Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Photoelectron spectra of copper oxide

cluster anions from first principles methods


Cite as: J. Chem. Phys. 149, 064306 (2018); https://doi.org/10.1063/1.5038744
Submitted: 04 May 2018 • Accepted: 24 July 2018 • Published Online: 13 August 2018

Bin Shi, Shira Weissman, Fabien Bruneval, et al.

ARTICLES YOU MAY BE INTERESTED IN

A consistent and accurate ab initio parametrization of density functional dispersion


correction (DFT-D) for the 94 elements H-Pu
The Journal of Chemical Physics 132, 154104 (2010); https://doi.org/10.1063/1.3382344

Practical GW scheme for electronic structure of 3d-transition-metal monoxide anions:


, , , and
The Journal of Chemical Physics 151, 134305 (2019); https://doi.org/10.1063/1.5118671

Toward reliable density functional methods without adjustable parameters: The PBE0
model
The Journal of Chemical Physics 110, 6158 (1999); https://doi.org/10.1063/1.478522

J. Chem. Phys. 149, 064306 (2018); https://doi.org/10.1063/1.5038744 149, 064306

© 2018 Author(s).
THE JOURNAL OF CHEMICAL PHYSICS 149, 064306 (2018)

Photoelectron spectra of copper oxide cluster anions from first


principles methods
Bin Shi,1 Shira Weissman,2 Fabien Bruneval,3 Leeor Kronik,2 and Serdar Öğüt1,a)
1 Department of Physics, University of Illinois at Chicago, Chicago, Illinois 60607, USA
2 Department of Materials and Interfaces, Weizmann Institute of Science, Rehovoth 76100, Israel
3 DEN, Service de Recherches de Métallurgie Physique, CEA, Université Paris-Saclay,

F-91128 Gif-sur-Yvette, France


(Received 4 May 2018; accepted 24 July 2018; published online 13 August 2018)

We present results and analyses for the photoelectron spectra of small copper oxide cluster anions
(CuO− , CuO−2 , CuO−3 , and Cu2 O− ). The spectra are computed using various techniques, including
density functional theory (DFT) with semi-local, global hybrid, and optimally tuned range-separated
hybrid functionals, as well as many-body perturbation theory within the GW approximation based
on various DFT starting points. The results are compared with each other and with the available
experimental data. We conclude that as in many metal-organic systems, self-interaction errors are a
major issue that is mitigated by hybrid functionals. However, these need to be balanced against a strong
role of non-dynamical correlation—especially in smaller, more symmetric systems—where errors
are alleviated by semi-local functionals. The relative importance of the two phenomena, including
practical ways of balancing the two constraints, is discussed in detail. Published by AIP Publishing.
https://doi.org/10.1063/1.5038744

I. INTRODUCTION resulting from the subtle interplay of their orbital, spin, and
charge degrees of freedom.13 Motivated by these observations,
Gas-phase photoelectron spectroscopy (PES) and its first-
here we examine the performance of various state-of-the-art
principles simulation using various computational approaches
DFT and GW methods in modeling the photoelectron spectra
have proven to be a powerful combination for probing the
of copper oxide anion clusters, by comparing their predic-
electronic structure of atomic and molecular clusters.1,2 In
tions with each other and with the available experimental
gas-phase PES, electrons are emitted upon absorption of
data.
(typically ultraviolet) light, allowing for the exploration of
We focus on four small clusters, Cu2 O− , CuO− , CuO−2 ,
electron removal energies. Computational methods based on
and CuO−3 , which span a relatively wide range of Cu con-
density functional theory (DFT) within the (generalized)
tent and exhibit strong electronic correlations. Photoelectron
Kohn-Sham scheme2–4 or many-body perturbation theory
spectra for these clusters have been available up to pho-
within the GW approximation4–9 are some of the most com-
ton energies of ∼5-6 eV since the pioneering studies of
monly used approaches to determine the energetics of such
Polak et al.14 and Wang et al.15,16 Among them, copper
charged excitations. The degree of accuracy with which var-
monoxide has been extensively studied both experimen-
ious computational methods can predict these excitations
tally14,16,17 and theoretically.10,18–31 Electronic and optical
depends not only on the fundamental limitations and practi-
properties of CuO2 in various charge states have also been
cal implementations of the underlying theoretical framework
of significant interest, especially with regards to its lowest
but also on the nature (e.g., chemical composition, size) of
energy structure (bent versus linear) in various experimen-
the system being studied. From this perspective, transition
tal16,32–36 and theoretical studies.37–43 There have been fewer
metal oxide clusters are stringent test cases for state-of-the-art
investigations on the electronic structures of CuO3 16,44,45 and
computational methods, owing to enhanced electron corre-
Cu2 O.10,15,46,47 The first-principles methods explored here
lations inherent in them, as well as their open-shell charac-
include DFT with semilocal and global hybrid exchange-
ter,10 which require a balanced and accurate description of
correlation functionals (based on shifted eigenvalue spectra),
dynamical and non-dynamical electron correlation. In addi-
DFT with optimally tuned range-separated hybrid (OT-RSH)
tion to the computational challenges in modeling their excited
functionals, and many-body perturbation theory techniques
states,5,11,12 bulk and nanostructured transition metal oxides
within the GW approximation using semilocal or global hybrid
have traditionally been systems of great technological and sci-
functional starting points.
entific interest, as they possess a wide range of complex, and
This article is arranged as follows. We begin with an
often desirable, structural, electronic, and optical properties
overview of the computational methods and their correspond-
ing parameters, given in Sec. II. This is followed in Sec. III
by a brief discussion on the relaxed structures of the copper
a)Electronic mail: ogut@uic.edu oxide clusters considered in this study and detailed analyses

0021-9606/2018/149(6)/064306/12/$30.00 149, 064306-1 Published by AIP Publishing.


064306-2 Shi et al. J. Chem. Phys. 149, 064306 (2018)

of the computed photoelectron spectra. Trends in the pho- polarizability computed within the random-phase approxi-
toelectron spectra of these clusters are analyzed in terms of mation, are used to compute the self-energy Σ = iG0 W 0 .
various factors, including self-interaction errors (SIEs), spa- The starting electronic structures were obtained from DFT
tial extent of the relevant orbitals, amount of exact exchange calculations with PBE and various hybrid functionals per-
in the DFT description, importance of non-dynamical corre- formed within the same package. Quasiparticle wavefunc-
lation, spin-splitting, and starting-point dependence in GW tions were represented using atom-centered Gaussian basis
calculations. Finally, we summarize our findings and analyses sets ranging from aug-cc-pVDZ to aug-cc-pV5Z58 with effec-
in Sec. IV. tive core potentials for Cu.59 The Coulomb interaction terms
were evaluated using the resolution-of-the-identity approxi-
mation.60,61 The GW energies in the complete basis set limit
II. COMPUTATIONAL DETAILS were obtained by fitting the extrapolated energy E ∞ and coef-
Computations within DFT were carried out using the ficient c to E(X) = E ∞ + cX −2 , where X = 2–5 for aug-cc-
NWChem code,48 Version 6.5, using the aug-cc-pVTZ basis pVDZ through aug-cc-pV5Z basis sets, as follows: For Cu2 O− ,
set. In these calculations, the geometries of the anions were CuO− , and CuO−3 , the variation in the computed GW quasi-
optimized using both the Perdew-Burke-Ernzerhof (PBE)49 particle energies with a PBE starting point was observed to
and PBE050 exchange-correlation functionals. The photoelec- be non-monotonic as a function of the basis set size when
tron spectra were simulated by convoluting the DFT eigenvalue aug-cc-pVQZ and aug-cc-pV5Z basis sets were included.
spectra of the anions with a 0.1-eV-wide Gaussian broadening Accordingly, for these three cluster anions, the G0 W 0 @PBE
function, without taking photoionization cross sections into quasiparticle energies were obtained by fitting to X = 2–3 only,
account. For the PBE and PBE0 calculations, the eigenvalue while for CuO−2 and all other starting points, we extrapolated
spectra were shifted so as to align the first peak with the vertical from X = 3–5.
ionization potential of the anion, computed as the total energy
difference between the anionic and the neutral cluster at the III. RESULTS AND DISCUSSION
fixed anion geometry.51,52 The amounts of the shifts and the
vertical ionization potentials for each case are provided in the The optimized structures of copper oxide cluster anions
supplementary material. OT-RSH calculations were based on are shown in Fig. 1. With the exception of one of the isomers
the combination of full long-range Fock exchange, a fraction of of CuO−3 , the computed inter-atomic distances and angles are
short-range Fock exchange, and PBE-based semi-local short- found to be mostly insensitive to the choice of the exchange-
range and long-range components, along with PBE correlation. correlation functional. Both CuO− and CuO−3 of C 2v symme-
The calculations were performed for two different values of try possess singlet ground states, whereas the linear CuO−2
the short-range Fock exchange fraction (α), 0.0 and 0.2,53 as (D∞h ) and the CuO−3 cluster of C s symmetry possess triplet
in the functionals long-range-corrected (LC)-ωPBE54 and LC- ground states. The bent Cu2 O− (C2v symmetry) is found to
ωPBE0,55 respectively. In difference to the parent functionals, have a doublet ground state. Importantly, the slight differences
the range-separation parameter, γ, was optimally tuned per observed in the bond lengths and angles did not have a signifi-
system so as to satisfy the ionization potential theorem,56 cant effect on the computed photoelectron spectra. Therefore,
γ opt differences in the photoelectron spectra computed at various
 H = −IPγ , where  H and IP are the highest occupied
opt

