Introduction to quantum many-boson theory

a theory of almost everything

Introduction to Quantum Many-boson Theory

a theory of almost everything

Xiao-Gang Wen Department of Physics, MIT

Contents
1 Principle of emergence. 1.1 1.2 1.3 1.4 1.5 1.6 Origin of elementary particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The emergence of the waves and the mechanical laws of liquids and solids. Waves = particles . . . . . The Maxwell equation, the Dirac equation and ether . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 4 5 8

One stone for two birds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 The vacuum is not empty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 13

2 Classical picture of quantum system 2.1 2.2 2.3

Picture vs. symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 **Mathematics, physics, and quantum gravity . . . . . . . . . . . . . . . . . . . . . . 14 Classical physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5 2.3.6 Classical states and equation of motion . . . . . . . . . . . . . . . . . . . . . 17 A particle in magnetic eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 The Lagrangian description of classical systems . . . . . . . . . . . . . . . . . 20 *A Lagrangian description of a 2D classical particle in magnetic eld . . . . . 22 *Magnetic eld in phase space . . . . . . . . . . . . . . . . . . . . . . . . . 24 **Generalized equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . 25 Quantum states and quantum motions . . . . . . . . . . . . . . . . . . . . . . 28 Classical-quantum correspondence . . . . . . . . . . . . . . . . . . . . . . . . 29 Equation-of-motion approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 *Coherent-state approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.4 2.5 2.6

Quantum physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.4.1 2.5.1 2.6.1 2.6.2 From classical physics to quantum physics . . . . . . . . . . . . . . . . . . . . . . . . 29 From quantum physics to classical physics . . . . . . . . . . . . . . . . . . . . . . . . 31

2.7 2.8

*Berrys phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 **Ambiguities and gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.8.1 2.8.2 2.8.3 **The ambiguity in the Lagrangian description of classical systems . . . . . . 40 **The ambiguity in the state-vector description of quantum states . . . . . . 40 **The gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 iii

3 A particle on a lattice 3.1 3.2

44

A particle hopping problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 *A particle hopping problem multi-band cases . . . . . . . . . . . . . . . . . . . . . 47 52

4 Boson systems 4.1 4.1.1 4.1.2 4.1.3 4.2 4.2.1 4.2.2 4.2.3 4.2.4 4.2.5 4.2.6 4.2.7 4.2.8 4.3 4.3.1 4.3.2 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.4.5 4.5 4.5.1 4.5.2 4.5.3 4.5.4

Two pictures for bosons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 A rst look at a free boson system . . . . . . . . . . . . . . . . . . . . . . . . 52 **A brief look at Fermi statistics . . . . . . . . . . . . . . . . . . . . . . . . . 54 A second look at the free boson system . . . . . . . . . . . . . . . . . . . . . 57 A vibrating-string picture of 1D boson system . . . . . . . . . . . . . . . . . . 58 Classical eld theory of free bosons . . . . . . . . . . . . . . . . . . . . . . . . 61 Physical quantities in the eld theory description . . . . . . . . . . . . . . . . 63 *The second quantized description of free bosons . . . . . . . . . . . . . . . . 64 Vacuum as a dynamical medium and the Casimir eect . . . . . . . . . . . . 66 *The second quantized description of interacting bosons . . . . . . . . . . . . 69 Classical eld theory of interacting bosons . . . . . . . . . . . . . . . . . . . . 71 *A coherent-state approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 A quantum phase transition in interacting boson system . . . . . . . . . . . . 74 Continuous phase transition and symmetry . . . . . . . . . . . . . . . . . . . 76 Collective modes sound waves . . . . . . . . . . . . . . . . . . . . . . . . . . 79 Quantized collective modes phonons . . . . . . . . . . . . . . . . . . . . . . 81 *Low energy eective Lagrangian for the sound waves . . . . . . . . . . . . . 83 *An oscillator picture for the sound wave . . . . . . . . . . . . . . . . . . . . 86 *An improved quantum variational approach . . . . . . . . . . . . . . . . . . 88 Superuid ow and momentum quantization . . . . . . . . . . . . . . . . . . 89 Stability of the owing state . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 . . . . . . . . . . . . . . . . . 91 *Collective excitations above the owing state

Field theory for many-boson systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

Phases and phase transitions of interacting bosons . . . . . . . . . . . . . . . . . . . 74

Low energy collective excitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Superuidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

Vortex and the decay of superow . . . . . . . . . . . . . . . . . . . . . . . . 92

iv

Chapter 1

Principle of emergence.
Why solids have xed shape and liquids can ow freely? Why some materials are magnetic? Why some materials are conductors and some are insulators? Why metals are shiny? Why liquid crystals can change their transparency? Condensed matter physics is a branch of physics that study the properties of various materials, including those mentioned above. Since there are so many dierent kinds of materials with vastly varying properties, not suppricingly, condensed matter physics is a very rich eld. Usually for each kind of material, we need to a dierent theory (or model) to explain its properties. So there are many dierent condensed matter theories and models to explain various properties of dierent materials. However, after seeing many dierent type of theories/models for condensed matter systems, a common theme among those theories start to emerge. The common theme is the principle of emergence,[Anderson 1972] which states that the properties of a material are mainly determined by how atoms are organized in the material. Dierent organizations of atoms lead to dierent materials and/or dierent phases of matter, which in turn leads to dierent properties of materials. For example, the xed-shape property of a solid is originated from the fact the atoms in a solid organize into a regular array. If atoms have a random uctuating distribution, then the material will behave like a liquid. However, the power of the principle of emergence is not limited in condensed matter physics. In this introduction, we will use the origin of elementary particles and the origin of waves and mechanical properties in solids/liquids to illustrate how principle of emergence oer a unied way to understand a vast amount of seemingly very dierent phenomena in condensed matter physics and in elementary particle physics. The later chapters of this book will oer a pedagogical and a more detailed exploration of the principle of emergence through a few simple model systems. We will also introduce some simple mathematical formalism to describe the principle of emergence in a quantitative way.

1.1

Origin of elementary particles

Elementary particles are not elementary. They are collective motions of a dynamical medium the vacuum. Physics laws are emergent too. The emergence explain the beauty of physics laws. The world around us is rich, complicated, and fascinating. Despite its unbounded richness and complexity, scientists have a faith that the world is reasonable and understandable. Here by reasonable and understandable we mean that the behavior of everything in our world can be 1

understood in terms of something very simple. We do not need something more mysterious than our would to understand it. But how do we understand our world? So far a reductionist point of view has dominated our approach to achieve such an understanding. At the beginning of the human civilization, people realized that matter can be divided into smaller and smaller parts. Some Chinese philosophers theorized that the division can be continued indenitely, and hence there are no elementary particles. Other Chinese philosophers theorized that all matter is formed by gold, wood, water, re and earth. Greek philosophers assumed that the division cannot be continued indenitely. As a result, there exist the ultimate and indivisible particles the building block of all matter. This may be the rst concept of elementary particles. Those ultimate particles were called atomos. Since all matter is formed by the atomos, an understanding of those atomos will allow us to understand everything in our world. Because of this reductionist point of view, a signicant part of science has been devoted to nding and understanding those atomos. Around 1900, chemists discovered that all matter is formed by a few dozens of dierent kinds of particles. People jumped the gun and named them atoms. After the discovery of electron, people realized that atoms are not atomos. The elementary particles are actually smaller than the atoms. Now, many people believe that photon, electron, quarks, and a few other particles are elementary particles. All matter is formed by those particles. But what is the origin of the those elementary particles? Are photons, electrons, quarks really indivisible? Can they be formed by even smaller more elementary particles (or superstrings [Green et al. 1988])? Can an even deeper understanding of nature be achieved by nding those smaller and more elementary objects? However, this line of questions and the concept of elementary particles itself may be fundamentally awed. They are based on the implicit assumptions that more basic understanding of matter can be achieved by dividing it into smaller and smaller pieces. This assumption is rooted in the belief that the vacuum is empty and that things placed in the empty vacuum are divisible. However, there are many examples from condensed matter physics indicating that sometimes this line of thinking does not make sense. In those examples, the low energy quasiparticles (for example phonons) cannot be understood by dividing them into smaller parts. Motivated by those condensed matter examples, in this book I will advocate a dierent view of the elementary particles based on the principle of emergence. In this approach, we view our vacuum as a dynamical medium formed by bosons or spins.1 The deformations of this medium (i.e. the collective motions of bosons/spins) are waves, which in quantum theory correspond to various particles. We will show that when the bosons/spins are organized in a certain way, the deformations of such organized medium correspond to wave described by the Maxwell equations and the the Dirac equations. Those waves correspond to photons and electrons in quantum theory. (A more detailed discussion of emergent waves and their corresponding particles will be given in the next a few sections.) The new point of view represents a dierent way to understand some fundamental questions of nature. This is because the interactions between the bosons/spins that form the vacuum medium can be quite arbitrary and ugly. But this does prevent the emergence of beautiful and reasonable photons and electrons from the such a messy medium. So instead of thinking that complex structures arise from simple beautiful building blocks, we think that beautiful elementary particles emerge from an arbitrary and messy structure at short distance.
We have been fascinated by the beauty of physics laws. Some of us have been believing that the fundamental physics laws must be beautiful. Such a faith is so deep that many physicists use the beauty of physics laws as a guide for future research direction: to nd more and more beautiful physics laws. People believe that if you found
1

Bosons and spins are simplest local objects in quantum theory.

a beautiful physics law, then very likely such a law will be the correct law that describes our world. However, if we say that elementary particles are emergent phenomena, we imply that the laws of physics that govern the behavior of elementary particles (and everything formed by the elementary particles) are also emergent. But why the emergent physics law must be beautiful. This is particularly puzzling knowing that the underlying boson/spin system (the vacuum medium) can be arbitrary, messy, and ugly. The principle of emergence also provides an answer to the origin of the beauty of the physics laws: the beauty comes from the separation of length and energy scales. The length scale of the boson/spin system that produces the elementary particles is very small. It is of order
G 33 the Planck scale: lP = cm. The associated energy is also very large EP = c/lP = 1041 eV. c3 = 1.61610 The collective excitations in the local bosonic system are in general interacting with each other. In general an interaction will give collective excitations an energy gap of order of the Planck energy EP . This will make the collective excitations and the corresponding elementary particles unobservable at energies much below EP .

However, in our world, the energy gaps for the observed elementary particles are less than 1012 eV. They are so small compare to the Planck energy so that the elementary particles can be treated as gapless (or massless) excitations. So the physics laws (or the interactions) for those elementary particles cannot be a generic one. Those laws must be very special (or beautiful) so that the interactions between the (almost) gapless elementary particles do not give them large energy gaps of order EP . We see that the simplicity and the beauty of the physics laws that govern the collective excitations do not come from the simplicity and the beauty of the local bosonic system, but from the fact that those laws have to allow the collective excitations to survive at low energies. The interactions between the bosons in the local bosonic system can be unreasonable and ugly. The interactions (or the physics laws) for the low energy collective excitations, on the other hand, are always beautiful, so that they do not open large energy gaps for (nearly) massless excitations. This is how beautiful physics laws emerge from ugly models at short distances. One example of beautiful physics laws is the Coulomb law. It says that the interaction force between two charges separated by a distance r is proportional to 1/r2 . We think such a law to be beautiful since the exponent is exactly 2. It is not 3, not 4.171 or one of billions of other possible values. Another example of beautiful physics laws is the interaction between two moving bubbles in a superuid. The interaction force is proportional to 1/r4 . Again the exponent 4 is exact. Traditionally, the 1/r4 law for the moving bubbles is understood as an emergent law in condensed matter physics, and the Coulomb law, on the other hand, is understood as a fundamental law of physics. However, in this book, we will take a dierent point of view. We will show that both the Coulomb law between two charges in vacuum and the 1/r4 law between two moving bubbles in a superuid are emergent phenomena. Both laws can be understood under the same framework of emergence. In chapter 4, we will explain the emergence of of 1/r4 law in superuid and in chapter ??, we will explain the emergence of the Coulomb law.

Earth use to be regarded as the center of the universe. It was viewed as the stage for almost all that happens. It had an untouchable and noble status. However, as times goes by, earths status was reduced to merely one of the billions of planets in the universe. It appears that the concept of elementary particles may have a similar fate. The view of emergence may bring the status of elementary particles from the building blocks of everything to merely collective modes of, possibly, a lowly boson/spin system. This is a good news for condensed matter experimentists, since it tells us that, in principle, we can make articial elementary particles, such as articial photons and articial electrons, in laboratories.
Problem 1.1.1 Where are we? On an axis of log scale, mark the positions of the size of universe, the size of a human, and the size of Planck length. Try to ll the axis by marking the length scales of other interesting/important objects/phenomena. You will discover an empty gap (the desert) in the axis. According to the theory of emergence, the desert is the origin of the beauty of the physical laws.

Figure 1.1: Liquids only have a compression wave a wave of density uctuations.

1.2

The emergence of the waves and the mechanical laws of liquids and solids.

Waves and wave equations in many-particle systems are determined by how the particles are organized in the ground state. In this section, we will use the origin of waves and mechanical properties to explain, in a more concrete way, what is the principle of emergence. From very early on, people realized that two kinds of matter, solids and liquids, have very dierent mechanical properties. Basically, a liquid has no shape and can ow freely. In contrast, a solid has a xed shape and can retain its shape. Using a more technical term, we say that a solid can resist shear deformations while a liquid cannot. We know that a deformation in a material can propagate just like the ripple on the surface of water. The propagating deformation corresponds to a wave traveling through the material. Since liquids can resist only compression deformation, so liquids can only support a single compression wave (see Fig. 1.1).2 Mathematically the motion of the compression wave is governed by the Euler equation3 2 2 v 2 2 = 0, (1.2.1) t2 x where is the density of the liquid. Solid can resist both compression and shear deformations. As a result, solids can support both compression wave and transverse wave. The transverse wave correspond to the propagation of shear deformations. In fact there are two transverse waves corresponding to two directions of shear deformations. The propagation of the compression wave and the two transverse waves in solids are described by the Navier equation 2 ui 2 uj Tjikl k l = 0 t2 x x where the vector eld ui (x, t) describes the local displacement of the solid. We would like to point out that the Navier equation and the Euler equations not only describe the propagation of waves, they actually describe all small deformations in solids and liquids. Thus, the two equations represent a complete mathematical description of the mechanical properties of solids and liquids. The Navier equation and the Euler equation are two very important equations. They lay the foundation of of many branches of science such as mechanical engineering, aerodynamic engineering, etc. But why do solids and liquids behave so dierently? What makes a solid solid and a liquid liquid? Many years after the discoveries of Navier and the Euler equations in 19th century, we started to realize in 20th century that both solids and liquids are formed by collections of atoms. The main dierence between the solids and liquids is that the atoms are organized very dierently.
Compression wave is also called longitudinal wave. To be precise, the liquids discussed in this chapter are liquids at zero temperature. The zero-temperature liquids are always superuids.
3 2

(1.2.2)

Figure 1.2: Drawing a grid on a sold helps us to see the deformation of the solid. The vector ui in eqn (1.2.2) is the displacement of a vertex in the grid. In addition to the compression wave (i.e. the density wave), solid also contains transverse wave (wave of shear deformation) as shown in the above gure.

(a)

(b)

Figure 1.3: (a) Atoms in liquids do not have xed relative positions. They uctuate freely and have a random but uniform distribution. (b) Atoms in solids form a xed regular lattice. In liquids, the positions of atoms uctuate randomly (see Fig. 1.3a), while in solids, atoms organize into a regular xed array (see Fig. 1.3b).4 It is the dierent organizations of atoms that lead to the dierent mechanical properties of liquids and solids. In solids, both the compression displacements (see Fig. 1.4a) and the transverse displacements (see Fig. 1.4b) lead to real physical changes of the atomic congurations. Such changes cost energies. As a result, solids can resist both kinds of deformations. This is why we have both the compression wave and the transverse wave in solid. In contrast, the shear displacements of atoms in liquids do not result in a new conguration since the atoms still have uniformly random positions. So shear deformation is a do-nothing operation for liquids. Only the compression deformation which changes the density of the atoms results in a new atomic conguration and costs energies. As a result, liquids can only resist compression and have only compression wave. We see that the properties of the propagating wave are entirely determined by how the atoms are organized in the materials. Dierent organizations lead to dierent kinds of waves and dierent kinds of mechanical laws. Such a point of view of dierent kinds of waves/laws originated from dierent organizations of possibly the same kind of particles is a central theme in condensed matter physics. This point of view is called the principle of emergence.

1.3

The Maxwell equation, the Dirac equation and ether

The Maxwell equation, the Navier equation and the Euler equation should be viewed at an equal footing. In addition to the Navier equation and the Euler equation, the Maxwell equation which describes the dynamics of electric eld E and magnetic eld B c E + t B = c B t E = 0
4

The solids here should be more accurately referred as crystals.

(a)

(b)

Figure 1.4: The atomic picture of (a) the compression wave and (b) the transverse wave in a crystal.

