Stavropoulos Panagiotis Peter PHD Thesis

Emergent phenomena in correlated materials with strong spin-orbit
coupling
by
Panagiotis Peter Stavropoulos
A thesis submitted in conformity with the requirements

for the degree of Doctor of Philosophy
Graduate Department of Physics
University of Toronto
© Copyright 2021 by Panagiotis Peter Stavropoulos

Emergent phenomena in correlated materials with strong spin-orbit coupling
Panagiotis Peter Stavropoulos

Doctor of Philosophy
Graduate Department of Physics
University of Toronto
2021
Abstract
Correlated materials have a long history in the research of the condensed matter
community, and in the past couple of decades effects of spin-orbit coupling have
been recognized for emergent unconventional phenomena in quantum materials. A
series of transition metal compounds have been proposed to exhibit exotic states
of mater. Among them, quantum spin liquids are highly entangled states of matter
where strongly interacting magnetic moments do not show long range magnetic order.
In general, strong spin-orbit coupling entangles spin and orbital degrees of freedom
into an effective pseudospin moment. For d5 materials like Sr2 IrO4 , β−Li2 IrO3 , and
α−RuCl3 , unconventional bond-dependent exchange interactions among jeff = 1/2
pseudospins are expected. Although stabilization of exotic states is possible within
these bond-dependent interactions, usually a unconventional magnetic order may set
in due to other small interactions. Beyond the pseudospin moment jeff = 1/2 pic-
ture, a bond-dependent interaction may exist among higher spin moments, but it is
unclear how such an interaction arise in transition metal systems. Understanding the
microscopics of how these interactions arise and how to manipulate them is impor-
tant in characterizing the nature of a phase, tuning the systems into different regimes,
and identifying new candidate materials. In this thesis, we provide the microscopic
theories for a certain set of pseudospin systems, and their implications. In the first
part, the unconventional counter-rotating magnetic order in β−Li2 IrO3 is addressed
from the perspective of a hidden symmetry of a minimal jeff = 1/2 model. Later, a
microscopic mechanism for a higher spin S model with bond-dependent interactions
ii
is proposed, where S=1 and S=3/2 cases are addressed explicitly. In the last part,
we present ab initio results that allow us to estimate and characterize the spin inter-
actions of related materials. Finally comments on open questions and directions for
future studies are summarized.
iii
Acknowledgments
I would like to begin by thanking my PhD supervisor Prof. Hae-Young Kee. For the
past few years she has guided and mentored me, and stuck with me through the best
and worst of times. I can only hope I have become a better man from our interactions,
and I will always carry with me the lessons she taught me.
I would like to acknowledge my supervisory committee, Prof. Yong-Baek Kim and
Prof. Stephen R. Julian for there support both as my committee members and as
teachers throughout the program.
To say I have met a lot of people is an understatement. Whether it be fleshing
out physics ideas or casual quality of life interactions I would like to thank in no spe-
cific order Andrei Catuneanu, Jacob S. Gordon, Austin Lindquist, Nazim Boudjada,
Emily Zinnia Zhang, Vijin Venu, Peihang Xu, CJ Woodford, Daniel Baker, Leonardo
Jose Uribe Castano, Sopheak Sorn, Wonjune Choi, Li Ern Chern, Geremia Mas-
sarelli, Adarsh Patri, Eli Bourassa, Ilan Tzitrin, Heung-Sik Kim, Yige Chen, Vijay
Shankar Venkataraman, Robert Schaffer, Ciaran Hickey, Cuauhtémoc Salazar, Daniel
A. Travo, Zachary Vernon, Sylvia Diana Swiecicki, Ilia Khait, Pranai Vasudev, Sergey
Eyderman. A very special thank you goes out to Heung-Sik Kim for his endless help,
I hope I didn’t bother you as much as I suspect I did.
I do not believe I would have come out alive if it where not for my friends, colleagues
and comrades Aris Spourdalakis and Dionysia Pitsili-Chatzi, to whom I am eternally
grateful to.
For their endless support in my endeavors I am deeply thankful to my mother
Panagiota Karouni and my sister Stavroula Stephi Stavropoulos. Thank you for
supporting and bearing with me all these years.
I am also grateful to have my friends over seas, Vasilis Rokaj, Petros Andreas
Pantazopoulos, George Batagiannis, Perseas Christodoulidis, Gilho Ahn, Savvas Ros-
tadis, Kalliopi Souvatzi, for keeping my spirits up.
Finally I would be remiss to not thank all the employed staff at the Department of
Physics, whose persistent work makes such a program possible for me and every other
graduate student. A special thank you to Isabella for being a friendly face during the
late work nights in the department.
iv
Contents
List of Tables vii
List of Figures ix
Author contributions xvi
1 Introduction 1
1.1 The jeff state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Finding an effective model: strong-coupling expansion . . . . . . . . . 4
1.3 Bond-dependent interactions . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Counter rotating spiral order 9

2.1 Hyperhoneycomb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Structure and spin model . . . . . . . . . . . . . . . . . . . . 10
2.1.2 SU(2) hidden symmetry: 12-site transformation . . . . . . . . 11
2.1.3 Counter-rotating spiral order: signature of SU(2) symmetry in
3D β − Li2 IrO3 . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.4 Phase diagram and moment direction . . . . . . . . . . . . . 14
2.1.5 Effects of interchain coupling: evolution from 1D to 3D . . . . 18
3 Higher-spin Kitaev models 23

3.1 Spin S=1 model from eg orbitals . . . . . . . . . . . . . . . . . . . . 24
3.1.1 Kanamori interaction . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.2 Tight-binding Hamiltonian . . . . . . . . . . . . . . . . . . . . 26
3.1.3 Strong-coupling expansion . . . . . . . . . . . . . . . . . . . . 28
3.1.4 Exact diagonalization of S=1 KH model . . . . . . . . . . . . 34
3.1.5 Kitaev candiate materials . . . . . . . . . . . . . . . . . . . . 34
3.2 Magnetic field effect in spin S=1 . . . . . . . . . . . . . . . . . . . . 35
3.3 Spin S=3/2 model from t2g orbitals . . . . . . . . . . . . . . . . . . . 37
v
3.3.1 Kanamori interaction . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.2 Tight-binding Hamiltonian . . . . . . . . . . . . . . . . . . . . 39
3.3.3 Ideal honeycomb structure . . . . . . . . . . . . . . . . . . . . 41
3.3.4 Effects of distortion: distorted octahdera . . . . . . . . . . . . 46
3.3.5 Application to CrI3 . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.6 Magnetic anisotropy . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.7 Spin gap and finite transition temperature . . . . . . . . . . . 54
4 Ab initio study of frustrated magnets 57

4.1 β−Li2 IrO3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1.1 Simulating pressure and structual relaxations . . . . . . . . . 58
4.1.2 Branching ratio . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1.3 Ab initio estimate of the branching ratio . . . . . . . . . . . . 61
4.2 NaNi2 BiO6−δ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.1 Ab initio results . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2.2 Local d7 moment . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Sr2 IrO4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3.1 Computational details . . . . . . . . . . . . . . . . . . . . . . 68
4.3.2 Results of atomic position optimization and Wannier orbitals . 69
4.4 α−RuCl3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.1 Computational details . . . . . . . . . . . . . . . . . . . . . . 72
4.4.2 Monolayer crystal modeling . . . . . . . . . . . . . . . . . . . 72
4.4.3 Hopping integral dependence on distortions: Wannier function
estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4.4 A leading order analysis . . . . . . . . . . . . . . . . . . . . . 76
4.4.5 Comparison to the bulk . . . . . . . . . . . . . . . . . . . . . 81
4.4.6 Spin model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5 Conclusions 85
A Exact diagonalization 90
B Spin wave theory of a ferromagnet on honeycomb lattice 94
C Details of the jeff infinite BR 97
Bibliography 101
*
vi
List of Tables
2.1 Axis and angle of rotation used in Eq. 2.6 to create the T12 counter-
rotating spiral ansatz. . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1 Energy spectrum of the on-site Hamiltonian H0 at the M and X sites.

′
The parameters Ud/p , Ud/p , JHd/p represent the Kanamori interactions
for d- and p-orbitals, and λp the spin-orbit coupling strength of the
X site, as discussed in the main text. The atomic potential energy of
M/X sites is denoted as ϵM/X . For the case of two holes at the X sites,
the two limits are presented: (1) Up′ ≡ Up − 2JHp and then JHp → 0,
and (2) λ → 0. In the ground state, eg orbitals have 2 holes in the spin
triplet state with energy EM,2,t while p-orbitals have zero holes with
energy EX,0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Spectrum of Hamiltonian Eq. (3.19). Listed states can contribute to
the second order strong-coupling expansion. . . . . . . . . . . . . . . 39
3.3 Effective hoppings to leading order in distortion-induced hoppings,
where δt are defined in Eq. (3.35). Distortion induced hoppings are
grouped in: δta = δt1 − δt4 , δtb = 2δt1 + δt4 , δtc = δt3 + δt5 , δtd =
δt1 − δt2 , δte = δt1 + δt4 , δtf = δt1 + δt2 . . . . . . . . . . . . . . . . . 49
3.4 Spin model terms, under distortions, to leading order in δt. Distortion-
induced δt defined in Eq. (3.35), and character subscripted δt defined
in caption of Tab. 3.3 and Eq. (3.38) . . . . . . . . . . . . . . . . . . 50
3.5 Indirect, direct and distortion induced hoppings in units of meV ob-
tained by ab initio calculations. The hopping integrals are defined in
Eq. (3.24), (3.31) and (3.35). . . . . . . . . . . . . . . . . . . . . . . . 51
4.1 Lattice parameters refer to the conventional orthorhombic unit cell and
are given in unit of Å. . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 BR values calculated from ab initio calculations. In all cases nh ≃ 5
as estimated from density of states. . . . . . . . . . . . . . . . . . . . 62
vii
4.3 Table is split into three sections: lattice optimization results (a) with-
out spin-orbit coupling or U, (b) including spin-orbit coupling (c) in-
cluding both spin-orbit coupling and U. First column is a list of the
substrates that Sr2 IrO4 is grown on, with the s labels standing for
s1 : GdScO3 (110), s2 : SrTiO3 (100), s3 : (LaAlO3 )0.3 (Sr2 AlTaO6 )0.7
(100), s4 : NdGaO3 (110), s5 : LaAlO3 (110). The second two columns
are the imposed lattice parameters a and c used as input in ab initio
calculations. The resulting Ir-O-Ir bond angles θ after atomic posi-
tion optimization, as well as t2g hopping parameters between first and
second nearest neighbors (n.n.) computed from Wannier calculations,
follow in the next three columns. The t2g hopping parameters are writ-
ten in matrix form in the {dxz , dyz , dxy } basis, with diagonal terms
being txz , tyz , txy and nonzero off-diagonal terms being tyzxz/xzyz . Us-
ing Eq. 4.3.2 (found in Ref. [158]) the resulting jeff hopping parameters
and spin interactions are given in the last six columns. All energy units
are in units of meV and length units are in Å. . . . . . . . . . . . . . 70
4.4 Tight-binding Hamiltonian HTZB of Eq. 4.4.3 as calculated from Wan-
nier functions for different δα and δβ. Energy units in meV and length
units in Å . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Comparison of tight-binding between the monolayer and the bulk.
Monolayer values taken directly from Tab. 4.4. . . . . . . . . . . . . . 82
4.6 Spin model parameters calculated from Tab. 4.4 using formulas from
Ref. [9]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
viii
List of Figures
1.1 Splitting of d5 due to octahedral Oh crystal field splitting ∆C and spin-

orbit coupling λ. When λ ≪ ∆C , the d-orbitals first split into t2g and eg
orbitals with five electrons in the t2g orbitals. The lobe structure of t2g
and eg orbitals are shown. The t2g forms effective angular momentum
leff = −1, which couples to spin through spin-orbit coupling leading
to half field jeff = 1/2 Kramers doublet. The real space density of
two jeff = 1/2 states mj = ±1/2 are shown with the color gradient
representing the spin contribution to the density from spin up being
red to spin down being blue. . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Honeycomb structure of black sites with edge-shared octahedra of lig-
ands in purple. The dxz to dyz hopping is mediated and greatly en-
hanced by the indirect process dxz − pz − dyz through the ligand p-
orbitals. Under this process the Kitaev interaction emerges as the
effective spin model. . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 Two conventional unit cells of the ideal hyper-honeycomb structure

are shown. The lattice vectors ⃗a, ⃗b, ⃗c for the conventional unit cell,
and xyz-axes are depicted by arrows in the left corner where ⃗a =
(−2, 2, 0), ⃗b = (0, 0, 4), ⃗c = (6, 6, 0) in the global coordinates xyz.
The positions of the 16 iridium (blue and yellow) atoms in the unit
cell can be generated from the F ddd space group starting with one
iridium atom at (1/8, 1/8, 17/24) in fractional coordinates. Iridium
atoms are located at the center of edge-sharing oxygen (light gray)
octahedra cages. The three types of nearest neighbor bonds X, Y, Z
are indicated on the left and the alternating sign carried by the σγij
symbol, is indicated by ±X and ±Y bonds. The 24-site sublattices
are labeled by a combination of 1, .., 6 and a, b. The spin directions
in the counter-rotating spiral at the hidden antiferromagnetic SU(2)
point are indicated by red arrows. See the main text for details. . . . 11
ix
2.2 Pictorial representation of the T12 transformation carried out on all
12-sites. The planes where spin moments rotate are indicated by blue
(A) and yellow (B) filled circles. Rotation axes are indicated by black
arrows, and the set of red arrows represents an example of moment
directions at the exactly solvable SU(2) point when the moment at 1a,b
is pinned along the (1, 1, 0)-direction. . . . . . . . . . . . . . . . . . . 14
2.3 N = 24 site cluster used for exact diagonalization. The blue and yellow
colors represent chains where the T6a and T6b are applied respectively.
The label na/b with n = 1, ..., 6 indicates the transformation at each
site. Periodic boundary condition (PBC) along the chains is imposed
by connecting the same color and style of the bonds. For different color
chains, it is imposed by connecting sites between na and nb via the Z-
bond with the same type black lines. This boundary condition respects
the T12 transformation. For the open boundary condition (OBC) dis-
cussed in the main text, half the Z-bonds are lost, denoted as black
lines along the ĉ-direction. The lighter and darker coloring of both yel-
low and blue chains indicate the staggering moment convention used
e2
in ⟨S stg,T 6 ⟩ discussed in Ch. 2.1.5. . . . . . . . . . . . . . . . . . . . 15
2.4 Phase diagram obtained by exact diagonalization on the 24-sites cluster

shown in Fig. 2.3 with (a) PBC and (b) OBC. There are four phases
identified. The S phase is the counter-rotating spiral ordered state,
which includes the hidden SU(2) point. The S ′ phase is another mag-
netically ordered phase and the nature of the S ′′ phase is discussed in
the main text. K refers to the Kitaev spin liquid phase. . . . . . . . . 16
2.5 Probability maps of P . (a) At the hidden SU(2) point −K = −Γ,

no direction is preferred and the probability is equal for every mo-
ment direction. (b) Perturbing away from the hidden SU(2) point
with |Γ| > |K| and J = 0, these bond-dependent terms favor the cu-
bic axis for the moment direction, indicated by green arrow for the
(1,0,0) direction. (c) Perturbing away from the hidden SU(2) point by
J > 0, the quantum fluctuations induced by J lifts the accidental clas-
sical U(1) degeneracy into six directions related by C3 : (1,1,0) (green
arrow), (0,1,-1), (-1,0,-1), (-1,-1,0), (0,-1,1), (1,0,1). . . . . . . . . . . 18
x
Order parameters ⟨S e2 e2 e2
2.6 stg,T12 ⟩ (red), ⟨Stot,T 6 ⟩ (blue), and ⟨Sstg,T 6 ⟩ (green)
for four values of interchain coupling X are shown. The dark solid line
is the second derivative of ground state energy with respect to ϕ, and it
represents the phase transitions. The hidden antiferromagnetic SU(2)
point from T12 transformation is shown as a filled circle in (a)-(d), and
the hidden ferromagnetic SU(2) point from T 6 as an empty circle in
the 1D limit where x = 0. . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Indirect hopping integrals between M and X sites are denoted by the
colored curve lines. The red, green, and blue colors represent t1 , t2 and
t3 respectively, and the sign of the hopping integrals is ignored for sim-
plicity. The M sites with eg orbitals are located in the center of each
octahedral cage formed by X sites occupied by three p-orbitals. Kitaev
bond-dependent interactions X-, Y-, and Z-bond are respectively rep-
resented by red, green, and blue shaded regions. For clarity, every X
site is drawn by two separated X sites to represent different hopping
contributions from different p and eg orbtials. The global coordinates
of x-, y-, and z-axes are shown in the center of the honeycomb plane. 27
3.2 Schematic representation of the three types of superexchange processes

that contribute to the spin interactions. The ring process involves
only one hole going around in a loop and related processes found by
permuting pairs 1, 2 ↔ 3, 4. The exchange process exchanges holes
between M sites via an intermediate state where each X site has one
hole at the same time, and related processes found by permuting 1 ↔ 2
and/or 3 ↔ 4. The two-hole process exchanges two holes between M
sites via an intermediate two hole state on the X site, and related
processes found by permuting 1 ↔ 2 and/or 3 ↔ 4. All processes have
a corresponding mirror processes obtained by permuting X1 ↔ X2
and/or M1 ↔ M2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
xi
3.3 (a) The phase diagram of the S=1 KH model. By tuning the ratio of
J/K, two transitions signalled by the singular behavior of first (blue)
and second (red) derivative of the ground state energy density ugs are
found on both 12- and 18-site exact diagonalization clusters shown in
(b) and (c), respectively. In the clusters, X- Y- Z- bonds are shown
in red, green and blue respectively, and the bonds imposing the peri-
odic boundary conditions are dashed. (d) Correlations for the 12 site
cluster discussed in main text. Site 3 is represented as a gray circle,
and real space correlations ⟨S3 · Si ⟩ are shown, with i running from
1 to 12. There are three phases identified by spin-spin correlators as
discussed in the main text. The Kitaev spin liquid (SL) appears near
J/K ∼ 0, and antiferromagnetic and zig-zag orderings are respectively
found in the antiferromagnetic and ferromagnetic Heisenberg interac-
tion regions. For J/K = 0.3 the correlations show antiferromagnetic
(AFM) behavior with alternating sign between any nearest neighbor
sites, while for J/K = −0.3 the correlations show zig-zag (ZZ) be-
havior with the correlations having the same sign within a XY chain
and opposite sign between the chains. The Kitaev point J/K = 0.0
correlations show Kitaev phase behavior with only nearest neighbor
correlations being finite and further neighbors being identically zero. . 33
3.4 Exact Diagonalization results on the 12 site ladder cluster for a S=1
Kitaev plus magnetic field h[111] model, with antiferromagnetic Ki-
taev K=1. The converged spectrum is shown along with the second
derivative of the ground state energy −∂J/K
2
uGS in blue. Two peaks
in the second derivative signify an intermediate state, with the spec-
trum collapsing to a very dense low energy spectrum, reminiscent of
the S=1/2 intermediate state. . . . . . . . . . . . . . . . . . . . . . . 36
3.5 The low lying spectrum of increasing ladder clusters from N=8 to 18.
The lowest states appear non degenerate with their gap closing with
increasing system size, while Wilson loop operator ±1 is found to dif-
ferentiate them. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Edge-shared octahedra honeycomb structure unit cell a, b, in global
coordinates xyz. Transition metal sites M in gray and non-magnetic
ligands X in purple. The nearest neighbor X-, Y-, and Z-bonds are
related by C3 symmetry. The sites M1 , M2 , X1 , and X2 are involved
in the second order strong-coupling expansion on the Z-bond. Indirect
hopping integrals t0 , t1 and t2 are shown. . . . . . . . . . . . . . . . . 40
xii
3.7 a) The distorted octahedra in R3̄ is shown. The octahedron made
of X ligand can be viewed as two yellow triangles normal to the ĉ =
√
(1, 1, 1)/ 3 direction. The blue arrows represent new positions of X
due to the staggered rotations of the two yellow triangular faces. The
change of position X due to the staggered rotations is parameterized
by δx. In addition, there is a compression of the two yellow triangles,
squeezed as shown by the orange arrows parallel to the ĉ-axis. The
change of position due to the compression is parameterized by δx′ . b)
A top view of distortions in a unit cell is shown. The dotted circles ⊙
indicate the new position of X moving out of the page along +ĉ while
the circled cross ⊗ indicate a new position of X moving into the page
toward −ĉ. The exact positions of X(1,2,3,4,5,6) as a function of δx, δx′
are found in the Appendix. . . . . . . . . . . . . . . . . . . . . . . . . 47
3.8 Spin model parameters J, K, Γ, Γ′ , and Ac estimated from ab initio

parameters and plotted against JH /U . Solid, dashed, and dotted lines
correspond to U = 3eV, 4eV and 5eV respectively. Shaded region
corresponds to JH /U ∼ 0.24 which is relevant to CrI3 microscopics. . 52
3.9 Moment pinning calculation for different values of (J, K, Γ) in the fer-
romagnetic phase: a) (−1.00, −0.20, 0.00), b) (−0.20, −1.00, 0.00), c)
(−0.20, −1.00, −0.02). The Kitaev interaction always prefers the cubic
axis as shown in panel a) and b). A small interaction like Γ is necessary
to pin along the ĉ direction as show in panel c). . . . . . . . . . . . . 54
3.10 X-, Y- and Z-bonds show in red, green and blue respectively. Dashed
bonds represent the periodic boundary conditions. . . . . . . . . . . . 54
4.1 Hyper-honeycomb structure of β−Li2 IrO3 . The conventional orthorhom-

bic unit cell is presented on the left, showing the edge-sharing struc-
ture of the nearest neighbor iridium bonds. On the right all sites are
stripped except for the hyper-honeycomb iridium network, and embe-
ded is the primitive unit cell as a yellow prism, with the primitive cell
atom basis indicated with pink. The relationship of the conventional
and primative lattice vectors are: α1 = (α2′ + α3′ )/2, α2 = (α1′ + α3′ )/2,
α3 = (α1′ + α2′ )/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2 Pictorial representation of excitation from p-core orbitals to valence

jeff states at five electron filling per iridium site. The p1/2 to jeff = 1/2
transition are dipole forbidden, leading to an infinite BR. . . . . . . . 59
xiii
4.3 Electronic band structures for β − Li2 IrO3 showing projected jeff =
1/2 and jeff = 2 3 states obtained for P = 0 GPa, Ueff = 2.5 eV (left
panel) and P = 3.08 GPa, Ueff = 1 eV (right panel) (no magnetic
order for both cases) and their respective projected density of states.
Experimentally determined lattice unit cells were used for both ambient
and high-pressure cases. The small size of the gap in the calculations
in the absence of magnetic order is due to the limitations of Hartree-
Fock in the ab initio calculations. When Mott physics is considered,
the gap should be bigger than shown. . . . . . . . . . . . . . . . . . . 63
4.4 Partial density of states of NaNi2 BiO6 . The left panel shows the band
structure and the right panel shows the density of states . . . . . . . 65
4.5 Partial density of states of Na3 Ni6 Bi3 O17 , which simulates NaNi2 BiO5.66 .
The left panel shows the band structure and the right panel shows the
density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 Local moment calculation for N i3+ d7 as a function of ϕ to interpolate
between trigonal distortion ∆trig and spin-orbit coupling λ as ∆trig =
Acosϕ, λ = Asinϕ. The left panel shows the effective moment within
the t2g by projecting out the eg orbitals, and the right panel shows the
total moment including eg orbitals. . . . . . . . . . . . . . . . . . . . 67
4.7 Left: Conventional unit cell of Sr2 IrO4 with the the characteristic cor-
ner shared staggered rotated IrO6 octahedra structure. Right: Top
view of one layer of IrO6 octahedra, with Ir-O-Ir angle θ measuring the
rotation angle between octahedra. . . . . . . . . . . . . . . . . . . . . 69
4.8 Pictorial representation on the two types of distortion: staggered rota-
tions shown in the top row and molded by δα, and compression shown
in the bottom row and modeled by δβ. δα and δβ are magnitudes,
while their direction is shown with blue arrows. The lattice parame-
ter is set to l = |a| = |b|. Crystal structure variables θI and wI are
the values in the ideal edge-shared octahedral geometry, while under
distortions we generally have θ, w (see text for detailed discussion on
these values). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.9 Conventions used in the model set up. X, Y, and Z label the three
kinds of nearest neighbor bonds. The bond Hamiltonians HTZB ,HTZB
and HTZB are related by C3 symmetry. . . . . . . . . . . . . . . . . . 74
4.10 Plots of hopping integrals against δβ, where: a) δα = 0.00Å, b) δα =
0.10Å. Energy units in meV and length units in Å. . . . . . . . . . . 75
xiv
4.11 Hopping channels in the ideal case for three different view points. Left
(i) (i) (i)
to right view point: generic, (0,0,1), (1,1,0). Top to bottom: t1 , t2 , t3 77
4.12 Hopping channels in the staggered rotations case for three different
view points. Left to right view point: generic, (0,0,1), (1,1,0). Top to
(r) (r) (r)
bottom: t1 , t2 , t3 . Only leading order corrections are shown (see
discussion in main text). . . . . . . . . . . . . . . . . . . . . . . . . . 79
view points. Left to right view point: generic, (0,0,1), (1,1,0). Top to
(c) (c) (c)
bottom: t1 , t2 , t3 . Only leading order corrections are shown (see
view points. Left to right view point: generic, (0,0,1), (1,1,0). All
cases are about t4 corrections, with top to bottom: ideal, stagered
rotations, compressions. Only leading order corrections are shown (see
4.15 Plots of J, K, Γ, Γ′ against δβ, where: a) δα = 0.00 Å, b) δα = 0.10 Å.
Energy units in meV and length units in Å. Plotting data from Tab.
4.6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
B.1 X Y and Z bonds of the J − K − Γ − Γ′ − Ac shown in red, green and

