosalt*cosdelt;
nSource[1] = sinalt*cosdelt;
nSource[2] = sindelt;
for (i = 0; i < sett.nifo; ++i){
modvir(sinalt, cosalt, sindelt, cosdelt,
sett.N, &ifo[i], &aux_arr, sigaa[i], sigbb[i]);
}
// F-statistic in given point
results = Fstatnet(&sett, sgnlo, nSource, sigaa, sigbb);
//printf("Fstatnet: %le %le %le %le %le %le\n", results[6], results[7], results[8], results[9], results[5], results[4]);
#pragma omp critical
if(results[5] < results_max[5]){
for (i = 0; i < 11; i++){
results_max[i] = results[i];
}
}
// Maximum search using simplex algorithm
if(opts.simplex_flag){
// puts("Simplex");
maximum = amoeba(&sett, &aux_arr, sgnlo, nSource, results, dim, tol, pc2, sigaa, sigbb);
printf("Amoeba: %le %le %le %le %le %le\n", maximum[6], maximum[7], maximum[8], maximum[9], maximum[5], maximum[4]);
// Maximum value in points searching
#pragma omp critical
if(maximum[5] < results_max[5]){
for (i = 0; i < 11; i++){
results_max[i] = maximum[i];
}
}
} //simplex
} // d - main outside loop
free_matrix(sigaa, sett.nifo, sett.N);
free_matrix(sigbb, sett.nifo, sett.N);
} //pragma
for(g = 0; g < 11; g++) results_first[g] = results_max[g];
// Maximum search using MADS algorithm
if(opts.mads_flag) {
// puts("MADS");
maximum = MADS(&sett, &aux_arr, results_max, mean, tol, pc2, bins);
}
//Time test
// tend = clock();
// tdiff = (tend - tstart)/(double)CLOCKS_PER_SEC;
printf("%le %le %le %le %le %le\n", results_max[6], results_max[7], results_max[8], results_max[9], results_max[5], results_max[4]);
} // while fread coi
// }
} //if coi
else {
perror (path);
return 1;
}
// Output information
/* puts("**********************************************************************");
printf("*** Maximum value of F-statistic for grid is : (-)%.8le ***\n", -results_first[5]);
printf("Sgnlo: %.8le %.8le %.8le %.8le\n", results_first[6], results_first[7], results_first[8], results_first[9]);
printf("Amplitudes: %.8le %.8le %.8le %.8le\n", results_first[0], results_first[1], results_first[2], results_first[3]);
printf("Signal-to-noise ratio: %.8le\n", results_first[4]);
printf("Signal-to-noise ratio from estimated amplitudes (for h0 = 1): %.8le\n", results_first[10]);
puts("**********************************************************************");
if((opts.mads_flag)||(opts.simplex_flag)){
printf("*** True maximum is : (-)%.8le ***\n", -maximum[5]);
printf("Sgnlo for true maximum: %.8le %.8le %.8le %.8le\n", maximum[6], maximum[7], maximum[8], maximum[9]);
printf("Amplitudes for true maximum: %.8le %.8le %.8le %.8le\n", maximum[0], maximum[1], maximum[2], maximum[3]);
printf("Signal-to-noise ratio for true maximum: %.8le\n", maximum[4]);
printf("Signal-to-noise ratio from estimated amplitudes (for h0 = 1) for true maximum: %.8le\n", maximum[10]);
puts("**********************************************************************");
}*/
// Cleanup & memory free
free(results_max);
free(results_first);
free(results);
free(maximum);
free(mean);
free_matrix(arr, ROW, 4);
cleanup_followup(&sett, &opts, &s_range, &aux_arr, F);
return 0;
}
/* least squares solver.
* returns number of iterations if solution converged, else -1.
*/
int
lstsqr (
double (*chisqr)(double p[]), /* function to evaluate chisqr with at p */
double params0[], /* in: guess: back: best */
double params1[], /* second guess to set characteristic scale */
int np, /* entries in params0[] and params1[] */
double ftol) /* desired fractional tolerance */
{
/* set up the necessary temp arrays and call the amoeba() multivariat
* solver. amoeba() was evidently transliterated from fortran because
* it indexes the arrays starting at [1].
*/
double **p; /* np+1 rows, np columns (+1 due to 1-based) */
double *y; /* np columns ( " ) */
int iter;
int i, j;
int ret;
/* save the caller's 0-based chi sqr function */
chisqr_0based = chisqr;
/* fill p[][] with np+1 rows of each set of guesses of the np params.
* fill y[] with chisqr() for each of these sets.
* chisqr() is actually the function the amoeba is minimizing.