Figure 1.5: The atomic picture of the compression wave in liquids. is also a very important equation. The Maxwell equation also describes a propagating wave which is called electromagnetic wave. Maxwell found that the velocity of the electromagnetic wave is close to the speed of light. He proposed that light is the electromagnetic wave, which turns out to be correct. So the Maxwell equation provides us a deeper understanding of light. It not only unies the electricity and magnetism, it also unies the electromagnetic wave with light. However, despite its stunning beauty and wide spread applications, the above description of the Maxwell equation is quite phenomenological and lack of a deep understanding. The this level of understanding for the Maxwell equation is the same as that of the Navier equation and the Euler equation in 19th century before the discovery of atoms. In 20th century, we gained a deeper understanding of the Navier equation and the Euler equation. We realized that the two equations are originated from two dierent organizations of atoms. Knowing that the Navier equation and the Euler equation come from the motion of atoms in solids and liquids, one naturally want to know: what is the origin of the Maxwell equation? The motion of what gives rise electromagnetic wave? When the Maxwell equation was rst introduced, people rmly believe that any wave must correspond to motion of something. Maxwell himself have tried to invent a mechanical model so that the electromagnetic wave correspond to the motion of the mechanical parts in the model (see Fig. 1.6). [Maxwell 1952] However, Maxwells mechanical model does not really work. One can also show that solids and liquids do not give rise to the Maxwell equation either. This is because the electromagnetic wave described by the Maxwell equation is very dierent from the compression wave and transverse wave in liquids and solids. More specically, the waves in liquids have only one compression mode and the waves in solids contain one compression mode and two transverse modes. On the other hand the electromagnetic waves contain two transverse modes corresponding to the two polarizations of light. Because of those dierences, we cannot regard to the electromagnetic wave as motion of atoms in liquids or solids. The two types of organizations of atoms in liquids and solids are unable to produce the electromagnetic wave. However, despite many failed attempts, people still insisted that the electromagnetic wave must be a motion of a certain medium. Such a hypothetical medium is called ether. For many years, no one knew what is ether and what does ether look like. But before anyone gured out what ether is, Michelson and Morley [Michelson and Morley 1887] performed their famous experiment in search of ether. They found that the speed of light is the same regardless if the earth moves along or perpendicular to the direction of light. Using the Galileo transformation that relates the velocities measured on dierent moving platforms, they 6

Figure 1.6: Maxwells ether: a mechanical model that might lead to the Maxwell equation and electromagnetic waves. conclude that ether does not exist. Based on Michelson-Morleys experiments and the symmetry of the Maxwell equation, Einstein developed his theory of special relativity and pointed out that the Galileo transformation between dierent moving platforms is incorrect. The correct transformation is the Lorentz transformation. But Michelson-Morleys conclusion of non-existence of ether stuck despite the basis of their analysis is no longer valid. Even worse, people commonly associate Einsteins special relativity with the non-existence of ether. In fact, it is Einsteins special relativity that allows ether to exist. According to Einsteins theory, an ether that has the Lorentz symmetry will be consistent with Michelson-Morleys observation. So we are back to square one: what does ether look like? In condensed matter physics, we know that if particles organize themselves into a liquid, then the collective motion of particles will be described by the Euler equation. If particles organize themselves into a lattice, their collective motion will be described by the Navier equation. So from this point of view, the question about the ether becomes In what form must particles organize in order to give rise to waves described by the Maxwell equation? In addition to the Maxwell equation, the Dirac equation is also a very important equation (despite it is less well known). the Dirac equation describes wave of electrons (and other fermions). The Dirac equation is much stranger than the three wave equations discussed above. This is because the waves of the Dirac equation suppose to describe fermions particles with Fermi statistics. The waves described by Euler, Navier, and the Maxwell equations can only correspond to particles with Bose statistics, such as phonons and photons. (For a discussion about the relation between waves and particles, see section 1.4.) In order to produce fermions, according to the standard quantum eld theory, the amplitude of the wave must be anticommuting Grassman numbers. Such a wave is very hard to visualize. Since electrons and their waves are so strange, few people regard electrons and the electron waves as collective motions of something. People accept without questioning that electrons are fundamental particles, one of the building blocks of all that exist. However, from a condensed matter physics point of view, all low energy excitations are collective motion of something. If we try to regard light as collective motions of proper organized particles, why can we not regard the strange electron wave as similar collective motions? So we can ask a similar question for electrons: What organization of particles can give rise to the Dirac equation and Fermi statistics? A recent study provides a unied answer to the above two questions. One nds that light wave and electron wave, just like sound wave, can indeed emerge from a properly organized particle system, or more precisely, a boson system. If the bosons in the system form large strings and if those strings can move around freely and form a liquid state (or more presicely, a quantum liquid state at zero temperature), then the collective motions of the such organized particles will

Figure 1.7: A quantum picture of ether: The uctuations of string-net give rise to electromagnetic waves (or light). The ends of strings correspond to electrons. correspond to waves described by the Maxwell equation and the Dirac equation.[Levin and Wen 2004, 2005a] The strings in the string liquid are free to join and cross each other. As a result, the strings look more like a network (see Fig. 1.7). For this reason, the string liquid is actually a liquid of string-nets, which is called string-net condensed state.5 Let us rst discuss the connection between string-net liquid and the light wave described by the Maxwell equation. We will discuss the relation between string-net liquid and the electron wave in the section 1.5. We like to understand why the waving of string-nets produces waves described by the Maxwell equation? We know that the particles in a solid organized into a regular lattice pattern. The waving of such organized particles produce a compression wave and two transverse waves. The particles in a liquid have a more random organization. As a result, the waves in liquids lost two transverse modes and contain only a single compression mode. The particles in a string-net liquid also have a random organization, but in a dierent way. The particles rst form string-nets and string-nets then form a random liquid state. Due to this dierent kind of randomness, the waves in string-net condensed state lost the compression mode and contain two transverse modes. A hand-waving way to understand how the compression mode is lost is given in Fig. 1.8 (for details, see chapter ??). Such a wave with only two transverse modes is exactly the electromagnetic wave described by the Maxwell equation.

1.4

Waves = particles

A understanding of a wave = a understanding of a kind of particles. After realizing the string-net origin of the Maxwell equation and the resulting electromagnetic
5 To be more precise, string-net theory of light (and other gauge bosons) contain two main points. (a) The Maxwell equation emerges from local bosonic systems. This is because the particles that form the string-nets are bosons. (b) In order for the Maxwell equation to emerge from local bosonic systems, the particles should organize into string-net and string-net should form a quantum liquid. Thus string-net is not a building block of the model, but a description of an ordering of particles. The electric eld and the magnetic eld in the Maxwell equation are called gauge elds. The string-net liquids demonstrate how gauge elds emerge from local bosonic systems. Many closely related earlier works led to such a picture of the Maxwell equation. String structures appear in the Wilson-loop characterization[Wilson 1974] of gauge theory. The Hamiltonian and the duality description of lattice gauge theory also reveal string structures.[Kogut and Susskind 1975; Banks et al. 1977; Kogut 1979; Savit 1980]. Lattice gauge theories are not local bosonic systems and the strings are unbreakable in lattice gauge theories. The unbreakability of the strings is the root of the gauge invariance in the gauge theory. Since the gauge invariance is the dening property of a gauge theory, so for a long time, people feel that gauge theory cannot emerge from a local bosonic system since the strings form by bosons are always breakable. String-net theory points out that even breakable strings can give rise to gauge theory with massless gauge bosons.[Hastings and Wen 2005] So we do not really need fundamental unbreakable strings. Bosonic particles themselves are capable of generating gauge elds and the associated the Maxwell equation. This phenomenon was discovered in several bosonic systems[Foerster et al. 1980; Baskaran and Anderson 1988; Wen 2002; Motrunich and Senthil 2002] before realizing their connection to the string-net liquids. The connection between ends of strings and Fermi statistics is a more recent realization.[Levin and Wen 2003]

(a)

(b)

Figure 1.8: A hand-waving way to understand why uctuations of condensed string-net have only two transverse modes. Here strings are formed by particles on a lattice. (a) A transverse motion of a string results in a new state and leads to a collective excitation. (b) A motion along the string does not result in any new states. Such a motion does nothing and does not lead to any collective excitations.
p

Figure 1.9: Quantized waves and number of particles. The solid curves represent the vibrations of a wave which correspond to one or two particles. The dashed curves represent the other allowed vibrations. The momentum p of each particle is related to the wave length of the wave p = h/ where h is the Planck constant. wave, we like to ask if such an understanding can shed any light on the question about the origin of elementary particles. At rst sight, the above question sounds strange. How can an understanding of the origin of a wave have anything to do with the origin of certain particles? However, our nature is such a strange reality that the waves and the particles in it are actually the same things. In other words, a wave is actually a collection of particles and a particle propagates and even interferes just like a wave. A theory quantum mechanics was formulated in early 1900 to describe such a strange behavior of our nature. The realization of the particle-wave equivalence together with its quantum mechanical description represents the single most important discovery in 20th century. Since waves and particles are the same things in our world, an understanding of the origin of a wave implies an understanding of the origin of the corresponding particles. The particle that correspond to the electromagnetic wave is called photon, which is one of the elementary particles. So the string-net picture also provides an origin of an elementary particle photon. Let us briey explain how particle and wave become the same thing in quantum theory and in our world. We know that a wave is a vibration of something. In contrary to a general belief, the amplitude of the vibration is not arbitrary. It can only take a sequence discrete values. If the of smallest amplitude is said to be 1, then the next allowed amplitude is 2. In general, the n-th allowed amplitude is n (see Fig. 1.9). A wave with the smallest allowed vibration corresponds to the presence of a single particle. The momentum p of the particle is determined by the wave 9

length of the wave: p = h/ where h is the Planck constant. Similarly, the wave with the nth allowed vibration amplitude corresponds to the presence of n particles, each carries the same momentum p = h/. This is how wave and particle become the same thing. Dierent types of waves correspond to dierent types of particles. The electromagnetic waves correspond to photons while the vibration waves in solids correspond to particles called phonons.
We like to point out that quantum mechanics plays a central role in the string-net picture for electromagnetic waves and photons. Without quantum theory, waving classical string-nets actually are not electromagnetic waves. This is why, at the time of Maxwell, it was hard to nd any model that produces electromagnetic waves. To understand the origin of the Maxwell equation, we must start with a quantum description of string-nets and view ether as a quantum liquid of string-nets. The waving of quantum string-nets directly gives rise to photons and then many photons put together form electromagnetic waves. This is a more correct way to understand how string-nets provide an origin for photons and the Maxwell equation.

1.5

One stone for two birds

Electrons as ends of strings. While the string-net picture for ether and the resulting light/photons is quite appealing, it may not be the only possible picture for ether and light. There may be other ways to understand the origin of light. So we are not sure if light is really the waving of string-nets. However, the string-net picture has a really nice bonus. It not only explains the origin of light, as we have mentioned earlier, it also explains the origin of electrons without any additional assumption. To see how electrons appear from string-nets, we would like to point out that if we only want photons and no other particles, the strings must be closed strings with no ends. The uctuations of closed strings produce only photons. If strings have open ends, those open ends can move around and just behave like independent particles. Those particles are not photons. In fact, the ends of strings are nothing but electrons. How do we know that ends of strings behave like electrons? First, since the waving of string-nets is an electromagnetic wave, a deformation of string-nets correspond to an electromagnetic eld. So we can study how an end of a string interacts with a deformation of string-nets. We nd that such an interaction is just like the interaction between a charged electron and an electromagnetic eld. Also electrons have a subtle but very important property Fermi statistics, which is a property that exists only quantum theory. Without their Fermi statistics, the electrons in atoms would have a very dierent organization. All atoms would have very similar chemical properties and behave like noble-gas atoms. Amazingly, the ends of strings reproduce this subtle quantum property of Fermi statistics.[Levin and Wen 2003, 2005b] Actually, string-net liquids explain why Fermi statistics should exist. (A more detailed discussion of emergent Fermi statistics from string-nets will be given in chapter ??.) We see that string-nets naturally explain both light and electrons. In other words, string-net theory provides a way to unify light and electrons.[Levin and Wen 2004, 2005a] So, the fact that our vacuum contains both light and electrons may not be a mere accident. It may actually suggest that the vacuum is indeed a string-net liquid. Here, we would like to point out that there are many dierent kinds of string-net liquids. The strings in those dierent liquids may have dierent numbers of types and may join in dierent ways. For some string-net liquids, the waving of the strings does not correspond to light and the ends of strings are not electrons. Only one kind of string-net liquids give rise to light and electrons. On the other hand, the fact that there are many dierent kinds of string-net liquids allows us to explain more than just light and electrons. We can nd a particular type of string-net liquids which not only gives rise to electrons and photons, but also gives rise to quarks and gluons.[Wen 2003b; Levin and 10

Wen 2005b] The waving of such type of string-nets corresponds to photons (light) and gluons. The ends of dierent types of strings correspond to electrons and quarks. It might even be possible to design a string-net liquid that produces all elementary particles! In this case, the ether formed by such string-nets would provide an origin of all elementary particles.6

1.6

The vacuum is not empty

Vacuum is a special kind of material. The string-net picture for elementary particles represents a very dierent way to view our would. We used to believe that the vacuum is empty. All objects are somethings placed in the empty vacuum. So objects and vacuum are viewed as separate entities. This point of view of empty vacuum is the fundamental reason why we try to understand all things by dividing them into smaller parts. This is also why we think the smaller pieces are simpler and more fundamental. However, according to the string-net theory, the vacuum is not empty. The vacuum itself is a dynamical medium formed by string-nets. The motions and the ends of strings give rise to the elementary particles, which in turn form all matter in our universe. In some sense, we like the shes in ocean who regard water as empty space. With such a comparison, light in our vacuum corresponds to sound wave in the water, and electrons may like bubbles in the water. Within the string-net picture, it does not make sense to ask if elementary particles (such as photons and electrons) are formed by even smaller particles. This is because if we look at photons and electrons really closely, we will not see smaller particles that form photons and electrons. We will see string-nets whose motion gives rise to photons (light), or we will see ends of strings which correspond to electrons. So the traditional way to gain a deeper understanding of nature by trying to nd the smallest building block may not be the right approach. We cannot study matter and the vacuum separately. The particles that form matter are just collective motions of things (for example, the string-nets) that form the vacuum. Matter and vacuum are two aspect of same thing. Also, within the string-net picture, the elementary particles and their properties are traced back to the organization of string-nets in our vacuum. Dierent organizations will lead to dierent kinds of elementary particles with dierent properties, which in turn lead to dierent worlds. So we attribute why our world behave as it does to the structure of string-nets that ll the space. The beauty, the ugliness, the reason and the arbitrariness in physics laws are all rooted in the organization of string-nets in the vacuum. To summarize, string-net liquid represents a dierent way to understand the deep structure of matter, where the elementary particles are not regarded as the building blocks of everything but as an emergent phenomenon from a deeper structure of our non-empty vacuum. String-net liquid provides a unied origin for almost all the elementary particles. In other words, if we say let there be string-net liquid, we will get almost everything.7 This is not just a fun thing to say. In
So far we can use string-net to produce almost all elementary particles, expect for two: the SU (2) gauge boson that is responsible for the weak interaction and the graviton that is responsible for the gravity. 7 We know that superstring theory is a potential theory of everything. One may want to ask what is the dierence between the string-net-liquid approach and the superstring approach? Our understanding of the superstring theory has been evolving. According to an early understanding of the superstring theory, all the elementary particles correspond to small segments of superstrings. Dierent vibration modes of a small superstring result in dierent types of elementary particles. This point of view is very dierent from that of the string-net liquid. According to the string-net picture, everything comes from simple bosonic particle. The bosons stick together to from string-nets which ll the whole space. The strings can be as long as the size of universe. Light (photons) correspond to the collective motion of the large string-nets and an electron corresponds to a single end of string. A modern understanding of the superstring theory is still under development. According to Witten, one of the most important questions in superstring theory is to understand what is superstring. [Witten 2004] So at this time, it is impossible to compare the
6

11

principle, we can realize string-net liquids in certain materials which will allow us to make articial elementary particles. So we can actually create an articial vacuum, and an articial world for that matter, by making a string-net liquid. This would be a fun experiment to do! We see that the principle of emergence not only can explain the rich properties of all dierent materials, it can also explain the facinating properties of one most important material our vacuum itself. In the rest of this book, we will rst use the principle of emergence to explain the properties of superuids and solids. Then we will use the principle of emergence to explain the emergence of electromagnism and Fermi statistics.

modern understanding of the superstring theory with the string-net theory. In particular it not clear if the superstring theory can be viewed as a local bosonic system. The string-net theory is fundamentally a local bosonic system.

12

Chapter 4

Boson systems
A boson system is the simplest system in nature. It demonstrates a wide variety of physical phenomena, such as superuidity, magnetism, crystals, etc. We can also use boson system to study many dierent phase transitions. Despite (or due to) its simplicity, a boson system may also be the deep fundamental structure that produces all the elementary particles including photons and electrons [Wen 2002, 2003b; Levin and Wen 2005a]. If this is true, a boson system will actually be a theory of everything. In this chapter, we will study interacting bosons using a classical picture. We rst develop a classical eld theory that describes the bosons. Then we consider the collective vibration modes of the eld. After quantizing those vibration modes, we gain a understanding of the low energy properties of the quantum interacting bosons.

4.1
4.1.1

Two pictures for bosons

A rst look at a free boson system

Many-boson Hamiltonian and many-boson energy eigenstates for a free boson system. There are two kinds of particles in nature, bosons and fermions. Photons and 4 He atoms are two examples of bosons. Photons hardly interact with each other. So the photon system is a noninteracting boson system, or a free boson system. 4 He atoms have a short range interaction. For a dilute 4 He gas there is little chance for two atoms to be close to each other. Thus the interaction between the 4 He atoms can also be ignored and we can treat the 4 He gas as a system of free bosons. In this section, we will study such free boson systems. To simplify our discussion even further, we will consider free bosons in one dimension. The generalization to higher dimension is often straight forward. To construct a quantum theory for many bosons, let us start with the simplest case: the state with no particle. Such a state is called a vacuum state and is denoted by |0 . The energy of such a state is zero. The next simplest state is a state with one particle. Actually there are many dierent oneparticle states. Those states form a Hilbert space H1 . One set of bases vectors for H1 is |x which describe a particle at x. |x s are normalized according to x|x = (x x ).

