You are on page 1of 27

Part III Physics

University of Cambridge

Why is the superconducting


critical temperature so high
in a FeSe/STO interface?
Research Project
May 2016
Abstract
The critical temperature (Tc ) below which bulk iron selenide (FeSe) is
superconducting is 7.9 K. However, its Tc can be increased to temperatures above the boiling point of nitrogen (77 K) if a single-unit-cell
layer of FeSe is placed on a strontium titanate (STO) substrate. There
is experimental evidence suggesting that interfacial electron-phonon
coupling is the main mechanism of enhancement of Tc in FeSe/STO.
A BCS model was applied to FeSe nanograins on STO and its predictions were compared to the experimental results presented by Zhi Li et
al. [Physical Review B 91, 060509(R) (2015)]. A coupling constant
= 0.27 was required to account for the experimentally measured normalised fluctuations of the gap. However, the expected Debye energy
~!D = 100 meV gives a bulk gap of 3.8 meV, far from the observed
16.5 meV. Electron-phonon coupling alone cannot therefore account
for the enhancement of Tc in FeSe/STO. Other mechanisms, such as
additional pairing channels and forward-scattering, have to be considered. An Eliashberg model that includes forward-scattering was
adapted to reduced dimensions.

Table of Contents
1 Introduction

2 Review of Superconductivity Mechanisms in FeSe/STO

3 The Project
3.1 The BCS Model . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Superconductivity in Reduced Dimensions . . . .
3.1.2 The Gap Equation . . . . . . . . . . . . . . . . .
3.1.3 FeSe Nanograins on STO Substrate . . . . . . . .
3.2 The Eliashberg Model . . . . . . . . . . . . . . . . . . .
3.2.1 The BCS Limit of the Eliashberg Gap Equation .
3.2.2 Low-momentum Transfer in the Eliashberg Model

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

9
9
9
10
15
17
19
20

4 Further Work

23

5 Conclusion

24

Introduction

Although the discovery of superconductivity dates back to 1911, when Onnes


cooled mercury below its critical temperature of 4.2 K, the development of
room-temperature superconductors still remains as one of the greatest challenges in physics. Nevertheless, two landmark events have brought us closer
to this ultimate goal: the discovery of high-temperature superconductivity
in Cu-based oxides by Bednorz and M
uller in 1986 [1], and the discovery of
iron-based superconductors (FeSC) in 2006 [2].
The discovery of copper-oxide superconductors (also known as cuprates)
allowed to significantly increase the highest critical temperature (Tc ) of superconductors. Indeed, the superconductor with the highest Tc confirmed
by peer-review at ambient pressure is currently a cuprate (HgBa2 Ca2 Cu3 O8 ,
Tc 133 K) [3], whereas in 1986 the highest reported Tc was of 23.3 K.
Crucially, whereas low-temperature superconductors (Tc 6 30 K) had to be
cooled using liquid helium in order to achieve the superconducting regime,
many high-temperature superconductors have higher critical temperatures
than the boiling point of nitrogen (77.4 K), so these can be cooled to superconductivity using liquid nitrogen. Current technological applications of
high-temperature superconductors include powerful electromagnets (used in
MRI and NMR machines, for example) and sensitive SQUID-based magnetometers.
In 2012, following the above mentioned discovery of FeSC superconductors, a high Tc was reported in one-unit-cell (1 UC) iron selenide (FeSe) films
grown on a Se-etched strontium titanate (STO) substrate [4], despite the very
low Tc of bulk FeSe (Tc 7.9 K). In particular, the large superconducting
gap (20 meV) and the magnetic-field-induced vortex state revealed by in situ
scanning tunneling microscopy (STM) led to a prediction of Tc 77 K for
this FeSe/STO interface. Actually, subsequent angle-resolved photoemission
spectroscopy (ARPES) studies for various substrates with dierent growth
conditions showed that the energy gap opens at temperatures ranging from
55 K to 75 K [5]. Nevertheless, this discovery introduced a powerful way
of developing new high-temperature superconductors by combining dierent
superconducting materials and substrates [6].
Thus far, the theoretical research on the superconducting properties of the
FeSe/STO interface has been exclusively focused on the full two-dimensional
layer of FeSe. This project considers instead FeSe nanograins on STO. The
finite-size eects that arise from reducing the dimensions of the supercon3

ducting system give rise to eects that are not observed in the bulk limit
and that can therefore provide extra information about the mechanism of
enhancement of Tc .
Zhi Li et al. [7] employed scanning tunneling microscopy (STM) and spectroscopy techniques to investigate the superconductivity in 1UC nanograins
of FeSe on STO substrates. These experiments have shown that for small
rectangular grains of area 50 - 150 nm2 the gap fluctuates significantly
with the grain size (Figure 1).