blue respectively. Second nearest neighbor bonds are shown in orange.
The orange arrows indicate when sgn(ij) = +1 in the Dzyaloshinskii-
Moriya term. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
xv
Author contributions
The works contained in this thesis are based on the following publications by the
author and collaborators:
• Ch. 2: “Counter-rotating spiral order in three-dimensional iridates: Signature
of hidden symmetry in the Kitaev-Γ model”, P. P. Stavropoulos, A. Catuneanu,
and H.-Y. Kee, Physical Review B 98, 104401 (2018).
• Ch. 3.1: “Microscopic Mechanism for a Higher-Spin Kitaev Model”, P. P.
Stavropoulos, D. Pereira, and H.-Y. Kee, Physical Review Letters 123, 037203
(2019).
• Ch. 3.2: “Field-driven gapless spin liquid in the spin-1 Kitaev honeycomb
model”, C. Hickey, C. Berke, P. P. Stavropoulos, H.-Y. Kee, and S. Trebst,
Physical Review Research 2, 023361 (2020).
“Characterizing spin-one Kitaev quantum spin liquids”, I. Khait, P. P.
Stavropoulos, H.-Y. Kee, and Y. B. Kim, Physical Review Research 3, 013160
(2021).
• Ch. 3.3: “Magnetic anisotropy in spin-3/2 with heavy ligand in honeycomb
Mott insulators: Application to CrI3 ”, P. P. Stavropoulos, X. Liu, and H.-Y.
Kee, Physical Review Research 3, 013216 (2021).
• Ch. 4.1: “Pressure tuning of bond-directional exchange interactions and
magnetic frustration in the hyperhoneycomb iridate β−Li2 IrO3 ”, L. S. I. Veiga,
M. Etter, K. Glazyrin, F. Sun, C. A. Escanhoela, G. Fabbris, J. R. L.
Mardegan, P. S. Malavi, Y. Deng, P. P. Stavropoulos, H.-Y. Kee, W. G. Yang,
M. van Veenendaal, J. S. Schilling, T. Takayama, H. Takagi, and D. Haskel,
Physical Review B 96, 140402(R) (2017).
• Ch. 4.2: “Counterrotating magnetic order in the honeycomb layers of
NaNi2 BiO6−δ ”, A. Scheie, K. Ross, P. P. Stavropoulos, E. Seibel, J. A.
Rodriguez-Rivera, J. A. Tang, Y. Li, H.-Y. Kee, R. J. Cava, and C. Broholm,
Physical Review B 100, 214421 (2019).
• Ch. 4.3: “Compressive strain induced enhancement of exchange interaction and
xvi
short-range magnetic order in Sr2 IrO4 investigated by Raman spectroscopy”, A.
Seo, P. P. Stavropoulos, H.-H. Kim, K. Fürsich, M. Souri, J. G. Connell, H.
Gretarsson, M. Minola, H.-Y. Kee, and B. Keimer, Physical Review B 100,
165106 (2019).
Exact diagonalization calculations in Ch. 3.2 are of motivating and supporting role,
and the lead authors are Dr. Ciarán Hickey and Dr. Ilia Khait respectively. Ab
initio estimate of tight-binding parameters discussed in Ch. 3.3 was performed by
Dr. Xiaoyu Liu. Ab initio calculations in Ch. 4.1, 4.2, 4.3 were performed in support
of experimental findings, with experiments carried out by the collaborators.
Publications/preprints the author was involved in but not included in this thesis:
• “Optical magnons with dominant bond-directional exchange interactions in the
honeycomb lattice iridate α−Li2 IrO3 ””, S. H. Chun, P. P. Stavropoulos, H.-Y.
Kee, M. M. Sala, J. Kim, J.-W. Kim, B. J. Kim, J. F. Mitchell, and Y.-J. Kim,
Physical Review B 103, L020410 (2021). I contributed spin wave analysis to
support experimental findings and conclusions.
• “Exchange interactions, Jahn-Teller coupling, and multipole orders in
pseudospin one-half 5d2 Mott insulators”, G. Khaliullin, D. Churchill,
P. P. Stavropoulos, H.-Y. Kee, arXiv:2105.09334 (2021). I performed spin wave
analysis and classical phase diagram calculations.
• “Spontaneous Chiral-Spin Ordering in Spin-Orbit Coupled Honeycomb
Magnets”, Q. Luo, P. P. Stavropoulos, H.-Y. Kee, arXiv:2010.11233. I
contributed dynamical spin structure factor calculations, and ab initio
calculations showing how to achieve the phase space presented in this work.
xvii
Chapter 1
Introduction
Condensed matter physics has the unique position of being on the cusp of the-
oretical physics and materials development, which is certainly the case for a lot of
magnetism related work. Early on, the focus of research was 3d transition metal com-
pounds. Large Coulomb electron-electron (e-e) interactions are expected due to the
small radial extent of 3d orbitals, leading to a scenario where a system expected to
be metallic from band theory ends up being insulating, i.e. the Mott insulator. For a
half-filled band with one electron per site, the large e-e interactions basically disfavor
the movement of electrons from atom to atom, as electrons would have to reside on
the same atom costing too much energy. Instead, the electrons prefer to settle locally
into each atom leading to an insulator. Within the 3d transition metals, large e-e in-
teractions and small band widths can lead to magnetic behavior, where the effective
moments of spin, orbital, or composite character, interact via exchange interactions.
The Goodenough-Kanamori-Anderson rules are especially useful in predicting these
interactions.
Looking at heavier 4d or 5d transition metals the picture is not immediately as
clear as in the 3d case. The larger radial extent of 4d and 5d orbitals complicates
the picture because of smaller e-e interactions. Nevertheless such compounds have
received an abundance of attention in the past two decades because of the presence
of strong spin-orbit coupling in these heavier transition metals. Spin-orbit coupling is
an interaction between the spin and orbital degrees of freedom of an electron moving
around an atom, which comes in as a correction term to the familiar non-relativistic
Schrodinger’s equation. An electron zooming around in the presence of the atomic
nucleus’s potential essentially acts like a current; ergo, a magnetic field that interacts
with its spin. A careful examination in the case of a hydrogen atom reveals that this
correction is proportional to (Zα)4 , where Z is the atomic number and α ≃ 1/137 the
fine structure constant.[1] For lighter atoms, the Z 4 correction suppresses the effects
1
2 CHAPTER 1. INTRODUCTION
of spin-orbit coupling, but once 4d and 5d compounds are considered with Z ≳ 40,
the spin-orbit coupling starts to become a significant energy scale that cannot be
disregarded. In broad strokes, adding spin-orbit coupling on top of the band structure
further splits the bands into even narrower band widths, which, relatively speaking,
makes the e-e interaction of the heavy compound appear large in comparison, leading
to a Mott insulator. The birth of this idea is clearly demonstrated in Sr2 IrO4 [2]. Thus,
the presence of spin-orbit coupling in heavier compounds has important implications
for the resulting exchange interactions, which can be viewed through the lens of
frustration in magnetic systems [3].
Frustration appears when we have physical interactions, which cannot be simul-
taneously satisfied, competing for the ground state of the system. This may lead
the system to resolve this frustration by a non-trivial compromise. A prototypical
example of frustration is seen in the antiferromagnetic Ising model on the triangular
lattice.[4] Here the frustration is due to geometrical constraints. Imagining a single
triangle with antiferromagnetic Ising interaction on each of the three bonds, one can
see that two sites can always be fully energetically satisfied when the spins are an-
tiparallel, while the third one dangles unhappily in either up or down state. These
“unhappy” states are energetically equal, which is to say they are degenerate, and
they quickly scale to a vast extensive degeneracy in the thermodynamic limit. Within
this extensive degeneracy unconventional physics may reside. In the triangular anti-
ferromagnetic Ising model this leads to a spin liquid, an unconventional paramagnet
where the the highly interacting Ising moments cooperate to create a non magnetic
state as a way to resolve frustration.
Another path to frustration is bond-dependent interactions. Here, the geometry
need not be the one constraining the system to “unhappy” states, however, a suffi-
ciently alternating interaction model on different bonds can bring about “unhappy”
frustration. Such is the case in Kitaev’s model on a honeycomb lattice, an exactly
solvable model that leads to a quantum spin liquid – a highly entangled state with
exotic properties like fractionalized excitations.[5]. An antiferromagnetic Ising model
on the honeycomb lattice is not geometrically frustrated, on account of the lattice
being a bipartite lattice, and a ferromagnetic Ising model is most certainly not frus-
trated. However, in the ferromagnetic or antiferromagnetic Kitaev model frustration
is born out of the alternating character of three different Ising interactions involving
only x, only y or only z spin components on different bonds.
Where does the role of spin-orbit coupling fit into the above? It comes into play
via the formation of bond-dependent interactions. The study of spin interactions had
been largely of Heisenberg type because the Heisenberg interaction is rotationally
1.1. THE JEFF STATE 3
invariant, consistent with the spin not being able to distinguish directions in the
absence of spin-orbit coupling. However, when one ties the spin to the orbitals through
spin-orbit coupling, directional information can get backed into the spin interactions.
The directional nature of d-orbitals residing on a lattice with particular neighbor
directions is passed onto the spin, resulting in bond-dependent exchange interactions.
This is the case in the seminal work of Jackeli and Khaliullin [6], who showed that
the Kitaev interaction arises in d5 systems with strong spin-orbit coupling. This led
to an explosion of research on Kitaev candidate materials based on this paradigm. It
is instructive to take a closer look at the spin-orbit entangled atomic state that leads
to such interactions, which we show in the subsection below.
1.1 The jeff state
Figure 1.1: Splitting of d5 due to octahedral Oh crystal field splitting ∆C and spin-orbit coupling λ.
When λ ≪ ∆C , the d-orbitals first split into t2g and eg orbitals with five electrons in the t2g orbitals.
The lobe structure of t2g and eg orbitals are shown. The t2g forms effective angular momentum
leff = −1, which couples to spin through spin-orbit coupling leading to half field jeff = 1/2 Kramers
doublet. The real space density of two jeff = 1/2 states mj = ±1/2 are shown with the color gradient
representing the spin contribution to the density from spin up being red to spin down being blue.
In perovskites like Sr2 IrO4 , as well as Kitaev candidates like Na2 IrO3 , Li2 IrO3 and
RuCl3 , the transition metal (Ir,Ru) resides in the center of octahedra formed by lig-
ands (O,Cl). This octahedral environment lowers the symmetry from a fully rotation-
ally invariant isolated atom down to the octahedral group Oh . As a result, the valence
d-orbitals split under Oh crystal field splitting into a doublet eg = {dx2 −y2 , d3r2 −z2 }
and a triplet t2g = {dyz , dxz , dxy }. Which manifold, eg or t2g , resides higher or lower
energetically can be understood by physical considerations of the orbital’s lobe di-
rection. The crystal field splitting occurs because the electrons on the center atom
interact with the surrounding octahedral environment. The eg lobes point towards
the surrounding sites, while t2g lobes point in-between surrounding sites. As a result
t2g are lower in energy, while eg are higher and separated by the crystal field splitting
denoted by ∆C .
In d5 systems, five electrons reside in the t2g orbitals because of the crystal field
splitting. Typical values of ∆C in transition metal material are of order ∆C ∼1-2eV.
For heavy compounds, like Ir and Ru, spin-orbit coupling λ enters as a secondary
energy scale, with λ ∼ 400meV. The crucial point is that although spin-orbit coupling
is strong, the energy scales still have the relation λ ≪ ∆C . As a result the five
electrons fill the t2g manifold, which then splits into jeff states due to spin-orbit
coupling. A schematic of this process is shown in Fig. 1.1.
The jeff = 1/2 and 3/2 states come out of the coupling of the t2g orbitals and
spin degree of freedom through spin-orbit coupling. When angular momentum L is
projected into the t2g orbitals with the projector Pt2g , the operators Leff = −Pt2g LPt2g
are a three dimensional representation that satisfy the algebra of angular momentum
[Lxeff , Lyeff ] = iℏLzeff , and the t2g states form a leff = −1. Coupling the leff = −1 to
s = 1/2 through spin-orbit coupling will result in total angular momentum jeff = 1/2
and 3/2. The jeff = 1/2 is higher when λ is positive, because of the minus sign
in the leff . Finally, the filling of five electrons results in fully occupied jeff = 3/2
manifold and a singly occupied jeff = 1/2 Kramers doublet. Under this picture, the
contributing degrees of freedom at the valence will be half filled jeff = 1/2 states.
To find the form of the jeff = 1/2 state, first transform t2g states into leff states;
1
|0⟩ = |xy⟩ | ± 1⟩ = √ (∓|yz⟩ − i|xy⟩). (1.1)
2
In this basis, the states |0⟩, | ± 1⟩ are eigenstates of the Lzeff operator. Finally, angular
momentum addition of leff = −1 and s = 1/2 will lead to:
1
|jeff = 1/2, +1/2⟩ = √ (|yz ↑⟩ − i|xz ↑⟩ − |xy ↓⟩) . (1.2)
3
with the |jeff = 1/2, −1/2⟩ being its time reversal partner.
1.2 Finding an effective model: strong-coupling expansion
When talking about the exchange interactions arising in transition metal com-
pounds, the interacting moments can be of spin, orbital, or even an entangled char-
acter like the jeff state. The electrons are viewed as trapped locally on each site,
unwilling to move around due to the large e-e interaction. This allows us to write
1.3. BOND-DEPENDENT INTERACTIONS 5
down an effective model to play with. So, what exactly is an effective model? Suppose
we have a Hamiltonian H living in some Hilbert space W. An effective model Heff is
a Hamiltonian that acts in a subspace Wo ⊂ W such that
H |Ψ⟩ = E |Ψ⟩ then Heff |Ψo ⟩ = E |Ψo ⟩ , with |Ψ⟩ ∈ W and |Ψo ⟩ ∈ Wo . (1.3)
Put into words, the effective model Heff replicates the eigenvalues E of the full model
H, but does so by acting only on a subset of states |Ψo ⟩. The hope is that the effective
model is more tractable to solve and characterize, so as to get insights about the full
model.
The prescription of arriving at an effective model is left open ended, and various
different schemes can be potentially applicable. A particularly useful prescription
in transition metal magnetism is through a perturbative approach. One can ask to
arrive at an effective model Heff which perturbatively approximates E up to some
order. Kato’s extensive work puts the perturbation approach on sound mathematical
footing.[7]
A simple case of large e-e interactions U localizing the electrons, with a kinetic term
t treated perturbatively, is illustrated in a two site, half filled, Hubbard model.[8] The
Hilbert space is spanned by W = {|↑, ↑⟩, |↑, ↓⟩, |↓, ↑⟩, |↓, ↓⟩, |↑↓, 0⟩, |0, ↑↓⟩}. The low
energy subspace, considering e-e interactions, would be single occupancy states Wo =
{|↑, ↑⟩, |↑, ↓⟩, |↓, ↑⟩, |↓, ↓⟩} with a cost of 0, while double occupancy states {|↑↓, 0⟩,
|0, ↑↓⟩} cost U . Perturbatively correcting against the small kinetic term t ≪ U , leads
to an effective spin model Heff = (4t2 /U )S1 · S2 , constrained in the single occupancy
subspace Wo . The absence of any spin-orbit coupling leads to a rotationally invariant
antiferromagnetic Heisenberg interaction of strength J = 4t2 /U between the two sites.
This procedure is normally called strong-coupling expansion, since the e-e interaction
is treated as the dominant Hamiltonian term.
Effective interactions for the pseudospin jeff = 1/2 state become richer due to
the complex nature of its wavefunction Eq. 1.2. Here orbital and spin degrees of
freedom are entangled. The e-e interactions happen on the orbital components, and
the different orientations of the orbital lobes mean several different hoppings t can be
involved at the same time. The spin component comes with different up and down
states tied to different orbital components effectively leading to a bond-dependent
pseudospin hopping resulting in a bond-dependent pseudospin model.
Figure 1.2: Honeycomb structure of black sites with edge-shared octahedra of ligands in purple. The
dxz to dyz hopping is mediated and greatly enhanced by the indirect process dxz − pz − dyz through
the ligand p-orbitals. Under this process the Kitaev interaction emerges as the effective spin model.
1.3 Bond-dependent interactions
In the seminal work of Jackeli and Khaliullin [6] a honeycomb structure was con-
sidered. Each honeycomb site is in the d5 configuration and is surrounded by a ligand
octahedra, with neighboring sites sharing octahedra edges. Naturally this leads to
half filled jeff = 1/2 states. In Fig. 1.2 the edge-shared honeycomb structure is shown,
with three distinct honeycomb nearest neighbor bonds X-, Y-, and Z- bonds. Focusing
on the Z-bond for the moment, the presence of these edge-shared octahedra greatly
enhances one of the hopping channels, specifically the indirect dxz −dyz . With ligands
consisting of valence p-orbitals, the ligand mediated dxz − pz − dyz process effectively
creates an indirect dxz − dyz hopping t = t2π /∆pd where tπ and ∆pd are the hopping
and the atomic energy difference respectively between metal d and ligand p orbitals.
Rewriting this hopping in the jeff basis, it is found that there is no hopping between
1.4. ORGANIZATION 7
pseudospin jeff = 1/2 states and, thus, no conventional Heisenberg interaction arises.
Nevertheless, jeff = 1/2 states at a site can still hop to jeff = 3/2 states at a next site,
and carrying out a strong-coupling expansion one arrives at H = KSiz Sjz with a fer-
romagnetic K = −(JH /U )4t2 /U where JH is the Hund’s coupling present due to the
d multiorbitals. Similarly, on X- and Y-bonds the same process would give KSix Sjx
and KSiy Sjy respectively, leading to the Kitaev model on the honeycomb lattice. The
authors speak clearly on these matters.
Later works showed that, due to the presence of other hopping channels, more
interactions arise. A generic nearest neighbor spin model on an ideal honeycomb
lattice derived by Rau et al [9] found a bond-dependent symmetric off-diagonal Γ-
term and a extended Kitaev-Γ-Heisenberg model was obtained
∑
H= [JSi · Sj + KSiγ Sjγ + Γ(Siα Sjβ + Siβ Sjα )], (1.4)
⟨i,j⟩∈γ−bond
where i, j are summing over first nearest neighbors, γ = x, y, z denotes the spin
component along the γ-bond and α, β are the remaining spin components. Roughly,
Γ ∝ (JH /U )ttd /U and J ∝ (t2d +4td t′d )/U , where td and t′d are direct hoppings between
dxy − dxy and dxz (dyz ) − dxz (dyz ) respectively. When there are only indirect hoppings
t, it reduces to the Kitaev model. The presence of the Γ interaction was confirmed
by ab initio [10] and quantum molecular calculations [11].
Since the spin interactions in general depend on hopping parameters, spin-orbit
coupling, and Coulomb interactions, their strengths vary among different materi-
als. Under the jeff = 1/2 picture, relevant materials include the iridates A2 IrO3
(A=Li,Na), α−RuCl3 , as well as three dimensional generalizations of the honeycomb
structure β− and γ−Li2 IrO3 . Although magnetic order eventually sets in at low tem-
peratures in all these compounds, capturing the magnetic orders of zig-zag (Na2 IrO3 ,
α−RuCl3 ) and unconventional counter-rotating spirals (α−, β−, γ−Li2 IrO3 ) has
required bond-dependent interactions including the Kitaev interaction.[12–14] This
leaves the possibility of tuning the systems via external means like applying pressure,
strain or a magnetic field, the latter having particularly intriguing, albeit debated,
results in α−RuCl3 [15].
1.4 Organization
The rest of the thesis is organized as follows: In Ch. 2, our work on the counter-
rotating spiral order in β−Li2 IrO3 is presented.[16] A minimal Kitaev-Γ model is
used and the experimental spiral is found to be related to a hidden symmetry. This
is based on the jeff = 1/2 paradigm. In Ch. 3, the idea of bond-dependent models
in higher spin systems is presented. A Kitaev-Heisenberg model is explicitly derived
for S=1 systems from half filled eg orbitals.[17] It is argued that the Heisenberg term
can be easily suppressed, and a numerical study clarifies nearby phases around the
Kitaev state. Preliminary results of the effects of a magnetic field on S=1 Kitaev
model and supporting exact diagonalization calculations are also presented.[18, 19]
The higher spin derivation is extended into S=3/2 systems from d3 in the t2g orbitals,
where effects of distortion are used to make a connection with CrI3 .[20] In Ch. 4,
ab initio studies of frustrated magnets are presented. Supporting results from ab
initio calculations for experimental work on β − Li2 IrO3 [21], NaNi2 BiO6−δ [22] and
Sr2 IrO4 [23] are summarized, as well as a tunability study of α−RuCl3 monolayer
through effects of distortions. Finally, in Ch. 5, findings are briefly summarized and
possible extensions are discussed. Some secondary information can be found in the
Appendices.
Chapter 2
Counter rotating spiral order
2.1 Hyperhoneycomb
Beyond the two dimensional (2D) layered honeycomb iridates [24, 25] and α−RuCl3
[26, 27], three dimensional (3D) hyper- and harmonic-honeycomb iridates β− and
γ−Li2 IrO3 [28, 29] have seen active research as Kitaev spin liquid candidates. At low
temperatures, these candidates order magnetically instead of becoming spin liquids,
but the magnetic ordering patterns are unconventional. The 3D iridates display in-
commensurate counter-rotating spiral ordering [30–32] with an ordering temperature
of 38K and 39.5K in β− and γ−Li2 IrO3 respectively. The magnetic order suggests
that other spin interactions apart from the Kitaev term must play a role, and thus the
Kitaev-Heisenberg (KH) model was proposed to explain the experimentally observed
magnetic orders. [25, 33–42]
The classical Kitaev-Γ-Heisenberg (KΓH) model was studied for 3D systems in
Ref. [43] and several ordered phases were found. Among them, the SPa− phase has
the same symmetry as the experimentally observed magnetic ordering pattern, as de-
termined by single crystal magnetic resonant X-ray diffraction experiments.[30] This
phase occurs when K and Γ are negative with a small positive Heisenberg interaction.
Ab initio studies and quantum molecular calculations on 3D iridates [44–46] reported
different values, and both Kitaev and Γ terms are negative and dominate over the
Heisenberg term.[45, 46]
It is worthwhile to note that in the 2D honeycomb lattice, the Kitaev-Γ (KΓ) model
has a hidden SU(2) symmetry via a 6-site transformation when K = Γ. This point in
phase space corresponds to the ordered phase denoted as the 120◦ state [9] or vortex
state [47]. Studies of the classical 3D KΓ model using the sublattice transformation
found two different phases near the same parameter space depending on the relative
strength of K and Γ .[48, 49]
9
10 CHAPTER 2. COUNTER ROTATING SPIRAL ORDER
Identifying spin liquids and nearby phases in quantum spin models in 3D is a

challenging task. Exact diagonalization is difficult to perform due to the large Hilbert
space. However, if there is an exactly solvable point in the KΓ model, one may limit
the parameter space around the solvable point and search for possible ordered phases
and spin liquids nearby. A symmetric point would suggest a natural way to construct a
suitable exact diagonalization cluster, as it may involve a finite number of sites similar
to the 4- or 6-site transformation found in the 2D honeycomb.[47] Before proceeding
in the aforementioned direction, the hyper-honeycomb β−Li2 IrO3 structure and a
minimal spin model are presented.
2.1.1 Structure and spin model
The hyper-honeycomb β−Li2 IrO3 , whose conventional unit cell is shown in Fig. 2.1,
can be viewed as the 3D generalization of the familiar 2D honeycomb structure. The
iridium sites form a tri-coordinated lattice with three bond types labeled by X, Y and
Z, similar to the 2D honeycomb lattice Fig. 1.2. The different color sites (yellow and
blue) which form chains in Fig. 2.1 denote the sign convention of the spin interactions
on the X- and Y-bonds, where different colored chains are connected by the Z-bonds.
A combination of crystal field splitting due to the octahedra cages and a strong
spin-orbit coupling leads to a pseudospin jeff = 1/2 as the relevant spin degree of
freedom in iridium oxides as described in the Introduction.[2, 50] Furthermore, the
edges of the octahedra cages are shared between two iridium atoms, resulting in
dominant bond-dependent interactions in the large Hubbard U limit. The minimal
nearest neighbor KΓH model as discussed above is written as [9, 43]:
∑
H= [JSi · Sj + KSiγ Sjγ + σγij Γ(Siα Sjβ + Siβ Sjα )], (2.1)
⟨ij⟩∈γ−bond
where i, j are summing over first nearest neighbors, γ = x, y, z denotes the spin
component along the γ-bond, and α, β are the remaining spin components. Due to
the rotation of octahedra, the Γ terms on X- and Y-bond have a sign change indicated
ij
by σx/y = ±1, while for the Z-bond, σzij = +1 as shown in Fig. 2.1. The space group
F ddd describing the hyper-honeycomb lattice allows for a different length of the Z-
bond from the X- and Y-bonds, as well as slight distortions of the oxygen octahedra
leading to anisotropic strengths between the bond interactions. Here, we will focus
on the ideal case with the same bond interaction strength among X-, Y- and Z-bonds.
2.1. HYPERHONEYCOMB 11
Figure 2.1: Two conventional unit cells of the ideal hyper-honeycomb structure are shown. The
lattice vectors ⃗a, ⃗b, ⃗c for the conventional unit cell, and xyz-axes are depicted by arrows in the left
corner where ⃗a = (−2, 2, 0), ⃗b = (0, 0, 4), ⃗c = (6, 6, 0) in the global coordinates xyz. The positions
of the 16 iridium (blue and yellow) atoms in the unit cell can be generated from the F ddd space
group starting with one iridium atom at (1/8, 1/8, 17/24) in fractional coordinates. Iridium atoms
are located at the center of edge-sharing oxygen (light gray) octahedra cages. The three types of
nearest neighbor bonds X, Y, Z are indicated on the left and the alternating sign carried by the σγij
symbol, is indicated by ±X and ±Y bonds. The 24-site sublattices are labeled by a combination of
1, .., 6 and a, b. The spin directions in the counter-rotating spiral at the hidden antiferromagnetic
SU(2) point are indicated by red arrows. See the main text for details.
2.1.2 SU(2) hidden symmetry: 12-site transformation
Previous studies on hidden symmetry via different sets of sublattice transforma-

tions on the 2D honeycomb lattice uncovered the zig-zag, stripy, and 120◦ (vortex)
ordered states, in addition to the conventional antiferromagnetic and ferromagnetic
states in the KΓH Hamiltonian.[47] This sublattice transformation idea was also ap-
plied to explain the magnetic orders in 3D iridates.[42, 48, 49] Here, we study a 12-site
transformation which leads to the SU(2) symmetric points in the KΓ model. This
transformation, dubbed the T12 transformation, is given by,
T6a (2.2)
sublattice 1a : (S x , S y , S z ) → ( Sex , Sey , Sez ),
sublattice 2a : (S x , S y , S z ) → ( Sez , −Sey , Sex ),
sublattice 3a : (S x , S y , S z ) → (−Sez , −Sex , Sey ),
sublattice 4a : (S x , S y , S z ) → ( Sey , Sex , −Sez ),
sublattice 5a : (S x , S y , S z ) → (−Sey , Sez , −Sex ),
sublattice 6a : (S x , S y , S z ) → (−Sex , −Sez , −Sey ).
T6b (2.3)
sublattice 1b : (S x , S y , S z ) → (−Sey , −Sex , −Sez ),
sublattice 2b : (S x , S y , S z ) → ( Sey , −Sez , −Sex ),
sublattice 3b : (S x , S y , S z ) → ( Sex , Sez , −Sey ),
sublattice 4b : (S x , S y , S z ) → (−Sex , −Sey , Sez ),
sublattice 5b : (S x , S y , S z ) → (−Sez , Sey , Sex ),
sublattice 6b : (S x , S y , S z ) → ( Sez , Sex , Sey ).
The T12 transformation can be considered as two 6-site transformations denoted by

T6a and T6b . As shown in Fig. 2.1, blue and yellow chains refer to the a- and b-subset
respectively. The Z-bonds connect the two chains by sites of the same numbering
of the a- and b-subset, i.e., na and nb where n = 1, ..., 6. Here, xyz refers to the
global axis shown in Fig. 2.1. A pair of two T12 transformations then makes a 24-site
sublattice: alternating T6a and T6b along the ⃗c-direction, i.e., x̂ + ŷ direction in xyz
coordinates. For the KΓ model, this 12-site transformation reveals a hidden SU(2)
symmetry when K = Γ, where the spin model takes the Heisenberg interaction form
in the transformed basis, i.e., H → H e = Je ∑ S e e e
⟨ij⟩ i · Sj with J = −K. For positive
K and Γ, it maps to the ferromagnetic Heisenberg model, while for negative K and
Γ it maps to the antiferromagnetic Heisenberg model. In the next section, we show
the corresponding spin ordering for the negative K and Γ case, after we transform
the antiferromagnetic ordering in the transformed basis back into the original basis.
2.1.3 Counter-rotating spiral order: signature of SU(2) symmetry in 3D

β − Li2 IrO3
As mentioned above, the SU(2) Heisenberg Hamiltonian is found when K = Γ in

the T12 transformed basis; H e = Je ∑ S e e e
⟨ij⟩ i · Sj with J = −K. For negative K and Γ,
this leads to antiferromagnetic ordering, which corresponds to the counter-rotating
spiral ordering in the original basis for a certain moment direction, as we now show.
It is straightforward to check that the T6a transformation on the sublattice {1a , 2a ,
3a ,4a ,5a ,6a } can be represented by the corresponding set of rotations {E, C2 , C32 ,
C3 C2 , C3 , C32 C2 }, where C3 is a 2π/3 rotation about the (−1, 1, 1)-axis and C2 is a π
rotation about the (1, 0, 1)-axis. These axes are depicted by black arrows on the blue
plane in Fig. 2.2. Similarly, the T6b transformation on the sublattice {1b , 2b , 3b ,4b
,5b ,6b } can be represented by a different set of rotations {C2′ , C2′ C2 , C2′ C32 , C2′ C3 C2 ,
C2′ C3 , C2′ C32 C2 } = C2′ × T6a where C2′ is a π rotation about the (−1, 1, 0)-axis.
Note that the sublattice {1a , 3a ,5a } follows C3 rotations while the sublattice {2a ,
4a , 6a } follows C3 rotations in the opposite order, creating a counter-rotating ordering
pattern. Furthermore, the C2 rotation (1, 0, 1)-axis is perpendicular to the C3 rotation
(−1, 1, 1)-axis, and thus all 6 moments lie inside one plane named A as shown with a
blue plane in Fig. 2.2. This generates a counter-rotating spiral order with wavevector
q = 2π/3 along the chain direction made of the subset-a.
A similar analysis for T6b with the sublattice {1b , 2b , 3b ,4b ,5b ,6b } transformation
could be made. It is the same as T6a , but with an additional π rotation around
the (−1, 1, 0)-axis denoted by C2′ . First, it contains the opposite orderings of C3
rotations within the 6 sites, like in the T6a transformation, and thus it possesses
the counter-rotating pattern. However, due to the additional C2′ rotation around
the (−1, 1, 0)-axis that is applied to all 6 sites, the spiral ordering moments lie in a
different plane named B, colored as a yellow plane in Fig. 2.2. Note that the blue
and yellow planes make an angle of Φ = arccos(1/3) ≃ 70.53◦ as shown in Fig. 2.2.
This result is remarkably close to the angle determined experimentally from magnetic
resonant X-ray diffraction measurements.
Given that the transformation maps into an antiferromagnetic Heisenberg model,
one then applies a bipartite sign factor (−1)n on all sites to generate the entire
magnetic pattern, where n = even for sites 1a , 2b , 3a , 4b 5a , 6b , and n = odd for the
other sites.
Depending on the magnetic ordering moment direction, this transformation gen-
erates several different orderings. The spiral ordering shown by red arrows in Fig. 2.1
occurs when the moment of 1a is pinned along the (1, 1, 0)-direction, which is similar
to the experimentally observed ordering pattern except for small deviations from the
Figure 2.2: Pictorial representation of the T12 transformation carried out on all 12-sites. The planes
where spin moments rotate are indicated by blue (A) and yellow (B) filled circles. Rotation axes are
indicated by black arrows, and the set of red arrows represents an example of moment directions at
the exactly solvable SU(2) point when the moment at 1a,b is pinned along the (1, 1, 0)-direction.
commensurate ordering wavevector.

Below we focus on the parameter space near the hidden symmetric point of the KΓ
model, and perform exact diagonalization on a 24-site cluster to determine nearby
phases. We also show how the magnetic moment direction is pinned when we go
slightly away from the SU(2) point.
2.1.4 Phase diagram and moment direction
The hidden antiferromagnetic SU(2) point has a central role in understanding the
counter-rotating spiral, therefore we explore the parameter space around this point by
using exact diagonalization on the reduced KΓ model of Eq. (2.1). Our calculations
were performed on the cluster shown in Fig. 2.3, which is a minimal cluster that
captures the T12 transformation.
Phase diagram
We normalize the exchange strengths by K = − sin ϕ and Γ = cos ϕ where

ϕ ∈ [π/2, 3π/2]. Phase transitions are identified by singular behavior in the second
derivative of the ground state energy density −∂ϕ2 uGS , where uGS = EGS /N . Four
phases are identified in the region of parameter space following −K → −Γ → K, as
presented in Fig. 2.4. Our results are insensitive to the choice of boundary conditions
as can be seen for periodic (PBC) and open boundary conditions (OBC) in Fig. 2.4(a)
and (b) respectively.
Figure 2.3: N = 24 site cluster used for exact diagonalization. The blue and yellow colors represent
chains where the T6a and T6b are applied respectively. The label na/b with n = 1, ..., 6 indicates
the transformation at each site. Periodic boundary condition (PBC) along the chains is imposed
by connecting the same color and style of the bonds. For different color chains, it is imposed by
connecting sites between na and nb via the Z-bond with the same type black lines. This boundary
condition respects the T12 transformation. For the open boundary condition (OBC) discussed in the
main text, half the Z-bonds are lost, denoted as black lines along the ĉ-direction. The lighter and
darker coloring of both yellow and blue chains indicate the staggering moment convention used in
e2
⟨S stg,T 6 ⟩ discussed in Ch. 2.1.5.
The Kitaev points −K and K exhibit the Kitaev spin liquid state as found in
Refs. [41, 51–53]. While the −K point is immediately unstable upon turning on Γ
interaction, the +K spin liquid, denoted by K, occupies a small area of the parameter
space. The counter-rotating spiral phase, denoted by S, includes a hidden antiferro-
magnetic SU(2) point and is extended from the −K limit to slightly beyond the −Γ
limit. Along −Γ → K, two phases S ′ and S ′′ appear in addition to the Kitaev spin
liquid. The S ′ phase is magnetically ordered as we discuss in the following section,
while the nature of S ′′ is difficult to pin down. To understand the S ′ and S ′′ phases,
we take the chain limit (1D) and study how the phases evolve as the strength of the
Z-bond increases in Ch. 2.1.5. Before that, let us understand how the spin moment
direction is pinned away from the SU(2) point.
24 site PBC 24 site OBC
(a) (b)
Figure 2.4: Phase diagram obtained by exact diagonalization on the 24-sites cluster shown in Fig. 2.3
with (a) PBC and (b) OBC. There are four phases identified. The S phase is the counter-rotating
spiral ordered state, which includes the hidden SU(2) point. The S ′ phase is another magnetically
ordered phase and the nature of the S ′′ phase is discussed in the main text. K refers to the Kitaev
spin liquid phase.
Pinning of magnetic moment direction
The 12-site transformation naturally gives rise to an ordered phase, however, the
precise ordering pattern depends on the magnetic moment direction on sites 1a,b .
Magnetic X-ray diffraction measurements found that the spiral is consistent with a
moment pinned along the (1, 1, 0) direction, as depicted by the red arrows on sites
1a,b in Fig. 2.2. In this section, we determine the preferred moment direction using
exact diagonalization, and discuss the importance of a finite Heisenberg interaction.
To this end, we follow the method used in Ref. [54] and construct a spin-coherent
product state to form the desired magnetic pattern on the cluster. This state is then
used as an ansatz for the ground state wavefunction. In general, for a cluster of N
sites labeled by j, this ansatz is parameterized by 2N parameters (θj , ϕj ) and takes
the form
∏ y
e−iϕj Sj e−iθj Sj |↑⟩j ,
z
|Ψansatz (θj , ϕj )⟩ = (2.4)
j
where there is a spin- 12 coherent state on each site, described by rotations θj and ϕj
about the y- and z-axes respectively. It requires a huge parameter space, however, in
practice one can reduce the set of 2N parameters if the form of the ground state is
known a priori. For example, a ferromagnetic ordering state needs only two param-
eters θ and ϕ since all the spins on all sites are aligned in the same direction. With
this ansatz, one can compute the probability P = |⟨Ψansatz |GS⟩|2 by varying θ and
ϕ, where |GS⟩ is obtained by exact diagonalization. The moment direction is then
determined by the values (θ, ϕ) which maximize P .[54]
In the same spirit, we can build a wavefunction ansatz at the hidden antiferro-
magnetic SU(2) point of the KΓ model on the hyper-honeycomb when K = Γ with
negative K and Γ. In the transformed basis, we only need two parameters (θ, ϕ)
to describe an antiferromagnetic ansatz. We then transform back into the original
basis via the T12 transformation, thereby generating a counter-rotating spiral ansatz.
Specifically, the antiferromagnetic ansatz in the transformed basis takes the form
∏ y y
e−iϕSj e−iθSj |↑⟩
˜ j e−iϕSj+1 e−iθSj+1 |↓⟩
z z
|Ψ̃(θ, ϕ)⟩ = ˜ j+1 , (2.5)
j∈evens
where up and down spins are on even and odd sublattices respectively. Transforming
back into the original basis we determine the counter-rotating spiral ansatz,
∏
N −1
e−iωT12 (i)⃗nT12 (i) ·S
⃗
UT12 = (2.6)
i=0
|Ψansatz (θ, ϕ)⟩ = UT12 |Ψ̃(θ, ϕ)⟩. (2.7)
Multiple consecutive rotations can always be re-written as one rotation, which is

equivalent to saying that rotations form a group. For interpreting the T12 transfor-
mation, the set of rotations given in Ch. 2.1.3 gave us an overview of the behavior of
the transformation, however, carrying out consecutive rotations in numerical calcu-
lations is inefficient and error prone, therefore we re-write the multiple rotations as
only one per site. In Tab. 2.1 the site dependent axis of rotation ⃗nT12 (i) and angle of
rotation ωT12 (i) for every site of the transformation are listed.
As expected, we find that all (θ, ϕ) are equally probable at the hidden SU(2) point,
indicating no preferred moment direction as shown in Fig. 2.5(a). On the other hand,
as one moves away from the SU(2) point, i.e., |Γ| > |K|, the probability is maximized
along the cubic axes, as shown by the representative green arrow in Fig. 2.5(b) for
the (1, 0, 0) direction. Interestingly, a finite positive Heisenberg term is necessary to
pin the direction of the moment along (1, 1, 0) as shown in Fig. 2.5(c). When a small
J > 0 is introduced at the SU(2) point, P (θ, ϕ) forms a ring-like shape normal to the
(−1, 1, 1) axis. This result is similar to a previous study, where it was shown that the
classical ground state solution in this parameter regime exhibits an accidental U (1)
symmetry.[48] In the present quantum calculation, we find that quantum fluctuations
lift this accidental classical degeneracy and selects six equivalent moment directions
connected by C3 rotations about the (−1, 1, 1) axis. Fig. 2.5(c) shows that the higher
Lable Rotations ⃗nT12 (i) ωT12 (i)

a
1 E any axis 0
a
√
2 C2 ( 1, 0, 1)/ 2 π
a
√
3 C32 (−1, 1, 1)/ 3 4π/3
a
√
4 C3 C2 ( 1, 1, 0)/ 2 π
√
5a C3 (−1, 1, 1)/ 3 2π/3
√
6 a
C32 C2 ( 0, −1, 1)/ 2 π
√
1 b
C2′ (−1, 1, 0)/ 2 π
√
2 b
C2′ C2 ( 1, 1, −1)/ 3 2π/3
3 b
C2′ C32 ( 1, 0, 0) 3π/2
4b C2′ C3 C2 ( 0, 0, 1) π
5b C2′ C3 ( 0, 1, 0) 3π/2
√
6 b
C2′ C32 C2 ( 1, 1, 1)/ 3 2π/3
Table 2.1: Axis and angle of rotation used in Eq. 2.6 to create the T12 counter-rotating spiral
ansatz.
probability regions (in red) associated with these moment directions lie on the ring,
with the green arrow pointing in the (1, 1, 0) direction.
(a) (b) (c)
Figure 2.5: Probability maps of P . (a) At the hidden SU(2) point −K = −Γ, no direction is
preferred and the probability is equal for every moment direction. (b) Perturbing away from the
hidden SU(2) point with |Γ| > |K| and J = 0, these bond-dependent terms favor the cubic axis
for the moment direction, indicated by green arrow for the (1,0,0) direction. (c) Perturbing away
from the hidden SU(2) point by J > 0, the quantum fluctuations induced by J lifts the accidental
classical U(1) degeneracy into six directions related by C3 : (1,1,0) (green arrow), (0,1,-1), (-1,0,-1),
(-1,-1,0), (0,-1,1), (1,0,1).
2.1.5 Effects of interchain coupling: evolution from 1D to 3D
As previously stated, in addition to the S phase, which encompasses a large region

of phase space in the KΓ model, we have found the two Kitaev spin liquid phases at
±K and have also found two magnetically ordered phases S ′ and S ′′ . To gain more
insight into the hyper-honeycomb phase diagram, we will investigate the remnants
of these phases in the decoupled 1D chain limit, and gradually couple them until
the 3D hyper-honeycomb structure is recovered. To this end, we have performed
exact diagonalization on the cluster in Fig. 2.3 with varying strengths of the Z-bond
that connect the yellow and blue species of 1D chains. We parameterize the Z-
bonds (leaving X- and Y-bonds the same as before) by Kz = −xK = −x sin ϕ and
Γz = xΓ = x cos ϕ, where x ∈ [0, 1] is the interchain coupling parameter with x = 0
and x = 1 being the 1D and 3D limits, respectively.
The x = 0 limit was studied with exact diagonalization on chains of up to N = 30

sites, and phase boundaries were determined by identifying singularities in −∂ϕ2 uGS .
Our results are shown in Fig. 2.6(a) when Γ < 0. Remarkably, the decoupled chain
phase diagram when Γ < 0 is very similar to the 3D limit. We find four phases in
total, two near the ±K limits and the 1D analogue of the S and S ′ phases, on which
we elaborate below.
The ±K limits, corresponding to the 1D Kitaev spin chain, have been studied
previously in Ref. [55]. In these limits, the model can be solved exactly using a
Jordan-Wigner transformation, and the resulting quasiparticle excitations are Majo-
rana fermions with a gapless dispersion. Here we find that the ferromagnetic Kitaev
limit is immediately unstable upon perturbation by any finite Γ; while by contrast,
the antiferromagnetic Kitaev limit is stable for sufficiently small Γ. This behavior in
1D is in agreement with our results in the 3D limit near the Kitaev limits.
To understand the S and S ′ phases in the 1D limit, we make use of two different
types of sublattice transformations which reveal four points of hidden SU(2) symmetry
in 1D. As alluded to previously, the 6-site transformations T6a and T6b defined in
Ch. 2.1.2 can be applied to each decoupled chain to yield two points of hidden SU(2)
symmetry when ±K = ±Γ. The negative case is in the 1D S phase and corresponds
to the antiferromagnetic Heisenberg limit, where the ground state is quasi-long-range-
ordered owing to the low dimensionality. Special to 1D, we may further define an
a,b
additional set of 6-site transformations T 6 by,
a
T6 (2.8)
sublattice 1a : (S x , S y , S z ) → ( Sex , Sey , Sez ),
sublattice 2a : (S x , S y , S z ) → (−Sez , −Sey , −Sex ),
sublattice 3a : (S x , S y , S z ) → ( Sez , −Sex , −Sey ),
sublattice 4a : (S x , S y , S z ) → ( Sey , Sex , −Sez ),
sublattice 5a : (S x , S y , S z ) → (−Sey , −Sez , Sex ),
sublattice 6a : (S x , S y , S z ) → (−Sex , Sez , Sey ).
b
T6 (2.9)
sublattice 1b : (S x , S y , S z ) → ( Sey , Sex , −Sez ),
sublattice 2b : (S x , S y , S z ) → (−Sey , Sez , −Sex ),
sublattice 3b : (S x , S y , S z ) → (−Sex , −Sez , −Sey ),
sublattice 4b : (S x , S y , S z ) → ( Sex , Sey , Sez ),
sublattice 5b : (S x , S y , S z ) → ( Sez , −Sey , Sex ),
sublattice 6b : (S x , S y , S z ) → (−Sez , −Sex , Sey ).
As a consequence of these transformations, the KΓ model on the chain is mapped

