UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
Density fluctuations in the 1D Bose gas
Panfil, M.K.
Publication date
2013
Document Version
Final published version
Link to publication
Citation for published version (APA):
Panfil, M. K. (2013). Density fluctuations in the 1D Bose gas.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
Miƚosz Panfil
Density Fluctuations in
the 1D Bose Gas
e
ns
ity
Flu
ctu
ati
ons i
n t
h
e
1
D
Bos
e
Gas
M.
P
anf
il
in the 1D Bose Gas
Academisch Proefschrift
ter verkrijging van de graad van doctor aan de Universiteit van Amsterdam
op gezag van de Rector Magnificus prof. dr. D.C. van den Boom
ten overstaan van een door het college voor promoties ingestelde commissie, in het openbaar te verdedigen in de Agnietenkapel
op woensdag 9 oktober 2013, te 12:00 uur door
Miłosz Krzysztof Panfil
Promotiecommissie
Promotor: Prof. dr. J.-S. Caux
Overige Leden: Prof. dr. G. Mussardo Prof. dr. B. Nienhuis Prof. dr. G. Shlyapnikov Prof. dr. ir. H. T. C. Stoof Prof. dr. E. Verlinde
Faculteit der Natuurwetenschappen, Wiskunde en Informatica
This work is part of the research programme of the Foundation for Fundamental Research on Matter (FOM), which is part of the Netherlands Organisation for Scientific Research (NWO).
Typeset by LATEX.
Plots were generated using the matplotlib library in the Python programming language. Printed and bound by SPS Print.
1 Introduction 1
1.1 The 3D Bose gas and the Bose-Einstein condensation . . . 3
1.2 Long-range Order and the MWH Theorem . . . 7
1.3 A Glimpse of the 1D World . . . 10
1.4 Structure of the Thesis . . . 10
2 Many-body Physics in 1D 13 2.1 1D Bose Gas . . . 13
2.1.1 Coordinate Bethe Ansatz . . . 14
2.1.2 Physical Picture . . . 19
2.2 The Density-density Correlation Function . . . 21
2.2.1 Real Space Correlation Function . . . 21
2.2.2 Momentum Space. . . 25
2.3 How to Compute Correlation Functions . . . 29
2.3.1 The ABACUS. . . 31
2.4 How to Measure Correlation Functions . . . 31
3 The Algebraic and the Thermodynamic Bethe Ansatz 35 3.1 The Algebraic Bethe Ansatz . . . 36
3.1.1 Construction of Eigenstates . . . 40
3.1.2 Repulsive Interactions . . . 42
3.1.3 Attractive Interactions . . . 45
3.1.4 Scalar Products and Form Factors . . . 47
3.1.5 The Density Operator . . . 49
3.2 Thermodynamic Bethe Ansatz . . . 50
3.2.1 Density of Quantum Numbers. . . 51
3.2.2 Density of Rapidities . . . 52
3.2.3 Excitations and the Back-flow Function . . . 56
3.2.4 Thermal Equilibrium . . . 59
4 Exact prefactors of the density-density correlation function in the (non-) linear Luttinger liquid model 65 4.1 Luttinger Liquid Theory . . . 66
4.2 Nonlinear Luttinger Liquid . . . 71
4.3 Thermodynamic Limit of the Density Operator . . . 77
4.3.1 Evaluation ofM1 . . . 86
4.3.2 Obtaining term M2 . . . 87
4.3.3 Fredholm Determinants . . . 101
4.4 Results. . . 104
4.4.1 Thermodynamic Limit of Form Factors. . . 104
5 Finite temperature density-density correlations 113 5.1 ABACUS and the Saddle-point Method . . . 114
5.2 Momentum Space . . . 116
5.3 Real Space . . . 121
5.4 Conclusions . . . 122
6 Correlations in the super Tonks-Girardeau Gas 123 6.1 Interaction Quench and the sTG gas . . . 124
6.2 Excitations in the sTG gas. . . 126
6.3 Momentum Space . . . 127
6.4 Real Space . . . 130
6.5 Conclusions . . . 131
7 Conclusions and Future Ideas 133 A Notation and Convention 135 A.1 Notation . . . 135
A.2 Fourier Transform . . . 135
A.3 Principal Value Integrals . . . 137
B Special Functions and their Properties 139 C Correlations in the weakly and strongly interacting regimes 143 C.1 Bogolyubov approximation . . . 143
C.2 Free Fermionic Gas . . . 147
Bibliography 150
Acknowledgments 159
Summary 161
Introduction
The purpose of physics is to describe natural phenomena in a mathematical language. Such a description should have two main features: an internal consistency and a predic-tive power. The requirement of internal consistency is rather obvious. The successful description should be cleared of any contradictions or paradoxes. The predictive power of physics means that the description should point to unambiguous results that are ver-ifiable by experiments. The importance of the predictive power caused specialization among physicists and division into two large classes (by no means disjoint): the experi-mental physicists and the theoretical. The experiexperi-mental physicists conduct experiments which discover new phenomena and verify the theoretical predictions. The theoreti-cal physicists provide quantitative description for already known phenomena and try to encapsulate a newly encountered situation into the existing mathematical framework1. This thesis focuses on theoretical research.
Within the theoretical physics we can still distinguish two approaches. The first one is a pursuit towards the mythological Atoms, the smallest constituents of nature2. The
premise of this approach is that once the Atoms are known and we understand interac-tions between them then we know everything. Everything follows from this Theory of Everything (ToE) [1,2]. As for now such a theory is not known and it is actually even disputed whether such a theory can exist at all [3]. The closest candidate for the ToE is the Standard Model of the elementary particles. Without going into details, the mere fact that the Standard Model cannot fully encompass the gravity signals that we are not quite there yet. The concept of the ToE leads also to an intriguing question. How can it be that, without knowing the ToE, we can still describe some things?
1Which leads from time to time to unexpected discoveries like Dirac’s prediction of the positron 2
The idea of an atom goes back at least to the ancient Greek’s time and to Democritus who first coined the term “Atom”, a physically indivisible and indestructible object out of which all matter is composed.
An answer to this question comes from the history of physics itself and specifically from a history of the pursuit towards the ToE. The first significant candidate for the Democritus Atoms were chemical elements (called atoms as it was believed that they are indeed the Democritus Atoms) discovered at the breakthrough of the 18th and the 19th century. More than one hundred years later it was discovered that atoms are in fact divisible and consist of nucleus and electrons. Later, in the 60’s of the previous century, quarks were discovered, the constituents of the nucleus and along with a few other particles, like the electron and the photon the building blocks of the Standard Model. Important for our discussion is the concept of a hierarchical structure. The fact that lack of knowledge of quarks did not stop us from an exploration and description of a rich world of chemical reactions, or from providing an astonishing agreement between the measurements and the predictions of atomic spectra in the early years of quantum mechanics. The hierarchy of physical theories is enforced by a protection mechanism: the macroscopic theories are blind to many microscopic details3. Partial explanation of this protection is provided through the Renormalization Group logic [4,5].
Therefore instead of following the reductionist ideas in the search of the ToE we can turn around and study how a microscopic theory relates to the macroscopic one. This is the second approach in the theoretical physics and this the point of view that we adopt in this thesis.
The indifference of a macroscopic description to many microscopic details has an im-portant consequence. It implies that microscopic theories cannot on their own faithfully explain macroscopic behavior, instead they provide a possible realization. For example, we could imagine substituting quarks with completely different particles still without modifying properties of chemical elements. The history of superconductivity provides another instructive example. Despite having a theory of everything it took physicists almost 30 years to find a mechanism behind the conventional superconductors. Recently also 30 years have passed since discovery of high temperature superconductors and we are almost as far as we were 30 years ago from understanding them. Therefore tracing the implications of the microscopic theories at the macroscopic level is not merely collecting stamps and involves serious difficulties [6–8].
