You are on page 1of 4

RAPID COMMUNICATIONS

PHYSICAL REVIEW B 83, 041104(R) (2011)

Geometrical frustration and the competing phases of the Sn/Si(111)

3R30 surface systems

Gang Li, Manuel Laubach, Andrzej Fleszar, and Werner Hanke


Institute for Theoretical Physics and Astrophysics, University of Wurzburg, Am Hubland, D-97074 Wurzburg, Germany
(Received 22 October 2010; published 19 January 2011)

We investigate the electronic correlation effects on the Sn/Si(111) 3 3R30 surface by combining the
ab initio density-functional approach with the dynamical mean-field theory, the variational cluster approach, and
the dual fermion (DF) method. A metal-insulator transition with first or second order at finite or zero temperature
is predicted at a critical on-site Hubbard Uc 0.65 eV and the system is proven to be short-range correlated. The
electron-electron interaction favors a row-wise antiferromagnetic (RW-AFM) order, although the noninteracting
system does not have a pronounced nesting at the Fermi surface. The stabilization of the RW-AFM order over the
120 AF one for certain interaction strengths is shown to be due to the longer-range single-particle hopping terms.
DOI: 10.1103/PhysRevB.83.041104

PACS number(s): 71.10.Fd, 71.27.+a, 71.30.+h

Low-dimensional adatom systems on semiconductor surfaces with the triangular structures are ideal, i.e., wellcontrolled, model systems to study geometrical frustration
effects and their interplay with magnetism in two dimensions
(2D). The ground state is largely degenerate and, as a
consequence, contains rich physics.1,2 Experimentally,
a 2D
triangular
system
can
be
realized
at
the
Sn/Si(111)
3

3R30 surface. This and also related adatom systems have


attracted much interest in the last few years. Theoretically, both
density-functional ab initio and many-body model calculations
have been carried out. The ground state has been predicted
to be metallic in the local-density approximation (LDA)
and insulating within the local spin-density approximation
(LSDA) + U method, when the electron interaction parameter
Uc becomes larger than about 1 eV.3 This is consistent with
experimental findings observing a metal-insulator transition
(MIT) for temperatures between 60 and 70 K.4 Model calculations employing the Hubbard model also point to the existence
of a MIT in this system. Moreover, these calculations also
indicate that a magnetically ordered 120 state may show up in
the insulating phase,5 as well as a spin-liquid state for reduced
Coulomb interactions.68 Despite the intensive study of various
related models, a reliable many-body calculation which can
equally describe various competing phases, i.e., paramagnetic
metal, 120 Neel order, and spin-liquid states, is still missing
for the material parameters relevant to experimentally realized
adatom systems, such as the Sn/Si(111) surface. Thus, in this
paper we focus on the investigation of the many-body effects
at this surface using complementary techniques, such as the
dynamical meant-field theory (DMFT),9 the variational cluster
approach (VCA),10 and the dual fermion (DF)11 methods.
As a starting point, ab initio density-functional theory
(DFT) calculations are performed within the LDA scheme,
using a plane-wave, pseudopotential, periodic-slab code.12
The geometrical and electronic structures obtained agree very
well with previous LDA calculations for this system.13 In
Fig. 1, the LDA KohnSham (KS) band structure k along
some directions of the planar Brillouin zone (BZ) is shown. A
characteristic feature is the presence of a single, half-filled
and rather narrow band, which crosses the Fermi surface
and is well separated from other bands. This band is mostly
due to the Sn(5pz ) and partly to the Sn(5s) orbitals. Based
on this observation, we construct a one-band tight-binding
1098-0121/2011/83(4)/041104(4)