∑ ei · S
e j with J = Γ when ±Γ = ∓K. For the same reasons as with
to H̄ = ⟨i,j⟩ J S
T6a,b , the different subsets of a- and b-type chains are due to the sign convention of X
and Y bonds as shown in Fig. 2.1, and one must apply the appropriate T 6 for each
chain. Thus we have found that the S ′ phase in 1D is ordered due to the point of
hidden ferromagnetic SU(2) symmetry when −Γ = K, with K > 0. It has no net
magnetization, but it possesses a finite sublattice magnetization.
To understand how the S and S ′ phases evolve from 1D to 3D, we compute the
expectation values of the squared total spin in the transformed basis for each of the
phases S and S ′ , denoted by ⟨S e2 e2
stg,T12 ⟩ and ⟨Stot,T 6 ⟩ respectively. In the calculation
of squared total spin for the case of the S phase, we further take care to introduce a
relative staggered sign (−1)i between the two sublattices of the 1D chain in order to
capture the antiferromagnetic nature of the phase owing to the hidden antiferromag-
netic SU(2) point. We plot ⟨S e2 e2
stg,T12 ⟩ (red line) and ⟨Stot,T 6 ⟩ (blue line) in Fig. 2.6(a)
and have normalized to the maximum total value expected for a ferromagnet, i.e.
( + 1). We find that S and S ′ are well differentiated by these order parameters
N N
2 2
and that the S ′ phase saturates the order parameter when K = −Γ, as expected due
(a) x = 0.0 (b) x = 0.2
(c) x = 0.5 (d) x = 1.0
Figure 2.6: Order parameters ⟨S e2 e2 e2

stg,T12 ⟩ (red), ⟨Stot,T 6 ⟩ (blue), and ⟨Sstg,T 6 ⟩ (green) for four values
of interchain coupling X are shown. The dark solid line is the second derivative of ground state energy
with respect to ϕ, and it represents the phase transitions. The hidden antiferromagnetic SU(2) point
from T12 transformation is shown as a filled circle in (a)-(d), and the hidden ferromagnetic SU(2)
point from T 6 as an empty circle in the 1D limit where x = 0.
to the hidden ferromagnetic nature of the phase. Accordingly, the S phase is also
well described, however the staggered squared total spin does not saturate the order
parameter because of the strong antiferromagnetic quantum fluctuations in 1D.
When the interchain coupling is turned on, i.e., x ̸= 0, the T12 transformations
maps the KΓ model to the antiferromagnetic Heisenberg model when K = Γ with K
and Γ < 0, as denoted by a black dot in Fig. 2.6. As a result, the S phase is present
from the 1D limit with x = 0 up to the 3D limit with x = 1. On the other hand,
when x ̸= 0, the T 6 transformations no longer map to the Heisenberg model because

the transformations do not satisfy the Z-bonds except on sites 1a,b and 4a,b . Despite
that, the S ′ phase survives for all x ̸= 0, as shown in Fig. 2.6(b)-(d), even though the
phase space becomes narrower. Due to the frustration of Z-bonds, a new phase S ′′ ,
between the S ′ and the antiferromagnetic Kitaev spin liquid phases appears.
In the S ′′ phase, since the a- and b-subset chains independently possess the hidden
ferromagnetic SU(2) symmetry, the X- and Y-bond interactions try to maximize
ferromagnetic ordering in the transformed basis, when the Z-bond interaction is small,
i.e., x ≪ 1. Thus, we investigate the tendency of ferromagnetic ordering in T 6 basis
along the chains. To avoid the frustration of Z-bond, it is likely that spins on different
chains may orient in opposite directions such as antiferromagnetic ordering between
chains. The antiferromagnetic ordering between chains is denoted by the lighter and
darker colouring of chains in Fig. 2.3. We compute the corresponding staggered
squared total spin ⟨S e2 ⟩. The green line referring to the order parameter is plotted
stg,T 6
in Fig. 2.6(b)-(d), which captures the nature of this ordered phase.
It is worthwhile to compare our results with the classical phase diagram presented
in Fig. 5 of Ref. [43]. Note that the KΓ line is a phase boundary between different
classical ordered states. Thus the quantum fluctuations have stronger effects along
the KΓ line, and may select one of the classical states or a different state from the
classical solutions.
For the negative K, as we presented, the S phase (mapping to SP a− ) occupies a
wide parameter space, including the pure −Γ limit, due to hidden SU(2) symmetry.
For the positive K region, S ′ phase is also one of classical states, corresponding to
SP a+ . On the other hand, S ′′ is different from any of the classical solutions, and
appears near the Kitaev spin liquid, possibly due to strong effects of fluctuations.
However, our cluster size is too small to rule out the possibility that M Q states are
appearing in some phase space of the KΓH model.
Chapter 3
Higher-spin Kitaev models
The prosperous work on candidate materials for S=1/2 Kitaev physics, which in-
cludes the iridates A2 IrO3 with A= Li, Na α−RuCl3 , as well as 3D candidates like
β−Li2 IrO3 has been built on the theoretical foundations of effective model deriva-
tions, as discussed in the introduction. These foundations were laid out in 2008 [2]
and 2009 [6], were it was discovered that large spin-orbit coupling on the transition
metal sites stabilize an pseudospin jeff = 1/2 moment, and the effective interaction
of these moments are bond-dependent interactions including the Kitaev interaction
and symmetric off-diagonal Γ interaction. Groundbreakingly, these works paved had
a path to S=1/2 Kitaev physics in real materials.
Along with the rapid progress on the S= 21 Kitaev spin liquids in solid state ma-
terials [6, 9, 12–14, 25–27, 33, 37, 56–65], the theoretical condensed matter physics
community has considered a higher spin S Kitaev model. A first attempt was made
by Baskaran and collaborators.[66] They showed that for arbitrary spin S, local-
ized Z2 flux excitations are present, as plaquette operators can be constructed, and
a vanishing spin-spin correlation beyond nearest-neighbors is found.[67] Unlike the
S= 12 model, the higher S Kitaev model is not exactly solvable, and several numer-
ical studies have been performed. In particular, the S=1 Kitaev model has been
studied by using exact diagonalization and thermal pure quantum techniques and it
was suggested that the ground state of the S=1 Kitaev model may be a gapless spin
liquid.[68] Using high-temperature series expansions and thermal pure quantum tech-
niques, a double peak structure in the specific heat similar to S= 12 and an incipient
entropy plateau at value of 12 ln3 were found in the S=1 model.[68, 69]. Dynamics
of the classical (S → ∞) Kitaev spin liquid were also studied and it was suggested
that the quantum model can be understood by fractionalization of magnons in one-
dimensional manifolds.[70] While these theoretical results promote another path to
quantum spin liquids, there has been a lack of microscopic routes to achieve a spin S
23
24 CHAPTER 3. HIGHER-SPIN KITAEV MODELS
Kitaev model in solid state materials.

Following the spirit of effective model building as outline in the introduction,
we aim to arrive at a higher spin model mechanism starting from the microscopics.
Stripping away all details, at the heart of the higher spin Kitaev model mechanism
sits the realization that one must forgo spin-orbit coupling on the transition metal
sites and allow Hund’s coupling to stabilize a higher local S moment, while spin-orbit
coupling remains active on the heavy ligand sites surrounding the transition metal. In
this way a higher spin moment can be stabilized unimpeded by virtue of large Hund’s
coupling on the transition metals. Interactions are then mediated through spin-orbit
coupled ligands, allowing bond directionality to be encoded on the resulting effective
interactions. In the following we pursue the above train of thought for S=1 and
S=3/2 systems
3.1 Spin S=1 model from eg orbitals
3.1.1 Kanamori interaction
Consider a 2D edge-shared octahedral system with two types of atoms. The honey-
comb (or triangular) network is made of transition metal (M) cations with half filled eg
orbitals such as d8 electronic configuration. The anion (X) atoms with fully occupied
p-orbitals form edge-shared octahedral cages around every M site as shown in Fig.
3.1. The Hamiltonian consists of the on-site interactions H0 and hopping between M
and X sites, Hkin . For 3d transition metals, such as Ni2+ , typical energy scales of the
hopping parameters are smaller than the energy scales of the on-site H0 , which allows
for the use of standard strong-coupling expansion. The on-site Hamiltonian of both
M and X sites is described by the Kanamori interaction [71] and spin-orbit coupling:
∑ U′ ∑
H0 = U nα↑ nα↓ + nασ nβσ′ + λl · s (3.1)
α
2 α̸=β,
σ,σ ′
JH ∑ † † ∑ † †
− cασ cβσ′ cβσ cασ′ + JH cα↑ cα↓ cβ↓ cβ↑ ,
2 α̸=β, α̸=β
σ,σ ′
where the density operator nασ is given by c†ασ cασ , and c†ασ is the creation operator
with orbital α and spin σ. U and U ′ are the intra-orbital and inter-orbital density-
density interaction respectively, and JH is the Hund’s coupling for the spin-exchange
and pair-hopping terms. Operators l and s respectively denote angular momentum
and spin for orbital α and spin σ, and λ denotes the strength of spin-orbit coupling.
3.1. SPIN S=1 MODEL FROM EG ORBITALS 25
Site M
1 hole
degeneracy=4 EM,1 = Ud + 2Ud′ − JHd + 3εM
2 holes
degeneracy= 3 EM,2,t = Ud′ − JHd + 2εM
degeneracy=1 EM,2,s = Ud′ + JHd + 2εM
degeneracy=1 EM,2,d1 = Ud + JHd + 2εM
degeneracy=1 EM,2,d2 = Ud − JHd + 2εM
3 holes
degeneracy=4 EM,3 = εM
Site X
0 hole
degeneracy = 1 EX,0 = 3Up + 12Up′ − 6JHp + 6εX
1 hole
degeneracy=2 EX,1, 12 = EX,1 + λp
λp
degeneracy=4 EX,1, 32 = EX,1 − 2
1 hole (when λp → 0)
degeneracy=6 EX,1 = 2Up + 8Up′ − 4JHp + 5εX
2 holes (when JHp → 0, Up′ = Up − 2JH )
degeneracy=1 EX,2,1 = 6Up + 2λp + 4εX
degeneracy=6 EX,2,2 = 6Up − λp + 4εX
λp
degeneracy=8 EX,2,3 = 6Up + 2 + 4εX
2 holes (when λp → 0)
degeneracy=1 EX,2,1 = 2Up + 4Up′ + 4εX
degeneracy=2 EX,2,2 = 2Up + 4Up′ − 3JHp + 4εX
degeneracy=3 EX,2,3 = Up + 5Up′ − JHp + 4εX
degeneracy=9 EX,2,4 = Up + 5Up′ − 3JHp + 4εX
Table 3.1: Energy spectrum of the on-site Hamiltonian H0 at the M and X sites. The parameters
′
Ud/p , Ud/p , JHd/p represent the Kanamori interactions for d- and p-orbitals, and λp the spin-orbit
coupling strength of the X site, as discussed in the main text. The atomic potential energy of M/X
sites is denoted as ϵM/X . For the case of two holes at the X sites, the two limits are presented: (1)
Up′ ≡ Up − 2JHp and then JHp → 0, and (2) λ → 0. In the ground state, eg orbitals have 2 holes in
the spin triplet state with energy EM,2,t while p-orbitals have zero holes with energy EX,0 .
In general the competition between the Hund’s coupling and spin-orbit coupling
leads to a different atomic state.[8] For the M sites with eg orbitals, the spin-orbit
coupling is inactive when the crystal field splitting between t2g and eg is bigger than
the spin-orbit coupling strength. Here we consider d8 systems, such as Ni2+ , where
t2g orbitals are fully filled, and the crystal field splitting is larger than the spin-orbit
coupling. In this case, the spin-orbit coupling does not mix the eg states, as the eg
orbitals are made of the z-component of angular momentum of ml = ±2 and 0. In
the half filled eg orbitals the Hund’s coupling selects the total spin S=1 state with
energy U ′ − JH . On the other hand, for the X sites with p-orbitals, the spin-orbit
coupling splits the p-orbitals into total angular momentum j = 23 and j = 12 states.
The Hund’s coupling for X sites is only relevant for excited states in the perturbation
theory. To differentiate U , U ′ , JH for d- and p-orbitals, we use subscript d/p for
′
Ud/p , Ud/p , and JHd/p , which refer to the on-site interactions for d/p-orbitals from now
on. Similarly we use d† and p† to represent creation operators for d- and p-orbital
respectively. For spin-orbit coupling, we have only λp because λd is inactive in the
eg orbitals when the crystal field splitting is larger than the spin-orbit coupling, as is
the case for 3d systems.
3.1.2 Tight-binding Hamiltonian
Let us consider the nearest neighbour hopping parameters between the M and X
sites to construct a minimal nearest neighbor spin model. Since the p-orbitals are fully
filled, and eg orbitals are half filled, we consider holes rather than electrons. Then
in the ground state of the atomic Hamiltonian H0 , there is no hole in the p-orbitals
while eg orbitals are half filled. It is straightforward to build the tight-binding model:
∑
d†i,ασ Mα,β pj,βσ + h.c.,
(i,j)
Hkin = (3.2)
⟨i,j⟩
σ
where d†i,ασ (p†j,βσ ) creates one of d(p)−orbitals denoted by α(β) and spin σ on site
i(j). The hopping matrix M (i,j) depends on the (i, j) bond. As shown in Fig. 3.1,
there are three distinct hopping integrals t1 , t2 and t3 denoted by the red, blue, and
green colored curves respectively. They appear on different bonds.
The hopping matrix M (i,j) on different bonds between one M connecting through
Figure 3.1: Indirect hopping integrals between M and X sites are denoted by the colored curve lines.
The red, green, and blue colors represent t1 , t2 and t3 respectively, and the sign of the hopping
integrals is ignored for simplicity. The M sites with eg orbitals are located in the center of each
octahedral cage formed by X sites occupied by three p-orbitals. Kitaev bond-dependent interactions
X-, Y-, and Z-bond are respectively represented by red, green, and blue shaded regions. For clarity,
every X site is drawn by two separated X sites to represent different hopping contributions from
different p and eg orbtials. The global coordinates of x-, y-, and z-axes are shown in the center of
the honeycomb plane.
six different directions to its surrounding X sites is
Hopping Matrix M(i,j)

( ) ( ) ( )
t1 0 0 0 −t1 0 0 0 0
+x +y +z
−t2 0 0 0 −t2 0 0 0 t3 (3.3)
( ) ( ) ( )
−t1 0 0 0 t1 0 0 0 0
−x −y −z
t2 0 0 0 t2 0 0 0 −t3
written in the basis of {dx2 −y2 , d3z2 −r2 } ⊗ {px , py , pz } in each direction of ±x, ±y
and ±z (xyz axes are shown in the center of honeycomb in Fig. 3.1). The hopping
integrals t1 , t2 and t3 are for dx2 −y2 ↔ px,y , d3z2 −r2 ↔ px,y and d3z2 −r2 ↔ pz hoppings
respectively. The relative signs of the hopping integrals are fixed by symmetry such
as a mirror. The tight-binding parameters can be represented by the Slater-Koster
√
3
parameters [72], i.e., t1 = t ,
2 pdσ
t2 = 12 tpdσ , t3 = tpdσ if the perfect cubic symmetry
is preserved.
3.1.3 Strong-coupling expansion
Treating the tight-binding Hamiltonian Hkin as a perturbation to the on-site Hamil-

tonian H0 , a nearest neighbor spin model for S=1 on the honeycomb lattice with
edge-sharing octahedra via superexchange processes is determined. Before we derive
the model explicitly, it is straightforward to check that the symmetry of the edge-
shared octahedral crystal allows Heisenberg J, Kitaev K, and symmetric off-diagonal
Γ interactions.[9–11].
1 2 1 3
4 3 4 2
3 4
1 2
Figure 3.2: Schematic representation of the three types of superexchange processes that contribute
to the spin interactions. The ring process involves only one hole going around in a loop and related
processes found by permuting pairs 1, 2 ↔ 3, 4. The exchange process exchanges holes between M
sites via an intermediate state where each X site has one hole at the same time, and related processes
found by permuting 1 ↔ 2 and/or 3 ↔ 4. The two-hole process exchanges two holes between M
sites via an intermediate two hole state on the X site, and related processes found by permuting
1 ↔ 2 and/or 3 ↔ 4. All processes have a corresponding mirror processes obtained by permuting
X1 ↔ X2 and/or M1 ↔ M2 .
There are several processes that contribute to the spin interaction and we cate-
gorize them by the number of holes at a given site. The one hole processes include
intermediate states with at most one hole on any X site and the two-hole processes
include intermediate states with two holes on an X site. In the one hole processes, the
spin-orbit coupling λp generates intermediate states of different energies, depending
on whether the one hole state is j = 21 or 32 . For the two-hole process, p-orbital Hund’s
coupling JHp becomes as important as the spin-orbit coupling, and we will consider
two limits of JHp → 0 and λp → 0 to show the origin of Kitaev interaction.
To derive the effective spin model between two M sites, M1 and M2 , one should
consider three types of superexchange processes shown in Fig. 3.2. The ring process
involves one hole that starts from M1 and hops around the ring made of M1 -X1 -M2 -
X2 . The intermediate states involve the creation of a one hole state on the X site
with either j = 21 or 32 . On the other hand, the exchange process exchanges two holes
between two M sites. The intermediate states involve single hole on both X sites. For
these two types of processes, the spin-orbit coupling λp allows different intermediate
paths leading to a finite Kitaev interaction. The two hole process also exchanges two
holes between two M sites, where intermediate two hole states on an X site appear.
Thus these processes involve both λp and p-orbital Hund’s coupling JHp . The Kitaev
interaction appears only when λp ≫ JHp as shown below.
An analytical expression is difficult to achieve when both JHp and λp are compa-
rable due to the complexity of intermediate states when there are two holes in the X
site. We first focus on the Z-bond and consider the two limits: 1. JHp → 0 and 2.
λp → 0, before addressing the X- and Y-bonds.
Z-bond Kitaev and indirect Heisenberg interactions
The Z-bond Kitaev interaction is given by S z S z , and thus the matrix element in
S=1 spin space with two sites are shown in the box below. The basis is chosen as
the on-site total spin S=1 and its S z -component m for two M sites m1 and m2 , i.e,
|S1 , S2 ; m1 , m2 ⟩ = |1, 1; m1 , m2 ⟩ ≡ |m1 , m2 ⟩.
z
Jind = ⟨+1, 0|Hef f |0, +1⟩
K z = ⟨+1, +1|Hef f | + 1, +1⟩ − Jind
z
(3.4)
Γz = −i⟨+1, +1|Hef f |0, 0⟩
1. When JHp → 0:
We present one hole (1h) and two hole (2h) contributions separately. The final spin
interaction is then given by a sum of the two contributions. The one hole processes
contribute the following to the Kitaev term,

[ ( )2 ] ( )2
4( 2 )
2 2 2 2 1 1 1 16 t21 t22 1 1
z
K1h = t − t2 + − + − −
9 1 E 1 , 1 E 21 E 3 , 3 E 23 E1,3 E1 E3 9 E1 E 1 E3
2 2 2 2 2 2 2 2 2 2 2 2
(3.5)
The two hole processes contribute the following

[ ( )2 ]
(
4 2 ) 2 1 1 1 2 2
z
K2h = t1 + t22 + − 2 − 2 (3.6)
9 E2,3 E 1 E3 E 3 E2,2 E 1 E2,1
2 2 2 2
Using the energy spectrum given in Tab. 3.1, we define Ei = Egs − (EM,2,t + EM,1 +
EX,1,i +EX,0 ) with i = 21 , 32 where the ground state energy Egs = 2EM,2,t +2EX,0 . Ei,j =
Egs − (2EM,1 + EX,1,i + EX,1,j ) with i, j = 21 , 32 , and E2,i = Egs − (2EM,1 + EX,2,i + EX,0 )
with i = 1, 2, 3 in the JHp → 0 case.
The indirect Heisenberg term is found to be half of the Kitaev term with a different
sign for both one hole and two holes processes:
1 z 1 z
z
J1h = − K1h z
, J2h = − K2h ; Γz = 0. (3.7)
2 2
When ∆ = εM −εX and Ud are the largest energy scales and assuming cubic symmetry,
i.e., t21 + t22 = t23 , the above expression can be further simplified:
( )
3 24 1 1
K ∼ λp t 3 + . (3.8)
2 (2Ud + ∆)5 2Ud (2Ud + ∆)4
Taking into account a small JHp may change some factor above, but it won’t alter
the main result of a finite Kitaev term.
2. When λp → 0:
In the limit λp → 0, the one hole X site states become six-fold degenerate denoted
by EX,1 in Tab. 3.1, and the Kitaev term vanishes.
z z
K1h = K2h = 0; Γz = 0. (3.9)
The indirect Heisenberg term originating from one hole processes vanishes, however
the two hole contribution survives in this limit leading to the Heisenberg term pro-
portional to the X site Hund’s coupling

[ ]
(t 2
+ t 2 2
) 1 1 (t2 + t2 )2 EX,2,4 − EX,2,3
z
J1h z
= 0, J2h =2 1 2 2 − =2 1 2 2 , (3.10)
EX,1 E2,4 E2,3 EX,1 E2,4 E2,3
where E2,i = Egs − (2EM,1 + EX,2,i + EX,0 ) with i = 3, 4 in the λp → 0 case in Tab. 3.1.
Since EX,2,4 < EX,2,3 the indirect Heisenberg term Jind is ferromagnetic.
X- and Y-bond Kitave and indirect Heisnerberg interactions
Given the symmetry of eg orbitals, X- and Y-bond have the same form of interac-
tion. Considering the Y-bond matrix element shown in the table below, the Kitaev
and Heisenberg interactions are derived.
y
Jind = ⟨+1, +1|Hef f | + 1, +1⟩
K y = 2(⟨+1, 0|Hef f |0, +1⟩ − Jind
y
) (3.11)
√
Γy = 2⟨+1, +1|Hef f |0, +1⟩
Similar to the derivation for the Z-bond above, we present one hole and two hole
y y
contributions separately. Then K y = K1h + K2h , and similarly for the X-bond.
1. When JHp → 0:
[ ( )2 ] ( )2
4 22 2 2 1 1 1 4 t21 t23 1 1
y
K1h = t2 t3 + − + − − .
9 E 1 , 1 E 21 E 3 , 3 E 23 E1,3 E1 E3 9 E1 E 1 E3
2 2 2 2 2 2 2 2 2 2 2 2
(3.12)
The two hole processes contribute the following

[ ( )2 ]
4( 2 ) 2 1 1 1 2 2
y
K2h = 2
t + t2 t3 + − 2 − . (3.13)
9 1 E2,3 E 1 E3 E 3 E2,2 E 21 E2,1
2 2 2 2
The indirect Heisenberg term for the Y-bond is a half of the Kitaev term with the
opposite sign, just like the Z-bond shown above:
1 y 1 y
y
J1h = − K1h y
, J2h = − K2h ; Γy = 0. (3.14)
2 2
Note that with the cubic symmetry, t21 = 34 t23 and t22 = 41 t43 , K x/y = K z , and the
isotropic Kitaev interaction is achieved. When the cubic symmetry is broken, K x/y
deviates from K z , and the above expressions can be used to compute the strength of
each bond interaction.
2. When λp → 0:
Similar to the Z-bond result, the Kitaev term vanishes, i.e.
y y
K1h = K2h = 0; Γy = 0. (3.15)
On the other hand, the indirect Heisenberg term is finite via two hole processes, given
by [ ]
t23 (t21 + t22 ) 1 1
y y
J1h = 0, J2h = 2 − . (3.16)
EX1
2
E2,4 E2,3
With the cubic symmetry the indirect Heisenberg term is isotropic, i.e. J x/y = J z .
The sign of this Heisenberg term is negative, leading to the ferromagnetic Heisenberg
interaction expected from the 90 degree edge-shared superexchange paths.
Results
Taking into account all possible fourth-order superexchange processes, the resulting
nearest neighbor spin model consists of the Kitaev and Heisenberg interactions:
γ
H⟨ij⟩ = K γ Siγ Sjγ + Jind Si · Sj , (3.17)
where i, j are nearest neighbor sites, and γ refers the X-, Y-, and Z-bond type. S is
the spin 1 operator and its bond-dependent interaction couples the γ = x, y, z spin
component. The spin components are directed along the cubic axes of the underlying
ligand octahedra, so the honeycomb layer lies in a plane perpendicular to the [111]
spin direction as shown in Fig. 3.1. Note that Γ term is exactly 0 within the fourth-
order term, and Jind = − 21 K γ .
The expressions of K γ and Jind can be simplified in certain limits. With the cubic
symmetry, i.e., t21 = 34 t23 and t22 = 14 t23 , K x/y = K z ≡ K. When Ud and the atomic
potential difference ∆ between the M (ϵM ) and X (ϵX ) sites (i.e, ∆ ≡ ϵM − ϵX ) are
the largest energy scales, it simplifies to
( )
3 24 1 1
K ∼ λp tpdσ + . (3.18)
2 (2Ud + ∆)5 2Ud (2Ud + ∆)4
2 4 2
3 λp tpdσ 3 tind
For a Mott insulator, i.e., ∆ > Ud , one can further simplify to K ∼ 4 Ud ∆4
≡ 4 Ud
,
λp t 2
where tind = ∆pdσ2 describes the effective hopping between the M and M site via the
X sites. When the cubic symmetry is slightly broken a slight difference between K z
and K x/y appears. The Heisenberg interaction Jind via the superexchange process is
ferromagnetic and its strength is half of K term. Interestingly, Jind is finite when the
large spin-orbit coupling of the anion sites is present, even when Hund’s coupling JHp
is absent. For the other limit of λp → 0, the ferromagnetic Heisenberg interaction

from two hole processes is found and the Kitaev term vanishes.
There is also a direct hopping t between the M sites, which leads to the antiferro-
magnetic Heisenberg term Jdir ∼ 4t2 /Ud . Given the distance between M and X sites
vs. M sites, the direct hopping integral t is an order of magnitude smaller than the
indirect t1 , t2 , t3 hoppings, however, the perturbation process involves second order
terms. Thus the antiferromagnetic Heisenberg term Jdir of similar strength to the
ferromagnetic term Jind may be generated via direct hopping. Since the direct and
indirect Heisenberg terms come with opposite signs, one may expect a small total
Heisenberg interaction J ≡ Jdir − |Jind | and that the Kitaev interaction dominates
the physics of the spin S=1 systems.
(a) (d)
(b) (c)
Figure 3.3: (a) The phase diagram of the S=1 KH model. By tuning the ratio of J/K, two transitions
signalled by the singular behavior of first (blue) and second (red) derivative of the ground state energy
density ugs are found on both 12- and 18-site exact diagonalization clusters shown in (b) and (c),
respectively. In the clusters, X- Y- Z- bonds are shown in red, green and blue respectively, and the
bonds imposing the periodic boundary conditions are dashed. (d) Correlations for the 12 site cluster
discussed in main text. Site 3 is represented as a gray circle, and real space correlations ⟨S3 · Si ⟩
are shown, with i running from 1 to 12. There are three phases identified by spin-spin correlators as
discussed in the main text. The Kitaev spin liquid (SL) appears near J/K ∼ 0, and antiferromagnetic
and zig-zag orderings are respectively found in the antiferromagnetic and ferromagnetic Heisenberg
interaction regions. For J/K = 0.3 the correlations show antiferromagnetic (AFM) behavior with
alternating sign between any nearest neighbor sites, while for J/K = −0.3 the correlations show
zig-zag (ZZ) behavior with the correlations having the same sign within a XY chain and opposite
sign between the chains. The Kitaev point J/K = 0.0 correlations show Kitaev phase behavior with
only nearest neighbor correlations being finite and further neighbors being identically zero.
3.1.4 Exact diagonalization of S=1 KH model
We show that the nearest neighbor spin model of two-electrons in the eg orbitals
surrounded by anions with strong spin-orbit coupling forming edge-shared octahedra
consists of the S=1 Kitaev and Heisenberg interactions. It is worthwhile to check
if the S=1 Kitaev spin liquid survives in the presence of the Heisenberg term. We
carry out exact diagonalization calculations to determine the phase diagram near the
antiferromagnetic Kitaev term. The exact diagonalization results are shown in Fig. 3.3
(a) on two clusters of N=12 and N=18 sites using periodic boundary conditions shown
in Fig. 3.3 (b) and (c), respectively. Phase transitions are identified by the singular
behavior of the second derivative of the ground state energy density (uGS ) with respect
to the variable J/K, i.e., −∂J/K
2
uGS .
Our results show three phases. A finite region of the Kitaev phase around the
antiferromagnetic K point appears with a J/K window of width ∼ 0.06, similar to
the spin- 12 with a J/K window of ∼ 0.07 [37]. While the Kitaev phase occupies a
narrow phase space at zero temperature, it would govern physical properties of finite
temperature and finite magnetic fields, similar to the spin- 12 systems, if the system
can be tuned closer to the Kitaev phase. To clarify the nature of the three phases,
we examine the spin-spin correlations of the three regions using the 12-site cluster,
which are shown in Fig. 3.3 (d). The spin-spin correlation is finite only on the nearest
neighbor bond, and zero for any further neighbors at J/K = 0, consistent with the
pure Kitaev S=1 phase.[66] For J/K ∼ 0.3 we find antiferromagnetic correlations,
while for J/K ∼ −0.3 the zig-zag correlations are present. These magnetically ordered
1
phases match the magnetic orderings found in the spin- and classical spin models.[37]
2
The phase transitions seem to be of first order, but due to finite size effects, an intrinsic
problem of the exact diagonalization technique, further studies are required to pin
down the nature of the transitions.
3.1.5 Kitaev candiate materials
A single layer of NiI2 is a candidate for S=1 Kitaev materials on triangular lattice.
The triangular lattice has X-,Y-, and Z-bond defined similarly to the honeycomb
lattice and the above derivation of the mechanism is applicable. The bulk compounds
form triangular layers of Ni cations and I anions form edge-shared octahedral cages
around Ni. While the bulk NiCl2 is ferromagnetic below 52 K [73], the heavier sister
compound NiI2 has helimagnetic order below 75 K [74, 75]. The helical ordering in the
bulk compound is related to the layer coupling as the ordering wave vector involves
the lattice vector perpendicular to the triangular layer.[75] Thus a single layer of NiI2
3.2. MAGNETIC FIELD EFFECT IN SPIN S=1 35
is desirable to test the dominant Kitaev interaction.