Usually these difficulties are connected with emergence of an unexpected behavior. An example of an emergent phenomenon is the description of the electrons in metal as essentially free particles. That such a description is emergent turned out really only after discovery of the BCS theory of superconductors [9,10]. In a nutshell, a BCS theory says the electrons in metals are nothing like free particles and the free particle picture
3
Here words microscopic and macroscopic are relative to the system we consider. If we want to describe gas of particles, microscopic means theory describing motion and interactions between all the particles, macroscopic means description in terms of few parameters characterizing the gas as a whole.
emerges only at large enough temperatures. A quite well understood group of emergent phenomena arises in a context of spontaneous symmetry breaking which we will describe in the following sections.
Studying emerging phenomena in physical systems containing many interacting particles is notoriously hopeless. On the other hand uncovering the universal features of emer-gence, or laws of emeremer-gence, is essential to deepen our understanding of the nature. In order to achieve this it is vital to have at hands an exactly solvable microscopic model, where at least in principle, the transition between “micro” and “macro” can be traced rigorously. Even better when such model is also realizable experimentally and theoretical predictions can be verified. And the 1D Bose gas, the main character of this dissertation is such a model.
Elaboration of the ideas of emergence rings through the rest of this chapter, and the rest of this dissertation. But let us start from the beginning.
1.1
The 3D Bose gas and the Bose-Einstein condensation
Before we dwell into the one-dimensional world it is useful to study a few properties of 3D gases. So let us take a gas-filled container of volumeV . The number of gas particles, N , is large and we assume that the gas is in equilibrium. There is, on average, no flow of the density from one place to another and the whole gas has a well-defined temperature T . System is in a well-defined thermodynamic state. Microscopically the particles move and collide but at larger scales there is not much happening. The great insight of Gibbs was that in such a situation the probability that the system is in a given state depends only on the energy of the state. Thus the probability of the system to be in a configuration A is equal to [11]
P (A) = 1 Ze
−(EA−hNA)/T, (1.1)
whereZ, a partition function, defined as
Z =X
B
e−β(EB−hNB), (1.2)
guarantees the proper normalization of the probability P (A). The summation is per-formed over all possible states of the gas. Here EA and NA are the energy and the number of particles in a configuration A. The chemical potential h allows us to implic-itly set the number of particles instead of fixing it from the beginning4.
4
We use letter "h" which is usually reserved for the Planck’s constant. Since we adopt system of units where ~ = 1 this choice should never lead to any confusion.
Let us specialize to the gas of bosonic atoms of equal mass m. We assume that the gas is dilute enough that we can neglect interactions between the atoms. Then the energy is equal to the sum of kinetic energies. Denoting byn~k,Anumber of particles with momentum ~k in a state A we write
EA= X ~k n~k,A(k), NA= X ~k n~k,A, (1.3)
where (k = |~k|) = k2/(2m) is the kinetic energy. Using the Gibbs expression (1.1) the probability of the gas to be in a state A, specified by set of momentum occupation numbers n~k,A, is given by
P (A) = 1 Z
Y ~k
e−n~k,A((k)−h)/T. (1.4)
Since the momentum modes are independent the total probability P (A) can be written as a product over probabilities of single momenta
P (A) =Y ~k
P (n~k,A), P (n~k) = 1 Z~ke
−n~k((k)−h)/T, (1.5)
with a new (partial) partition function
Z~k = ∞ X n~k=0
e−n~k((k)−h)/T. (1.6)
Now let us ask the following question: what is the average occupation of the momentum ~k at a given temperature T ?
Before considering the general case it is instructive to first consider the classical limit. In the classical limit the probability of occupation of a single momentum mode is small. To put it another way the probability that a given mode is unoccupied is close to1. This implies that
1≈ P (n~k = 0) = 1
Z~k, (1.7)
and therefore Zk≈ 1. The probability that there is a single particle then equals
P (n~k = 1) = e−((k)−h)/T, (1.8)
1)≈ 0 we obtain that average number of particles with momentum k follows the Boltz-mann distribution hn~kicl = ∞ X n~k=0 n~kP (n~k) = e−((k)−h)/T. (1.9)
The chemical potentialh should be chosen such thatP~khn~kicl= N . With this constraint in mind the momentum distribution is a function of temperatureT and particle number N .
Let us now move to the general case when we do not assume that occupation numbers are small. Therefore instead of approximating the partition functionZ~k we write
Z~k = ∞ X n~k=0 e−n~k((k)−µ)/T = ∞ X n~k=0 e−((k)−µ)/Tn~k = 1 1− e−((k)−h)/T (1.10)
and the average occupation follows from the equalityhn~ki = T∂h∂ logZ~k,
hn~ki =
1
e((k)−h)/T − 1. (1.11)
This is the Bose-Einstein distribution. The classical case is recovered at large tempera-tures. The transition between the classical and general expression is a smooth transition. As the temperature is decreased the classical formula becomes a worse and worse ap-proximation to the general one.
In the derivation of formula (1.10) it is crucial to assume that the chemical potential is smaller than or equal to zero. Otherwise the geometric series cannot be summed and we cannot assign probabilities to states of the system. On the other hand we see that the chemical potential is a decreasing function of temperature. This is required for the equality N = Pkhnki to hold. Therefore let us assume that at a certain temperature, let us call itTc, the chemical potential h equals to zero. This temperature can be easily computed as follows
AtT = Tcthe following equality holds N V = 1 V X ~k 1 e(k)/Tc− 1. (1.12)
Assuming that the summand is a smooth function of momentum we can approximate the summation by an integration up to terms proportional to inverse powers of the (large)
volumeV . We obtain N V = 1 (2π)3 Z V d3k e(k)/Tc− 1+O(1/V ), (1.13)
Writing the integral in terms of dimensionless variables and using equality (B.7) in Ap-pendixBwe obtain Tc= (2π)2 2m α 2/3 N V 2/3 , α = ζ(3/2)Γ(3/2) ≈ 0.736. (1.14)
At temperatures lower thanTceq. (1.13) does not have solutions. However the equality N = Pkhnki must still be fulfilled. The resolution to this apparent paradox is the assumption that the summation overnkcan be replaced by integration. Indeed note that if nk has a discontinuity around k = 0 the integral does not capture it (d3k = 4πk2dk for spherically symmetric functions). Therefore atT < Tcthe integral captures only the k6= 0 part of the distribution. The k = 0 part can be written as
hnk=0i = N 1 − N1 X ~k6=0 hn~ki , (1.15)
and, after expressing the summation as an integral and using the definition of the critical temperature Tc (1.14), it yields hnk=0i = N 1− T Tc 3/2! . (1.16)
Thus we see that at temperatures below the critical temperature Tc bosonic particles condense in a single momentum state which acquires a macroscopic occupation. This phenomenon is called Bose-Einstein condensation5 (BEC). This natural tendency of bosonic particles at low temperatures to form BEC survives also in a presence of repulsive interactions [12], however with lowerTc.
The BEC phenomenon was theoretically predicted by Einstein in 1926 and experimen-tally realized for the first time in 1995 [13], [14]. Bose-Einstein condensate rapidly started to be created in laboratories around the world (for a review from the “early days” of BEC in atomic gases see [15]) and became a standard manifestation of the quantum nature of the world.
5
It is probably worth emphasizing that a Bose-Einstein condensate is a condensate in momentum space and therefore due to Heisenberg’s uncertainty principle the position of particles is not determined.
