> Hi, siesta users:
> I noticed that one line in Siesta manual, saying:
> "The Ordern(N) subsystem is quite fragile and only
> works for systems with clearly separated occupied
> and empty states."
>
> Is there any obvious reason that it cannot handle
> metallic systems?
>
> Best regards,
> Bing
>

Sure, Bing, it is.
In metal, the states which are close to the Fermi energy may shift
back and forth between occupied and unoccupied domains. On each iteration,
the eigenvalues are ordered by their increased values, and the
AUFBAU PRINCIPLE applied -
the so many lowest states (counted over the Brillouin zone)
are declared OCCUPIED
(and contribute to the density in the next iteration),
and the rest thrown out.

The Order(N) approach (at list as it is implemented in Siesta, or more
general, as far as I remember) avoids diagonalization; instead it works
directly on orbitals or  density matrix elements, iterating them according
to a certain criterion. Hence the method does not know eigenvalues and
goes ahead for density, total energy, and other related properties.
(But does NOT provide the density of states, say).
Therefore the method cannot systematically order the eigenvalues
and find out where the occupied ones end. Instead, it needs to know this
in advance. In practice, in addition to the number of electrons
(which is known anyway), the chemical potential has to be provided
IN ADVANCE as a measure to identify the occupied states.
If you have a gap and safely put the chemical potential inside,
the calculation may go stable. If you have a metal, it ends up in a mess;
the iteration algorithms go astray.

Best regards

Andrei Postnikov

Responder a