Firefly and PC GAMESS-related discussion club



Learn how to ask questions correctly


Re^4: NCONV

Alex Granovsky
gran@classic.chem.msu.su


Dear Masoud,
Dear Andrei,

NCONV is an integer variable. Setting it to 4.5 actually
sets it to 4 (as the fractional part is truncated).
The initial convergence threshold is equal to 10-NCONV
for conventional SCF and to 2*10-NCONV for direct SCF.
However, for SCF geometry optimization runs with large
initial gradients, these thresholds are multiplied by some
factor depending on the magnitude of gradient. The idea is that
far from equilibrium geometry you do not need to know gradient
precisely. At later steps, threshold will be decreased to match
NCONV input.

Hope this helps,
Alex

On Fri Jan 28 '11 7:00pm, Andrei V Scherbinin wrote
---------------------------------------------------
>All that seems far from real quantum chemistry problems, as for me.

>I guess that it all depends on how the variable NCONV is actually defined
>in the source code of FF.

>If it is integer indeed, the answer could depend on how the FF executable
>reads in a real format value supplied by the user instead of the expected
>integer format value. It may depend, e.g., on the version of the compilator
>used to build the executable, or some other technical reasons. So, the
>result can be to some extent unpredictable.

>I think that using recommended (or even default) values is the best way
>to avoid problems like that, at least if you are a beginner. If you try to
>change them, do it carefully, following the lines in the program manual.
>
>
>Andrei
>
>
>
>On Fri Jan 28 '11 6:03pm, Masoud Nahali wrote
>---------------------------------------------
>>Dear Andrei

>>You think that the NCONV should be an integer, please note that 0.000385881 > 10^(-4) too (for NCONV=4.5). It seems that there is a reason that I can't find. Also, if the FF reads only an integer for NCONV why doesn't give any warning or error ? Many Thanks � � �

>>Best Wishes

>>Masoud
>>--------------
>>Masoud Nahali, Sharif University of Technology �
>>
>>
>>On Fri Jan 28 '11 3:10pm, Andrei V Scherbinin wrote
>>---------------------------------------------------
>>>It is likely that NCONV is an integer value, so I would set the values recommended in the manual,

>>>> (default is n=5, except CI or MP2 gradients n=6)

>>>if you use default OPTTOL value, and if you do not experience problems with basis set dependence,
>>>for instance.

>>>Of course, if you decreased OPTTOL (say, ~0.00001) for tighter geometry optimization, then it would
>>>be wise to converge your SCF tighter, say, up to NCONV=7 or so.

>>>Andrei
>>>

>>Dear Firefly Users

>>I have used "NCONV=4.5" in my input for the SCF convergence. It means that the convergence is reached when the density change between two consecutive SCF cycles is less than 10^(-4.5) or 3*10^(-5).

>>the lines below is one of my SCF steps :

>>ITER EX DEM �TOTAL ENERGY � � �E CHANGE �DENSITY CHANGE � �DIIS ERROR � � �INTEGRALS � SKIPPED
>> � � � � * * * � INITIATING DIIS PROCEDURE � * * *
>> �1 �0 �0 -3654.010061008 -3654.010061008 � 0.027738005 � 0.006150221 � � � � � � -1 � � � �-1
>> �2 �1 �0 -3654.010871171 � �-0.000810163 � 0.023518159 � 0.022479261 � � � � � � -1 � � � �-1
>> �3 �2 �0 -3654.013930379 � �-0.003059209 � 0.008856269 � 0.008333109 � � � � � � -1 � � � �-1
>> �4 �3 �0 -3654.013534182 � � 0.000396198 � 0.005316080 � 0.003565101 � � � � � � -1 � � � �-1
>> �5 �4 �0 -3654.014225050 � �-0.000690868 � 0.001903469 � 0.005148299 � � � � � � -1 � � � �-1
>> �6 �5 �0 -3654.014368343 � �-0.000143293 � 0.000956684 � 0.002133792 � � � � � � -1 � � � �-1
>> �7 �6 �0 -3654.014404861 � �-0.000036518 � 0.000503151 � 0.000987793 � � � � � � -1 � � � �-1
>> �8 �7 �0 -3654.014412598 � �-0.000007738 � 0.000385881 � 0.000328227 � � � � � � -1 � � � �-1

>> � � � � -----------------
>> � � � � DENSITY CONVERGED
>> � � � � -----------------

>>As it is clear 0.000385881 > 3*10^(-5) ; so why the density was converged ? Is there another criteria to reach the convergence ?

>>Is the "NCONV=4.5" logical when my "OPTTOL" value is 0.0001 ?

>>I appreciate your help in advance to understand the convergence criteria better.
>>
>>
>>
>>Best Wishes

>>Masoud
>>--------------
>>Masoud Nahali, Sharif University of Technology �


[ Previous ] [ Next ] [ Index ]           Mon Jan 31 '11 9:46pm
[ Reply ] [ Edit ] [ Delete ]           This message read 964 times