See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/332305028

Bulk vs Layer Confined RNiO 3 : Disentangling Lattice and Electronic

Contributions to the Metal-Insulator Transition

Preprint · April 2019

CITATIONS READS

0 228

5 authors, including:

Alexandru Bogdan Georgescu Ankit S Disa

Northwestern University Max Planck-NYC Center for Non-equilibrium Quantum Phenomena
39 PUBLICATIONS   300 CITATIONS    38 PUBLICATIONS   775 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Metal-Insulator Transition Compounds and Correlated Electrons View project

Quantum Materials View project

All content following this page was uploaded by Alexandru Bogdan Georgescu on 10 April 2019.

The user has requested enhancement of the downloaded file.

Bulk vs Layer Confined RNiO3: Disentangling
Lattice and Electronic Contributions to the
Metal-Insulator Transition
Alexandru B. Georgescu1 , Oleg E. Peil2 , Ankit S. Disa3 , Antoine Georges1,4,5,6 , and Andrew J. Millis1,7
1
Center for Computational Quantum Physics, Flatiron Institute, 162 5th Avenue, New York, NY 10010, USA; 2 Materials Center Leoben, Leoben, Austria; 3 Max Planck Institute
for the Structure and Dynamics of Matter, Hamburg, Germany; 4 Collège de France, 11 place Marcelin Berthelot, 75005 Paris, France; 5 Centre de Physique Théorique Ecole
Polytechnique, CNRS, Universite Paris-Saclay, 91128 Palaiseau, France; 6 Department of Quantum Matter Physics, University of Geneva, 24 Quai Ernest-Ansermet, 1211
Geneva 4, Switzerland; 7 Columbia University, NYC, NY, United States

This manuscript was compiled on April 9, 2019

In complex oxide materials, changes in electronic properties are of-

ten associated with changes in crystal structure, raising the ques-
tion of the relative roles of the electronic and lattice effects in driving
the metal-insulator transition. This paper presents a combined the-
oretical and experimental analysis of the dependence of the metal-
insulator transition of NdNiO3 on crystal structure, specifically com-
paring properties of bulk materials to one and two layer samples of
NdNiO3 grown between multiple electronically inert NdAlO3 counter-
layers in a superlattice. The comparison amplifies and validates a
theoretical approach developed in previous papers and disentangles

FT
the electronic and lattice contributions. In bulk NdNiO3 the correla-
tions are not strong enough to drive a metal-insulator transition by
themselves: a lattice distortion is required. Ultra-thin films exhibit
two additional electronic effects and one lattice-related effect. The
electronic effects are quantum confinement, leading to dimensional
reduction of the electronic Hamiltonian, and an increase in electronic
RA
bandwidth due to counterlayer induced bond angle changes. We find Fig. 1. : Heterostructuring NdNiO3 with NdAlO3 : structural distortions are repre-
that the confinement effect is much more important. The lattice effect sented as motion (not to scale) of oxygen ions (red circles) away from center of Ni-Ni
is an increase in stiffness due to the cost of propagation of the lattice and Ni-Al bonds; note that the distortions do not propagate significantly into the Al
disproportionation into the confining material. layers are are of reduced amplitudes along the Ni-Al bonds. The different colors of
the Ni atoms represent the electronic disproportionation. The kinetic energy of the Ni
transition metal oxide | metal-insulator transition | heterostructure | eg electrons is reduced by confinement, as electrons are not allowed to hop through
the insulating NAO layers (|t0 |«|t‹ |,|tÎ |), while |tÎ | < |t‹ | due to propagation of bond
epitaxial constraint | structural modulation | layer confinement
angles from NdAlO3 .
D

Introduction:. Metal insulator transitions (MIT) in correlated

electron materials typically involve changes in both the elec-
lattice distortion Q, both defined more precisely below. The
tronic and atomic structure. The relative importance of the
two effects has been the subject of extensive discussion(1–8).
In this paper, using a recently developed theoretical approach
(3, 8), we argue that comparison of few-layer and bulk materi- Significance Statement
als yields considerable insight into the relative importance of
Our combined theoretical and experimental study of bulk and
electronic and lattice contributions, essentially because these
heterostructured forms of a correlated electron material leads
are affected by heterostructuring in opposite ways. Motivated
to new insights into the metal-insulator transition. Comparison
by recent experimental (9–25) and theoretical (8, 13, 26–36)
of single layer, bilayer and very thick samples validates a com-
results, we focus here on the rare earth nickelate family of
bined ab-initio/many-body theoretical approach and enables
materials. The concepts, formalism and findings are applicable
a clear disentangling of electronic and lattice contributions to
to wide classes of materials.
the transition. Analysis of the lattice relaxations associated
The rare earth nickelates have chemical formula RNiO3 (R
with the metal-insulator transition highlights the importance of
is a rare earth metal of the lanthanide rare earth series). In
the elastic properties of and propagation of distortions into the
bulk, at high T, they are metallic and form an orthorhombic
electronically inert counterlayer, defining a new control parame-
Pbnm structure (except for R=La for which the structure is
ter for tuning electronic properties. Counterlayer induced bond
rhombohedral) that is a distorted ABO3 cubic perovskite in
angle changes and electronic confinement provide separate
which the Ni ions are equivalent up to a rotation and trans-
tuning parameters, with bond angle changes found to be a
lation. For all R except for La, the bulk materials undergo a
much less effective tuning parameter.
metal to insulator transition (MIT) as T is decreased. The
transition is first order and the low T phase has a P21 /n
structure with two fundamentally inequivalent Ni sites charac-
terized by an electronic charge disproportionation N and a 2
To whom correspondence should be addressed. E-mail: ageorgescuflatirioninstitute.org

www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX PNAS | April 9, 2019 | vol. XXX | no. XX | 1–6