0 2 4 6 8 10 12 14 0.0 0.2 0.4 0.6 0.8 1.0 hn k i T = 10 Tc 0 2 4 6 8 10 12 14 0.0 0.2 0.4 0.6 0.8 1.0 T = 3 Tc 0 2 4 6 8 10 12 14 k 0.0 0.2 0.4 0.6 0.8 1.0 hnk i T = Tc 0 2 4 6 8 10 12 14 k 0.0 0.2 0.4 0.6 0.8 1.0 T = 1/3 Tc
Figure 1.1: Plots of the momentum distribution functionhnki = V−1
R
dΩhn~ki. Dots
on the first two plots show the classical approximation (1.9) which at large enough temperatures is indistinguishable from the exact result (1.11). At temperature below Tc (1.14) we observe Bose-Einstein condensation. The zero momentum mode has a
macroscopic occupation. Compare the first plot with the last one to see the quantum effects at work.
Fig. 1.1containing plots of distribution of momenta at various temperature summarizes the results of this section.
1.2
Long-range Order and the MWH Theorem
Let us look now a bit deeper into the BEC phenomenon. The condensate is a macroscopic quantum object: at temperatures belowTc a macroscopic number of particles share the same wave function. This has interesting consequences that we are going to investigate now.
Recall first that wave functions in quantum physics are not determined uniquely, multi-plying the wave function of the whole system by a phase factoreiφ, φ∈ R does not change observables. Therefore the system possesses a U (1) symmetry: each phase of the total wavefunction is as good as any other. On the other hand in the Bose-Einstein condensate a macroscopic number of particles share the same wavefunction and this wavefunction has a specific phase. It means that while lowering the temperature below Tc the U (1) symmetry gets broken: out of all possible phases that the total wavefunction could have,
the condensate function acquires a single well-defined phase. This phenomenon, of a system to be in a state that has less symmetry than the theory describing it, is called spontaneous symmetry breaking.
Spontaneous symmetry breaking is a phenomenon that occurs not only during the Bose-Einstein condensation. Actually many different complex systems can be analyzed from this perspective. Let us take, for example, water. In a liquid state it has translational symmetry. If we observe a homogenous liquid at a positionx and at any other position x + δx the picture is the same. But let us lower the temperature below the Tc = 0◦C. Then the water freezes and forms a crystal, a highly symmetric structure but still less symmetric than the water. What kind of symmetry is broken in this case? The translational symmetry. If we look at the ice we will see molecules of water regularly spaced. The δx that could be any real number for a water, now must be a multiple of lattice spacing.
It was a great insight of Noether, who showed that symmetries in physics are inevitably connected with conservation laws. The existence of a certain symmetry implies the exis-tence of quantity which is constant under time evolution. For example the translational symmetry is connected with momentum conservation. What kind of conservation law is connected with the U (1)? The conservation of particle number. Therefore is the BEC a system which does not conserve a number of particles?
Let us turn the question around and ask which effects a lack of conservation of number of particles would cause. Let us denote by |BECi a state of the gas representing the BEC. Let us also define the operation of removing (annihilating) a particle from the gas at certain placex, Ψ(x). Since the BEC state does not conserve the number of particles we can expect that the state Ψ(x)|BECi is not orthogonal to |BECi6. In fact we can
claim that both states are identical7
Ψ(x)|BECi = |BECi. (1.17)
Of course there is nothing special about positionx that we have chosen in the expression above. This could be as well any other position, let us say x0. Thus denoting byΨ†(x0) an operator conjugated to Ψ(x0) we obtain
g1(x0, x)≡ hBEC|Ψ†(x0)Ψ(x)|BECi = hBEC|BECi = 1, (1.18)
6
Contrary, in the presence of conservation of particle number, states with different number of particles are orthogonal.
7
where in the second equality we assumed that the BEC state is normalized. Amazingly this equality holds even in the presence of interactions8 [12]. Function g1(x, x0) for homogenous systems depends only on the difference of argumentsg1(x, x0) = g1(x− x0) and the fact that it does not vanish for large separations is a remarkable feature. To see this let us consider the gas at T > TC, when there is no BEC. Then all the momentum modes are smoothly occupied (see Fig. 1.1) and if we are to annihilate a particle at position x we never know in advance what value of momentum it has. Therefore an action of Ψ(x) leads to a huge superposition of states with very different momenta. The same of course happens with annihilation at x0. What g
1(x, x0) in fact measures is how similar both superpositions are. But ifx and x0 are far from each other the phases of individual terms in both superpositions are very different and therefore the overlap quickly vanishes. In fact it can be shown that for T > TC, g1(x− x0) decays exponentially with the distance and the larger the temperature the faster the decay [16].
Therefore the condensate is characterized by a coherent phase: the one-body function (g1(x− x0)) does not vanish at large distances but acquires a constant value (1.18)
g1(x→ ∞) → const. (1.19)
This behavior is a signature of a long-range order and as we have seen vanishes immedi-ately atT > TC.
Interestingly this is only true in 3 and (to a much smaller extent) 2 dimensions.
When we consider two or one dimensional systems the restricted dimensionality rein-forces thermal fluctuations. As was shown by Hohenberg [17] at any finite temperature and in the infinite system the relation (1.17) cannot hold. Together with Mermin and Wagner who reached similar conclusions for the Heisenberg model the Mermin-Wagner-Hohenberg (MWH) theorem states that: No spontaneous continuous symmetry breaking can occur at finite temperature and in an infinite system in 1D or 2D.
The MWH theorem does not rule out the possibility of BEC at zero temperature in infinite systems. In fact it can be shown that in 2D there is a phase transition at T = 0. In 1D however there is never a truly BEC condensate in infinite systems. At T = 0 we obtain a quasi condensate, that is a condensate with a slowly varying phase. The one-body functiong1(x) vanishes for large x but it decays only as a power law. The power-law decay of a correlation function signals a quasi long-range order and the system is critical.
8In the presence of interactions or finite temperature only a fraction of particles form the condensate
So far we have described the phenomenon of the Bose-Einstein condensation and shown how it is connected with breaking of the conservation of particle number. We also showed how BEC can be characterized by the existence of a long-range order. Finally we have recalled the MHW theorem claiming that none of these happens in one dimension. So, why the 1D Bose gas?
1.3
A Glimpse of the 1D World
Throughout the previous two sections we were considering a free system, we were ne-glecting the interactions between particles. From a time to time we were claiming that the conclusion holds also in the presence of interactions. But actually the logic should be opposite. We should have start with an interacting system, as in reality they are always there, and thus first fully establish the interacting theory. If in the limit of interaction strength approaching zero the results would qualitatively be the same than we would ar-gue the it makes sense to consider the free theory as an approximation of reality. What is interesting about 1D that the proximity of the free 1D Bose to an interacting theory is completely broken. In 1D the effect of interactions is so strong that it immediately takes us far from the non-interacting case.
This property is easy to understand. In 1D any motion is inherently collective,which is also the essence of the MWH theorem. Any slightest perturbation propagates through the whole system breaking the long-range order. Therefore the 1D gas is a strongly correlated system. It is much stronger correlated than the 3D Bose gas at high temperatures but at the same time is not as perfectly correlated as the BEC. Thus what makes the 1D Bose gas interesting is its position between the classical-like system and the purely quantum, coherent state of matter.
And since this intermediate position is caused by correlations in fluctuation, by studying them we can deepen our knowledge of the 1D Bose gas as a quantum many-body system.
1.4
Structure of the Thesis
The thesis is organized in the following way
• Chapter2: Contains an introduction to the Lieb-Liniger model and to the density-density correlations which we study in various circumstances throughout the rest of the thesis.
• Chapter 3: Contains a brief introduction to the method of Algebraic and Ther-modynamic Bethe Ansatz. Focuses on highlighting key concepts, sacrificing the completeness of discussion whenever it seems appropriate. For the complete expo-sure to the method we refer to the bible of the Algebraic Bethe Ansatz [18]. • Chapter4: This is the first chapter containing the author’s contribution to the field.
