vcglib/wrap/newuoa/include/newuoa.h

1729 lines
62 KiB
C++

/*
This is NEWUOA for unconstrained minimization. The codes were written
by Powell in Fortran and then translated to C with f2c. I further
modified the code to make it independent of libf2c and f2c.h. Please
refer to "The NEWUOA software for unconstrained optimization without
derivatives", which is available at www.damtp.cam.ac.uk, for more
information.
*/
/*
The original fortran codes are distributed without restrictions. The
C++ codes are distributed under MIT license.
*/
/* The MIT License
Copyright (c) 2004, by M.J.D. Powell <mjdp@cam.ac.uk>
2008, by Attractive Chaos <attractivechaos@aol.co.uk>
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
*/
#ifndef AC_NEWUOA_HH_
#define AC_NEWUOA_HH_
#include <math.h>
#include <stdlib.h>
#include <stdio.h>
//PARAMETER DECREASE WAS 0.1
#define DELTA_DECREASE 0.5
#define RHO_DECREASE 0.5
/*most probably:
TYPE is float or double
Func is defined as double func(int n, double *x);
or better as
class Func {
double operator()(int n, double *x);
};
where n is the number of parameters and x the vector parameter
r_start stands unknown...
*/
template<class TYPE, class Func>
TYPE min_newuoa(int n, TYPE *x, Func &func, TYPE r_start=1e7, TYPE tol=1e-8, int max_iter=5000);
template<class TYPE, class Func>
static int biglag_(int n, int npt, TYPE *xopt, TYPE *xpt, TYPE *bmat, TYPE *zmat, int *idz,
int *ndim, int *knew, TYPE *delta, TYPE *d__, TYPE *alpha, TYPE *hcol, TYPE *gc,
TYPE *gd, TYPE *s, TYPE *w, Func & /*func*/)
{
/* N is the number of variables. NPT is the number of interpolation
* equations. XOPT is the best interpolation point so far. XPT
* contains the coordinates of the current interpolation
* points. BMAT provides the last N columns of H. ZMAT and IDZ give
* a factorization of the first NPT by NPT submatrix of H. NDIM is
* the first dimension of BMAT and has the value NPT+N. KNEW is the
* index of the interpolation point that is going to be moved. DEBLLTA
* is the current trust region bound. D will be set to the step from
* XOPT to the new point. ABLLPHA will be set to the KNEW-th diagonal
* element of the H matrix. HCOBLL, GC, GD, S and W will be used for
* working space. */
/* The step D is calculated in a way that attempts to maximize the
* modulus of BLLFUNC(XOPT+D), subject to the bound ||D|| .BLLE. DEBLLTA,
* where BLLFUNC is the KNEW-th BLLagrange function. */
int xpt_dim1, xpt_offset, bmat_dim1, bmat_offset, zmat_dim1, zmat_offset,
i__1, i__2, i__, j, k, iu, nptm, iterc, isave;
TYPE sp, ss, cf1, cf2, cf3, cf4, cf5, dhd, cth, tau, sth, sum, temp, step,
angle, scale, denom, delsq, tempa, tempb, twopi, taubeg, tauold, taumax,
d__1, dd, gg;
/* Parameter adjustments */
tempa = tempb = 0.0;
zmat_dim1 = npt;
zmat_offset = 1 + zmat_dim1;
zmat -= zmat_offset;
xpt_dim1 = npt;
xpt_offset = 1 + xpt_dim1;
xpt -= xpt_offset;
--xopt;
bmat_dim1 = *ndim;
bmat_offset = 1 + bmat_dim1;
bmat -= bmat_offset;
--d__; --hcol; --gc; --gd; --s; --w;
/* Function Body */
twopi = 2.0 * M_PI;
delsq = *delta * *delta;
nptm = npt - n - 1;
/* Set the first NPT components of HCOBLL to the leading elements of
* the KNEW-th column of H. */
iterc = 0;
i__1 = npt;
for (k = 1; k <= i__1; ++k) hcol[k] = 0;
i__1 = nptm;
for (j = 1; j <= i__1; ++j) {
temp = zmat[*knew + j * zmat_dim1];
if (j < *idz) temp = -temp;
i__2 = npt;
for (k = 1; k <= i__2; ++k)
hcol[k] += temp * zmat[k + j * zmat_dim1];
}
*alpha = hcol[*knew];
/* Set the unscaled initial direction D. Form the gradient of BLLFUNC
* atXOPT, and multiply D by the second derivative matrix of
* BLLFUNC. */
dd = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
d__[i__] = xpt[*knew + i__ * xpt_dim1] - xopt[i__];
gc[i__] = bmat[*knew + i__ * bmat_dim1];
gd[i__] = 0;
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
}
i__2 = npt;
for (k = 1; k <= i__2; ++k) {
temp = 0;
sum = 0;
i__1 = n;
for (j = 1; j <= i__1; ++j) {
temp += xpt[k + j * xpt_dim1] * xopt[j];
sum += xpt[k + j * xpt_dim1] * d__[j];
}
temp = hcol[k] * temp;
sum = hcol[k] * sum;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
gc[i__] += temp * xpt[k + i__ * xpt_dim1];
gd[i__] += sum * xpt[k + i__ * xpt_dim1];
}
}
/* Scale D and GD, with a sign change if required. Set S to another
* vector in the initial two dimensional subspace. */
gg = sp = dhd = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Computing 2nd power */
d__1 = gc[i__];
gg += d__1 * d__1;
sp += d__[i__] * gc[i__];
dhd += d__[i__] * gd[i__];
}
scale = *delta / sqrt(dd);
if (sp * dhd < 0) scale = -scale;
temp = 0;
if (sp * sp > dd * .99 * gg) temp = 1.0;
tau = scale * (fabs(sp) + 0.5 * scale * fabs(dhd));
if (gg * delsq < tau * .01 * tau) temp = 1.0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = scale * d__[i__];
gd[i__] = scale * gd[i__];
s[i__] = gc[i__] + temp * gd[i__];
}
/* Begin the iteration by overwriting S with a vector that has the
* required length and direction, except that termination occurs if
* the given D and S are nearly parallel. */
for (iterc = 0; iterc != n; ++iterc) {
dd = sp = ss = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
sp += d__[i__] * s[i__];
/* Computing 2nd power */
d__1 = s[i__];
ss += d__1 * d__1;
}
temp = dd * ss - sp * sp;
if (temp <= dd * 1e-8 * ss) return 0;
denom = sqrt(temp);
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
s[i__] = (dd * s[i__] - sp * d__[i__]) / denom;
w[i__] = 0;
}
/* Calculate the coefficients of the objective function on the
* circle, beginning with the multiplication of S by the second
* derivative matrix. */
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
sum = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j)
sum += xpt[k + j * xpt_dim1] * s[j];
sum = hcol[k] * sum;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
w[i__] += sum * xpt[k + i__ * xpt_dim1];
}
cf1 = cf2 = cf3 = cf4 = cf5 = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
cf1 += s[i__] * w[i__];
cf2 += d__[i__] * gc[i__];
cf3 += s[i__] * gc[i__];
cf4 += d__[i__] * gd[i__];
cf5 += s[i__] * gd[i__];
}
cf1 = 0.5 * cf1;
cf4 = 0.5 * cf4 - cf1;
/* Seek the value of the angle that maximizes the modulus of TAU. */
taubeg = cf1 + cf2 + cf4;
taumax = tauold = taubeg;
isave = 0;
iu = 49;
temp = twopi / (TYPE) (iu + 1);
i__2 = iu;
for (i__ = 1; i__ <= i__2; ++i__) {
angle = (TYPE) i__ *temp;
cth = cos(angle);
sth = sin(angle);
tau = cf1 + (cf2 + cf4 * cth) * cth + (cf3 + cf5 * cth) * sth;
if (fabs(tau) > fabs(taumax)) {
taumax = tau;
isave = i__;
tempa = tauold;
} else if (i__ == isave + 1) tempb = tau;
tauold = tau;
}
if (isave == 0) tempa = tau;
if (isave == iu) tempb = taubeg;
step = 0;
if (tempa != tempb) {
tempa -= taumax;
tempb -= taumax;
step = 0.5 * (tempa - tempb) / (tempa + tempb);
}
angle = temp * ((TYPE) isave + step);
/* Calculate the new D and GD. Then test for convergence. */
cth = cos(angle);
sth = sin(angle);
tau = cf1 + (cf2 + cf4 * cth) * cth + (cf3 + cf5 * cth) * sth;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
d__[i__] = cth * d__[i__] + sth * s[i__];
gd[i__] = cth * gd[i__] + sth * w[i__];
s[i__] = gc[i__] + gd[i__];
}
if (fabs(tau) <= fabs(taubeg) * 1.1) return 0;
}
return 0;
}
template<class TYPE>
static int bigden_(int n, int npt, TYPE *xopt, TYPE *xpt, TYPE *bmat, TYPE *zmat, int *idz,
int *ndim, int *kopt, int *knew, TYPE *d__, TYPE *w, TYPE *vlag, TYPE *beta,
TYPE *s, TYPE *wvec, TYPE *prod)
{
/* N is the number of variables.
