Phase diagram of a distorted kagome antiferromagnet and application to Y-kapellasite

– Supplementary Information –
Max Hering,1, 2 Francesco Ferrari,3 Aleksandar Razpopov,3 Igor I.
Mazin,4 Roser Valentı́,3 Harald O. Jeschke,5 and Johannes Reuther1, 2
1
Helmholtz-Zentrum Berlin für Materialien und Energie, Hahn-Meitner Platz 1, 14109 Berlin, Germany
2
Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universität Berlin, 14195 Berlin, Germany
3
Institute for Theoretical Physics, Goethe University Frankfurt,
Max-von-Laue-Straße 1, 60438 Frankfurt am Main, Germany
4
Department of Physics and Astronomy, and Quantum Science and Engineering Center,
George Mason University, Fairfax, VA 22030, USA
5
Research Institute for Interdisciplinary Science, Okayama University, Okayama 700-8530, Japan
(Dated: November 29, 2021)

Supplementary Note 1. Finite-size clusters

optimal value of φ (in degrees)
210◦ ◦
240→
−−
Most of the numerical calculations of this work are per-
formed on finite-size clusters with periodic boundary condi-
200◦
tions. We consider two classes of fully symmetric finite-size
clusters, type-I and type-II, which are defined by two differ-
ent choices for the boundary conditions. Type-I clusters are
constructed by imposing periodic boundary conditions along 190◦
the vectors T~1 = L~a1 and T~2 = L~a2 , for a total of N = 9L2
sites. Only the Type-I clusters in which L is a multiple of 3
can accommodate the Q ~ = (1/3, 1/3) magnetic order. For
type-II clusters, instead, we take periodic boundary condi- 180◦
0.0 0.5 1.0 1.5 2.0 2.5 3.0
tions along T~1 = L(~a1 − ~a2 ) and T~2 = L(2~a1 + ~a2 ), which
are parallel to the axes of the ideal (undistorted) kagome J/J9
lattice. Type-II clusters contain N = 27L2 sites, and they
can always accommodate the Q ~ = (1/3, 1/3) order, regard-
less of the value of L. In Supplementary Fig. 1 we show Supplementary Figure 2. Optimal value of the angle φ for the
Q~ = (1/3, 1/3) magnetic order of Fig. 1c of the main text as a
the small finite-size clusters employed for the iterative min-
function of the J/J9 ratio (J 0 = 0).
imization calculations, i.e. the L = 2 type-I cluster with
N = 36 sites and the L = 1 type-II cluster with N = 27
sites.
~ = (1/3, 1/3)
Supplementary Note 2. More on the Q
ordered phase

In the limit J 0 = 0, the energy of the classical spin con-

figuration is given by Eq. (7) of the main text. The optimal
value of φ is obtained by the minimization of this energy
and its dependency on the ratio J/J9 is shown in Supple-
N = 27 sites mentary Fig. 2.
(type-II, L = 1) Further insight is gained by looking at two extreme lim-
its of the ratio J/J9 . When J  J9 (strong hexagons
limit), the angle φ → π, namely the spins of A sites form
an antiferromagnetic pattern around the J9 -hexagons. On
the other hand, in the opposite limit, J  J9 (trimer
N = 36 sites
(type-I, L = 2) limit), the angle φ → 4π/3 and the spins of sublattice
B are antiferromagnetically aligned with respect to the two
nearest-neighboring A sites, thus forming antiferromagnetic
trimers as shown in the main text, Fig. 1d. Consider-
Supplementary Figure 1. Finite-size clusters used for iterative ing the trimers as single effective spins, we note that they
minimization, classical Monte Carlo, and variational Monte Carlo form a kagome lattice and they interact through an effec-
calculations. tive antiferromagnetic coupling (each trimer is connected
 2

name dCu−Cu (Å) assignment

J1 ≡ J 0 3.24978 inplane 1nn
J2 ≡ J9 3.36832 inplane 1nn
J3 ≡ J 3.37619 inplane 1nn
J4 5.67876 interlayer
J5 5.70735 inplane 2nn
J6 5.76750 inplane 2nn
J8 5.86065 interlayer

