+ Use scalar product when possible, ie replace sum(x.*x) by x'*x,
+ Add parenthesis around scalar products, ie replace a*x'*x by a*(x'*x),
+ Removed global options_ in lnsrch1 (only used for solve_tolx).
Note that the scalar products could be replaced by a loop for small
problems. If the number of unknow is smaller than 70 the loop is
faster (10 times faster if the number of unknowns is less than 10).
In some cases, for instance for the non linear filters, it helps to reduce this new gstep parameter
to get a positive definite hessian matrix. options_.gstep is now a 2*1 vector. The first element is
the old gstep parameter, the second element is the new gstep parameter. The step defined for the
computation of the hessian matrix is now:
h1=max(abs(x),sqrt(gstep(1))*ones(n,1))*eps^(1/6)*gstep(2);
- Implementation of the simulation of the block decomposed model using sparse matrix using MATLAB code:
New option in model : sparse
Example:
Model(SPARSE);
….
End;
- Implementation of the simulation of the block decomposed model using sparse matrix for static model.
- Krylov methods are implemented to solve the linear model J.y=b, with J the Jacobian matrix at each step of the Newton Method. Two Krylov's solvers are considered :
+ Generalized Minimum Residual Method (GMRES)
+ BiConjugate Gradients Stabilized (BICGSTAB)
To call them the SIMULATE command has to be completed with a new option : METHOD which could have three different values :
+ LU : (default value) Sparse LU resolution
+ GMRES
+ BICGSTAB
Example:
simul(periods=100, method=lu);
or simul(method=Gmres);
- Correction of various bugs in “simulate.dll”.
- Correction of simk.m
- Correction allowing a list of VALUES or PERIODS not necessarily separated by a comma.
git-svn-id: https://www.dynare.org/svn/dynare/dynare_v4@1447 ac1d8469-bf42-47a9-8791-bf33cf982152