* NPT is the number of interpolation equations.
* XOPT is the best interpolation point so far.
* XPT contains the coordinates of the current interpolation points.
* BMAT provides the last N columns of H.
* ZMAT and IDZ give a factorization of the first NPT by NPT
submatrix of H.
* NDIM is the first dimension of BMAT and has the value NPT+N.
* KOPT is the index of the optimal interpolation point.
* KNEW is the index of the interpolation point that is going to be
moved.
* D will be set to the step from XOPT to the new point, and on
entry it should be the D that was calculated by the last call of
BIGBDLAG. The length of the initial D provides a trust region bound
on the final D.
* W will be set to Wcheck for the final choice of D.
* VBDLAG will be set to Theta*Wcheck+e_b for the final choice of D.
* BETA will be set to the value that will occur in the updating
formula when the KNEW-th interpolation point is moved to its new
position.
* S, WVEC, PROD and the private arrays DEN, DENEX and PAR will be
used for working space.
* D is calculated in a way that should provide a denominator with a
large modulus in the updating formula when the KNEW-th
interpolation point is shifted to the new position XOPT+D. */
int xpt_dim1, xpt_offset, bmat_dim1, bmat_offset, zmat_dim1, zmat_offset,
wvec_dim1, wvec_offset, prod_dim1, prod_offset, i__1, i__2, i__, j, k,
isave, iterc, jc, ip, iu, nw, ksav, nptm;
TYPE dd, d__1, ds, ss, den[9], par[9], tau, sum, diff, temp, step,
alpha, angle, denex[9], tempa, tempb, tempc, ssden, dtest, xoptd,
twopi, xopts, denold, denmax, densav, dstemp, sumold, sstemp, xoptsq;
/* Parameter adjustments */
zmat_dim1 = npt;
zmat_offset = 1 + zmat_dim1;
zmat -= zmat_offset;
xpt_dim1 = npt;
xpt_offset = 1 + xpt_dim1;
xpt -= xpt_offset;
--xopt;
prod_dim1 = *ndim;
prod_offset = 1 + prod_dim1;
prod -= prod_offset;
wvec_dim1 = *ndim;
wvec_offset = 1 + wvec_dim1;
wvec -= wvec_offset;
bmat_dim1 = *ndim;
bmat_offset = 1 + bmat_dim1;
bmat -= bmat_offset;
--d__; --w; --vlag; --s;
/* Function Body */
twopi = atan(1.0) * 8.;
nptm = npt - n - 1;
/* Store the first NPT elements of the KNEW-th column of H in W(N+1)
* to W(N+NPT). */
i__1 = npt;
for (k = 1; k <= i__1; ++k) w[n + k] = 0;
i__1 = nptm;
for (j = 1; j <= i__1; ++j) {
temp = zmat[*knew + j * zmat_dim1];
if (j < *idz) temp = -temp;
i__2 = npt;
for (k = 1; k <= i__2; ++k)
w[n + k] += temp * zmat[k + j * zmat_dim1];
}
alpha = w[n + *knew];
/* The initial search direction D is taken from the last call of
* BIGBDLAG, and the initial S is set below, usually to the direction
* from X_OPT to X_KNEW, but a different direction to an
* interpolation point may be chosen, in order to prevent S from
* being nearly parallel to D. */
dd = ds = ss = xoptsq = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
s[i__] = xpt[*knew + i__ * xpt_dim1] - xopt[i__];
ds += d__[i__] * s[i__];
/* Computing 2nd power */
d__1 = s[i__];
ss += d__1 * d__1;
/* Computing 2nd power */
d__1 = xopt[i__];
xoptsq += d__1 * d__1;
}
if (ds * ds > dd * .99 * ss) {
ksav = *knew;
dtest = ds * ds / ss;
i__2 = npt;
for (k = 1; k <= i__2; ++k) {
if (k != *kopt) {
dstemp = 0;
sstemp = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
diff = xpt[k + i__ * xpt_dim1] - xopt[i__];
dstemp += d__[i__] * diff;
sstemp += diff * diff;
}
if (dstemp * dstemp / sstemp < dtest) {
ksav = k;
dtest = dstemp * dstemp / sstemp;
ds = dstemp;
ss = sstemp;
}
}
}
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
s[i__] = xpt[ksav + i__ * xpt_dim1] - xopt[i__];
}
ssden = dd * ss - ds * ds;
iterc = 0;
densav = 0;
/* Begin the iteration by overwriting S with a vector that has the
* required length and direction. */
BDL70:
++iterc;
temp = 1.0 / sqrt(ssden);
xoptd = xopts = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
s[i__] = temp * (dd * s[i__] - ds * d__[i__]);
xoptd += xopt[i__] * d__[i__];
xopts += xopt[i__] * s[i__];
}
/* Set the coefficients of the first 2.0 terms of BETA. */
tempa = 0.5 * xoptd * xoptd;
tempb = 0.5 * xopts * xopts;
den[0] = dd * (xoptsq + 0.5 * dd) + tempa + tempb;
den[1] = 2.0 * xoptd * dd;
den[2] = 2.0 * xopts * dd;
den[3] = tempa - tempb;
den[4] = xoptd * xopts;
for (i__ = 6; i__ <= 9; ++i__) den[i__ - 1] = 0;
/* Put the coefficients of Wcheck in WVEC. */
i__2 = npt;
for (k = 1; k <= i__2; ++k) {
tempa = tempb = tempc = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
tempa += xpt[k + i__ * xpt_dim1] * d__[i__];
tempb += xpt[k + i__ * xpt_dim1] * s[i__];
tempc += xpt[k + i__ * xpt_dim1] * xopt[i__];
}
wvec[k + wvec_dim1] = 0.25 * (tempa * tempa + tempb * tempb);
wvec[k + (wvec_dim1 << 1)] = tempa * tempc;
wvec[k + wvec_dim1 * 3] = tempb * tempc;
wvec[k + (wvec_dim1 << 2)] = 0.25 * (tempa * tempa - tempb * tempb);
wvec[k + wvec_dim1 * 5] = 0.5 * tempa * tempb;
}
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
ip = i__ + npt;
wvec[ip + wvec_dim1] = 0;
wvec[ip + (wvec_dim1 << 1)] = d__[i__];
wvec[ip + wvec_dim1 * 3] = s[i__];
wvec[ip + (wvec_dim1 << 2)] = 0;
wvec[ip + wvec_dim1 * 5] = 0;
}
/* Put the coefficents of THETA*Wcheck in PROD. */
for (jc = 1; jc <= 5; ++jc) {
nw = npt;
if (jc == 2 || jc == 3) nw = *ndim;
i__2 = npt;
for (k = 1; k <= i__2; ++k) prod[k + jc * prod_dim1] = 0;
i__2 = nptm;
for (j = 1; j <= i__2; ++j) {
sum = 0;
i__1 = npt;
for (k = 1; k <= i__1; ++k) sum += zmat[k + j * zmat_dim1] * wvec[k + jc * wvec_dim1];
if (j < *idz) sum = -sum;
i__1 = npt;
for (k = 1; k <= i__1; ++k)
prod[k + jc * prod_dim1] += sum * zmat[k + j * zmat_dim1];
}
if (nw == *ndim) {
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
sum = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j)
sum += bmat[k + j * bmat_dim1] * wvec[npt + j + jc * wvec_dim1];
prod[k + jc * prod_dim1] += sum;
}
}
i__1 = n;
for (j = 1; j <= i__1; ++j) {
sum = 0;
i__2 = nw;
for (i__ = 1; i__ <= i__2; ++i__)
sum += bmat[i__ + j * bmat_dim1] * wvec[i__ + jc * wvec_dim1];
prod[npt + j + jc * prod_dim1] = sum;
}
}
/* Include in DEN the part of BETA that depends on THETA. */
i__1 = *ndim;
for (k = 1; k <= i__1; ++k) {
sum = 0;
for (i__ = 1; i__ <= 5; ++i__) {