levels of theory, reported next, reflect true electronic struc-


molecular orbital (HOMO) energy and (vertical) ionization
ture effects rather than differences inherited from different
potential of the anion, respectively.
geometries.
GW calculations were performed using molgw57 within
the perturbative “one-shot” G0 W 0 methodology. In this
A. Cu2 O−
approximation, a Green’s function, G0 , computed using
Kohn-Sham wavefunctions and eigenvalues, and the screened We start our analysis of computed photoelectron spectra
Coulomb interaction, W 0 , obtained from a dynamical with Cu2 O− , which has the largest Cu content of the anions

FIG. 1. Structures and symmetries of


the Cux O−y clusters considered in this
study. O atoms are denoted by smaller
red balls and Cu atoms are denoted
by larger blue balls. Optimized bond
lengths and angles have been computed
with the PBE and PBE0 (in parenthe-
ses) functionals. For CuO−3 , the structure
with C s symmetry on the right is lower
in energy.
064306-3 Shi et al. J. Chem. Phys. 149, 064306 (2018)

considered in this study. Figure 2 shows the experimental pho- than the experimental value of 1.43 eV. Introducing a frac-
toelectron spectrum of Cu2 O− along with spectra computed at tion of Fock exchange, either globally or in range-separated
various levels of theory. The experimental spectrum consists form (the latter to correct the asymptotic potential), signifi-
of a low-energy peak at 1.1 eV (X), followed by a sizeable cantly improves the agreement with the experiment, with the
gap, and sharp peaks at 2.53 (A), 2.66 (B), 2.85 (C), 2.95 OT-RSH approach additionally removing the need for shifting
(D), and 3.08 eV (E). Wang et al.15 interpreted the large gap the spectra.53,56 Specifically, the mean absolute errors (MAEs)
between the X and A peaks as an indicator of a large high- averaged over the six measured (X–E) peaks for shifted PBE,
est occupied molecular orbital-lowest unoccupied molecular shifted PBE0, OT-RSH (α = 0), and OT-RSH (α = 0.2) calcu-
orbital (HOMO-LUMO) gap for neutral Cu2 O, suggesting that lations are 0.60, 0.19, 0.12, and 0.10 eV, respectively. These
the neutral cluster has a closed-shell electronic structure. The trends are qualitatively the same as those observed previously
similarity in the spectral features and intensity ratios between for a metal-organic Cu-based complex, copper phthalocyanine
the A and C bands, as well as between the B and D bands, led (CuPc).53,62,63
Wang et al. to suggest that these bands arise from the removal The predictions from the G0 W 0 approximation with a PBE
of either a spin-down or a spin-up electron from orbitals of the starting point (Fig. 2) are strikingly poor when compared with
same character. the experimental data. For example, the X and A peaks, as
With both the PBE and PBE0 functionals, the ground state well as their separation, are underestimated severely, by 0.6,
of Cu2 O− is found to be a doublet and the neutral cluster 1.44, and 0.59 eV, respectively. In fact, comparison of the spec-
is found to be a closed-shell singlet in its ground state. An tra displayed in Fig. 2 reveals that the G0 W 0 @PBE performs
overall comparison of the experimental data with the DFT- even worse than the shifted PBE spectrum, leading to a MAE of
computed spectra (Fig. 2) shows that the (shifted) PBE does 1.36 eV. Starting the G0 W 0 calculation with a DFT calculation
not provide quantitatively accurate predictions for the pho- that contains some fraction of exact exchange leads to better
toelectron peak positions, while various hybrid functionals agreement with experiment. For example, the G0 W 0 @PBE0
(with or without range separation) perform significantly bet- predictions are considerably better than the G0 W 0 @PBE
ter. The most obvious discrepancy between experimental data results. In spite of this improvement, the G0 W 0 @PBE0 still
and PBE predictions is the separation of the X band from leads to underbound quasiparticle levels. Similar to the shifted
the A–E bands, which is severely underestimated. The X–A PBE case, the shifted PBE0 predictions are more accurate than
separation of 0.56 eV computed with PBE is much smaller the G0 W 0 @PBE0 predictions, for which the MAE over the six
experimentally measured peaks is 0.44 eV. To investigate the
effect of the amount of exact exchange in the G0 W 0 start-
ing point further, we performed G0 W 0 calculations on top of a
BHLYP64 starting point (with 50% exact exchange). As shown
in Fig. 2, the G0 W 0 @BHLYP results are in nearly perfect
agreement with the experimental data, with the largest dif-
ference being 0.07 eV (for the X peak) and an overall MAE
of less than 0.04 eV. While the importance of using hybrid
functional starting points for the ensuing G0 W 0 calculations
has been highlighted in previous studies of sp-bonded molec-
ular systems,7,65–68 of metal-organic complexes,63,69 and of
bulk metal oxides,5,70 the results presented here show just
how poorly the PBE starting point can be and that basing a
GW calculation on it can in fact make things worse rather than
better.
To gain further insight into our computational results,
Fig. 2 additionally compares them with the quantum chemistry
calculations reported in Ref. 10, which were performed within
the equation-of-motion coupled-cluster (EOM-CC) approach,
at the singles and doubles (CCSD) level, as well as from
differences in CCSD with perturbative triples total energies
between the doublet anion and the lowest neutral state (to
determine the X peak). These calculations predict the photo-
FIG. 2. Experimental photoelectron spectrum of Cu2 O− (adapted from electron peaks X–E at energies of 1.00, 2.62, 2.68, 3.09, 3.19,
Ref. 15), along with spectra computed with PBE, PBE0, OT-RSH (α = 0, and 3.33 eV. Overall, these agree well with the experimental
γ opt = 0.221 a.u.−1 ), OT-RSH (α = 0.2, γ opt = 0.188 a.u.−1 ), G0 W 0 @PBE, spectrum, with an MAE of 0.16 averaged over the six mea-
G0 W 0 @PBE0, G0 W 0 @BHLYP, and equation-of-motion coupled-cluster
(EOM-CC) (adapted from Ref. 10). The PBE and PBE0 spectra are shifted to sured values. Remarkably, both the GW calculations (with an
align the first peak with the IP of the anion cluster. Contour plots of the selected appropriately chosen starting point) and the DFT ones (based
molecular orbitals (with symmetries), in both majority (↑) and minority (↓) on OT-RSH) are overall as accurate as the EOM-CC ones,
spin channels, are shown on the right-hand side, with matching color codes
establishing that quantitative interpretation of the experimen-
displayed in the spectra (for G0 W 0 @PBE, the eigenvalues are color-coded).
For PBE and G0 W 0 @PBE, the spectra are scaled by half compared with the tal data can be obtained for this cluster anion even without
others to fit them in the panels. resorting to expensive wavefunction-based approaches.
064306-4 Shi et al. J. Chem. Phys. 149, 064306 (2018)