52

A generic one-particle state | is described by a complex wave function (x): | = dx (x)|x

Let us assume that the particle with momentum k has an energy (k). Then the one-particle Hamiltonian will have a form H1 = (ix ) (4.1.1) The energy eigenstates of such a Hamiltonian are plane waves |k = dxe i kx |x

with energy (k). Certainly the statistics is not important here. The one-particle states for a boson and for a fermion are identical. For a particle in three dimensions, H1 becomes H1 = (ix , iy , iz ) where (kx , ky , kz ) is the energy of a particle with momentum (kx , ky , kz ). The energy eigenstates are 3D plain wave state |k with energy (k). For non-relativistic particles, (k) has a form k2 (k) = 2m + U where m is the mass of the particle and U the potential seen by the particle. For relativistic particles (k) has a form (k) = c2 k2 + m2 c4 where c is the speed of light. For a single massless photon, the dispersion can be obtained by taking m = 0 and is given by (k) = c|k|. The two-particle states form a bigger Hilbert space H2 . One set of bases vectors for H2 is |x1 x2 with a understanding that |x1 x2 and |x2 x1 are the two names for the same physical state. So we have |x1 x2 = |x2 x1 (4.1.2) The equivalence of |x1 x2 and |x2 x1 is very important. If |x1 x2 and |x2 x1 describe two dierent quantum states, then there are two dierent states with one particle at x1 and one particle at x2 . In this case the system will be a system of non-identical particles. The condition that there is only a single state with one particle at x1 and one particle at x2 makes the particles in our system identical particles. A generic two-particle state is given by |two-particles =

x1 x2

dx1 dx2 (x1 , x2 )|x1 x2

(4.1.3)

Note that the integration is only over the region x1 x2 to avoid double counting, since |x1 x2 and |x2 x1 represent the same state. So the the two-particle wave function (x1 , x2 ) is only dened for x1 x2 . Using eqn (4.1.2), we can extend the wave function (x1 , x2 ) to the region with x1 > x2 through the relation (x1 , x2 ) = (x2 , x1 ) This allows us to rewrite eqn (4.1.3) as |two-particles = 1 2 dx1 dx2 x1 ,x2 |x1 x2

where the integration is over the whole 2D plane (x1 , x2 ). We see that the states of two identical particles can be described by symmetric wave functions (x1 , x2 ) = (x2 , x1 ). 53

A careful reader may note that so far we only specied that the two particles are identical particles. We did not specify if the two particles are bosons or fermions. So the above reasoning implies that both bosonic and fermionic identical particles are described by symmetric wave functions. But what determines the statistics of the identical particles? It turns out that the statistics is not determined by the symmetry or antisymmetry property of the wave function, but by the Hamiltonian that governs the dynamics of the two particles. If we choose the Hamiltonian that acts on the two-particle state |two-particles to be the sum of two one-particle Hamiltonian (4.1.1) H2 = (ix1 ) + (ix2 ) (4.1.4)

then the two identical particles will be bosons. Further more such a Hamiltonian also implies that there is no interaction between the two particles. So H2 describes our free 1D boson system with two bosons. We note that H2 is invariant under the exchange x1 x2 . So when it acts on a symmetric wave function (x1 , x2 ), H2 will generate another symmetric wave function. Since the identical particles (bosons or fermions) are always described by symmetric wave functions, the two-particle Hamiltonian for identical particles are always invariant under the exchange, so that the action of the Hamiltonian on the allowed wave functions can only generate allowed wave functions. The energy eigenstates of H2 are plain waves (x1 , x2 ) = e i (k1 x1 +k2 x2 ) + e i (k1 x2 +k2 x1 ) (which is symmetric under the exchange of x1 and x2 ) or |k1 k2 =
x1 x2

dx1 dx2

e i (k1 x1 +k2 x2 ) + e i (k1 x2 +k2 x1 ) |x1 x2

We note that |k1 k2 = |k2 k1 . So |k1 k2 s are also redundant names: |k1 k2 and |k2 k1 are two names for the same plain wave state. The energy of the plain wave state is Ek1 k2 = (k1 ) + (k2 ). The above discussion can be easily generalized to n-particles. The n-particle Hamiltonian have a form
n

Hn =
a=1

(ixa )

(4.1.5)

For non-relativistic particles, (k) =

k2 2m

+ U . The n-particle Hamiltonian is reduced to

n

Hn =
a=1

1 2 + U) 2m xa

(4.1.6)

Such a Hamiltonian determines the statistics of the particles to be bosonic. The energy eigenstates are |k1 k2 kn with energy n (ki ). We will call such a way to describe bosons as the particle i=1 picture of the boson gas. The dierent orders of k1 k2 kn in |k1 k2 kn correspond to the same state, for example |k1 k2 k3 = |k2 k1 k3 = |k3 k1 k2 . In contrast, for non-identical particles The dierent orders of k1 k2 kn in |k1 k2 kn correspond dierent quantum states: |k1 k2 k3 = |k2 k1 k3 = |k3 k1 k2 . So non-identical particles have a very dierent Hilbert space than the identical particles.

4.1.2

**A brief look at Fermi statistics

Identical particles can always be described by symmetric wave functions. The statistics of the identical particles is determined by n-particle Hamiltonians. This provides a unied way to understand Bose, Fermi, and fractional statistics. 54

y x (x,y)

Figure 4.1: The denition of the function (x, y). We have stressed that both bosons and fermions can be described by symmetric wave functions. The statistics of the identical particles are determined by the many-particle Hamiltonian. The particular two-particle Hamiltonian (4.1.4) gives rise to Bose statistics. A curious reader may wonder what kind of two-particle Hamiltonian gives rise to Fermi statistics. The following two-particle Hamiltonian is a concrete example that gives rise to Fermi statistics in 2D (in non-relativistic limit):
2 2 2 2 ferm = (x1 + iax ) + (y1 + iay ) (x2 iax ) + (y2 iay ) H2 2m 2m y1 y2 x1 x2 ax = , ay = , 2 + (y y )2 (x1 x2 ) (x1 x2 )2 + (y1 y2 )2 1 2

(4.1.7)

ferm is invariant under the exchange (x1 , y1 ) (x2 , y2 ). Such a two-particle HamilWe note that H2 tonian when acting on symmetric wave functions describes two fermions in two dimensions. How to understand such a result? H ferm can be simplied by the following transformation
2

(x1 , y1 , x2 , y2 ) asym (x1 , y1 , x2 , y2 ) = e i (x1 x2 ,y1 y2 ) (x1 , y1 , x2 , y2 ) H ferm H asym-ferm = e i (x1 x2 ,y1 y2 ) H ferm e i (x1 x2 ,y1 y2 )
2 2 2

(4.1.8)

where (x, y) is the angle between the vector (x, y) and the positive x direction (see Fig. 4.1). y For positive x and y, (x, y) = arctan x . Although (x, y) is discontinuous on the positive x-axis with a discontinuity of 2, the function e i (x,y) is a smooth function of (x, y) (except at (x, y) = (0, 0)). Using the relation e i (x,y) x e i (x,y) = x + i x2 y , + y2 e i (x,y) y e i (x,y) = y i x2 x , + y2

asym-ferm has a simple form we nd that the transformed Hamiltonian H2

2 2 2 2 asym-ferm = 1 (x + y ) 1 (x + y ). H2 1 2 2m 1 2m 2

From e i (x,y) = e i (x,y) , we can show that the transformed wave function is antisymmetric asym (x1 , y1 , x2 , y2 ) = asym (x2 , y2 , x1 , y1 ). We know that when acting on antisymmetric wave functions asym (x1 , y1 , x2 , y2 ), the simple asym-ferm describes two fermions in two dimensions. This way we contwo-particle Hamiltonian H2 clude that when acting on symmetric wave functions (x1 , y1 , x2 , y2 ), the complicated two-particle ferm describes two fermions in two dimensions. Hamiltonian H2 In the standard way to understand fermions, fermions are dened as particles described by antisymmetry wave functions. Through the above discussion, we see that this standard understanding of fermions did not capture the essence of Fermi statistics. This is because fermions can 55

be described by both symmetric wave functions (with a complicated many-particle Hamiltonian) or antisymmetric wave functions (with a simpler many-particle Hamiltonian). I personally believe that symmetric wave functions plus complicated many-particle Hamiltonian is a correct way to understand fermions, at least physically. The standard understanding using antisymmetric wave function is very formal and misleading despite its mathematical simplicity. The confusion caused by the standard understanding is reected in the following conversation: A: I have two fermions. One at x1 and the other at x2 . I wonder what is the amplitude of such a state. B: Well it depends on how you say it. If you say one fermion at x1 and one fermion at x2 , the amplitude will be . If you say one fermion at x2 and one fermion at x1 , the amplitude will be . A: This is ridiculous. The two ways of saying mean exactly the same thing. How come it leads to two dierent results. B: Well, it is not that ridiculous. You know that two wave functions dier by a total phase factor e i actually describe the same quantum state. So the amplitudes and actually correspond to the same physical state. There is no contradiction. A: But then why is the minus sign important? Why does the minus sign characterize the Fermi statistics? Saying one particle at x2 and the other at x1 may very well leads to an amplitude e i instead of . The phase should have no physical meaning, less to determine the statistics of the identical particles. B: Well, we should look at the wave function of identical particles (x1 , x2 , ..., xn ) as a whole. Imposing an arbitrary exchange phase, say (x1 , x2 , ..., xn ) = e i (x2 , x1 , ..., xn ), may result in a discontinuous many-particle wave function (x1 , x2 , ..., xn ). Only when = 0 or can we have a continuous wave function. This is why we have only Bose statistics and Fermi statistics. A: But the continuity of the wave function should not be essential. If the identical particles are dened on a lattice, the continuity of the wave function will be meaningless. On the other hand, identical particles on lattice still have well dened statistics. In 2D we can even have fractional statistics. Using symmetric/anti-symmetric wave function to understand statistics misses the fractional statistics in 2D. So symmetric/anti-symmetric wave function does not represent a correct way to understand statistics (at least physically). I hope that I have made my point. The statistics of identical particles is a very tricky subject. Using exchange symmetry of many-particle wave function to understand statistics is formal and misleading. It misses the essence of statistics. If we understand the statistics that way, the origin of statistics will appear to be very mysterious. Such a understanding does not tell us how to make identical particles with dierent statistics. It does not encourage us to think how to make identical particles with dierent statistics. It suggests that the statistics is fundamental and is given. We just have to accept it. In contrast, the description of identical particle using symmetric wave function and encoding the statistics in the many-particle Hamiltonian leads to completely dierent picture. I believe that it is a more correct picture that captures more essence of statistics despite its mathematical complexity. Within such a picture, statistics of identical particles is a dynamical property determined by manyparticle Hamiltonian. We can change the Hamiltonian to obtain dierent statistics. We can also naturally obtain fractional statistics in two dimensions within this picture (see Prob. 4.1.2). Such an understanding tells us how to make dierent statistics and we can have phase transitions that change the statistics of particles. We will have a more detailed discussion of Fermi statistics later.
Problem 4.1.1 (a) Find a 3-particle Hamiltonian that acts on symmetric wave functions and describes Fermi statistics. (Hint: You may generalize eqn (4.1.7)). (b) Find a transformation that transform symmetric wave function to antisymmetric wave function. Find the transformed 3-particle Hamiltonian. (Try to choose the 3-particle Hamiltonian in (a) such that the transformed

56

Figure 4.2: Exchanging two particles.

3-particle Hamiltonian is a sum of three one-particle Hamiltonians.) Problem 4.1.2 By rescaling ax and ay in eqn (4.1.7), we can obtain the Hamiltonian that describes two particles with fractional statistics. As an example, the following Hamiltonian, when acting on symmetric wave functions, describes two such particles conned by a harmonic potential K (x2 + y 2 ): 2 1 1 1 1 frac (x1 + iax )2 (y1 + iay )2 (x2 iax )2 (y iay )2 H2 = 2m 2m 2m 2m 2 K 2 2 + (x2 + y1 + x2 + y2 ) 2 2 1 y1 y2 x1 x2 ax = , ay = , (x1 x2 )2 + (y1 y2 )2 (x1 x2 )2 + (y1 y2 )2

(4.1.9)

where is the statistical angle. = 0 correspond to bosons and = correspond to fermions. frac (a) Find the ground state energy of H2 for = 0, /2, and . (b) The ground state energy for two bosons/fermions in the harmonic potential K (x2 + y 2 ) can also be obtained 2 by lling energy levels. Check the correctness of your results in (a) against the results obtained by lling energy levels. (c) Can we still use a transformation similar to eqn (4.1.8) to encode the fractional statistics in the exchange symmetry of the wave function?

4.1.3

A second look at the free boson system

The bosons states with dierent numbers of bosons can all be labeled by occupation numbers. A free boson system is equivalent to collection of harmonic oscillators. In the section 4.1.1, we have viewed vacuum as an empty stage. The bosons are actors on the stage. In this picture, the existence and the origin of identical particles are very mysterious. To appreciate this point, let us consider a state with particle 1 at x and particles 2 at y. After exchanging the two particles we get another state with particle 1 at y and particles 2 at x (see Fig. 4.2). If the two states are dierent, then the two particles are not identical. If the two states before and after the exchange are actually the same state, then the two particles are called identical particles But why the two states have to be the same? It appears that we can always follow the trajectories of the particles in time history to distinguish the two states before and after the exchange. We wonder where do identical particles come from? Why do they have to exist? The mystery of identical particles is one of most fundamental mystery of our nature. It reects certain deep structures in physics laws and in our vacuum. But what is the message? What does the existence of identical particle tell us about physics laws and the vacuum? In this section, we will try to provide an answer to those fundamental questions. We will use the 1D free boson system to illustrate our points. We assume the bosons live on circle of length L. In this case, the wave vectors k are quantized: k = n 57 2 n L

4 3 2 1 0 1 2 3 4 k

Figure 4.3: The circles on the dispersion curve (k) correspond to allowed k-levels labeled by n . The empty dots represent unoccupied k-levels with nn = 0, and the lled dots represent occupied k-levels with nn = 1. Two lled dots together represent doubly occupied levels with nn = 2. The state in the above graph is described by | n1 n0 n1 n2 with n3 = 1, n2 = 2 and other nn = 0 in the occupation-number notation. It is a three-boson state described by |k1 k2 k3 with k1 = 6/L and k2 = k3 = 4/L in the wave-vector notation. where n is an integer. The total Hilbert space of arbitrary number of bosons is formed by the no-boson state, one-boson states, two-boson states, etc: H = H0 H1 H2 = {|0 , |k1 , |k1 k2 , } Instead of using ki that describes the momentum of each boson, we can use a set of the occupation numbers nn to label each bosonic state in H | n1 n0 n1 n2 . nn is the number of bosons in the level with momentum k = n (see Fig. 4.3). Such an level will be called klevel. The vacuum state |0 is given by the state with all nn = 0: |0 = | 0000 . The one-boson state |k1 is given by the state with all nn = 0, except nk1 = 1. A more general state is represented in Fig. 4.3. In this way, all the states in H are labeled by | n1 n0 n1 n2 with nn = 0, 1, 2, . From the relation between the two ways of labeling states, |k1 k2 and | n1 n0 n1 n2 , one can show that i (ki ) = n nn (n ). So if the bosons are free, the state | n1 n0 n1 n2 is an energy eigenstate with an energy Etot =
n

nn (n ).

(4.1.10)

The expression (4.1.10) tells us that the free boson system can be viewed as a collection of harmonic oscillators. The dierent oscillators in the collection are labeled by n . The eigenstates of 1 the oscillator n are labeled by nn . The energy of |nn is (nn + 2 ) n where n is the oscillation angular frequency of the oscillator n . If we put the oscillators together, a state of such a collection can be denoted as | n1 n0 n1 n2 , where the oscillator n is in the nn th excited state. The 1 energy of such a state is n (nn + 2 ) n . We see that if we choose n = (n ), then the above energy will reproduce the boson energy n nn (n ) apart from an overall constant 1 n (n ). 2 Also the oscillator states and the many-boson states have an one-to-one correspondence. Thus the free many-boson system can be viewed as a collection of harmonic oscillators.

4.2
4.2.1

Field theory for many-boson systems

A vibrating-string picture of 1D boson system

A 1D boson system is equivalent to quantized vibrating string. A classical vibrating string provides a classical picture of 1D bosons. The vibrating string has a more formal name 1D eld theory. 58

In this section, we would like to show that the collection of harmonic oscillators can be viewed as the vibrating modes of a string. The connection between the oscillators and the vibration modes of string leads to a vibrating-string picture or a eld theory of the bosons.
If the bosons have a relativistic dispersion relation (k) = c2 k 2 + m2 c4 , then the corresponding collection of the oscillators will has the following set of frequencies n = (n ) = c2 2 + m2 c4 . Such a collection of n the oscillators correspond to the vibration modes of an ordinary string. Even for bosons with a generic dispersion (k), the corresponding collection of the oscillators can still be viewed as vibration modes of an string, but such a string will have some unusual dynamics.

To obtain the vibrating string picture of bosons, let us introduce the coordinate Xn and the momentum Pn to describe the oscillator labeled by n . The energy of the oscillator is given by En = 1 1 2 2 Pn + Kn Xn . 2Mn 2 (4.2.1)

where Mn is the mass and Kn is the spring constant of the oscillator. The classical equation of motion for (Xn , Pn ) is given by En Xn (t) = = Pn (t)/Mn Pn En Pn (t) = = Kn Xn (t) Xn Since the oscillation frequency is n = (n ), Mn and Kn must satisfy (n ) =

(4.2.2) Kn /Mn .