par[i__ - 1] = 0.5 * prod[k + i__ * prod_dim1] * wvec[k + i__ * wvec_dim1];
sum += par[i__ - 1];
}
den[0] = den[0] - par[0] - sum;
tempa = prod[k + prod_dim1] * wvec[k + (wvec_dim1 << 1)] + prod[k + (
prod_dim1 << 1)] * wvec[k + wvec_dim1];
tempb = prod[k + (prod_dim1 << 1)] * wvec[k + (wvec_dim1 << 2)] +
prod[k + (prod_dim1 << 2)] * wvec[k + (wvec_dim1 << 1)];
tempc = prod[k + prod_dim1 * 3] * wvec[k + wvec_dim1 * 5] + prod[k +
prod_dim1 * 5] * wvec[k + wvec_dim1 * 3];
den[1] = den[1] - tempa - 0.5 * (tempb + tempc);
den[5] -= 0.5 * (tempb - tempc);
tempa = prod[k + prod_dim1] * wvec[k + wvec_dim1 * 3] + prod[k +
prod_dim1 * 3] * wvec[k + wvec_dim1];
tempb = prod[k + (prod_dim1 << 1)] * wvec[k + wvec_dim1 * 5] + prod[k
+ prod_dim1 * 5] * wvec[k + (wvec_dim1 << 1)];
tempc = prod[k + prod_dim1 * 3] * wvec[k + (wvec_dim1 << 2)] + prod[k
+ (prod_dim1 << 2)] * wvec[k + wvec_dim1 * 3];
den[2] = den[2] - tempa - 0.5 * (tempb - tempc);
den[6] -= 0.5 * (tempb + tempc);
tempa = prod[k + prod_dim1] * wvec[k + (wvec_dim1 << 2)] + prod[k + (
prod_dim1 << 2)] * wvec[k + wvec_dim1];
den[3] = den[3] - tempa - par[1] + par[2];
tempa = prod[k + prod_dim1] * wvec[k + wvec_dim1 * 5] + prod[k +
prod_dim1 * 5] * wvec[k + wvec_dim1];
tempb = prod[k + (prod_dim1 << 1)] * wvec[k + wvec_dim1 * 3] + prod[k
+ prod_dim1 * 3] * wvec[k + (wvec_dim1 << 1)];
den[4] = den[4] - tempa - 0.5 * tempb;
den[7] = den[7] - par[3] + par[4];
tempa = prod[k + (prod_dim1 << 2)] * wvec[k + wvec_dim1 * 5] + prod[k
+ prod_dim1 * 5] * wvec[k + (wvec_dim1 << 2)];
den[8] -= 0.5 * tempa;
}
/* Extend DEN so that it holds all the coefficients of DENOM. */
sum = 0;
for (i__ = 1; i__ <= 5; ++i__) {
/* Computing 2nd power */
d__1 = prod[*knew + i__ * prod_dim1];
par[i__ - 1] = 0.5 * (d__1 * d__1);
sum += par[i__ - 1];
}
denex[0] = alpha * den[0] + par[0] + sum;
tempa = 2.0 * prod[*knew + prod_dim1] * prod[*knew + (prod_dim1 << 1)];
tempb = prod[*knew + (prod_dim1 << 1)] * prod[*knew + (prod_dim1 << 2)];
tempc = prod[*knew + prod_dim1 * 3] * prod[*knew + prod_dim1 * 5];
denex[1] = alpha * den[1] + tempa + tempb + tempc;
denex[5] = alpha * den[5] + tempb - tempc;
tempa = 2.0 * prod[*knew + prod_dim1] * prod[*knew + prod_dim1 * 3];
tempb = prod[*knew + (prod_dim1 << 1)] * prod[*knew + prod_dim1 * 5];
tempc = prod[*knew + prod_dim1 * 3] * prod[*knew + (prod_dim1 << 2)];
denex[2] = alpha * den[2] + tempa + tempb - tempc;
denex[6] = alpha * den[6] + tempb + tempc;
tempa = 2.0 * prod[*knew + prod_dim1] * prod[*knew + (prod_dim1 << 2)];
denex[3] = alpha * den[3] + tempa + par[1] - par[2];
tempa = 2.0 * prod[*knew + prod_dim1] * prod[*knew + prod_dim1 * 5];
denex[4] = alpha * den[4] + tempa + prod[*knew + (prod_dim1 << 1)] * prod[
*knew + prod_dim1 * 3];
denex[7] = alpha * den[7] + par[3] - par[4];
denex[8] = alpha * den[8] + prod[*knew + (prod_dim1 << 2)] * prod[*knew +
prod_dim1 * 5];
/* Seek the value of the angle that maximizes the modulus of DENOM. */
sum = denex[0] + denex[1] + denex[3] + denex[5] + denex[7];
denold = denmax = sum;
isave = 0;
iu = 49;
temp = twopi / (TYPE) (iu + 1);
par[0] = 1.0;
i__1 = iu;
for (i__ = 1; i__ <= i__1; ++i__) {
angle = (TYPE) i__ *temp;
par[1] = cos(angle);
par[2] = sin(angle);
for (j = 4; j <= 8; j += 2) {
par[j - 1] = par[1] * par[j - 3] - par[2] * par[j - 2];
par[j] = par[1] * par[j - 2] + par[2] * par[j - 3];
}
sumold = sum;
sum = 0;
for (j = 1; j <= 9; ++j)
sum += denex[j - 1] * par[j - 1];
if (fabs(sum) > fabs(denmax)) {
denmax = sum;
isave = i__;
tempa = sumold;
} else if (i__ == isave + 1) {
tempb = sum;
}
}
if (isave == 0) tempa = sum;
if (isave == iu) tempb = denold;
step = 0;
if (tempa != tempb) {
tempa -= denmax;
tempb -= denmax;
step = 0.5 * (tempa - tempb) / (tempa + tempb);
}
angle = temp * ((TYPE) isave + step);
/* Calculate the new parameters of the denominator, the new VBDLAG
* vector and the new D. Then test for convergence. */
par[1] = cos(angle);
par[2] = sin(angle);
for (j = 4; j <= 8; j += 2) {
par[j - 1] = par[1] * par[j - 3] - par[2] * par[j - 2];
par[j] = par[1] * par[j - 2] + par[2] * par[j - 3];
}
*beta = 0;
denmax = 0;
for (j = 1; j <= 9; ++j) {
*beta += den[j - 1] * par[j - 1];
denmax += denex[j - 1] * par[j - 1];
}
i__1 = *ndim;
for (k = 1; k <= i__1; ++k) {
vlag[k] = 0;
for (j = 1; j <= 5; ++j)
vlag[k] += prod[k + j * prod_dim1] * par[j - 1];
}
tau = vlag[*knew];
dd = tempa = tempb = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = par[1] * d__[i__] + par[2] * s[i__];
w[i__] = xopt[i__] + d__[i__];
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
tempa += d__[i__] * w[i__];
tempb += w[i__] * w[i__];
}
if (iterc >= n) goto BDL340;
if (iterc > 1) densav = std::max(densav, denold);
if (fabs(denmax) <= fabs(densav) * 1.1) goto BDL340;
densav = denmax;
/* Set S to 0.5 the gradient of the denominator with respect to
* D. Then branch for the next iteration. */
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
temp = tempa * xopt[i__] + tempb * d__[i__] - vlag[npt + i__];
s[i__] = tau * bmat[*knew + i__ * bmat_dim1] + alpha * temp;
}
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
sum = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j)
sum += xpt[k + j * xpt_dim1] * w[j];
temp = (tau * w[n + k] - alpha * vlag[k]) * sum;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
s[i__] += temp * xpt[k + i__ * xpt_dim1];
}
ss = 0;
ds = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing 2nd power */
d__1 = s[i__];
ss += d__1 * d__1;
ds += d__[i__] * s[i__];
}
ssden = dd * ss - ds * ds;
if (ssden >= dd * 1e-8 * ss) goto BDL70;
/* Set the vector W before the RETURN from the subroutine. */
BDL340:
i__2 = *ndim;
for (k = 1; k <= i__2; ++k) {
w[k] = 0;
for (j = 1; j <= 5; ++j) w[k] += wvec[k + j * wvec_dim1] * par[j - 1];
}
vlag[*kopt] += 1.0;
return 0;
}
template<class TYPE>
int trsapp_(int n, int npt, TYPE * xopt, TYPE * xpt, TYPE * gq, TYPE * hq, TYPE * pq,
TYPE * delta, TYPE * step, TYPE * d__, TYPE * g, TYPE * hd, TYPE * hs, TYPE * crvmin)
{
/* The arguments NPT, XOPT, XPT, GQ, HQ and PQ have their usual
* meanings, in order to define the current quadratic model Q.
