This action might not be possible to undo. Are you sure you want to continue?

Chetan Nayak November 5, 2004

2

Contents

I

Preliminaries

1

3 3 3 4 4 6 11

1 Conventions, Notation, Reminders 1.1 Mathematical Conventions . . . . . . . . . . . . . . . . . . . . 1.2 Plane Wave Expansion . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Transforms deﬁned on the continuum in the interval [−L/2, L/2] . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Transforms deﬁned on a real-space lattice . . . . . . . 1.3 Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . 1.4 Statistical Mechanics . . . . . . . . . . . . . . . . . . . . . . .

II

Basic Formalism

17

19 19 20 21 23 23 23 25 28

2 Phonons and Second Quantization 2.1 Classical Lattice Dynamics . . . . . . . . . . . . . . . . . . . 2.2 The Normal Modes of a Lattice . . . . . . . . . . . . . . . . . 2.3 Canonical Formalism, Poisson Brackets . . . . . . . . . . . . 2.4 Motivation for Second Quantization . . . . . . . . . . . . . . 2.5 Canonical Quantization of Continuum Elastic Theory: Phonons 2.5.1 Review of the Simple Harmonic Oscillator . . . . . . . 2.5.2 Fock Space for Phonons . . . . . . . . . . . . . . . . . 2.5.3 Fock space for He4 atoms . . . . . . . . . . . . . . . . 3

4 3 Perturbation Theory: Interacting Phonons 3.1 Higher-Order Terms in the Phonon Lagrangian . 3.2 Schr¨dinger, Heisenberg, and Interaction Pictures o 3.3 Dyson’s Formula and the Time-Ordered Product 3.4 Wick’s Theorem . . . . . . . . . . . . . . . . . . 3.5 The Phonon Propagator . . . . . . . . . . . . . . 3.6 Perturbation Theory in the Interaction Picture .

CONTENTS 31 31 32 33 35 37 38 43 43 46 51 53 56 57 59 61 61 64 66 67 69 71 75 75 78 81 82 84 86 88 90 90

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

4 Feynman Diagrams and Green Functions 4.1 Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . 4.2 Loop Integrals . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Green Functions . . . . . . . . . . . . . . . . . . . . . . . 4.4 The Generating Functional . . . . . . . . . . . . . . . . . 4.5 Connected Diagrams . . . . . . . . . . . . . . . . . . . . . 4.6 Spectral Representation of the Two-Point Green function 4.7 The Self-Energy and Irreducible Vertex . . . . . . . . . . 5 Imaginary-Time Formalism 5.1 Finite-Temperature Imaginary-Time Green Functions 5.2 Perturbation Theory in Imaginary Time . . . . . . . . 5.3 Analytic Continuation to Real-Time Green Functions 5.4 Retarded and Advanced Correlation Functions . . . . 5.5 Evaluating Matsubara Sums . . . . . . . . . . . . . . . 5.6 The Schwinger-Keldysh Contour . . . . . . . . . . . . 6 Measurements and Correlation Functions 6.1 A Toy Model . . . . . . . . . . . . . . . . 6.2 General Formulation . . . . . . . . . . . . 6.3 The Fluctuation-Dissipation Theorem . . 6.4 Perturbative Example . . . . . . . . . . . 6.5 Hydrodynamic Examples . . . . . . . . . 6.6 Kubo Formulae . . . . . . . . . . . . . . . 6.7 Inelastic Scattering Experiments . . . . . 6.8 Neutron Scattering by Spin Systems-xxx . 6.9 NMR Relaxation Rate . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

7 Functional Integrals 93 7.1 Gaussian Integrals . . . . . . . . . . . . . . . . . . . . . . . . 93 7.2 The Feynman Path Integral . . . . . . . . . . . . . . . . . . . 95 7.3 The Functional Integral in Many-Body Theory . . . . . . . . 97

CONTENTS 7.4 7.5 Saddle Point Approximation, Loop Expansion . . . . . . . . . The Functional Integral in Statistical Mechanics . . . . . . . 7.5.1 The Ising Model and ϕ4 Theory . . . . . . . . . . . . 7.5.2 Mean-Field Theory and the Saddle-Point Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Transfer Matrix** . . . . . . . . . . . . . . . . . . . . . .

5 99 101 101 104 105

7.6

**III Goldstone Modes and Spontaneous Symmetry Breaking 107
**

8 Spin Systems and Magnons 8.1 Coherent-State Path Integral for a Single Spin . 8.2 Ferromagnets . . . . . . . . . . . . . . . . . . . . 8.2.1 Spin Waves . . . . . . . . . . . . . . . . . 8.2.2 Ferromagnetic Magnons . . . . . . . . . . 8.2.3 A Ferromagnet in a Magnetic Field . . . . 8.3 Antiferromagnets . . . . . . . . . . . . . . . . . . 8.3.1 The Non-Linear σ-Model . . . . . . . . . 8.3.2 Antiferromagnetic Magnons . . . . . . . . 8.3.3 Magnon-Magnon-Interactions . . . . . . . 8.4 Spin Systems at Finite Temperatures . . . . . . . 8.5 Hydrodynamic Description of Magnetic Systems 8.6 Spin chains** . . . . . . . . . . . . . . . . . . . . 8.7 Two-dimensional Heisenberg model** . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 109 114 114 115 117 118 118 119 122 122 126 127 127

9 Symmetries in Many-Body Theory 129 9.1 Discrete Symmetries . . . . . . . . . . . . . . . . . . . . . . . 129 9.2 Noether’s Theorem: Continuous Symmetries and Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 9.3 Ward Identities . . . . . . . . . . . . . . . . . . . . . . . . . . 134 9.4 Spontaneous Symmetry-Breaking and Goldstone’s Theorem . 137 9.4.1 Order parameters** . . . . . . . . . . . . . . . . . . . 141 9.4.2 Conserved versus nonconserved order parameters** . . 141 9.5 Absence of broken symmetry in low dimensions** . . . . . . . 141 9.5.1 Discrete symmetry** . . . . . . . . . . . . . . . . . . . 141 9.5.2 Continuous symmetry: the general strategy** . . . . . 141 9.5.3 The Mermin-Wagner-Coleman Theorem . . . . . . . . 141 9.5.4 Absence of magnetic order** . . . . . . . . . . . . . . 144 9.5.5 Absence of crystalline order** . . . . . . . . . . . . . . 144

6 9.5.6 9.5.7 Proof 9.6.1 Generalizations** . . . . . . . . . . . Lack of order in the ground state** of existence of order** . . . . . . . . Infrared bounds** . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CONTENTS . . . . . . . . . . . . . . . . . . . . 144 144 144 144

9.6

IV

Critical Fluctuations and Phase Transitions

145

10 The 10.1 10.2 10.3 10.4 10.5

Renormalization Group and Eﬀective Field Theories 147 Low-Energy Eﬀective Field Theories . . . . . . . . . . . . . . 147 Renormalization Group Flows . . . . . . . . . . . . . . . . . . 149 Fixed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 Phases of Matter and Critical Phenomena . . . . . . . . . . . 153 Inﬁnite number of degrees of freedom and the nonanalyticity of the free energy** . . . . . . . . . . . . . . . . . . . . . . . 155 10.5.1 Yang-Lee theory** . . . . . . . . . . . . . . . . . . . . 155 10.6 Scaling Equations . . . . . . . . . . . . . . . . . . . . . . . . . 155 10.7 Analyticity of β-functions** . . . . . . . . . . . . . . . . . . . 157 10.8 Finite-Size Scaling . . . . . . . . . . . . . . . . . . . . . . . . 157 10.9 Non-Perturbative RG for the 1D Ising Model . . . . . . . . . 159 10.10Dimensional crossover in coupled Ising chains** . . . . . . . . 160 10.11Real-space RG** . . . . . . . . . . . . . . . . . . . . . . . . . 160 10.12Perturbative RG for ϕ4 Theory in 4 − ǫ Dimensions . . . . . 160 10.13The O(3) NLσM . . . . . . . . . . . . . . . . . . . . . . . . . 166 10.14Large N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 10.15The Kosterlitz-Thouless Transition . . . . . . . . . . . . . . . 175 10.16Inverse square models in one dimension** . . . . . . . . . . . 181 10.17Numerical renormalization group** . . . . . . . . . . . . . . . 181 10.18Hamiltonian methods** . . . . . . . . . . . . . . . . . . . . . 181 183 183 185 188 191 195 197 198 201

11 Fermions 11.1 Canonical Anticommutation Relations . . . . . . . . . . . 11.2 Grassmann Integrals . . . . . . . . . . . . . . . . . . . . . 11.3 Solution of the 2D Ising Model by Grassmann Integration 11.4 Feynman Rules for Interacting Fermions . . . . . . . . . . 11.5 Fermion Spectral Function . . . . . . . . . . . . . . . . . . 11.6 Frequency Sums and Integrals for Fermions . . . . . . . . 11.7 Fermion Self-Energy . . . . . . . . . . . . . . . . . . . . . 11.8 Luttinger’s Theorem . . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . .

CONTENTS 12 Interacting Neutral Fermions: Fermi Liquid Theory 12.1 Scaling to the Fermi Surface . . . . . . . . . . . . . . . 12.2 Marginal Perturbations: Landau Parameters . . . . . 12.3 One-Loop . . . . . . . . . . . . . . . . . . . . . . . . . 12.4 1/N and All Loops . . . . . . . . . . . . . . . . . . . . 12.5 Quartic Interactions for Λ Finite . . . . . . . . . . . . 12.6 Zero Sound, Compressibility, Eﬀective Mass . . . . . . 13 Electrons and Coulomb Interactions 13.1 Ground State . . . . . . . . . . . . . . . . 13.2 Screening . . . . . . . . . . . . . . . . . . 13.3 The Plasmon . . . . . . . . . . . . . . . . 13.4 RPA . . . . . . . . . . . . . . . . . . . . . 13.5 Fermi Liquid Theory for the Electron Gas 14 Electron-Phonon Interaction 14.1 Electron-Phonon Hamiltonian 14.2 Feynman Rules . . . . . . . . 14.3 Phonon Green Function . . . 14.4 Electron Green Function . . . 14.5 Polarons . . . . . . . . . . . .

7 205 . 205 . 207 . 211 . 214 . 216 . 217 223 223 225 228 233 234

. . . . . .

. . . . . .

. . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

237 . 237 . 237 . 237 . 237 . 239

15 Rudiments of Conformal Field Theory 241 15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 15.2 Conformal Invariance in 2D . . . . . . . . . . . . . . . . . . . 242 15.3 Constraints on Correlation Functions . . . . . . . . . . . . . . 244 15.4 Operator Product Expansion, Radial Quantization, Mode Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 15.5 Conservation Laws, Energy-Momentum Tensor, Ward Identities248 15.6 Virasoro Algebra, Central Charge . . . . . . . . . . . . . . . . 251 15.7 Interpretation of the Central Charge . . . . . . . . . . . . . . 253 15.7.1 Finite-Size Scaling of the Free Energy . . . . . . . . . 254 15.7.2 Zamolodchikov’s c-theorem . . . . . . . . . . . . . . . 256 15.8 Representation Theory of the Virasoro Algebra . . . . . . . . 258 15.9 Null States . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264 15.10Unitary Representations . . . . . . . . . . . . . . . . . . . . . 268 15.11Free Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 15.12Free Bosons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 15.13Kac-Moody Algebras . . . . . . . . . . . . . . . . . . . . . . . 280 15.14Coulomb Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

8

CONTENTS 15.15Interacting Fermions . . . . . . . . . . . . . . . . . . . . . . . 287 15.16Fusion and Braiding . . . . . . . . . . . . . . . . . . . . . . . 287

V

**Symmetry-Breaking In Fermion Systems
**

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

289

291 291 292 299 300 301 303 303 304 306 308 308 308 310 313 314 317 320 320 320 320 320 321 321 321 321 321 321 321

16 Mean-Field Theory 16.1 The Classical Limit of Fermions . . . . . . . 16.2 Order Parameters, Symmetries . . . . . . . 16.3 The Hubbard-Stratonovich Transformation 16.4 The Hartree and Fock Approximations . . . 16.5 The Variational Approach . . . . . . . . . .

17 Superconductivity 17.1 Instabilities of the Fermi Liquid . . . . . . . . . 17.2 Saddle-Point Approximation . . . . . . . . . . . 17.3 BCS Variational Wavefunction . . . . . . . . . 17.4 Condensate fraction and superﬂuid density** . 17.5 Single-Particle Properties of a Superconductor 17.5.1 Green Functions . . . . . . . . . . . . . 17.5.2 NMR Relaxation Rate . . . . . . . . . . 17.5.3 Acoustic Attenuation Rate . . . . . . . 17.5.4 Tunneling . . . . . . . . . . . . . . . . . 17.6 Collective Modes of a Superconductor . . . . . 17.7 The Higgs Boson . . . . . . . . . . . . . . . . . 17.8 Broken gauge symmetry** . . . . . . . . . . . . 17.9 The Josephson Eﬀect-xxx . . . . . . . . . . . . 17.10Response Functions of a Superconductor-xxx . 17.11Repulsive Interactions . . . . . . . . . . . . . . 17.12Phonon-Mediated Superconductivity-xxx . . . . 17.13The Vortex State*** . . . . . . . . . . . . . . . 17.14Fluctuation eﬀects*** . . . . . . . . . . . . . . 17.15Condensation in a non-zero angular momentum 17.15.1 Liquid 3 He*** . . . . . . . . . . . . . . 17.15.2 Cuprate superconductors*** . . . . . . 17.16Experimental techniques*** . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . state*** . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

18 Density waves in solids 323 18.1 Spin density wave . . . . . . . . . . . . . . . . . . . . . . . . . 323 18.2 Charge density wave*** . . . . . . . . . . . . . . . . . . . . . 323

CONTENTS

9

18.3 Density waves with non-trivial angular momentum-xxx . . . . 323 18.4 Incommensurate density waves*** . . . . . . . . . . . . . . . 323

VI

Gauge Fields and Fractionalization

325

327 . 327 . 330 . 332 . 333 337 . 337 . 340 . 344 . 349 . 352 . 353 . 354 . 357

19 Topology, Braiding Statistics, and Gauge Fields 19.1 The Aharonov-Bohm eﬀect . . . . . . . . . . . . 19.2 Exotic Braiding Statistics . . . . . . . . . . . . . 19.3 Chern-Simons Theory . . . . . . . . . . . . . . . 19.4 Ground States on Higher-Genus Manifolds . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

20 Introduction to the Quantum Hall Eﬀect 20.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 20.2 The Integer Quantum Hall Eﬀect . . . . . . . . . . 20.3 The Fractional Quantum Hall Eﬀect: The Laughlin 20.4 Fractional Charge and Statistics of Quasiparticles . 20.5 Fractional Quantum Hall States on the Torus . . . 20.6 The Hierarchy of Fractional Quantum Hall States . 20.7 Flux Exchange and ‘Composite Fermions’ . . . . . 20.8 Edge Excitations . . . . . . . . . . . . . . . . . . .

. . . . . . . . States . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

21 Eﬀective Field Theories of the Quantum Hall Eﬀect 361 21.1 Chern-Simons Theories of the Quantum Hall Eﬀect . . . . . . 361 21.2 Duality in 2 + 1 Dimensions . . . . . . . . . . . . . . . . . . . 364 21.3 The Hierarchy and the Jain Sequence . . . . . . . . . . . . . 369 21.4 K-matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 21.5 Field Theories of Edge Excitations in the Quantum Hall Eﬀect375 21.6 Duality in 1 + 1 Dimensions . . . . . . . . . . . . . . . . . . . 379 22 Frontiers in Electron Fractionalization 385 22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385 22.2 A Simple Model of a Topological Phase in P, T -Invariant Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 22.3 Eﬀective Field Theories . . . . . . . . . . . . . . . . . . . . . 389 22.4 Other P, T -Invariant Topological Phases . . . . . . . . . . . . 391 22.5 Non-Abelian Statistics . . . . . . . . . . . . . . . . . . . . . . 393

. . . . . .3. . . . . . 451 24. . . .1 Impurity States . . . .5. .4. .2 Lifetime. . . . . .4. . . . .1 Introduction . . . . . . . . . . . . . . . . . . . . 429 23. . . .2 Localization . . . .4. . . . . . . . . . . 408 23. . . . . . . . . . . . . . . . 463 . . . . . . . . . 403 23. . 452 24. . . . . . . . . .1 Derivation of the σ-model . . . . . . . . . . 439 23. .7 Electron-Electron Interactions in the Diﬀusive Fermi Liquid . .3. . . . . . . .3. . . . . . . 452 24. .4. . . . . .3 Non-Interacting Electrons . . . . . . . .1 Percolation . . . . . . . . . . . . . . 449 24.4 Diﬀusion . .2 Lifschitz Tails . .2. Minimum Metallic Conductivity . . . . . . . . . . . . . . . . . 443 23. . . 447 24 Non-Linear σ-Models for Diﬀusing Electrons and Anderson Localization 449 24. . . . .5 Weak Localization . . . . . . . . . . Mean-Free-Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401 23. . . . . . . . Mott Insulators . . . . 406 23. . . . . . . . 434 23. .3. . . . 418 23. .3 Scaling Theory for Non-Interacting Electrons . . . .5 The Metal-Insulator Transition . . . . . . 441 23. . . . . . . . . . . . . . . . . .4. . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Magnetic Properties . . . .4 Physics of the Metallic State . . . . . . . . . . . . .6 Weak Magnetic Fields and Spin-Orbit Interactions: the Unitary and Symplectic Ensembles .3 Conductivity . . . . . . . . . . . . . 407 23. .5. . . . . . . . . .2 AC Conductivity . . . . . . . . . .10 CONTENTS VII Localized and Extended Excitations in Dirty Systems 399 23 Impurities in Solids 401 23. . . .3 Anderson Insulators vs. .1 Anderson Model . . . . . . . . . . . . . . 462 24. . . . .3 RG Equations for the NLσM . . . . . . 412 23.2 The Replica Method . . . . . . . . . . . . . . . . . . . 416 23. . .3.2. 460 24. . . . .5. . . . . . .3. . 405 23. . . . 410 23. . . . . . . . . .3. .6 The Integer Quantum Hall Plateau Transition . . . . . . . . . .2 Mobility Edge. . . . . . .4. . . . . . . 439 23. . . . 434 23. . . . .3 Physics of the Insulating State . . . .4 Interacting Electrons . . . . . . . . . 420 23. . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . 409 23. .1 Disorder-Averaged Perturbation Theory . . . .1 Variable Range Hopping . 423 23. . . Analogies with Classical Critical Phenomena . . . . . . . . . .3 Eﬀect of Coulomb Interactions . . . . . . . . . . . . . . . . 403 23. . . . . . .2 Interpretation of the σ-model. . 416 23.

. . . . . . . . . . . . . 468 24. .6 Mesoscopic ﬂuctuations*** . . . 468 . . .CONTENTS 11 24.5 The Metal-Insulator Transition . . . . . . . . . . . . . . . . .

12 CONTENTS .

Part I Preliminaries 1 .

.

.CHAPTER 1 Conventions. . repeated indices are summed over. The diversity of notations appear confusing. The problem is that the normalizations are often chosen diﬀerently for transforms deﬁned on the real space continuum and transforms deﬁned on a real space lattice. i = 1. so that the reader is not confused when confronted with varied choices of normalizations. . d. jµ .g. x. We shall do the same. . e. Unless otherwise speciﬁed. y. Occasionally. Unless otherwise noted. Notation. d where the 0-component is the time-component as in xµ = (t. 2. or with a Latin subscript xi . . ai bi = a1 b1 +a2 b2 +a3 b3 = a·b We will use the following Fourier transform convention: dω ˜ f (ω) e−iωt 1/2 −∞ (2π) ∞ dt ˜ f (t) eiωt f (ω) = (2π)1/2 −∞ f (t) = ∞ (1. z).1 Mathematical Conventions Vectors will be denoted in boldface. x. µ = 0. E.2 Plane Wave Expansion A standard set of notations for Fourier transforms does not seem to exist. we will use Greek subscripts. Ei . 3 . 1. .1) 1. we will work in d = 3 dimensions.g. Reminders 1. e. . .

e.2. . The plane waves form a complete orthogonal set.2) Because the function has the period L. where the integer n = 0.6) These results are simple to generalize to the multivariable case. L/2] which we wish to expand in a Fourier series. q must be given by 2πn/L.2. with a suitable choice of the normalization at the very beginning. xn = na. .5) √ ˜ If we deﬁne f (q) = limL→∞ Lfq . x−N/2 = . 1. (1. (1. f (x) = f (x + L).3) fq = √ L −L/2 Let us now take the limit L → ∞. ±2. ±1.. NOTATION. so that both integrals would symmetrically involve a √ factor of (1/ 2π). xN/2 = L/2. . Note that n takes all integer values between −∞ and +∞.2 Transforms deﬁned on a real-space lattice Consider now the case in which the function f is speciﬁed on a periodic lattice in the real space. but note the asymmetry: the factor (1/2π) appears in one of the integrals but not in the other. For the ﬁrst choice of the normalization we get f (x) = lim and L→∞ L→∞ √ L ∞ −∞ ∞ dq fq eiqx . e. ∆q = 2π/L then tend to zero. everything is ﬁne. and we can convert the q-sum to an integral.4 CHAPTER 1. REMINDERS 1. of spacing a. 1 f (x) = √ L fq eiqx .4) lim √ Lfq = dxeiqx f (x). i. 2π (1. i. So the inverse is L/2 1 dxeiqx f (x). (1. CONVENTIONS. although we could have arranged.. q (1. We can write. L/2] Consider a function f (x) deﬁned in the interval [−L/2.1 Transforms deﬁned on the continuum in the interval [−L/2..q′ → 2πδ(q − q ′ ). We shall restrict ourselves to the commonly used periodic boundary condition. so that the interval between the successive values of q. Note also that L→∞ −∞ lim Lδq.

this equation becomes π a −π a 1 dq iqxn e = δn. we started with a function which was only deﬁned on a discrete set of lattice points. despite the fact that the lattice is inﬁnitely large. (1.7) and Eq.2) are diﬀerent. the distinct set of integers m can be restricted in the interval −N/2 < m ≤ N/2. in the limit N → ∞.1.10) eiqxn = N δn. (1. 2π a (1. and N a = L . the Fourier series now reads f (xn ) = 1 L fq eiqxn . (1.11) The integration runs over a ﬁnite range of q. Because of the periodic boundary condition. (1. It is only in the limit a → 0 that we recover the results of the continuum given above.0 . G.q′ −→ ′ N →∞ 2π a δ(q − q ′ ). Thus.9). Na (1. The completeness can be written as (1. the lattice periodicity has not disappeared.9) Note the consistency of Eq.6) and Eq.7) Note that the choice of the normalization in Eq. To see this note that our complete set of functions are invariant with respect to the shift q → q + G. are ±(2π/a). q (1. To summarize. q∈1stBZ In the limit that N → ∞. The periodic boundary condition implies that f (xn ) = f (xn + L). Therefore the number of distinct q’s is equal to N . The function fq is periodic with respect . (1. where the smallest such reciprocal vectors. accordingly. q is restricted to q= 2πm . What about the orthogonality and the completeness of these set of plane waves? It is easy to see that N n=0 ei(q−q )xn = N δq. exactly the same as the number of the lattice sites in the real space. Why shouldn’t it? No matter how large the lattice is.8) but the integers m constitute a ﬁnite set.2.0 . PLANE WAVE EXPANSION 5 −L/2. this discreteness does not go away but the set [q] approaches a bounded continuum. Thus the distinct set of q’s can be chosen to be within the 1st Brillouin zone −(π/a) < q ≤ (π/a).

|ψ gives a complete description of a system through the expectation values. which would be the average values of the corresponding physical observables if we could measure them on an inﬁnite collection of identical systems each in the state |ψ .e. |ψ . to any pair of states. i. whose vectors are states.16) . Just see how ugly it would be if we were to deﬁne the scalar product of two vectors as (|χ . You can now generalize all this to three dimensions and work out the consequences of various normalizations. there is an inner product. NOTATION. Here c∗ is the complex conjugate of the complex number c. (1. The prefactor a in front of this sum is actually the volume of the unit cell in real space. |ψ ) = (|ψ . (1. 1. f (xn ) = fq = a n π a −π a dq iqxn e fq .6 CHAPTER 1. which assigns a complex number. it is a very elegant and useful concept. 2π (1. ψ|Oi |ψ (assuming that |ψ is normalized so that ψ|ψ = 1). There are linear operators. the entire q space can be divided up into periodic unit cells.14) e−iqxn f (xn ). |ψ . Oi which act on this Hilbert space. Finally. |χ )∗ . χ|ψ . |χ .15) then the operator L is the Hermitian adjoint of O and will be denoted by O† . An Hermitian operator satisﬁes O = O† (1.13) (1. CONVENTIONS. A state vector.3 Quantum Mechanics A quantum mechanical system is deﬁned by a Hilbert space. the inverse Fourier series is given by fq = a n e−iqxn f (xn ). REMINDERS to the reciprocal lattice vectors. Finally. but clearly not in an unique manner. These operators correspond to physical observables. χ|L|ψ = ψ|O|χ ∗ . Although the introduction of the dual vector space could be avoided. The notation follows Dirac and tacitly uses the dual vector space of bras { ψ | } corresponding to vector space of kets { | ψ }. H..12) In the limit N → ∞. If for all vectors |ψ and |χ .

λ′ |O|λ = λ λ′ |λ = λ′ λ′ |λ For λ = λ′ . Schr¨dinger’s equation tells us that H determines o how a state of the system will evolve in time. O.1. Hence. then. i ∂ |ψ = H|ψ ∂t (1.23) λ cλ |λ (1. we can use the eigenstates |λ as a basis for Hilbert space. (1. within the subspace spanned by these states. Given an Hermitian operator. QUANTUM MECHANICS while a unitary operator satisﬁes OO† = O† O = 1 If O is Hermitian. which we will denote by H. then eiO 7 (1. H|E = E|E (1. or total energy.22) The Hamiltonian.18) is unitary. λ′ |λ = 0 (1.17) (1.24) which evolve in time according to: |E(t) = e−i Et |E(0) (1. we can pick a set of n mutually orthogonal states. its eigenstates are orthogonal.20) If there are n states with the same eigenvalue.25) An arbitrary state can be expanded in the basis of energy eigenstates: |ψ = ci |Ei . Any state |ψ can be expanded in the basis given by the eigenstates of O: |ψ = with cλ = λ|ψ .19) (1.21) If the Hamiltonian is independent of time. (1. is a particularly important operator.26) i . then we can deﬁne energy eigenstates.3.

CONVENTIONS. inner product. The classical variables p.31) implies the commutation relations. This procedure is known as canonical quantization. REMINDERS It will evolve according to: |ψ(t) = cj e−i j Ej t |Ej .28) (1. [H. q are promoted to quantum operators and the Poisson bracket relation [p.31) (1. q.27) The usual route for constructing the quantum mechanical description of a physical system (Hilbert space.8 CHAPTER 1. a† ] = ωa† . [p. q] = −i. operators corresponding to physical observables) leans heavily on the classical description. we can determine the state of the system at an arbitrary time t. Let us carry this out explicitly in the case of a simple harmonic oscillator. given its state at some initial time t0 . NOTATION. q] = −i We deﬁne raising and lowering operators: √ a = (q + ip) /√2 a† = (q − ip) / 2 The Hamiltonian and commutation relations can now be written: H = ω a† a + [a. = 1 becomes the commutator of the corresponding operators p.26) and (1. H= and the commutation relations. The harmonic oscillator is deﬁned by the Hamiltonian.27). (1. a† ] = 1 1 2 (1. according to (1. Hilbert space is then constructed as the representation space for the algebra of the operators p. The theory is then “solved” by ﬁnding the eigenstates and eigenvalues of the Hamiltonian. The solution of the harmonic oscillator will be useful preparation for the Fock space construction of quantum ﬁeld theory.B. q]P. With these in hand.29) 1 ω p2 + q 2 2 (1. q: [p.30) We construct the Hilbert space of the theory by noting that (1.

we write a† |E = cE |E + ω . we see that a|E0 = 0 if E0 = ω/2. Then. imply that there is a ladder of states.3. |cE |2 = E|aa† |E ω =E+ 2 Hence. we will be concerned with systems composed of many particles. they are formulated in precisely the same way as any other system.33) This ladder will continue down to negative energies (which it can’t since the Hamiltonian is manifestly positive deﬁnite) unless there is an E0 ≥ 0 such that a|E0 = 0 (1. we can label the states of a harmonic oscillator by their integral † a eigenvalues.32) (1. with n ≥ 0 such that a H|n = ω n + and √ a† |n = √n + 1|n + 1 a|n = n|n − 1 (1.36) From the second of these equations. in terms of a Hilbert space with an .e. we know that a† |E ∝ |E + ω . a† |E = a|E = ω |E + ω 2 ω E − |E − ω 2 E+ (1. To get the normalization.37) These relations are suﬃcient to determine the probability of any physical observation at time t given the state of the system at time t0 . in turn. |n . a† on energy eigenstates |E . From the commutation relations. QUANTUM MECHANICS [H. At the most general and abstract level. a] = −ωa These.1. Thus. we need to ﬁnd the precise action of a. In this book. Ha† |E = (E + ω) a† |E Ha|E = (E − ω) a|E 9 (1.35) (1.38) 1 2 |n (1. i.34) To ﬁnd E0 .

Then the two-particle Hilbert space is spanned by the set: |i. j = e2iα |i.e. . REMINDERS inner product acted on by operators corresponding to observables. The two-particle Hilbert spaces of bosons and fermions are respectively spanned by: |i. 1 ⊗ |α.43) so eiα = ±1.45) and |i. 1 where i = 0. i (1. 1. j ≡ |i. j and |j. Applying this relation twice implies that |i. When we have a system with many particles. we must be more careful because |i. j (1. We will brieﬂy review this aspect of quantum many-particle systems. are the states of particle 2. 2 (1. i (1. . are the states of particle 1 and |α. If we have two distinguishable particles whose Hilbert spaces are spanned by the bases |i. The former corresponds to bosons. |i.44) (1. CONVENTIONS.41) Suppose that the two single-particle Hilbert spaces are identical.46) . i. 1. . However. while the latter corresponds to fermions. we must now specify the states of all of the particles.g.42) If the particles are identical.40) (1. j + |j. 1.10 CHAPTER 1. α. Then the two-particle Hilbert space is: |i. 2 (1. 1 ⊗ |j. however. j − |j. 2 (1. . the two particles are in the same box. i . 2 ≡ |i. e. j = eiα |j. 2. . NOTATION. .39) where α = 0. there is one feature of this description which is peculiar to many-particle systems composed of identical particles and has no real classical analog: Hilbert space must furnish an irreducible representation of the permutation group acting on identical particles. i must be physically the same state.

. in terms of the available phase space volume.4. Alternatively. . .48) Here π denotes a permutation of the particles. deﬁne entropy. . . . . . STATISTICAL MECHANICS 11 The n-particle Hilbert spaces of bosons and fermions are respectively spanned by: |iπ(1) . and kB is the Botzmann constant. . . Given the partition function.53) But how do we ﬁnd Γ(E)? We must solve the equations of motion. . xi . . however. xj . which is S = kB ln Γ(E). . Z= (1. . .1. in some cases it is not clear if we can reasonably achieve this condition. . (1. xn ) = −ψ(x1 . . xn ) (1. .52) F = − ln Z. e−βEn . . . 1 (1. S. and the issue of equilibration . . that is. is 1/kB β. . . we may. xj . we must make the scale of observation considerably greater than all the relevant time scales of the problem. xn ) = ψ(x1 . β To make sure that a system is in equilibrium. . . .50) (1. . xi . Γ(E). . . following Boltzmann.49) 1. . . we must know the dynamics of the system. iπ(n) (1.51) n where the temperature of the ensemble. . . xj . it is given by Z. . xi . . . . . . . T . . . . . this means that a bosonic wavefunction must be completely symmetric: ψ(x1 . iπ(n) (1. . xn ) while a fermionic wavefunction must be completely antisymmetric: ψ(x1 .47) π and π (−1)π |iπ(1) . . . . . F . . . . the macroscopic properties can be calculated from the free energy.4 Statistical Mechanics The concept of partition function is central to equilibrium statistical mechanics. xi . . . . . In position space. Here En are the energy eigenvalues of the Hamiltonian. xj . For a canonical ensemble that we shall frequently use.

They are of order N (∼ 1023 ). Although the ensemble approach is quite elegant and convenient. we simply have to count the number of states. According to quantum mechanics each point in the phase space corresponds to a quantum state. for the moment. They cannot wander very much in the phase space. uncritical use of it can be misleading. the slow degrees of freedom can be assumed to be constant. (1. No matter how complicated the motion may be. in the canonical ensemble.55) where β. combined with Eq. In contrast.55) we seem to sum over all states. (1. combined with Eq. in . this is impossible without computing the trajectory of the system in the phase space. the sum is so sharply peaked that practically all the weight is concentrated at E. Now. (1. but in (1. a very fast one and a very slow one. The recourse is to assume that Γ(E) is the entire volume of the phase space allowed by the conservation laws. If the observation scale is longer than the shorter time scale. Suppose that you are given a Hamiltonian which has two widely separated scales. if the system. all we need to do is to calculate the measure in the phase space corresponding to the conserved quantities.55). In general. This is curious. We have now arrived at the cannonical ensemble. Thus. The Boltzmann formula can be reconciled with the ensemble approach of Gibbs. The reason for this miracle is the extensive nature of E and S. in the course of time. You can show that the entropy deﬁned by Γ′ (E) leads to the same thermodynamics as the one deﬁned by Γ(E). So.54).54) we only sum over states of energy E. is an unknown positive number. It is convenient to introduce quantum mechanics at this step to simplify the argument. (1. We can go further and ask what would happen if we replaced the above formula by the following: Γ′ (E) = n e−β(En −E) . provided β = 1/kB T . But in “deriving” it. 1. the calculation of the partition function is a counting problem. visits every point in the phase space.12 CHAPTER 1. Consequently. NOTATION. leads to the same thermodynamics as you would obtain from a canonical ensemble. deﬁnes the microcannonical ensemble of Gibbs. we did not have to invoke the notion of an ensemble.53).54) Equation (1. CONVENTIONS. (1. and we write Γ(E) = n δ(E − En ). REMINDERS must be addressed. in Eq. Eq.53. but smaller than the longer time scale. We must determine Γ(E).

it is not suﬃcient to know all the states and sum over all of them. When there are a few discrete set of widely separated scales. In this respect.56) wn |n n| . ˆ where the density matrix ρ is given by ρ= ˆ n (1. (1. The spins of the nuclei in a hydrogen molecule can be either in a triplet state. We still have to understand what we mean by an ensemble average when experiments are done on a single system. A simple well known example of two distinct time scales is the problem of ortho. however. Therefore. If glass is to be described by a Hamiltonian. In principle. the number of nuclei in the singlet state and the number of nuclei in the triplet state are separately constants of motion on the time scale of a typical experiment. the Boltzmann formula. falls into this category. and there is a continuum of of time scales. Of course. (1. STATISTICAL MECHANICS 13 calculating the relevant volume of the phase space we must ignore the slow degrees of freedom. while the momenta of the molecules equilibrate in a microscopic time scale. There is no need to use the hypothesis that Γ(E) is the total volume allowed by the conservation laws.4. we could calculate the actual trajectories to determine the volume of the phase space sampled during the observation time.1. where this is not the case. the ortho-para conversion takes time. on the order of days. we must examine the actual dynamics of the system. extending from very short microscopic scales to very long macroscopic scales.and para-hydrogen. because it is clear what the relevant region of the phase space is. There are instances. as far as we know.57) = 1 Z n . Experimental observations strikingly conﬁrm this fact. including a time dependent speciﬁc heat. e−βEn |n n| . Glass is known to exhibit many anomalous thermal properties. this formula is a postulate as well and is not derived from any other known laws of physics. The interaction between the nuclei is negligible and so is the interaction between the nuclei and the electronic spins that are in a singlet state. it is easy to apply our formulae.53). and the free energies of these two subsytems must be added rather than the partition functions. Eq. window glass. Thus. can still be used. The ensemble average of an observable O is deﬁned to be O = Tr ρO. or in a singlet state. The common amorphous material. otherwise we would get an answer that will not agree with observations.

that is.59) In this ensemble. the most probable value is the only value. As in mechanical equilibrium. Another useful partition function is the grand canonical partition function ZG deﬁned by. that an experiment yields the most probable value of O.60) All thermodynamic quantities can be calculated from these deﬁnitions. (1. similar to a mechanical system in which a force ﬁxes the conjugate coordinate. . in a statistical equilibrium the chemical potentials for all the components must balance. and deﬁne more and more ensembles. However. For every such extension. ZG = Tr e−β(H−µN ) . and the system is assumed to be in contact with a particle bath as well as a heat bath. in addition. pressure P is not constant and deﬁne a pressure ensemble. in which we add a term −P V in the exponent. In the deﬁnition of the trace one must also include a sum over a number of particles. in which all the forces must balance. the number of particles is not ﬁxed. we may assume that. V → ∞ such that ρ = N is a given number. must be equal. It is convenient to think of chemical potential as a “force” and the number of particles as a “coordinate”. the value shared by most members of the ensemble.14 CHAPTER 1. We could also consider an ensemble in which the linear momentum is not ﬁxed etc.that is. we would add a “force” multiplied by the corresponding conjugate “coordinate”. Ω: Ω=− 1 ln ZG β = F − µN. we could go on. hence the mean value. Thus. The relative ﬂuctuations in O is given by V < O 2 > − < O >2 ∼O < O >2 1 √ N (1. we deﬁne the grand potential. the distribution of the members in the ensemble is so strongly peaked for a macroscopic system that roughly only one member matters. For the grand canonical ensemble.58) which is insigniﬁcant when N ∼ 1023 . For example. ﬂuctuations are insigniﬁcant in the thermodynamic limit deﬁned by N → ∞. (1. CONVENTIONS. REMINDERS It is more likely. Actually. NOTATION. It is also possible to give a similar interpretation to our formula for the canonical ensemble where we can take the temperature as the “force” and the entropy as the corresponding “coordinate”. The average number of particles is determined by the chemical potential µ. however.

from the deﬁnition of the deﬁnition of the thermodynamic potential Ω.4.64) (1.69) .N (1. T. V.67) (1. Then. we get dF = −SdT − P dV + µdN.66) .65) (1.V .V Similarly. Since F = E − TS (1.µ (1. we can derive the same relations as S=− P =− µ=− ∂Ω ∂T ∂Ω ∂V ∂Ω ∂N . T.63) (1.µ .1. V. STATISTICAL MECHANICS 15 The deﬁnition of the free energies allow us to calculate various thermodynamic quantities.68) (1. S=− P =− µ=− ∂F ∂T ∂F ∂V ∂F ∂N .N .62) (1.61) and dE = T dS − P dV + µdN. T. T.

16 CHAPTER 1. REMINDERS . NOTATION. CONVENTIONS.

Part II Basic Formalism 17 .

.

a crystalline solid can be modelled as an elastic medium. 1 0 1 0 1 0 0 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 1 1 0 0 0 1 1 1 0 0 0 1 0 1 0 1 0 1 0 1 0 1 1 0 0 1 0 1 0 1 0 1 0 1 0 r+u(r) 1 1 0 0 1 1 0 0 1 0 1 0 1 0 1 0 R i+ u(Ri ) Figure 2. Let us describe small displacements from these sites by a displacement ﬁeld u(Ri ). 19 . equivalently. at wavevectors q ≪ 2π/a. We replace u(Ri ) by u(r) (i. by a continuous variable. We will imagine that the crystal is just a system of masses connected by springs of equilibrium length a. Suppose the equilibrium positions of the ions are the sites Ri . At length scales much longer than its lattice spacing.CHAPTER 2 Phonons and Second Quantization 2. Such an approximation is valid at length scales much larger than the lattice spacing.e. we replace the lattice vectors. r). Ri .1 Classical Lattice Dynamics Consider the lattice of ions in a solid. or.1: A crystalline solid viewed as an elastic medium. a.

however. Translational invariance implies V [u + u0 ] = V [u]. is a surface term and can be ignored. it is α2 µ/2.3) (2. viewed as an elastic medium. and make the approximation of full rotational invariance. but let’s focus on the long-wavelength physics for now). 2. ∂i uj . In a crystal – which has only a discrete rotational symmetry – there may be more parameters than just µ and λ.4) The solutions.1) There are only two possible such terms.5) 2µ + λ k ≡ ±vl k ρ (2. uy = uz = 0. In a crystal with cubic symmetry. for instance. ukk . PHONONS AND SECOND QUANTIZATION The potential energy of the elastic medium must be translationally and rotationally invariant (at shorter distances. t) = ǫi ei(k·r−ωt) where is a unit polarization vector. under a shear stress. three independent parameters.2) where ρ is the mass density of the solid and µ and λ are the Lam´ coeﬃcients. so V can only be a function of the derivatives. uij ≡ 1 (∂i uj + ∂j ui ) 2 (2. ui (r. the action of a crystalline solid to quadratic order. depending on the symmetry of the lattice. uij uij and u2 (repeated indices are kk summed). Hence. the change in the energy density of the elastic medium is α2 (λ + 2µ/3)/2. e Under a dilatation. u(r) = αr. there are.2 The Normal Modes of a Lattice Let us expand the displacement ﬁeld in terms of its normal-modes. ux = αy. A third term.6) . is: S0 = dtd3 rL = 1 2 dtd3 r ρ(∂t ui )2 − 2µuij uij − λu2 kk (2. The equations of motion which follow from (2. these symmetries are broken to discrete lattice symmetries.20 CHAPTER 2. l ωk = ± (2. in general. We will make life simple. Rotational invariance implies that it can only be a function of the symmetric combination. satisfy −ρω 2 ǫi = − (µ + λ) ki (kj ǫj ) − µk2 ǫi For longitudinally polarized waves.2) are: 2 ρ∂t ui = (µ + λ) ∂i ∂j uj + µ∂j ∂j ui (2. ki = kǫi .

Poisson Brackets ∂L = ρ∂t ui ∂(∂t ui ) The canonical conjugate to our classical ﬁeld.s ei(k·r−ωk t) + a† e−i(k·r−ωk t) s i k. instead. ui . we can take the inﬁnite-volume limit by making the replacement: f (k) = k 1 k (∆k)3 V = (2π)3 f (k) (∆k)3 d3 k f (k) (2.3 Canonical Formalism.8) s = 1. t) = k.3) is of the form: ui (r. Hence. s The normalization factor. 3 corresponds to the longitudinal and two transverse polarizations.3. kj ǫj = 0 have t ωk = ± 21 µ k ≡ ±vs k ρ (2. t) = d3 k (2π)3/2 1 s s ǫs ak. CANONICAL FORMALISM. is πi ≡ The Hamiltonian is given by H= d3 r πi ∂t ui − L (2. which is consistent with our Fourier transform convention: ui (r.7) Hence. 2. POISSON BRACKETS while transverse waves. n3 ).2. Hence. 1/ 2ρωk . the allowed k’s are 2π (n1 .10) s 2. use the following.s 1 s s ǫs ak. The allowed k values are determined by the boundary conditions in a ﬁnite system. For periodic boundary conditions in a cubic system of size V = L3 . was chosen for later convenience.9) It would be natural to use this in deﬁning the inﬁnite-volume limit. n2 . the k-space volume L per allowed k is (2π)3 /V .s 2ρωk (2. but we will. the general solution of (2.11) .s 2ρωk (2.s ei(k·r−ωk t) + a† e−i(k·r−ωk t) s i k.

[. we will promote the Poisson brackets to commutators.22 CHAPTER 2. ]. πi (r ′ )]PB = δ r − r ′ and ∂t πi = [πi .12) Let us deﬁne the functional derivative.16) δui δπi δπi δui With this deﬁnition.17) (2. [uj (r). the normalization chosen above for the normal mode expansion of ui (r) is particularly convenient since it leads to: [ak. ]PB → i [. PHONONS AND SECOND QUANTIZATION 1 2 1 = 2 = d3 r ρ(∂t ui )2 + 2µuij uij + λu2 kk d3 r 1 2 π + 2µuij uij + λu2 kk ρ i (2. V ]PB = d3 r − (2.19) When we quantize a classical ﬁeld theory.s . H]PB ∂t ui = [ui .14) From these equations.s (2.15) δH δui (2. we see that it is natural to deﬁne the Poisson brackets: δU δV δU δV [U. H]PB (2. δ δη d3 r F(r) = ∂F ∂F − ∂η ∂(∂i η) (2. .13) Then the equation of motion for πi can be written ∂t πi = − while ∂t ui = δH δπi (2.18) As we will see shortly. a† ′ ′ ]PB = −i δss′ δ k − k′ k .

4. rN ). MOTIVATION FOR SECOND QUANTIZATION 23 2. whose number is ﬁxed). It has 3 degrees of freedom per spatial point – i. However. r2 .2. .2) deﬁnes a classical ﬁeld theory. Here. the collective excitations of an elastic medium are particle-like objects – phonons – whose number is not ﬁxed. the basic quantities will be the normal modes of the displacement ﬁeld.1 Canonical Quantization of Continuum Elastic Theory: Phonons Review of the Simple Harmonic Oscillator No physics course is complete without a discussion of the simple harmonic oscillator.5 2. it has inﬁnitely many degrees of freedom. .5. . r2 . r1 . we can handle the electromagnetic ﬁeld and an elastic medium in a parallel fashion which greatly facilitates calculations. E = −∇ϕ − ∂t A. . ∂t A(r) rather than ra . . Phonons are an example of the quasiparticle concept. Hence. By going to the continuum limit. This is a consequence of the continuum limit which we took. rather than the ionic coordinates. As we will see below. there are 3N degrees of freedom. we will also be concerned with the electromagnetic ﬁeld. ∂t u(r) or A(r). rN . We thereby make a transition from classical particle mechanics (with a discrete number of degrees of freedom) to classical ﬁeld theory (with continuously many degrees of freedom): ra ↔ u(x. which does have inﬁnitely many degrees of freedom (2 per spatial point when gauge invariance is taken into account). In this approach. pa . it is extremely convenient to take the continuum limit and ignore the diﬀerence between 3N and ∞. A real ﬁnite-size sample of a solid has a ﬁnite number of degrees of freedom: if there are N ions.e. . we will recall the operator formalism which will lead naturally to the Fock space construction of quantum ﬁeld theory.20) At the quantum level. we will have to develop the formalism of second quantization. t) t↔t a↔x (2. The coordinates r are no more than indices (but continuous ones) on the ﬁelds. B = ∇ × A. In order to deal with particles whose number is not ﬁxed (in contrast with the ions themselves. Furthermore. . the operators of the theory will be u(r). we will be dealing with wavefunctionals of the form Ψ[u(r)] or Ψ[A(r)] rather than ψ(r1 . . . 1 2.4 Motivation for Second Quantization The action (2.

a† ] = 1 The commutation relations.25) 1 2 (2.22) (2. |cn |2 = n|aa† |n . [H. To get the normalization. |n . such that H|n = ω n + 1 2 |n (2. Then. [p.27) Such a state has E0 = ω/2. q] = −i We deﬁne raising and lowering operators: √ a = (q + ip) /√2 a† = (q − ip) / 2 The Hamiltonian and commutation relations can now be written: H = ω a† a + [a. PHONONS AND SECOND QUANTIZATION The harmonic oscillator is deﬁned by the Hamiltonian.21) and the commutation relations.24 CHAPTER 2. H= 1 ω p2 + q 2 2 (2. Ha† |E = (E + ω) a† |E Ha|E = (E − ω) a|E (2.23) This ladder will continue down to negative energies (which it can’t since the Hamiltonian is manifestly positive deﬁnite) unless there is an E0 ≥ 0 such that a|E0 = 0 (2. a† ] = ωa† [H. a] = −ωa imply that there is a ladder of states. We have a complete set of H eigenstates.26) (2.24) (2.28) and (a† )n |0 ∝ |n . We label the states by their a† a eigenvalues. we write a† |n = cn |n + 1 .

s k. a† ′ k .s . 0) e ωk ωk ′ . √ a† |n = √n + 1|n + 1 a|n = n|n − 1 25 (2. t) .2 Fock Space for Phonons A quantum theory is made of the following ingredients: • A Hilbert Space H of states |ψ ∈ H.2. 0) + i i ik·x .s (2.32) ωk ui (x. CANONICAL QUANTIZATION OF CONTINUUM ELASTIC THEORY: PHONONS =n+1 Hence. t) .s′ d3 x (2π)3/2 d3 x′ (2π)3/2 e−ik ·x ′ ′ ui (x. πi by quantum operators satisfying the canonical commutation relations: ui (x. • An Inner Product χ|ψ which must be deﬁned so that Oi is Hermitian with respect to it if the corresponding physical observable is real. t = πi (x. uj x′ . a† ak. ui .31) according to: 1 s ǫ 2 i 1 = ǫs 2 i s 2ρωk s 2ρωk d3 x (2π)3/2 d3 x (2π)3/2 i ik·x s ∂t ui (x. t = 0 ui (x.29) (2.72).s . we see that ak. We replace the classical variables. 0 − s′ ∂t uj x′ . 0) − s ∂t ui (x. t) . πj x′ .s = a† k.s satisfy the commutation relations: 1 = ρ 2 s s ω k ω k ′ ǫs ǫs i j ′ ′ ak. πj x′ . uj x′ . (8. t = i δij δ x − x′ We can now deﬁne the operators ak.s and k.5. 0) e ωk i ui (x. • Operators Oi on H. corresponding to physical observables. 0) e−ik·x (2. 0) + a† These expressions can be inverted to give the normal-mode expansion.30) 2. 0 s ∂t ui (x. Using πi = ρ∂t ui . and the above commutation relations. In order to construct these objects for an elastic medium – thereby quantizing our classical ﬁeld theory – we employ the following procedure.5.

s e 2ρωk + ak. a† ′ k. 0) .si i (2.37) are an orthogonal basis of Fock space.35) Hence. The full set of energy eigenstates can built on the vacuum state: a† 2 . ak′ . uj x . thereby deﬁning the inner product. which has energy ni ωki .s s d3 k ωk 1 + δ(0) 2 (2.s s e2iωk t −ρ(ωk )2 + µk2 + δs1 µ + λ k2 ) s + a† ak.s′ = a† .33) ak.s k. |0 which satisﬁes: ak.s′ =0 k.s k.s . . in terms of ak.34) by substi- We can re-write the Hamiltonian. There is a ground state. PHONONS AND SECOND QUANTIZATION 1 ρ 2 s s ω k ω k ′ ǫs ǫs i j ′ ′ = d3 x (2π)3/2 = 1 2 s s ω k ω k ′ ǫs ǫs i j ′ ′ d3 x (2π)3/2 = δss′ δ k − k′ We can similarly show that d3 x′ ik·x−ik′ ·x′ i e ( ui (x.37) The Hilbert space spanned by these states is called Fock space. H= 1 2 d3 k s s −2iωk t −ρ(ωk )2 + µk2 + δs1 µ + λ k2 s (ak.s a−k.s ρ(ωk )2 + µk2 + δs1 µ + λ k2 + a† a† 1 = 2 = d 3 s k ωk ak.s k .72) into (2. 0 ωk d3 x′ 1 1 ′ ′ δij δ x − x′ eik·x−ik ·x s + s′ 3/2 ωk ωk ′ (2π) + (2.36) for all k.37).s k.sj k nj |0 (2.s a† k.s k. one for each k.s1 k n1 (2. s.s −k.s |0 = 0 a† 1 . We demand that the states of the form (2. The state (2. 0 3/2 (2π) ωk ′ i ′ ) s ∂t ui (x.38) . the elastic medium can be treated as a set of harmonic oscillators. a† j .s k.26 CHAPTER 2. or vacuum state.12).s s ρ(ωk )2 + µk2 + δs1 µ + λ k2 s + a† ak. . H. 0) .s and a† tuting (8.s (2.s2 k n2 .s a† a† ak. − s′ ∂t uj x′ .

In most of the problems which we will be discussing in this course. You may have observed that the above Hamiltonian has an inﬁnite constant. the cutoﬀ will be at a scale of electron volts.s2 above is unimportant. In fact.sj ’s is the product with all of the a† i . hence the name phonons. .s1 and ak2 .2. When we turn to electrons. The normal ordered product can be deﬁned for any free ﬁelds. The creation operator a† i . then we eliminate the zero-point energy.39) Since creation operators commute with one another and annihilation operators do as well. however.si ’s k and akj .si ’s to the left and all of the k aki . is ﬁnite.e. The normal-ordered product of a set of a† i . The divergent zero-point energy is the ﬁrst of many ultra-violet divergences which we will encounter.s1 ak1 . This can be done by introducing the notion of a normal-ordered product. It is denoted by a pair of colons. this cutoﬀ is the inverse of the lattice scale. .s2 k k (2. Hence in the above example. i. which is due to the zero-point motion of the ions.s1 annihilates such a phonon. we will not be interested in this energy (see. For the most part.s1 ak2 . this is the wavevector corresponding to the Debye energy. In the above example. For example. the problem set)..si creates k a phonon of momentum ki and polarization si while the annihilation operator ak1 . If we deﬁne the quantum Hamiltonian to be : H :. for any ﬁelds which can be expanded in creation and annihilation operators with timedependence of the form (8. Then we can always write A = A(+) + A(−) where A(+) is the part of the expansion of A which contains positive frequencies. When ﬁeld theory is applied to electrodynamics. Suppose A is such an operator. we will ﬁnd that this energy. They occur when we extend the upper limit of k-integrals to inﬁnity. This constant is the zero-point energy of the system. . nj phonons of momentum kj and polarization sj .si ’s to the right. the ordering of ak1 . the normalmode sound-wave oscillations have aquired a particle-like character.s1 a† 1 . Normal-ordering puts the A(+) ’s to the left and the A(−) ’s to the right. If we go back to our underlying ionic lattice. eiωt and A(−) is the part which contains the negative frequencies. which will be useful later.5.72).s2 : = a† 1 . it is inﬁnite because we have taken the continuum limit in an inﬁnite system. CANONICAL QUANTIZATION OF CONTINUUM ELASTIC THEORY: PHONONS 27 can be thought of as a state with n1 phonons of momentum k1 and polarization s1 .s1 ak2 . : ak1 . At the quantum level. so we will drop it. The sum over k really terminates when ωk is the Debye energy. n2 phonons of momentum k2 and polarization s2 . these integrals are always cutoﬀ at some short length scale. it must be cutoﬀ at the scale at which . we do not need to speﬁcy their orderings.

|0 . k2 (2.5. There is also a no-particle state: the vacuum. k2 2m 2m = k1 + k2 |k1 .3 Fock space for He4 atoms We can use the same formalism to discuss a system of bosons. we have eigenstates |k1 . a† ′ = 0 so that |k = a† |0 k (2. approximately 100GeV . k2 = P |k1 . 4-particle. k2 satisfying k1 .42) They are orthogonal: k′ |k = δ(k − k′ ) If we have two particles. is allowed to vary. a† ′ = δ(k − k′ ) k k k ak . The Hilbert space spanned by all of these states is Fock space. states. etc.. .44) 2 k2 k1 + 2 |k1 . k2 |k3 .40) The energy eigenstates |k have energies and momenta H|k = k2 |k 2m P |k = k|k (2. . This is particularly convenient when the number of He4 atoms is not ﬁxed.43) We can continue in this way to 3-particle. say He4 atoms. k3 = δ(k1 − k3 ) δ(k2 − k4 ) + δ(k1 − k4 ) δ(k2 − k3 ) (2. is ﬁxed and the number of particles. where the chemical potential.41) (2. . as for instance in the grand canonical ensemble. ak′ = a† . 2. µ. ak satisfying: k ak . k2 with H|k1 . Suppose we have a He4 atom with Hamiltonian H= P2 2m (2. We can deﬁne creation and annihilation operators a† .45) . PHONONS AND SECOND QUANTIZATION it becomes uniﬁed with the weak interactions.28 CHAPTER 2. N .

5. .42) (2. CANONICAL QUANTIZATION OF CONTINUUM ELASTIC THEORY: PHONONS |k1 . ψ(x). We can now construct the ﬁelds. we can second quantize this wave equation and arrive at the Fock space description. ψ † (x): ψ(x) = ψ † (x) = ψ(x) satisﬁes the equation: i ∂ 1 2 ψ(x) = − ∇ ψ(x) ∂t 2m (2.48) In other words.3) – which can be derived from the action S= dt d3 r ψ † i 1 2 ∂ + ∇ ψ ∂t 2m (2.2. Writing H= P = d3 k ωk a† ak k 29 (2.50) Then.44) are also recovered. k2 = a† a† |0 k1 k1 etc. we see that we recover the correct energies and momenta.47) where ωk = k2 /2m.46) d3 k k a† ak k (2. we see that the correct orthonormality properties (2. suppose we view the Sch¨dinger equation as a classical wave o equation – analogous to the wave equation of an elastic medium (2.49) d3 k a e−i(ωk t−k·x) (2π)3/2 k d3 k a† ei(ωk t−k·x) (2π)3/2 k (2. From the commutation relations.

PHONONS AND SECOND QUANTIZATION .30 CHAPTER 2.

it is not intended to be a realistic phonon Lagrangian.1) is: H= 1 2 1 g π + 2µuij uij + λu2 + d3 r kk 2 ρ i 4! = H0 + H ′ dt d3 x (∂k uk )4 (3. most systems have Lagrangians with higher-order terms.1 Higher-Order Terms in the Phonon Lagrangian The second quantization procedure described in the previous chapter can be immediately applied to any classical ﬁeld theory which has a Lagrangian which is quadratic in its basic ﬁelds. there is no normal mode expansion. and we cannot follow the steps which we took in chapter 2. the presence of such terms means that diﬀerent solutions can no longer be superposed. Hence. there are certainly terms in the phonon Lagrangian which we have neglected which are cubic.1) The Hamiltonian corresponding to (3. An example of a phonon Lagrangian with such a term included is S = S0 − g 4! dt d3 x (∂k uk )4 (3.2) We use this phonon Lagrangian as an illustrative example. quartic. When g is small. However.CHAPTER 3 Perturbation Theory: Interacting Phonons 3. we 31 . ui . For instance. and higher-order in the displacement ﬁelds. Classically.

U (t. U (t. If H is time-independent. and Wick’s theorem. t′ )|ψ(t′ ) S (3. t))† (3. t′ ) = H(t)U (t. is time-dependent. states are time-independent. In this chapter. Suppose OS (t) is an operator in the Schr¨dinger picture (we have allowed for explicit time deo pendence as in H(t) above).5) S = U (t. on the other hand. t′ ) = e−i(t−t )H ′ (3.3) while operators are time-independent unless they have explicit time dependence.7) while operators contain all of the time-dependence. hope to use perturbation theory to solve this Hamiltonian. the timeordered product. Heisenberg. t) = 1.6) In the Heisenberg picture. In order to facilitate the construction of the perturbation theory. higher-order terms in the phonon Lagrangian lead to interactions between the phonons which cause them to scatter oﬀ each other. t′ ) acts on states in the following way: |ψ(t) It satisﬁes the equation i ∂ U (t. we will need several technical preliminaries: the interaction picture. states evolve in time according to: o i S = H(t) |ψ(t) S (3. |ψ(t) H = |ψ(0) S = |ψ(0) H (3.2 Schr¨dinger. Then the corresponding Heisenberg picture operator is: OH (t) = U (0. we develop a perturbation theory for H ′ using the solution of H0 presented in chapter 2. H(t) = p2 /2m + x cos ωt. PERTURBATION THEORY: INTERACTING PHONONS can. then U (t.4) subject to the initial condition. 3. however. For example. if we have a particle in 1D. but we can switch on a time-dependent driving force in which case the Hamiltonian.32 CHAPTER 3. and Interaction Pico tures ∂ |ψ(t) ∂t In the Schr¨dinger picture.8) . t′ ) ∂t (3. t) OS (t) (U (0. The timeevolution operator. As we will see. p and x do not depend on time.