relative roles of the two has been the subject of debate. The
issue has typically been addressed by calculations (typically
performed at fixed crystal structure) and experiments on a
specific material or on members of a family of materials, and
has not been resolved. Similar issues arise in many other
transition metal oxide materials.
Recent experiments (9) report that in NdNiO3 /NdAlO3
(NNO/NAO) superlattices in which one or two monolayers
of NNO are separated by many layers of the wide-gap insu-
lator NAO, the MIT occurs at a much higher temperature
than in the bulk, while the X-ray signatures of the lattice
distortion are much less pronounced in the superlattices than
in bulk. These experiments suggest that heterostructuring
affects electronic and lattice properties differently and thus
that a comparative examination of the two material forms
can help disentangle the relative importance of electronic and
lattice contributions to the metal-insulator transition. In this
paper we theoretically investigate the differences between bulk
NdNiO3 and superlattice NdNiO3 /NdAlO3 materials using
a theoretical approach previously applied to bulk nickelates
(5, 8, 27, 29, 31–33, 37–39) and to ruthenates (3).
In figure 1 we represent the main phenomenology that we
disentangle in this paper, as exemplified on the bilayer NNO.
Namely, the structural distortions in the material become
inhomogenous due to the presence of the NdAlO3 counterlayer
FT
and the absence of a driving force on the interfacial oxygen
from the aluminum atom. In our effective model, this leads to
an increased effective stiffnes of the bond disproportionation
mode as the same force from the nickel atoms leads to a lower
average oxygen displacement. On the electronic structure,
the layer confinement of the material leads to suppressed
RA
hopping along the z direction, while the propagation of bond
angles from NAO to NNO leads to a small increase of the
Ni-Ni hopping in-plane as compared to the out-of-plane Ni-Ni
hopping in the bilayer.
Energy. Central to our discussion is an expression for the en-
ergy difference E between the insulating and metallic phases
D
as a function of lattice distortion Q and charge disproportion-
ation N (3, 8):
kQ2 1
E(Q, N) = ≠ gQ N + Eel ( N ) [1]
2 2
The first term is the elastic energy cost of establishing the
lattice distortion, the middle term is the leading symmetry
allowed coupling between the structural and electronic order
parameters, and the final term is the energy associated with
the electronic transition. The three control parameters are
thus k, g and the combination of interaction parameters and
bandwidths that determines Eel ( N ). A detailed description
of the process by which we fix the control parameters is given
in the Appendix B; here we summarize the findings and give
physical interpretations.
In a previous work on bulk perovskites, the structural stiff-
ness parameter k was found to vary only slightly as the rare
earth ion was changed (8). We find that heterostructuring
has a stronger effect, with k increasing from k=15.86eV/Å2
for bulk NNO, to 17.71eV/Å2 for the bilayer structure and
20.18eV/Å2 for the monolayer. The fundamental difference
between bulk and layered systems appears at the interface be-
tween the two components of the heterostructure. A schematic
of the bond disproportionation mode in the bilayer as obtained
2 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX
from DFT+U structural relaxations is shown in Fig. 2. The
essential point is that the lattice distortion propagates a short
distance into the counter-layer, and the stiffness to this in-
tertwined layer-counterlayer distortion is larger than for the
nickelate material alone.
We represent the bond disproportionation mode and its
propagation in Fig 2 for a particular Q for the bilayer. The
structural disproportionation is inhomogeneous: the in-plane
NiO6 bonds show disproportionation of about the same ampli-
tude we would expect in the bulk. The interfacial bonds are
less disproportionated as the driving force on the apical oxygen
atom comes only from the Ni, the disproportionation is lower.
We can then estimate the relative stiffness of the Al-O bond
relative to the Ni-O bonds from the relative displacements in
a simple elastic spring model, to approximately 88% of the the
stiffness of the Ni-O bonds. However, the additional energy
cost per octahedron due to propagation in the NAO (or, equiv-
alently, the lack of an Al driving force on the Oxygen) leads to
a higher effective stiffness per octahedron. Finally, the out-of-
plane bonds between nickelate layers in the bilayer structure
disproportionate even more than the in-plane bonds, likely to
compensate for the decreased interfacial disproportionation.
The analysis is almost identical for the monolayer, with the
exception of the inter-layer bond which does not exist.
Fig. 2. Left: Schematic of bulk NNO and NAO (ignoring octahedral rotations).
Right: Schematic of the bond disproportionation modes in the NNO bilayer and its
propagation into nearby NAO layers, projected on the Ni-Ni and Ni-Al direction, as
discussed in the main text for a bilayer structure with the average Q=0.078 Å- similar
to the Q of the experimental bulk low T structure.
The parameter g that relates the bond disproportionation
Q to the on-site field difference between the two inequivalent
Ni atoms ( S = gQ) is obtained from its value as obtained
from DFT alone g DF T as described in the supplement, and
adjusted for double counting as g = g DF T (1 + (U ≠ 53 J)‰0 )
with ‰0 = ˆˆ NS
the electronic susceptibility as extracted from
DFT. The value of g DF T is relatively constant between the
bulk and heterostructured materials, and has been previously
shown to be constant throughout the RNiO3 family. However,
as ‰0 is related to the inverse of the bandwidth, the ratio of
the g across the materials changes slightly depending on the
choice of U,J.
Finally, N is obtained from dynamical mean field cal-
culations that require as input values of on-site interaction
Georgescu et al.
parameters U and J as well as band structure information
obtained by Wannier function fits to density functional theory
band structures. In Figure 3 we plot the metal-insulator phase
diagram calculated at Q = 0 for the different materials on the
plane of interaction parameters U and J. , i.e. determined by
Eel ( N ) exclusively. We also plot a dashed line corresponding
to the MIT transition for the experimental bulk low tempera-
ture, insulating structure, calculated as in Ref. (33), to show
the effect of the lattice distortion. We choose parameters U,J
in the range in which the bulk undisproportionated material
is metallic. We note that the resulting J needed to obtain an
insulating state is significantly higher than in a more realistic
calculation involving U,J and an inter-site Hartree term V(32),
however for simplicity and to avoid overfitting multiple param-
eters we work within a U, J model as in previous work (33).
Consistent with previous work we find insulating solutions for
values U,J that are to the right of U-3J<0 (which corresponds
to spontaneous electronic disproportionation in the atomic
limit), insulating solutions, in the absence of any structural
disproportionation. The transitions as a function of U and J
are strongly first order. We find that spontaneous dispropor-
tionation ( N ”= 0) is achieved for a slightly smaller critical J
for a given U in the monolayer and the bilayer than in the bulk.
Despite minor differences in bandwidth, the phase diagrams for
the monolayer and bilayer are nearly identical. An additional
previous work (40), the bond angles from the NAO propagate
into the NNO leading to straighter in-plane bond angles and
slightly higher in-plane bandwidths in the heterostructures
than for bulk NNO. This leads to a x2 -y2 bandwidth of 2.72eV
for the monolayer and 2.68eV for the bilayer. Previous work
has shown that one can use the bond angles of the counterlayers
as a control parameter to tune the metal-insulator transition
temperature in nickelate heterostructures (40–42).
For a lower number of layers as in this work however, the
electronic confinement dominates and leads to an increased
tendency to disproportionate. A third effect of heterostruc-
turing on the electronic structure is that of the crystal-field
splitting induced by the inequivalence of the bonds and the
relative ionicity of the material. Finally, within the eg Wannier
picture, the monolayer also shows a crystal field splitting of
‘¯x2 ≠y2 ≠ ‘¯3z2 ≠r2 = 0.14eV in DFT. We’ve performed calcu-
lations for the monolayer with the crystal-field splitting set
to 0 for U=2.1eV and found that the critical J for the MIT
transition line is the same as with the crystal-field splitting
set to the DFT relaxed value, within an accuracy of J=0.01eV,
thus showing a negligible effect.

