Author Topic: calculation not converge at zero bias.  (Read 7074 times)

0 Members and 1 Guest are viewing this topic.

Offline Jenny

  • Heavy QuantumATK user
  • ***
  • Posts: 61
  • Reputation: 0
    • View Profile
calculation not converge at zero bias.
« on: March 24, 2014, 17:17 »
Dear All,

I use quantumwise atk to do the parallel calculation under windows system.

And last week I met a problem that when I tried to get the transmission spectrum of my device at zero bias, it says the calculation did not converge shown as below. I've already checked the length of electrode and it says "Truncation of some second-order interactions will occur in both left and right electrode." I think this is good enough to do the calculation right? I think the problem is my k-point sampling. So how can I adjust my k-point sampling to make the calculation to converge?

btw: I've attached my script in this post.

+------------------------------------------------------------------------------+
|  99 E = -990.739 dE =  4.022227e+01 dH =  6.510678e+00                       |
+------------------------------------------------------------------------------+
################################################################################
#                                                                              #
# Warning: The calculation did not converge to the requested tolerance!        #
#                                                                              #
################################################################################


Thank you.

Jenny

Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5565
  • Country: dk
  • Reputation: 93
    • View Profile
    • QuantumATK at Synopsys
Re: calculation not converge at zero bias.
« Reply #1 on: March 25, 2014, 00:22 »
The electrodes are short but probably ok-ish. The main problem is more likely that you have only as many Cu layers in the central region as in the electrode, that's a bit too few.

Offline Jenny

  • Heavy QuantumATK user
  • ***
  • Posts: 61
  • Reputation: 0
    • View Profile
Re: calculation not converge at zero bias.
« Reply #2 on: March 25, 2014, 19:04 »
The electrodes are short but probably ok-ish. The main problem is more likely that you have only as many Cu layers in the central region as in the electrode, that's a bit too few.

Dr. Anders Blom,

Thank you.

So you mean I need to add more Cu layers in the central region?

Jenny


Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5565
  • Country: dk
  • Reputation: 93
    • View Profile
    • QuantumATK at Synopsys
Re: calculation not converge at zero bias.
« Reply #3 on: March 25, 2014, 22:14 »
Yes

Offline Jenny

  • Heavy QuantumATK user
  • ***
  • Posts: 61
  • Reputation: 0
    • View Profile
Re: calculation not converge at zero bias.
« Reply #4 on: March 31, 2014, 21:37 »
Yes

Thank you Dr. Anders Blom.

But there is another problem I have.

When I tried to optimize the bulk system, how could I choose the optimization condition? Usually, bulk system is larger than unit, so it may take a long time to get the optimization geometry. So can I set the max force and max stress larger than unit cell? for example, my optimization condition for the unit cell is max force=0.05eV/A and max stress=0.05eV/A. When I use the unit cell to form the central region of my device and Cu as electrode to do bulk system relaxation, I choose max force=1eV/A and max stress=1eV/A. And if it is acceptable, how can I decide which kind of condition is good enough for a further analysis?
 
And another thing is that when I got the optimized geometry, the copper layers in the bulk system get torsion, I am not able to form the device from bulk using the function in quantumwise. what should I do? The optimized bulk configuration is attached.

Thank you.

Jenny

Offline Umberto Martinez

  • Supreme QuantumATK Wizard
  • *****
  • Posts: 479
  • Country: dk
  • Reputation: 26
    • View Profile
Re: calculation not converge at zero bias.
« Reply #5 on: April 3, 2014, 16:21 »
Quote

When I tried to optimize the bulk system, how could I choose the optimization condition? Usually, bulk system is larger than unit, so it may take a long time to get the optimization geometry. So can I set the max force and max stress larger than unit cell? for example, my optimization condition for the unit cell is max force=0.05eV/A and max stress=0.05eV/A. When I use the unit cell to form the central region of my device and Cu as electrode to do bulk system relaxation, I choose max force=1eV/A and max stress=1eV/A. And if it is acceptable, how can I decide which kind of condition is good enough for a further analysis?

You can not increase the convergence criteria in this way when increasing you system size.
max forces 0.05 eV/Å is a good choice which you should keep also for large systems.
Indeed, if you start form a "bad" initial configuration the calculation will take a while.


Quote
And another thing is that when I got the optimized geometry, the copper layers in the bulk system get torsion, I am not able to form the device from bulk using the function in quantumwise. what should I do? The optimized bulk configuration is attached.

when you run the bulk optimization fix the last three layers of your Cu slabs in the OptimizeGeometry block.
see for example here: http://quantumwise.com/documents/tutorials/latest/Au_Pentacene/index.html/chap.au111pc.1layer.html#sect2.relax.au111-2x3s3

in this way the device from bulk plugin will recognize the periodicity of the system and build a proper device.

