[sharp-discuss] ask SHARP questions

[Claus Flensburg]

On Fri, Oct 01, 2004 at 02:26:13PM +0800, ???? wrote:
> Dear Sir:
> 
> I use the SHARP first time. I got trouble in my result. The result
> shows Section 2 : MAXIMUM-LIKELIHOOD REFINEMENT - NO CONVERGENCE YET
> (QARNING: SMALL CYCLES = NUM CYCLES!) Do you tell me what is wrong
> with my setting, please? Thank you
> 
> Sincerely
> 
> cholien

Hi Cholien,

usually a SHARP run consists of a number of BIG CYCLES, with each BIG
CYCLE having a varying number of SMALL CYCLES. There's a maximum
number of SMALL CYCLES pr. BIG CYCLES to protect against a run-away
optimisation.

This maximum number of SMALL CYCLES is set via the CYCLES card in the
SIN file usually done with SUSHI : Calculation Options : ML Parameter
refinement : using a maximum of _N_ small cycles for each.

The maximum number of SMALL CYCLES is usually 20. When your job does
not converge you should check for :

A) very small eigen-values (in the "Auxiliary Cycle Information")

B) is the optimisation on the right track - ie. are the steps getting
  smaller

If A) you need to check for a possibly wrong parametrisation eg.
missing fixed coordinate in polar space group.

If B) increase the allowed number of SMALL CYCLES to, say, 30.

Note that the message is only a warning, but if you see this for all
your BIG CYCLES it is more troublesome. You are welcome to send the
LIST.html, check_input.asc, cycle_info*, eigen* files to
sharp-develop at globalphasing.com if you need more help.

With best wishes,

ClAuS