‰bulk
0 0 = 1.25/eV , ‰monolayer
0
FT
way to estimate the effect of the change in bandwidth is by
comparing the static electronic response to an on-site field in
our eg tight-binding model, ‰0 as defined previously. We get
= 1.16/eV , ‰bilayer = 1.39/eV .
RA
D

Fig. 3. Solid lines: boundaries of region of spontaneous electronic disproportionation

in the (U,J) plane for DFT-relaxed bulk, bilayer and monolayer NNO in the absence of
structural distortion. Insulating phases are to the right of the lines shown. The dotted
line to the left shows the Metal-Insulator transition line for the bulk experimental Low
T disproportionated structure calculated as in Ref (33). The stars correspond to the
points at which the constructions in Fig. 5 are plotted.

The dominant effect of the layering in the case of the
heterostructures is electronic confinement: electron hopping
is confined to be in-plane only for the NNO monolayer, and
confined between the two layers for the bilayer. While the bulk
orbitals have a bandwidth of 2.6eV, the 3z2 -r2 orbital for the
monolayer has a bandwidth of 1.85eV and for the bilayer 2.15eV.
Two other, more minor effects appear as well. Similar to
Fig. 4. Projected density of states of low-energy eg Wannier bands for GGA-relaxed
structures for bulk (top), bilayer (middle) and monolayer (bottom) structures. Dotted
horizontal lines show approximate cutoff for determining bandwidths mentioned in
main text. Confinement greatly reduces the bandwidth of the 3z2 -r2 orbital, however
the bond angle propagation leads to a slightly wider bandwidth of the x2 -y2 orbital.
We have then found three main effects of heterostructuring
on the electronic structure. The effect of layer confinement

Georgescu et al. PNAS | April 9, 2019 | vol. XXX | no. XX | 3

strongly lowers the kinetic energy of the electrons and favors an
insulating state, with its effect primarily on the 3z2 -r2 orbital.
The bond angle propagation leads to a small effect in the
opposite direction, primarily on the x2 -y2 band. Finally, the
crystal field splitting is only significant in the monolayer,
however it does not affect the electronic transition.

Equilibrium bond and electronic disproportionation from to-

tal energy model. We now turn to determining equlibrium
points in the energy functional from equation 1. Stationarity
of E with respect to variations in N and Q implies the
two equilibrium conditions:
1
0 = kQ ≠ g N [2]
2
and:
1 ˆEel ( N )
0 = ≠ gQ + [3]
2 ˆ N
Equation 2 gives Q as a function of N and requires determin-
ing k and g which we obtain from density functional theory
(DFT) calculations as described in the Methods section and
the Appendix. However, its meaning is very simple: for a
particular value of the electronic disproportionation N one
can obtain the equilibirum disproportionation of the oxygen
atoms as a result of the resulting electrostatic forces. Equation

2k
Q=
FT
3 gives N as a function of Q, as obtained via the density
functional plus dynamical mean field theory (DFT+DMFT)
method.
Combining the two, we have an equation of state:

N [Q] [4] Fig. 5. N versus bond disproportionation Q within DFT+DMFT for bulk, bilayer
RA
g and monolayer structures as well as Q versus N lines from the total energy model
calculation in different areas of the phase space as determined by equation 2 and
The N (Q) relation is shown in Figure 5 as large symbols equation 3 via DFT+DMFT as described in the main text. The thick circles mark the
connected by lines for U = 2.1eV and three J values. For intersections that respect the equation of state 4. The panels marked with A,B,C
the smallest J value neither the bulk nor the superlattice correspond to the points marked with stars in the phase diagram in Fig. 3
materials show a spontaneous disproportionation at Q = 0;
for small Q there is a regime in which the disproportionation
is linear in Q and the solution remains metallic. Above a is nearly constant in the insulating phase N ¥ Ninsulating , Q
is then set by optimizing the structure Q[ N ] in equation [4] as
D