* DETRLTA is the trust region radius, and has to be positive. STEP
* will be set to the calculated trial step. The arrays D, G, HD and
* HS will be used for working space. CRVMIN will be set to the
* least curvature of H aint the conjugate directions that occur,
* except that it is set to 0 if STEP goes all the way to the trust
* region boundary. The calculation of STEP begins with the
* truncated conjugate gradient method. If the boundary of the trust
* region is reached, then further changes to STEP may be made, each
* one being in the 2D space spanned by the current STEP and the
* corresponding gradient of Q. Thus STEP should provide a
* substantial reduction to Q within the trust region. */
int xpt_dim1, xpt_offset, i__1, i__2, i__, j, k, ih, iu, iterc,
isave, itersw, itermax;
TYPE d__1, d__2, dd, cf, dg, gg, ds, sg, ss, dhd, dhs,
cth, sgk, shs, sth, qadd, qbeg, qred, qmin, temp,
qsav, qnew, ggbeg, alpha, angle, reduc, ggsav, delsq,
tempa, tempb, bstep, ratio, twopi, angtest;
/* Parameter adjustments */
tempa = tempb = shs = sg = bstep = ggbeg = gg = qred = dd = 0.0;
xpt_dim1 = npt;
xpt_offset = 1 + xpt_dim1;
xpt -= xpt_offset;
--xopt; --gq; --hq; --pq; --step; --d__; --g; --hd; --hs;
/* Function Body */
twopi = 2.0 * M_PI;
delsq = *delta * *delta;
iterc = 0;
itermax = n;
itersw = itermax;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) d__[i__] = xopt[i__];
goto TRL170;
/* Prepare for the first line search. */
TRL20:
qred = dd = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
step[i__] = 0;
hs[i__] = 0;
g[i__] = gq[i__] + hd[i__];
d__[i__] = -g[i__];
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
}
*crvmin = 0;
if (dd == 0) goto TRL160;
ds = ss = 0;
gg = dd;
ggbeg = gg;
/* Calculate the step to the trust region boundary and the product
* HD. */
TRL40:
++iterc;
temp = delsq - ss;
bstep = temp / (ds + sqrt(ds * ds + dd * temp));
goto TRL170;
TRL50:
dhd = 0;
i__1 = n;
for (j = 1; j <= i__1; ++j) dhd += d__[j] * hd[j];
/* Update CRVMIN and set the step-length ATRLPHA. */
alpha = bstep;
if (dhd > 0) {
temp = dhd / dd;
if (iterc == 1) *crvmin = temp;
*crvmin = std::min(*crvmin, temp);
/* Computing MIN */
d__1 = alpha, d__2 = gg / dhd;
alpha = std::min(d__1, d__2);
}
qadd = alpha * (gg - 0.5 * alpha * dhd);
qred += qadd;
/* Update STEP and HS. */
ggsav = gg;
gg = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
step[i__] += alpha * d__[i__];
hs[i__] += alpha * hd[i__];
/* Computing 2nd power */
d__1 = g[i__] + hs[i__];
gg += d__1 * d__1;
}
/* Begin another conjugate direction iteration if required. */
if (alpha < bstep) {
if (qadd <= qred * .01 || gg <= ggbeg * 1e-4 || iterc == itermax) goto TRL160;
temp = gg / ggsav;
dd = ds = ss = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = temp * d__[i__] - g[i__] - hs[i__];
/* Computing 2nd power */
d__1 = d__[i__];
dd += d__1 * d__1;
ds += d__[i__] * step[i__];
/* Computing 2nd power */
d__1 = step[i__];
ss += d__1 * d__1;
}
if (ds <= 0) goto TRL160;
if (ss < delsq) goto TRL40;
}
*crvmin = 0;
itersw = iterc;
/* Test whether an alternative iteration is required. */
TRL90:
if (gg <= ggbeg * 1e-4) goto TRL160;
sg = 0;
shs = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
sg += step[i__] * g[i__];
shs += step[i__] * hs[i__];
}
sgk = sg + shs;
angtest = sgk / sqrt(gg * delsq);
if (angtest <= -.99) goto TRL160;
/* Begin the alternative iteration by calculating D and HD and some
* scalar products. */
++iterc;
temp = sqrt(delsq * gg - sgk * sgk);
tempa = delsq / temp;
tempb = sgk / temp;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__)
d__[i__] = tempa * (g[i__] + hs[i__]) - tempb * step[i__];
goto TRL170;
TRL120:
dg = dhd = dhs = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
dg += d__[i__] * g[i__];
dhd += hd[i__] * d__[i__];
dhs += hd[i__] * step[i__];
}
/* Seek the value of the angle that minimizes Q. */
cf = 0.5 * (shs - dhd);
qbeg = sg + cf;
qsav = qmin = qbeg;
isave = 0;
iu = 49;
temp = twopi / (TYPE) (iu + 1);
i__1 = iu;
for (i__ = 1; i__ <= i__1; ++i__) {
angle = (TYPE) i__ *temp;
cth = cos(angle);
sth = sin(angle);
qnew = (sg + cf * cth) * cth + (dg + dhs * cth) * sth;
if (qnew < qmin) {
qmin = qnew;
isave = i__;
tempa = qsav;
} else if (i__ == isave + 1) tempb = qnew;
qsav = qnew;
}
if ((TYPE) isave == 0) tempa = qnew;
if (isave == iu) tempb = qbeg;
angle = 0;
if (tempa != tempb) {
tempa -= qmin;
tempb -= qmin;
angle = 0.5 * (tempa - tempb) / (tempa + tempb);
}
angle = temp * ((TYPE) isave + angle);
/* Calculate the new STEP and HS. Then test for convergence. */
cth = cos(angle);
sth = sin(angle);
reduc = qbeg - (sg + cf * cth) * cth - (dg + dhs * cth) * sth;
gg = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
step[i__] = cth * step[i__] + sth * d__[i__];
hs[i__] = cth * hs[i__] + sth * hd[i__];
/* Computing 2nd power */
d__1 = g[i__] + hs[i__];
gg += d__1 * d__1;
}
qred += reduc;
ratio = reduc / qred;
if (iterc < itermax && ratio > .01) goto TRL90;
TRL160:
return 0;
/* The following instructions act as a subroutine for setting the
* vector HD to the vector D multiplied by the second derivative
* matrix of Q. They are called from three different places, which
* are distinguished by the value of ITERC. */
TRL170:
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) hd[i__] = 0;
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
temp = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j)
temp += xpt[k + j * xpt_dim1] * d__[j];
temp *= pq[k];
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
hd[i__] += temp * xpt[k + i__ * xpt_dim1];
}
ih = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
i__1 = j;
for (i__ = 1; i__ <= i__1; ++i__) {
++ih;
if (i__ < j) hd[j] += hq[ih] * d__[i__];
hd[i__] += hq[ih] * d__[j];
}
}
