| Re: Release of OpenMX Ver. 3.8 ( No.1 )|
- Date: 2016/04/05 20:37
- Name: Artem Pulkin
- The list of new features is actually amazing. The transport properties is a strong side of OpenMX from my point of view and it is nice to see that the functionality has been extended. VCR is a must-have for the DFT code so well done.
Though, looking back to geometry optimization with OpenMX, the LCAO-computed forces are considerably less accurate than plane-wave DFT. Also, I do not understand (because I did not use) Wannier functionality: wasn't it already present in 3.7?
Then, there were several requests about orbital decomposition of Bloch states in the band structure. Was it implemented at some point?
Finally, the functionality of eigenchannels (and transmission matrix). There is a well-known Caroli expression under "Transmission matrix" at http://www.openmx-square.org/tech_notes/tech20-1_0/tech20-1_0.html . I am indeed confused why is it called "transmission matrix": it is a square matrix of the size of the scattering region while the transmission matrix is a rectangular matrix with state-to-state transmission probabilities between left and right Bloch modes, see last pages of www.acmm.nl/molsim/han/2005/lnotes.pdf . Is there a simple relation between those?
| Re: Release of OpenMX Ver. 3.8 ( No.2 )|
- Date: 2016/04/06 13:44
- Name: T. Ozaki
> Looking back to geometry optimization with OpenMX, the LCAO-computed forces are
> considerably less accurate than plane-wave DFT.
What do you mean by "less accurate than plane-wave DFT"? In OpenMX the forces are calculated
by analytically differentiating the total energy with respect to the atomic position.
So if the total energy well converges with respect to basis functions, the calculated forces
can also be considered to be accurate on an equal footing. An exceptional case happens for
a metal with a high electric temperature. Since the effect of smearing by the Fermi function
is not taken into account in the formula, the calculated analytic forces deviate from the numerical force
calculated by differentiating the total energy numerically. It should be also noted that the geometry
optimization may suffer from the egg box effect if the fineness of the numerical grid is coarse,
implying the use of a higher cutoff energy of numerical grid is important to avoid the local minimum
> I do not understand (because I did not use) Wannier functionality: wasn't it already
> present in 3.7?
OpenMX Vers. 3.7 and 3.8 provide a tool to construct MLWFs, but not many functionalities
to calculate physical properties using the MLWFs. Instead, Wannier90 does. So, that's why
we interface OpenMX with Wannier90.
> Then, there were several requests about orbital decomposition of Bloch states in the band
> structure. Was it implemented at some point?
This functionality is supported in OpenMX Ver. 3.8.
Take a look at
| Re: Release of OpenMX Ver. 3.8 ( No.3 )|
- Date: 2016/04/08 04:53
- Name: Riemann <email@example.com>
- Dear professor Ozaki,
I'm very thankful for your excellent package OpenMX. It's really fast and acurate.
It would be very complete if it done phononics calculations. For this reason I have a question.
Do You have any plan for adding phononics part to OpenMX ?
| Re: Release of OpenMX Ver. 3.8 ( No.4 )|
- Date: 2016/04/08 08:48
- Name: T. Ozaki
Some of functionalities related to phonon will be available in the next release.
| Re: Release of OpenMX Ver. 3.8 ( No.5 )|
- Date: 2016/05/06 10:56
- Name: Mitsuaki Kawamura <firstname.lastname@example.org>
- Dear Artem
I am sorry that I was late to answer your question about the transmission matrix.
> I am indeed confused why is it called "transmission matrix": it is a square matrix of the size of the scattering region
It is my mistake that I called that matrix as "transmission matrix".
That matrix should not be called so.
> transmission matrix is a rectangular matrix with state-to-state transmission probabilities between left and right Bloch modes
You are right; this matrix is called "transmission amplitude matrix" in PRB 76, 115117.
Eigenchannels diagonalaize the square of the transmission amplitude matrix (it is called "transmission probability matrix" and its size is the DOS of the Bloch state of the left lead).
However the practical scheme to compute eigenchannels looks like the diagonalization of
G_C \Gamma_L G_C^\dagger \Gamma_R .
I am planning to improve on the technical note.