Revisiting Artifacts of Kohn-Sham Density

Functionals for Biosimulation

arXiv:2406.01520v2 [physics.chem-ph] 11 Jul 2024

Samuel A. Slattery, Jaden C. Yon, and Edward F. Valeev∗

Department of Chemistry, Virginia Tech, Blacksburg, VA 24061

E-mail: [email protected]

Abstract

We revisit the problem of unphysical charge density delocalization/fractionalization

induced by the self-interaction error of common approximate Kohn-Sham Density Func-

tional Theory functionals on simulation of small to medium-size proteins in vacuum.

Aside from producing unphysical electron densities and total energies, the vanishing

of the HOMO-LUMO gap associated with the unphysical charge delocalization leads

to an unphysical low-energy spectrum and catastrophic failure of most popular solvers

for the Kohn-Sham (KS) self-consistent field (SCF) problem. We apply a robust quasi-

Newton SCF solver [Phys. Chem. Chem. Phys. 26, 6557 (2024)] to obtain solutions

for some of these difficult cases. The anatomy of the charge delocalization is revealed

by the natural deformation orbitals obtained from the density matrix difference be-

tween the Hartree-Fock and KS solutions; the charge delocalization can occur not only

between charged fragments (such as in zwitterionic polypeptides) but also may involve

neutral fragments. The vanishing-gap phenomenon and troublesome SCF convergence

are both attributed to the unphysical KS Fock operator eigenspectra of molecular frag-

ments (e.g., amino acids or their side chains). Analysis of amino acid pairs suggests that

the unphysical charge delocalization can be partially ameliorated by the use of some

1
 range-separated hybrid functionals, but not by semilocal or standard hybrid function-

als. Last, we demonstrate that solutions without the unphysical charge delocalization

can be located even for semilocal KS functionals highly prone to such defects, but such

solutions have non-Aufbau character and are unstable with respect to mixing of the

non-overlapping “frontier” orbitals. Caution should be exercised when unexpectedly

small (or vanishing) HOMO-LUMO gaps and atypical SCF convergence patterns (e.g.,

oscillatory) are observed in KS DFT simulations in any context (bio or otherwise).

1 Introduction

The unphysical behavior of all Kohn-Sham density functional approximations (DFAs) causes
a number of artifacts, such as:

• Unrealistically small HOMO-LUMO gaps 1,2 and excitation energies, 3 especially for
states with charge transfer 4 and/or Rydberg 5 characters (although some dispute these
notions 6 ),

• charge fractionalization/delocalization, such as partial detachment of an electron in

anions as the basis approaches completeness, 7,8

• predicted reaction barriers too low, 9–13

• incorrect dissociation curves for both cations 14 and some neutrals, 15

• overbinding of opposite-charged moieties, 16 and charge transfer complexes. 17–19

While these failures are well-recognized by the community, some artifactual behaviors of KS
DFT in the context of biosimulations are less understood and invite more controversy. For
example, it was discovered that SCF convergence problems appeared for some biomolecules
when using semilocal DFAs in vacuum. 20–23 Subsequent work concluded that no fundamental
problem exists for semilocal DFAs when applied to such systems, and that the problem was
rather with the way the system was “prepared”. 24,25 Ultimately this came down to two major

2
factors: missing solvent effects and geometry relaxation. Thus, commonly proposed solutions
for the problem include: (1) use of a solvent model (particularly implicit 25–27 or even point
charges 21 ), (2) application of geometry relaxation, 24,25 and (3) change to range-separated
functionals 23,28,29 (e.g. ωB97). We agree that the problem is in the functional, but note
here that even range-separated functionals do not systematically solve these problems for
all cases. 30 The idea that the fundamental problem is electrostatic in nature, rather than a
result of more essential problems with the approximate functionals has persisted. 26,31
Many researchers have realized that some of these problems can have striking conse-
quences for the use of KS DFT for systems containing charged moieties. For example,
Jensen 7 recognized that the artificially high HOMO energy of the negatively charged depro-
tonated carboxylic acid group in a peptide zwitterion in vacuum could result in fractional
charge transfer. While it has long been known that approximate KS DFT struggles with an-
ions, 7,21,32–38 the work of Jensen indicated the possibility of transfer of electron density within
a single molecule. This possibility relates to the idea of “delocalization error” 39–41 where the
density is too spread out due to the nonlinearity of the KS DFT total energy with respect
to fractional electron number. 8,14,39,42,43 Clearly, this can result in incorrect electrostatic de-
scriptions, 24,44 but it can also cause the aforementioned SCF convergence problems. 20,24 One
proposed solution to some of these issues involves evaluating KS DFT energies using HF
densities, 37,38,45 although the lack of self-consistency would then be a substantial concern.
In this work, we revisit the convergence problems of SCF for local and semilocal DFAs
applied to medium-sized biomolecules in vacuum. In Section 3.1 we examine solutions for
a set of 17 polypeptides considered in Ref. 21 using a robust local-convergence SCF solver
(QUOTR 46 ), thus allowing us to characterize the “true” (energy minimized) solutions even
for cases where no solutions could be found in Ref. 21; the SCF convergence issues are
correlated to the unphysically small (or even vanishing) HOMO-LUMO gap. Although the
incorrect density predicted by approximate functionals has been investigated for zwitteri-
ons, 27,44 the orbital structure (at least for the semilocal functionals) was not examined in

3
detail. Section 3.1 includes details of orbital-based analysis of KS→HF deformation densities
revealing in pinpoint detail the anatomy of unphysical KS fractionalization/delocalization of
charges in such systems. The use of deformation density to decompose a change in density
into orbitals contributions has precedent in the context of fragment to complex interaction, 47
but it seems to be unknown for comparing different quantum chemistry methods. Addition-
ally, we show that by using a local solver we can obtain SCF solutions with properly-localized
charges in some cases, which are actually non-Aufbau (and non-energy minimizing). We fol-
low this in Section 3.2 by scanning through several popular functionals for all 20 naturally
occurring amino acids to reveal which combinations of amino acids with which DFAs can
cause such issues. In particular, we note that range-separation, while almost always helpful,
does not eliminate the possibility of unphysical charge delocalization and problematic SCF
convergence.