Another group of potential materials is the layered transitional metal oxide com-
pounds A3 Ni2 X′ O6 (A=Li, Na, X′ =Bi, Sb). Unlike the simple binary NiI2 , the M sites
are surrounded by edge-shared oxygen octahedral cages, forming layers of honeycomb
networks sandwiched between layers of the alkali A sites. X′ sites reside in the center
of the honeycomb. A3 Ni2 X′ O6 exhibits zig-zag ordering at low temperatures.[76, 77]
There are two electrons in eg orbitals making total spin S=1 states, a good example
for the proposed mechanism. The strong spin-orbit coupling may occur via proximity
to the heavy X′ atoms. While the oxygen has a weak atomic spin-orbit coupling,
the heavy X′ atoms with strong spin-orbit coupling λX′ induce splitting among the p
orbitals of the oxygen atoms leading to similar effects presented above. For instance,
the effective spin-orbit coupling could be enhanced when one considers hopping be-
tween X′ and O sites denoted by tpp . Using a perturbative approach, the strength
of effective
( spin-orbit )coupling in the p orbitals of O sites is then determined by
t2pp t2
λ̃p ∼ ˜ X′
λ − pp
˜
∆+λ
˜ is an atomic potential difference between X and O
where ∆
∆− X′
2
atoms. While it is difficult to quantify λ̃p in this case, we note that the specific heat
measurements resulting in entropy of ∼ 12 log3 per Ni above the Neel temperature
for both Li3 Ni2 SbO6 and Na3 Ni2 SbO6 [76] strongly hinting that they are promising
candidates for S=1 Kitaev honeycomb materials.
3.2 Magnetic field effect in spin S=1
There are various experimental ways to test the Kitaev interactions in these candi-
date materials. The magnetic field is a way to induce or reveal the Kitaev spin liquids
and its effects have been widely studied in the jeff = 1/2 candidate α−RuCl3 . [78–93]
Applying a magnetic field to the spin S=1/2 Kitaev model leads to a non-trivial
topological state with non-Abelian anyon excitations [94]. This state hosts chiral
Majorana edge states which would give a half-quantized thermal Hall conductance.
[5] This is believed to be related to the observed half-integer quantized thermal Hall
conductivity in α−RuCl3 [15], however the measurements have not been reproduced
yet by other groups.
In the jeff = 1/2 candidates, the Kitaev interaction appears as ferromagnetic.[14]
Under an increasing magnetic field in the ferromagnetic Kitaev model the system
quickly goes from a quantum spin liquid state to a polarized state. [95] Interestingly
in the case of antiferromagnetic Kitaev model a moderate magnetic field leads to an
intermediate state [83], between the topological and polorized states, an emergent
U(1) gapless quantum spin liquid state. [84, 85, 87, 90, 92, 96] This would appear
inaccessible in the jeff = 1/2 candidates. Note however that the S=1 Kitaev materials
suggested here have the antiferromagnetic Kitaev interaction dominant, unlike the
jeff =1/2 Kitaev candidate. Thus, the magnetic field along [111] direction may induce
the U(1) spin liquid with Fermi surface, similar to S=1/2 case [83, 86, 87, 90–92].
We carry out exact diagonalization calculations on the 12 site ladder cluster (clus-
ter in Fig. 3.3 (b) ) to investigate if an intermediate state appears in the S=1 case.
The Kitaev K is set to 1 and we vary the magnetic field strength h, with the field
direction fixed along the [111] direction. The second derivative of the ground state
energy and converged eigenstate energies results are shown in Fig. 3.4. Under a
high field the state eventually polarizes with eigenenergies scaling linearly with field
strength h. Within intermediate values h ∼ 0.35 − 0.45, two peaks enclose an in-
termediate state. In this region, the spectrum collapses to a very dense low lying
manifold. This collapse is reminiscent of the behavior in the intermediate U(1) spin
liquid in S=1/2 [87], and we believe the S=1 also hosts an intermediate state. Several
works have been carried out for the S=1 under magnetic field [19, 97] including work
motivated by these results [18].
Figure 3.4: Exact Diagonalization results on the 12 site ladder cluster for a S=1 Kitaev plus magnetic
field h[111] model, with antiferromagnetic Kitaev K=1. The converged spectrum is shown along
with the second derivative of the ground state energy −∂J/K 2
uGS in blue. Two peaks in the second
derivative signify an intermediate state, with the spectrum collapsing to a very dense low energy
spectrum, reminiscent of the S=1/2 intermediate state.
In the zero field case, the ground state of the 12 site cluster appears singly degen-
erate. However, there is a nearby state, which is then separated by a much larger gap
from the rest of the spectrum. We carry out a system size study by diagonalizing
the Kitaev S=1 model for ladders of 8 to 18 sites, with results shown in Fig. 3.5.
3.3. SPIN S=3/2 MODEL FROM T2G ORBITALS 37
The small gap between the ground and the first excited state decreases with system
size, and in the thermodynamic limit it is expected to be a doubly degenerate ground
state. These two states come with opposite Wilson loop operator ±1, similarly to
the S=1/2 model, which is consistent with the two degenerate ground states on the
cylinder for a Z2 spin liquid state [19].
Figure 3.5: The low lying spectrum of increasing ladder clusters from N=8 to 18. The lowest states
appear non degenerate with their gap closing with increasing system size, while Wilson loop operator
±1 is found to differentiate them.
3.3 Spin S=3/2 model from t2g orbitals
Ferromagnetism in a single-layer CrI3 has generated excitement in recent years.

[98–112] CrI3 is a ferromagnetic insulator with Tc ∼ 61K for bulk samples.[113–115]
Single-layer CrI3 was successfully synthesized, and showed an ferromagnetic order-
ing with Tc ∼ 45K. [116] The two-dimensional ferromagnetic Heisenberg model is
insufficient to explain finite Tc , i.e., the Mermin-Wagner theorem [117], and several
theoretical models were proposed to explain the magnetic anisotropy. They include
the XXZ model [118, 119], single-ion anisotropy and Kitaev [120], and large Kitaev
with small symmetric off-diagonal Γ interactions [121].
While the Heisenberg, Kitaev, and Γ interactions are allowed by symmetry, and
found to be significant in the earlier derivations for lower-spins [6, 9, 17], their
strengths may not be significant in S=3/2 systems. Thus, a microscopic derivation
of S=3/2 model is necessary to find the sources of the magnetic anisotropy. Here we
derive a nearest neighbor spin model for S=3/2 with three electrons in t2g orbitals of
transition metal sites and strong spin-orbit coupling in p-orbitals of ligands.
3.3.1 Kanamori interaction
The honeycomb network is made of metal M site d-orbital sites with half filled
t2g orbitals and octahedra cages of non-magnetic ligand X sites with fully occupied
p-orbitals. The full Hamiltonian is composed of the on-site Kanamori interaction and
the tight-binding Hamiltonian between two sites.
The on-site Hamiltonian of the M sites is described by the Kanamori interac-

tion [71] as well as crystal field splitting:
∑ ∑ U′ ∑
Hee = ∆c c†ασ cασ + U nα↑ nα↓ + nασ nβσ′
α∈eg , α
2 α̸=β,
σ σ,σ ′
JH ∑ ∑
− c†ασ c†βσ′ cβσ cασ′ + JH c†α↑ c†α↓ cβ↓ cβ↑ , (3.19)
2 α̸=β, α̸=β
σ,σ ′
where the density operator nασ is given by c†ασ cασ , and c†ασ is the creation operator
with orbital α and spin σ. U and U ′ are the intra-orbital and inter-orbital Hubbard
interactions respectively, and JH is the Hund’s coupling for the spin-exchange and
pair-hopping terms. ∆C is a crystal field splitting on the M sites, originating from
the surrounding octahedra, leading to the splitting of the d-orbitals into t2g and eg
orbitals. In a d3 system one has half-filled t2g orbitals, where the Hund’s coupling JH
selects for the S=3/2 configuration as the ground state, and the angular momentum
is quenched.
The energies of the exited states are larger than the hopping integrals, which
allows us to treat the tight-binding hopping integrals as a perturbation. A table of
the excited state energy spectrum contributing to exchange processes are shown in
Tab. 3.2. In the lowest energy state there are three electrons at each metal site M.
The exchange process then involves one electron hopping between M sites. Thus the
intermediate states have two electrons in t2g -orbitals on one M site and four electrons
in either t2g - or eg -orbitals on the other M site. On the other hand, the single-ion
anisotropy would follow from a single M site, where the three electrons in the ground
state interact with an exited three electron state. To arrive at the exchange processes
we need to consider the pertubative kinetic term connecting M sites which we do
immediately below.
Degen. Energy Microscopics

2 electrons (only in t2g )
1 EA,2,1 = U + 2JH + 2ϵd
2 EA,2,2 = U − JH + 2ϵd
3 EA,2,3 = U ′ + JH + 2ϵd
9 EA,2,4 = U ′ − JH + 2ϵd
3 electrons (GS)
4 EA,3,gs = 3U ′ − 3JH + 3ϵd
4 EA,3,1 = 3U ′ + 3ϵd
6 EA,3,2 = U + 2U ′ − 2JH + 3ϵd
6 EA,3,3 = U + 2U ′ + 3ϵd
3 electrons (2-t2g , 1-eg )
8 EA,3,e1 = U + 2U ′ − 2JH + ∆c + 3ϵd
4 EA,3,e2 = U + 2U ′ + JH + ∆c + 3ϵd
24 EA,3,e3 = 3U ′ − 3JH + ∆c + 3ϵd
24 EA,3,e4 = 3U ′ + ∆c + 3ϵd
1 EA,4,1 = 2U + 4U ′ + 4ϵd
2 EA,4,2 = 2U + 4U ′ − 3JH + 4ϵd
3 EA,4,3 = U + 5U ′ − JH + 4ϵd
9 EA,4,4 = U + 5U ′ − 3JH + 4ϵd
4 electrons (3-t2g , 1-eg )
4 EA,4,e1 = 6U ′ + ∆c + 4ϵd
6 EA,4,e2 = U + 5U ′ + ∆c + 4ϵd
10 EA,4,e3 = 6U ′ − 6JH + ∆c + 4ϵd
18 EA,4,e4 = U + 5U ′ − 4JH + ∆c + 4ϵd
18 EA,4,e5 = 6U ′ − 2JH + ∆c + 4ϵd
24 EA,4,e6 = U + 5U ′ − 2JH + ∆c + 4ϵd
Table 3.2: Spectrum of Hamiltonian Eq. (3.19). Listed states can contribute to the second order
strong-coupling expansion.
3.3.2 Tight-binding Hamiltonian
In the edge-shared octahedra structure, each bond between nearest neighbor M

sites involves two adjacent ligands, as shown in Fig. 3.6. A tight-binding Hamiltonian
between two transition metal sites M1 and M2 including the two adjacent ligands X1
Figure 3.6: Edge-shared octahedra honeycomb structure unit cell a, b, in global coordinates xyz.
Transition metal sites M in gray and non-magnetic ligands X in purple. The nearest neighbor X-,
Y-, and Z-bonds are related by C3 symmetry. The sites M1 , M2 , X1 , and X2 are involved in the
second order strong-coupling expansion on the Z-bond. Indirect hopping integrals t0 , t1 and t2 are
shown.
and X2 is given below

 
 05×5 TM1 M2 TM1 X1 TM1 X2 
 
 
 TM1 M2 † 05×5 TM2 X1 TM2 X2 
HT B =

,
 (3.20)
 TM X † TM2 X2 † 03×3 03×3 
 1 1 
 
TM1 X2 † TM2 X2 † 03×3 03×3
† †
where 0n×n refers the n×n null matrix. The basis is chosen as (CM , CM , CX† 1 ,p , CX† 2 ,p ),
( ) 1 ,d 2 ,d
†
where CM i ,d
= c†i,x2 −y2 , c†i,3z2 −r2 , c†i,yz , c†i,xz , c†i,xy , are five d-orbitals at site Mi , and
( )
CX† m ,p = c†m,px , c†m,py , c†m,pz are three p-orbitals at ligand site Xm . Each block of
indirect hopping between Mi and Xm is denoted by TMi Xm and the direct hopping
between M sites by TM1 M2 . The details of each block matrix will be presented later.
To account for the indirect d to p hoppings, we integrate out the p-orbitals through
a perturbative procedure truncated at second order, leading to an effective d to d
hopping model:
∑ TMi Xm |a⟩⟨a|TXm Mj
TeffMi Mj = , (3.21)
∆Ea
(a,m)
where (a, m) represent a sum over all single hole states a of all sites Xm . The hole
states are spin-orbit coupled states, thus creating two energy costs ∆Ea = ∆ −
λp /2 or ∆ + λp , where ∆ = ϵd − ϵp is the atomic energy difference between M and
X sites, and λp is the spin-orbit coupling in p-orbitals. The spin-orbit coupling will
introduce explicit spin dependence in the effective d to d hopping. The total effective
hopping between the two M sites now reads
 
eff
 010x10 TM1 M2 + TM1 M2 
HTeffB =  , (3.22)
† eff †
TM1 M2 + TM1 M2 010x10
where we still retain the bare direct hoppings TM1 M2 . Below, we focus on the ideal
honeycomb structure and first examine the effects of TeffM1 M2 before adding the direct
hoppings TM1 M2 and summarizing the resulting spin model.
3.3.3 Ideal honeycomb structure
To understand the microscopic origin of the spin model, we start with the ideal
honeycomb network surrounded by perfect edge-shared octahedra. It was shown
that the symmetry of the edge-shared octahedra Z-bond allows Heisenberg J, Kitaev
K, and symmetric off-diagonal Γ interactions [9–11]. However, since their strengths
depend on various exchange processes, we perform the strong-coupling perturbation
theory to determine the exchange terms.
Truncating at second order in perturbation theory, we arrive at the following
Heisenberg-Kitaev (J − K) spin model for the ideal honeycomb octahedra.
∑
H= J0 Si · Sj + K0 Siγ Sjγ , (3.23)
⟨ij⟩∈γ
where γ = x, y, z bond, and J0 and K0 refer to Heisenberg and Kitaev interactions

for the ideal octahedra.
Below we present the details of the derivation of Heisenberg and Kitaev interac-
tions. An explanation of the absence of the Γ interaction within the second order
perturbation theory is also discussed. The exchange processes include contributions
from both indirect and direct hoppings. We focus on the Z-bond of the honeycomb
Fig. 3.6, as the other two bonds are related by C3 symmetry.
Superexchange path: indirect hopping
We first consider indirect hoppings between the M and X sites, which are the largest
hopping integrals. They enter through the effective hoppings of Eq. (3.22). The non-
zero indirect hopping between M1 and X1 sites t0 , t1 , t2 are shown in Fig. 3.6. These
hoppings are incorporated in Teff
M1 M2 , which can be simplified using the Slater-Koster
decomposition and symmetry related M − X bonds. On the M1 −X1 bond we have
the hopping matrix  
t1 0 0
 
 −t2 0 0 
 
TM1 X1 =   0 0 0 
 (3.24)
 
 0 0 t0 
0 t0 0
The other indirect bonds TM2 X2 , TM2 X1 , TM1 X2 are obtained from TM1 X2 by con-
secutively applying the symmetry operation of the octahedron C4 (0, 0, 1). Eq. (3.24)
and symmetry related bonds are then used in Eq. (3.22) to arrive at effective hop-
pings Eq. (3.27) and Eq. (3.29). Slater-Koster analysis [72] allows us to decompose
t0 , t1 , t2 and t3 into σ- and π-bonding integrals between d- and p-orbitals
√
3tpdσ tpdσ
t0 = tpdπ , t1 = , t2 = , t3 = tpdσ ,
2 2 (3.25)
tpdσ < 0, tpdπ > 0, |tpdπ | < |tpdσ | .
Use of Eq. (3.21) leads to Teff

M1 M2 used in the perturbation calculation. We can split
up the contributions to both Heisenberg and Kitaev interactions into two, i.e., each
interaction is composed of two exchange terms. One is from t2g − t2g hoppings and
the other is from eg − t2g hoppings
t e t e
J0 = J02g + J0 g , K0 = K02g + K0 g , (3.26)
where the superscript t2g and eg indicate its corresponding hopping processes. Below
t t e e
we present each exchange path leading to J02g , K02g , J0 g , and K0 g .
t2g − t2g contributions
( )
t20 2 1
Introducing the effective hopping integral teff = + between
3 ∆ − λp /2 ∆ + λp
2λp
M1 and M2 via p-orbitals, and the ratio r = between spin-orbit coupling λp
2∆ + λp
and the atomic energy difference ∆, the hopping matrix involving only t2g orbitals,
denoted by t0 in Fig. 3.6, can be simplified in block form to

 
r
 02×2 σo i σx 
 2 
 
 

M1 M2 (t2g ⊗ t2g ) = teff 
Teff r 
σo 02×2 −i σy  , (3.27)
 2 
 
 r r 
−i σx i σy 02×2
2 2
where t2g = {dyz , dxz , dxy } and σi with i = x, y, z are the Pauli matrices carrying
the spin degrees of freedom and σo is the 2 × 2 identity matrix. The holes in the
intermediate states at the X site and their indirect hopping TMi Xm determine the
type of σ matrices in the effective hopping. For example, in the limit λp → 0,
only the dyz − dxz term teff σ0 → t20 /∆ is present which contributes to the direct
hopping channel. The new terms present for non-zero λp are the spin-flip (SF) terms
between dyz /dxz and dxy . Such terms would normally not appear in the second order
perturbation process, as they involve a dyz /dxz − pz hopping followed by a px /py − dxy
hopping, which will only occur if pz is entangled with px /py . The spin-orbit coupling
among the px , py , pz generates such an entanglement. Furthermore, when spin-
orbit coupling is the dominant energy scale of the hole states, the wavefunctions are
inevitably mixtures of p-orbitals and their spin, leading to σi dependence proportional
to the r ratio of the spin-orbit coupling and atomic energy difference.
The superexchange process involving only t2g orbitals Eq. (3.27) results in
t 8t2eff t 4(rteff )2 (3.28)

J02g = , K02g = − .
9 (U + 2JH ) 9 (U + 2JH )
The spin-dependent hoppings have generated a Siz Sjz Kitaev interaction. This can
be rudimentarily understood by the following steps. To simplify the steps, we focus
on one spin 1/2 electron hopping along the Z-bond, through only a SF hopping.
Imagining two sites starting in (↑, ↑) state, the SF hopping can lower the energy by
the process: (↑, ↑) − SF − (0, ↓↑) − SF − (↑, ↑), at a energy cost of −(rteff )2 /U . On
the other hand, if the two sites start in (↑, ↓) the SF process is forbidden from Pauli
exclusion principle. Thus, the exchange path starting from (↑, ↑) and ending in (↑, ↑)
lowers the energy, which generates the ferromagnetic Siz Sjz interaction. Carrying out
the details for the S=3/2 leads to the expression of the spin exchange terms shown
in Eq. (3.28).
Among the symmetry allowed terms, the symmetric off-diagonal term with opera-
i( − − )
tor Six Sjy +Siy Sjx = Si Sj − Si+ Sj+ does not occur in the second order perturbation
2
results. This operator would connect (↑, ↑) to (↓, ↓), which is definitely possible from
a SF process, however, there is a subtle cancellation. Having a single Pauli matrix
in the SF term creates such cancellation among the two paths within the spin block,
resulting in a null Γ term. Thus, it becomes finite only when higher order perturba-
tion terms are included, or when the octahedra are no longer ideal, as we will show
in the later Ch. 3.3.4.
eg − t2g contributions
Given that the p-orbital’s hybridization with eg is larger than with t2g , denoted
by t1 and t2 in Fig. 3.6, this contribution is essential. The final form of the effective
M1 M2 (eg ⊗ t2g ) hopping includes SF terms in the dyz , dxz to eg blocks, as well
Teff
as spin-independent −2teff (t2 /t0 )σ0 hopping between dxy and d3z2 −r2 . The effective
hopping matrix involving eg − t2g hoppings, in block matrix form, reads
 
 r t1 r t1 
 i σy i σx 02×2 
TM1 M2 (eg ⊗ t2g ) = teff 
eff

2 t0 2 t0 ,
 (3.29)
 r t2 r t2 t2 
−i σy i σx −2 σo
2 t0 2 t0 t0
where eg = {dx2 −y2 , d3z2 −r2 } and t2g = {dyz , dxz , dxy }. Carrying out the strong-
coupling expansion, these hoppings lead to the additional contribution to the Heisen-
berg and Kitaev interactions
e 16 JH t2eff t22
J0 g = − ,
3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH ) t20
(3.30)
e 2 JH (r teff )2 t21 + t22

K0 g = .
3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH ) t20
Similar to the t2g − t2g case, the SF terms contribute to the Kitaev interaction, while
the spin-independent hopping, −2teff (t2 /t0 )σ0 , generates a ferromagnetic Heisenberg
term. The ferromagnetic Heisenberg interaction originates from the competition of
two excited states separated by Hund’s coupling, i.e, eg paths consistent with the
earlier findings obtained via first principle calculations.[112, 122]
Direct hopping
The direct hopping, denoted by td1 , td2 , td3 and t̃d0 in TM1 M2 between M1 and M2
is given by  
td1 td2 0
 
TM1 M2 (t2g ⊗ t2g ) =  td2 td1 0  ⊗ σo ,
0 0 td3 (3.31)
( )
0 0 0
TM1 M2 (eg ⊗ t2g ) = ⊗ σo .
0 0 t̃d0
The direct hoppings include δ-bonding in addition to σ- and π-bondings
tddπ + tddδ −tddπ + tddδ

td1 = , td2 =
2 √ 2
3tddσ + tddδ 3 (tddσ − tddδ )
td3 = , t̃d0 = ,
4 4
(3.32)
tddσ < 0, tddπ > 0, tddδ < 0,
|tddδ | < |tddπ | < |tddσ | ,
with the Anderson prediction tddσ : tddπ : tddδ = −6 : 4 : −1. Since there are no SF
hopping terms, this does not generate the Kitaev interaction, but does change the
Heisenberg interaction.
Summary and Comments
Combining both indirect and direct hopping contributions, the two exchange in-
teractions for the ideal octahedra environment Eq. (3.23) are found to be
( )2
4 (2t2d1 + 2(teff + td2 )2 + t2d3 ) 4JH 2teff (t2 /t0 ) − t̃d0
J0 = − ,
9 (U + 2JH ) 3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH )
(3.33)
2 2
4 (r teff ) 2 JH (r teff ) t21 + t22
K0 = − + .
9 (U + 2JH ) 3 (∆c + U − JH ) (∆c + U ′ + 3JH )
′ t20
Thus for the ideal honeycomb structure, the J −K model is derived within the second
order perturbation theory.
Some comments are useful, which will also motivate further investigation on the
effects of trigonal distortion presented in the next section. Kitaev and Heisenberg
interactions have been generated while the Γ interaction has not appeared at second
order due to a subtle cancellation. The Kitaev has been generated purely from SF
hoppings and has a prefactor of r2 = (2λp /(2∆ + λp ))2 , while the Heisenberg includes
spin-independent hopping contributions. Nevertheless, both K and J have two con-

tributions which come with opposite signs: one from t2g only paths and the other
from the paths involving eg . The opposite contributions lead to a smaller Kitaev
compared to Heisenberg interaction, unless the contribution from eg paths reduces
the overall strength of the Heisenberg interaction. So long as the crystal field splitting
is not excessively large, the naturally larger eg hoppings will drive the system to a
ferromagnetic Heisenberg interaction.
The ferromagnetic J − K model pins the moment along the cubic axis when the
quantum fluctuations are taken into account [54, 123], not along the observed [111]
direction in CrI3 . We thus investigate if other interaction terms may be generated.
The Γ interaction allowed by the symmetry can be finite if higher order perturbation
terms are included. However, one may ask if there are other interactions allowed by
a slight distortion of the lattice within the second order perturbation theory with-
out invoking higher order terms. Indeed, transition metal trihalide materials do not
have ideal octahedra, but have either rhombohedral R3̄ or monoclinic C2/m struc-
tures, and their magnetism strongly depends on structural differences and number
of layers.[124–127] Below, we study the effects of distorted octahedra, which induce
additional hopping integrals that were forbidden without the distortion.
3.3.4 Effects of distortion: distorted octahdera
CrI3 goes through a structural transition from C2/m to R3̄ structure at low
temperatures.[115] In the rhombohedral structure, there are two types of X ligand
position deviations from the ideal octahedra structure. As shown in Fig. 3.7(a), a
single octahedron can be viewed as two shaded triangles. One distortion is the stag-
gered rotations of the two triangles denoted by δx blue arrows, with displacements
√
of X sites perpendicular to the ĉ = (1, 1, 1)/ 3 direction. The other distortion is the
compression of the distance between these two triangles along the ĉ-axis denoted by
δx′ orange arrows, with displacements along the ĉ direction. Here, the dimensionless
parameters δx and δx′ are in units of the distance between the nearest neighbor M
sites dM . One can track the positions rm of Xm , as it is related to the undistorted
position ro,m , as a function of δx and δx′ :
a) b)
Figure 3.7: a) The distorted octahedra in R3̄ is shown. The √ octahedron made of X ligand can be
viewed as two yellow triangles normal to the ĉ = (1, 1, 1)/ 3 direction. The blue arrows represent
new positions of X due to the staggered rotations of the two yellow triangular faces. The change of
position X due to the staggered rotations is parameterized by δx. In addition, there is a compression
of the two yellow triangles, squeezed as shown by the orange arrows parallel to the ĉ-axis. The change
of position due to the compression is parameterized by δx′ . b) A top view of distortions in a unit
cell is shown. The dotted circles ⊙ indicate the new position of X moving out of the page along +ĉ
while the circled cross ⊗ indicate a new position of X moving into the page toward −ĉ. The exact
positions of X(1,2,3,4,5,6) as a function of δx, δx′ are found in the Appendix.
a
rm = ro,m + δx dM (ξm b
â + ξm b̂) + δx′ dM ζm ĉ
( a b)
m ro,m ξm , ξm ζm
−d
X 1 ( √2 M
0 0 ) ( −1 , −1 ) 1
X2 ( 0 d√M
0 ) ( 1 , 1 ) −1
2 (3.34)
−d
√M
X3 ( 0 2
0 ) ( 1 , 0 ) 1 ,
X4 ( 0 0 d√M
2
) ( −1 , 0 ) −1
−d
√M
X5 ( 0 0 2
) ( 0 , 1 ) 1
X6 ( d√M
2
0 0 ) ( 0 , −1 ) −1
where Xm label sites as in Fig. 3.7(a) and â, b̂ and ĉ are unit vectors along the lattice
vectors of Fig. 3.7(b).
In the R3̄ space group there are other types of distortions, namely the M1 −X2,4,6
bond length can be different from the M1 −X1,3,5 bond length. This type of distortion
is generally exceedingly small compared to δx and δx′ and we neglect it in the following
analysis. Fig. 3.7(b) shows a top view of the the honeycomb unit cell with two
such distorted octahedra forming the Z-bond, and the staggered rotation of the right
octahedron is the mirror image of the left octahedron.
The distortion-induced hopping matrices have all elements non-zero in TMi Xm

as a result of lowering the local symmetry of the octahedron from Oh to D3 . To
get the distortion-induced hoppings one needs to find the directional cosines of the
indirect d − p bonds using the ligand positions Eq. (3.34). With the metal site
located at rM1 and ligand Xm located at rm , the required directional cosines follow
from rm − rM1 . They are the inputs in the Slater-Koster formula for rotated bonds
Ref. [72]. All elements become finite, and the distortion-induced hopping between t2g
and p-oribtials are denoted by δti and eg and p-orbitals by δt′i
 
t1 δt′1 δt′2
 
 −t2 δt′3 δt′4 
 
TM1 X1 =
 δt1 δt2 δt3 ,
 (3.35)
 
 δt4 δt1 t0 
δt5 t0 δt1
where X1 labels the site as in Fig. 3.7(a). The distorted octahedron realizes the
D3 point group, which contains C3 (1, 1, 1) and C2′ (−1, 1, 0) rotations. The hoppings
TM1 X3 and TM1 X5 are recovered by applying C3 (1, 1, 1) to TM1 X1 . Furthermore,
TM1 X1 relates to TM1 X2 by a C2′ (−1, 1, 0). Finally, TM1 X2 relates to TM1 X4 and
TM1 X6 by C3 (1, 1, 1). The direct hopping integrals denoted by TM1 M2 is same as the
Eq. (3.31).
Starting with the distortion-induced hoppings, we follow the procedure described

in Ch. 3.3.3, namely we use the distortion-induced TMi Xm matrices to derive the
M1 M2 (t2g ⊗ t2g ) and TM1 M2 (eg ⊗ t2g ). The distortion induced effective
effective Teff eff
hoppings to leading order in δt and listed in Tab. 3.3.
Due to the distortion and spin-orbit coupling, the single-ion anisotropic term is
also generated via the hopping to anions which can hop back to create spin-dependent
on-site terms denoted by Teff eff
M1 M1 = TM2 M2 . Without the trigonal distortion, the
effective hopping integrals between t2g at M1 are given by
 
4σo irσz −irσy
 
 
Teff
M1 M1 (t2g ⊗ t2g ) = teff  −irσz 4σo irσx  . (3.36)
 
irσy −irσx 4σo
 
TA TC TD
M1 M2 (t2g ⊗ t2g )
Teff = teff  TC† TA U TD U † 
† †
TD U † TD U TB
( )
T̃A −U T̃A U † T̃C
M1 M2 (eg ⊗ t2g )
Teff = teff
T̃B U T̃B U † T̃D
U = e
iπ √12 ( σ2x −
σy
2 ) = √i (σ − σ )
x y
2
( )
δt3 td1
TA = 2 + σo
t0 teff
( )
δt5 td3
TB = 4 + σo
t0 teff
( )
td2 r δta
TC = 1+ σo + i (σx + σy )
teff 2 t0
( )
δtb r δtc δtd
TD = σo + i σx − σy − σz
t 2 t0 t
(0 ′ ) [( 0 ) ]
δt2 δtd t1 r δt3 t1 δt′1 t1 δte t1
T̃A = + 2 σo + i − σx + σy − 2 σz
t0 t0 2 t20 t0 t0 t0
( ′ ) [( ′
) ]
δt4 δtf t2 r δt3 t2 δt3 t2 δta t2
T̃B = − 2 σo + i − σx − σy − σz
t0 t0 2 t20 t0 t0 t20
[( ) ( ) ]
r δt1 t1 δt′ δt5 t1 δt′
T̃C = i 2 − 2 (σx + σy ) − 2 2 − 1 σz
2 t0 t0 t0 t0
( ) ( ′
)
t2 t̃d0 r δt1 t2 δt
T̃D = −2 + σo + i 2 + 4 (σx − σy )
t0 teff 2 t0 t0
Table 3.3: Effective hoppings to leading order in distortion-induced hoppings, where δt are defined
in Eq. (3.35). Distortion induced hoppings are grouped in: δta = δt1 − δt4 , δtb = 2δt1 + δt4 , δtc =
δt3 + δt5 , δtd = δt1 − δt2 , δte = δt1 + δt4 , δtf = δt1 + δt2 .
Similarly, effective hopping between t2g and eg is found as

 
 t1 t1 t1 
 ir σx ir σy −i2r σz 

M1 M1 (eg ⊗ t2g ) = teff 
Teff t0 t0 t0 . (3.37)

 t2 + t3 t2 + t3 
ir σx −ir σy 02×2
t0 t0
They lead to an Si · Si term, which is just equal to S(S + 1), an irrelevant constant to
the spin model. Thus there is no single-ion anisotropy without the trigonal distortion.
t2g ⊗ t2g eg ⊗ t2g
( ) ( )2
4 2t2d1 + 2(teff + td2 )2 + t2d3 4JH 2teff (t2 /t0 ) − t̃d0
J= −
9 (U + 2JH ) 3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH )
32teff td1 δt3 + td3 δt5
+
9(U + 2JH ) t0
4 (r teff )2 2 JH (r teff )2 t21 + t22

K= − +
9 (U + 2JH ) 3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH ) t20
( ) [
8(rteff )2 δtc 4JH (rteff )2 t1 (t1 δt3 − t0 δt′1 )
Γ= − − ′ ′
9 (U + 2JH ) t0 3 (∆c + U − JH ) (∆c + U + 3JH ) t30
]
t2 (t2 δt3 − t0 δt′3 )
−
t30
( ) ( 2 )
′ 4(rteff )2 δtd 2JH (rteff )2 t1 δte t22 δta
Γ = − − − 3 + 3
9 (U + 2JH ) t0 3 (∆c + U ′ − JH ) (∆c + U ′ + 3JH ) t0 t0
( ) [ ( )
4(rteff )2 δtα 4JH (rteff )2 t1 δtβ t1 δt′α
Ac = − − +
(U + 3JH − U ′ ) t0 ∆c (∆c + 3JH ) t0 t20 t0
( )]
t2 + t3 δtβ t2 + 2δt1 t3 δt′β
− +
t0 t20 t0
Table 3.4: Spin model terms, under distortions, to leading order in δt. Distortion-induced δt defined
in Eq. (3.35), and character subscripted δt defined in caption of Tab. 3.3 and Eq. (3.38)
With the trigonal distortion, the single-ion anisotropy along the ĉ-axis is generated:
( ) [ ( )
4(rteff )2 δtα 4JH (rteff )2 t1 δtβ t1 δt′α
Ac = − − +
(U + 3JH − U ′ ) t0 ∆c (∆c + 3JH ) t0 t20 t0
( ) ] (3.38)
t2 + t3 δtβ t2 + 2δt1 t3 δt′β
− + ,
t0 t20 t0
√ ′
3δt4 − δt′2
where δtα = −2δt1 + δt2 + δt3 + δt4 + δt5 , δtβ = δt2 + = δt3 , δt′α −δt′1 + ,
√ ′ 2
3δt2 + δt′4
and δt′β = −δt′3 + .
2
Treating the distortion-induced effective hoppings as a perturbation for the inter-
site terms as well the minimal nearest neighbor spin model is finally given by
∑ [
H= JSi · Sj + KSiγ Sjγ + Γ(Siα Sjβ + Siβ Sjα )
⟨ij⟩∈αβ(γ) ] ∑ (3.39)
+ Γ′ (Siα Sjγ + Siβ Sjγ + Siγ Sjα + Siγ Sjβ ) + Ac (Si · ĉ)2 ,
i
where α, β, (γ) refers to the γ bond taking α and β spin components [9, 57], and
√
ĉ = (1, 1, 1)/ 3. In addition to the J − K terms, two symmetric off-diagonal terms
Γ and Γ′ have been generated, as well as the single-ion term Ac allowed by the
C3 symmetry present on every site. Γ, Γ′ and Ac are proportional to distortion
induced hoppings δt as well as r, thus both distortion-induced hoppings and spin-
orbit coupling is needed to give rise to these terms. The form of the spin model
J − K − Γ − Γ′ − Ac to leading order in δti as well as eg contributions are listed in
Tab. 3.4. Note that both Heisenberg and Kitaev interactions are renormalized by the
distortion, but the Heisenberg interaction has a linear term in δt, while the Kitaev
interaction does not.
3.3.5 Application to CrI3
To apply the above model to understand the magnetism in CrI3 , we use ab ini-
tio calculations to obtain the microscopic parameters. The calculation is performed
with Vienna ab initio simulation package (VASP) [128]. We use the Perdew-Burke-
Ernzerhof (PBE) functional [129] in our calculations. The experimental bulk struc-
ture [115] is used for the calculations. We use WANNIER90 [130] to extract the
hopping integrals.
From the density functional theory results, we estimate the crystal field splitting
∆c = 1253 meV from on-site Cr d-orbitals as well as a crystal field splitting from
the on-site I p-orbitals of ∆p = 528 meV. From Cr d- and I p-orbitals we extract
the atomic energy difference between Cr and I ∆ = 2070 meV. Finally we find the
dominant indirect p − d hopping integrals t0 , t1 and t2 and the direct d − d hoppings
parameters as well as the distortion induced p − d hoppings listed in Tab 3.5. We
verified that t0 , t1 and t2 obtained by the ab initio calculation match well with the
Slater-Koster expectations of Eq. (3.25).
In the ab initio calculations, we find a sizable crystal field splitting ∆p on the
ligand I sites, which we have not taken into account in the earlier analysis. It is
about 528 meV, which is comparable to an estimated spin-orbit coupling parameter
of λp = 630 meV. To capture its effects, we revisit the effective hopping derivation, and
add the crystal field splitting ∆p on the X sites. Then the energy level in Eq. (3.21)
t0 t1 t2 td1 td2 td3 t̃d0