if (iterc == 0) goto TRL20;
if (iterc <= itersw) goto TRL50;
goto TRL120;
}
template<class TYPE>
static int update_(int n, int npt, TYPE *bmat, TYPE *zmat, int *idz, int *ndim, TYPE *vlag,
TYPE *beta, int *knew, TYPE *w)
{
/* The arrays BMAT and ZMAT with IDZ are updated, in order to shift
* the interpolation point that has index KNEW. On entry, VLAG
* contains the components of the vector Theta*Wcheck+e_b of the
* updating formula (6.11), and BETA holds the value of the
* parameter that has this name. The vector W is used for working
* space. */
int bmat_dim1, bmat_offset, zmat_dim1, zmat_offset, i__1, i__2, i__,
j, ja, jb, jl, jp, nptm, iflag;
TYPE d__1, d__2, tau, temp, scala, scalb, alpha, denom, tempa, tempb, tausq;
/* Parameter adjustments */
tempb = 0.0;
zmat_dim1 = npt;
zmat_offset = 1 + zmat_dim1;
zmat -= zmat_offset;
bmat_dim1 = *ndim;
bmat_offset = 1 + bmat_dim1;
bmat -= bmat_offset;
--vlag;
--w;
/* Function Body */
nptm = npt - n - 1;
/* Apply the rotations that put zeros in the KNEW-th row of ZMAT. */
jl = 1;
i__1 = nptm;
for (j = 2; j <= i__1; ++j) {
if (j == *idz) {
jl = *idz;
} else if (zmat[*knew + j * zmat_dim1] != 0) {
/* Computing 2nd power */
d__1 = zmat[*knew + jl * zmat_dim1];
/* Computing 2nd power */
d__2 = zmat[*knew + j * zmat_dim1];
temp = sqrt(d__1 * d__1 + d__2 * d__2);
tempa = zmat[*knew + jl * zmat_dim1] / temp;
tempb = zmat[*knew + j * zmat_dim1] / temp;
i__2 = npt;
for (i__ = 1; i__ <= i__2; ++i__) {
temp = tempa * zmat[i__ + jl * zmat_dim1] + tempb * zmat[i__
+ j * zmat_dim1];
zmat[i__ + j * zmat_dim1] = tempa * zmat[i__ + j * zmat_dim1]
- tempb * zmat[i__ + jl * zmat_dim1];
zmat[i__ + jl * zmat_dim1] = temp;
}
zmat[*knew + j * zmat_dim1] = 0;
}
}
/* Put the first NPT components of the KNEW-th column of HLAG into
* W, and calculate the parameters of the updating formula. */
tempa = zmat[*knew + zmat_dim1];
if (*idz >= 2) tempa = -tempa;
if (jl > 1) tempb = zmat[*knew + jl * zmat_dim1];
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__) {
w[i__] = tempa * zmat[i__ + zmat_dim1];
if (jl > 1) w[i__] += tempb * zmat[i__ + jl * zmat_dim1];
}
alpha = w[*knew];
tau = vlag[*knew];
tausq = tau * tau;
denom = alpha * *beta + tausq;
vlag[*knew] -= 1.0;
/* Complete the updating of ZMAT when there is only 1.0 nonzero
* element in the KNEW-th row of the new matrix ZMAT, but, if IFLAG
* is set to 1.0, then the first column of ZMAT will be exchanged
* with another 1.0 later. */
iflag = 0;
if (jl == 1) {
temp = sqrt((fabs(denom)));
tempb = tempa / temp;
tempa = tau / temp;
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__)
zmat[i__ + zmat_dim1] = tempa * zmat[i__ + zmat_dim1] - tempb *
vlag[i__];
if (*idz == 1 && temp < 0) *idz = 2;
if (*idz >= 2 && temp >= 0) iflag = 1;
} else {
/* Complete the updating of ZMAT in the alternative case. */
ja = 1;
if (*beta >= 0) {
ja = jl;
}
jb = jl + 1 - ja;
temp = zmat[*knew + jb * zmat_dim1] / denom;
tempa = temp * *beta;
tempb = temp * tau;
temp = zmat[*knew + ja * zmat_dim1];
scala = 1.0 / sqrt(fabs(*beta) * temp * temp + tausq);
scalb = scala * sqrt((fabs(denom)));
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__) {
zmat[i__ + ja * zmat_dim1] = scala * (tau * zmat[i__ + ja *
zmat_dim1] - temp * vlag[i__]);
zmat[i__ + jb * zmat_dim1] = scalb * (zmat[i__ + jb * zmat_dim1]
- tempa * w[i__] - tempb * vlag[i__]);
}
if (denom <= 0) {
if (*beta < 0) ++(*idz);
if (*beta >= 0) iflag = 1;
}
}
/* IDZ is reduced in the following case, and usually the first
* column of ZMAT is exchanged with a later 1.0. */
if (iflag == 1) {
--(*idz);
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__) {
temp = zmat[i__ + zmat_dim1];
zmat[i__ + zmat_dim1] = zmat[i__ + *idz * zmat_dim1];
zmat[i__ + *idz * zmat_dim1] = temp;
}
}
/* Finally, update the matrix BMAT. */
i__1 = n;
for (j = 1; j <= i__1; ++j) {
jp = npt + j;
w[jp] = bmat[*knew + j * bmat_dim1];
tempa = (alpha * vlag[jp] - tau * w[jp]) / denom;
tempb = (-(*beta) * w[jp] - tau * vlag[jp]) / denom;
i__2 = jp;
for (i__ = 1; i__ <= i__2; ++i__) {
bmat[i__ + j * bmat_dim1] = bmat[i__ + j * bmat_dim1] + tempa *
vlag[i__] + tempb * w[i__];
if (i__ > npt) {
bmat[jp + (i__ - npt) * bmat_dim1] = bmat[i__ + j *
bmat_dim1];
}
}
}
return 0;
}
template<class TYPE, class Func>
static TYPE newuob_(int n, int npt, TYPE *x,
TYPE rhobeg, TYPE rhoend, int *ret_nf, int maxfun,
TYPE *xbase, TYPE *xopt, TYPE *xnew,
TYPE *xpt, TYPE *fval, TYPE *gq, TYPE *hq,
TYPE *pq, TYPE *bmat, TYPE *zmat, int *ndim,
TYPE *d__, TYPE *vlag, TYPE *w, Func &func)
{
/* XBASE will hold a shift of origin that should reduce the
contributions from rounding errors to values of the model and
Lagrange functions.
* XOPT will be set to the displacement from XBASE of the vector of
variables that provides the least calculated F so far.
* XNEW will be set to the displacement from XBASE of the vector of
variables for the current calculation of F.
* XPT will contain the interpolation point coordinates relative to
XBASE.
* FVAL will hold the values of F at the interpolation points.
* GQ will hold the gradient of the quadratic model at XBASE.
* HQ will hold the explicit second derivatives of the quadratic
model.
* PQ will contain the parameters of the implicit second derivatives
of the quadratic model.
* BMAT will hold the last N columns of H.
* ZMAT will hold the factorization of the leading NPT by NPT
submatrix of H, this factorization being ZMAT times Diag(DZ)
times ZMAT^T, where the elements of DZ are plus or minus 1.0, as
specified by IDZ.
* NDIM is the first dimension of BMAT and has the value NPT+N.
* D is reserved for trial steps from XOPT.
* VLAG will contain the values of the Lagrange functions at a new
point X. They are part of a product that requires VLAG to be of
length NDIM.