We show how the Luttinger liquid (macroscopic theory) can be fine-tuned with the microscopic (1D Bose gas) input. We focus on the zero temperature correlation functions.
• Chapter 5: Here we present results on finite temperature correlation functions obtained directly from the microscopic description of the 1D Bose gas.
• Chapter 6: Finally we describe a metastable super Tonks-Girardeau gas, that is a specific out-of-equilibrium regime in which correlations between particles are specifically strong. In a sense this describes a regime of the 1D Bose gas that is far away from the free gas.
• Chapter7: We conclude by giving an outlook on further interesting directions that could be undertaken in a future and would extend the results presented here.
Many-body Physics in 1D
This chapter provides an introduction to the physics of the 1D Bose gas and its correlation functions. We start with a construction of exact wave functions of the gas by means of the Coordinate Bethe Ansatz. This creates a firm basis on which further discussions in this chapter rely. After highlighting a few properties of the wavefunctions of the 1D Bose gas and connecting them with physical phenomena we will jump to the topic of correlation functions. The main focus here, as well as in the rest of the thesis, lies on the density-density correlation function which we discuss throughout the second part of this chapter. We conclude by reviewing different available methods to study correlation functions in the 1D Bose gas and discussing the relation between theoretical studies and experimental realizations.
2.1
1D Bose Gas
Let us start with a simple problem involving two particles of the same mass m moving along a line. We assume the particles interact through a short-ranged repulsive potential and thus whenever they are far from each other their energy is purely kinetic. In order to build a bit of intuition it is instructive to consider the classical case. Then the momentum and energy conservation in 1D yield that the only possible effect of a short-ranged interaction is an exchange of momenta (velocities) (see Fig. 2.1). If we close the line by placing two walls from which particles bounce elastically then they will keep on exchanging momenta every time they collide and beside that nothing more will ever happen. Interestingly even after increasing number of particles N but without reaching too high densities n = N/L the conclusions do not change1. Although we have more
particles all the collisions still lead only to exchange of momenta and that is it. Thus
1
We assume here that the density is small enough that we can neglect chances of 3-body collisions.
the classical 1D gas is not the most interesting system to devote attention to2. And if there is anything that emerges out of such studies, it is a curiosity to see whether the quantum nature of particles modifies this picture.
k1 k2
k1
k2
Figure 2.1: Schematic depiction of two parti-cles moving along a closed line. The two mo-menta(k1, k2) before collision together with the
momentum and the energy conservation deter-mine the final momenta to be equal (k2, k1).
Quantum physics modifies this picture by in-cluding one more degree of freedom: the relative phase between two particles (which is interac-tion dependent). Generally speaking this is the source of complexity of the quantum version of the 1D gas.
In fact, it modifies it drastically. The 1D Bose gas has such rich physics with so many beautiful phenomena that are neatly expressible in the mathematical language that despite 50 years of studies only re-cently we have started to be able to grasp many of them. It is worth remembering that the complexity that we will encounter during these studies is all due to quantum effects. Therefore studying 1D gases we learn a lot not only about them but also about quantum many-body physics.
2.1.1 Coordinate Bethe Ansatz
Therefore without further delay let us start immediately with a quantum version of the 1D gas. In order to be more specific
let us write down the general Hamiltonian describing a one-dimensional system of mas-sive, bosonic particles with two-body interactions. Usually in the context of cold atoms the interaction is due to spontaneous polarization and results in a short-ranged Van der Waals potential [16]. In the field theory language we write
H = Z L 0 dx ∂xΨ†(x)∂xΨ(x) + Z L 0
dxdy V (|x − y|)Ψ†(x)Ψ†(y)Ψ(x)Ψ(y), (2.1)
with the bosonic fields Ψ(x), Ψ†(x) obeying the canonical commutation relations ([Ψ(x), Ψ†(y)] = δ(x− y) and zero otherwise). We also assume the existence of the vacuum state (Ψ(x)|0i = 0) and its dual. Moreover we impose periodic boundary condi-tions: Ψ(x+L) = Ψ(x). As usual we define the particle number and the total momentum
2
This rather harsh statement should be understood in the context of correlations: 1D classical gases with short interactions do not exhibit any long-range correlations [19].
operators as follows ˆ N = Z L 0 dx Ψ†(x)Ψ(x), (2.2a) ˆ P =−i Z L 0 dx Ψ†∂xΨ(x). (2.2b)
Both operators commute with the Hamiltonian and therefore the particle number and the total momentum are conserved quantities. Conservation of momentum does not depend on the specific form of the interaction potentialV (x, y) as long as the potential is a function of the relative distance only, that isV (x, y) = V (|x − y|).
In what follows we assume that the range of the potential is extremely short and for all practical purposes at low enough temperatures the potential is proportional to the Dirac δ(x) function3. The strength of interaction depends not only on the intrinsic properties
of the atoms but actually can be tuned by various experimental methods (see section
2.4). This in principle allows us to study values of interaction strength extending from minus to plus infinity. For simplicity in this chapter we assume repulsive interactions unless otherwise stated. For a complete discussion of both cases we refer to Chapter 3. Coming back to the Hamiltonian describing the 1D Bose gas we write
H = Z L 0 dx∂xΨ†(x)∂xΨ(x) + cΨ†(x)Ψ†(x)Ψ(x)Ψ(x) , (2.3)
where c sets the aforementioned interaction strength and is positive for repulsive in-teractions. This Hamiltonian was first introduced by E. Lieb and W. Liniger [21] as a generalization of the Tonks-Girardeau gas [22] which is a strongly repulsive limit of (2.3). Varying the density n = N/L of the gas the effective interaction strength changes and thus the correct parameter describing the thermodynamics is γ = c/n. This leads to a counter-intuitive conclusion that the particles are more free for larger densities contrary to the 3D gases. The Tonks-Girardeau gas corresponds to γ → ∞.
Employing the conservation of particle number one can recast the Hamiltonian on a fixed particle number subspace. To this end we write a general state as
|ΨNi = Z L 0 dNx ΨN {xi}Ni=1 YN i=1 Ψ†(xi)|0i, (2.4) 3
This assumption is easily justifiable by noting that the scattering length produced by a δ-potential is the same as the scattering length produced by the Van der Waals potential in the low energy limit. Thus this simplification can be viewed as a projection of the original Hamiltonian on a low energy subspace of the Hilbert space. [20]
where ΨN {xi}Ni=1
is an N-particle wave function. Then an action of H on |ΨNi is equivalent to an action of the following Hamiltonian
HN =− N X i=1 ∂x2i+ 2cX i>j δ(xi− xj), (2.5)
on the wave functionΨN {xi}Ni=1
and thus the eigenvalue problems for H and HN are equivalent. Amazingly the eigenvalue problem
HNΨN {xi}Ni=1
= ENΨN {xi}Ni=1
(2.6)
can be solved exactly [21].
The main idea, which goes back to Bethe’s solution of the 1D magnet [23], follows from the observation that because interactions are point-like, scattering is already asymptotic for any nonzero separation. Moreover since particles are in 1D they can be labeled with respect to their ordering in x. Thus if we restrict ourselves to the domain where x1 < x2<· · · < xN then the wavefunction can be guessed using the following reasoning. Let us start with set of N rapidities {λj}Nj=1 entering the ansatz function (which is not yet an eigenfunction) formed as a product of plane waves
Ψ(x1< x2 <· · · < xN)∼ exp N X j=1 λixi . (2.7)
The conservation of energy and momentum in 1D implies that after collisions the particles keep or interchange rapidities. Thus after all possible collisions the initial set of rapidities will get permuted among all the particles. Thus refining the ansatz we write
ΨN(x1< x2 <· · · < xN) = X P A(P ) exp i N X j=1 λPjxj , (2.8)
where the sum is taken over all permutations of set ofN elements. A(P ) are functions of permutation and of rapidities {λj}Nj=1 but do not depend on spatial variables. The meaning of A(P ) is clear, it captures the effect of interactions. Indeed consider two permutations P and P0 that differ only by transposition of the first two rapidities. We can write
A(P ) = S(λ1, λ2)A(P0), (2.9)
where S(λ1, λ2) describes the interaction (scattering) of two particles with rapidities λ1 and λ2. Since the collisions are elastic S(λ1, λ2) must be a purely phase function.