590.1 -992.13 -558.3 44.67 -41.26 -147.44 -20.75
δt1 δt2 δt3 δt4 δt5 δt′1 δt′2 δt′3 δt′4

2.34 21.76 -22.81 -61.74 65.86 70.97 -43.50 -49.41 29.65
Table 3.5: Indirect, direct and distortion induced hoppings in units of meV obtained by ab initio
calculations. The hopping integrals are defined in Eq. (3.24), (3.31) and (3.35).
Figure 3.8: Spin model parameters J, K, Γ, Γ′ , and Ac estimated from ab initio parameters and
plotted against JH /U . Solid, dashed, and dotted lines correspond to U = 3eV, 4eV and 5eV
respectively. Shaded region corresponds to JH /U ∼ 0.24 which is relevant to CrI3 microscopics.
( √ )
split into three levels as ∆Ea = ∆ − λ
2
,∆ + 1
4
2∆p + λ ± (2∆p − λ) + 8λ2 .
2
While the holes in I sites are no longer pure total angular momentum states, the
spin-orbit coupling is sizable enough to carry the spin entanglement through to the
spin model. We found that second order perturbation theory, including the splitting
∆p in the effective hoppings, does not change the form of the spin model Eq. (3.39).
The analytic formulas would be vastly more complex in this case, so we proceed to
a direct numerical evaluation of J − K − Γ − Γ′ − Ac exchange terms using effective
hoppings TeffM1 M2 obtained by the ab-inito calculations listed in the Tables.
Assuming spherical symmetry, i.e., U ′ = U − 2JH , we are left with two unknown
parameters U and JH . We plot J, K, Γ, Γ′ , and Ac as a function of JH /U for several
U values, with results shown in Fig. 3.8. The dominant term is Heisenberg except
near the range 0.11 ≲ JH /U ≲ 0.13 where J is almost zero before it changes the sign.
The sign of J(K) is sensitive to the ratio JH /U , and an adequately large Hund’s cou-
pling allows the ferromagnetic Heisenberg (antiferromagnetic Kitaev) interaction to
persevere. This reflects the competition of the t2g vs eg terms seen in Eq. (3.33), with
eg eventually winning over t2g leading to the ferromagnetic Heisenberg interaction.
The distortion induced Γ−Γ′ −Ac come in with ferromagnetic sign, and the single-ion
anisotropy Ac is the sizable term. Adopting the ratio of JH /U ∼ 0.24 obtained by
cRPA clculations [131], CrI3 sits in the shaded area in Fig. 3.8. CrI3 is a ferromagnet
with non-negligible K − Γ − Γ′ − Ac , and we now examine the implication of these
exchange terms on the observed moment.
3.3.6 Magnetic anisotropy
We have shown how the ideal honeycomb structure leads to J − K model up to

second order in perturbation, while distortions additionally generate Γ − Γ′ − Ac . The
magnetic moment of the ferromagnetic state obtained with the ferromagnetic J − K
model is pinned along the cubic axis such as [100] and C3 equivalent directions via
quantum fluctuations [54, 123]. We take a closer look at the effects of Γ − Γ′ − Ac on
the moment direction. We show three examples in Fig. 3.9 to illustrate the moment
pinning direction.
Following the method in Ref. [54], we perform Exact Diagonalization calculations
on an eight site honeycomb cluster, with the cluster setup shown in Fig. 3.10. Once
the ground state wavefunction |GS⟩ is obtained, the probability distribution P =
|⟨ΨF M (θ, ϕ)| GS⟩|2 is computed, where |ΨF M (θ, ϕ)⟩ is a ferromagnetic ansatz with
moment direction pointing at (θ, ϕ) on the sphere. Results are shown in Fig. 3.9
for different values of J, K, and Γ. Independent of the ratio between J and K, the
moment is along the cubic axis as shown in the panel (a) and (b). On the other hand,
when a small ferromagnetic Γ is introduced, the moment is along the ĉ direction as
shown in panel (c). This effect can be anticipated from the classical analysis. The
classical J − K − Γ model in the ferromagnetic state with moment S0 = (S0x , S0y , S0z )
has an energy density per unit cell uc = (3J + K) + 2Γ(S0x S0y + S0y S0z + S0z S0x ),
which demands min [uc ] ⇒ min [sgn(Γ)(S0x S0y + S0y S0z + S0z S0x )]. When Γ < 0 we
have min [−(S0x S0y + S0y S0z + S0z S0x )] ⇒ S0 = ĉ leading to a moment pinned along the
[1,1,1]. We choose Γ/K ≪ 1 to show that a tiny ferromagnetic Γ anisotropy results
in the ĉ-axis moment shown in Fig. 3.9 (c).
Ferromagnetic Γ′ has the identical effect as ferromagnetic Γ with respect to the
moment direction of the ferromagnet. Clearly ferromagnetic Ac also promotes the
ĉ-axis moment direction. All three interactions individually prefer the moment on
the ĉ-axis. This effect is bigger than the quantum fluctuation effect of cubic axis
pinning from the Kitaev term resulting in the observed ĉ-axis moment direction.
a) b) c)
Figure 3.9: Moment pinning calculation for different values of (J, K, Γ) in the ferromagnetic phase:
a) (−1.00, −0.20, 0.00), b) (−0.20, −1.00, 0.00), c) (−0.20, −1.00, −0.02). The Kitaev interaction
always prefers the cubic axis as shown in panel a) and b). A small interaction like Γ is necessary to
pin along the ĉ direction as show in panel c).
Figure 3.10: X-, Y- and Z-bonds show in red, green and blue respectively. Dashed bonds represent
the periodic boundary conditions.
3.3.7 Spin gap and finite transition temperature
The preceding results suggest that it is likely that CrI3 has dominant ferromagnetic
Heisenberg and smaller Kitaev interaction, and the magnetic anistoropy originates
from the distortion of the octahedra together with the spin-orbit coupling of heavy
ligands. The magnetic ordering pattern changes from bulk to films [116, 127, 131–
134], suggesting a strong-coupling between magnetism and crystal structures, which
further implies the importance of the distorted octahedra in the presence of spin-orbit
coupling. Our microscopic model has an interesting connection to the previous stud-
ies. The Kitaev and single-ion anistoropy in addition to the Heisenberg interaction
were found in Ref. [120] using density functional theory. If Γ = Γ′ , the J − Γ − Γ′
model maps to the XXZ model [118, 119] in the a − b − c crystallographic coordinate
JSi ·Sj +Γ(Six Sjy +Siy Sjx +Six Sjz +Siz Sjx +Siy Sjz +Siz Sjy ) =⇒ Jab Si ·Sj +Jc Sic Sjc , (3.40)
where Jab = J − Γ and Jc = 3Γ.

Since the exchange parameters strongly depend on JH /U , ∆c /U , and tight-binding
parameters, it is useful to obtain experimental inputs to determine some parameters.
The inelastic neutron scattering experiments and magneto-Raman spectroscopy have
reported a spin gap of approximately 0.36 meV at the Brillouin zone (BZ) center
Γ-point.[135, 136] This is also consistent with a small anistoropy found in the ferro-
magnetic resonance experiment [121], which is about 0.07 meV leading to the spin
gap of 0.3 meV.
Based on our spin wave analysis using the J − K − Γ − Γ′ − Ac model including the
second nearest neighbor Dzyaloshinskii-Moriya interaction, the spin wave dispersion
ωk is expressed as ω0 + ρ k 2 around the Γ-point. Here, ω0 and ρ are the spin gap and
stiffness, respectively, and they are given by
ωo = −S(3Γ + 6Γ′ + 2Ac ),

S (K+2Γ−2Γ′ )2

(3.41)
ρ= 3J + K − Γ − 2Γ′ − 2(2Γ+4Γ′ +2Ac +3J+K)
12
The details of the linear spin wave theory (LSWT) are presented in Appx. B.
Kitaev and Dzyaloshinskii-Moriya interactions do not generate a gap at the Γ-
point within the LSWT. The classical ferromagnetic ground state under the Kitaev
and Dzyaloshinskii-Moriya terms have a continuous degeneracy, and as a result ex-
panding around this ground state in the LSWT will not result in a spin gap from
these terms. The spin gap is rather small, which is expected because it originates
from a combination of slightly distorted octahedra and spin-orbit coupling, i.e., Γ,
Γ′ , and Ac . While small, it is essential for a finite Tc in a single layer CrI3 . In the
ferromagnetic ordered phase, at low temperatures the magnons are excited and their
∫ ∫
number is given by Ns (T ) = d2 k eβω1k −1 = βρ π dx
βω0 ex −1
. Without the spin gap ω0 ,
Ns diverges in two-dimensions at any temperature except T = 0, i.e., the celebrated
Mermin-Wagner theorem [117]. Thus, one can understand the essential role of ω0
which cuts the divergence, and allows the ferromagnetic ordering at finite tempera-
tures, as long as ω0 (T ) remains finite for T ≤ Tc . While quantifying the transition
temperature requires further analysis [137], the temperature dependence of ρ(T ) and
ω0 (T ) from the inelastic neutron scattering measurement [135] indicates the crucial
role of ω0 (T ) which vanishes at Tc .
Another important parameter is the Kitaev interaction K, which leads to a gap
at the the BZ corner K-point known as the Dirac gap [121], reported in the neu-
tron scattering.[135, 138] However, the second nearest neighbor Dzyaloshinskii-Moriya
term also generates the Dirac gap.[135, 138] We would like to point out that Γ and
Γ′ also play a part in the Dirac gap as shown in Eq. (B.6). Estimating the Kitaev
interaction by an independent experimental measurement and further analysis on the
individual role of the Kitaev and Dzyaloshinskii-Moriya interactions remain to be
resolved in future studies.
Chapter 4
Ab initio study of frustrated

magnets
4.1 β−Li2 IrO3
The compound β−Li2 IrO3 [29] has attracted attention as a potential platform
for 3D frustrated magnetism and the possibility of a 3D quantum spin liquid. The
iridium sites arrange themselves in a hyper-honeycomb network, a 3D network with
tri-coordination on every site. Every iridium site is surrounded by an octahedron
cage of oxygen atoms, leading to Ir5+ with a jeff = 1/2 state, by ways described in
Ch. 1.1. These jeff = 1/2 moments connect through nearest neighbor edge-shared
octahedra bonds, and one would expect Kitaev as well as Γ interactions to appear as
dominant contributions in the jeff = 1/2 spin model [6, 9]. Experimental evidence has
shown that the compound magnetically orders below 38 K with an unconventional
spiral ordering [29, 30]. However, pressure studies have shown that applying pressure
to this compound results in a sharp disappearance of magnetic moment, while still
remaining insulating.[21, 29]
β−Li2 IrO3 in the low pressure regime, below 4 GPa, crystallizes in the F ddd space-
group, and the Ir sites are arranged in a nearly perfect hyper-honeycomb network.
The three bond types are labeled X-, Y- and Z-bond. This tri-bond configuration
results in similarities to its 2D counterpart, the honeycomb lattice, however, there
is no one-to-one map between these two structures, and the analogy should aid in
an intuitive understanding only. A visual aid is presented in Fig. 4.1 where the
conventional choice and primitive choice of unit cell are shown. The space-group
symmetries demand that X-bonds and Y-bonds be the same length, but Z-bonds can
differ. Nevertheless, X-/Y-bonds end up having nearly identical length with the Z-
bonds. Above 4 GPa, a structural transition takes place, and the compound is found
57
58 CHAPTER 4. AB INITIO STUDY OF FRUSTRATED MAGNETS
Figure 4.1: Hyper-honeycomb structure of β−Li2 IrO3 . The conventional orthorhombic unit cell is
presented on the left, showing the edge-sharing structure of the nearest neighbor iridium bonds. On
the right all sites are stripped except for the hyper-honeycomb iridium network, and embeded is the
primitive unit cell as a yellow prism, with the primitive cell atom basis indicated with pink. The
relationship of the conventional and primative lattice vectors are: α1 = (α2′ +α3′ )/2, α2 = (α1′ +α3′ )/2,
α3 = (α1′ + α2′ )/2
in the C2/c space group. The hyper-honeycomb like network persists, but is far from
ideal since the X-/Y-bonds differ from Z-bond substantially.
The branching ratio (BR) is an observable of X-ray magnetic circular dichroism
measurements, allowing one to resolve magnetic properties of the valence by exciting
core states with circularly polarized light and measuring the excitation intensities.
The measurements over a pressure range show an intriguing behavior, which involves
a sharp drop of the BR around 2 GPa, well before the structural transition at
4 GPa, with the low BR persisting for even higher pressure, all while remaining
insulating.[21, 29]. This opens the possibility of a new state in the 2 − 4 GPa window.
In the following the low pressure region is approached from an ab initio analysis.
4.1.1 Simulating pressure and structual relaxations
For the structural relaxations, we employed the Vienna ab initio Simulation Pack-
age (VASP) version 5.4.3, which uses the projector-augmented wave basis set [139,
140]. A plane wave energy cutoff of 400 eV was used, with PREC set to accurate. For
k-point sampling a 9x9x9 Gamma-Centered grid was adopted for a primitive unit cell.
The revised Perdew-Burke-Ernzerhof generalized gradient approximation [129, 141]
was used for total energy calculations, with a force convergence criteria of 10−3 eV/Å.
In all computations that follow, we use the primitive cell to save on computational
cost.
Pressure in the structure is simulated by adjusting the lattice parameters and
letting only the atomic positions relax, including effects of spin-orbit coupling. In
4.1. β−LI2 IRO3 59
Figure 4.2: Pictorial representation of excitation from p-core orbitals to valence jeff states at five
electron filling per iridium site. The p1/2 to jeff = 1/2 transition are dipole forbidden, leading to an
infinite BR.
Tab. 4.1 we indicate the lattice parameters provided by experiment [21] in the two
cases of ambient pressure and 3.08 GPa pressure. Results of bond angles and bond
lengths are in good agreement with experimental findings. We proceed with the
relaxed structures in all further calculations.
P (Gpa) α1′ α2′ α3′
∼0 5.9104 8.4562 17.8271
3.08 5.8324 8.3110 17.5620
Table 4.1: Lattice parameters refer to the conventional orthorhombic unit cell and are given in unit
of Å.
4.1.2 Branching ratio
The BR is the fraction of two absorption lines, and can be analytically estimated
when working with a single atom and considering dipole transitions. We expect
this approximation to adequately work in the case of the lattice structure since the
process involves local excitations from core ion states to narrow 5d bands. Consider
the iridium core p-orbitals exciting to valence d-orbitals. These are dipole allowed
transitions, so we expect the dipole term to dominate over higher order terms. In
the case of heavy iridium, large spin-orbit coupling will split the p-orbitals into total
angular momentum j = 1/2 and j = 3/2 core states. The excitations from these
two core states to the valence d-orbitals are labeled as the L2 edge (transition from
j = 1/2 to d) and the L3 edge (transition from j = 3/2 to d). The BR is expressed
as the ratio
IL
BR = 3 . (4.1)
I L2
The intensity IL2,3 will be given by the matrix element of the dipole operator
∑
ILj ∝ ⟨Ψ| b q |λj⟩⟨λj|D
D b q |Ψ⟩. (4.2)
λ,q
Normalizing over the total intensity, it can be found that [142, 143]:
ILj 2j + 1 2 − l(l + 1) − c(c + 1)

= ± ⟨L · S⟩, (4.3)
IL2 + IL3 2(2c + 1) l(l + 1)(2c + 1)(4l + 2 − n)
where j = 1/2 ± c is the total angular momentum of the core states (for the L2 edge
j = 1/2 and for the L3 edge j = 3/2), c is the angular momentum of the core states,
l is the angular momentum of the valence states, n is the number of occupied valence
states and ⟨L · S⟩ is the expectation value of the spin-orbit coupling of the valence
states. For core p-orbitals (c = 1), valence d-orbitals (l = 2) and measuring the filling
of the d-orbital valence by nh = 10 − n the BR simplifies to
IL3 2nh − ⟨L · S⟩ 2−r ⟨L · S⟩

BR = = = , r= . (4.4)
IL2 nh + ⟨L · S⟩ 1+r nh
Lets examine two cases relevant to valence d-orbitals. First consider the case of
strong spin-orbit coupling and no crystal field splitting. The d-orbitals would split
into lower energy j = 3/2 states and higher j = 5/2 states. One body expectation
values are:
j(j + 1) − l(l + 1) − s(s + 1)
⟨L · S⟩ = ,
2
3 (4.5)
j = 3/2 ⟨L · S⟩ = − ,
2
j = 5/2 ⟨L · S⟩ = 1.
For a filling of 5 electrons, the j = 3/2 are totally filled and there is a singly occupied
j = 5/2 sate, thus ⟨L · S⟩ = −5. A filling of 5 electrons means there are nh = 5
available states, so BR → ∞.
Now consider the case of strong crystal field splitting under the octahedral sym-
metry group. The d-orbitals split into lower energy t2g and higher energy eg states.
The t2g acts like an leff = −1 manifold, and spin-orbit coupling will split the t2g into
lower energy jeff = 3/2 and higher energy jeff = 1/2. One body expectation values
4.1. β−LI2 IRO3 61
are:
jeff (jeff + 1) − leff (leff + 1) − s(s + 1)

⟨L · S⟩ → − ⟨Lef f · S⟩ = − ,
2
1 (4.6)
jeff = 3/2 ⟨L · S⟩ = − ,
2
jeff = 1/2 ⟨L · S⟩ = 1.
For a filling of 5 electrons, the jeff = 3/2 are totally filled and there is a singly
occupied jeff = 1/2 sate, thus ⟨L · S⟩ = −1. A filling of 5 electrons in the strong
crystal field splitting means eg are not accessible, so there are nh = 1 available states,
thus BR → ∞.
In a typical setting of edge-shared octahedral environments in transition metal
compounds like β − Li2 IrO3 , crystal field splitting is an important energy scale, fol-
lowed by spin-orbit coupling, thus a large BR will be an indicator of the jeff = 1/2
state. A schematic of the jeff BR behavior is shown in Fig. 4.2 with details about the
algebra of calculating the transitions in Appx. C.
4.1.3 Ab initio estimate of the branching ratio
The first question one would ask themselves concerns what happens to the jeff
state under pressure. Is it still reasonable to work within a jeff model as pressure
increases? Experiments found that the BR has a sharp drop from 4.5 to 3, at around
the same pressure as the transition from a magnetic to non-magnetic state [21, 29].
A large BR value is an indicator of the presence of the jeff state (Fig. 4.2), so under
these circumstances one may question the jeff regime as a reasonable model.
Within our ab initio calculations we can estimate the expectation value of the
spin-orbit coupling, and apply the above Eq. 4.4 to find the BR. We use the ab initio
package OpenMX [144–146] which outputs the Bloch wavefunctions in a pseudoatomic
basis. We simply need to compute
∑
occ
⟨L · S⟩ = ⟨ψnk |L · S|ψnk ⟩ (4.7)
nk
We neglect all non iridium atoms, which will have order of magnitude weaker spin-
orbit coupling. We also neglect all orbitals apart from the partially filled d-orbitals.
P (Gpa) Ueff =1.0 eV Ueff =2.0 eV Ueff =2.5 eV

∼ 0 (Paramagnet) 3.51 4.01 4.34
∼ 0 (Magnetic ord.) 3.70 4.32 4.66
3.08 (Paramagnet) 3.46 4.26 4.25
Table 4.2: BR values calculated from ab initio calculations. In all cases nh ≃ 5 as estimated from
density of states.
Finally, we transform into the total angular momentum basis, where the spin-orbit
coupling operator is diagonal. Therefore, we only need to sum the diagonal terms.
The Bloch wavefunctions read as follows
∑ ∑
5/2
5 ∑
3/2
3
|ψnk ⟩ = Ir,d |ϕIr,d ⟩
cnk = mj | , mj ⟩
ank + mj | , mj ⟩,
bnk (4.8)
Ir,d
2 2
mj =−5/2 mj =−3/2
and summing over the diagonal finally gives us

∑ 3 nk 2
⟨L · S⟩ = mJ | − |bmJ | ).
(|ank 2
(4.9)
ϵ<ϵF ,mJ
2
Electronic band structures and BR computation are performed for ambient pres-
sure and P = 3.08 GPa, by using the relaxed atomic positions as input structures
(Ch. 4.1.1). To study the behavior of the BR as a function of electron correlation
and magnetic ordering, several cases are computed and summarized in Tab. 4.2. Both
magnetically ordered and paramagnetic states are converged with varying on-site elec-
tronic correlations. The electronic correlations in OpenMX are incorporated by Du-
darev’s rotationally invariant DFT+U formalism [147] with an effective Ueff = U −JH ,
where U is the on-site Coulomb repulsion and JH is the Hund’s coupling. All results
are above the static value of BR = 2 (the value of BR when ⟨L · S⟩ = 0), signaling
the jeff = 1/2 state under pressure. As a further verification we compute the density
of states of jeff =1/2 and 3/2 states in the band structure, shown in Fig. 4.3. A clear
separation of jeff =1/2 and 3/2 states can be seen for the ambient and pressurized
structure.
There is a slight difference in the BR when only pressure is increased while keeping
Ueff constant. On the other hand, Ueff has a significant impact on the BR. For example,
with U = 1 eV, the BR decreases from 3.51 to 3.46 as P increases from 0 to 3.08
GPa. When Ueff is increased to 2.5 eV, the BR changes to 4.34 and 4.25 for P = 0
and 3.08 GPa respectively. Furthermore, when the magnetic ordering is imposed, the
BR further increases to 4.66 at P = 0 GPa with Ueff = 2.5eV.
While the estimation of effective interaction Ueff is beyond our current compu-
tation, we expect that the Hubbard U decreases under pressure, as β−Li2 IrO3 will
eventually turn to a metal at high pressure. Thus, it is not unreasonable to choose a
4.2. NANI2 BIO6−δ 63
Ambient pressure β-Li2IrO3 Jeff projected DOS 3.08 GPa pressure β-Li2IrO3 Jeff projected DOS
1 Jeff=1/2
0
eV
-1
-2
Jeff=3/2
L Γ Y T Z Γ X A1 Y TX1 X A Z (arb. unit) L Γ Y T Z Γ X A1 Y TX1 X A Z (arb. unit)
(arb. unit)
Figure 4.3: Electronic band structures for β − Li2 IrO3 showing projected jeff = 1/2 and jeff =
2 3 states obtained for P = 0 GPa, Ueff = 2.5 eV (left panel) and P = 3.08 GPa, Ueff = 1 eV
(right panel) (no magnetic order for both cases) and their respective projected density of states.
Experimentally determined lattice unit cells were used for both ambient and high-pressure cases.
The small size of the gap in the calculations in the absence of magnetic order is due to the limitations
of Hartree-Fock in the ab initio calculations. When Mott physics is considered, the gap should be
bigger than shown.
weaker Ueff as pressure increases. Then one can obtain the BR change from 4.66 at
P = 0 GPa (Ueff = 2.5 eV with a magnetic ordering) to 3.46 at P = 3.08 GPa (with
Ueff = 1 eV and no magnetic ordering).
The above analysis explains a visible change in the BR measurements, but cannot
explain the sharp drop around 1.5 GPa. We speculate that such a sudden drop is re-
lated to changes in the strongly interacting electronic state, which cannot be captured
by ab initio calculations. Since this sudden drop accompanies the disappearance of
magnetic order while keeping a finite charge gap (with no crystal structure change),
a 3D spin liquid may be achieved via a pressure induced phase transition.
4.2 NaNi2 BiO6−δ
NaNi2 BiO6−δ is a compound with partial oxygen vacancies and magnetic Ni sites in
a honeycomb network.[148] At low temperatures two phase transitions are observed
at 6.3 K and 4.8 K with neutron scattering showing first an onset of c-axis order
(6.3 K) and then in plane counter-rotating spiral magnetic order (4.8K).[22, 148]
Such a counter-rotating spiral order is consistent with an extended Kitaev model,
however it is not apparent if and how that would occur in this compound. It was
originally proposed that a mixture of 2/3 Ni3+ (S=1/2) and 1/3 Ni2+ (S=1) are
caused by oxygen vacancies.[148] In the following it is argued that the Ni sites are in
the trivalent state Ni3+ and bond-dependent interactions like Kitaev are present in
this material despite the light Ni ions.
4.2.1 Ab initio results
To understand the valence of Ni, we computed the band structure and partial
density of states of NaNi2 BiO6 and NaNi2 BiO5.66 using OpenMX ab-initio package.
OpenMX [149] is a density functional theory code based on the linear combination
of psuedoatomic orbitals formalism.[145] The exchange-correlation potential used is
the Perdew-Burke Ernzerhof generalized gradient approximation.[129, 141] An energy
cutoff of 400 Ry is used for real-space integrations and a 8 × 8 × 8 k-grid is used for
sampling the Brillouin zone.
We found Bi s- and O p-orbitals are heavily mixed most likely due to covalent
bond character. For NaNi2 BiO6 the partial density of states results are shown in
Fig. 4.4, where there is one band with mainly Bi s-orbital character deep below the
Fermi energy around -10.5 eV, while one of O p-orbital band appears above Ni eg
orbitals, leading to 1/4 filling of Ni eg orbitals, in other words d7 . For the valence
of Ni in NaNi2 BiO5.66 , we first triple the unit cell of NaNi2 BiO5.66 and remove one
oxygen, i.e., Na3 Ni6 Bi3 O17 . The partial density of states of Na3 Ni6 Bi3 O17 shown in
Fig. 4.5 has three Bi s-orbitals near -10.5 eV, but only two O p-orbitals above the
eg bands, and one p-orbital below the eg bands. This charge redistribution keeps 1/4
filling of eg orbitals of Ni. For both cases, electron charge is redistributed between
Bi and O in such a way as to keep 1/4 filling of Ni eg orbitals and filled t2g orbitals,
i.e. d7 . This feature is independent of the strength of spin-orbit coupling, and we
propose that all ions are Ni3+ . When Hubbard U and Hunds coupling are included,
i.e, LDA+SOC+U, the system develops a local moment for any finite U, suggesting
that Hunds coupling is strong, and thus a high spin state S = 3/2 is preferred. Note
that eg bands are heavily mixed with O p-orbitals near the Fermi energy, suggesting
that indirect hopping paths are important in determining a microscopic spin model.
4.2.2 Local d7 moment
To understand the local moments on the Ni3+ sites, consider one site of d-orbitals
surrounded by an octahedral environment. The octahedral crystal field splitting in-
duced on the d-orbitals splits them into higher eg and lower t2g orbitals. A subleading
crystal field splitting due to trigonal distortion may be present with an energy scale
that can compete with the small spin-orbit coupling expected of light 3d atoms. The
on-site Hamiltonian is taken to be the Kanamori interaction [71] plus the above crystal
Figure 4.4: Partial density of states of NaNi2 BiO6 . The left panel shows the band structure and the
right panel shows the density of states
field splitting and spin-orbit coupling

∑ U′ ∑ JH ∑ † † ∑ † †
H = U nα↑ nα↓ + nασ nβσ′ − cασ cβσ′ cβσ cασ′ + JH cα↑ cα↓ cβ↓ cβ↑
α 2 α̸=β, 2 α̸=β, α̸=β
σ,σ ′ σ,σ ′
 
0 1 1
∑ ∑ †  
+∆oct nασ + ∆trig cασ (Dtrig )αβ cβσ + λl · s , Dtrig = 1 0 1
α∈eg , α,β∈t2g ,
σ σ 1 1 0
(4.10)
† †
where the density operator nασ is given by cασ cασ , and cασ is the creation operator
with orbital α and spin σ. U and U ′ are the intra-orbital and inter-orbital density-
density interaction respectively, and JH is the Hund’s coupling for the spin-exchange
and pair-hopping terms. ∆oct > 0 is the octahedral crystal field splitting strength
separating eg and t2g orbitals and ∆trig is the subleading crystal field splitting due
Figure 4.5: Partial density of states of Na3 Ni6 Bi3 O17 , which simulates NaNi2 BiO5.66 . The left panel
shows the band structure and the right panel shows the density of states
to trigonal distortion, where ∆trig > 0 describes compressive distortion. Operators l

and s respectively denote angular momentum and spin for orbital α and spin σ, and
λ denotes the strength of spin-orbit coupling.
In the limit of λ → 0 and ∆trig ≪ ∆oct , a S = 3/2, L = 0 configuration is selected
as the ground state. However it is not clear what configuration is preferred when ∆trig
and λ are of comparable size. To study this we diagonalize the above Hamiltonian
Eq. 4.10, parameterized by ∆trig = Acosϕ, λ = Asinϕ, with A = 50 meV, ϕ ∈ [0, π/2],
while fixing the rest of the parameters to U = 6 eV, U ′ = U − 2JH , JH = 1.0 eV,
∆oct = 1.5 eV. The resulting total angular momentum J and decomposed total S
and L moments are shown in Fig. 4.6
⟨ ⟩
Looking at the results we can see that the total Jztotal moment drops below 3/2
when spin-orbit coupling is competing with trigonal distortion, but fails to reach 1/2.
⟨ ⟩
On the other hand, the effective Jzeff moment saturates at a value of 1/2. In the J eff
Figure 4.6: Local moment calculation for N i3+ d7 as a function of ϕ to interpolate between trigonal
distortion ∆trig and spin-orbit coupling λ as ∆trig = Acosϕ, λ = Asinϕ. The left panel shows the
effective moment within the t2g by projecting out the eg orbitals, and the right panel shows the total
moment including eg orbitals.
picture, eg and t2g orbitals are treated as completely separated, with orbital moment
partially surviving in the t2g orbitals, which act like an effective leff = −1 manifold
(Ch. 1.1). In this picture, Hund’s rules plus perturbative spin-orbit coupling give the
J = 1/2 configuration as the ground state [150]. Therefore Jzeff = 1/2 signals the onset
of important contributions from the spin-orbit coupled configuration J = 1/2 in our
ground state wavefunction. In the competing regime, both the S = 3/2 and the spin-
orbit coupled J = 1/2 configurations are mixed into the ground state wavefunction
⟨ ⟩
resulting in an intermediate value of the total measured Jztotal moment.
Although it is difficult to untangle the contributions from the S = 3/2 and the
spin-orbit coupled J = 1/2 configurations, one can consider their effects separately
for a qualitative understanding. The d7 system with the t2g manifold acting as an
effective leff = −1 resulting in a spin-orbit coupled J = 1/2 configuration has been
show to bring about large Kitaev interaction [150]. On the other hand, the S = 3/2
can generate the Kitaev interaction when hopping through intermediate anions with
strong spin-orbit coupling, an effect which can be present by the mixing of O ligands
with the heavy Bi.[17] In either case a large Kitaev interaction naturally arises. The
spiral order can then be captured by the extended Kitaev-Γ-Heisenberg model in the
regime of ferromagnetic Kitaev and small Γ term, with a small antiferromagnetic Γ′
term induced by the trigonal distortion.
4.3 Sr2 IrO4
Controlling properties of strongly spin-orbit coupled solid state materials can be

non-trivial, it is however desirable to be able to predict tunability. Sr2 IrO4 is a Mott
insulator with local jeff = 1/2 moments on account of the heavy Ir+5 atoms. The
nature of the jeff state can lead to unexpected results when microscopic parameters are
tuned through external means like strain. All d-orbitals must be collectively accounted
for in order to cartograph the effects of the multi-orbital and spin entangled jeff
state. Strain will affect the atomic positions, which in turn determine the underlying
hoppings of the valence d-orbitals in Ir. Raman spectroscopy experiments on Sr2 IrO4
under in-plane misfit strain found that the two-magnon peak increases with increasing
compressive strain, which would suggest an increase of exchange interactions.[23]
At first glance this behavior is unexpected. The Sr2 IrO4 compound is a perovskite
compound with the structure shown in Fig. 4.7. Under strain, perovskites tend to
alleviate the strain by decreasing the rotations θ between IrO6 octahedra. This would
lead to hopping integrals, expected to be proportional to sin(θ/2) [151], to decrease
and by extension exchange interactions to also decrease. In the following, we carry
out an ab initio study to understand these observations.
4.3.1 Computational details
For the electronic structure calculations and the atomic position optimization, we
employ the Vienna ab initio Simulation Package, which uses the projector-augmented
wave basis set.[140, 152] The plane-wave cutoff energy used is 500 eV and the k-
point sampling used is a 9 × 9 × 9 Monkhorst-Pack grid for the primitive unit cell.
A revised Perdew-Burke-Ernzerhof generalized gradient approximation is used for
atomic position optimization and total-energy calculations.[153, 154] Optimization is
performed with a force criterion of 1 meV/Å and without any symmetry constraints.
The lattice parameters are kept fixed due to the substrate and atomic position
optimization is carried out for three cases: (1) without spin-orbit coupling or on-site
Coulomb interaction U, (2) including only spin-orbit coupling, and (3) including spin-
orbit coupling and U for the iridium sites. The on-site interactions are incorporated
4.3. SR2 IRO4 69
Figure 4.7: Left: Conventional unit cell of Sr2 IrO4 with the the characteristic corner shared
staggered rotated IrO6 octahedra structure. Right: Top view of one layer of IrO6 octahedra, with
Ir-O-Ir angle θ measuring the rotation angle between octahedra.
using Dudarev’s rotationally invariant DFT+U formalism [147] with effective Ueff =
U −JH = 2 eV, causing the system to magnetically order. Experimentally it has been
found that a weak ferromagnetic ordering is energetically very close to the ground
state of a canted antiferromagnetic ordering when a small magnetic field of 0.2 T
is applied.[155] For the sake of efficiency in structural relaxations, we have used the
primitive unit cell which captures a weak ferromagnetic order.
Using structures with optimized atomic positions, the hopping parameters between
the iridium t2g orbitals are computed by employing maximally localized Wannier
orbital formalism [156, 157] implemented in the WANNIER90 package[130]. Owing
to the layered structure of the material, a unit cell including only one layer of rotated
octahedra is used for converging maximally localized Wannier orbitals.
4.3.2 Results of atomic position optimization and Wannier orbitals
Optimized Ir-O-Ir bond-angles θ without spin-orbit coupling or U, hopping param-