A formal solution was written down by Dyson: UI (t. UI (t. DYSON’S FORMULA AND THE TIME-ORDERED PRODUCT 33 Finally. since we will know the time dependence of both operators and states. |ψ(t) I = eitH0 |ψ(t) S (3.10) (We will drop the prime and simply call it HI . which we will use extensively. t′ )|ψ(t′ ) I (3. t′ ) = HI (t) UI (t. t′ ) ∂t which evolves states according to |ψ(t) I = UI (t.13) i UI (t. t′ ).14) 3. satisfying ∂ (3.9) This includes the interaction Hamiltonian. states satisfy the diﬀerential equation i ∂ |ψ(t) ∂t I ∂ eitH0 |ψ(t) S ∂t = eitH0 (−H0 + HS ) |ψ(t) S ′ = eitH0 HS (t) e−itH0 |ψ(t) I = HI (t) |ψ(t) I = i (3.3 Dyson’s Formula and the Time-Ordered Product If we can ﬁnd UI (t. we turn to the interaction picture. The interaction picture interpolates between the Heisenberg and Schr¨dinger pictures. t′ ) = T e−i Rt ′′ t′ dt HI (t′′ ) (3. t′ ).3. H ′ which now has time-dependence due to H0 : ′ HI (t) = eitH0 HS (t) e−itH0 (3. then we will have solved to full Hamiltonian H0 + H ′ .) The states lack this part of the time-dependence. Operators o have time-dependence given by H0 : OI (t) = eitH0 OS (t) e−itH0 (3.15) .12) We can deﬁne an Interaction picture time-evolution operator.3. This picture can be deﬁned when the Hamiltonian is of the form H = H0 + H ′ and H0 has no explicit time-dependence.11) Hence.

Hence. . of a string of operators. Suppose we perform a neutron scattering experiment in which a neutron is ﬁred into the system with momentum P at time t′ and then interacts with our system according to HI . . O1 (t1 )O2 (t2 ) . all of the Oi ’s are HI .13).compute transition amplitudes. if we prepare an initial state |i at t′ = −∞. On (tn ). > tin (3.19) Of course. . we can pull it out to the left: i R t ′′ ′′ ∂ T e−i t′ dt HI (t ) ∂t = HI (t) T e−i Rt ′′ t′ dt HI (t′′ ) (3. T. we measure the system in a ﬁnal state f | at t = ∞ with amplitude: f |UI (∞.16) There is some ambiguity if ti = tj and O(ti ) and O(tj ) do not commute. so we must often expand the exponential. −∞)|i = −i f | ∞ −∞ dt HI (t)|i (3. let us suppose that we have a system which is in its ground state. t′ )|0. The probability (which is the square of the amplitude) for the system to undergo a transition to an excited state 1| so that the neutron is detected with momentum P ′ at time t is: 2 1. .20) Hence. we can rarely evaluate UI (t. so we do not need to worry about this. PERTURBATION THEORY: INTERACTING PHONONS where the time-ordered product. with operators with earlier times to the right of operators with later times: T {O1 (t1 )O2 (t2 ) . we can – at least in principle . it is greater than or equal to any other t′′ which appears under the time-ordering symbol. .17) Since t is the upper limit of integration. however. In (3. To see that it satisﬁes the diﬀerential equation (3. . P ′ |UI (t.34 CHAPTER 3. is their product arranged sequentially in the order of their time arguments. The ﬁrst-order term in the expansion of the exponential is: t −i t′ dt1 HI (t1 ) (3. t′ ) exactly. For example.21) . On (tn )} = Oi1 (ti1 )Oi2 (ti2 ) . observe that all operators commute under the time-ordering symbol. |0 .18) With Dyson’s formula in hand. so we can take the derivative naively: i R t ′′ ′′ ∂ T e−i t′ dt HI (t ) ∂t = T HI (t)e−i Rt ′′ t′ dt HI (t′′ ) (3. . Oin (tin ) if ti1 > ti2 > .15). P (3. .

3. There is a slight subtlety in that the t integral leads to an amplitude proportional to δ(Ei − Ef ). we recover Fermi’s Golden Rule. This can be done eﬃciently if we can reduce the time-ordered products to normal-ordered products (which enjoy the relative simplicity of annihilating the vacuum).4 Wick’s Theorem We would like to evaluate the terms of the perturbation series obtained by expanding Dyson’s formula (3.4. Hence. This is a c-number (i. To get a sense of the meaning of the T symbol. we see that: A(t1 )B(t2 ) = A(−) . A(+) (3. which we will denote by an overbrace: A(t1 )B(t2 ) = T (A(t1 )B(t2 )) − : A(t1 )B(t2 ) : (3. B (+) if t1 > t2 and A(t1 )B(t2 ) = B (−) .24) Dividing A and B into their positive. however. . A(±) .25) if t1 < t2 .15). .23) 3. a† ] = 1. HI }. this implies that the transition rate is: dP = | f| dt ∞ −∞ dt HI (t)|i |2 (3.26) (3. We understand. it is an ordinary number which commutes with everything) since [a. in the interaction picture. To do this.and negative-frequency parts. Hence. WICK’S THEOREM 35 Squaring this. the operators are free and the states have complicated time-dependence). that this is a result of taking the limits of integration to inﬁnity carelessly: the square of the δfunction is actually a single δ-function multiplied by the diﬀerence between the initial and ﬁnal times. it is equal to its vacuum . To do this. This appears to lead to a transition probability which is proportional to the square of a δ-function. we deﬁne the notion of the contraction of free ﬁelds (remember that.e.22) with one δ-function dropped. we need to compute timeordered products T {HI HI . it is instructive to consider the second-order term in the expansion of the exponential: (−i)2 2! t t′ t dt1 t′ dt2 T (HI (t1 )HI (t2 )) = (−i)2 t t′ t1 dt1 t′ dt2 HI (t1 )HI (t2 ) (3. B (±) .

. un ) (+) (3. .j=1 2 ui uj ∂ ∂ ∂ui ∂uj u1 u2 . . . un } = : u1 u2 .36 CHAPTER 3. . . . . Suppose that it is true for time-ordered products of n − 1 ﬁelds. . It contains all possible terms with all possible contractions appear. . . . The proof proceeds by induction. . due to Gian-Carlo Wick. . . . . . . 2. un ) = u1 w (u2 . un ) + u1 w (u2 . Let us further suppose. Then. T {u1 u2 . un ) (+) (−) = u1 w (u2 . un ).29) (−) The equality between the last two lines follows from the fact that the ﬁnal expression is normal ordered and contains all possible contractions: the ﬁrst two terms contain all contractions in which u1 is not contracted while the third term contains all contractions in which u1 is contracted. . . Let us call the right-hand-side w(u1 u2 . un : + other terms with one contraction + : u1 u2 u3 u4 . . . . . uses the contraction to reduce time-ordered products to normal-ordered products: T {u1 u2 . un ) + w (u2 . + : u1 u2 . un−1 un : + other such terms if n is even + : u1 u2 . . . . un : + other terms with two contractions . . u2 . . . un : + : u1 u2 .27) The following theorem. un } = u1 w (u2 . . un : (3. . without loss of generality. . un ) u1 = w (u1 . .30) . . . . The equality of the left and right-hand sides is trivial for n = 1. each with coeﬃcient 1. . A concise way of writing down Wick’s theorem is the following: T {u1 u2 . .28) (−) + u1 . . PERTURBATION THEORY: INTERACTING PHONONS expectation value: A(t1 )B(t2 ) = 0| A(t1 )B(t2 ) |0 = 0|T (A(t1 )B(t2 )) |0 − 0| : A(t1 )B(t2 ) : |0 = 0|T (A(t1 )B(t2 )) |0 (3. . . . . un } = u1 T {u2 . . un−2 un−1 un : + other such terms if n is odd The right-hand-side is normal-ordered. . that t1 is the latest time. un } = : e 1 Pn i. . w (3.

31) T (uk (x1 .33) By Fourier transforming this equation. t1 ) (3.32) and acting on this with (2. t1 )uj (x2 . we can rewrite the phonon propagator as: T (ui (x1 . t2 )) = θ(t1 −t2 ) ui (x1 . the contraction of two phonon ﬁelds. t1 )uj (x2 . In the problem set. using ωp = vl p. t2 )ui (x1 . uj (−) You will also calculate it by noting that T (ui (x1 . t2 )) = ui (+) . t1 )uj (x2 .5 The Phonon Propagator As a result of Wick’s theorem. The contraction ui uj is also called the phonon propagator. t2 )) = 1 ρ i ki kj /k2 d3 p dω i(p·(x1 −x1 )−ω(t1 −t2 )) e (2π)3 2π ω 2 − vl2 p2 + iδ .34) s ǫs = δ . ui uj . t2 )) = 1 ρ 1 T (ui (x1 . you will compute the propagator in two diﬀerent ways. we have used ǫi j ij p are unresolved by this expression.5. t1 )uj (x2 . t2 )) = ρ iǫs ǫs d3 p dω i(p·(x1 −x2 )−ω(t1 −t2 )) i j e s (2π)3 2π ω 2 − (ωp )2 + iδ (3. t1 )uj (x2 . However. you will calculate it directly from: T (ui (x1 . the singularities at ω 2 = (ω s )2 Here. t1 )uj (x2 . First. 2 ρδik ∂t − (µ + λ) ∂i ∂k − µδik ∂l ∂l = −iδ (x1 − x2 ) δ (t1 − t2 ) δij (3. THE PHONON PROPAGATOR 37 3. t1 )uj (x2 . t2 ) +θ(t2 −t1 ) uj (x2 .35) Since ǫ1 ki = k while ǫ2.3) to obtain.3 ki = 0. ωp = vs p. As you will show in the problem set. the correct expression is: T (ui (x1 .3.3 1 Hence. we ﬁnd: iǫs ǫs d3 p dω i(p·(x1 −x1 )−ω(t1 −t2 )) i j e s (2π)3 2π ω 2 − (ωp )2 (3. Matrix elements of the time-evolution operator will be given by integrals of products of contractions.36) 2. is the basic building block of perturbation theory. i i ki kj k2 ki kj ǫ2 ǫ2 + ǫ3 ǫ3 = δij − 2 i j i j k ǫ1 ǫ1 = i j (3. t2 )) (3.

t2 ) : (3.1). l.6 Perturbation Theory in the Interaction Picture We are now in position to start looking at perturbation theory. l.1). t = ∞|U (−∞. PERTURBATION THEORY: INTERACTING PHONONS + 1 ρ d3 p dω i(p·(x1 −x1 )−ω(t1 −t2 )) i (δij − ki kj /k2 ) e 2 (2π)3 2π ω 2 − vt p2 + iδ (3. t1 ))4 (∂k uk (x2 . If we look at k3 . Since transverse phonons are unaﬀected by the interaction (3. it will be more convenient to consider a slightly diﬀerent phonon ﬁeld. we get such terms as: (−ig/4!)2 2! d3 x1 dt1 d3 x2 dt2 : ∂k uk ∂k uk ∂k uk ∂k uk (x1 . Consider the second-order contribution in our theory of phonons with a quartic anharmonicity (3.38 CHAPTER 3. l. U (−∞. k4 .41) This term will contribute to such physical processes as the scattering between two longitudinal phonons.37) For some purposes.s e−i(k·r−ωk t) i ρ (3. k2 . t2 )) = 1 ρ 1 = ρ s 2iωp ǫs ǫs d3 p dω i(p·(x1 −x1 )−ω(t1 −t2 )) i j e s (2π)3 2π ω 2 − (ωp )2 + iδ iǫs ǫs iǫs ǫs i j i j 3 i(p·(x1 −x1 )−ω(t1 −t2 )) + d pdω e s s ω − ωp + iδ ω + ωp − iδ (3.s ei(k·r−ωk t) + ak.39) 3. ∞) = ∞ (−i)2 ∞ dt2 T (HI (t1 )HI (t2 )) dt1 2! −∞ −∞ (−ig/4!)2 = d3 x1 dt1 d3 x2 dt2 T (∂k uk (x1 .40) When we apply Wick’s theorem. This ﬁeld has propagator: T (ui (x1 . ∞)|k1 . t2 ))4 2! (3. we only need to discuss longitudinal phonons. t1 ) ∂k uk ∂k uk ∂k uk ∂k uk (x2 . t = −∞ = . l. ϕi (r.38) s s The diﬀerence with ui is the missing 1/ 2ωk . t) = d3 p (2π)3 1 s s † √ ǫs ak. t1 )uj (x2 .

t2 ).and energyconservation. (3. t2 ) × 2! k3 . t1 ) and the outgoing phonons are (+) (+) created by the ∂k uk ∂k uk at (x2 .45) The x and t integrals give δ functions which enforce momentum. l. t2 ). l.. t1 )uk (x2 .43) |0 e l l −i((k3 +k4 )·x2 −(ωk +ωk )t2 ) 3 4 × (3. we ﬁnd: (−ig/4!)2 2! d3 x1 dt1 d3 x2 dt2 d3 p1 dω1 d3 p2 dω2 {|k1 | |k2 ||k3 | |k4 | × (2π)3 2π (2π)3 2π 3 e e 1 2 1 2 1 k1 k2 1 2 1 1 1 i i } |p1 |2 2 |p2 |2 2 l l ρ ω1 − (ωp1 )2 + iδ ρ ω2 − (ωp2 )2 + iδ i((k +k −p −p )·x −(ω l +ω l −ω −ω )t ) l l −i((k +k4 −p1 −p2 )·x2 −(ωk +ωk −ω1 −ω2 )t2 ) 3 4 × (3.46) 2 1 ω 2 − (ω l )2 + iδ 2 ω 2 − (ω l )2 + iδ ρ p1 p2 1 2 .42) . t = ∞| : ∂k uk ∂k uk (x1 . l. t1 ) ∂k uk ∂k uk (x2 .. Let’s suppose that the incoming phonons (−) (−) are annihilated by ∂k uk ∂k uk at (x1 . t2 ) : |k1 . k2 . t1 )∂k uk ∂k uk (x2 . PERTURBATION THEORY IN THE INTERACTION PICTURE 39 (−ig/4!)2 d3 x1 dt1 d3 x2 dt2 ∂k uk ∂k uk (x1 . t2 ) ∂k uk (x2 . t2 )} Substituting the expression for uk (x1 . t1 ) ∂k uk (x1 . t = −∞ = − |k1 | |k2 | e we obtain a contribution to (3.. k4 .42) of the form: (−ig/4!)2 2! e d3 x1 dt1 d3 x2 dt2 {|k1 | |k2 ||k3 | |k4 | × l l i((k1 +k2 )·x1 −(ωk +ωk )t1 ) 1 2 (−) (−) l l i(k1 ·x1 −ωk t1 ) i(k2 ·x1 −ωk t1 ) 1 e 2 (3. l. t1 )|k1 . (−ig/4!)2 2! d3 p1 dω1 d3 p2 dω2 {|k1 | |k2 ||k3 | |k4 | × (2π)3 2π (2π)3 2π l l (2π)3 δ(k1 + k2 − p1 − p2 ) 2πδ(ωk1 + ωk2 − ω1 − ω2 ) l l (2π)3 δ(k3 + k4 − p1 − p2 ) 2πδ(ωk3 + ωk4 − ω1 − ω2 ) i 1 i |p |2 |p |2 } (3. k2 .3. + this will give a non-vanishing contribution since two of the uncontracted ui ’s can annihilate the phonons in the initial state and the other two can create the phonons in the ﬁnal state. l.6. t1 )∂k uk (x2 . l.. Since ∂k uk ∂k uk (x1 . t = −∞ + .44) ∂k uk (x1 .

and the outgoing phonons are created at the same point.47) There are actually several ways in which an identical contribution can be obtained. t2 ). t1 ) and the outgoing phonons are annihilated by the ui ’s at (x2 . The incoming phonons are annihilated by the ui ’s at (x1 . ﬁnally. This gives an identical contribution. by contracting ﬁelds at the same point or by annihilating the incoming phonons at diﬀerent points. It is now clear why we included a factor of 1/4! in our deﬁnition of g: the above multiplicity almost cancels the two factors of 1/4!. By an identical contribution. There are 2 ways in which we can choose how the remaining ui ’s at (x1 . PERTURBATION THEORY: INTERACTING PHONONS which. t2 ).48) There are. If we permute (x1 . then the incoming phonons are annihilated by the ui ’s at (x2 . of course. gives us (−ig/4!)2 1 2! ρ2 ‘ d3 p1 dω1 {|k1 | |k2 ||k3 | |k4 | × (2π)3 2π i i |k1 + k2 − p1 |2 l |p1 |2 2 l − ω )2 − (ω l l )2 + iδ 2 ω1 − (ωp1 (ωk1 + ωk2 1 k1 +k2 −p1 ) + iδ l l l l (2π)3 δ(k1 + k2 − k3 − k4 ) 2πδ(ωk1 + ωk2 − ωk3 − ωk4 )} (3.40 CHAPTER 3. t1 ) and (x2 . Ther is a contirbution of the form: (−ig)2 1 2 ρ2 d3 p1 dω1 {|k1 | |k2 ||k3 | |k4 | × (2π)3 2π i i |p1 |2 2 |k1 − k3 − p1 |2 l l − ω )2 − (ω l l )2 + iδ 2 ω1 − (ωp1 (ωk1 − ωk3 1 k1 −k3 −p1 ) + iδ . the incoming phonons are annihilated at the same point (which can be either (x1 . distinct second-order contributions to the two phonon → two phonon transition amplitude which result. we mean one in which there are two contractions of the form uk (x1 . t2 ) and the outgoing phonons are annihilated by the ui ’s at (x1 . say. which can be done in (4 · 3)(4 · 3) ways. Consider the latter contributions. giving us a multiplicity of (4 · 3)(4 · 3)2. t1 ) or (x2 . thereby cancelling the 1/2! which we get at second-order. the sum of all such contributions is: (−ig)2 1 2 ρ2 d3 p1 dω1 {|k1 | |k2 ||k3 | |k4 | × (2π)3 2π i i |p1 |2 2 |k1 + k2 − p1 |2 l l − ω )2 − (ω l l 2 ω1 − (ωp1 )2 + iδ (ωk1 + ωk2 1 k1 +k2 −p1 ) + iδ l l l l (2π)3 δ(k1 + k2 − k3 − k4 ) 2πδ(ωk1 + ωk2 − ωk3 − ωk4 )} (3. t2 ) since these are dummy variables which are integrated over). t1 )uk (x2 . Hence. t2 ). t1 ) are contracted with the remaining ui ’s at (x2 . other. t2 ). Only a factor of 1/2 remains. t1 ).

Again. but they will give a diﬀerent contribution which is diﬀerent in form from the one above. + 2! ∂k uk (x1 .49) and one with k3 → k4 . ti )’s does not always occur. + 2! ∂k uk ∂k uk ∂k uk ∂k uk 0.. (3. ti )’s. t2 ) ∂k uk (x1 . an overall factor of 1/2!4!. there will be a factor. t1 )∂k uk (x2 . In the above. we will see that the symmetry factor. there are 2 ways of contracting the remaining ﬁelds at (x1 .3. t = ∞|0. there will be factors arising from the incomplete cancellation of the (1/4!)n . In addition. t2 ) 0. t = ∞|U (−∞. t = −∞ = (−ig/4!)2 d3 x1 dt1 d3 x2 dt2 ∂k uk (x1 . There are no distinct permutations. There are 4 · 3/2 ways of choosing the two ﬁelds at (x1 . so there is nothing to cancel the 1/2!. S = 2). t = −∞ + . t2 ) ∂k uk (x1 .. this additional factor is 1/4!. For instance. how this works at nth order.. t2 ) × . t = ∞|U (−∞. If this is incompletely cancelled by permutations of the (xi . Finally. ∞)|0. t = −∞ + . In the next chapter.. t1 ) ∂k uk (x2 . This multiplicity incompletely cancels the 1/2!(4!)2 . t1 ) which are contracted and 4 · 3/2 ways of choosing the two ﬁelds at (x2 . there are other second-order contributions to the vacuum-tovacuum amplitude which result from contracting ﬁelds at the same point.. t2 )∂k uk (x2 . 1/S (in the above.. and hence. S. The cancellation which we obtained by permuting the diﬀerent (xi . the following contraction at secondorder makes a contribution to the amplitude for the vacuum at t = −∞ to go into the vacuum at t = ∞: 0. We have written this term with contracted ﬁelds adjacent in order to avoid clutter.. In addition. There will be a factor of n!. t2 ) × . t2 ) which are contracted. t2 ). ∞)|0. One such is the following: 0. t1 )∂k uk (x1 . t = ∞|0..50) . so there is an uncancelled factor of 1/4! as well. Consider. t = −∞ = (−ig/4!)2 d3 x1 dt1 d3 x2 dt2 ∂k uk (x1 .51) (3. there are only 4! ways to do the contractions. t1 )∂k uk (x2 . for a moment. is related to the symmetries of Feynman diagrams. t1 ) with those at (x2 . t1 )∂k uk (x2 . PERTURBATION THEORY IN THE INTERACTION PICTURE 41 l l l l (2π)3 δ(k1 + k2 − k3 − k4 ) 2πδ(ωk1 + ωk2 − ωk3 − ωk4 )} (3.6. t1 )∂k uk (x2 .

This gives an overall factor of 4 · 4 · 3! · 2. Finally. consider (−ig/4!)2 2! d3 x1 dt1 d3 x2 dt2 : ∂k uk ∂k uk ∂k uk ∂k uk (x1 . t1 ) ∂k uk ∂k uk ∂k uk ∂k uk (x2 . t2 ) : (3. (x1 . which incompletely cancels the (1/2!) · (1/4!)2 . t2 ). t1 ) and (x2 . leaving 1/3!. t1 ) and annihilate the outgoing phonon at (x2 . t2 ) can be permuted.52) This contributes to the amplitude for processes in which both the initial and ﬁnal states contain one longitudinal phonon. . PERTURBATION THEORY: INTERACTING PHONONS As another example.42 CHAPTER 3. There are 4·4 ways of choosing the ui ’s which create the incoming phonon at (x1 . There are 3! ways of contracting the remaining ui ’s.

Each such vertex represents a factor of (∂k uk )4 .47) can be represented as: Given such a diagram – a Feynman diagram – you can immediately reconstruct the expression which it represents according to the following rules: • Assign a directed momentum and energy to each line.1 Feynman Diagrams Feynman introduced a diagrammatic notation which will help us systematically enumerate all of the perturbative contributions which we generate using Wick’s theorem. We will call such a line an internal line. This diagrammatic notation will have the added beneﬁt of having a simple physical interpretation which will guide our intuition about physical processes. We draw a diagram containing n vertices with 4 lines emanating from each vertex. The remaining (uncontracted) lines – external lines – represent incoming and outgoing phonons.CHAPTER 4 Feynman Diagrams and Green Functions 4. The ﬁrst contribution which we considered in chapter 3 (3. Suppose we want to construct a matrix element at nth order in perturbation theory. The lines emanating from the vertices can be connected. For external 43 . Each such connection represents a contraction. We will adopt the convention that incoming phonon lines enter at the left of the diagram while outgoing phonon lines exit at the right of the diagram.

we can ignore the transverse phonons – since they don’t interact – and consider only the longitudinal phonons. we assign a factor 1/S to the diagram. . for all vertices i. . jm (m ≤ 4) and to external momenta p1 . . . ω4 ) directed into the vertex. • For each external line with momentum k. . incoming and outgoing phonons.47). FEYNMAN DIAGRAMS AND GREEN FUNCTIONS k 1 k 3 k 2 Figure 4. respectively. p4−m . . write |k|. . n. . . energies (p1 . . For the particular interaction we have chosen. You can verify that by applying these rules to ﬁgure (4. ω write: 1 ρ i d3 p dω 2 |p| 2 (2π)3 2π ω − vl2 p2 + iδ • For each vertex with momenta. • For each internal line with momentum and energy p. . .44 CHAPTER 4. write: g (2π)3 δ(p1 + p2 + p3 + p4 ) 2πδ(ω1 + ω2 + ω3 + ω4 ) • Imagine labelling the vertices 1. 2. . Such a permutation leaves the diagram invariant if. Vertex i will be connected to vertices j1 . . . k 4 lines. i is still connected to vertices j1 . we assign a factor 1/l! to the diagram. the momentum is directed into or out of the diagram for. . . ω1 ).47). .1) we recover (3. If we were to consider a model in which both longitudinal .1: The diagram corresponding to (3. (p4 . jm (m ≤ 4) and to external momenta p1 . If S is the number of permutations which leave the diagram invariant. Consider a permutation of these labels. • If two vertices are connected by l lines. . p4−m . . . . . .

External lines correspond to ‘real phonons’ or simply phonons. k2 interact at x1 . propagate as virtual phonons . uj . For the diagram of ﬁgure 4.2) we display all of the connected diagrams which appear to O(g2 ) in the theory given by (3.. our Feynman diagrams would have to have internal and external indices corresponding to the vector indices of the ﬁelds ui . etc. In ﬁgure (4.4.1. In (b). we have the diagrams of order g. In (a).1) to O(g2 ). while internal lines correspond to ‘virtual phonons’.1.2: All connected Feynman diagrams for the theory (3. you will write down expressions for them. we say that the incoming phonons with momenta k1 . The Feynman diagram representation for transition amplitudes suggests a beautiful visualization of perturbative processes. and our Feynman rules would have to tell us how to contract or route these indices. and transverse phonons interact.1). FEYNMAN DIAGRAMS (a) 45 (b) Figure 4. we have the diagrams of order g2 . In the problem set.

The expansion in loops is actually an expansion in powers of Planck’s constant. k2 exchange a pair of virtual phonons. let’s imagine connecting all of the external lines at a single point so that the Feynman diagram deﬁnes a polyhedron with E + I edges. and there are no remaining integrals to be evaluated. as you will show in the problem set each propagator comes with a factor of and each vertex . Virtual phonons need not be ‘on-shell’.i. we say that the incoming phonons with momenta k1 . FEYNMAN DIAGRAMS AND GREEN FUNCTIONS with momenta p1 . and so on. The treelevel diagrams are indicated in ﬁgure 4. k4 .minus the number of vertices. and ﬁfth have two loops.with E faces formed as a result of connecting the external lines. further. while the third. Then. Indeed. I internal lines.3 These diagrams can be evaluated without doing any integrals.the extra vertex being the one at which the external lines are connected. each coming with momentum and energy-conserving δ-functions. i. 4. These satisfy ω 2 = (ωp )2 . thereby signalling a resonance. to evaluate a two-loop diagram we need to do two such integrals. dω1 d3 p1 dω2 d3 p2 . energies. energy l (p. and L + E faces . In short.). the ﬁrst diagram in ﬁgure 4.46 CHAPTER 4. ω) pairs to be integrated over as there are loops. Suppose. thereby scattering into k3 . Note that most of these are not connected diagrams. For the ﬁrst diagram of ﬁgure 4. External lines correspond to initial or ﬁnal states with phonons of momentum.2 has one loop. In order to evaluate a one-loop diagram.2b.g.e. Such a phonon is said to be ‘on-shell’. k1 + k2 − p1 .e. ω). V + 1 vertices .2 Loop Integrals Suppose we have a Feynman diagram with E external lines. fourth. thereby scattering into the outgoing phonons with momenta k3 . the number of propagators. that this diagram has L loops (e. and V vertices. each coming with an integral .1) or. since. k4 . A diagram with no loops has as many δ-functions as integrals. The second has no loops.2) The number of loops is given by the number of internal lines . Such a diagram is said to be a tree level diagram. (# f aces) + (# vertices) − (# edges) = 2 (4. the phonon propagator diverges if a virtual phonon is on-shell. According to Euler’s fomula. there are as many (p. plus 1 for the overall δ-functions satisﬁed by the external momenta. and ﬁnally interact again at x2 . so the integrals can all be evaluated trivially. L=I −V +1 (4. we need to do one dωd3 p integral.

4.2. LOOP INTEGRALS 47 (a) (b) (c) + . .3: The tree-level Feynman diagrams of the theory (3... Figure 4.1).

FEYNMAN DIAGRAMS AND GREEN FUNCTIONS comes with a factor of 1/ .48 CHAPTER 4. we will use the following formula: ∞ Γ(α) = dt tα−1 e−At (4. Changing variables from ti to xi according to tj = sxj .1. Turning now to the evaluation of multi-loop diagrams.you guessed it . These are more easily evaluated if we combine the denominators of the propagators using the following formula: 1 dx 0 1 [ax + b(1 − x)] 2 = 1 ab (4.4) Aα 0 Using this formula.7) . dxn δ(1 − xj ) j j xj Aj αj −1 jx P αj (4.Feynman) very useful. we have: 1 αj = j Aj = j ∞ 0 tj ) (4.6) To see why these formulas are useful. we ﬁnd the following trick (due to . which can occur in higher orders of perturbation theory.3) For the more general case of the product of several propagators. the integral in brackets is equal to 1. An L-loop diagram comes with a coeﬃcient of L−1 . we can write: 1 αj = j Aj 1 Γ(αj ) ∞ 0 dtj tj j α −1 −Aj tj e j ∞ 0 ds δ(s − j Here. We have g2 2ρ dω1 d3 p1 2π (2π)3 |k1 + k2 − p1 |2 2 (ǫ1 + ǫ2 − ω1 )2 − vl2 k1 + k2 − p1 + iδ |p1 |2 2 ω1 − vl2 p2 + iδ 1 (4. . consider the evaluation of diagram 4.5) ds j 1 Γ(αj ) 1 0 ∞ 0 dxj sαj xj j ∞ 0 α −1 −sAj xj e 1 δ(1 − s P j Aj x j xj ) j 1 Γ(αj ) j αj j Γ(αj ) dxj xj j 1 0 α −1 ds s P j αj −1 e−s δ(1 − xj ) j Γ = dx1 . When we integrate the momenta in closed loops. . we often encounter integrals of products of propagators.

If we’re careful and we remember that these integrals are cutoﬀ in the ultraviolet. ωi ≪ Λω . This is a real. On the other hand. i.e. physical eﬀect: phonon Green functions do diverge when the . In condensed matter physics. q = p1 − x(k1 + k2 ). ki ≪ Λ. Λ. ki vanish. is the inverse lattice spacing and the frequency cutoﬀ. then these integrals are ultraviolet divergent. then we will get ﬁnite (albeit cutoﬀ-dependent) answers.4. is the Debye energy. in our phonon theory. Thus. these integrals are always cutoﬀ. we can shift the variables of integration and neglect the eﬀect of this shift on the range of integration.3) with a = (ǫ1 + ǫ2 − ω1 )2 − vl2 k1 + k2 − p1 b= 2 ω1 2 49 + iδ (4. so we must be a little more careful.e. we proceed by changing the variables of integration to ω = ω1 − x(ǫ1 + ǫ2 ). if ǫi . Λω .9) Integrals of this form are often divergent. Writing a = x(1 − x)((ǫ1 + ǫ2 )2 − vl2 (k1 + k2 )2 ) we can write the loop integral as: q4 dω d3 q 2π (2π)3 ω 2 − v 2 q 2 + a l 2 + + dω d3 q x2 (1 − x)2 (k1 + k2 )4 2 2π (2π)3 ω 2 − vl2 q 2 + a 4 dω d3 q q 2 (k1 + k2 )2 x2 + (1 − x)2 + 3 x(1 − x) 2 2π (2π)3 ω2 − v2 q 2 + a l (4. we can write |k1 + k2 − p1 |2 |p1 |2 2 ω1 − vl2 p12 + iδ |p1 |2 |k1 + k2 − p1 |2 2 2 + iδ x + ω1 − vl2 p12 + iδ (1 − x) 2 2 (ǫ1 + ǫ2 − ω1 )2 − vl2 k1 + k2 1 2 + iδ = 0 dx 1 (ǫ1 + ǫ2 − ω1 )2 − vl2 k1 + k2 − p1 dx = 0 2 ω1 − vl2 p12 + x((ǫ1 + ǫ2 )2 − vl2 p12 ) − 2xω1 (ǫ1 + ǫ2 ) + 2xvl2 p1 · k1 + k2 − p1 + iδ |p1 |2 |k1 + k2 − p1 |2 If these integrals were from −∞ to ∞.3). However. so long as the external momenta and energies are much smaller than the cutoﬀs. then we could shift the variables of integration without worrying. these integrals are infrared divergent if a = 0 – i.8) − vl2 p12 + iδ Using (4. LOOP INTEGRALS This can be brought into a more useful form using (4.2. If we forget about the momentum and frequency cutoﬀs. the momentum cutoﬀ.

energies tend to zero. Λ.9) is equal to i vl7 d4 q i q4 + 5 4 2 + a)2 (2π) (−q vl i + 3 vl 3 d4 q 4 q 2 (k1 + k2 )2 x2 + (1 − x)2 + 4 x(1 − x) 3 (2π)4 (−q 2 + a)2 d4 q x2 (1 − x)2 (k1 + k2 )4 (4.0 (a) (4.12) or. we can simplify (4. and V (S 3 ) = 2π 2 . If we write q4 = −iω and q 2 = q4 + vl2 q 2 .1 (a) 3 vl7 vl 4 1 + 3 x2 (1 − x)2 (k1 + k2 )4 I2.50 CHAPTER 4.9) by using analytic continuation. For either sign of |q|2 − a. It will sometimes be important to distinguish between the frequency cutoﬀ. Hence. Often. however.10) (2π)4 (−q 2 + a)2 or 4 1 3 1 I2. Λω = vl Λ.m (a) = Λ z m+1 dz i 2 16π 0 (−z + a)n (−1)n dn−1 i = n−1 (n − 1)! da 16π 2 2 Λ2 0 z m+1 dz (−z + a) (u + a)m+1 du u =− =− (−1)n dn−1 (n − 1)! dan−1 (−1)n i 16π 2 dn−1 Λ2 −a −a m+1 k=0 i 16π 2 (n − 1)! dan−1 m + 1 k−1 m+1−k u a k m+1 k=0 i (−1)n dn−1 =− am+1 ln u + 16π 2 (n − 1)! dan−1 1 m + 1 k m+1−k u a k k Λ2 −a −a . In. then (4. the poles in ω are in the second and fourth quadrants. We will study the power-law forms of these divergences when we turn to the renormalization group in chapter 11.11) vl where the integrals which we need to study are: In.m (a) = i d4 q q 2m (2π)4 (−q 2 + a)n (4.2 (a) + 5 (k1 + k2 )2 x2 + (1 − x)2 + x(1 − x) I2. setting z = q 2 . Λω . and the momentum cutoﬀ. the distinction is unimportant and we can assume that there is a single cutoﬀ. In such a case. FEYNMAN DIAGRAMS AND GREEN FUNCTIONS momenta. the contour of integration can be harmlessly rotated in an anti-counter-clockwise direction from the real ω axis to the imaginary ω 2 axis.

however. • Combine the denominators using Feynman’s trick.3 Green Functions In the preceding discussion. One reason is that we may want to deﬁne a Feynman diagram – or .4. It does. GREEN FUNCTIONS i (−1)n dn−1 m+1 Λ2 − a (a ln + 16π 2 (n − 1)! dan−1 −a m+1 1 m + 1 m+1−k k a Λ2 − a − (−a)k ) k k k=0 51 =− (4. i. 4.e. they satisfy ω 2 = (ωp )2 . • Use rotational invariance to reduce the integral to an integral over a single variable for each ω. • Analytically continue to imaginary frequencies. we have implicitly assumed that external phonon l lines are ‘on shell’. • Shift the variables of integration to make the denominator invariant under ω → ω. p → −p. p. we evaluate a Feynman diagram by the following steps: • Use the Feynman rules to obtain a loop integral. we ﬁnally obtain: − i 16π 2 1 dx[ 0 1 1 2 2 4 {− Λ4 − 3Λ2 x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) 7 x (1 − x) (k1 + k2 ) 2 vl − x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) 2 2 +3 x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) + ln Λ2 1 3 4 2 x2 + (1 − x)2 + x(1 − x) {Λ2 + x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) 5 4 (k1 + k2 ) 3 vl Λ2 } −x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) ln x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) + 1 x2 (1 − x)2 (k1 + k2 )4 vl3 ln Λ2 x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) −1 ] (4. make sense to relax this requirement and allow even the external phonons to be “oﬀshell”.13) Hence.14) x(1 − x)(vl2 (k1 + k2 )2 − (ǫ1 + ǫ2 )2 ) } To summarize.3.

The shaded circle represents all possible diagrams with 4 external legs.4a.4b. p3 . . Consider the diagram of ﬁgure 4. p2 . .16) . pn ) (4. ω. FEYNMAN DIAGRAMS AND GREEN FUNCTIONS = +. p2 . p2 .. p3 .) G(p1 .52 CHAPTER 4.4: The deﬁnition of the 4 point Green function a set of diagrams – which can be part of a larger diagram. We will call such an object G(p1 . In such a case.15) (We will use p as a shorthand for p.. p4 ) (4. . We can deﬁne similar objects – called Green functions – for any number of external legs: G(p1 . The ﬁrst few are shown in ﬁgure 4. . the lines which enter this part might not be on-shell. p4 ) is deﬁned to include the momentum conserving δ functions.+ + + + Figure 4. (2π)3 δ(p1 + p2 + p3 + p4 ) 2πδ(ω1 + ω2 + ω3 + ω4 ) and a propagator |pi |2 i 2 ωi − vl2 p2 + iδ i on each external leg.

j. we can study the propagation of sound waves. x2 ) = ∂i ∂j T (ui (x1 .e.19) as you showed in the ﬁrst problem set. j . x2 ) = iδ (x1 − x2 ) δ (t1 − t2 ) (4. is some arbitrary.17) This follows since the two-point Green function is just the derivative of the propagator: G(x1 . t) as a knob which we can turn in order to set up compressional waves in the solid. prescribed function. pn with a propagator assigned to each external leg. t) (4.4 The Generating Functional Let’s modify our Hamiltonian by adding a ‘source term’. We can Fourier transform the Green function to obtain the real-space n-point Green function. . . . so our Feynman rules must be expanded to include a new vertex – which we will call a ‘source vertex’ –with only one line emerging from it. . . which is more natural in some contexts. the two-point Green function is a Green function of the diﬀerential operator 2 ρδik ∂t − (µ + λ) ∂i ∂k − µδik ∂j ∂j (4. p2 . energy p. The name ‘Green function’ is due to the fact that when the interaction is turned oﬀ. the phonon two-point Green function is often denoted D(x1 . xn ). t1 )uj (x2 .4. i. x2 ). . t)∂k uk (x. we assign −i ˜ ω) to it (˜ is the Fourier transform j(p.20) The source. g = 0. 4. G(x1 . While the notation G(x1 . . H→H+ d3 x j(x.4. However. t2 )) iǫs ǫs d3 p dω i(p·(x1 −x1 )−ω(t1 −t2 )) 1 i j e (4. Our interaction Hamiltonian is now HI + d3 x j(x)∂k uk (x). we will reserve this notation for the two-point function of the other phonon ﬁeld ϕi . We can think of j(x. If this line has momentum. xn ) is generically used for Green functions. THE GENERATING FUNCTIONAL 53 It is given by the sum of all diagrams with n external legs with (possibly oﬀ-shell) momenta and energies p1 . t′ ). . . . By measuring the system at (x′ . . ω.18) = ∂i ∂j 3 2π 2 − (ω s )2 + iδ ρ (2π) ω p It therefore satisﬁes 2 ρδik ∂t − (µ + λ) ∂i ∂k − µδik ∂j ∂j G(x1 . .

22) Hence. Let us now look at the vacuum-to-vacuum amplitude. . pn ) (4.5). Several of these are shown in ﬁgure (4.54 CHAPTER 4. . We then have to integrate over all of the pi ’s with a factor of 1/n! to avoid overcounting due to permutations of the pi ’s. . d3 pn dωn j(p1 ) .23) . Hence. H → H + E0 and choose E0 so that: Z[0] = 1 (4. . pn ). . FEYNMAN DIAGRAMS AND GREEN FUNCTIONS 11 00 11 j00 11 00 11 00 j 11 j 00 11 00 11 j00 11 00 11 00 j 11 00 11 00 11 00 j 11 j 00 11 00 11 00 11 00 j Figure 4. ω1 ) . j(pn ) G(p1 . the sum of all of the diagrams with no source vertices is 1. . . which we will call Z[j]: Z[j] = 0|T e−i R HI +j∂k uk |0 (4. .5: Some vacuum-to-vacuum diagrams in the presence of an external source. . . . . Each ˜ i . Consider a diagram with n source vertices. It will have an amplitude proportional to ˜ 1 . To make life easy. ωn ).21) This is given by the sum of all diagrams with no external legs. of j). . ˜ n . ωi ) creates a phonon with momentum j(p j(p j(p pi . ωi . let us shift the zero of energy by adding a constant to the Hamiltonian. These n phonons enter the diagram along the external legs of the Green function G(p1 . We have denoted the new vertex by a large dot with a j next to it. . Z[j] = 1 + (−i)n n! n=1 ∞ d3 p1 dω1 .

Using Dyson’s formula.4.t)∂k uH (x. .4. ∂k uH (xn ) |0 = k k . we can express Z[j] as: Z[j] = 0|T e−i =1+ ∞ n=1 R d3 x dt j(x. THE GENERATING FUNCTIONAL We can Fourier transform this expression into real space: Z[j] = 1 + (−i)n n! n=1 ∞ 55 d3 x1 dt1 . . j(xn ) G(x1 . d3 xn dtn j(x1 ) . . . the present deﬁnition as a vacuum expectation value is far more general since it is non-perturbative and can be compared with experiments. . . . The ‘interaction’ representation for this ‘free’ Hamiltonian is actually what we would call the Heisenberg representation for j = 0. xn ) (4. .26) Comparing this with our earlier expression for Z[j]. H → H + j∂k uk = H0 + H ′ + j∂k uk = (H0 + H ′ ) + j∂k uk (4.t) k (−i)n n! |0 d3 x1 dt1 . . These vacuum expectation values are also called timeordered correlation functions. G(x1 . so we will put an H superscript on all ﬁelds. we can’t use Wick’s theorem. .24) We can understand the Green function in another way by considering the Hamiltonian with a source term. .25) We can now treat H0 + H ′ as our ‘free’ Hamiltonian and j∂k uk as our interaction Hamiltonian. d3 xn dtn j(x1 ) . . . .27) In other words. . . While our earlier deﬁnition – as a sum of Feynman diagrams – is convenient for perturbative calculation. . we see that the Green function is given by: δn Z[j] δj(x1 ) . . . . ∂k uH (xn ) |0 k k (4. . the Green functions are the vacuum expectation values of the T -ordered product of a string of (Heisenberg picture) ﬁelds. but we we can still use Dyson’s formula. . Since H0 + H ′ is not actually free. δj(xn ) (4. xn ) = 0|T ∂k uH (x1 ) . These vacuum expectation values are the coeﬃcients of the Taylor expansion of the vacuumto-vacuum transition amplitude in the presence of an external source ﬁeld. j(xn ) 0|T ∂k uH (x1 ) . x2 .

.29) The nr ! in the denominator is the symmetry factor resulting from the permutations of the nr identical connected components of type r.56 CHAPTER 4.30) This theorem – sometimes called the linked cluster theorem – will be particularly useful when we construct a diagrammatic expansion for the partition function. in general.28) where Z[j] is the sum of vacuum-to-vacuum Feynman diagrams we deﬁned above and W [j] is the sum of connected vacuum-to-vacuum diagrams. . etc. The Taylor expansion of W [j] is: W [j] = W [0]+ (−i)n n! n=1 ∞ d3 x1 dt1 .. j(xn ) Gc (x1 .31) . Commuting the sums and the product: Z[j] = = ∞ r=1 ∞ ∞ nr =0 n Cr r nr ! eCr = e r=1 Cr = eW [j] r=1 P∞ (4. . FEYNMAN DIAGRAMS AND GREEN FUNCTIONS 4. ∞ r=1 n1 =0 n2 =0 n Cr r nr ! (4. of its connected components: Z[j] = ∞ ∞ . We can assemble the set of all vacuum-to-vacuum diagrams by putting together n1 connected diagrams of type 1. observe that a given diagram which contributes to Z[j] will. be made up of several diﬀerent connected diagrams and it will factor into the contributions from each of them.5 Connected Diagrams There is a very useful theorem which states that Z[j] = eW [j] (4. To prove this theorem. n2 connected diagrams of type 2. d3 xn dtn j(x1 ) . . F = −T ln Z. In order to compute the free energy. . nr connected diagrams of type r.. Cr . . Z = tr(e−βH ). . we need only compute the connected diagrams. The contribution of such a diagram to Z[j] will be the product of the contributions. . xn ) (4. .. . .

ω) = dt d3 P dE |P |2 B(P . t)| i = eipi ·x−iωi t 0 |∂i ui (0.34) | i |∂j uj (0.6. 0)| i i |∂i ui (x1 . 0)| 0 |2 δ(P − pi ) δ(ωi − E) + θ(−t)e−i(P ·x−Et) = × ≡ θ(t)e i i(P ·x−Et) d3 P dE |P |2 B(P .33) By translational invariance.32) |0 This correlation function is often more useful since it measures ﬂuctuations around mean values.6 Spectral Representation of the Two-Point Green function The spectral representation of the two-point Green function has the advantage of being intuitive yet well-suited for rigorous statements. 0)∂i ui (x.35) Here. E) θ(t)ei(P ·x+(ω−E+iδ)t) + θ(−t)e−i(P ·x−(ω+E−iδ)t) . 0)| 0 |2 θ(t)ei(pi ·x−ωi t) + θ(−t)e−i(pi ·x−ωi t) d3 P dE | i |∂j uj (0. 0) + θ(−t) ∂j uj (0. t1 )| 0 (4. B(P . 4. E) θ(t)ei(P ·x−Et + θ(−t)e−i(P ·x−Et (4. 0)| i where pi and ωi are the momentum and energy of the state |i . so that |P |2 B(P . The two-point connected Green function is given by: Gc (x1 . SPECTRAL REPRESENTATION OF THE TWO-POINT GREEN FUNCTION 57 where the Gc ’s are connected Green functions. Hence. so that we obtain: G(x. t)uj (0. 0)) = θ(t) ∂i ui (x. 0 |∂i ui (x.4. It is obtained by inserting a complete set of states into T (∂i ui (x. t)uj (0. we have introduced the spectral function. t)∂j uj (0. t)| i i |∂j uj (0. E) is given by the quantity in brackets. E). 0)) = i (4. In order to take the Fourier transform. we have to add iδs to make the t integral convergent. 0)| 0 i + θ(−t) i 0 |∂j uj (0. we can write the Green function as T (∂i ui (x. x2 ) = 0|T (∂k uk (x1 )∂k uk (x2 )) |0 − 0|∂k uk (x1 )|0 0|∂k uk (x2 )|0 = 0|T ((∂k uk (x1 ) − ∂k uk (x1 ) ) (∂k uk (x2 )) − ∂k uk (x2 ) )(4. t) = θ(t) 0 |∂i ui (x.

t1 ) − 0|ui (x1 . E) = Zδ(E 2 − ωp ) + Binc (p. We could shift ui (x1 . t1 )| 0 = 0 (4. E) is non-negative. E) θ(t)ei(ω−E)t−δt eiP ·x + θ(−t)ei(ω+E)t+δt e−iP ·x e−iP ·x eiP ·x − ω − E + iδ ω + E − iδ (4. B(p. this would be the statement that there is some kind of static distortion present in the ground state. so G(p. 0)| p |2 (4. E) = δ(E 2 −ωp ). B(p. If the phonons are non2 interacting. E) − ω − E + iδ ω + E − iδ (4. and the remaining ‘incoherent’ weight: 2 B(p.41) The phonon propagates as a free phonon with probability Z and as a multiphonon state of energy E with probability Binc (p. carrying weight Z. We can split the sum over i into the vacuum state. E). we will write: |p|2 Z = | 0 |∂i ui (0. E) B(−p. Now. . ω) = dE i|p|2 2E B(p. B(p.40) Rotational and Galilean invariance imply that the left-hand-side is independent of the direction of p. which looks like a non-interacting phonon. E) (4.36) d3 P dE i|p|2 B(p. t1 ) → ui (x1 .39) If it didn’t. Then.58 CHAPTER 4. the one-phonon states. t1 )|0 and we would have the above for our new displacement ﬁeld. FEYNMAN DIAGRAMS AND GREEN FUNCTIONS = = or G(p. let us assume that 0 |ui (x1 .38) From its deﬁnition. ω) = dE i|p|2 B(p. E). and all other states. and we recover the free-phonon two-point function.37) dt d3 P dE |P |2 B(P . E) For a parity-invariant system. E) = B(−p. Consider a one-phonon state of momentum p. t1 ) by ui (x1 . E) ω 2 − E 2 + iδ (4. Then the spectral function can be broken into a piece.

There is an O(g) correction given by the diagram of ﬁgure (4.. p2 ) = (2π)3 δ(p1 + p2 ) 2πδ(ω1 + ω2 ) |p1 |2 2 ω1 − vl2 p2 1 i − Π(p1 . p2 ) = (2π)3 δ(p1 + p2 ) 2πδ(ω1 + ω2 ) ( |p1 |2 + g i |p1 |2 2 2 ω1 − vl2 p2 + iδ 1 2 i 2 ω1 − + iδ i dω 2 |p| 2 + O(g2 )) (2π)3 2π ω − vl2 p2 + iδ (4. To zeroth order in g. Let us deﬁne the one-particle irreducible.42) d3 p vl2 p2 1 For the two-point Green function. ω) is called the self-energy because the two-point Green function can be expressed in terms of it according to the graphical relation of ﬁgure 4.43) . THE SELF-ENERGY AND IRREDUCIBLE VERTEX 59 Π= G= + + + . ω)/p2 . or 1PI n-point Green function as the sum of all the Feynman graphs contributing to the n-point Green function which cannot be made disconnected by cutting a single internal line (this is a subset of the set of connected diagrams).. ω1 ) + iδ (4.4.7 The Self-Energy and Irreducible Vertex The two-point Green function G(p1 . Figure 4.+ .. 4. Π(p. For the 1PI n-point function. we have: G(p1 . p2 ) is given by the diagrams in ﬁgure (??).6: The relation between Π and G.. we do not include propagators on the external legs. we can do better without doing much more work. it is simply the free phonon propagator. + .6. Summing this geometrical series.2) which leads to G(p1 .7. The 1PI two-point Green function is given by Π(p...

is given by: Z −1 = 1 − ∂ Re(Π) ∂ω 2 (4. p4 ). ω) = g 2 |p| 2 d3 q dǫ 2 i 2 |q| 2 2 q 2 + iδ + O(g ) 3 2π (2π) ǫ − vl (4. p2 ) according to the graphical relation of ﬁgure 4. p2 . The full Green function G(p1 . Γ(p1 .44) In the problem set. p3 . we see that the self-energy is given by: Π(p. p3 . FEYNMAN DIAGRAMS AND GREEN FUNCTIONS Figure 4. Z.7: The relation between the regular and 1PI four point Green functions.45) ω=vp We can also deﬁne a 1PI 4-point Green function. p3 . . From our calculation above. ω)} is related to the phonon lifetime. you will show that Im{Π(p. The coherent weight in the phonon spectral function. p2 . p4 ) and the two-point Green function G(p1 .60 CHAPTER 4. p2 .7. p4 ) can be expressed in terms of Γ(p1 .

As we will demonstrate in this chapter. ω → iω. We make the analytic continuation it → τ and deﬁne the following object for 0 < τ < β: G(x − x′ . τ ′ )∂k uk (x. τ )∂j uj (x′ . it is extremely convenient to work with imaginary-time from the outset. we found that the mathematical trick of analytically continuing to imaginary frequencies. τ ′ ) 61 +θ(τ ′ − τ )T r e−βH ∂j uj (x′ . in many cases. τ ) (5. Such an imaginary-time formalism will have the advantage of having a natural extension to arbitrary temperature. It can also.1 Finite-Temperature Imaginary-Time Green Functions In the previous chapter. τ − τ ′ ) = θ(τ − τ ′ )T r e−βH eτ H ∂k uk (x)e−τ H eτ +θ(τ ′ − τ )T r e−βH eτ ′H ′H ∂j uj (x′ )e−τ ′H ∂j uj (x′ )e−τ ′H eτ H ∂k uk (x)e−τ H = θ(τ − τ ′ )T r e−βH ∂k uk (x.1) . serve as a preliminary step in the calculation of retarded correlation functions which – as we will discuss in the next chapter – are the quantities most closely related to physical measurements. τ ′ ) ≡ Tτ ∂k uk (x.CHAPTER 5 Imaginary-Time Formalism 5. τ )∂j uj (x′ . facilitated the calculation of the integrals arising from Feynman diagrams.

0)}. Z[j] = T r{e−βH } = T r{U (β. If τ1 > τ2 > . In a similar way. Tτ . In the perturbative expansion of U (β. weighted by e−βEn . by analogy with the real-time symbol. |n . We will also exploit the formal analogy between the time-evolution operator e−itH → e−τ H and the Boltzmann weight. Since 0 < τ. In the ﬁnal line. .3) We will add a source to the Hamiltonian and view the partition function. If we had used it rather than τ . we would have the ﬁnite-temperature timeordered Green function. then Tτ (O1 . The ﬁnal equality follows from the fact that τ − τ ′ + β > 0. . On ) = T r e−βH O1 . > τn . β]: β G(iωn ) = 0 dτ eiωn τ G(τ ) (5. IMAGINARY-TIME FORMALISM We have passed from the Schr¨dinger representation in the ﬁrst line to an o imaginary-time Heisenberg representation in the third line. Now suppose that τ < τ ′ . we will only encounter ﬁelds with imaginary-time arguments in the interval [0. . τ1 ) = e−(τ2 −τ1 )H (5. τ ′ < β. 0). it follows that −β < τ − τ ′ < β. we can deﬁne the imaginary-time-ordered product of strings of ﬁelds. we can take the Fourier transform over the interval [0. . .2) Rather than take the expectation value in the gound state. as the generating functional for imaginary-time Green functions.5) . we will construct a Green function from which the retarded Green function can be constructed by analytic continuation.4) The ﬁrst equality follows from the cyclic property of the trace. e−βH . In analogy wth the real-time case. we write U (τ2 . G(τ − τ ′ < 0) = T r e−βH eτ ′H ∂j uj (x′ )e−τ ′H eτ H ∂k uk (x)e−τ H ′H = T r eτ H ∂k uk (x)e−τ H e−βH eτ ∂j uj (x′ )e−τ ′H ′H ′H = T r e−βH eβH eτ H ∂k uk (x)e−τ H e−βH eτ = G(τ − τ ′ + β) ∂j uj (x′ )e−τ (5. T : operators are arranged from right to left in order of increasing τ . β]. . There is a further condition which follows from the cyclic property of the trace. we are averaging the expectation value over all energy eigenstates. As a result of periodicity in imaginary-time. we have deﬁned the imaginary-time-ordering symbol.62 CHAPTER 5. Then. On (5. By working with τ .

τ ) = θ(τ ) T r e−βH0 ∂k uk (x. τ ) n e−ik·x−iωn τ 2 ωn + vl2 k2 (5. τ ) = − 1 β d3 k k (2π)3 2 |p|2 = 2ωk |p|2 = 2ωk dτ eiωn τ G(x. τ )∂j uj (0. we have: G(x. T r e−βH0 a† ak k = nB (ωk ) = 1 eβωk − 1 (5. τ ) 2 d3 k k [θ(τ ) (nB (ωk ) + 1)eik·x−ωk τ + nB (ωk )e−ik·x+ωk τ 3 2ω (2π) k (5. iωn ) = d3 xeip·x 0 β (nB (ωp ) + 1)(e−βωp − 1) nB (ωp )(eβωp − 1) + iωn − ωp iωn + ωp 1 −1 + iωn − ωp iωn + ωp 1 2 = − |p| 2 2 ωn + ωp 1 = − |p|2 2 (5. are given by: ωn = 2nπ β 63 (5.6) Inverting the Fourier transform. FINITE-TEMPERATURE IMAGINARY-TIME GREEN FUNCTIONS where the Matsubara frequencies ωn .10) ωn + vl2 p2 In real-space. we have: G(τ ) = 1 β G(iωn ) e−iωn τ (5. 0) +θ(−τ ) T r e−βH0 ∂j uj (0.7) n In the absence of interactions.11) .8) and substituting the mode expansion of uk . 0)∂k uk (x. we can evaluate the imaginary-time twopoint Green function directly.9) +θ(−τ ) nB (ωk )eik·x−ωk τ + (nB (ωk ) + 1)e−ik·x+ωk τ ] = We can now compute the Fourier representation of the Green function: G(p. Using the Planck distribution.5. this is: G(x.1.

2 Perturbation Theory in Imaginary Time Following our real-time development. as G(x.64 CHAPTER 5. τ − τ ′ ) = T r e−βH0 Tτ U (β. 0) U −1 (τ ′ . 0)∂k uk (x. τ ′ )∂k uk (x)U (τ. only imaginary times τ ∈ [0. U (τ1 . 0)∂j uj (0.12) Hence. 0) U −1 (τ. we have the decaying exponential e−ωτ rather than the oscillatory eiωt . we expand U (β. 0)× U −1 (τ. 0) (5. sadly. 0)} ′ = θ(τ − τ )T r{e−βH0 U (β. 5. 0)∂k uk (x.15) As we noted earlier. we consider a Hamiltonian of the form H = H0 + Hint. 0) ∂k uk (x. . as in the real-time case: G(x. IMAGINARY-TIME FORMALISM As we can see from the above derivation of the Green function. In this representation. τ ′ ) U (τ ′ . 0)} ′ +θ(τ − τ )T r{e−βH0 U (β. 0). τ − τ ′ ) = (−1)n n! n=0 ∞ β β . imaginary-time is simpler than real-time. τ2 ) = Tτ e − Rτ 1 dτ H int (τ ) τ2 (5. To evaluate this perturbatively. in the imaginary-time formalism. τ ) U (τ. τ ′ )U (τ ′ . so we can dispense with the iδ’s which are needed to ensure convergence in the real-time formulation. 0)∂j uj (0. τ )U (τ.13) eτ H ∂k uk (x)e−τ H can be written in the interaction picture as: G(x. τ ′ ) ∂j uj (0. . τ − τ ′ ) = θ(τ − τ ′ )T r{e−βH0 U (β. dτn .. 0) ∂k uk (x. 0)× U −1 (τ ′ . τ ′ ) (5.14) or. τ ) ∂j uj (0. the imaginary-time Green function . and go to the imaginary-time interaction representation. τ ′ ) ≡ Tτ U (β. 0)} +θ(τ ′ − τ )T r e−βH0 U (β. 0 0 dτ1 . τ ) ∂k uk (x. simply. Indeed. τ − τ ′ ) = θ(τ − τ ′ )T r e−βH eτ H ∂k uk (x)e−τ H eτ +θ(τ ′ − τ )T r e−βH eτ ′H ′H ∂j uj (0)e−τ ′H ∂j uj (0) e−τ ′H (5. τ ′ ) U (τ ′ .. from a mathematical point of view. which takes the Schr¨dinger o picture form: G(x. β] appear. τ ′ ) ∂j uj (0) U (τ. nature forces us to live in real-time. τ ) ∂j uj (0. τ ) U (τ.