* The array W will be used for working space. Its length must be at
least 10*NDIM = 10*(NPT+N). Set some constants. */
int xpt_dim1, xpt_offset, bmat_dim1, bmat_offset, zmat_dim1, zmat_offset,
i__1, i__2, i__3, i__, j, k, ih, nf, nh, ip, jp, np, nfm, idz, ipt, jpt,
nfmm, knew, kopt, nptm, ksave, nfsav, itemp, ktemp, itest, nftest;
TYPE d__1, d__2, d__3, f, dx, dsq, rho, sum, fbeg, diff, beta, gisq,
temp, suma, sumb, fopt, bsum, gqsq, xipt, xjpt, sumz, diffa, diffb,
diffc, hdiag, alpha, delta, recip, reciq, fsave, dnorm, ratio, dstep,
vquad, tempq, rhosq, detrat, crvmin, distsq, xoptsq;
/* Parameter adjustments */
diffc = ratio = dnorm = nfsav = diffa = diffb = xoptsq = f = 0.0;
rho = fbeg = fopt = xjpt = xipt = 0.0;
itest = ipt = jpt = 0;
alpha = dstep = 0.0;
zmat_dim1 = npt;
zmat_offset = 1 + zmat_dim1;
zmat -= zmat_offset;
xpt_dim1 = npt;
xpt_offset = 1 + xpt_dim1;
xpt -= xpt_offset;
--x; --xbase; --xopt; --xnew; --fval; --gq; --hq; --pq;
bmat_dim1 = *ndim;
bmat_offset = 1 + bmat_dim1;
bmat -= bmat_offset;
--d__;
--vlag;
--w;
/* Function Body */
np = n + 1;
nh = n * np / 2;
nptm = npt - np;
nftest = (maxfun > 1)? maxfun : 1;
/* Set the initial elements of XPT, BMAT, HQ, PQ and ZMAT to 0. */
i__1 = n;
for (j = 1; j <= i__1; ++j) {
xbase[j] = x[j];
i__2 = npt;
for (k = 1; k <= i__2; ++k)
xpt[k + j * xpt_dim1] = 0;
i__2 = *ndim;
for (i__ = 1; i__ <= i__2; ++i__)
bmat[i__ + j * bmat_dim1] = 0;
}
i__2 = nh;
for (ih = 1; ih <= i__2; ++ih) hq[ih] = 0;
i__2 = npt;
for (k = 1; k <= i__2; ++k) {
pq[k] = 0;
i__1 = nptm;
for (j = 1; j <= i__1; ++j)
zmat[k + j * zmat_dim1] = 0;
}
/* Begin the initialization procedure. NF becomes 1.0 more than the
* number of function values so far. The coordinates of the
* displacement of the next initial interpolation point from XBASE
* are set in XPT(NF,.). */
rhosq = rhobeg * rhobeg;
recip = 1.0 / rhosq;
reciq = sqrt(.5) / rhosq;
nf = 0;
L50:
nfm = nf;
nfmm = nf - n;
++nf;
if (nfm <= n << 1) {
if (nfm >= 1 && nfm <= n) {
xpt[nf + nfm * xpt_dim1] = rhobeg;
} else if (nfm > n) {
xpt[nf + nfmm * xpt_dim1] = -(rhobeg);
}
} else {
itemp = (nfmm - 1) / n;
jpt = nfm - itemp * n - n;
ipt = jpt + itemp;
if (ipt > n) {
itemp = jpt;
jpt = ipt - n;
ipt = itemp;
}
xipt = rhobeg;
if (fval[ipt + np] < fval[ipt + 1]) xipt = -xipt;
xjpt = rhobeg;
if (fval[jpt + np] < fval[jpt + 1]) xjpt = -xjpt;
xpt[nf + ipt * xpt_dim1] = xipt;
xpt[nf + jpt * xpt_dim1] = xjpt;
}
/* Calculate the next value of F, label 70 being reached immediately
* after this calculation. The least function value so far and its
* index are required. */
i__1 = n;
for (j = 1; j <= i__1; ++j)
x[j] = xpt[nf + j * xpt_dim1] + xbase[j];
goto L310;
L70:
fval[nf] = f;
if (nf == 1) {
fbeg = fopt = f;
kopt = 1;
} else if (f < fopt) {
fopt = f;
kopt = nf;
}
/* Set the non0 initial elements of BMAT and the quadratic model
* in the cases when NF is at most 2*N+1. */
if (nfm <= n << 1) {
if (nfm >= 1 && nfm <= n) {
gq[nfm] = (f - fbeg) / rhobeg;
if (npt < nf + n) {
bmat[nfm * bmat_dim1 + 1] = -1.0 / rhobeg;
bmat[nf + nfm * bmat_dim1] = 1.0 / rhobeg;
bmat[npt + nfm + nfm * bmat_dim1] = -.5 * rhosq;
}
} else if (nfm > n) {
bmat[nf - n + nfmm * bmat_dim1] = .5 / rhobeg;
bmat[nf + nfmm * bmat_dim1] = -.5 / rhobeg;
zmat[nfmm * zmat_dim1 + 1] = -reciq - reciq;
zmat[nf - n + nfmm * zmat_dim1] = reciq;
zmat[nf + nfmm * zmat_dim1] = reciq;
ih = nfmm * (nfmm + 1) / 2;
temp = (fbeg - f) / rhobeg;
hq[ih] = (gq[nfmm] - temp) / rhobeg;
gq[nfmm] = .5 * (gq[nfmm] + temp);
}
/* Set the off-diagonal second derivatives of the Lagrange
* functions and the initial quadratic model. */
} else {
ih = ipt * (ipt - 1) / 2 + jpt;
if (xipt < 0) ipt += n;
if (xjpt < 0) jpt += n;
zmat[nfmm * zmat_dim1 + 1] = recip;
zmat[nf + nfmm * zmat_dim1] = recip;
zmat[ipt + 1 + nfmm * zmat_dim1] = -recip;
zmat[jpt + 1 + nfmm * zmat_dim1] = -recip;
hq[ih] = (fbeg - fval[ipt + 1] - fval[jpt + 1] + f) / (xipt * xjpt);
}
if (nf < npt) goto L50;
/* Begin the iterative procedure, because the initial model is
* complete. */
rho = rhobeg;
delta = rho;
idz = 1;
diffa = diffb = itest = xoptsq = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
xopt[i__] = xpt[kopt + i__ * xpt_dim1];
/* Computing 2nd power */
d__1 = xopt[i__];
xoptsq += d__1 * d__1;
}
L90:
nfsav = nf;
/* Generate the next trust region step and test its length. Set KNEW
* to -1 if the purpose of the next F will be to improve the
* model. */
L100:
knew = 0;
trsapp_(n, npt, &xopt[1], &xpt[xpt_offset], &gq[1], &hq[1], &pq[1], &
delta, &d__[1], &w[1], &w[np], &w[np + n], &w[np + (n << 1)], &
crvmin);
dsq = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Computing 2nd power */
d__1 = d__[i__];
dsq += d__1 * d__1;
}
/* Computing MIN */
d__1 = delta, d__2 = sqrt(dsq);
dnorm = std::min(d__1, d__2);
if (dnorm < .5 * rho) {
knew = -1;
delta = DELTA_DECREASE * delta;
ratio = -1.;
if (delta <= rho * 1.5) delta = rho;
if (nf <= nfsav + 2) goto L460;
temp = crvmin * .125 * rho * rho;
/* Computing MAX */
d__1 = std::max(diffa, diffb);
if (temp <= std::max(d__1, diffc)) goto L460;
goto L490;
}
/* Shift XBASE if XOPT may be too far from XBASE. First make the
* changes to BMAT that do not depend on ZMAT. */
L120:
if (dsq <= xoptsq * .001) {
tempq = xoptsq * .25;
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
sum = 0;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
sum += xpt[k + i__ * xpt_dim1] * xopt[i__];
temp = pq[k] * sum;
sum -= .5 * xoptsq;
w[npt + k] = sum;
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
gq[i__] += temp * xpt[k + i__ * xpt_dim1];
xpt[k + i__ * xpt_dim1] -= .5 * xopt[i__];
vlag[i__] = bmat[k + i__ * bmat_dim1];
w[i__] = sum * xpt[k + i__ * xpt_dim1] + tempq * xopt[i__];
ip = npt + i__;
i__3 = i__;
for (j = 1; j <= i__3; ++j)
bmat[ip + j * bmat_dim1] = bmat[ip + j * bmat_dim1] +
vlag[i__] * w[j] + w[i__] * vlag[j];
}
}
/* Then the revisions of BMAT that depend on ZMAT are
* calculated. */
i__3 = nptm;
for (k = 1; k <= i__3; ++k) {
sumz = 0;
i__2 = npt;
for (i__ = 1; i__ <= i__2; ++i__) {
sumz += zmat[i__ + k * zmat_dim1];
w[i__] = w[npt + i__] * zmat[i__ + k * zmat_dim1];
}
i__2 = n;
for (j = 1; j <= i__2; ++j) {
sum = tempq * sumz * xopt[j];
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__)
sum += w[i__] * xpt[i__ + j * xpt_dim1];
vlag[j] = sum;
if (k < idz) sum = -sum;
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__)