U (eV) J1 (K) J2 (K) J3 (K) J4 (K) J5 (K) J6 (K) J8 (K) TCW (K)
4 11.3(9) 161.1(6) 189.1(9) -0.5(1.0) 4.2(6) 0.5(6) 2.0(9) -122
4.5 10.2(7) 149.3(5) 174.0(7) -0.9(8) 3.8(5) 0.5(5) 1.4(7) -113
5 9.1(6) 138.7(4) 160.3(6) -1.2(7) 3.4(4) 0.4(4) 0.9(6) -104
5.23 8.7(6) 134.2(4) 154.4(6) -1.3(7) 3.2(4) 0.4(4) 0.7(6) -100
5.5 8.2(5) 129.1(3) 147.8(5) -1.4(5) 3.0(3) 0.4(3) 0.5(5) -96
6 7.4(4) 120.2(3) 136.4(4) -1.4(5) 2.7(3) 0.3(3) 0.2(4) -89
6.5 6.6(3) 112.0(2) 125.8(3) -1.5(4) 2.3(2) 0.3(2) 0.0(3) -82
7 5.9(3) 104.4(2) 116.0(3) -1.4(3) 2.0(2) 0.2(2) -0.2(3) -76
7.5 5.2(2) 97.2(2) 106.9(2) -1.4(3) 1.7(2) 0.2(2) -0.3(2) -70
8 4.5(2) 90.5(1) 98.5(2) -1.3(2) 1.5(2) 0.1(2) -0.4(2) -65

Supplementary Table 1. Exchange couplings of Y3 Cu9 (OH)19 Cl8 (SXRD structure), calculated within GGA+U and 6 × 6 × 6 k points.
The experimental value TCW = −100 K of the Curie Weiss temperature is matched by the line in bold face. Statistical errors are
indicated.

name dCu−Cu (Å) assignment

J1 ≡ J 0 3.25310 inplane 1nn
J2 ≡ J 3.36068 inplane 1nn
J3 ≡ J9 3.37591 inplane 1nn
J4 5.68423 interplane 1nn
J5 5.68831 inplane 2nn
J6 5.76630 inplane 2nn
J7 5.77832 inplane 2nn

U (eV) J1 (K) J2 (K) J3 (K) J4 (K) J5 (K) J6 (K) J7 (K) TCW (K)
4 11.2(7) 151.4(8) 154.3(5) -1.0(7) 5.7(5) 0.8(5) 3.3(7) -108
4.46 10.3(7) 139.8(8) 143.9(5) -1.3(7) 5.2(5) 0.8(5) 2.6(7) -100
4.5 10.3(6) 138.8(7) 143.0(4) -1.3(6) 5.2(4) 0.7(4) 2.6(6) -99
5 9.4(5) 127.4(6) 132.7(3) -1.5(5) 4.6(3) 0.6(3) 1.9(5) -91
5.5 8.5(4) 117.1(5) 123.4(3) -1.6(4) 4.1(3) 0.5(3) 1.4(4) -84
6 7.7(3) 107.7(4) 114.9(2) -1.6(3) 3.6(2) 0.5(2) 1.1(3) -78
6.5 6.8(3) 99.1(3) 107.0(2) -1.6(3) 3.2(2) 0.4(2) 0.7(3) -72
7 6.1(2) 91.1(3) 99.6(2) -1.5(2) 2.8(2) 0.3(2) 0.5(2) -66
7.5 5.3(2) 83.8(2) 92.7(1) -1.4(2) 2.4(2) 0.2(2) 0.3(2) -61
8 4.5(2) 77.0(2) 86.3(1) -1.3(2) 2.1(1) 0.2(1) 0.1(2) -56

Supplementary Table 2. Exchange couplings of Y3 Cu9 (OH)19 Cl8 (SND structure), calculated within GGA+U and 6 × 6 × 6 k points.
The experimental value TCW = −100 K of the Curie Weiss temperature is matched by the line in bold face. Statistical errors are
indicated.