2 Technical Details

HF and KS DFT computations on polypeptides in Section 3.1 were performed with a devel-
opmental version of the Massively Parallel Quantum Chemistry (MPQC) version 4 program
package, 48 using the recently developed QUOTR SCF solver. 46 The maximum L-BFGS his-
tory size was set to 15 (parameter m), and the initial guess was the unperturbed version of
the extended-Hückel-like guess used previously, except for 1RVS which used the perturbed
guess. The regularizer threshold (tr ) was lowered to 0.15, and the history was also reset
whenever the RMS gradient crossed 1 × 10−6 (either crossing below or coming back up
again). However, the calculations on the glycine systems in Section 3.2.1 used the same pa-
rameters as in ref. 46. One final difference in the solver is that the history data was trimmed
whenever the lowest eigenvalue of the matrix VT V was below −1×10−15 until that condition
was no longer true. The KS DFT implementation in MPQC employs GauXC 49 (which calls
LibXC 50 ) for calculation of the exchange-correlation potentials and energies. The integra-

4
tion grid for evaluation of these quantities for the PDB systems was the “superfine” grid
(250 radial Mura-Knowles 51 points, 974 angular Lebedev-Laikov 52 for all atoms except hy-
drogen, which has 175 radial points.). Density functionals used include LDA (SVWN5), 53,54
BLYP, 55 PBE, 56 B3LYP, 57 PBE0, 58,59 for the PDB systems and additionally for the peptide
pair screening in Section 3.2.2: revPBE, 60 MN15, 61 TPSS, 62 SCAN, 63 revSCAN, 64 CAM-
B3LYP, 65 ωB97, 66 ωPBE. 67,68 To match the calculations performed in ref. 21 we used the
6-31G** basis. 69–73 Density fitting was performed with the def2-universal-J basis. 74 Geome-
tries for all systems (except the individual amino acids used in Section 3.2) were obtained
from the Protein Data Bank (PDB). 75
The electron densities of the converged KS DFT solutions were analyzed by comparing
them with the corresponding HF electron densities. This was done by finding the eigenvalues
and eigenvectors of the difference of the density matrices (in orthonormal basis).

γ diff = γ HF − γ KS (1)

These eigenvectors were then transformed back to the AO basis; these are the HF-KS natural
deformation orbitals (NDOs). The eigenvalues associated with each NDO will be referred to
as natural deformation charges qnd . Plotting NDOs with negative qnd identifies the regions
where KS has gained density (relative to HF), and vice versa. Although the use of a post-HF
reference density would be preferred to include correlation effects (e.g., MP2), for the cases of
qualitative failure of KS DFT the HF-KS and MP2-KS natural deformation orbitals should
be qualitatively similar.
Amino acid calculations in Section 3.2.2 used Psi4 76 for both geometry optimization
(B3LYP/6-31G* 69–73 ) and single point KS DFT energy evaluation (along with eigenspec-
trum), using default parameters, including density fitting. The side chains of the amino
acids were uncharged in all cases, except for arginine, which had a +1 charge.

5
3 Results

3.1 Small polypeptides

Our analysis starts with the set of 17 small polypeptides used by Rudberg to illustrate SCF
convergence failures. 21 For 12 of these systems the standard diagonalization-based Roothaan-
Hall (RH) SCF solver could not locate a solution for at least one functional; the chemical
structures for these 12 “difficult” systems are presented in the SI. Using our QUOTR solver
we managed to obtain converged SCF solution for all system/DFA combinations considered
by Rudberg (BHandHLYP was excluded since it posed no convergence issues). For all RH-
converged SCF solutions in Ref. 21 QUOTR confirmed the HOMO-LUMO gaps to within
0.04 eV (the largest deviation was observed for the 1XT7 system with the LDA functional).
For the cases where the RH solver could not locate a solution the QUOTR solver located
solutions with vanishing HOMO-LUMO gap. These vanishing-gap QUOTR solutions were
then analyzed by the HF-KS deformation density analysis (DDA) described in Section 2.

3.1.1 HF-LDA deformation density analysis summary

Although Hartree-Fock is known to artificially localize charges, 40 it provides qualitatively

correct electron distributions for all these systems; specifically, the HF charges agree with
the expected integral values of formal charges on charged functional groups like -CO−
2 . In all

cases HF predicts large positive HOMO-LUMO gaps. In contrast, KS DFT can “delocalize”
charges across large distances, or alternatively it can lead to fractional charges on charged
functional group; such states have vanishing HOMO-LUMO gap. The HF-KS DDA reveals
where the fractional charges are located in the KS solution. The summary of HF-LDA DDA
performed for the 12 systems with vanishing LDA gap (Table 2) reports which and how
many functional groups donate (Nd ) and accept (Na ) electron density in LDA relative to
HF. Usually, but not always, the donor and acceptor sites correspond to the location of
the Hartree-Fock HOMO and LUMO, or other states near the Fermi level. See the SI for

6
 HF PBE0 B3LYP BLYP PBE LDA
PDB ID RH QUOTR RH QUOTR RH QUOTR RH QUOTR RH QUOTR RH QUOTR
2P7R 12.03 12.04 4.65 4.67 4.16 4.18 2.12 2.14 2.10 2.11 2.12 2.14
1BFZ 11.96 11.96 6.18 6.18 5.77 5.77 3.97 3.98 3.93 3.93 3.79 3.79
2IGZ 11.81 11.81 5.70 5.70 5.27 5.27 3.27 3.28 3.23 3.24 3.12 3.13
1D1E 10.14 10.12 3.96 3.96 3.47 3.47 1.56 1.57 1.58 1.59 1.54 1.55
1SP7 9.13 9.11 1.65 1.64 0.87 0.87 — < 0.01 — < 0.01 — < 0.01
1N9U 9.12 9.12 1.12 1.14 0.57 0.59 — < 0.01 — < 0.01 — < 0.01
1MZI 8.77 8.74 1.29 1.29 0.54 0.54 — < 0.01 — < 0.01 — < 0.01
1XT7 8.51 8.48 3.32 3.29 2.65 2.63 1.02 1.00 1.24 1.23 1.30 1.34
1PLW 7.25 7.24 0.36 0.36 0.29 0.28 — < 0.01 — < 0.01 — 0.01

7
1FUL 6.95 6.93 0.20 0.20 0.16 0.16 — −0.01 — −0.01 — −0.01
1EDW 6.89 6.90 0.26 0.26 0.21 0.21 — < 0.01 — < 0.01 — < 0.01
1EVC 5.82 5.84 0.30 0.30 0.24 0.24 — < 0.01 — < 0.01 — < 0.01
1RVS 5.60 5.59 — 0.10 — 0.08 — < 0.01 — < 0.01 — < 0.01
2FR9 5.48 5.50 0.26 0.26 0.21 0.21 — < 0.01 — < 0.01 — < 0.01
2JSI 5.26 5.27 0.24 0.24 0.19 0.19 — < 0.01 — < 0.01 — < 0.01
1LVZ 5.05 5.03 0.31 0.31 0.25 0.25 — < 0.01 — < 0.01 — < 0.01
1FDF 3.64 3.62 0.13 0.14 0.11 0.11 — < 0.01 — < 0.01 — < 0.01

