Author Topic: Back Engine exception 92 and some ather questions  (Read 4514 times)

0 Members and 1 Guest are viewing this topic.

Offline ziand

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: de
  • Reputation: 5
    • View Profile
Back Engine exception 92 and some ather questions
« on: March 8, 2011, 11:28 »
Hallo,
I am testing the new version (still ATK 11.2.b2 in the moment, we will buy new hardware for replacing our extremly old and unreliable license server soon and ask for a 11.2.0 license after that).

1.) Are there major differences between 11.2.b2 and 11.2.0 / 11.2.r1?

2.) At the beginning of the transmission spectrum calculation, I got the following error:
Quote
** Back Engine Exception 92 : Too large a difference in calculated transmissions has been detected among processes
** Location : calculatetransmissioncoefficient.cpp:166
What could be the reason?
I have to mention that the device density matrix calculation did not converge and there were large difference density values on some atoms.

3.) As I'm often dealing with long metallic systems I am curious about the new Kerker preconditioner. The documentation there is a bit confusing to me. In the explanation of energy_q0 what do you mean with "i.e 0.5*q0*q0" ? Or "energyQ0(): Return 0.5*q0*q0".   Is enegry_q0 equal to q0 or is energy_q0 equal to 0.5*q0*q0 ?

4.) Some things about the ReferenceManual: In the DeviceCalculator sections there is no hint about the new FFT2D solver. On some pages (e.g. 97, 98, 101, ...) in the new Manual there is lots of red text that looks as it should be formatted in a different way.


Online Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5575
  • Country: dk
  • Reputation: 96
    • View Profile
    • QuantumATK at Synopsys
Re: Back Engine exception 92 and some ather questions
« Reply #1 on: March 8, 2011, 13:18 »
1.) Are there major differences between 11.2.b2 and 11.2.0 / 11.2.r1?

There are certainly a few differences, including several crucial bug fixes of problems in the b2 version. Moreover, you cannot use the b2 version beyond 1 April, it will simply not work any more.

2.) At the beginning of the transmission spectrum calculation, I got the following error:
** Back Engine Exception 92 : Too large a difference in calculated transmissions has been detected among processes
** Location : calculatetransmissioncoefficient.cpp:166
What could be the reason?
I have to mention that the device density matrix calculation did not converge and there were large difference density values on some atoms.

For instance, this is fixed in 11.2.0 :)

3.) As I'm often dealing with long metallic systems I am curious about the new Kerker preconditioner. The documentation there is a bit confusing to me. In the explanation of energy_q0 what do you mean with "i.e 0.5*q0*q0" ? Or "energyQ0(): Return 0.5*q0*q0".   Is enegry_q0 equal to q0 or is energy_q0 equal to 0.5*q0*q0 ?

I agree it's a bit cryptic 8) The algorithm uses q0, which is an inverse length. If you compute E=hbar^2 q0^2 / 2m, this will have the dimension of energy. In atomic units, hbar and the electron mass m are both 1, so we can compute q0 from E=0.5*q0*q0. The input to Kerker is energy_q0=E, and from this we can compute q0 internally.

4.) Some things about the ReferenceManual: In the DeviceCalculator sections there is no hint about the new FFT2D solver. On some pages (e.g. 97, 98, 101, ...) in the new Manual there is lots of red text that looks as it should be formatted in a different way.

From a formal perspective, the Poisson solver is completely uncoupled from the geometry; you can run the same geometry with many different boundary conditions. We also do not discuss the other Poisson solvers on this page. So I don't think it's an appropriate place for it, although there is a general need for a more detailed discussion on the two-probe geometry, which partly is available here: http://quantumwise.com/publications/tutorials/mini-tutorials/99

The red things should be fixed - we'll look into it, thanks!

Offline ziand

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: de
  • Reputation: 5
    • View Profile
Re: Back Engine exception 92 and some ather questions
« Reply #2 on: March 8, 2011, 18:44 »
Thanks for your answer. I mentioned the Poisson solver for device calculators, because on page 124 and 131 is written "The poisson_solver must be set to the MultigridSolver ...". This suggests that you can ONLY use multigrid, but as I think also FFT is possible (in some situations) and now FFT2D should be the best choice.

Online Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5575
  • Country: dk
  • Reputation: 96
    • View Profile
    • QuantumATK at Synopsys
Re: Back Engine exception 92 and some ather questions
« Reply #3 on: March 8, 2011, 21:06 »
Indeed, this advise should be updated. FFT2D is your best choice; full multigrid is only needed when you have gates that break the periodicity in X/Y, and FFT is not recommended for two-probes at all (except perhaps for quick-and-dirty tests at zero bias).