to four other trimers by a J9 bond). In this limit, the Supplementary Note 3. More on the classical spin liquid
~ = (1/3, 1/3) magnetic order of our model reduces to the
Q phase
√ √
3 × 3 order [3] of the effective kagome lattice made by
the antiferromagnetic trimers. It is worth mentioning that the J = J 0 line represents
a special case within this region, where coplanar ground
As discussed in the main text, the linear spin wave dis- states with L~ 4 = 0 can be analytically defined. Indeed, in
persion of the Q ~ = (1/3, 1/3) order in the strong hexagon
this special limit, for each triangle we can write
(J  J9 , Fig.3a) and trimer (J  J9 , Fig.3c) limits share
similarities with those of the√triangular
√ lattice 120◦ order [1] ~ 4 = JS
L ~b + S
~a + S
~ a0 , (1)
and of the kagome lattice 3 × 3 order [3], respectively.
In Supplementary Fig. 3 we show the linear spin wave dis- where a and a0 are two neighboring sites of sublattice A,
persions for the latter two cases, for comparison. b is a site of sublattice B and we have set J9 = 1 for
 3

Supplementary Figure 3. Linear spin wave dispersions for a the

120◦ magnetic order in the antiferromagnetic
√ Heisenberg
√ model
on the triangular lattice [1] and b the 3 × 3 order in the
antiferromagnetic Heisenberg model on the kagome lattice [2].
We note that in the latter case the true ground state of the
S = 1/2 system is not magnetically ordered. A zero-energy
dispersionless
√ √mode is present in the linear spin-wave spectrum
of the 3 × 3 order on the kagome lattice [2]. In both panels
Supplementary Figure 5. Third band of the coupling matrix
the spin wave energies are plotted along the Γ − K − K10 − Γ
[see Eq. (2)] depicted in the first Brillouin zone (gray hexagon).
line (see Fig.1b of the main text).
Energies E are indicated relative to the bottom of the lowest
band. The dashed line marks a valley in the band structure
which may explain the system’s unusual thermal fluctuations.

t9 eiϕ9
t9 ei(ϕ9 +π)
teiϕ
tei(ϕ+π)
t0

Supplementary Figure 6. Optimal hopping pattern of the best

variational Ansatz [Eq. (5)] for the effective spin model of Y-
Kapellasite. The empty (full) dots represent the sites of sublat-
tice A (B). The arrows encode the convention for complex hop-
Supplementary Figure 4. Value of h|L ~ 4 |i (averaged over tri- pings: i → j represents the hopping terms ti,j c†i,σ cj,σ (σ =↑, ↓).
angles) in the optimal classical ground state found by iterative A total of five real parameters is used to parametrize the hopping
minimization. The results are obtained in the classical spin liquid part of the variational Ansatz (t9 , t, t0 , ϕ9 , ϕ).
phase, on the N = 36 sites cluster sketched in Supplementary
Fig. 1 (type-I, L = 2, see Supplementary Note 1 for details).
Close to the inner boundaries of the spin-liquid phase, we find
a number of points where the system is characterized by two
distinct non-coplanar degenerate ground states with h|L ~ 4 |i > 0. the J9 -hexagons, where we have Sa = (cos(θ), sin(θ), 0)
We note that the N = 36 sites cluster cannot accommodate the for odd sites and Sa0 = (cos(θ), − sin(θ), 0) for even sites,
~ = (1/3, 1/3) order.
Q with θ = arccos(−J/2). The same analysis can be repeated
for each unit cell separately and coplanar ground states can
be constructed, as discussed also in Ref. [4], where the J =
simplicity. Since this expression is valid for all triangles of J 0 model is investigated in connection to the physics of
the system, simple solutions with L ~ 4 = 0, ∀4 can be found Volborthite. We note that this solution is connected to the
by focusing on a single unit cell. There, we can set, e.g., Q~ = 0 coplanar order, since for J = J 0 = 2J9 we get θ = π,
Sb = (1, 0, 0) for all sites of sublattice B and we can choose i.e. the spins inside the J9 -hexagons are ferromagnetically
an alternating pattern for the sites of sublattice A forming arranged, and antiparallel to the spins of sublattice B.
 4