Table 1: HOMO-LUMO gaps (eV) for 17 small polypeptides from Ref. 21. The standard (RH) SCF solver values are from 21,
the quasi-Newton (QUOTR) SCF solver 46 values are from this work. For all cases where the standard solver failed to converge
to a solution, the quasi-Newton solver located a solution has a vanishing HOMO-LUMO gap.
images comparing the Hartree-Fock HOMO and LUMO for each system with the natural
deformation orbitals used in our analysis. In all cases shown in Table 2, the electron donors
are formally negatively charged moieties (mostly CO−
2 ). The electron acceptors are often

positively charged (mostly NH+

3 ), however, a few cases involve neutral functional groups as

acceptors, most notably 2FR9, 1EDW, 1EVC, and 2JSI. Sometimes the natural deformation
orbitals are not so clear to interpret. The column labeled “backbone?” is used to indicate
when there are some density change contributions that are hard to classify which involve
other parts of the system. In most cases, these contributions include carbonyl groups in the
peptide bonds.
Table 2 also characterizes the HF-LDA NDOs with nonnegligible qnd , such orbitals will
be referred to as frontier NDOs (FNDOs). For all systems studied, the smallest qnd was
simply the negative of the largest one, thus we only display the largest |qnd | for each system.
In many cases there is a single pair of FNDOs, which typically corresponds to the case
of electrons transferred between single functional groups. But in some systems there are
multiple FNDOs (e.g. 1FUL); in such cases we only display the largest |qnd |. We only record
donor/acceptor groups in Table 2 contributing to the FNDOs with |qnd | ≥ 0.2. Only three
systems had multiple |qnd | over this threshold: 1FUL, 2F9R, and 1FDF. Images of all natural
deformation orbitals used in the analysis are provided in the SI.
We next examine three different cases that provide insight into the possible scenarios.

3.1.2 Case 1FUL: 2 donors, 2 acceptors

First, we consider a zwitterion that exemplifies the classic case of charge separation, where
the delocalization error is known to be a problem. 27,44 In Figure 1 we display the natural
deformation orbitals with the largest deformation charges, qnd = ±0.411. In this case there
are actually at least two regions that appear to be accepting electrons: the NH+
3 group and

the protonated guanidine group of arginine. Also, there are two deprotonated carboxylic acid
groups that donate electrons. Although this case is very similar to the classic zwitterions,

8
Table 2: HF-LDA natural deformation density analysisa for the 12 systems with SCF con-
vergence difficulties from Ref. 21, indicating donor group types and acceptor group types.b

PDB ID |qnd | Nd donor types Na acceptor types backbone?

1SP7 0.211 1 −CO− 2 1 −NH+3 minimal
1N9U 0.282 2 −CO− 2 1 −NHC(NH2 )+ 2 yes
1MZI 0.284 2 −CO− 2 1 −NH+3 minimal
1PLW 0.379 1 −CO− 2 1 −NH+3 some
1FUL 0.411 2 −CO− 2 2 −NH+ 3 ,−NHC(NH +
2 )2 some;S2
1EDW 0.414 2 −CO− 2 1 phenyl near −NH+ 3 some
1EVC 0.447 1 −CO− 1 phenyl yes
1RVS 0.439 1 −CO− 2 1 −NH+ 3 and phenyl minimal
2FR9 0.566 2 −CO− ,−NH− 2 phenyl,amide yes;S2
2JSI 0.468 1 −NH− 1 phenol yes
1LVZ 0.687 3 −CO− 2 2 −NH3 ,−NHC(NH2 )+
+
2 some
1FDF 0.564 2 −CO− ,−PRO− 3 −NH+ +
3 ,−NHC(NH2 )2 ,amide yes
a
Nd is “number of donors”; Na is “number of acceptors”. See text for further details.
b
−NHC(NH2 )+2 is protinated guanidine; −PRO is proline.

+0.411 -0.411
Figure 1: The HF-LDA NDOs of 1FUL with largest-magnitude deformation charges. This
case illustrates charge delocalization over multiple donor and multiple acceptor functional
groups.

9
the analysis is made more complicated because there is density transfer occurring with the
peptide backbone and at least one of the disulfide bridges.

3.1.3 Case 1EVC: neutral acceptor

The problems with zwitterion convergence using semilocal DFAs are relatively well known.
Here we show that it is possible to have the same problem without an explicitly positive
group receiving the electron density. In Figure 2 the natural deformation orbitals of largest
magnitude are displayed for 1EVC.

+0.447 -0.447

Figure 2: The HF-LDA natural deformation orbitals of 1EVC with largest-magnitude defor-
mation charges.

10
 Closer examination of the structure reveals that electron density is being donated from
a CO− (not CO−
2 ) group! Clearly this system has an invalid structure generated to fit the

experimental NMR data. For such unphysical structures it is possible to have an extremely
unstable anion whose density will be transferred to a neutral group. However, such exceptions
are rare since this is one of the few examples of anion-to-neutral charge transfer in the current
test set.

3.1.4 Case 1RVS: non-Aufbau solution

Although the near-zero gap displayed for 1FUL in Section 3.1.2 is typical for zwitterions in
vacuum, we have found that for a different zwitterion our local SCF solver is able to converge
to the “physically correct” charge-separated solution. In Figure 3 we display the HOMO and
LUMO for 1RVS when the QUOTR solver is given the usual extended-Hückel-like initial
guess for the orbitals, but without perturbation. 46 We see that there is no unphysical mixing
of the occupied orbital on the CO− +
2 with the unoccupied orbital on the NH3 . For this solution

the HOMO has an energy of -0.037 eV while the LUMO has an energy of -0.130 eV, giving a
nonzero negative HOMO-LUMO gap. Thus, this is a non-Aufbau state! Now, this solution
is not the best solution in a variational sense, because the total energy could be lowered by
mixing the HOMO with the LUMO (note Janak’s theorem 77 ). However, it does avoid the
major contribution to the incorrect charge delocalization. We conclude that we have been
able to find a local stationary point for 1RVS (a zwitterion system) using B3LYP that has
qualitatively the physically correct HOMO and LUMO, but such solution is not the lowest
energy state. When the initial guess is perturbed, QUOTR is able to converge to the nearly
zero-gap solution. The energy of this solution is 0.012 Eh below the physically-reasonable
non-Aufbau solution.

11
Figure 3: For the 1RVS zwitterion it was possible to locate a non-Aufbau B3LYP SCF
soluton that does not suffer from charge delocalization and qualitatively matches the charge
distribution of the exact ground state. The non-Aufbau HOMO (bottom) and LUMO (top)
are localized on the CO− +
2 and NH3 termini, as expected.