particular Q there is a very rapid crossover to an insulating

solution with a N which is large and only weakly dependent
on Q. In the insulating regime the monolayer has a larger
disproportionation than the bulk, with the bilayer in between.
For an intermediate J, the monolayer and bilayer exhibit a
spontaneous disproportionation at Q = 0 but the bulk material
exhibits a Q-driven first order transition. At the larger J all
three systems spontaneously distort at Q = 0.
Also shown in Figure 5 are straight lines corresponding
to the Q( N ) relation from equation 2. The intersection
of these lines with the DMFT N (Q) curves defines the
actual values of N and Q. We see from the relative positions
of the intersections that Qmonolayer < Qbilayer < Qbulk and
Nmonolayer > Nbilayer > Nbulk . From an electronic point
of view the monolayer and bilayer are more disproportionated
( N is larger) as N does not depend strongly on Q, however
the higher stiffness of the heterostructures leads to a lower Q.
Further, as shown in the middle figure in 2, there is a range
of U,J for which the heterostructures will stay insulating even
at a very small Q while the bulk becomes metallic.
The relative roles of the lattice and electronic structure
are easily disentangled from the above: while the electronic
disproportionation has a first order transition, presenting a
very slowly varying N in the insulating phase. Assuming N
approximately Q ¥ . This allows the seemingly
g Ninsulating
2k
paradoxical solutions with the amplitude of Q and N showing
opposite trends between the bulk and heterostructure. If
we assume that the experimentally obtained metal-insulator
transition temperature is more strongly correlated to N than
Q, while the XAS spectra splitting is more strongly correlated
with Q, we can thus explain the seemingly paradoxical results
in previous work (9).
One of the signatures associated with the bond dispropor-
tionated phase of the RNO nickelates is an increased peak-
prepeak splitting of the XAS Ni L3 edge, which in the mono-
layer and bilayer were found to be in between the values of the
bulk disproportionated and undisproportionated structures
throughout the insulating temperature range scanned. Con-
sistent with this result, we find that the predicted value of
the structural disproportionation of the monolayer is lower
than the bulk, with the bilayer in between the two. Further,
XAS integration of the monolayer in-plane and out-of-plane
Ni L3 edge has found an orbital polarization of 8% favoring
the 3z2 -r2 orbital. Within our insulating solutions we find that
orbital polarization is strongly suppressed ( <2%) however
we consistently find that the long bond site has an orbital
polarization of 5-8% in a direction consistent with experiment,

4 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX Georgescu et al.