Figure 2 also shows the DFT-computed molecular orbitals


for the highest seven (counting both spin channels) occupied
states of Cu2 O− . Based on the nodal structure along the Cu–O
and Cu–Cu directions, these can be viewed as anti-bonding
orbitals. The HOMO, possessing a1 symmetry, is a relatively
extended orbital with significant Cu content, with appreciable
contributions from 3d and 4s atomic orbitals and some from
4p, and smaller (∼15%) O 2p character. In going from PBE
to PBE0, the shape of the molecular orbital remains virtually
the same, but the Cu 4s content increases at the expense of a
decrease in the Cu 3d content. A similar behavior is observed
for the lower lying states of b2 , b1 , and a1 symmetry, for which
the Cu 3d content decreases at the expense of an increase in the
O 2p content. These observations are in accordance with the
expectation that introduction of exact exchange should push
the localized Cu 3d orbitals down in energy due to mitigation
of self-interaction error (SIE).71 Another difference between
PBE and PBE0 predictions is related to the ordering of the
various orbitals. In particular, the ordering of the first five
occupied orbitals in increasing the binding energy (BE) at the
PBE level is a1 ↑ b2 ↓ b2 ↑ b1 ↓ b1 ↑ , while the ordering at the PBE0
level is a1 ↑ b1 ↓ b2 ↓ b2 ↑ b1 ↑ . While in some cases the relevant
energy differences are small, such rearrangement of orbitals is
generally related to (i) the spatial extent (degree of localiza-
tion) of the relevant orbitals, which determines the amount that
the introduction of exact exchange shifts the orbital eigenval-
ues in going from PBE to PBE0 to partially correct for SIE,2
and (ii) the tendency of exact exchange to segregate (energeti-
cally) orbitals belonging to the same spin channel. In this case, FIG. 3. Experimental photoelectron spectrum of CuO− (Ref. 16), along
with spectra computed with PBE, PBE0, OT-RSH (α = 0, γ opt = 0.225
PBE0 leads to a configuration in which HOMO-1/HOMO- a.u.−1 ), OT-RSH (α = 0.2, γ opt = 0.138 a.u.−1 ), G0 W 0 @PBE, G0 W 0 @PBE0,
3 (and HOMO-2/HOMO-4), associated with A/C (and B/D) G0 W 0 @BHLYP, and EOM-CC (Ref. 10). The PBE and PBE0 spectra are
bands in the experiment, are in opposite spin channels (unlike shifted to align the first peak with the IP of the anion cluster. Contour plots of
molecular orbitals discussed in the text are shown on the right, with matching
PBE). They also have different symmetries (b1 and b2 ), in
color codes displayed in the spectra.
agreement with the experimental suggestion that A/C and B/D
bands correspond to removal of spin-up/spin-down electrons
from orbitals of the same character. When viewed both from state of CuO− was found to be a closed-shell singlet (1 Σ+ )
the perspective of low MAE and the experimentally suggested that could roughly be described as 3d 10 2pσ 2 2pπ 4 . Based on
ordering of the orbitals, the G0 W 0 @BHLYP prediction with these studies and the relative intensity ratios, X and Y states
the a1 ↑ b1 ↓ b2 ↓ b1 ↑ b2 ↑ ordering and the OT-RSH (α = 0, and α were interpreted as arising from the removal of 2pπ and 2pσ
= 0.2 being very close) prediction with the a1 ↑ b2 ↓ b1 ↓ b2 ↑ b1 ↑ electrons, leading to the X2 Π ground state and Y2 Σ+ excited
ordering have the best overall agreement with the state of CuO, respectively. The experiments of Wu et al. also
experiment. revealed a broad and noisy band (labeled Z in Fig. 3) in the
In summary, the photoelectron spectra of Cu2 O− com- 4–6 eV energy range, which was interpreted as being due to
puted with hybrid functionals within DFT, or within the G0 W 0 the detachment of Cu 3d electrons. In addition, weak features
approximation with hybrid functional starting points, are in at 1.27 and 3.18 eV were observed, which were attributed
much better agreement with the experimental data and EOM- to transitions from an electronically excited state of CuO− .
CC results, as compared with the PBE (or G0 W 0 @PBE) In the following discussion, we only focus on the X, Y, and
predictions. This is a somewhat expected finding, because as Z bands, corresponding to transitions from the ground state
discussed above the introduction of Fock exchange mitigates of CuO− .
SIE and binds orbitals of large Cu 3d content more strongly, We first discuss DFT and GW results for the X and Y
while PBE typically underbinds such orbitals. bands. A comparison of the computed spectra with the exper-
imental data surprisingly shows that the PBE provides the
best predictions (within 0.05 eV of experiment) for the posi-
B. CuO−
tions of the first two peaks. The PBE0 and OT-RSH (α = 0.2)
Figure 3 shows the computed and experimental photo- predictions are virtually the same, but they significantly under-
electron spectra of CuO− . In the experiments of Polak et al.14 estimate both the IP and the X–Y separation, by ∼0.5 eV. The
and Wu et al.,16 the first two main peaks (labeled X and Y OT-RSH predictions with α = 0 for the positions of the first
in Fig. 3) were reported at 1.78 and 2.75 eV, respectively. two peaks (both IP and X–Y separation underestimated by
In previous electronic structure calculations,18–21 the ground ∼0.2 eV) lie between those of PBE and those of other hybrid
064306-5 Shi et al. J. Chem. Phys. 149, 064306 (2018)

functionals. Similar to the case of Cu2 O− , the G0 W 0 @PBE


results are poor, with both X and Y peaks underestimated by
∼1.4 eV. The use of a PBE0 starting point for G0 W 0 calcula-
tions significantly improves the quasiparticle energies, but the
X and Y levels are still underestimated by ∼0.4 eV compared
with the experiment. Better agreement with the experiment
can be obtained at the G0 W 0 @BHLYP level, where the X
and Y peaks are predicted to be 0.07 and 0.32 eV below the
experimental values.
As in the discussion of Cu2 O− , we additionally compare
our results with those obtained from the EOM-CC calcula-
tions.10 The best results obtained from such calculations find
transitions at 1.89 and 2.56 eV from the CuO− reference. These
values are within ∼0.1 and ∼0.2 eV of experimental data. The
EOM-CC calculations also provide a convenient point of ref-
erence for the Z-band calculations, where the broad and noisy FIG. 4. Unshifted eigenvalue spectra of CuO− computed for various values
of exact exchange and PBE correlation, defined by Eq. (1). Orbitals of π ? ,
nature of the Z band precludes direct comparisons with the σ ? , non-bonding d, π, and σ character (same as in Fig. 3) are color-coded to
experiment. These higher energy transitions, arising from the highlight changes in their energies as a function of α and β. See text for more
ionization of orbitals with large Cu 3d character, were found to details.
be 4.83, 5.11, and 5.57 eV.10 These values fall within the range
(4–6 eV) of the Z band observed in the photoemission exper- respectively, we performed a series of computations by vary-
iments and therefore indeed provide an adequate benchmark. ing α and β from 0 to 1. As discussed above, in general, one
Comparing this benchmark with the computational approaches excepts stronger effects of adding Fock exchange to PBE cal-
employed here, we observe that PBE significantly underesti- culations for more localized orbitals, as the SIE is more severe
mates the BEs of these higher energy states, while the hybrid in spatially localized orbitals compared with the more extended
functionals do slightly better. The G0 W 0 @PBE predictions are ones. This is indeed what we observe in CuO− too. While all
again poor, while hybrid functional starting points, especially orbitals become more bound (DFT eigenvalues decreasing),
G0 W 0 @BHLYP, perform much better. Overall, taking all five the magnitude of eigenvalue decrease correlates directly with
peaks (with BEs less than 6 eV) into account, the EOM-CC the spatial extent of the molecular orbital involved. In particu-
performs quite well. The G0 W 0 @BHLYP results are also in lar, the spatial extent of the σ? orbital is larger than that of the
good agreement with the experiment and quantum chemistry π? orbital. As a result, as more Fock exchange is introduced,
results. The PBE predictions are excellent for the two most the eigenvalue of the π? orbital decreases more significantly
loosely bound states, but they do not perform as well for states (by ∼5 eV) than that of the σ? orbital (by ∼3.2 eV) in going
with higher BEs, while the opposite trend is observed for the from α = 0 to 1 (at β = 1). For the more localized lower
case of hybrid functionals. lying orbitals with large d character, the eigenvalue shifts to
Figure 3 also provides the orbital characters of the molecu- lower energies upon increasing Fock exchange are consider-
lar orbitals involved. The doubly degenerate HOMO/HOMO-1 ably larger than those of σ? and π? orbitals. Among them,
has π? character: These antibonding states can be described the bonding σ orbital (HOMO-7 at the PBE level) is the least
well as nearly equal mixtures of Cu d xz (d yz ) and O 2px localized one, and this orbital has the smallest eigenvalue shift
(2py ) atomic orbitals. The non-degenerate HOMO-2 has in going from α = 0 to α = 1, at which point this σ orbital
anti-bonding σ? character with significant Cu dz2 and 4s becomes the least bound one among orbitals of large d char-
contribution and O 2pz admixture. The doubly degenerate acter, similar to what is observed for the case of σ? and π?
HOMO-3/HOMO-4 are non-bonding Cu d xy and dx2 −y2 type orbitals.
orbitals. HOMO-5/HOMO-6 are also doubly degenerate, cor- Of particular importance in Fig. 4 is the ordering of the
responding to bonding π orbitals, while HOMO-7 is a bonding orbitals at the HF level (equivalent to α = 1, β = 0), where
σ orbital, both with large contributions from Cu d atomic the HOMO is incorrectly predicted to be of σ? character and
orbitals. the doubly degenerate HOMO-1/HOMO-2 has π? character.
The surprising observation that PBE outperforms PBE0 Adding more semilocal correlation to HF decreases the mag-
for the position of the first two (X, Y ) peaks can be inter- nitude of the π? − σ? separation slightly, but the HOMO still
preted in terms of the spatial extent of the relevant orbitals and has σ? character even at β = 1. Upon removing some of the
the compatibility of exact exchange and correlation. Figure 4 exact exchange, however, the π? − σ? ordering gets reversed,
shows the unshifted eigenvalue spectra of CuO− , computed for and for α . 0.5, HOMO has π? character, in agreement with
various values of exact exchange and semilocal (PBE) correla- the experimental data. Therefore, the small π? − σ? separation
tion. More specifically, using an exchange-correlation energy predicted by PBE0 (α = 0.25, β = 1) can be traced to the incor-
E xc (α, β) expressed as rect description of the ordering at the HF level, with PBE0 still
Exc (α, β) = αEx,HF + (1 − α)Ex,PBE + βEc,PBE , (1) having “too much” exact exchange or “not enough” semilocal
exchange. Since semilocal exchange is known to partially rep-
where E x ,HF is the Hartree-Fock exchange and E x ,PBE resent non-dynamical correlation,72–76 we attribute the appar-
and E c,PBE are semilocal PBE exchange and correlation, ent success of the PBE predictions (for the first two peaks) to
064306-6 Shi et al. J. Chem. Phys. 149, 064306 (2018)

