Spin polarised calculation

Gryffindor

Hello,

I am currently working on calculating the response properties and density of states (DOS) for a spin-polarized system involving Cobalt, ZIF-9. To familiarize myself with the process, I conducted a small SCF test calculation. Despite specifying an increased number of SCF cycles, the calculation consistently terminates after 50 steps without achieving convergence.

I have referred to the CRYSTAL tutorials on SCF convergence tools and magnetic systems . However, I am still encountering the same problem. Could you kindly look into the input and output and let me know what might be going wrong, which keywords could I use?

Thanks

INPUT.d12.txt

OUTPUT.d12.txt

Aleks

Dear Gryffindor,

I had a quick look at your input/output files. You have increased the number of cycles for which the number of (up-down) electrons will be fixed, without increased the actual number of SCF cycles. This can be done with the MAXCYCLE keyword. For example, in your input file:

...
SHRINK
1 1
MAXCYCLE
200
SPINLOCK
0 100
...

You can always verify that by checking the exact value in the output file, which in your case reads:

...
*******************************************************************************
 MAX NUMBER OF SCF CYCLES      50  CONVERGENCE ON DELTAP        10**-16               <---
 WEIGHT OF F(I) IN F(I+1)      30% CONVERGENCE ON ENERGY        10**- 6
 SHRINK. FACT.(MONKH.)    1  1  1  NUMBER OF K POINTS IN THE IBZ      1
 SHRINKING FACTOR(GILAT NET)    1  NUMBER OF K POINTS(GILAT NET)      1
 *******************************************************************************
...

Also, a final note, care is advised to not have the number of SPINLOCK cycles for too long as you might converge while locking is active. Ideally, the SPINLOCK directive should remain active until the SCF process has found a numerically stable path along for which the system converges to the required configuration.

Hope this helps,
Cheers,
Aleks

Gryffindor

Dear Aleks,

Thank you for your response and suggestions. I tried using MAXCYCLE, and it worked.

However, I have another issue with optimizing the same structure; I get an intel error. I do not understand what it means. I have attached the input and output files below.

Thanks

INPUT.d12

OUTPUT.d12

Aleks

Hi,

from a quick look at your input/output files, it doesn't seem that the intel error is the final one, as the code started struggling with converging the SCF solution. You can monitor the behaviour with each cycle, where at cycle 8 you end up with:

...
 TTTTTTTTTTTTTTTTTTTTTTTTTTTTTT NUMDFT      TELAPSE      156.79 TCPU      151.49
 CYC   8 ETOT(AU)                 NaN DETOT       NaN tst  7.97E+01 PX  1.03E+02
 TTTTTTTTTTTTTTTTTTTTTTTTTTTTTT MPP_TRAF    TELAPSE      157.35 TCPU      152.05
...

after which the code stops. Without unnecessarily repetition, I would rather point you to a great source itself:
https://tutorials.crystalsolutions.eu/tutorial.html?td=SCF_conv&tf=conv_tut

Give a shout if you still struggle after trying those things.
And perhaps this part could be moved to the geometry optimization sub-topic.

Cheers,
A

eascrizzi

Dear Gryffindor,

As Aleks correctly pointed out, the problem in your calculation is that the SCF does not manage to converge. Particular attention has to be paid when dealing with spin polarized calculations: you should monitor in your output file how the spin density eveolves during the SCF procedure.

If you look in your OUTPUT.d12.txt file, you will see that at CYCLE 0 the spin density on your cobalt atoms looks like this

TOTAL ATOMIC SPINS  :
  3.0000000  -3.0000000   3.0000000  -3.0000000   3.0000000  -3.0000000

And it is totally coherent with your antiferromagnetic input setting. But at CYCLE 2 it looks like this

TOTAL ATOMIC SPINS  :
   0.6795143  -0.5759257   0.5922359  -0.6379062   0.6630575  -0.6747715

This means that the spin setting you wanted to obtain is already lost, the spin density on your cobalt atom is close to zero and the calculation is not going towards the solution you would like.
One thing that can be done is incresing the FMIXING, which helps mantaining your input spin setting. Furthermore, SPINLOCK and the DIIS convergence accelerator (present by deafault) do not go well together.

From the file I attach (in which both input and output are reported), you can see that if FMIXING is set equal to 95 and if the DIIS accelerator is activated only when the SPINLOCK is disabled, your system SCF procedure converges smoothly in 30 cycles. Your can try to optimize the geometry with these settings.

zif9.out

Hope it helps,
Eleonora

Gryffindor

Dear Eleonora and Aleks,

Thank you so much for the detailed explanation and for sharing the corrected input/output file. That really clarified the issue. I tried your suggested settings with FMIXING and SPINLOCK adjustments, and the geometry optimization is running well with no errors. Have a nice day!

Best,
Aparajita