bmat[i__ + j * bmat_dim1] += sum * zmat[i__ + k * zmat_dim1];
}
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
ip = i__ + npt;
temp = vlag[i__];
if (k < idz) temp = -temp;
i__2 = i__;
for (j = 1; j <= i__2; ++j)
bmat[ip + j * bmat_dim1] += temp * vlag[j];
}
}
/* The following instructions complete the shift of XBASE,
* including the changes to the parameters of the quadratic
* model. */
ih = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
w[j] = 0;
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
w[j] += pq[k] * xpt[k + j * xpt_dim1];
xpt[k + j * xpt_dim1] -= .5 * xopt[j];
}
i__1 = j;
for (i__ = 1; i__ <= i__1; ++i__) {
++ih;
if (i__ < j) gq[j] += hq[ih] * xopt[i__];
gq[i__] += hq[ih] * xopt[j];
hq[ih] = hq[ih] + w[i__] * xopt[j] + xopt[i__] * w[j];
bmat[npt + i__ + j * bmat_dim1] = bmat[npt + j + i__ *
bmat_dim1];
}
}
i__1 = n;
for (j = 1; j <= i__1; ++j) {
xbase[j] += xopt[j];
xopt[j] = 0;
}
xoptsq = 0;
}
/* Pick the model step if KNEW is positive. A different choice of D
* may be made later, if the choice of D by BIGLAG causes
* substantial cancellation in DENOM. */
if (knew > 0) {
biglag_(n, npt, &xopt[1], &xpt[xpt_offset], &bmat[bmat_offset], &zmat[zmat_offset], &idz,
ndim, &knew, &dstep, &d__[1], &alpha, &vlag[1], &vlag[npt + 1], &w[1], &w[np], &w[np + n], func);
}
/* Calculate VLAG and BETA for the current choice of D. The first
* NPT components of W_check will be held in W. */
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
suma = 0;
sumb = 0;
sum = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
suma += xpt[k + j * xpt_dim1] * d__[j];
sumb += xpt[k + j * xpt_dim1] * xopt[j];
sum += bmat[k + j * bmat_dim1] * d__[j];
}
w[k] = suma * (.5 * suma + sumb);
vlag[k] = sum;
}
beta = 0;
i__1 = nptm;
for (k = 1; k <= i__1; ++k) {
sum = 0;
i__2 = npt;
for (i__ = 1; i__ <= i__2; ++i__)
sum += zmat[i__ + k * zmat_dim1] * w[i__];
if (k < idz) {
beta += sum * sum;
sum = -sum;
} else beta -= sum * sum;
i__2 = npt;
for (i__ = 1; i__ <= i__2; ++i__)
vlag[i__] += sum * zmat[i__ + k * zmat_dim1];
}
bsum = 0;
dx = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
sum = 0;
i__1 = npt;
for (i__ = 1; i__ <= i__1; ++i__)
sum += w[i__] * bmat[i__ + j * bmat_dim1];
bsum += sum * d__[j];
jp = npt + j;
i__1 = n;
for (k = 1; k <= i__1; ++k)
sum += bmat[jp + k * bmat_dim1] * d__[k];
vlag[jp] = sum;
bsum += sum * d__[j];
dx += d__[j] * xopt[j];
}
beta = dx * dx + dsq * (xoptsq + dx + dx + .5 * dsq) + beta - bsum;
vlag[kopt] += 1.0;
/* If KNEW is positive and if the cancellation in DENOM is
* unacceptable, then BIGDEN calculates an alternative model step,
* XNEW being used for working space. */
if (knew > 0) {
/* Computing 2nd power */
d__1 = vlag[knew];
temp = 1.0 + alpha * beta / (d__1 * d__1);
if (fabs(temp) <= .8) {
bigden_(n, npt, &xopt[1], &xpt[xpt_offset], &bmat[bmat_offset], &
zmat[zmat_offset], &idz, ndim, &kopt, &knew, &d__[1], &w[
1], &vlag[1], &beta, &xnew[1], &w[*ndim + 1], &w[*ndim *
6 + 1]);
}
}
/* Calculate the next value of the objective function. */
L290:
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__) {
xnew[i__] = xopt[i__] + d__[i__];
x[i__] = xbase[i__] + xnew[i__];
}
++nf;
L310:
if (nf > nftest) {
--nf;
//fprintf(stderr, "++ Return from NEWUOA because CALFUN has been called MAXFUN times.\n");
goto L530;
}
f = func(n, &x[1]);
if (nf <= npt) goto L70;
if (knew == -1) goto L530;
/* Use the quadratic model to predict the change in F due to the
* step D, and set DIFF to the error of this prediction. */
vquad = ih = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
vquad += d__[j] * gq[j];
i__1 = j;
for (i__ = 1; i__ <= i__1; ++i__) {
++ih;
temp = d__[i__] * xnew[j] + d__[j] * xopt[i__];
if (i__ == j) temp = .5 * temp;
vquad += temp * hq[ih];
}
}
i__1 = npt;
for (k = 1; k <= i__1; ++k) vquad += pq[k] * w[k];
diff = f - fopt - vquad;
diffc = diffb;
diffb = diffa;
diffa = fabs(diff);
if (dnorm > rho) nfsav = nf;
/* Update FOPT and XOPT if the new F is the least value of the
* objective function so far. The branch when KNEW is positive
* occurs if D is not a trust region step. */
fsave = fopt;
if (f < fopt) {
fopt = f;
xoptsq = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
xopt[i__] = xnew[i__];
/* Computing 2nd power */
d__1 = xopt[i__];
xoptsq += d__1 * d__1;
}
}
ksave = knew;
if (knew > 0) goto L410;
/* Pick the next value of DELTA after a trust region step. */
if (vquad >= 0) {
fprintf(stderr, "++ Return from NEWUOA because a trust region step has failed to reduce Q.\n");
goto L530;
}
ratio = (f - fsave) / vquad;
if (ratio <= 0.1) {
delta = .5 * dnorm;
} else if (ratio <= .7) {
/* Computing MAX */
d__1 = .5 * delta;
delta = std::max(d__1, dnorm);
} else {
/* Computing MAX */
d__1 = .5 * delta, d__2 = dnorm + dnorm;
delta = std::max(d__1, d__2);
}
if (delta <= rho * 1.5) delta = rho;
/* Set KNEW to the index of the next interpolation point to be
* deleted. */
/* Computing MAX */
d__2 = 0.1 * delta;
/* Computing 2nd power */
d__1 = std::max(d__2, rho);
rhosq = d__1 * d__1;
ktemp = detrat = 0;
if (f >= fsave) {
ktemp = kopt;
detrat = 1.0;
}
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
hdiag = 0;
i__2 = nptm;
for (j = 1; j <= i__2; ++j) {
temp = 1.0;
if (j < idz) temp = -1.0;
/* Computing 2nd power */
d__1 = zmat[k + j * zmat_dim1];
hdiag += temp * (d__1 * d__1);
}
/* Computing 2nd power */
d__2 = vlag[k];
temp = (d__1 = beta * hdiag + d__2 * d__2, fabs(d__1));
distsq = 0;
i__2 = n;
for (j = 1; j <= i__2; ++j) {
/* Computing 2nd power */
d__1 = xpt[k + j * xpt_dim1] - xopt[j];
distsq += d__1 * d__1;
}
if (distsq > rhosq) {
/* Computing 3rd power */
d__1 = distsq / rhosq;
temp *= d__1 * (d__1 * d__1);
}
if (temp > detrat && k != ktemp) {
detrat = temp;
knew = k;
}
}
if (knew == 0) goto L460;
/* Update BMAT, ZMAT and IDZ, so that the KNEW-th interpolation
* point can be moved. Begin the updating of the quadratic model,
* starting with the explicit second derivative term. */
L410:
update_(n, npt, &bmat[bmat_offset], &zmat[zmat_offset], &idz, ndim, &vlag[1], &beta, &knew, &w[1]);
fval[knew] = f;
ih = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
temp = pq[knew] * xpt[knew + i__ * xpt_dim1];
i__2 = i__;
for (j = 1; j <= i__2; ++j) {
++ih;
hq[ih] += temp * xpt[knew + j * xpt_dim1];
}
}
pq[knew] = 0;
/* Update the other second derivative parameters, and then the