We have some freedom to dene what we call coordinate and momentum. We note that the following rescaling (Xn , Pn ) (Xn , 1 Pn ) does not change the form of the equation of motion (4.2.2) nor the energy (4.2.1). So the rescaled (Xn , Pn ) remain to be an coordinate-momentum pair.1 The rescaling only changes the values of Mn and Kn : (Mn , Kn ) (2 Mn , 2 Kn ), but not the oscillation frequency. Thus we can use the rescaled (Xn , Pn ) to describe the same oscillator. Using the freedom of rescaling, we can choose Mn = 1/ (n ) and Kn = (n ). The equation of motion is simplied to Xn (t) = (n ) Pn (t) Pn (t) = (n ) Xn (t) and the energy of the oscillator becomes En = (n ) 2 2 (Pn + Xn ) 2

(4.2.3)

Now introduce a complex eld to encode the coordinates and the momenta of all the oscillators: (x, t) =
n

n (t)L1/2 e i n x

(4.2.4)

1 n = (Xn + iPn ). 2 Since n is quantized as n = 2n/L, (x, t) is periodic in x (x, t) = (x + L, t).

1

where

(4.2.5)

This can be seen more clearly from the phase-space Lagrangian of the oscillator: L = Pn Xn

The form of the Lagrangian remains the same under the rescaling. In particular, the invariance of the rst term Pn Xn indicates that the rescaled Xn and Pn remains to be a coordinate-momentum pair.

2 1 K n X n . 2

1 2Mn

2 Pn

59

Using (x, t) we can rewrite the total energy of the oscillators as Etot =
n

En =
n

L n n

=
0

dx (x) (ix ) (x)

(4.2.6)

and rewrite the equation of motion as it n (t) = (n ) n (t) or it (x, t) = (ix ) (x, t) which has a form of Schrdinger equation! In particular, for non-relativistic particles, (k) = o and the eqn (4.2.7) becomes the standard Schrdinger equation o it (x, t) = 1 2 +U 2m x (x, t) (4.2.7)
k2 2m +U

(4.2.8)

The expression of the energy (4.2.6) and the equation of motion (4.2.7) provide a complete description of the oscillators as a classical system. To show that such a classical system is actually a vibrating string, we introduce two real elds h(x, t) and p(x, t) which are proportional to the real and the imaginary parts of (x, t): (x, t) = h(x, t) + ip(x, t) 2 (4.2.9)

In terms of h(x, t) and p(x, t) the equation of motion (4.2.7) becomes h(x, t) = (ix ) p(x, t) p(x, t) = (ix ) h(x, t) which can be rewritten as For relativistic bosons, (ix ) = h(x, t) = 2 (ix )h(x, t)
2 c2 x + m2 c4 and the above equation reduces to

(4.2.10) (4.2.11)

2 h(x, t) = c2 x h(x, t) m2 c4 h(x, t)

(4.2.12)

If we view h(x, t) as the vibration amplitude of the string then eqn (4.2.12) is the standard wave equation for an ordinary vibrating string. Since h(x, t) is periodic in x h(x, t) = h(x + L, t), the string form a loop of length L. We see that a quantum many-boson system in 1D is related to a vibrating string. Such a system of vibrating string is also called a classical eld theory (in one dimension) where h or is the eld. From the order of the time derivative, we nd that eqn (4.2.11) is a coordinate-space equation of motion while eqn (4.2.7) is a phase-space equation of motion. If we quantize the classical vibrating string (i.e. quantize the oscillators from the vibration modes of the string), then the quantized vibrating string is equivalent to the 1D quantum many-boson system. The above result is quite amazing. The quantum theory of a vibrating string described by eqn (4.2.11) is the same as the quantum theory of bosons described by eqn (4.1.5)! So instead of using the picture Fig. 4.4a to describe a quantum boson gas, we can also use the picture Fig. 4.4b to describe the same quantum boson gas. The picture Fig. 4.4b represents a eld theory description 60

(a)

(b)

Figure 4.4: (a) A boson gas in the particle picture and (b) the same boson gas in the vibratingstring picture. One may wonder which picture represents the truth or the reality. Is a boson gas a collection of particles or a collections of vibrations?
k k k k

(a)

(b)

(c)

Figure 4.5: Three pictures of a two-boson state where both bosons have the same momentum k = 2 2 . (a) The particle picture, (b) the occupation picture, and (c) the vibrating-string picture. L We know that for an oscillator, not all amplitudes of oscillation are allowed in quantum theory. In (c), dierent curves represent dierent allowed quantized vibration amplitude of a particular vibration mode. The solid curve represents the quantized vibration that correspond to two bosons. The momentum of the bosons is determined by the wave length of the vibration mode. of the boson gas. Dierent many-boson states correspond to dierent quantized vibrating states of the string (see Fig. 4.5). The vibrating-string picture of the boson gas only works in 1D. In 2D, we need to replace string by membrane: a boson gas in 2D can also be regarded a vibrating membrane. Similarly in d-dimensions, a boson gas is equivalent to a vibrating d-brane (a d-dimensional membrane).
Problem 4.2.1 Following the example of Fig. 4.5, draw three pictures for a two-boson state with one boson carrying momentum k = 4/L and the other k = 6/L.

4.2.2

Classical eld theory of free bosons

Phase-space Lagrangian for the classical eld theory that describes free bosons. Generalizing the results eqn (4.2.8) and eqn (4.2.6) in the last section to d-dimensions, we nd that free non-relativistic bosons in d-dimensions described by the Hamiltonian (4.1.6) is related to a d-dimensional eld theory described by the following phase-space equation of motion it (x, t) = and the total energy Etot = dd x 1 2 +U 2m x , (4.2.14) 1 2 +U 2m x (x, t) (4.2.13)

The d-dimensional eld theory describes a vibrating d-brane with the complex eld which encodes both the amplitude and the velocity of the vibration (see eqn (4.2.9) and eqn (4.2.10)).2
2

The classical eld theory (the vibrating brane) is also described by a coordinate-space equation of motion h=


1 2 x + U 2m

2

61

Both the phase-space equation of motion (4.2.13) and the total energy (4.2.14) can be derived from the following phase-space Lagrangian L= dd x (i t Etot ) = dd x i t 1 2 +U 2m x . (4.2.15)

To see how the equation of motion (4.2.13) is derived from the Lagrangian (4.2.15), let us consider a variation of the action S = dt L: S[ + ] S[] = dtdd x it 1 2 +U 2m x + c.c. + O(2 )

For stationary path , the variation S[ + ] S[] contains no linear order term in . Thus stationary path must satises eqn (4.2.13). This is how the wave equation eqn (4.2.13) is derived from the Lagrangian (4.2.15). We note that the potential U in the classical energy (4.2.14) or in the phase-space Lagrangian (4.2.15) may have a spatial dependence U = U (x). The resulting theory will describe bosons moving in a non-uniform potential (see problem 4.2.2). In particular, a hard wall in a certain region can be realized by an innite potential U (x) in that region. In order for the total energy (4.2.14) to be nite, an innite potential in a region will force (x) to be zero in that region. Thus a hard wall can be represented by the boundary condition (x) = 0 or h(x) = 0. This result will be used in section 4.2.5.
1 2 We also note that although the total energy Etot is always real, the integrand 2m x + U in eqn (4.2.14) is in general complex and cannot be interpreted as an energy density. To obtain the energy density, we rewrite (4.2.14) as

Etot =

dd x

1 |x |2 + U ||2 , 2m

(4.2.16)

through integration by part (assuming (x) 0 as x ). We nd the energy density to be Etot = 1 |x |2 + U ||2 . 2m (4.2.17)

which is always real. Similarly, the Lagrangian density can also be chosen to be real and has a form 1 1 L = (i t it ) |x |2 U ||2 . 2 2m (4.2.18)

Problem 4.2.2 Field theory for bosons in non-uniform potential and magnetic eld: Non-interacting bosons in a non-uniform potential and a non-zero magnetic eld is described by Hamiltonian
n

Hn =
a=1

1 [x iAi (xa )]2 + U (x) 2m a

(4.2.19)

where Ai (x) is the vector potential of the magnetic eld. (a) Show that the above many-boson system can be described by a collection of harmonic oscillators. Determine the oscillation frequencies of those oscillators.
where h is the amplitude of the vibration and a Lagrangian
Z    

L=

dd x

1 1 1 1 2 h 1 2 h h x + U 2 2m x + U 2 2m

h .

62

(b) Show that the collection of the oscillators can be described by a eld theory with a phase-space Lagrangian density 1 1 L = (i (t + i U ) i(t + i U ) ) |(x i Ai )|2 . (4.2.20) 2 2m (c) Show that the above phase-space Lagrangian density is invariant under the gauge transformation U U t , Ai Ai + i , e i .

4.2.3

Physical quantities in the eld theory description

How to express physical quantities in terms of the boson eld . After obtain the eld theory of bosons, we calculate the dynamics (i.e. the time evolution) of the boson eld . But this not what we really want. What we want is the dynamics of measurable quantities, such a the density of the bosons. In the section, we will discuss how to express some physical quantities in term of the boson eld. This allows us to obtain the dynamics of physical quantities from the dynamics of the boson eld. From the phase-space Lagrangian, we see that the total energy of the bosons is given by Etot = dd x 1 |x |2 + U ||2 2m

According to Prob. 4.2.2, the above expression is correct even for non-uniform potential U (x). We know that if we change the potential by a constant U , the total energy for the bosons will change by Etot = U N , where N is the total number of the bosons. From the expression of Etot we see that N can be expressed in term of the boson eld : N = dd x ||2 . This suggests that the boson density is (x) = |(x)|2 . (4.2.21) The key to obtain the above result is to realize that the terms that couples U (x) in the expression of the energy is the boson density. This is a general method to obtain physical quantities. To obtain the expression a certain quantities, we may turn on a certain eld (such as the potential U ) that couple to the quantity (such as the boson density ). Then we calculate the total energy in the presence of the eld. By identifying the term that couple to the eld in the expression of the energy, we can obtain the express for desired physical quantity. As an application of the above method, let us nd the expression for the current density of the k bosons. We know that the velocity of a boson is given by v = (k ) . So the current density of a k boson with momentum k is j = v = V 1 (k ) , where V is the volum of space. Here we treat the boson as a plane wave and hence its density is = 1/V. When we have many bosons, the total boson current density is given by N (ki ) j = V 1 k
i=1

where ki is the momentum of the

ith

boson.

To obtain a eld that couples to the current density, let us consider a boson system with a dispersion (k + a). The total energy is given by Etot = N (ki + a). For small a, we see that i=1
N N

Etot =
i=1

(ki ) +
i=1

(ki ) a= k 63

(ki ) +
i=1

dd x a j

Thus a is the eld that couple the current density. For bosons with a dispersion (k +a), the total energy of the corresponding classical eld theory is given by (see eqn (4.2.6)) Etot = dd x (ix + a) = dd x |
1/2

(ix + a)|2

Here we write Etot in a form in which the integrand is real. Expand Etot to the linear order of a, we get Etot = dd x | + = where u(k) =
1/2

(ix )|2
1/2

dd x [
1/2

(ix )] [u(ix )] + [u(ix )] [ dd x Re[ v(ix )] a

1/2

(ix )] a

dd x |

(ix )|2 +
1/2 (k)

= 1/2 (k)v(k), k We see that the boson current density is given by j = Re[ v(ix )]

v(k) =

(k) k

k k2 For bosons with dispersion (k) = 2m , we nd v(k) = m . So the boson current density becomes
j = Re[ i ix ] = [(x ) x ] m 2m (4.2.22)

A more general way to obtain the boson current density is to start with the Lagrangian density (4.2.20) for bosons in magnetic eld (assuming that bosons charry charge 1). The current density is then simply given by ji = L Ai

When A = 0, the above reproduces eqn (4.2.22). So more generally, the vector potential is the eld that couple to the current density. Problem 4.2.3 Show that the dynamics of the boson eld is such that the boson density and the boson current density satisfy k2 the continuity equation + x j = 0. (You may assume (k) = 2m .)

Problem 4.2.4 Find an expression of the total momentum and the momentum density of free bosons in term of the boson eld k2 . (You may assume (k) = 2m .)

4.2.4

*The second quantized description of free bosons

Quantization of the vibrating string. A Hamiltonian without xing the number of bosons. Boson creation and annihilation operators. 64

The vibrating string discussed in the section 4.2.1 is a classical theory. Such a classical theory does not directly describe the quantum boson gas. Only quantized vibrating string describe the quantum bosons. To quantize the vibrating string, we note that a vibrating string is a collection of oscillators described by (Xn , Pn ). Since (Xn , Pn ) is a canonical coordinate-momentum pair, a quantized theory can be obtained by replacing (Xn , Pn ) by a pair of operators (Xn , Pn ) that satisfy [Xn , Pn ] = i So after quantization, the classical modes n and the classical eld (x) all become operators: 1 n n = (Xn + i Pn ) 2 (x) (x) = L1/2 e i xn
n

(4.2.23)

One can check that n and (x) satisfy the following algebra [n , m ] = n m , and [(x), (x )] = (x x ), [n , m ] = 0. [(x), (x )] = 0, (4.2.24) (4.2.25)

The Hamiltonian of the quantized vibrating string can be obtained from the classical energy 2 2 Etot = n 1 (n )(Pn + Xn ) (see eqn (4.2.6)) after replacing Xn and Pn by the corresponding 2 operators: H0 =
n

1 2 2 (n )(Pn + Xn ) 2 1 (n )( n n + ). 2 (4.2.26)

=
n

Since n n with dierent n commute with each other, the eigenstates of H0 are common eigen states of n n . Let nn be the eigenvalues of n n , then the eigenstates of H0 has a form 1 | n1 n0 n1 n2 which has an eigenvalue n (nn + 2 ) (n ). So the above Hamiltonian can also be regarded as the Hamiltonian of the free quantum many-boson system if we interpret nn as the boson occupation number. Such a description of the bosons is called the second quantized description. In contrast, the description of n-boson states using wave functions of n variables (x1 , , xn ) and the n-particle Hamiltonian (4.1.5) is called the rst quantized description. From eqn (4.2.24) and eqn (4.2.26), we see that n and n are the raising and the lowing operators of the oscillator n . Since the eigenvalues of n n correspond to occupation numbers, the total boson number operator is given by N=
n

n n =

dx (x)(x)

Thus we may interpret (x)(x) as the boson number-density operator. One can also show that [N , ] = , [N , ] = .

where is (x) or n . Thus increases the boson number by one while decreases the boson the creation operator and the annihilation operator. number by one. For this reason we also call In contrast to the Hamiltonian (4.1.5) in the particle picture which only acts on n-particle states, the Hamiltonian (4.2.26) in the second quantized description acts on states with any numbers of bosons. 65

Problem 4.2.5 (a) Derive eqn (4.2.24) and eqn (4.2.25). (b) Show that [(x ), (x)] = (x x ) (x)

(4.2.27)

where (x) = (x)(x) is the boson density operator. (c) Show that when acts on the no boson state, (x) produces an eigenstate of (x )s with eigenvalues (xx ). That is (x )|x = (x x )|x , |x (x)| 000 where | 000 is the state with all nn = 0. Thus |x is the state with one boson at x, and (x) creates a boson at x. (d) Show that |x1 x2 xN = (x1 ) (x2 ) (xN )| 000 is an N -boson state with bosons at x1 , x2 , etc.

Problem 4.2.6 Find the N -particle wave function (x1 , , xN ) = x1 x2 xN | for a state | which has N bosons with momentum k and no bosons with other momenta. Find the N -particle wave function of the bosons condensed state.

4.2.5

Vacuum as a dynamical medium and the Casimir eect

The dierences between the two views of bosons: the particle picture and the vibrating-string picture. Vacuum is not empty. It is a dynamical medium just like any materials encountered in condensed matter physics. We have discussed two ways to view a many-boson system: the particle picture and the vibrating-brane picture (see Fig. 4.4). The essence of the particle picture is the assumption that the vacuum is empty. In this picture, bosons are things placed in the empty vacuum. However, in the vibrating-brane picture, the vacuum is regarded as a dynamical non-empty medium. The vibration of such a medium gives rise to bosons in the vacuum. In the last a few sections, we have stressed the mathematical equivalence of the two pictures. However, the two pictures are not really equivalent. First, the vibrating-brane picture (or the eld theory picture) provides an origin and an explanation of identical particles. The particles arising from the vibrations of a d-brane are naturally and always identical bosons. Actually, it is impossible to obtain non-identical particles from the vibrations of the brane. In contrast, particles do not have to be identical particles in the particle picture. In this case, the existence and the appearance of identical particles is very mysterious. We can also turn our argument around. The very existence of identical particles in our vacuum suggests that we should not view particles as things placed in the vacuum. We should instead view our vacuum as a 3-brane and the particles as the vibrations of the 3-brane. Our vacuum is not empty. It is a dynamical medium whose collective motion give rise to elementary particles. So our vacuum just like a material studied on condensed matter physics. The theory of elementary particles is actually a theory about one material the vacuum material.
Here we would like to remark that the vibrating-brane only reproduce a particular kind of identical particles scalar bosons. The photons in our vacuum are actually vector bosons (or gauge bosons) due to its two polarizations and electrons are fermions. Those particles cannot arise from vibrating 3-branes. So if we view vacuum as a 3brane, then the vibrations of the 3-brane will not give rise to observed light waves. However, as we will see later in this book, the above philosophy of viewing vacuum as a dynamical medium is still correct. However, the vacuum

66

lP

Figure 4.6: According to quantum gravity, a continuous string cannot be a physical reality. A real string may look more like a sequence discrete beads.
has a more complicated internal structure than the simple 3-brane. It is this more complicated internal structure that makes the vibrations in the medium behave like photons, electrons, gluons, quarks,[Wen 2002, 2003b] and possibly all other elementary particles observed in our vacuum. In this chapter, for simplicity, we treat photons as massless scalar bosons and regard them as vibrations of a 3-brane.