(TB) model using maximally localized Wannier functions


(MLWFs). The Wannier functions build the basis for the
determination of the TB hopping parameters tij , i.e., ti,j
i|HKS |j , where |i and |j  denote MLWFs. The resulting
TB energy band kW , with the TB-MLWF hopping parameters
of Table I, is virtually identical with the original band k . In
Fig. 1, an intensity plot of the kW band is plotted in the first
BZ. No pronounced nesting of the Fermi surface is observed
for this half-filled band.
In principle, we could use in the second step an approximate, e.g., constrained random-phase approximation,
calculation using the MLWFs to obtain reasonable estimates
for the interaction parameters, in particular, the Hubbard U ,
(see, e.g., Ref. 14 and references therein). Here, we are more
interested in the critical Uc values for a MIT. Thus, we use
a different scenario; i.e., we explore U in a relatively wide
range. This is motivated by the still missing experimental
results concerning the energy gap of the insulating phase. On
the basis of this Hubbard model, we address next the effect
of the electron Coulomb interactions by applying the DMFT,
the VCA, and the DF method. We start our calculations from
the paramagnetic phase and identify the instability toward a
magnetic order, which is discussed later in this paper, with an
instability of the corresponding susceptibilities.
For a system which is dominated by local fluctuations,
the DMFT is a very good approximation. However, the
neglect of nonlocal correlations can result in different physics
in finite dimensions.15 In the DF method, both the local
and nonlocal correlations are included, and, as a result,
the self-energy function is also momentum dependent. The
correlations of the DF variables are normally smaller than
those of the lattice fermions. Therefore, we can determine the
nonlocal self-energy from the perturbation expansion of the
DF variables.16 Our calculations are based on the first two DF
self-energy diagrams. The DMFT equation was solved with the
continuous-time quantum Monte Carlo method.17 The vertex
functions were obtained for a large enough frequency cutoff,
the first BZ was discretized to 32 32 momentum grids, and
the full circle was employed to ensure the convergence of both
the hybridization function and the DF self-energy.18
Insulating phase. Figure 2 shows the one-particle local
density of states (DOS) determined from the two methods.
We explore U in a wide range around the critical value. In
Fig. 2(a), the green curves display the DOS A() for different

041104-1

2011 American Physical Society

RAPID COMMUNICATIONS

0.6

T = 300 K
T = 61 K

(a)

0.4
0.2

Double Occupancy

PHYSICAL REVIEW B 83, 041104(R) (2011)

Quasiparticle Weight

LI, LAUBACH, FLESZAR, AND HANKE

0.07
T = 61 K
0.06
(b)
0.05
0.04
0.03
0.02

0.3 0.4 0.5 0.6 0.7 0.8 0.9

0.6

U (eV)

A() (eV-1)

5
4

U=0.52eV, DMFT
U=0.52eV, DF

0.64 0.68
U (eV)

0.72

U=0.63eV, DMFT
U=0.63eV, DF

(c)

2
1
0
-0.8

-0.4

0.4

0.8

interactions at T 61 K. The dotted and solid lines represent


the DMFT and DF solutions, respectively. With the increase
of the interaction strength, the charge gap starts to open at
UcDMFT 0.75 eV in the DMFT method, and UcDF 0.65 eV
in the DF method. Thus, there is clear evidence of the existence
of the MIT in this system. Moreover, the nonlocal correlations
contained in the DF method have a sizable effect on the MIT.
They reduce Uc by about 0.1 eV.
At U 0.52 eV, DMFT predicts the surface state to be
metallic, which is confirmed by the DF calculation. However,
A() at the Fermi level in the DF approach is suppressed
due to the inclusion of the nonlocal correlations. This effect
is significantly more pronounced when U is close to Uc . As
shown in Fig. 2(c), at U 0.63 eV, the nonlocal correlations
included in the DF calculations change the surface from
metallic, as predicted by the DMFT, to nearly insulating. This
clearly shows the nonlocal nature of the MIT at the Sn/Si(111)
surface. In order to further understand to what extent this
system is short- or long-range correlated, we additionally
carried out VCA calculations. This technique takes short-range
electronic correlations via the exact diagonalization of the
reference cluster. Here the first three hopping parameters
in Table I are used, where the three-site cluster with three
additional bath sites was employed as a reference system.
Notice, the charge gap opens at UcVCA  UcDF . We conclude,
therefore, that the Sn/Si(111) surface is dominated by shortrange correlations and that the three-site cluster is sufficient
for addressing correctly the physics of the MIT.
TABLE I. The TB-MLWF hopping parameters for Sn/Si(111)
given in eV.
t00
0.0105

t10

t11 /t10

t20 /t10

t21 /t10

t22 /t10

0.0527

0.3881

0.1444

0.0228

0.0318

-1

FIG. 1. (Color online) (a) The KohnSham Bloch bands along


the high-symmetry points determined from the LDA calculation for
the Sn/Si(111) surface. The band crossing the Fermi level is used
to construct a single-band TB model. (b) The intensity plot of the
single-band crossing the Fermi level. (c) The first BZ.