Figure 1: Experimental results of STM measurements performed by Zhi Li et al. [7]


for FeSe nanograins on STO. (Left) Superconducting gap versus area of nanograins
(Right) Gap asymmetry = +
versus nanograin area, with + and
indicating the deviations of the two peaks in dierential conductance above and
below EF from EF , respectively. [Picture Credits: Zhi Li et al., [7]]

Two models, within the BCS and Eliashberg theories of superconductivity, were adapted to reduced dimensions. The BCS model accounted for the
fluctuations of the gap but not for the value of the bulk gap from the experimental results above shown. Other mechanisms beyond electron-phonon
coupling, such as forward scattering or additional pairing channels, are required to explain the enhancement of Tc . The Eliashberg model includes a
cut-o in the momentum transfer.
An outline of the state of the art in this field is presented in the next
chapter. Section 3 includes a detailed description of the two models developed
in this project and the discussion of the respective results. Suggestions of
further work and the concluding remarks of the project are presented in
chapters 4 and 5, respectively.
4

Review of Superconductivity Mechanisms


in FeSe/STO

Several mechanisms that could justify this enhancement of the Tc of FeSe have
been proposed, including charge transfer from STO to FeSe, strain eects
due to the substrate and vibrational eects due to enhanced electron-phonon
coupling within the FeSe layer itself or across the interface. Dung-Hai Lee [8]
made a comprehensive review of these mechanisms in terms of the currently
available experimental observations. The conclusions are summarised below.
The absence of hole pockets at the Brillouin Zone centre, which is demanded by charge neutrality (since FeSe has two valence electrons per unit
cell), suggests that the FeSe single-layer is electron doped. However, even
though 1 UC FeSe/STO and K-doped 3 UC FeSe/STO have very similar
Fermi surfaces, the critical temperature of the former is considerably greater,
thereby suggesting that electron doping cannot be entirely responsible for the
Tc enhancement.

Figure 2: Schematic representation of the crystal structure of FeSe monolayer


grown (a) on T iO2 -terminated STO surface and (b) on O-deficient STO surface.
The O-deficiency due to Se-etching leads to an excess of negative charge around
the Sr atoms. [Picture Credits: Junhyeok Bang et al. [9]]

The transferred charge is thought to arise from the oxygen deficiency


at the STO surface, which is caused by the Se-etching as the FeSe layer is
5

grown on top of the substrate. Such absence of O atoms leads to an excess


of negative charge around the Sr atoms, which can be transferred across the
interface.
By changing the detailed properties of the interface, Peng et al. [10] concluded that tensile strain has a negligible eect on this material. This observation is in accordance with the fact that the lattice constant of the FeSe
single-layer is nearly the same as for bulk FeSe.
The fact that FeSe layers grown on a graphene/SiC(0001) substrate have
a maximum Tc close to that of bulk FeSe [11] reveals that the choice of STO
as a substrate is crucial and that the relevant enhancement mechanism(s)
must involve both the FeSe layer and the substrate. Another evidence that
corroborates this observation is the fact that simply thinning a FeSe sample
down to a 1 UC layer does not give rise to a high Tc . As a result, enhanced
electron-phonon coupling within the FeSe layer is not likely to be a crucial
mechanism to account for the rise of Tc .
In addition to the measurements of the band gap for the FeSe/STO interface, the ARPES results reported in [5] reveal the existence of replica bands
approximately 100 meV away from the main electron-like and hole-like bands
(Figure 3). Given that pure STO has a very flat optical phonon band centred
around 100 meV [12], and since these oxygen vibrational modes are widely
separated from other phonon modes, the occurrence of these replica bands
may be explained in terms of the coupling between the FeSe 3d electrons and
the optical oxygen phonon branch in the STO substrate.
Another important feature of the replica bands is how closely these follow
the dispersion of the original bands. The resolution of replicas of an entire
band with such a clear dispersion is a consequence of both the substantial electron-phonon coupling and, most crucially, the forward-focused (i.e.
low-momentum transfer) nature of this coupling. Indeed, if electron-phonon
coupling occurred over a wide range of momentum transfer, the dispersion
of the replica bands would be blurred out due to the superposition of several
replica bands with relative displacements in momentum space.

Figure 3: Band structure of FeSe on STO substrate: (Left) about point (Right)
about M point. Replica bands are identified by the letter of the respective original
band followed by a prime. Replica bands are well resolved, which implies lowmomentum transfer. The separation from original bands is of approximately 100
meV. [Picture Credits: J. J. Lee et al. [5]]

Two conceptually dierent approaches to this problem in the bulk limit


(i.e. the full two-dimensional layer of FeSe rather than nanograins) were
proposed by Johnston et al. [13] and Gorkov [14]. A description of the two
models and the main dierences between them is given below.
Noting that forward scattering plays a decisive role in the mechanism of
enhancement of superconductivity in FeSe/STO, Johnston et al. solved the
momentum-dependent Eliashberg equations for a matrix element containing
an exponential decay in momentum exp( q/q0 ), with q0 = C/a, where a
is the lattice constant and C = O(0.1) = constant. In addition to this
numerical calculation, an analytical expression for the critical temperature
was found in the perfect forward-scattering limit (where the cut-o is set by
a Dirac Delta-function (q)); importantly, the critical temperature was found
to be linearly proportional to the dimensionless coupling constant , which
contrasts with the usual BCS expression Tc exp( 1/ ).
Gorkov, on the other hand, emphasised the role played by charge transfer in addition to the electron-phonon coupling mechanism. In fact, charge
7

transfer produces two distinct eects:


Only a fraction of the transferred electrons appear at the chemical
potential and hence go into the conduction band of FeSe, thereby contributing towards the superconducting current.
Below a threshold charge density, the transferred electrons are localised
and give rise to an additional polarisation at the interface, which changes
the dielectric properties of the surface.
Gorkov therefore regarded the dielectric constant of the surface as one
of the parameters of his model, in addition to the coupling constant and
the Debye energy of the STO phonon modes.
The Eliashberg equations assume the validity of Migdals Theorem, which
states that vertex corrections may be neglected if the phonon energy scale
is much smaller than the electron energy scale. Gorkov claimed that this
condition is violated for ~!D 100 meV and F = 60 meV. Hence, Gorkov
considered the anti-adiabatic limit ~!D >> F instead, treating the ratio
F /~!D as the small parameter in the perturbative expansion.
In reality, though, the parameter in which the perturbative expansion is
performed in conventional Eliashberg Theory is ~!D /F . Therefore, in the
weak-coupling regime the usual Eliashberg equations are still valid. Johnston
et al. obtained numerical results consistent with the experimentally observed
band structure of FeSe on STO for 2 [0.15, 0.2], for which range of values
~!D /F < 1, as required for the perturbative expansion to hold.
Gorkov also highlighted the existence of additional pairing channels beyond electron-phonon coupling, e.g. due to antiferromagnetic fluctuations.
This would give rise to a larger coupling constant, which justifies the option
for the anti-adiabatic expansion described above.