3.2 Amino Acids

The simplest analogous situation where the HOMO could be above the LUMO is in a system
composed of two amino acids (in vacuum) separated by a large distance so that they are
essentially non-interacting. This simple model is explored in Section 3.2.1 where we show
that the non-Aufbau solution is obtainable using the QUOTR SCF solver. By perturbing
the guess orbitals (importantly mixing orbitals on the separated fragments), we can also
obtain the near zero-gap solution. The underlying principle thus indicates that this will
occur whenever the HOMO on one fragment is above the LUMO of the other fragment.
Therefore, in Section 3.2.2 we calculate HOMO and LUMO energies for a variety of popular
functionals, to see which combinations are possibly problematic.

3.2.1 Separated glycine zwitterion

We demonstrate the utility of QUOTR by finding a solution for a system that is impossible
for a diagonalization-based solver: a non-Aufbau filled system. For this analysis we use KS-

12
DFT with the local density approximation (LDA) to the exchange-correlation functional.
The HOMO and LUMO for glycine in two protonation states (labled gly− and gly+ , see
Figure 4) were calculated separately. These could represent the two ends of a peptide chain
in zwitterion form. Then a supersystem was constructed with these two fragments together,
separated by approximately 200 angstrom. Thus, the two fragments should be physically
isolated from each other.

Figure 4: Glycine in protonated (overall positive charge, gly+ ) and deprotonated (overall
negative charge, gly− ) states.

From the data in Table 3 we see that a non-Aufbau solution is expected: although each
charged fragment has a sizable positive HOMO-LUMO gap, the HOMO of the anion is much
higher than the LUMO of the cation. When these two fragments are calculated together,
the initial guess for the MOs is of utmost importance for any direct minimization solver. In
the column labeled “unperturbed” the initial guess is our standard extended-Hückel guess.
Due to the large separation of about 200 angstroms, there is no overlap between the AOs of
the different fragments. Thus, the guess MOs are disjoint: each orbital is either associated
with gly− or gly+ . The gradient for mixing the MOs on different fragments should therefore
be zero. Minimization should then result in nearly the same orbitals on each fragment as
in the isolated calculations, implying a solution with the HOMO higher than the LUMO!
In contrast, the column labeled “perturbed” takes the extended-Hückel guess and applies a
small, random unitary matrix that allows mixing between all orbitals, regardless of which
fragment they belong to. Thus, there are likely some orbitals in the initial guess that span

13
both fragments. This is physically incorrect, but LDA is actually able to find a lower total
energy for the supersystem with some MOs spanning both fragments. The HOMO is then
lowered and the LUMO raised, until the gap is essentially zero.

Table 3: Frontier orbital energetics (eV) of LDA/6-31G** SCF solutions for isolated depro-
tonated/protonated glycine molecules and their noninteracting supersystem. Quasi-Newton
solver locates a non-Aufbau supersystem solution, unless the initial (atomic density-based)
guess is perturbed to mix the orbitals of the monomers. Perturbed guess leads to the solution
with a vanishing HOMO-LUMO gap, and an unphysical charge delocalization.

gly− gly+ gly− . . . gly+

unperturbed perturbed
LUMO: 5.72 -6.79 -6.72 -3.31
HOMO: 0.64 -12.59 0.57 -3.29
HOMO-LUMO gap: 5.08 5.80 -7.29 -0.02

This striking example is due to the self-interaction error in the LDA functional, and it
demonstrates LDA’s tendency to prefer fractional electron systems. In this case the fractional
electron on one of the fragments is caused by having an MO with substantial density on the
other fragment too.

3.2.2 Screening natural amino acid pairs

To systematically investigate when the problem of vanishing HOMO-LUMO gap is likely

to appear, we do not need to run the same calculations as in the previous section for each
pair of amino acids. The principle is clear: when an isolated fragment has a HOMO above
the LUMO of a separate fragment, a non-Aufbau filled solution is possible. The idea that
having the LUMO on one fragment above the HOMO on another other fragment can cause
excessive electron transfer and binding strength has been examined before in the context
of radical-molecules complexes, 78 and charge transfer complexes. 79 We performed a series
of tests on different functionals for all 20 naturally occurring amino acids in three different
protonation states: neutral, anionic, and cationic (except for Arginine, where the charges
are shifted to 0, +1, and +2). Table 4 displays the number of predicted non-Aufbau pairs

14
of amino acids, grouped by charge state, for a representative set of local, gradient-corrected
(GGA), meta-GGA, and hybrid DFAs (both standard and range-separated).
As expected, including larger fractions of Hartree-Fock exchange reduces the number of
problematic combinations. The performance of the semilocal functionals is almost unchanged
going from LDA to GGA. The meta-GGAs provide some improvement, but not as much
improvement as hybrid functionals. Finally, the range-separated functionals perform best of
all, however, there are some cases where they still predict non-Aufbau filled systems. The
most difficult systems are when the donor is negatively charged and the acceptor has a +2
charge (Arginine), where even ωB97 fails.

Figure 5: Overall number of non-Aufbau states predicted by various KS functionals for

noninteracting pairs of the 20 natural amino acids in their neutral/protonated/deprotonated
states. See Table 4 for the further breakdown by the aminoacid charges.

15
Table 4: Number of non-Aufbau states predicted by various KS functionals for noninteracting
pairs of the 20 natural amino acids in their neutral/protonated/deprotonated states. The
data is broken down by charges Qd /Qa of the “donor”/“acceptor” fragments, respectively
(i.e., containing the HOMO and LUMO, respectively).

Semilocal
LDA Qa BLYP Qa revPBE Qa
-1 0 +1 +2 -1 0 +1 +2 -1 0 +1 +2
-1 0 379 380 19 -1 0 380 380 19 -1 0 380 380 19
0 0 0 306 20 0 0 0 286 20 0 0 0 273 20
Qd Qd Qd
+1 0 0 0 3 +1 0 0 0 4 +1 0 0 0 2
+2 0 0 0 0 +2 0 0 0 0 +2 0 0 0 0

meta-GGA
TPSS Qa SCAN Qa revSCAN Qa
-1 0 +1 +2 -1 0 +1 +2 -1 0 +1 +2
-1 0 368 380 19 -1 0 275 380 19 -1 0 265 380 19
0 0 0 176 20 0 0 0 74 20 0 0 0 63 20
Qd Qd Qd
+1 0 0 0 1 +1 0 0 0 0 +1 0 0 0 0
+2 0 0 0 0 +2 0 0 0 0 +2 0 0 0 0

Hybrid
a
MN15 Qa B3LYP Qa PBE0 Qa
-1 0 +1 +2 -1 0 +1 +2 -1 0 +1 +2
-1 0 0 380 19 -1 0 0 380 19 -1 0 0 380 19
0 0 0 0 1 0 0 0 0 20 0 0 0 0 10
Qd Qd Qd
+1 0 0 0 0 +1 0 0 0 0 +1 0 0 0 0
+2 0 0 0 0 +2 0 0 0 0 +2 0 0 0 0

Range-separated
CAM-B3LYP Qa ωB97 Qa ωPBE Qa
-1 0 +1 +2 -1 0 +1 +2 -1 0 +1 +2
-1 0 0 374 19 -1 0 0 0 17 -1 0 0 0 19
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Qd Qd Qd
+1 0 0 0 0 +1 0 0 0 0 +1 0 0 0 0
+2 0 0 0 0 +2 0 0 0 0 +2 0 0 0 0

a
MN15 is both a hybrid functional and a meta functional, but we list it under “Hybrid”.