eters, and spin interactions are listed in Tab. 4.3(a). The substrate and the resulting
Sr2 IrO4 lattice parameters a, c from Ref. [159] label each strain case we examine.
The angle θ after optimization is found to correlate predominantly with the variation
t2g hoppings t2g hoppings

a c θ t tz t′ J1 |D| J2
first n.n. second n.n.
(a) Optimized

structures without
 
spin-orbit coupling

or U
−166 −25 0 6 −3 0
s1 5.5875 25.3619 157.03  25 −55 0   −3 43 0  -147 -17 -24 43 10 1.1
 0 0 −221   0 0 −120 
−205 −29 0 4 −4 0
s2 5.5188 25.7022 154.74  29 −54 0   −4 44 0  -157 -19 -25 49 12 1.2
 0 0 −212   0 0 −122 
−211 −25 0 4 −5 0
s2 5.5129 25.7996 154.36  25 −55 0   −5 44 0  -159 -17 -25 50 11 1.2
 0 0 −210   0 0 −122 
−269 −33 0 3 −4 0
s4 5.3979 25.9004 151.76  33 −55 0   −4 45 0  -177 -22 -27 61 16 1.5
 0 0 −206   0 0 −129 
−254 −30 0 3 −4 0
s5 5.4324 25.9709 152.33  30 −54 0   −4 44 0  -171 -20 -26 58 14 1.4
0 0 −205 0 0 −126
(b) Optimized

structures including
 
spin-orbit 
coupling
−211 −27 0 5 −5 0
s2 5.5188 25.7022 155.50  27 −54 0   −5 44 0  -161 -18 -25 51 12 1.3
 0 0 −219   0 0 −124 
−274 −32 0 3 −5 0
s3 5.3979 25.9004 152.15  32 −55 0   −5 44 0  -179 -21 -27 63 15 1.4
0 0 −208 0 0 −127
(c) Optimized

structures including
 
spin-orbit coupling

+U
−206 −26 0 5 −5 0
s2 5.5188 25.7022 155.29  26 −55 0   −5 44 0  -159 -17 -25 50 11 1.2
 0 0 −217   0 0 −123 
−271 −32 0 3 −4 0
s3 5.3979 25.9004 152.08  32 −55 0   −4 45 0  -178 -21 -26 62 15 1.4
0 0 −207 0 0 −127
Table 4.3: Table is split into three sections: lattice optimization results (a) without spin-orbit
coupling or U, (b) including spin-orbit coupling (c) including both spin-orbit coupling and U. First
column is a list of the substrates that Sr2 IrO4 is grown on, with the s labels standing for s1 : GdScO3
(110), s2 : SrTiO3 (100), s3 : (LaAlO3 )0.3 (Sr2 AlTaO6 )0.7 (100), s4 : NdGaO3 (110), s5 : LaAlO3
(110). The second two columns are the imposed lattice parameters a and c used as input in ab
initio calculations. The resulting Ir-O-Ir bond angles θ after atomic position optimization, as well
as t2g hopping parameters between first and second nearest neighbors (n.n.) computed from Wannier
calculations, follow in the next three columns. The t2g hopping parameters are written in matrix
form in the {dxz , dyz , dxy } basis, with diagonal terms being txz , tyz , txy and nonzero off-diagonal
terms being tyzxz/xzyz . Using Eq. 4.3.2 (found in Ref. [158]) the resulting jeff hopping parameters
and spin interactions are given in the last six columns. All energy units are in units of meV and
length units are in Å.
4.3. SR2 IRO4 71
of lattice parameter a, which is consistent with the layered nature of the material,
namely layers of corner sharing octahedra perpendicular to c.
The weak inter-layer coupling results in dispersion along kz being much less then in
the kx , ky plane which further justifies our isolated layer Wannier orbital calculation.
We find hopping parameters between iridium t2g states of first and second nearest
neighbors. The relevant spin model was derived in terms of jeff = 1/2 moments
in Ref. [158] where one can also find formulas of hopping parameters between jeff
moments from the hopping parameters between t2g orbitals, i.e. the following formulas
e , |D| = 8ttz /U
J1 = 4(t2 − t2z )/U e , J2 = 4t′2 /U
e,
(4.11)
t = (txz + tyz + txy )/3, tz = (tyz,xz − txz,yz )/3, t′ = (t′xz + t′yz + t′xy )/3,
where txz , tyz , and txy are the intra-orbital hoppings, tyz,xz/xz,yz is the inter-orbital
hopping, and t′ refers to second nearest neighbors hopping. First nearest neighbor spin
interactions consist of Heisenberg J1 and Dzyaloshinskii-Moriya D terms because of
the absence of an inversion center along the nearest neighbor bonds, however, second
nearest neighbor bonds do have inversion resulting in no second nearest neighbor
Dzyaloshinskii-Moriya term.
Looking at the intra-orbital hoppings under increasing compressive strain, the txy
hopping between dxy orbitals slightly decreases as a result of the decreasing angle
θ. On the other hand the txz (tyz ) hopping between orbitals dxz (dyz ) increases.
This results in the jeff hopping t = (txz + tyz + txy )/3 increasing. At the same time
the inter-orbital hoppings tyz,xz = −txz,yz also increase leading to the jeff hopping
tz = (tyz,xz − txz,yz )/3 increasing. The above means that both J1 and D increase
as a result of the increasing t and tz . One can write the effective SU(2) Heisenberg
√
model J = J12 + D2 because tz can be absorbed into t by a unitary transformation
[160]. The increasing J under compressive strain show qualitatively consistent results
with the experiments.[23] The second nearest neighbor Heisenberg interaction J2 is
of order 1 meV, so we ignored its contributions.
Optimization of atomic positions including effects of spin-orbit coupling and on-site
U make for a heavier calculation, thus we optimize atomic positions for only two cases
of substrates, SrTiO3 and NdGaO3 . Results when including only spin-orbit coupling
are found in Table 4.3(b), and results when including both spin-orbit coupling and
U are found in Table 4.3(c). Comparing these results to the case without spin-orbit
coupling or U, we find no appreciable difference between resulting angles θ and spin
interactions.
4.4 α−RuCl3
Along with Kitaev’s seminal paper [5] a lot of research was poured into material
candidates. After Jackeli and Khaliullin demonstrated how to get this model in d5
Mott insulators [6], where the jeff = 1/2 state forms from strong spin-orbit coupling,
many candidates were proposed, with one of the most promising ones being α−RuCl3
[26, 27, 59–61]. Later, it was shown that two more terms – the off-diagonal symmet-
ric Γ and Γ′ interactions – also arise in these materials. Much like other candidates,
α−RuCl3 orders magnetically at low temperatures.[58, 62–65] This leaves the possi-
bility of tuning the system close to an exotic spin liquid state. Here, the structural
properties and how they tune the spin model interactions is studied.
4.4.1 Computational details
For the electronic structure calculations, we employed the Vienna ab-initio Simu-
lation Package (VASP) version 5.4.4, which uses the projector-augmented wave basis
set [139, 140]. A plane wave energy cutoff of 400 eV was used, with PREC set to
accurate. For k-point sampling a 12x12x1 Gamma-Centered grid was adopted for a
single-layer primitive cell. The revised Perdew-Burke-Ernzerhof generalized gradient
approximation [129, 141] was used for total energy calculations, with a convergence
criteria of 10−8 . Tight-binding parameters were obtained by employing maximally-
localized Wannier orbital formalism [156, 157], on a 6x6x6 k-grid, implemented in
WANNIER90 package [161] version 3.1.0 with a convergence criteria of 10−8 .
4.4.2 Monolayer crystal modeling
The monolayer α−RuCl3 calculations are carried out by introducing an artificially

enlarged c lattice vector unit cell, such that there is a vacuum of more than 20Å
between the layers, while a and b lattice vectors are set from bulk experimental
results. We use the experimental results in Ref. [162] where the bulk α − RuCl3
is found to belong to the R3 space-group. Under this space-group, one does not
necessarily have the ideal edge-shared octahedral geometry of the Cl sites, as it allows
for specific distortions away from this.
The allowed distortions can be separated into two types: staggered rotations and
compression. In Fig. 4.8 a visual representation of the two types of distortions are
shown. It is noted that isolating a layer out of the R3 bulk results in the monolayer
belonging to P 31m space-group, however, this does not alter the allowed distortions
and symmetry analysis of the tight-binding model.
4.4. α−RUCL3 73
Figure 4.8: Pictorial representation on the two types of distortion: staggered rotations shown in
the top row and molded by δα, and compression shown in the bottom row and modeled by δβ. δα
and δβ are magnitudes, while their direction is shown with blue arrows. The lattice parameter is
set to l = |a| = |b|. Crystal structure variables θI and wI are the values in the ideal edge-shared
octahedral geometry, while under distortions we generally have θ, w (see text for detailed discussion
on these values).
In the ideal edge-shared octahedral geometry, one can distinguish two Cl planes,
each forming a triangular lattices, while the two Cl planes sandwich the honeycomb
network of Ru sites, collectively creating the edge-shared octahedral cages. In the
ideal case, each Cl plane has triangular Cl faces with all angles in the triangular face
being π/3, while the angle between adjacent triangular faces is also θI = π/3. The
staggered rotation distortions involve rotations of triangular Cl faces in a staggered
fashion. Lets examine the the top Cl plane in Fig. 4.8. Two neighboring triangular
faces are rotated in opposite directions which maintains π/3 angle within each trian-
gle, while the angles between triangular faces θ deviates from π/3. Given a lattice
parameter l = |a| = |b|, the analytic form is
( )
−1 1 9 δα l
θ = cos − , (4.12)
2 2(9δα2 + l2 − 3 δα l)
where δα is the magnitude of displacement of the Cl sites under staggered rotations,

while the direction of displacement has a complicated site dependent structure shown
in the top row of Fig. 4.8 with blue arrows. Note that flipping the sign of δα simply
reverses all the rotations in opposite orientations (flip all blue arrows). One trivially
recovers the bottom Cl plane’s staggered rotations by performing a C2 rotation along
a − b.
In the ideal case the two Cl planes that sandwich the Ru sites are separated by a
√
width of wI = 2l/3. The compressive distortions involve the two Cl planes moving
closer to each other, such that the width deviates by
w = wI − 2β, (4.13)
where δβ is the magnitude of displacement of the Cl sites under compression, while

the direction of displacement is shown in the bottom row of Fig. 4.8 with blue arrows.
Note that flipping the sign of δβ simply reverses the compression to an expansion (flip
all blue arrows).
Applying this type of analysis to the experimental bulk structure of Ref. [162]
where l = 5.973 Å, one finds that in that case: δα = 0.102 Å, δβ = 0.093 Å. In the
following, we vary δα and δβ to study the effects of these small movements on the
tight-binding parameters.
4.4.3 Hopping integral dependence on distortions: Wannier function es-

timation
dyz dxz dxy

( t1 t2 t4 ) dyz
HTZB = t2 t1 t4 dxz
t4 t4 t3 dxy
HTXB = C3 HTZB C3−1
HTYB = C32 HTZB C3−2

 
0 0 1
C3 =  1 0 0 
0 1 0
Figure 4.9: Conventions used in the model set up. X, Y, and Z label the three kinds of nearest
neighbor bonds. The bond Hamiltonians HTZB ,HTZB and HTZB are related by C3 symmetry.
In Fig. 4.9 one can see the system setup. The three types of nearest neighbor
4.4. α−RUCL3 75
Ru-Ru bonds are labeled X, Y, and Z, while the xyz axes are oriented such that
x(y,z)-axis is perpendicular to X(Y,Z)-bonds. We work in the t2g basis as the valence
electrons of Ru+3 would leave one hole in the t2g manifold. One needs to construct
only the tight-binding term HTZB on a Z-bond, as the other bonds are related by C3
rotations. The symmetry analysis of the tight-binding model is well established [9]
and leads to
dyz dxz dxy

( t1 t2 t4 ) dyz
. (4.14)
HTZB = t2 t1 t4 dxz
t4 t4 t3 dxy
In Tab. 4.4, one can find the hopping integrals calculated from maximally-localized
Wannier orbitals for several values of δα and δβ. In Fig. 4.10 the hopping integrals
are plotted against δβ.
a)
b)
Figure 4.10: Plots of hopping integrals against δβ, where: a) δα = 0.00Å, b) δα = 0.10Å. Energy
units in meV and length units in Å.
δβ \ δα 0.00 0.10
   
45.8 144.4 17.6 34.1 193.2 −11.0
0.15  144.4 45.8 17.6   193.2 34.1 −11.0 
17.6 17.6 −137.9 −11.0 −11.0 −55.2
   
50.4 139.4 12.6 38.2 184.3 −15.7
0.12  139.4 50.4 12.6   184.3 38.2 −15.7 
12.6 12.6 −157.0 −15.7 −15.7 −77.0
   
54.3 133.9 7.9 41.7 175.1 −20.1
0.09  133.9 54.3 7.9   175.1 41.7 −20.1 
7.9 7.9 −174.5 −20.1 −20.1 −97.1
   
57.7 127.7 3.3 44.6 165.5 −24.2
0.06  127.7 57.7 3.3   165.5 44.6 −24.2 
3.3 3.3 −190.5 −24.2 −24.2 −115.8
   
60.5 121.1 −1.0 47.1 155.6 −28.0
0.03  121.1 60.5 −1.0   155.6 47.1 −28.0 
−1.0 −1.0 −204.9 −28.0 −28.0 −133.2
   
62.8 114.1 −5.1 49.2 145.4 −31.4
0.00  114.1 62.8 −5.1   145.4 49.2 −31.4 
−5.1 −5.1 −217.8 −31.4 −31.4 −149.4
   
64.6 106.8 −9.0 51.0 134.9 −34.6
−0.03  106.8 64.6 −9.0   134.9 51.0 −34.6 
−9.0 −9.0 −229.3 −34.6 −34.6 −164.6
   
66.1 99.3 −12.6 52.4 124.4 −37.4
−0.06  99.3 66.1 −12.6   124.4 52.4 −37.4 
−12.6 −12.6 −239.3 −37.4 −37.4 −178.2
   
67.2 91.6 −15.9 53.5 114.1 −39.8
−0.09  91.6 67.2 −15.9   114.1 53.5 −39.8 
−15.9 −15.9 −248.1 −39.8 −39.8 −190.5
Table 4.4: Tight-binding Hamiltonian HTZB of Eq. 4.4.3 as calculated from Wannier functions for
different δα and δβ. Energy units in meV and length units in Å
4.4.4 A leading order analysis
In Fig. 4.10 we plot hopping integrals for several values of δα, δβ. First, we note
that the behavior is monotonic. This is to be expected as the imposed distortions
0.1 Å < δα < 0.0 Å and 0.15 Å < δβ < −0.09 Å are much smaller than the
characteristic length scale dRu−Cl ≃ 2.4 Å. This allows us to do a leading order
analysis to understand the monotonic behavior. For notation brevity we adopt the
4.4. α−RUCL3 77
following notation for the dependence of hopping integrals on distortions t(δα, δβ):
ideal t(i) ≡ t(δα = 0, δβ = 0)
staggered rotations t(r) ≡ t(δα ̸= 0, δβ = 0) (4.15)
compresions t(c) ≡ t(δα = 0, δβ ̸= 0)
Ideal case: δα = δβ = 0
Figure 4.11: Hopping channels in the ideal case for three different view points. Left to right view
(i) (i) (i)
point: generic, (0,0,1), (1,1,0). Top to bottom: t1 , t2 , t3
(i) (i)
Looking at the ideal case first, t1 and t3 are determined from their Slater-Koster
(i)
analysis, while t2 is dominated by the effective hopping path dyx − pz − dxz as seen
in Fig. 4.11. The Slater-Koster decomposition reads
(i) tddδ + tddπ

t1 = ,
2
(i)
(i) tddδ − tddπ (tπ )2 (4.16)
t2 = + ,
2 ∆
(i) tddδ + 3tddσ

t3 = ,
4
where tddσ(π,δ) are the independent Slater-Koster hopping integrals for d − d hoppings,
(i)
tπ < 0 is the π Slater-Koster hopping integral for d−p, and ∆ > 0 is the atomic energy
difference between p- and d-orbitals. Under distortions, dRu−Ru is fixed, therefore tdd
need not carry the identifier (i) as they do not change.
The δ hopping integral tddδ will be much smaller than tddπ and tddσ . Under this
(i) (i)
observation, t1 ∼ tddπ /2 > 0. The hopping parameter t2 is heavily renormalized
by the indirect virtual process dyz − pz − dxy giving a large positive contribution
(i) (i)
(tπ )2 /∆ > 0. The hopping parameter t3 is dominated by the large σ integral
tddσ < 0 which is negative. Therefore, we expect t1 , t2 > 0, t3 < 0 and t2 , |t3 | > t1
which is indeed observed in Fig. 4.10.
Staggered rotations case: δα effects
Under staggered rotations the bond length dRu−Cl becomes smaller for the two Cl
sites that are adjacent to both Ru sites of the bond. As a result of dRu−Cl becoming
(r) (r) (r)
shorter, tπ will slowly increase (see top row Fig. 4.12), resulting in t2 ∼ (tπ )2 /∆
increasing. Indeed, t2 is increasing as a function of δα in Fig. 4.10.
t1 and t3 were set by the d − d Slater-Koster hoppings in the ideal case, therefore
their leading order corrections will arise from new d − p hopping channels tn , that
become allowed under staggered rotations. From the plethora of new paths, the
leading order effects will come from virtual paths that involve one tπ hopping and
one tn hopping. This leading order term is shown in Fig. 4.12.
(r)
The tn1 coming from pz − dxy (see top row Fig. 4.12) is originally forbidden in
(r)
the ideal case. Under staggered rotations, we expect tn1 > 0 as it involves overlap
(r)
of opposite sign orbital lobes. This results in tn1 having a leading order correction
(r) (r)
tπ tn1 /∆ < 0, thus reducing the value of t1 as a function of δα, which is indeed
(r)
observed in Fig. 4.10. Focusing on t3 now, the tn2 coming from px −dxy (see Fig. 4.12)
(r)
is originally forbidden in the ideal case. Under staggered rotations, we expect tn2 < 0
(r)
as it involves overlap of same sign orbital lobes. This results in t3 having a leading
(r) (r) (r) (r)
order correction tπ tn2 /∆ > 0. Since t3 starts off as a negative value, the tπ tn2 /∆ > 0
4.4. α−RUCL3 79
will make the absolute value of t3 smaller. Furthermore, due to the directional nature
of p- and d-orbitals, looking at Fig. 4.12 one would expect |tn2 | > |tn1 |, so the leading
order effect for t3 will be much more drastic than the one we saw for t1 . This is indeed
observed in Fig. 4.10.
Figure 4.12: Hopping channels in the staggered rotations case for three different view points. Left
(r) (r) (r)
to right view point: generic, (0,0,1), (1,1,0). Top to bottom: t1 , t2 , t3 . Only leading order
corrections are shown (see discussion in main text).
Compression case: δβ effects
Under compression, the bond length dRu−Cl again becomes smaller. Although exact
values of the new allowed hopping integrals will vary from the staggered rotations
case, the same arguments presented above also apply here for the paths shown in
Fig. 4.13, and we qualitatively expect similar behavior: t1 and |t3 | drop while t2 rises
as a function of δβ.
The behavior of t4
Hopping integral t4 is expected to become non-zero only in the distorted case.

(i)
However, in Fig. 4.10 we clearly see t4 ̸= 0. The fine point here is symmetry. It is true
(c) (c) (c)
to right view point: generic, (0,0,1), (1,1,0). Top to bottom: t1 , t2 , t3 . Only leading order
corrections are shown (see discussion in main text).
that looking only at one isolated bond, and assuming ideal octahedra, the symmetry
(i)
forbids t4 because of three perpendicular C2 rotations of the bond. However the
space-group symmetry of the monolayer only has one C2 out of the three of the
(i)
isolated bond. As a result, t4 is symmetry allowed. Since the single bond will not
(i) (i)
allow t4 from symmetry, then t4 is set by several higher order virtual paths that go
beyond the bond, thereby making its value small. This reasoning also makes clear why
(i)
the value of t4 is hard to pin down, as many higher order subleading contributions
add up. A more rigorous treatment is needed to account for all these subleading
(i)
paths, which is beyond the scope of our analysis. Having said that, assuming t4
acquires some non-zero value, the functional behavior under staggered rotations and
compression can still be understood from a leading order contribution.
The leading order corrections will involve a tπ hopping and the largest subleading
hopping present only under distortions tn . The active leading correction under
distortions can be seen in the dxy − px − dxz paths with subleading hopping tn3
(r)
(see figure Fig. 4.14). Under rotations we expect tn3 > 0 as it involves overlap of
4.4. α−RUCL3 81
(r) (r)
opposite sign orbital lobes. This results in the leading order correction tπ tn3 /∆ < 0,
thus driving t4 into negative values as a function of δα, which is indeed observed
(c)
in Fig. 4.10. Under compressions however, we expect tn3 < 0 as now it involves
the overlap of same sign orbital lobes. This results in the leading order correction
(r) (r)
tπ tn3 /∆ > 0, thus driving t4 into positive values as a function of δβ, which is indeed
observed in Fig. 4.10.
to right view point: generic, (0,0,1), (1,1,0). All cases are about t4 corrections, with top to bottom:
ideal, stagered rotations, compressions. Only leading order corrections are shown (see discussion in
main text).
4.4.5 Comparison to the bulk
In Tab. 4.5 we compare the hopping integrals of the monolayer to the bulk for four
distortion cases. Note that the bulk is in the R3 space-group. Lack of rotations allow
for tyz−yz ̸= txz−xz and tyz−xy ̸= txz−xy , thus introducing t1 , t′1 and t4 , t′4
dyz dxz dxy

( t1 t2 t4 ) dyz
(4.17)
HTZB,bulk = t2 t′1 t′4 dxz
t4 t′4 t3 dxy
A similar argument can be made as in Ch. 4.4.4 subsection “The behavior of t4 ”

about these differences coming from higher order processes going beyond the bond,
thus making t1 ≃ t′1 and t4 ≃ t′4 as observed. We see that the values of the artificiality
isolated monolayer and the bulk, given the same distortion parameters, are very close.
This justifies use of the monolayer results in the following section.
(δα, δβ) monolayer bulk

   
62.8 114.1 −5.1 73.7 116.1 2.4
   
(0, 0)  114.1 62.8 −5.1   116.1 56.2 −9.1 
−5.1 −5.1 −217.8 2.4 −9.1 −212.6
   
49.2 145.4 −31.4 66.2 145.9 −28.3
   
(0.10, 0)  145.4 49.2 −31.4   145.9 35.6 −29.4 
−31.4 −31.4 −149.4 −28.3 −29.4 −145.2
   
54.3 133.9 7.9 64.0 135.0 12.1
   
(0, 0.09)  133.9 54.3 7.9   135.0 49.4 6.4 
7.9 7.9 −174.5 12.1 6.4 −171.6
   
41.7 175.1 −20.1 55.3 175.5 −20.3
   
(0.1, 0.09)  175.1 41.7 −20.1   175.5 31.9 −16.2 
−20.1 −20.1 −97.1 −20.3 −16.2 −94.6
Table 4.5: Comparison of tight-binding between the monolayer and the bulk. Monolayer values
taken directly from Tab. 4.4.
4.4.6 Spin model
The spin model can be derived from a strong-coupling expansion, where one treats
the on-site Hubbard U and Hund’s coupling JH as dominant terms, and the tight-
binding as perturbing terms. We use the derived formulas for the case of one hole
t2g orbitals from Ref. [9]. Setting JH /U = 0.15, the resulting spin model parameters
are shown in Tab. 4.6 and the data is plotted in Fig. 4.15. Since inversion symmetry
is present the Dzyaloshinskii-Moriya term would never appear at nearest-neighbor,
however, the symmetric anisotropic Γ and Γ′ terms are allowed. Given the analysis of
the tight-binding parameters in the previous section one can understand the behavior
of the spin model parameters as discussed in Ref. [9].
4.4. α−RUCL3 83
a)
b)
Figure 4.15: Plots of J, K, Γ, Γ′ against δβ, where: a) δα = 0.00 Å, b) δα = 0.10 Å. Energy units
in meV and length units in Å. Plotting data from Tab. 4.6.
The parameters depend strongly on the compression and staggered rotations, and
offer a tuning mechanism for the underlying spin model. Although J and Γ appear
more robust to the distortions, owing to the strong t3 dependence, K and Γ′ are very
susceptible and can even switch sign. Since J remains small, this allows one to tune
the system close to the Γ limit. Specifically in the region of δα = 0, δβ = 0.09 there is
a dominant Γ term along with a small antiferromagnetic Γ′ , opening the possibility for
a chiral spin liquid.[163] Stabilizing a monolayer and controlling its distortions could
be achieved with a substrate, and further studies of spin interactions for a mono-layer
α−RuCl3 on various substrates are necessary.
δβ(Å) J(meV) K(meV) Γ(meV) Γ′ (meV)
δα = 0.00Å
0.15 −1.760 −2.647 5.133 0.418
0.12 −2.084 −1.411 5.544 0.254
0.09 −2.394 −0.114 5.842 0.129
0.06 −2.678 1.204 6.029 0.042
0.03 −2.930 2.511 6.111 −0.010
0.00 −3.145 3.776 6.096 −0.030
−0.03 −3.322 4.978 5.993 −0.023
−0.06 −3.461 6.100 5.813 0.009
−0.09 −3.564 7.130 5.567 0.061
δα = 0.10Å
0.15 −0.504 −9.852 3.316 −0.514
0.12 −0.911 −8.361 4.107 −0.655
0.09 −1.295 −6.798 4.736 −0.740
0.06 −1.645 −5.198 5.215 −0.774
0.03 −1.954 −3.589 5.557 −0.763
0.00 −2.217 −1.993 5.772 −0.710
−0.03 −2.434 −0.433 5.869 −0.621
−0.06 −2.597 1.040 5.856 −0.507
−0.09 −2.711 2.404 5.745 −0.372
Table 4.6: Spin model parameters calculated from Tab. 4.4 using formulas from Ref. [9].
Chapter 5
Conclusions
In this chapter, I will summarize the main results of my thesis, and list a few open
questions.
The exotic nature of the Kitaev spin liquid has motivated the community to re-
alize the phase in candidate materials from the 2D honeycomb materials A2 IrO3
(A = Li, Ni) and α-RuCl3 to the 3D hyper- and harmonic-honeycomb materials
β−, γ− Li2 IrO3 . Despite this intensive search, all these materials are magnetically
ordered and, in particular, an intriguing counter-rotating spiral is found in the Li-
based iridates. It is, therefore, important to understand the role of the competing
bond-dependent interactions in the minimal nearest neighbor and potentially further
neighbor interactions. The bond-dependent interactions are ubiquitous in Mott insu-
lators with edge-shared octahedral environment and strong spin-orbit coupling. This
is because strong spin-orbit coupling mixes different orbitals and spin components at
a given site, and bond-dependent spin interactions rely on the hopping integrals of
the bond, whose size is determined from the overlap of relevant orbitals. However, the
approach taken in the compass [6] and generic spin model [9] does not work for higher
spins. Hunds coupling, which maximizes the total spin, is necessary for a higher spin
such as S=1 from two electrons in eg or S=3/2 from three electrons in t2g . On the
other hand, the bond-dependent interaction requires a mixture of spins via spin-orbit
coupling, which works against the Hund’s coupling. Thus, it has been unclear how
to generate a higher spin Kitaev model.
In the first part of this work, we focused on a counter-rotating non-coplanar spiral
order in 3D iridates β-Li2 IrO3 . We make use of a 12-site transformation (T12 ) that
gives an exactly solvable point of hidden antiferromagnetic SU(2) symmetry when
−K = −Γ. Using this as our guide, we have been able to construct the minimal
cluster necessary to use in exact diagonalization to understand the counter-rotating
spiral phase as well as nearby phases. In addition to the Kitaev spin liquid phase,
85
86 CHAPTER 5. CONCLUSIONS
three spiral ordered phases S, S ′ and S ′′ are found. Apart from the S ′′ phase, each
of these phases has a 1D analogue if we tune the hyper-honeycomb structure to the
limit of decoupled chains. The S phase is a counter-rotating spiral ordered state,
which is a direct consequence of the T12 sublattice transformation. Furthermore, at
−K = −Γ, when a small antiferromagnetic J is added, the local moment is pinned
to in the [1,1,0] direction or its symmetry equivalents. These results are consistent
with the magnetic structure determined experimentally by magnetic resonant X-ray
diffraction on β-Li2 IrO3 . The nearby S ′ and S ′′ phases are also interesting. We see
that the S ′ phase is a spiral whose pattern is determined by another set of 6-site
a,b
transformations T 6 , which give rise to two new points of hidden SU(2) symmetry
when ±Γ = ∓K in the decoupled chain limit. However, the mappings of Heisenberg
a,b
model by T 6 transformations work only for isolated chains, and the Z-bonds are
frustrated when the chains are coupled. As a result of Z-bond frustration, another
ordered phase S ′′ , distinct from S and S ′ , emerges between S ′ and Kitaev spin liquid
phase. The phase transition between S ′′ and the spin liquid phase is rather unclear
and may be related to finite size clusters. Note that the 24-site exact diagonalization
with OBC in Fig. 2.4(b) shows a clearer signature of a phase transition than that
of PBC in (a). It will be interesting to study the effects of anisotropy by cutting
Z-bonds in both S ′′ and the Kitaev spin liquid phases in the 3D system, which may
change the transition between the two, similar to a study from the positive Γ to a
negative Kitaev point in 2D honeycomb lattice.[164]
β-Li2 IrO3 is likely within the magnetically ordered S spiral phase. The mapping to
the Heisenberg model by a T12 sublattice transformation is exact at ±K = ±Γ without
J. The corresponding ordered S phase indeed captures the spiral ordering, but how
the S phase becomes incommensurate in the presence of J is not clear due to the
finite size of the cluster, a problem that plagues all finite size exact diagonalization
calculations. Previous classical studies have shown that the S spiral phase, when
considering only negative K and Γ terms, has a constant ordering vector of (2/3, 0, 0)
and a finite J is enough to induce incommensuration. [41] To push the system into
a spin liquid state within a KΓ model with a small J, one has to shift towards the
relatively stable +K point. Going beyond nearest neighbors J, a second nearest
neighbor Heisenberg J2 term has also been found to cause incommensuration [44,
165] with an estimated J2 close to a spin-liquid phase [44]. It was reported that
applying hydrostatic pressure causes the magnetic order to disappear for pressure
above ∼ 1.5−2 GPa, while a structure transition occurs at a higher pressure, above ∼
4 GPa.[21, 29, 166] However, a follow-up work suggested that the structural transition
actually occurs at around 1.5 GPa for sufficiently cooled samples below 50 K.[167].
87
Instead of hydrostatic pressure, uniaxial stress may be a way to effectively tune the
exchange interactions in the laboratory setting, which remains a challenging task.
In the second part of this work, a S=1 Kitaev interaction is derived via superex-
change processes between half-filled eg orbital metals mediated by p-orbital ligands
with strong spin-orbit coupling, by using a standard strong-coupling expansion. We
find the dominant interaction is the antiferromagnetic Kitaev term, whose strength is
twice as big as that of the ferromagnetic Heisenberg interaction. Taking into account
the direct exchange process that results in an antiferromagnetic Heisenberg term, we
expect that the Kitaev interaction dominates spin physics of these Mott insulators.
A small region of S=1 Kitaev phase with only nearest neighbor spin-spin correlation
is found in 12- and 18-site exact diagonalization calculations. A finite ferromagnetic
Heisenberg interaction stabilizes the zig-zag magnetic ordering nearby the spin liquid.
The analysis presented in the current work is extended to a higher spin S=3/2 Ki-
taev model. For example, Cr3+ leaves three electrons in the t2g orbitals creating spin
S = 3/2 via Hund’s coupling, and the superexchange processes via strong spin-orbit
coupled ligands lead to the Kitaev term. It is found that there are only Heisenberg
and Kitaev interactions for the ideal honeycomb lattice among the three symmetry
allowed interactions (J, K, Γ), because Γ is zero up to the second order perturbation.
The exchange paths between t2g and t2g vs. t2g and eg via ligands generate opposite
signs for both Heisenberg and Kitaev interactions. The Heisenberg interaction is of
order t2eff /U , while the Kitaev is smaller by a factor of r2 ∼ (λp /∆)2 . The ferromag-
netic Heisenberg interaction is originated from the eg paths, with the hopping integral
between eg and p-orbitals being larger compared to t2g and p-orbitals.
S=1 Kitaev materials include a single layer of NiI2 on the triangular lattice and
A3 Ni2 X′ O6 with X′ =Bi, Sb and A=Li, Na on the honeycomb lattice. A group of van
der Waals transition metal halides, such as MX2 and MX3 , provides a rich family of
magnetic materials. The dihalides MX2 and trihalides MX3 are made of triangular
and honeycomb networks of transition metal cations respectively, surrounded by edge-
shared ligands X.[168] When X is heavy, the strong spin-orbit coupling at X sites
plays a role in the magnetic mechanism presented here. Theoretical studies on these
magnetic materials have been limited to the first, second, and third nearest neighbor
Heisenberg model. We propose that one revisits these layered 3d transitional metal
compounds with edge-shared heavy ligands from a new perspective of bond-dependent
interactions. Furthermore, note that the S=1 Kitaev materials suggested here would
have a dominant antiferromagnetic Kitaev interaction, unlike the jeff = 1/2 Kitaev
candidates like α−RuCl3 that have a dominant ferromagnetic Kitaev interaction.
Thus, the magnetic field along the [111] direction may induce the proposed U(1) spin
88 CHAPTER 5. CONCLUSIONS
liquid with spinon Fermi surface, similar to the 1/2 case [87]. Theoretical studies of
the S=1 Kitaev model with a magnetic field were motivated by this idea.[18, 19]
In the S=3/2 case from half-filled t2g orbitals, the ferromagnetic Heisenberg and
Kitaev interactions lead to ferromagnetic ordering, but the moment direction is pinned
along the cubic x-, y-, or z-axis, e.g., [100] (and C3 equivalent directions) via quantum
fluctuations. The [111] moment pinning found in CrI3 should, thus, originate from
other interactions, which are also responsible for the spin gap at the Γ-point in the
neutron scattering measurements [135]. Including the distorted octahedra present in
the rhombohedral structure, three additional spin interactions are found, Γ, Γ′ and
single-ion anisotropy Ac containing terms linear in the distortion-induced hopping
integrals. The Heisenberg interaction is heavily modified by distortions, but the
Kitaev is not, implying that it is possible to fine-tune a system closer to the Kitaev
dominant regime via trigonal distortions. Focusing on the monolayer system of CrI3 ,
the t2g −t2g vs. eg −t2g contributions to the intra-layer Heisenberg term J are opposite
in sign. This property should hold, when considering a bulk CrI3 , for inter-layer
Heisenberg interaction J⊥ , because it is determined from superexchange processes.
Thus, the magnetic ordering pattern between layers depends on the details of orbital
compositions. From the ab initio calculation, we found that the inter-layer hopping
ranges from 10 meV to 30 meV. This leads to J⊥ of order 0.1 meV. While it does
not affect the intra-layer magnetism presented here, it is important for the ordering
pattern in the bulk [116, 127, 131–134].
In this thesis it was demonstrated that spin-orbit coupling of X ligands can lead to
higher spin S bond-dependent interactions between nearest neighbor M sites. How-
ever, other interactions of the X ligands are not fully appreciated yet. Since the heavy
X ligand is at the core of this work, one wonders about the role of e-e interactions at
the ligand site, and their interplay with spin-orbit coupling. After all this has proven
to be the case in jeff = 1/2 spin model derivations [6, 9] with Kitaev interaction,
and other bond-dependent interactions, appearing as proportional to the Hund’s cou-
pling of the transition metal. It is thus reasonable to expect a similar effect from
the heavy X ligand Hund’s coupling. The heavier X atoms will have more extended
atomic wavefunctions, and one may also consider enhanced hopping paths. Hopping
enhancement depends on the system details as seen in α−RuCl3 , where our ab initio
study of small distortions enhances certain indirect hopping channels leading to a
significant tunability of the jeff = 1/2 spin model. Further neighbor exchange inter-
actions may become relevant because of enhanced hoppings through several X ligand
sites, and hopping through spin-orbit coupled ligands could enhance certain bond-
dependent interactions. This can be relevant in CrI3 where a symmetry allowed, next
89
nearest neighbor, Dzyaloshinskii-Moriya interaction has been used to explain nuclear

magnetic resonance results [135], and quantifying the interaction strength may be
important to determine the origin of the measured Dirac gap. The origins of effective
interactions is a natural research question that will continue to shape our funda-
mental understanding of materials, and I hope my thesis will contribute to further
developments in this research direction.
Appendix A
Exact diagonalization
The holy grail of any physics problem is finding and characterizing the spectrum of
the Hamiltonian. If we can do that, then we know everything about the system. This
is akin to saying “I want to find the eigenvalues and eigenvectors of the Hamiltonian”.
Very few problems can have an exact analytical solution, and even fewer can have a
sufficiently simple solution to extract much intuition from. It is therefore desirable
to try and numerically solve the Hamiltonian. Exact diagonalization is one method
to solve the problem that tries to approximate extremal eigenvalues and eigenvectors
by use of the Lanczos algorithm.
Starting with our Hamiltonian H, we aim to transform it into a tridiagonal matrix.
Tridiagonal matrices enjoy a plethora of specialized numerical eigenvalue solvers that
are very efficient, and so long as we bring it in such a form, we can let mathematical
software libraries like LAPACK take over the diagonalization. So we are looking for
 
a0 b0
 
 b0 a1 b1 
 
 b1 ... ... 
U HU = T = 
†

,

 ... ai bi 
  (A.1)
 bi ... ... 
...
†
ai ∈ R, bi ≥ 0, U U = I,
U = (|u0 ⟩ , |u1 ⟩ , ..., |ui ⟩ , ...).
U , ai , and bi can all be found from the relation
U T = HU ⇒ |ri ⟩ ≡ bi+1 |ui+1 ⟩ = H |ui ⟩ − ai |ui ⟩ − bi−1 |ui−1 ⟩ . (A.2)
Suppose |u0 ⟩ is chosen randomly, and let us define b−1 = 0. Then everything is
90
91
determined because from the above relation Eq. A.2 it follows that
⟨ui |ri ⟩ = 0 ⇒ ai = ⟨ui |H|ui ⟩ ,