a more accurate accounting of non-dynamical correlation in character. However, for the first two peaks, the semi-local
PBE compared with PBE0. PBE functional provides a much better prediction. This some-
In order to investigate the effect of the degree of orbital what surprising finding is interpreted in terms of a stronger
localization on the difference between semilocal and hybrid SIE effect for the lower lying states, as compared with a
functional predictions, we have also performed PBE and PBE0 larger role of non-dynamical correlation for the higher lying
calculations for diatomic molecules CuS− and AgO− and com- states.
pared with CuO− . Substituting Cu and O with isovalent ele-
ments in the next row should result in more extended d− and
C. CuO−2
p−like orbitals, respectively. Figure 5 shows the predicted pho-
toelectron spectra of CuS− , AgO− , and CuO− at the PBE and For the linear CuO−2 structure shown in Fig. 1, DFT cal-
PBE0 levels. For CuS− , the agreement between PBE and PBE0 culations at both the PBE and the PBE0 level show that the
for the first two peaks (π? and σ?) is very good, much bet- molecule does not possess a closed-shell singlet configuration
ter than the case for CuO− , both in terms of the position of as originally assumed,16,34 but rather it has a triplet ground
the first peak (doubly degenerate HOMO and HOMO-1) and state. At the PBE level, the (3 Σg− ) triplet state of CuO−2 is
the π? − σ? separation. Since these three orbitals making up energetically favorable than the 1 Σg+ singlet configuration by
the first two peaks have large S 3p character, they are signifi- 0.71 eV, in good agreement with the value of 0.68 eV obtained
cantly more spatially extended than the corresponding orbitals previously by Deng et al.39 At the PBE0 level, the triplet
in CuO− with large O 2p character. Indeed, we have observed state is energetically favorable by a larger energy difference
no π? − σ? crossing as a function of α for CuS− , which has of 1.22 eV.
a doubly degenerate HOMO even at the HF level of theory. Removing an electron from the triplet CuO−2 to find its IP
Lower lying orbitals of large Cu 3d character, on the other requires more careful considerations, as the electron removal
hand, are still significantly underbound at the PBE level com- can lead to either a doublet or a quartet configuration of the
pared with the PBE0, similar to what is observed in CuO− . neutral cluster. With PBE, the quartet state (4 Π u ) is lower
For AgO− , on the other hand, the first two peaks have very in energy than the doublet state (2 Π g ), by a relatively small
similar behavior to that observed in CuO− , with significant 0.09 eV. However, we find that the quartet configuration pos-
differences between PBE and PBE0 predictions. Also similar sesses unstable vibrational modes, again in agreement with
to CuO− , the HOMO of AgO− at HF level is found to be the the results of Deng et al.39 Furthermore, the doubly degener-
singly degenerate orbital of σ? character. For the lower lying ate HOMO of CuO−2 is in the majority spin channel, and the
orbitals of largely d character, on the other hand, the agree- charge density difference between the triplet state of CuO−2
ment between PBE and PBE0 predictions is quite good, as and the doublet state of neutral CuO2 indeed corresponds to
these molecular orbitals primarily derived from Ag 4d atomic the charge density distribution associated with the HOMO of
orbitals are significantly more extended than their counterparts CuO−2 . PBE0 also predicts the quartet state of the neutral to be
in CuO− and undergo less severe self-interaction corrections lower in energy than the doublet state but with a much larger
when a fraction of Fock exchange is mixed in at the PBE0 energy difference of 1.36 eV. Furthermore, PBE0 calculations
level. do not reveal any unstable vibrational modes, and because the
In summary, as in the case of Cu2 O− , the photoelectron doubly degenerate HOMO of CuO−2 is in the minority spin
spectrum of CuO− is well captured by the G0 W 0 @BHLYP channel, the charge density difference between the triplet state
and the hybrid functionals do well in predicting the excita- of CuO−2 and the quartet state of neutral CuO2 does correspond
tion energies of lower lying states of predominantly Cu 3d to the charge density distribution associated with the HOMO
of CuO−2 .
Figure 6 shows the computed and experimental photo-
electron spectra of CuO−2 . In the experiments of Wu et al.,16
sharp photoelectron peaks were observed at 3.47 (X), 3.79 (A),
4.10 (B), 4.28 (C), 4.67 (D), and 5.16 eV (E). In light of the
above discussion, the PBE and PBE0 spectra have been shifted
according to the total energy difference between the triplet
3 Σ − state of CuO− and the doublet or quartet state of CuO ,
g 2 2
respectively. As explained above, for the OT-RSH results, no
shift of eigenvalues is needed. Nevertheless, the neutral energy
still comes into play as part of the optimal tuning process.
As in PBE0, the OT-RSH calculations also found the quartet
to be lower in energy and with a charge density difference
compatible with the HOMO of the anion.
Comparison of the computed peaks with the measured
photoelectron spectrum reveals interesting trends. With PBE,
the onset energy of 4.01 eV is much too high in comparison
FIG. 5. Eigenvalue spectra of CuS− , AgO− , and CuO− computed with PBE
with the measured value at 3.47 eV. However, the overall line
and PBE0 functionals. The spectra are shifted to align the first peak with the
IP of the anion cluster. Orbitals of π ? , σ ? , non-bonding d, π, and σ character shape of the spectrum appears to correspond to the experiment
are same as in Figs. 3 and 4. fairly well. Indeed, Fig. 6 also reveals that if the PBE data are
064306-7 Shi et al. J. Chem. Phys. 149, 064306 (2018)