The Project

3.1
3.1.1

The BCS Model


Superconductivity in Reduced Dimensions

When the dimensions are sufficiently reduced, quantum size eects (QSEs)
lead to changes in the superconducting properties. The most significant
one is that the material ceases to be superconducting when the mean energy
level spacing becomes comparable to the superconducting gap [15]. However,
before this limit is attained, there are two main changes produced by QSEs
on the superconducting gap:
Global fluctuations of the gap due to the quantisation of the energy
levels: varying the area of the grain leads to a change in the spacing
between the energy levels, which in turn changes the number of levels
within the interaction range.
Local fluctuations in the gap due to the non-trivial matrix elements.
The Bardeen-Cooper-Schrieer (BCS) theory [16] was originally applied
in the bulk. Given the translational invariance, the one-particle eigenstates
could be chosen to be plane waves, a set of orthogonal states. Since the
interaction potential is taken to be constant for a range 2 [ ~!D , ~!D ],
where is the chemical potential and is the electron energy, the squared
matrix element simply becomes a constant times a Dirac-delta function that
imposes conservation of momentum, which is trivially satisfied because of
the usual choice of trial wavefunction in the BCS theory - a coherent state of
Cooper pairs consisting of two electrons of opposite spin and momentum. For
orthonormal states, the non-zero overlap integrals are equal to 1 (neglecting
the factors of 2). However, in reduced dimensions, the overlap integrals
may take dierent non-zero values depending on the quantum numbers of
the interacting particles. For the case of rectangular nanograins, the singleparticle wavefunctions are:
r)
n,m (~

n
m
2
= p sin
x sin
y
a
b
A

(1)

where A = ab is the area of the nanograin and (n, m) are quantum numbers.
The overlap integrals are defined as:
Z
2
In,n0 A
r) n2 0 (~r)d~r
(2)
n (~
where n are the eigenstates of the free-particle Hamiltonian. This takes the
following form for rectangles:

1
1
In,n = 1 + n,n0 1 + m,m0
(3)
2
2
where is the Kronecker delta and n (n, m), n (n0 , m0 ). Hence, when
n = n or m = m, the overlap integral becomes 3/2 rather than 1. When n
= n and m = m, I = 2.
Hence, in reduced dimensions the BCS Hamiltonian takes the form:
X
X
H=
n cn cn
In,n0 cn" cn# cn0 # cn0 "
(4)
n,n0

where cn is the annihilation operator for an electron of spin in state n,


is the BCS coupling constant and n are the eigenvalues of the free-particle
Hamiltonian, which are measured with respect to the chemical potential .
In order to make dimensionless, there is an extra factor , which is the
mean level spacing, with = 1/T F (0), where T F (0) is the spectral density
at the Fermi energy within the Thomas-Fermi approximation.
For a rectangular grain of sides a and b, the eigenvalues are:
~2 2
[(n/a)2 + (m/b)2 ]
2m
where m is the eective mass of the electrons.
n =

3.1.2

(5)

The Gap Equation

The self-consistency equation of BCS theory gives the following expression


for the gap energy:
n

|n0 |<~!D

0I
0
1
p n n,n
2
2
n0 + n0 T F (0)

(6)

where ~!D is the relevant Debye energy. The superconducting gap was
found by solving this equation numerically using the MATLAB R built-in
10

function fzero, which finds the roots of a non-linear single-variable equation. The initial point to search for the root was set to be the bulk gap
0 = ~!D / sinh(1/ ).
Although the chemical potential is expected to be close to the Fermi
energy EF , the corrections induced by quantum size eects to the spectral
density lead to a change in that is not negligible for the size of the grains
considered
R in this project. was computed exactly by inverting the relation
N/2 =
()d, where () is the spectral density.
Correction to Chemical Potential due to Quantum Size Effects
1.35
Exact Numerical
Analytical Estimate

1.3

/ EF

1.25

1.2

1.15

1.1

1.05
50

100

150

200

250

300

350

400

450

500

550

600

Area of Nanograin / nm2

Figure 4: Ratio of chemical potential to Fermi energy EF = 60 meV against


area of nanograin: (Blue) Exact numerical solution (Orange) Analytical estimate
to leading-order. Because of the correction to the spectral density due to the
quantum size eects, is not exactly equal to EF even at T = 0.