• The propagator assigned to each internal line is: − 1 β d3 p 1 |p|2 2 3 (2π) ωn + vl2 p2 n • For each vertex with momenta. 0)}] = All connected diagrams with no external legs (5. p4−m . . Using our result on connected Feynman diagrams from chapter 5. . we write g (2π)3 δ(p1 + p2 + p3 + p4 ) β δn1 +n2 +n3 +n4 . . . . . . • To each line. . . Matsubara frequencies (p1 . Consider a permutation of these labels. . . ωn . . we assign a factor 1/S to the diagram. (p4 . .16) We can show that Wick’s theorem holds for the imaginary-time-ordered product and use it to evaluate the above expectation values. .5. • If two vertices are connected by l lines. F . Such a permutation leaves the diagram invariant if. . we can use Feynman diagrams to evaluate the perturbation series. we associate a momentum. for all vertices i. we see that the free energy. p and a Matsubara frequency. i is still connected to vertices j1 .0 • Imagine labelling the vertices 1.17) . . ωn4 ) directed into the vertex. is given by −βF = ln [T r {U (β. . 2. jm (m ≤ 4) and to external momenta p1 . . PERTURBATION THEORY IN IMAGINARY TIME 65 Tτ ∂k uk (x. . Vertex i will be connected to vertices j1 . . . τ ′ ) Hint (τ1 ) . . . . ωn1 ). we assign a factor 1/l! to the diagram. τ ) ∂j uj (0. . n. If S is the number of permutations which leave the diagram invariant. Following our real-time development in this way. write |p|. jm (m ≤ 4) and to external momenta p1 . Hint (τn ) (5.2. The diﬀerences are that in imaginary-time. . p4−m . • For each external line with momentum p.

it is useful to consider τ as a complex variable. IMAGINARY-TIME FORMALISM 5. Hence. G(τ ) is not.21) e−βEn − e−βEm | m |∂i ui (0. in lieu of (4. imaginary-time Green functions cannot be directly measured in experiments. When τ lies on the real axis. 0)| n (5. τ − τ ′ ) = θ(τ − τ ′ )T r e−βH e−τ H ∂k uk (x)eτ H e−τ +θ(τ ′ − τ )T r e−βH e−τ ′H ′H ∂j uj (x′ )eτ ′H ∂j uj (x′ )eτ ′H e−τ H ∂k uk (x)eτ H (5. G(x. we have. We follow our earlier derivation of the spectral representation for the T = 0 real-time ordered Green function. however.m δ(p − pm + pn )δ(ω − ωnm )(θ(τ )eip·x−ωτ e−βEn +θ(−τ ))e−ip·x+ωτ e−βEm )| m |∂i ui (0.66 CHAPTER 5. τ ) = d3 p dω[ n. we have the imaginary-time Green function: G(x − x′ . we must contemplate real-time Green functions. Hence. by inserting a complete set of intermediate states.m d3 x 0 β dτ G(x. we have the real-time Green function: G(x − x′ . G(p.35).18) When τ is on the imaginary axis. iωj ) = dE [ n. an analytic function over the entire complex plane. G(τ ) interpolates between these two. |m m|. t − t′ ) = θ(t − t′ )T r e−βH e−itH ∂k uk (x)eitH e−it H ∂j uj (x′ )eit H ′ ′ ′ ′ +θ(t′ − t)T r e−βH e−it H ∂j uj (x′ )eit H e−itH ∂k uk (x)eitH (5.3 Analytic Continuation to Real-Time Green Functions In spite of their many charms.19) For arbitrary complex τ . as we can see from its spectral representation. iωj ) = is given by: G(p. and to analyze the properties of G(τ ) as τ varies through the complex plane.20) |2 ] The Fourier transform. τ = it. In fact. The principal diﬀerence is that we now have e−βEn |n rather than |0 . 0)| n |2 . τ ) eiωj τ (5.

x2 . Gret (p. t1 )∂j uj (x2 . Hence. E) E − iωj (5. t1 ) (5. t2 )∂i ui (x1 . we dealt with the time-ordered two-point correlation function.26) Note that the spectral function is the diﬀerence between the retarded and advanced correlation functions.27) 5. ω) = dE ∞ (5. ω) − Gadv (p. E) E − ω − iδ −∞ ∞ p2 B(p. the spectral function B(p. these are the advanced and retarded correlation functions deﬁned in the next section: p2 B(p.4. t2 )= θ(t1 − t2 ) ∂i ui (x1 . it does not satisfy the Kramers-Kronig relations.5. the functions ∞ −∞ dE p2 B(p. ω) = dE E − ω + iδ −∞ Gret (p. E) is real and positive. t1 . 0)| n |2 δ(p−pm +pn ) δ(E−Emn ) (5. E) = n.53).24) As usual.25) are analytic functions of ω in the lower and upper-half planes respectively. As you will show in the problem set.28) . G is not analytic as a result of the singularities in (11.22) e−βEn − e−βEm | m |∂i ui (0. ω) = 2πip2 B(p. they do satisfy the Kramers-Kronig relations. iωn ) = ∞ −∞ dE p2 B(p.23) we have the spectral representation of G: G(p. Consequently. However. E) E − ω ± iδ (5. E) Gadv (p. G(x1 . ω) (5. t2 ) + θ(t2 − t1 ) ∂j uj (x2 . RETARDED AND ADVANCED CORRELATION FUNCTIONS 67 × δ(p − pm + pn )δ(E − Em + En )] Writing p2 B(p.4 Retarded and Advanced Correlation Functions In the previous chapter.m 1 E − iωj (5.

33) (ω − iδ)2 − vl2 p2 For interacting phonons. t1 . t1 )](5. At zero temperature. this procedure works. t2 )] Gadv (x1 .53). we have introduced the imaginary-time two-point correlation function: G(x − x′ . t1 ). t1 ). . τ − τ ′ ) = θ(τ − τ ′ )T r e−βH ∂k uk (x. t1 )]| 0 (5. x2 . ∂i ui (x1 . IMAGINARY-TIME FORMALISM In this chapter. In other words. τ ′ )∂k uk (x. τ )∂j uj (x′ . ∂j uj (x2 . t1 . the retarded correlation function is often more useful for comparison with experiments. we can obtain G from Gret by taking ω → iωn . t2 ). t2 ) = θ(t1 − t2 ) 0 |[∂i ui (x1 . t1 . x2 . t2 ) = d3 pdω ei(p·(x1 −x1 )−ω(t1 −t2 )) d3 pdω ei(p·(x1 −x1 )−ω(t1 −t2 )) ip2 (5. the zero-temperature advanced and retarded correlation functions can be obtained by choosing the correct iδ prescription for the poles: Gret (x1 .68 CHAPTER 5. τ ′ ) + θ(τ ′ − τ )T r e−βH ∂j uj (x′ . we have now added the retarded and advanced correlation function. t1 . In the speciﬁc examples which we will look at. the retarded and advanced correlation functions are given by: Gret (x1 . we see that G(−iωn ) = G ∗ (iωn ). What we would like to do. t2 ) = θ(t1 − t2 ) T r e−βH [∂i ui (x1 .30) At ﬁnite temperature. x2 .32) (ω + iδ)2 − vl2 p2 ip2 (5. from which we can obtain G for iωn in the lower-half-plane. x2 . is obtain Gret from G. However. From (11. As we will see in the next chapter. t1 . In the next chapter. there is no general theorem stating that this can always be done. ∂j uj (x2 . however. the situation is not so simple. we can always obtain G from Gret .29) To this family of Green functions. x2 . these are generalized to: Gret (x1 . t2 ) = θ(t2 − t1 ) T r e−βH [∂j uj (x2 . ∂i ui (x1 . we see that for iωn in the upper-half-plane. τ ) (5. t2 ) = Gadv (x1 . x2 . This analytic continuation from the Matsubara frequencies iωn to the entire upper-half-plane can often be done by simply taking iωn → ω + iδ. From (11. t2 )]| 0 Gadv (x1 . we will see why retarded correlation functions are intimately related to experimental measurements. t2 ) = θ(t2 − t1 ) 0 |[∂j uj (x2 .53). t1 .31) For free phonons. t2 ).

q) n (5. as may be seen from the spectral representation (11. thereby leading to the Matsubara sum.1: The contour of integration in (5.35). it is natural to rewrite these integrals in terms of advanced and retarded Green functions. The only nonvanishing segments of the contour integral are those which run on either . p − q) G(iωn .5. p − q) G(ω. q) (5. to convert sums over Matsubara frequencies into integrals. q) = 2πi β C G(iΩm − iωn . EVALUATING MATSUBARA SUMS 69 5. Consider the sum 1 β n G(iΩm − iωn .5 Evaluating Matsubara Sums We can use contour integration and the fact that the poles of nB (ω) are precisely the Matsubara frequencies.53).34) This sum is equal to the following contour integral (see ﬁgure 5.5. p − q) G(iωn . consequently. ωn = 2nπ/β. ω ω = ιΩ + Ε ω=Ε ω = 2πi/n Figure 5. it picks up only the poles of nB (ω). dω 1 nB (ω) G(iΩm − ω.1) since the integral avoids the singularities of the Green functions. As we will see.35) The singularities of the Green functions occur when ω or iΩm − ω are real.

q) (p − q)2 B(E. p − q) (5. they arise from the fact that E and ω are oppositely directed.37) It is important that we did the analytic continuation of the external frequency. p − q) Since nB (iΩm − E) = −(1 + nB (E)) for Matsubara frequencies iΩn .70 CHAPTER 5. then we can use (11. q) dE (nB (E) + 1) G(iΩm − E. as you will see in the problem set. q) dE nB (iΩm − E) G(iΩm − E. rather than during some intermediate step. then we have: 1 β G(iΩm − iωn . q)(p − q)2 B(E. q) (p − q)2 B(E. . IMAGINARY-TIME FORMALISM side of the lines ω = E (the ﬁrst term on the right-hand-side below) or ω = iΩm − E (the second term) where E is real: ∞ 1 dω nB (ω) G(ω + iΩm ) G(ω) = dE nB (E) G(iΩm − E) (G(E + iδ) − G(E − iδ)) 2πi −∞ C 2πi −∞ 1 + dE nB (iΩm − E) (G(E + iδ) − G(E − iδ)) G(iΩm − E) 2πi ∞ Note the reverse limits of integration in the second integral. p (5. If we assume that the analytic continuation is straightforward (as it often is). q) dE (nB (E) + 1) Gret (Ωm − E. p − q) q 2 B(E. This type of contour integral trick can be used rather generally to bring Matsubara sums to a convenient form. p − q)q 2 B(E. p − q) G(iωn .36) − q) If we also continue iΩm → Ω + iδ. Ω.56) to write this as: dω nB (ω) G(ω + iΩm ) G(ω) = 2πi C − ∞ −∞ ∞ −∞ dE nB (E) G(iΩm − E. we ﬁnally have: 1 β G(iωn + iΩm ) G(iωn ) = + ∞ −∞ ∞ −∞ n dE nB (E) G(iΩm − E. at the end of the calculation. q) = + ∞ −∞ ∞ −∞ n dE nB (E) Gret (Ω − E. p − q) q 2 B(E.

41) i.e. The evolution operator for such a process is: U (∞. this is simply equal to 1 and2 i|U (−∞. −∞)|i = 1 (5. ∞) U (∞. Then i|U (−∞.39) (5. by diﬀerentiating with respect to j(x.42) (5. We may.6.38) Suppose. t) and. ∞) Uj (∞. We may then wish to compute the correlation functions of some observable such as ∂i ui (x. wish to consider a situation in which the system is in a given intial state – say the vacuum state or the state at thermal equilibrium at inverse temperature β – but we make no assumptions about the ﬁnal state. hence. we evolve the system from t = −∞ to t = ∞ then back to t = −∞ and thence to t = −∞ − iβ.40) and. we have Z[j] = T r e−βH U (−∞. −∞) or Z[j] = T r {U (−∞ − iβ. however. we were able to relate these amplitudes to the amplitude for a system to remain in its vacuum state in the presence of an external source j(x. −∞)|0 At ﬁnite temperature. −∞) Clearly. that we switch on a source ﬁeld. In the previous chapter. to correlation functions in the vacuum state. we imagine evolving the system from some initial state |i at t = −∞ to t = ∞ and then back to t = −∞. only during the forward propagation in time. −∞)} (5. j(x. however. t). we can obtain correlation functions. t). ∞) Uj (∞. t) at time t. If we wish to work with a system at zero-temperature which is in its ground state at t = −∞. . −∞) U (−∞.43) (5. then we deﬁne the generating functional: Z[j] = 0|U (−∞. ∞) Uj (∞. −∞) U (∞. THE SCHWINGER-KELDYSH CONTOUR 71 5.6 The Schwinger-Keldysh Contour The formalism which we have thus far constructed was designed to determine the transition amplitudes from some given intial state at t = −∞ to some ﬁnal state at t = ∞. −∞)|i = 1 (5. ∞) Uj (∞.5. In order to do this.

72 CHAPTER 5. which we won’t prove here. Using Dyson’s formula for U (tf .44) where Z = T r {U (−∞ − iβ. There is an important factorization property. IMAGINARY-TIME FORMALISM C1 ti C3 ti −i σ ti −i β C4 C2 tf −i σ tf Figure 5. ∞) Uj (∞. To construct the resulting perturbation theory. however.2: Real-time contour separated into four parts that factorize into separate contributions: C1 ∪ C2 and C3 ∪ C4 . C. it is useful to denote the ﬁelds on the . We can use Wick’s theorem to evaluate Tc -ordered products. There is a more general class of contours. the second copy of each ﬁeld propagates backwards in time. The price which must be paid is the doubling of the number of the ﬁelds in the theory. from ∞ to ∞ − iσ. ti ). this is a matter of taste. ∞ − iσ) U (∞ − iσ. We make the choice σ = β/2 for which the propagator takes a particularly simple form. which go from −∞ to ∞. and thence to ∞ − iβ. we can expand Z = Z C1 ∪C2 perturbatively as we did in the equilibrium zero-temperature and imaginary-time formalisms. −∞)} = T r Tc e−i R C dt Hint (t) (5. −∞ − iσ) U (−∞ − iσ. for the contributions from each piece of the contour to the functional integral Z: Z = Z C1 ∪C2 Z C3 ∪C4 . All choices of σ share the advantage of being real-time formulations and thereby obviating the need for potentially ill-deﬁned analytical continuations. t) can drive the system out of equilibrium since we make no assumptions about the ﬁnal state of the system. from ∞ − iσ to −∞ − iσ.45) Only C1 and C2 are important in obtaining correlation functions. The Keldysh contour which we just described is just one example of a possible contour along which the time-evolution is performed. (5. This generating functional allows us to calculate non-equilibrium correlation functions: j(x.

51) The matrix u contains the information about the temperature. For the Keldysh contour. where i∆0 (ω.5. 5.50) ¯ −iG∗ (t − t′ . THE SCHWINGER-KELDYSH CONTOUR upper (C1 ) and the lower (C2 ) pieces of the countour by u1 (t) = ui (t). i u2 (t) = ui (t − iσ). 2 and the amplitude is assigned a factor of −1 for each a = 2 vertex. x − x′ ) =< Tc [∂i ua (t. x)] 0 (5. x′ )∂i ui (t. the dynamical information contained in the zero-temperature propagator and the information about occupation numbers can be untangled by the parametrization: i∆(ω. the diagonal elements of the propagator are the real-time zero-temperature time. the anti-time-ordered one 0 (5. x′ )] )∗ = T [∂i ui (t′ . consequently.and anti-time ordered propagators. T cosh2 θω = Notice that at zero temperature u = 1. For our choice. k) u† (ω) .49) with iG0 (ω. x)∂j uj (t′ . k) = iG0 (ω. x′ )] > i j (5. This matrix is given by u(ω) = cosh ∆θω sinh ∆θω sinh ∆θω cosh ∆θω and . but they can be more general. k) 0 ∗ (ω. x)∂j ub (t′ . To summarize. k) = u(ω) i∆0 (ω. k) 0 −iG0 . k). k) the usual time-ordered propagator iG0 (t − t′ . x − x′ ) = T [∂i ui (t. The oﬀ-diagonal element contains all of the information about the occupation numbers of states. in thermal equilibrium the occupation numbers are given by nB (ω). 1 − e−ω/T (5. The vertex resulting from the source ﬁeld j(x. x − x′ ) = ( T [∂i ui (t.6.47) where Tc denotes ordering of ﬁelds according their position along the contour of Fig. 1 .46) The Feynman rules are similar to those of our equilibrium zero-temperature theory with the major diﬀerence being each vertex is labelled by an index a = 1. T T where ∆θω = θω − θω =0 .48) (5. The propagator between two vertices with indices a and b is given by: −i∆ab (t − t′ . x′ )] and −iG∗ (ω. t) is assigned a = 1.52) .2. (5. i (t = real) 73 (5. σ = β/2. x)∂i ui (t′ .

. . If S is the number of permutations which leave the diagram invariant. . • For each external line with momentum p. . • To each vertex we assign an index a = 1. 2. . Consider a permutation of these labels. IMAGINARY-TIME FORMALISM • To each line. . the Schwinger-Keldysh formalism gives results which are identical to those of the Matsubara formalism. ω4 ) directed into the vertex. In equilibrium.74 CHAPTER 5. p4−m . p and a frequency. we associate a momentum. Out of equilibrium. . the two formalisms give diﬀerent results. . . only the Schwinger-Keldysh is correct. frequencies (p1 . . . . . p4−m . jm (m ≤ 4) and to external momenta p1 . . i is still connected to vertices j1 . Such a permutation leaves the diagram invariant if. ω) 0 • For each vertex carrying index a with momenta. . . . however. for all vertices i. 2. jm (m ≤ 4) and to external momenta p1 . n. . • If two vertices are connected by l lines. . ω. write |p|. . . • The propagator assigned to an internal line carrying momentum p and frequency ω which connectes vertices labelled by indices a and b is: ∆ab (p. . External lines are assigned a = 1 at their free end. we assign a factor 1/l! to the diagram. we assign a factor 1/S to the diagram. ω1 ). . we write (3 − 2a) g (2π)3 δ(p1 + p2 + p3 + p4 ) δ(ω1 + ω2 + ω3 + ω4 ) • Imagine labelling the vertices 1. (p4 . Vertex i will be connected to vertices j1 . . .

with equation of motion dx d2 x 2 +γ + ω0 x = F ext (t) (6.CHAPTER 6 Measurements and Correlation Functions 6. symmetries.2) into the equation of motion (6. (6.1 A Toy Model We will now take a break from our development of techniques for calculating correlation functions to relate retarded correlation functions to experimental measurements. Substituting the deﬁnition of χret (t − t′ ).1) 2 dt dt We deﬁne the retarded response function. Consider a single damped harmonic oscillator. We will also discuss those properties of retarded correlation functions which follow from causality. χret (t−t′ ) must vanish for t−t′ < 0.2) By causality. we see that χret (t − t′ ) satisﬁes: d d2 2 χ (t − t′ ) + γ χret (t − t′ ) + ω0 χret (t − t′ ) = δ(t − t′ ) 2 ret dt dt 75 (6. χret (t − t′ ) by x(t) = ∞ −∞ dt′ χret (t − t′ )F ext (t′ ) (6. Let us take a look at a toy model to see why retarded correlation functions are useful objects which are simply related to experimentally measurable quantities.3) . and conservation laws.1).

χret (t − t′ ) is the Green function of the diﬀerential operator on the left-hand-side of (6.7) From the above deﬁnition.8) dteiωt 1 (χ(t) + χ(−t)) 2 (6. χ′′ (ω) is the Fourier transform of the part of χ(ω) which is not invariant under t → −t while χ′ (ω) is the Fourier transform of the part of χ(ω) which is invariant under t → −t. χ′ (ω) = ∞ −∞ (6. the integral is well-deﬁned for ω anywhere in the upper-half-plane. we ﬁnd χ(ω): 1 (6. Substituting the Fourier representation of χ(t) in the equation of motion. χ(ω) is an analytic function in the upperhalf-plane. Therefore.4) Since χ(t) vanishes for t < 0.1) subject to the boundary condition that χret (t − t′ ) = 0 for t − t′ < 0. We can deﬁne the Fourier transform of χ: χ(ω) = ∞ −∞ dteiωt χ(t) = 0 ∞ dteiωt χ(t) (6. In other words.9) Thus.76 HAPTER 6. χ′′ (ω) knows about . we have in our toy model: 2 ω0 − ω 2 − ω 2 )2 + (γω)2 γω ′′ χ (ω) = 2 (ω0 − ω 2 )2 + (γω)2 (6.5). χ′′ (ω) = Im = ∞ ∞ −∞ dteiωt χ(t) 1 iωt e − e−iωt χ(t) 2i −∞ ∞ 1 dteiωt (χ(t) − χ(−t)) = 2i −∞ dt Similarly.5) χ(ω) = 2 ω0 − ω 2 − iγω We can break χ(ω) into its real and imaginary parts: χ(ω) = χ′ (ω) + iχ′′ (ω) From (6. MEASUREMENTS AND CORRELATION FUNCTIONS C Thus.6) χ′ (ω) = 2 (ω0 (6.

χ′ (ω). on resonance. the energy dissipation. A TOY MODEL 77 the arrow of time which is implicit in the condition χ(t − t′ ) vanishes for t − t′ < 0.1. the dissipation: dx dW = F (t) dt dt d = F (t) dt = F (t) = ∞ −∞ dω iωt e iωχ(ω) F (ω) −∞ 2π dω dω ′ i(ω+ω′ )t e iωχ(ω) F (ω)F (ω ′ ) 2π 2π ∞ dt′ χret (t − t′ )F (t′ ) (6. This is reﬂected in the fact that χ′′ (ω) determines the dissipative response. we have a Lorentzian lineshape: γ 1 2 dW = F0 dt 2 4(ω0 − Ω0 )2 + γ 2 (6.6.10) If we assume F (t) = F0 cos Ω0 t and compute the zero-frequency part of dW/dt (rather than the part which oscillates at 2ω0 ). does not.13) with full-width γ at half-maximum.11) The essential reason that χ′ doesn’t enter the dissipation is that the timereversal symmetry of χ′ implies that the energy gain and loss due to χ′ are the same. we ﬁnd: 1 2 dW (ω = 0) = F0 i (Ω0 χ(Ω0 ) − Ω0 χ(−Ω0 )) dt 4 1 2 = F0 Ω0 χ′′ (Ω0 ) 2 (6. 1 2 γΩ0 dW = F0 Ω0 2 2 )2 + (γΩ )2 dt 2 (ω0 − Ω0 0 (6. χ′ is often called the reactive part of the susceptibility while χ′′ is the dissipative or absorptive part. F (t). For our toy model.e. Approximating Ω0 + ω0 ≈ 2ω0 . To see this.14) .e.12) is maximum at Ω0 = ±ω0 . Consider the resonance Ω0 ≈ ω0 . The work done by this force is the energy which is tranferred to the system – i. as a result of causality – we have the Kramers-Kronig relations. The analyticity of χ(z) in the upper-half-plane implies that: dz χ(z) =0 C πi z − ω (6. suppose that we apply a force. on the other hand. i.e. As a result of the analyticity of χ in the upper-half-plane – i.

78 HAPTER 6.16) Comparing real and imaginary parts. this can be done – at least at a formal level – for any theory.18) 6.2 General Formulation We can cull the essential features of the toy model and apply them to a many-body system such as our phonon theory. for the contour of ﬁgure 6. As we will see momentarily.1 The integral of the semicircular part of the contour vanishes. just a set of harmonic oscillators. we have the Kramers-Kronig relations: dω ′ χ′′ (ω ′ ) ′ −∞ π ω − ω ∞ ′ χ′ (ω ′ ) dω χ′′ (ω) = −P ′ −∞ π ω − ω χ′ (ω) = P ∞ dω ′ χ(ω ′ ) ′ −∞ π ω − ω (6. MEASUREMENTS AND CORRELATION FUNCTIONS C z C 111111111111111111 000000000000000000 Figure 6. so we have: χ(ω ′ ) dω ′ =0 ′ −∞ πi ω + iǫ − ω Using the relation: ω′ we have χ(ω) = −iP 1 =P + iǫ − ω ω′ 1 −ω ∞ ∞ (6. It is fairly clear that this can be done for a free ﬁeld theory which is.17) (6. . after all.1: The countour of integration taken for the Kramers-Kronig relation.15) − iπ δ(ω ′ − ω) (6.

t′ ). t) = u3 (x. t) | 0 Then. Alternatively. We will work – as in the last chapter when we deﬁned Green functions – in an interaction representation in which Hprobe is the interaction Hamiltonian and the rest of the Hamiltonian is the ‘free’ Hamiltonian. t′ ) 0 φ(x′ . GENERAL FORMULATION 79 Suppose our external probe couples to our system through a term in the Hamiltonian of the form: Hprobe = d3 x φ(x. t) 0 (6. and φ(x. η(x. t) f (x. causing displacements along the 3-direction. η(x.22) 0 −∞ We have added a subscript 0 to emphasize that these are interaction picture states – i. t) (6. t) is some quantum ﬁeld describing the system. these are expectation values in the absence of the probe.19) f (x.21) If we keep only terms up to ﬁrst-order in Hprobe . we could take φ(x. then we have: t η(x.24) . t) in which case our probe compresses the solid. t) = 0 UI (t. we could take φ(x.e. Its expectation value is given by: −1 η(x. Let us suppose that we now measure the ﬁeld η(x. t) = i d3 x′ t −∞ 0 =0 (6. −∞) 0 (6.6.2. which we control. t) 0 t 0 dt′ f (x′ . t′ ). which may or may not be the same as φ(x. t). (6. .23) dt′ f (x′ . η(x. t′ ) 0 φ(x′ . t) = ∂k uk (x. t). as is usually the case. −∞) = T e−i =1−i Rt t ′ −∞ dt Hprobe (t′ ) −∞ dt′ Hprobe (t′ ) + . t). t) 1−i −∞ dt′ Hprobe (t′ ) 0 = 0 | η(x. t) (6. Let us assume. t) | 0 = 0 | η(x. η(x. . that 0 | η(x. t) = 0 1+i −∞ 0 dt′ Hprobe (t′ ) t −∞ t η(x. In our phonon theory.20) where UI (t. t) UI (t. t) is our external probe. −∞) η(x. t) | 0 + 0 i +i d x 3 ′ dt′ Hprobe (t′ ).

φ(x′ . t. t − t′ ) (6. is an example of a response function.28) d3 x′ ∞ −∞ (6. t − t′ ) = −χ′′ (x′ − x. t′ − t) χ′′ (x′ − x. φ(x′ . t′ ) 0 (6.29) Following the steps of (6. t′ ) χ(x − x′ . t).25) If the Hamiltonian is space.26) dt′ f (x′ . χ′′ . t) = fω (x) cos ωt is given by dw = dt d3 x d3 x′ ωχ′′ (x′ − x.32) dω −iω(t−t′ ) ′′ e χ (x − x′ . the dissipation rate under the inﬂuence of a periodic probe. t′ ). we write: η(x. φ(x. χ(x − x′ . η(x. t − t′ ) = 0 As in our toy model. t)]|0 . ω) + iχ′′ (x − x′ . f (x. t) = φ(x′ . t′ ) (6. ω) and its real and imaginary parts. t − t′ ) We can also extend the dt′ integral to ∞ η(x. t − t′ ) = where χ′′ (x − x′ .and time-translationally in the absence of Hprobe . ω) (6. ω) = −χ′′ (x′ − x. t′ ) χ(x. Following the steps which we took in our toy model. ω) = χ′ (x − x′ .33) .80 HAPTER 6. 0| [φ(x′ . x. t. t − t′ ) ≡ iθ(t − t′ ) 0 φ(x. x′ .31) satisﬁes the antisymmetry properties: (6. ω) 2π (6.8). t) = −i d3 x′ t −∞ dt′ f (x′ . t). −ω) These properties follow from the fact that χ′′ is a commutator.27) As in our toy model. Then. t) = −i if we deﬁne χ(x − x′ . ω) is the Fourier transform of the commutator without a θ-function: χ′′ (x − x′ . t′ ). ω)fω (x)fω (x′ ) (6. Let us specialize to the case η(x. t′ ) 0 (6. then we can write: χ(x. t′ ) = χ(x − x′ . MEASUREMENTS AND CORRELATION FUNCTIONS C The commutator on the right-hand-side. χ(x − x′ . we see that χ′′ (x−x′ .30) χ′′ (x − x′ . we can deﬁne the Fourier transform with respect to time.

t) = T r e−βH η(x. t) = θ(t) (Sηφ (x. χ′′ (q.41) (6. According to (6. t) φ(0.18).6. 0) Consider the correlation function: (6. ω) = (Sηφ (x. ω) = where ρ(q. E) −∞ π E − ω − iδ ∞ (6. −t)) or χηφ ′′ (x. t) = (Sηφ (x.3.38) The response function can be expressed in terms of the correlation function: χηφ (x. t) − Sφη (−x. THE FLUCTUATION-DISSIPATION THEOREM 81 We can also follow the derivation for our toy model to show that χ(x′ − x. ω) is the spectral function for χ(q.35) If we construct a spectral representation for χ′′ (q. E) = m ∞ −∞ dE ρ(q. −ω)) (6. ω) is non-vanishing.40) (6. ω) (6.28). ω): χ(q. we have: χ(q. ω) satisﬁes the Kramers-Kronig relations (6.34) | m |φ(0. we see that: χ′′ (q.37) χ(q. Following our earlier derivations of spectral representations. ω) − Sφη (−x.3 The Fluctuation-Dissipation Theorem Sηφ (x. E) E − ω − iδ (6. −t)) and its dissipative part is simply χηφ ′′ (x.36) In other words. ω) = dE χ′′ (q. ω) is singular whenever χ′′ (q.34). thereby causing dissipation. ω) – which can be done trivially since there are no θ-functions – or simply take the imaginary part of (6. t) − Sφη (−x. ω) = π ρ(q.39) . 0)| 0 |2 δ(q − pm ) δ(E − ωm ) (6. These are the regions of phase space where there are states of the system with which the external probe can resonate. The Kramers-Kronig relations can be approached from a diﬀerent angle by constructing a spectral representation for the response function (6.37) which is the Kramers-Kronig relation. 6.

neutron scattering experiments measure Sρρ (q. 0)e−βH η(x. 0)e−βH η(x. so we are driving the solid in the x1 -direction and measuring the subsequent ﬂow of the positive ions in this direction. t) φ(0. (6. t) = T r e−βH eβH φ(0. −ω) Thus.42) Since the right-hand-side is a measure of the dissipation and the left-handside is a measure of the ﬂuctuation. Let us make the simplifying assumption that vl = vt . i. there will only be dissipation if the compressional force has a component with ω 2 = vl2 q 2 . 6.e. ω) = 1 − e−βω Sηφ (x. −t − iβ) Hence. χ′′ = ω 2 sgn(ω)δ(ω 2 − vl2 q 2 ) φφ (6. According to the iδ prescription of (5. 0) = T r φ(0. t) = Sφη (−x. As we will see shortly. ω) = eβω Sφη (−x. Sηφ (x.32). for free phonons this is: χφφ (ω. of the phonon spectrum.44) is called the ﬂuctuation-dissipation theorem. ω). ω). χρρ ′′ (q.4 Perturbative Example Let us consider the case in which φ = η = ∂t u1 .46) In other words.82 HAPTER 6. t) = T r e−βH η(x. ∂t u1 is the current in the x1 -direction carried by the ions in the solid.44) (6. ω2 (ω + iδ)2 − vl2 q 2 (6.45) . Sηφ (x. −iβ) η(x. ω) (6. Then χφφ is given by the retarded correlation function of ∂t u1 with itself. q) = Hence. t) = T r e−βH φ(0. MEASUREMENTS AND CORRELATION FUNCTIONS C By the cyclic property of the trace. Thus a measurement of this response function is a direct measurment of vl .43) (6. and the ﬂuctuation-dissipation theorem relates this to a quantity which we can attempt to calculate using the imaginary-time formalism: the imaginary part of a retarded correlation function. we ﬁnally have: χηφ ′′ (x.

p1 ) G(iωn2 . χφφ (ω. The analytic continuation to the retarded Green function is trivial.47) and the δ-function is broadened: χ′′ = φφ ω 2 − vl2 q 2 + Re {Πret (ω. q)} = Im{q 2 g2 1 6 β2 n d3 p1 d3 p2 G(iωn − iωn1 − iωn2 .6.) g Λ4 /v.49) n In the problem set. q) = g 2 d3 p p q + O(g2 ) 2 (2π)3 v = (const. q) = − q 2 2 β d3 p 1 + O(g2 ) |p|2 2 3 (2π) ωn + vl2 p2 (6. This diagram gives a purely real contribution 1 g Π(iωn . When we convert the two Matsubara sums to integrals. but we can make some simple observations. q)})2 (6.48) Consider a perturbative computation of Πret (ω.51) Evaluating this integral is tedious. At zero temperature. At O(g).4.) g q 2 Λ4 /v + O(g2 ) = (δvl2 )q 2 (6.n2 . q)} 2 83 (6. taking the imaginary part will convert the ﬁnal one to a spectral function: Im {Π(iωn . q)} + (Im {Πret (ω. q). q) = ω2 (ω + iδ)2 − vl2 q 2 + Πret (ω. you will compute the Matsubara sum. PERTURBATIVE EXAMPLE If the phonons are interacting.50) where (δvl2 ) = (const. q)} = Im{q 2 g2 6 g2 = Im{q 2 6 dω1 dω1 dω2 dω2 d3 p1 (2π)3 d3 p1 (2π)3 d3 p2 G B B}iωn →ω+iδ + O(g3 ) (2π)3 d3 p2 B B B}iωn →ω+iδ (2π)3 1 . we ﬁnd: Π(iωn . The ﬁrst contribution to the imaginary part of the self-energy comes from diagram (b) of ??. there is diagram (a) of ﬁgure ??. q) ω 2 Im {Πret (ω. p2 )}iωn →ω+iδ + O(g3 ) (6. we will convert the Green functions to spectral functions. It is given by Im {Π(iωn . q − p1 − p2 ) (2π)3 (2π)3 G(iωn1 .

and current.55) Unlike in the case of a propagating mode. ρ. we would like to compute D (perturbatively or by going beyond perturbation theory). MEASUREMENTS AND CORRELATION FUNCTIONS C +O(g3 ) (6.56) where D is the diﬀusion constant and Bext is an external force acting on the particles (such as gravity). is a diﬃcult problem. In general. we can often say something about χ since some of its general features follow from conservation laws and symmetries. q) are small and lead to a small damping of a propagating mode. ρ and J M will satisfy a constitutive relation: J = −D ∇ρ + fext (6. χ.)g2 Λ7 . There is no phase space for the δ-functions to be satisﬁed if ω = 0 (since p1 and p2 will have to be collinear). Nevertheless. so the integral is propotional to ω. we have: Im {Π(iωn . 6. we can now write the spectral function as: χ′′ = φφ ω 2 Dq 2 ω ω 2 − vl2 q 2 ˜ 2 + (Dq 2 ω)2 (6. In such a case. q) can be calculated perturbatively. The resulting equations satisﬁed by physical quantities (including response functions) are hydrodynamic equations.84 HAPTER 6. we have put the three internal phonon lines on-shell. Hence. J of these particles will satisfy a conservation law: ∂ ρ+∇·J =0 ∂t (6. the corrections due to Πret (ω. Keeping only the ﬁrst term. The density. the calculation of the response function.54) where vl2 = vl2 + δvl2 . q)} = Dq 2 ω + O(q 4 ω) + O(g3 ) (6. ˜ This is the form of the response function which we expect when g is small and Πret (ω. however.5 Hydrodynamic Examples Let us consider as an example some particles dissolved in a ﬂuid. Ideally.53) where D = (const. but in many cases we must .52) By taking the imaginary part.

but with vl = 0.54) above.56) hold. Note that such constraints usually only hold for conserved quantities. As a result of the constitutive relation. In this section and the last. When both of these conditions are satisﬁed. low-frequency – equations which they satisfy. Similarly.59) Thus. q) = and χ′′ (ω. Hydrodynamic laws hold in the long-wavelength. we have seen how the low q.58) (6.6. χρρ (ω. The conservation law.61) is of precisely the same form as (6. ˜ In this example. leads to hydrodynamic equations satisﬁed by the conserved quantity and its current. q) = JJ iω −iω + Dq 2 Dq 2 ω ω 2 + (Dq 2 )2 (6. there will be one hydrodynamic mode for each conservation law satisﬁed by the system. in turn. ω behavior of response functions of conserved quantities and their associated currents can be determined by a knowledge of the hydrodynamic modes of the system. determine the correlation functions.57) (6. we can sometimes perturbatively calculate response functions which satisfy the constraints imposed by . q) = ρρ iq −iω + Dq 2 Dq 3 ω 2 + (Dq 2 )2 (6. These equations. together with a constitutive relation. we have seen how response func˜ tions which are calculated perturbatively (as we imagined doing in the ﬁrst example) are often of the same form as those which are deduced from the hydrodynamic – or long-wavelength. In general.54) for v = 0.5. correlation functions of arbitrary ﬁelds are typically unconstrained. q) = and χ′′ (ω.61) Note that this is precisely the same as (6. χ′′ /q is a Lorentzian centered at ω = 0 with width Dq 2 . Observe that (6.60) (6. Linear response theory holds in the limit of small fext . HYDRODYNAMIC EXAMPLES 85 leave it as a phenomenological parameter. ρρ χJJ (ω. low-frequency limit in which local equilibrium is maintained so that constitutive relations such as (6. ρ satisﬁes the diﬀusion equation: ∂ ρ − D ∇2 ρ = ∇ · fext ∂t Hence.

t) m (6. t) 1−i −∞ dt′ Hprobe (t′ ) 0 = 0 J(x.54) and (6. χρE . for the electrical conductivity of a system. A and ϕ.86 HAPTER 6. Let j denote the current in our condensed matter system when the external vector and scalar potentials. in general.59). t) = 0 1+i −∞ dt′ Hprobe (t′ ) t t J (x. These coeﬃcients – or the corresponding response functions – may be derived by following the steps of section 2. We have already encountered one example of a transport coeﬃcient which can be obtained from a response function. m Meanwhile. MEASUREMENTS AND CORRELATION FUNCTIONS C hydrodynamic relations. let us derive the corresponding relation. we have d3 x −ρ(x. Let ρ be the charge density. J (x. t) + t J(x. t) 0 0 + 0 i −∞ dt′ Hprobe (t′ ). when 2 we turn on the electromagnetic ﬁeld. or Kubo formula. χρρ . ω) q 2 JJ (6. and χEE are given by linear combinations of functions of the form of (6. Hprobe is given by: Hprobe = e ρ(x.64) . t)ϕ(x. the current is given by J = j − ne A.54) or (6.63) Following our derivation of the response function in section 2. Transport coeﬃcients relate the resulting currents to the applied gradients. In chapter 7. t) + j(x. be related to such limits of response functions.62) ω→0 q→0 To see why transport properties should. It turns out that these quantities are coupled so that the “normal modes” of the solid are a combination of the energy and mass. D. 6. we will see examples of this in the context of spin systems. Then. In a solid. which can be obtained from (6.6 Kubo Formulae Transport measurements ﬁt naturally into the paradigm of linear response theory: a weak external probe – such as a potential or temperature gradient – is applied and the resulting currents are measured. we might be interested in the response functions for the energy density and the mass density. t) 0 (6. Consequently. t) · A(x. t)A2 (x. One of thse normal modes diﬀuses while the other is a (damped) propagating mode. are zero.61) by: D = lim lim ω ′′ χ (q. namely the diﬀusion constant.

so let’s take ϕ = 0 gauge. so we naively have: Ji (x. KUBO FORMULAE Using 0 J 0 = 0 j 0 ne2 A =− m − ne2 A m 87 0 0 (6. ω) = 1 ω ∞ 0 d3 x′ ∞ −∞ dt′ σij (x − x′ . t′ ). ji (q. t′ ) (6. ρ(x. t′ ) t dt′ ji (x′ . t) = or Ji (q. ji (x. ω) = χ′′ (q. t′ ).72) ω jj .66) ne2 Ai (x. t) = i d3 x′ ∞ −∞ dt′ θ(t − t′ ) jj (x′ . ji (x.65) and the expression (6. t − t′ ) E(x′ . t) ϕ(x′ . Then. 0). t) Ai (x′ . Ji (x.63) for Hprobe and keeping only terms linear in A.70) In terms of the response function.6. t′ ) − ne2 Ai (x. t) Ai (x′ . ω) Ej (q. t) − m −∞ We are free to choose any gauge we want.69) dt e−iωt [jj (−q. ω) (6. ω) = σij (q.67) In this gauge. ω) with σij (q.6. t′ ) (6. ω) − iω iω m (6. we have: Ji (x. t′ ). E = dA/dt.71) The ﬁrst term on the right-hand-side leads to the real part of the conductivity: 1 Re σij (q. ω) = 1 ne2 1 χjj (q. t) m (6. t) = i −i d3 x′ d x t −∞ 3 ′ dt′ jj (x′ .68) (6. σij (q. t)] − 1 ne2 iω m (6.

n i M 2π 2 kf . The DC conductivity is obtained by taking the q → 0 limit ﬁrst. n 2 e−βEn δ (ω + En − Em ) (6. Let us assume that our system is in thermal equilibrium at inverse temperature β and that neutrons interact with our system via the Hamiltonian H ′ . and then taking the DC limit. DC σij = lim lim ω→0 q→0 1 ′′ χ (q.88 HAPTER 6. let us assume that there are simple δ-function interactions . 6. and energy ωi are scattered by our system. Suppose that neutrons of momentum ki . we can compute the total ﬁeld and thereby ﬁnd the correction to (6.74) where ω = ωi − ωf and n and m label the initial and ﬁnal states of our system. The diﬀerential cross-section for the neutrons to be scattered into a solid angle dΩ centered about kf and into the energy range between ωf ± dω is: d2 σ = dΩ dω kf k m. ω → 0. inhomogenous charge distribution and the q → 0 limit won’t tell us anything about the conductivity. we want the response to the total electric ﬁeld. In fact. m H ′ ki .73). The above formulas are almost right.73) If we take ω → 0. MEASUREMENTS AND CORRELATION FUNCTIONS C while the second term – if it’s not cancelled by the imaginary part of the ﬁrst term – leads to superconductivity. then we’ll get a static. to set up a spatially uniform current. The problem with them is that they give the response to the applied electric ﬁeld.7 Inelastic Scattering Experiments Another way of experimentally probing a condensed matter system involves scattering a neutron oﬀ the system and studying the energy and angular dependence of the resulting cross-section. ω) ω jj (6. This is typically (but not exclusively) done with neutrons rather than photons – for which the requisite energy resolution has not yet been achieved – or electrons – which have the complication of a form factor arising from the long-range Coulomb interactions. Using Maxwell’s equations and our linear response result for J. so we will defer a thorough discussion of it to that chapter. This issue is most relevant in the context of interacting electrons. For simplicity.

ω) δ(ω + ω ′ ) = T r e−βH ρ(q.77) and pass from the Schr¨dinger to the Heisenberg representation. INELASTIC SCATTERING EXPERIMENTS between the neutrons and the particles in our system: H′ = V j 89 δ(x − Xj ) = V ρ(x) (6.6. S(q. t)| m then we can rewrite (6. t)ρ(−q. n kf ki M 2π 2 2 e−βEn δ (ω + En − Em ) m |ρ(−q. n = V d3 x eiq·x m |ρ(x)| n = V m |ρ(q)| n 1 2π ∞ −∞ (6. m H ′ ki .82) According to the ﬂuctuation-dissipation theorem.80) If we deﬁne the dynamic structure factor.74) as d2 σ = dΩ dω = kf ki ∞ −∞ (6. ω)ρ(−q. 0) (6.75) Then kf . ω): S(q. ω) ≡ Sρρ (q. ω) (6. 0)| n (6.83) .81) S(q. t)| m We can now use |m m| = 1 and write this as d2 σ 1 kf = dΩ dω 2π ki M 2π 2 ∞ −∞ dt eiωt T r e−βH ρ(q.7. this can be written d2 σ 1 kf = dΩ dω 2π ki M 2π 2 1 1 − e−βω χ′′ (q. ω) (6.78) m M 2π 2 kf . m H ′ ki .76) If we use the Fourier representation of the δ-function. δ(ω + En − Em ) = dt ei(ω+En −Em )t (6.79) 1 2π dt eiωt m e−βEn n |ρ(q. ω ′ ) then we can write the inelastic scattering cross-section as: 1 kf d2 σ = dΩ dω 2π ki M 2π 2 (6. o ei(En −Em )t m |ρ(q)| n = n eiHt ρ(q)e−iHt m = n |ρ(q.

ω)} δ(q − Q)δ(ω) + dΩ dω 2π ki 2π 1 − e−βω Q where Q is the set of reciprocal lattice vectors. ω) q · u(−q.8 6. ρ(q) = i eiq·Ri 1 − iq · u(Ri ) δ(q − Q) − iq · u(q) (6. ω) is precisely the quantity which is measured in inelastic scattering experiments.87) Recognizing our longitudinal phonon Green function on the right-hand-side and using the ﬂuctuation-dissipation theorem. Hence.86) ≈ Q (6. experiments. there is . MEASUREMENTS AND CORRELATION FUNCTIONS C In our elastic theory of a solid. As a result of this magnetic ﬁeld. ρ(x) = i δ x − Ri − u(Ri ) eiq·(Ri −u(Ri )) i (6.88) Hence. we are neglecting multi-phonon emission processes.84) (6.9 Neutron Scattering by Spin Systems-xxx NMR Relaxation Rate In nuclear magnetic resonance. then there will be Lorentzian peaks at ω = ±˜l q of width Dq 2 . a material is placed in a constant magnetic ﬁeld.54). the scattering cross-section is given by the sum of the contributions of the Bragg peaks together with the contributions of one-phonon emission processes: 2 2σ 1 kf M d δ(q − Q)δ(ω) + T r e−βH q · u(q. By dropping the higher-order terms in the expansion of the exponential. If we assume a self-energy as we did in (6. or NMR. the quantity which our imaginary-time perturbation theory is designed to compute – Gret (q. −ω) = dΩ dω 2π ki 2π Q 6.90 HAPTER 6. v (6.85) ρ(q) = Let us assume that the displacements of the ions are small and expand the exponential. we can write this as: d2 σ 1 kf M 2 1 = Im {Gret (q.

91) or.n n |S− (−q. 1/T1 . therefore. −ω0 )S+ (q. ω0 )| n e−βEn (2π)2 n d2 q A(q) T r e−βH S− (−q. n 2 −βEn e m.6. the up-spin state has a ﬁnite lifetime. T1 . using the ﬂuctuation-dissipation theorem. ω0 ) (2π)2 (6. ω0 ) d2 q A(q) +− −βω0 (2π)2 1−e (6.n n |S− (q)| m m |S+ (q)| n e−βEn dt eiω0 t m. phonons. If an up-spin state were an energy eigenstate.9. t)| n e−βEn d2 q A(q) n |S− (−q. m H ′ ↑. The width of the resonance is. The lifetime of the up-spin state is given by: 1 = T1 ↓.90) Following the steps which we used in the derivation of the scattering crosssection. the absorption cross section would have a δ-function at ω0 . A measurement of T1 is an important probe of the spin-carrying excitations of a system. As a result of the interaction between nuclear spins and the other excitations in the system (electrons.89) A(q) is the hyperﬁne coupling between the the nuclear spin I and the spin density S(q) due to the excitations of the system. NMR RELAXATION RATE 91 an energy splitting ω0 between the up-spin excited state and the downspin ground state of the nuclei (let’s assume spin-1/2 nuclei). −ω0 )| m m |S+ (q.92) . ω0 )| n e−βEn (2π)2 m.n d2 q A(q) n |S− (−q. 1 = T1 χ′′ (q. The interaction Hamiltonian for the coupling between a nuclear spin and the other excitations is: Hint = d2 q A(q) [I+ S− (q) + I− S+ (q)] (2π)2 (6. 0)| m m |S+ (q. magnons). then electromagnetic radiation at frequency ω0 would be perfectly resonant with the nuclear spins. −ω0 ) S+ (q.n δ (ω0 + En − Em ) (6. we rewrite this as: 1 = T1 = = = = d2 q 1 A(q) (2π)2 2π 1 A(q) 2 (2π) 2π d2 q ∞ −∞ ∞ −∞ dt ei(ω0 +En −Em )t m.

MEASUREMENTS AND CORRELATION FUNCTIONS C ω0 is usually a very small frequency. spin waves.93) .92 HAPTER 6. so we can take ω0 → 0: 1 = T1 T d2 q 1 A(q) lim χ′′ (q. compared to the natural frequency scales of electrons. ω) +− 2 ω→0 ω (2π) (6. etc..

∞ −∞ dxe− 2 ax = 1 2 2π a 1/2 (7.1 Gaussian Integrals We will now shift gears and develop a formalism which will give us a fresh perspective on many-body theory and its associated approximation methods.CHAPTER 7 Functional Integrals 7. We can generalize this to integration over n variables. Consider the Gaussian integral. This formalism – functional integration – will also reveal the underlying similarity and relationship between quantum and classical statistical mechanics. 93 .1) This integral is well-deﬁned for any complex a so long as Re{a} > 0.2) and even to integration over complex variables zi with dn zdn z ∗ ≡ dn (Rez)dn (Imz). dn xe− 2 xi Aij xj = (2π)n/2 (detA)−1/2 1 (7. dn zdn z ∗ e−zi ∗A ij zj = (4π)n (detA)−1 (7.3) so long as Aij is a symmetric matrix with Re{Aij } > 0.

. . we can also do the integrals of polynomials multiplying Gaussians: dn x P (x1 . the generating functional. ∂j f (x0 ) = 0 for all j.7) Where x0 is a stationary point of f (xi ).. Such an expression for the generating functional will facilitate many formal manipulations such as changes of variables and symmetry transformations. we can do the integral of the exponential of a i quadratic form: dn xe− 2 xi Aij xj +bi xi = (2π)n/2 (detA)−1/2 e 2 1 1 bi (A−1 )ij bj (7.6) A non-Gaussian integral can often be approximated by a Gaussian integral using the saddle-point approximation: dn x e−λf (xi ) ≈ dn x e −λf (x0 )−λ(∂i ∂j f )x i 0 0 i =xi (xi −x0 )(xj −x0 ) j i 1 −2 = (2π)n/2 e−λf (xi ) det (λ∂i ∂j f )xi =x0 i (7.e. i. xn ) e− 2 xi Aij xj +bi xi = P 1 ∂ ∂ . FUNCTIONAL INTEGRALS By completing the square.5) By diﬀerentiating with respect to bi . . If we blithely allow n to be inﬁnite (ignoring the protests of our mathematician friends). Nothing that we have done so far depended on having n ﬁnite. In the next section. ∂b1 ∂bn (2π)n/2 (detA)−1/2 e 2 1 bi (A−1 )ij bj (7.. can be expressed in this way. . Z. and d xe n 1 − 2 xi Aij xj → Dx(t) e − R tf ti “ dt 1 x(t) − 2 d2 dt2 ” x(t) (7. This i i approximation is good in the λ → ∞ limit where the minimum of f (xi ) dominates the integral. we have the Gaussian functional integral.94 CHAPTER 7. we will do this by making the replacement i → t.4) where x0 = (A−1 )ij bj . .8) As we will see. . 1 1 1 xi Aij xj + bi xi = xi − x0 Aij xj − x0 − bi A−1 i j 2 2 2 b ij j (7. xi → x(t).. It will also guide our intuition about quantum mechanical processes and emphasize the connections with classical statistical mechanics.

.. We will now take the desired matrix element (7. . e−δτ H xi and insert a resolution of the identity. ti ) = e−(τf −τi )H We would like to compute xf |U (τf . the integration measure will be non. . τi )| xi In order to do this. We could just as well write down the functional integral without any further ado and justify it by the fact that it gives the same result as canonical quantization . .give an argument relating the matrix elements of the evolution operator.7. our reason for discussing Feynman’s derivation lies its heuristic value and its intuitive appeal. we will insert a resolution of the identity in terms of an overcomplete set of states. This derivation can be made more or less rigorous. Here.following Feynman .9) H= 2m Then the imaginary-time evolution operator is given by U (tf . we will . U .. THE FEYNMAN PATH INTEGRAL 95 7. but shouldn’t be taken overly seriously.10) (7.2. .11) (7. .ultimately. x1 e−δτ H xi (7. dxNτ −1 xf e−δτ H xNτ −1 .12) in the limit δτ = (τf − τi )/Nτ → 0. dx|x x| = 1 between each factor of e−δτ H : e−δτ H = . we will write (the Trotter product formula) e−(τf −τi )H → e−δτ H Nτ (7. (7. Then p2 + V (x) (7.11). of a free particle to a Gaussian functional integral. . this is its real justiﬁcation. As a result.trivial.2 The Feynman Path Integral In this section.14) (7. dx1 .13) xf e−δτ H .15) When we repeat this derivation for a quantum mechanical spin in the next chapter. Suppose that we have a particle in one dimension moving in a potential V (x).

Observe that e−δτ ( 2m +V (x)) = e−δτ ( 2m ) e−δτ (V (x)) + O(δτ 2 ) Hence. we have e Hence. . FUNCTIONAL INTEGRALS We now make an approximation which is accurate to O(δτ 2 ). . we write δτ → dτ (7. we can write xn e−δτ H xn−1 = xn e−δτ ( 2m ) xn−1 e−δτ (V (xn−1 )) + O(δτ 2 ) (7. xf |U (τf .20) nδτ → τ xn → x(τ ) n In the δτ → 0 limit. τi )| xi = .23) .21) (but we make no assumption that x(τ ) is diﬀerentiable or even continuous) and xf |U (τf .18) dpn eipn (xn −xn−1 ) e−δτ e−δτ V (xn−1 ) Note that the second line is an ordinary integral of c-numbers.96 CHAPTER 7. dx1 .. In the δτ → 0 limit.22) where SE is the imaginary-time – or ‘Euclidean’ – action: tf 2 SE [x(τ )] = ti dτ dx 1 m 2 dτ + V (x) (7. we will have an exact (though formal for arbitrary V (x)) expression for the desired matrix element. dxNτ −1 e − P m n δτ ( 2 −δτ „ m 2 “x n −xn−1 δτ ”2 « +V (xn−1 ) (7. Doing the Gaussian pn integral..16) We now insert a complete set of momentum eigenstates into the righthand-side of this expression: xn e−δτ p2 2m xn−1 e−δτ V (xn−1 ) = = dpn xn |pn pn e−δτ p2 2m p2 n 2m xn−1 e−δτ V (xn−1 ) (7.19) “x n −xn−1 δτ ”2 +V (xn−1 )) (7. τi )| xi = Dx(τ ) e−SE (7.17) p2 p2 p2 (7.

we will simply demonstrate that the generating functional. is given by3 : Z[j] = N Du e−SE [j] (7. x(τn ) e − R tf ti dτ “ 1 2 ” 2 m( dx ) +V (x) dτ (7. For a more general potential. τ ) and N is a normalization factor. . 7. etc. x(tf ) = xf and N is a ‘normalization constant’ into which we have absorbed factors of m. THE FUNCTIONAL INTEGRAL IN MANY-BODY THEORY 97 The path integral representation suggests a beautiful interpretation of the quantum-mechanical transition amplitude: the particle takes all possible trajectories with each trajectory x(τ ) contributing e−SE [x(τ )] to the amplitude.3 The Functional Integral in Many-Body Theory Instead of following the steps of the previous section to derive the functional integral for a ﬁeld theory.6) Dx(τ ) e− (−1)n n! R tf ti dτ tf ti 1 2 m( dx ) +V (x) dτ ′ 2 = n ∞ dt V n=0 ∂ ∂b(t′ ) Dx(τ ) e − R tf ti dτ “ 1 2 ” dx 2 m( dτ ) +b(t)x(t) (7. π. . Z[j].3. V (x) = 2 x2 /2. x(τn ) U (τf . . this follows because the T -symbol puts the operators in precisely the right order so that they can act on the appropriate resolutions of the identity and become c-numbers. .7. If the particle is free or is in a harmonic oscillator potential. τi ))| xi = Dx(τ ) x(τ1 ) .25) Time-ordered expectation values can be simply handled by the pathintegral formalism: xf |T (x(τ1 ) . . the path integral can be deﬁned perturbatively using (7.26) As you will show in the problem set. this is a Gaussian functional integral: mω Dx(τ ) e − R tf ti dτ 1 2 m( dx ) + 1 mω 2 x2 dτ 2 2 = N det − d2 + ω2 dt2 (7.24) where the determinant is taken over the space of functions satisfying x(ti ) = xi .27) where SE [j] is the imaginary-time – or Euclidean – action in the presence of an external source ﬁeld j(x.

30) which follows because 2 G0 = ∂l ∂l −δij ∂τ − (µ + λ)∂i ∂j − µδij ∂k ∂k −1 If we choose N to cancel the determinant. we will ﬁrst show that it is true for a free ﬁeld. we can write the functional integral as: N Du e R −S0 [j]− Lint (∂k uk ) N − R Lint = e N0 “ δ δj ” eW0 [j] (7. FUNCTIONAL INTEGRALS To show that this is true. is given by a single diagram: Z0 [j] = eW0 [j] = e 2 1 R j(x)G0 (x−y)j(y) Z[0] (7. where imaginary-time integrals run from 0 to β. on the other hand.27) is true even for an interacting theory. S = S0 + Lint (∂k uk ). we have precisely the same thing for the generating functional. we have the desired result. With the functional integral representation of Z[j] in hand. Now consider the interacting case.33) .e. g = 0 in our phonon Lagrangian. F (x) = eia·x G(y) = eib·y (7. is given by N 2 Du e−SE [j] = N det −δij ∂τ − (µ + λ)∂i ∂j − µδij ∂k ∂k −1 2 e2 1 R j(x)G0 (x−y)j(y) (7. The inverse is almost always most easily taken in momentum space. Z[j] is given by the exponential of the generating functional for connected Green functions. An important result which follows from this discussion is that the propagator is simply the inverse of the diﬀerential operator in the quadratic term in the action.28) The functional integral.32) This identity can be proven by expanding F and G in plane waves. The generating functional. i.6). we can show that the same relation holds at ﬁnite-temperature. Hence (7. we can give simple proofs of Wick’s theorem and the Feynman rules. W [j]. it’s helpful to use the following identity (which we state for ﬁnite-dimensional vector spaces) F −i ∂ ∂x G (x) = G −i ∂ ∂y F (y) eix·y y=0 (7. To do this. By straightforward extension. in turn. following (7. which.29) (7. Then.98 CHAPTER 7.31) According to Dyson’s formula.

.32) follows. the loop expansion is an expansion in powers of . we can obtain another perspective on the expansion in powers of .35) from which (7. . . Using this. SADDLE POINT APPROXIMATION. e 2 j(x)G0 (x−y)j(y) = δj(x1 ) δj(x1 ) = e 1 2 j=0 R G0 (x−y) δ(∂ δ δ i ui (x)) δ(∂j uj (y)) ∂i ui (x1 ) .. (7.4 Saddle Point Approximation. . Using our expression (7. . .37) Using our identity.4.32) we can rewrite this as: Z[j] = N e 1 2 R G0 (x−y) δ δ δ∂k uk (x) δ∂j uj (y) e− R Lint (∂k uk )+j∂k uk (7. LOOP EXPANSION The left-hand-side of (7.38) which is a compact expression of the Feynman rules for the generating functional. In the same way. we can derive the Feynman rules for an interacting theory. Using the functional integral.7.32) is eia·∂/∂x eib·x = eib·(x+a) while the right-hand-side is: eib·∂/∂y ei(x+a)·y = ei(x+a)·(y+b) 99 (7. Z[j] = N e − R Lint R “ δ δj ” eW0 [j] ” =N e − Lint “ δ δj e2 1 R j(x)G0 (x−y)j(y) (7.4.34) (7. 7.39) . the functional integral is Z[j] = N 1 Du e− (SE [u]+ R j∂k uk ) (7. ∂i ui (xn ) (7. δj(xn ) R 1 δ δ .31). Restoring the .36) ui =0 This is Wick’s theorem in the form in which we rewrote it at the end of section 4. Loop Expansion As we pointed out in chapter 5. we can compute the time-ordered product of a string of ﬁelds in a free ﬁeld theory: 0|T (∂k uk (x1 ) . ∂k uk (xn )) |0 = δn Z[j] δj(x1 ) . .

a propagator Kij which is not attached to a ∂i j. In the terms not shown.41) (7. .e. (7. this is given by: 1 c Z[j] = N e− (SE [u ]+ R j∂k uc ) k (7.) (7.46) This gives the following contribution to the generating functional of connected diagrams: g −1 g T r ln −Kij + ∂i ∂j (∂k uc )2 = T r ln (−Kij ) + T r ln 1 − Kji ∂i ∂j (∂k uc )2 k k 2 2 g 2 −1 = T r ln (−Kij ) − T r Kji ∂i ∂j (∂k uc ) + k 2 .40) where uc is a classical solution of the equation of motion: Kij (x − y)uj (y) − and 2 Kij (x − x′ ) = δ(x − x′ ) δij ∂τ + (µ + λ)∂i ∂j + µδij ∂k ∂k g ∂i (∂k uk )3 = −∂i j(x) 3! (7.43) and solve this iteratively: −1 ui (x) = −Kij (x−y)∂j j(y)+ g −1 −1 K (x−x′ )∂l ∂k Kkj (x′ − x′′ )∂j j(x′′ ) 3! il 3 +.100 CHAPTER 7.45) This is the contribution to the generating functional for connected diagrams coming from tree-level diagrams. FUNCTIONAL INTEGRALS At saddle-point level. .42) If we use the classical solution which is obtained by starting with the g = 0 solution: −1 ui (x) = −Kij (x − y)∂j j(y) (7. The Gaussian ﬂuctuations about the saddle point contribute to the functional integral a factor of g T r ln det −Kij + ∂i ∂j (∂k uc )2 = e k 2 “ ” 2 −Kij + g ∂i ∂j (∂k uc ) k 2 (7.44) we obtain: W [j] = ln Z[j] = 1 1 −1 ( ∂i j(x)Kij (x − y)∂j j(y) 2 1 g −1 −1 − ∂i j(x) Kil (x − x′ )∂l ∂k Kkj (x′ − x′′ )∂j j(x′′ ) 2 3! 3 + . . . for each additional −1 vertex carrying a g we have one extra internal line – i.

k k 2 2 2 (7. . the uc ’s attach all possible tree diagrams to this loop. observe that each propagator comes with a factor of (since it is the inverse of the quadratic part of the action) while each vertex comes with a factor of 1/ (from the perturbative expansion of eS/ ). t) and can be absorbed in the normalization. t))2 × k 2 −1 c ′ ′ 2 ′ ′ g Kli (x − x . namely the i one which can be obtained by solving the classical equations perturbatively about g = 0.5 7. . t))2 k 2 + 7. The k third term is the loop obtained by connecting two points by two propagators and again attaching all possible tree diagrams.1 The Functional Integral in Statistical Mechanics The Ising Model and ϕ4 Theory The functional integral representation of the generating functional of a quantum mechanical many-body system bears a strong resemblance to the parti- . t ) ) + . which proves the claim.5.48) 2 The ﬁrst term is independent of j(x. Hence. and so on. I − V = L − 1.7. . .5. To see that the L-loop diagrams give the O( L−1 ) contribution to W [J]. the tree-level diagrams give the O(1/ ) contribution to W [J] while the one-loop diagrams give the O(1) contribution. writing out the traces. THE FUNCTIONAL INTEGRAL IN STATISTICAL MECHANICS 101 1 −1 g −1 g T r Knm ∂m ∂l (∂k uc )2 Kli ∂i ∂j (∂k uc )2 + . The rest of the series gives the connected one-loops contributions. The second term is the loop obtained by connecting a point to itself with a propagator. g −1 d3 xdt Kji (0) ∂i ∂j (∂k uc (x. we must. (7. In principle. W1−loop [J] = d3 xdt ln (−Kii ) (0) − 1 2 g −1 d3 xdtd3 x′ dt′ (Kjm (x′ − x. The next term (not written) is the loop obtained by connecting three points with three propagators. According to the graphical argument we gave in chapter 5. Note that we have chosen a particular saddle-point. uc . t − t ) ∂i ∂j ∂k uk (x . t′ − t) ∂m ∂l (∂k uc (x.47) or. sum the contributions from all saddle-points of the functional integral. Hence a diagram with I internal lines and V vertices is O( I−V ). of course.