√
⟨ri |ri ⟩ = b2i ⇒ bi = ⟨ri |ri ⟩, (A.3)
|ui+1 ⟩ = |ri ⟩ /bi .
Given a |u0 ⟩, and b−1 = 0, the above equations for i = 0 determine a0 , then r0 , then
b0 , and then |u1 ⟩. From here we can now work out i = 1, and so on and so forth.
This is the Lanczos algorithm. If we carry it out for dim(H) iterations we could
tridiagonalize the entire Hamiltonian. In pseudocode the algorithm is carried out as
follows
Lanczos pseudocode
1. u0 = normalized random vector

2. r = Hu0
3. a0 = u0 · r
4. r = u0 − a0 ∗ r
√
5. b0 = r·r
(A.4)
start for i=1,2,...
6. ui = r/bi−1
7. r = ui−1 − bi−1 Hui
8. ai = ui · r
9. r = r − ai ∗ r
√
10. bi = r·r
end for
In real problems, the iteration is not carried out to dim(H), but is rather stopped
when a sufficient number of eigenstates have converged. Intuitively, we are find-
ing better and better approximations to the extremal eigenvalues as the iterations
progress. This follows from the fact that every iteration involves a H |ui ⟩ oper-
ation. If we were to unfold the recursive iteration at the ith step, it would in-
volve a factor of H i |u0 ⟩. Decompose |u0 ⟩ in the eigenvectors |ψm ⟩ of H, to get
∑ ∑
H i |u0 ⟩ = m=0 cm H |ψm ⟩ =
i
m=0 cm εm |ψm ⟩. Thus, the largest eigenvalues εm
i
are coming in as successively larger polynomial powers, and eventually the largest
eigenvalues will be singled out of the sum as the largest contributions. Inevitably the
Lanczos algorithm selects for the extremal eigenvalues.
The Lanczos algorithm has one big problem. In exact arithmetic it all pans out
nicely as above, however, in carrying it out on a computer, we have to deal with finite
92 APPENDIX A. EXACT DIAGONALIZATION
precision arithmetic and rounding errors. It turns out this algorithm is numerically
unstable, and this manifests by loss of orthogonality in the |ui ⟩ vectors. A simple but
costly remedy is to insert an extra reorthogonalization step between 9 and 10
Reorthogonalization pseudocode between 9 and 10
start for j=0,...,i-1

′ (A.5)
1 . q = ui · r
2′ . r = (r − q ∗ ui )/(1 − q)
end for
As the iterations i progress the number of j reorthogonalization steps increase, making

this a very expensive procedure overall. However, the loss of orthogonality also has
an upside. Assume we are carrying out the Lanczos procedure, and at the nth step
we get bn = 0. This means we have reached an invariant subspace of the Hilbert
space spanned by u0 , ..., un , and further iterations are now ill defined. One would
have to stop, find a new u′0 vector orthogonal to the previously found subspace, and
then continue with a new Lanczos procedure. In finite precision arithmetic with the
loss of orthogonality, the likelihood of this occurring is practically zero. The code can
keep on executing, occasionally running into small but not exactly zero bi , and the
small errors will just push it along onto the rest of the Hilbert space never being fully
trapped in an invariant subspace.
The remaining question then, is what is H? How do we do H |u⟩ in step 7 in the

code? That is left entirely up to us. One would immediately scream out at this point
“just use matrices”. However consider an N site model, with spin S=1/2 on each site.
In total the Hilbert space is dim(H) = 2N . Try writing a 2N × 2N matrix in RAM
memory and you will run out of it exponentially fast. A first remedy to this problem
is to use sparse matrices. This special structure holds only matrix elements that are
non-zero, and is particularly useful for matrices with elements that are mostly zero.
For a first or even second nearest neighbor spin model this is definitely the case.
However, this will also very quickly eat up all our RAM, and after all, we also need
to store the vectors during the Lanczos procedure so we wish to economize as much
as possible. Thus, we inevitably arrive at coding the Hamiltonian.
The arrays u in code are simply indexed from 0 to 2N − 1. The µth element u[µ]
of the array u is nothing more than the vector coefficient of the µth basis element,
N∑
−1
and the array u is really representing |u⟩ = ⟨µ|u⟩ |µ⟩, with u[m] = ⟨µ|u⟩ for an
µ=0
orthogonal basis choice |µ⟩. Thus the Hamiltonian operation |w⟩ = H |u⟩ means we
93
are trying to find a new array w such that w[ν] = ⟨ν|w⟩. Expanding |w⟩ we have
w[ν] = ⟨ν|w⟩ = ⟨ν|H|u⟩

N∑
−1
= ⟨µ|u⟩ ⟨ν|H|µ⟩
µ=0 (A.6)
N∑
−1
= u[µ] ⟨µ|H|ν⟩∗ .
µ=0
If we were to stop at the second line above, this will be vastly expensive. To create
just one of the elements of array w, we need to do a for loop of length N to carry out
the above sum, and then keep on doing that N more times for every element of w, for a
total of N 2 iterations. Thankfully, we can do much better in the third line. Note that
H |ν⟩ would produce a small amount of new basis states, and this fact is connected to
the sparsity mentioned above. So rather than doing N 2 iterations, we would rather
do N iterations over ν, and for every ν, compute the small number of produced |µ′ ⟩
and subsequent ⟨µ′ |H|ν⟩∗ , reducing the overall computation significantly.
To perform H |ν⟩ and find ⟨µ|H|ν⟩∗ as efficiently as possible, we take advantage
of the simple indexing of integers n = 0 to 2N − 1. Thinking about these numbers
in binary basis, each digit (0 or 1) is just a bit in memory. For example lets say we
were considering a two site spin S=1/2 system. Assign spin up to 1 and spin down
to 0. Then the four basis states will be indexed as (0, 1, 2, 3) → (00, 01, 10, 11) →
(|↓↓⟩ , |↓↑⟩ , |↑↓⟩ , |↑↑⟩). With this correspondence, we can now write code for any
Hamiltonian operation H |ν⟩ by making use of bit operations. There are a handful of
such operations commonly found in programming languages, bit wise AND, bit wise
OR, bit wise XOR, the bit’s compliment, and left and right bit shifts. All we need to
do given a spin model is rewrite it in ladder operators S + , S − and S z . These operators
measure or flip spin states, which corresponds to bit flips and permutations. At that
point, we can come up with a simple combination of bit operations to perform these
tasks. Such customize Hamiltonians and other operators were devised in this thesis
as needed for the system at hand.
Appendix B
Spin wave theory of a ferromagnet

on honeycomb lattice
Figure B.1: X Y and Z bonds of the J − K − Γ − Γ′ − Ac shown in red, green and blue respectively.
Second nearest neighbor bonds are shown in orange. The orange arrows indicate when sgn(ij) = +1
in the Dzyaloshinskii-Moriya term.
We consider the J − K − Γ − Γ′ − Ac as well as the second nearest neighbor
94
95
Dzyaloshinskii-Moriya term (Dc )
∑ [ ( )
H = JSi · Sj + KSi Sj + Γ Si Sj + Si Sj
γ γ α β β α
⟨i,j⟩∈αβ(γ) ( )]
′
+Γ Siα Sjγ
+ + Siβ Sjγ + Siγ Sjα Siγ Sjβ (B.1)
∑ ∑
+ Dc · (Si × Sj ) + Ac (Si · ĉ)2 ,
⟨⟨i,j⟩⟩ i
where Dc = Dc sgn(ij)ĉ and sgn(ij) = +1 when i to j points along the orange arrows
in Fig. B.1. The standard Holstein-Primakoff transformation [169] expanded to linear
order in S reads as
( )1
√ a† a 2 √
S +
= 2S 1 − a ≃ 2Sa,
2S
( )1
√ a† a 2 √ (B.2)
S− = †
2Sa 1 − ≃ 2Sa† ,
2S
Sz = S − a† a.
Using the above approximation, and Fourier transforming Eq. (B.1), this leads to
∑
H = ECL + x†k hk xk ,
k∈BZ
x†k = (a†k , b†k , a−k , b−k ),
 
ho− (k) h1 (k) 0 h2 (k) (B.3)
 
 h1 (k)∗ ho+ (k) h2 (−k) 0 
hk = 
 ∗
,

 0 h2 (−k) ho+ (k) h1 (k) 
h2 (k)∗ 0 h1 (k)∗ ho− (k)
where the two species of bosons ak and bk correspond to the two sublattices of the
unit cell. The h(k) terms are
ho± (k) = ho ± hDM (k),

ho = −S (2Ac + 2Γ + 4Γ′ + 3J + K) ,
hDM (k) = 2SDc (sin(a · k) + sin(b · k) − sin((a + b) · k)) , (B.4)
′
S (Γ + 2Γ − 3J − K)
h1 (k) = − (1 + e−ia·k + eib·k ),
3
S (2Γ − 2Γ′ + K) ( √ √ )
h2 (k) = (1 − i 3)e−ia·k + (1 + i 3)eib·k − 2 ,
6
96 APPENDIX B. SPIN WAVE THEORY OF A FERROMAGNET ON HONEYCOMB LATTICE
where a and b are the lattice vectors in Fig. B.1. Following standard methods of
diagonalizing BdG Hamiltonians [170], we find the lowest eigenvalue around the Γ
point in the Brillouin zone, and upon expanding to orders of k, we get the spin gap
ωo and spin stiffness ρ:
ωk = ωo + ρk 2
ωo = S |3Γ + 6Γ′ + 2Ac |

(B.5)

S (K + 2Γ − 2Γ ′ 2
)

ρ = 3J + K − Γ − 2Γ′ − .
12 2 (2Γ + 4Γ′ + 2Ac + 3J + K)
K
At the K point in the BZ the Dirac gap is ω+ − ω−
K
where
{
ω+K
= S (6Γ′ + 2Ac + 3J) (4Γ + 2Γ′ + 2Ac + 3J + 2K)
√ }1/2
+6 3Dc (2Γ + 4Γ′ + 2Ac + 3J + K) + 27Dc2 , (B.6)
√
K
ω− = S 2Γ + 4Γ′ + 2Ac + 3J + K − 3 3Dc .
Appendix C
Details of the jeff infinite BR
We go into some more detail about the behavior of branching ratio BR → ∞ in the
IL
jeff picture. We go back to the definition of branching ratio BR = 3 . When the jeff
IL2
state is present, it so happens that the relative phase structure of the wavefunction
is such that IL2 = 0. To see this effect, first lest write out our initial state. In
principle we need to work with a multi-electron wavefunction. However, for a filling
of 5 electrons in the jeff states, and fully occupied core p-orbitals, we always have
a single unoccupied state, so we can view the problem as a single hole problem. In
the initial state, we have one hole in the jeff = 1/2 states and fully occupied core
p-orbitals ⟩
1 1

|Ψi ⟩ = , + . (C.1)
2 2 eff
Our final state will be fully occupied jef f states and one hole in p1/2 states. There
are two such final states we need to sum over:
⟩
⟩
Ψf+ = 1 , + 1 . (C.2)
2 2
⟩
⟩ 1 1
Ψf− = , −
2 2 . (C.3)
The intensity of the L2 edge (for unpolarized light) caused by a dipole allowed tran-
sition through the dipole operator D̂ will be proportional to:
∑ ⟨ ⟩2
IL2 ∝ Ψf D̂Ψi . (C.4)
f
The dipole operator is a Cartesian vector operator. It suits our purpose to re-write
97
98 APPENDIX C. DETAILS OF THE JEFF INFINITE BR
it as a spherical tensor of rank one:

(1) (1) (1) (1)
D̂ + D̂ D̂ − D̂
D̂x = +1√ −1 , D̂y = D̂0 , D̂z = +1 √ −1 ,
(1)
(C.5)
2 i 2
D̂x + iD̂y D̂x − iD̂y

(1)
D̂+1 = √ (1) (1)
, D̂0 = D̂z , D̂−1 = √ . (C.6)
2 2
Let us write jeff = 1/2 states and p1/2 states in terms of angular momentum states:
⟩ √
1 1 1 2
,+ = − √ |1eff , 0; ↑⟩ + |1eff , +1; ↓⟩
2 2 3 3
eff
( )
i √ 1 1
= √ 2 |2, −1; ↓⟩ + √ |2, −2; ↑⟩ − √ |2, 2; ↑⟩ ,
3 2 2
⟩ √
1 1 1 2
,− = √ |1eff , 0; ↓⟩ − |1eff , −1; ↑⟩
2 2 3 3
eff
( ) (C.7)
i √ 1 1
= −√ 2 |2, 1; ↑⟩ + √ |2, 2; ↓⟩ − √ |2, −2; ↓⟩ ,
3 2 2
⟩ √
1 1 1 2
,+ = − √ |1, 0; ↑⟩ + |1, +1; ↓⟩ ,
2 2 3 3
⟩ √
1 1 1 2
,− = √ |1, 0; ↓⟩ − |1, −1; ↑⟩ .
2 2 3 3
Finally, we make use of the Wigner-Eckart theorem, namely, matrix elements of

spherical tensors are equal to a reduced matrix element independent of the magnetic
quantum numbers and a Clebsch-Gordan coefficient:
⟨ ⟩
⟨j ′ , m′ | Tq(k) |j, m⟩ = ⟨j, m; k, q|j ′ m′ ⟩ j ′ T (k) j . (C.8)
The dipole operator only connects states that satisfy m + q = m′ . Also since the
dipole operator doesn’t involve spin, we have that ∆Sz = 0 . We use this to get the
99
following matrix elements

⟨ ⟩ ⟨ ⟩
1 1 1 1
Ψf + D̂Ψi = , + D̂ , +
2 2 2 2 eff
( √ ) (
1 2 i (√
= − √ ⟨1, 0; ↑| + ⟨1, +1; ↓| D̂ (1) √ 2 |2, −1; ↓⟩
3 3 3 ))
1 1
+ √ |2, −2; ↑⟩ − √ |2, 2; ↑⟩
2 2
= 0.
(C.9)
The above results is understood considering that ∆Sz = 0 and no q can connect the
states, so this is immediately zero.
⟨ ⟩ ⟨ ⟩
1 1 1 1
Ψf − D̂Ψi = , + D̂ , +
2 2 2 2 eff
( √ ) (
1 2 i (√
= √ ⟨1, 0; ↓| − ⟨1, −1; ↑| D̂ (1) √ 2 |2, −1; ↓⟩
3 3 3 ))
1 1
+ √ |2, −2; ↑⟩ − √ |2, 2; ↑⟩
2 2
1 i √ ⟨ (1) ⟩ √2 i 1 ⟨ ⟩
(1)
= √ √ 2 1, 0D̂+1 2, −1 − √ √ 1, −1D̂+1 2, −2
3 3 3 3 2
√ ( )⟨ ⟩
i 2 1
= ⟨2, −1; 1, +1|1, 0⟩ − √ ⟨2, −2; 1, +1|1, −1⟩ 2 D̂(1) 1 .
3 2
(C.10)
To go from the second to third line, consider that ∆Sz = 0 and than only q = +1 can
connect the states. Then, from the third line to the fourth we used the Wigner-Eckart
theorem. Finally, the Clebsch-Gordan coefficients are found to be
√ √
3 3
⟨2, −1; 1, +1|1, 0⟩ = ⟨2, −2; 1, +1|1, −1⟩ = , (C.11)
10 5
⟨ ⟩

allowing us to conclude that Ψf − D̂Ψi = 0.
1
Observe that the cancellation boils down to the factor of − √ which is coming
2
from the relative phase structure of the jeff = 1/2 state. The calculation follows
similarly for |Ψf + ⟩. Thus, we have that the intensity for p1/2 → Jeff = 1/2 transition
100 APPENDIX C. DETAILS OF THE JEFF INFINITE BR
is exactly zero.
Note that for unpolarized light, the dipole selection rule states ∆J = ±1, 0 but
always J = 0 ↛ J = 0, which is coming from the Wigner-Ekcart theorem of a rank
one spherical tensor operator, and parity must change as πi πf = −1 coming from
the parity transformation of the dipole operator P−1 D̂P = −D̂. These selection rules
alone would imply that J = 1/2 → Jeff = 1/2 is allowed, however, we explicitly found
J = 1/2 ↛ Jeff = 1/2. This must be a result of the relative phase structure of the
Jeff = 1/2.
Bibliography
[1] S. Gasiorowicz, Quantum Physics, 3rd Ed (Wiley John and Sons Incorporated,
2003).
[2] B. J. Kim, H. Jin, S. J. Moon, J.-Y. Kim, B.-G. Park, C. S. Leem, J. Yu, T. W.
Noh, C. Kim, S.-J. Oh, J.-H. Park, V. Durairaj, G. Cao, and E. Rotenberg,
“Novel Jeff = 1/2 Mott State Induced by Relativistic Spin-Orbit Coupling in
Sr2 IrO4 ,” Physical Review Letters 101, 076402 (2008).
[3] L. Balents, “Spin liquids in frustrated magnets,” Nature 464, 199 (2010).
[4] G. H. Wannier, “Antiferromagnetism. The Triangular Ising Net,” Physical Re-
view 79, 357 (1950).
[5] A. Kitaev, “Anyons in an exactly solved model and beyond,” Annals of Physics
(New York) 321, 2 (2006), January Special Issue.
[6] G. Jackeli and G. Khaliullin, “Mott Insulators in the Strong Spin-Orbit Cou-
pling Limit: From Heisenberg to a Quantum Compass and Kitaev Models,”
Physical Review Letters 102, 017205 (2009).
[7] T. Kato, Perturbation theory for linear operators; 2nd ed., Grundlehren der
mathematischen Wissenschaften : a series of comprehensive studies in mathe-
matics (Springer, Berlin, 1976).
[8] P. Fazekas, Lecture notes on electron correlation and magnetism (World Scien-
tific, 1999).
[9] J. G. Rau, E. K.-H. Lee, and H.-Y. Kee, “Generic Spin Model for the Honey-
comb Iridates beyond the Kitaev Limit,” Physical Review Letters 112, 077204
(2014).
[10] Y. Yamaji, Y. Nomura, M. Kurita, R. Arita, and M. Imada, “First-Principles
Study of the Honeycomb-Lattice Iridates Na2 IrO3 in the Presence of Strong
Spin-Orbit Interaction and Electron Correlations,” Physical Review Letters
113, 107201 (2014).
101
102 BIBLIOGRAPHY
[11] V. M. Katukuri, S. Nishimoto, V. Yushankhai, A. Stoyanova, H. Kandpal,

S. Choi, R. Coldea, I. Rousochatzakis, L. Hozoi, and J. van den Brink, “Kitaev
interactions between j = 1/2 moments in honeycomb Na2 IrO3 are large and
ferromagnetic: insights from ab initio quantum chemistry calculations,” New
Journal of Physics 16, 013056 (2014).
[12] W. Witczak-Krempa, G. Chen, Y. B. Kim, and L. Balents, “Correlated Quan-

tum Phenomena in the Strong Spin-Orbit Regime,” Annual Review of Con-
densed Matter Physics 5, 57 (2014).
[13] J. G. Rau, E. K.-H. Lee, and H.-Y. Kee, “Spin-Orbit Physics Giving Rise to
Novel Phases in Correlated Systems: Iridates and Related Materials,” Annual
Review of Condensed Matter Physics 7, 195 (2016).
[14] S. M. Winter, A. A. Tsirlin, M. Daghofer, J. van den Brink, Y. Singh, P. Gegen-

wart, and R. Valentı́, “Models and materials for generalized Kitaev magnetism,”
Journal of Physics: Condensed Matter 29, 493002 (2017).
[15] Y. Kasahara, T. Ohnishi, Y. Mizukami, O. Tanaka, S. Ma, K. Sugii, N. Kurita,

H. Tanaka, J. Nasu, Y. Motome, T. Shibauchi, and Y. Matsuda, “Majorana
quantization and half-integer thermal quantum Hall effect in a Kitaev spin
liquid,” Nature 559, 227 (2018).
[16] P. P. Stavropoulos, A. Catuneanu, and H.-Y. Kee, “Counter-rotating spiral or-

der in three-dimensional iridates: Signature of hidden symmetry in the Kitaev-Γ
model,” Physical Review B 98, 104401 (2018).
[17] P. P. Stavropoulos, D. Pereira, and H.-Y. Kee, “Microscopic Mechanism for a

Higher-Spin Kitaev Model,” Physical Review Letters 123, 037203 (2019).
[18] C. Hickey, C. Berke, P. P. Stavropoulos, H.-Y. Kee, and S. Trebst, “Field-

driven gapless spin liquid in the spin-1 Kitaev honeycomb model,” Physical
Review Research 2, 023361 (2020).
[19] I. Khait, P. P. Stavropoulos, H.-Y. Kee, and Y. B. Kim, “Characterizing spin-

one Kitaev quantum spin liquids,” Physical Review Research 3, 013160 (2021).
[20] P. P. Stavropoulos, X. Liu, and H.-Y. Kee, “Magnetic anisotropy in spin-3/2

with heavy ligand in honeycomb Mott insulators: Application to CrI3 ,” Physical
Review Research 3, 013216 (2021).
[21] L. S. I. Veiga, M. Etter, K. Glazyrin, F. Sun, C. A. Escanhoela, G. Fabbris,

J. R. L. Mardegan, P. S. Malavi, Y. Deng, P. P. Stavropoulos, H.-Y. Kee,
BIBLIOGRAPHY 103
W. G. Yang, M. van Veenendaal, J. S. Schilling, T. Takayama, H. Takagi, and

D. Haskel, “Pressure tuning of bond-directional exchange interactions and mag-
netic frustration in the hyperhoneycomb iridate β − Li2 IrO3 ,” Physical Review
B 96, 140402 (2017).
[22] A. Scheie, K. Ross, P. P. Stavropoulos, E. Seibel, J. A. Rodriguez-Rivera, J. A.

Tang, Y. Li, H.-Y. Kee, R. J. Cava, and C. Broholm, “Counterrotating magnetic
order in the honeycomb layers of NaNi2 BiO6−δ ,” Physical Review B 100, 214421
(2019).
[23] A. Seo, P. P. Stavropoulos, H.-H. Kim, K. Fürsich, M. Souri, J. G. Connell,

H. Gretarsson, M. Minola, H. Y. Kee, and B. Keimer, “Compressive strain
induced enhancement of exchange interaction and short-range magnetic order in
Sr2 IrO4 investigated by Raman spectroscopy,” Physical Review B 100, 165106
(2019).
[24] Y. Singh and P. Gegenwart, “Antiferromagnetic Mott insulating state in single

crystals of the honeycomb lattice material Na2 IrO3 ,” Physical Review B 82,
064412 (2010).
[25] Y. Singh, S. Manni, J. Reuther, T. Berlijn, R. Thomale, W. Ku, S. Trebst, and

P. Gegenwart, “Relevance of the Heisenberg-Kitaev Model for the Honeycomb
Lattice Iridates A2 IrO3 ,” Physical Review Letters 108, 127203 (2012).
[26] K. W. Plumb, J. P. Clancy, L. J. Sandilands, V. V. Shankar, Y. F. Hu, K. S.

Burch, H.-Y. Kee, and Y.-J. Kim, “α − RuCl3 : A spin-orbit assisted Mott
insulator on a honeycomb lattice,” Physical Review B 90, 041112 (2014).
[27] H.-S. Kim, V. V. Shankar, A. Catuneanu, and H.-Y. Kee, “Kitaev magnetism
in honeycomb RuCl3 with intermediate spin-orbit coupling,” Physical Review
B 91, 241110 (2015).
[28] K. A. Modic, T. E. Smidt, I. Kimchi, N. P. Breznay, A. Biffin, S. Choi, R. D.

Johnson, R. Coldea, P. Watkins-Curry, G. T. McCandless, J. Y. Chan, F. Gan-
dara, Z. Islam, A. Vishwanath, A. Shekhter, R. D. McDonald, and J. G. Ana-
lytis, “Realization of a three-dimensional spin-anisotropic harmonic honeycomb
iridate,” Nature Communications 5, 4203 (2014).
[29] T. Takayama, A. Kato, R. Dinnebier, J. Nuss, H. Kono, L. S. I. Veiga, G. Fab-

bris, D. Haskel, and H. Takagi, “Hyperhoneycomb Iridate β−Li2 IrO3 as a
Platform for Kitaev Magnetism,” Physical Review Letters 114, 077202 (2015).
104 BIBLIOGRAPHY
[30] A. Biffin, R. D. Johnson, S. Choi, F. Freund, S. Manni, A. Bombardi, P. Manuel,

P. Gegenwart, and R. Coldea, “Unconventional magnetic order on the hyper-
honeycomb Kitaev lattice in β−Li2 IrO3 : Full solution via magnetic resonant
x-ray diffraction,” Physical Review B 90, 205116 (2014).
[31] A. Biffin, R. D. Johnson, I. Kimchi, R. Morris, A. Bombardi, J. G. Analytis,

A. Vishwanath, and R. Coldea, “Noncoplanar and Counterrotating Incom-
mensurate Magnetic Order Stabilized by Kitaev Interactions in γ−Li2 IrO3 ,”
[32] S. C. Williams, R. D. Johnson, F. Freund, S. Choi, A. Jesche, I. Kimchi,

S. Manni, A. Bombardi, P. Manuel, P. Gegenwart, and R. Coldea, “Incom-
mensurate counterrotating magnetic order stabilized by Kitaev interactions in
the layered honeycomb α−Li2 IrO3 ,” Physical Review B 93, 195158 (2016).
[33] J. Chaloupka, G. Jackeli, and G. Khaliullin, “Kitaev-Heisenberg Model on a

Honeycomb Lattice: Possible Exotic Phases in Iridium Oxides A2 IrO3 ,” Physi-
cal Review Letters 105, 027204 (2010).
[34] J. Reuther, R. Thomale, and S. Trebst, “Finite-temperature phase diagram of

the Heisenberg-Kitaev model,” Physical Review B 84, 100406 (2011).
[35] I. Kimchi and Y.-Z. You, “Kitaev-Heisenberg-J2 -J3 model for the iridates
A2 IrO3 ,” Physical Review B 84, 180407 (2011).
[36] C. C. Price and N. B. Perkins, “Critical Properties of the Kitaev-Heisenberg

Model,” Physical Review Letters 109, 187201 (2012).
[37] J. Chaloupka, G. Jackeli, and G. Khaliullin, “Zigzag Magnetic Order in the

Iridium Oxide Na2 IrO3 ,” Physical Review Letters 110, 097204 (2013).
[38] E. Sela, H.-C. Jiang, M. H. Gerlach, and S. Trebst, “Order-by-disorder and

spin-orbital liquids in a distorted Heisenberg-Kitaev model,” Physical Review
B 90, 035113 (2014).
[39] J. Reuther, R. Thomale, and S. Rachel, “Spiral order in the honeycomb iridate
Li2 IrO3 ,” Physical Review B 90, 100405 (2014).
[40] Y. Sizyuk, C. Price, P. Wölfle, and N. B. Perkins, “Importance of anisotropic

exchange interactions in honeycomb iridates: Minimal model for zigzag antifer-
romagnetic order in Na2 IrO3 ,” Physical Review B 90, 155126 (2014).
[41] E. K.-H. Lee, R. Schaffer, S. Bhattacharjee, and Y. B. Kim, “Heisenberg-Kitaev

model on the hyperhoneycomb lattice,” Physical Review B 89, 045117 (2014).
BIBLIOGRAPHY 105
[42] I. Kimchi and R. Coldea, “Spin dynamics of counterrotating Kitaev spirals via
duality,” Physical Review B 94, 201110 (2016).
[43] E. K.-H. Lee and Y. B. Kim, “Theory of magnetic phase diagrams in hyper-
honeycomb and harmonic-honeycomb iridates,” Physical Review B 91, 064407
(2015).
[44] V. M. Katukuri, R. Yadav, L. Hozoi, S. Nishimoto, and J. van den Brink, “The
vicinity of hyper-honeycomb β-Li2IrO3 to a three-dimensional Kitaev spin liquid
state,” Scientific Reports 6, 29585 (2016).
[45] H.-S. Kim, E. K.-H. Lee, and Y. B. Kim, “Predominance of the Kitaev interac-
tion in a three-dimensional honeycomb iridate: From ab initio to spin model,”
EPL (Europhysics Letters) 112, 67004 (2015).
[46] H.-S. Kim, Y. B. Kim, and H.-Y. Kee, “Revealing frustrated local moment
model for pressurized hyperhoneycomb iridate: Paving the way toward a quan-
tum spin liquid,” Physical Review B 94, 245127 (2016).
[47] J. Chaloupka and G. Khaliullin, “Hidden symmetries of the extended Kitaev-

Heisenberg model: Implications for the honeycomb-lattice iridates A2 IrO3 ,”
Physical Review B 92, 024413 (2015).
[48] S. Ducatman, I. Rousochatzakis, and N. B. Perkins, “Magnetic structure and

excitation spectrum of the hyperhoneycomb Kitaev magnet β−Li2 IrO3 ,” Phys-
ical Review B 97, 125125 (2018).
[49] I. Rousochatzakis and N. B. Perkins, “Magnetic field induced evolution of inter-

twined orders in the Kitaev magnet β−Li2 IrO3 ,” Physical Review B 97, 174423
(2018).
[50] H. Gretarsson, J. P. Clancy, X. Liu, J. P. Hill, E. Bozin, Y. Singh, S. Manni,

P. Gegenwart, J. Kim, A. H. Said, D. Casa, T. Gog, M. H. Upton, H.-S. Kim,
J. Yu, V. M. Katukuri, L. Hozoi, J. van den Brink, and Y.-J. Kim, “Crystal-
Field Splitting and Correlation Effect on the Electronic Structure of A2 IrO3 ,”
[51] S. Mandal and N. Surendran, “Exactly solvable Kitaev model in three dimen-
sions,” Physical Review B 79, 024426 (2009).
[52] I. Kimchi, J. G. Analytis, and A. Vishwanath, “Three-dimensional quantum

spin liquids in models of harmonic-honeycomb iridates and phase diagram in an
infinite-D approximation,” Physical Review B 90, 205126 (2014).
106 BIBLIOGRAPHY
[53] R. Schaffer, E. K.-H. Lee, Y.-M. Lu, and Y. B. Kim, “Topological Spinon
Semimetals and Gapless Boundary States in Three Dimensions,” Physical Re-
view Letters 114, 116803 (2015).
[54] J. Chaloupka and G. Khaliullin, “Magnetic anisotropy in the Kitaev model

systems Na2 IrO3 and RuCl3 ,” Physical Review B 94, 064435 (2016).
[55] X.-Y. Feng, G.-M. Zhang, and T. Xiang, “Topological Characterization of

Quantum Phase Transitions in a Spin-1/2 Model,” Physical Review Letters 98,
087204 (2007).
[56] S. K. Choi, R. Coldea, A. N. Kolmogorov, T. Lancaster, I. I. Mazin, S. J.