16
4 Summary

This work reexamined the SCF convergence problems for polypeptides in gas phase in con-
junction with modern non-hybrid and hybrid DFAs. While standard SCF solvers typically
fail catastrophically, 21,27 using a robust quasi-Newton SCF solver 46 we were able to obtain
SCF solutions when the conventional solvers fail; in such cases the KS solutions always had
a vanishing HOMO-LUMO gap. Deeper analysis of these solutions using novel natural de-
formation orbitals obtained from the HF-KS density matrix difference reveals which regions
of the system donate and accept electron density in the unphysical KS DFT solution, and
are thus the culprits in the delocalization error. In general the unphysical charge delocal-
ization can involve not only charged moieties but also formally neutral fragments (e.g. a
phenyl ring), demonstrating that zwitterions are not the only problematic cases for semilo-
cal DFAs. The origin of the unphysical charge delocalization is the misalignment of the
KS Fock operator eigenspectrum between molecular fragments. A systematic study of pairs
of 20 naturally occurring amino acids in various protonation states suggested that the un-
physical charge delocalization is only partially reduced by the use of a more sophisticated
functional. The range-separated functionals employing 100% exact exchange at long range
were a nearly perfect remedy, albeit not totally immune from the problem. The rest of the
representative functional families (standard hybrid, meta-GGA, GGA) all suffered from the
unphysical charge delocalization to various extents.
The most direct lesson from our work is the need for caution when unexpectedly small
(or vanishing) HOMO-LUMO gaps and atypical SCF convergence patterns (e.g., oscillatory)
are observed in KS DFT simulations in any context (bio or otherwise). Such anomalies
should call for probing the solution (e.g., population analysis) and trying more advanced KS
functionals. Although our work focused on a specific, and somewhat artificial, biosimula-
tion context, namely isolated polypeptides, such systems continue to serve as components
and even as the sole focus of benchmark datasets 80 for training more approximate models.
There are also lessons here for the broader class of Kohn-Sham DFT simulations. As afford-

17
able source of first-principles potentials for training more approximate models, KS DFT is
increasingly used to generate massive datasets for benchmarking and training purposes, 81
with scale too large for validatation of even a nonnegligible portion of the dataset. So as the
role of KS DFT as the “ground truth” model rises so should our expectations of its accuracy
and robustness. Although we can expect that the continuing improvement of KS functionals
will reduce the occurrence of the artifacts discussed here, the need for robust solvers will
only rise as the degree of automation of KS DFT-based workflows continues to rise. The

Acknowledgements

This work was supported by the U.S. Department of Energy via award DE-SC0022327.
We also acknowledge Advanced Research Computing at Virginia Tech (www.arc.vt.edu) for
providing computational resources and technical support that have contributed to the results
reported within this paper.

References

(1) Tsuneda, T.; Song, J.-W.; Suzuki, S.; Hirao, K. On Koopmans’ Theorem in Density
Functional Theory. The Journal of Chemical Physics 2010, 133, 174101.

(2) Rudberg, E.; Rubensson, E. H.; Salek, P. Kohn-Sham Density Functional Theory Elec-
tronic Structure Calculations with Linearly Scaling Computational Time and Memory
Usage. J. Chem. Theory Comput. 2011, 7, 340–350.

(3) Tozer, D. J. Relationship between Long-Range Charge-Transfer Excitation Energy Er-

ror and Integer Discontinuity in Kohn–Sham Theory. J. Chem. Phys. 2003, 119, 12697–
12699.

(4) Dreuw, A.; Weisman, J. L.; Head-Gordon, M. Long-Range Charge-Transfer Excited

18
 States in Time-Dependent Density Functional Theory Require Non-Local Exchange. J.
Chem. Phys. 2003, 119, 2943–2946.

(5) Casida, M. E.; Jamorski, C.; Casida, K. C.; Salahub, D. R. Molecular Excitation Ener-
gies to High-Lying Bound States from Time-Dependent Density-Functional Response
Theory: Characterization and Correction of the Time-Dependent Local Density Ap-
proximation Ionization Threshold. J. Chem. Phys. 1998, 108, 4439–4449.

(6) Baerends, E. J.; Gritsenko, O. V.; Van Meer, R. The Kohn–Sham Gap, the Fundamental
Gap and the Optical Gap: The Physical Meaning of Occupied and Virtual Kohn–Sham
Orbital Energies. Phys. Chem. Chem. Phys. 2013, 15, 16408.

(7) Jensen, F. Describing Anions by Density Functional Theory: Fractional Electron Affin-
ity. J. Chem. Theory Comput. 2010, 6, 2726–2735.

(8) Peach, M. J. G.; Teale, A. M.; Helgaker, T.; Tozer, D. J. Fractional Electron Loss
in Approximate DFT and Hartree–Fock Theory. J. Chem. Theory Comput. 2015, 11,
5262–5268.

(9) Dobbs, K. D.; Dixon, D. A. Ab Initio Prediction of the Activation Energy for the
Abstraction of a Hydrogen Atom from Methane by Chlorine Atom. J. Phys. Chem.
1994, 98, 12584–12589.

(10) Johnson, B. G.; Gonzales, C. A.; Gill, P. M.; Pople, J. A. A Density Functional Study
of the Simplest Hydrogen Abstraction Reaction. Effect of Self-Interaction Correction.
Chemical Physics Letters 1994, 221, 100–108.

(11) Zhang, Q.; Bell, R.; Truong, T. N. Ab Initio and Density Functional Theory Studies of
Proton Transfer Reactions in Multiple Hydrogen Bond Systems. J. Phys. Chem. 1995,
99, 592–599.

19
(12) Vydrov, O. A.; Scuseria, G. E. A Simple Method to Selectively Scale down the Self-
Interaction Correction. J. Chem. Phys. 2006, 124, 191101.

(13) Janesko, B. G.; Scuseria, G. E. Hartree–Fock Orbitals Significantly Improve the Reac-
tion Barrier Heights Predicted by Semilocal Density Functionals. J. Chem. Phys. 2008,
128, 244112.