while the LB site is orbitally polarized of about the same mag-
nitude but in the opposite direction. This suggests that the
XAS spectra may sample primarily the LB, however further
theoretical and experimental work is needed.
Conclusion and Outlook. Using a combination of
DFT+DMFT and many-body theory we have eluci-
dated the relative importance of lattice and electronic effects
in heterostructured materials. We have found that the higher
lattice energy cost in the heterostructured materials decreases
the structural signatures of the symmetry-broken phase within
the correlated material going through an MIT but that the
distortion associated with it can propagate into the epitaxial
layer. We have found that, as the effect of interactions is
increased in a layer-confined structure, electronic dispropor-
tionation can be higher despite lower structural distortions in
a heterostructure. Through comparison with experiment(9),
our study suggests that the electronic disproportionation is
more likely to be correlated to the metal insulator transition
temperature than the structural disproportionation, which is
suppressed by the higher structural stiffness of the material.
At the same time, our work suggests that the structural
disproportionation is more strongly connected to the XAS
splitting observed experimentally, likely via the induced
on-site electrostatic potential difference.
FT
These general results can be used both to understand other
similar heterostructures (for example LaNiO3 /LaAlO3 ) as well
as to design new materials. Our analysis of the bond dispro-
portionation mode on the interfacial structure in this class
of materials as well as in related classes of materials (vana-
dates, manganites etc) can be studied both theoretically and
RA
experimentally. The combination of bond angles, confinement
and relative structural stiffness can be used to fine tune metal-
insulator transition temperatures. Based on the methodology
in this work and previous work (3, 8), future work involving
DFT+DMFT, DFT+U studies and model calculations, can
address the relative roles of lattice and electronic dispropor-
tionation.
D
Methods. For our calculations, we use structures obtained from
fully relaxed DFT+U calculations (43) and impose 0% strain
relative to the theoretical bulk NNO lattice constant on the
heterostructures. We use Quantum Espresso, ultrasoft pseu-
dopotentials, either from the GBRV or generated using the
Vanderbilt ultrasoft pseudopotential generator as described in
previous work (43–47) and benchmark our results against ex-
perimental bulk structures. The disproportionated structures
have two inequivalent Ni sites, one with relatively long Ni-O
bonds (’LB’) and one with relatively short Ni-O bonds (’SB’).
We define the structural order Q as:
Úq
(i) (i)
(lLB ≠ lSB )2
Q= i
[5]
6 of the European Research Council (ERC-319286-QMAC). OEP
where l(i) are the lengths of the Ni-O bonds.
For each structure we then perform a self consistent DFT
calculation and fit the bands arising from the frontier eg orbitals
using maximally localized Wannier functions as implemented
in Wannier90 (48, 49). Representative structures are shown
in Figure 1 while their corresponding bands near the Fermi
level and Wannier fits are shown in Figure 4. We define N
as the difference in occupancy between the Wannier orbitals
Georgescu et al.
on the two inequivalent Ni atoms:
N = NHF ≠ NLF [6]
with HF = Higher Filling and LF=Lower Filling. When there
is structural disproportionation, HF corresponds to LB and
LF corresponds to SB.
The parameter g in equation 2 is defined in terms of
the on-site energy difference S = gQ entering our DMFT
calculations. In our one-shot DMFT, g is corrected from
the DFT value by a double counting term (8, 32), so g =
g DF T (1 + (U ≠ 53 J)‰0 ).
The Wannier fits define a low energy tight binding model to
which we add standard Slater-Kanamori interactions and solve
using dynamical mean field theory (using the triqs library(50),
ct-hyb solver(51) and dfttools (52) interface) with the two
inequivalent Ni treated as different embedded atoms. Note
that in the DMFT theory, the calculated lattice distortion
enters as a site-dependent energy difference gQ.
The parameter k is the stiffness to lattice distortions at
fixed N . We argue, following (3, 8) that since the stiffness
comes from the full electronic structure at fixed N , the fron-
tier orbitals play a relatively minor role and for the purpose
of calculating k may be treated at the DFT level. We there-
fore obtain k from the dependence of the DFT energy on Q
ˆEDF T
ˆQ
= cQ. However in the DFT calculations N is relaxed
at each Q . Referring to equation 2 we have on the DFT level
(and noting the stationarity with respect to N ):
cQ =
ˆEDF T
ˆQ
1
= kQ ≠ g DF T
2
N (Q) [7]
The parameter g DF T is defined in terms of the average on
site energy DF
S
T
obtained from our Wannier fits to DFT band
DF T
structures as g DF T = SQ and N (Q) is obtained from the
occupancy difference of the Wannier orbitals and is found to be
linear in Q, N = ‰0 g DF T Q. This relation defines the on-site
susceptibility ‰0 . g DF T can be read off from the on-site energy
difference and is nearly identical for all three materials, namely
2.89ev/Å for bulk and 2.972ev/Å for the bilayer and 2.962ev/Å
for the monolayer. This means that a similar movement of
the ions leads to a similar change of electrostatic potential,
which is something we would expect as the local environment
is similar. The static electronic response however is different,
with ‰0 = 1.16eV ≠1 for bulk, ‰0 = 1.25eV ≠1 for bilayer and
‰0 = 1.39eV ≠1 for the monolayer, with a larger response for
the materials with lower bandwidth.
ACKNOWLEDGMENTS. We thank Sohrab Ismail-Beigi, Jean-
Marc Triscone, Hugo U.R. Strand, Manuel Zingl, Alexander Hampel
and Claude Ederer for helpful conversations and Nick Carriero and
the Scientific Computing Core division of the Flatiron Institute
for invaluable technical assistance. The Flatiron Institute is a
division of the Simons Foundation. AG acknowledges the support
acknowledges funding from Forschungsförderungsgesellschaft (FFG)
COMET program IC-MPPE (project No 859480)
References.
1. Park H, Millis A, Marianetti C (2014) Total energy calculations using DFT+ DMFT: computing
the pressure phase diagram of the rare earth nickelates. Physical Review B 89:245133.
2. Park H, Millis AJ, Marianetti CA (2014) Computing total energies in complex materials using
charge self-consistent DFT+DMFT. Physical Review B - Condensed Matter and Materials
Physics 90(23):1–17.
3. Han Q, Millis A (2018) Lattice Energetics and Correlation-Driven Metal-Insulator Transitions:
The Case of Ca2RuO4. Physical Review Letters 121(6):67601.
PNAS | April 9, 2019 | vol. XXX | no. XX | 5
 4. Haule K, Pascut GL (2016) Forces for structural optimizations in correlated materials within a
DFT+embedded DMFT functional approach. Physical Review B 94(19):1–16.
5. Haule K, Pascut GL (2017) Mott transition and magnetism in rare earth nickelates and its
fingerprint on the x-ray scattering. Nature Scientific Reports (7):10375.
6. Mandal S, Zhang P, Ismail-Beigi S, Haule K (2017) How Correlated is the FeSe/SrTiO3 Sys-
tem? Physical Review Letters 119(6):1–6.
7. Amadon BA (2012) A self-consistent DFT+DMFT scheme in the projector augmented wave
method: Applications to cerium, Ce2O3and Pu2O3with the Hubbard i solver and comparison
to DFT+U. Journal of Physics Condensed Matter 24(7).
8. Peil OE, Hampel A, Ederer C, Georges A (2018) Mechanism and Control Parameters of the
Metal-Insulator Transition in Nickelates. (arXiv: 1809.03720).
9. Disa AS, et al. (2017) Control of hidden ground-state order in NdNiO3 Superlattices. Physical
Review Materials 1(2):024410.
10. Catalano S, et al. (2018) Rare-earth nickelates RNiO3: Thin films and heterostructures. Re-
ports on Progress in Physics 81(4):046501.
11. Ismail-Beigi S, Walker FJ, Disa AS, Rabe KM, Ahn CH (2017) Picoscale materials engineer-
ing.
12. Boris AV, et al. (2011) Dimensionality Control of Electronic Phase Transitions in Nickel-Oxide
Superlattices. Science Reports 332(May):937–940.
13. Kumah DP, et al. (2014) Tuning the structure of nickelates to achieve two-dimensional electron
conduction. Advanced Materials 26(12):1935–1940.
14. Middey S, et al. (2018) Disentangled Cooperative Orderings in Artificial Rare-Earth Nicke-
lates. Physical Review Letters 120(15):156801.
15. Gray AX, et al. (2011) Insulating state of ultrathin epitaxial LaNiO3 thin films detected by
hard x-ray photoemission. Physical Review B - Condensed Matter and Materials Physics
84(7):1–6.
16. Stemmer S, Millis AJ (2013) Quantum confinement in oxide quantum wells. MRS Bulletin
38(12):1032–1039.
17. Phillips PJ, et al. (2017) Experimental verification of orbital engineering at the atomic scale:
Charge transfer and symmetry breaking in nickelate heterostructures. Physical Review B
95(20):205131.
18. Zhang JY, Kim H, Mikheev E, Hauser AJ, Stemmer S (2016) Key role of lattice symmetry in
the metal-insulator transition of NdNiO3films. Scientific Reports 6(November 2015):23652.
the metal-insulator transition. Physical Review Letters 112(10):106404.
40. Liao Z, et al. (2018) Metal-insulator-transition engineering by modulation tilt-control in per-
ovskite nickelates for room temperature optical switching. Proceedings of the National
Academy of Sciences of the United States of America (September):201807457.
41. Chang SH, et al. (2011) Thickness-dependent structural phase transition of strained Sr-
RuO3ultrathin films: The role of octahedral tilt. Physical Review B - Condensed Matter and
Materials Physics 84(10):104101.
42. Gao R, et al. (2016) Interfacial Octahedral Rotation Mismatch Control of the Symmetry and
Properties of SrRuO3. ACS Applied Materials and Interfaces 8(23):14871–14878.
43. Anisimov VI, Aryasetiawan F, Liechtenstein A (1997) First-principles calculations of the elec-
tronic structure and spectra of strongly correlated systems: the LDA + U method. Journal of
Physics: Condensed Matter 9:767–808.
44. Giannozzi P, et al. (2009) Quantum ESPRESSO: a modular and open-source software project
for quantum simulations of materials. Journal of Physics: Condensed Matter 21:395502.
45. Vanderbilt D (1990) Soft Self-Consistent Pseudopotentials in a generalized eigenvalue for-
malism. Physical review B 41(11):7892–7895.
46. Garrity KF, Bennett JW, Rabe KM, Vanderbilt D (2014) Pseudopotentials for high-throughput
DFT calculations. Computational Materials Science 81:446–452.
47. Chen H, et al. (2014) Reversible Modulation of Orbital Occupations via an Interface- Induced
Polar State in Metallic Manganites. Nano. Lett 14(9):4965–4970.
48. Mostofi AA, et al. (2014) An updated version of wannier90: A tool for obtaining maximally-
localised Wannier functions. Computer Physics Communications 185(8):2309–2310.
49. Marzari N, Mostofi AA, Yates JR, Souza I, Vanderbilt D (2012) Maximally localized Wannier
functions: Theory and applications. Reviews of Modern Physics 84(4):1419–1475.
50. Parcollet O, et al. (2015) TRIQS: A toolbox for research on interacting quantum systems.
Computer Physics Communications 196(196):398–415.
51. Seth P, Krivenko I, Ferrero M, Parcollet O (2016) TRIQS/CTHYB: A continuous-time quan-
tum Monte Carlo hybridisation expansion solver for quantum impurity problems. Computer
Physics Communications 200:274–284.
52. Aichhorn M, et al. (2016) TRIQS/DFTTools: A TRIQS application for ab initio calculations of
correlated materials. Computer Physics Communications 204:200–208.