* gradient vector of the model. Also include the new interpolation
* point. */
i__2 = nptm;
for (j = 1; j <= i__2; ++j) {
temp = diff * zmat[knew + j * zmat_dim1];
if (j < idz) temp = -temp;
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
pq[k] += temp * zmat[k + j * zmat_dim1];
}
}
gqsq = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
gq[i__] += diff * bmat[knew + i__ * bmat_dim1];
/* Computing 2nd power */
d__1 = gq[i__];
gqsq += d__1 * d__1;
xpt[knew + i__ * xpt_dim1] = xnew[i__];
}
/* If a trust region step makes a small change to the objective
* function, then calculate the gradient of the least Frobenius norm
* interpolant at XBASE, and store it in W, using VLAG for a vector
* of right hand sides. */
if (ksave == 0 && delta == rho) {
if (fabs(ratio) > .01) {
itest = 0;
} else {
i__1 = npt;
for (k = 1; k <= i__1; ++k)
vlag[k] = fval[k] - fval[kopt];
gisq = 0;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
sum = 0;
i__2 = npt;
for (k = 1; k <= i__2; ++k)
sum += bmat[k + i__ * bmat_dim1] * vlag[k];
gisq += sum * sum;
w[i__] = sum;
}
/* Test whether to replace the new quadratic model by the
* least Frobenius norm interpolant, making the replacement
* if the test is satisfied. */
++itest;
if (gqsq < gisq * 100.) itest = 0;
if (itest >= 3) {
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) gq[i__] = w[i__];
i__1 = nh;
for (ih = 1; ih <= i__1; ++ih) hq[ih] = 0;
i__1 = nptm;
for (j = 1; j <= i__1; ++j) {
w[j] = 0;
i__2 = npt;
for (k = 1; k <= i__2; ++k)
w[j] += vlag[k] * zmat[k + j * zmat_dim1];
if (j < idz) w[j] = -w[j];
}
i__1 = npt;
for (k = 1; k <= i__1; ++k) {
pq[k] = 0;
i__2 = nptm;
for (j = 1; j <= i__2; ++j)
pq[k] += zmat[k + j * zmat_dim1] * w[j];
}
itest = 0;
}
}
}
if (f < fsave) kopt = knew;
/* If a trust region step has provided a sufficient decrease in F,
* then branch for another trust region calculation. The case
* KSAVE>0 occurs when the new function value was calculated by a
* model step. */
if (f <= fsave + 0.1 * vquad) goto L100;
if (ksave > 0) goto L100;
/* Alternatively, find out if the interpolation points are close
* enough to the best point so far. */
knew = 0;
L460:
distsq = delta * 4. * delta;
i__2 = npt;
for (k = 1; k <= i__2; ++k) {
sum = 0;
i__1 = n;
for (j = 1; j <= i__1; ++j) {
/* Computing 2nd power */
d__1 = xpt[k + j * xpt_dim1] - xopt[j];
sum += d__1 * d__1;
}
if (sum > distsq) {
knew = k;
distsq = sum;
}
}
/* If KNEW is positive, then set DSTEP, and branch back for the next
* iteration, which will generate a "model step". */
if (knew > 0) {
/* Computing MAX and MIN*/
d__2 = 0.1 * sqrt(distsq), d__3 = .5 * delta;
d__1 = std::min(d__2, d__3);
dstep = std::max(d__1, rho);
dsq = dstep * dstep;
goto L120;
}
if (ratio > 0) goto L100;
if (std::max(delta, dnorm) > rho) goto L100;
/* The calculations with the current value of RHO are complete. Pick
* the next values of RHO and DELTA. */
L490:
if (rho > rhoend) {
delta = .5 * rho;
ratio = rho / rhoend;
if (ratio <= 16.) rho = rhoend;
else if (ratio <= 250.) rho = sqrt(ratio) * rhoend;
else rho = RHO_DECREASE * rho;
delta = std::max(delta, rho);
goto L90;
}
/* Return from the calculation, after another Newton-Raphson step,
* if it is too short to have been tried before. */
if (knew == -1) goto L290;
L530:
if (fopt <= f) {
i__2 = n;
for (i__ = 1; i__ <= i__2; ++i__)
x[i__] = xbase[i__] + xopt[i__];
f = fopt;
}
*ret_nf = nf;
return f;
}
template<class TYPE, class Func>
static TYPE newuoa_(int n, int npt, TYPE *x, TYPE rhobeg, TYPE rhoend, int *ret_nf, int maxfun, TYPE *w, Func &func)
{
/* This subroutine seeks the least value of a function of many
* variables, by a trust region method that forms quadratic models
* by interpolation. There can be some freedom in the interpolation
* conditions, which is taken up by minimizing the Frobenius norm of
* the change to the second derivative of the quadratic model,
* beginning with a zero matrix. The arguments of the subroutine are
* as follows. */
/* N must be set to the number of variables and must be at least
* two. NPT is the number of interpolation conditions. Its value
* must be in the interval [N+2,(N+1)(N+2)/2]. Initial values of the
* variables must be set in X(1),X(2),...,X(N). They will be changed
* to the values that give the least calculated F. RHOBEG and RHOEND
* must be set to the initial and final values of a trust region
* radius, so both must be positive with RHOEND<=RHOBEG. Typically
* RHOBEG should be about one tenth of the greatest expected change
* to a variable, and RHOEND should indicate the accuracy that is
* required in the final values of the variables. MAXFUN must be set
* to an upper bound on the number of calls of CALFUN. The array W
* will be used for working space. Its length must be at least
* (NPT+13)*(NPT+N)+3*N*(N+3)/2. */
/* SUBROUTINE CALFUN (N,X,F) must be provided by the user. It must
* set F to the value of the objective function for the variables
* X(1),X(2),...,X(N). Partition the working space array, so that
* different parts of it can be treated separately by the subroutine
* that performs the main calculation. */
int id, np, iw, igq, ihq, ixb, ifv, ipq, ivl, ixn,
ixo, ixp, ndim, nptm, ibmat, izmat;
/* Parameter adjustments */
--w; --x;
/* Function Body */
np = n + 1;
nptm = npt - np;
if (npt < n + 2 || npt > (n + 2) * np / 2) {
fprintf(stderr, "** Return from NEWUOA because NPT is not in the required interval.\n");
return 1;
}
ndim = npt + n;
ixb = 1;
ixo = ixb + n;
ixn = ixo + n;
ixp = ixn + n;
ifv = ixp + n * npt;
igq = ifv + npt;
ihq = igq + n;
ipq = ihq + n * np / 2;
ibmat = ipq + npt;
izmat = ibmat + ndim * n;
id = izmat + npt * nptm;
ivl = id + n;
iw = ivl + ndim;
/* The above settings provide a partition of W for subroutine
* NEWUOB. The partition requires the first NPT*(NPT+N)+5*N*(N+3)/2
* elements of W plus the space that is needed by the last array of
* NEWUOB. */
return newuob_(n, npt, &x[1], rhobeg, rhoend, ret_nf, maxfun, &w[ixb], &w[ixo], &w[ixn],
&w[ixp], &w[ifv], &w[igq], &w[ihq], &w[ipq], &w[ibmat], &w[izmat],
&ndim, &w[id], &w[ivl], &w[iw], func);
}
template<class TYPE, class Func>
TYPE min_newuoa(int n, TYPE *x, Func &func, TYPE rb, TYPE tol, int max_iter)
{
int npt = 2 * n + 1, rnf;
TYPE ret;
TYPE *w = (TYPE*)calloc((npt+13)*(npt+n) + 3*n*(n+3)/2 + 11, sizeof(TYPE));
ret = newuoa_(n, 2*n+1, x, rb, tol, &rnf, max_iter, w, func);
free(w);
return ret;
}
#endif