*/
p = (double **) malloc ((np+2)*sizeof(double *));
y = (double *) malloc ((np+2)*sizeof(double));
for (i = 1; i <= np+1; i++) {
double *pi = p[i] = (double *) malloc ((np+1)*sizeof(double));
for (j = 1; j <= np; j++)
pi[j] = (j == i-1) ? params1[j-1] : params0[j-1];
y[i] = chisqr_1based(pi);
}
/* solve */
ret = amoeba (p, y, np, ftol, chisqr_1based, &iter);
/* on return, each row i of p has solutions at p[1..np+1][i].
* average them?? pick first one??
*/
if (ret == 0) {
for (i = 0; i < np; i++) {
#ifdef USE_AVERAGE
double sum = 0.0;
for (j = 1; j <= np+1; j++)
sum += p[j][i+1];
params0[i] = sum/(np+1);
#else
params0[i] = p[1][i+1];
#endif
}
ret = iter;
}
/* free the temp space */
for (i = 1; i < np+2; i++)
free ((void *)p[i]);
free ((void *)p);
free ((void *)y);
return (ret);
}
int main(int argc, char* argv[])
{
double *params, **xi, **pxi, mlk;
double *y;
long n, iter, i, j, nfunk;
char* fileName = "armasq.dat";
xobs = loadmat(fileName, &nn, &mm);
nn *= mm;
if(argc < 3)
{
printf("Need to specify p an q\n");
exit(0);
}
else
{
p = atoi(argv[1]);
q= atoi(argv[2]);
// printf("%d %d\n", p, q);
}
n = p + q;
if(argc > 5)
{
fileName = argv[4];
}
if (argc < 4)
{
printf(
"need to specify the minimization method:\n\tarma p\tfor Powell\n\tarma n\tfor Nelder and Mead\n");
exit(0);
}
else if (argv[3][0] == 'p')
{
printf("Using Powell method\n");
params = vector(n);
xi = (double**) malloc(n * sizeof(double*));
//intialize the initial point and directions
for (i = 0; i < n; i++)
{
params[i] = 0.0;
xi[i] = vector(n);
for (j = 0; j < n; j++)
{
if (i == j)
xi[i][j] = .1;
else
xi[i][j] = 0;
}
}
powell(params, xi, n, TOLERANCE, &iter, &mlk, armalk_wrapper);
printf("Maximum LK %E\t after %ld iterations with parameters:\n", 1/mlk,
iter);
for (i = 0; i < n; i++)
{
printf("%E\n", params[i]);
}
free_vector(params);
for (i = 0; i < n; i++)
{
free_vector(xi[i]);
}
free(xi);
}
else if (argv[3][0] == 'n')
{
printf("Using Nelder & Mead method\n");
y = vector(n+1);
xi = (double**) malloc((n + 1) * sizeof(double*));
//intialize the initial points
for (i = 0; i < n + 1; i++)
{
xi[i] = vector(n);
for (j = 0; j < n; j++)
{
if (i == j)
xi[i][j] = .1;
else
xi[i][j] = 0;
}
y[i] = armalk_wrapper(xi[i]);
}
amoeba(xi, y, n, TOLERANCE, armalk_wrapper, &nfunk);
printf("After %ld function evaluations, the converged new points are:\n", nfunk);
for (i = 0; i < n+1; i++)
{
printf("\tPoint %ld: MLK = %E, parameters:", i, 1/y[i]);
for (j = 0; j < n; j++)
printf(" %E ", xi[i][j]);
printf("\n");
}
free_vector(y);
for (i = 0; i < n + 1; i++)
{
free_vector(xi[i]);
}
free(xi);
}
free(xobs);
return 1;
}
double max_rz_arr(fcomplex * data, int numdata, double rin, double zin,
double *rout, double *zout, rderivs * derivs)
/* Return the Fourier frequency and Fourier f-dot that */
/* maximizes the power. */
{
double y[3], x[3][2], step = 0.4;
float locpow;
int numeval;
maxdata = data;
nummaxdata = numdata;
locpow = get_localpower3d(data, numdata, rin, zin, 0.0);
/* Now prep the maximization at LOWACC for speed */
/* Use a slightly larger working value for 'z' just incase */
/* the true value of z is a little larger than z. This */
/* keeps a little more accuracy. */
max_kern_half_width = z_resp_halfwidth(fabs(zin) + 4.0, LOWACC);
/* Initialize the starting simplex */
x[0][0] = rin - step;
x[0][1] = zin / ZSCALE - step;
x[1][0] = rin - step;
x[1][1] = zin / ZSCALE + step;
x[2][0] = rin + step;
x[2][1] = zin / ZSCALE;
/* Initialize the starting function values */