FT
19. Shamblin J, et al. (2018) Experimental evidence for bipolaron condensation as a mechanism
for the metal-insulator transition in rare-earth nickelates. Nature Communications 9:86.
20. Meyers D, et al. (2016) Pure electronic metal-insulator transition at the interface of complex
oxides. Scientific Reports 6:27934.
21. Först M, et al. (2017) Multiple Supersonic Phase Fronts Launched at a Complex-Oxide Het-
erointerface. Physical Review Letters 118(2):027401.
22. Forst M, et al. (2015) Spatially resolved ultrafast magnetic dynamics initiated at a complex
oxide heterointerface. Nature Materials 14(9):883–888.
23. Medarde M, Lacorre P, Conder K, Fauth F, Furrer A (1998) Giant ˆ{16}O-ˆ{18}O Isotope Effect
RA
on the Metal-Insulator Transition of RNiO_{3} Perovskites ( R=Rare Earth). Physical Review
Letters 80:2397–2400.
24. Hu W, Catalano S, Gibert M, Triscone JM, Cavalleri A (2016) Broadband terahertz spec-
troscopy of the insulator-metal transition driven by coherent lattice deformation at the SmNi
O3/LaAl O3 interface. Physical Review B 93(16):161107(R).
25. Caviglia AD, et al. (2013) Photoinduced melting of magnetic order in the correlated elec-
tron insulator NdNiO3. Physical Review B - Condensed Matter and Materials Physics
88(22):220401(R).
26. Guzman-Verri G, Brierley R, Littlewood P (2017) Elastic interactions and control of the Mott
transition. arXiv:1701.02318.
D

27. Mandal B, et al. (2017) The driving force for charge ordering in rare earth nickelates. arxiV:
1701.06819.
28. Park H, Millis AJ, Marianetti CA (2016) Influence of quantum confinement and strain on orbital
polarization of four-layer LaNiO3 superlattices: A DFT+DMFT study. Physical Review B
93(23):1–10.
29. Ruppen J, et al. (2015) Optical spectroscopy and the nature of the insulating state of
rare-earth nickelates. Physical Review B - Condensed Matter and Materials Physics
92(15):155145.
30. Peil OE, Ferrero M, Georges A (2014) Orbital polarization in strained LaNiO3: Structural
distortions and correlation effects. Physical Review B - Condensed Matter and Materials
Physics 90(4):045128.
31. Han MJ, Wang X, Marianetti CA, Millis AJ (2011) Dynamical mean-field theory of nickelate
superlattices. Physical Review Letters 107(20):206804.
32. Seth P, et al. (2017) Renormalization of effective interactions in a negative charge transfer
insulator. Physical Review B 96(20):205139.
33. Subedi A, Peil OE, Georges A (2015) Low-energy description of the metal-insulator transition
in the rare-earth nickelates. Physical Review B - Condensed Matter and Materials Physics
91(7):1–16.
34. Strand HUR (2014) Valence-skipping and negative- U in the d -band from repulsive lo-
cal Coulomb interaction. Physical Review B - Condensed Matter and Materials Physics
90(15):155108.
35. Blanca-Romero A, Pentcheva R (2011) Confinement-induced metal-to-insulator transition in
strained LaNiO 3/LaAlO3 superlattices. Physical Review B - Condensed Matter and Materials
Physics 84(19):195450.
36. Janson O, Held K (2018) inite-temperature phase diagram of ( 111 ) nickelate bilayers. Phys-
ical review B 98:115118.
37. Park H, Millis AJ, Marianetti Ca (2012) Site-selective Mott transition in rare-earth-element
nickelates. Physical Review Letters 109(15):156402.
38. Mercy A, Bieder J, Iniguez J, Ghosez P (2017) Structurally triggered metal-insulator transition
in rare-earth nickelates. Nature Communications 8:1677.
39. Johnston S, Mukherjee A, Elfimov I, Berciu M, Sawatzky GA (2014) Charge disproportiona-
tion without charge transfer in the rare-earth-element nickelates as a possible mechanism for

6 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX Georgescu et al.