The above discussion about the vibrating-brane picture and the particle picture sounds philosophical. Actually, the vibrating-brane picture and the particle picture has a measurable dierence. The vibrating-brane picture produces Casimir eect while the particle picture does not.[Casimir 1948] The Casimir eect has been observed in our vacuum, conrming that the vibrating-brane picture is a correct picture while the particle picture is an incorrect picture for bosons. To understand the Casimir eect, we start with the energy of vacuum. In the particle picture, the vacuum is just an empty stage or a reference point. We naturally assign a zero energy to it. In contrast, the vacuum energy in the vibrating-brane picture is naturally non-zero and is given by 1 the zero-energies 2 k = 1 (k) of the oscillators. For 1D free boson system, the vacuum energy is 2 given by 1 Uvac = (n ) 2
n

where (k) is the dispersion of the bosons. One may say that Uvac is just a constant term in the total energy which dene the reference point of the energy. Such a constant term has no measurable eect. Indeed, Uvac cannot be measured directly. However, if we change the dispersion (k) and/or change the distribution of quantized momentum n , then the change in Uvac has physical eects and can be measured. Before discussing how to calculate the change of Uvac , let us calculate Uvac itself. At rst sight, the calculation of Uvac appear to be very simple since Uvac = 1 n (n ) = + due to the innite 2 many positive (n )s that appear in the sum n . Certainly, such a simple result of innity is meaningless. The innite Uvac is the famous innity problem that plague all forms of eld theories. It is so annoying that it once led people to abandon eld theories. One can use this innity problem to argue that the vibrating-brane picture for bosons is wrong. Actually, the problem of the innity is not the problem of the vibrating-brane picture but the problem of regarding space (or the vibrating-brane) as a continuous manifold. As has been argued in section 2.2, continuous manifold simply does not exist in our universe. It is meaningless and impossible to have two points separated by a distance less the Planck length lP . Similarly, a wave vector larger than 1/lP is also meaningless. At moment we have no clue about what structure replaces manifold at short distance. So we do not know what the string becomes at length scale less than the Planck scale lP . But at least, it is more correct to view the string as a sequence discrete beads (see Fig. 4.6). Our vibrating-brane picture is meaningful only for wave vectors less then 1/lP . The wave-vector (or the momentum) scale is called the cut-o scale. So we should limit the wave vector summation n to over the meaningful wave vectors |n | < . Therefore, the vacuum energy is really nite Uvac =
n =0,2/L,4/L, ,

1 (n ) 2

67

(a)

(b)

Figure 4.7: (a) A boson gas between two hard walls. (b) The same boson gas is described by a vibrating string with boundary condition h(0) = h(L) = 0.
L

Figure 4.8: The vibrating string with h(0) = h(a) = h(L) = 0 describes a boson gas between three hard walls. Certainly, we are not sure that the energy levels should suddenly disappear for k above as implied by the above formula. We may very well have a softer cut-o Uvac =
n =0,2/L,4/L,

1 (n )e|n |/ 2

But which one is the correct way to calculate Uvac ? The physics at short distance is still unknown to us (since we still do not have a theory of quantum gravity). So it is not clear what is the correct way to cut-o the wave vector summation n . However, we will see later that the low energy and long distance eects do not depend on how we cut-o the momentum summation. This allows us to make prediction without a complete understanding of the theory. The ignorance of the short distance physics does not prevent us from gaining a (partial) understanding of long distance physics. Let us consider a 1D boson system between two hard walls (see Fig. 4.7a). One wall is at x = 0 and the other at x = L. Such a boson system is described by a string (4.2.12) that satisfy the boundary condition h(0) = h(L) = 0 (see Fig. 4.7b).3 In contrast to the periodic boundary condition h(0) = h(L), the string vibration modes for the hard-wall case has a form h(x) sin(nx/L) and are labeled by n = n/L, n = 0, 1, 2, . In this case, the vacuum energy is dierent from that for periodic boundary condition
hw Uvac (L) = n =0,/L,2/L,

1 ( n )e|n |/ 2

hw For massless relativistic particles (such as photons), (k) = c|k|. Uvac (L) can be calculated easily. We nd c 2 L c hw Uvac (L) = = + O(2 ) (4.2.28) 2 c 4c 24L 8Lsinh ( 2L )

Now let us add the third hard walls at x = a (see Fig. 4.8). The third hard wall modies the distribution of the vibration modes which changes the vacuum energy. Using the result (4.2.28),
3

The connection between the hard wall and the boundary condition h(0) = h(L) = 0 is discussed in section 4.2.2.

68

we nd the modied vacuum energy to be

hw Uvac (a, L) =

2 a c 2 (L a) c + 4c 24a 4c 24(L a) 2L c c = 4c 24(L a) 24a

When L is very large (i.e. L a), we nd that the total vacuum energy depends on the separation c hw between the two walls at x = 0 and a: Uvac (a, L) = 24a +Const. This causes an attractive force between the two walls separated by a distance a c 24a2 This force between two walls in vacuum is called the Casimir eect.[Casimir 1948] It is interesting to note that the force does not depend on the cut-o. F = The above result is for massless bosons in 1D. For massless photons in 3D, the attractive force between two metallic or superconducting plates separated by a is 2 c A 240a4 where A is the area of the plates. Such a force was measured by Spamaay, 1958 and Lamoreaux, 1997, indicating that our vacuum is really a dynamical medium. The study of our vacuum and its elementary particles is really a material science. F =

4.2.6

*The second quantized description of interacting bosons

The many-body Hamiltonian that describes interacting bosons. The boson density operator. Now let us turn to a more complicated problem of interacting bosons. Let V (x) be the potential energy of two bosons separated by x. The Hamiltonian that describe n interacting bosons in ddimension is 1 2 Hn = ( + U) + V (xi xj ) (4.2.29) 2m xi
i i<j

We know that the free bosons have a second quantized description (4.2.26) which describes 0boson, 1-boson, 2-boson, and general n-boson cases with a single Hamiltonian. What is the second quantized description (or the oscillator description) of the interacting bosons? Again the second quantized description is written in terms of the boson creation/annihilation operators and . To obtain the form of the second quantized Hamiltonian for interacting bosons, we rst note that the total potential energy can be rewritten in terms of the boson density V (xi xj ) =
i<j

dd xdd x

1 (x)V (x x )(x ) 2

Since the boson density operator is given by (x) = (x)(x) as discussed in section 4.2.4, the interaction Hamiltonian is given by Hint = = 1 2V dd xdd x 1 (x)V (x x )(x ) 2

q,k,k

+q k Vq q k k k

69

where Vq =

dd x e i qx V (x)

Here we have assumed that our system is a d-dimensional cube of volume V = Ld . The wave vectors q, k, k are quantized, i.e. their components are 2/L times integers. The summation q,k,k sums over those quantize wave vectors. k is given by (x) =

k
and satises the algebra [k , ] = kk , k

V 1/2 e i kx k

(4.2.30)

[k , k ] = 0.

Putting Hint and H0 in eqn (4.2.26) together, we nd the following Hamiltonian H= 1 1 (k)( k + ) + k 2 2V +q k Vq q k k k (4.2.31)

q,k,k

that describes the interacting bosons. Naively, one expects (k) describes the boson dispersion and k2 should be taken to be (k) = 2m + U . As we will see below, this naive expectation is incorrect. We k2 need to choose a dierent (k) in order to reproduce the single-boson dispersion 2m + U . To determine the proper form of (k), let us discuss a few known eigenstates of H in eqn (4.2.31). is dened by the algebraic relation One eigenstate |0 of H k |0 = 0 (4.2.32)

Such a state is an eigenstate of the boson number operator with zero eigenvalue N |0 = 0. Thus 1 |0 is the vacuum state with no bosons. The energy of |0 is E0 = k 2 (k). Another class of the eigenstates is given by |k = |0 k To show |k is an eigenstate, we commute through in H using the commutation relation . Such a procedure is called normal ordering. This allows us (4.2.30) to put all a to the right of a to rewrite H as H=

k
where (k) is given by

1 1 ( (k) k + (k)) + k 2 2V

q,k,k
V (0) 2

Vq +q q k k k k

(4.2.33)

(k) = (k) +

(4.2.34)

and V (0) is V (x) at x = 0. Using eqn (4.2.33), we easily see that |k is an eigenstate of H with eigenvalue Ek = (k) + E0 . |k is a state with one boson that carries a momentum k. The excitation energy of such a one-boson state is Ek E0 = (k). We see that the single-boson k|2 dispersion is given by (k). To reproduce the dispersion (k) = |2m + U , we need to choose (k) to 2 be (k) = |k| + U V (0) .
2m 2

A generic state described by a collection of boson occupation numbers {nk } can be created by the boson creation operators from the vacuum state |0 : k |{nk }

k
70

( )nk |0 k

 Such a state is an eigenstate of the total boson number operator N = k k with eigenvalue k k nk . The state |{nk } also carry a denite total momentum P = k knk . However, for interacting bosons the state |{nk } is not an energy eigenstate. In general, it is hard to calculate energy eigenstates. 2 To summarize, if we choose (k) = k + U , eqn (4.2.33) and eqn (4.2.29) will describe the same interacting boson system! Since H is not quadratic in k and , H describes a collection k of anharmonic quantum oscillators. So interacting bosons are described by anharmonic oscillators. k2 When (k) = 2m + U , eqn (4.2.33) can be written in a more compact form H= dd x (x) x + U (x) 2m 1 + dd xdd x V (x x ) (x) (x )(x )(x) 2
2 2m

(4.2.35)

where (x) satises the following algebra [(x), (x )] = (x x ), [(x), (x )] = 0,

Problem 4.2.7 Show that [N , H] = 0. So the total boson number is conserved and the Hamiltonian is invariant under the U (1) transformation generated by N : H = e i N H e i N .

Problem 4.2.8 Find the total momentum operator P of the bosons in terms of k . Show that [P , H] = 0 and hence the total momentum is conserved. (Hint: the total momentum operator P does not depend on the interaction between the bosons and is the same for free bosons and interacting bosons.)

Problem 4.2.9 It is too hard to nd the eigenstates and eigenvalues of H (4.2.33) in the 2-boson sector. Here we simplify the problem by limiting ourselves to 1D and consider only three k-levels: k = 0, 2/L. Find the eigenstates and the eigenvalues of H in the 2-boson sector with the above simplication.

Problem 4.2.10 Derive eqn (4.2.33).

4.2.7

Classical eld theory of interacting bosons

A classical picture for interacting bosons. The free boson system (4.1.6) is easy to solve. We do not need a classical vibrating brane picture to understand and to visualize the behavior of free bosons. However, an interacting boson system described by 1 2 Hn = ( + U) + V (xi xj ) (4.2.36) 2m xi
i i<j

(or eqn (4.2.33)) is entirely a dierent matter. The interacting Hamiltonian eqn (4.2.36) is so hard to solve that we have no clue what do the low energy eigenstates and the eigenvalues look like. So 71

how can we understand the properties of the interacting bosons without being able to diagonalize the Hamiltonian eqn (4.2.36)? One way to understand interacting bosons is to nd the corresponding classical system and study its low energy collective motions. Then we can quantize those low energy classical motions to obtain low energy quantum properties. This way, we can gain a understanding of the low energy properties of the quantum interacting boson system. To obtain the corresponding classical system for interacting bosons, we note that the boson density is given by (x) = (x)(x). The total interaction potential energy can be rewritten in terms of the boson density V (xi xj ) =
i<j

dd xdd x

1 (x)V (x x )(x ) 2

This allows us to guess that the classical eld theory for interacting bosons to have a modied total energy Etot = 2 dd x x + U 2m + dd xdd x |(x)|2 V (x x ) |(x )|2 2 (4.2.37)

and hence a modied Lagrangian L= dd x i t Etot = dd x 1 2 +U 2m x V (x x ) dd xdd x |(x)|2 |(x )|2 2 i t

(4.2.38)

The corresponding equation of motion can be obtained from L and is given by it (x, t) = 1 2 + Ue (x) (x, t) 2m x dd x V (x x )|(x )|2 (4.2.39)

Ue (x) = U +

We note that eqn (4.2.39) is a non-linear equation. So the interacting bosons are described by a non-harmonic eld theory.
Problem 4.2.11 Symmetry and conservation: (a) Derive the equation of motion (4.2.39) from the Lagrangian (4.2.38). (b) The Lagrangian (4.2.38) is invariant under a U (1) transformation e i . Show that the particle number d is conserved for such a system, i.e. dt dd x ||2 = 0. d (c) We include a term d x g( + ) in the Lagrangian to break the U (1) invariance. Show that the particle number is no longer conserved for the new system.

Problem 4.2.12 In this section we have guessed the classical eld theory (4.2.38) or (4.2.39) for the interaction bosons. In fact the classical wave equation (4.2.39) can be derived using the equation-of-motion approach described in section 2.6.1 from the second quantized Hamiltonian (4.2.35). Find the operator equation of motion for (x, t) e i Ht (x) e i Ht using t (x, t) = i[H, (x, t)]. Replace and by = and = to obtain the corresponding classical equation of motion.

72

4.2.8

*A coherent-state approach

A state in the classical eld theory is described by a complex function (x). A classical state described by (x) corresponds to a quantum boson state |(x) (4.2.41). The average energy of the quantum boson state |(x) is exactly the total energy (4.2.37) of the classical state (x). Since the classical eld theory is equivalent to a collection of (anharmonic) oscillators, a classical state of the eld theory is described by the coordinates and momenta of the oscillators (Xk , Pk ). Due to the relation between (Xk , Pk ) and the complex eld (x) (see eqn (4.2.5) and eqn (4.2.4) for an 1D example), the classical state of the eld theory can also be described by the complex eld (x). That is each complex eld (x) correspond to a classical state. Since the eld theory describes the boson gas, the complex function should also correspond to a state of the bosons. But the quantum states of bosons are described by |{nk } (see section 4.2.6) or their superpositions. How to relate the above two seemingly very dierent descriptions of the boson states? From eqn (4.2.23), we see that the complex eld (x) correspond to quantum operators (x). Since (x) at dierent x commute with each other, we can have a common eigenstate of the operators (x) with (x) as the eigenvalues: (x)|(x) = (x)|(x) . (4.2.40)

Since (x) is not hermitian, the eigenvalues (x) are in general complex. We may say the quantum state |(x) is the state where the classical eld is equal to (x). So it is natural to relate the quantum state |(x) to the classical state described by the complex function (x). The quantum state |(x) can be constructed explicitly (see problem 4.2.13) |(x) = e
R

dd x (x) (x) |0

N [(x)]

N [(x)] = e

dd x |(x)|2

(4.2.41)

where |0 is the vacuum state with no bosons (see eqn (4.2.32)). Such type of states are called coherent states. The coherent states are particular superpositions of |{nk } . We may view the state |(x) as a variational state with the complex function (x) as the variational parameters. An approximate ground state and the ground state energy of the interacting bosons can be obtained by minimizing the average energy H = (x)|H|(x) . Using eqn (4.2.40) easily. We nd that H is exactly the classical total energy and eqn (4.2.35), we can calculate H (4.2.37) for the classical state (x)! So the calculation of the classical ground state is the same as the calculation of the variational parameters that minimize the average energy H . The classical ground state described by (x) = grnd correspond to the quantum variational state |grnd = N 1/2 [grnd ]e
Vgrnd 0

|0

(4.2.42)

where is the boson creation operator k at the k = 0 (see eqn (4.2.30)) and V the volume of the 0 boson system. It is clear that only the occupation on the k = 0 level is non zero and nk=0 = grnd | 0 |grnd = V grnd grnd 0 So the coherent state |grnd is a boson condensed state with (on average) V|grnd |2 = VU/V bosons in the k = 0 level. 73

For (x) close to ground state (x) = grnd + (x), the quantum state |(x) describes the low energy excitations above the ground state |grnd . As in section 2.6.2, we treat the quantum uctuations |(x) as classical waves described by the complex eld (x). The dynamics of the classical waves is determined by the Lagrangian (see eqn (2.6.12)) L(t (x, t), (x, t)) = (x, t)|i d H|(x, t) dt (4.2.43)

where H is given in eqn (4.2.35). We nd, up to a total time derivative term the Lagrangian derived from the coherent state is exactly the classical Lagrangian (4.2.38) that we guessed in section 4.2.7.
Problem 4.2.13 Coherent states and phase-space Lagrangian: (a) Show that P |{k } e k k k |0 is an common eigenstate of k s with eigenvalues k . (b) Show that R d R d R 0| e d x 1 (x)(x) e d x 2 (x) (x) |0 = e
A B [A,B] B A

dd x (x)2 (x) 1

(4.2.44)

(Hint: e e = e e e is [A, B] commute with A and B.) (c) Show that |(x) in eqn (4.2.41) is a normalized common eigenstate of (x)s with the eigenvalues (x). (d) Show that Lagrangian (4.2.43) is given by eqn (4.2.38). (Hint: you may want to use eqn (4.2.44).)

Problem 4.2.14 The coherent state |grnd is not an eigenstate of the total boson number operator N = k k . Calculate the k (N N )2 . Compare your result with the number quantum uctuation of the total boson number N uctuation of a typical thermal dynamical system with a xed chemical potential where N N .

