dynare/matlab/optimization/lnsrch1.m

142 lines
4.4 KiB
Matlab

function [x,f,fvec,check]=lnsrch1(xold, fold, g, p, stpmax, func, j1, j2, tolx, varargin)
% function [x,f,fvec,check]=lnsrch1(xold,fold,g,p,stpmax,func,j1,j2,tolx,varargin)
% Computes the optimal step by minimizing the residual sum of squares
%
% INPUTS
% xold: actual point
% fold: residual sum of squares at the point xold
% g: gradient
% p: Newton direction
% stpmax: maximum step
% func: name of the function
% j1: equations index to be solved
% j2: unknowns index
% tolx: tolerance parameter
% varargin: list of arguments following j2
%
% OUTPUTS
% x: chosen point
% f: residual sum of squares value for a given x
% fvec: residuals vector
% check=1: problem of the looping which continues indefinitely
%
%
% SPECIAL REQUIREMENTS
% none
% Copyright © 2001-2018 Dynare Team
%
% This file is part of Dynare.
%
% Dynare is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% Dynare is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with Dynare. If not, see <https://www.gnu.org/licenses/>.
alf = 1e-4 ;
alam = 1;
x = xold;
nn = length(j2);
summ = sqrt(p'*p);
if ~isfinite(summ)
if ~isequal(func,@perfect_foresight_problem)
eq_number_string=[];
for ii=1:length(j1)-1
eq_number_string=[eq_number_string, num2str(j1(ii)), ', '];
end
eq_number_string=[eq_number_string, num2str(j1(end))];
var_string=[];
M_=evalin('base','M_');
for ii=1:length(j2)-1
var_string=[var_string, M_.endo_names{j2(ii)}, ', '];
end
var_string=[var_string, M_.endo_names{j2(end)}];
fprintf('\nAn infinite element was encountered when trying to solve equation(s) %s \n',eq_number_string)
fprintf('with respect to the variable(s): %s.\n',var_string)
fprintf('The values of the endogenous variables when the problem was encountered were:\n')
label_width=size(char(M_.endo_names),2)+2;
label_string=sprintf('%%-%us %%8.4g \\n',label_width);
for ii=1:length(xold)
fprintf(label_string, M_.endo_names{ii}, xold(ii));
end
skipline();
end
error(['Some element of Newton direction isn''t finite. Jacobian maybe' ...
' singular or there is a problem with initial values'])
end
if summ > stpmax
p = p*stpmax/summ ;
end
slope = g'*p ;
test = max(abs(p)'./max([abs(xold(j2))';ones(1,nn)])) ;
alamin = tolx/test ;
if alamin > 0.1
alamin = 0.1;
end
while 1
if alam < alamin
check = 1 ;
return
end
x(j2) = xold(j2) + (alam*p) ;
fvec = feval(func,x,varargin{:}) ;
fvec = fvec(j1);
f = 0.5*(fvec'*fvec) ;
if any(isnan(fvec))
alam = alam/2 ;
alam2 = alam ;
f2 = f ;
fold2 = fold ;
else
if f <= fold+alf*alam*slope
check = 0;
break
else
if alam == 1
tmplam = -slope/(2*(f-fold-slope)) ;
else
rhs1 = f-fold-alam*slope ;
rhs2 = f2-fold2-alam2*slope ;
a = (rhs1/(alam^2)-rhs2/(alam2^2))/(alam-alam2) ;
b = (-alam2*rhs1/(alam^2)+alam*rhs2/(alam2^2))/(alam-alam2) ;
if a == 0
tmplam = -slope/(2*b) ;
else
disc = (b^2)-3*a*slope ;
if disc < 0
error ('Roundoff problem in nlsearch') ;
else
tmplam = (-b+sqrt(disc))/(3*a) ;
end
end
if tmplam > 0.5*alam
tmplam = 0.5*alam;
end
end
alam2 = alam ;
f2 = f ;
fold2 = fold ;
alam = max([tmplam;(0.1*alam)]) ;
end
end
end
% 01/14/01 MJ lnsearch is now a separate function
% 01/12/03 MJ check for finite summ to avoid infinite loop when Jacobian
% is singular or model is denormalized