=
a) b)
Figure 2.2: a) Different representation of collision between two particles (c.f. Fig.
2.1). b) The graphical representation Yang-Baxter equation (2.13). The aftermath of 3 consecutive collisions is independent of the order in which the collisions occurred.
Moreover translational invariance implies that it must be a function of the rapidity difference only. Therefore we write (the minus signs are for convenience only)
S(λ1, λ2) =− exp (−iθ(λ1− λ2)) . (2.10) The function θ(λ) can be easily found. To do this we only need to consider a two-body problem which can be further simplified by noting that the interaction term in the Hamiltonian (2.5) is equivalent to the following boundary condition on the wave function
∂x2Ψ(x1< x2)|x1=x2 − ∂x1Ψ(x1< x2)|x1=x2 = cΨ(x1, x2). (2.11)
Thus substituting the two-body wave function of the form (2.8) in eq. (2.11) we find
θ(λ) = 2 arctan (λ/c) . (2.12)
The functionθ(λ) describes a phase shift caused by interactions. It has two simple limits: for c → 0 it equals π sgn(λ)/2 which is consistent with absence of interactions as then S(λ) = 1 and all A(P )’s are the same. In the other limit, c → ∞, it also becomes constant but this time equals 0. In this case S(λ) =−1 which causes the wavefunction to vanish each time xi = xj,i6= j.
The factorsA(P ) fulfill an interesting relation that in the present context can be viewed as a consistency check. Let us consider two factors A(P ) and A(P00) that differ by transposition of the first and third element instead of the first and the second (see Fig.
2.2). We can obtainA(P00) from A(P ) by successive application of eq. (2.9). Consistency requires that two ways in which we can do it yield the same result. This is guaranteed
once the following relation (Yang-Baxter) is fulfilled
S(λ1− λ2)S(λ1− λ3)S(λ2− λ3) = S(λ2− λ3)S(λ1− λ3)S(λ1− λ2). (2.13) In the present context, withS(λ) given by eq. (2.10), this is naturally the case.
The factorsA(P ) are fully determined once we choose their value for an identity permu-tation. In the end this choice is physically irrelevant since it is equivalent to multiplying the whole wavefunction by a constant (recall the U (1) symmetry). Motivated by the simplicity of the final expression, we choose
A(I) = exp i 2 N X i>j θ (λi− λj) , (2.14)
from which all the others readily follow
A(P ) = (−1)[P ]exp i 2 N X i>j θ λPi− λPj . (2.15)
This specifies the wave function in the fundamental domain (eq. (2.8)). Extension of the wave function to the full domain follows simply from symmetrization of the wave function (as is required for bosonic particles) and yields
Ψ {xj}Nj=1 = Y j<k sgn(xj − xk) ×X P (−1)[P ]exp N X j=1 λPjxj+ 1 2 N X i>j sgn(xi− xj)θ(λPi− λPj) . (2.16) The eigenvalues of the energy and the total momentum are now easy to calculate
EN = N X j=1 λ2j, (2.17a) PN = N X j=1 λj, (2.17b)
where we used that the total momentum in the wave function representation reads
ˆ P =−i N X j=1 ∂x. (2.18)
rapidities. The equality Ψ(x1, . . . , xj, . . . , xN) = Ψ(x1, . . . , xj + L, . . . , xN) yields the Bethe equations λj = 2π LIj− 1 L N X k=1 θ (λj− λk) , j = 1, . . . , N, (2.19)
with quantum numbersIj being integers (N odd) or half-odd integers (N even). In terms of quantum numbers the total momentum of the system is easily expressible
PN = 2π L N X j=1 Ij, (2.20)
which follows from the antisymmetry ofφ(λ) =−φ(−λ). The energy stays as an implicit function of quantum numbers.
Finally let us comment on the name rapidities instead of momenta as the letter sounds plausible in view of eq. (2.17b). The reason is that although the sum of rapidities is an eigenvalue of the total momentum operator, a single rapidity is not an eigenvalue of a momentum operator of a single particle (pˆi = −i∂xj). In fact pˆi does not even
commute with the Hamiltonian (because of the interaction part) and thus the concept of momentum of a single particle does not carry on from the free theories to interacting ones. In order to avoid confusion using the name rapidities instead of momenta seems appropriate.
Empowered now with an exact solution of the Lieb-Liniger model we can discuss some of the physical implications. Leaving the completeness of the discussion for the next chapters (mainly Chapter 3) we focus here just on three aspects: the structure of the ground state and two types of excitations that reign over the low energy sector of 1D Bose gas.
2.1.2 Physical Picture
First of all the quantum numbers {Ij} obey a Pauli principle since the wave function vanishes identically whenever any two quantum numbers coincide. We will prove this important result in the next chapter. What is important for now is that the Pauli principle significantly restricts the structure of eigenstates and yields that the ground state is formed by a Fermi sea (Fig. 2.3). The following choice of quantum numbers minimizes the energy
IjGS=−N + 1
a) b) c)
Figure 2.3: Pictorial representation of concepts introduced in section 2.1.2. a) The Fermi sea: rapidities form the lowest energy configuration, the ground state of the gas. b) the low momentum excitation of the Fermi sea: when rapidity moves slightly off the edge of the Fermi sea, the energy increase is proportional to the momentum. c) the backscattering excitation: can be viewed as either small shift of all rapidities or as a single rapidity jumping from one edge of the Fermi sea to the other side. The large momentum is accompanied by a negligible energy increase.
This can be easily seen when one notices that rapidities are strictly monotonic functions of quantum numbers (see the Bethe equations (2.19)) and the energy (2.17a) is quadratic in rapidities. Therefore the ground state is achieved by the most compact, allowed choice of quantum numbers, hence the Fermi sea structure.
Now let us consider low momentum excitations above the ground state (see Fig. 2.3). For simplicity let us start with the c → ∞ case in which the phase shifts φ(λ) vanish and the Bethe equations (2.19) decouple from each other
λ∞j = 2π
L Ij, j = 1, . . . , N. (2.22)
Then the energyω of a low-momentum k excitation over the ground state is proportional to the momentum: ω = vsk with vs = vF wherevF = πn/m is the Fermi velocity4. In the presence of finite interactions the equality between the sound velocity and the Fermi velocity is broken and in general 0 < vs< vF. The non-zero value of the sound velocity implies that the 1D Bose gas has a liquid like behavior, the low energy excitations are given by sound waves, collective waves of density spreading over the whole system and involving all particles.
Finally there is also a group of very special excitations with finite momentum but with a very small energy (i.e. with the energy vanishing in the thermodynamic limit). These excitations can be created by shifting all the ground state quantum numbers by a constant integerm. From the form of the Bethe equations (2.19) it follows that the rapidities with
4It can be defined as a velocity of the fastest particle in the ground state of the spinless free Fermi
shifted quantum numbers are related to the ground state rapidities byλj = λGSj +2πm/L. It is easy to see that it leads to a state with momentum larger byk = 2πmN/L = 2mkF on one hand, on the other hand the energy is larger only byω = k2/(2N m) and vanishes in the thermodynamic limit. The system thus has low energy excitations also with finite momentum (the umklapp excitations). The appearance of umklapp excitations is a peculiarity of 1D, in higher dimensions there are low-energy excitations with arbitrary momentum, not only 2mkF . It is only in 1D, when the Fermi sea is a line, that an excitation from one edge to the other is connected through high energy excitations (see Fig.2.3).