(14) Zhang, Y.; Yang, W. A Challenge for Density Functionals: Self-interaction Error In-
creases for Systems with a Noninteger Number of Electrons. J. Chem. Phys. 1998, 109,
2604–2608.

(15) Dutoi, A. D.; Head-Gordon, M. Self-Interaction Error of Local Density Functionals for
Alkali–Halide Dissociation. Chem. Phys. Lett. 2006, 422, 230–233.

(16) Otero-de-la-Roza, A.; Johnson, E. R. Analysis of Density-Functional Errors for Nonco-

valent Interactions between Charged Molecules. J. Phys. Chem. A 2020, 124, 353–361.

(17) Ruiz, E.; Salahub, D. R.; Vela, A. Charge-Transfer Complexes: Stringent Tests for
Widely Used Density Functionals. J. Phys. Chem. 1996, 100, 12265–12276.

(18) Isborn, C. M.; Mar, B. D.; Curchod, B. F. E.; Tavernelli, I.; Martı́nez, T. J. The Charge
Transfer Problem in Density Functional Theory Calculations of Aqueously Solvated
Molecules. J. Phys. Chem. B 2013, 117, 12189–12201.

(19) Otero-de-la-Roza, A.; Johnson, E. R.; DiLabio, G. A. Halogen Bonding from Dispersion-
Corrected Density-Functional Theory: The Role of Delocalization Error. J. Chem.
Theory Comput. 2014, 10, 5436–5447.

(20) Rubensson, E. H.; Rudberg, E. Bringing about Matrix Sparsity in Linear-scaling Elec-
tronic Structure Calculations. J Comput Chem 2011, 32, 1411–1423.

20
(21) Rudberg, E. Difficulties in Applying Pure Kohn–Sham Density Functional Theory Elec-
tronic Structure Methods to Protein Molecules. J. Phys.: Condens. Matter 2012, 24,
072202.

(22) Antony, J.; Grimme, S. Fully Ab Initio Protein-ligand Interaction Energies with Dis-
persion Corrected Density Functional Theory. J Comput Chem 2012, 33, 1730–1739.

(23) Kulik, H. J.; Luehr, N.; Ufimtsev, I. S.; Martinez, T. J. Ab Initio Quantum Chemistry
for Protein Structures. J. Phys. Chem. B 2012, 116, 12501–12509.

(24) Lever, G.; Cole, D. J.; Hine, N. D. M.; Haynes, P. D.; Payne, M. C. Electrostatic
Considerations Affecting the Calculated HOMO–LUMO Gap in Protein Molecules. J.
Phys.: Condens. Matter 2013, 25, 152101.

(25) Cole, D. J.; Hine, N. D. M. Applications of Large-Scale Density Functional Theory in

Biology. J. Phys.: Condens. Matter 2016, 28, 393001.

(26) Zuehlsdorff, T. J.; Haynes, P. D.; Hanke, F.; Payne, M. C.; Hine, N. D. M. Solvent Ef-
fects on Electronic Excitations of an Organic Chromophore. J. Chem. Theory Comput.
2016, 12, 1853–1861.

(27) Ren, F.; Liu, F. Impacts of Polarizable Continuum Models on the SCF Convergence
and DFT Delocalization Error of Large Molecules. J. Chem. Phys. 2022, 157, 184106.

(28) Sepunaru, L.; Refaely-Abramson, S.; Lovrinčić, R.; Gavrilov, Y.; Agrawal, P.; Levy, Y.;
Kronik, L.; Pecht, I.; Sheves, M.; Cahen, D. Electronic Transport via Homopeptides:
The Role of Side Chains and Secondary Structure. J. Am. Chem. Soc. 2015, 137,
9617–9626.

(29) Sharley, J. N. Amino Acid Preference against Beta Sheet through Allowing Backbone
Hydration Enabled by the Presence of Cation. 2016.

21
(30) Vydrov, O. A.; Scuseria, G. E. Assessment of a Long-Range Corrected Hybrid Func-
tional. J. Chem. Phys. 2006, 125, 234109.

(31) Li, J.-H.; Zuehlsdorff, T. J.; Payne, M. C.; Hine, N. D. M. Identifying and Tracing
Potential Energy Surfaces of Electronic Excitations with Specific Character via Their
Transition Origins: Application to Oxirane. Phys. Chem. Chem. Phys. 2015, 17, 12065–
12079.

(32) Shore, H. B.; Rose, J. H.; Zaremba, E. Failure of the Local Exchange Approximation
in the Evaluation of the H - Ground State. Phys. Rev. B 1977, 15, 2858–2861.

(33) K Schwarz, First Ionisation Potentials of Atoms Obtained with Local-Density Schemes.
J. Phys. B: Atom. Mol. Phys. 1978, 11, 1339–1351.

(34) Schwarz, K. Instability of Stable Negative Ions in the Xα Method or Other Local
Density Functional Schemes. Chemical Physics Letters 1978, 57, 605–607.

(35) Rösch, N.; Trickey, S. B. Comment on “Concerning the Applicability of Density Func-
tional Methods to Atomic and Molecular Negative Ions” [J. Chem. Phys. 105 , 862
(1996)]. J. Chem. Phys. 1997, 106, 8940–8941.

(36) Peach, M. J. G.; De Proft, F.; Tozer, D. J. Negative Electron Affinities from DFT:
Fluorination of Ethylene. J. Phys. Chem. Lett. 2010, 1, 2826–2831.

(37) Lee, D.; Furche, F.; Burke, K. Accuracy of Electron Affinities of Atoms in Approximate
Density Functional Theory. J. Phys. Chem. Lett. 2010, 1, 2124–2129.

(38) Kim, M.-C.; Sim, E.; Burke, K. Communication: Avoiding Unbound Anions in Density
Functional Calculations. J. Chem. Phys. 2011, 134, 171103.

(39) Mori-Sánchez, P.; Cohen, A. J.; Yang, W. Many-Electron Self-Interaction Error in

Approximate Density Functionals. J. Chem. Phys. 2006, 125, 201102.

22
(40) Mori-Sánchez, P.; Cohen, A. J.; Yang, W. Localization and Delocalization Errors in
Density Functional Theory and Implications for Band-Gap Prediction. Phys. Rev. Lett.
2008, 100, 146401.

(41) Cohen, A. J.; Mori-Sánchez, P.; Yang, W. Insights into Current Limitations of Density
Functional Theory. Science 2008, 321, 792–794.

(42) Perdew, J. P.; Parr, R. G.; Levy, M.; Balduz, J. L. Density-Functional Theory for
Fractional Particle Number: Derivative Discontinuities of the Energy. Phys. Rev. Lett.
1982, 49, 1691–1694.