the G0 W 0 @BHLYP, which was shown earlier to lead to very


good agreement with the experiment for Cu2 O− (Fig. 2) and
CuO− (Fig. 3), does not perform well for CuO−2 : The sepa-
ration of 1.53 eV between the first two peaks computed with
G0 W 0 @BHLYP is much too large compared with the exper-
imental X − A separation, resulting in a spectrum where all
(A–E) but the lowest energy (X) transitions are predicted above
4.75 eV, approximately 1 eV higher than the experimental
spectrum.
As the differences between the PBE and PBE0 predic-
tions are striking, it is instructive to examine them in more
detail. At the PBE level, the HOMO of CuO−2 occurs in the
majority spin channel and is a doubly degenerate orbital of
eg symmetry that is a nearly equal mixture of O 2p and Cu
3d states. The LUMO in this case corresponds to the same
(spin-split) eg orbital in the minority spin channel. At the
PBE0 level, on the other hand, while LUMO is still the eg ↓
orbital, HOMO occurs in the minority spin channel and has
eu symmetry, which is almost purely composed of O 2p states
(in a slightly π bonding configuration). In Fig. 7, we plot the
(unshifted) eigenvalue spectra of CuO−2 at the PBE, PBE0, and
HF levels. As observed in the figure, this switch in the orbital
character of HOMO occurs as the eg ↑ orbital gets pushed down
by as much as 1.98 eV, while the eu ↓ orbital is pushed down
by only 0.6 eV in going from PBE to PBE0. At first sight, it
is tempting to interpret this large difference in the downward
FIG. 6. Experimental photoelectron spectrum of CuO−2 (Ref. 16) along with shift of the eigenvalues in terms of the character of the relevant
spectra computed with PBE, PBE0, OT-RSH (α = 0, γ opt = 0.314 a.u.−1 ), orbitals, namely, since the eg orbital has approximately 50%
OT-RSH (α = 0.2, γ opt = 0.230 a.u.−1 ), G0 W 0 @PBE, G0 W 0 @PBE0, and Cu 3d character, while the eu orbital has practically no con-
G0 W 0 @BHLYP. The PBE and PBE0 spectra are shifted to align the first peak tribution from the Cu 3d states, the SIE is expected to affect
with the IP of the anion cluster. Also shown is the PBE spectrum shifted by
the IP of the quartet configuration. See the text for details. Contour plots of the
molecular orbitals discussed in the text, with matching color codes displayed
in the spectra, are shown on the right.

shifted by the IP (3.92 eV) of the “incompatible” quartet con-


figuration anyway, improvement with the experiment improves
drastically. We note that in this scenario the PBE spectrum is
shifted so as to align the doubly degenerate HOMO-1/HOMO-
2 in the minority spin channel with the IP of the quartet
configuration, so that the removal of one of the electrons in
HOMO-1/HOMO-2 of CuO−2 would lead to the 4 Π u configu-
ration of CuO2 . However, this comes at the cost of predictive
power, as strict validity of the theoretical procedure demands
that the doublet be used. With PBE0 or OT-RSH, while the
onset energy is close to the measured value [especially with
OT-RSH (α = 0) for which the onset is 3.50 eV], separations
between the first two computed peaks, which are 0.85, 1.03,
and 1.27 eV for PBE0, OT-RSH (α = 0), and OT-RSH (α = 0.2),
respectively, are much too large compared with the experimen-
tal X − A separation of 0.32 eV and, in fact, are in much worse
agreement with the experiment than the prediction of PBE.
Turning to the GW results, as before the G0 W 0 @PBE quasi-
particle energies with an onset energy of 2.81 eV are signif-
icantly underestimated with respect to the experimental data.
However, unlike the case for Cu2 O− and CuO− , G0 W 0 @PBE0 FIG. 7. (Unshifted) Eigenvalue spectra of CuO−2 (along with symmetries of
the associated orbitals) in the majority (↑) and minority (↓) spin channels
provides very good agreement with the experimental data: The
at the PBE, PBE0, and HF level of theory. Contour plots of the molecular
onset energy is only 0.09 eV higher than the X peak, and the orbitals discussed in the text, with color codes matching the levels shown in
MAE over the six measured peaks is 0.17 eV. Interestingly, the spectra, are shown on the left.
064306-8 Shi et al. J. Chem. Phys. 149, 064306 (2018)

the eg orbital more than the eu orbital, leading to the observed


change in the HOMO orbital character. However, SIE cannot
be the sole explanation, as the orbital of eu symmetry in the
majority spin channel (eu ↑ ), which has the same nearly pure O
2p composition as eu ↓ , is observed to shift down by the large
amount of 1.87 eV, unlike its eu ↓ counterpart, in going from
PBE to PBE0. Note that this difference in the amount by which
spin-split orbitals get pushed down in energy becomes most
obvious in the extreme case of HF level of theory; e.g., the
eu ↑ and eu ↓ orbitals shift down by the significantly different
amounts of 7.88 and 1.93 eV, respectively, in going from PBE
to HF.
In addition to SIE, another factor that contributes to the
different ordering of orbitals at the PBE and PBE0 level is
the amount of spin-splitting. In general, introduction of exact
exchange is expected to increase spin-splitting. This is indeed FIG. 8. DFT eigenvalue centers (average of majority and minority spin chan-
what we observe in CuO−2 , especially for doubly degenerate nel eigenvalues) of CuO−2 orbitals mentioned in the text computed with
orbitals for which the increase in spin-splitting as a func- differing amounts of exact exchange [α in Eq. (1) varying from 0 to 1, with
β = 1]. The orbitals have the same color code as in Figs. 6 and 7. The eigenvalue
tion of the amount of exact exchange is particularly large. centers for HF calculations (α = 1, β = 0) are also shown.
For example, the 1.22 eV spin-splitting of the eu orbitals (of
predominantly O 2p character) at the PBE level increases
to 2.50 and 7.20 at the PBE0 and HF levels, respectively. spin-channel-dependent, we focus on how the centers of spin-
Similarly, the 1.38 eV spin-splitting of the eg orbital men- split states change as a function of exact exchange (α), as
tioned earlier (HOMO and LUMO at PBE) increases to 4.53 shown in Fig. 8. This averaging over spin-splitting provides a
and 15.35 eV at PBE0 and HF, respectively. Accordingly, as clearer effect of SIE. Indeed, the trends observed in Fig. 8 are
more exact exchange is introduced, the eigenvalue spectrum consistent with the orbital characters of the states. The states
becomes more segregated with respect to the spin channel; with symmetries of a1g (σ bonding), b1g , b2g (non-bonding
namely, the spectrum consists of clusters of several states in Cu d xy and dx2 −y2 orbitals), and eg (π bonding), which have
the majority/minority spin channels before switching to the pure or very large Cu 3d content, fall into one category, which
other spin channel. For example, as shown in Fig. 7, the six are affected the most by exact exchange. Above them are three
highest energy-occupied orbitals at the HF level all occur in the states with symmetries of a2u , a1g (σ anti-bonding), and eu
minority spin channel, followed by six orbitals in the majority (π bonding of O 2p orbitals), which have very little or (practi-
spin channel, and so on. This is significantly different from the cally) no Cu 3d content. Finally, although the center of the eg
much more homogeneous ordering of orbitals (with respect to state (nearly equal mixture of O 2p and Cu 3d orbitals in an
spin channel) at the PBE level. This observation can be qual- anti-bonding π configuration) does not show a significant vari-
itatively understood as follows: Removing an electron from ation as a function of α, the reason for this is that the unoccu-
the minority (as opposed to majority) spin channel of a given pied eg ↓ orbital (LUMO at PBE and PBE0 levels and LUMO+5
orbital in the anion cluster would result in a neutral cluster at HF) is pushed up in energy to maintain the symmetry of
configuration with a larger spin imbalance and hence be ener- spin-splitting. As shown in Fig. 7, the occupied eg ↑ orbital
getically more favorable due to the increased magnitude of does experience significant SIE, as the level moves down
the exchange energy. Because the HF eigenvalues are related by ∼8.3 eV.
to (unrelaxed) electron removal energies, it makes sense that The above analysis shows that on the one hand, there
in all the occupied spin-split orbitals, those in the minority is a significant role played by SIE, indicating a preference
channel appear at higher energies (less negative eigenvalues) for hybrid functionals. But on the other hand, the disagree-
than those in the majority spin channel, as observed in Fig. 7. ment between the eigenvalue spectra produced by the hybrid
However, one should notice that this exchange interaction has functionals and experiment, along with the good agreement
a profound effect, at the HF level, not just on spin-split orbitals (up to a shift) between the PBE data and the experiment,
but also on orbitals of significantly different shape and atomic suggests a strong role of non-dynamical correlation. This sug-
orbital content. For example, Fig. 7 shows that at the HF level gests that a useful compromise between these two constraints
the six highest eigenvalues all occur in the minority spin chan- may be sought by using a Fock exchange fraction α that is
nel. In particular, the doubly degenerate HOMO-2/HOMO-3 lower than the default 25% in PBE0. This idea has prece-
of eg symmetry in the minority spin channel, where the Cu dence in the context of energy differences between high-spin
atom is in a pdπ bonding configuration with the O atoms, is and low-spin energy splittings in spin-crossover molecules,
located ∼3.30 eV higher than the eu orbitals in the majority such as Fe(ii)–S complexes, where an exact exchange admix-
spin channel, in spite of the fact that eg ↓ orbitals have ∼85% ture of 10%-15% was recommended.77–79 Figure 9 shows the
Cu 3d character, while the eu ↑ orbitals have almost purely O shifted spectra computed with α values ranging from 0 (PBE)
2p content. to 0.25 (PBE0). We observe that it is indeed possible to get
Because the amount by which spin-split levels are very good agreement with the experimental data if one uses
pushed down in energy with more exact exchange is a Fock exchange fraction of 15%. In this case, the largest
064306-9 Shi et al. J. Chem. Phys. 149, 064306 (2018)