y[0] = power_call_rz(x[0]);
y[1] = power_call_rz(x[1]);
y[2] = power_call_rz(x[2]);
/* Call the solver: */
numeval = 0;
amoeba(x, y, 1.0e-7, power_call_rz, &numeval);
/* Restart at minimum using HIGHACC to get a better result */
max_kern_half_width = z_resp_halfwidth(fabs(x[0][1]) + 4.0, HIGHACC);
/* Re-Initialize some of the starting simplex */
x[1][0] = x[0][0] + 0.01;
x[1][1] = x[0][1];
x[2][0] = x[0][0];
x[2][1] = x[0][1] + 0.01;
/* Re-Initialize the starting function values */
y[0] = power_call_rz(x[0]);
y[1] = power_call_rz(x[1]);
y[2] = power_call_rz(x[2]);
/* Call the solver: */
numeval = 0;
amoeba(x, y, 1.0e-10, power_call_rz, &numeval);
/* The following calculates derivatives at the peak */
*rout = x[0][0];
*zout = x[0][1] * ZSCALE;
locpow = get_localpower3d(data, numdata, *rout, *zout, 0.0);
get_derivs3d(data, numdata, *rout, *zout, 0.0, locpow, derivs);
return -y[0];
}
void max_rz_arr_harmonics(fcomplex* data[], int num_harmonics,
int r_offset[],
int numdata, double rin, double zin,
double *rout, double *zout, rderivs derivs[],
double power[])
/* Return the Fourier frequency and Fourier f-dot that */
/* maximizes the power. */
{
double y[3], x[3][2], step = 0.4;
float *locpow;
int numeval;
int i;
locpow = gen_fvect(num_harmonics);
maxlocpow = gen_fvect(num_harmonics);
maxr_offset = r_offset;
maxdata_harmonics = data;
//FIXME: z needs to be multiplied by i everywhere
for (i=1;i<=num_harmonics;i++) {
locpow[i-1] = get_localpower3d(data[i-1], numdata, (r_offset[i-1]+rin)*i-r_offset[i-1], zin*i, 0.0);
maxlocpow[i-1]=locpow[i-1];
}
nummaxdata = numdata;
max_num_harmonics = num_harmonics;
/* Now prep the maximization at LOWACC for speed */
/* Use a slightly larger working value for 'z' just incase */
/* the true value of z is a little larger than z. This */
/* keeps a little more accuracy. */
max_kern_half_width = z_resp_halfwidth(fabs(zin*num_harmonics) + 4.0, LOWACC);
/* Initialize the starting simplex */
x[0][0] = rin - step;
x[0][1] = zin / ZSCALE - step;
x[1][0] = rin - step;
x[1][1] = zin / ZSCALE + step;
x[2][0] = rin + step;
x[2][1] = zin / ZSCALE;
/* Initialize the starting function values */
y[0] = power_call_rz_harmonics(x[0]);
y[1] = power_call_rz_harmonics(x[1]);
y[2] = power_call_rz_harmonics(x[2]);
/* Call the solver: */
numeval = 0;
amoeba(x, y, 1.0e-7, power_call_rz_harmonics, &numeval);
/* Restart at minimum using HIGHACC to get a better result */
max_kern_half_width = z_resp_halfwidth(fabs(x[0][1]*num_harmonics) + 4.0, HIGHACC);
/* Re-Initialize some of the starting simplex */
x[1][0] = x[0][0] + 0.01;
x[1][1] = x[0][1];
x[2][0] = x[0][0];
x[2][1] = x[0][1] + 0.01;
/* Re-Initialize the starting function values */
y[0] = power_call_rz_harmonics(x[0]);
y[1] = power_call_rz_harmonics(x[1]);
y[2] = power_call_rz_harmonics(x[2]);
/* Call the solver: */
numeval = 0;
amoeba(x, y, 1.0e-10, power_call_rz_harmonics, &numeval);
/* The following calculates derivatives at the peak */
*rout = x[0][0];
*zout = x[0][1] * ZSCALE;
for (i=1; i<=num_harmonics; i++) {
locpow[i-1] = get_localpower3d(data[i-1], numdata, (r_offset[i-1]+*rout)*i-r_offset[i-1], (*zout)*i, 0.0);
x[0][0] = (r_offset[i-1]+*rout)*i-r_offset[i-1];
x[0][1] = *zout/ZSCALE * i;
maxdata = data[i-1];
power[i-1] = -power_call_rz(x[0]);
get_derivs3d(data[i-1], numdata, (r_offset[i-1]+*rout)*i-r_offset[i-1], (*zout)*i, 0.0, locpow[i-1], &(derivs[i-1]));
}
vect_free(locpow);
vect_free(maxlocpow);
}