1

2 Supplementary Information for

3 Bulk vs Layer Confined RNiO3 : Disentangling Lattice and Electronic Contributions to the
4 Metal-Insulator Transition

5 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

6 Alexandru B. Georgescu.
7 E-mail: ageorgescu@flatironinstitute.org

8 This PDF file includes:

9 Supplementary text
10 Figs. S1 to S5
11 References for SI reference citations

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 1 of 11

12 Supporting Information Text
13 Appendix A - Technical Details
14 DFT. We use the Quantum Espresso (1) software package for our DFT calculations, Wannier90(2) to generate the extended
15 Wannier eg orbitals, and the TRIQS software package(3) for the DMFT calculations with its various applications: DFTTools
16 (4) for the interface with Wannier90 and the CT-HYB DMFT solver(5) to solve the DMFT equations using the Hybridization
17 expansion impurity solver (6). We use ultrasoft pseudopotentials (7) and the PBE (8) version of the GGA exchange-correlation
18 functional formalism. To simulate the bulk material, we used a 20 atom c(2x2)x2 unit cell, relaxed starting from the experimental
19 structure; for the m=1 superlattice we used 1 layer of NNO and 3 layers of NAO, while for m=2 we used 2 layers of each material,
20 making sure that the total number of layers is even in order to allow the octahedra to relax. For the bulk undisproportionated
21 calculation, we find that our high-temperature relaxed structure within GGA has a bandwidth that is only slightly smaller
22 than the experimental high temperature structure. In the phase diagram in figure 2, this leads to a small shift to the left in the
23 critical J for a given U of at most 0.05eV. Of the available functionals we have tried however (LDA, PBE and PBEsol), we find
24 that relaxations using GGA in the PBE formalism provide a bandwidth that is very close to the one using the experimental
25 high temperature structure as well as a phase diagram that is closer to the one obtained by using said structure. We also
26 find that the phase diagram difference between the undisproportionated structures is robust with respect to changing the
27 functional used in order to obtain the relaxed structures. For the initial DFT relaxations without structural disproportionation,
28 we found a k-point mesh of 5x5x4 for the 20 atom bulk unit cell or 5x5x4 for the heterostructures (equivalent to 7x7x8 and
29 7x7x12) per formula unit are sufficient to generate quality structures. For consistency with the DFT+U results, we also
30 performed higher k-point calculations but found they do not affect results. For the various DFT+U relaxations involving
31 structural disproportionation, we impose ferromagnetic order. The k-point mesh has to be higher to differentiate between the
32 various possible local minima. Thus, for the 20 atom bulk is 7x7x5, while for the 40 atom heterostructures it is 7x7x4 (roughly
33 equivalent to a 10x10x10 k-mesh per formula unit) in order to have a 1meV accuracy in the total energies of the relaxed
34 structures, however we have found that structures obtained using 5x5x2 and 7x7x3 k-point grids give nearly identical structural
35 results. For all calculations, the wavefunction energy cutoff is 35 Ry, and the density cutoff is 280Ry. A Marzari-Vanderbilt
36 smearing of 0.027 eV is used for the electron occupation function for all calculations. The energy convergence threshold for
37 relaxations is 10≠9 eV and the force covergence threshold is 10≠4 a.u., the electronic energy convergence threshold within the
38 electronic self-consistency loop is 10≠12 eV.
39 To extract the tight-binding parameters for DMFT calculations, we obtain the antibonding eg orbitals as a postprocessing
40 step using Wannier90 within the Maximally Localized Wannier Function framework.

41 DFT+U. For the DFT+U calculations, we performed calculations in the ferromagnetic phase, as our unit cells are not large
42 enough to reproduce the full experimentally determined magnetic structure in the bulk (with vector q=(1/4,1/4,1/4)) or bilayer
43 (q=(1/4,1/4,0) while the magnetic structure in the monolayer is not known but with ferromagnetism and a q=(1/4,1/4,0) vector
44 excluded. We use the FM state as a way to break symmetry leading to bond disproportionation and an alternating S=1, S=0
45 state, as more important in the physics of the final structure than the exact magnetic order. A Hund J is included implicitly
46 within spin-polarized DFT, however within the version of Quantum Espresso used for these calculations, only the effective
47 Dudarev Uef f = (U ≠ 2J) scheme is available, not allowing us to control U, J separately. For the two inequivalent Ni sites, as a
48 proxy for charge disproportionation, we use the difference in local magnetic moment. We find that the total magnetization
49 for the supercell is always equal to one Bohr magneton times the number of Nickel sites, as we would expect. However, the
50 projected magnetic moment on the Löwdin orbitals does not add up to this total magnetization, as the Löwdin orbitals are not
51 sufficient to describe the eg valence states. However, making the assumption that the Löwdin magnetic moment is directly
52 proportional to the magnetic moment per site, we can then write mN i ¥ mLowdin≠total ú 2µB/mtotal where mLowdin≠total is
53 the total magnetization per pair of Ni sites: this is the M plotted in Appendix C.

DFT+DMFT. The Slater-Kanamori Hamiltonian solved by DMFT using the tight binding model extracted from the Wannier
functions has the form on each lattice site:
ÿ ÿ ÿ ÿ
ĤU = U n̂mø n̂m¿ + (U ≠ 2J) n̂mø n̂mÕ ¿ + (U ≠ 3J) n̂m‡ n̂mÕ ‡ ≠ J c†mø cm¿ c†mÕ ¿ cmÕ ø
m m”=mÕ m<mÕ ,‡ m”=mÕ
ÿ
+J c† mø c†m¿ cmÕ ¿ cmÕ ø [1]
m”=mÕ

54 In the DFTTools interface between DFT and DMFT we use a k-mesh of 14x14x14 for the bulk material, and a k-mesh of
55 12x12x6 for the heterostructure to obtain quality local density of states. The DMFT calculations in this paper are performed
56 in a single-shot formalism, i.e. the new density is not updated into the DFT loop. Most DMFT calculations in this paper,
57 unless mentioned otherwise, are done at an inverse temperature — = 40eV ≠1 , equivalent to a temperature of 290K. Most
58 DMFT calculations are run for 15 steps using 107 QMC cycles. Calculations that require higher precision are ran for 35
59 steps, the first 30 using around 107 QMC cycles, and the final 5 using 3 ◊ 108 cycles. To compare with previous literature,
60 we’ve also performed density-density only calculations to benchmark our calculations. We find that using the Slater-Kanamori
61 Hamiltonian shifts the metal-insulator phase boundary to slightly higher values of J given a particular U (¥ 0.2eV ).