As a sanity check, the numerical solutions of this equation were compared


to an analytical estimate of the correction to using the leading-order term
of the semiclassical expansion of the spectral density: the Weyl correction
11

[17], which for a p


rectangle takes the form g() = L/2kF A, where L is the
perimeter, kF = 2m EF /~ is the Fermi wavevector and A is the area. This
correction is due to the boundary conditions; the minus sign arises from
assuming Dirichlet boundary conditions. The analytical estimate and the
exact numerical calculation are plotted in figure 4.
Figure 5 shows the correction due to the overlap integrals In,n0 (blue) and
the Weyl correction (black). It is therefore clear that, apart from making a
small contribution to the global fluctuations of the gap, the main eect of the
overlap integrals is to cancel the eect caused by the change in the spectral
density due to the boundary conditions, which is in agreement with [18].
Comparison of Correction due to Overlap Integrals and Weyl Correction to Spectral Density
1.4

Correction due to Overlap Integrals


1 / Weyl Correction (Dirichlet BCs)
Weyl Correction (Dirichlet BCs)

1.3

Size of Correction

1.2

1.1

0.9

0.8

0.7
50

100

150

200

250

Area of Nanograin / nm2

Figure 5: Comparison of correction due to overlap integrals (blue) and Weyl correction to spectral density (black). The reciprocal of the Weyl correction is also
plotted (orange) to facilitate the comparison. The main eect of the overlap integrals is to cancel the Weyl correction.

Before considering the unusual case of the FeSe/STO interface, where the
12

Debye energy ~!D is of the same order of magnitude as the Fermi energy EF ,
in order to confirm that the program was working correctly, the gap equation
was solved for a range of dierent areas of nanograins in the conventional
BCS limit, setting EF = 5 eV, ~!D = 50 meV and = 0.5. The numerical
results were compared to the analytical results obtained by periodic orbit
theory [17]. Since this method was also used in the following part of the
project, a brief outline of this technique is given below.

Figure 6: Gap equation solutions in the conventional BCS limit, with EF = 5


eV, ~!D = 50 meV and = 0.5. (Blue dots) Numerical results (Red crosses)
Analytical results from periodic orbit theory. Although there is a good agreement
between the numerical and analytical results, the former appear to have slightly
greater fluctuations, which is explained by the fact that the contribution from the
overlap integrals was not accounted for in the analytical calculation.

Periodic orbit theory establishes a link between the spectral density of


13

a degenerate system and the periodic orbits of the corresponding classical


system. Within this semiclassical theory, the spectral density may be expanded as () = T F (0)(1 + g(0) + g()), where the monotonous correction
g(0) corresponds to the Weyl correction above mentioned and the oscillatory
correction, g() arises from the Gutzwiller trace formula [19].
In two dimensions, these two corrections are given by:
g(0) =
(2)

g() = g1,2 ()
=

1
X

L
2kF A

(7)

1 X (1)
g () =
2 i i

J0 (k()L1,2
n )

Ln 6=0

2
1
1 X 4Li X
cos(k()Lin )
2 i=1 kF A L 6=0

(8)

where the terms in the Weyl correction have the meaning mentioned above,
the plus and minus signs corresponding to Neumann and Dirichlet boundary
conditions, respectively. J0 is the zeroth-order Bessel function
p of first kind.
1,2
L1 = a and L2 = b are the sides of the rectangle, Ln = 2 a2 n2 + b2 m2 is
the length of the periodic orbit (n,m) and L1n = 2na and L2n = 2mb are the
p
(2)
lengths of the single-integer periodic orbits. g1,2 () is of O(1/ kF L), whereas
p
(1)
g(0) and gi are both of O(1/kF L), where L A is the linear size of the
system)
By converting the sum over energy levels into an integral over energy in
the gap equation, replacing the spectral density by the expression above and
expanding the gap as = 0 (1 + f (1/2) + f (1) ), where f (n) O(1/(kF L)n ),
the gap equation can be solved order by order in the small parameter 1/kF L.
The zeroth-order equality gives the bulk gap; the remaining equalities give
the leading-order corrections:
f

(1/2)

1
X

1,2
J0 (kF L1,2
n )K0 (Ln /)

Ln 6=0

14

(9)

(1)

L
2kF A

2
1
X
2Li X
cos(kF Lin )K0 (Lin /) +
k
A
F
i=1
L 6=0
n

(10)
i
1,2
L
n
+ f (1/2)
f (1/2)
J0 (kF L1,2
K1 (L1,2
n )
n /)
6

Ln 6=0
p
R +1
where K0 (Ln /) 1/2
dx
cos(L
x/)/
1 + x2 and Ln / K1 (Ln /)
n
1
R +1
1/2
dx cos(Ln x/)/(1 + x2 )3/2 are cut-o functions that suppress the
1
contribution of periodic orbits Ln longer than the coherence length
~2 kF /m 0 . These cut-o functions were obtained by expanding the wavevector of the electron about the Fermi wavevector, k() = kF (1+/(2EF )), since
the relevant interactions take place at electron energies close to EF . (9) and
(10) are consistent with the expressions published in [18] apart from a numerical factor in one of the terms in f (1) .
Given that for this calculation EF = p
5 eV, kF 20 nm 1 . For the smallest
grain considered, A = 50 nm2 and L A 7 nm, so 1/kF L . 1/140 1.
The semiclassical expansion is therefore valid.
h1