This structure of the low-lying excitations is not only specific to the 1D Bose gas but is general for the 1D models with gapless (thus linear at small momentum) spectrum. It forms a basis for an effective theory of 1D quantum liquids (Luttinger liquid theory, see Chapter4).
In the next chapter we extend and supplement the sketched here solution of the 1D Bose gas with an algebraic method.
2.2
The Density-density Correlation Function
The main topic of this dissertation is correlations between particles in the 1D gas. Since we assume that there is no external potential acting on the gas we expect particles to homogeneously spread over the length L: the density of the gas n = N/L does not depend on the position. Still the gas is a dynamic system and the density is subject to fluctuations. In order to quantify these fluctuations we can ask the following question: What is the correlation between the expectation value of the density operator at position x2 and time t2 and expectation value of the density operator at position x1 and time t1. The answer to this question is given by the density-density correlation functions (thus the function that describes correlations between density fluctuations, also known as dynamical structure factor or DSF). The pursuit to compute this function for the 1D Bose gas in different physical situations occupies most of what follows in this and the next chapters.
2.2.1 Real Space Correlation Function
Let us start by studying some general properties of the correlation functions (We have already introduced one correlation function, the one-body function, in the first Chapter1, eq. (1.18)). First of all since all the positions are equivalent (the system is homogeneous)
and there is no reference in time the correlation functions are actually functions of the relative distance x = x2− x1 and elapsed timet = t2− t1. Thus we define the density-density correlation function as follows
S(x, t) =hˆρ(x, t)ˆρ(0, 0)i, (2.23) where ρ(x, t) = Ψˆ †(x, t)Ψ(x, t) is the density operator. Since the interaction between particles is short ranged we expect the fluctuations to be uncorrelated for large x or t: S(x, t) −−−−→
x,t→∞ n
2. Moreover in the maximally correlated case S(x, t) = 2n2 and in the completely anticorrelated case S(x, t) = 0. The averaging symbol h. . . i in the definition (2.23) represents expectation value with respect to either some state |Ωi of the Hamiltonian (h. . . i = hΩ| . . . |Ωi) or to an ensemble of states. The most important example of the later is the thermal ensemble following the Gibbs weight (1.1): h. . . i = Z−1P
Ωe−(EΩ−hNΩ)/ThΩ| . . . |Ωi with the partition function Z (see eq. (1.2)).
Equal-time Correlation
For simplicity let us focus on the equal-time correlation functionS(x)≡ S(x, t = 0) also known as the pair correlation function. It supplies us with information about the spatial patterns (are lack thereof) in the density fluctuations. For example looking at smallx we can infer whether particles are likely to be spatially separated and how it is influenced by interaction or temperature.
Exactly atx = 0 this correlation function can be computed by employing the Hellmann-Feynman theorem5. Indeed we have
S(0) =hΨ†(0)Ψ†(0)Ψ(0)Ψ(0)i + δ(0)hΨ†(0)Ψ(0)i = 1 L
∂E
∂c + nδ(0), (2.24) where E = h ˆHi. In general the expectation value of the energy (and its derivative) depends on both the interaction parameter and the temperature. As compared with the large distance limit, the short distance correlation function carries a lot of information about the interplay between quantum correlations (coming from the quantum statistics of the particles), thermal correlations and the effect of the interactions.
For convenience one usually considers the correlation function S(x) minus the discon-nected part nδ(0)
g2(x)≡ (S(x) − nδ(0)) /n2. (2.25)
5The theorem says that the expectation value of any operator appearing in the Hamiltonian is equal to
the derivative of energy with respect to a parameter multiplying this operator (in our case the parameter is c).
0 2 4 6 8 10 12 14 γ 0.0 0.5 1.0 1.5 2.0 g2 (x = 0)
Figure 2.4: Local correlation function at zero temperature computed using eq. (2.24). The energy of the ground state followed from numerical solution to the Bethe equations (2.19) and formula (2.17a). The derivative with respect to the interaction strength was also computed numerically. The dot is an exact result for free bosons.
In Fig. 2.4 we plot g2(0) at zero temperature. One clearly sees that by changing the value of c the local (x = 0) pair correlation function interpolates between uncorrelated value at smallc towards completely anti-correlated, fermionic like behavior of the Tonks-Girardeau gas. It is also worth pointing out that exactly atc = 0 the free bosons exhibit the bunching phenomenon andg2(x) = 2. This effect vanishes immediately at any finite value of c and signals a quantum phase transition between the c = 0 gas with a broken phase symmetry and c 6= 0 gas with a fluctuating phase. Increase of this fluctuation ultimately leads to an anti-bunching phenomenon forc =∞.
To get more intuition about the correlation function it is instructive to consider two limiting situations: weakly interacting gas γ < 1 and strongly interacting gas (γ > 1). Both cases are solvable by referring to either the Bogolyubov approximation (for weakly interacting gas) or by mapping to free fermions (for the strongly interacting case). Let us start with the former case.
The above-observed discontinuity in g2(0) at c = 0 disqualifies immediately any pertur-bative expansion at smallc. However progress can be made by following the Bogolyubov approximation. To this end we write the field operator as a sum of a small momentum (∼ 0) component and the rest
Ψ(x) = 1 LΨ0+ 1 L X k6=0 eikxΨk. (2.26)
0.0 0.2 0.4 0.6 0.8 1.0 x 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 S (x ) γ =∞ γ = 0.2
Figure 2.5: Non-local pair correlation (eq. (2.25)) in the weakly and strongly inter-active regimes at zero temperature.
Noting that the Hamiltonian is minimized by maximizing to occupancy of the lowest momentum mode we replace Ψ0 with a c-number hΨ0i ∼ L√n. This allows for di-agonalization of the Hamiltonian and computation of the correlation functions. The computation itself is shown in the Appendix C, here we show only the final results. Before that however let us consider the opposite limit.
When c→ ∞ the Bethe wavefunctions (2.19) start to resemble the wave functions of a fermionic system. Indeed up to a symmetrization factor the wave function can be written as a Slater determinant Ψc=∞(x1, . . . , xN) = Y j<k sgn(xj− xk) det N exp i2π L Ijxk . (2.27)
Now since the density operator does not depend on the statistic of the particles its expectation values will be the same for a fermionic and bosonic system. Therefore the symmetrization factor does not play any role and the density-density correlation function of the 1D Bose gas at c = ∞ is identical to the density-density correlation function of free fermions. This allows for a straightforward computation of the correlation function which again is performed in the Appendix C.
Fig. 2.5shows the density-density correlation function of weakly and strongly interacting 1D Bose gases. The small x values agree with the earlier calculations based on the Hellmann-Feynman theorem. Beyond that we see now also the fullx dependence together with the largex limit. Interestingly the strong repulsive interaction leads to the formation of a quasi-crystal. The oscillation in S(x) signals the existence of more and less likely
distances between particles. No symmetry breaking occurs but the density fluctuates much more than in the weakly interacting case. Thus the c → ∞ limit is dominated by the density fluctuations whereas thec → 0 limit is dominated by phase fluctuations (sometimes the weakly interacting gas is called a quasi-condensate exactly because the phase still fluctuates so there is no complete condensation). As we should see later in the metastable super Tonks-Girardeau gas the density fluctuations are even stronger. Now let us move to the momentum space and see what kind of information about the 1D Bose gas we can extract from studying the Fourier transform of S(x, t). This will provide us with a different point of view on the nature of correlations and also allow for deriving a few important relations.
