TURBOMOLE Modules > Riper
Convergence problem
antti_karttunen:
Hi,
you should remove the option "lenonly on" under $riper. When this option is included, riper skips the calculation of gradients (needed for optimization).
Best,
Antti
eliseo:
Hi again,
Thank you Antti. Now, I have a problem with the orbital occupation in the control file I have
alpha shells
a 1-332 ( 1 )
$beta shells
a 1-324 ( 1 )
all the parameters seem to perform an unrestricted calculation (define creates alpha and beta files....)
but in the job.last
------------------ Fractional Occupations ----------------
UHF calculation with a single Fermi level for both spins
Fermi level = -0.1608418607
alpha Electrons(/UC) = 328.0000000000
beta Electrons(/UC) = 328.0000000000
Current HOMO-LUMO gap = 0.90223E-01
----------------------------------------------------------
+--------------------------------------------------+
| SCF iteration 1 |
+--------------------------------------------------+
Number of electrons from P*S = 655.99999999978832
Numerical integration of the XC term:
Number of alpha electrons = 328.00008873660437
Number of beta electrons = 328.00008873660437
and the energies are equal to those of the restricted calculations. Any suggestion??
best wishes, thanks a lot
Eliseo
antti_karttunen:
Hi,
In your $riper data group, you have the keyword "sigma 0.01" (Fermi smearing). I have not personally checked out the behavior of this keyword, but at least in the molecular case, the use of Fermi smearing typically changes the occupation to the lowest-eneryg occupation. You can force a certain multiplicity with the "desnue" keyword:
The optional keyword desnue can be used within the $riper data group to constrain the number of unpaired electrons. This can be used to force a certain multiplicity in case of an unrestricted calculation, e.g., desnue 0 for singlet and desnue 1 for dublet.
Please see the manual for further information (Chapter 7.2.4)
Best,
Antti
eliseo:
Hi again,
using the desnue keyword, I can perform the unrestricted calculations for the periodic systems with two high-spin Fe(II) in the unit cell (desnue 8). However, in the restricted low-spin calculation the scf convergence is relatively fast but in the unrestricted high-spin I cannot reach the convergence and after some cycles the energy becomes completely crazy:
| TOTAL ENERGY = -6977.7539894147 |
| TOTAL ENERGY = -6948.8891499268 |
| TOTAL ENERGY = -6966.1812162700 |
| TOTAL ENERGY = -7003.9950034516 |
| TOTAL ENERGY = -6996.5229609826 |
| TOTAL ENERGY = -6982.0301324688 |
.....
| TOTAL ENERGY = -6140.3417339576 |
| TOTAL ENERGY = -2265.1918160646 |
| TOTAL ENERGY = -2289.9961494947 |
| TOTAL ENERGY = -4836.2935906069 |
| TOTAL ENERGY = 2140.9916183059 |
| TOTAL ENERGY = 2595.2900097573 |
| TOTAL ENERGY = 2758.9662267957 |
| TOTAL ENERGY = -1287.5215532921 |
| TOTAL ENERGY = 11508.6752014039 |
| TOTAL ENERGY = 15459.9397246313
I think that I don't have geometry problems, I tried as starting geometry the one employed
for the low-spin state and also the experimental high-spin structure. I was trying many scfdamp
and scforbitalshift but there are not significant differences in the convergence.
best wishes, thanks in advance
Eliseo
antti_karttunen:
Hi, did you also check the SCF Convergence FAQ at http://www.turbo-forum.com/index.php/topic,195.0.html?
Your original control file had "$scforbitalshift closedshell=.05". Better to use automatic shifting. Here are some settings that are still rather "mild":
--- Code: ---$scfdamp start=4.500 step=0.200 min=0.500
$scforbitalshift automatic 0.4
$scfiterlimit 200
--- End code ---
By the way, is there some particular reason why you are using "gridsize 5" and "radsize 50" for DFT grids? Typically m-grids like "gridsize m4" offer good accuracy and performance.
In the end it could also be that your system is just rather difficult to converge with GGA-PBE.
Best,
Antti
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version