Hence.102 CHAPTER 7. and we can write 1 J(q) = J0 − J2 q 2 + O(q 4 ) 2 so long as J(i − j) falls oﬀ suﬃciently rapidly. we have gone back to real space and taken the continuum limit.56) .j (J −1 )ij ϕi ϕj (7.j ϕϕ = ij i j 1 dd q ϕ(q)ϕ(−q) d J(q) (2π) (7. To see the correspondance.j (J −1 )ij ϕi ϕj + P i ln cosh ϕi (7. Expanding the other term.49) H=− 2 i. 1 Jij σi σj (7.51) σi The sum over the σi ’s can be done. giving: Z=N Dϕi e −T 2 P i. ϕi → ϕ(x).53) The momenta are cutoﬀ at large q by the inverse lattice spacing.54) In the second line.50) We can introduce auxiliary variables. Indeed the formal similarity between the two allows us to use the same language and calculational techniques to analyze both. we can write dd q 1 ϕ(q)ϕ(−q) = d J(q) (2π) = dd q (2π)d dd x 1 J2 ϕ(q)ϕ(−q) + 2 q 2 ϕ(q)ϕ(−q) + O(q 4 ) J0 2J0 J2 1 2 2 4 (7. let’s consider the Ising model.55) 2 (∇ϕ) + J ϕ + O (∇ϕ) 2J0 0 (7. FUNCTIONAL INTEGRALS tion function of a classical statistical mechanical system. then the ﬁrst term in the exponential can be brought to a more convenient form by Fourier transforming: J −1 i.52) σi If Jij = J(i − j).j Jij σi σj (7. to rewrite this as: Z=N Dϕi e P i ϕi σi e −T 2 P i. 1 1 ln cosh ϕi = ln 2 + ϕ2 − ϕ4 + O ϕ6 2 12 (7. ϕi . The classical partition function is: Z= e 2T σi 1 P i.j where the spins σi = ±1 lie on a lattice. this cutoﬀ is unimportant. If we are interested in small q.

. we assign a factor 1/S to the diagram. . . jm (m ≤ 4) and to external momenta p1 . r = T /J0 − 1. write: u (2π)d δ(p1 + p2 + p3 + p4 ) • Imagine labelling the vertices 1.57) 2 where K = T J2 /2J0 . . jm (m ≤ 4) and to external momenta p1 . . . . we assign a factor 1/l! to the diagram. 2. . For external lines. . .7. Z=N Dϕ e− R 1 1 dd x ( 1 K(∇ϕ)2 + 2 rϕ2 + 4! uϕ4 ) 2 (7. . . THE FUNCTIONAL INTEGRAL IN STATISTICAL MECHANICS 103 If we neglect the O((∇ϕ)4 ) and higher gradient terms – which seems reasonable in the small q limit – as well as powers of ϕ higher than the quartic term – which is not obviously the right thing to do. Consider a permutation of these labels. u = 2. .5. The similarity between the functional integral (7. . .57) and the functional integral for our theory of interacting phonons allows us to immediately deduce its Feynman rules: • Assign a directed momentum to each line. . . If S is the number of permutations which leave the diagram invariant. Such a permutation leaves the diagram invariant if. . . The weighted sum over all possible classical histories is in direct analogy with the sum over all classical conﬁgurations. the imaginary-time functional integral which we have introduced for the generating functional of quantum-mechanical correlation functions is analogous to the classical partition function. Hence. Vertex i will be connected to vertices j1 . • For each internal line with momentum q write: − 1 dd p d Kp2 + r (2π) • For each vertex with momenta p1 . for all vertices i. . . . n. p4−m . i is still connected to vertices j1 . p4−m . but later will be shown to be reasonable – then we can write the partition function as the following functional integral. • If two vertices are connected by l lines. . p4 directed into the vertex. . the momentum is directed into the diagram. .

7. Much of what we have to say about quantum-mechanical manybody systems can be applied to classical systems with little modiﬁcation. β.2 Mean-Field Theory and the Saddle-Point Approximation We can apply the saddle-point – or 0-loop – approximation to our functional integral (7. If r > 0. let’s use the full . as we have seen. ϕ = 0. there is only ϕ=0 However. √ for T < Tc there is spontaneous magnetizaand tion in the system. Of course.60) (7. FUNCTIONAL INTEGRALS In classical equilibrium statistical mechanics. at T = Tc . According to the saddle-point approximation.104 CHAPTER 7. So long as the system is rotationally invariant. Of course. you will ﬁnd the Ginzburg criterion which determines whether higher-loop processes invalidate the saddle-point analysis. To make this more obvious. For T > Tc . The saddle-point analysis reproduces and is completely equivalent to the standard mean-ﬁeld-theory. the theory will have Euclidean invariance in d spatial dimensions. time plays no role. classical statistical mechanics should be contained within quantum statistical mechanics in the small β limit.57). r = 0 occurs at T = Tc = J0 .59) These latter solutions have larger saddle-point contribution: exp(3r 2 /2u) compared to 1 and are therefore more important.5. and therefore be formally the same as the imaginary-time description of a quantum system in d − 1-spatial dimensions (and one time dimension). In the problem set. ϕ ∼ ± Tc − T . there is only one saddle-point solution. The classical equation of motion is: −K∇2 ϕ(x) + rϕ(x) + u 3 ϕ (x) = 0 6 (7. so there are momenta but no frequencies (unlike in quantum statistical mechanics where. a phase transition occurs. in one direction.4 The classical analog of a quantum system at ﬁnite-temperature is a classical system which has a ﬁnite extent. statics and dynamics are intertwined).58) Let’s look for spatially uniform solutions ϕ(x) = ϕ. for r < 0 there are also the solutions ϕ=± −6r u (7. we shouldn’t stop with the saddle-point approximation but should include higher-loop processes.

61) The saddle-point equation is: T ϕ = tanh ϕ J0 (7. For purposes of comparison.56) rather than the one truncated at quartic order: Z=N Dϕ e − R dd x “ 1 T K(∇ϕ)2 + 2J ϕ2 −ln cosh ϕ 2 0 105 ” (7. THE TRANSFER MATRIX** potential (7. this is: σ = tanh J0 σ T (7.j by a mean-ﬁeld.66) Using the self-consistency condition. h: H=− with h given by h= i h In this ﬁeld.65) σ = tanh h T (7.7. let’s recapitulate mean-ﬁeld theory.62) This is the usual self-consistency condition of mean-ﬁeld theory which predicts Tc = J0 . We replace the eﬀective ﬁeld which each spins sees as a result of its interaction with its neighbors. 7. 1 H=− Jij σi σj (7. the partition function is just 2 cosh T and hσi i (7.6.63) 2 i.64) Jij σi = J0 σi (7.6 The Transfer Matrix** .67) which is the saddle-point condition above.

106 CHAPTER 7. FUNCTIONAL INTEGRALS .

Part III Goldstone Modes and Spontaneous Symmetry Breaking 107 .

.

φ of Ω z= e−iφ/2 cos θ 2 eiφ/2 sin θ 2 109 (8.5) (8. . One basis is: S z |sz = sz |sz .CHAPTER 8 Spin Systems and Magnons 8. α = ±1/2: |Ω = z α |α (8. sz = −s. s (8.2) where Ω is a unit vector.1 Coherent-State Path Integral for a Single Spin Let us follow Feynman’s derivation of the functional integral to formulate a functional integral for a quantum-mechanical spin. .1) For the functional integral representation. A quantum mechanical spin of magnitude s has a 2s + 1-dimensional Hilbert space of states. . −s + 1.4) . s − 1. For s = 1/2. we can write this basis in terms of the spinor z α . . it is more convenient to use the overcomplete coherent state basis |Ω : S · Ω|Ω = s|Ω (8.3) where: Ω = z ∗α σαβ z β in terms of the spherical angles θ.

. σ2s σ1 + . we can simply symmetrize 2s spin−1/2’s: |Ω = z α1 . the resolution of the identity is given by: 2s + 1 I= d2 Ω |Ω Ω| (8. SPIN SYSTEMS AND MAGNONS and z ∗α z α = 1. ˆ|Ω = z = Hence the general relation is: Ω′ | Ω = 1 1 + Ω′ · Ω e−iφ 2 s ∗α 2s ∗α (8.7) (8. α2s (8.110 CHAPTER 8. . . . .6) (8. .8) 1 (1 + cos θ) e−iφ 2 1 1 + ˆ · Ω e−iφ z 2 s s (8. but so long as we choose a phase and stick with it. .13) (8. there is no problem. The usefulness of this basis lies in the following property: Ω|f (S)|Ω = f (sΩ) To see this. . In this basis. α2s with Ω′ |Ω = z ′ z α In terms of the spherical angles. Of course. . z α2s |α1 .. .10) where φ is the phase of z ′ ∗α z α . there is arbitrariness in our choice of the overall phase of z α . z ∗β2s β1 . .7) to write this as: Ω|f (S)|Ω = z ∗β1 . z α2s |α1 . β2s |f = f (sΩ) where we have used Ω = z ∗α σαβ z β in the second line. + 2 2 z α1 . Therefore. . . . .11) 4π as may be seen by taking its matrix elements between states s| and |s − n . . the states |Ω and |Ω′ have overlap: Ω′ |Ω = z ′ z α To obtain larger s.. .9) (8. .12) . use (8.

19) term can be dropped and we can write the partition function as T r e−βH(S) = N DΩ(τ ) e−Sspin [Ω] dz α dτ (8.20) where Sspin [Ω] = dτ H(sΩ) − 2sz ∗α dφ cos θ(τ ) dτ (8.14) where N ∆τ = β. Then we can write the partition function as: Tr e −∆τ H(S) N N = i=1 2s + 1 2 d Ωi Ωi+1 e−∆τ H(S) Ωi 4π (8. COHERENT-STATE PATH INTEGRAL FOR A SINGLE SPIN 111 Let us construct the functional integral representation for the partition function of a single spin.21) In terms of the spherical angles.17) while the third term follows from esφ ≈ es sin φ ∗α α ∗α α = es(zi zi+1 −zi+1 zi ) Hence. Following our derivation of the path integral in chapter 8. we have Tr The e −∆τ H(S) N N (8. the second term can be written 2s z ∗α dz α = is dτ dτ .1.16) The second term in the exponent was obtained by making the approximation 1+Ω ·Ω ′ s = 1 Ω − Ω′ 2− 2 2 s ≈2 s 1 dΩ 1 − (∆τ )2 4 dτ 2 s 2 dΩ s ≈ 2s e− 4 (∆τ ) ( dτ ) 2 (8.15) Taking ∆τ → 0. we write the imaginary-time evolution operator as e−βH(S) = e−∆τ H(S) N (8. we have Ωi+1 e−∆τ H(S) Ωi 2 dΩ 1 ≈ e−∆τ H(sΩi )−s 4 (∆τ ) ( dτ ) 2 ∗α α ∗α α +s(zi zi+1 −zi+1 zi ) (8.18) = N lim ∆τ →0 i=1 P 2s + 1 2 d Ωi e i 4π “ ” α 2 s −∆τ H(sΩi )− 4 (∆τ )2 ( dΩ ) +2s∆τ z ∗α dz dτ dτ (∆τ )2 (8.8.

Actually. If the spin is in a magnetic ﬁeld. these two areas add up to 4π and e4πis = 1. δ z ∗α while δΩ = δz ∗α σαβ z β + z ∗α σαβ δz β so that δ z ∗α dz α dτ = δΩ · Ω × dΩ dτ (8.22) where A is the area of the region of the sphere enclosed by the curve traced out by Ω(τ ). We can check that we get the correct equation of motion from this Lagrangian. so there is some ambiguity in the deﬁnition of A. As you will show in the problem set. S=s dτ −i cos θ dφ dτ (8. SPIN SYSTEMS AND MAGNONS = is = is dφ cos θ(φ) 1 dφ 1− d (cos θ) cos θ(φ) = i s (2π − A) (8. H].25) dz α dτ = δz ∗α dz ∗α dz α − δz α dτ dτ (8. However.23) Hence. then the spin precesses about the magnetic ﬁeld: dΩ i = µΩ × B (8.29) . the curve traced out by Ω(τ ) divides the surface of the sphere into two pieces. but is otherwise free. Ω and dΩ/dτ are perpendicular and the equation of motion can be written is ∂H dΩ =Ω× dτ ∂Ω (8.24) (8.28) dτ The time-derivative term. H = µ B · S.26) Since Ω is a unit vector. the equation of motion following from our Lagrangian is: isΩ × ∂H dΩ = dτ ∂Ω (8. so we can take either choice of A.27) which is what we expect from dΩ/dτ = [Ω.112 CHAPTER 8.

we have δΩ(τ ) = δr(dΩ/dr) and: β δS = s 0 β dτ δr dτ 0 −i dΩ dΩ ·Ω× dr dτ dΩ dτ (8.33) so that δΩ(τ ) = Ω(r ′ . Ω(r. z z Ω(1. Then.30) S=s 0 dτ 0 dr −i dΩ dΩ ·Ω× dr dτ (8. τ ). τ ) = ˆ. the action depends only on Ω(1.1. observe that the integrand is equal to the Jacobian of the map from the (r. τ ) = Ω(τ ). ˆ and Ω(τ ): Ω(0.32) where Ω(r.31) This clearly leads to the same equation of motion and leads to the following interpretation of a spin: the spin can be modelled by a charged particle moving on the surface of a sphere with a magnetic monopole of magnetic charge s located at the origin. .8. writing δr = r ′ − 1. the action is given by the magnetic ﬂux through the area enclosed by the orbit.34) =s −i δΩ · Ω × from which the correct equation of motion follows. τ ) thereby covers the region of the sphere enclosed by Ω(τ ). τ ) − Ω(1. By Stokes’ theorem. τ ) = Ω(τ ) and is independent of the particular interpolating function Ω(r. τ ) interpolates between the north pole. To see that the same equation of motion results from this form of the action.32) gives the area of this region. Hence. τ ) plane to the surface of the sphere. imagine varying the upper limit of integration: β r′ S=s 0 dτ 0 dr −i dΩ dΩ ·Ω× dr dτ (8. Terms of this form are often called Wess-Zumino terms. To see that the action (8. COHERENT-STATE PATH INTEGRAL FOR A SINGLE SPIN 113 can be written in the alternative form: S=s where A(Ω) satisﬁes ∇Ω × A Ω = s Ω (8. It can also be written in the form β 1 dτ −i A Ω · dΩ dτ (8. τ ).

is 2d. z.114 CHAPTER 8.38) where the sum is over the z nearest neighbors.j with J > 0 and the sum restricted to nearest neighbors.36) is dτ j The ground state of the Heisenberg ferromagnet is one in which the SU (2) spin-rotational symmetry is spontaneously broken.2 8.40) .39) In the small q limit. For a square lattice of spacing a in d-dimensions. the spin-waves have quadratic dispersion: E(q) = 2Jsa2 q 2 (8. Ωi = Ω0 + δΩi . SPIN SYSTEMS AND MAGNONS 8. The system chooses a direction along which to order. Linearizing about the uniform ground state. we have the dispersion relation of spin-wave theory: z E(q) = Js z − eiq·δi i=1 (8. the cordination number.35) i.37) j Substituting a plane-wave solution. For a d-dimensional hypercubic lattice. δΩ = ǫ eiq·Ri . we have: is dδΩi = −Js2 dτ Ω0 × δ Ωj − δ Ωi (8. this gives: d E(q) = 2Js d − cos qi a i=1 (8. Let us call this direction Ω0 . The equation of motion is: dΩi = −Js2 Ωi × Ωj (8.2.1 Ferromagnets Spin Waves Suppose that we have the ferromagnetic Heisenberg Hamiltonian: H = −J Si · Si (8.

Since Si = s(s + 1). τ )m− (0.44) Introducing the ﬁelds m± = mx ± imy . my ) = s(0.2. we linearize the Lagrangian about an ordered state which. Si · Si = 1 Si − Sj 2 2 + const. without loss of generality. my . 1) = sΩ Hence. We write z Ω = (mx . 1 − m2 − m2 ) x y (8. (8. mx .41) the action takes the following form in the continuum limit: S=s dd x dτ −i A Ω · dΩ 1 + D ∇Ω dτ 2 2 (8.46) (8. 0.2. 0) 2 (8. FERROMAGNETS 115 8. my are small so that we can neglect all terms in the action higher than quadratic.42) where D = 2Jsa2 .48) n .2 Ferromagnetic Magnons The small q behavior can be obtained directly from the continuum limit of 2 the Lagrangian. we learned that the propagator is simply the inverse of the diﬀerential operator in the quadratic part of the action. The inverse can be taken trivially in momentum space: Tτ (m+ (x. my ) = since ∇m × A(mx .43) and assume that mx .47) In chapter 8. 0)) = 2 1 s β dd p ei(p·x−ωn τ ) (2π)d iωn − Dp2 (8. As in the previous section.45) s (−my . we take to be Ω = ˆ. we can write: S=s dd x dτ ∂m− 1 1 m+ + D ∇m+ · ∇m− 2 ∂τ 2 (8. we can write the action as: S=s dd x dτ ∂mj 1 1 iǫji mi + D(∇mi )2 2 ∂τ 2 (8. We can now write A(mx .8.

Magnons are the quantum particles which correspond to spin waves in analogy with the correspondence between phonons and sound waves or photons and electromagnetic waves. τ ) = 1 − 1 m+ (x. τ )m− (x. SPIN SYSTEMS AND MAGNONS Alternatively. Note that at T = 0. the exact . Ωz (x. the magnetization decreases as Mz (0) − Mz (T ) ∼ T 2 . Ωz = Hence.116 CHAPTER 8.49) and compute the propagator directly as we did for phonons. To lowest order in m± . In d ≤ 2. we have converted the sum over Matsubara frequencies to an integral.50) 1 dω 1 dd p 1 nB (ω) − d 2 s (2π) 2πi ω + iδ − Dp ω − iδ − Dp2 dp d 1 =1− dω nB (ω)δ(ω − Dp2 ) s (2π)d dd p 1 nB Dp2 =1− s (2π)d d d 1 2π2 T d/2 ∞ x 2 −1 dx =1− (8.48). 3 Hence in d = 3. τ ) 2 1 1 dd p 1 =1− sβ n (2π)d iωn − Dp2 1 1 − m+ m− ≈ 1 − m+ m− 2 (8. τ ) = dd k 2 a† e−Dk τ +ik·x d/2 k (2π) dd k 2 a eDk τ −ik·x d/2 k (2π) (8. a† and ak k are called magnon creation and annihilation operators. to annihilate it. as usual. the integral is divergent so we cannot trust the approximation (8. however. In the ground state. In fact. we ﬂip the spin back up. this divergence is a sign that the magnetization vanishes for any ﬁnite temperature in d ≤ 2.51) 2s (2π)d Γ(d/2) D ex − 1 0 =1− In the third line. we ﬂip one spin down with a† . we can compute the magnetization as a function of temperature. To create a magnon. τ ) = m+ (x. we can expand m± in normal modes: m− (x. obtaining a contribution only from the real axis.50). Using the propagator (8. all of the spins point up.

55) Unlike in the B = 0 case.53) The propagator is now S=s dd x dτ Tτ (m+ (x. we would have interactions between the magnons. 0)) = 2 1 sβ dd p ei(p·x−ωn τ ) (2π)d iωn − Dp2 − µsB/2 (8. Magnon-magnon interactions aﬀect. τ )m− (x. but we will discuss the analogous interactions in the next section in the context of antiferromagnetism. FERROMAGNETS 117 ground state of the ferromagnetic Heisenberg Hamiltonian is fully polarized for arbitrary d. For d > 2.8. there is a minumum energy cost µB to ﬂip a spin.2. . Then the action is: z S=s dd x dτ −i A Ω · dΩ 1 + D ∇Ω dτ 2 2 + µ s B Ωz (8. the magnetization decreases continuously from its full value as the temperature is increased. If we repeat the calculation of the magnetization as a function of temperature. Thus far. B. By including these terms. then we have the quadratic action: z ∂m− 1 1 m+ + D ∇m+ · ∇m− − µ s B m+ m− 2 ∂τ 2 (8.2. τ )m− (0. At zero temperature. we have neglected anharmonic terms in the magnon Lagrangian. We will not discuss these interactions here. τ ) = 1 − 1 m+ (x. we ﬁnd: Ωz (x.54) n As a result of the magnetic ﬁeld. for instance. τ ) 2 1 dd p =1− nB Dp2 + µsB/2 s (2π)d (8. the magnetization will line up along the direction of the ﬁeld.52) where µ is the gyromagnetic ratio.3 A Ferromagnet in a Magnetic Field Suppose we place our ferromagnet in a magnetic ﬁeld. this integral is not infrared divergent: the magnetic ﬁeld stabilizes the ferromagnetic state against thermal ﬂuctuations. Let us suppose that the ﬁeld is in the ˆ direction. For d ≤ 2. Ω = Ωz ˆ. 8. the magnetization as a function of temperature. Mz discontinuously jumps to zero as the temperature is raised above zero. If we expand about the ordered state.

57) and 1 ∇Ω × (−1)i n × l = (−1)i+1 n + l s 1 i = (−1) n + l + (−1)i+1 ∂x n s . .j Si · Si (8. we can write H = Js2 i.(8. we have: H = va−d where v = 2Jsa. n and l are assumed to be slowly varying and n · l = 0.1 Antiferromagnets The Non-Linear σ-Model Let us now consider the antiferromagnetic case.j −n(xi ) · n(xj ) + 1 l(xi ) · l(xj ) s2 1 l(xi ) + l(xj ) s2 2 = Js2 i.58) going to the continuum limit.j (n(xi ) − n(xj ))2 + + const.57) l is the q = 0 part of the spin ﬁeld while n is the q = (π/a. .3 8. . We only keep the Fourier modes near these wavevectors. Then Ω2 = 1 is satisﬁed to O(1/s2 ) if i n2 = 1. SPIN SYSTEMS AND MAGNONS 8. H=J i. We expand about the state in which the spins are staggered – the N´el state: e Ωi = (−1)i n(xi ) + 1 l(xi ) s (8.118 CHAPTER 8.56) with J > 0 and the sum restricted to nearest neighbors.59) Using (8. π/a) part.3. . With this decomposition.60) dd x 1 2 sa l + (∇n)2 sa 4 (8. The corresponding action is: S = a−d 1 i dd x dτ (− A (−1)i n + l · a s 1 sa + v l2 + v (∇n)2 ) sa 4 (−1)i dn 1 dl + dτ s dτ (8.

d. . ny . Let us also employ the notation µ = 0. so we can perform the l integral.8.2 Antiferromagnetic Magnons Let us. 1 − n2 − n2 ) x y (8. . Neglecting oscillatory terms in the action. we have: S = a−d dd x dτ dn 1 sa i n×l· + v l2 + v (∇n)2 a dτ sa 4 D l Dn e−S[l.65) This action is called the O(3) Non-Linear σ Model. n2 = 1.61) we can express A in terms of n and l if we drop the gradient term in the penultimate line (this cannot be done in d = 1.3. or O(3) NLσM for short. Integrating out l. 8. . we have: S = a−d dd x dτ sa dn 4v dτ 2 + sa v (∇n)2 4 (8. with 0 referring to the time direction so that ∂0 = ∂t . Then we can write the action of the O(3) NLσM as: 1 S= dd x dτ (∂µ n)2 (8. S= 1 g dd x dτ 1 2v 2 dn dτ 2 + 1 (∇n)2 2 (8. as in the ferromagnetic case. where it is absolutely crucial.n] (8.67) . we expand about an ordered state.63) is a Gaussian integral in l.66) g If. The model is non-linear because there is a non-linear constraint. for simplicity work in a system of units in which v = 1. but can in higher dimensions).64) Or. n = (nx . We can always rescale our time coordinate at the end of any calculation so as to restore v. ANTIFERROMAGNETS ≈Ω 119 (8. 1.62) The functional integral (8.3. O(3). . The O(3) refers to the fact that the ﬁeld n is a three-component ﬁeld which transforms as a vector under the rotation group. writing g = ad−2 /Js2 .

i eik·r+ωk τ + a† e−ik·r−vkτ k. in which there was only one type of magnon. up. 0)) = δij 1 β dd p ei(p·x−ωn τ ) 2 (2π)d ωn + v 2 p2 (8.and down-spin antiferromagnetic x y ± magnons. Note the diﬀerence with the ferromagnetic case. Sfree is very similar to the phonon action. Then the action becomes: S= dd x dτ (∂µ ni )2 + g ni ∂µ ni nj ∂µ nj 1 − gni ni (8.68) where i = 1. as in that case. 2 and n1 = nx . Sfree = and Sint = dd x dτ g dd x dτ (∂µ ni )2 ni ∂µ ni nj ∂µ nj 1 − gni ni (8. respectively.73) n n . restoring v. 0)) = δij or. we can expand ni in normal modes: ni (r. SPIN SYSTEMS AND MAGNONS then we can write the action as: S= 1 g dd x dτ (∂µ ni )2 + ni ∂µ ni nj ∂µ nj 1 − ni ni (8. we can either ﬂip a spin up – thereby creating an up-spin magnon – or ﬂip a spin down – thereby creating a down spin magnon. Let us rescale the ﬁelds so that √ ni → gni . τ )nj (0.120 CHAPTER 8. which we can hope to do when it is small.69) In order to do perturbation theory in g. Since there is no net uniform magnetization in the ground state.71) Note that Sint contains all powers of g when the denominator is expanded in a geometric series.70) (8. we divide the action into two parts.i s (8. We can obtain the antiferromagnetic magnon propagator using this mode expansion or by inverting Sfree: Tτ (ni (x. Tτ (ni (x. n2 = ny .74) 1 β dd p ei(p·x−ωn τ ) 2 (2π)d ωn + p2 (8. t) = 1 dd k √ d/2 (2π) 2k ak.72) a† = a† ± a† create. τ )nj (0.

It is given by nz = Hence. the integral is actually logarithmically divergent.3. τ ) 2 1 dd p 1 1 = 1−g ·2· d ω 2 + v 2 p2 2 β n (2π) n dd p (2π)d dd p (2π)d dd p (2π)d dω nB (ω) 2πiB(ω. Again. This divergence hints at the impossibility of antiferromagnetic order in d = 1. τ )ni (x. we have: nz (x. the staggered magnetization of an antiferromagnet is less than 1 even at T = 0. just as in the ferromagnetic case. .78) If we approximate the Brillouin zone by a sphere. For T = 0.77) Unlike in the ferromagnetic case. In d = 1. the second term does not vanish in the T → 0 limit.76) = 1−g = 1−g = 1−g (8. unlike the uniform magnetization in the ferromagnetic case. with neighboring spins oppositely oriented. τ ) ≈ 1 − g dd p 1 (2π)d 2vp (8. τ ) ≈ 1 − g d−1 a d−1 (2π) 2 (2π)d Γ(d/2) d (8. we can compute the staggered magnetization in the g → 0 limit. the integral (8.79) Hence. The N´el state. |p| < π/a. then we ﬁnd 1 π nz (x.75) With this propagator in hand. ANTIFERROMAGNETS The corresponding spectral function is: B(ω. p) 2πi 1 (nB (vp) − nB (−vp)) 2vp βvp 1 coth 2vp 2 1 − g ni ni ≈ 1 − 1 g ni ni 2 (8. τ ) ≈ 1 − g 1 ni (x. which is a consequence of a theorem which we will prove in the next chapter. is not the exact e ground state. p) = 1 1 δ(ω − vp) − δ(ω + vp) 2vp 2vp 121 (8. For T ﬁnite.77) is logarithmically divergent at small p in d = 2.8. it is a sign of the impossibility of antiferromagnetic order at ﬁnite temperatures in d = 2. nz (x.

. SPIN SYSTEMS AND MAGNONS 8. Matsubara frequencies (p1 .122 CHAPTER 8. . iωn write: − 1 β dd p 1 d ω 2 + v 2 p2 (2π) n n • For each 4-leg vertex with momenta. ωn1 ).0 • For each 6-leg vertex with momenta momenta. ωn1 ). . .3 Magnon-Magnon-Interactions Thus far. ωn4 ) directed into the vertex and indices i1 . . . . we saw that ferromagnetic and antiferromagnetic order are suppressed by thermal ﬂuctuations. (p6 . For external lines. we have ignored the higher-order terms in the NLσM.4 Spin Systems at Finite Temperatures In the previous two sections. . 8. . let’s expand Sint to O(g2 ): Sg 2 = dd x dτ (∂µ ni )2 + g ni ∂µ ni nj ∂µ nj + g2 nk nk ni ∂µ ni nj ∂µ nj 1 − gni ni + . The Feynman rules for this action are: • Assign a directed momentum and Matsubara frequency to each line. (8. . 2 to each line.0 • We assign a factor 1/S to the diagram if there are S permutations of the vertices and external momenta which leave the diagram invariant. To get an idea of the nature of these terms. . . i6 associated to these incoming lines. i4 associated to these incoming lines. .80) We need only these terms in order to do computations to order O(g2 ). . These terms lead to interactions between magnons.3. Assign an index i = 1. . . Let us examine this . ωn6 ) directed into the vertex and indices i1 . . write: g2 δi1 i2 δi3 i4 δi5 i6 (ωi1 ωi3 + p1 · p3 ) (2π)d δ(p1 +p2 +p3 +p4 +p5 +p6 ) β δn1 +n2 +n3 +n4 +n5 +n6 . Matsubara frequencies (p1 . (p4 . • For each internal line with momentum. frequency q. write: g δi1 i2 δi3 i4 (ωi1 ωi3 + p1 · p3 ) (2π)d δ(p1 + p2 + p3 + p4 ) β δn1 +n2 +n3 +n4 . . the momentum and frequency are directed into the diagram.

which have to do with their dynamics. The diﬀerences between ferromagnets and antiferromagnets. consider the case of antiferromagnetism.q)) 2 ” (8. Let us re-write the action in the following dimensionless form: βv/a 1 dn 2 1 ad−1 (8. which is the limit which we have just taken. more generally. the functional integrals for the ferromagnet and antiferromagnet are identical at temperatures large compared to v/a or D/a2 .81) dτ + (∇n)2 dd y S= gv 2 du 2 0 where u = vτ /a and y = x/a.83) We can similarly write the ferromagnetic functional integral with momentum cutoﬀ of order 1 and Matsubara frequencies ωn = 2πna2 /βD: Z= = DΩ e −sad R dd q (2π)d Rβ 0 ” “ 2 dτ −i A(Ω)·∂τ Ω+ 1 (q Ω) 2 1 2 DΩ e −sad β R dd q (2π)d “ (qΩ(ωn =0. Hence. Both systems are described by the d-dimensional NLσM. SPIN SYSTEMS AT FINITE TEMPERATURES 123 more closely.82) then the cutoﬀ is just π (for a spherical Brillouin zone. S= ad−1 gv π 0 dd q 1 (2π)d β n 1 1 |ωn n(ωn . it’s some number of order 1). The Matsubara frequencies are ωn = 2πna/βv. q)|2 + |qn(ωn . When a/βv ≫ 1 – i.8.4. Therefore. for the sake of concreteness.84) Hence. we may make the approximation: Z= ≈ Dn e − a gv d−1 R π dd q 0 (2π)d d−1 R π 0 Rβ 0 1 dτ ( 2 (∂τ n)2 + 1 |qn(q)|2 ) 2 Dn e − βagv 1 dd q ( 2 |qn(ωn =0. q)|2 2 2 (8. the functional integral is dominated by conﬁgurations β which are independent of τ and we can replace 0 dτ → β. at temperatures which are large compared to v/a (the energy of a magnon at the cutoﬀ wavevector) – the conﬁgurations with ωn = 0 are strongly suppressed and give very little contribution to the functional integral. Thus we can kill two birds with one stone by studying the functional integral Z= Dn e−β R 1 dd x ( 2 (∇n)2 ) (8.85) .q)|2 ) (8. Let us.e. If we go to momentum space. are unimportant in the limit of classical statistical mechanics.

SPIN SYSTEMS AND MAGNONS This functional integral would be a trivial Gaussian integral if it were not for the constraint n2 = 1. then this is simply d dλ or. For high-temperature. there is no spatially homogeneous saddle-point solution.88) If we look for a saddle-point solution λ0 for which λ(x) is independent of position. however. let’s intoduce a Lagrange multiplier: Z= Dn Dλ e−β R 1 dd x ( 2 (∇n)2 +λ(x)(n2 −1)) (8.93) . if d > 2. then we have: d−2 − λ0 2 ∼ β (8. there is always a solution: λ0 ∼ 1 ad−2 −β (8. we can’t do the resulting integral exactly.124 CHAPTER 8.90) dd q ln q 2 + λ − βλ (2π)d =0 (8.89) If we aproximate the integral by using λ0 as an infrared cutoﬀ. To impose this constraint.91) For T → 0. The saddle-point of the argument of the exponential is given by: δ δλ T r ln −∇2 + λ(x) − β dd x λ(x) =0 (8.86) Now the functional integral is.) β e a2 (8. λ0 . a Gaussian integral in n which we can do: Z= = Dλ det ∇2 + λ(x) Dλ e− 2 T r ln(−∇ 1 −1/2 e−β R R dd x λ(x) 2 +λ(x) )−β dd x λ(x) (8.87) Unfortunately. but we can try to use the saddle-point approximation. indeed.92) 1 − (const. there is a solution of the form: d−2 λ0 2 ∼ In d = 2. dd q 1 =β d q2 + λ (2π) 1 ad−2 (8.

94). R d 2 1 2 Z = Dn e−β d x ( 2 (∇n) +λ0 n ) with λ0 > 0. This Gaussian theory is called a linear σ-model This describes the high-temperature phase in which thermal ﬂuctuations have disordered the magnet.97) n(x)n(0) ∼ |x|(d−1)/2 at the critical point. • The critical point. T = Tc . there are 4 regimes for a ferro. using the real-space Green function which you have calculated in the problem set. we can approximate it by: Z= Dn e−β R 1 dd x ( 2 (∇n)2 +λ0 n2 ) (8. where the system is described by a linear σ-model. However.95) Furthermore. a transition takes place and the magnet orders. this occurs at λ0 = 0.4.8.or antiferromagnet: • High temperature. as we did in the previous two sections. we can see that n(x) = 0 (8. the correlation length grows. SPIN SYSTEMS AT FINITE TEMPERATURES 125 When the functional integral is dominated by a non-zero saddle-point value λ0 . we see that correlation functions of the magnetization decay exponentially with distance n(x)n(0) ∼ 1 e−|x|/ξ |x|(d−1)/2 (8. Correlation functions fall oﬀ exponentially with correla√ tion length ξ ∼ λ0 .96) √ where we have deﬁned the correlation length ξ ∼ λ0 . From (8. To summarize. Finally. .94) which is a Gaussian integral. T > Tc . T = Tc . In the saddle-point – or mean-ﬁeld – approximation. at which correlation functions have powerlaw falloﬀ. As the temperature is lowered and λ0 → 0. as we will discuss in chapter 11 – and as you have investigated in the problem set – the saddle-point approximation is often incorrect. and we can expand about the ordered state. The saddle-point approximation would tell us that 1 (8. For temperatures below Tc the magnet is ordered.

126

CHAPTER 8. SPIN SYSTEMS AND MAGNONS • The ordered phase, 0 < T < Tc , where the magnetization (or staggered magnetization) has a non-zero expectation value. This regime is described by the d-dimensional NLσM: Z= Dn e−β

R

1 dd x ( 2 (∇n)2 )

which can be expanded perturbatively about the ordered state. • The ordered state at T = 0 which is described by the d+1-dimensional NLσM in the antiferromagnetic case, Z= Dn e

1 −g

R

dd x dτ

“

1 2v 2

( dn ) dτ

2

1 + 2 (∇n)2

”

**and by the following functional integral in the ferromagnetic case. Z=Z= DΩ e
**

−s R “ ” 2 dd x dτ −i A(Ω)· dΩ + 1 D (∇Ω) dτ 2

(8.98)

8.5

Hydrodynamic Description of Magnetic Systems

In the limit in which magnon-magnon interactions are strong, it is hopeless to try to expand perturbatively about a quadratic action in either the ferromagnetic or antiferromagnetic cases. However, some properties of correlation functions can be deduced from hydrodynamic equations. A ferromagnet satisﬁes hydrodynamic equations very similar to those of a conserved particle density which we discussed in chapter 7. The magnetization, Ω is a conserved quantity, so the deviation from an ordered state, δΩ = Ω − Ω0 satisﬁes a conservation law: ∂ δΩi + ∇ · J i = 0 ∂t and a constitutive relation: J i = −χ−1 (T )∇δΩi + ∇Bi 0 (8.100) (8.99)

(χ0 is the static magnetic susceptibility) from which it follows that the magnetization has diﬀusive correlation functions: χδΩi δΩi (ω, q) = q2 −iω + χ−1 (T )q 2 0 (8.101)

8.6. SPIN CHAINS**

127

In the case of an antiferromagnet, however, the staggered magnetization, n, is not conserved, so it does not diﬀuse. If the system is ordered with n = ˆ, z then the correct hydrodynamic equations for i = 1, 2 are: ∂li = ρs (T )∇ · ǫijk nj ∇nk ∂t ǫijk nj ∇nk = χ−1 (T )∇li ⊥

∂ ∂t

(8.102)

These equations are the equations of motion for n, l which follow from the NLσM with ρs = 1/g and χ⊥ = 1/gv 2 . Rotational invariance dictates that the hydrodynamic equations must hold generally albeit with diﬀerent values of ρs and χ⊥ . These equations can be combined to give: ∂2 ǫijk nj ∇nk = ρs (T )χ−1 (T ) ∇2 ǫijk nj ∇nk ⊥ ∂t2 (8.103)

from which it follows that there is a propagating mode of velocity ρs (T )χ−1 (T ). ⊥ A more reﬁned analysis, which includes higher-order terms which have been neglected above leads to a small damping of this mode which goes as q 2 .

8.6 8.7

Spin chains** Two-dimensional Heisenberg model**

128

CHAPTER 8. SPIN SYSTEMS AND MAGNONS

CHAPTER

9

Symmetries in Many-Body Theory

9.1

Discrete Symmetries

Symmetries constrain the behavior of physical systems and are, therefore, a useful tool in the analysis of a many-body system. In general, a more symmetrical system is more highly constrained and, consequently, more easily solved. The most useful symmetries are continuous symmetries – ie. symmetries which belong to continuous families – which we discuss in the remainder of this chapter. The simplest symmetries are discrete, and we focus on them in this section. We will focus on the archetypal discrete symmetries, parity, P , and time-reversal, T . A discrete symmetry is a transformation of the ﬁelds in a theory, ϕ(x, τ ) → ′ (x, τ ) which leaves the action unchanged. Since the classical equations of ϕ motion are just the stationarity conditions on the action, a discrete symmetry takes one solution of the equations of motion and transforms it into another. At the quantum level, we would like such a transformation to be eﬀected by a unitary operator, U : U † ϕ(x, τ ) U = ϕ′ (x, τ ) (9.1)

Parity is an example of such a symmetry. We will call any transformation of the form ϕa (x, τ ) → Mab ϕb (−x, τ ) (9.2) 129

130

CHAPTER 9. SYMMETRIES IN MANY-BODY THEORY

a parity transformation, but Mab is typically ±δab . Let us consider, as an example, our action for interacting phonons. S= dτ d3 r g 1 1 ρ(∂t ui )2 + µuij uij + λu2 + (∂k uk )4 kk 2 2 4! (9.3)

**The parity transformation, ui (x, τ ) → −ui (−x, τ ) leaves (9.3) invariant. If we deﬁne the unitary operator UP by
**

† UP ui (x, τ )UP = −ui (−x, τ )

(9.4)

(9.5)

**then U has the following eﬀect on the creation and annihilation operators:
**

† UP ak,s UP = −a−k,s k,s −k,s

† UP a† UP = −a†

(9.6)

Hence the vacuum state of a free phonon system, g = 0, which is deﬁned by: ak,s |0 = 0 is invariant under parity: U |0 = |0 (9.8) If we assume that the ground state evolves continuously as g is increased from 0 so that the g = 0 ground state is also invariant under parity, then parity constrains the correlation functions of the interacting theory: 0 |Tτ (ui1 (x1 , τ1 ) . . . uin (xn , τn ))| 0 = (−1)n 0 |Tτ (ui1 (−x1 , τ1 ) . . . uin (−xn , τn ))| 0 (9.9) Note that ui (x, τ ) → −ui (x, τ ) (9.10) is also a symmetry of (9.3), so we can take any combination of this and parity, such as ui (x, τ ) → ui (−x, τ ) (9.11) It doesn’t really matter what we call these various symmetries so long as we realize that there are two independent symmetries. Realistic phonon Lagrangians have cubic terms which are not invariant under (9.10), so usually (9.7)

9.1. DISCRETE SYMMETRIES

131

the parity transformation (9.4) is the only symmetry. The symmetry (9.10), when it is present, leads to the relation 0 |Tτ (ui1 (x1 , τ1 ) . . . uin (xn , τn ))| 0 = (−1)n 0 |Tτ (ui1 (x1 , τ1 ) . . . uin (xn , τn ))| 0 (9.12) which implies that correlation functions of odd numbers of phonon ﬁelds must vanish. A spin, on the other hand, transforms as:

† UP Ω(x, t) UP = Ω(−x, t)

(9.13)

The time-derivative term in the Lagrangian is not invariant under Ω → −Ω, so there is no arbitrariness in our choice of parity transformation. Time-reversal is a symmetry which does not quite ﬁt into this paradigm. Reversing the direction of time, t → −t takes one solution of the equations of motion into another, but it does not necessarily leave the action S = τf τi dτ L(τ ) invariant. Nevertheless, we might expect that there is a unitary operator, UT , which transforms the phonon ﬁeld ui (t) → ui (−t)

−1 UT ui (t) UT = ui (−t)

(9.14)

**In fact, this operator cannot be unitary. To see this, diﬀerentiate both sides of (9.14): −1 (9.15) UT ∂t ui (t) UT = −∂t ui (−t) Act on
**

−1 with UT and UT :

ui (x, t), ρ∂t uj (x′ , t) = i δij δ(x − x′ )

(9.16)

−1 −1 UT ui (x, t), ρ∂t uj (x′ , t) UT = UT i δij δ(x − x′ ) UT ′ −1 ui (x, −t), −ρ∂t uj (x , −t) = i δij δ(x − x′ ) UT UT − ui (x, t), ρ∂t uj (x′ , t) = i δij δ(x − x′ ) − i δij δ(x − x′ ) = i δij δ(x − x′ )

(9.17)

which is a contradiction. The time-reversal operator must actually be an antiunitary operator. An antiunitary operator is a type of antilinear operator. While a linear operator O satisﬁes: O (α|ψ + β|χ ) = αO|ψ + βO|χ (9.18) an antilinear operator satisﬁes O (α|ψ + β|χ ) = α∗ O|ψ + β ∗ O|χ (9.19)

132

CHAPTER 9. SYMMETRIES IN MANY-BODY THEORY

˜ An antiunitary operator is an antilinear operator O which satisﬁes ˜ ˜ Oχ, Oψ = (χ, ψ) (9.20)

where we have used the notation (χ, ψ) to denote the inner product between the states |ψ and |χ . The time-reversal operator, UT , is an anti-unitary operator, which explains how the paradox (9.17) is avoided. While the phonon ﬁeld has the time-reversal property (9.14), a spin must transform under time-reversal so as to leave invariant the time-derivative term in the action:

β 1

s

0

dτ

0

dr −i

dΩ dΩ ·Ω× dr dτ

(9.21)

**Evidently, the correct transformation property is:
**

−1 UT Ω(t) UT = −Ω(−t)

(9.22)

Hence, the ferromagnetic, Ω(x, t) = Ω0 , and antiferromagnetic, Ω(x, t) = (−1)i n0 , ground states are not time-reversal invariant, i.e. they spontaneously break time-reversal invariance, unlike the phonon ground state which does not. The antiferromagnetic state also breaks the discrete symmetry of translation by one lattice spacing. However, the product of T and a translation by one lattice spacing is unbroken.

9.2

Noether’s Theorem: Continuous Symmetries and Conservation Laws

Before looking at continuous symmetries in quantum systems, let us review one of the basic results of classical ﬁeld theory: Noether’s theorem. This theorem relates symmetries of the action to the existence of conserved currents. Suppose we have a classical ﬁeld theory deﬁned by an action and Lagrangian density: S= dtd3 r L(φ, ∂φ, r) (9.23)

where φ is the classical ﬁeld. Consider a transformation φ(r, t) → φ(r, t, λ) , φ(r, t, 0) = φ(r, t) (9.24)

**9.2. NOETHER’S THEOREM: CONTINUOUS SYMMETRIES AND CONSERVATION LAWS 133 and deﬁne the inﬁnitesimal transformation DK = ∂K ∂λ (9.25)
**

λ=0

Then, this transformation is a symmetry of the action if and only if DL = ∂µ Fµ (9.26)

for any φ (i.e. not only for φ satisfying the equations of motion). (Greek indices take the values 0, 1, 2, 3 where 0 = iτ ; relativistic invariance is not implied.) Now, a general expression for DL can be obtained from the chain rule: DL = ∂L Dφ + πµ D (∂µ φ) ∂φ = ∂µ πµ Dφ + πµ D (∂µ φ) = ∂µ (πµ Dφ)

(9.27)

We used the equations of motion to go from the ﬁrst line to the second and the equality of mixed partials, D∂φ = ∂Dφ, to go from the second to the third. Setting these two expressions equal to each other, we have Noether’s theorem: for every transformation which is a symmetry of the action – i.e. DL = ∂µ Fµ – there is a current jµ = (ρ, j), jµ = πµ Dφ − Fµ which is conserved, ∂µ jµ = ∂t ρ + ∇ · j = 0 (9.29) The extension to theories with multiple ﬁelds is straightforward and can be accomodated by decorating the preceding formulas with extra indices. As an example, let’s consider space and time translations: φ(xµ ) → φ(xµ + λ eµ ) where xµ = (t, r) and eµ is an arbitrary 4-vector. Then, Dφ = eα ∂α φ DL = ∂α (eα L) Hence, the conserved current is jβ = eα Tαβ (9.32) (9.31) (9.30) (9.28)

134 where

CHAPTER 9. SYMMETRIES IN MANY-BODY THEORY

Tαβ = πα ∂β φ − δαβ L

(9.33)

This is the stress-energy tensor. T0µ is the 4-current corresponding to timetranslation invariance: T00 is the energy density and T0i is the energy 3current. Tiµ are the 4-currents corresponding to spatial translational invariance: Ti0 are the momentum densities and Tij are the momentum currents. In our theory of an elastic medium, Tαβ is given by: T00 = H T0i = ρ ∂t uj ∂i uj Tij = −2µ ∂i uk ∂j uk − ∂k uk ∂j ui − δij L

(9.34)

Tij is the stress tensor of the elastic medium. Our action for a spin – as well as our actions for ferro- and anti-ferromagnets – is invariant under spin rotations, Ωa → Rab Ωb . In the case of an ferromagnet, this leads to 3 conserved quantities corresponding to spin rotations about the three diﬀerent axes:

i J0 , J i = Ωi , Dǫijk Ωj ∇Ωk

(9.35)

**In the antiferromagnetic case, they are:
**

i Jµ = ǫijk nj ∂µ nk

(9.36)

In general, symmetries are related to unobservable quantities. In the above, the conservation of momentum follows from the unobservability of absolute position; the conservation of energy, from the unobservability of absolute temporal position. Angular momentum is a consequence of the unobservability of absolute direction.

9.3

Ward Identities

In the previous section, we discussed the consequences of continuous symmetries and conservations laws for classical systems. We now turn to the quantum theory, where the existence of continuous symmetries and their associated conservation laws leads to important constraints on correlation functions. These constraints are called Ward identities. The Ward identity relates the divergence of a time-ordered correlation function of a conserved current, jµ , with some other ﬁelds, ϕi to the variations of those ﬁeld under the symmetry generated by j0 . The variation of ϕ(x, t) under such a

. . . ϕn (xn . τn ) + . we consider a correlation function of jµ with the ϕi ’s: Tτ (jµ (x. . τ1 ) jµ (x. τ ) ϕ1 (x1 . τn )) + . . ϕ1 (x1 . . . τn ) − δ(τ1 − τ )θ(τ1 − τ2 ) .. τn ) + θ(τ1 − τ )θ(τ1 − τ2 ) . τ1 )] . ϕ(x. ϕn (xn . τ ) = 0. then we can ordinarily conclude that ∂µ jµ (x. WARD IDENTITIES symmetry operation is: Dϕ(x. it is possible for this equation to be violated in the quantum theory when there is a cutoﬀ (the conservation law can be violated when the conserved quantity ﬂows to wavevectors beyond the cutoﬀ) and this violation can remain ﬁnite even as the cutoﬀ is taken to inﬁnity. However. (9.37) To derive the Ward identities. τn )) + . ϕn (xn . . τn )) = δ(x − x1 )δ(τ − τ1 ) Tτ (Dϕ1 (x1 . . the derivative operator can act on a θ-function which has τ in its argument or it can act on jµ (x. . ϕn (xn . . t) = dd x′ J 0 (x′ . ϕn (xn . (9. . θ(τn−1 − τn ) jµ (x. τn ) + . .40) As an example of the Ward identity. . . τ ) ϕ1 (x1 . τ1 ) jµ (x. t) 135 (9. . . for which the spin currents are: i Jµ = ǫijk nj ∂µ nk (9. . ϕn (xn . τ ) ϕ1 (x1 . ϕn (xn . τ2 ) .41) . τn )) + δ(x − x1 )δ(τ − τ1 ) Tτ (ϕ1 (x1 . .. τ1 ) .9. . . τn )) = δ(τ − τ1 )θ(τ1 − τ2 ) . . τ ) ϕ1 (x1 . . τ1 ) Dϕ2 (x2 . Such a symmetry is called anomalous. consider an antiferromagnet. θ(τn−1 − τn ) ϕ1 (x1 .3. If the symmetry is conserved in the classical ﬁeld theory. ϕn (xn . . τ1 ) . . . . τn )) + δ(x − x2 )δ(τ − τ2 ) Tτ (ϕ1 (x1 . . θ(τn−1 − τn ) [jµ (x. . τ1 ) . = δ(τ − τ1 )θ(τ1 − τ2 ) . ϕn (xn .. τn )) = θ(τ − τ1 )θ(τ1 − τ2 ) . . τ1 ) . τ ) ϕ1 (x1 . . θ(τn−1 − τn ) jµ (x. τ1 ) . ϕn (xn . τn ) + . = δ(x − x1 )δ(τ − τ1 ) Tτ (Dϕ1 (x1 . τ1 ) . . τ1 ) Dϕ2 (x2 . τ2 ) . τ ) . . τ ) . then the right-hand-side contains only terms resulting from the derivative acting on the θ-function to give a δ-function: ∂µ Tτ (jµ (x. τ ).39) The ﬁnal equality is the Ward identity: ∂µ Tτ (jµ (x. ϕn (xn . ϕn (xn . (9.. .38) If we diﬀererentiate this with respect to xµ . If the symmetry is not anomalous. . θ(τn−1 − τn ) ϕ1 (x1 . . t). . . τ1 ) . τ ). . .

43) with J0 = Ωx and ∂µ → pµ = (iωn . 0) = 1 iωn (9. Ωx (iωn . 0)Ωy (iωn .1b. SYMMETRIES IN MANY-BODY THEORY (a) (b) Figure 9.45) We found the same result earlier for a linearized theory in which magnonmagnon interactions. 0) (9. 0)Ωy (iωn . τ1 ) nl (x2 . consider a ferromagnet which is ordered along the ˆ axis. Then the Ward identity tells us that: ∂µ Tτ (ǫijs nj (x. τ ) nk (x1 .1: Diagrams contributing to the (a) left-hand-side and (b) righthand side of the Ward identity (9.1a while the right-hand-side is given by diagrams such as those of 9.136 CHAPTER 9.44) or. The Ward identity shows that this result is exact. 0). Hence. As another example. . τ2 )) + δ(x − x2 )δ(τ − τ2 )ǫilr Tτ (nl (x1 . τ1 ) nl (x2 . so the following z correlation function is of the form for which the Ward identity is applicable: Ωx (iωn . 0) = Ωz = 1 (9. τ1 ) nr (x2 . τ2 )) = δ(x − x1 )δ(τ − τ1 )ǫikm Tτ (nm (x1 . the Ward identity tells us that: iωn Ωx (iωn . Ωx generates rotations about the x-axis. since the correlation function on the left-hand-side is given by diagrams such as those in ﬁgure 9. τ )∂µ ns (x. 0)Ωy (iωn .42).42) ( This is a non-trivial constraint when imposed order-by-order in perturbation theory. τ2 ))9. Ωz = 1:.

τ ) U . τ ) = 0 (9.It is possible that there is not an invariant ground state. but rather a multiplet of degenerate symmetry-related ground states. the ground state is invariant under the symmetries of the Lagrangian. The ferromagnetic and antiferromagnetic ground states are two more examples of spontaneous symmetry breaking: the Heisenberg model and the ﬁeld theories derived from it. either as a result of a random ﬂuctuation or some weak external perturbation.47) is invariant under the Z2 symmetry ϕ → −ϕ which is broken for T < Tc (i. the tree-level result is unchanged by the inclusion of magnon-magnon interactions. this is not the only possibility. S=s and S= 1 g dd x dτ 1 2v 2 dn dτ dd x dτ −i A Ω · dΩ 1 + D ∇Ω dτ 2 2 2 (9. Our phonon Lagrangian. As the temperature is lowered through Tc .9. τ ) is a ﬁeld which is not invariant under the symmetry. as we will see in the next section. τ ) = U † φ(x.48) + 1 (∇n)2 2 (9.46) where φ(x.4 Spontaneous Symmetry-Breaking and Goldstone’s Theorem Often. For instance. the statement of spontaneous symmetrybreaking is φ(x. ϕ = ± 6r/u and unbroken for T > Tc . as we have already seen. SPONTANEOUS SYMMETRY-BREAKING AND GOLDSTONE’S THEOREM 137 i. for instance. our ﬁeld theory for the Ising model. Such a ﬁeld is called an order parameter. The divergence of this correlation function at low frequency is an example of Goldstone’s theorem in action. in which case we say that the symmetry is spontaneously broken.e.4. ϕ = 0. the system spontaneously chooses one of the two symmetry-related conﬁgurations ϕ = ± 6r/u. In terms of correlation functions. Z=N Dϕ e− R 1 1 dd x ( 1 K(∇ϕ)2 + 2 rϕ2 + 4! uϕ4 ) 2 (9. and the ground state is as well. r < 0).49) . φ(x. is invariant under parity and timereversal.e. 9. However.

the reason for the existence of Goldstone bosons is that by applying the generator of the broken symmetry. The signal of spontaneous symmetry breakdown is the non-invariant expectation value Ω = 0 or n = 0. and H is a discrete subgroup. ni → Rij nj . since it is continuous. For instance. . These expectation values also break the discrete T symetry. if we take a magnet aligned along the ˆ axis and rotate all of the spins away from the ˆ axis then z z we obtain another state of the same energy. A ferromagnetic in a magnetic ﬁeld does not have an SU (2) or T -invariant Lagrangian. If H = G. so dimG − dimH = 0. the ui ’s. If the spins instead vary slowly in space with wavevector q. i. leads to gapless excitations – magnons and phonons. This is a general feature of ﬁeld theories with broken symmetries: broken continuous symmetries lead to gapless excitations – called Goldstone modes or Goldstone bosons – but broken discrete symmetries do not.138 CHAPTER 9. then the energy of the resulting state vanishes as q → 0. and there are no gapless excitations in the symmetry-broken phase. The phonon Lagrangian is actually another example: the translational symmetry of the continuum is broken to the discrete translational symmetry of the lattice. the symmetry is completely unbroken. dimH = 0. A ferromagnet in zero ﬁeld has only one Goldstone mode while dimG − dimH = 2. The µB · Ω term is called a symmetry-breaking term. dimG = 6. T > Tc . the broken symmetry. SYMMETRIES IN MANY-BODY THEORY are invariant under SU (2) spin-rotational symmetry. In the other two examples. the Ising model. while G is the group of translations and rotations. At high-temperature. Ωi → Rij Ωj . the broken symmetry is discrete. The number of Goldstone modes is at most dimG − dimH if G is the symmetry group of the theory and H is subgroup of G which is left unbroken. In the case of an antiferromagnet. G = SU (2) and H = U (1) – the group of rotations about staggered magnetization axis – so there are dimG − dimH = 2 gapless modes. but the ground states are not invariant since the magnetization or staggered magnetization chooses a particular direction. the lattice melts and the resulting ﬂuid state has the full translational symmetry. so its ferromagnetic ground state is an example of an explicitly broken symmetry rather than a spontaneously broken one.e. We will now give a precise statement and proof of Goldstone’s theorem. Physically. the symmetry is restored and Ω = 0 or n = 0. magnets and ionic lattices. we obtain another state with the same energy as the ground state. then there no Goldstone bosons. G = H = U (1). At high temperature (when our continuum elastic theory is no longer valid). A ferromagnet in a ﬁnite ﬁeld has no Goldstone modes. These states are the Goldstone modes. A crytal has only three Goldstone modes. the group of rotations about the direction of the ﬁeld. In the ﬁrst example.

These gapless excitations are the Goldstone modes. implies that ρ(ω. t) = dd x′ J 0 (x′ . ϕ(x. k) = 0 for ω < ∆. then 0 |Dϕ(k = 0. 0) n n |ϕ(0. 0)| 0 δ(ω − En ) δ(k − Pn ) (9. then there must be gapless excitations. t) 0 = 0 Fourier transforming and taking the k → 0 limit. t) be some ﬁeld in the theory. if 0|Dϕ(k = 0.9. we have: 0 ∂µ J µ (x′ .58) (9. t) (9. The proof proceeds by constructing a spectral representation for 0|J 0 (x′ . k) ≥ 0. J 0 . t). t′ )ϕ(x.55) (9. ω) = 0 or ω=0 (9.50) under an inﬁnitesimal symmetry operation corresponding to the conserved quantity J 0 .57) (9. t)|0 : 0 J 0 (x′ . k) = n dd k dω e−i(k·(x−x )−ω(t−t )) ρ(ω. t)| 0 = 0 (9. t)|0 = 0. −ω) ϕ(k = 0. Let ϕ(x.52): ω ρ(k = 0. ∆. t′ )ϕ(x.51) Conversely. ϕ(x. ∆. t) 0 = where ρ(ω. so that ∂µ J µ = 0. ω) = 0 which implies that ρ(k = 0. t) transforms as Dϕ(x. ω) 0 = 0 (9. ρ(ω.52). and its associated current. Goldstone’s Theorem: If there is an energy gap.4. SPONTANEOUS SYMMETRY-BREAKING AND GOLDSTONE’S THEOREM 139 Suppose we have a conserved quantity. Then. t′ )ϕ(x. Applying the conservation law to the correlation function of (9. The existence of an energy gap. J i . k) ′ ′ (9.56) . the following theorem holds. we have: ω 0 J 0 (k = 0.54) The left-hand-side can be rewritten using our spectral representation (9.52) 0 J 0 (0.53) By unitarity.