Offline Jenny

  • Heavy QuantumATK user
  • ***
  • Posts: 61
  • Reputation: 0
    • View Profile
Re: calculation not converge at zero bias.
« Reply #6 on: April 10, 2014, 20:32 »
The electrodes are short but probably ok-ish. The main problem is more likely that you have only as many Cu layers in the central region as in the electrode, that's a bit too few.

Hi, Dr. Anders Blom.

I've tried to extend the Cu layers in the central region, now the calculation converges now. Thank you. But I was wondering why the length of Cu layers would influence the convergence of the calculation? Would you briefly tell me the reason or give me some references?

Thank you again.

Best,

Jenny


Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5565
  • Country: dk
  • Reputation: 93
    • View Profile
    • QuantumATK at Synopsys
Re: calculation not converge at zero bias.
« Reply #7 on: April 11, 2014, 14:30 »
I've tried to extend the Cu layers in the central region, now the calculation converges now.

Great!

But I was wondering why the length of Cu layers would influence the convergence of the calculation? Would you briefly tell me the reason or give me some references?

This is because in the middle of the system your electron density is very different from the electrodes. The electrodes are made of bulk Cu, the middle is a nanotube. The Cu layers in the central region are intermediate - they are not pure Cu bulk-like, but provide a gradual transition from "Cu surface connected to a CNT" to bulk Cu, and if this transition region is too short, you get into convergence problems.

Offline Jenny

  • Heavy QuantumATK user
  • ***
  • Posts: 61
  • Reputation: 0
    • View Profile
Re: calculation not converge at zero bias.
« Reply #8 on: April 18, 2014, 19:01 »
Quote

When I tried to optimize the bulk system, how could I choose the optimization condition? Usually, bulk system is larger than unit, so it may take a long time to get the optimization geometry. So can I set the max force and max stress larger than unit cell? for example, my optimization condition for the unit cell is max force=0.05eV/A and max stress=0.05eV/A. When I use the unit cell to form the central region of my device and Cu as electrode to do bulk system relaxation, I choose max force=1eV/A and max stress=1eV/A. And if it is acceptable, how can I decide which kind of condition is good enough for a further analysis?

You can not increase the convergence criteria in this way when increasing you system size.
max forces 0.05 eV/Å is a good choice which you should keep also for large systems.
Indeed, if you start form a "bad" initial configuration the calculation will take a while.


Quote
And another thing is that when I got the optimized geometry, the copper layers in the bulk system get torsion, I am not able to form the device from bulk using the function in quantumwise. what should I do? The optimized bulk configuration is attached.

when you run the bulk optimization fix the last three layers of your Cu slabs in the OptimizeGeometry block.
see for example here: http://quantumwise.com/documents/tutorials/latest/Au_Pentacene/index.html/chap.au111pc.1layer.html#sect2.relax.au111-2x3s3

in this way the device from bulk plugin will recognize the periodicity of the system and build a proper device.


Hi, Umberto Martinez.

Thank you very much for your instruction. But I have several questions regarding to the link you put in the previous reply.
1.It suggests to do parallel calculation of geometry optimization, is it possible? In the different nodes, Can atoms move to the same spots at the same time?
2.In the example, it used two step OptimizeGeometry, why? I thought it should be easier to get relaxed geometry with only two layer of gold atoms fixed as the second step suggested.
3.If we set the optimized condition to 0.01eV/A for such a large system, it usually takes a long time to finish. Can we first set the condition to a dummy value like 1.0eV/A and then use the "optimizedgeometry" to do futher relaxation with the final value 0.01eV/A?

Looking forward to your reply.

Jenny

Offline Umberto Martinez

  • Supreme QuantumATK Wizard
  • *****
  • Posts: 479
  • Country: dk
  • Reputation: 26
    • View Profile
Re: calculation not converge at zero bias.
« Reply #9 on: April 23, 2014, 10:47 »

1. yes it is possible. Take a look to our Parallel Guide
and in particular http://quantumwise.com/documents/tutorials/latest/ParallelGuide/index.html/chap.strategy.html

2. You do not necessarily need these two steps but the strategy follow the same idea as you also suggest in the next question.

3. You are right, 0.01 eV/Å is pretty low value and yes, it may take long time to reach this convergence limit.
Indeed you can follow the strategy you suggest but increase the value to max 0.05 eV/Å to have reasonable results.

Note that you will not really save time (total computational time) by doing two calculations (0.05+0.01) instead of one (0.01).
However, you will have the 0.05 output files for further analysis.
You may use this value for "screening" or test calculations and increase the convergence criteria only for your final results.

Finally, note that this value also depends on your system. You can find that a value of 0.05 or 0.02 gives you converged results.