Firefly and PC GAMESS-related discussion club

 
Learn how to ask questions correctly  
 
 

 

Patrick SK Pang
pangsiukwong@gmail.com

Besides DIIS and CONJG, does Firefly include other solvers for the analytic Hessian calculation? If so, which cost less memory than CONJG?

Can the analytic Hessian calculation be applied to B3LYP?

Thanks!

Patrick

On Thu Sep 20 '12 4:20pm, Alex Granovsky wrote
----------------------------------------------
>Dear Patrick,

>The default solver is SOLVER=CONJG, a preconditioned conjugate
>gradient solver. The convergence criteria can be changed using
>$cphf cptol variable. The default is cptol=1.0d-5. While this
>criterion is interpreted a bit differently for SOLVER=CONJG and
>for SOLVER=DIIS, its default value guarantees that residual errors
>in the approximate solution of the CPHF equations do not result
>in errors in the computed wavenumbers larger than, say, 0.01 cm^-1 or so.

>The DIIS solver is usually more accurate than GONJG solver.
>It also uses less memory. However, it is slower than conjugate
>gradient solver.

>Some additional documentation on the keywords of the $cphf input
>group can be found here

>Hope this helps.

>Kind regards,
>Alex Granovsky
>
>
>
>
>On Wed Sep 19 '12 3:08am, Patrick SK Pang wrote
>-----------------------------------------------
>>Dear Alex,

>>What is the default solver for the analytic Hessian calculation?
>>How could I change the convergence criteria? Does the change in the convergence criteria affect the results significantly? I have many results calculated by the default solver. However, my computer has not sufficient memory so I use the DIIS solver to calculate several large molecules.

>>Thanks!

>>Regards,

>>Patrick

>>On Mon Sep 17 '12 9:31pm, Alex Granovsky wrote
>>----------------------------------------------
>>>Hi,

>>>>1. I have added $CPHF SOLVER=DIIS $END in the input file. The analytic Hessian calculation is still proceeding when MAXIMUM RESPONSE ERROR is down to E-6. However, without SOLVER=DIIS, the calculation can normally terminate and complete at MAXIMUM RESPONSE ERROR down to E-6. Why?

>>>Because different solvers use different convergence criteria.

>>>>2. How can I restart the analytic Hessian calculation?

>>>Sorry, it's impossible.

>>>Regards,
>>>Alex Granovsky