Hence. only the former has a δ(ω) contribution. then the symmetries generated by Ωx and Ωy are broken. Note that this proof depended on unitarity and translational invariance.τ ) (9. J0 = ϕ. t) 0 = 0 Similarly. The order parameters of broken translational invariance in a crystal are: ρG = eiG·u(x. In the case of an antiferromagnet. ρ(k = 0. t)| 0 = 0 (9. SYMMETRIES IN MANY-BODY THEORY Hence. consequently. ϕ. the latter vanishes. the existence of an energy gap. 0 J 0 (x′ .62) is an order parameter: it signals the development of an ordered state.60) (9. t)| 0 = 0 (9. for ω = 0. 0 |Dϕ(k = 0. .63) where G is a reciprocal lattice vector of the crystal and u is the phonon ﬁeld. ρ(k = 0. k = 0) does not vanish when ω = 0. ρ(ω. have δ(ω) contributions. t)| 0 = σ. The order parameter of a ferromagnet is Ω while the order parameter of an antiferromagnet is n. 0 ϕ(k = 0. It is not necessary for the order parameter of a theory to be the fundamental ﬁeld of the theory. Then 0 |Dϕ(k = 0. in particular. This is the reason why a ferromagnet has only 1 Goldstone mode. ω) = 0 for all ω. If we z look at the spectral functions for Ωx Ωy and Ωy Ωy . ω) = 0 for ω < ∆ and. is itself a conserved quantity. then Dϕ vanishes identically and the associated Goldstone boson doesn’t exist. k = 0) coming from the Goldstone modes of the form ρ(ω. t′ )ϕ(k = 0. t′ ) 0 = 0 and.61) (9.140 CHAPTER 9. If the order parameter. ∆ implies that ρ(k = 0. However. which generates a spontaneously broken symmetry. If the ferromagnet is ordered along the ˆ axis.59) When the symmetry is broken. the spectral functions for both Lx ny and Ly ny . Therefore. ω) = 0 for all ω > 0. instead. there is a contribution to ρ(ω. t)J 0 (x′ . k = 0) = σδ(ω). A ﬁeld ϕ which satisﬁes 0 |Dϕ(k = 0.

and antiferromagnets have the z same description at ﬁnite temperature.4. the energy cost of a ﬂuctuation is independent of the size of the ﬂuctuation. It is simply 4J. this energy cost implies that such ﬂuctuations occur with very low probability Pﬂuc ∼ e−(const. At low temperature. we encountered hints that neither ferromagnets nor antiferromagnets could order at ﬁnite T in d ≤ 2 and that antiferromagnets could not even order at zero temperature in d = 1. As a result of these ﬂuctuations.)βL d−1 (9. a ﬂuctuation in which a large fraction of the system consists of reversed spins can occur with probability ∼ exp(−4βJ). (Recall that ferro.3 Absence of broken symmetry in low dimensions** Discrete symmetry** Continuous symmetry: the general strategy** The Mermin-Wagner-Coleman Theorem In chapter 9.5. In d = 1. however. so large regions of reversed spins are energetically disfavored. namely the Ising model. What is the energy cost to create a size Ld region of down spins in d-dimensions? It is simply Eﬂuct ∼ Ld−1 (9. Let us now consider a continuous symmetry at ﬁnite temperature.9. This is clearly true for any discrete symmetry.) The energy cost for a size Ld region of reversed magnetization is less than in the case of a discrete symmetry .64) i.1 9. For d > 1. Let us now discuss the diﬃculties involved in breaking a symmetry in a low dimensional system. it is impossible for the system to order at any ﬁnite temperature.1 9.e. the energy cost of a domain of reversed spins is proportional to the surface area of the domain wall since that is the only place where unlike spins are neighbors. Suppose the system is ordered with all of the spins pointing up.65) and hence the orderd phase is stable. For the sake of concreteness.4.5.5 9. this grows with L. Hence. Consider the simplest example.2 9. let us consider a d-dimensional magnet in an ordered phase in which the magnetization (or staggered magnetization) is aligned along the ˆ axis. ABSENCE OF BROKEN SYMMETRY IN LOW DIMENSIONS**41 1 9.2 Order parameters** Conserved versus nonconserved order parameters** 9.5.5.

ϕ(x. Pﬂuc ∼ e−(const. Rather.66) For a ﬂuctuation of linear size L. The energy cost will be the gradient energy. in d = 2.67) Hence. The conservation law ki Fi = 0 implies that Fi (k) = σ ki δ(k2 ) + ǫij kj ρ(k) (9.72) .)βL (9. the spins can interpolate continuously between one ground state and another. This conﬁrms our suspicion that magnets can’t order for T > 0 in d = 2. The dimension below which symmetry-breaking is impossible is called the lower critical dimension. We will show that Dϕ = 0 (9. dd x 1 (∇n)2 2 (9. the lower critical dimension is 2 for continuous symmetries and 1 for discrete dymmetries.71) This decomposition is clearly special to two dimensions. SYMMETRIES IN MANY-BODY THEORY since the magnetization at the domain wall need not jump from one degenerate ground state to another. Let us consider our ﬁnite-temperature order parameter. Deﬁne: F (k) = Fi (k) = Fil (k) = d2 xeik·x ϕ(x)ϕ(0) d2 xeik·x ji (x)ϕ(0) d2 xeik·x ji (x)jl (0) (9. These qualitative considerations can be made rigorous.70) where i = 0. but not for d ≤ 2. we have: Dϕ = σ (9. ω = 0). Substituting this decomposition into the deﬁnition of Dϕ. For T > 0. so dd x Ld 1 (∇n)2 2 ∼ Ld−2 d−2 (9. 1.68) and we conclude that a continuous symmetry can be broken for T > 0 in d > 2.69) i.e.142 CHAPTER 9. the symmetry is necessarily unbroken for d = 2. (∇n)2 ∼ 1/L2 .

Thus. The order parameter for a ferromagnet. The proof works by essentially taking the spatial points very far apart in the correlation functions on the left-hand-side.75) If we take an h(k) which is even in x1 – and. Thus far. on the other hand. ωn = 0) is a conserved quantity: the components of Ω(q = 0. This can be said somewhat diﬀerently as follows. In the presence of long-range forces. Consequently. d2 x h(x) (aj0 (x) + bϕ(x)) |0 (9.73) has positive norm. |k0 |)|2 (9. in d > 1. our discussion has focussed on thermal ﬂuctuations. can order in any number of dimensions. σ = 0 and the symmetry is unbroken. The reason that the above arguments about ﬂuctuations do not apply to the ferromagnet is that it has a ﬂuctuationless ground state. i. A ferromagnet. ABSENCE OF BROKEN SYMMETRY IN LOW DIMENSIONS**43 1 By unitarity. therefore. while it is true that there is very little energy cost for a state with reversed spins. such a state will never be . the left-hand-side need not vanish.9. and spontaneous symmetry-breaking is possible. the answer is clearly yes. but not in d = 1. The quantum theory of an d-dimensional antiferromagnet at T = 0 is the same as the classical statistical theory of a magnet in d + 1-dimensions. The exact ground state of a Heisenberg ferromagnet is a state in which all of the spins are aligned.76) We can make the left-hand-side vanish by making |h(k)|2 sharply peaked at very high k where F (k) and F00 (k) must vanish. we see that d2 k |h(k)|2 F (k) ≥ 0 d2 k |h(k)|2 F00 (k) ≥ 0 positivity of the norm also implies that d k |h(k)| F (k) 2 2 (9. even in k1 – then the right-hand-side will be: σ d2 k |h(k)|2 k0 δ(k2 ) = σ dk0 |h(k0 .5. Can quantum ﬂuctuations prevent order at T = 0? In the case of an antiferromagnet.74) d k |h(k)| F00 (k) 2 2 ≥ d k |h(k)| F0 (k) 2 2 2 (9. ωn = 0) are the components of the total spin along the diﬀerent axes. From the special cases a = 0 and b = 0.e. we conclude that a quantum antiferromagnet can order at T = 0 in d + 1 > 2. Ω(q = 0. Hence.

it occurs when d+z = 2 for a continuous symmetry or d + z = 1 for a discrete symmetry. then there can always be order at T = 0 in any d. the dynamics of the system can lead to ﬂuctuations which destroy the order. we must average over all of the states in the canonical ensemble. so the ﬂuctuations can destroy the ordered state of the ferromagnet.5.5.6 9. If it is not. At ﬁnite temperature.1 Proof of existence of order** Infrared bounds** . More generally. In the case of an antiferromagnet. 9. SYMMETRIES IN MANY-BODY THEORY reached at T = 0 since the dynamics conserves the the total spin.5 9. In the case of antiferomagnets – or phonons – this occurs in d = 1. if the order parameter is a conserved quantity. hence. n is not conserved.7 Absence of magnetic order** Absence of crystalline order** Generalizations** Lack of order in the ground state** 9. then quantum ﬂuctuations can destroy the order at T = 0.5.4 9.144 CHAPTER 9.6 9. To summarize.5.6. on the other hand.

Part IV Critical Fluctuations and Phase Transitions 145 .

.

2) Dφ e−S[φ] (10. engineering. 147 . applications).g. . independent of many of the details of the systems.1 Low-Energy Eﬀective Field Theories In our earlier discussions. Suppose we have a system deﬁned by the following functional integral: Z= with correlation functions: φ(p1 ) . universal properties of the system are determined by these correlation functions in the pi → 0 limit.e.CHAPTER 10 The Renormalization Group and Eﬀective Field Theories 10. i. φ(pn ) e−s[φ] (10. low ω. Why? The principal reason is that these properties are universal – i.1) The long-wavelength. . in the limit that the pi ’s are smaller than any other scales in the problem. e. . Sometimes universal properties are the most striking and interesting aspect of a physical system.e. . but not always (certainly not for many practical. we focussed on the low (compared to some cutoﬀ Λ) T . q properties of the systems at which we looked. We like universal properties because we can understand them using eﬀective ﬁeld theories. φ(pn ) = Dφ φ(p1 ) .

Seﬀ . We would like an eﬀective action. we can make some statements about it on general grounds. On the other hand.148 CHAPTER 10.7) In general.φH ] (10. At a formal level.10) (10. . (10.6) (10. We have to use any and all tricks available to us (sometimes we can ﬁnd a small parameter which enables us to get Zeﬀ approximately and often we simply have to guess. even without deriving Seﬀ . at least.8) Seﬀ [φL ] has cutoﬀ Λ′ Occasionally.φH ] (10. we can make the division: φL (p) = φ(p) θ(Λ′ − |p|) φH (p) = φ(p) θ(|p| − Λ′ ) so that φ(p) = φL (p) + φH (p) where Λ′ is some scale such that we’re interested in |p| < Λ′ . Then Z= DφL DφH e−S[φL . so that Zeﬀ = Dφ e−Seﬀ [φ] (10. simpler than Z. is Zeﬀ = where e−Seﬀ [φL ] = DφH e−S[φL .9) DφL e−Seﬀ [φL ] (10.4) The eﬀective ﬁeld theory for long-wavelength correlation functions. φH ] = SL [φL ] + SH [φH ] + λSint [φL . we have no such luck. correlation functions of φL . and we have to work much harder to derive Seﬀ . The reason that this a worthwhile program to pursue is that Zeﬀ is often simple or. i. . this is not a comletely straightforward program because no one tells us how to derive Zeﬀ .3) only contains the information necessary to compute the long-wavelength correlation functions. φH ] with λ small so that we can compute Seﬀ perturbatively in λ: Seﬀ = SL [φL ] + λS1 [φL ] + λ2 S2 [φL ] + .e.5) (10. we will be in the fortunate situation in which S[φL . However. . THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES Z contains a great deal of ‘extraneous’ information about correlation functions for large pi .

10. Typically.2. dd x dτ (∂µ ni )2 dd x dτ ni ∂µ ni nj ∂µ nj dd x dτ ni ∂µ ni nj ∂µ nj ni ni (10. . S[φ] itself may have had all kinds of structure. If we choose Ofree = O4 then [ui ] = 0. but this doesn’t interest us now.14) We now pick one term in the action – call it Ofree even though it need not be quadratic – and use this term to assign dimensions to the various ﬁelds in the theory by requiring Ofree to be dimensionless. We expand SΛ [φ] as SΛ [φ] = gi Oi (10. RENORMALIZATION GROUP FLOWS 149 10. if we choose Ofree = O1 in our phonon theory.12) 1 2 1 O4 = 4! O3 = while the NLσM for an antiferromagnet can be written as: S = O0 + gO1 + g2 O2 + .11) i where the gi ’s are ‘coupling constants’ and the Oi are local operators. we choose the term which we believe to be most . our phonon Lagrangian can be written as: S = ρO1 + µO2 + λO3 + gO4 with O1 = O2 = 1 2 dτ d3 x(∂t ui )2 dτ d3 x uij uij dτ d3 x u2 kk dτ d3 x (∂k uk )4 (10.13) (10. we’re only interested in SΛ [φ]. .2 Renormalization Group Flows Let’s suppose that we have somehow derived Seﬀ with cutoﬀ Λ. with O0 = O1 = O2 = . then [ui ] = 1. For instance. For instance. Let’s call it SΛ [φ]. .15) (10. .

If δi = 0. We now rescale all momenta and ﬁelds by the cutoﬀ Λ.150 CHAPTER 10.17) Ordinarily. this term becomes less important at low energies. δfree = 0 by construction. At lower and lower energies. If we choose the ‘wrong’ one (i. the dimensionless couplings λi will be O(1). If Seﬀ is simple. • The next step is to obtain (by hook or by crook) SbΛ e−SbΛ [φL ] = • We now rescale q → qb DφH e−SΛ [φL . SΛ [φ] and a choice of Ofree . it becomes a better and better approximation to simply drop the irrelevant operators. Oi ∼ (ω/Λ)δi . • We divide φ(p) = φL (p) θ(bΛ′ − |p|) + φH (p) θ(|p| − bΛ′ ) where b < 1.16) gi Λδi Oi = i λi Oi (10. ω ≪ Λ. q → qΛ φ → φΛδφ so that the momenta and ﬁelds are now dimensionless. and the action SΛ [φ] = i λi ω Λ δi .18) If δi > 0. Let’s formalize this by putting together the notion of a low-energy eﬀective ﬁeld theory with the above considerations about scaling: • We have an eﬀective action. it is only because there might be a ﬁnite number of relevant operators.e. Then SΛ [φ] = i (10. it grows in importance at low energies. it remains constant as ω → 0.φH ] . the term is called marginal. Such a term is called irrelevant. On dimensional grounds. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES ‘important’. an inconvenient one) then we will ﬁnd out in the next step. at energy scale ω. If δi < 0. Let’s call δφ the dimension of the ﬁeld φ and δi the dimension of the operator Oi . (10. the term is relevant.

For δi > 0. If we can neglect the . i. These equations are called Renormalization Group (RG) equations or ﬂow equations.2. Then we can deﬁne ﬂow equations: dg = −δi gi + . gi grows as ℓ increases. the correlation length is given by (ξ)SΛ = • If then b SΛ [φ] = 1 (ξ)S b Λ b 0 gi Oi SΛ [φ] = i gi (b)Oi i 0 where gi (1) = gi . if dg = g2 dℓ Then g(ℓ) = so g(ℓ) grows as ℓ grows. . one must allow for arbitrary z.. .20) . it can dominate the ﬁrst term. In fact.e. i. gi decreases. In general. dℓ which describe the evolution of SΛ under an inﬁnitesimal transformation. δi = 0. . . need not be small.e. z = z(gi ). so we do not need to worry about g. In the case of a marginal operator. ζ and z will depend on the couplings gi . so z is ﬁxed to z = 1. the . . Oi is more important at low energies (ℓ → ∞. the . In general. ζ = ζ(gi ). b • The physics of the system can be obtained from SΛ [φ] by a rescaling. Let b = e−ℓ . b however.10. In equilibrium classical statistical mechanics. The rescaling yields SΛ [φ] which also has cutoﬀ Λ. in the theories which have examined thus far ω and q are on the same footing. g0 1 − g0 (ℓ − ℓ0 ) (10. there are no frequencies. .19) (10. Oi is less important at low energies. . For instance. is the whole story. RENORMALIZATION GROUP FLOWS ω → ωbz φ → φbζ 151 where ζ and z are chosen to preserve Ofree . Of course. . For example. then for δi < 0.

but irrelevant at another ∗ . What this really means is that Oi has this property at the ﬁxed point S ∗ = Ofree . SΛ = SΛ . Hence. If yi = 0. Oi = Oij Oi j.21) b ∗ then we call gk = gk a ﬁxed point. At a ﬁxed point. At a ﬁxed point. ui is irrelevant at this ﬁxed point. It is possible for a coupling ∗ constant.24) (10. these systems have the same asymptotic long-wavelength physics.25) ˜ are called eigenoperators. S2 . so the physics is the same at all scales. SΛ . ﬁxed point. marginal. to be relevant at one ﬁxed point.152 CHAPTER 10. we can linearize the RG equations: d ∗ ∗ (gi − gi ) = Aij gj − gj dℓ This can be diagonalized to give: dui = yi ui dℓ (10. gi = gi . S1 . g. The corresponding operators. we characterized Oi as relevant. If this happens. as shown in ﬁgure 10. If yi < 0. or irrelevant according to whether δi < 0. SΛ [φ] = i ˜ ui Oi (10. The notion of universality is encapsulated by the observation that diﬀer′ ′′ ent physical systems with very diﬀerent ‘microscopic’ actions SΛ . δi = 0. Earlier. we say that ui and Oi are relevant at this ﬁxed point. SΛ .22) i.1. dgi dℓ =0 ∗ gk =gk ′ (10. If yi > 0.23) ∗ ˜ where ui = Oij (gj − gj ). ξ= 1 ξ b (10.3 Fixed Points ∗ If. or δi > 0. ξ = ∞ – the low-energy physics is gapless – or ξ = 0 – there is no low-energy physics. SΛ can ∗ all ﬂow into the same ﬁxed point action. for some values of the couplings.e.e. i. the same universal behavior. we say that ui is marginal. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES 10.

both are stable for d > 2. Our phonon theory.10. to ferromagnetic and antiferromagnetic phases. For T > 0. Theories will generically ﬂow into such a ﬁxed point. Stable ﬁxed points represent phases of matter.28) are also stable ﬁxed points corresponding. we have already looked at a number of stable ﬁxed points. Similarly.27) (10.4. For T > 0. This stable ﬁxed point corresponds to a stable phase of matter: a crystal. In this course. 10. while the antiferromagnet is stable at T = 0 for d > 1. it is stable for d > 2.4 Phases of Matter and Critical Phenomena If yi < 0 for all i at a given ﬁxed point.26) is a stable ﬁxed point (you can check that g is an irrelevant coupling) at T = 0 for d > 1. The ferromagnet is always stable at T = 0. Our theories of non-interacting magnons S=s dd x dτ ∂m− 1 1 m+ + D ∇m+ · ∇m− 2 ∂τ 2 S= dd x dτ (∂µ ni )2 (10. XY magnets and superﬂuid 4 He S= dd x dτ (∂µ θ)2 (10. then we call this ﬁxed point an attractive or stable ﬁxed point.29) . repectively.1: The coupling g1 is relevant at the ﬁxed point on the left but irrelevant at the ﬁxed point on the right. S0 = dtd3 xL = 1 2 dtd3 r ρ(∂t ui )2 − 2µuij uij − λu2 kk (10. PHASES OF MATTER AND CRITICAL PHENOMENA 153 * g * 2 g 1 Figure 10.

The unstable directions correspond to the parameters which must be tuned to reach the critical point. Similarly. The low q. the high-temperature disordered states of magnets are stable phases with gaps.e. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES are phases of matter at T = 0 for d > 1. It makes sense to do perturbation theory in an irrelevant coupling because this perturbation theory gets better at low q. ξ = ∞ which are a consequence of spontaneous symmetry breaking. Essentially. the eﬀective expansion parameter grows at low q. In such a case.154 CHAPTER 10.31) has two stable phases – ordered and disordered phases – corresponding to the ﬁxed points r → ±∞. the expansion parameter for perturbation theory is the dimensionless combination gω −yg for some correlation function at characteristic frequency ω. At both of these ﬁxed points. If some of the yi > 0 then the ﬁxed point is repulsive or unstable. ϕ4 theory dd x 1 1 1 K(∇ϕ)2 + rϕ2 + uϕ4 2 2 4! (10. The 4 He action with µ < µc (in the saddle-point approximation. There are also stable phases without gapless modes – i.30) Similarly.e. ξ = 0. S= dτ dd x ψ ∗ ∂τ ψ + |∇ψ|2 + V |ψ|2 − µ 2V 2 (10. ω. they are stable for d ≥ 2 (d = 2 is a special case). Similarly. The relevant couplings must be tuned to zero in order for the theory to ﬂow into an unstable ﬁxed point. Hence. Unstable ﬁxed points represent (multi-)critical points separating these phases. For T > 0. ω. in fact controlled by some other ﬁxed point. However. it does not make sense to perturb in a relevant coupling such as u in ϕ4 for d < 4. This critical point separates two stable phases: the superﬂuid . perturbation theory in the coupling u in ϕ4 is sensible for d > 4 (the Ginzburg criterion) as you showed in the problem set. The stable phases described above are all characterized by gapless modes – i. The corresponding ﬁxed point is at µ = µ∗ . µc = 0) describes an empty system with a gap µc − µ to all excitations. ω physics is. V = V ∗ . Superﬂuid 4 He S= dτ dd x ψ ∗ ∂τ ψ + |∇ψ|2 + V |ψ|2 − µ 2V 2 (10. with ξ = 0. our perturbative calculations of correlation functions in the phonon and magnon theories were sensible calculations.32) has a critical point at µ = µc .

and the empty system. (gi − gi ) b−λi b (10. Suppose C(pi . Suppose that the couplings are all close ∗ ∗ to their ﬁxed point values. 10.5. gi − gi ) = b−nζ C ∗ pi .5 10. in general. Similarly. gi − gi ) and suppose that the linearized ﬂow equations read: d ∗ ∗ (gi − gi ) = λi (gi − gi ) dℓ Then we can perform an RG transformation. in the b → 0 limit ∗ C (pi .35) .10. INFINITE NUMBER OF DEGREES OF FREEDOM AND THE NONANALYTICITY OF THE FREE ENERGY** 155 * S * C 1 * S 2 Figure 10.2 depicts the ﬂow diagram for a critical point and two stable ﬁxed points which it separates. we can pass from the ordered state through the critical point and into the disordered state. By tuning the relevant coupling. gi ≈ gi . There is one relevant direction at this ﬁxed point. S2 which it separates.2: The ﬂow diagram of a critical point C and two stable ﬁxed points S1 .6 Scaling Equations Let us consider the the implications of this framework for physical quantities. depend on some momenta pi and on the coupling constants of the system.0 b (10. By tuning this relevant direction. so we will write C as C(pi . It will.1 Inﬁnite number of degrees of freedom and the nonanalyticity of the free energy** Yang-Lee theory** 10.33) pi ∗ .5. gi − gi ) = b−nζ C ∗ (10. Figure 10. Then. we can pass from one phase through the critical point to the other phase.34) Suppose that we are in the vicinity of a stable ﬁxed point. the Ising model has a ﬁxed point with one relevant direction which we discuss in a later section. so that all of the λi < 0. according to which: ∗ C (pi . gi ) is some physical quantity such as a correlation function of n ﬁelds ϕ.

gi − gi ) ∼ p−2ζ Λ−2δφ +2ζ ∗ ∗ (10.36) A similar result follows if we are in the vicinity of an unstable ﬁxed point. βi = 0 and η is a constant. The β functions are the right-hand-sides of the ﬂow-equations for the couplings: dgi βi (gj ) = (10. this correlation function is equal to its value after an RG transformation: Cn (pi . gi (ℓ)) = 0 (10. gi − gi ) → p−2ζ C (1. This scaling relation may seem to contradict simple dimensional analysis. Consider a correlation function Cn (pi . so diﬀerentiating both sides with respect to ℓ yields the RG equation: 1 ∂ ∂ + n δϕ − η (gj ) + βi (gj ) ∂ℓ 2 ∂gi where 2δϕ − η (gj ) = Cn (pi eℓ . gi − gi ) ∼ p−2δφ .38) The left-hand-side is independent of ℓ. The missing powers of p are made up by the dependence on the cutoﬀ: ∗ C (p. there is no contradiction. .156 CHAPTER 10.43) In other words.42) so the RG equation reads: ∂ 1 − n δϕ − η ∂ℓ 2 Cn (pi e−ℓ ) = 0 (10.41) dℓ At a ﬁxed point. 0) ∗ (10. Up to a rescaling. ∗ 2δϕ − η = 2ζ (gi ) (10. the correlation function is a power-law in pi with exponent nδϕ − nη/2. the β-functions vanish. we can take b = p and: ∗ C (p.37) These observations can be reformulated as follows. gi ) of n ϕ ﬁelds. gi ) = eℓ nζ(gi ) Cn (pi eℓ . THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES If.40) dℓ δϕ is the naive scaling dimension of ϕ. In fact. which would ∗ predict C (p. gi (ℓ)) (10. η is called the anomalous dimension of ϕ. for instance. we are interested in the two-point correlation function at low p.39) d ℓ ζ(gi ) e (10. but we have set all of the relevant couplings equal to zero.

u − u∗ . that we are near a ﬁxed point with one relevant direction.7 10. temperature is one of the couplings in the theory.48) 10. the least irrelevant of the irrelevant couplings) so that we are interested in the g dependence of the correlation function. (u − u∗ ) b−λu .45) or ∗ C (pi . or. gi − gi ) → ∗ 1 (u − ∗ u∗ )nζ /λu C pi (u − u∗ )−1/λu . (gi − gi ) b−λi (10. gi − gi ) = b−nζ C pi /b.7. 0 can be calculated and does not have any interesting structure. 1. g. If we are interested in a correlation function at pi = 0 such as the magnetization of a ferromagnet.8 Analyticity of β-functions** Finite-Size Scaling Temperature plays a very diﬀerent role in classical and quantum statistical mechanics. u − u∗ . In the classical theory. The temperature dependence of physical quantities can be . gi − gi ) → 1 (u − ∗ u∗ )nζ /λu ∗ F (gi − gi ) (u − u∗ )−λi /λu (10. 0 (10.e. Then. u − u∗ . setting the other couplings to their ﬁxed point values in the small u − u∗ limit: ∗ C0 (u − u∗ . instead. even that g is the leading irrelevant coupling (i. gi − gi ) → 1 (u − ∗ u∗ )nζ /λu F pi (u − u∗ )−1/λu (10. ANALYTICITY OF β-FUNCTIONS** 157 Suppose.44) ∗ is small.e.10.47) Now imagine that there is a second relevant coupling. gi − gi ) → 1 (u − u∗ )nζ ∗ /λ u F ∗ gi − gi (u − u∗ )−λi /λu (10. ∗ ∗ C (pi . Call this coupling u and the other irrelevant couplings gi . has a gap – then C pi (u − u∗ )−1/λu . then we can write: ∗ C0 (u − u∗ . 1. then we can take b = (u − u∗ )1/λu and be in the b → 0 If u − u limit: ∗ C (pi .46) where F (x) is called a scaling function. The non-trivial physics is entirely in the prefactor. If the stable phase to which the system ﬂows is trivial – i. Then.

1) 1 (10. namely the size of the system (in our case. related way of dealing with ﬁnite-temperature is dicussed in the context of the NLσM in the last section of this chapter. β. g(ℓ). An alternative. 1) 1 (10. so (10.51) will give the temperature dependence of Green functions in the low-temperature limit. In the quantum theory. and we ﬁnd G(n) (pi . which satisﬁes a renormalization group equation with a low-energy ﬁxed point: ∂ 1 ∂ + β(g) + n δϕ − η (g) ∂ℓ ∂g 2 G(n) (pi eℓ .51) We will be primarily concerned with the case in which the ﬁnite size. g∗ . Finite-size scaling can be used in the limit in which β is large. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES determined from the scaling behavior of the temperature. Le−ℓ ) = 0 (10. While the equations themselves are unchanged. As a result. ﬁnite-size scaling allows us to compare numerics to analytical calculations which are more easily done in the inﬁnite-size limit. we have g(ln L) → g∗ . we have the scaling form: G(n) (pi . L. g. Since numerical calculations must be done in such systems. and the temperature dependence of physical quantities can be determined from ﬁnite-size scaling which we discuss below. For simplicity. L) = Ln(δϕ − 2 η(gj )) G(n) (pi L.49) G(n) is an n-point Green function. Since renormalization group equations describe the evolution of eﬀective Lagrangians as one integrates out short-distance physics. g. We may take eℓ = L. L is the ﬁnite size of the system. temperature is the size of the system in the imaginary time direction.50) Then in the large-size limit. β). β) ∼ β n(δϕ − 2 η) G(n) (pi β.52) . 1) 1 (10. the solutions are modiﬁed because they depend on an additional dimensionful parameter. φ4 theory). g. let us consider a theory with a single relevant coupling (say. G(n) (pi .158 CHAPTER 10. L → ∞. it is clear that these equations are insensitive to ﬁnite-size eﬀects so long as the ﬁnitesize is much larger than the inverse of the cutoﬀ. Finite-size scaling is also useful for dealing with systems which are ﬁnite in one or more spatial directions. L) = Ln(δϕ − 2 η(gj )) G(n) (pi L. will be the inverse temperature. Classical statistical mechanics can be used to calculate a correlation function at wavevector the temperature is larger than the important excitation energies since the n = 0 Matsubara frequencies can then be ignored. g∗ . g(ln L).

we look at the 1D Ising model. T = 0 and T = ∞.9. T is J J relevant at the T = ∞ ﬁxed point but irrelevant at the T = 0 ﬁxed point. Whereas the original model has wavevectors −π/a < k < π/a. . The ﬂow diagram is shown in 10. 10.9 Non-Perturbative RG for the 1D Ising Model In the next two sections. Z= = σ2+1 =±1 i e T σi σi+1 σi =±1 i J 2 cosh ′ J σ2i+1 + σ2(i+1)+1 T (10. First.10.3: The ﬂow diagram of the 1D Ising model.3. This procedure is called decimation. Only at T = 0 can the system be ordered.54) = σ2+1 =±1 i ′ J Ke( T ) σ2i+1 σ2(i+1)+1 J where K = 2e( T ) and J T ′ = 1 J ln cosh 2 2 T (10. H=J i σi σi+1 8 (10. we will look at two examples of RG transformations. We can then rescale momenta by 2 to obtain an RG tranformation. one non-perturbative and one perturbative. NON-PERTURBATIVE RG FOR THE 1D ISING MODEL * J/T=0 * J/T= 159 Figure 10. the resulting theory has −π/2a < k < π/2a.53) Our RG transformation will be done in ‘real space’ by integrating out the spins on the even sites. the system is controlled by the disordered T = 0 ﬁxed point.55) J J J This RG transformation has only 2 ﬁxed points. This ﬂow diagram shows that for any J T > 0.

we must take ϕ → ϕ b− d+2 2 (10. dℓ (10. . r has scaling dimension 2.12 Dimensional crossover in coupled Ising chains** Real-space RG** Perturbative RG for ϕ4 Theory in 4 − ǫ Dimensions dd q 1 dd q 1 2 q |ϕ(q)|2 + r |ϕ(q)|2 (2π)d 2 (2π)d 2 dd q1 dd q2 dd q3 u ϕ (q1 ) ϕ (q2 ) ϕ (q3 ) ϕ (−q1 − q2 − (10.10 10. At the Gaussian ﬁxed point. When the quartic interaction is irrelevant.58) We immediately ﬁnd one ﬁxed point. the Gaussian ﬁxed point: r = u = 0. the correct theory of the critical point is simply.56) q3 ) + 4! (2π)d (2π)d (2π)d Our second example is ϕ4 theory. You will recognize that this is the same as the Ginzburg criterion which determines when the saddle-point approximation is valid for this theory: the saddle-point approximation is valid when the quartic interaction is irrelevant. S= We take the ﬁrst term as Ofree . Under a rescaling q → qb. . . r.59) which has critical correlation functions ϕ(x) ϕ(0) ∼ (10. . dℓ du = (4 − d) u + . At this ﬁxed point. S= dd q 2 q |ϕ(q)|2 (2π)d 1 |x|d−2 (10. we immediately see that the leading terms in the RG equations are: dr = 2r + .60) The one relevant direction at the Gaussian ﬁxed point in d > 4 is the temperature. r is always relevant while u is irrelevant for d > 4 and relevant for d < 4.160 CHAPTER 10. .11 10. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES 10.57) Using this rescaling.

the Gaussian ﬁxed point has two unstable directions. etc.65) (2π)d (2π)d (2π)d Sint also contains ϕH ϕH ϕH ϕL and ϕL ϕL ϕL ϕH terms.64) Sint [ϕL . there is a contribution of the form: e−SΛ [ϕL ] = e−SL [ϕL ] b DϕH e−SH [ϕH ] (1− 0 . but the phase space for these terms is very small since it is diﬃcult for three large momenta to add up to a small momentum or the reverse. we can safely ignore these terms. Expanding perturbatively. the ϕ8 operator is only relevant below 8/3 dimensions. they are also irrelevant at the ﬁxed point which we ﬁnd below.ϕH ] 0 (10. Hence. the ϕ6 operator is only relevant below 3 dimensions. We can compute the RG equations to the one-loop level to ﬁnd other ﬁxed points. this works as follows: e−SΛ [ϕL ] = e−SL [ϕL ] where SL [ϕL ] = bΛ dd q 1 dd q 1 2 q |ϕL (q)|2 + r |ϕL (q)|2 (2π)d 2 (2π)d 2 0 0 u bΛ dd q1 dd q2 dd q3 + ϕL (q1 ) ϕL (q2 ) ϕL (q3 ) ϕL (q4 ) (10.63) 4! 0 (2π)d (2π)d (2π)d bΛ b DϕH e−SH [ϕH ] e−Sint [ϕL . (Actually. At a schematic level.10. r and u. We make the division of ϕ into ϕL and ϕH and integrate out ϕH at the one-loop level. they will be generated by the RG transformation. we can ignore the higher order terms in the asymptotic q → 0 limit because they are irrelevant. in fact. Hence. However.62) 0 SH [ϕH ] = dd q 1 2 q |ϕH (q)|2 + (2π)d 2 bΛ u 4! + u 4 Λ dd q 1 r |ϕH (q)|2 (2π)d 2 bΛ Λ (10. Hence. etc. r ∼ T − Tc . 1 (10. ϕ8 . As we discussed in the context of the Ising model.12. for d > 3. terms. ξ ∼ 1/ r. ϕH ] = dd q1 dd q2 dd q3 ϕH (q1 ) ϕH (q2 ) ϕH (q3 ) ϕH (q4 ) (2π)d (2π)d (2π)d dd q1 dd q2 dd q3 ϕH (q1 ) ϕH (q2 ) ϕL (q3 ) ϕL (q4 ) (10.61) ξ∼ |T − Tc |1/2 Of course. If we don’t include them initially in our action. we should also allow ϕ6 . we have only shown that they are irrelevant at the Gaussian ﬁxed point.) For d < 4. PERTURBATIVE RG FOR ϕ4 THEORY IN 4 − ǫ DIMENSIONS 161 √ Hence.

dd q1 dd q2 dd q3 ϕH (q1 ) ϕH (q2 ) ϕL (q3 ) ϕL (q4 ) + . 1− 4 (2π)d (2π)d (2π)d « u 4 dd q1 dd q2 dd q3 (2π)d (2π)d (2π)d = e−SL [ϕL ] =e „ R − SL [ϕL ]+ u 4 ϕH (q1 ) ϕH (q2 ) ϕL (q3 ) ϕL (q4 ) + O(u2 ) (10. (Note that the contribution to SbΛ is the negative of the value of the diagram since we are absorbing b it into e−SΛ [ϕL ] .. . 1 4! . Diagrams with four external legs give a contribution to SbΛ of the form: dd q1 dd q2 dd q3 v (q1 .4: The one-loop diagrams which determine the RG equations for (a) r and (b) u.4(a) gives the contribution of (10.) The contribution described above results from the ﬁrstorder diagram with two external legs.162 CHAPTER 10.66) We can do this diagrammatically by computing one-loop diagrams with internal momenta restricted to the range bΛ < |q| < Λ. i. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES (a) (b) Figure 10. q2 . must have momenta q < bΛ. The diagram of 10. Such diagrams give a contribution to SbΛ of the form: dd q c(q) |ϕL (q)|2 (10.4. The external legs must be ϕL ﬁelds. . . .e. . .67) d (2π) where c(q) = c0 + c2 q 2 + . q3 ) ϕ (q1 ) ϕ (q2 ) ϕ (q3 ) ϕ (−q1 − q2 − q3 ) (2π)d (2π)d (2π)d (10.68) The one-loop contribution to the RG equations is given by the diagrams of ﬁgure 10.) (2π)d (2π)d (2π)d dd q1 dd q2 dd q3 u ϕH (q1 ) ϕH (q2 ) ϕL (q3 ) ϕL (q4 ) + .66).

71) or. PERTURBATIVE RG FOR ϕ4 THEORY IN 4 − ǫ DIMENSIONS 163 namely 1 δSbΛ = − u 2 b 1 dd q 1 d q2 + r (2π) dd p 1 |ϕ(p)|2 (2π)d 2 (10.72) Observe that there is no one-loop contribution to the dd q 2 q |ϕ(q)|2 (2π)d term.69) Dropping higher-order terms in r (because we are interested in the vicinity of the Gaussian ﬁxed point.74) As a result. we ﬁnd the one-loop RG equations: r + dr = b−2 (r + 2π2 1 u 2 (2π)d Γ d d 2 1 2π2 − ur 2 (2π)d Γ d 2 1 1 − bd−4 ) d−4 1 1 − bd−2 d−2 .12.73) (10.70) Meanwhile. we can rewrite the integral as 1 b dd q 1 = (2π)d q 2 + r 1 dd q 1 dd q 1 −r d q2 (2π) (2π)d q 2 (q 2 + r) b b 1 1 dd q 1 dd q 1 −r + O(r 2 ) = (2π)d q 2 (2π)d q 4 b b d 1 2π2 1 − bd−2 = (2π)d Γ d d − 2 2 1 2π2 − (2π)d Γ d d 2 r 1 1 − bd−4 d−4 (10. adding them together and evaluating the integral in the r = 0 limit. the three diagrams of ﬁgure 10. r = u = 0). 1 2 u 2 1 dd q (2π)d (q 2 + r)2 − 1 3 2 u d 2 (2π) 2 Γ d 2 1 1 1 − bd−4 d−4 dd q1 dd q2 dd q3 1 ϕ (q1 ) ϕ (q2 ) ϕ (q3 ) ϕ (q4 ) (2π)d (2π)d (2π)d 4! (10. Hence the correct rescaling is still ϕ → ϕ b− d d+2 2 (10.4(b) each give a contribution − dd q1 dd q2 dd q3 1 ϕ (q1 ) ϕ (q2 ) ϕ (q3 ) ϕ (−q1 − q2 − q3 ) (2π)d (2π)d (2π)d 4! b (10.10.

and taking the limit of small ǫ = 4 − d we have: (10. . These RG equations have a ﬁxed point at O(ǫ): 1 r∗ = − ǫ 6 16π 2 u∗ = ǫ 3 At this ﬁxed point. (10.5.5: The ﬂow diagram of a ϕ4 theory in 4 − ǫ dimensions. .78) dℓ There is only one relevant direction (corresponding to the temperature) with 1 scaling dimension 1/ν = 2 − 3 ǫ. . . u + du = bd−4 (u − 2π2 3 2 u 2 (2π)d Γ d d 2 1 1 − bd−4 ) d−4 (10.79) Writing b = e−dℓ . THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES r*= −ε/6 u*= 16π ε/3 2 * * r*= u*= 0 Figure 10. 2 dℓ 16π 16π 2 du 3 = ǫu − u2 + . The correlation length scales as: ξ∼ 1 |T − Tc |ν (10.75) dr 1 1 = 2r + u− ur + .164 CHAPTER 10. the eigenoperators are: 1 d (r − r ∗ ) = 2 − ǫ (r − r ∗ ) dℓ 3 d ∗ (u − u ) = − ǫ (u − u∗ ) (10.76) dℓ 16π 2 The corresponding ﬂow diagram is shown in ﬁgure 10.77) .

. (10. However.e. we ﬁnd a non-vanishing contribution: η = ǫ2 /54. We can now justify their neglect: these terms are irrelevant. our neglect of higher-loop contributions is justiﬁed. . such as exponents. in the RG equation dg = λg (g − g∗ ) + .g. as we have seen. are scheme independent.12. PERTURBATIVE RG FOR ϕ4 THEORY IN 4 − ǫ DIMENSIONS 165 At the critical point. but g∗ is scheme-dependent. • Both RG equations and the ﬁxed point values of the couplings are scheme dependent. this is an artifact of the O(ǫ) calculation. η = 0.81) dℓ λg is scheme independent and independent of all microscopic details. Consequently they are scheme independent. however. by introducing a hard cutoﬀ (as in ϕ4 theory above). • Corresponding to these diﬀerent regularization schemes. For ǫ small. by multiplying all momentum 2 2 integrals by e−q /Λ – just to name a few. or by introducing a soft cutoﬀ – e. e. the correlation function has the power law decay ϕ(x) ϕ(0) ∼ 1 xd−2+η (10. Several remarks are in order: • As we showed in chapter 8. • There are many diﬀerent ways of implementing the cutoﬀ. the Ising model can be mapped onto a ϕ4 theory if higher powers of ϕ are neglected. ultimately. higher-order in ǫ. i. . and can depend on microscopic details such as the cutoﬀ. equivalently.g. At the next order. into the same ﬁxed point under the diﬀerent RG transformations. we must go to higher loops or. The universal properties. such as real-space decimation on the lattice or momentum shell-integration for a momentumspace cutoﬀ.10. there are different renormalization group transformations. Since all of these diﬀerent cutoﬀ theories will ﬂow. it is a matter of convenience which scheme we choose. Integrals which are more strongly ultra-violet divergent are scheme-dependent.80) At order ǫ. regularizing a theory: by putting the theory on a lattice (as in the 1D Ising model above). To compute in d = 3. • Integrals which are logarithmically divergent at large q are proportional to ln b and are independent of the cutoﬀ Λ.

the internal momenta are restricted to the shell e−ℓ Λq < Λ.e. Instead. In fact. . This is diﬀerent from the RG which we deﬁned earlier. . it vanishes. we would like to consider even the high temperatures at which classical phase transitions can occur. but it is still perfectly well-deﬁned. but the Matsubara frequencies can run . It does not aﬀect the scaling exponents.13 The O(3) NLσM In order to study the phase diagram of a quantum system we would like to consider both T = 0 and ﬁnite T in the same phase diagram. r ∗ .85) We now deﬁne an RG transformation in which we integrate out ni (q. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES • The term in the RG equation for r which is independent of r determines only the ﬁxed point value of r. dℓ 16π 2 (10. so it can be dropped and we can work with: 1 dr = 2r − ur + . we rewrite the action as: S= dd x 0 β dτ (∂µ ni )2 + g ni ∂µ ni nj ∂µ nj 1 − gni ni (10.84) but no relevant terms for d ≥ 1. We will show how this is done for an antiferromagnet: Dn e 1 −g R dd x Rβ 0 dτ (∂µ n)2 (10. .82) 10. dℓ 16π 2 3 du = ǫu − u2 + . In particular. in some schemes.166 CHAPTER 10.83) The requirement of O(3) symmetry together with the constraint n2 = 1 implies that we can add to this action irrelevant terms such as dd x dτ 0 β dτ (∂µ n · ∂µ n)2 (10. it is renormalization scheme dependent. i. In the evaluation of diagrams. Finite-size scaling – which is useful for asymptotically low temperatures – is not appropriate for such an analysis. As usual. ωn ) with wavevectors e−ℓ Λ < |q| < Λ but arbitrary Matsubara frequency ωn . we carry out the renormalization group transformation directly on the ﬁnitetemperature quantum-mechanical functional integral. .

dβ = −β dℓ (10. we can write a scaling equation for t: d dℓ g β = d 1 dg +g β dℓ dℓ 1 β . THE O(3) NLσM 167 t tc * * * gc g Figure 10.86) (10.88) In the problem set.89) d 2 β changes trivially since it is only aﬀected by the rescaling.6: The ﬂow diagram of an antiferromagnet in d > 2. we rescale both momenta and frequencies: q → qe−ℓ ωn → ωn e−ℓ The second equation means that the temperature is rescaled: β → βeℓ ni must also be rescaled: ni → ni e−ℓζ 2π2 (2π)d Γ d (10.90) Hence. It is: dg 1 = (1 − d) g + dℓ 2 g2 coth β 2 (10. you will compute the one-loop RG equation for g. in the rescaling step.10. from n = −∞ to n = ∞.87) (10. However.13. if we deﬁne the parameter t = g/β.

91) g 2π2 1 = (2 − d) + β 2 (2π)d Γ d 2 g In other words. the antiferromagnet is disordered by quantum ﬂuctuations. At this ﬁxed point.92) At zero temperature.168 CHAPTER 10. t = 0. Such ﬁxed points are called quantum . g small. 1 = β 2π2 1 (1 − d) g + 2 (2π)d Γ d d d 2 g2 coth g β coth β 2 β 2 + g β (10.7: The ﬂow diagram of an antiferromagnet in d = 2. we can rewrite the RG equation for g and the trivial rescaling for β as the two equations: dg 1 = (1 − d) g + dℓ 2 1 dt = (2 − d) t + dℓ 2 2π2 (2π)d Γ 2π2 (2π)d Γ d d d 2 d 2 g2 coth g t coth g 2t g 2t (10. This is the antiferromagnetically ordered phase. For t. the system ﬂows into the g∗ = t∗ = 0 ﬁxed point. the ﬁrst equation shows that there is a stable ﬁxed point at g∗ = 0 for d > 1. g ﬂows to g = ∞. We will discuss the basin of attraction of this ﬁxed point below. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES t * * gc g Figure 10.93) For g > gc . There is an unstable ﬁxed point at gc = (d − 1)(2π)d Γ π2 d d 2 (10.

The region underneath the dark line in ﬁgure 10.95) In the problem set. an antiferromagnet described by the O(3) NLσM exhibits the following physics: • An antiferromagnetic phase controlled by the g∗ = t∗ = 0 ﬁxed point. you will calculate νd . at ﬁnite temperture.13. At ﬁnite temperature. At the ﬁnite-temperature critical point.e. you will calculate ηd . the correlation functions exhibit power-law decay. Tc → ∞. For given g – i. the divergence is: ξ ∼ |g − gc |−νd+1 while. all ﬂows go to t = ∞. For g → 0. At both the zero and ﬁnite-temperature ﬁxed points. n =0 (10. 0 < T < Tc . there is no ﬁxed point at ﬁnite temperature. the correlation length diverges. they decay as: ni (x. g can be varied by introducing a next-neighbor coupling J ′ which frustrates the nearest-neighbor coupling J. This phase is characterized by an O(3) symmetry which is spontaneously broken to U (1). In the zero-temperature case. t∗ = 0 ﬁxed point.10. To summarize. τ ) nj (0) ∼ 1 (x2 + τ 2 )(d−1+ηd+1 )/2 δij (10.7.6 is the antiferromagnetically ordered phase controlled by the g∗ = 0. for a given system – there is a range of t for which the system is antiferromagnetically ordered.96) (10. This range of t translates into a range of temperatures. The ﬂow diagrams are shown in ﬁgures 10.6 and 10.97) d x while at the zero-temperature critical point. g∗ = 0 t∗ = tc = (d − 2)(2π)d Γ d 2 (10. THE O(3) NLσM 169 critical points. As these ﬁxed points are approached. it is: ξ ∼ |t − tc |−νd (10. t > 0.98) In the problem set. . the correlation functions have power-law decay: 1 ni (x) nj (0) ∼ d−2+η δij (10.94) d 2π2 For d ≤ 2. Increasing J ′ increases g. in d > 2.99) yielding two Goldstone modes. there is a ﬁxed point for d > 2 at.

the correlation length diverges as: ξ ∼ |g − gc |−νd+1 (10.170 CHAPTER 10. In the next section. g∗ = 0 and characterized by exponentially-decaying correlation functions: e−|x|/ξ ni (x) nj (0) ∼ d−2 δij (10. .100) • A zero-temperature paramagnetic phase controlled by a ﬁxed point at g∗ = ∞. we have an antiferromagnetic phase only at T = 0 for 0 < g < gc .102) • A ﬁnite-temperature critical point at t∗ = tc .103) Near 4 dimensions. t∗ = 0 and characterized by exponentially-decaying correlation functions: ni (x.104) x As t → tc . we will generalize the O(3) NLσM to the O(N ) NLσM and derive the RG equations to lowest order in 1/N . τ ) nj (0) ∼ 1 (x2 + τ 2 )(d−1+ηd+1 )/2 δij (10. g∗ = 0 characterized by power-law correlation functions: ni (x) nj (0) ∼ 1 δij xd−2+ηd (10.101) As gc is approached at t = 0.105) In d = 2. The calculations of this section and the problem set are all to lowest order in d − 1 at zero-temperature and d − 2 at ﬁnite temperature. the correlation length diverges as. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES • A zero-temperature quantum critical point g∗ = gc . ξ ∼ |t − tc |−νd (10. t∗ = 0 characterized by power-law correlation functions: ni (x. this critical point can can be studied with an O(3) (ϕa ϕa )2 theory using a an ǫ = 4 − d expansion. The system is paramagnetic in the rest of the phase diagram. τ ) nj (0) ∼ e−|x|/ξ (x2 + τ 2 )(d−1)/2 δij (10. • A ﬁnite-temperature paramagnetic phase controlled by a ﬁxed point at t∗ = ∞.

Figure 10. ζ).14 Large N Suppose we generalize ϕ4 theory to S= dd x 1 1 1 u ∇ϕa ∇ϕa + r ϕa ϕa + (ϕa ϕa )2 2 2 N 8 (10. the theory has the O(2) = U (1) symmetry of 4 He or an XY magnet. we get a factor of N . LARGE N 171 (a) (b) Figure 10. such as that of ﬁgure 10. For N = 2. Every time we sum over an index a. Other diagrams. To organize the diagrams in powers of 1/N . First.8a contains some O(1) two-leg diagrams. the action has O(N ) symmetry. 2. this theory has the Z2 symmetry of the Ising model.9b are down by powers of 1/N . These diagrams renormalize r (and. . Each vertex gets a factor of 1/N . 10. Let’s classify diagrams according to powers of N . .107) .10.9a contains some O(1/N ) diagrams with 4 external legs. The RG equations simplify for N → ∞ as we now show. . For arbitrary N .14.106) where a = 1. . N . possibly. Let’s now turn to the diagrams with 4 external legs. Figure 10. it is useful to perform a Hubbard-Stratonovich transformation. We introduce a ﬁeld σ and modify the action by adding a term: S→S− u N σ− ϕa ϕa 2u 2N 2 (10. For N = 1.8: (a) O(1) diagrams and (b) O(1/N ) diagrams with two external legs. let’s consider the diagrams with two external legs.

110) Since there is a factor of N in front of Seﬀ [σ]. it is also possible to expand the square. . a diagram goes as N V −I−E = N −E+1−L . σ(pn ) δ (pi + ki − ki+1 ) (10. which leads to the action: S= dd x 1 N 2 1 1 ∇ϕa ∇ϕa + r ϕa ϕa − σ + σϕa ϕa 2 2 2u 2 (10. the logarithm can be expanded to give: Seﬀ [σ] = − dd p N σ(p) σ(−p) 2π 2u n d (−1)n d pi dd ki 1 +N 2 n 2π 2π ki + r n i=1 σ(p1 ) . THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES (a) (b) Figure 10.9: (a) Some O(1/N ) diagrams and (b) an O(1/N 2 ) diagram with four external legs. This is now a quadratic action for ϕa . Hence. Since the action is quadratic in σ.109) Dropping a constant. while each vertex carries an N . The lowest order in 1/N for the E-point σ correlation function is a tree-level in Seﬀ [σ] diagram. each σ propagator carries a 1/N . we can integrate out ϕa : Seﬀ [σ] = dd x (2π)d − N 2 σ + N T r ln ∇2 + r + σ 2u (10. However. To compute the ϕa k-point correlation . Hence. we could integrate out σ without aﬀecting the functional integral for ϕa . .108) Notice that integrating out σ restores the (ϕa ϕa )2 term.172 CHAPTER 10.

Since the σ one-point function is: N 1 dd q d q2 + r (2π) −1 (10. It is O(1/N ) and it is given by the graphs of ﬁgure ??b. Imposing the constraint n2 = 1 with a Lagrange multiplier. the one-loop RG equations are valid even when ǫ is not small. dℓ 16π 2 3 du = ǫu − u2 + . . (10. It is O(1) and it is given by the graph of ﬁgure ??a. we obtain the RG equations: dr 1 = 2r − ur + . We can also consider the O(N ) generalization of the NLσM: Dn e− g 1 R dd x Rβ 0 dτ (∂µ n)2 (10.116) .14. we have: S= 1 2g dd x 0 β dτ (∇n)2 + λ n2 − 1 (10.113) Diﬀerentiating these equations.112) − −1 u + du = bd−4 N − 3N 2u (2π)d (q 2 + r)2 (10. .115) where n is an N component vector.114) dℓ 16π 2 In other words. LARGE N 173 function.10. we need to compute a diagram with k/2 external σ legs. the one-loop RG equations contain the same information as the geometric series of O(1) and O(1/N ) diagrams! In the N → ∞ limit. Hence. The lowest order in 1/N contribution to the ϕa four-point correlation function is obtained from the σ two-point correlation function. the lowest order in 1/N contribution to the ϕa two-point correlation function is obtained from the σ one-point correlation function (which is determined by the diagram obtained by joining the one-point function to one end of the two-point function). . .111) while the σ two-point function at zero momentum is N −N 2u we have: r + dr = b−2 r + N N 2u 1 dd q 1 b (2π)d q 2 +r 1 dd q N b (2π)d 2 1 2 (q +r) 1 dd q 1 b dd q 1 d (2π) (q 2 + r)2 (10.

124) . we have: S= 1 2g dd x 0 β dτ 1 1 λ n2 − 1 + N T r ln −∇2 + λ(x)(10. First. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES Integrating out n. we have: 1 1 − bd−1 d−1 (10. assuming that λ << Λ2 : b N or.117) 2g 2 In the N → ∞ limit. b dω dd q 1 +N d ω2 + q2 + λ 2π (2π) 1 b dω dd q 1 1 = d ω2 + q2 + λ 2π (2π) g (10. the saddle-point approximation becomes exact. so: N n 1 dd q 1 = d ω2 + q2 + λ g (2π) n (10. q → qb.118) Let’s specialize to the case T = 0: Λ N 1 dω dd q 1 = d ω2 + q2 + λ 2π (2π) g (10. we integrate out momenta bΛ < |q| < Λ.123) In other words.119) This integral equation can be solved using the RG transformation. ω → ωb we have: N d 2 1 b 1 1 2π2 dω dd q = −N d ω2 + q2 + λ 2π (2π) g (2π)d Γ d N dω dd q 1 = bd−1 2π (2π)d ω 2 + q 2 + λb2 1 2π2 −N g (2π)d Γ d d 2 1 1 − bd−1 d−1 (10.120) N dω dd q 1 2π2 +N 2π (2π)d ω 2 + q 2 + λ (2π)d Γ d d 2 1 1 1 − bd−1 = (10.174 CHAPTER 10.122) Rescaling the momenta in the integral.121) d−1 g If we bring the second term on the left-hand-side to the right-hand-side. 1 +d g 1 g =b d−1 2π2 1 −N g (2π)d Γ d d 2 1 1 − bd−1 d−1 (10.

129) (10. equivalently. Clearly.130) R d2 x d2 x′ J(x)G(x−x′ )J(x′ ) = eG(x)−G(0) (10. This is an O(2) non-linear sigma model with K = 1/g. THE KOSTERLITZ-THOULESS TRANSITION writing b = e−dℓ .128) ρs 2T d2 x (∇θ)2 (10.126) or at ﬁnite temperature in 2D. this gives: 2π2 dg = (d − 1) g − N dℓ (2π)d Γ d 175 d 2 g2 (10. 10.15. S= We will use the notation S= 1 K 2 d2 x (∂µ θ)2 (10. the large-N limit is one in which RG equations can be calculated with minimum fuss.127) to encompass both cases. eiθ(x) e−iθ(0) = e 2 1 Dθ e R 1 d2 x( 2 K(∂µ θ)2 +J(x)θ(x)) (10. 4 He at zero-temperature in 1D S= 1 ρs 2 dτ dx (∂µ θ)2 (10.15 The Kosterlitz-Thouless Transition We turn now to the RG analysis of an XY magnet or. the large-N RG equation is essentially a one-loop RG equation. dK/dℓ = 0 to all orders in K. As we will see again in the context of interacting fermions.10. The two-point correlation function of the order parameter may be calculated using: eiθ(x) e−iθ(0) = where J(y) = iδ(y − x) − iδ(y) Hence it is given by.131) .125) Again.

n.176 Now.136) implies that ∂µ θ ∼ 2πn r (10. for |x| large. we have neglected the periodicity of θ. n is the winding number of the vortex. However. i.133) = 0. In other words. Thus far. ei P i ni θ(xi ) = e− 2πK 1 P ln |xi −xj | (10.e. A vortex is a singular conﬁguration of the ﬁeld θ(x) such that the vector ﬁeld ∂µ θ(x) twists around an integer number. CHAPTER 10. the fact that 0 < θ < 2π. This means that θ must wind around 2π. which encloses the vortex. it is a point about which the spins rotate. the correlation function has the form of the Boltzmann weight for a Coulomb gas.e. then (10. THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES G(x) − G(0) = θ(x) θ(0) − θ(0) θ(0) 1 1 d2 q eiq·x − 1 = (2π)2 K q2 1 Λ d2 q 1 =− K 1 (2π)2 q 2 |x| 1 =− ln |x/a| 2πK where a = 1/Λ is a short-distance cutoﬀ. ∂µ θ dxµ = 2πn P (10. times as the vortex is encircled.134) In other words.137) . a vortex is a swirl of current.135) tells us that (θ(x) − θ(0))2 becomes large for |x| large. Hence. In the context of 4 He. eiθ(x) e−iθ(0) = Similarly if i ni (10. P . e− 2 1 (θ(x)−θ(0))2 = 1 |x| 2πK 1 (10. If there is a vortex at the origin with winding number n. i. We can understand this qualitatively by calculating the contribution of a vortex conﬁguration to the functional integral.132) 1 |x| 2πK i. In an XY magnet. that there are vortices.136) for any path.j ni nj 1 (10.

Using ∂µ θV (x) = i ni ǫµν (x − xi )µ (x − xi )2 (10. The proliferation of vortices destroys the power-law correlation functions. this is a large contribution. and a piece that contains the vortices θV . THE KOSTERLITZ-THOULESS TRANSITION So the action of a vortex at the origin is: Av = 2πn 2 2 d r r R = πKn2 ln + Ec a K 2 177 (10. Let us now study this transition more systematically. so vortices can proliferate.10. Hence. θs .138) where R is the size of the system.140) For K < 2/π. Meanwhile a vortex-anti-vortex pair separated by a distance r has energy Apair = πK ln r + 2Ec a (10. we must take into account the fact that the vortex can be placed anywhere in the system.15.139) To calculate the contribution of a vortex to the functional integral.141) where the ith vortex has winding number ni and position xi .142) = i ni ǫµν ∂µ ln |x − xi | we can rewrite the action as (the cross term between θs and θV vanishes upon integration by parts) S= 1 K 2 d2 x (∂µ θ)2 . a is the size of the core of the vortex and Ec is the core energy. We break θ into a smooth piece. the contribution to the functional integral is proportional to the area of the system: Zv ∼ R a 2 e−Av R ∼ e(2−πK) ln a (10. θV (x) = i ni arctan (x − xi )2 (x − xi )1 (10.

THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES d2 x (∂µ (θs + θV ))2 1 K 2 1 d2 x (∂µ θs )2 + K 2 d2 x (∂µ θs )2 + d2 x (∂µ θs )2 − 2πK d2 x (∂µ θV )2 d2 x i. nv =0 (10. the partition function is: Z= Dθs e− 2 K e−nv Ec i=1 ni =0.134) implies that this can be rewritten as: Z[φ] = Dθs Dφe − R d2 x “ 1 8π 2 K (∂µ φ)2 + P m ym cos mφ ” e− 2 K 1 R d2 x (∂µ θs )2 (10.134).±1. This is an example of a duality transformation. ..145) where ym = e−nv Ec is the vortex fugacity. i y1 1 −1 y2 2 −2 .147) Notice that we have transformed the partition function for the vortices in the ﬁeld θ into the partition function for another scalar ﬁeld. The action for φ is a sine-Gordon model. 2 −πK Pi. Hence. .j ni nj ∂µ ln |x − xi | ∂µ ln |x − xj | (10. we have restored the core energies of the nv vortices. i 1 R d2 x (∂µ θs )2 ∞ nv × P 1 d2 xi e−πK i. N +N N +N (10. . The perturbative expansion of Z[φ] function is the sum over all vortex conﬁgurations of Z[θ].146) Integrating out θs . we are left with: Z= Dφ e − R d2 x “ 1 8π 2 K (∂µ φ)2 + P m ym cos mφ ” (10.. . φ.j ni nj ln |xi − xj | + nv Ec In the last line. Expanding perturbatively in the yi ’s and using (10. Let us consider the cos mφ term in the action: S= d2 x 1 (∂µ φ)2 + 8π 2 K ym cos mφ m (10.±1. . Observe that (10. we have: Z[φ] = ∞ nv =0 nv e −nv Ec i=1 ni =0.144) where Nk is the number of vortices of strength k in a given conﬁguration.±2.148) . ..±2.143) i..178 1 K 2 1 = K 2 1 = K 2 = = 1 K 2 CHAPTER 10.j ni nj ln |xi −xj | d xi e N1 ! N−1 !N2 ! N−2 ! ...j ni nj ln |xi −xj | N1 ! N−1 !N2 ! N−2 ! .

Z= Dφ e − R d2 x “ 1 8π 2 K (∂µ φ)2 + y cos φ ” = ...149) which tells us that cos mφ has dimension πmK...150) dℓ Consequently.. Hence.. y ≡ y1 is the most relevant operator.151) In order to study the ﬂow of K resulting from the presence of y. . + y2 Dφ y2 2 y2 Dφ[ 2 d2 x cos φ(x) d2 y cos φ(y) e − R d2 x “ 1 8π 2 K (∂µ φ)2 ” + . y1 becomes relevant ﬁrst – i.. + Dφ y 2 ×e − R d2 x “ d2 x 1/Λ<|x−y|<1/bΛ 1 8π 2 K d2 y + . . d y (∂φ(y)) 2 2 )e − R d2 x “ 1 8π 2 K (∂µ φ)2 + y cos φ ” + . Let us. focus on the action with only the m = 1 term: S= d2 x 1 (∂µ φ)2 + y cos φ 8π 2 K (10. + = . As K is decreased.15.. therefore.... + × Dφ(y 2 d2 x 1/Λ<|x−y|<1/bΛ R 1 − d2 x 1 + i(x − y)∂φ(y) − (x − y)2 (∂φ(y))2 ) e 2 1 1 d2 x Dφ(− y 2 4 (x − y)2πK−2 1/Λ<|x−y|<1/bΛ 1 1 2 (x − y)2πK “ 1 8π 2 K (∂µ φ)2 + y cos φ ” + .. This can be determined from the correlation function: cos θ(x) cos θ(0) ∼ 1 |x|2πmK (10. .. d2 y (∂µ φ)2 + y cos φ ” 1 1 eiφ(x)−iφ(y) 2 (x − y)2πK = . (10... + × = . let us expand the functional integral perturbatively in y.. = .e... we can determine the ﬁrst term in the RG equation for ym from its scaling dimension.. d2 x cos φ(x) |x−y|>1/bΛ d2 y cos φ(y) + “ 1 8π 2 K d2 x 1/Λ<|x−y|<1/bΛ d2 y R 1 iφ(x) −iφ(y) − d2 x ]e e e 2 (∂µ φ)2 ” + .10. at K = 2/π. THE KOSTERLITZ-THOULESS TRANSITION 179 Is this term relevant or irrelavant? At ym = 0. the RG equation for y is: dym = (2 − πmK)ym + ..

and re-exponentiated the result. Hence. At the critical point. The resulting Kosterlitz-Thouless ﬂow diagram is shown in ﬁgure 10.180 = . THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES Dφe − R d2 x h“ 1 +π 2 8π 2 K ” i y 2 ln b (∂µ φ)2 + y cos φ (10.152) In the last line... These RG ﬂows feature a line of ﬁxed points – or a ﬁxed line – y ∗ = 0.10. in terms of which the RG equations are: du = −v 2 + O(uv 2 ) dℓ dv = −uv + O(v 3 ) dℓ Observe that u2 − v 2 is an RG invariant to this order: d u2 − v 2 = 0 dℓ (10. Any point below the the asymptote u = v – or.154) (10. the RG trajectories in the vicinity of K = Kc = 2/π are hyperbolae which asymptote the lines u = ±v. Correlation functions exhibit power-law falloﬀ at these ﬁxed points: eiθ(x) e−iθ(0) = 1 |x| 2πK 1 (10. dy = (2 − πK)y + O(y 3 ) dℓ (10. + CHAPTER 10. πK − 2 = 4πy – ﬂows into one of these ﬁxed points. equivalently. we have done the integral at πK = 2 (since we are interested in the vicinity of the transition) where it is logarithmic.158) 1 |x| 4 .156) (10. 1 eiθ(x) e−iθ(0) = (10. K −1 ﬂows as a result of y: d −1 K = 4π 3 y 2 + O(y 4 ) dℓ Together with the ﬂow equation for y. These equations may be analyzed by deﬁning u = πK − 2 and v = 4πy. K > Kc .157) The line πK − 2 = 4πy which separates these power-law phases from the exponentially decaying phase is called the Kosterlitz-Thouless separatrix.155) Hence.153) these RG equations determine the physics of an XY model in 2 dimensions at ﬁnite temperature or in 1 dimension at zero temperature.

4 He and XY magnets have a phase transition between a disordered phase and a power-law ordered ‘phase’. the correlation length diverges as: 1 (10. INVERSE SQUARE MODELS IN ONE DIMENSION** y 181 Kc K Figure 10. it bounds the ﬂuctuations of φ. When the system is above the line πK −2 = 4πy. 10.159) ξ ∼ e (Kc −K)1/2 Hence.16. at ﬁnite temperature in 2D or at zero-temperature in 1D.17 10.16 10. In the problem set. you will show that as Kc is approached from below. Since the cos φ term is relevant. just as an rφ2 term would. it ﬂows away to large y: the system is disordered by the proliferation of vortices and has exponentially decaying correlation functions.18 Inverse square models in one dimension** Numerical renormalization group** Hamiltonian methods** .10.10: The Kosterlitz-Thouless ﬂow diagram.

THE RENORMALIZATION GROUP AND EFFECTIVE FIELD THEORIES .182 CHAPTER 10.

ψ (x. we must modify the Hamiltonian by H → H − µN . t Performing a mode expansion ψ(x) = d3 k c eξk τ +ik·x (2π)3/2 k 183 = δ(x − x′ ) =0 (11. we will be applying the ﬁeld-theoretic techniques which we have developed to systems of interacting electrons.1 Canonical Anticommutation Relations In the remainder of this course. ψ † x′ . i. This can be accomplished by imposing the canonical anticommutation relations.3) = ψ † (x. ψ x′ . In order to do this.1) The diﬀerence is that we want the associated Fock space to be fermionic. we will have to make a detour into formalism so that we can handle systems of fermions. The action is the same as for a system of free bosons: S= dτ d3 x ψ † 1 2 ∂ + ∇ −µ ψ ∂τ 2m (11. we would like the Pauli exclusion principle to hold. Let us ﬁrst consider a system of non-interacting spinless fermions at chemical potential µ.e. t .2) (11. t) . ψ † x′ .CHAPTER 11 Fermions 11. t) . As in the case of 4 He. t ψ (x. t) .

i. FERMIONS d3 k c† e−ξk τ −ik·x (2π)3/2 k (11. c† ′ k ck . The ﬁnal equality follows from the fact that τ − τ ′ + β > 0. τ ) = δ(τ ) δ(x) ∂τ 2m (11. Hence.6) Note the −sign in the deﬁnition of the Green function. It is necessary because the fermions satisfy canonical anticommutation relations. Since 0 < τ. (c† )2 = c2 = 0. we ﬁnd a further condition which follows from the cyclic property of the trace. k k The Green function is: G(x. τ )ψ(0. 0) −θ(−τ ) T r e−β(H0 −µN ) ψ(0.5) Hence.7) As in the bosonic case. c† ′ k k = δ(k − k′ ) =0 (11.8) G(iǫn ) = 0 dτ e−iǫn τ G(τ ) (11. Then. ck′ = c† .e. we can take the Fourier transform over the interval [0. Now suppose that τ < τ ′ . You may verify that G as deﬁned above satisﬁes: 1 2 ∂ + ∇ − µ G(x.184 ψ † (x) = CHAPTER 11. β]: β = −T r eτ (H−µN ) ψ † (x)e−τ (H−µN ) e−β(H−µN ) eτ ψ(x′ )e−τ ′ (H−µN ) (11. As a result of antiperiodicity in imaginary-time.4) where ξk = ǫk − µ = k2 /2m − µ. no state can be doubly occupied. τ ) (11. it follows that −β < τ − τ ′ < β. τ ) = θ(τ ) T r e−β(H0 −µN ) ψ † (x. 0)ψ † (x. G(τ − τ ′ < 0) = −T r e−β(H−µN ) eτ ′ (H−µN ) ψ(x′ )e−τ ′ (H−µN ) eτ (H−µN ) ψ † (x)e−τ (H−µN ) ′ (H−µN ) = −T r{e−β(H−µN ) eβ(H−µN ) eτ (H−µN ) ψ † (x)e−τ (H−µN ) e−β(H−µN ) ′ ′ eτ (H−µN ) ψ(x′ )e−τ (H−µN ) } = −G(τ − τ ′ + β) The ﬁrst equality follows from the cyclic property of the trace. a fermion Green function is anti-periodic in imaginary time.9) . τ ′ < β. we see that the creation and annihilation operators satisfy: ck .

2.10) Inverting the Fourier transform. 0) = nF (ξk ) = eβξk 1 +1 (11. τ )ψ(0. τ ) = θ(τ ) T r e−β(H0 −µN ) ψ † (x.13) ))) (2π)3 2ωk We can now compute the Fourier representation of the Green function: G(p. 0)ψ † (x. In order to do this.11) n Using the mode expansion and the Fermi-Dirac distribution. T r e−β(H0 −µN ) c† ck k we can compute the propagator: G(x.16) . τ ) = d3 k e−ik·x+ξk τ (θ(τ ) nF (ξk ) − θ(−τ ) (1 − nF (ξk(11.14) 11.15) and 2 ψi = 0 (11. iǫn ) = d3 xeip·x 0 β dτ e−iǫn τ G(x.11.12) −θ(−τ ) T r e−β(H0 −µN ) ψ(0.2 Grassmann Integrals Fermionic systems can also be described by functional integrals. Grassmann numbers are objects ψi which can be multiplied together and anticommute under multiplication: ψi ψj = −ψi ψj (11. GRASSMANN INTEGRALS where the Matsubara frequencies ǫn . we will need the concept of a Grassmann number. τ ) nF (ξk ) eβ(−iǫn +ξk ) − 1 =− −iǫn + ξk 1 = iǫn − ξk (11. are given by: ǫn = (2n + 1)π β 185 (11. we have: G(τ ) = 1 β G(iǫn ) eiǫn τ (11.

20) (11. dψ dψ = 0 dψ dψ ψ = 0 dψ dψ ψ = 0 dψ dψ ψ ψ = 1 (11. the order is important in the last line: dψ dψ ψ ψ = −1 (11.17) ψ = ψ1 + iψ2 and ψ = ψ1 − iψ2 can be treated as independent Grassmann variables.18) ψψ = −ψψ Since the square of a Grassmann number vanishes. dθn dθ n e P i. we can do Grassmann integrals of Gaussians. FERMIONS Grassmann numbers can be multiplied by complex numbers. .j eθi Aij θj .186 CHAPTER 11.22) Since the square of a Grassmann number vanishes.j θ i Aij θj = dθ1 dθ 1 . multiplication by a complex number is distributive: a (ψ1 + ψ2 ) = aψ1 + bψ2 (11. . Then dθ1 dθ1 . these rules are suﬃcient to deﬁne integration. With these deﬁnitions. Suppose θi and θi are independent Grassmann variables. For instance. (11. . the Taylor expansion of a function of Grassmann variables has only two terms. eψ = 1 + ψ Integration is deﬁned for Grassmann numbers as follows: dψ = 0 dψ ψ = 1 Similarly. dθn dθ n i.19) As a result of the anticommutation.21) (11. .

θ ik θjk e e P 1 det(A) (−1) σ ∂ ∂ ∂ ∂ . η i ) = i dθi dθi e P i.27) In other words.23) × θ1 θσ(1) . η i ) ∂ηi1 ∂η j1 ∂ηik ∂η jk Ajσ(1) i1 Ajσ(2) i2 . Thus far. we have considered ﬁnite-dimensional Grassmann integrals. Anσ(n) σ dθ1 dθ 1 . Anσ(n) (11. . θ i1 θj1 .. . we sum over all possible Wick contractions. . . . θik θjk = = = σ (A−1 )ij ηj (11. multiplying by −1 every time the contraction necessitates a reordering of the ﬁelds by an odd permutation. . . dθn dθn i. . . η i ) = i ′ η ij j −1 η j A ji ” “ ′ θ i Aij θj ′ +ηi (A−1 )ij ηj (11.26) i dθi dθ i θ i1 θj1 . . the generalization to functional Grassmann integrals is straightforward. .j θ i Aij θj + P i (ηi θi +θi ηi ) (11. .j η i ′ ′ P i.j = det(A) e Hence. .j 187 1 + θi Aij θj A1σ(1) A2σ(2) .11.j θ i Aij θj P i. . .2. θn θσ(n) = det (A) We can prove Wick’s theorem for Grassmann integrals: Z(ηi . . Ajσ (k)ik θ i1 θjσ(1) . θik θjσ(k) i dθi dθ i i. dθn dθn σ = σ (−1) A1σ(1) A2σ(2) .25) dθi dθi e P i.. .j θ i Aij θj ηi =ηi =0 = σ (−1)σ (11. θi = θi ′ + A−1 θ i = θi + we get Z(ηi .24) By making the change of variables. . Z(ηi . However. GRASSMANN INTEGRALS = = dθ1 dθ 1 . .

They are depicted in ﬁgure ??.j i.j i. The Grassmann variables will live on the sites of the dual lattice. so it can be performed using the deﬁnitions in this chapter.j h. These 8 conﬁgurations will give a graphical representation of the possible non-vanishing terms in the Grassmann integral.j .j + χv χh + ηi. Consider the following Grassmann integral.j dχh.j (11.j P i. χh.j + z χv ηi. j labels a site on the square lattice.188 CHAPTER 11.j variables.29) we produce terms which are strings of χh.j v h v h +χh ηi.j + χv ηi.v s. We will do this shortly.3 Solution of the 2D Ising Model by Grassmann Integration In this section. Thus.v dηi.29) where z = eβJ .j ηi.j i. When we expand the exponential of (11. and ηi.j (11.29). we will present the solution of the 2D Ising model as an application of Grassmann integration. It is straightforward to allow anisotropic coupling constants Jh .j h v +χh ηi. the Grassmann integral will be equal to the partition function of the Ising model.j+1 i. We can introduce the following graphical representation for the terms which are generated by expanding the exponential of (11. The integral of such a term is non-zero if and only if every single χh. Our strategy will be to represent a state of the Ising model in terms of the domain walls which separate upspins from down-spins. The ﬁrst two terms are . if we weight each of these terms in the Grassmann integral with the appropriate Boltzmann weight. Ai.j = z χh ηi+1.v appears exactly once in the term.j i. but ﬁrst we must establish the raison d’ˆtre of this integral. The h. v index stands for horizontal and vertical.j is given by h v Ai.j i.29) is quadratic.v e i. namely that it reproduces e 2D Ising model conﬁgurations.j i. We will then concoct a Grassmann integral which generates all allowed domain wall conﬁgurations. We will assume that the Ising model is on a square lattice. The ﬁrst thing to observe is that there are 8 possible conﬁgurations of the four spins surrounding a given site on the dual lattice.j + χv ηi. FERMIONS 11.j Ai.v where i.v are Grassmann i. up to an overall ﬂip of the four spins. Jv by replacing the coeﬃcient of the ﬁrst term by zh and the second by zv . The Grassmann integral (11. Z= i.v s and η h.v and η h.j i.28) h.

The only concern is that some of them may not come with the correct sign. The reader may check that they do.j would appear twice. In a conﬁguration in which there is no domain wall passing through (i. this term places a corner at site (i. This can only happen if the second term in Ai. They come with a coeﬃcient z.j also appear. essentially because every loop of domain wall contains an even number of segments and and even number of corners. j).v η h.j . j) or (i.ky (11. χh ηi.j−1 and Ai. j). j) and (i + 1.j cannot i. v Consider the ﬁrst such term. respectively. SOLUTION OF THE 2D ISING MODEL BY GRASSMANN INTEGRATION 189 represented by lines of domain wall connecting sites (i. both will appear.3. compute the partition function of the 2D Ising model by evaluating the Grassmann integral above.11.j to appear in the v h string because either χh or ηi.ky kx . an incoming line from the South to an outgoing line to the East (ﬁg ??).30) with a similar equation for χ. a horizontal domain wall can cross a vertical domain wall.j and the ﬁrst term in Ai−1.j and Ai. It is equal to the square root of the determinant of the quadratic form in i. which is the Boltzmann weight for such a segment of domain wall. this occurs when the ﬁrst term in Ai−1. j) and heading to (i + 1. We can almost diagonalize it by going to momentum space: 1 h. and an outgoing line to the North to an outgoing line to the East (ﬁg ??). then the ﬁrst of these terms will appear (as will the second term of Ai. j). Therefore. The last two terms simply prevent. which accounts for the conﬁguration in ﬁg ??. an incoming line from the West to an incoming line from the South (ﬁg ??). j − 1) to arrive at (i. χv or ηi.j appear and the second term in Ai. j).j Ai.j . Thus.j appear a second time in the string. The four types of corners connect. However. It is also impossible for a segment of wall coming from (i. The next four terms place a corner at site (i. j). It is also impossible for any of the ﬁnal ﬁve terms in Ai. indeed. j).v ei(kx i+ky j) ηi. However. Thus. j). we see that by expanding the exponential of bilinears in Grassmann variables we can reproduce all possible domain wall conﬁgurations. an incoming line from the West to an outgoing line to the North (ﬁg ??). we can. When it appears in a string. Hence.j appear.j i. If there is a vertical domain wall passing through (i. a horizontal or vertical domain wall from passing through (i. j + 1). with its concomitant z).j must appear once in the string.j = √ N kx . it is impossible to have a segment of domain wall originating at (i. respectively. Now the exponent in the Grassmann integral . These terms implement these corners in the following way.j i. Note that two such corners cannot occur at the same site.

and its concomitant singularities and power-law correlation function occur when r = 0.−ky ηkx . ky ))1/2 kx .e. Expanding about this critical point.−ky ηkx .−ky χhx .ky −k k h v + χh x .−ky ηkx .ky + z eiky χv x . Performing the integral above.ky −k −k h v + χv x . √ Such a system has correlation length ξ ∼ 1/ r.−ky ηkx .ky ) (11.31) Up to a normalization N. we have ξ ∼ 1/|T − Tc |. ky ) is the 4 × 4 matrix of coeﬃcients in (11. can be obtained by studying small k. the speciﬁc heat diverges logarithmically.ky 1 2 =Ne V R d2 k (2π)2 ln det M (kx . ν = 1.ky + z eiky + 1 χv x . − T (sinh 2βJ − 1)2 ln |sinh 2βJ − 1| 2π (11.−ky ηkx .ky −k −k h v + χh x .ky ] −k −k = kx .33) The infrared behavior of this integral.32) where M (kx . where possible singularities lurk.−ky ηkx .ky CHAPTER 11.−ky ηkx . FERMIONS v h [z eikx χh x . There is a critical point.−ky ηkx . the argument of the logarithm is of the form Kp2 +r. the Grassmann integral is given by the: Z=N (det M (kx .−ky ηkx .j = i. which is rather diﬀerent from the mean-ﬁeld result ν = 1/2.−ky χhx . i. Computing the determinant of this 4 × 4 matrix.ky −k −k v h + χh x .ky + χv x . α = 0− .ky k −k (11.34) Hence.ky −k −k v h + χv x . i.190 can be rewritten: Ai. where r = (sinh 2βJ − 1)2 . sinh 2βc J = 1.j kx .ky + χv x . . we ﬁnd: ln Z = 1 V 2 d2 k ln cosh2 2βJ − sinh 2βJ (cos kx + cos ky ) (2π)2 (11.−ky ηkx .31) and V is the area of the system.−ky ηkx . we obtain the free energy density: F = Fnon−sing. This is precisely what we would expect for a system of particles with propagator Kp2 + r.−ky ηkx .e. In this limit.ky h v [z eikx + 1 χh x .ky + η−kx .ky + η−kx .ky + χv x .

ψ † → e−iθ ψ † . this is the free fermion functional integral: Dψ Dψ e † − Rβ 0 (11.4 Feynman Rules for Interacting Fermions Let us now turn to a system of fermions with a δ-function interaction.41) The diﬀerence between this Green function and the Green function of a bosonic system with the same Hamiltonian is that this is the inverse of this operator on the space of functions with antiperiodic boundary conditions on [0. there is a conserved density. τ )ψ(0.42) −µ .38) (11. The Fourier transform of the Green function is: G(k.39) dτ R “ “ 2 ”” dd x ψ† ∂τ − ∇ −µ ψ 2m = det ∂τ − ∇2 −µ 2m (11. According to Noether’s theorem.40) The Green function is: G(x.36) This action has the U (1) symmetry ψ → eiθ ψ. β].4. The grand canonical partition function is given by: Z=N Dψ Dψ † e−S (11. τ ) = N = Dψ Dψ ψ (x.35) where the functional integral is over all Grassmann-valued functions which are antiperiodic in the interval [0.11. FEYNMAN RULES FOR INTERACTING FERMIONS 191 11. β] (so that ǫn = (2n + 1)π/β).37) ∇ ψ− m ∇ † ψ ψ m (11. ǫn ) = 1 iǫn − k2 2m (11. ρ = ψ† ψ and current j = ψ† satisfying the conservation law ∂ρ +∇·j = 0 ∂t For V = 0. 0) e ∇2 −µ 2m −1 † † − Rβ 0 dτ R ”” “ “ 2 ∇ dd x ψ† ∂τ − 2m −µ ψ ∂τ − (11. β S= 0 dτ dd x ψ † ∂τ − ∇2 −µ 2m ψ + V ψ † ψψ † ψ (11.

192

CHAPTER 11. FERMIONS

**Figure 11.1: The graphical representation of the fermion propagator and vertex. In the presence of source ﬁelds η, η , ¯ Z0 [η, η ] = N ¯ = e− Dψ Dψ e
**

dτ dτ ′ R † − Rβ

0

dτ

R

”” “ “ 2 ∇ ¯ dd x ψ† ∂τ − 2m −µ ψ+ηψ† +ψη

Rβ

0

dd x dd x′ η (x,τ )G(x−x′ ,τ −τ ′ )η(x′ ,τ ′ ) ¯

(11.43)

**For interacting fermions, it is straightforward to generalize (7.31) to Grassmann integrals, so that Z[η, η ] = ¯ N − R Lint e N0
**

“

δ , δ¯ δη δ η

”

Z0 [η, η ] ¯

(11.44)

In applying this formula, we must remember that η and η are Grassmann ¯ numbers so a − sign results every time they are anticommuted. As in the bosonic case, we can use (7.32) to rewrite this as: Z[η, η ] = e ¯

− Rβ

0

dτ dτ ′

R

dd x dd x′

δ δ G(x−x′ ,τ −τ ′ ) δψ δψ †

e−

R

Lint (ψ† ,ψ)+ηψ† +ψη ¯

(11.45) (11.46)

**By expanding the e−
**

R Lint (ψ† ,ψ)+ηψ† +ψη ¯

we derive the following Feynman rules for fermions with δ-function interactions. The lines of these Feynman diagrams have a direction which we denote by an arrow. Each vertex has two lines directed into it and two lines directed out of it. Momenta and Matrsubara frequencies are directed in the direction of the arrows. The propagator and vertex are shown in ﬁgure 11.1 • To each line, we associate a momentum, p and a Matsubara frequency, ǫn . • The propagator assigned to each internal line is:

11.4. FEYNMAN RULES FOR INTERACTING FERMIONS

193

−

1 β

n

d3 p 1 3 (2π) iǫ − k2 − µ n 2m

• For each vertex with momenta, Matsubara frequencies (p1 , ǫn1 ), (p2 , ǫn2 ) directed into the vertex and (p3 , ǫn3 ), (p4 , ǫn4 ) directed out of the vertex, we write V (2π)3 δ(p1 + p2 − p3 − p4 ) β δn1 +n2 ,n3 +n4 • Imagine labelling the vertices 1, 2, . . . , n. Vertex i will be connected to vertices j1 , . . . , jm (m ≤ 4) and to external momenta p1 , . . . , p4−m by directed lines. Consider a permutation of these labels. Such a permutation leaves the diagram invariant if, for all vertices i, i is still connected to vertices j1 , . . . , jm (m ≤ 4) and to external momenta p1 , . . . , p4−m by lines in the same direction. If S is the number of permutations which leave the diagram invariant, we assign a factor 1/S to the diagram. • If two vertices are connected by l lines in the same direction, we assign a factor 1/l! to the diagram. • To each closed loop, we assign a factor of −1. The ﬁnal rule follows from the necessity of performing an odd number of anticommutations in order to contract fermion ﬁelds around a closed loop. For µ < 0, (11.36) describes an insulating state. There is a gap to all excited states. For V = 0, the gap is simply −µ. In the problem set, you will compute the gap for V = 0 perturbatively. For µ > 0, the ground state has a Fermi surface. For V = 0, this √ Fermi surface is at kF = 2mµ. In the problem set, you will compute the Fermi momentum for V = 0 perturbatively. Since this phase has gapless excitations, we must worry whether the interaction term is relevant. If the interactions are irrelevant, then we can perturbatively compute corrections to free fermion physics. If interactions are relevant, however, the system ﬂows away from the free fermion ﬁxed point, and we must look for other ﬁxed points. Such an analysis is taken up in the next chapter, where we see the importance of the new feature that the low-energy excitations are not at k = 0, but, rather, at k = kF . We construct the renormalization group which is appropriate to such a situation, thereby arriving at Fermi

194

CHAPTER 11. FERMIONS

a.

b.

Figure 11.2: A one-loop diagram with an intermediate (a) particle-hole pair and (b) particle-particle pair.

11.5. FERMION SPECTRAL FUNCTION

195

liquid theory. First, however, we will investigate the two-point function of the interacting Fermi gas perturbatively. Fermion lines with arrows that point to the right represent fermions above the Fermi surface. Those which point to the left represent holes below the Fermi surface. This is analogous to electrons and positrons in QED. However, unlike in QED, where a positron can have any momentum, fermions must have k > kF and holes must have k < kF at T = 0 (at ﬁnite-temperature, this is smeared out by the Fermi function). Hence, the diagram of ﬁgure 11.2a corresponds to the expression dd q 1 (2π)d β G(iΩm + iǫn , p + q) G(iǫn , q) (11.47)

n

When q < kf , the second Green function represents the propagation of a hole at q while the ﬁrst Green function represents the propagation of a fermion at p + q. If p + q isn’t above the Fermi surface (smeared by the Fermi function), then this expression vanishes, as we will see shortly. Similarly, when q is above the Fermi surface, p + q must be a hole below the Fermi surface. Meanwhile, the diagram of ﬁgure 11.2b corresponds to the expression dd q 1 (2π)d β G(iΩm − iǫn , p − q) G(iǫn , q) (11.48)

n

where q and p − q are now both fermions above the Fermi surface.

11.5

Fermion Spectral Function

Following our earlier derivation of the phonon spectral representation, we construct a spectral representation for the fermion two-point Green function. By inserting a complete set of intermediate states, |m m|, we have, G(x, τ ) = d3 p dǫ[

n,m

**δ(p − pm + pn )δ(ǫ − ǫnm )(θ(τ )e−ip·x+ǫτ e−βEn ] −θ(−τ ))eip·x−ǫτ e−βEm ) m ψ † (0, 0) n (11.49)
**

2

**The Fourier transform, G(p, iǫj ) = d3 x
**

0 β

dτ G(x, τ ) e−iǫj τ

(11.50)

**196 is given by: G(p, iǫj ) = [ e−βEn + e−βEm
**

n,m

CHAPTER 11. FERMIONS

m ψ † (0, 0) n

2

× δ(p − pm + pn )δ(E − Em + En )] Writing A(p, E) =

n,m

1 E − iǫj

(11.51)

e−βEn + e−βEm

m ψ † (0, 0) n

2

δ(p−pm +pn ) δ(E−Emn ) (11.52)

**we have the spectral representation of G: G(p, iǫn ) =
**

∞ −∞

dE

A(p, E) E − iǫj

(11.53)

As usual, the spectral function A(p, E) is real and positive. It also satisﬁes the sum rule: ∞ dE A(p, E) = 1 (11.54) −∞ 2π G is not analytic since it does not satisfy the Kramers-Kronig relations. However, the advanced and retarded correlation functions, A(p, E) E − ǫ − iδ −∞ ∞ A(p, E) dE Gadv (p, ǫ) = E − ǫ + iδ −∞ Gret (p, ǫ) = dE

∞

(11.55)

are analytic functions of ǫ in the upper- and lower-half-planes, respectively. As usual, the spectral function is the diﬀerence between the retarded and advanced correlation functions. Gret (p, ǫ) − Gadv (p, ǫ) = 2πiA(p, ǫ) The spectral function of a free Fermi gas is a δ-function: A(p, ǫ) = δ ǫ − p2 −µ 2m (11.57) (11.56)

In an interacting Fermi gas, the spectral weight is not concentrated in a δ function but spread out over a range of frequencies as in ﬁgure 11.3.

11.6. FREQUENCY SUMS AND INTEGRALS FOR FERMIONS

197

Α(ε)

Α(ε)

p2/2m - µ

ε

p2/2m - µ

ε

Figure 11.3: The spectral function in free and interacting Fermi systems.

11.6

Frequency Sums and Integrals for Fermions

To compute fermion Green functions perturbatively, we will need to do summations over Matsubara frequencies. Sums over fermion Matsubara frequencies can be done using contour integrals, as in the bosonic case. Consider the Matsubara sum: dǫ nF (ǫ) G(iΩm + ǫ, p + q) G(ǫ, q) 2πi C n (11.58) where the contour avoids the singularties of the Green functions, as in chapter 6. ǫn and Ωm + ǫn are fermionic Matsubara frequencies, so Ωm is a bosonic one. The contour integration is given by two contributions: ǫ real and iΩm + ǫ real. Hence, 1 β G(iΩm + iǫn , p + q) G(iǫn , q) = 1 β G(iΩm + iǫn , p + q) G(iǫn , q) = + 1 2πi 1 2πi

∞ −∞

n

dE nF (E) G(E + iΩm ) (G(E + iδ) − G(E − iδ))

∞ −∞

dE nF (E − iΩm ) (G(E + iδ) − G(E − iδ)) G(E − iΩm )

**Analytically continuing the imaginary-time Green functions, we have: 1 β G(iΩm + iǫn , p + q) G(iǫn , q) =
**

∞ −∞

n

dE nF (E) G(E + iΩm , p + q) A(E, q)

198 +

∞ −∞

CHAPTER 11. FERMIONS dE nF (E) G(E − iΩm , q) A(E, p + q) (11.59)

In the case of free fermions, the spectral function is a δ-function, so the dE integrals can be done: 1 β G(iΩm + iǫn , p + q) G(iǫn , q) = nF (ξq ) − nF (ξp+q ) iΩm + ξq − ξp+q (11.60)

n

so

**At zero-temperature, the discrete frequency sum becomes a frequency integral, ∞ dǫ 1 → (11.61) β n −∞ 2π 1 β G(iΩm + iǫn , p + q) G(iǫn , q) → dǫ G(iΩ + iǫ, p + q) G(iǫ, q) (11.62) −∞ 2π
**

∞ ∞ −∞ ∞ −∞ ∞

n

**Using the spectral representation of G we can rewrite this as: dǫ G(iΩ + iǫ, p + q) G(iǫ, q) = −∞ 2π
**

∞

dǫ −∞ 2π

dE1

dE2

A(p + q, E1 ) A(q, E2 ) (11.63) E1 − iΩ − iǫ E2 − iǫ

The dǫ integral can be done by closing the contour in the upper-half-plane. The pole at ǫ = −iE1 − Ω is enclosed by the contour when E1 < 0; the pole at ǫ = −iE2 is enclosed when E2 < 0. Hence, dǫ G(iΩ + iǫ, p + q) G(iǫ, q) = 2π −∞

∞ ∞ ∞ −∞

dE1

∞ −∞

dE2

θ(−E2 ) − θ(−E1 ) A(p + q, E1 ) A(q, E2 ) (11.64) E2 − E1 + iω

In the case of free fermions, the dEi integrals may be done: dǫ θ(−ξq ) − θ(−ξp+q ) G(iΩ + iǫ, p + q) G(iǫ, q) = iΩ + ξq − ξp+q −∞ 2π which is the zero-temperature limit of (11.60). (11.65)

11.7

Fermion Self-Energy

We can begin to understand the role played by the Fermi surface when we start computing perturbative corrections to the behavior of free fermions. Let us look ﬁrst at the fermion two-point Green function. As in the bosonic

11.7. FERMION SELF-ENERGY

199

case, we can deﬁne the self-energy, Σ(ǫ, k), as the 1PI two-point function and sum the geometric series to obtain: G(p, iǫn ) = 1 iǫn −

p2 2m

(11.66)

− µ − Σ(ǫn , p)

**The retarded Green function is deﬁned by analytic continuation: Gret (p, ǫ) = 1 ǫ−
**

p2 2m

(11.67)

− µ − Σret (ǫ, p)

**The spectral function can be written as: A(p, ǫ) = 1 π −2ImΣret (ǫ, p) − µ − ReΣret (ǫ, p)
**

2

ǫ−

p2 2m

+ (ImΣret (ǫ, p))

2

(11.68)

When ImΣret (ǫ, p) = 0, the spectral function can be rewritten as: A(p, ǫ) = Z(p) δ (ǫ − ξp ) p2 − µ − ReΣret (ξp , p) 2m (11.69)

**where ξp is the location of the pole, deﬁned by the implicit equation ξp = and Z(p) is its residue Z(p) = 1 − ∂ ReΣret (ǫ, p) ∂ǫ
**

−1

(11.70)

(11.71)

ǫ=ξp

**ξp can be expanded about the Fermi surface:
**

∗ ξp = vF (p − pF ) + O (p − pF )2

(11.72)

where

pF (11.73) m∗ ∗ and m∗ is the eﬀective mass. ξp and vF deﬁne the one-particle density of states, N (ǫF ), of the interacting problem.

∗ vF =

dd k 2π 2 = (2π)d (2π)d Γ

d

d 2

kd−1 dk

200 2π 2 ≈ (2π)d Γ = 2π 2 (2π)d Γ

d d

CHAPTER 11. FERMIONS

d 2 d 2

d−1 kF d−1 kF ∗ vF

dk dξk (11.74)

≡ N (ǫF )

dξk

The lowest-order contribution to ImΣret (ǫ, p) comes from the diagram of ﬁgure ??. We can do the zero-temperature calculation by contour integration: Σ(iǫ, k) = V 2 dζ G(iζ, p) G(iζ+iω, p+q) G(iǫ−iω, k−q) 2π (11.75) The dζ integral may be done by contour integration, as in (11.65): Σ(iǫ, k) = V 2 dd q (2π)d dd p (2π)d dd q (2π)d dd p (2π)d dω 2π

dω θ(−ξp ) − θ(−ξp+q ) G(iǫ − iω, k − q) 2π iω + ξp − ξp+q (11.76) The dω integral may be done the same way: Σ(iǫ, k) = V 2 dd p (θ(−ξp ) − θ(−ξp+q )) (θ(ξk−q ) − θ(ξp − ξp+q )) (2π)d iω + ξp − ξp+q − ξk−q (11.77) Hence, the imaginary part of the self-energy at zero-temperature is: dd q (2π)d

2(d−1) ∞ 0 dΩp 2 kF dξk−q dξp 2 (2π)d vF 0 −∞ 2(d−1) ǫ 0 2 2 kF Sd−1 dξp dξk−q 2 vF 0 −ǫ 2(d−1) 2 kF 2 Sd−1 ǫ2 2 vF

ImΣret (ǫ, k) = V ≤V =V

dΩk−q δ (ξp − ξp+q − ξk−q + ǫ) (2π)d

(11.78)

Hence, we have seen that the phase space restrictions (imposed by the δ function above) due to the existence of a Fermi surface severely restricts ImΣret (ǫ, k). For ǫ → 0, ImΣret (ǫ, k) ∼ ǫ2 . In other words, for ǫ small, the decay rate is much smaller than the energy: near the Fermi surface, single-fermion states are long-lived.

concept.82) .79) In the second line. ǫ) = −i (2π)d 2π (11. the advanced and retarded Green functions vanish at equal times. Inside the Fermi surface. If we write G(k. ǫ) ∂ k2 −µ ǫ− ∂ǫ 2m ∂ G(k. ǫ) = ǫ− then N = −i V dd k dǫ (2π)d 2π dd k dǫ = −i (2π)d 2π dd k dǫ = −i (2π)d 2π dd k dǫ =i (2π)d 2π G(k. k) is positive. we have the time-ordered Green function. ǫ) Σ(ǫ. we begin with N = V dd k ψ † (k. k) diverges at the Fermi surface. which we now discuss. However. k) = 0.) According to Luttinger’s theorem. G(0. G(0. the Fermi surface has essentially been a tree-level.80) − µ − Σ(ǫ. G(0. k) (11. N/V . k) changes sign. ǫ) Σ(ǫ. k) ∂ǫ ∂ǫ G(k. In a free fermion system. LUTTINGER’S THEOREM 201 11. ǫ) G−1 (k. Luttinger’s theorem deﬁnes the Fermi surface as the surface in k-space at which G(0.11. ǫ) + Σ(ǫ. k) is negative. k) = 0 d 2π (2π) ∂ǫ (11. the notion of a Fermi surface is not tied to perturbation theory.8. the volume enclosed by the Fermi surface is equal to the electron density. In fact. ǫ) − G(k. k) ∂ǫ ∂ ∂ ln G(k. t)ψ(k. as we will see later. G(0. so long as ImΣ(0. ǫ) 1 k2 2m (11. or free fermion. t) (2π)d dd k dǫ G(k. outside the Fermi surface.81) We will now use the following ‘lemma’ which we will prove later: ∂ dd k dǫ G(k. (In a superconductor.8 Luttinger’s Theorem Up until now. the existence and location of a Fermi surface is constrained by a non-perturbative theorem due to Luttinger. To prove this. k) vanishes at the Fermi surface.

85) and the normalization property of the spectral function. ǫ) ln (11. the volume enclosed by the Fermi surface is equal to the electron density. we must prove that ∂ dd k dǫ G(k. ǫ) ∂ǫ ∂ ln Gret (k. N =− V dd k [ϕ(0.88) . ǫ) = ∞ −∞ dE A(p. ǫ) 0 ln (2π)d Gret (k.86) Since ImΣ(0. G(p. ǫ) Since Gret is analytic in the upper-half-plane. ǫ) + i ∂ǫ CHAPTER 11. To complete the proof of the theorem. The integral only receives contributions from the former case: N = V dd k d R (2π) (11. Hence. N =i V =i =− 0 dd k dǫ ∂ G(k. while G∗ = Gret for ǫ < 0. k)] (2π)d (11.89) (11. we see that ϕ(−∞. ϕ(0. Also note that G = Gret for ǫ > 0. k) is equal to 0 or π. k) (2π)3 (11. k) = 0 d 2π (2π) ∂ǫ To do this. E) ǫ − E + iδ sgn(ǫ) (11. ǫ) ln d (2π) −∞ 2π ∂ǫ Gret (k. the ﬁrst integral vanishes. k) > 0}. k) δG(ω. In other words. k) = 0 by assumption. k) = π.87) where R = {k|G(0. ǫ) Σ(ǫ. ǫ) dd k G(k.83) d 2π ∂ǫ (2π) Gret (k. FERMIONS dd k dǫ ∂ G(k. ǫ) −∞ dd k [ϕ(0. k) − ϕ(−∞. k) − π] (2π)d (11.84) From the spectral representation. Hence. we prove that there exists a functional X[G] deﬁned by: δX = dω 2π d3 k Σ(ω.202 Then N =i V =i dd k (2π)d dd k (2π)d dǫ 2π dǫ 2π ∂ ln G(k.

so δX/δǫ = 0 which proves (11. k) δG(ω + ǫ. To see that X actually exists. δ2 X δΣ(p) = δG(p)δG(q) δG(q) X exists if and only if the derivatives can be commuted: δ2 X δ2 X = δG(p)δG(q) δG(q)δG(p) Since δΣ(p) = Γ(p. k) is ﬁxed by the density to be at kF . k) δG(ω.90) (11. nk = k θ(kF − k) with nk = N/V . One might imagine that. q and Γ(p.91) However. LUTTINGER’S THEOREM According to this deﬁnition. q) = Γ(q. Luttinger’s theorem tells us that this cannot happen.11. k) δG(ω. . p.93) (11.8. it would be possible to have nk = λθ(kF − kλ) with λ < 1 which would preserve nk = N/V while moving the location of the singularity in nk to kF /λ. p).95) (11. the existence of X follows.92) where Γ(p.88). X is independent of ǫ. q. k) (2π)3 d3 k Σ(ω + ǫ. k) δG(ω + ǫ. nk = c† ck is a step function. k) (2π)3 203 (11. δX = = Hence. k) 3 (2π) ∂ǫ ∂ d3 k Σ(ω.94) (11. observe that δX = Σ(p) δG(p) Hence. in an interacting Fermi gas. q) δG(q) (11. q) is the irreducible 4-point function with external momenta p. In the case of a free Fermi gas. δX = δǫ = dω 2π dω 2π ∂ d3 k Σ(ω + ǫ. k) (2π)3 ∂ω dω 2π dω 2π d3 k Σ(ω. The singularity in nk = dǫG(ǫ.

FERMIONS .204 CHAPTER 11.

First.e. The Fermi sea is ﬁlled up to some kF . . . i. . the vicinity of the Fermi surface. Polchinski.1) If the kinetic energy is much larger than the potential energy. . The RG analysis of such systems was pioneered by Shankar. Since we will be interested in low-energy. then it makes sense to focus ﬁrst on it and use it to determine the scaling of ψ from the kinetic energy. we make the approximation k2 − µ ≈ vF (k − kF ) 2m ≡ vF l 205 (12.CHAPTER 12 Interacting Neutral Fermions: Fermi Liquid Theory 12. in momentum space. let us examine the free part of the action.2) ∇2 −µ 2m ψ (12. dτ dd x ψ † ∂τ − or.3) . dǫ dd k † ψ iǫ − 2π (2π)d k2 −µ 2m ψ (12.1 Scaling to the Fermi Surface We now consider a rotationally invariant system of interacting spinless fermions with µ > 0 in D ≥ 2.

Ω. • Rescale: ω → bω l → bl Ω→Ω 3 ψ → b− 2 ψ (12. ǫ) for bΛ < |l| < Λ and ǫ. Ω. To this action. k = 0. as we would expect.8) . we do not scale to the origin of momentum space. The angular integral has no cutoﬀ.4) Hence. Hence. ψ † (l.5) Restoring the cutoﬀs. Thus. Ω arbitrary.7) The principal diﬀerence between the renormalization group applied to a system with a Fermi surface and its application to more familiar contexts is that the low-energy degrees of freedom are located in the neighborhood of a surface in momentum space. we can add the following perturbation: d−1 kF (2π)d Λ −Λ dl dd Ω dǫ δµ ψ † ψ −∞ 2π ∞ (12. of course.206 CHAPTER 12. d−1 kF (2π)d Λ −Λ dl dd Ω dǫ † ψ (iǫ − vF l) ψ 2π −∞ ∞ (12. Our RG transformation now takes the following form: • Integrate out ψ(l. but to the Fermi surface. We leave the frequency integral unrestricted (as we did in the case of the O(3) NLσM). l = 0. INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY where l = k − kF .5) is evidently a ﬁxed point of this RG transformation. ǫ). We can also make the approximation d−1 dd k = kF dk dd Ω d−1 = kF dl dd Ω (12.6) The momentum integral is restricted to a shell of thickness 2Λ about the Fermi surface. The free fermion action (12. rather than in the vicinity of a point. a free fermion system looks the same at any energy scale: it is always just a free theory. the action can be written as: d−1 kF (2π)d dl dd Ω dǫ † ψ (iǫ − vF l) ψ 2π (12.

which is a generalization of a coupling constant. k4 ) = u(k1 . If it depends on the |ki | − kF |’s. k2 . then it is irrelevant. k2 . k3 ) δ k1 + k2 − k3 − k4 (12. then we recover (12.5). δµ scales as: δµ → b−1 δµ 207 (12. Such a perturbation can change the shape of the Fermi surface.2. and δµ is no diﬀerent. δµ Ω . etc. i. MARGINAL PERTURBATIONS: LANDAU PARAMETERS Under the scaling (12. If u(k1 . k3 . ω4 )ψ † (k3 . k1 ) is a function only of the angular variables. The RG equations for a coupling function are called functional RG equations. then it is marginal.2 Marginal Perturbations: Landau Parameters Let us now consider four-fermion interactions. Ω3 . In a system which is not rotationally invariant. so we ignore this possibility. Similarly. δµ Ω is an example of a ‘coupling function’.11) Rather than a single coupling constant. It is clearly a marginal perturbation. Changing the chemical potential shifts the Fermi δµ surface. k3 . u. that the fermions have short-ranged interactions. A second perturbation is d−1 kF (2π)d Λ −Λ dl dd Ω dǫ δvF l ψ † ψ 2π −∞ ∞ (12. We will deal with the complications resulting from Coulomb interactions in the next section. 12. Consider the term S4 = dω1 dω2 dω3 dd k1 dd k2 dd k3 u(k1 .e. ω2 )ψ(k1 .9) Since δµ is a relevant operator. Relevant operators typically bring about a fundamental change of the ground state. Ω4 ). If we change coordinates to l′ = l + vF . k4 ). momentum conservation implies that u(k1 . k4 ) is non-singular or. it does not scale under (12. Ω2 . We will assume that u(k1 . in other words. k3 . six-fermion. as in 3 He.7).12. k3 . k2 . Furthermore. we cannot study it perturbatively. k4 )× ψ † (k4 . we have a coupling function. u(k1 . eight-fermion. k2 . u(Ω1 .12) . k2 .10) δvF shifts the Fermi velocity. k2 . ω3 )ψ(k2 . k2 . interactions are neglected because they are highly irrelevant. ω1 ) (12.7). δµ can depend on the angle around the Fermi surface.

b. Figure 12. k2 can only scatter into k3 = k1 . . unless (b) k1 = −k2 .1: (a) If all of the momenta are constrained to lie on the Fermi surface.208 CHAPTER 12. INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY k1 + k 2 k3 k 2 k1 k4 k 2 k1 a. incoming momenta k1 . k4 = k1 . k4 = k2 or k3 = k2 .

17) . k3 ). Ω2 ) = −F (θ1 − θ2 ) In three dimensions. ﬁnally. k2 ) and exchange scattering. Ω2 . At small but non-zero Λ. For Λ ≪ kF . then k3 = −k4 is arbitrary. Ω3 . we considered the case of δ function interactions. depicted in ﬁgure 12. k2 . As Λ is decreased. This is the Cooper pairing channel. The Fourier components of F are called Landau Parameters. we should focus on: F (Ω1 · Ω2 .2.15) in the Λ → 0 limit. for simplicity. k2 . the angle between k3 and k4 is the same as the angle between k1 and k2 θ(k1 . φ) ≡ u(Ω1 . Ω2 . Ω2 . There is one loophole in the preceeding analysis. k3 because k4 typically does not lie within the cutoﬀ. Here.1b. Ω4 ) (12. k2 . It is this drastic simpliﬁcation which makes Fermi liquid theory soluble. In three dimensions. some of these are set discontinuously to zero. k4 ) (12.2. k3 . the case of D = 2. fewer non-zero u’s remain until. The crucial observation underlying Fermi liquid theory. at Λ = 0. only the three mentioned above remain. As we scale to the Λ → 0 limit. the rest do not scale. As Λ becomes smaller. −k1 . Consider. Ω1 ) (12. The constraint of momentum conservation.1. only forward scattering. a small subset of the u’s are non-zero. k1 . k3 ) = V . u(k1 . k2 . k2 ). MARGINAL PERTURBATIONS: LANDAU PARAMETERS 209 In the last section. k3 .16) (12. k2 .13) but the plane of k3 and k4 can be rotated relative to the plane of k1 and k2 by and angle φ as in ﬁgure 12. we are considering the more general case of arbitrary (non-singular) u(k1 . k3 ) = 0 for generic k1 . u(k1 . is the following. can satisfy momntum conservation. k2 . Ω2 . Ω1 . We write: V (Ω1 · Ω3 ) ≡ u(k1 . k2 . These phase space restrictions imply that in two dimensions.12. k2 . k1 ) = −u(k1 . k2 ) = ±θ(k3 . u(k1 . Ω2 . If k1 = −k2 .14) Fermi statistics dictates that exchange scattering is related to forward scattering by: u(Ω1 . k4 lie within Λ of the Fermi surface severely limits the phase space for scattering. k1 . −k3 ) (12. we should focus on F (θ1 − θ2 ) ≡ u(Ω1 . k1 + k2 = k3 + k4 together with the restriction that k1 . Ω1 ) = −u(Ω1 . k2 . which is depicted in Figure 12. for which u(k1 . Ω2 .

ǫ3 )ψ(k2 . we have the following action: S= dl dd Ω dǫ † ψ (iǫ − vF l) ψ 2π + dǫ1 dǫ2 dǫ3 dk1 dk2 dk3 dd Ω1 dd Ω2 F (Ω1 · Ω2 .2: In three dimensions. ǫ3 )ψ(k2 .18) Then. Ω4 ) with Λ Ω1 − Ω3 < (12. at tree-level. ǫ2 )ψ(k1 . in the Λ → 0 limit. ǫ1 ) dǫ1 dǫ2 dǫ3 dk1 dk2 dk3 dd Ω1 dd Ω3 V (Ω1 · Ω3 )× + ψ † (k4 .210 CHAPTER 12. ǫ1 ) (12.19) For Λ ﬁnite. Ω2 . Ω3 . φ)× ψ † (k4 . this can be written as: V (θ1 − θ3 ) d−1 kF (2π)d (12. In D = 2.20) kF or Λ Ω2 − Ω3 < (12.21) kF . INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY k1 k3 k4 k2 φ Figure 12. we have to keep the full coupling function u(Ω1 . ǫ4 )ψ † (k3 . ǫ2 )ψ(k1 . ǫ4 )ψ † (k3 . the outgoing momenta can be rotated relative to the incoming momenta.

3 12. consider the renormalization of F . The kinematic restriction .3 One-Loop At tree-level. First.3. The condition on |p| restricts its magnitude. The one-loop diagrams are in ﬁgure 12.4. The ﬁrst diagram gives a contribution dF (Ω1 − Ω2 ) = dl dΩ dǫ F (Ω)F (Ω + Ω1 − Ω2 ) 1 1 (12. ONE-LOOP 211 k1 k2 k3 k4 Figure 12. as in ﬁgure 12. The internal momenta in these diagrams must lie in thin shells at the cutoﬀ. the condition on |p+k1 −k2 | restricts the direction of p.12. we can set the external frequencies to zero and put the external momenta on the Fermi surface. Λ − dΛ < |p| − kF < Λ. In the second diagram.3: The phase space available to k3 and k4 when Λ is small but non-zero is the region bounded by the arcs which has area ∼ Λ2 . We would now like to compute the one-loop RG equations for F (Ω1 − Ω2 ) and V (Ω1 − Ω3 ). p and p + k1 − k2 must both satisfy this condition.22) iǫ − vF l iǫ − vF l which vanishes since both poles in ǫ are on the same side of the axis. F (Ω1 − Ω2 ) and V (Ω1 − Ω3 ) are marginal. Since F is independent of the frequencies and the li .

at one-loop.4: The one-loop diagrams which can contribute to the renormalization of F . is essentially the same as that depicted in ﬁgure 12.3. INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY k1 k3 k1 b. p+Q p k3 a. We now turn to the one-loop RG equations for V . The relevant diagrams are analogous to those of 12. this gives a vanishing contribution to dF/dℓ. d F (Ω1 − Ω2 ) = 0 (12.4. The third diagram gives a vanishing contribution for the same reason. k1 p k3 k2 Q-p k4 Figure 12. As a result. Hence. p p k2 k4 k2 k4 c. therefore. do not contribute to the RG equation. The ﬁrst two diagrams are proportional to (dΛ)2 and. the third diagram gives the contribution dV (Ω1 − Ω3 ) = − 1 1 dǫ dl dd Ω V (Ω1 − Ω)V (Ω − Ω3 ) d−1 2π 2π (2π) iǫ − vF l −iǫ − vF l . However.24) dℓ The Landau parameters are strictly marginal.23) in the dΛ → 0 limit. they remain constant as we scale to lower energies. and therefore dF ∼ (dΛ)2 (12.212 CHAPTER 12. the dl and dΩ integrals each give a contribution proportional to dΛ.

There are no relevant interactions other than F (Ω1 .28) Therefore. Ω3 ). BCS theory and its reﬁnement by Nambu. A more general formalism. Gor’kov.30) Bardeen.3. These theories will be discussed in chapter 16. until it reaches the scale: Λ ∼ Λ0 e−2πvF /|Vl (Λ0 )| (12. and Eliashberg. Ω2 ) and V (Ω1 . ONE-LOOP 1 dl dd Ω V (Ω1 − Ω)V (Ω − Ω3 ) d−1 2π (2π) 2vF p dd Ω 1 dℓ V (Ω1 − Ω)V (Ω − Ω3 ) = 2πvF (2π)d−1 213 = (12. Gor’kov.28). In the BCS theory of a phonon-mediated superconductor. which can be used when the interactions are retarded was pioneered by Nambu. Both of these approaches owe their success to the kinematic constraints of the problem. and Schrieﬀer found the paired ground state using a variational ansatz. so the diagram of Figure ?? is essentially the only diagram which must be taken into account. F (Ω1 . we write 2π Vm = 0 dθ imθ e V (θ) 2π (12. Ω2 ) does not contribute to the running of V (Ω1 . Cooper.12. this leads to a critical temperature or zero-temperature gap given by Tc ∼ ∆(T = 0) ∼ ωD e−2πvF /|V0 | (12. while attractive BCS interactions are marginally relevant. Ω3 ). we see that an attractive BCS interaction will grow as we go to lower scales. . pairing takes place. and Eliashberg are mean-ﬁeld theories which evaluate this diagram selfconsistently. at this scale.26) The renormalization group ﬂow equation for Vl is: 1 dVl =− V2 d ln Λ 2πvF l Vl (Λ) = Vl (Λ0 ) Vl (Λ0 ) ln (Λ0 /Λ) (12. From (12.25) In two dimensions.27) 1+ 1 2πvF (12. repulsive BCS interactions are marginally irrelevant.29) As we will discuss later.

To see this. consider the case D = 2.214 CHAPTER 12. to scattering from one point to another within the same patch. ǫ1 ) δ k1 + k2 − k3 − k4 2Λ We have broken the angular integral into a summation over Fermi surface ‘patches’ and an integral over each patch. Hence. we can write S= i d2 k dǫ † ψ (iǫ − vF l) ψi (2π)2 2π i dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π † Vij ψj (k3 . ǫ2 ) ψi (k4 . ǫ1 ) δ k1 + k2 − k3 − k4 + i. ǫ2 )ψi+ kF (k1 . the δ-function. kF /Λ. ǫ3 )ψ † F j+ 2Λ k + i.31) Then. therefore. ∆θ = Λ/kF with θj = 2πj(Λ/kF ) and j = 0.e. 1. the ki ’s live within patches and. ǫ4 )ψi (k1 . we have automatically restricted to nearly forward scattering – i.4 1/N and All Loops The one-loop structure of a system of interacting fermions is actually stable to all orders in perturbation theory.33) does not contain any momenta of O(kF ). The essential reason for this (which was ﬁrst recognized in this language by Shankar) is that Λ/kF is a small parameter like 1/N . 2π 0 θi+1 dθ → dθ i θi (12. By restricting to Fij rather than allowing uijkl .34) . . δ k1 + k2 − k3 − k4 (12. ǫ3 )ψj (k2 . INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY 12.j (12. Furthermore. . . d2 k = = −Λ Λ 2πΛ/kf dl −Λ Λ 0 πΛ dθ dk −πΛ dk⊥ (12. ǫ4 ) ψi (k2 . are all less than the cutoﬀ. . Break the angular integration into pieces.j † † Fij ψj (k3 .32) (k4 . In this expression. F (θi −θj ) has been replaced by Fij .

j=0 dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π Λ † (12. ǫ1 ) δ k1 + k2 − k3 − k4 N dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π + i. Hence. The only O(1) corrections to the two-point function are diagrams of the form of ﬁgure . Recall the analysis of the O(N ) model in the large N limit. ǫ1 ) δ k1 + k2 − k3 − k4 (12.12. ǫ3 )ψj (k2 . ǫ4 ) ψi (k2 .j=0 kF /Λ + i.35) we can rewrite the action as: kF /Λ S= i=0 kF /Λ d2 k dǫ † ψ (iǫ − vF l) ψi (2π)2 2π i dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π Λ † † Fij ψj (k3 .j=0 1 † Vij ψj (k3 . ǫ3 )ψj (k2 . ǫ2 )ψi+ kF (k1 .4. 1/N AND ALL LOOPS 215 so both momenta have cutoﬀ ∼ Λ. ǫ1 ) δ k1 + k2 − k3 − k4 j+ 2Λ kF 2Λ In other words. as well: ω ω→ Λ k⊥ k⊥ → Λ k k⊥ → Λ ψ → Λ2 ψ (12. ǫ3 )ψ † kF (k4 .36) Vij ψj (k3 .37) j+ 2Λ N 2Λ then we see that we have a model in the large N limit. N S= i=0 N d2 k dǫ † ψ (iǫ − vF l) ψi (2π)2 2π i dǫ1 dǫ2 dǫ3 d2 k1 d2 k2 d2 k3 d2 k4 × 2π 2π 2π 2π 2π 2π 2π 1 † † Fij ψj (k3 . ǫ4 )ψi (k1 . ǫ4 ) ψi (k2 . ǫ2 ) ψi (k4 . ǫ2 ) ψi (k4 . if we write N = kF /Λ. ǫ1 ) δ k1 + k2 − k3 − k4 kF + i. ǫ2 )ψi+ kF (k1 . if we rescale all momenta by Λ and the ﬁeld. ǫ3 )ψ † kF (k4 . ǫ4 )ψi (k1 . ψ.j=0 N + i.

INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY ??a. using the scaling ky → sky . Suppose Λ is small but ﬁnite. Consider now the correction to the four-point function. The non-trivial diagrams such as ??b are O(1/N ). ǫ(kx . The only diagrams which need to be considered are the bubble diagrams. the scaling of a physical quantity which depends on the u’s is determined not by the scaling of the u’s.37) is a stable ﬁxed point if Vi j > 0. we focus on these points. kx → s1/2 kx . but on the number of non-zero u’s. These shift the chemical potential. ??b. Ω2 .38) We now scale to Ω1 . in two dimensions. In the case of forward scattering. it’s a less natural renormalization group transformation because it involves selecting preferred points on the Fermi surface and scaling diﬀerently at diﬀerent points on the Fermi surface. k4 with |k1 − k3 | < Λ. or Cooper scattering – which do scale continuously). In the vicinity of these points. The other corrections.5 Quartic Interactions for Λ Finite The scaling of generic four-fermi interactions is quite awkward for calculations at a ﬁnite frequency or temperature scale because the u’s don’t scale continuously. such as those of ??c are O(1/N 2 ). kx (tangent to the Fermi surface) and ky (perpendicular to the Fermi surface). Consequently. which is marginal. Then. at these two points on the Fermi surface. |k2 − k4 | < Λ. Only diagrams such as those of ??a. k2 to k3 . We construct cartesian coordinates.216 CHAPTER 12. Ω2 . the action (12.) The same answers are obtained with either scaling transformation. As we learned earlier in the context of O(N ) models. a diﬀerent scaling transformation is useful. in d > 2. For such calculations. Thus. gives a non-trivial contribution. (We have assumed d = 2. ω → sω. we can consider nearly forward scatering. it’s just that some calculations are easier with this one. ky ) = vF ky + 2 kx 2kF (12. exchange. the large-N limit introduces the following simpliﬁcations. In the case of Cooper scattering. On the other hand. which is scale dependent (except in the important case where the quantity is determined by forward. A system of fermions which is controlled by this ﬁxed point is called a Fermi liquid. there are d − 1 momenta which scale as kx . The one-loop RG is the full story. Let’s brieﬂy see how this . ??a vanishes because both poles are on the same side of the axis.b are O(1/N ). from k1 . Since all of the action is taking place in the neighborhods of Ω1 . 12.