2 2'

1 1'
1 1' 2 2 2 2
2' 2' 1' 1'

2 2' 3 3' 2 4' 4 2'

4 3' 3 4' 4 3' 3 4'
2 2'
4' 3 3' 4 4' 3 3' 4
1 1' 1 3' 3 1'
1 1'

1 1' 1 1' 1 2' 1 2'

Supplementary Figure 7. Diagrammatic representation of the PFFRG flow equations for ΣΛ (circle) and ΓΛ (square). Slashed (bare)
arrows depict the single-scale propagator S Λ (dressed propagator GΛ ). Since the two-particle vertex couples to the three-particle
vertex (hexagon), the flow equations are truncated at this level. External legs contain one full set of quantum numbers each and
internal quantum numbers have to be summed up or integrated.

name dCu−Cu (Å) Ji (K) assignment

J1 3.24978 8.4(9) inplane 1nn Cu1-Cu2
J2 3.36832 133.1(6) inplane 1nn Cu1-Cu1
J3 3.37619 152.0(9) inplane 1nn Cu1-Cu2
J4 5.67876 0.3(8) interlayer
J5 5.70735 1.0(5) inplane 2nn Cu1-Cu1
J6 5.76750 0.6(6) inplane 2nn Cu2-Cu2
J7 5.83040 3.0(5) inplane 2nn Cu1-Cu1
J8 5.86065 0.0(6) interlayer
J9 6.49956 -0.6(5) inplane 3nn (2J1 , Cu1-Cu1)
J10 6.51314 -0.5(6) interlayer
J11 6.57249 -0.5(5) interlayer
J12 6.61523 -1.7(7) inplane 3nn (J1 +J2 , Cu1-Cu2)
J13 6.62488 -0.3(7) inplane 3nn (J1 ,J3 hex., Cu1-Cu2)
J14 6.63443 -1.6(9) interlayer
J15 6.66444 0.2(7) interlayer
J16 6.69872 -0.2(5) interlayer
J17 6.72699 0.4(6) interlayer
J18 6.73343 1.2(5) inplane 3nn (J2 hex., Cu1-Cu1)
J19 6.74065 0.2(8) inplane 3nn (J2 +J3 , Cu1-Cu2)
J20 6.75238 -0.1(6) inplane 3nn (2J3 , Cu1-Cu1)
J21 8.04603 0.7(8) interlayer
J22 8.09002 3.2(5) interlayer
Supplementary Figure 8. Exchange couplings of Y3 Cu9 (OH)19 Cl8 J23 8.09791 0.4(7) interlayer
(SND structure) extracted with GGA+U at JH = 1 eV as func- J24 8.14162 1.0(7) interlayer
tion of interaction strength U . Solid
√ symbols:
√ P 1 cell, 7 cou-
plings extracted. Empty symbols: 2 × 2 × 1 supercell, 24 Supplementary Table 3. Exchange couplings of
couplings extracted. Y3 Cu9 (OH)19 Cl8 (SXRD structure), calculated within GGA+U
and 4 × 4 × 4 k points. The couplings are reported at U =
5.28 eV; this U value is obtained by demanding that the model
matches the experimental Curie-Weiss temperature. Statistical
Supplementary Note 4. Analytical approach to explain errors are indicated.
classical Monte Carlo data

