Hi,
The difficulty with charge assignment
schemes is largely due to historical artifacts.
All of the
earliest schemes for assigning net atomic charges were seriously flawed:
Mulliken, Lowden, Hirshfeld, etc.
Even Bader's AIM theory, which
has many beautiful theoretical properties, can give non-nuclear attractors in
some materials.
The persistent use of methods like Mulliken and
Lowden, which have been known for decades (longer than most of us have been
alive)
to be highly unstable and have no well-defined basis set limit,
has contributed to the confusion.
From a historical
perspective, this is similar to people's early aversion to "flying
machines".
For many decades, the concept of heavier than air
flying machines (what we now call airplanes) was considered a crackpot idea (a
"fiction").
This is because early attempts to build
"flying machines" were dismal failures.
It was only after
people were able to construct successful demonstrations that the societal
resistance to airplanes began to reverse.
Nowadays, it is hard for us
to imagine or justify the profound societal resistance to building airplanes
that existed around 1900.
Some of the modern charge
assignment methods are getting much better than the earliest
ones.
There is mounting experimental and computational evidence that
some of the most recent schemes
are orders of magnitude more reliable
than those developed decades ago.
Hopefully, this recent progress will
go a long way towards convincing computational chemistry researchers
that
the limitations we have become accustomed to can be
reversed after all.
I realize all of this will take time, but as a
computational chemistry community we need to make an effort for
progress.
The continual use of explicitly basis set
dependent methods like Mulliken and Lowden is a bit weary at times, and comes
off as a
serious shame in the face of better available recent
methods. As a computational chemistry community what can we do to support
progress?
Do you think it is appropriate at this time for
computational chemistry journals to develop standards that Mulliken and
Lowden
populations are not considered to be of publishable quality and
that published populations should be from methods with
a
well-defined basis set limit? One could argue that it is
important for published articles to show the fundamental limitations
of
Mulliken and Lowden schemes, but this has been established in
countless papers over the past 50 years.
I recall that my
master's thesis advisor, who is a purely experimental catalysis researcher,
warned us many years ago
about the limitations of Mulliken charges. If
experimentalists knew about this all the way back then, why as a
leading
computational chemistry community are we still fighting the
obvious decades later?
It may be that the computational
chemistry community has been slow to develop better alternatives.
I
believe the continued use of Mulliken and Lowden schemes as default methods in
some
quantum chemistry programs is due to the inherent limitations of
other available charge assignment schemes.
Until recently, the NPA/NBO
scheme, which has many great properties, was non-convergent for many
systems
near the complete basis set limit. My experience and tests
have shown this improved within the last year.
The NBO website and
manual says that recently they have improved the projection algorithm to make
it
more reliable. So, at least until very recently, the convergence of
the NPA/NBO scheme was not sufficiently
reliable to use it as a
default charge assignment method. The issue of non-nuclear attractors has
precluded
using Bader's scheme as a default method for assigning
net atomic charges. The Hirshfeld method assigns
net atomic charges
that are too small in magnitude. Methods such as CHELP, CHELPG, ESP, etc. do not
work
for dense nonporous materials that lack a van der Waals surface.
Computing the APT and Born effective charges
requires a
computationally expensive frequency (first-order perturbation theory)
computation or a series of energy calculations with
various atomic
displacements. Thus, it seems that until recently the computational chemistry
community has been stuck with
seriously flawed methods such as
Mulliken and Lowden as defaults in popular quantum chemistry
programs.
For the past couple of years, I and one of my
Ph.D. students have been developing a better alternative.
The goal of
our project is to develop a general-purpose atomic population analysis method
that works
nearly flawlessly across a broad range of diverse material
classes, including both molecular and solid-state
materials. We have
made extensive comparisons to experimental properties across diverse
materials.
Our objective is to develop a method that converges
reliably and efficiently and would be ideally suited
for use as a
default atomic population analysis method in popular quantum chemistry
programs
irrespective of the basis set type. We currently have a
manuscript on this new method under review.
We are hoping that it is
not too late for the computational chemistry community to improve.
We
earnestly solicit your support to enable the computational chemistry community
to make progress in the important
area of atomic population
analysis
methods.
Sincerely,
Tom