A() (eV )

(eV)
7
U=0.47 eV
6
U=0.53 eV
5
4
3
2
1
0
-0.6
-0.4

U=0.63 eV
U=0.69 eV

-0.2

(d)

0.2

0.4

0.6

(eV)

FIG. 2. (Color online) Figs. (a) shows the DMFT and DF quasiparticle weights as a function of U for two different temperatures,
(b) shows the double occupancy for T 61 K. The comparison in
(c) is for the local density of state obtained from the DMFT and DF
methods for different interaction strengths. Dashed and solid lines
correspond to the DMFT and DF solutions. Fig. (d) displays A()
extracted from the VCA for a variety of U -values around Uc .

We also study the temperature dependence of the MIT in


this system. At Fig. 2(a), the quasiparticle weight Z  [1
Im(i T )/ T ]1 at two different temperatures is plotted. At
lower temperature, the Hubbard U interaction is the dominating reason for the transition. Further lowering temperature
essentially has no effect on the value of Uc . However, at a
significantly higher temperature, for example, T = 300 K, at
and around Uc the quasiparticle weight drops to small but
finite values. Correspondingly, no coherent peak is observed
in the local DOS and a pseudogap opens. The presence of the
temperature-dependent MIT was indeed observed in a recent
experiment4 at temperatures between 60 and 70 K. The MIT
is a bad first-order transition at finite temperature, which
follows from the numerical estimation of the quasiparticle
weight and the double occupancy. The double occupancy, as
shown in Fig. 2(b), shows a narrow coexistence region, in
which the DMFT convergence became worse. The unstable
DMFT solutions and the narrow coexistence region may come
from the fact that the temperature studied here is only slightly
below the temperature of the second-order crossover endpoint.
Thus, it appears a tough task to determine reliable values of
Uc1 and Uc2 for the MIT for the temperature studied. Uc values
obtained from the measurements of the quasiparticle weight
and the double occupancy, overall, agree with each other. The

041104-2

RAPID COMMUNICATIONS

GEOMETRICAL FRUSTRATION AND THE COMPETING . . .

PHYSICAL REVIEW B 83, 041104(R) (2011)


(a). U = 0.42 eV

-0.3

(b). U = 0.58 eV

0.9

2.5

0.8

-0.4

0.7

-0.5

PM
NMI

-0.6

0.6

1.5

0.5
0.4

0.3

-0.7
M

-0.8

0.5

0.2

0.1

0.4

0.45

0.5

0.55 0.6
U (eV)

0.65

0.7

0.75

(c). U = 0.63 eV

FIG. 3. (Color online) The grand potential from T = 0 VCA


calculation at different phases.

MIT tends to become second order at zero temperature as


shown by the VCA calculations in Fig. 3, which display a
smooth connection of the free energy at the transition point
between two phases.
Magnetic order. Several previous many-body calculations
have suggested that the triangular Hubbard model favors the
magnetically ordered 120 Neel state5,8,19 at large values of
U/t 10, which is also in the range of our Uc calculations.
Also a possibility of the spin-liquid (nonmagnetic insulating)
state was reported.6,7 Motivated by this,
the mag
we examine
netic correlations in the Sn/Si(111) 3 3R30 surface.
To this end, the static spin susceptibility q (im = 0) has
been calculated within the DF scheme, in such a way that
both the bubble and the vertex contributions are included in a
controlled approximation, which has previously been shown to
lead to a good agreement with quantum Monte Carlo (QMC)
calculations in the 2D square lattice.18 Here we also show
(Fig. 4) a test calculation of q for the triangular model with
the same parameters as used in QMC studies.20 Although we
only used the two leading diagrams11 in the construction of
the DF self-energy, the overall agreement of our results with
the QMC solutions on a 12 12 lattice is good. Both the peak
at q = K and the shoulder at q = M are reproduced.
In Fig. 5, the static spin susceptibilities at T 61 K
for different interactions are plotted. As the Fermi surface
is not perfectly nested (Fig. 1), we do not expect the spin
susceptibility to peak at K in a weak-interaction picture.
Indeed, this is what is shown in Fig. 5(a): q peaks at positions
close to but different from K, as already discussed by Santoro
et al. in Ref. 21. The amplitude of (K) is increased by
further increasing the Coulomb interaction U , as shown in
4.5

<n> = 1.0
<n> = 1.3

4
3.5
(q)

3
2.5
2
1.5
1
0.5
0

FIG. 4. (Color online) Spin susceptibility as a function of momentum for T = 0.333|t| and U = 8t at two different dopings, n = 1.0
and 1.3. The comparison to the QMC data from Ref. 20, which are
plotted as dots, shows the overall agreement.