In this Supplementary Note we discuss the origin of the which forms the basis for various analytical techniques for
unusual peak shift in the classical Monte Carlo data for the studying classical spin systems. Here, the sites i are char-
SXRD structure as a function of temperature, as observed acterized by the two indices a and ρ, specifying the unit cell
in Fig. 5 of the main text. A possible explanation for this and the sublattice of site i, respectively. Similarly, j → b, σ
behavior comes from the system’s coupling matrix in mo- and Jij → Jaρbσ , yielding the 9×9 matrix Jρσ (~k). Further-
mentum space ~ a, R
~ b specify the positions of the unit cells a and
more, R
b. Diagonalizing this matrix for the couplings of the SXRD
~ ~ ~
X
Jρσ (~k) = eik(Ra −Rb ) Jaρbσ , (2) structure one finds that the third band (when counting from
ra −~
~ rb the lowest one) is in an energy range from E ∼ 0.6J to
 5

E ∼ 1.3J relative to the bottom of the lowest band. This on top of the Gutzwiller-projected fermionic state [see
intermediate energy range roughly corresponds to the tem- Eq. (4)]. The optimal parametrization of the hoppings,
perature regime at which the peak shift in Monte Carlo is which yields the best variational energy for the exchange
observed, i.e., the third band can be expected to be con- couplings of the SXRD structure, is illustrated in Supple-
nected to these unusual thermal fluctuations. Indeed, as mentary Fig. 6.
shown in Supplementary Fig. 5 the third band exhibits a
pronounced valley between the K and Γ points (dashed line
in Supplementary Fig. 5) such that fluctuations preferably
occur along this direction. Note that the magnetic band
structure is defined in the first Brillouin zone such that all
points Γ0 , Γ00 are folded back onto Γ and, equivalently, K10 ,
K20 , K30 all map onto K. Supplementary Note 6. Additional DFT results

Supplementary Note 5. Details on the variational Monte In this Supplementary Note we present the full set of
Carlo calculations Heisenberg exchange coupling constants up to nth neigh-
bors. These values were extracted from spin-polarized ab
The variational Monte Carlo (VMC) calculations for initio DFT calculations by making use of the total energy
the Hamiltonian of Y-kapellasite make use of Gutzwiller- mapping method as explained in the main text. Supple-
projected fermionic wave functions as variational Ansätze. mentary Table 1 shows the results for the SXRD structure
The definition of these states is based on the Abrikosov (see Fig. 6 of the main text for the definition of J9 , J and
fermion representation of S = 1/2 spins, J 0 ) up to 8th neighbors, and Supplementary Table 2 shows
the results for the SND structure up to 7th neighbors, where
~i = 1 the definitions of J9 , J and J 0 are indicated in Supplemen-
X †
S ci,α~σα,β ci,β . (3)
2 tary Fig. 8. As explained in the method section of the main
α,β
text, we perform the energy mapping method by calculating
Here ~σ = (σx , σy , σz ) is a vector of Pauli matrices and far more total energies than would be minimally required
ci,α (c†i,α ) are fermionic annihilation (creation) operators. to extract the exchange interactions; for example, for ex-
Within this formalism, suitable variational states for spin tracting 7 exchange interactions, we calculate 24 energies
systems can be Pconstructed by applying the Gutzwiller pro- instead of the minimal number 8. Fitting the Hamiltonian
jector, PG = i (ni,↑ − ni,↓ )2 , to a fermionic wave func- using singular value decomposition yields a statistical error
tion, e.g. a Slater determinant [5]. The Gutzwiller op- for the exchange couplings which is indicated in Tables 1 to
erator enforces the single occupation of each lattice site 3. These errors give an indication of the uncertainty of the
and projects the fermionic wave function onto the spin Hamiltonian parameters calculated for a particular crystal
Hilbert space. In this work, we consider Gutzwiller-projected structure; however, the impact of the uncertainties in the
Jastrow-Slater variational Ansätze of the form crystal structure on the exchange couplings is not included.
|Ψ0 i = J PG |Φ0 i. (4) In order to check the presence of possible long-ranged
coupling constants, we also performed calculations for larger
Here, |Φ0 i is the ground state of an auxiliary quadratic
supercells of the SXRD structure and extracted values of
Hamiltonian of Abrikosov fermions, named H0 , which
the exchange coupling constants up to 24th neighbors, as
includes first-neighbor hopping terms (ti,j ) and a site-
shown in Table 3. This calculation is computationally very
dependent magnetic field of strength h:
demanding and cannot be repeated for all structures and
ti,j c†i,σ cj,σ + h.c. + h values of U . However, it is an important sanity check be-
XX X Q
H0 = ~i .
n̂i · S (5)
hi,ji σ i
cause it is the nature of the energy mapping method that
a calculation with a small cell yielding few exchange in-
The local orientation of the magnetic field, given by the unit teractions effectively provides information on each coupling
vector n̂Qi , is chosen such that it induces the optimal single-
with some average of unresolved longer range exchange cou-
~ classical order in the (Sx , Sy )-plane. Since we focus on
Q plings added to it. By resolving more couplings as shown in
the exchange couplings derived for the SXRD structure of Supplementary Table 3, we have verified that there are no
Y-kapellasite, our magnetic field corresponds to the best important long range exchange couplings that could have in-
classical order with Q ~ = (1/3, 1/3) periodicity. Non-trivial troduced errors into the couplings shown in Supplementary
quantum fluctuations are introduced by the hopping terms Tables 1 and 2 or that would significantly modify the mag-
ti,j of the auxiliary Hamiltonian H0 , and by the application netic properties. Comparing the coupling J5 at U = 5.23 eV
of a long-range spin-spin Jastrow obtained in the smaller cell (J5 = 3.2(4) K) to that obtained
  in the large supercell (J5 = 1.0(5) K), we find that the larger
X value 3.2 K is not substantiated, and it is reasonable to fo-
J = exp  v(i, j)Siz Sjz  . (6) cus all further analysis on the first three exchange couplings
i,j J9 , J and J 0 .
 6