3.1.3

1
X

FeSe Nanograins on STO Substrate

For the case of FeSe nanograins on top of a STO substrate, EF = 60 meV


and ~!D O(100 meV). Figure 7 shows the numerical results for ~!D = 100
meV and = 0.27 (blue dots) and the experimental results obtained by Zhi
Li et al. [7] (red crosses).
It should be noted that Zhi Li et al. only measured the area of the
nanograins and not both sides of the rectangles.
p However, Zhi Li informed
me that the ratio of the two sides is less than 2 and
p that the grains
p were
definitely rectangles and
p not squares, so I set a = 1.1 A and b = A / 1.1.
If I had set a = b = A, there would have been an additional symmetry
that would have given rise to a further enhancement of the fluctuations due
to the overlap integrals.
In addition, Zhi Li et al. made more than one measurement of the gap in
each grain. Such measurements revealed the already mentioned local fluctuations of the superconducting gap due to the overlap integrals. However, in
this part of the project, only the global fluctuations of the gap were considered, so each of the experimental data plotted in figure 7 corresponds to the
average of the measurements of the gap for each grain.
15

Normalized Fluctuations in FeSe Nanograins on STO Substrate

1.5

Numerical Results
Experimental Data

1.4

1.3

/ 0

1.2

1.1

0.9

0.8

0.7

0.6
60

80

100

120

140

160

Area of Nanograin / nm2

Figure 7: Normalized gap fluctuations (gap / bulk gap) versus area of nanograins
for FeSe grains on STO susbtrate. (Blue dots) Numerical solutions of BCS gap
equation for EF = 60 meV, ~!D = 100 meV and = 0.27 (Red crosses) Experimental results obtained by Zhi Li et al. [7].

Although figure 7 suggests a good agreement between the numerical and


experimental results, in truth the choice of parameters = 0.27 and ~!D =
100 meV only allows to fit the size of the fluctuations.
p Replacing these values
in the BCS expression for the bulk gap, 0 = 2 ~!D exp( 1/ )1 with
EF = 60 meV, gives only 3.8 meV, which is far from the experimentally
measured 16.5 meV.
Originally, I attempted to fit both the fluctuation size and the bulk gap
by varying the two parameters of the model, and ~!D . However, for a
coupling constant of (0.25, 0.30) required to get fluctuations of the right
1
=
0
p For the case ~!D > , the correct BCS bulk gap expression is
2 ~!D exp( 1/ ) instead of the usual 0 = 2~!D exp( 1/ ), which applies for ~!D

16

size, the Debye energy would be several orders of magnitude greater than
the expected 100 meV, which would be an unphysical value. Hence, the BCS
model is not sufficient to account for the experimental observations made by
Zhi Li et al. There are two possible justifications:
There might be additional pairing channels, for example, due to spin
fluctuations, as suggested by Gorkov [14]. This could give rise to an
additional parameter: the coupling constant for the new channel. Provided that the eective coupling constant that accounts for all pairing
channels was 0.5, the bulk gap would be consistent with the experimental measurement.
The BCS theory puts no constraints on the momentum transferred
in the phonon-mediated electron-electron pairing. However, as pointed
out in section 2, low-momentum transfer appears to be necessary to explain the band structure of FeSe on STO. In fact, solving the Eliashberg
momentum-dependent equations for perfect forward scattering gives a
gap linear in both the Debye energy and the coupling constant, contrary to the usual BCS expression 0 exp( 1/ ). Such a linear
relation would allow the bulk gap to be obtained for a Debye energy of
the expected order of magnitude.
Assuming the phonon-mediated pairing channel is the dominant one, the
next step to understand the experimental observations in FeSe nanograins
on STO involves considering the eects of forward scattering, which requires
going beyond the BCS model.

3.2

The Eliashberg Model

The second part of the project corresponds to adapting the Eliashberg gap
equation in the weak-coupling limit to reduced dimensions. Johnston et al.
included an outline of the Eliashberg theory in [13], which is summarised
below.
The Eliashberg theory is a diagrammatic perturbation theory that extends the BCS theory by taking into account the intrinsic time scale of the
superconducting interactions (i.e. the the finite temporal existence of the
phonon). In BCS theory, on the other hand, the paring interactions are
assumed to be instantaneous.
17

Within the Eliashberg theory the electron self-energy can be written in


the Nambu notation as:
(k, i!n ) = i!n [1

Z(k, i!n )]0 + (k, i!n )3 + (k, i!n )1

(11)

where !n = (2n+1)/ is a Matsubara frequency, i (i = 1,2,3) represent the


Pauli matrices and 0 = 1. Z(k, i!n ) and (k, i!n ) renormalise the singleparticle mass and band dispersion, respectively. (k, i!n ) is the so-called
anomalous self-energy, which is zero in the normal state.
The self-energy is found by self-consistently evaluating the loop diagram
shown in figure 8 below:

Figure 8: Loop Feynman diagram to compute the electron self-energy within


Migdal-Eliashberg Theory. G(k, i!n ) is the dressed Greens function of the electron. D(0) (q, i!m ) is the bare phonon propagator. [Picture Credits: Johnston et
al. [13]]

(k, i!n ) =

1X
q,m

|g(k, q)|2 D(0) (q, i!n

i!m )3 G(k + q, i!m )3

(12)

where G 1 (k, i!n ) = i!n 0 k 3 (k, i!n ) is the dressed Greens function
2
of the electron and D(0) (q, i!m ) = 2~!D /((~!D )2 + !m
) is the bare phonon
propagator (assuming a flat phonon mode of Debye energy ~!D ). k is the
dispersion of the electron (relative to ), N is the number of momentum grid
points and
= 1/kB T is the inverse temperature. In the weak-coupling
limit, Z(k, i!n ) = 1, (k, i!n ) = 0 and (k, i!n ) can be identified with the

18

superconducting gap (i!n ). Comparing equations (11) and (12) within this
limit gives the Eliashberg gap equation:

