Chapter 2 Theory background
2.3 Quantum mechanics
2.3.2 Density Functional Theory
DFT was formulated in the 1960s based on the Hohenberg-Kohn
theorem
27,28, which states that the energy of a system is a functional of the
electron density, and therefore the electron density is the only requirement
to calculate the energy of the system. This reduces the dimensionality of
the problem (the dimension of the wavefunction is 3N while the density is
three-dimensional), but the exact functional is not known, for which
reason approximations are derived by assuming that the electronic part of
the system can be seen locally as a uniform electron gas, from which all
properties can be derived.
This initial approach, called Local Density Approximation
29(LDA) did not
consider different α and β spin densities (open shell systems), and was
later replaced by the Local Spin Density Approximation
30(LSDA). In the
Generalised Gradient Approximation (GGA) methods
31the exchange and
correlation energy depends on the electron density and its first derivative.
Second order derivatives of the electron density (represented by the
Laplacian, ∇
2ρ) have also been implemented in the so-called “meta-GGA
methods”, which also include the orbital kinetic energy density
32. Hybrid
functionals are built to improve the exchange term. They contain a mix of
GGA and LSDA exchange and correlation terms plus exact HF exchange.
One example of this is the very popular B3LYP functional
33,34which
combines the Becke88 exchange functional and the Lee-Yang-Parr
correlation functional with 20% of exact HF exchange. Although DFT was
initially found to perform well for a variety of circumstances
35, failure has
since become evident when calculating energies (see for example
Izgorodina
et al.
36and references therein) although they remain good
methods for geometries. The main reason why DFT is inaccurate for
energy calculations is the exchange and correlation (XC) functional, for
which no exact form is known and therefore needs to be approximated.
There are two main problems of the approximations. The first of these is
the self-interaction (SI) error
37, which is a spurious interaction of each
electron with itself that causes large errors in the computation of energies.
The second problem is the improper description of the tail of the electronic
wavefunction
38,39. The first problem arises from the construction of the
Coulombic energy functional and results in unreliable energies
32, the
second affects also geometries, and will be discussed next.
Dispersion forces, also known as London forces, are attractive forces that
originate when a fluctuation in the electronic density in one molecule
creates an instantaneous dipole, which in turn induces a dipole moment in
another molecule. Although weak, these interactions play an important
role in the structure of bigger molecules and complexes, such as proteins
and nucleic acids. They decay as R
-6and are therefore long-range forces
(Grimme et al.
40reviews dispersion-corrected DFT-D methods and the
importance of dispersion in biomolecules). Recent work has pointed out
the importance of dispersion energy when analyzing protein-substrate
binding
41and protein folding
42. There is evidence
43that the recently
developed DFT-D methods provide satisfactory results for the interaction
energies of biologically relevant groups.
Pauli’s exclusion principle states that two fermions cannot occupy the
same space at a time, which implies that the two-particle wavefunction
must be antisymmetric to exchange of the particles. This is the origin of
the exchange energy. This is part of the HF method but it needs to be
added in the case of DFT. HF, however, includes no electron correlation –
which DFT does by construction. Because of the lack of dynamic effects,
the electron repulsion is underestimated by HF. These effects give a
reduced probability of finding a second electron close to where a first
electron is, and a higher chance for this event far from the first electron in
HF methods. This is known as the exchange-correlation hole, which is
non-local – that is to say, it is not limited to one atom but rather
delocalized over the entire system. The exchange and correlation (XC)
functional in DFT is local, and so classical DFT methods fail to describe
long-range interactions properly. This inability of DFT methods to
reproduce dispersion effects was noted at the beginning of this century
39,44.
Since then, new functionals with this added capability have been created
45- 48which show improved performance
49. In these methods, the total energy
E
DFT-Dis the DFT (Kohn-Sham, KS) energy plus a correction for the
dispersion energy E
disp47.
E
DFT-D= E
DFT(KS)+ E
disp(2.3)
where E
dispis calculated as
Edisp = -s6
!!!" !!"! ! !!!!! !!! !!!fdamp(Rij)
(2.4)
The dispersion correction is scaled by the scaling factor s
6, which is
empirically determined for each density functional. N is the number of
atoms, !
!!"is the dispersion coefficient for a given atom pair (ij) and R
ijis
the distance between atoms i and j. The analytical form of the dispersion
energy was shown by Alonso et al
50and to have asymptotic behaviour,
which will produce unphysical effects at short distances. This was
addressed long time ago by Ahlrichs
et al.
51who added dispersion
corrections to a semi-empirical method, and who proposed the use of a
damping function to eliminate the spurious effects of adding dispersion
energies at short distances
51. The damping function f
dampis as follows:
!
!"#$(!)
=
! !!!!!(!
!!!!)
(2.5)
An analytical expression for the dispersion forces has been derived for use
with DFT functionals
50although the empirical correction just discussed is
the one of choice for dispersion-corrected DFs so far – likely due to its
lower cost as indicated in Chai et al.
46.
In this work, the following functionals are used. ωB97XD
46is a hybrid
functional based on the ωB97X long-range corrected functional
52(that was
developed from B97
53with long-range corrections) with an unscaled
dispersion correction, which is equivalent to setting the factor s
6to 1. This
functional is claimed to be free of long-range self-interaction, although
there is some degree of self-interaction at short range. It is important to
note that the long-range correlation is only based on empirical dispersion
corrections and therefore the dispersion effects do not enter the KS
orbitals. The M06 suite of methods
54are hybrid meta-GGAs with exchange
and correlation functionals based on the M05 suite
55. All these include a
percentage of HF exchange, which is around 25-27% for M05 and M06 and
double that for the -2X version of the functional. DFT-D is still a work in
progress, with more accurate and widely applicable functionals expected
in the future.
In document
Computational studies of the E3 carboxylesterase from Lucilia cuprina
(Page 43-47)