5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0

0
(d). U = 0.67 eV

10
9
8
7
6
5
4
3
2
1
0

FIG. 5. (Color online) The spin susceptibilities q at T = 61 K


are plotted in the first BZ (the diamond zone of Fig. 1) at half-filling,
as a function of Columb interaction. Increasing the interaction, the
peak of q gradually moves from K to M, with (K) having an
amplitude comparable to that of (M).

Fig. 5(b). We would expect (q) to peak at wave vector


q = K for sufficiently large U corresponding to the 120
antiferromagnetism, as the model calculations have predicted.
However, our calculations show that further increasing U does
not result in peaking at K but at M (with no divergence),
indicating the formation of a different magnetic order, i.e., the
row-wise antiferromagnetism (RW-AFM). Note that, although
q peaks at M, the amplitude of (K) remains comparable
with (M). Thus, we believe that the true magnetic order
of the Sn/Si(111) system cannot be simply described by
either 120 -AFM or RW-AFM, which seems to coincide with
the cellular-cluster rotational invariant slave-boson study in
Ref. 22.
To understand why in this system the 120 -AFM is not
dominating for large U values, or why it is magnetically
different from previous model predictions, we performed
two calculations with different dispersion relations. In one
calculation, we only used the t10 parameter in the TB-MLWF
Hamiltonian in Table I, whereas the t11 one was additionally
included in the second calculation. As seen in Figs. 6(a) and
6(b), the noninteracting density of state changes dramatically,
when t11 is included. The first model is the widely studied
nearest-neighbor (NN) triangular model, while the second one
includes longer-range hopping and has a very similar noninteracting DOS as in our actual surface system. The static spin
susceptibilities of both systems show totally different magnetic
orders as seen from Fig. 6. In the calculation with only NN hopping, two peaks are formed at two equivalent K points in the
first BZ corresponding to the 120 -AFM order. After including
the next-nearest-neighbor (NNN) hopping, the peak moves
to M points, indicating a formation of the RW-AFM order.
Thus, the extra frustration included in the second calculation
suppresses the 120 magnetic order. Our study shows that the
NNN hopping process is essential for the correct description
of the Sn/Si(111) surface and its neglect in previous model
calculations might be one of the reasons for the differences in
the prediction of the magnetic properties of this system.

041104-3

RAPID COMMUNICATIONS

-1

9
8 (a). NN
7
6
5
4
3
2
1
0
-0.3 -0.2 -0.1

A() (eV )

-1

A() (eV )

LI, LAUBACH, FLESZAR, AND HANKE

0 0.1 0.2 0.3 0.4


(eV)

PHYSICAL REVIEW B 83, 041104(R) (2011)

14
12
(b). NNN
10
8
6
4
2
0
-0.4 -0.3 -0.2 -0.1 0
(eV)

0.1 0.2 0.3

M
qx

qy

M
qx

qy

FIG. 6. (Color online) The noninteracting DOS and the spin


susceptibility q at T = 100 K and U = 0.63 eV for the triangular
model with nearest-neighbor (NN) and next-nearest-neighbor (NNN)
hopping terms.

It is also worth noting that there is experimental evidence for


the presence of a small splitting of the photoemission Sn(4d)
line, which may be interpreted as being caused by the buckling
within the Sn monolayer.2325 However, previous ab initio
density-functional calculations excluded such a geometrical
buckling and the low-temperature photoelectron-diffraction
measurements of Modesti et al.4 indicate that all Sn
atoms have the same bonding geometry. Therefore, the
photoemission core-level splitting might alternatively be
related to the presence of a magnetic order, possibly the one