(43) Hait, D.; Head-Gordon, M. Delocalization Errors in Density Functional Theory Are
Essentially Quadratic in Fractional Occupation Number. J. Phys. Chem. Lett. 2018,
9, 6280–6288.

(44) Jakobsen, S.; Kristensen, K.; Jensen, F. Electrostatic Potential of Insulin: Exploring the
Limitations of Density Functional Theory and Force Field Methods. J. Chem. Theory
Comput. 2013, 9, 3978–3985.

(45) Sim, E.; Song, S.; Vuckovic, S.; Burke, K. Improving Results by Improving Densities:
Density-Corrected Density Functional Theory. J. Am. Chem. Soc. 2022, 144, 6625–
6639.

(46) Slattery, S. A.; Surjuse, K. A.; Peterson, C. C.; Penchoff, D. A.; Valeev, E. F. Econom-
ical Quasi-Newton Unitary Optimization of Electronic Orbitals. Phys. Chem. Chem.
Phys. 2024, 26, 6557–6573.

(47) Pakiari, A. H.; Fakhraee, S.; Azami, S. M. Decomposition of Deformation Density into
Orbital Components. Int. J. Quantum Chem. 2008, 108, 415–422.

(48) Peng, C.; Lewis, C. A.; Wang, X.; Clement, M. C.; Pierce, K.; Rishi, V.; Pavošević, F.;
Slattery, S.; Zhang, J.; Teke, N.; Kumar, A.; Masteran, C.; Asadchev, A.; Calvin, J. A.;

23
 Valeev, E. F. Massively Parallel Quantum Chemistry: A High-Performance Research
Platform for Electronic Structure. J. Chem. Phys. 2020, 153, 044120.

(49) Petrone, A.; Williams-Young, D. B.; Sun, S.; Stetina, T. F.; Li, X. An Efficient Imple-
mentation of Two-Component Relativistic Density Functional Theory with Torque-Free
Auxiliary Variables. Eur. Phys. J. B 2018, 91, 169.

(50) Lehtola, S.; Steigemann, C.; Oliveira, M. J.; Marques, M. A. Recent Developments
in Libxc — A Comprehensive Library of Functionals for Density Functional Theory.
SoftwareX 2018, 7, 1–5.

(51) Mura, M. E.; Knowles, P. J. Improved Radial Grids for Quadrature in Molecular
Density-Functional Calculations. J. Chem. Phys. 1996, 104, 9848–9858.

(52) Lebedev, V. I.; Laikov, D. N. A Quadrature Formula for the Sphere of the 131st Alge-
braic Order of Accuracy. Dokl. Math. 1999, 59, 477–481.

(53) Dirac, P. A. M. Note on Exchange Phenomena in the Thomas Atom. Math. Proc. Camb.
Phil. Soc. 1930, 26, 376–385.

(54) Vosko, S. H.; Wilk, L.; Nusair, M. Accurate Spin-Dependent Electron Liquid Correla-
tion Energies for Local Spin Density Calculations: A Critical Analysis. Can. J. Phys.
1980, 58, 1200–1211.

(55) Miehlich, B.; Savin, A.; Stoll, H.; Preuss, H. Results Obtained with the Correlation
Energy Density Functionals of Becke and Lee, Yang and Parr. Chem. Phys. Lett. 1989,
157, 200–206.

(56) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made
Simple. Phys. Rev. Lett. 1996, 77, 3865–3868.

(57) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Ab Initio Calculation

24
 of Vibrational Absorption and Circular Dichroism Spectra Using Density Functional
Force Fields. J. Phys. Chem. 1994, 98, 11623–11627.

(58) Adamo, C.; Barone, V. Toward Reliable Density Functional Methods without Ad-
justable Parameters: The PBE0 Model. J. Chem. Phys. 1999, 110, 6158–6170.

(59) Ernzerhof, M.; Scuseria, G. E. Assessment of the Perdew–Burke–Ernzerhof Exchange-

Correlation Functional. J. Chem. Phys. 1999, 110, 5029–5036.

(60) Zhang, Y.; Yang, W. Comment on “Generalized Gradient Approximation Made Sim-
ple”. Phys. Rev. Lett. 1998, 80, 890–890.

(61) Yu, H. S.; He, X.; Li, S. L.; Truhlar, D. G. MN15: A Kohn–Sham Global-Hybrid
Exchange–Correlation Density Functional with Broad Accuracy for Multi-Reference
and Single-Reference Systems and Noncovalent Interactions. Chem. Sci. 2016, 7, 5032–
5051.

(62) Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E. Climbing the Density Func-
tional Ladder: Nonempirical Meta–Generalized Gradient Approximation Designed for
Molecules and Solids. Phys. Rev. Lett. 2003, 91, 146401.

(63) Sun, J.; Ruzsinszky, A.; Perdew, J. P. Strongly Constrained and Appropriately Normed
Semilocal Density Functional. Phys. Rev. Lett. 2015, 115, 036402.

(64) Mezei, P. D.; Csonka, G. I.; Kállay, M. Simple Modifications of the SCAN Meta-
Generalized Gradient Approximation Functional. J. Chem. Theory Comput. 2018, 14,
2469–2479.

(65) Yanai, T.; Tew, D. P.; Handy, N. C. A New Hybrid Exchange–Correlation Functional
Using the Coulomb-attenuating Method (CAM-B3LYP). Chem. Phys. Lett. 2004, 393,
51–57.

25
(66) Chai, J.-D.; Head-Gordon, M. Systematic Optimization of Long-Range Corrected Hy-
brid Density Functionals. J. Chem. Phys. 2008, 128, 084106.

(67) Henderson, T. M.; Janesko, B. G.; Scuseria, G. E. Generalized Gradient Approxima-

tion Model Exchange Holes for Range-Separated Hybrids. J. Chem. Phys. 2008, 128,
194105.

(68) Weintraub, E.; Henderson, T. M.; Scuseria, G. E. Long-Range-Corrected Hybrids Based

on a New Model Exchange Hole. J. Chem. Theory Comput. 2009, 5, 754–762.

(69) Ditchfield, R.; Hehre, W. J.; Pople, J. A. Self-Consistent Molecular-Orbital Meth-

ods. IX. An Extended Gaussian-Type Basis for Molecular-Orbital Studies of Organic
Molecules. J. Chem. Phys. 1971, 54, 724–728.

(70) Hehre, W. J.; Ditchfield, R.; Pople, J. A. Self—Consistent Molecular Orbital Methods.
XII. Further Extensions of Gaussian—Type Basis Sets for Use in Molecular Orbital
Studies of Organic Molecules. J. Chem. Phys. 1972, 56, 2257–2261.