4.3
4.3.1

Phases and phase transitions of interacting bosons

A quantum phase transition in interacting boson system

The classical ground state is obtained by minimizing the total energy. Singularities in the ground state energy as we change the parameters of the system implies quantum phase transitions. Instead of directly studying the very dicult quantum interacting boson system (4.2.36), in the next a few sections, we will study the corresponding classical eld theory (or the anharmonic brane) described by eqn (4.2.38). The classical eld theory is much easier to deal with. The physical properties of the classical eld theory will give us some good ideas about the physical properties of the quantum interacting bosons. We start with a classical state described by a complex eld (x). The energy of such a classical state is given by eqn (4.2.37). So looking for a classical ground state is equivalent to looking for a complex function that minimize the total energy (4.2.37). We note that a constant function minimize the kinetic energy term dd x x = dd x x x . 2m 2m So if the interaction V is not too large, we may assume the function that minimize the total energy is a constant. In this case the energy density becomes 1 E = U ||2 + V ||4 2 74
2

 where V dd x V (x). When V < 0, we note that the total energy Etot = dd x E is not bound from below and has no minimum. So the ground state of boson gas with attractive interaction is not well dened.4 When V > 0, we nd that the total energy is minimized by the following eld (or the classical state) grnd = 0, U/V , U >0 U <0 (4.3.1)

This eld corresponds to the classical ground state of the system. The ground state energy density is 0, U >0 E= (4.3.2) 1 2 2 U /V , U < 0 Remember that U is the potential experienced by a boson and V characterize the interaction between two bosons. Eqn (4.3.2) tells us that how the ground state energy density depends on those parameters. Because the ground state is the state at zero temperature, how ground state energy depends on those parameters will tell us if there are quantum phase transitions5 or not. As we change the parameters that characterize a system, if the ground state energy changes smoothly, we will say that there is no quantum phase transition. If the ground state energy encounter a singularity, the singularity will represent a quantum phase transition. From eqn (4.3.2) we see that there is a quantum phase transition at U = 0 if V > 0. Since it is the second order derivative 2 E/U 2 that has a discontinuity, the phase transition is a second order phase transition. The phase for U > 0 contains no bosons since it costs energy have a boson. The phase for U < 0 has a non-zero boson density = U/V . This is because for negative U the system can lower its energy by having more bosons. But if there are too many bosons, there will be a large cost of interaction energy due to the repulsive interaction between the bosons. So the density = U/V is a balance . Later we will see that between the potential energy due to U and the interaction energy due to V the phase with non-zero is a superuid phase. It is also called the boson condensed phase. We would like to remark that the eqn (4.3.2) is only the ground state energy density of the classical eld theory. It is an approximation of the real ground state energy density of the interaction boson system. So we are not sure if the real ground state energy density contains a singularity or not. Even if the singularity does exist, we are not sure if it is the same type as described by eqn (4.3.2). A more careful study indicates that the real ground state energy density of the interaction boson system does have a singularity that is of the same type as in eqn (4.3.2) if the dimensions of the space is 2 or above. In 1D, the real ground state energy density has a singularity at U = 0 but the form of the singularity is dierent from that in eqn (4.3.2).
Problem 4.3.1 In Bose-Einstein condensation experiment of ultra cold atoms, the bosonic atoms are conned by a 3D non-uniform potential K U (x) = x2 . 2 Assume the atomic interaction potential to have a form V (x) = V (x) with V > 0. Find the boson density prole for the many-boson ground state for a positive chemical potential > 0. The exact result is hard to get. So nd the best approximate result.
Here we just pointed out a mathematical inconsistency of a particular mathematical description of an attractive boson gas. It is interesting to think physically what really going to happen to a chamber of boson gas with an attractive interaction? 5 A quantum phase transition, by denition, is a phase transition at zero temperature.
4

75

Ea a>0 EA (a) EB x

Ea a<0 a EA (b) x EB EB (c) EA

Figure 4.9: The energy function Ea (x) Ea,1,0,1 (x) = ax x2 + x4 with b = 1, c = 0, and d = 1. (a) Ea (x) for a > 0 and (b) for a < 0. EA and EB are the energies of the two local minima. The global minimum switch from one local minimum to the other as a passes 0. (c) The global minimum E0 (a) has a singularity at a = 0.

4.3.2

Continuous phase transition and symmetry

Two mechanisms for phase transitions. A continuous phase transition is a symmetry breaking transition. The concept of order parameter. We know that ground state energy (4.3.2) is obtained by minimizing the energy functional (4.2.37). U and V are the parameters in the energy functional. Can we have a more general and a deeper understanding when the minimum of an energy functional has a singular dependence on the parameters in the energy functional? Let us consider a simpler question: when the minimum of an energy function has a singular dependence on the parameters in the energy function? To be concrete, let us consider a real function parametrized by a, b, c, d (where d > 0): Eabcd (x) = ax + bx2 + cx3 + dx4 (4.3.3)

Let E0 (a, b, c, d) be the minimum of Eabcd (x). How can the minimum E0 (a, b, c, d) to have a singular dependence on a, b, c, d, knowing that the function Eabcd (x) itself has no singularity. One mechanism for generating singularity in E0 (a, b, c, d) is through the minima switching as shown in Fig. 4.9. When Eabcd (x) has multiple local minima, a singularity in the global minimum E0 (a, b, c, d) is generated when the global minimum switch from a local minimum to another. The singularities generated by minimum-switching always correspond to rst order phase transitions since the rst order derivative of of the ground state energy E0 is discontinuous at the singularities. When energy function has a symmetry, there can be another mechanism for generating singularities in the ground state energy. The energy function Eabcd (x) has a x x symmetry if a = c = 0. For such a symmetric energy function, the single minimum at the symmetric point x = 0 for positive b splits into two minima at x0 and x0 as b decreases below 0 (see Fig. 4.10). The shifting from the single minimum to one of the two minima generate the singularity in ground state energy E0 (b) at b = 0. We will call such a mechanism minimum-splitting. Minimum-splitting always generate continuous phase transitions. This is because the minima before and after the transition are connected continuously (see Fig. 4.11). We would like to stress that the x x symmetry in the energy function E0b0d (x) is crucial for the existence of the continuous transition caused by the minimum-slitting. Even a small symmetry breaking term, such as the ax term, will destroy the continuous transition by changing it into a smooth cross-over or a rst order phase transition.

76

b>0

Eb

b<0

Eb b

E0 (a)

x (b)

E0

E0 (c)

Figure 4.10: The energy function Eb (x) E0,b,0,1 (x) = bx2 + x4 with a = 0, c = 0, and d = 1 has a x x symmetry. (a) Eb (x) for b > 0 and (b) for b < 0. (c) The global minimum E0 (b) has a singularity at b = 0.
x

Figure 4.11: Trace of the positions of the minimum/maximum of Eb (x) as we vary b. The solid lines represent minima and the dash line represents the maximum. The single minimum for b > 0 splits continuously into two minima when b is lower below zero. The solid curve also shows how the order parameter x becomes non zero after the phase transition. When the energy function has a symmetry, one may expect that the minimum (i.e. the ground state) also has the symmetry. In our example Fig. 4.10, when b > 0 the ground state of Eb (given by x = 0) indeed has the x x symmetry (see Fig. 4.10a). However, after the phase transition (i.e. when b < 0), the new ground state no longer have the x x symmetry despite the energy function continues to have the same symmetry. Under the x x transformation, the new ground state is changed into another degenerate ground state (see Fig. 4.10b). This phenomenon of ground state having less symmetry than the energy function is called spontaneous symmetry breaking. Our example suggests that the continuous phase transition (caused by the minimum-splitting) always changes the symmetry of the ground state. So such a transition is also called symmetry breaking transition. Our simple example (4.3.3) reects a general phenomenon. The picture described above also applies to more general energy functional such as eqn (4.2.37). It is a deep insight by Landau [Landau 1937; Landau and Lifschitz 1958] that the singularity in the ground energy (or the free energy) is intimately related to the spontaneous symmetry breaking. This leads to a general theory of phase and phase transition based on symmetry and symmetry breaking. Within such a theory, we can introduce an order parameter to characterize dierent phases. The order parameter is a variable that transform non-trivially under the symmetry transformation. In our example Fig. 4.10, we may choose x or x3 as the order parameter, since they both change signs under x x transformation. In the symmetry unbroken phase, the order parameter x = 0. In the symmetry breaking phase, the order parameter is non-zero x = 0. The continuous phase transition is characterized by the order parameter acquiring a non-zero value. Landaus symmetry breaking theory is so general and so successful that for a long time it was believed that all continuous phase transitions are described by symmetry breaking. In our eld theory description of interacting bosons, the energy functional (4.2.37) has many symmetries, which include a U (1) symmetry e i and a translation symmetry x x + a. The phase = 0 (for U > 0) is invariant under both the transformations e i and x x + a. 77

E( )

Figure 4.12: The shape of the energy function (4.2.37) for a uniform when U < 0. The dot represents one ground state = U/V and the think circle represents the innitely many i U/V , 0 < < 2. degenerate ground states = e

supersolid

solid He

liquid HeI liquid HeII

He gas

Figure 4.13: Phases of 4 He. Thus the = 0 phase breaks no symmetries. The phase = U/V (for U < 0) is invariant under the translation x x + a but not the U (1) transformation e i . So the = 0 phase breaks the U (1) symmetry spontaneously. Under the U (1) transformation, = U/V is changed to = e i U/V which corresponds to one of the innitely many degenerate ground states (see Fig. 4.12). According to Landaus symmetry breaking theory, the two phases, = 0 and = 0, having dierent symmetries, must be separated by a phase transition. Fig. 4.13 describes phases of 4 He. We may use the symmetry breaking picture to understand those phases. The liquid He-I phase and the He gas phase do not break any symmetry. They have both the U (1) symmetry and the translation symmetry. They are actually the same phase. The liquid He-II phase breaks the U (1) symmetry of the liquid He-I phase. So the phase transition (the -line) between the liquid He-II and the liquid He-I phases is a continuous transition. The solid He phase breaks the translation symmetry but not the U (1) symmetry. As we we go from the solid He phase to the liquid He-II phase, the translation symmetry is restored and, at the same time, the U (1) symmetry is broken. A continuous symmetry breaking transition can only break or restore one symmetry at a time. So the transition between the solid He phase and the liquid He-II phase is not a continuous symmetry breaking transition. In fact the transition is a rst order transition. The solid He phase and the liquid He-II phase dier only by the translation symmetry. However, here the transition is actually rst order. Recently, a supersolid phase of 4 He is discovered. The 78

supersolid phase breaks both the U (1) and the translation symmetry. So the solid He and the supersolid phase dier by the U (1) symmetry. The transition between the two phases appears to be a continuous transition.
After understanding the above two mechanism which generate rst order phase transitions and continuous phase transitions, one may wonder is there a third mechanism for the singularity in the ground state energy? If you do nd the third mechanism, it will represent a new type of phase transitions beyond Landaus symmetry breaking theory! Problem 4.3.2 Show that if we include a term h dd x ( + ) in the energy functional (4.2.37) to explicitly break the U (1) symmetry,6 the continuous transition cause by changing U will change into a smooth cross-over no matter how small h is.

Problem 4.3.3 Adding h dd x ( + ) term in eqn (4.2.37) completely breaks the U (1) symmetry and destroys the continuous phase transition. Show that adding h2 dd x (2 + c.c.) term in eqn (4.2.37) breaks the U (1) symmetry down to a Z2 symmetry, i.e. the resulting energy functional is still invariant under . Study the phase and the phase transition in the resulting Z2 symmetric system. Show that the Z2 symmetry in the energy functional allows a continuous phase transition.

4.4
4.4.1

Low energy collective excitations

Collective modes sound waves

Small uctuations around the ground state have a wave-like dynamics. The uctuations around the symmetry breaking ground state have a linear dispersion for small k. Those uctuations are called sound waves. For our interacting boson system (4.2.37), the symmetric phase and the symmetry breaking phase have very dierent (classical) ground states grnd = 0 and grnd = U/V . As a result, the collective excitations above the ground states are also very dierent. The collective uctuations around the ground state are described = grnd . The equation of motion for that describes the classical dynamics of the uctuations can be obtain by substituting = + grnd into eqn (4.2.39). To simplify our calculation, we assume the interaction potential V (x x ) is short ranged and approximate it by V (x x ) = g(x x ). The wave equation (4.2.39) and the energy (4.2.37) are simplied to it (x, t) = and Etot = 1 2 + U + g|(x)|2 (x, t) 2m x (4.4.1)

2 g dd x x + U + ||2 . 2m 2

(4.4.2)

The Lagrangian (4.2.38) is simplied to L= dd x 2 g i t ( x + U ) ||4 . 2m 2 (4.4.3)

79

(a)
0 4 0

(b) k 4

Figure 4.14: (a) The dispersion of the collective modes (the sound waves) in the symmetry breaking state. (b) The dispersion of the collective modes in the liquid 4 He-II phase at T = 0. The dash-dot line is the dispersion of a 4 He atom. Let us rst discuss the equation of motion of in the symmetry breaking phase grnd = U/g for U < 0 (see eqn (4.3.1) and note that V = g). Substituting = + U/g into eqn (4.4.1), we nd i = 1 2 U 2m x U

Since we are interested in low lying uctuations, we can assume to be small. So in the above equation, we have only kept the terms linear in . Separating into real and imaginary part: 1 = 2 (h + ip), we rewrite the above equation as p = 1 2 2U 2m x 1 2 h = p 2m x h

1 2 From h = 2m x p, we nd

h =

1 2 1 2 x 2U 2m 2m x

(4.4.4)

We see that the equation of motion that describe the weak uctuations is a linear wave equation. The solutions of eqn (4.4.4) are of form h = C e i kx i k t + e i kx+ i k t . The dispersion relation of the wave is (See Fig. 4.14) k = |k|2 2m |k|2 2U 2m = |k|2 2m |k|2 + 2grnd g 2m (4.4.5)

where the grnd = |grnd |2 is the boson density in the ground state. For small k, we nd a linear dispersion ggrnd U k = v|k|, v= = (4.4.6) m m
Changing the symmetry of the energy functional (or the Hamiltonian) is called explicit symmetry breaking, which should not be confused with the spontaneous symmetry breaking. A spontaneous symmetry breaking refers a change in the symmetry of the ground state with the symmetry of the energy functional (or Hamiltonian) unchanged.
6

80

 k symmetry breaking

k 0 symmetric

k U

Figure 4.15: The dispersions of the low energy collective modes in the symmetry-breaking (superuid) phase , the symmetric phase, and at the phase transition point. where v is the velocity of the uctuating wave. Such a wave in the symmetry breaking phase is called the sound wave in the superuid. In the symmetric phase grnd = 0 for U > 0, the equation of motion of = grnd = is it = 1 2 +U 2m x

if we ignore the higher order terms in . The solutions have a form = C e i kx i k t . The k|2 k |2 dispersion relation of the wave is k = |2m + U . For the dispersion is k = |2m U . (Note that, strictly speaking, the uctuations are described by a pair (, ).) Combining the two, we get |k|2 k = +U (4.4.7) 2m Fig. ?? summarizes the dispersions of the low energy collective modes in the symmetry-breaking (superuid) phase and in the symmetric phase (only the positive-frequency branches are plotted). The dispersions of the low lying modes at the transition point can be obtained from eqn (4.4.5) or from eqn (4.4.7): k2 k = . 2m We see that there are more low energy modes at the transition point than those in the either phases on the two sides of the transition. All continuous quantum phase transitions have this property.

4.4.2

Quantized collective modes phonons

Waves = collection of oscillators. Quantized waves = quantized oscillators = free phonons. Phonons are a new type of bosons, completely dierent from the original interacting bosons that form the superuid. The emergence of phonons is the simplest example that completely new types of particles can emerge from collective uctuations. The interacting bosons are described anharmonic oscillators, which are very hard to solve. However, for small uctuations (or low lying excitations) around the ground state, the the anharmonic eects are small and can be ignored. So, those small uctuations are waves which correspond to a collection of harmonic oscillators.7
7

We will show explicitly how a wave is related to a collection of oscillators in section 4.4.4.

81

To specify the collection of the oscillator more concretely, let us assume the space to be a d-dimensional cube of volume V = Ld . Then the wave vectors of the wave are quantized: k = 2 L (n1 , n2 , ). Each quantized wave vector labels an oscillator. The frequency of the oscillator k is k for a wave with a dispersion k . After identifying the wave (the collective modes) as a collection of oscillators, the quantum theory for the collective modes can then be obtained by quantizing those oscillators. The eigenstates of the oscillator k are labeled by a positive integer: |nph , whose energy is given by nph k where k k

k = |k |.

