Why scf calculation for a two probe system only converge for the SZ basis set?

[songll0808]

Hi, I have set up a two probe system consisting of aluminum electrode and Li2Pc molecule, and the Li2Pc molecule has been optimized strictly. But the system can't converge except for the SZ basis set?

What 's the problem ?

Thank for your attention.

[Nordland]

In general, it is often harder to converge a problem with more degrees of freedom than one with few degrees of freedom. Hence the hydrogen atom does not really have any degrees of freedom, and therefore it is practical impossible for it to notconverge, where a complex two-probe system with many extra degrees of freedom from a larger basis set is somewhat more complex.

Depending on what you have done so far, there is a number of options you could try to continue on, but in my opinion the best approach would be first to determine if this is an "element" problem or if it is a twoprobe problem.

Therefore I would investigate if it is possible to converge the system as a normal bulk system. If yes, then it is most likely a twoprobe problem, and you can then find a lot of inspiration for dealing with this by browsing the forum. If it is not possible to make it converge for bulk, then it is most likely due insufficient accuracy of the calculation, and hence the parameters should be tuned for better accuracy.

[songll0808]

Thank for your timely reply. 

However, I'm still in doubt.

Firstly, whether the second part in scf calculation (Equivalent Bulk Calculation (Initial Density for TwoProbe)) is the normal bulk calculation? This part is converged well. But the third part ( Twoprobe calculation) can't be converged.

Secondly, I have made another calculation, in which some parameters for accuracy are increased, mesh_cutoff = 250, and K points (10,10,100). Unfortunately, the third part still is not converged.

Can you give me some more suggestion?

Thank you very much! 

[zdhlover]

 
I think you'd better send your error message or the log file to here,which may  be easy see where is the problem.

By the way ,I think your system or the parameters that you choose in your scf may be so big which make the scf out of memory in your computer.

[songll0808]

I think you'd better send your error message or the log file to here,which may  be easy see where is the problem.

By the way ,I think your system or the parameters that you choose in your scf may be so big which make the scf out of memory in your computer.

  Thank you for your advice!

The attach is the output file of scf and error message.

[zdhlover]

In the error txt, it is clear see where is the problem with the last sentence "ATKError: Exceeded maximum number of self-consistent iterations."
so please increase the Max Steps in the Iteration Control option with default number "100".

By the way ,from the scf.out file I also doubt whether the two-probe system you built was correct or not, for example, do you have enough screen layers in surface or do you have enough vacuum in the transverse plane at central region to screen the interaction of the repeated Li2Pc molecules? I suggest you check your geometry careful. And if you don't want others to see your structure,you can send your geometry to Anders Blom , if not,you can demonstrate it here,which will be convenient to see the real problem.

[nori]

According to "scf.out", SCF calculation seems to be converged with more dozens of iterations.
But unfortunately converged density matrix will be unphysical because of electron reduction in the whole system.
The system has 275 electrons as neutral, on the other hand, SCF calculation goes toward the state that loses 32 electrons.

It's probably due to "electrode_constraint".
If you use "ElectrodeConstraints.Off", I recommend that you use "ElectrodeConstraints.RealSpaceDensity" instead of it.

[songll0808]

According to "scf.out", SCF calculation seems to be converged with more dozens of iterations.
But unfortunately converged density matrix will be unphysical because of electron reduction in the whole system.
The system has 275 electrons as neutral, on the other hand, SCF calculation goes toward the state that loses 32 electrons.

It's probably due to "electrode_constraint".
If you use "ElectrodeConstraints.Off", I recommend that you use "ElectrodeConstraints.RealSpaceDensity" instead of it.

Thank you for your advice!

Though scf calculation is still unconverged, the calculation results seem better than before. So I doubt there is some problem with my system.

hehe, I have done my work again, from structure relaxation. After adopting your advice, the scf is converged on the SZP basis set. But when I change the exchange_correlation_type from LDA to GGA, Error happens.

# ----------------------------------------------------------------
# Electrodes Calculation
# ----------------------------------------------------------------
# sc  0 : Fermi Energy =    0.00000 Ry
# sc  1 : Fermi Energy =   -0.27930 Ry  dRho =  9.5071E-01
# sc  2 : Fermi Energy =   -0.28228 Ry  dRho =  9.9687E-03
# sc  3 : Fermi Energy =   -0.29443 Ry  Etot = -194.74281 Ry  dRho =  3.4378E-02
# sc  4 : Fermi Energy =   -0.29895 Ry  Etot = -194.74312 Ry  dRho =  9.7081E-03  dEtot = -3.1220E-04 Ry
# sc  5 : Fermi Energy =   -0.29900 Ry  Etot = -194.74317 Ry  dRho =  4.7630E-03  dEtot = -4.7948E-05 Ry
# sc  6 : Fermi Energy =   -0.29958 Ry  Etot = -194.74317 Ry  dRho =  1.7574E-03  dEtot = -4.8249E-06 Ry
# ----------------------------------------------------------------
# Equivalent Bulk Calculation (Initial Density for TwoProbe)
# ----------------------------------------------------------------
rank 5 in job 1  cu069-ib_34585   caused collective abort of all ranks
  exit status of rank 5: return code 137

Meanwhile, the error output :
/vnl-2008.10.0/atk/bin/atk: line 3: 12068 Killed                  LD_LIBRARY_PATH=$EXEC_DIR/../lib $EXEC_DIR/atk_exec $*

~What's wrong?

when I execute the atk command, the python file can't be executed.

the error output:
Traceback (most recent call last):
  File "scf.py", line 53, in ?
    runtime_parameters = runtime_params
KeyError: 'Electrode'

Why?

[Anders Blom]

First problem is related to your cluster. Might be temporary, might be out of memory.

Second script (different one???) contains a typo or other mistake, and you need to submit it so we can see what the problem is.

[songll0808]

First problem is related to your cluster. Might be temporary, might be out of memory.

Second script (different one???) contains a typo or other mistake, and you need to submit it so we can see what the problem is.

Yes, It should be problem with the cluster. When it was calculated by serial not by parallel, it worked well. No error occured.

I'm truly grateful for you and other people's help. Thank you very much!