(71) Hariharan, P. C.; Pople, J. A. The Influence of Polarization Functions on Molecular

Orbital Hydrogenation Energies. Theoret. Chim. Acta 1973, 28, 213–222.

(72) Francl, M. M.; Pietro, W. J.; Hehre, W. J.; Binkley, J. S.; Gordon, M. S.; DeFrees, D. J.;
Pople, J. A. Self-consistent Molecular Orbital Methods. XXIII. A Polarization-type
Basis Set for Second-row Elements. J. Chem. Phys. 1982, 77, 3654–3665.

(73) Gordon, M. S.; Binkley, J. S.; Pople, J. A.; Pietro, W. J.; Hehre, W. J. Self-Consistent
Molecular-Orbital Methods. 22. Small Split-Valence Basis Sets for Second-Row Ele-
ments. J. Am. Chem. Soc. 1982, 104, 2797–2803.

(74) Weigend, F. Accurate Coulomb-fitting Basis Sets for H to Rn. Phys. Chem. Chem.
Phys. 2006, 8, 1057.

(75) Berman, H. M. The Protein Data Bank. Nucleic Acids Res. 2000, 28, 235–242.

26
(76) Smith, D. G. A.; Burns, L. A.; Simmonett, A. C.; Parrish, R. M.; Schieber, M. C.;
Galvelis, R.; Kraus, P.; Kruse, H.; Di Remigio, R.; Alenaizan, A.; James, A. M.;
Lehtola, S.; Misiewicz, J. P.; Scheurer, M.; Shaw, R. A.; Schriber, J. B.; Xie, Y.;
Glick, Z. L.; Sirianni, D. A.; O’Brien, J. S.; Waldrop, J. M.; Kumar, A.; Hohen-
stein, E. G.; Pritchard, B. P.; Brooks, B. R.; SchaeferIII, H. F.; Sokolov, A. Y.;
Patkowski, K.; DePrinceIII, A. E.; Bozkaya, U.; King, R. A.; Evangelista, F. A.;
Turney, J. M.; Crawford, T. D.; Sherrill, C. D. PSI4 1.4: Open-source Software for
High-Throughput Quantum Chemistry. J. Chem. Phys. 2020, 152, 184108.

(77) Janak, J. F. Proof That ∂ E ∂ n i = ε in Density-Functional Theory. Phys. Rev. B

1978, 18, 7165–7168.

(78) Johnson, E. R.; Salamone, M.; Bietti, M.; DiLabio, G. A. Modeling Noncovalent
Radical–Molecule Interactions Using Conventional Density-Functional Theory: Beware
Erroneous Charge Transfer. J. Phys. Chem. A 2013, 117, 947–952.

(79) Sini, G.; Sears, J. S.; Brédas, J.-L. Evaluating the Performance of DFT Functionals in
Assessing the Interaction Energy and Ground-State Charge Transfer of Donor/Acceptor
Complexes: Tetrathiafulvalene-Tetracyanoquinodimethane (TTF-TCNQ) as a Model
Case. J. Chem. Theory Comput. 2011, 7, 602–609.

(80) Prasad, V. K.; Otero-de-la-Roza, A.; DiLabio, G. A. PEPCONF, a Diverse Data Set
of Peptide Conformational Energies. Scientific Data 2019, 6, 180310.

(81) Culka, M.; Kalvoda, T.; Gutten, O.; Rulı́šek, L. Mapping Conformational Space of All
8000 Tripeptides by Quantum Chemical Methods: What Strain Is Affordable within
Folded Protein Chains? J. Phys. Chem. B 2021, 125, 58–69, PMID: 33393778.

27
 Supporting Information for “Revisiting Artifacts
of Kohn-Sham Density Functionals for
arXiv:2406.01520v2 [physics.chem-ph] 11 Jul 2024

Biosimulation”

Samuel A. Slattery, Jaden C. Yon, and Edward F. Valeev∗

Department of Chemistry, Virginia Tech, Blacksburg, VA 24061

E-mail: [email protected]

1 Structures of Systems from the PDB

Chemical structures of the 12 systems investigated in section 3.1 are presented here.

2 Natural Deformation Orbitals: Systems from the

PDB using LDA

Images of the HF-LDA natural deformation orbitals with magnitudes of deformation charges
greater than 0.2 (“Frontier Natural Deformation Orbitals”, or FNDO) are presented in this
section, along with the Hartree-Fock HOMO and LUMO.

1
Figure 1: 1SP7

2
Figure 2: 1N9U

3
Figure 3: 1MZI

4
Figure 4: 1PLW

5
Figure 5: 1FUL

6
Figure 6: 1EDW

7
Figure 7: 1EVC

8
Figure 8: 1RVS

9
Figure 9: 2FR9

10
Figure 10: 2JSI

11
Figure 11: 1LVZ

12
Figure 12: 1FDF

13
 NDO (+0.221) NDO (-0.221)

HF HOMO HF LUMO

Figure 13: 1SP7: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

14
 NDO (+0.282) NDO (-0.282)

HF HOMO HF LUMO

Figure 14: 1N9U: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

15
 NDO (+0.284) NDO (-0.284)

HF HOMO HF LUMO

Figure 15: 1MZI: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

16
 NDO (+0.379) NDO (-0.379)

HF HOMO HF LUMO

Figure 16: 1PLW: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

17
 NDO (+0.411) NDO (-0.411)

HF HOMO HF LUMO

Figure 17: 1FUL: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

18
 NDO (+0.247) NDO (-0.247)

HF HOMO HF LUMO

Figure 18: 1FUL: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

19
 NDO (+0.414) NDO (-0.414)

HF HOMO HF LUMO

Figure 19: 1EDW: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

20
 NDO (+0.447) NDO (-0.447)

HF HOMO HF LUMO

Figure 20: 1EVC: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

21
 NDO (+0.439) NDO (-0.439)

HF HOMO HF LUMO

Figure 21: 1RVS: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

22
 NDO (+0.566) NDO (-0.566)

HF HOMO HF LUMO

Figure 22: 2FR9: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

23
 NDO (+0.384) NDO (-0.384)

HF HOMO HF LUMO

Figure 23: 2FR9: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

24
 NDO (+0.468) NDO (-0.468)

HF HOMO HF LUMO

Figure 24: 2JSI: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

25
 NDO (+0.687) NDO (-0.687)

HF HOMO HF LUMO

Figure 25: 1LVZ: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

26
 NDO (+0.564) NDO (-0.564)

HF HOMO HF LUMO

Figure 26: 1FDF: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

27
 NDO (+0.476) NDO (-0.476)

HF HOMO HF LUMO

Figure 27: 1FDF: HF-LDA FNDOs, juxtaposed with the Hartree-Fock HOMO and LUMO.

28