G. H. Wannier, Phys. Rev. 79, 357 (1950).


L. Balents, Nature (London) 464, 199 (2010).
3
G. Profeta and E. Tosatti, Phys. Rev. Lett. 98, 086401 (2007).
4
S. Modesti, L. Petaccia, G. Ceballos, I. Vobornik, G. Panaccione,
G. Rossi, L. Ottaviano, R. Larciprete, S. Lizzit, and A. Goldoni,
Phys. Rev. Lett. 98, 126401 (2007).
5
T. Yoshioka, A. Koga, and N. Kawakami, Phys. Rev. Lett. 103,
036401 (2009).
6
B. Kyung and A.-M. S. Tremblay, Phys. Rev. Lett. 97, 046402
(2006).
7
L. F. Tocchio, A. Parola, C. Gros, and F. Becca, Phys. Rev. B 80,
064419 (2009).
8
T. Ohashi, T. Momoi, H. Tsunetsugu, and N. Kawakami, Phys. Rev.
Lett. 100, 076402 (2008).
9
A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg, Rev. Mod.
Phys. 68, 13 (1996).
10
M. Potthoff, Eur. Phys. J. B 32, 429 (2003).
11
A. N. Rubtsov, M. I. Katsnelson, A. I. Lichtenstein, and A. Georges,
Phys. Rev. B 79, 045133 (2009).
12
P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car,
C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni,
I. Dabo et al., J. Phys. Condens. Matter 21, 395502 (2009);
[http://www.quantum-espresso.org].
13
G. Profeta, A. Continenza, L. Ottaviano, W. Mannstadt, and A. J.
Freeman, Phys. Rev. B 62, 1556 (2000).
2

proposed in this work. This issue is not definitely settled


and should be further investigated both experimentally
and theoretically.

we have investigated the Sn/Si(111) 3


In conclusion,
3R30 surface by combining ab initio calculation with
many-body techniques, i.e., DMFT, VCA, and DF methods.
The phase diagram confirms the existence of the insulating
phase. The first order at finite temperature and the second
order at T = 0 MIT occur at Uc  0.65 eV, which barely
depends on T in a lower-temperature regime. The comparison
of the DMFT and DF methods indicates that the nonlocal
correlations included in the DF method have important effects
in the system. By restricting our VCA calculation to a
three-site cluster and comparing to the DF results, we further
establish that this surface system is short-range correlated.
The magnetic correlations in the system were investigated via
the calculation of the static spin susceptibilities. The electron
interactions favor the RW-AFM order at the experimentally
relevant temperature regime for certain interactions. Unlike the
normal triangular lattice with only NN hopping, the presence
of the additional, e.g., longer-range, hopping processes in
the Sn/Si(111) surface system dramatically modifies the
noninteracting DOS, and as a result, the magnetic order is
changed from the 120 -AFM to the RW-AFM.
The work was supported by the Deutsche Forschungsgemeinschaft within Project FOR 1162, GZ: HA 1537123-1,
and by the Bavarian Supercomputing Network KNOWIHR II.
G. L. thanks S. Biermann, A. I. Lichtenstein, M. I. Katsnelson,
V. I. Anisimov, and A. N. Rubtsov for many useful discussions.
W. H. gratefully acknowledges discussions with E. Tosatti and
S. Sorrela.

14

T. Miyake, K. Nakamura, R. Arita, and M. Imada, J. Phys. Soc. Jpn.


79, 044705 (2010).
15
H. Park, K. Haule, and G. Kotliar, Phys. Rev. Lett. 101, 186403
(2008).
16
H. Hafermann, G. Li, A. N. Rubtsov, M. I. Katsnelson,
A. I. Lichtenstein, and H. Monien, Phys. Rev. Lett. 102, 206401
(2009).
17
A. N. Rubtsov, V. V. Savkin, and A. I. Lichtenstein, Phys. Rev. B
72, 035122 (2005).
18
G. Li, H. Lee, and H. Monien, Phys. Rev. B 78, 195105
(2008).
19
H. Lee, G. Li, and H. Monien, Phys. Rev. B 78, 205117
(2008).
20
N. Bulut, W. Koshibae, and S. Maekawa, Phys. Rev. Lett. 95,
037001 (2005).
21
G. Santoro, S. Scandolo, and E. Tosatti, Phys. Rev. B 59, 1891
(1999).
22
S. Schuwalow, D. Grieger, and F. Lechermann, Phys. Rev. B 82,
035116 (2010).
23
M. Gothelid, M. Bjorkqvist, T. M. Grehk, G. Le Lay, and U. O.
Karlsson, Phys. Rev. B 52, R14352 (1995).
24
R. I. G. Uhrberg, H. M. Zhang, T. Balasubramanian, S. T. Jemander,
N. Lin, and G. V. Hansson, Phys. Rev. B 62, 8082 (2000).
25
A. A. Escuadro, D. M. Goodner, J. S. Okasinski, and M. J. Bedzyk,
Phys. Rev. B 70, 235416 (2004).

041104-4

You might also like