Guangping, The differences with respect to Gaussian results could com from pseudos and bases - if you could port both to your siesta calculation, then in principle with a suitable cutoff and a large enough cell you would have a structure which already relaxed. Unfortunately I don't know exactly how Gaussian differs from Siesta when it comes to representing the charge density and wavefunctions, so I can't say more.
You can try two options: the first is to make MD.MaxCGDispl 0.04 Ang or even a bit less. This will make the initial CG displacements over the PES smaller, and then maybe you'll get to convergence faster. The second option is to try a Broyden relaxation with MD.Broyden.Initial.Inverse.Jacobian 0.1, which will also make the initial displacements over the PES surface smaller (and the subsequent ones as well, hopefully). You could try both and **do** check that both give consistent final geometries. Marcos On Wed, May 5, 2010 at 12:52 PM, Guangping Zhang <[email protected]> wrote: > Dear siesta users and developers: > When I do a geometry optimization,even a small molecular about a dozen > atoms,it needs more than one hundred steps to find the stationary point.It > is the same even if I pre-optimized in Gaussion 03.My force convergence is > 0.02eV/A,and the DM tolerance is 1.0d-4,MaxDisplLength 0.1Ang.I use > siesta-3.0-b,is it related to the above parameters used? > So who can help me ? > Thanks in advance. > > 2010-05-05 > ------------------------------ > Guangping Zhang >