FIG. 9. The experimental photoelectron spectrum of CuO−2 along with DFT


eigenvalue spectra computed with Fock exchange fractions α varying from 0
(PBE) to 0.25 (PBE0). In each case, the spectrum is shifted to align the first
peak with the IP of the anion cluster.

discrepancy of 0.25 eV is in the predicted position of the A


peak and the MAE averaged over the six experimental peaks FIG. 10. Experimental photoelectron spectrum of CuO−3 (Ref. 16), along with
is 0.09 eV. spectra computed using PBE, PBE0, OT-RSH (α = 0, γ opt = 0.349 a.u.−1 ),
OT-RSH (α = 0.2, γ opt = 0.284 a.u.−1 ), G0 W 0 @PBE, and G0 W 0 @PBE0 for
In summary, unlike in the previous cases, the photo- the lower energy (Iso1) C s structure of CuO−3 . The PBE and PBE0 spectra are
electron spectrum of CuO−2 is not well captured by PBE, or shifted to align the first peak with the IP of the anion cluster.
PBE0/OT-RSH, or even by GW based on a BHLYP starting
point. This is likely owing to a large role of both non-dynamical
correlation and SIE considerations, with the former benefiting possibility of feature B (separated from A by 0.12 eV) being
from lack of Fock exchange and the latter suffering from it. a vibrational replica of A. The spectrum of CuO−3 is basically
However, using a smaller fraction of exact exchange (15% divided into two main groups, with features X, A, and B form-
for PBE0 and 25% as the starting point for GW ) results in a ing the first group and features C and D forming the second.
quantitatively useful compromise between the two conflicting These two groups are followed by the broader and weaker fea-
requirements. ture E. The observed separation between the two main groups
of peaks is roughly the same as the X–Y separation in CuO− .
Accordingly, Wu et al. suggested that these two groups could
D. CuO−3
be interpreted as derived from the X–Y states of CuO− per-
For CuO−3 , the two lowest energy isomers we have found, turbed by an O2 molecule and assigned features X through D
shown in Fig. 1, have C s (Iso1) and C2v (Iso2) symmetries. Iso1 to O 2p orbitals and the E feature at high BE due to Cu 3d
possesses a triplet ground state, while the ground state of Iso2 orbitals.
is a singlet. Within both PBE and PBE0, Iso1 is lower in energy Comparison of the spectra computed with PBE, PBE0,
than Iso2, by 0.05 and 0.43 eV, respectively. Both Iso1 and Iso2 OT-RSH, G0 W 0 @PBE, and G0 W 0 @PBE0 for Iso1 (Fig. 10)
feature an O2 unit bonded to a CuO unit. This bonding scenario with the experimental data shows that none of the DFT or GW
is consistent with the work of Wu et al.,16 who suggested it methods results in satisfactory agreement with the experiment.
based on photodissociation events observed at a 355 nm photon For example, the onset energy with PBE (2.67 eV) is much
energy for both CuO−3 and Cu(O2 )− complex. smaller than the position of the X peak. While it may be possi-
Figures 10 and 11 show the experimental photoelectron ble to assign some of the peak positions predicted with hybrid
spectrum for the CuO−3 cluster, as compared with the theoret- functionals and G0 W 0 @PBE0 to photoelectron data (e.g., the
ical spectra computed for Iso1 (Fig. 10) and Iso2 (Fig. 11). first two peaks of PBE0 and G0 W 0 @PBE0 are within ∼0.15
In the experiments of Wu et al.,16 sharp photoelectron peaks and ∼0.04 eV of the experimental X and A bands, respectively),
were recorded at 3.39 (A) and 4.34 eV (D), as well as weaker the overall shapes of the predicted spectra do not resemble the
ones at 3.19 (X), 3.51 (B), 4.02 (C), and ∼5.8 eV (E), with the experimental spectrum that has two broad peaks centered at
064306-10 Shi et al. J. Chem. Phys. 149, 064306 (2018)

poor, with the first two peaks underestimated by 0.8-1.0 eV


and an averaged MAE of 0.73 eV. The G0 W 0 @PBE0, on the
other hand, provides excellent agreement with the experimen-
tal data for the first four peaks with an MAE of 0.07 eV. The
fifth peak, however, lies too low in the G0 W 0 @PBE0 cal-
culation. Similar to the case of CuO−2 , the G0 W 0 @BHLYP
results, while better than the G0 W 0 @PBE, are not good; the
predicted quasiparticle levels are found to be much more
bound compared with the experiment, and the averaged MAE
is 0.51 eV.
Analysis of the different orbital characters for Iso2, also
shown in Fig. 11 at various levels of theory, additionally lends
support to the assumption of a CuO− unit perturbed by an O2
unit: In particular, at the PBE level, HOMO-1 of b1 symmetry
and HOMO-2 of a1 symmetry have very similar shapes and
orbital content as those of X (π?) and Y (σ?) states of CuO−
(see Fig. 3), respectively, as they contain very little (less than
3%) contribution from the O2 unit. We note, however, that
these b1 and a1 states of CuO−3 do have appreciable (∼40 and
∼70%, respectively) Cu 3d content, similar to π? and σ? states
of CuO− . Orbitals of large O 2p character at the PBE level are
HOMO (a2 symmetry) and HOMO-3 (b2 symmetry) with very
FIG. 11. Experimental photoelectron spectrum of CuO−3 (Ref. 16), along with little (less than 10%) Cu 3d contribution.
spectra computed with PBE, PBE0, OT-RSH (α = 0, γ opt = 0.246 a.u.−1 ), One of the significant changes in the ordering of the
OT-RSH (α = 0.2, γ opt = 0.165 a.u.−1 ), G0 W 0 @PBE, G0 W 0 @PBE0, and
G0 W 0 @BHLYP for the Y-shaped Iso2 structure of CuO−3 (with C2v symme- orbitals in going from PBE to PBE0 for Iso2 of CuO−3 is the
try). The PBE and PBE0 spectra are shifted to align the first peak with the IP switching of a1 and b2 orbitals. This behavior can again be
of the anion cluster. Contour plots of the molecular orbitals (with associated understood in terms of the degree of spatial localization of the
symmetries) discussed in the text are shown on the right, with matching color relevant orbitals. In going from PBE to PBE0, orbitals of a2 ,
codes displayed in the computed spectra.
b1 , and a1 symmetry are pushed down in energy by the sim-
ilar amounts of 1.67, 1.60, and 1.45 eV, respectively, while
3.39 and 4.34 eV and no appreciable signals between 4.34 and the b2 orbital becomes more bound by only 0.76 eV (keep in
5.80 eV. mind that the spectra plotted in Fig. 11 are shifted eigenvalue
Agreement between theory obtained for Iso2 and experi- spectra). From the shapes and spatial extents of the molecular
ment, though not perfect, is much more satisfactory. Assuming orbitals plotted in Fig. 11, one can easily observe that the b2
that the experimental feature B is indeed a vibrational level is the least localized orbital, which accordingly gets affected
of feature A, we can compare the first four computed peaks the least by SIE, consistent with the difference in the shifts of
at each level of theory with features X, A, C, and D in the the corresponding eigenvalues.
experiment. The MAEs for PBE, PBE0, OT-RSH (α = 0), and Another important observation about the PBE and PBE0
OT-RSH (α = 0.2) are then found to be 0.17, 0.15, 0.24, and spectra of CuO−3 is related to the separation of the b1 and a1
0.34 eV, respectively. We therefore focus on Iso2, although orbitals, which correspond to the π? and σ? orbitals of CuO− ,
we comment further on the isomer issue below. We also note as mentioned above. With PBE, the b1 − a1 separation in CuO−3
that using a Fock exchange fraction α between PBE and PBE0 has exactly the same value of 0.88 eV as the π? − σ? separation
values leads to only slightly better agreement with the experi- of CuO− . However, with PBE0, while the π? − σ? separation
ment. The MAEs for α = 0.05, 0.10, 0.15 are found to be 0.085, of CuO− drops significantly to 0.46 eV, the b1 − a1 separation
0.12, and 0.10 eV, respectively. The computed binding ener- in CuO−3 is only reduced to 0.73 eV and remains close to the
gies are reported in the supplementary material. Therefore, for PBE value. Earlier, we attributed the large change in the π?
the analysis and discussion to follow, we focus on PBE, PBE0, − σ? separation of CuO− upon addition of exact exchange
OT-RSH, and G0 W 0 levels of theory. to a more accurate accounting of non-dynamical correlation
For Iso2, the first peak is predicted quite well (within ±0.1 in PBE compared with PBE0. Our finding of similar b1 − a1
eV) with both PBE and hybrid functionals. While PBE and separation in CuO−3 with PBE and PBE0 is then in accordance
PBE0 predictions for the first two peaks are very close to each with the expectation that non-dynamical correlation should
other and agree well with the experiment, the positions of the play a less significant role in larger and less symmetric molec-
next two peaks are overestimated and underestimated by PBE ular systems. Accordingly, both the PBE and the PBE0 do a
and PBE0, respectively, by 0.2-0.3 eV. Surprisingly, the exper- reasonably good job of describing the first two main (broad)
imental peak at ∼5.8 eV is best predicted with PBE, while the experimental peaks, while PBE is clearly much better for the
predictions from PBE0 and range-separated hybrid functionals CuO− molecule.
are not as good, even though the relevant orbitals in this energy The results additionally show that while starting points
range indeed have large (>90%) Cu 3d character. Similar to that involve some fraction of exact exchange are defi-
the other clusters considered, the G0 W 0 @PBE predictions are nitely needed for satisfactory agreement of one-shot G0 W 0
064306-11 Shi et al. J. Chem. Phys. 149, 064306 (2018)