5. Eﬀective Mass As a result of the preceeding analysis. If we insert a δ(k1x − k3x ) or δ(k1x − k4x ) into the integrand. 12.The scaling is perfectly continuous.5 To see why this is a useful scaling. If k1 . Compressibility. k4 = k1 +k2 −k3 lie within the cutoﬀ Λ. namely forward scattering. as before.39) Hence. a result which is more cumbersome to derive with the other RG transformation. It has a real part. in agreement with the explicit calculation which we did in chapter 13. proportional to Ff ω which comes from the marginal 2 forward scattering interaction. Other diagrams are down by powers of 1/N .6 Zero Sound. COMPRESSIBILITY. S4 = dω1 dω2 dω3 d2 k1 d2 k2 d2 k3 u(k1 . in which F is a constant. so four-fermi interactions. the ﬁeld now scales as ψ → s−7/4 ψ. and an imaginary part. ω1 ) (12. ω3 )ψ(k2 . proportional to Fnf ω 2 coming from irrelevant non-forward processes. EFFECTIVE MASS 217 Figure 12. k3 ) ψ † (k4 . ω2 )ψ(k1 . works. . k2 . The above scaling immediately yields 2 2 the suppression of Fnf with respect to Ff by one power of ω. the density-density correlation function can be computed by summing bubble diagrams. then we get a marginal interaction. The approximation which consists of neglecting these other diagrams is called the RPA.5: A one-loop self-energy correction. The bubble diagrams form a geometric series which may be summed.12. k2 . k3 . ZERO SOUND.6. ω4 )ψ † (k3 . then they continue to do so under this renormalization group transformation. consider the diagram in ﬁgure 2 12. Let us consider the simplest case. The quadratic part of the Lagrangian is of the form: S0 = dω dky dkx {ψ † iω − vF ky + 2 kx 2kF ψ} (12.40) scale as s1/2 . or random-phase approximation (for historical reasons).

6: The geometric series of bubble diagrams which determine ρ(q. iω))2 F0 + . −iω) to O(1). If we imagine changing the chemical potential by a frequency. . −iω) = I(q. k + q) 2π (2π)3 1 1 dǫ dl dϕ 2 d(cos θ) = kF 3 2π (2π) iǫ − vF l iǫ + iω − vF l − vF q cos θ dl θ(l) − θ(l + q cos θ) 2 = kF d(cos θ) (2π)2 iω − vF q cos θ 2 kF q cos θ d(cos θ) = (2π)2 iω − vF q cos θ 1 2 kF x = 2 dx iω 4π vF −1 vF q − x 2 kF iω + vF q 1 iω = 2 ln −1 (12.42) 2π vF 2 vF q iω − vF q The retarded density-density correlation function is a response function. k) G(iǫ + iω. iω)F0 where I is the value of a single particle-hole bubble. . I(q.41) = 1 − I(q. iω) ρ(−q. iω))n+1 F0 + . iω) + (I(q. iω) ρ(−q.. iω) (12. INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY + + + . this is: I(q. In the limit of q ≪ kF . ..218 CHAPTER 12. iω) = dǫ d3 k G(iǫ. + (I(q.and wavevector-dependent amount. of the type which we discussed in chapter 7. −χρρ . . . Figure 12. F (Ω1 · Ω2 ) = F0 : n ρ(q.

44) Since it reﬂects the density change resulting from a variation of the chemical potential. following the steps of chapter 7.7 shows the allowed phase space for a particle-hole pair.41). In other words.43) Hence.41). q) = χρρ (ω. but the interactions which would allow such decay are six-fermion and higher interactions which are highly irrelevant. it cannot decay. Since the zero-sound mode lies outside this continuum for q small. i. According to (12. As usual. ω) ρρ χρρ (q. 0). we have δρ(ω. it is called the compressibility. EFFECTIVE MASS δµ(ω. q) (12. Let us conρρ sider the static compressibility. svF . From (12.47) where χ0 is the compressibility in the absence of interactions.e.46) F0 2π vF 2 s−1 where s = ω/vF q The solution of this equation occurs for s > 1.45) F0 2π vF 2 vF q ω − vF q or. χ0 (q. there is a mode with ω = svF q. The continuum of states composed of a particle-hole pair lies beneath the line ω = vF q for q small. 2 kF 1 s+1 1 = 2 s ln −1 (12. According to (12. q) δµ(ω. ω) ρρ (12. called zero sound. ω) = 1 + F0 χ0 (q. Figure 12. then the action changes by S→S− dω dd q δµ(ω. this explains why it is a propagating mode. ω) on the real axis is a propagating mode.6. χ0 (q → 0. Energy and momentum conservation do allow it to decay into multiple particle-hole pairs. χρρ (q → 0. COMPRESSIBILITY.42). 0) = ρρ 2 kF 2π 2 vF m∗ kF = 2π 2 (12. there is such a pole when: 2 kF ω + vF q 1 ω 1 ln = 2 −1 (12. which is greater than the Fermi velocity. ZERO SOUND. this mode. for both the interacting and non-interacting systems. has a velocity of propagation.12. q). q) 2π (2π)d 219 (12. q) ρ(ω. a pole of χρρ (q.48) .

INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY ω ω = sv q F ω= v q F q Figure 12. H= dd x ψ† ∇2 −µ 2m ψ + ψ † (x)ψ(x) V (x − x′ ) ψ † (x′ )ψ(x′ ) (12. χρρ (q → 0.51) so the energy transforms as: δE = P · δp/m (12.49) The compressibility is decreased by interactions if we assume that m∗ is the same in both the interacting and non-interacting systems. S= dτ dd x ψ † ∂τ − ∇2 −µ 2m ψ + ψ † (x)ψ(x) V (x − x′ ) ψ † (x′ )ψ(x′ ) (12. this is usually not the case. However. 0) ρρ 1 + 2π2 vF F0 (12.52) . Hence. where m∗ ≡ kF /vF .50) Similarly for the Hamiltonian. 0) 1 = 2 kF χ0 (q → 0.7: The allowed ω.220 CHAPTER 12. Consider the behavior of a Fermi liquid under a Galilean boost by δv = δp/m. q values of the the zero sound mode and the continuum of particle-hole excitations. but the potential energy is invariant. The kinetic term in the action transforms.

If we consider a state with a ﬁlled Fermi sea – which has momentum zero – and a quasiparticle at the Fermi energy. Hence. but it does change ψi ψi by δp cos θi .12. ZERO SOUND.53) On the other hand. . 0) = 2 free (q → 0. comparing these two expressions and using m∗ = kF /vF .56) a ratio which depends on the relative strengths of F0 and F1 . and we boost the system in the direction of the quasiparticle’s momentum. the energy shift of this state is also δE = vF δp + δp d3 Ω F (θ.55) Consequently. The boost shifts the quasiparticle momentum by δp and also moves the Fermi sea by this amount. COMPRESSIBILITY. then δE = kF δp/m (12. φ) cos θ (2π)3 (12. the ratio of the interacting and free compressibilities is: 1 1 + 3 2πF1 F 2v χρρ (q → 0. This doesn’t aﬀect its momentum to lowest † order in δp. we have 1 m∗ =1+ m 3 F1 2π 2 vF (12.6.54) 1 = vF δp + δp F1 3 Hence. EFFECTIVE MASS 221 where P is the total momentum. we can compute the energy change using Fermi liquid theory. 0) k χρρ 1 + 2π2F F F0 v (12.

222 CHAPTER 12. INTERACTING NEUTRAL FERMIONS: FERMI LIQUID THEORY .

CHAPTER 13 Electrons and Coulomb Interactions 13.) rs EKinetic (13. we expect the electrons to form a Wigner crystal. later in this chapter.1) where rs is ratio of the interparticle spacing to the eﬀective Bohr radius in the metal 1 3 3 a−1 (13.1 Ground State Thus far. Stated diﬀerently. This assumption is appropriate for 3 He. we have assumed that there are only short-range interactions between the fermions in our system. we will show that this liquid is a Fermi liquid. on the other hand. If. charge e is enclosed within a sphere of radius rs a0 .2) rs = 0 4πn and a0 = 1/mκe2 is the eﬀective Bohr radius in the metal. V (r) = κe2 /r. we expect the electrons to form some kind of liquid state. A naive comparison of these energies estimates the kinetic energy by the kinetic energy of a free Fermi gas and the interaction energy from the average Coulomb energy of a system of electrons at that density: ECoulomb = (const. When the Coulomb interaction energy is large compared to the kinetic energy. the Coulomb energy is small compared to the kinetic energy. but not for electrons in metals which interact through the Coulomb interaction. rs is the controlling 223 .

the zeroth-order term is the kinetic energy. parameter for many of the approximations which we make in this chapter.7 ≈ − 2 n rs rs (13. this can be computed perturbatively from the Lagrangian: S= + dτ dτ 4πκe2 q2 d3 k ψ † ∂τ ψ + 1 2 † k ψ (k)ψ(k) 2m d3 k d3 k′ d3 q ψ † (k + q)ψ(k) − nδ(q) × ψ † (k′ − q)ψ(k′ ) − nδ(q) (13. this can be estimated to be: W Eg C 2.6eV .and high-density limits for a model of electrons in a ﬁxed uniform background of positive charge (the jellium model).2. rs is small in the high-density limit where we expect a Fermi liquid and large in the low-density limit where we expect Wigner crystallization.2099 1. The ﬁrst-order terms come from diagrams of (a) and (b) of ﬁgure 13.224 CHAPTER 13. 13. ELECTRONS AND COULOMB INTERACTIONS k k+q q k’ k’-q Figure 13.1: The graphical representation of Coulomb interactions. Better estimates of the Wigner crystal and electron liquid energies can be obtained in the low.3) in units of the Rydberg. In the liquid state.1 If we expand the ground state energy perturbatively. In the Wigner crystal state. as in ﬁgure 13. The ﬁrst – or Hartree – term vanishes as a result of .4) The Coulomb interaction is represented by a dotted line.

2: The (a) Hartree and (b) Fock contributions to the ground state energy of the electron gas. it is justiﬁed for a calculation of the ground state energy in the small rs limit since the neglected diagrams give contributions of O(rs ). The next terms in the expansion in rs come from summing the diagrams of ﬁgure 13. In the Hartree-Fock approximation.5) Eg = 2 rs rs where the ﬁrst term is the kinetic energy and the second term is the exchange energy. For rs large.094 + 0. the neutralizing background (i. 13.e.9163 L − (13.2099 0. The second – or Fock – term is non-vanishing. but the sum is convergent: L Eg = 2. the nature of which we discuss in this chapter.3. the ground state of the electron gas is the Wigner crystal. the ground state energy is given by: 2. naive perturbation theory is infrared divergent because the interaction V (q) = 4πκe2 /q 2 (unless otherwise . it is the liquid state.2.2 Screening In the presence of Coulomb interactions.13.6) This is the sum over bubble diagrams – the Random Phase Approximation – which we encountered in the small Λ/kF approximation for a Fermi liquid.0622 ln rs 2 rs rs (13. The ﬁrst term in this series is infrared divergent. SCREENING 225 a) b) Figure 13.2099 0. In this context.9163 − − 0. the nδ(q)). For rs small.

however. However. a simple model for screening which illustrates the basic physics in the low q. In the language of the last chapter. we’ll have to use more sophisticated approximations such as the RPA. we cannot divide the Fermi surface into N patches and justify Fermi liquid theory in the large N limit because the interaction V (q) = 4πκe2 /q 2 is singular within a single patch when q → 0. In fact. the ‘bare’ Coulomb interaction. is not the actual interaction between two electrons. ω limit. The correct strategy for dealing with electrons with Coulomb interactions is to do perturbation theory in the screened Coulomb interaction.7) κ κ is the dielectric constant due to the ions and the core electrons. we recall the Thomas-Fermi model.3: The RPA contributions to the ground state energy of the electron gas. they will induce a charge distribution ρind .226 CHAPTER 13. we work in d = 3 in this chapter) is singular in the q → 0 limit. we assume that φ is very slowly-varying so that we can make the approximation that the local density is given by n(µ + eφ(r)): ρind (r) ≈ −e (n(µ + eφ(r)) − n(µ)) . V (q) = 4πκe2 /q 2 .. To understand the physics at q → 2kF . Figure 13. First. Let us imagine that we have test charges described by ρext . the interaction between any two electrons will be far weaker because all of the other electrons will act to screen the Coulomb interaction.. as we show in the next two sections. According to the Laplace equation 1 2 ∇ φ = 4πρext + 4πρind (13. speciﬁed. This can be done systematically. In a metal. In the Thomas-Fermi aproximation. ELECTRONS AND COULOMB INTERACTIONS + + + .

ω) = V (q) + V (q) I(q. ω) V (q) + V (q) I(q. ω)V (q) .13. Restricting attention to the sum of bubble diagrams is.2.9) In other words. .14) = 1 − I(q.8) (13.1 by the sum of the diagrams of ﬁgure ??. ω) V (q) + . the screening length is large and we do not expect the Thomas-Fermi approximation to be valid.e. ∂n 1 2 q + 4πe2 κ ∂µ φ = 4πρext ∂n φ(r) ∂µ 227 (13. µ = (3πn)2/3 /2m. Veﬀ (q.12) For a free Fermi gas. so the screening length is −1 k0 = π 4 4 9 1/3 1 2 1 2 a0 rs (13. When rs is large. A more reﬁned result may be obtained by by replacing the bare Coulomb interaction of ﬁgure 13.13) When rs is small. . ω) V (q) I(q. the screening length is short and the Coulomb interaction is eﬀectively screened.10) (13. the RPA approximation. SCREENING ≈ −e2 Then.11) ∂µ (13. k0 = 4πκe 2 ∂n 1 2 (13. when the density is large. however. again. we expect the potential to be slowly-varying and the Thomas-Fermi approximaton to be reasonable. is: RPA Veﬀ (q. When this is true. V (q) (13. ω). i. RPA The eﬀective interaction. the bare Coulomb interaction has been replaced by a screened Coulomb interaction: 4πκe2 4πκe2 → 2 2 q2 q + k0 or 1 e−k0 r → r r where k0 is the inverse of the Thomas-Fermi screening length.

1 2 ∇ φ = 4πρ κ dj m = ne2 ∇φ dt dρ +∇·j =0 dt (13.3 The Plasmon Although Coulomb interactions are ultimately screened and therefore allow a Fermi liquid treatment.228 CHAPTER 13. RPA Veﬀ (q. for ω = 0.17) Hence. for q → 2kF . Of course. 0) = V (q) ∗ 1 + m kF 2π 2 4πκe2 V (q) m∗ kF 2π 2 (13. the RPA result contains information about the Fermi surface. the RPA result contains additional information about the dynamics of the electrons. This may be seen at a classical level from Maxwell’s equations together with the continuity equation. we have ω2 + 4πκne2 m ρ(q. ω) = 0 (13. The zero-sound mode no longer has linear dispersion. 13. j = |j|q/|q|. Also. there are non-trivial diﬀerences with the case of short-range interactions. For q small. However. ELECTRONS AND COULOMB INTERACTIONS where I(q. this sum gives the leading contribution in rs in the limit of small ω.16) Combining these equations for a longitudinal disturbance. ωp ωp = 4πκne2 m 1 2 (13.15) = q2 4πκe2 = 2 2 q + k0 + 4πκe2 which is the same asthe Thomas-Fermi result. ω = vs q. As we will see below. q. it is not clear why we can restrict attention to the sum of bubble diagrams. ω) is the particle-hole bubble which we evaluated in the last chapter. the frequency of a longitudinal density modulation is the plasma frequency.18) . the RPA approximation can be called into question. For ω = 0 and q → 2kF .

In this section. THE PLASMON 229 On the right-hand-side. ω) = ρ(q. However. φind (q. When this is done. One might be tempted to conclude that the compressibility of the electron gas vanishes. −ω) δφext (q. considering the RPA sum of bubble diagrams which determines the density-density correlation fucntion. again.3.22) so the compressibility is ﬁnite as q → 0. gauge invariance tells us that longitudinal photons – whose exchange gives . 2 kF 1 q2 = 2 4πκe2 2π vF 2 2 1 kF = 3 2π 2 vF rather than the gapless dispersion of zero sound in a neutral Fermi liquid. ω) = − Hence.21) which is the same as (13. ω) φ(q. ω) ρ(−q. ω) In linear response. Essentially. screened Coulomb interaction.18). ω) q2 (13.41) by the Coulomb interaction.. δ ρ(q. the pole in this correlation function now occurs at: 2 kF ω + vF q 1 ω 1 = 2 ln −1 (13.23) (13. the remaining electronic degrees of freedom interact through a short-ranged. ω) ρ(−q. ρ(q. ω) δφ(q. ω) = ρ(q. we will show. ω) is given by. −1 (13. F0 . Since V (q) → ∞ as q → 0.24) 4πκe2 0 χρρ (q.. φind (q. ω) ρ(−q. The Landau parameter. following Bohm and Pines. the density-density correlation function gives the compressibility in response to the applied ﬁeld: δ ρ(q. −ω) 1+ 4πκe2 0 χρρ (q. Consequently. take the q → 0 limit with q ≪ ω.13. The same result may be seen by. ω) q2 (13. V (q). is replaced in (12.19) V (q) 2π vF 2 vF q ω − vF q ω vF q vF q ω vF q 2 vF q + ω 3 ω 3 2 2 + .20) or. how to separate the plasma oscillation from the rest of the degrees of freedom of an electronic system. ω 2 = 4πκe2 2 vF kF 6π 2 (13. −ω) → 0 in this limit.

26) d3 k The magnetic part of the electromagnetic action has been dropped since we assume that all velocities are much smaller than the speed of light. We begin with the action of a system of electrons with Coulomb interactions: S= + dτ dτ d3 k 1 2 † k ψ (k)ψ(k) 2m 4πκe2 † ′ ψ (k − q)ψ(k′ ) (13. which do not enter here. A0 = 0. we will choose Coulomb gauge. In doing so we must. a tree-level analysis misses the fact that A(k) is. To do this. S= + dτ dτ d3 k 2 1 † ψ (k) k + eA ψ(k) 2m 1 1 † 3 (∂τ A(k))2 + A0 ∇ · E − 4πκeψ(13.26) by integrating out the electromagnetic ﬁeld. the Coulomb interaction becomes short-ranged. To exhibit this clearly. If we were to integrate out A(k).230 CHAPTER 13. in fact. Consequently. we keep only the longitudinal modes of the electromagnetic ﬁeld. not gapless. we make the following manipulations: • Electrons with Coulomb Interactions. However. since A(k) is gapless at tree-level. When long-wavelength density ﬂuctuations aquire a mass gap as a result of their self-interaction. we would get Coulomb interactions between electrons. The real dynamics of the electromagnetic ﬁeld is in its transverse components. however. We could equivalently write this as S= + dτ dτ d3 k ψ † ∂τ ψ + eψ † A0 ψ + 1 E(k)E(−k) 8πκ 1 † ψ (k) k + eA 2m 2 ψ(k) (13. they (and the longitudinal photons to which they are equivalent) can no longer propagate over long-distances.27) ψ d k 8πκ 4πκ ψ † ∂τ ψ + Note that A(k) is a scalar ﬁeld because it is only the longitudinal part of the electromagnetic ﬁeld – which not independent of the density ﬂuctuations of the electrons. impose the Gauss’ law constraint which is the A0 equation of motion. Equation (13.25) is obtained from (13. then.25) d3 k d3 k′ d3 q ψ † (k + q)ψ(k) q2 ψ † ∂τ ψ + • Electrons interacting with Longitudinal Photons. The long-range 1/q 2 interaction results from integrating out the longitudinal part of the electromagnetic ﬁeld. ELECTRONS AND COULOMB INTERACTIONS rise to the Coulomb interaction – and density ﬂuctuations are not distinct objects. .

let us instead only integrate out those modes of A(k) with k > Q for some Q. by applying a canonical transformation of the form: U =e √ RΛ 2 R 3 d k ψ† (k+q)ψ(k) −i Q d3 q a(q) 4πκe q (13. Bohm and Pines did this at the Hamiltonian level.13. we obtain an action of the form: S= + + dτ 2 1 † ψ (k) k + eA ψ(k) 2m Q 1 1 d3 q dτ (∂τ A(q))2 + A0 ∇ · E(q) − 4πκeψ † ψ 8πκ 4πκ 0 Λ 4πκe2 † ′ ψ (k − q)ψ(k′ ) (13. Λ is the upper cutoﬀ and Q is a wavevector to be determined. Then. • Isolate the term which gives a gap to long-wavelength Photons. and a part resulting from ﬂuctuations in the density: dτ d3 k 0 Q d3 q 0 Q d3 q ′ e2 A(q)A(q ′ ) ψ † (k − q − q ′ )ψ(k) = 2m .29) dτ d3 k d3 k′ d3 q ψ † (k + q)ψ(k) q2 Q d3 k ψ † ∂τ ψ + Notice that the four-fermion interaction is now short-ranged since it is restricted to |q| > Q. THE PLASMON 231 • Integrate out Short-Wavelength Photons. Instead of integrating out A(k) fully. We now expand (k + eA)2 : S= + + + + dτ dτ dτ d3 k ψ † ∂τ ψ + Λ Q k2 † ψ (k)ψ(k) 2m d3 q ψ † (k + q)ψ(k) k+ 4πκe2 † ′ ψ (k − q)ψ(k′ ) q2 d3 k d3 k′ d3 k Q q · A(−q) ψ † (k + q)ψ(k) 2 0 Q Q e2 A(q)A(q ′ ) ψ † (k − q − q ′ )ψ(k) d3 q ′ d3 q dτ d3 k 2m 0 0 Q 1 1 |∂τ A(q)|2 + A0 ∇ · E(q) − 4πκeψ † ψ (13. n.3.27).30) d3 q dτ 8πκ 4πκ 0 d3 q e m We split the third line into a part which comes from the average density.28) to the Hamiltonian corresponding to (13.

232 CHAPTER 13. therefore. To understand the electron gas at a quantitative level. The basic physics is already clear from (13. we have obtained a theory of electrons with short-range interactions. If we were to integrate them out now. these degrees of freedom are not gauge invariant and are. SInt describes the interaction between electrons and plasmons. Q. and is not the Thomas-Fermi interaction in the ω. SP is the action of plasmon modes A(q) with |q| < Q. The interaction depends on a free parameter. these modes have frequency ωp . we are not yet in a position to make quantitative predictions. we must consider SInt and SC .32). .32) dτ dτ d3 k ψ † ∂τ ψ + Λ Q Q k2 † ψ (k)ψ(k) 2m d3 q ψ † (k + q)ψ(k) 4πκe2 † ′ ψ (k − q)ψ(k′ ) q2 d3 k d3 k′ dτ 0 2 d3 q |∂τ A(q)|2 + ωp |A(q)|2 Q dτ dτ d k Q 0 0 3 d3 q Q 0 e m d3 q ′ k+ q 2 · A(−q) ψ † (k + q)ψ(k) d3 kψ † (k − q − q ′ )ψ(k) − nδ(q + q ′ ) (13. unphysical.33) d3 q e2 A(q)A(q ′ ) 2m Q d q 3 1 A0 ∇ · E(q) − 4πκeψ † ψ 4πκ SFL is the action of electrons with short-range interactions. However. SC imposes the constraints which eliminate the additional degrees of freedom introduced with the plasmons.31) the ﬁrst term on the right-hand-side can be combined with the |∂τ A(q)|2 term to give SP in the action: S = SFL + SP + SInt + SC with SFL = + 1 SP = 8πκ SInt = + SC = dτ 0 (13. they would mediate a short-range interaction. not the long-range Coulomb interaction. ELECTRONS AND COULOMB INTERACTIONS ne2 A(q)A(−q) 2m 0 Q Q e2 A(q)A(q ′ ) d3 q ′ d3 q dτ 2m 0 0 d3 q Q dτ + d3 kψ † (k − q − q ′ )ψ(k) − nδ(q + q ′ ) (13. q → 0 limit. By separating the plasmon from the other electronic degrees of freedom.

13.34) Physicaly. for q large.35) in SInt . To justify the RPA. or RPA and completely neglect the term: Q dτ 0 d3 q 0 Q d3 q ′ e2 A(q)A(q ′ ) 2m d3 kψ † (k − q − q ′ )ψ(k) − nδ(q + q ′ ) (13. The RPA is justiﬁed in the limit of small rs and the limit q → 0. RPA 233 13. the random phase approximation is valid in the small rs limit since the energy shift is small compared to the plasmon zero-point energy. we consider SInt and the RPA approximation which simpliﬁes it. We have used the term RPA in several contexts. In the computation of a correlation function at q. we only consider diagrams in which the dotted Coulomb interaction line carries momentum q. We neglect the coupling between ρ(q) and ρ(q ′ ) if q = q ′ . the plasmon mixes with the particle-hole continuum and is no longer a well-deﬁned mode. We are now left with the action SRPA = dτ d3 k ψ † ∂τ ψ + k2 † ψ (k)ψ(k) 2m .4 RPA The conclusions which we drew at the end of the previous section were based on a neglect of SInt and SC . In other words. It shifts this energy by ∆E ∼ Q3 3 3 Q ωp 2kF rs 3 3 4 ∼ Q ωp 4 (13. Without proof. thereby making the eﬀect of SInt as small as possible. The deﬁnition of the RPA is the following. The ﬁrst step is to choose a Q which optimizes SFL +SP . We now make the Random Phase Approximation. this amounts to keeping only the bubble diagrams and neglecting other diagrams. V (q ′ ) does not appear in these diagrams unless q = q ′ . In this section. we state that we can minimize the energy of the ground state of SFL + SP (computed to lowest order in the screened Coulomb interaction) by taking Q ≈ kF rs 4 1 4 (13.4. consider the eﬀect of this term on the ground state energy.36) Hence. For the density-density response function or the ground state energy. Q must be ﬁnite since.

What we can do. (13. 13.38) generated by S: S= e m Q 0 d3 q d3 k 1 ω(q. ψ † ) = ωp 1+ d3 k k2 ψ † (k)ψ(k) + q4 2 8m2 ωp (13. ψ. we can ignore the plasmons. we now perform a canonical transformation. the constraint will only involve the particles. When this is done.40) The principal results of this canonical transformation in the limit rs → 0 are: . is decouple the plasmon from the electrons. we would have a theory of electrons with short-range interactions together with gapped plasmons. If we ignore the constraint – which is now a constraint on the electrons alone – then we can apply Fermi liquid theory to the electronic action. At frequencies or temperatures much less than ωp .5 Fermi Liquid Theory for the Electron Gas O → e−iS OeiS |χ → e−iS |χ Following Bohm and Pines. ψ † ) − q · k + q 2 /2m q2 2 2nm2 ωp { k+ q 2 · A(−q) ψ † (k + q)ψ(k)} (13. ELECTRONS AND COULOMB INTERACTIONS d3 k d3 k′ Q Λ Q + dτ d3 q ψ † (k + q)ψ(k) 4πκe2 † ′ ψ (k − q)ψ(k′ ) q2 1 + 8πκ + + dτ dτ dτ 0 2 d3 q |∂τ A(q)|2 + ωp |A(q)|2 Q d k Q 0 0 3 d3 q e m k+ q 2 · A(−q) ψ † (k + q)ψ(k) (13. instead. Treating the electrons and plasmons as fully independent would be a double-counting of the degrees of freedom of the system. so we would have a Fermi liquid.37) d q 3 1 A0 ∇ · E(q) − 4πκeψ † ψ 4πκ If we could ignore the last line.234 CHAPTER 13. Fermi liquid theory (as we saw in the last chapter) instructs us to compute only bubble diagrams to obtain the screened Coulomb interaction. ψ. the constraint cannot be ignored. However.39) where ω(q.

ψ † )|A(q)|2 (13. The contraints reduce the number of degrees of freedom of the electrons. ψ. this is assumed to have a small eﬀect on the electronic degrees of freedom. we now have a theory of weakly-coupled electrons and plasmons. Λ Q d3 q ψ † (k+q)ψ(k) 4πκe2 † ′ ψ (k −q)ψ(k′ ) → q2 d3 q ψ † (k+q)ψ(k) VRPA (q) ψ † (k′ −q)ψ(k′ ) (13.13. ψ † )) − q · k − q 2 /2m for |q| < Q. 2 2 ψ † (k + q)ψ(k) = 0 (13.5. ψ. ψ. The constraints now read: d3 k ω(q. For Q small. ψ † ) (ω(q.42) • The elimination of the plasmons from the contraints. Hence. FERMI LIQUID THEORY FOR THE ELECTRON GAS 235 • The elimination of the Aψ † ψ interaction between plasmons and electrons • The modiﬁcation of the plasmon action to: SP = 1 8πκ Q dτ 0 d3 q |∂τ A(q)|2 + ω(q.43) . The electrons interact through a short-ranged interaction (which can be obtained by summing bubble diagrams).41) • The replacement of the cutoﬀ Coulomb interaction by the RPA screened Coulomb interaction.

236 CHAPTER 13. ELECTRONS AND COULOMB INTERACTIONS .

k) = g2 = g2 dω 2π dω 2π d3 q G(iǫ − iω.2) .CHAPTER 14 Electron-Phonon Interaction 14. q) (2π)3 d3 q q2 1 (2π)3 iǫ − iω − ξk−q −ω 2 − v 2 q 2 237 (14.2 14. The ﬁrst diagram just shifts the chemical potential.1) which couples electrons to transverse phonons. k − q) D(iω.3 14. Sel−ph = g dτ d3 x ψ † ψ ∂i ui (14. What eﬀect does this have on the electron Green function? The one-loop electron self-energy is given by the diagrams of ﬁgure 14.4 Electron-Phonon Hamiltonian Feynman Rules Phonon Green Function Electron Green Function Let us consider the electron-phonon interaction. At zero-temperature.1. the second diagram gives a contribution: Σ(iǫ.1 14.

k) = g2 d3 q (2π)3 At small ǫ.3) −2qv iǫ + vq − ξk−q (iǫ − ξk−q )2 − v 2 q 2 q 2 θ (ξk−q ) 1 q + (14. k) = g2 = d3 q q [− (1 − θ (ξk−q )) δ (ǫ + vq − ξk−q ) − θ (ξk−q ) δ (ǫ − vq − ξk−q )] (2π)3 2v q 3 dq d(cos θ) [− (1 − θ (ξk−q )) δ (ǫ + vq − ξk−q ) − θ (ξk−q ) δ (ǫ − vq − ξk−q )] For ǫ small. ImΣret (ǫ. k) ∼ g2 ǫ Meanwhile.6) Analytically continuing iω → ω + iδ to obtain the retarded self-energy.1: The one-loop diagrams contributing to the electron-self-energy. ELECTRON-PHONON INTERACTION Figure 14. we pick up the pole at iω = iǫ − ξk−q if ξk−q > 0 and the pole at iω = −vq: Σ(iǫ. Closing the contour in the upper-half-plane. k) = g2 d3 q (2π)3 q 2 θ (ξk−q ) 1 q2 + (14. dropping a constant shift in the chemical potential.7) g2 2(2π)2 .4) −2v ǫ + vq − ξk−q (ǫ − ξk−q )2 − v 2 q 2 2ξk−q q 2 θ (ξk−q ) 1 q − 2 (14. ReΣret (ǫ.5) 2v (vq − ξk−q )2 ξk−q 2 − v 2 q 2 (14. k) = g2 ǫ d3 q (2π)3 We can take k → kF in this integral to obtain the leading behavior: ReΣret (ǫ. we have: ReΣret (ǫ.238 CHAPTER 14. and q 2 terms in the δ function can be dropped: ImΣret (ǫ. q ∼ ǫ. k) = g2 2(2π)2 q 3 dq d(cos θ) [− (1 − θ (ξk−q )) δ (ǫ + (v + vF cos θ) q) (14.

5. POLARONS − θ (ξk−q ) δ (ǫ − (v − vF cos θ) q)] 239 ∼g ǫ 2 3 (14.8) 14.5 Polarons .14.

ELECTRON-PHONON INTERACTION .240 CHAPTER 14.

Some special conformal ﬁeld theories – but. SU(2) symmetry.CHAPTER 15 Rudiments of Conformal Field Theory 15. In two or 1 + 1 dimensions. Conformal ﬁeld theory is a highly-developed subject which has taken on a life of its own. it places strong constraints on the critical spectrum and correlation functions – much stronger than those due to. The techniques which we will use in this section. the Virasoro algebra). on which this part of the book is focused. We also hope that our treatment of conformal ﬁeld theory will help orient the reader who is interested in delving more deeply into the literature. in part due to applications to string theory. say. Such an in-depth treatment would be out of place here. by no means. its quantum version. rather. Rather. we would like to emphasize some of the key points. which are primarily algebraic. are the ﬁrst truly non-perturbative ones which we have encountered 241 . this symmetry algebra is inﬁnite-dimensional.1 Introduction In ﬁnite-temperature equilibrium statistical mechanical systems at their critical points or quantum systems at T = 0 with z = 1. scale invariance and rotational invariance (or pseudo-Lorentz invariance) give rise to the larger symmetry algebra of conformal transformations. so it is the subject of many full-length books and long review articles. Thus. all – can be completely solved essentially through the representation theory of the conformal algebra (or. and use conformal ﬁeld theory to illustrate some important concepts relating to critical points.

. and special conformal transformations.5) If we write z. . Once interesting feature. We will be working in ﬂat Euclidean space or Lorentzian spacetime. ǫ = ǫ1 ± iǫ2 . with metric ηµν = diag(1.242 CHAPTER 15. . is that the analysis is most easily done in real space rather than momentum space. scale transformations. z = x1 ± ix2 . and ǫ. 15.4) (15. ǫ = ǫ(z).2) For d > 2. RUDIMENTS OF CONFORMAL FIELD THEORY so far in this book.3) (15. which will recur in the nonperturbative methods discussed in later parts of the book.1) (15. 1. The advantages of momentum space for perturbation theory – the ease with which diﬀerential operators are inverted and the simple expression of translational invariance as momentum conservation – are outweighed by the advantages of real space which we will discover in the following pages. there is a ﬁnite-dimensional group of such transformations comprised of translations. . 1). 1. xµ → xµ + ǫµ ds2 = ηµν dxµ dxν → ηµν (dxµ + ∂α ǫµ dxα ) dxν + ∂β ǫν dxβ = ηµν dxµ dxν + (∂µ ǫν + ∂ν ǫµ ) dxµ dxν Such a transformation will be a conformal transformation if (∂µ ǫν + ∂ν ǫµ ) ∝ ηµν Comparing the traces of both sides. this condition is simply the Cauchy-Riemann equations: ∂1 ǫ2 = −∂2 ǫ1 ∂1 ǫ1 = ∂2 ǫ2 (15. . 1) or ηµν = diag(−1. rotations. this can only be satisﬁed if (∂µ ǫν + ∂ν ǫµ ) = 2 (∂α ǫα ) ηµν d (15. ǫ = ǫ(z). while ǫ is anti-holomorphic. then ǫ is a holomorphic function of z.2 Conformal Invariance in 2D A conformal transformation is any coordinate transformation which only changes the metric by a scale factor. that all of the . which are inversion-translation-inversion combinations. . . Writing our coordinates and ﬁelds in terms of holomorphic and antiholomorphic quantities will prove to be such a powerful tool. In d = 2. . so this means that ηµν transforms as ηµν → λ(x) ηµν .

. which contains real space as a section.15. ηzz = ηzz = 1 . Thus. Making the transformation z = −1/w in order to study the z → ∞ limit. ℓ−1 . in the extension of it to the two complex dimensional space (z. we see that the ℓn s and ℓn s generate two independent copies of the same algebra: [ℓm . ℓn ] = 0 (15. Indeed. As we will see. Its equation of motion is ∂µ ∂ µ ϕ = 0. which is used to lower indices. ℓ1 generate special conformal transformations. By direct calculation. We have been careful to write upper and lower indices because the metric tensor. ℓn = −z n+1 ∂ is non-singular as z → 0 only for n ≥ −1. n = 0.6) The analogous operator ℓn generates transformations of the zs. we ﬁnd that ℓn = −w1−n ∂w is non-singular as w → 0 only for n ≤ 1. we can take this as a deﬁnition of a conformal ﬁeld theory: a 2D quantum ﬁeld theory which has correlation functions which decouple in this way. In particular. however. Consider the transformation z → z − ǫ z n+1 . CONFORMAL INVARIANCE IN 2D 243 subsequent development will be carried out in real space – or. Such a transformation is generated by the linear operator ℓn = −z n+1 ∂: δf (z) = f (z − ǫ z n+1 ) − f (z) = f (z) − ∂f · ǫ z n+1 − f (z) = −ǫ z n+1 ∂f = ǫ ℓn f (15. ∂ = ∂z . Hence. where we have introduced the notation ∂ = ∂z . i ℓ0 − ℓ0 generates rotations. ℓn ] = (m − n)ℓm+n [ℓm . and ℓ1 . z) ∈ C 2 . rather. ℓn ] = (m − n)ℓm+n [ℓm . Hence. ℓ0 + ℓ0 generates scale transformations. this means that the correlation functions have simple scaling properties. Consider a free scalar ﬁeld.2. ±1 are the only globally well-deﬁned transformations. the equation 2 of motion takes the form ∂z ∂z ϕ ≡ ∂∂ϕ = 0. that these transformations are not all globally welldeﬁned. z = (z)∗ – rather than in momentum space where most of our previous discussion of ﬁeld theory has taken place. ϕ is the sum of an arbitrary holomorphic function and an arbitrary anti-holomorphic function. takes the form ηzz = ηzz = 0. ℓ−1 generate translations. This decoupling between holomorphic and anti-holomorphic – or rightand left-moving – degrees of freedom is a general feature of conformal ﬁeld theory. The (classical) algebra of conformal transformations also decouples in this way.7) Note. in complex coordinates. where ǫ is inﬁnitesimal.

invariance under special conformal transformations is automatically a property of any theory which is scale and translationally invariant. A primary ﬁeld of weight (h. consider the transformation properties of the basic ﬁelds in a given theory. h = 0 and h = 0. f (z) (15.h (z. b. Thus. z → λz it transforms as Φ → λh+h Φ. C)/Z2 (the Z2 is modded out because a. The special conformal transformations. This is the group SL(2. and translational and rotational invariance are nice features which one would want most theories to have (at least in their continuum limits). a scalar under rotation has h = h. −b.h f (z). z → e−iθ z a primary ﬁeld of weight h transforms as Φ → ei(h−h)θ Φ. RUDIMENTS OF CONFORMAL FIELD THEORY The general form of a transformation generated by these operators is z→ az + b cz + d (15. z) → ∂f ∂z h ∂f ∂z h Φh. d → −a.3 Constraints on Correlation Functions Now. the anti-holomorphic part is analogous). the correlation function Φ1 (z1 )Φ2 (z2 ) transforms as δǫ Φ1 (z1 Φ2 (z2 ) = [(h1 ∂1 ǫ(z1 ) + ǫ(z1 )∂1 ) + (h2 ∂2 ǫ(z2 ) + ǫ(z2 )∂2 )] Φ1 (z1 )Φ2 (z2 ) (15. while a vector has two components with h = 1.8) with ad − bc = 1. z → z/(az + 1) are somewhat less familiar than the others.10) . Under a rotation z → eiθ z. −c.11) (15. h) transforms as Φh.244 CHAPTER 15. a primary ﬁeld transforms as δǫ Φh (z) = (1 + ∂ǫ)h Φh (z + ǫ(z)) − Φh (z) = (h∂ǫ + ǫ∂)Φh (z) Hence.9) under a conformal transformation z → f (z). −d leaves the transformation unchanged). Scale invariance is the deﬁning property of a critical theory. h = 1. 15. z → f (z). under a scale transformation z → λz. but why special conformal transformations? As we will show in our discussion of the energy-momentum tensor. Primary ﬁelds generalize the notions of vectors and tensors under the rotation group to the conformal algebra. c. Under an inﬁnitesimal conformal transformation (on the holomorphic part of the theory.

we must determine the numbers cijk .3. when we turn to the four-point correlation function. Similarly. . correlations functions will be invariant under these transformations. the functions f (x. Hence. the vacuum state will be invariant under translations. and their counterparts for higher correlation functions. the four-point function takes the form: Φ1 (z1 . the correlation function is reduced to a single unknown constant (which remains unknown once the normalization is determined by the twopoint function). we see that there is real freedom here.13) Combining this with the analogous equation for the anti-holomorphic dependence of the correlation function.11): (∂1 ) + ∂2 ) Φ1 (z1 )Φ2 (z2 ) = 0 (15. its anti-holomorphic dependence is only on z 12 = z 1 − z 2 The implication of scale and rotational invariance may be seen by substituting ǫ(z) = z in (15.15. the two-point correlation function is reduced to single unknown constant.14) z12 z 12 Thus. z 2 )Φ3 (z3 . z 2 ) = 2h 2h (15. We can also apply the constraint of conformal invariance to the threepoint function. However. z 1 )Φ2 (z2 . the two-point function is completely constrained. In order to solve a conformal ﬁeld theory.12) In other words. we see that the correlation function vanishes unless h1 = h2 and c12 Φ1 (z1 . x).11): [(z1 ∂1 ) + h1 ) + (z2 ∂2 + h2 )] Φ1 (z1 )Φ2 (z2 ) = 0 (15. We ﬁnd that it must take the form c123 (15.16) × (z → z) where h = j hj . rotations. the cross ratio (or anharmonic ratio) x = z12 z34 /z13 z24 is left invariant by the group of global conformal transformations SL(2. CONSTRAINTS ON CORRELATION FUNCTIONS 245 In a ﬁeld theory which is invariant under the group of conformal (global) transformations. C)/Z2 . z 3 ) = h1 +h2 −h3 h2 +h3 −h1 h1 +h3 −h2 × (z → z) z12 z23 z13 Again. The implication of translational invariance may be seen by substituting ǫ(z) = 1 in (15. z 2 )Φ3 (z3 . the correlation function is a function of z12 = z1 − z2 . and special conformal transformations. x) i>j zij −hi ihj +h/3 (15.15) Φ1 (z1 . z 1 )Φ2 (z2 . Since we can always redeﬁne our ﬁelds in order to absorb this constant into their normalization. z 4 ) = f (x. As a result. z 1 )Φ2 (z2 . z 3 )Φ3 (z4 .

the operator product expansion expresses the notion that two nearby particles will appear to be a single composite particle when viewed from a great distance.18) Now. the operator product expansion is written in the form: φi (z. Then. if we perform a conformal transformation to coordinates z = eτ −ix . is that other operators which are formally unrelated to φk will also appear because they are generated under renormalization by φi . Mode Expansions As we have already discussed. w) (15. C2 is a contour encircling the origin at radius smaller than |z|. The new element. The operator product expansion is a very powerful tool when combined with contour integration. Hence. ∂ 2 φk will appear. The leading term on the right-hand-side is the most singular one. compared to an ordinary Taylor expansion. all of its derivatives ∂φk . the variation in B(z) is: δA B(z) = dz ′ A(z ′ )B(z) − B(z)A(z ′ ) 2πi dz ′ dz ′ = R A(z ′ )B(z) − R A(z ′ )B(z) 2πi 2πi C1 C2 dz ′ ′ R A(z )B(z) = Cz 2πi (15.e.19) where C1 is a contour encircling the origin at radius larger than |z|. so that the spatial coordinate x is restricted to the interval [0. In addition to a given operator φk . i. the operator φk of lowest dimensions hk .17) The coeﬃcients cijk are precisely the same as appear in the three-point function. 2π]. If A(z ′ ) is purely holomorphic. and Cz is a contour . The right-hand-side can be thought of as somewhat like a Taylor expansion. Time-ordering is now radial ordering: R A(z ′ )B(z) = A(z ′ )B(z) if |z ′ | > |z| B(z)A(z ′ ) if |z| > |z ′ | (15.246 CHAPTER 15. z) φj (w. Radial Quantization. w) = k cijk (z − w)hk −hi −hj (z − w)hk −hi −hj φk (w. this doesn’t depend on the speciﬁc circle along which we do the integral. φj . hk . constant radius circle in the complex plane are constant time slices. In a 2D conformal ﬁeld theory. consider the transformation property of B(z) under a transformation generated by A = A(z ′ )dz ′ /2πi.4 Operator Product Expansion. This is most nicely done if one considers a system with periodic boundary conditions. RUDIMENTS OF CONFORMAL FIELD THEORY 15.

the correlation function of ∂ϕ is 1 (15. According to Wick’s theorem.. Hence.25) (Recall that a radial ordering symbol is implicit. : ∂ϕ(z)∂ϕ(z) := −2T (z). in fact. h = 0. . Let us consider. as a simple example.26) (z − w)2 As we will see in the next section. .21) the ﬁeld can be written as the sum of a holomorphic ﬁeld and an antiholomorphic ﬁeld. The latter contour integral is determined by the shortdistance singularity of R (A(z ′ )B(z)). Thus. where T ≡ Tzz is a component of the energy-momentum tensor. in other words. but the notation hides this fact. with action S= 1 2π ∂ϕ∂φ d2 x (15. ϕ(z)ϕ(w) = − ln(z − w) (15.) Expanding ∂ϕ(z) = ∂ϕ(w) + (z − w)∂ 2 ϕ(w) + . φ(z. One customarily drops the radial ordering symbol. or.20) According to the equation of motion. by the operator product expansion of A(z ′ ) and B(z). The correlation functions of these 2 ﬁelds take the form 1 (15. in other words. we have ∂ϕ(z)∂ϕ(w) = − 1 + : ∂ϕ(w)∂ϕ(w) : + (z − w) : ∂ 2 ϕ(w)∂ϕ(w) : (15. OPERATOR PRODUCT EXPANSION. RADIAL QUANTIZATION. (15.24) ∂ϕ(z)∂ϕ(w) = − (z − w)2 ϕ is a primary ﬁeld of dimension h = 1. the expansion which is useful to us is. and simply writes A(z ′ )B(z). Thus. with the analogous equation for ϕ(z). z) = 1 (ϕ(z) + ϕ(z)). not the ordinary product. z)ϕ(w. ∂∂φ = 0 (15.27) (z − w)2 . MODE EXPANSIONS 247 encircling the point z. and expansion of the radial product. . ∂ϕ(z)∂ϕ(w) = − 1 − 2T (w) − (z − w)∂T (w) + .4.23) ∂ϕ(z)∂ϕ(w) = ∂ϕ(z)∂ϕ(w) + : ∂ϕ(z)∂ϕ(w) : 1 =− + : ∂ϕ(z)∂ϕ(w) : (z − w)2 (15.15. . which is understood. a free scalar ﬁeld. w) = − ln |z − w| 2 or.22) φ(z.

15. this looks like φ = ikx−ωt . i. ±2. Translational invariance implies that ∂ µ Tµν = 0 (15. implies that 0 = ∂ µ (Tµν xν ) = (∂ µ Tµν ) xν + Tµν (∂ µ xν ) µ = Tµ (15.29) since ǫµ is constant for a translation. and its derivatives. ±1. the allowed momenta are simply the integers k = 0. Ward Identities According to N¨ther’s theorem (see chapter ). 2π].. . In radial quantization. where Tµν is the energy-momentum tensor. satisfying the conservation law ∂µ j µ = 0. the energy-momentum operator.248 CHAPTER 15. for which ǫµ = λxµ . it is often useful to rewrite the ﬁeld operators in terms of creation and annihilation operators. if we take periodic boundary conditions. there is an associated conserved quantity. The operator product expansion of two operators is equivalent to the commutation relation between their modes. We will see an explicit example of this two sections hence. Scale invariance. Since x ∈ [0. In free ﬁeld theories. to each symmetry of the aco tion. Energy-Momentum Tensor. This operator plays a special role in conformal ﬁeld theories. .30) . z = eix+τ .e. Indeed.5 Conservation Laws. the operator product of ∂ϕ(z) with ∂ϕ(w) contains the identity operator. this was the starting point of our development of quantum ﬁeld theory way back in chapter 2.28) The −1 in the exponent is a matter of convention whereby we separate explicitly the weight of a primary ﬁeld. as may be seen using contour integration. . Very schematically. RUDIMENTS OF CONFORMAL FIELD THEORY Hence. Consider the case of the primary ﬁeld ∂ϕ: ∂ϕ(z) = n∈Z ϕn z −n−1 (15. only integer powers of z appear in the expansion. so the expansion in modes k ak e is an expansion in powers of z. which are their Fourier modes. Invariance under the coordinate transformation xµ → xµ + ǫµ is associated with the current jµ = Tµν ǫµ . as we will see in the next section.

As an aside. Tzz . Under a transformation z → z + ǫ(z). a ﬁeld Φ(z) transforms as δΦ(w) = dz ǫ(z) T (z) Φ(w) 2πi (15. a primary ﬁeld Φ of weight h must.19). Hence. .31) Let us rewrite the trace-free and divergence-free conditions in complex notation.36) . Tzz = Tzz . . T (z) ≡ Tzz and T (z) ≡ Tz z are. as usual) At the same time. WARD IDENTITIES 249 Hence. transform as: δΦ = h∂ǫ Φ + ǫ ∂Φ (15. CONSERVATION LAWS. gzz = gzz = 0. we have used the fact that the energy-momentum tensor is symmetric.35) Hence. such a primary ﬁeld must have the following operator product expansion with the energy-momentum tensor: T (z) Φ(w) = 1 h ∂Φ(w) + .5. The associated conserved quantity is Kµν = x2 Tµν − xµ xα Tαν : ∂ ν Kµν = ∂ µ x2 Tµν − ∂ µ (xµ xα Tαν ) ν = 2xν Tµν + x2 ∂ ν Tµν − δµ xα Tαν − xµ gαν Tαν − xµ xα ∂ ν Tαν ν ν α = 2x Tµν − δµ x Tαν − xµ gαν Tαν =0 (15. consequently. respectively. Thus. gzz = gzz = 1 zz = g zz = 2.33) Similarly. there are only two non-zero components. ENERGY-MOMENTUM TENSOR. holomorphic and antiholomorphic.34) (with radial ordering understood. ds2 = (dx1 )2 + (dx2 )2 = dz dz. ∂Tzz = 0. Φ(w) + 2 (z − w) z−w (15. by deﬁnition. 2 and. Tzz . T (z) and T (z) must implement conformal transformations according to (15. we note that invariance under special conformal transformations comes for free once we have scale and translational invariance. At the quantum level.32) In the last line. Hence. g µ Tµ = 0 gzz Tzz + gzz Tzz = 0 Tzz = 0 (15. Hence. the energy-momentum tensor is both divergenceless and traceless. They are constrained by the divergencelessness of the energymomentum tensor: gαµ ∂α Tµν = 0 g ∂Tzz + gzz ∂Tzz = 0 ∂Tzz = 0 zz (15.15.

20)) is: 1 (15. Let’s check that this does.250 CHAPTER 15. indeed.41) Ci . The operator product of T (z) with ∂ϕ(w) can be computed using Wick’s theorem. this is the usual type of ambiguity which is encountered in quantizing a classical theory.36). . Either of the ∂ϕ(z) factors in T (z) can be contracted with ∂ϕ(w): T (z) ∂ϕ(w) = − 1 : ∂ϕ(z) ∂ϕ(z) : ∂ϕ(w) 2 1 1 ∂ϕ(z) − =2 − 2 (z − w)2 1 1 ∂ϕ(w) + ∂ 2 ϕ(w) = (z − w)2 (z − w) (15.34) and (15. RUDIMENTS OF CONFORMAL FIELD THEORY We can take this to be the deﬁning relation for a primary ﬁeld of weight h. ǫ(z)T (z)φi (wi ) .35). φn (wn ) = 2πi n dz φ1 (w1 ) .37) T (z) = − : ∂ϕ(z) ∂ϕ(z) : 2 This expression has been normal ordered in order to make it well-deﬁned. .38) The factor of 2 in the second line comes from the two diﬀerent choices for contraction. dz ǫ(z)T (z)φi (wi ) = ǫ(wi )∂φi (wi ) + hi ∂ǫ(wi )φi (wi ) 2πi (15. . . Thus. indeed. This can be derived by inserting the generator of the symmetry transformation into a correlation function and allowing it to act on all of the other ﬁelds in the correlation function.39) The contour is taken to encircle all of the wi s.40) 2πi Ci i=1 C From (15. of the form (15. with h = 1. This is. This contour can be deformed into the sum of contour integrals along small circles encircling each of the wi s. The energymomentum tensor for a free scalar ﬁeld (with action normalized as in (15. . hold for a free scalar ﬁeld. dz ǫ(z)T (z)φ1 (w1 ) . Consider the following contour integral dz ǫ(z)T (z)φ1 (w1 ) . φn (wn ) 2πi (15. . . The quantum mechanical expression of the existence of a symmetry is the corresponding Ward identity which correlation functions satisfy. φn (wn ) (15. There are other ways of deﬁning the energy-momentum tensor. .

. φn (wn ) (15. then this OPE will take the form (15. . . which can be done in two diﬀerent ways. . . VIRASORO ALGEBRA. This is almost what we would get if T (z) were primary. The leading term on the right-hand-side of the OPE prevents the energymomentum tensor from being primary.43) 15. − =2· 2 4 (z − w) 4 (z − w)2 2 1 1/2 + T (w) + ∂T (w) + . . φn (wn ) (15. . φn (wn ) = n ∂ 1 h + 2 (z − wi ) z − wi ∂wi i=1 φ1 (w1 ) .46) 2 1 c/2 + T (w) + ∂T (w) + . . dz ǫ(z)T (z)φ1 (w1 ) . .36) with h = 2. but not quite. 4 2 (z − w) (z − w) z−w (15. we can drop the integral and simply write T (z)φ1 (w1 ) . (15. Central Charge Let’s now consider the OPE of the energy-momentum tensor with itself. The calculation can be done explicitly in the case of a free scalar ﬁeld: T (z) T (w) = 1 1 : ∂ϕ(z) ∂ϕ(z) : − : ∂ϕ(w) ∂ϕ(w) : 2 2 2 1 1 1 1 + 4· · : ∂ϕ(z) ∂ϕ(w) : + . CENTRAL CHARGE Hence. . If T (z) is a primary ﬁeld. φn (wn ) = 2πi dz h ∂ 1 ǫ(z) + 2 (z − wi ) z − wi ∂wi Ci 2πi 251 C n i=1 φ1 (w1 ) . so it is allowed in general: T (z) T (w) = Similarly.6. The second term comes from contracting a single ∂ϕ in T (z) with a single one in T (w). . 4 2 (z − w) (z − w) z−w (15.15. .42) Since this holds for all ǫ(z). T (z) T (w) = 2 c/2 1 + T (w) + ∂T (w) + . . .6 Virasoro Algebra. which can be done in four diﬀerent ways. . A term of this form scales in the correct way. .45) .44) = 4 2 (z − w) (z − w) (z − w) − The ﬁrst term on the second line comes from contracting both ∂ϕs in T (z) with both of them in T (w).

z. Let us consider some of the consequences of the existence of c. z 2 . . up to a constant coeﬃcient. By taking the contour integral of (15. z → f (z). T (z) transforms as a primary ﬁeld. For now. for the algebra of global conformal transformations. for reasons which will become clear momentarily. It is one of the numbers which characterizes any given conformal ﬁeld theory. T (z) → ∂f ∂z 2 T (f (z)) + c ∂f ∂ 3 f − 3 ∂ 2 f 2 12 (∂f )2 2 (15. we will consider conformal ﬁeld theories with other values of c. We can become a bit more comfortable with the Schwartzian derivative by considering the example of a free scalar ﬁeld.47) 12 For ǫ(z) = 1. which is a primary ﬁeld: T (z) → − 1 a lim f ′ z + 2 a→0 2 f′ z − a 2 ∂ϕ f z + a 2 ∂ϕ f z − a 2 + 1 a2 . although the converse is clear. let us continue to proceed with full generality and take c.e.49) The transformation properties of T (z) under a conformal transformation are determined by those of ∂ϕ. it is the unique such quantity. we have just shown that c = 1.47). Later. a similar calculation shows that c = 1. we can ﬁnd the transformation property of T (z) under the conformal transformation z → z + ǫ(z): δǫ T (w) = dz ǫ(z) T (z) T (w) 2πi c/2 2 1 dz ǫ(z) ∂T (w) + . c to be arbitrary. Some insight may be gained by checking that it vanishes for a global conformal transformation z → (az + b)/(cz + d). . RUDIMENTS OF CONFORMAL FIELD THEORY In the special case of a free scalar ﬁeld. In fact. For a ﬁnite transformation.252 CHAPTER 15. which is usually called the central charge.48) The odd-looking second term is called the Schwartzian derivative. It is not so obvious that this is the ﬁnite transformation corresponding to the inﬁnitesimal transformation (15. c. T (z) = − 1 : ∂ϕ(z) ∂ϕ(z) : 2 1 a = − lim ∂ϕ z + 2 a→0 2 ∂ϕ z − a 1 + 2 2 a (15. + T (w) + = 4 2 2πi (z − w) (z − w) z−w c = ∂ 3 ǫ(w) + 2∂ǫ(w) T (w) + ǫ(w) ∂T (w) (15.45) with ǫ(z). i.

Ln = n dz n+1 z T (z) 2πi (15.45) over z and w after multiplying by z n+1 and wn+1 . . Thus. Note that the real culprit here is the normalordering which must be done in order to deﬁne T (z) in a quantum theory.6 15. Let us rewrite (15. which is proportional to the identity operator – hence. c. we see that the classical algebra of inﬁnitesimal conformal transformations has been modiﬁed.15. T (z) = Ln z −n−2 (15. let’s think about the physical interpretation of the central charge.50) which is (15.45) in terms of the commutator between the modes of T (z). It is extended by an additional term. Ln .0 + 2(n + 1) Lm+n − (m + n + 2) Lm+n 12 c [Ln .53) 12 Thus. Lm ] = (n − m) Lm+n + n3 − n δm+n. INTERPRETATION OF THE CENTRAL CHARGE a 2 a 2 1 a 2 2 253 1 a2 = lim a→0 f′ z + 2 f′ z − a→0 T (f (z)) − f′ z + a 2 a 2 = f ′ (z) T (f (z)) − lim = f ′ (z) T (f (z)) − 2 1 12 f ′ f ′′′ f z+ −f z− 3 ′′ )2 − 2 (f (f ′ )2 f z+ ′ z− a f 2 a 2 −f z− − 1 a2 a 2 2 + (15. we ﬁnd dz n+1 z 2πi dz n+1 z 2πi [Ln .52) If we take the integral of (15.7.51) n This expansion may be inverted to give the modes. classically. Lm ] = dw m+1 w T (z) T (w) = 2πi 2 1 dw m+1 c/2 w ∂T (w) + T (w) + 4 2 2πi (z − w) (z − w) z−w c n3 − n δm+n.0 (15. the ordering ambiguities associated with the quantization of a classical theory are responsible for making T (z) a non-primary ﬁeld.48) with c = 1. it is primary.7 Interpretation of the Central Charge Before taking up the representation theory of the Virasoro algebra. a central extension.

. N free scalar ﬁelds have c = N .1 Finite-Size Scaling of the Free Energy Further insight into the central charge. . we will see that the particular form of accounting which is done by the central charge is the ‘right’ one for the purposes of giving insight into RG ﬂows. At the end of this section. and the central charges add (since there won’t be any cross terms in their OPEs).55) This follows from the deﬁnition of the energy-momentum tensor. etc. then their energy-momentum tensors add. Hence.254 CHAPTER 15. If we take two ﬁeld theories and simply add together their actions without coupling them. Thus. it is somewhat surprising that the third term in this expansion is proportional to ln L rather than a constant. In a large but ﬁnite system of linear scale L. the free energy scales as: βF = A2 L2 + A1 L + A0 ln L + . (15. This means that it depends on the geometrical shape of the ﬁnite system in way which is independent of the particular theory under consideration and is completely independent of the short-distance cutoﬀ and other microscopic details. Fermions have diﬀerent central charges from bosons. As we will see. the size of the system changes by L → (1 + ǫ)L. However.56) . However. it weighs such modes diﬀerently. Under such a rescaling. A0 can be – and is – universal. there is a sense in which the central charge measures the number of gapless modes which a system has. the action varies by δS = − ǫ 2π d2 xTzz (15. Under such a transformation. it can be seen that such a term is present by considering the eﬀect on the free energy of an inﬁnitesimal scale transformation.7. the free energy changes by F (L + dL) − F (L) = dL ∂F ∂L ∂F = ǫL ∂L (15. is given by the following interpretation in terms of the ﬁnite-size scaling of the free energy. xµ → (1 + ǫ)xµ . RUDIMENTS OF CONFORMAL FIELD THEORY The ﬁrst obersvation is that the central charge is additive.54) The bulk free energy density and boundary free energy are clearly nonuniversal since they are dimensional quantities. 15. and a way of calculating it for an arbitrary system. At ﬁrst glance.