2 of 11 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

62 Appendix B - Model Parameter Extraction. In order to extract the relevant parameters we need, we use the structures generated
63 within DFT+U. We need to extract the following: g DF T = ˆˆQS , ‰0 = ˆˆ N S
, ⁄ = g DF T ‰0 and finally the stiffness k, which can
64 then be extracted by doing DFT calculations on said structures. We then get: EDF T (Q) = ( k2 ≠ 14 (g DF T )2 ‰0 )Q2 = 2c Q2 with c,
65 g, ‰0 constants we extract from the structures we generated as a function of Q. In order to benchmark this method as compared
66 to explicitly imposing a Q, we compared our results on the bulk material with previous work (9) in which multiple structures
67 were generated by modulating the bond disproportionation Q alone, and find that our results are consistent. We also note
68 that while using bulk structures obtained through GGA relaxation does lead to slightly lower bandwidths than those obtained
69 using the experimental structure, this only affects the results by shifting all the phase diagrams to lower J for a given U by
DF T
about 0.05eV as compared to the experimental structure. As the stiffness k and the static response functions g DF T =
ˆ S
70
ˆQ
71 and ‰0 = ˆ N DF T
ˆ DF T do not depend on the bandwidth by more than a fraction of a percent within our calculations, and the
S
72 bond angles of the disproportionated structures remains the same up to within < 1 degree within our relaxed spin-polarized
73 calculations, we can simulate a large number of disproportionated structures with varying Q and assuming that the DFT
74 Hamiltonian would look like the one corresponding to the undisproportionated, nonmagnetic structure plus an on-site energy
75 difference DFS
T
= g DF T Q. The spin-polarized relaxed structures have bond angles that are slightly more bent than those in
76 the non-magnetic calculations, however as was shown in (9), neither k nor g depend on the bandwidth; we do expect a slightly
77 larger ‰0 in the structures obtained using spin-polarized DFT+U by about 3-5% due to the slightly reduced bandwidth.

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 3 of 11

Fig. S1. Plot of energy per 2 formula units of NNO for the different structures for different Q, using structures obtained from DFT+U calculations. The energy and Q2 are
plotted relative to the energy of the smallest Q that are included in the plot.

4 of 11 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

 Fig. S2. Fit to obtain the static electronic response for the three materials, ‰0

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 5 of 11

 Fig. S3. Fit to obtain the linear coupling coefficient within DFT, gDF T

6 of 11 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

78 Appendix C - DFT+U Bulk and Heterostructure Results. Further supporting the picture of stronger electronic disproportionation
79 in the heterostructures as a driving mechanism, we present results for the bulk, bilayer and monolayer structures as obtained
80 by DFT+U using spin-polarized calculations in the FM state and full relaxations in Figure S4, with further details at the
81 end of Appendix C. We find that, for small U, the heterostructures have a critical U under which the lattice stiffness is too
82 strong to allow disproportionation. As expected, the critical U at which the monolayer disproportionates is higher than for the
83 bilayer. Consistent with the DFT+DMFT results, however, once electronic disproportionation is allowed the monolayer is more
84 electronically disproportionated than the bulk, with a similar result for the bilayer with the exception of U=1eV. Q are very
85 similar for all three structures, with the exception of large U for which both the bulk and bilayer structural disproportionation
86 stop increasing, while the monolayer disproportionation increases past that of both.

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 7 of 11

87 Appendix D - Experimental Hole Determination. Experimental values of the Ni eg hole ratio were extracted from x-ray linear
88 dichroism (XLD) measurements on NNO/NAO superlattices, which can be directly compared to theory. The sample growth
89 and measurement procedures are described in Ref. (10). To extract r experimentally, the XLD sum rule (11) is applied for the
3I3z 2 ≠r2
90 Ni L edge, r= 4I ≠I3z 2 ≠r2
, where Ip corresponds to the integrated absorption measured with the polarization parallel to p.
x2 ≠y 2
91 Figure 1 shows the absorption for in-plane and out-of-plane polarizations as well as the difference spectrum of the monolayer
92 NNO/NAO superlattice. The values of r for the monolayer NNO/NAO superlattice, bilayer NNO/NAO superlattice, and bulk
93 NNO samples are 0.91 ± 0.03, 1.00 ± 0.02, and 1.02 ± 0.02, respectively.

8 of 11 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

 Fig. S4. Electronic and structural disproportionation within DFT+U for the ground state relaxed structures for the bulk, monolayer, and bilayer.

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 9 of 11

Fig. S5. (top) X-ray absorption spectra and (bottom) x-ray linear dichroism of monolayer NNO/NAO superlattice. Blue and pink curves in top panel correspond to absorption
with polarization in-plane and out-of-plane, respectively.

10 of 11 Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis

94 References
95 1. Giannozzi P, et al. (2009) Quantum ESPRESSO: a modular and open-source software project for quantum simulations of
96 materials. Journal of Physics: Condensed Matter 21:395502.
97 2. Mostofi AA, et al. (2014) An updated version of wannier90: A tool for obtaining maximally-localised Wannier functions.
98 Computer Physics Communications 185(8):2309–2310.
99 3. Parcollet O, et al. (2015) TRIQS: A toolbox for research on interacting quantum systems. Computer Physics Communica-
100 tions 196(196):398–415.
101 4. Aichhorn M, et al. (2016) TRIQS/DFTTools: A TRIQS application for ab initio calculations of correlated materials.
102 Computer Physics Communications 204:200–208.
103 5. Seth P, Krivenko I, Ferrero M, Parcollet O (2016) TRIQS/CTHYB: A continuous-time quantum Monte Carlo hybridisation
104 expansion solver for quantum impurity problems. Computer Physics Communications 200:274–284.
105 6. Werner P, Millis AJ (2006) Hybridization expansion impurity solver: General formulation and application to Kondo lattice
106 and two-orbital models. Physical Review B - Condensed Matter and Materials Physics 74(15):155107.
107 7. Vanderbilt D (1990) Soft Self-Consistent Pseudopotentials in a generalized eigenvalue formalism. Physical review B
108 41(11):7892–7895.
109 8. Perdew JP, Burke K, Ernzerhof M (1996) Generalized gradient approximation made simple. Physical Review Letters
110 77(18):3865–3868.
111 9. Peil OE, Hampel A, Ederer C, Georges A (2018) Mechanism and Control Parameters of the Metal-Insulator Transition in
112 Nickelates. (arXiv: 1809.03720).
113 10. Disa AS, et al. (2017) Control of hidden ground-state order in NdNiO3 Superlattices. Physical Review Materials 1:024410.
114 11. van der Laan G (1994) Sum Rules and Fundamental Spectra of Magnetic X-Ray Dichroism in Crystal Field Symmetry.
115 Journal of the Physical Society of Japan 63:2393.

Alexandru B. Georgescu, Oleg E. Peil, Ankit S. Disa, Antoine Georges,Andrew J. Millis 11 of 11

View publication stats