(k, i!n ) =

3.2.1

1X
q,m

|g(k, q)|2 D(0) (q, i!n

i!m )

2
!m

(k, i!m )
+ 2 (k, i!m )

2k+q

(13)

The BCS Limit of the Eliashberg Gap Equation

For an isotropic s-wave gap, the gap is momentum-independent, therefore


(k, i!n ) = (i!n ), In addition, there is no constraint on momentum transfer in the BCS theory, so |g(k, q)|2 = g02 = constant. Setting (i!n )
2g02 D(0) (i!n )(0), which is dimensionless, and assuming that (i!n i!m )
and (i!m ) are constants and 0 for a range of electron energies 2
[ ~!D , ~!D ], the Matsubara frequency summation can be calculated by contour integration. After replacing the sum over momentum by an integral
over energy, we get:
p 2 2
+ 0
Z ~!D
2
2
() tanh
p
=
d
(14)
(0)
2 + 20
~!D

which is the BCS gap equation at a given temperature kB T = 1/ .


Although this may seem to be a trivial derivation, in the literature (e.g.
[13], [20]) an incorrect connection between the sum over Matsubara frequencies in the Eliashberg equation and the integral over electron energy in the
BCS equation is frequently found. This is clearly not correct, since the spacing between the Matsubara frequencies is temperature dependent, which is
certainly not the case for the electron energy levels. Furthermore, the Matsubara frequencies are linear in the quantum number n, whereas electron
energies are parabolic.
The understanding of this transition from Eliashberg to BCS theory is
required to be able to solve the Eliashberg gap equation analytically using
periodic orbit theory. It is the density of states that arises from the sum over
momentum and not from the sum over Matsubara frequencies that is to be
expanded to account for the quantum size eects.
In attempting to solve the Eliashberg gap equation analytically, the only
correction due to reduced dimensions that will be introduced will be the
19

quantisation of the energy levels through the correction to the spectral density. The overlap integrals will be ignored, since, as demonstrated in the first
part of the project, their contribution to the fluctuations is very small.
3.2.2

Low-momentum Transfer in the Eliashberg Model

The extreme case of low-momentum transfer corresponds to perfect forward


scattering, for which no momentum transfer is allowed and hence the matrix
element can be written as |g(q)|2 = g02 N q , where q is the Kronecker Delta
function. The sum over momentum is trivial and the sum over Matsubara
frequencies can be converted into an integral at T = 0 since the spacing between the Matsubara frequencies becomes infinitesimal. This integral can be
calculated assuming 0 ~!D and using the ansatz proposed by Johnston
et al. (i!n ) = 0 /(1 + (!n /!D )2 ), in which case the bulk gap is found to
be2 :
2
~!D
(15)
2+3
However, in this limit, no corrections due to quantum size eects can be
expected. As discussed in section 3.1.1, the fluctuations due to the quantisation of the energy levels are due to the variation in the number of states
that contribute to the interaction as the area of the grain is changed; such
change cannot be observed in this case because the Kronecker delta picks a
single momentum state for the interaction.
As a result, we must consider a finite cut-o in order to observe fluctuations. For simplicity, to attempt an analytical solution of the gap equation in reduced dimensions, I considered
p the nearly-perfect forward scattering
limit, with a sharp cut-o q0 = C/ A, where C = O(0.1) = constant.
Since the smallest area considered is of 50 nm2 , ~2 q02 /2m . 1 meV, so
~2 q02 /2m 0 ~!D
0

The expression for the bulk gap 0 can be written in terms of the dimensionless
coupling constant instead of the matrix element constant g0 by making use of the relation
= g02 /(~!D )2 derived by Johnston et al. in the weak-coupling limit of the Eliashberg
theory.

20

Equation (13) may be rewritten as:


2 (~!D )3 X

~2 q02 /2m

()
2 + 2 +
2 (i! )
!m
m
~2 q02 /2m
m
(16)
By replacing the spectral density () by the semiclassical expansion that
includes the finite-size eects, expanding the gap to leading-order (i!m ) =
(1/2)
) and comparing terms of equal order in 1/kF L, we obtain:
0 (i!m )(1 + f
(i!n ) =

0 (i!n )

(i!m )
(~!D )2 + (!m !n )2

2 (~!D )3 X
m

0 (i!m )

(~!D

)2

f (1/2) = P

+ (!m

0 (i!m )
2
m (~!D )2 +!m
2 30 (i!m )
2
m (~!D )2 +!m

!n

)2

R ~2 q02 /2m

~2 q02 /2m

d
~2 q02 /2m

(2)

d !2 +21,2
+
~2 q 2 /2m
0

R ~2 q02 /2m

2
!m

(0)
+ + 20 (i!m )
(17)

()
2 (i! )
m

d [!2 +2 +1 2 (i!m )]2


~2 q 2 /2m
0

(18)

P
(2)
1,2
where g1,2 () = 1
Ln 6=0 J0 (k()Ln ), as before. After solving the integrals over
energy assuming ~2 q02 /2m 20 and the sum over Matsubara frequencies at
absolute zero temperature, 0 and f (1/2) can be expressed as:
0

(1/2)

=W

(Aq02 )
~!D

1
X

J0 (kF L1,2
n )

(19)
(20)

Ln 6=0

where W =

(~!D )2 (~2 q02 /2m)