Supplementary Note 7. Comparison of VMC and PFFRG magnetic order are discretized. It is interesting to note that
results taking a pure hopping Ansatz within VMC, i.e. discarding
the magnetic field variational parameter (cfr. Supplemen-
tary Note 5), we obtain a broad susceptibility profile with
While VMC and PFFRG both find magnetic long range shifted peaks, which is very similar to the one provided by
order for Y-kapellasite, the small peak shift and the rather PFFRG. However, the corresponding variational state has a
broad susceptibility profile in PFFRG are not seen in VMC, considerably higher energy. Possibly, this energy could be
which detects sharp Bragg peaks at the ordering vectors reduced if one were able to implement a magnetic field varia-
(compare Fig. 7 and Fig.8 b of the main text). Given that tional parameter for an incommensurate structure, while at
the two techniques are conceptually very different and rely the same time maintaining broad susceptibility peaks and
on different approximations, such discrepancies are not sur- thereby reaching better agreement with PFFRG. On the
prising. For example, detecting incommensurate magnetic other hand, it is possible that PFFRG and VMC results will
order within VMC is a hard task, because the calculations resemble each other more closely when promoting the PF-
are restricted to finite-size clusters and thus the vectors of FRG towards the aforementioned multi-loop scheme.

[1] A. L. Chernyshev and M. E. Zhitomirsky, Spin waves in a tri- 48, 9539 (1993).
angular lattice antiferromagnet: Decays, spectrum renormal- [4] F. Wang, A. Vishwanath, and Y. B. Kim, Quantum and clas-
ization, and singularities, Phys. Rev. B 79, 144416 (2009). sical spins on the spatially distorted kagomé lattice: Applica-
[2] A. B. Harris, C. Kallin, and A. J. Berlinsky, Possible Néel tions to volborthite Cu3 V2 O7 (OH)2 ·2H2 O, Phys. Rev. B 76,
orderings of the Kagomé antiferromagnet, Phys. Rev. B 45, 094421 (2007).
2899 (1992). [5] F. Becca and S. Sorella, Quantum Monte Carlo Approaches
[3] J. N. Reimers and A. J. Berlinsky, Order by disorder in the for Correlated Systems (Cambridge University Press, 2017).
classical Heisenberg kagomé antiferromagnet, Phys. Rev. B