(4.4.8)

We like to point out that the frequency of a vibration can be both positive or negative. In fact a classical vibration correspond to quantum transitions between two quantum states |nph and k |nph + 1 . The transition |nph |nph + 1 increases energy and the transition |nph + 1 |nph k k k k k reduces energy. The two transitions correspond to frequencies with opposite sign. So the frequencies of classical oscillations always appear in pairs with opposite signs. The absolute value of the frequency correspond to the level spacing of the corresponding quantum oscillator, which leads to eqn (4.4.8). The energy eigenstates for the collection of the oscillator are labeled by a set of integers {nph }: k |{nk } , where k runs over all quantized wave vectors. The energy of such a state |{nph } is k ph k nk k where k sums over all the quantized wave vectors. The low energy eigenstates are labeled by a few non-zero nph s. This way we obtain the low energy eigenstates and their eigenvalues k of the very complicated interacting bosons system described by eqn (4.2.36) (or eqn (4.2.33)). This is quite an achievement for the simple classical picture!
ph

If we interpreted nph as the occupation number of a kind of bosons at the k-level, then the colk lection of the quantum oscillators can also be viewed as the system of free bosons with a dispersion (4.4.8) (see section 4.1.3). This is quite amazing. We start with an interacting boson system. At the end, we nd the low energy excitations of the system are described by free bosons. To distinguish the two kinds of bosons, we will call the free bosons that describe the low lying excitations emergent bosons. But what are the emergent bosons? In the symmetric phase, the emergent bosons have the k|2 dispersion k = |2m + U , which is exactly the same as the dispersion of the original bosons. In fact, in the symmetric phase, the emergent bosons are the original bosons. This is because the ground state for the symmetry phase, |{nph = 0} is the state with no original bosons. nph in this case is k k identical to the occupation number nk of the original bosons. For the low energy excitations, only few nph s are non-zero, which correspond to a dilute gas of the original bosons. In this limit, the k interactions between the original bosons can be ignored and the original bosons become the free emergent bosons. However, in the symmetry breaking phase, the emergent bosons have a linear dispersion for small ks, k = v|k|, which is very dierent from the original bosons. The occupation numbers nph of the emergent bosons is not related to the occupation numbers nk of the original bosons. k In fact, since grnd = 0 and the original bosons have a nite density, the interaction between the original bosons cannot be ignored. In this case, the occupation numbers nk of the original bosons are not even well dened, i.e. the energy eigenstates do not have a denite occupation numbers nk . In contrast, the energy eigenstates do have a denite occupation numbers nph for the emergent k bosons. Since the emergent bosons are completely dierent from the original bosons, we will give the emergent bosons a new name: phonons. We note that phonons are gapless excitations above the superuid ground state. We like to pointed out that gapless excitations are very rare in nature and in condensed matter systems. Therefore, if we see gapless excitations, we should to ask why do they exists? 82

V()

gapless mode

Figure 4.16: The gapless phonons are uctuations between degenerate ground states. One mechanism for gapless excitations is spontaneously breaking of a continuous symmetry. As discussed in section 4.3.2, the superuid phase spontaneously break the U (1) symmetry: the energy functional has the U (1) symmetry: Etot [(x)] = Etot [e i (x)] while the ground state does not: grnd = e i grnd . Only in this case, do the gapless excitations exist. If a symmetry (continuous or discrete) is spontaneously broken, then the ground states must be degenerate (see Figs. 4.10b and 4.12). This is because, for a spontaneously broke symmetry, the Hamiltonian has the symmetry but the ground state does not have the symmetry. Thus the dierent degenerate ground states can be obtained by acting the symmetry transformations on one of the ground states. As a result, the spontaneous breaking of continuous symmetry gives rise to a continuous manifold of degenerate ground states. The uctuations between the degenerate ground states correspond to the gapless excitations (see Fig. 4.16). Nambu and Goldstone have proved a general theorem: if a continuous symmetry is spontaneously broken in a phase, the phase must contain gapless excitations [Nambu 1960; Goldstone 1961]. Those gapless excitations are usually called the Nambu-Goldstone modes. The gapless phonon in the superuid is a Nambu-Goldstone mode. In the next section, we will give an explicit discussion of the relation between the spontaneous U (1) symmetry breaking and the gapless phonons.
Problem 4.4.1 (a) The dispersion (4.4.5) for the collective modes is calculated for -interaction-potential. Find the dispersion of the collective modes for a more general interaction potential V (x). (Hint: it may useful to introduce V (k) = dd x e i kx V (x).) (b) Apply the result in (a) to an 1D interacting boson system with an interaction potential V (x) = g(|x| a). Show that when g is bigger than a critical value gc , the boson superuid state becomes unstable. Find the value of gc . Problem 4.4.2 (a) Let N (E) be the density of states of a many-body system: N (E) dE is the number of many-body energy h eigenstates with energy between E and E + dE. Assume N (E) has a form eg V , where is the energy density = E/V and V the volume of the system. The exponent h and the coecient g are constants. Find the specic heat of the system. (From this problem, you see that the low temperature specic heat measures the number of low energy excitations.) (b) To measure the number of low energy excitations in dierent phases of an interacting boson system, calculate the low temperature specic heat of the symmetry breaking phase, the symmetric phase, and at the transition point.

4.4.3

*Low energy eective Lagrangian for the sound waves

The sound wave correspond to free bosons. Their eective theory should be simple. The simple eective theory allows us to easily calculate dynamical properties of interacting bosons. 83

In section 4.4.1, we have used equation of motion to study the dynamics of the collective modes of the classical ground state. We can also use Lagrangian approach to achieve the same goal. Since the sound waves are described by the small uctuations = grnd , we can rewrite the Lagrangian (4.4.3) in terms of and drop the higher order terms in to obtain the simplied eective Lagrangian for the sound waves. However, to make the U (1) symmetry: e i more explicit, we will instead use (, ) to describe the uctuations around the ground state grnd = U/g. (, ) are dened through (x, t) = + (x, t)e i (x,t)
grnd

where grnd = U/g is the boson density in the ground state. The U (1) symmetry is now realized through (x) (x) + (4.4.9) To the quadratic order in (, ), eqn (4.4.3) becomes L= dd x [(grnd + ) grnd (x )2 (x )2 g 2 ]. 2m 8mgrnd 2 (4.4.10)

The invariance of eqn (4.4.3) under the U (1) transformation e i implies that Lagrangian (4.4.10) is invariant under eqn (4.4.9). So the U (1) symmetry is explicit in the above eective 4gmgrnd , we can Lagrangian. If we only interested in the modes with small wave vectors: |k| 1 2 since it is much smaller than g2 . The Lagrangian is simplied to drop the term 4mgrnd (x ) L=
2 grnd (x ) g 2 ]. dd x [(grnd + ) 2m 2

(4.4.11)

From the Lagrangian (4.4.11), we obtain the following equation of motion + grnd 2 =0 m x g = 0

From the second of the above two equations, we obtain = /g. The rst of the above two equations can be simplied to
2 = v 2 x ,

v=

ggrnd m

(4.4.12) ggrnd /m. We note that

which describes a linear dispersing wave k = v|k| with a velocity v = such a result agrees with eqn (4.4.6).

Since it contains only rst-order time derivative terms, the eective Lagrangian (4.4.11) is a phase-space Lagrangian. If we treat as the coordinate, then = L = t /g will be the corresponding canonical momentum. Substituting = t /g into eqn (4.4.11), we can rewrite the phase-space Lagrangian in terms of and Le = dd x [ 1 2 grnd (x )2 ], 2g 2m (4.4.13)

which is a coordinate-space eective Lagrangian. The wave equation (4.4.12) can be obtained directly from eqn (4.4.13). After quantization, the waves with dispersion k = v|k| give rise to phonons with energy (k) = v|k|. Those phonons are gapless. The gaplessness of the phonons are protected by the 84

U (1) symmetry. To see this, we note that in order for the phonon to gain an energy gap, we need to add a 2 term in the eective Lagrangian: Le = dd x [ 1 2 grnd h (x )2 2 ], 2g 2m 2

The modied equation of motion becomes

2 = v 2 x gh

which gives rise to a dispersion k = v 2 k2 + gh with an energy gap gh. However, the U (1) transformation is realized through (x, t) (x, t) + in the eective theory (4.4.13). The U (1) symmetry, i.e. the invariance of Le under (x, t) (x, t) + , prevents the terms like 2 from appearing in Le , and hence prevent the phonons from gaining an energy gap.
Problem 4.4.3 The superuid ow in the symmetry breaking phase (a) The phase-space Lagrangian (4.4.11) describes the collective modes in the symmetry-breaking phase. Find the expressions of boson density and boson current density in terms of the eld and . (b) The coordinate-space Lagrangian (4.4.13) for the collective modes contains only one eld . Find the expressions of boson density and boson current density in terms of . (c) Generalize the coordinate-space Lagrangian (4.4.13) to include a time dependent potential V (x, t). (d) For t < 0, the boson system is in the symmetry breaking ground state. At t = 0, we turn on a weak oscillating force F = F0 cos(t)z that acts on every bosons. Find the time-dependent boson current density j(t). Show that the maximum current density diverges as 0 (the static force limit) regardless how small F0 is. (The phenomenon that an arbitrary small static force can generate an arbitrary large current is called superuidity. We see that the symmetry breaking phase of the bosons is a superuid. The essence of superuidity will be discussed in section 4.5.)

Problem 4.4.4 Adding h2 dd x (2 + c.c.) term to eqn (4.4.3) breaks the U (1) symmetry down to a Z2 symmetry. The resulting system still has two zero-temperature phases as we tune U : one breaks the Z2 symmetry and the other does not. (a) Show that the collective excitations in both phases have an energy gap. (b) Show that the energy gap of the excitations approach to zero as we approach the continuous phase transition between the two phases. (c) Find the low energy eective Lagrangian that describe the low energy collective excitations near the Z2 symmetry breaking transition. (d) Find the low temperature specic heat at the transition point.

Problem 4.4.5 two-component boson superuid Consider a many-boson system which contain two types of bosons in d-dimensions. The quantum Hamiltonian of the system is given by H=
i

1 2 1 +U 2m xi

+
j

1 2 2 +U 2m xj V2 (x2 x2 ) + j j V12 (x1 x2 ) i j

ij

+
ii

V1 (x1 x1 ) + i i
jj

where xa are the coordinates of the type-a bosons. (Such a system can be realized by ultra cold atoms.) i (a) Assume Va = V12 = 0, nd the phase-space Lagrangian (which generalizes eqn (4.2.15)) that describes the corresponding classical eld theory for the above many-boson system. (b) Find the phase-space Lagrangian (which generalizes eqn (4.2.15)) that describes the corresponding classical eld theory for non-zero Va and V12 .

85

(c) Assume V1 = V2 = V12 , show that the phase-space Lagrangian has a U (1) SU (2) symmetry. (d) Assume V1 (x) = V2 (x) = V12 (x) = V (x), V > 0, and U < 0, nd the classical ground state of the system. (Hint: you may assume that the ground state does not break the translation symmetry.) (e) Find the dispersions of low energy collective modes in the above ground state. (f) Find the low temperature specic heat of the system. (g) Describe the low energy modes and the symmetry of the ground state assuming V1 (x) = V2 (x) = V (x), V12 (x) = V (x), V, V > 0, and U < 0. Discuss the two cases V < V and V > V .

4.4.4

*An oscillator picture for the sound wave

A derivation of the oscillator picture for the sound wave. A relation between the physical quantities of the original bosons, such as the boson density, and the oscillator variables. Such a relation will allow us to calculate physical properties of interacting bosons using simple oscillators. The sound waves in the symmetry breaking phase are described by a collection of oscillators. Here we like to understand such a relation in more detail. We like to know how the amplitude of sound wave is related to the coordinates and the momenta of the oscillators. We start with the eld theory Lagrangian (4.4.3) for interacting bosons. Since the sound waves are described by the small uctuations = grnd , we can rewrite (4.4.3) in terms of to obtain the Lagrangian for the sound waves. To the quadratic order in , eqn (4.4.3) becomes L= = dd x dd x 2 i t x + U ggrnd 2||2 + Re(2 ) 2m 2 i t x + ggrnd ggrnd Re(2 ) 2m

(4.4.14)

where we have used U = ggrnd and dropped some total time derivative terms. The equal coecient in the front of the two terms and Re(2 ) is a consequence of the U (1) symmetry. Let us expand the above Lagrangian in terms of k-modes. We rst write (x) as (x) = V 1/2 e i kx k

(4.4.15)

In terms of k , the Lagrangian for the sound wave takes a form L=

k
where

k i t k k k k k k k . k k k 2 2

(4.4.16)

|k|2 + ggrnd , k = k = ggrnd 2m In the following, we will assume k to a generic real function and k to be a generic complex function with k = k . If we ignore the term k k k + k k in eqn (4.4.16), then each term in the sum k k = Lk = i t k k k k k (4.4.17)

will describe a decoupled harmonic oscillator where the real part of k corresponds to the coordinate and the imaginary part of k corresponds to the momentum of the oscillator. The term 86

k (k k + k ) couples the oscillator k to the oscillator k. So eqn (4.4.16) describes a k collection of coupled oscillators. But we can choose a dierent set of variables to obtain a set of decoupled oscillators. Since the mixing is between the k-mode and the k-mode only, let us introduce ak = uk k + vk k One can show that, up to a total time derivative term,

k
if One can also show that

a ak = k

t k , k

|uk |2 |vk |2 = 1, Ek a ak k

uk vk = uk vk .

(4.4.18)

k
=

k
=

Ek |uk |2 k + |vk |2 k k + vk uk k k + vk u k k k k

1 1 k k + k k k + k k k k 2 2

(4.4.19)

if (Ek , uk , vk ) satisfy Ek |uk |2 + Ek |vk |2 = k Ek uk vk + Ek uk vk = k . One solution of eqn (4.4.20) and eqn (4.4.18) is given by uk = k |k | k 1 + , k 2 2E vk = Ek =
k |k |

(4.4.20)

k 1 , k 2 2E

Ek = Ek + k , k + k k = , 2 So in terms of ak , the Lagrangian has a form L=

2 k |k |2 , k k k = . 2

(4.4.21)

k
To show that the term

(ia ak Ek a ak ) k k

(4.4.22)

which describes a collection of decoupled oscillators. Lk = ia ak Ek a ak k k

describes a single harmonic oscillator, we introduce (Xk + iPk ) = a total time derivative term) Lk = Pk Xk
2 2 Pk Kk Xk + 2Mk 2

2ak and rewrite Lk as (up to

87

where Mk = 1/Ek and Kk = Ek . We see that Lk is a phase-space Lagrangian that describes an oscillator of mass Mk and spring constant Kk . The oscillation frequency (or more precisely, the energy level spacing) is k = which agrees with eqn (4.4.5). We note that in the expression of Ek (4.4.21), the rst ggrnd comes from the coecient in front of the term in eqn (4.4.14). The second ggrnd comes from the coecient in front of the Re() term in eqn (4.4.14). We see that the equal coecient in front of and Re(2 ) makes Ek 0 as k 0. The U (1) symmetry protects the gapless excitations in the superuid phase.
Problem 4.4.6 In this section, we write the quadratic phase-space eective Lagrangian (4.4.14) as a sum of Lagrangians of independent harmonic oscillators: L = k Lk (see eqn (4.4.22)). We can do a similar thing to the quadratic coordinate-space eective Lagrangian (4.4.13). Write eqn (4.4.13) as a sum of Lagrangians of independent harmonic oscillators.

Kk = Ek = Mk

|k|2 + ggrnd 2m

(ggrnd )2

Problem 4.4.7 Quantum uctuations of the bosons eld 2 2 2 (a) For a classical oscillator, Xk = Pk = 0 in the ground state. However, for a quantum oscillator, Xk = 0 and 2 Pk = 0 even in the ground state due to the zero point uctuations. Similarly, the boson eld |(x)|2 = 0 in the classical ground state. Use the relation between (x) and (Xk , Pk ) to nd the value of |(x)|2 in quantum ground state after including the zero point uctuations. (b) Show that |(x)|2 is nite in dimensions d 2 while it is innite in one dimension. Discuss the meaning of diverging |(x)|2 . (Hint: remember the short-distance cut-o.)

4.4.5

*An improved quantum variational approach

In the last section, we have seen that the low energy collective modes of interacting bosons are described by ak = uk k + vk k for non-zero k.8 After quantizing ak , we get the lowing operator of the oscillator or the annihilation operator of the phonon ak = uk k + vk k . The quantum ground state |0 of the interacting bosons is a state with no phonon excitations and should satises ak |0 = (uk k + vk k )|0 = 0 for all non-zero k. But the variational ground state |grnd introduced in section 2.6.2 satises k |grnd = 0. This suggests that |grnd is not the best variational state. A better variational state can be found by requiring (uk k + vk k )|0 = 0. In this section, we will discuss such an improved variational state.
We note that the unitary transformation ek (k k k k ) mixes k and k : ek (k k k k ) k ek (k k k k ) = cosh(k )k + sinh(k ) k

(4.4.23)

provided that k = k and k = 0. So the mixture (cosh(k )k + sinh(k ) k ) annihilate ek (k k k k ) |0 . This motivates us to generalize the variational ground state (4.2.42) to
|k ,N0 = eN0 /2 e k=0 k (k k k k ) e
8

N0 0

|0 ,

(4.4.24)

The real part and the imaginary part of ak correspond to the coordinate and momentum of the oscillator k. Also note that k = k for non-zero k.

88

Figure 4.17: The owing state for bosons in a pipe is described by a periodic boson eld (x) = 0 e i n . The quantization of n is the reason for the stability of the ow. where the real numbers k and N0 can be viewed as variational parameters. (To be completed)
Problem 4.4.8 Derive eqn (4.4.23). (Hint: you may want to take derivative k on both sides of eqn (4.4.23).

4.5
4.5.1

Superuidity
Superuid ow and momentum quantization

The essence of persistent superow is the stability of the owing state. The bosons in the symmetry breaking phase have a very special property: they can ow without friction. Once bosons start to ow, they can ow forever all by themselves. What is really remarkable is that such a persistent ow can exist even in the presence of a random potential, for example, in a pipe with a rough surface. This kind of ow is called superow and the uids with such a property is called superuid. The symmetry breaking ground state grnd has a vanishing boson current, and hence bosons do not ow in such a state. To understand superow, we need to study state where bosons do ow. The state described by (x) = 0 e i xK (4.5.1) is one such state. The coherent |(x) for the above (x) (see eqn (4.2.41)) is a state in which all the bosons carry the same momentum K. The boson density and the current density are given by (see and eqn (4.2.21) and eqn (4.2.22)) = |0 |2 , The velocity of the bosons is V = We will call such a state owing state. The issue here is that why bosons in the state 0 e i xk want to ow forever. Why does the state 0 e i xK refuse to evolve into another state with a small K and eventually to the ground state with a vanishing K? To answer those questions, let us consider a nite system with a volume Lx Ly Lz . We also assume the bosons to have a periodic boundary condition. In this case, the boson eld is periodic (see Fig. 4.17) (x) = (x + Lx x) = (x + Ly y) = (x + Lz z) 89 j = |0 |2 j K = m K m

2 2 2 Therefore, K in the owing state 0 e i xk is quantized: K = (Kx , Ky , Kz ) = (nx Lx , ny Ly , nz Lz ). So K cannot change continuously. If K is pinned at a quantized value, the owing state will be stable despite it has a higher energy than the ground state. This is why the bosons want to ow forever.