3
0

. As expected, the expression for the bulk gap is


p
linear in and ~!D and for a choice of parameters q0 = 0.4/ A, = 0.3
and ~!D = 100 meV we get a bulk gap of 15 meV. However, the leadingorder correction does not include a decaying cut-o that suppresses terms
involving periodic orbits greater than some coherence length. It is therefore
not clear if this term converges, but, assuming it does, its value cannot be
calculated as the convergence is too slow. Hence, this expression is of no use
as far as comparison to the experimental results is concerned.
21

The reason why I present this expression is to show that the difficulty in
obtaining the analytical expression for the gap using periodic orbit theory
in this low-momentum transfer limit is to get a decaying cut-o. Indeed, in
the BCS case, the cut-o
p was found by using the asymptotic limit of the
Bessel function J0 (x) = 2/x cos(x /4) and expanding the wavevector
k() = kF (1 + /(2EF )) to get:
1
2

~!D

P1

Ln 6=0

J0 (k()L1,2
n )

2 +
Z
1
X
1,2 1
=
J0 (kF Ln )
2
L 6=0
~!D

2
0

kF 1,2
cos( 2E
Ln )
d p F 2 =
2 + 0
~!D

~!D

h1 Z

1,2

cos( Ln x ) i
1,2
=
J0 (kF Ln )
dx p

2
2
1
+
x
1
Ln 6=0
1
L1,2
X
n

J0 (kF L1,2
)K
0
n

L 6=0
1
X

(21)

The decaying cut-o arises from taking the limits of integration of the
cut-o integral to 1, which is possible since ~!D
0 . In the Eliashberg
case, however, the energy integral for the leading-order correction is of the
form:
Z

~2 q02 /2m

d
~2 q02 /2m

1
X
J0 (k()L1,2
n )
2
! 2 + 0 + 2
L 6=0 m

(22)

so in this case after a change of variables the limits of integration become


~2 q 2 /2m
p 20 2 , which cannot be taken to 1 for any finite value of q0 because
!m +

the Matsubara frequencies extend from 1 to +1.


To overcome this problem, assuming ~!D
~2 q02 /2m,
0 and ~!D
before solving the sum over momentum in (13) I solved the Matsubara fre-

22

quency summation at T = 0 K:
1X
m

1
=
2

2
(!m

+1
1

(i!m )
+ 2 +

2
(~!D )2 )(!m

2 (i! ))
m

1
d!m 2
2
2
2 )2 + (~! )4
(!m + )((~!D ) + !m
D

(23)
2
0

in which case (13) becomes:

~!D
1=
N

|q|<q0

= ~!D

q
2q +

2
0

q
2
3 2q +

~!D

~2 q02 /2m

1
d () p
2 +
~2 q02 /2m

2
0

2
0

p
3 22 +
2
~!D

(24)
2
0

As a sanity check, assuming ~!D


~2 q02 /2m as before, we ob0
tain (19) for the bulk gap. As expected, the result is the same regardless of
the order by which the sum over Matsubara frequencies and the sum over
momentum are performed. However, as discussed earlier, this approximation does not give rise to the decaying cut-o in the finite-size corrections.
However, the limit ~!D
~2 q02 /2m
0 , which in principle would solve
this problem, gives rise to a divergence in the bulk gap equation. Hence, the
analytical calculation of the superconducting gap and the respective leadingorder finite size corrections using periodic orbit theory is not possible within
this model.
Alternatively, since the Matsubara frequency summation has been solved,
the sum over momentum in (24) can now be solved numerically.

Further Work

Since the analytical approach chosen to include quantum size eects in the
Eliashberg model was not successful, the next step would correspond to solving equation (24) numerically in a similar way to the BCS gap equation in
the first part of the project. Given the limited time available to complete
the project, this could not be finished.
23

The full understanding of the mechanism of enhancement of superconductivity in FeSe/STO will require further experiments. For example, Gorkov
[14] related the momentum cut-o q0 introduced by Johnston et al. [13] to
the dielectric properties of the FeSe/STO interface. Therefore, varying the
dielectric constant of this interface could provide more information about the
forward-scattering mechanism.
In addition, since electron-phonon coupling appears to be a crucial mechanism of superconductivity in FeSe/STO, and given that the phonon mode
responsible for this pairing channel is likely to be associated with oxygen oscillations transverse to the interface, experimental tests of the isotope eect
could also be valuable.
Dung-Hai Lee [8] also proposed an interesting structure: a FeSe monolayer between two STO substrates. This structure could reveal a further
enhancement of the superconducting properties since the phonon-mediated
pairing should be twice as strong.

Conclusion

A BCS model was applied to FeSe nanograins on a STO substrate and could
account for the normalized fluctuations of the gap measured by Zhi Li et
al. [7] for a choice of = 0.27. However, a Debye energy of 100 meV leads
to a very low estimate of the bulk gap ( 4 meV instead of the experimentally
measured 16.5 meV).
This inconsistency between the BCS model and the experimental data
leads to the conclusion that the phonon-mediated pairing mechanism cannot explain by itself the enhancement of the superconducting properties of
FeSe/STO. Additional pairing channels, such as spin fluctuations and lowmomentum transfer, are also required to understand this mechanism.
In order to account for forward scattering, an Eliashberg model was
adapted to reduced dimensions. Although the attempt to derive an analytical expression for the bulk gap and the leading-order corrections due to the
quantum size eects using periodic orbit theory was not successful, a simple
non-linear gap equation was derived, which could be solved numerically.