predictions with experiment, the amount of that fraction is these findings call for careful assessment of the opposing roles
highly dependent on the cluster composition: Clusters with of non-dynamical correlation and self-interaction errors in
larger Cu content (Cu2 O− and CuO− ) appear to need DFT functional choice for small transition metal oxide and metal-
starting points with a larger exact exchange fraction, such organic clusters. They also call for further development of DFT
as BHLYP, than clusters with a smaller Cu content (CuO−2 and GW approaches that can balance the two issues in a natural
and CuO−3 ) for which the PBE0 starting point works quite way.
well.
Finally, we note that because Iso1 is slightly lower in SUPPLEMENTARY MATERIAL
energy than Iso2, one could have expected it to be the exper-
imentally pertinent isomer. Here, we note that generally pho- See supplementary material for tabulated PBE and PBE0
toemission measurements of clusters are not necessarily taken atomic coordinates, and experimental, DFT (PBE, PBE0, OT-
at thermal equilibrium and that higher energy clusters may RSH), and G0 W 0 (with PBE, PBE0, and BHLYP starting
dominate the experimental spectrum owing to kinetic con- points) binding energies for all copper oxide cluster anions
siderations.51 Still, the remaining differences between theory considered.
for Iso2 and experiment may indicate some presence of Iso1
under the experimental setup. Furthermore, the Iso1 structure ACKNOWLEDGMENTS
is inherently floppy and its orbital arrangement may depend B.S. and S.Ö. would like to acknowledge support from the
on the exact geometry, and therefore change with temperature. U.S. Department of Energy Grant No. DE-SC0017824, as well
The overall observed spectrum may then reflect contributions as the National Energy Research Scientific Computing Cen-
from both the isomers.80 Exploring this scenario, however, ter, a DOE Office of Science User Facility supported by the
would require extensive molecular dynamics simulations that Office of Science of the U.S. Department of Energy under Con-
are outside the scope of this article. tract No. DE-AC02-05CH11231, for computational resources.
F.B. acknowledges High Performance Computing resources
IV. SUMMARY from GENCI-CCRT-TGCC (Grant No. 2017-096018). Work
in Rehovoth was supported by the European Research Council.
In summary, we provided a systematic comparison of We also would like to thank Natalie Orms and Anna I. Krylov
DFT- and GW -computed eigenvalue spectra to experimen- for useful discussions.
tal photoelectron spectra of four small copper oxide cluster
anions–Cu2 O− , CuO− , CuO−2 , and CuO−3 —with a focus on 1 Theory of Atomic and Molecular Clusters: With a Glimpse at Experiments,

comparison between semi-local and hybrid functional (within edited by J. Jellinek (Springer, Berlin, 1999).
2 L. Kronik and S. Kümmel, Top. Curr. Chem. 347, 137 (2014).
DFT and as a starting point for a G0 W 0 calculation). A main 3 I. Dabo, A. Feretti, and N. Mazrari, First Principles Approaches to Spectro-
theme found in the comparison between theory and experiment scopic Properties of Complex Materials, edited by C. DiValentin, S. Botti,
is that the theoretical success hinges on a number of constraints and M. Cococcioni (Springer-Verlag, 2014), p. 193.
4 C. Faber, P. Boulanger, C. Attaccalite, I. Duchemin, and X. Blase, Philos.
that are difficult to fulfill simultaneously. On the one hand,
Trans. R. Soc., A 372, 20130271 (2014).
hybrid functionals have a clear advantage in mitigating self- 5 P. Rinke, A. Qteish, J. Neugebauer, C. Freysoldt, and M. Scheffler, New J.
interaction errors. On the other hand, removal of some semi- Phys. 7, 126 (2005).
6 M. L. Tiago and J. R. Chelikowsky, Phys. Rev. B 73, 205334 (2006).
local exchange, which is necessary for creating a hybrid func- 7 N. Marom, F. Caruso, X. Ren, O. T. Hofmann, T. Körzdörfer, J. R. Che-
tional, results in a less successful treatment of non-dynamical likowsky, A. Rubio, M. Scheffler, and P. Rinke, Phys. Rev. B 86, 245127
correlation. Generally, we found non-dynamical correlation to (2012).
increase for smaller, more symmetric clusters but to bear some 8 M. J. van Setten, F. Caruso, S. Sharifzadeh, X. Ren, M. Scheffler, F. Liu,

importance also for the larger ones. As a consequence, for J. Lischner, L. Lin, J. R. Deslippe, S. G. Louie, C. Yang, F. Weigend, J.
B. Neaton, F. Evers, and P. Rinke, J. Chem. Theory Comput. 11, 5665
some clusters (notably Cu2 O− ), conventional hybrid function- (2015).
als were very successful, for others (notably CuO−2 ), a semi- 9 L. Hung, F. H. da Jornada, J. Souto-Casares, J. R. Chelikowsky, S. G. Louie,

local hybrid (with some rigid shift) was quite accurate. For and S. Öğüt, Phys. Rev. B 94, 085125 (2016).
10 N. Orms and A. I. Krylov, J. Phys. Chem. A 122, 3653 (2018).
others yet, a hybrid functional with a reduced amount of Fock 11 F. Bruneval, N. Vast, and L. Reining, Phys. Rev. B 74, 045102 (2006).
exchange turned out to be a successful compromise. These 12 L. I. Bendavid and E. A. Carter, First Principles Approaches to Spectro-
trends reflected themselves in the GW calculations as well: scopic Properties of Complex Materials, edited by C. DiValentin, S. Botti,
Whereas Cu-rich clusters, where SIE is a major issue, required and M. Cococcioni (Springer-Verlag, 2014), p. 47.
13 P. A. Cox, Transition Metal Oxides: An Introduction to Their Electronic
more (50%) exact exchange in the hybrid functional starting
Structure and Properties (Oxford University Press, 2010).
point, Cu-poor clusters required less exact exchange (25%). 14 M. L. Polak, M. K. Gilles, J. Ho, and W. Lineberger, J. Phys. Chem. 95,
These results obtained with the one-shot G0 W 0 method suggest 3460 (1991).
15 L.-S. Wang, H. Wu, S. R. Desai, and L. Lou, Phys. Rev. B 53, 8028 (1996).
that it would be interesting future work to examine the effects
16 H. Wu, S. R. Desai, and L.-S. Wang, J. Phys. Chem. A 101, 2103 (1997).
of various levels of GW self-consistency as well as the effects 17 G. V. Chertihin, L. Andrews, and C. W. Bauschlicher, J. Phys. Chem. A
of beyond-GW corrections. Generally, the importance of non- 101, 4026 (1997).
dynamical correlation is expected to decrease with the sys- 18 P. S. Bagus, C. J. Nelin, and C. W. Bauschlicher, J. Chem. Phys. 79, 2975

tem size, explaining the well-documented success of DFT for (1983).


