A question about atk-se

[albatross]

  I found that the time computing transmission spectrum with no scf in atk-se is much longer than that in atk-DFT for the same structure. Is it right, or there is something wrong with my structure or my atk-se script.
 

[Anders Blom]

What you may actually have observed is the difference between the Krylov and Direct methods. In ATK 11.8, the faster Krylov method is the default, but in 12.2 we switched the default to Direct for safety reasons. The Krylov method can on occasion give negative transmission values, which is unphysical, and occurs due to numerical instabilities. When it works, however, the Krylov method is significantly faster (about 10 times in systems I have tested).

So, if this is indeed the case (and it would be if you built the ATK-SE script in VNL 12.2 - you should be able to see this by inspecting the script), you can always try the Krylov method. If you don't see any negative transmission values, all is fine.

[nori]

the time computing transmission spectrum with no scf in atk-se is much longer than that in atk-DFT for the same structure.

It hardly (never?) occurs in SCF but it can be possible in transmission.
Because the calculation time of transmission mainly depends on the size of transmission matrix, in other words, the total number of orbitals.
For instance, if you use SZ in DFT and Cerda in SE for Silicon, the number of orbitals per atom is 4(s:1, p:3) in DFT while 9(s:1, p:3, d:5) in SE.
In such a case, the calculation of transmission for SE costs time more than that for DFT.

[albatross]

OK, I got it. Thanks very much for the answers