So the key to understand persistent superow is to understand the stability of the owing state 0 e i xK . Since we cannot smoothly change K to reduce the energy of the owing state, we like to ask are there other ways to reduce the energy of the owing state and make it unstable. In the owing, we will discuss two mechanism which may destabilize the owing state and stop the superow.

4.5.2

Stability of the owing state

If the velocity of the owing state is less than a critical velocity, all samll uctuations around the owing state cost positive energies. One way for the owing state to become unstable is through non-uniform uctuations of (x). Those uctuations represent collective excitations above the owing state. By studying the dynamics of those uctuations, we can understand the stability of the owing state. In sections 4.4.1 and 4.4.4, we study the uctuations around the symmetry breaking ground state grnd . Here we will use the similar approach to study the uctuations around the owing state 0 e i xK . Since our boson system has a xed chemical potential, = U , the owing state may have a dierent boson density = |0 |2 than the ground state. To determine the value of 0 , we simply minimize the total energy of the owing state (see eqn (4.4.2)) Etot = V We nd that 0 = g K2 |0 |2 + U |0 |2 + |0 |4 2m 2
2 U K 2m g

(note that U < 0 in the symmetry breaking phase). We then introduce to describe the uctuations (x) = (0 + )e i xK Substitute the above into the phase-space Lagrangian (4.4.3), we can obtained the eective Lagrangian that describe the dynamics of . To make the calculation easier, we note that under the transformation e i xK the Lagrangian density (4.4.3) changes to L = i t (x + iK)2 +U 2m g ||4 . 2

Using the above result, we nd that to the quadratic order in , the Lagrangian density (4.4.3) becomes L = i t = i t (x + iK)2 + U g0 2||2 + Re(2 ) 2m 2 + 2iKx x + g0 g0 Re(2 ) 2m 90

(4.5.2)

Figure 4.18: Each quantized K gives rise to a locally stable owing state. where 0 = |0 |2 is the boson density in the owing state and we have used U + K + g0 = 0. 2m In the k-space (see eqn (4.4.15)), the Lagrangian L = dd x L obtained from eqn (4.5.2) can be rewritten in a form as in eqn (4.4.16) with
2

k =

k2 kK + g0 + , 2m m

k = k = g0

(4.5.3)

To determine the stability of the owing state, we note that the energy of the uctuation is given by

k 1 = 2

(k k + k

k k k + k k ) k 2 2
k k k k

k k

k k

as one can see from the phase-space Lagrangian eqn (4.4.16). When k k > |k |2 , k + k > 0, for all k (4.5.4)

the energy is positive denite since the 2 2 matrix have positive eigenvalues. Substituting eqn (4.5.3) into eqn (4.5.4), we get |k|2 + g0 2m which is equivalent to
2

K k m

> (g0 )2 ,

for all k

|K| = |V | < v = ggrnd /m. (4.5.5) m Here V = K/m is the owing velocity of the bosons in the owing state and v the sound velocity in the superuid. We see that when the owing velocity V of the owing state is less than a critical velocity Vc = v, no small uctuations around the owing state can lower its energy despite the owing state has a higher energy than the ground state. So the owing state is a local minimum, or a locally stable state. Since the owing state cannot decay into the ground state, the bosons in the owing state will ow forever. Fig. 4.18 summarizes the local stability of the owing state. But if owing velocity is bigger than Vc , then certain uctuations can lower the energy and the owing state becomes unstable.

4.5.3

*Collective excitations above the owing state

To obtain the dynamics of the collective modes, we can use the result in section 4.4.4. We nd that the collective excitations above the owing state has the following dispersion

k = Ek =

|k|2 + g0 2m 91

(g0 )2 + V k

(4.5.6)

V<V c
4

V=V c
0 V>V c 4

k 4

Figure 4.19: The superuid ow aects the dispersion of collective modes. When V > Vc the negative energy of the collective excitations destabilize the superow.
k

=V c k k

Figure 4.20: Critical velocity Vc of a superuid with a generic dispersion. It is interesting to note that the stability condition (4.5.5) can be rewritten as

k=

|k|2 2m

|k|2 + 2g0 2m

+V k >0

(4.5.7)

So the owing state is stable as long as the collective excitations have a positive energy. This is a very reasonable result. As we increase the superuid ow velocity V , the dispersion evolves as shown in Fig. 4.19. We see that as |V | passes a critical velocity Vc , the excitations start to have negative energies and the superuid ow becomes unstable. So the superow can only exist at a velocity below Vc . The value of Vc can be obtained from the dispersion k (see Fig. 4.20). For our boson model described by the dispersion (4.4.5) (see Fig. 4.14), the critical velocity happen to be the sound velocity: Vc = v = ggrnd /m. When g = 0, our boson system becomes a free boson system. The symmetry breaking state grnd is a free-boson condensed state. However, the critical velocity for a free-boson condensed state is Vc = 0. So a free-boson condensed state is not a superuid! We also see that even a small interaction will make Vc = 0. So an interacting-boson condensed state is a superuid, regardless how weak is the boson interaction.

4.5.4

Vortex and the decay of superow

Superuid contains a topological excitation vortex line. Motion of vortex line can cause a owing state to quantum tunnel into the ground state. In last section, based on the stability of owing states, we argue that bosons can ow forever in a superuid. However, nothing last forever in physics. Even protons may decay away. So if we believe that superow will eventually decay to zero, the issue becomes identifying the mechanism that causes the decay of superow. 92

f(r) grnd
Vortex core

Figure 4.21: f (r) and the vortex core. Note that the f 2 (r) is the boson density. We have argued that K is quantized. So the superow cannot slow down via reducing K continuously. We also show that small uctuations always cost positive energies and the owing states are locally stable. So the small uctuations cannot slow down the superow either. However, our analysis of stability is based on classical physics. It is well known that a locally stable classical state can quantum tunnel into a lower energy state. From Fig. 4.18, we see that a owing state can tunnel through a barrier and decay into a state with a smaller K. To understand the tunneling process that change the value of quantized K, we like to rst discuss a new type of excitation in superuid a vortex line. We known that a constant boson eld grnd describes the ground state state. A non-uniform eld (x) describes an excited state. A vortex-line excitation is described by a particular non-uniform boson eld v (x) = v (r, , z) = f (r)e i (4.5.8)

where (r, , z) is the polar coordinate. Such a eld describes a vortex line located along the z-axis. The function f (r) need to be chosen properly such that v is a locally stable state. In other words, any small uctuations around v cost a positive energy. A static state that minimize the energy is also stationary path of the phase-space action (4.4.14). So the vortex state v is a time independent solution of eqn (4.4.1) which satises the following non-linear equation 1 2 + U + g|(x)|2 v (x) = 0 2m x

For large r, x e i 0. We nd that f (r) grnd as r . The region where f (r) is signicantly less than grng is called vortex core (see Fig. 4.21). As r 0, f (r) must vanish. Otherwise, the energy of the vortex line will diverge. Vortex-line excitations are important in understanding several phenomena of superuid. As an example, let us consider the rotation of superuid. The rotation of uid is characterized by vorticity the curl of the velocity eld v(x) w = x v(x) If we write the boson eld as (x) = i e , we nd the velocity eld for a superuid to be v(x) = j(x) x = m m (4.5.9)

This implies that the vorticity of a superuid is always zero. A superuid cannot rotate! The above result, although sounds ridiculous, is not quite wrong. A uid that rotates uniformly around the z-axis has the following velocity eld v(x) = z x where is the angular velocity of the rotation. Such a rotation has a constant vorticity w = z. There is no boson eld that can reproduce the above velocity eld. So a uniform rotation is 93

(m)

(n)

Figure 4.22: (a) (l) Quantized rotations of superuid 4 He (liquid He-II) are described by stable vortex arrays (top view). Each white spot corresponds to a single vortex line. (m) The rotating liquid He-II in a beaker (side view). (n) A trianglar lattice of vortex lines best approximates a uniform rotation. impossible for superuid. However a non-uniform rotation is possible. Non-uniform rotations are generated by creating vortex lines and those rotations are quantized. To understand the quantization of the rotation in superuid, let us view the vorticity as ux and consider the amount of ux (the vorticity) going through a disk D W =
D

d2 S w

where

d2 S

is the surface integral on the disk. Using the Stoke theorem, we nd W =

D

d2 S w =
D

dx v

(4.5.10)

where D dx is the line integral along the boundary of the disk. Using eqn (4.5.9), we nd that in superuid W = dx v = m D where is the change of along the loop D. Since the change of can only be multiple of 2 along a loop, the integral of vorticity is quantized in superuid. The rotation is also quantized. There is no innite small rotation in superuid. From the boson eld that describes a vortex line (4.5.8), we see that changes by 2 around the vortex line. So the integral of vorticity is 2 if the disk D is pierced by a vortex line. This result contradicts with the previous result which states that the vorticity in superuid is always zero. However, the previous argument is not quite correct. In that argument, we have implicitly assumed that |(x)| > 0 so that the phase e i is well dened everywhere. In fact, the phase e i is singular at the center of the vortex. This results in a vorticity distribution which has a form of -function at the center of the vortex line. We note that uniformly distributed vortex lines give rise to uniform vorticity (after a local smearing) and approximately correspond to a uniform rotation. Now we see that it is hard to make superuid rotate. We must create at least a single vortex line to generate the minimal amount of rotation. The energy cost for generating a vortex line is of order L ln L where L is the size of the superuid. Fig. 4.22 shows several quantized non-uniform rotations of superuid 4 He. To see how can vortex lines reduce superow, let us consider superuid in a circlar pipe (see Fig. 4.17). As we go around the pipe, the phase of the owing state 0 en i changes by 2n. A non-zero n corresponds to nite superow. We will call n the winding number of the owing state. 94

 1 2

Figure 4.23: Superow in a pipe. The winding numbers of the owing state along the loop-1 and the loop-2 are dierent due to the presence of a vortex line. If we pierce the superuid with a vortex line as in Fig. 4.23, we see the change of the phase along the two loops in Fig. 4.23 will dier by 2. So if we move the vortex line across the pipe, the winding number of the owing state can change by 1. Hence, the vortex line tunnel across the pipe provides a path for the owing state to decay into the ground state. It is a mechanism for the superow to stop at zero temperature. We note that the energy of the vortex line is of order d ln d where d is the diameter of the pipe. So for a thick pipe, the tunneling barrier is very high and the superuid can indeed ow forever (here forever means longer than the age of universe). At nite temperatures, vortex lines are thermally excited. The thermal motion of vortex lines will cause decay of superow. In contrast, the thermal uctuations of a form 0 e i K x + (i.e. the small uctuations) cannot cause any decay of the superow. This is because the small uctuations can never change the winding number.
Problem 4.5.1 Size of vortex core Consider a vortex with winding number n v (r, , z) = fn (r)e i n . (a) fn (r) can be obtained through variational approach by minimizing the energy of the vortex state. Use this method to obtain an approximate fn (r) and determine the size of vortex core. (b) We may view the motion of bosons around a vortex line as a superow. Use the critical velocity of the superow to estimate the size of vortex core.

Problem 4.5.2 Angular momentum of the minimal rotation Fig. 4.22a describes the minimal rotation of superuid 4 He in a beaker. Find the total angular momentum of the system if the total mass of the uid is 4g and the radius of the beaker is 1cm. Now we heat up the system to cause a superuid to normal uid transition. Assume the beaker is hanged by a thin string which conserves anguler momentum in the heat-up process. Find the angular velocity of the beaker after the phase transition. (Ignore the mass of the beaker.)

Problem 4.5.3 Superuid 4 He ows in a pipe of diameter 1cm. Estimate roughly how long the superow can last at zero temperature. (Here we assume that the pipe has a rough surface and the angular momentum does not conserve. The density of liquid He is 0.125g/cm3 and the sound velocity is 230m/s.)

95

Bibliography
Anderson, P. W., 1972, More Is Dierent, Science 177, 393. Banks, T., R. Myerson, and J. B. Kogut, 1977, Phase Transitions In Abelian Lattice Gauge Theories, Nucl. Phys. B 129, 493. Baskaran, G., and P. W. Anderson, 1988, Gauge theory of high-temperature superconductors and strongly correlated Fermi systems, Phys. Rev. B 37, 580. Casimir, H. B. G., 1948, Proc. Kon. Ned. Akad. Wetensch. B 51, 793. Foerster, D., H. B. Nielsen, and M. Ninomiya, 1980, Dynamical stability of local gauge symmetry Creation of light from chaos, Phys. Lett. B 94, 135. Goldstone, J., 1961, Field Theories With Superconductor Solutions, Nuovo Cimento 19, 154. Green, M. B., J. H. Schwarz, and E. Witten, 1988, Superstring Theory (Cambridge University Press). Hastings, M. B., and X.-G. Wen, 2005, Quasi-adiabatic Continuation of Quantum States: The Stability of Topological Ground State Degeneracy and Emergent Gauge Invariance, cond-mat , 0503554. Hermele, M., M. P. A. Fisher, and L. Balents, 2004, Pyrochlore Photons: The U(1) Spin Liquid in a S=1/2 Three-Dimensional Frustrated Magnet, Phys. Rev. B 69, 064404. Kogut, J., and L. Susskind, 1975, Hamiltonian formulation of Wilsons lattice gauge theories, Phys. Rev. D 11, 395. Kogut, J. B., 1979, An introduction of Lattice gauge theory and spin systems, Rev. Mod. Phys. 51, 659. Lamoreaux, S. K., 1997, Demonstration of the Casimir Force in the 0.6 to 6 Range, Phys. Rev. Lett. 78, 5. Landau, L. D., 1937, Theory of phase transformations, Phys. Z. Sowjetunion 11, 26. Landau, L. D., and E. M. Lifschitz, 1958, Statistical Physics - Course of Theoretical Physics Vol 5 (Pergamon, London). Laughlin, R. B., 1983, Anomalous Quantum Hall Eect: An Incompressible Quantum Fluid with Fractionally Charged Excitations, Phys. Rev. Lett. 50, 1395. Levin, M., and X.-G. Wen, 2003, Fermions, strings, and gauge elds in lattice spin models, Phys. Rev. B 67, 245316. Levin, M., and X.-G. Wen, 2005a, hep-th/0507118 . 96

Levin, M., and X.-G. Wen, 2005b, String-net condensation: A physical mechanism for topological phases, Phys. Rev. B 71, 045110. Levin, M. A., and X.-G. Wen, 2004, A unication of light and electrons based on bosonic models, cond-mat , 0407140. Maxwell, J. C., 1952, in The Scientic Papers of James Clerk Maxwell, edited by W. D. Niven (Dover, New York), volume 1, pp. 155229. Michelson, A. A., and E. W. Morley, 1887, On the Relative Motion of the Earth and the Luminiferous Ether, Amer. J. Sci. 34, 333. Moessner, R., and S. L. Sondhi, 2003, Three-dimensional resonating-valence-bond liquids and their excitations, Phys. Rev. B 68, 184512. Motrunich, O. I., and T. Senthil, 2002, Exotic order in simple models of bosonic systems, Phys. Rev. Lett. 89, 277004. Nambu, Y., 1960, Axial Vector Current Conservation In Weak Interactions, Phys. Rev. Lett. 4, 380. Savit, R., 1980, Duality in eld theory and statistical systems, Rev. Mod. Phys. 52, 453. Spamaay, M. J., 1958, Physica 24, 751. Tsui, D. C., H. L. Stormer, and A. C. Gossard, 1982, Two-Dimensional Magnetotransport in the Extreme Quantum Limit, Phys. Rev. Lett. 48, 1559. Wen, X.-G., 1995, Topological Orders and Edge Excitations in FQH States, Advances in Physics 44, 405. Wen, X.-G., 2002, Origin of Gauge Bosons from Strong Quantum Correlations (Origin of Light), Phys. Rev. Lett. 88, 11602. Wen, X.-G., 2003a, Articial light and quantum order in systems of screened dipoles, Phys. Rev. B 68, 115413. Wen, X.-G., 2003b, Quantum order from string condensations and origin of light and massless fermions, Phys. Rev. D 68, 065003. Wen, X.-G., 2004, Quantum Field Theory of Many-Body Systems From the Origin of Sound to an Origin of Light and Electrons (Oxford Univ. Press, Oxford). Wilson, K. G., 1974, Connement of quarks, Phys. Rev. D 10, 2445. Witten, E., 2004, Future of string theory, http://online.itp.ucsb.edu/online/kitp25/witten/ .

97

Index
U (1) symmetry, 95, 96 annihilation operator, 63 Berrys phase, 36 bosons, 52 Brillouin zone, 43 Casimir eect, 67 coherent state, 34, 71 coordinate-space equation of motion, 19 coordinate-space Lagrangian, 19 creation operator, 63 crystal momentum, 43 eective Lagrangian, 82 eective mass, 44 elementary particles, 2 emergent bosons, 80 fermions, 53 rst order phase transition, 74 rst quantized description, 63 fractional statistics, 55 functional, 18 gauge gauge gauge gauge gauge eld, 37 xing condition, 128 invariance, 37 potential, 20 transformation, 37, 39 phase-space Lagrangian, 19 phonons, 80 Planck energy, 3 Planck length, 3 principle of emergence, 1, 5 quantum phase transition, 73 reductionist point of view, 1 second order phase transition, 73 second quantization, 63 sound wave in the superuid, 79 spontaneous symmetry breaking, 75 state vector, 25 stationary path, 19 stationary point, 18 string-net, 112 string-net condensation, 114 superow, 87 superuid, 87 symmetry breaking transition, 75 vacuum medium, 2 winding number of a string, 125

Hamiltonian (classical), 17 Hilbert space, 25 identical particles, 51, 55 Mott insulator, 99 Nambu-Goldstone modes, 81 normal ordering, 68 order parameter, 75 phase space, 16 phase-space equation of motion, 17 98