24

References
[1] J. Bednorz and K. Mller, Possible high Tc superconductivity in the
BaLaCuO system, Zeitschrift fr Physik B Condensed Matter, vol. 64,
no. 2, pp. 189193, 1986.
[2] Y. Kamihara, H. Hiramatsu, M. Hirano, R. Kawamura, H. Yanagi,
T. Kamiya, and H. Hosono, Iron-based layered superconductor:
LaOFeP, Journal of the American Chemical Society, vol. 128, no. 31,
pp. 1001210013, 2006. PMID: 16881620.
[3] A. Schilling, M. Cantoni, J. D. Guo, and H. R. Ott, Superconductivity
above 130 K in the Hg-Ba-Ca-Cu-O system, Nature, vol. 363, pp. 56
58, May 1993.
[4] W. Qing-Yan, L. Zhi, Z. Wen-Hao, Z. Zuo-Cheng, Z. Jin-Song, L. Wei,
D. Hao, O. Yun-Bo, D. Peng, C. Kai, W. Jing, S. Can-Li, H. Ke, J. JinFeng, J. Shuai-Hua, W. Ya-Yu, W. Li-Li, C. Xi, M. Xu-Cun, and X. QiKun, Interface-Induced high-temperature superconductivity in single
unit-cell FeSe films on SrTiO3 , Chinese Physics Letters, vol. 29, no. 3,
p. 037402, 2012.
[5] J. J. Lee, F. T. Schmitt, R. G. Moore, S. Johnston, Y.-T. Cui, W. Li,
M. Yi, Z. K. Liu, M. Hashimoto, Y. Zhang, D. H. Lu, T. P. Devereaux,
D.-H. Lee, and Z.-X. Shen, Interfacial mode coupling as the origin of
the enhancement of Tc in FeSe films on SrTiO3 , Nature.
[6] F. Li, H. Ding, C. Tang, J. Peng, Q. Zhang, W. Zhang, G. Zhou,
D. Zhang, C.-L. Song, K. He, S. Ji, X. Chen, L. Gu, L. Wang, X.-C.
Ma, and Q.-K. Xue, Interface-enhanced high-temperature superconductivity in single-unit-cell FeTe1 x Sex films on SrTiO3 , Phys. Rev. B,
vol. 91, p. 220503, Jun 2015.
[7] Z. Li, J.-P. Peng, H.-M. Zhang, C.-L. Song, S.-H. Ji, L. Wang, K. He,
X. Chen, Q.-K. Xue, and X.-C. Ma, Visualizing superconductivity in
FeSe nanoflakes on SrTiO3 by scanning tunneling microscopy, Phys.
Rev. B, vol. 91, p. 060509, Feb 2015.
[8] D.-H. Lee, What Makes the Tc of FeSe/SrTiO3 so High?, ArXiv eprints, Aug. 2015.
25

[9] J. Bang, Z. Li, Y. Y. Sun, A. Samanta, Y. Y. Zhang, W. Zhang, L. Wang,


X. Chen, X. Ma, Q.-K. Xue, and S. B. Zhang, Atomic and electronic
structures of single-layer FeSe on SrTiO3 (001): The role of oxygen deficiency, Phys. Rev. B, vol. 87, p. 220503, Jun 2013.
[10] R. Peng, H. C. Xu, S. Y. Tan, H. Y. Cao, M. Xia, X. P. Shen, Z. C.
Huang, C. Wen, Q. Song, T. Zhang, B. P. Xie, X. G. Gong, and D. L.
Feng, Tuning the band structure and superconductivity in single-layer
FeSe by interface engineering, Nature Communications, vol. 5, sep
2014.
[11] C.-L. Song, Y.-L. Wang, Y.-P. Jiang, Z. Li, L. Wang, K. He, X. Chen, X.C. Ma, and Q.-K. Xue, Molecular-beam epitaxy and robust superconductivity of stoichiometric FeSe crystalline films on bilayer graphene,
Phys. Rev. B, vol. 84, p. 020503, Jul 2011.
[12] N. Choudhury, E. J. Walter, A. I. Kolesnikov, and C.-K. Loong, Large
phonon band gap in SrTiO3 and the vibrational signatures of ferroelectricity in ATiO3 perovskites: First-principles lattice dynamics and
inelastic neutron scattering, Phys. Rev. B, vol. 77, p. 134111, Apr 2008.
[13] L. Rademaker, Y. Wang, T. Berlijn, and S. Johnston, Enhanced superconductivity due to forward scattering in FeSe thin films on SrTiO3
substrates, ArXiv e-prints, July 2015.
[14] L. P. Gorkov, On mechanisms of superconductivity at the single-layer
FeSe/SrTiO3 interface, ArXiv e-prints, Oct. 2015.
[15] P. Anderson J. Phys. Chem. Solids, vol. 11, 1959.
[16] J. Bardeen, L. N. Cooper, and J. R. Schrieer, Theory of superconductivity, Phys. Rev., vol. 108, pp. 11751204, Dec 1957.
[17] M. Brack and R. Bhaduri, Semiclassical Physics. Frontiers in physics,
Addison-Wesley, Advanced Book Program, 1997.
[18] J. Mayoh and A. M. Garca-Garca, Number theory, periodic orbits,
and superconductivity in nanocubes, Phys. Rev. B, vol. 90, p. 014509,
Jul 2014.

26

[19] M. Gutzwiller, Chaos in Classical and Quantum Mechanics. Interdisciplinary Applied Mathematics, Springer New York, 1991.
[20] F. Marsiglio and J. P. Carbotte, Superconductivity: Conventional and
Unconventional Superconductors, ch. Electron-Phonon Superconductivity, pp. 73162. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008.

27

You might also like