2.2.2 Momentum Space
Let us get back to the full distance and time dependent correlation function. Imagine that for some specific values ofx and t the correlation function varies significantly from the uncorrelated value. This means that a fluctuation in density is likely to travel a distance x within a time t. Therefore we can think about the density fluctuation as a density wave with a wave vector (or momentum) k and a frequency (or energy) ω. Transformation of the correlation function from real space (thus function ofx and t) to momentum space is simply given by the Fourier transform. Thus we define
S(k, ω) = Z L 0 dx Z ∞ −∞ dt ei(kx−ωt)S(x, t), (2.28) wherek = 2πm/L with m∈ Z.
TheS(k, ω) function fulfills a simple sum rule 1 L X k Z ∞ −∞ dω 2πS(k, ω) = n 2, (2.29)
which follows fromS(x = 0, t = 0) = n2. Moreover for parity-invariant systemsS(k, ω) = S(−k, ω).
A very practical representation of the correlation function in momentum space can be obtained by explicitly carrying out the Fourier transform in (2.28). To this end it is useful to note that since the total momentum operator commutes with the Hamiltonian we have an equality
ˆ
Inserting now a resolution of the identity ˆ1 = Pλ|λihλ| between the density operators in (2.23) we obtain (when averaging is performed with respect to a given eigenstate |Ωi ∈ HN)
S(x, t) = X λ∈HN
ei[(EΩ−Eλ)t−(PΩ−Pλ)x]hΩ|ˆρ(x = 0)|λihλ|ˆρ(x = 0)|Ωi
= X
λ∈HN
ei[(EΩ−Eλ)t−(PΩ−Pλ)x]|hλ|ˆρ(x = 0)|Ωi|2, (2.31)
where in the last step we used also that the density operator is hermitian. The Fourier transform is easily performed and we obtain
SΩ(k, ω) = 2πL X λ∈HN
|hλ|ˆρ(x = 0)|Ωi|2δ (ω− (Eλ− EΩ)) δk,Pλ−PΩ. (2.32)
This is the Lehmann representation of the correlation function. In what follows it is convenient to have an expression in terms of the matrix elements of the Fourier trans-formed density operator. Explicitly inserting the Fourier transform ofρ(0) = Lˆ −1P
kρˆk we obtain6 SΩ(k, ω) = 2π L X λ∈HN |hλ|ˆρk|Ωi|2δ (ω− (Eλ− EΩ)) . (2.33) For a thermal correlation function we also need to perform the Gibbs trace with respect to |Ωi ST(k, ω) = 2π LZ X Ω,λ∈HN e−(EΩ−hNΩ)/T|hΩ|ˆρ k|λi|2δ (ω− (Eλ− EΩ)) , (2.34)
whereZ is the partition function.
The Lehmann representation is useful for a number of reasons. First of all it gives us a nice interpretation of the correlation functions7. The value of S(k, ω) can be easily connected with the probability of creating an excitation in the gas with given k and ω following the Fermi golden rule logic8.
The Lehmann representation also allows to establish two important equalities fulfilled by the density correlation function: detailed balance and the f-sum rule.
6
The Kroneceker delta δk,Pλ−PΩ is absorbed by the form factor of ρkwhich vanishes for k 6= Pλ− PΩ.
7The Lehmann representation does not depend on the specific choice of the operator (in this case the
density operator) and therefore it easily generalizes to any operator which fluctuations one would like to study.
8
We remind that according to the Fermi golden rule the probability of excitation from a state |Ai to a state Bi due to a given perturbation ˆV is proportional to ∼ |hA| ˆV |Bi|2.
Detailed Balance
Exchanging the meaning of |λi and |Ωi in the summand of eq. (2.34) and using that, due to the presence of the Dirac delta, EΩ = Eλ+ ω we obtain an important equality: the detailed balance condition
ST(k, ω) = e−ω/TST(−k, −ω) = e−ω/TST(k,−ω), (2.35) where the second equality holds for the parity-invariant gas. This relation has a simple interpretation, the negative energy part of the correlation is a dimmed picture of the positive part. The “dimming” factor exp (−ω/T ) depends on both the temperature and the energy. At infinite temperature the correlation function is symmetric aroundω = 0 and at any finite temperature becomes exponentially skewed towards positive energies. The region inω of the negative part of the correlation shrinks rapidly as the temperature is decreased and at zero temperature the correlation with negative energy is exactly equal to zero. The effect of finite temperatures is visible only for ω < T . For higher ω the correlation is almost identical to the zero-temperature case. We will get back to this observation and to the detailed balance in Chapter 5 where we consider finite temperature correlation functions.
The f-sum Rule
There is one more relation that is fulfilled by the density-density correlation: the f-sum rule Z ∞ −∞ dω 2πωSΩ(k, ω) = nk 2. (2.36)
Starting from the left hand side and performing the integral we obtain
I(k) = Z ∞ −∞ dω 2πωSΩ(k, ω) = 1 L X λ∈HN (EΩ− Eλ)|hλ|ˆρk|Ωi|2 =− 1 2LhΩ| [H, ˆρ−k] ˆρk− ˆρ−k[H, ˆρk]|Ωi. (2.37) The expression on the right hand side can be written in terms of double commutators: fromS(k, ω) = S(−k, ω) it follows that I(k) = I(−k), thus
I(k) =− 1
4LhΩ| [[H, ˆρ−k] , ˆρk] + [[H, ˆρk] , ˆρ−k]|Ωi. (2.38) Since the double commutators are hermitially conjugated it is enough to consider one of them. To this end we write the Hamiltonian and the density operator in momentum
space as H = 1 L X k k2Ψ†kΨk+ c L3 X k1,k2,q Ψ†k 1+qΨ † k2−qΨk1Ψk2, (2.39a) ˆ ρk= 1 L X q Ψ†k+qΨk. (2.39b)
which follows from the Fourier transform of eq. (2.3) and definition of the density operator ˆ
ρ(x) = Ψ†(x)Ψ(x) (see also AppendixA). Evaluation of the commutators simplifies upon noting that the density operator commutes with the interaction part of the Hamiltonian and the kinetic part yields simply
[H, ˆρ−k] = 1 L X q k2− 2kqΨ†−k+qΨq. (2.40)
The double commutator gives then
[[H, ˆρ−k] , ˆρk] =−2k2 1 L X q Ψ†qΨq, (2.41)
which proves the f-sum rule relation. The f-sum rule states that the average energy of the fluctuations with momentumk is always the same irrespective of interaction strength (the r.h.s does not depend onc) and equal to k2 just as if it were carried by a single particle. This shows that the only effect of interactions is to change the energy dependence of S(k, ω) while keeping the average energy constant. In fact as we will see later this dependence can be modified drastically and one of the purposes of the work presented in this thesis is to give quantitative predictions forS(k, ω) in various circumstances.
The Static Correlation Function
As for the real space correlator it is also useful to consider correlation functions reduced to one variable. This is achieved by integrating out either ω or k and yields a static correlator (static structure factor) or an autocorrelator
S(k) = Z ∞ −∞ dω 2πS(k, ω), (2.42) S(ω) = 1 L X k S(k, ω). (2.43)
This is then consistent with the Fourier transform of static correlatorS(k) which equals the pair correlation function S(x).
As discussed before, the excitations in the 1D Bose gas are gapless and thus at low momentum they are like sound waves with the sound velocity vs that depends on the interaction and temperature. At zero temperature, the gas occupies the lowest energy state and thus only excitations with positive energy are possible. In this case we can write that
S(k, ω) = Akδ(ω− vsk), (2.44)
where Ak is an unknown function of momentum. This function can be uniquely fixed using the f-sum rule (2.36), we obtain
S(k) = |k| vs
+O(k2), T = 0. (2.45)
Thus at zero temperature the static correlator vanishes linearly withk. The proportion-ality constant is the inverse sound velocity.