Blundell, P. G. Radaelli, Y. Singh, P. Gegenwart, K. R. Choi, S.-W. Cheong,
P. J. Baker, C. Stock, and J. Taylor, “Spin Waves and Revised Crystal Structure
of Honeycomb Iridate Na2 IrO3 ,” Physical Review Letters 108, 127204 (2012).
[57] J. G. Rau and H.-Y. Kee, “Trigonal distortion in the honeycomb iridates: Prox-
imity of zigzag and spiral phases in Na2IrO3,” (2014), arXiv:1408.4811 [cond-
mat.str-el] .
[58] J. A. Sears, M. Songvilay, K. W. Plumb, J. P. Clancy, Y. Qiu, Y. Zhao, D. Par-

shall, and Y.-J. Kim, “Magnetic order in α−RuCl3 : A honeycomb-lattice quan-
tum magnet with strong spin-orbit coupling,” Physical Review B 91, 144420
(2015).
[59] L. J. Sandilands, Y. Tian, A. A. Reijnders, H.-S. Kim, K. W. Plumb, Y.-J. Kim,

H.-Y. Kee, and K. S. Burch, “Spin-orbit excitations and electronic structure of
the putative Kitaev magnet α − RuCl3 ,” Physical Review B 93, 075144 (2016).
[60] L. J. Sandilands, Y. Tian, K. W. Plumb, Y.-J. Kim, and K. S. Burch, “Scatter-

ing Continuum and Possible Fractionalized Excitations in α-RuCl3 ,” Physical
Review Letters 114, 147201 (2015).
[61] A. Banerjee, C. A. Bridges, J.-Q. Yan, A. A. Aczel, L. Li, M. B. Stone, G. E.

Granroth, M. D. Lumsden, Y. Yiu, J. Knolle, S. Bhattacharjee, D. L. Kovrizhin,
R. Moessner, D. A. Tennant, D. G. Mandrus, and S. E. Nagler, “Proximate Ki-
taev quantum spin liquid behaviour in a honeycomb magnet,” Nature Materials
15, 733 (2016).
[62] R. D. Johnson, S. C. Williams, A. A. Haghighirad, J. Singleton, V. Zapf,

P. Manuel, I. I. Mazin, Y. Li, H. O. Jeschke, R. Valentı́, and R. Coldea,
“Monoclinic crystal structure of α − RuCl3 and the zigzag antiferromagnetic
ground state,” Physical Review B 92, 235119 (2015).
BIBLIOGRAPHY 107
[63] H.-S. Kim and H.-Y. Kee, “Crystal structure and magnetism in α − RuCl3 : An
ab initio study,” Physical Review B 93, 155143 (2016).
[64] H. B. Cao, A. Banerjee, J.-Q. Yan, C. A. Bridges, M. D. Lumsden, D. G. Man-

drus, D. A. Tennant, B. C. Chakoumakos, and S. E. Nagler, “Low-temperature
crystal and magnetic structure of α − RuCl3 ,” Physical Review B 93, 134423
(2016).
[65] L. Janssen, E. C. Andrade, and M. Vojta, “Magnetization processes of zigzag

states on the honeycomb lattice: Identifying spin models for α-RuCl3 and
Na2 IrO3 ,” Physical Review B 96, 064430 (2017).
[66] G. Baskaran, D. Sen, and R. Shankar, “Spin-S Kitaev model: Classical ground
states, order from disorder, and exact correlation functions,” Physical Review
B 78, 115116 (2008).
[67] G. Baskaran, S. Mandal, and R. Shankar, “Exact Results for Spin Dynamics
and Fractionalization in the Kitaev Model,” Physical Review Letters 98, 247201
(2007).
[68] A. Koga, H. Tomishige, and J. Nasu, “Ground-state and Thermodynamic Prop-

erties of an S = 1 Kitaev Model,” Journal of the Physical Society of Japan 87,
063703 (2018).
[69] J. Oitmaa, A. Koga, and R. R. P. Singh, “Incipient and well-developed entropy

plateaus in spin-S Kitaev models,” Physical Review B 98, 214404 (2018).
[70] A. M. Samarakoon, A. Banerjee, S.-S. Zhang, Y. Kamiya, S. E. Nagler, D. A.

Tennant, S.-H. Lee, and C. D. Batista, “Comprehensive study of the dynamics
of a classical Kitaev spin liquid,” Physical Review B 96, 134408 (2017).
[71] J. Kanamori, “Electron Correlation and Ferromagnetism of Transition Metals,”

Progress of Theoretical Physics 30, 275 (1963).
[72] J. C. Slater and G. F. Koster, “Simplified LCAO Method for the Periodic Po-
tential Problem,” Physical Review 94, 1498 (1954).
[73] R. H. Busey and W. F. Giauque, “The Heat Capacity of Anhydrous NiCl2 from
15 to 300◦ K. The Antiferromagnetic Anomaly near 52◦ K. Entropy and Free
Energy,” Journal of the American Chemical Society 74, 4443 (1952).
[74] D. Billerey, C. Terrier, N. Ciret, and J. Kleinclauss, “Neutron diffraction study

and specific heat of antiferromagnetic NiI2 ,” Physics Letters A 61, 138 (1977).
108 BIBLIOGRAPHY
[75] S. Kuindersma, J. Sanchez, and C. Haas, “Magnetic and structural investiga-

tions on NiI2 and CoI2 ,” Physica B+C 111, 231 (1981).
[76] E. A. Zvereva, M. I. Stratan, Y. A. Ovchenkov, V. B. Nalbandyan, J.-Y. Lin,

E. L. Vavilova, M. F. Iakovleva, M. Abdel-Hafiez, A. V. Silhanek, X.-J. Chen,
A. Stroppa, S. Picozzi, H. O. Jeschke, R. Valentı́, and A. N. Vasiliev, “Zigzag
antiferromagnetic quantum ground state in monoclinic honeycomb lattice anti-
monates A3 Ni2 SbO6 (A = Li, Na),” Physical Review B 92, 144401 (2015).
[77] A. I. Kurbakov, A. N. Korshunov, S. Y. Podchezertsev, A. L. Malyshev, M. A.

Evstigneeva, F. Damay, J. Park, C. Koo, R. Klingeler, E. A. Zvereva, and
V. B. Nalbandyan, “Zigzag spin structure in layered honeycomb Li3 Ni2 SbO6 : A
combined diffraction and antiferromagnetic resonance study,” Physical Review
B 96, 024417 (2017).
[78] R. Yadav, N. A. Bogdanov, V. M. Katukuri, S. Nishimoto, J. Van Den Brink,

and L. Hozoi, “Kitaev exchange and field-induced quantum spin-liquid states
in honeycomb α-RuCl3 ,” Scientific Reports 6, 37925 (2016).
[79] S.-H. Baek, S.-H. Do, K.-Y. Choi, Y. S. Kwon, A. U. B. Wolter, S. Nishimoto,
J. van den Brink, and B. Büchner, “Evidence for a Field-Induced Quantum
Spin Liquid in α-RuCl3 ,” Physical Review Letters 119, 037201 (2017).
[80] A. U. B. Wolter, L. T. Corredor, L. Janssen, K. Nenkov, S. Schönecker, S.-H.

Do, K.-Y. Choi, R. Albrecht, J. Hunger, T. Doert, M. Vojta, and B. Büchner,
“Field-induced quantum criticality in the Kitaev system α-RuCl3 ,” Physical
Review B 96, 041405(R) (2017).
[81] J. Zheng, K. Ran, T. Li, J. Wang, P. Wang, B. Liu, Z.-X. Liu, B. Normand,
J. Wen, and W. Yu, “Gapless Spin Excitations in the Field-Induced Quantum
Spin Liquid Phase of α-RuCl3 ,” Physical Review Letters 119, 227208 (2017).
[82] N. Jansa, A. Zorko, M. Gomilsek, M. Pregelj, K. W. Krämer, D. Biner, A. Biffin,

C. Rüegg, and M. Klanjsek, “Observation of two types of fractional excitation
in the Kitaev honeycomb magnet,” Nature Physics 14, 786 (2018).
[83] Z. Zhu, I. Kimchi, D. N. Sheng, and L. Fu, “Robust non-Abelian spin liquid
and a possible intermediate phase in the antiferromagnetic Kitaev model with
magnetic field,” Physical Review B 97, 241110(R) (2018).
[84] M. Gohlke, R. Moessner, and F. Pollmann, “Dynamical and topological prop-

erties of the Kitaev model in a [111] magnetic field,” Physical Review B 98,
014418 (2018).
BIBLIOGRAPHY 109
[85] J. Nasu, Y. Kato, Y. Kamiya, and Y. Motome, “Successive Majorana topologi-

cal transitions driven by a magnetic field in the Kitaev model,” Physical Review
B 98, 060416(R) (2018).
[86] D. C. Ronquillo, A. Vengal, and N. Trivedi, “Signatures of magnetic-field-driven

quantum phase transitions in the entanglement entropy and spin dynamics of
the Kitaev honeycomb model,” Physical Review B 99, 140413 (2019).
[87] C. Hickey and S. Trebst, “Emergence of a field-driven U(1) spin liquid in the
Kitaev honeycomb model,” Nature Communications 10, 530 (2019).
[88] S. Liang, M.-H. Jiang, W. Chen, J.-X. Li, and Q.-H. Wang, “Intermediate
gapless phase and topological phase transition of the Kitaev model in a uniform
magnetic field,” Physical Review B 98, 054433 (2018).
[89] P. Lampen-Kelley, L. Janssen, E. C. Andrade, S. Rachel, J. Q. Yan, C. Balz,

D. G. Mandrus, S. E. Nagler, and M. Vojta, “Field-induced intermediate phase
in α-RuCl3 : Non-coplanar order, phase diagram, and proximate spin liquid,”
(2021), arXiv:1807.06192 [cond-mat.str-el] .
[90] H.-C. Jiang, C.-Y. Wang, B. Huang, and Y.-M. Lu, “Field induced quan-
tum spin liquid with spinon Fermi surfaces in the Kitaev model,” (2018),
arXiv:1809.08247 [cond-mat.str-el] .
[91] L. Zou and Y.-C. He, “Field-induced QCD3 -Chern-Simons quantum criticalities
in Kitaev materials,” Physical Review Research 2, 013072 (2020).
[92] N. D. Patel and N. Trivedi, “Magnetic field-induced intermediate quantum spin

liquid with a spinon Fermi surface,” Proceedings of the National Academy of
Sciences 116, 12199 (2019).
[93] J. S. Gordon, A. Catuneanu, E. S. Sørensen, and H.-Y. Kee, “Theory of the

field-revealed Kitaev spin liquid,” Nature Communications 10, 2470 (2019).
[94] C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, “Non-

Abelian anyons and topological quantum computation,” Reviews of Modern
Physics 80, 1083 (2008).
[95] H.-C. Jiang, Z.-C. Gu, X.-L. Qi, and S. Trebst, “Possible proximity of the
Mott insulating iridate Na2 IrO3 to a topological phase: Phase diagram of the
Heisenberg-Kitaev model in a magnetic field,” Physical Review B 83, 245104
(2011).
110 BIBLIOGRAPHY
[96] D. A. S. Kaib, S. M. Winter, and R. Valentı́, “Kitaev honeycomb models in

magnetic fields: Dynamical response and dual models,” Physical Review B 100,
144445 (2019).
[97] Z. Zhu, Z.-Y. Weng, and D. N. Sheng, “Magnetic field induced spin liquids
in S = 1 Kitaev honeycomb model,” Physical Review Research 2, 022047(R)
(2020).
[98] W.-B. Zhang, Q. Qu, P. Zhu, and C.-H. Lam, “Robust intrinsic ferromagnetism
and half semiconductivity in stable two-dimensional single-layer chromium tri-
halides,” Journal of Materials Chemistry C 3, 12457 (2015).
[99] D. Torelli and T. Olsen, “Calculating critical temperatures for ferromagnetic

order in two-dimensional materials,” 2D Materials 6, 015028 (2018).
[100] S. S. Pershoguba, S. Banerjee, J. C. Lashley, J. Park, H. Ågren, G. Aeppli,

and A. V. Balatsky, “Dirac Magnons in Honeycomb Ferromagnets,” Physical
Review X 8, 011010 (2018).
[101] F. Zheng, J. Zhao, Z. Liu, M. Li, M. Zhou, S. Zhang, and P. Zhang, “Tunable
spin states in the two-dimensional magnet CrI3 ,” Nanoscale 10, 14298 (2018).
[102] L. Webster and J.-A. Yan, “Strain-tunable magnetic anisotropy in monolayer

CrCl3 , CrBr3 , and CrI3 ,” Physical Review B 98, 144411 (2018).
[103] J. Liu, M. Shi, P. Mo, and J. Lu, “Electrical-field-induced magnetic Skyrmion

ground state in a two-dimensional chromium tri-iodide ferromagnetic mono-
layer,” AIP Advances 8, 055316 (2018).
[104] M. Wu, Z. Li, T. Cao, and S. G. Louie, “Physical origin of giant excitonic
and magneto-optical responses in two-dimensional ferromagnetic insulators,”
Nature Communications 10, 2371 (2019).
[105] T. Olsen, “Theory and simulations of critical temperatures in CrI3 and other 2D
materials: easy-axis magnetic order and easy-plane Kosterlitz-Thouless transi-
tions,” MRS Communications 9, 1142 (2019).
[106] O. Besbes, S. Nikolaev, N. Meskini, and I. Solovyev, “Microscopic origin of

ferromagnetism in the trihalides CrCl3 and CrI3 ,” Physical Review B 99, 104432
(2019).
[107] V. K. Gudelli and G.-Y. Guo, “Magnetism and magneto-optical effects in bulk
and few-layer CrI3 : a theoretical GGA+U study,” New Journal of Physics 21,
053012 (2019).
BIBLIOGRAPHY 111
[108] C. Xu, J. Feng, M. Kawamura, Y. Yamaji, Y. Nahas, S. Prokhorenko, Y. Qi,

H. Xiang, and L. Bellaiche, “Possible Kitaev Quantum Spin Liquid State in
2D Materials with S = 3/2,” Physical Review Letters 124, 087205 (2020).
[109] M. Pizzochero and O. V. Yazyev, “Inducing Magnetic Phase Transitions in

Monolayer CrI3 via Lattice Deformations,” The Journal of Physical Chemistry
C 124, 7585 (2020).
[110] M. Pizzochero, R. Yadav, and O. V. Yazyev, “Magnetic exchange interactions

in monolayer CrI3 from many-body wavefunction calculations,” 2D Materials 7,
035005 (2020).
[111] E. Aguilera, R. Jaeschke-Ubiergo, N. Vidal-Silva, L. E. F. Foa Torres, and A. S.

Nunez, “Topological magnonics in the two-dimensional van der Waals magnet
CrI3 ,” Physical Review B 102, 024409 (2020).
[112] D. Soriano, M. I. Katsnelson, and J. Fernández-Rossier, “Magnetic Two-

Dimensional Chromium Trihalides: A Theoretical Perspective,” Nano Letters
20, 6225 (2020).
[113] L. L. Handy and N. W. Gregory, “Structural Properties of Chromium(III) Iodide

and Some Chromium(III) Mixed Halides,” Journal of the American Chemical
Society 74, 891 (1952).
[114] J. F. Dillon and C. E. Olson, “Magnetization, Resonance, and Optical Properties

of the Ferromagnet CrI3 ,” Journal of Applied Physics 36, 1259 (1965).
[115] M. A. McGuire, H. Dixit, V. R. Cooper, and B. C. Sales, “Coupling of Crys-

tal Structure and Magnetism in the Layered, Ferromagnetic Insulator CrI3 ,”
Chemistry of Materials 27, 612 (2015).
[116] B. Huang, G. Clark, E. Navarro-Moratalla, D. R. Klein, R. Cheng, K. L. Seyler,

D. Zhong, E. Schmidgall, M. A. McGuire, D. H. Cobden, W. Yao, D. Xiao,
P. Jarillo-Herrero, and X. Xu, “Layer-dependent ferromagnetism in a van der
Waals crystal down to the monolayer limit,” Nature 546, 270 (2017).
[117] N. D. Mermin and H. Wagner, “Absence of Ferromagnetism or Antiferromag-

netism in One- or Two-Dimensional Isotropic Heisenberg Models,” Physical
Review Letters 17, 1133 (1966).
[118] J. L. Lado and J. Fernández-Rossier, “On the origin of magnetic anisotropy in

two dimensional CrI3 ,” 2D Materials 4, 035002 (2017).
112 BIBLIOGRAPHY
[119] D.-H. Kim, K. Kim, K.-T. Ko, J. Seo, J. S. Kim, T.-H. Jang, Y. Kim, J.-Y.
Kim, S.-W. Cheong, and J.-H. Park, “Giant Magnetic Anisotropy Induced by
Ligand LS Coupling in Layered Cr Compounds,” Physical Review Letters 122,
207201 (2019).
[120] C. Xu, J. Feng, H. Xiang, and L. Bellaiche, “Interplay between Kitaev interac-
tion and single ion anisotropy in ferromagnetic CrI3 and CrGeTe3 monolayers,”
npj Computational Materials 4, 57 (2018).
[121] I. Lee, F. G. Utermohlen, D. Weber, K. Hwang, C. Zhang, J. van Tol, J. E.

Goldberger, N. Trivedi, and P. C. Hammel, “Fundamental Spin Interactions
Underlying the Magnetic Anisotropy in the Kitaev Ferromagnet CrI3 ,” Physical
Review Letters 124, 017201 (2020).
[122] I. V. Kashin, V. V. Mazurenko, M. I. Katsnelson, and A. N. Rudenko,

“Orbitally-resolved ferromagnetism of monolayer CrI3 ,” 2D Materials 7, 025036
(2020).
[123] G. Jackeli and A. Avella, “Quantum order by disorder in the Kitaev model on
a triangular lattice,” Physical Review B 92, 184416 (2015).
[124] D. R. Klein, D. MacNeill, J. L. Lado, D. Soriano, E. Navarro-Moratalla,

K. Watanabe, T. Taniguchi, S. Manni, P. Canfield, J. Fernández-Rossier, and
P. Jarillo-Herrero, “Probing magnetism in 2D van der Waals crystalline insula-
tors via electron tunneling,” Science 360, 1218 (2018).
[125] T. Song, X. Cai, M. W.-Y. Tu, X. Zhang, B. Huang, N. P. Wilson, K. L. Seyler,

L. Zhu, T. Taniguchi, K. Watanabe, M. A. McGuire, D. H. Cobden, D. Xiao,
W. Yao, and X. Xu, “Giant tunneling magnetoresistance in spin-filter van der
Waals heterostructures,” Science 360, 1214 (2018).
[126] D. R. Klein, D. MacNeill, Q. Song, D. T. Larson, S. Fang, M. Xu, R. A. Ribeiro,

P. C. Canfield, E. Kaxiras, R. Comin, and P. Jarillo-Herrero, “Enhancement of
interlayer exchange in an ultrathin two-dimensional magnet,” Nature Physics
15, 1255 (2019).
[127] H. H. Kim, B. Yang, S. Li, S. Jiang, C. Jin, Z. Tao, G. Nichols, F. Sfigakis,

S. Zhong, C. Li, S. Tian, D. G. Cory, G.-X. Miao, J. Shan, K. F. Mak, H. Lei,
K. Sun, L. Zhao, and A. W. Tsen, “Evolution of interlayer and intralayer
magnetism in three atomically thin chromium trihalides,” Proceedings of the
National Academy of Sciences of the U. S. A. 116, 11131 (2019).
BIBLIOGRAPHY 113
[128] G. Kresse and J. Hafner, “Ab initio molecular dynamics for liquid metals,”
Physical Review B 47, 558(R) (1993).
[129] J. P. Perdew, K. Burke, and M. Ernzerhof, “Generalized Gradient Approxima-

tion Made Simple,” Physical Review Letters 77, 3865 (1997).
[130] A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari,

“wannier90: A tool for obtaining maximally-localised Wannier functions,” Com-
puter Physics Communications 178, 685 (2008).
[131] S. W. Jang, M. Y. Jeong, H. Yoon, S. Ryee, and M. J. Han, “Microscopic un-

derstanding of magnetic interactions in bilayer CrI3 ,” Physical Review Materials
3, 031001(R) (2019).
[132] N. Sivadas, S. Okamoto, X. Xu, C. J. Fennie, and D. Xiao, “Stacking-Dependent

Magnetism in Bilayer CrI3 ,” Nano Letters 18, 7658 (2018).
[133] P. Jiang, C. Wang, D. Chen, Z. Zhong, Z. Yuan, Z.-Y. Lu, and W. Ji, “Stacking
tunable interlayer magnetism in bilayer CrI3 ,” Physical Review B 99, 144401
(2019).
[134] D. Soriano, C. Cardoso, and J. Fernández-Rossier, “Interplay between inter-

layer exchange and stacking in CrI3 bilayers,” Solid State Communications 299,
113662 (2019).
[135] L. Chen, J.-H. Chung, B. Gao, T. Chen, M. B. Stone, A. I. Kolesnikov,

Q. Huang, and P. Dai, “Topological Spin Excitations in Honeycomb Ferro-
magnet CrI3 ,” Physical Review X 8, 041028 (2018).
[136] J. Cenker, B. Huang, N. Suri, P. Thijssen, A. Miller, T. Song, T. Taniguchi,

K. Watanabe, M. A. McGuire, D. Xiao, and X. Xu, “Direct observation of two-
dimensional magnons in atomically thin CrI3 ,” Nature Physics 17, 20 (2021).
[137] P. A. Maksimov and A. L. Chernyshev, “Rethinking α−RuCl3 ,” Physical Re-

view Research 2, 033011 (2020).
[138] L. Chen, J.-H. Chung, T. Chen, C. Duan, A. Schneidewind, I. Radelytskyi, D. J.

Voneshen, R. A. Ewings, M. B. Stone, A. I. Kolesnikov, B. Winn, S. Chi, R. A.
Mole, D. H. Yu, B. Gao, and P. Dai, “Magnetic anisotropy in ferromagnetic
CrI3 ,” Physical Review B 101, 134418 (2020).
[139] G. Kresse and J. Furthmller, “Efficiency of ab-initio total energy calculations

for metals and semiconductors using a plane-wave basis set,” Computational
Materials Science 6, 15 (1996).
114 BIBLIOGRAPHY
[140] G. Kresse and J. Furthmüller, “Efficient iterative schemes for ab initio total-
energy calculations using a plane-wave basis set,” Physical Review B 54, 11169
(1996).
[141] J. P. Perdew, K. Burke, and M. Ernzerhof, “Generalized Gradient Approxima-

tion Made Simple [Phys. Rev. Lett. 77, 3865 (1996)],” Physical Review Letters
78, 1396 (1997).
[142] B. T. Thole and G. van der Laan, “Linear relation between x-ray absorption
branching ratio and valence-band spin-orbit expectation value,” Physical Re-
view A 38, 1943 (1988).
[143] G. van der Laan and B. T. Thole, “Local Probe for Spin-Orbit Interaction,”
[144] T. Ozaki and H. Kino, “Numerical atomic basis orbitals from H to Kr,” Physical
Review B 69, 195113 (2004).
[145] T. Ozaki and H. Kino, “Efficient projector expansion for the ab initio LCAO
method,” Physical Review B 72, 045121 (2005).
[146] M. J. Han, T. Ozaki, and J. Yu, “O(N ) LDA + U electronic structure calcula-
tion method based on the nonorthogonal pseudoatomic orbital basis,” Physical
Review B 73, 045110 (2006).
[147] S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. Humphreys, and A. P.

Sutton, “Electron-energy-loss spectra and the structural stability of nickel oxide:
An LSDA+U study,” Physical Review B 57, 1505 (1998).
[148] E. M. Seibel, J. H. Roudebush, M. N. Ali, K. Ross, and R. Cava, “Structure

and magnetic properties of the spin-1/2-based honeycomb NaNi2 BiO6−δ and its
hydrate NaNi2BiO6−δ · 1.7H2 O,” Inorganic chemistry 53, 10989 (2014).
[149] T. Ozaki, “Variationally optimized atomic orbitals for large-scale electronic

structures,” Physical Review B 67, 155108 (2003).
[150] H. Liu and G. Khaliullin, “Pseudospin exchange interactions in d7 cobalt com-

pounds: Possible realization of the Kitaev model,” Physical Review B 97,
014407 (2018).
[151] A. M. Glazer, “The classification of tilted octahedra in perovskites,” Acta Crys-

tallographica Section B 28, 3384 (1972).
BIBLIOGRAPHY 115
[152] G. Kresse and D. Joubert, “From ultrasoft pseudopotentials to the projector

augmented-wave method,” Physical Review B 59, 1758 (1999).
[153] J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria,

L. A. Constantin, X. Zhou, and K. Burke, “Restoring the Density-Gradient
Expansion for Exchange in Solids and Surfaces,” Physical Review Letters 100,
136406 (2008).
[154] J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria, L. A.

Constantin, X. Zhou, and K. Burke, “Erratum: Restoring the Density-Gradient
Expansion for Exchange in Solids and Surfaces [Phys. Rev. Lett. 100, 136406
(2008)],” Physical Review Letters 102, 039902 (2009).
[155] B. J. Kim, H. Ohsumi, T. Komesu, S. Sakai, T. Morita, H. Takagi, and

T. Arima, “Phase-Sensitive Observation of a Spin-Orbital Mott State in
Sr2IrO4,” Science 323, 1329 (2009).
[156] N. Marzari and D. Vanderbilt, “Maximally localized generalized Wannier func-

tions for composite energy bands,” Physical Review B 56, 12847 (1997).
[157] I. Souza, N. Marzari, and D. Vanderbilt, “Maximally localized Wannier func-

tions for entangled energy bands,” Physical Review B 65, 035109 (2001).
[158] J.-M. Carter, V. Shankar V., and H.-Y. Kee, “Theory of metal-insulator tran-
sition in the family of perovskite iridium oxides,” Physical Review B 88, 035111
(2013).
[159] J. Nichols, J. Terzic, E. G. Bittle, O. B. Korneta, L. E. De Long, J. W. Brill,

G. Cao, and S. S. A. Seo, “Tuning electronic structure via epitaxial strain in
Sr2IrO4 thin films,” Applied Physics Letters 102, 141908 (2013).
[160] F. Wang and T. Senthil, “Twisted Hubbard Model for Sr2 IrO4 : Magnetism and
Possible High Temperature Superconductivity,” Physical Review Letters 106,
136402 (2011).
[161] G. Pizzi, V. Vitale, R. Arita, S. Blgel, F. Freimuth, G. Géranton, M. Gibertini,

D. Gresch, C. Johnson, T. Koretsune, J. Ibañez-Azpiroz, H. Lee, J.-M. Lihm,
D. Marchand, A. Marrazzo, Y. Mokrousov, J. I. Mustafa, Y. Nohara, Y. No-
mura, L. Paulatto, S. Poncé, T. Ponweiser, J. Qiao, F. Thle, S. S. Tsirkin,
M. Wierzbowska, N. Marzari, D. Vanderbilt, I. Souza, A. A. Mostofi, and
J. R. Yates, “Wannier90 as a community code: new features and applications,”
Journal of Physics: Condensed Matter 32, 165902 (2020).
116 BIBLIOGRAPHY
[162] S. Y. Park, S. H. Do, K. Y. Choi, D. Jang, T. H. Jang, J. Schefer, C. M.

Wu, J. S. Gardner, J. M. S. Park, J. H. Park, and S. Ji, “Emergence of the
Isotropic Kitaev Honeycomb Lattice with Two-dimensional Ising Universality
in α-RuCl3 ,” (2016), arXiv:1609.05690 [cond-mat.mtrl-sci] .
[163] Q. Luo, P. P. Stavropoulos, and H.-Y. Kee, “Spontaneous Chiral-Spin Ordering

in Spin-Orbit Coupled Honeycomb Magnets,” (2020), arXiv:2010.11233 [cond-
mat.str-el] .
[164] A. Catuneanu, Y. Yamaji, G. Wachtel, Y. B. Kim, and H.-Y. Kee, “Path

to stable quantum spin liquids in spin-orbit coupled correlated materials,” npj
Quantum Materials 3, 23 (2018).
[165] I. Kimchi, R. Coldea, and A. Vishwanath, “Unified theory of spiral magnetism

in the harmonic-honeycomb iridates α, β, and γ Li2 IrO3 ,” Physical Review B
91, 245134 (2015).
[166] M. Majumder, R. S. Manna, G. Simutis, J. C. Orain, T. Dey, F. Fre-

und, A. Jesche, R. Khasanov, P. K. Biswas, E. Bykova, N. Dubrovinskaia,
L. S. Dubrovinsky, R. Yadav, L. Hozoi, S. Nishimoto, A. A. Tsirlin, and
P. Gegenwart, “Breakdown of Magnetic Order in the Pressurized Kitaev Iri-
date β−Li2 IrO3 ,” Physical Review Letters 120, 237202 (2018).
[167] L. S. I. Veiga, K. Glazyrin, G. Fabbris, C. D. Dashwood, J. G. Vale, H. Park,

M. Etter, T. Irifune, S. Pascarelli, D. F. McMorrow, T. Takayama, H. Takagi,
and D. Haskel, “Pressure-induced structural dimerization in the hyperhoney-
comb iridate β−Li2 IrO3 at low temperatures,” Physical Review B 100, 064104
(2019).
[168] M. A. McGuire, “Crystal and Magnetic Structures in Layered, Transition Metal

Dihalides and Trihalides,” Crystals 7, 121 (2017).
[169] T. Holstein and H. Primakoff, “Field Dependence of the Intrinsic Domain Mag-
netization of a Ferromagnet,” Physical Review 58, 1098 (1940).
[170] J. Colpa, “Diagonalization of the quadratic boson hamiltonian,” Physica A 93,

327 (1978).

Stavropoulos Panagiotis Peter PHD Thesis

Uploaded by

Copyright:

Available Formats

Stavropoulos Panagiotis Peter PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stavropoulos Panagiotis Peter PHD Thesis

Uploaded by

Copyright:

Available Formats

Emergent phenomena in correlated materials with strong spin-orbit

Panagiotis Peter Stavropoulos

A thesis submitted in conformity with the requirements

© Copyright 2021 by Panagiotis Peter Stavropoulos

Panagiotis Peter Stavropoulos

List of Tables vii

Author contributions xvi

2 Counter rotating spiral order 9

3 Higher-spin Kitaev models 23

4 Ab initio study of frustrated magnets 57

B Spin wave theory of a ferromagnet on honeycomb lattice 94

C Details of the jeﬀ infinite BR 97

3.1 Energy spectrum of the on-site Hamiltonian H0 at the M and X sites.

1.1 Splitting of d5 due to octahedral Oh crystal field splitting ∆C and spin-

2.1 Two conventional unit cells of the ideal hyper-honeycomb structure

2.4 Phase diagram obtained by exact diagonalization on the 24-sites cluster

2.5 Probability maps of P . (a) At the hidden SU(2) point −K = −Γ,

3.2 Schematic representation of the three types of superexchange processes

3.8 Spin model parameters J, K, Γ, Γ′ , and Ac estimated from ab initio

4.1 Hyper-honeycomb structure of β−Li2 IrO3 . The conventional orthorhom-

4.2 Pictorial representation of excitation from p-core orbitals to valence

B.1 X Y and Z bonds of the J − K − Γ − Γ′ − Ac shown in red, green and

1.1 The jeﬀ state

1.2 Finding an eﬀective model: strong-coupling expansion

1.3 Bond-dependent interactions

Counter rotating spiral order

Identifying spin liquids and nearby phases in quantum spin models in 3D is a

2.1.1 Structure and spin model

2.1.2 SU(2) hidden symmetry: 12-site transformation

Previous studies on hidden symmetry via diﬀerent sets of sublattice transforma-

transformation, dubbed the T12 transformation, is given by,

The T12 transformation can be considered as two 6-site transformations denoted by

2.1.3 Counter-rotating spiral order: signature of SU(2) symmetry in 3D

As mentioned above, the SU(2) Heisenberg Hamiltonian is found when K = Γ in

commensurate ordering wavevector.

2.1.4 Phase diagram and moment direction

We normalize the exchange strengths by K = − sin ϕ and Γ = cos ϕ where

24 site PBC 24 site OBC

Pinning of magnetic moment direction

|Ψansatz (θ, ϕ)⟩ = UT12 |Ψ̃(θ, ϕ)⟩. (2.7)

Multiple consecutive rotations can always be re-written as one rotation, which is

Lable Rotations ⃗nT12 (i) ωT12 (i)

2.1.5 Eﬀects of interchain coupling: evolution from 1D to 3D

As previously stated, in addition to the S phase, which encompasses a large region

The x = 0 limit was studied with exact diagonalization on chains of up to N = 30

As a consequence of these transformations, the KΓ model on the chain is mapped

(a) x = 0.0 (b) x = 0.2

(c) x = 0.5 (d) x = 1.0

Figure 2.6: Order parameters ⟨S e2 e2 e2

when x ̸= 0, the T 6 transformations no longer map to the Heisenberg model because

Higher-spin Kitaev models

Kitaev model in solid state materials.

3.1 Spin S=1 model from eg orbitals

3.1.1 Kanamori interaction

3.1.2 Tight-binding Hamiltonian

six diﬀerent directions to its surrounding X sites is

Hopping Matrix M(i,j)

3.1.3 Strong-coupling expansion

Treating the tight-binding Hamiltonian Hkin as a perturbation to the on-site Hamil-