19 G. Igel, U. Wedig, M. Dolg, P. Fuentealba, H. Preuss, H. Stoll, and R. Frey,
larger clusters. However, larger clusters can and very often do
J. Chem. Phys. 81, 2737 (1984).
encounter a significant problem of multiple isomers, which has 20 P. V. Madhavan and M. D. Newton, J. Chem. Phys. 83, 2337 (1985).

also been reflected here in the case of CuO−3 . Taken together, 21 S. R. Langhoff and C. W. Bauschlicher, Chem. Phys. Lett. 124, 241 (1986).
064306-12 Shi et al. J. Chem. Phys. 149, 064306 (2018)

22 G. L. Gutsev, B. K. Rao, and P. Jena, J. Phys. Chem. A 104, 5374 (2000). 52 M. Moseler, B. Huber, H. Häkkinen, U. Landman, G. Wrigge, M. A. Hoff-
23 H. Xian, Z. X. Cao, X. Xu, X. Lu, and Q. E. Zhang, Chem. Phys. Lett. 326, mann, and B. v. Issendorff, Phys. Rev. B 68, 165413 (2003).
485 (2000). 53 D. A. Egger, S. Weissman, S. Refaely-Abramson, S. Sharifzadeh, M. Dauth,
24 B. Dai, K. Deng, J. Yang, and Q. Zhu, J. Chem. Phys. 118, 9608 (2003). R. Baer, S. Kümmel, J. B. Neaton, E. Zojer, and L. Kronik, J. Chem. Theory
25 T. Baruah, R. R. Zope, and M. R. Pederson, Phys. Rev. A 69, 023201 (2004). Comput. 10, 1934 (2014).
26 Y. Pouillon and C. Massobrio, Appl. Surf. Sci. 226, 306 (2004). 54 O. A. Vydrov and G. E. Scuseria, J. Chem. Phys. 125, 234109 (2006).
27 S. Midda, N. C. Bera, and A. K. Das, Int. J. Quantum Chem. 105, 43 (2005). 55 M. A. Rohrdanz, K. M. Martins, and J. M. Herbert, J. Chem. Phys. 130,
28 S. Midda, N. C. Bera, I. Bhattacharyya, and A. K. Das, J. Mol. Struct.: 054112 (2009).
THEOCHEM 761, 17 (2006). 56 L. Kronik, T. Stein, S. Refaely-Abramson, and R. Baer, J. Chem. Theory
29 G.-T. Bae, B. Dellinger, and R. W. Hall, J. Phys. Chem. A 115, 2087 (2011). Comput. 8, 1515 (2012).
30 L. Hung, F. Bruneval, K. Baishya, and S. Öğüt, J. Chem. Theory Comput. 57 F. Bruneval, T. Rangel, S. M. Hamed, M. Shao, C. Yang, and J. B. Neaton,

13, 2135 (2017). Comput. Phys. Commun. 208, 149 (2016).


31 L. Hung, F. Bruneval, K. Baishya, and S. Öğüt, J. Chem. Theory Comput. 58 K. A. Peterson and C. Puzzarini, Theor. Chem. Acc. 114, 283 (2005).

13, 5820 (2017). 59 D. Figgen, G. Rauhut, M. Dolg, and H. Stoll, Chem. Phys. 311, 227 (2005).
32 G. A. Ozin, S. A. Mitchell, and J. Garcia-Prieto, J. Am. Chem. Soc. 105, 60 X. Blase, C. Attaccalite, and V. Olevano, Phys. Rev. B 83, 115103 (2011).

6399 (1983). 61 X. Ren, P. Rinke, V. Blum, J. Wieferink, A. Tkatchenko, A. Sanfilippo,


33 V. E. Bondybey and J. H. English, J. Phys. Chem. 88, 2247 (1984). K. Reuter, and M. Scheffler, New J. Phys. 14, 053020 (2012).
34 H. Wu, S. R. Desai, and L.-S. Wang, J. Chem. Phys. 103, 4363 (1995). 62 N. Marom, O. Hod, G. E. Scuseria, and L. Kronik, J. Chem. Phys. 128,
35 N. Caspary, E. V. Savchenko, A. Thoma, A. Lammers, and V. E. Bondybey, 164107 (2008).
Low Temp. Phys. 26, 744 (2000). 63 N. Marom, X. Ren, J. E. Moussa, J. R. Chelikowsky, and L. Kronik, Phys.
36 R. Mabbs, N. Holtgrewe, D. B. Dao, and J. Lasinski, Phys. Chem. Chem. Rev. B 84, 195143 (2011).
Phys. 16, 497 (2014). 64 A. D. Becke, J. Chem. Phys. 98, 1372 (1993).
37 T.-K. Ha and M. T. Nguyen, J. Phys. Chem. 89, 5569 (1985). 65 T. Körzdörfer and N. Marom, Phys. Rev. B 86, 041110 (2012).
38 Y. Mochizuki, K. Tanaka, and H. Kashiwagi, Chem. Phys. 151, 11 (1991). 66 F. Bruneval and M. A. L. Marques, J. Chem. Theory Comput. 9, 324 (2013).
39 K. Deng, J. Yang, L. Yuan, and Q. Zhu, J. Chem. Phys. 111, 1477 (1999). 67 D. Lüftner, S. Refaely-Abramson, M. Pachler, R. Resel, M. G. Ramsey,
40 G. L. Gutsev, B. K. Rao, and P. Jena, J. Phys. Chem. A 104, 11961 (2000). L. Kronik, and P. Puschnig, Phys. Rev. B 90, 075204 (2014).
41 Y. Pouillon and C. Massobrio, Chem. Phys. Lett. 331, 290 (2000). 68 J. W. Knight, X. Wang, L. Gallandi, O. Dolgounitcheva, X. Ren, J. V. Ortiz,
42 Y. Pouillon, C. Massobrio, and M. Celino, Comput. Mater. Sci. 17, 539 P. Rinke, T. Körzdörfer, and N. Marom, J. Chem. Theory Comput. 12, 615
(2000). (2016).
43 M. Güell, J. M. Luis, L. Rodriguez-Santiago, M. Sodupe, and M. Sola, 69 E. Salomon, P. Amsalem, N. Marom, M. Vondracek, L. Kronik, N. Koch,

J. Phys. Chem. A 113, 1308 (2009). and T. Angot, Phys. Rev. B 87, 075407 (2013).
44 K. Deng, J. Yang, and Q. Zhu, J. Chem. Phys. 113, 7867 (2000). 70 H. Jiang, R. I. Gomez-Abal, P. Rinke, and M. Scheffler, Phys. Rev. B 82,
45 Z. Cao, M. Solà, H. Xian, M. Duran, and Q. Zhang, Int. J. Quantum Chem. 045108 (2010).
81, 162 (2001). 71 J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).
46 B. Dai, L. Tian, and J. Yang, J. Chem. Phys. 120, 2746 (2004). 72 D. K. W. Mok, R. Neumann, and N. C. Handy, J. Phys. Chem. 100, 6225
47 M. Jadraque and M. Martin, Chem. Phys. Lett. 456, 51 (2008). (1996).
48 M. Valiev, E. Bylaska, N. Govind, K. Kowalski, T. Straatsma, H. V. Dam, 73 A. D. Becke, J. Chem. Phys. 104, 1040 (1996).

D. Wang, J. Nieplocha, E. Apra, T. Windus, and W. de Jong, Comput. Phys. 74 O. V. Gritsenko, P. R. T. Schipper, and E. J. Baerends, J. Chem. Phys. 107,

Commun. 181, 1477 (2010). 5007 (1997).


49 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 75 V. Polo, E. Kraka, and D. Cremer, Mol. Phys. 100, 1771 (2002).

(1996). 76 S. Kümmel and L. Kronik, Rev. Mod. Phys. 80, 3 (2008).


50 K. Burke, M. Ernzerhof, and J. P. Perdew, Chem. Phys. Lett. 265, 115 77 M. Reiher, O. Salomon, and B. A. Hess, Theor. Chem. Acc. 107, 48 (2001).

(1997). 78 M. Reiher, Inorg. Chem. 41, 6928 (2002).


51 L. Kronik, R. Fromherz, E. Ko, G. Ganteför, and J. R. Chelikowsky, Nat. 79 G. Prokopiou and L. Kronik, Chem. Eur. J. 24, 5173 (2018).

Mater. 1, 49 (2002). 80 N. Binggeli and J. R. Chelikowsky, Phys. Rev. Lett. 75, 493 (1995).

You might also like