[CPMD-list] Restarting interrupted TDDFT geometry optimization

[Eung-Gun Kim]

Dear Salomon,

Thanks very much for kindly responding to my message.  Restarting an interrupted
TDDFT geometry optimization run with LSSTAT allowed my calculation to continue
seamlessly and to produce results consistent with those of a non-stop run.  It
seems that in TDDFT geometry optimization, proper restarting is crucial.

Best regards,

EG

Quoting Salomon Billeter <ZRLSRB at ch.ibm.com>:

> Dear Eung-Gun,
>
> if you are using L-BFGS, restarting with HESSIAN has no effect on the
> optimization because L-BFGS does not use or update any Hessian but
> approximates the application of the inverse Hessian to the current gradient
> from the history of displacements and gradients. This history, along with
> the current trust radius, is read from the RESTART file using LSSTAT in the
> RESTART line:
> RESTART COORDINATES WAVEFUNCTION LSSTAT ADPTTL ...
> ADPTTL will make CPMD read the status of the adaptive tolerance from the
> RESTART file.
> Still, the difference between the final coordinates should be normally very
> small even if you don't read the mentioned information.
>
> Best regards,
> Salomon
>
>