The opposite limit of the static structure factor is when k → ∞. In this case the correlation function becomes independent of temperature and interaction strength and approaches the universal value1. This can be easily understood. First from the physical point of view very large momentum means that we are looking at the ultra local (in real space) correlation, and this in turn can depend only on a single particle property
9. Thus the static correlator should be equal to this of free bosons (or free fermions)
which indeed is equal to 1. Mathematically speaking at large momentum, the Fourier transform of S(x) (that is S(k)) depends only on the discontinuity of S(x) at x = 0 (see eq. 2.24) what yields1 for S(k→ ∞).
Fig. 2.6 summarizes the prediction on the limits of the static structure factor together with the results of Bogolyubov approximation and the Tonks-Girardeau gas (see Ap-pendixB). Similar analysis could be performed for the autocorrelation function however we save it for discussions in further chapters.
2.3
How to Compute Correlation Functions
In the previous chapter we have introduced correlation functions, discussed their certain, universal limits, and what is most important showed how understanding of correlation functions enriches our understanding of physical systems. Despite invoking calculations in certain limits when the theory becomes essentially free (the Tonks-Girardeau limit and Bogolyubov approximation) the question how to compute the full dynamical correlation
9For short-range interactions and not to large densities there is a region around every particle where
0.0 0.5 1.0 1.5 2.0 2.5 3.0 k[kF] 0.0 0.2 0.4 0.6 0.8 1.0 1.2 S (k ) γ =∞ γ = 0.2
Figure 2.6: The static correlator at the zero temperature. We see a linear dependence at low momenta and approaching asymptotically1 for large momenta. Comparing with eq. (2.45) we see that the sound velocityvs≤ vF with the equality holding forc =∞.
functions remains open. In this section we would like to address this problem by briefly surveying the most suitable methods that one has at hand.
The available methods can be divided roughly in two groups: the numerical methods or analytic methods. We will start with the first group.
The 1D Bose gas being a continuum model restricts the accessible numerical methods to various types of Monte Carlo algorithms. The other methods relying on a finite basis of the Hamiltonian like Density Matrix Renormalization Group [24, 25] or Stochastic Series Expansion [26] or even straightforward diagonalization are not available since the Hilbert space of a continuum model is strictly infinite dimensional. A method that was successfully used in studies of 1D Bose gas and its variants (such as with an external potential or crossover between 3D and 1D) is the Diffusive Monte Carlo method [27,28]. Unfortunately the natural limitation of the method is to static properties of the system such as equal-time correlation functions [29,30]. Similar limitations are also present in the second method that we would like to mention here, the Stochastic P-gauge which was used to computeg2(x) in the almost full γ− T phase space [31].
The second group of methods relies on having some form of an analytic solution of the theory. The full S(k, ω) function was studied by perturbative expansion around the Tonks-Girardeau limit [32, 33]. The spatial correlation function g2(x) was studied at various interaction strengths at zero and at finite temperatures by employing either perturbative methods or Bogolyubov approximation [31]. The more complete treatment of the correlation functions but limited only to large-distance and long-time asymptotes
is given by the Luttinger liquid theory. We will briefly summarize the idea behind this theory and discuss it recent developments that allow to gain insights also in the momentum dependent correlations in a separate chapter (see Chapter 4). The most successful attempt in computation of correlation functions starting from the exact N− body Hamiltonian of the 1D Bose gas is presented in the work of Lyon group [34–37]. This method, that relies on the Algebraic Bethe Ansatz (see the next chapter), allows for capturing the large-distances long-times asymptotes of correlation function and proved to be consistent with the Luttinger liquid predictions.
2.3.1 The ABACUS
Finally we would like to use the last paragraph briefly describe an alternative approach. This approach was already used in the past [38, 39] and this thesis documents small further developments in the same direction. The main idea is as follows. The Lehmann representation of the correlation function (2.33) that for convenience is repeated here
S(k, ω) = 2π L
X λ∈HN
|hλ|ˆρk|Ωi|2δ (ω− (Eλ− EΩ)) (2.46) gives in fact a simple prescription. All that we need to compute a correlation function is to: (1) Be able to span the Hilbert space, that is enumerate (in some consistent way) all the eigenstates. (2) Compute their energy. (3) Compute the matrix elements. (4) Perform summation. This simple prescription relies on two pillars. First computation of matrix elements of an interesting operator (like density) in an interacting theory is a challenge on its own. Fortunately for the 1D Bose gas these matrix elements are known exactly thanks to the Algebraic Bethe Ansatz. The second pillar is the complexity of the summation problem. The number of states that needs to be considered in order to achieve good results (the quality of results can be assured by the f-sum rule, for example) should not be exponentially large in the system size. Fortunately the number of states contributing significantly to the summation is only polynomial in system size and thus is relatively small. This makes the method feasible. We will not go here into details of how the algorithm in fact works and for more information we refer to [40].
2.4
How to Measure Correlation Functions
Finally we would like to conclude this chapter by addressing the experimental relevance of correlation functions. In this last section we will briefly show how the correlation functions explicitly enter measurable quantities and we will show also two experimental
configurations that are widely used nowadays to prepare and perform experiments with cold atomic gases and specifically with the 1D Bose gas.
The most straightforward way to measure the density-density correlation function is Bragg spectroscopy [41, 42]. In this setup two lasers beams are used to create a time dependent potential
VBragg= V0 Z L
0
dx cos (kx− ωt) ˆρ(x, t), (2.47) whereV0 is the intensity of laser beams,k is the momentum reflecting the geometry of the setup and ω is a result of a slight detuning of the beams. The perturbation caused by VBragg couples to the density of atoms and transfers energy and momentum into the system. Within the logic of linear response theory the transferred momentum and energy are directly related to the density-density correlation function and are given by [41]
dP dt =L V0 2 2 k (ST(k, ω)− ST(k,−ω)) , (2.48a) dE dt =L V0 2 2 ω (ST(k, ω)− ST(k,−ω)) . (2.48b) In principle eq. (2.48) supply us with a direct relation between the density-density cor-relation function and measurable quantities like change in the momentum or the energy of the gas.
What is left is a problem of creating a 1D gas. There are two popular setups: using an atomic chip, or using a highly anisotropic optical lattice. Let us start with the former. In an atom chips experiments one uses a pre-manufactured chip with nano wires engraved on its surface. The geometry of the wires together with a control over currents allows for the creation of a magnetic field that acts as a trap for atoms. Leaving out the experimental details this allows for trapping atoms in the effective 1D geometry like a single tube. Unfortunately trapping is realized through magnetic field and thus one cannot use a Feshbach resonance to control the interaction strength. This means that the atom chip experiments are limited to weakly interacting 1D Bose gas [43,44].
The second popular approach is to use optical lattices. Using a configuration of 6 lasers one creates standing waves that trap atoms at the minima
Vopt(x, y, z) = X i=x,y,z
Visin2(qixi) . (2.49)
By choosing highly anisotropicVi, for exampleVz Vx, Vy, the trap becomes elongated in one direction and narrow in the other two. This results in an array of 1D tubes. Usually
the experimental procedure populates many of these tubes and results of measurements are an average over the whole array. Therefore the signal is not as clear as in atom chip trap experiments especially because different tubes contain different number of atoms. Thus the densities n, and as a result also the interaction strength γ, vary among the tubes and smear the contribution from a single tube at a single value ofγ. The advantage of optical lattices however is that they allow to use the Feshbach or confinement induced resonances. Moreover the optical setting allows for flexible tuning of radial confinement which also influence the interaction strength [20]. Altogether the effective tuning of interactions allows to reach even the Tonks-Girardeau limit [45–47].