INTERPRETATION OF THE CENTRAL CHARGE 255 However. since the free energy is the logarithm of the functional integral of the action. The simplest example of this is a wedge of the plane of angle γ. In an inﬁnite system in ﬂat space. on a curved space or in a ﬁnite region of ﬂat space. Tzz = 0 in a critical theory.57) (15.58) Hence. ﬁnite constant. T (w) is singular. As a result of this singularity of T (w). γ γ −1 zπ Tuhp (z) = π Since Tuhp(z) = 0.15. We can compute Tzz in such a wedge using the conformal mapping w = z γ/π from the upper-half-plane to the wedge and the transformation law for T (z). ∂ c 24w2 ∂ 1− c 24w2 π γ 2 + ∂Tzz = 0 1− π γ 2 = −∂Tzz . c Twedge (w) = 24w2 1− π γ 2 2 c Twedge (w) − 24z 2 γ π 2 −1 (15. the expectation value of the change in the action is equal to the change in the free energy. it need not vanish.60) At the corner of the wedge. However.59) (15. L 1 ∂(βF ) =− ∂L 2π d2 x Tzz (15. if the integral over the entire ﬁnite-size system of the right-hand-side of this equation is a non-zero. to lowest order in ǫ: e−βF (L+dL) = = e−S−dS e−S 1 − dS + O(ǫ2 ) = e−βF (L) 1 − dS + O(ǫ2 ) = e−βF (L)− dS Hence. it is not true that ∂T = 0‘. w = 0.7. then the free energy has a ln L term in its expansion. so the conservation law will require that Tzz be a delta-function situated at the corner.

z) Θ(0. c is obtained. Let Θ ≡ Tzz . Once. Consider an arbitrary two-dimensional ﬁeld theory. the the free energy can be computed analytically. z) T (0. due to Zamolodchikov. RUDIMENTS OF CONFORMAL FIELD THEORY c 24 π γ 2 1− δ(2) (w) = −Tzz (15. we get a measure of c. translational invariance implies that: ∂T + ∂Θ = 0 (15. 0) = (15. we deﬁne: F (zz) z4 G (zz) T (z. 0) = 2 2 z z T (z. .2 Zamolodchikov’s c-theorem The interpretation of the central charge as a measure of the number of degrees of freedom in a theory has an interesting corollary. of how many degrees of freedom there are in the system. hence.64) . By unitarity.256 CHAPTER 15. z) Θ(0. .61) This gives a ﬁnite contribution to ∂F/∂(ln L): F = . one can hope to compute all of the correlation functions of the theory using the techniques which we will discuss in the following sections – which would be a much more involved numerical calculation. 15. Meanwhile. (15. If the theory is soluble by the Bethe ansatz. something which the Bethe ansatz solution does not give. − cγ 24 1− π γ 2 ln L + . The resulting central charge might enable one to compute the correlation functions of the theory.63) The forms of the right-hand-sides are dictated by rotational invariance.62) By measuring how large a ln L contribution there is to the free energy.. and. F (zz) ≥ 0. but doesn’t know what low-energy conformal ﬁeld theories are associated with its critical points. Then. If one has a lattice model.7. then one could compute the free energy numerically and see how it scales with system size. H (zz) ≥ 0. 0) = z3z H (zz) Θ(z..

The following . • C(t) is a monotonically decreasing function of t since: d C(t) = − 12 H(t) < 0 dt (15. i.theorem should hold in higher dimensions. there can’t be limit cycles or other exotic ﬂows. the interpretation of the c-theorem as a statement about the information loss (or entropy gain) associated with course-graining is probably a little too naive. 0) and taking the correlation function. where t = ln(zz). This is sometimes justiﬁed by saying that the renormalization group procedure involves course graining a system. Thus. An RG transformation takes a system with cutoﬀ Λ and transforms it into another system with the same cutoﬀ Λ.65) ˙ ˙ where F ≡ zz F ′ .66) If we deﬁne the quantity C(t) = 2F (t) − 4G(t) − 6H(t). However. No such theorem has been found. of course.e.e.) Thus. Similarly.15. so ‘information’ should be lost in this procedure. related by a rescaling. and C(t) decreases along the ﬂows. i. A little further thought suggest that another hole in this intuitive interpretation of the c-theorem.7. Θ = 0. then we can immediately make two observations about C(t): • C(t) = c when the theory is critical. (The two systems are. there are precisely the same number of degrees of freedom before and after an RG transformation. INTERPRETATION OF THE CENTRAL CHARGE 257 Multiplying the left-hand-side of this conservation law by T (0. renormalization group ﬂows always go from ﬁxed points of large central charge to those of smaller central charge. which would imply that RG ﬂows should ﬂow ‘downhill’ according to some measure of ‘information’ or entropy. F = dF/dt. One consequence is that RG ﬂows are necessarily gradient ﬂows in unitary 2D theories. 0=∂ G (zz) H (zz) +∂ z3z z2z2 ˙ − G + H − 2H ˙ =G (15. 0) and take the correlation function. we ﬁnd 0=∂ F (zz) G (zz) +∂ 4 z z3z ′ ′ = zz F + zzG − 3G ˙ ˙ = F + G − 3G (15.67) Hence. this would imply that some version of the c. we multiply the conservation law by Θ(0.

In such a case. by their thermal conductivity or speciﬁc heat. their number is reduced and C(t) decreases. RUDIMENTS OF CONFORMAL FIELD THEORY is a more accurate interpretation.69) (15. as may be seen from the commutation relation: [L0 . it is not so clear in D > 2 that a low-energy theory will have fewer gapless modes than its high-energy parent. We will use these raising and lowering operators to build up representations of the Virasoro algebra. Ln is a lowering operator. In more than two dimensions. Thus. Ln ] = −nLn (15. Since the energy is bounded below. . One could imagine having N species of fermions with SU (N ) symmetry which form bilinear order parameters which condense. The latter term is suggestive of a primary ﬁeld. Hermiticity implies that L† = L−n since T (z) = Ln z −n−2 in radial n quantization implies that T (x. there must be some state |h in the representation such that Ln |h = h|h Ln |h = 0 ∀n>0 (15. 15. Let us consider some representation. Since the right.and left-handed parts of the energy-momentum tensor decouple. which results in Goldstone modes.70) Such a state is called a highest weight state or a primary state.68) For n > 0. we can separately consider L0 which is E + P . t) = Ln e−in(x−t) in the original Minkowski space. as measured by their contribution to the T − T correlation function or. The c-theorem may be nothing more than the statement that systems like to form gaps by allowing the gapless degrees of freedom to interact and form gaps. loosely speaking. Thereby. which is plays the role of the Hamiltonian for the right-handed part of the theory. For n < 0. The function C(t) is the number of lowenergy degrees of freedom. The Ln s are raising and lowering operators. Ln is a raising operator.258 CHAPTER 15. there would be N 2 − 1 Goldstone modes. it is possible to generate new gapless degrees of freedom by breaking a continuous symmetry.8 Representation Theory of the Virasoro Algebra L0 + L0 is the Hamiltonian. thereby completely breaking the SU (N ) symmetry. so the number of gapless modes – by any measure – would have to increase at low-energies for N suﬃciently large. to which a primary state correspond.

there is a simple correspondence between states and operators. we see that Ln 1 |0 = Ln |0 = 0 for all n > 0 since the identity .72) Acting on the vacuum state and expanding T (z). Another perspective on this condition on the vacuum state may be gained by considering the trivial OPE: T (z) 1 = T (0) + z∂T (0) + z2 2 ∂ T (0) + . z z h 1 |h + ∂Φh (0)|0 + non-sing. we expect the vacuum state to be invariant under the global symmetries of the theory.76) Thus.8. we see that Ln Φh (0)|0 = Ln |h = 0 and L−1 Φh (z) = ∂Φh (z) (15.77) ∀n>0 (15. we have T (z) Φh (0)|0 = 1 h Φ (0)|0 + ∂Φh (0)|0 + non-sing.73) (15. z = 0 is actually t = −∞. In this case. We act on the vacuum with with the operator φ(z) at the origin z = 0 |φ ≡ φ(0)|0 (15. this means L0 |0 = L±1 |0 = 0.71) In radial quantization.74) 2 z z n>0 Ln z −n−2 |h h |h + z2 + Ln z −n−2 |h = n<0 Comparing the right. according to (15.and left-hand sides of the last line. so this state should be thought of as a state with a single φ quanta in the distant past. as advertised.15. 2 (15. primary states are created by primary ﬁelds. Implicit in this derivation. .75) where the identity operator is taken to be nominally at the origin. As usual. From this OPE. We deﬁne |h ≡ Φ(0)|0 Then. (15.36) T (z) Φh (0) = 1 h Φ (0) + ∂Φh (0) + non-singular terms 2 h z z (15. Suppose that Φh (z) is a primary ﬁeld of weight h. was the notion of a vacuum state |0 . 2 h z z h 1 T (z)|h = 2 |h + ∂Φh (0)|0 + non-sing. . REPRESENTATION THEORY OF THE VIRASORO ALGEBRA 259 In conformal ﬁeld theory.

has vanishing inner product with itself and all other states. |ψ is given by |ψ = L−m1 .74). we saw that L−1 Φh = ∂Φh By keeping the less singular terms of (15. we see that L−2 |0 = T (0)|0 and. as required by global conformal invariance. . . . For each primary ﬁeld. L−4 |h −1 −1 −2 . If the theory has a fnite number of primary ﬁelds. . L−2 |h −1 h+3 L3 |h . . . Any given conformal ﬁeld theory will be characterized by a central charge c. L−mn |h for some −m1 . L−1 L−3 |h . From our earlier discussion of the identity operator. . L2 |h . Φhn . . . A primary descendent state. L−mn ) |χ = ψ| (Lm1 . it is called a rational conformal ﬁeld theory..260 CHAPTER 15. . It will contain some number of primary ﬁelds. . . Then. the inner product ψ|χ = ( ψ|L−m1 . more generally. The actual vacuum is then simply one of these states. . Lmn |χ ) = 0. . L2 L−2 |h . The ﬁelds to which they correspond are descendent ﬁelds. The states which are obtained from the primary states by acting with the Ln s are called descendent states. This is one way of understanding why it is not a primary ﬁeld. . L−3 |h −1 h+4 L4 |h . . L−1 L−2 |h . .74). we can ﬁnd the other descendents of Φh : Φ−n (w) ≡ L−n Φh (w) = h 1 dz T (z)Φh (w) 2πi (z − w)n−1 (15. −mn < 0. A primary state |χ is deﬁned by the condition Lm |χ = 0 for all m > 0. a state which is simultaneously a primary state and a descendent state. Φh1 .78) The descendent state can thus be obtained by acting on the vacuum with L−n |h = Φ−n (0). so it should be set to zero in any unitary representation of the Virasoro algebra. . In (15. we have a sector of Hilbert space which is built on the highest weight state |hn by acting with raising operators. RUDIMENTS OF CONFORMAL FIELD THEORY is a primary ﬁeld. . Finally. Descendent states must be orthogonal to primary states. it is useful to think of the primary states as ‘vacuum’ states of sectors – each sector corresponding to a diﬀerent primary ﬁeld – of the Hilbert space. i. we h see that the energy-momentum tensor is a descendent of the identity. the one corresponding to the identity operator |0 = 1 |0 . We further see that L0 |0 = L−1 |0 = 0. Φh2 . L−n |0 = ∂ n−2 T (0)|0 . For some purposes. A descendent state.e. These states form a Verma module: L0 eigenvalues States h |h h+1 L−1 |h h+2 L2 |h .

REPRESENTATION THEORY OF THE VIRASORO ALGEBRA 261 Each Verma module is a single irreducible representation of the Virasoro algebra. so it imposes far more structure on Hilbert space. which is not enough to determine the spectrum of a theory with an inﬁnite-dimensional Hilbert space.15. . the Verma module built on the vacuum state |0 . If there are inﬁnitely many primary ﬁelds. The simplest example of this is the Verma module built on the identity operator – or. in particular. the Virasoro algebra reduces the problem of determining the spectrum of the theory to that of determining a ﬁnite set of numbers h1 . L0 . The latter was merely an oﬀset from zero by which the entire spectrum was rigidly shifted. L−1 |0 = 0. the Virasoro algebra is a spectrum-generating algebra of the theory. This should not be surprising since the Virasoro algebra is inﬁnite-dimensional. hn . Rather. SU (2) symmetry introduces some simpliﬁcation. In fact. be many of them for each s. these commutation relations require that some of the states in a Verma module with a given L0 eigenvalue are not linearly independent. if the theory is a rational conformal ﬁeld theory) of such representations. The Virasoro algebra is inﬁnite-dimensional. There are inﬁnitely many states in each representation. C) global conformal transformations so. there is no state with L0 eigenvalue 1 in the vacuum sector of the theory. Thus. The vacuum state of a conformal ﬁeld theory should be invariant under SL(2.8. spins-1 representations. it turns out that for some c. in other words. then this may still not be enough to render the theory tractable. Our discussion of Verma modules appeared not to depend at all on c or h. there is only one state with L0 eigenvalue 2 since .. a† for the simple harmonic oscillator. analogous to a. h. in general.e. . This is misleading. Hence. Ln ] commutation relations. etc. As a consequence. therefore. the algebra of inﬁnitesimal conformal transformations – is not a symmetry algebra of the theory in the usual sense (i. The operators in this algebra do not commute with the Hamiltonian. in the sense in which we used it above for SU (2)). some Verma modules are smaller than one would naively expect from the above construction. Furthermore. for the case of rational conformal ﬁeld theories. A theory which is invariant under the SU (2) symmetry of spin roations will have a Hilbert space which can be broken into irreducible representations of SU (2): spin-0 representations. It is useful to keep in mind the analogy with SU(2). A spin-s representation will be 2s + 1-dimensional and there will. Note that the Virasoro algebra – and. h2 . We have not imposed any of the commutation relations other than the [L0 . but only a ﬁnite amount of simpliﬁcation. However. spin-1/2 representations. The Hilbert space of any conformal ﬁeld theory will contain some number (a ﬁnite number. .

so the correlation function must vanish. More generally. and the condition L1 |0 = 0. we remove such states from the Hilbert space which we are constructing. there will be a contribution proportional to ǫα1 α2 ǫα3 α4 and a contribution proportional to (σ y σ)α1 α2 · (σ y σ)α3 α4 . This correlation function will only α be non-vanishing if these ﬁelds are taken in some spin-singlet combination. ↓ are spin-1/2 ﬁelds. we will use the constraints imposed by the Virasoro algebra for a given c on a representation with a given h to ﬁnd other null states. φinn |0 α α α (15. at every level. for n = 4. It is useful to consider the analogy with SU (2). for n = 2. we can focus on primary operators because the correlation functions of descendent operators can be obtained from them. Knowing the spectrum of a quantum ﬁeld theory is not a full solution of the theory. for n = 3. In a similar way. In order to have a unitary representation of the Virasoro algebra. let’s pause for a moment to see why this is so important. there is no way of making an invariant combination. Before we do so. In a conformal ﬁeld theory. the result is proportional to ǫα1 α2 .82) . Suppose that we have the correlation function 0|φi11 φi22 .262 CHAPTER 15. RUDIMENTS OF CONFORMAL FIELD THEORY L2 |0 = 0. we also need the matrix elements of the operators of interest.81) where the φi . For example. we need to know how to decompose the tensor products of irreducible representations of the Virasoro algebra into a sum of irreducible representations. we can consider correlation functions of ﬁelds of arbitrary spins. again. Indeed. In short. and so on. L−1 ] = 2L0 . α =↑. C) invariance of the vacuum. . A state with vanishing norm is called a null state.79) Using the commutation relation [L1 . . which can be generalized to other primary ﬁelds is by condiering the inner product |L−1 |0 |2 = 0|L1 L−1 |0 (15.80) since the vacuum state has h = 0. we simply need to know how to construct invariants. there are fewer states than one would −1 naively expect. In order to compute correlation functions. In order to compute correlation functions of primary operators. there is a simple decomposition j1 ⊗ j2 ⊕j1 +j21−j2 | j3 j3 =|j (15. Another way of deriving it. In the case of SU (2). we can determine the spin structure of this correlation function if we know how to construct SU (2) invariants out of tensor products of spin-1/2 ﬁelds. The condition L−1 |0 = 0 follows from the SL(2. we see that |L−1 |0 |2 = 0|2L0 |0 = 2h = 0 (15.

we can write this as Φ1 (z1 ) Φ2 (z2 ) Φ3 (z3 ) Φ4 (z4 ) = k c12k (z1 − z2 )hk −h1 −h2 Φk (z2 ) × n c34n (z3 − z4 )hn −h3 −h4 Φn (z4 ) Φk (z2 ) Φn (z4 ) (15.15. we need to know whether the OPE of Φ1 and Φ2 contains Φ3 . and are. Consider the four-point function Φ1 Φ2 Φ3 Φ4 . h2 ) and one or both of them have null states. how can we determine the cijk s? If we knew more about the theory. Φ1 Φ2 Φ3 . to a sum over primary ﬁelds.8. unmanageable. If we had additional symmetries in the theory. then we might be able to deduce which primary ﬁelds appear in the operator product of two others. suppose that we wish to proceed purely algebraically. hence. This is still unmanageable if there are inﬁnitely many primary ﬁelds. Using the OPE of Φ1 with Φ2 and the OPE of Φ3 with Φ4 . h1 ) and (c. keeping only the spin j = 0 piece after the last step.83) = c12k c34n hk −h1 −h2 hn −h3 −h4 z34 k. h3 ). the sums on the right-hand-side run over all ﬁelds in the theory. we know that we need h1 = h2 . In these theories. If we take the tensor product of two representations of the Virasoro algebra (c. Even when there is a ﬁnite number of primary ﬁelds. In order to calculate correlation functions in a conformal ﬁeld theory. In the case of three-point functions.n z12 In principle. hk ) which can appear in their product. for instance if we knew its action and had some sense of how the diﬀerent primary ﬁelds were related physically. the sum on the right-hand-side can be reduced to a ﬁnite sum and some progress can be made. h1 ) and (c. formally. we need to know how to multiply representations together in this way in order to get invariants. In such a case. Theories with a ﬁnite number of primary ﬁelds are called rational conformal ﬁeld theories. However. knowing only c and h and no further information about the theory. we must take advantage of the existence of null states. then this limits the representations (c. h2 ) contains (c. In the case of two-point functions Φ1 Φ2 . REPRESENTATION THEORY OF THE VIRASORO ALGEBRA 263 By applying this relation n − 1 times to a correlation function of n ﬁelds and. However. This is is essentially the question of whether the product of the representations (c. . and thus reduce the sum. we can group each primary ﬁeld with its descendents. If we know this. ﬁnally. then we can compute the correlation function because the spatial dependence is essentially determined by conformal invariance. we obtain the spin structure of the correlation function. then we might be able to derive further restrictions.

we need to determine the primary ﬁelds of the theory. By reducing the size of the Verma module. there may be a consistent OPE involving some subset.264 CHAPTER 15. it will require the existence of negative norm states. Note that. the uses of negative norm states and null states are diﬀerent: the former determines the allowed hs. Hence. In general. as we saw in the previous section. However. so that there’s a null state at this level. and at certain speciﬁc levels which are predicated by those of (c. the Virasoro algebra will not allow unitary representations for most values of h. h2 ). L2 |h = 0 and L−2 |h = 0. Then. two diﬀerent conformal ﬁeld theories with the same c could include diﬀerent subsets of the allowed hs. Here. From a logical standpoint. the existence of null states is a boon. the decomposition of the products of these hs. Let us see how null states can be used in a particular example. for certain values of c. these constrain the OPE. at these values of c. the spectrum of hs. Suppose that they are not −1 linearly independent. the latter. It may turn out that the OPE requires that all of the allowed hs must actually be in the theory in order for the OPE algebra to close. and discuss them somewhat separately and in the reverse of the normal order. If h = 0. hk ) must also contain null states. 15. which will turn out to be the Ising model. in principle. we will eschew this approach. However. which corresponds to the identity operator.9 Null States In order to solve a conformal ﬁeld theory with a given central charge. h1 ) and (c. let’s consider some representation in this theory with h = 0. Consider the two states at level 2. they are usually discussed together because the existence of both can be derived with one fell swoop. and also how these diﬀerent irreducible representations are tensored together.84) . Thus. then there is a null state at level one since L−1 |0 = 0. we can ﬁgure out the allowed primary ﬁelds in the theory since they correspond to those values of h ate which a unitary representation is possible. c. there is no way to solve either problem purely algebraically. L−2 |h + aL2 |h = 0 −1 (15. We consider a theory with c = 1/2. RUDIMENTS OF CONFORMAL FIELD THEORY (c.e. In order to solve the second problem – how to decompose the tensor product of representations – we can. not a nuisance. alternatively. Now. i. make progress in those fortunate situations in which the allowed representations have null states. as we discussed in the previous section.

the physical ﬁelds are left-right symmetric cominations. L2 −1 |h = (3L−1 + aL−1 (2L0 ) + a (2L0 ) L−1 ) |h = (3 + 2a(2h + 1)) |h Hence.86) (15. we can obtain the critical exponents η and ν from the two-point functions of the dimension 1/16 and 1/2 operators. then it must also be true that 0 = L2 L−2 |h + aL2 |h −1 = 0 + [L2 . the state (15. L2 −1 |h c = 4L0 + + aL−1 (3L1 ) + a (3L1 ) L−1 |h 2 c = 4h + + 6ah |h 2 Substituting the value of a obtained above. so these dimensions are eﬀectively doubled by the anti-holomorphic dependence. we ﬁnd 0 = L1 L−2 |h + aL2 |h −1 = 0 + [L1 . indeed. the well-known critical exponents of the 2D Ising model.84) is a null state only if a=− 3 2(2h + 1) (15.85) If this state is null. σ(z. L−2 ] + a L2 . ε(z. Acting on this with L1 . These are. and using the commutation relations of the Virasoro algebra together with the fact that L1 |h = 0. .87) For c = 1/2. ε in the Ising model.89) from which we see that η = 1/4. L−2 ] + a L1 . NULL STATES 265 for some a. 1/2.15.9. 0) ∼ 1 1 ∼ 2(d−1/ν zz r (15. these correspond to the spin ﬁeld. According to this identiﬁcation. In the Ising model.88) (15.90) from which we see that ν = 1. As we will see. z) σ(0. Similarly. 0) ∼ 1 1 ∼ η 1/8 r z 1/8 z (15. σ and the energy operator. this implies that 16h2 − (10 − 2c)h + c = 0 (15. this has the solutions h = 1/16. z) ε(0.

95) (z) k k−1 ∂w wj − z) (wj − z) j Applying this. we begin by noting that the following correlation function. we obtain φ1 (w1 ) n−1 − j=1 . . However. φn−1 (wn−1 ) L−k φn (z) (15. .93) would vanish if we move the L−k to the left.78). this means that φ1 (w1 ) . . . φn−1 (wn−1 ) φn(15. in conjunction with (15. this will give us a diﬀerential equation satisﬁed by the four-point function.266 CHAPTER 15. Consider.94) can be simpliﬁed by taking breaking the contour into small circles encircling each of the wi s. the four-point function of the spin ﬁeld σ. . As we will see in a moment. .91) +1 L2 −1 σ=0 (15.96) = 0 . We can use the existence of null states is the h = 1/16 and h = 1/2 representations. φn−1 (wn−1 ) 1 dz T (z ′ )φn (z) 2πi (z ′ − z)n−1 (15. .92) to the σ four-point function. φn−1 (wn−1 ) L−k φn (z) = (1 − k)hj 1 ∂ + φ1 (w1 ) . φ1 (w1 ) . Thus. where it could act on the vacuum state and annihilate it. Since L−2 − it is equivalently true that L−2 − 3 2 2· 1 16 3 2 2· 1 16 +1 L2 −1 |1/16 = 0 (15. in the process of moving it to the left.92) when σ(z) is inside of some correlation function. it must commute with the φi (wi )s. we have n−1 2 ∂ 1 1/16 4 ∂ − + σ (z1 ) σ (z2 ) σ (z3 ) σ (z4 ) 2 2 3 ∂zi z4 − zj ) z4 − zj ∂zj j=1 (15. RUDIMENTS OF CONFORMAL FIELD THEORY Suppose that we want to compute a more non-trivial correlation function in the critical Ising model. Using equation (15. . In order to derive this diﬀerential equation. for example.

contains only descendent ﬁelds. z 2 ) σ (z3 . z 2 ) + . In order to get a single-valued result.(15.101) At this stage.15. as we will see later. we do not yet know whether there are other primary ﬁelds in the c = 1/2 theory. we can determine a1 .2 (x) = 1 ± √ 1−x 1/2 (15. could. but equivalent to it by translational invariance. . so the . z 4 ) = 1/4 √ √ z13 z24 x 1 + 1 − x + 1 − 1 −(15. . NULL STATES 267 In the ﬁrst term on the left.95).96). a2 as well as determine the OPE of σ with itself by considering the behavior of this correlation function in various limits.99) This is clearly a multiple-valued function. If we take z12 → 0 and z34 → 0 in the four-point function and use these OPEs. we must combine the holomorphic and anti-holomorphic parts of the theory. Using global conformal invariance. let’s conside the OPE of σ with itself: σ (z1 . . Substituting this form into the diﬀerential equation (15. . z 1 ) σ (z2 . in principle include both primary and descendent ﬁelds. First. z 2 ) = 1 |z12 | 1/4 + Cσσε |z12 |3/4 ε (z2 . . we have used L−1 φ = ∂φ which is simpler than (15.97) where x = z12 z34 /z13 z24 . The only way of doing this is by taking the left-right symmetric combination f1 (x)f1 (x) + f2 (x)f2 (x): σ (z1 . z 1 ) σ (z2 . we have σ (z1 . we can write σ (z1 ) σ (z2 ) σ (z3 ) σ (z4 ) = z13 z24 z12 z23 z34 z41 1/8 F (x) (15. we have the ordinary diﬀerential equation x(1 − x) ∂2 + ∂x2 1 −x 2 ∂ 1 + ∂x 16 =0 (15. so the .98) This equation has two independent solutions. In fact. .9. z 3 ) σ (z4 . there are none. z 4 ) = .100) a z12 z23 z34 z41 for some a. Now. We have suppressed the dependence on z for simplicity. f1. z 1 ) σ (z2 . z 3 ) σ (z4 . z 2 ) σ (z3 .

RUDIMENTS OF CONFORMAL FIELD THEORY 1 |z12 | 1/4 1 |z34 |1/4 2 + Cσσε |z12 |3/8 |z12 |3/8 ε (z2 . z 4 ) + .1 = 1/2 are the null states .104) ( = det where hp. At level 2. (As we will discuss in the next part of the book. taking the same limit in (15.268 CHAPTER 15. we see that a = 2 . . systems with quenched random disorder are described by non-unitary ﬁeld theories.. which must propagate to higher levels. z 2 ) ε (z4 . . comparing the leading terms. h) < ( 1 . it would be non-unitary. 1/16. = 1 1 |z12 |1/4 |z34 |1/4 2 + Cσσε |z12 |3/4 |z12 |3/4 |z24 |2 + . we also see that Cσσε = 1 . 1/2 representations are the only ones in the c = 1/2 theory. We also note that there are no operators 2 with (h.102) Meanwhile.1 (c)) × (h − h1. h1. z 2 ) σ (z3 .q = [(m + 1)p − mq]2 − 1 4m(m + 1) 1 1 25 − c m=− ± 2 2 1−c (15. 15.103) z12 z34 2 z24 2 z24 1 Hence. 1 ) appearing in this OPE (which would be natural if there 2 2 were no other primary ﬁelds in the theory).10 Unitary Representations In fact.100).2 (c)) (h − h2. so the requirement of unitarity does not help us there. the h = 0. At c = 1/2.1 = 0 corresponds to the null state at level 1.) Let us see why this is so.105) h1. z 1 ) σ (z2 . Comparing the next terms. the existence of a null state can be determined by taking the determinant of the matrix det h|L2 L−2 |h h|L2 L2 |h −1 h|L2 L−2 |h 1 h|L2 L2 |h 1 −1 4h + c/2 6h 6h 4h(1 + 2h) = 16h3 − 10h2 + 2h2 c + hc = 32 (h − h1. If we tried to construct a representation with any other value of h. so it should not arise in most physical theories. z 4 ) = 1 z12 z34 1 1/4 1 z12 z34 2− a + 2 2 (15.2 = 1/16 and h2. z 3 ) σ (z4 .1 (c)) 15. we have σ (z1 . (15..

this means that there is a single negative norm state at this level for this (c.104) detMN (c. Each of these null states leads to P (N − pq) null states L−n1 . but is nowhere negative. . h). h > 0. h. .106) gives the product of eigenvalues. . which generalizes (15. One can check that the null states constructed are the full set by comparing the highest power of h on both sides. the eigenvalues are strictly positive since they are just the diagonal elements (which are positive) in this limit. . the eigenvalues must remain positive. The determinant vanishes at these values. but the basic idea is to write down for each p. we had considered a representation at h = 1/4. Keep in mind that the factorization (15. h) = αN pq≤N (h − hp. Since the determinant never changes sign. For large h. L−nk |h + pq at level N (where P (n) is the number of partitions of n). Hence. m is not real so the hp. h > 0 not lying on the curves hp. When the determinant is negative. 4.q (c))P (N −pq) (15. There are null states.˜(c). For c = 1. It can be shown that there is a single negative norm state along these curves except at the “ﬁrst intersection” points where two of these curves intersect. This would imply that there are negatie norm states in the representations. so hp.q = (p − q)2 /4. h) are positive.10. in the c = 1/2 theory. If.. so there are unitary representations at all hs. We will not prove this formula here. at these hs but there are no negative norm states. h) can be connected to the region c > 1 by ˜˜ crossing precisly one of the curves hp. The only exceptions are the the curves hp. At these values of .q (c). m = ∞. the determinant never vanishes. but the factors in (15. up to an overall constant. UNITARY REPRESENTATIONS 269 at level 2. For 1 < c < 25. q an explicit null state at level pq.15. so that it could not be unitary. In order to ﬁnd the full set of restrictions on the allowed hs. for instance. then we would have found this determinant would be negative. A polynomial is determined. there are an odd number of eigenvalues.q (c) themselves. which leads to the result above. there exists some level N = pq such that the point (c. For c < 1.106) are not the eigenvalues themselves. Since an eigenvalue changes sign ˜q at these curves and the region c > 1 has only positive eigenvalues. .q s either have an imaginary part or are negative (the latter only occurs for p = q). by its zeroes. we must also consider the analogous determinant at levels 3. Let us now consider the question of whether the eigenvalues of detMN (c.106) where αN is a constant independent of c. which must be set to zero. The determinant of inner products at the N th level is given by the following formula.

Any other h would be a non-unitary representation of the Virasoro algebra for c = 1/2. m = 4 is the tricritical Ising model (in 4 − ǫ dimensions.270 CHAPTER 15. we can examine the three-point correlation functions to determine which primary ﬁelds can appear in the OPE of two primary ﬁelds. 15. and ′ the analogous rule holds for j3 .108) These theories are called ‘minimal models’. Such a relation is called a fusion rule: φp1 . This is almost the same as the decomposition of the product of the spin j1 and j2 representations in SU (2). there are only three possible primary ﬁelds in the c = 1/2 theory. m − 2 − j1 − j2 ).q3 (15. The case m = 3 is the Ising model.q1 × φp2 . q3 can appear on the right-hand-side of this ′ equation can be expressed most neatly by writing pi = 2ji + 1. qi = 2ji + 1. the only diﬀerence is that the upper limit j1 +j2 is replaced by min (j1 + j2 . As we claimed earlier. the action of a Dirac fermion ψ = in the form: S= = ψ † γ 0 γ µ ∂µ ψ † † ψR ∂ψR + ψL ∂ψL can be written (15. 1/16. as we mentioned earlier. Using techniques analogous to those which we used in our discussion of the c = 1/2 theory. Focussing on the chiral part of these theories. RUDIMENTS OF CONFORMAL FIELD THEORY c.109) The rule which speciﬁes which p3 . These crossings occur at a discrete series of c: 6 (15. the 2 tricritical point is at r = u = 0). this is a scalar ﬁeld with potential V (φ) = r φ2 + uφ4 + vφ6 . with h = 0. while m = 5 is the 3-state Potts model.q2 = φp3 . m − 2 − j1 − j2 ).110) .107) c=1− m(m + 1) At each such c.q (m) = [(m + 1)p − mq]2 − 1 4m(m + 1) (15. The origin of this rule will be clearer when we discuss Kac-Moody algebras. there is a ﬁnite set of allowed hs: hp. 1/2. we can compute the OPEs of ﬁelds in the minimal models. Then the allowed j3 s are |j1 − j2 | ≤ j3 ≤ min (j1 + j2 .11 Free Fermions ψR ψL In 1 + 1 dimensions. there are null states but no negative norm states. h.

FREE FERMIONS 271 where γ 0 = σx . one right-moving and the other left-moving. A right-moving Dirac fermion can be written as the sum of two rightmoving Majorana fermions which are its real and imaginary parts ψ = ψ1 + iψ2 . Thus. In radial quantization.. and the extra 1/2 in the exponent which results from the passage from the cylinder to the plane in radial quantization. the fermion splits into two independent ﬁelds. This is what one usually means by a real ﬁeld. z = eτ −ix . The right-moving ﬁeld satisﬁes the equation of motion ∂ψR = 0 while the left-moving ﬁeld satisﬁes ∂ψL = 0. essentially.111) † satisﬁes ψn = ψ−n . γ 1 = σy .) From the action (or. The inverse Fourier transform is: dz iψ(z) z n−1/2 (15. but it is somewhat masked by the fact that we’re in imaginary time.15.116) = 4 (z − w) (z − w)2 1 : ψ∂ψ : 2 (15. we can either have a right-moving fermion (a Weyl fermion) or a Majorana fermion..11. In general.115) . a right-moving real fermion is one whose mode expansion i ψ(z) = ψn z −n−1/2 (15. by scaling). (15. we can compute the OPE of a Majorana fermion: 1 + .112) ψn = 2πi The action for a right-moving Majorana fermion ψ(z) is S= ψ∂ψ (15.114) ψ(z) ψ(w) = − z−w The energy-momentum tensor is T =− The central charge is obtained from 1 −1 1 1 1 2 T (z)T (w) = (−1) − + + 3 2 (z − w)2 4 z − w (z − w) 4 (z − w) 1 T (w) 2 (z − w)2 2T (w) 1/4 + (15.113) (The fact that we can simultaneously diagonalize chirality and charge conjugation is special to 4k + 2 dimensions.

Physically measurable quantities are always bosonic. This should not come as an enormous surprise if one recalls our rewriting of the partition function of the 2D Ising model as a Grassman integral. while n ∈ Z gives antiperiodic boundary conditions. we need n ∈ Z + 2 in the mode-expansion (15. composed of two Majorana fermions. z → e2πi z. For periodic boundary conditions. Note that for anti-periodic boundary conditions. RUDIMENTS OF CONFORMAL FIELD THEORY The (−1) in the ﬁrst term on the ﬁrst line results from an anticommutation.0 (15.118) For antiperiodic boundary conditions. From the OPE (15. which are single-valued. we see that c = 1/2. Hence. ψ(z) ψ(w) = i2 =− =− =− ∞ ψn z −n−1/2 z −n−1/2 wn−1/2 w z n ψm w−m−1/2 n=1/2 ∞ 1 z 1 z−w n=0 (15. For 1 periodic boundary conditions. Accord2 ing to the canonical anticommutation relations. we can compute the anticommutator of the modes of ψ: {ψm . {ψ0 . From this calculation. Thus. a theory of a free Majorana fermion is the same as the critical theory of the Ising model.117) Fermionic ﬁelds can be either periodic or antiperiodic as one goes around the cylinder. we will discuss the zero mode operator a little further. ψn } = i2 = i2 dw m−1/2 w 2πi dw m−1/2 w 2πi dz n−1/2 z ψ(z) ψ(w) 2πi dz n−1/2 −1 z 2πi z−w = δn+m. ψ0 . there is a zero mode. has c = 1. a Dirac fermion. n ∈ Z.272 CHAPTER 15.114). A little bit later. ψ0 } = 1. Let us compute the fermion propagator in these two cases. or ψ0 = 1/2.111). so they must be composed of fermion bilinears. ψ(z) ψ(w) = i2 =− ∞ ψn z −n−1/2 ψm w−m−1/2 1 2 z −n−1/2 wn−1/2 + √ ψ0 zw n=1 .

AP (15. P = ∂w 0.120) As expected. We can introduce a ‘twist’ operator. hσ = 1/16. Thus. σ.125) 16 z 2 Hence. but the global analytic structure is diﬀerent. P = 0.123) We know the right-hand-side. is the spin ﬁeld of the Ising model. this will exchange periodic and antiperiodic boundary conditions. P |σ(∞) ψ(z) ψ(w) σ(0)| 0.124) ( =− (z − w)2 8 w2 AP z w (15. T by taking z → w and subtracting an inﬁnite constant −1/(z − w)2 . Thus. the short-distance behavior of the propagator is the same for both boundary conditions.15. T (z) = Thus. AP |ψ(z) ψ(w)| 0. when placed at z1 and z2 introduces a branch cut for the fermions which extends from z1 to z2 . 0. The twist ﬁeld. AP = − + w 1 1 z + 2 3/2 z 1/2 (z − w) 4w 1 1 1 + + O(z − w)15.11. we have 0. we have: 2 273 w 2 + ψ0 z−w (15. 0. which. The z 1/2 on the right-hand-side ensures that a minus sign results when ψ(z) is taken around σ(0).121) where µ(z) is an ‘excited’ twist ﬁeld of the same dimension as σ (we will 1 discuss this more later). AP |ψ(z) ψ(w)| 0. If we place twist ﬁelds at the origin and at ∞. AP Diﬀerentiating both sides with respect to w. P |σ(∞) ψ(z) ∂w ψ(w) σ(0)| 0. σ. AP |ψ(z) ∂w ψ(w)| 0. The left-hand-side can be re-expressed in terms of the energy-momentum tensor. FREE FERMIONS 1 =−√ zw 2 using ψ0 = 1 .122) 1 1 (15. . The OPE of a Majorana fermion ψ(z) with its twist ﬁeld σ(z) is: ψ(z) σ(0) = 1 z 1/2 µ(0) (15. our use of the notation σ is more than merely suggestive.119) ψ(z) ψ(w) = 1 2 + z−w z w w z (15.

RUDIMENTS OF CONFORMAL FIELD THEORY To summarize. 0). 1/16). . h = 1/16 Representation L0 eigenvalues States 1 1 | 16 16 1 1 ψ−1 | 16 1 + 16 1 1 ψ−2 | 16 2 + 16 1 1 3 + 16 ψ−3 | 16 . . . 0). . thereby leaving the state unchanged. σ(0)|0 . h = 1/16 Representation L0 eigenvalues States 1 1 ψ0 | 16 16 1 1 ψ−1 ψ0 | 16 1 + 16 1 1 ψ−2 ψ0 | 16 2 + 16 1 1 3 + 16 ψ−3 ψ0 | 16 . These representations are obtained by acting with the L−n s. However. h = 1/2 Representation h = 0 Representation L0 eigenvalues States L0 eigenvalues States 1/2 ψ−1/2 |0 0 |0 3/2 ψ−3/2 |0 1 2 ψ−3/2 ψ−1/2 |0 5/2 ψ−5/2 |0 3 ψ−5/2 ψ−1/2 |0 7/2 ψ−7/2 |0 . 1/2). . . In this sector. |0 . σ is the spin ﬁeld.274 CHAPTER 15. (0. The Ising model itself has both right. One of the two states is σ(0) |0 (zero mode unoccupied) while the other is µ(0) |0 (zero mode occupied). (1/2. . etc. on.and left-moving Majorana fermions. 1/2). The constraints associated with consistently . or order operator. . . One might naively think that its Hilbert space could have representations of all possible right-left combinations. respectively. there are actually two states with L0 eigenvalue 1/16 because there is a zero mode ψ0 . (1/16. of course. . and two copies of the h = 1/16 representation. this is not the case. (1/16. . . ψ−1/2 |0 . ψ0 σ(0)|0 = µ(0) |0 . This zero mode can be either occupied or unoccupied. . . up to a factor of 1/2. In fact. ψ is a linear combination of a creation and an √ annihilation operator (with coeﬃcient 1/ 2). Thus. so acting twice with ψ0 will create and then annihilate (or the reverse if the zero mode were initially occupied) a fermion in this mode. the c = 1/2 theory of a free Majorana fermion has one copy of the h = 0 representation. . For a Majorana fermion. 0). This can be derived by direct computation on the Ising model. (1/16. . . . antiperiodic boundary conditions are simply the sector of the full c = 1/2 theory given by h = 1/16 Verma modules. . . there are only the right-left symmetric combinations: (0. one copy of the h = 1/2 representation. . while µ is the disorder operator to which it is dual. It can also be obtained by putting the theory on a torus.

these operators are among the primary ﬁelds of 2 2R . Now. In fact. as claimed. 0) and (0. the OPE with the energy-momentum tensor. (z − w)2 (15.. In order to derive the full set of primary ﬁelds. but not all exponential operators are singlevalued under φ → φ + 2πR. FREE BOSONS 275 putting a theory on a torus (‘modular invariance’) restrict the allowed rightleft combinations of the set of possible (h. By Wick’s theorem. the operator : eiαϕ : is a dimension α2 /2 primary ﬁeld. 1) ﬁelds ∂ϕ and ∂ϕ.126) Hence. This theory has (1.12 Free Bosons 1 Earlier. Consider operators of the form : eimφ/R :=: eim(ϕ+ϕ)/2R :..12. : eiαϕ : is a dimension α2 /2 operator.. is: − 2 iα 1 1 : ∂ϕ(z)∂ϕ(z) : : eiαϕ(0) : = − + . h)s at a given (c. In such a case.. Let us now consider the exponential operators : eiαϕ :. They are clearly well-deﬁned.15. it is a primary operator. as we now show. the only allowed combinations are the symmetric ones. As we will see shortly. we considered a theory of a free boson. ∂ϕ and ∂ϕ are still ﬁne. Thus.129) . . 15. . suppose that φ is an angular variable φ ≡ φ + 2πR. they are not the only primary ﬁelds in such a theory. In the case of the Ising model. z (15. which has c = c = 1. h) = 1 m 2 1 m 2 . and have dimensions (h. 2 2 z−w α2 /2 = + . These exponential operators have the following OPE with i∂ϕ i∂ϕ(z) : eiαϕ(0) : = α : eiαϕ(0) : + . it is useful to consider the mode expansion for a system of free bosons: i ∂ϕ(z) = n αn z −n−1 (15.127) Hence. 2 2 : eiαϕ(z) : : eiαϕ(0) : = eα ( ϕ(z)ϕ(0) − ϕ(0) ) 1 = α2 z (15. c). T (z).128) Thus. φ = 2 (ϕ(z) + ϕ(z)). 2 2R the theory.

commutes with itself. The zero modes ϕR. we can go back to the cylinder with coordinates x. τ ) = φ0 + (α0 + α0 ) t+ (α0 − α0 ) x+ 2 2 1 αn enτ −inx + αn enτ +inx n (15.L can be derived from the OPE of ϕ.130) The commutators of the modes are: [αm . (α0 − 0 α0 )] = 0.276 CHAPTER 15. and φ itself: ϕ(z) = ϕR + α0 ln z + 0 n=0 1 αn z −n n 1 αn z −n n 1 2 1 αn z −n + αn z −n (15. but not from the exponential operators : eimφ/R :.∂ϕ: 0 ϕR . τ with z = eτ −ix .132) n ϕ(z) = ϕL 0 + α0 ln z + n=0 φ(z.134) n=0 . α0 = i.133) By similar steps. Consider the mode expansions of ϕ. z) = φ0 + α0 ln z + α0 ln z + n=0 where φ0 = (ϕR + ϕL )/2. ϕ with ∂ϕ. ϕ. Hence. the zero mode.L disappear from ∂ϕ and 0 0 0 ∂ϕ.131) As a result of the n on the right-hand-side. α0 = 0 = =i dz ϕ(z)∂ϕ(0) 2πi dz i 2πi 2z (15. RUDIMENTS OF CONFORMAL FIELD THEORY with αn = dz n z i ∂ϕ(z) 2πi (15. as it had better. [φ0 . Alternatively. αn ] = i2 dz n dw m w z ∂ϕ(z) ∂ϕ(w) 2πi 2πi dw m dz n −1 = i2 w z 2πi 2πi (z − w)2 dw m w nwn−1 = 2πi = n δn+m.0 (15. α0 . ϕL . The commutation relations of ϕR. the mode expansion is: 1 1 φ(x. Then. (α0 + α0 )] = i while [φ0 .

Since T (z) = − 1 : ∂ϕ∂ϕ :. Let’s deﬁne a set of states |α. ∂0 φ(0.137) n For m = 0. deﬁned by α0 |0 = 0 . α 2 (15. L0 eigenvalues are: 1 1 L0 |α.136) With this commutation relation in hand. (α0 + α0 )] = i 277 (15. we see that α0 commutes with the Hamiltonian. The simplest is the vacuum state itself. as we show below. α = α|α. we can label states by their α0 . α = α|α. let us turn to the energy-momentum tensor and its eigenvalues. α α0 |α. [φ(x. α = α2 |α. equivalently. Their L0 . αm−n and αn commute.12. α =0 for n > 0 =0 for n > 0 (15. with the Ln s for n < 0 and also with their anti-holomorphic counterparts. α 2 0 2 1 L0 |α. according to the canonical commutation relations for φ. α are highest weight states. in fact. FREE BOSONS Meanwhile. α0 eigenvalues.131). Hence. There is a similar expression for the anti-holomorphic modes. α = α2 |α.139) These highest weight states are created by exponential operators. Verma modules. Now. and the states |α. α by α0 |α. These are. so we do not need to worry about normal ordering in the second line. 2 L0 = Lm=0 = 1 2 n>0 1 α−n αn + α2 2 0 αm−n αn (15.135) (15. α = α2 |α.15. From the commutation relations (15.138) We can build towers of states on these by acting with the αn s for n < 0 or. we can derive the spectrum of α0 + α0 . α αn |α. τ )] = i δ(x) from which we obtain [φ0 . α αn |α. τ ).

the eigenvalues of (α0 + α0 ) are quantized. α0 . the highest weight states of a free boson with angular identiﬁcation φ ≡ φ + 2πR are: m m m m + nR.146) Thus. R Periodic boundary conditions around the cylinder require. the diﬀerent Verma modules of the theory correspond to diﬀerent values of the zero modes. according to (15.278 CHAPTER 15. (α0 + α0 ) = m . − nR = : ei( 2R +nR)ϕ ei( 2R −nR)ϕ : |0 2R 2R . acting with α0 .140) (15.145) or α0 = m + nR . φ ≡ φ + 2πR. |α. 2R α0 = m − nR 2R (15. consider the state : eiβφ(0) : |0 .143) 1 eiβφ0 α0 + eiβφ0 β |0 2 1 β : eiβφ(0) : |0 2 (15. Then. α . φ0 ≡ φ0 + 2πR.144) Let us ﬁnd the other primary ﬁelds in the case in which φ is an angular variable. that α0 − α0 = 2nR for some integer n.142) To summarize. α0 : eiβφ(0) : |0 = 1 β 2 : eiβφ(0) : |0 (15. we have m α0 + α0 = R α0 − α0 = 2nR (15. Putting these two together. Since [φ0 . while the n < 0 terms in the expansion vanish when we take z = 0.134). β/2 (15. so that φ → φ + 2πnR when x → x + 2π. Some of these highest weight states are created by the primary operators : eiβφ(0) : acting on the vacuum: : eiβφ(0) : |0 = |β/2. : eiαφ(0) : |0 = eiαφ0 |0 (15. (α0 + α0 )] = i. Thus. α0 : eiβφ(0) : |0 = α0 eiβφ0 |0 = = Similarly.141) the right-hand-side follows from the fact that αn |0 = 0 for n ≥ 0. RUDIMENTS OF CONFORMAL FIELD THEORY α0 |0 = 0 Now.

simply the product of the twist ﬁelds for the Dirac fermions ψR . m − n Let us divide these into the states in which m ∈ 2Z + 1 2 2 and those in which m ∈ 2Z.L = with m ∈ 2Z + 1 are of the form 1 L0 |nR . we see that † ψR ψR = e−iϕ(z) eiϕ(w) 1 e−i(ϕ(z)−ϕ(w)) = z−w .1 σL.1 + iψL. FREE BOSONS = : eimφ/R e2inRφ : |0 ˜ where we have deﬁned the dual ﬁeld φ = (h.1. AP = 2 1 nR + 2 2 m 2 |nR . nL . : e2inRφ(0) : creates an n-fold vortex at the origin of the complex plane.149) − nR 2 1 2 (ϕ .1 + iψR. ψR = ψR.2 = ψR ψL † = ψL ψR = σR. nR. AP with the obvious counterpart for L0 . σL. σL.148) ± n ∈ Z. Let us consider the special case R = 1. σR.147) + nR 2 1 . ψL .2 m 2R The action of the exponential of the dual ﬁeld.2 . nL . These are the primary states of a Dirac fermion. : e2inRφ : is to increase (α0 − α0 )/2 by nR. This causes φ to wind n more times around 2π as x winds around the cylinder or z winds around the origin in the complex ˜ plane.15.12. 8 8 From these formulae. nL .L.2 = ψL = ψL. These states have ˜ with the obvious counterpart for L0 . The states with m ∈ 2Z are of the form L0 |nR . nL .150) σR. h) = 1 2 m 2R ˜ 279 (15. If we make the identiﬁcations eiϕ eiϕ e2iφ ˜ e2iφ eiφ = ψR = ψR.2 σL. The same holds for the anti-holomorphic part of the theory.2 or.1 σR. with c = 1. Here. In other words. the primary states are m + n. 1 ). This product has dimension ( 1 .1 .2 are the twist ﬁelds for ψR.2 . The states (15. . At this radius. P = 1 2 n |nR . − ϕ). each with c = 1/2.2 (15.1 . which also has c = 1 since it is composed of two Majorana fermions.1 + iψR. P 2 R (15.

They satisfy some symmetry algebra Qa .) z−w 1 = + i∂ϕ z−w = Hence. In such a case. RUDIMENTS OF CONFORMAL FIELD THEORY 1 (1 − i(w − z)∂ϕ(w) + . which render the theory soluble. . The associated conserved charges are Qa = a dx J0 . The primary states correspond to diﬀerent nR and nL in the twisted and untwisted sectors. inﬁnitely many primary states. two conserved charges. (The commutator must equal a linear combination of charges because any state which is invariant under the symmetry transformations generated by the charges must be annihilated by the charges and also by their commutators. 2. . the symmetry generators form a Kac-Moody algebra. There is a big diﬀerence between a Dirac fermion and a Majorana fermion because the former has a conserved charge (or. rather. a a Consider a theory with conserved currents Jµ . 15. where the fabc s are the structure constants of some Lie algebra G. ∂ µ Jµ = 0. upon normal ordering.13 Kac-Moody Algebras We now consider conformal ﬁeld theories which have other symmetries – such as charge conservation – in addition to symmetry under conformal transformations. One slightly peculiar thing about this theory is that we started with two copies of the c = 1/2 theory. fabc = ǫabc . when these other symmetries mesh with conformal invariance in a certain way. rather.) In the case G = SU (2). one right-moving and one left-moving). . they lead to the existence of null states. nL = i∂ϕ. we don’t get 9 primary states but. This is not really so peculiar when we recall that we are now talking about primary under the c = 1 Virasoro algebra of the combined theory. As we will see. .280 CHAPTER 15. n label the currents. . we get the conserved quantity † nR =: ψR ψR := i∂ϕ (15. Qb = ifabc Qc . . while the latter does not.151) (15.152) similarly. which has only 3 primary states. where a = 1. . When we tensor these two copies of the theory together. corresponding to the (1) even and (2) odd fermion number in the untwisted sector and the (3) twisted sector (where even and odd fermion number are not distinguished because the zero mode moves one between them). not under the individual c = 1/2 Virasoro algebras.

15.153) ∂J = ∂J = 0 where J = J0 + iJ1 . .0 (15. As a result of the decoupling of the right. ∂ µ ǫµν Jν = 0. . in which they vanish. suppose that the dual current ǫµν Jν is also conserved. ϕ → ϕ + f (z). in turn. there is actually a much large set of symmetries of the classical equation of motion. The OPE of the currents is J(z) J(0) = − ∂ϕ(z) ∂ϕ(0) . z2 z (15. The charges QR = dzJ and QR = dzJ must be dimensionless since. 0) and (0. This. 1) ﬁelds. the right-moving and leftmoving charges are separaely conserved. respectively. ϕ → ϕ + f (z). The simplest example of a theory with a Kac-Moody algebra is the free boson. This OPE can be translated into the commutation relations of the modes a (z) = a −n−1 . The ﬁrst term is allowed by the global symmetry (generated by the Qa s) and by scaling.156) These are the currents associated with the global symmetry φ → φ+c. J n Jn z c a b Jm .154) The second term on the right-hand-side is dictated by the commutators of the Qa s. KAC-MOODY ALGEBRAS 281 a a Now. J = J0 − iJ1 . they commute with L0 and L0 .155) The m = 0 modes still satisfy the commutation relations of the Lie algebra G. There two conserved charges. where c is a constant. as conserved quantities. except in the Abelian case. (15. J(z) is holomorphic while J (z) is antiholomorphic. In other words. Jn = i f abc Jm+n + k m δab δm+n. The conserved currents are: J(z) = i∂ϕ J (z) = i∂ϕ (15. The most general form for the OPE for J a is J a (z) J b (0) = if abc c kδab ++ J (0) + .and left-handed parts of the theory. The normalization of the central extension k is ﬁxed by the normalization of the structure constants in the second term. Then.13. associated with the integrals of the current or the dual current. implies that J and J are dimension (1. In more physical terms.

2. More generally. . which also has a U (1) symmetry.R ∂ψA.157) There is no 1/z term. 2 z z i z ± z J (z) J (0) = ± J (0) + . which is promoted to a Kac-Moody symmetry with † † J = ψR ψR and J = ψR ψL . is ‘compactiﬁed’). 3 are Pauli matrices. irrespective of whether the boson is an angular variable (i. U ∈ SU (2). Such an algebra will be called SU (2)1 .282 CHAPTER 15. . This can be generalized straighforwardly to a theory of N Dirac fermions.R (15. (In this case.R (15.R where U † U = 1. a = 1. e±iϕ 2 = e±i(φ/R+2φR) . J ± = e±iϕ 2 and similarly for the antiholomorphic currents.R ταβ ψα. The associated currents are † a J a = ψα. This action is invariant under the symmetry ψα. an SU (2) Kac-Moody algebra at level k wil be called SU (2)k . √ ˜ Now. . .160) a where ταβ .e. there are additional currents. † ψA.158) Again. i. z (15. then these operators have the OPE expected for an SU (2) Kac-Moody algebra: J + (z) J − (0) = 1 2i + J z (0) + . . If we write J z = i∂ϕ. .) The special case of a boson with R = 1 is equivalent to a Dirac fermion. 1). as we would expect since the U (1) algebra is Abelian. 0). Gk . RUDIMENTS OF CONFORMAL FIELD THEORY = 1 + 2T (z) + . which have di√ ˜ e mension (1. √ A more interesting example is aﬀorded by a free boson at R = 1/ 2.159) where α = 1.e. and e±iϕ 2 = √±i(φ/R−2φR) which have dimension (0.R (15. The same algebra arises in a theory of 2 Dirac fermions. the level is k = 1. 2.161) . this is somewhat a matter of convention because there is no term on the right-hand-side linear in the currents so we can rescale J(z) to change k to whatever we wish. that associated with an arbitrary Lie algebra G. ψ → eiθ ψ. We focus on the holomorphic part for simplicity: † ψα.R → Uαβ ψβ. We see that k = 1.R ∂ψα. z2 (15.

2J0 2 1 ˜ ˜ ˜ dimensional unitary representation. β (15. I 3 ≡ .165) Not all representations (r) are allowed in the theory at level k. then this algebra realizes an SU (N )k Kac-Moody algebra since the level is clearly additive when we take k copies of a theory. α = T(r) αβ |(r). J a (z) = m Jm z −m−1 . For the sake of concreteness. Hence.155). primary ﬁelds create highest weight states. we see that I a ≡ J0 3 has integer eigenvalues in any ﬁniteform an SU (2) Lie algebra.R TAB ψB. Let us now consider the representation theory of Kac-Moody algebras. This theory is invariant under SU (N ) transformations. (15. KAC-MOODY ALGEBRAS 283 where A = 1. As usual. We will return to them later.13. Note that these fermionic theories all have U (1) Kac-Moody algebras as well. we consider the case of SU (2). α = ϕα |0 (r) (15. Similarly. 2. ψAR → UAB ψB. . They satisfy the SU (N )1 Kac-Moody algebra. A restriction on the allowed j’s in the SU (2)k theory can be found by considering the mode a a expansion of J a . especially in the k > 1 case. We deﬁne a primary ﬁeld ϕ(r) to be a ﬁeld which transforms in representation r of the group G and is primary under the Kac-Moody algebra. I 1 ≡ J1 . J (z) ϕ(r) (0) = a a T(r) ϕ(r) (0) z + . The bosonic representations of these Kac-Moody algebras are rather interesting and non-trivial. N . kN Dirac fermions. but if we focus on the SU (N ) subgroup of SU (kN ). The simplest way of obtaining a Kac-Moody algebra with a level other than 1 is to simply take k copies of this theory or.164) Here.163) a where the T(r) s are the matrices representing the generators in representation (r).R (15. The Jm s satisfy the commutation a relations (15. we have explicitly written the representation vector index α which we suppressed earlier for convenience. .162) a and the TAB s are the generators of the fundamental representation of SU (N ).R with associated currents † a J a = ψA. in other words. Such a theory has an SU (kN )1 symmetry. The highest weight states form a a multiplet under the global symmetry generated by the J0 s: a a J0 |(r). From these commutation relations. I 2 ≡ J−1 .. . .15.. |(r).

166) Hence. This satisﬁes the Virasoro algebra. Then 1 2 ˜ ˜ 0 ≤ j.168) z4 = = = = = with c= k|G| k + CA (15. . m = j| I + .284 CHAPTER 15. Consider T (z) deﬁned by T = 1/2 : Ja Ja : k + CA (15.167) where CA is the quadratic Casimir in the adjoint representation if the highest root is normalized to length 1. . . ˜ ˜ Since the I a s form an SU (2) algebra. . m = j . from which we conclude that k must be an integer. k + CA 2 1/2 kδab if abc c ×2× + J (0) + . × k + CA z2 z kδab if abd d J (0) + . + . T (z) T (0) = = = 2 1/2 : J a (z) J a (z) : : J b (0) J b (0) : k + CA 2 1/2 × 2 × (J a (z)J a (0)) J b (z)J b (0) + . Consider a spin j highest weight state |j. . m = j|I + I − |j. with I 3 |j. m = j ˜ ˜ = j. + z2 z k2 δab δab f abc f abd c d 1/2 − J J 2 z4 z2 (k + CA ) k2 |G| CA δcd c d 1/2 + J J z4 z2 (k + CA )2 1/2 k2 |G| CA δcd kδcd + z4 z2 z2 (k + CA )2 1 k|G| 1 2 k + CA z 4 c/2 (15. .169) . I 3 must have integer eigenvalues. the SU (2)k theory contains the representations j = 1/2. . . m = j = k − 2j (15. . One remarkable feature of a Kac-Moody algebra is that it automatically includes the structure of the Virasoro algebra. . . I − |j. RUDIMENTS OF CONFORMAL FIELD THEORY 3 k − J0 also form an SU (2) Lie algebra. m = j = j. . m = j = j|j. m = j|k − 2I 3 |j. k/2 and only these representations. 1. m = j .

172) a We act with this equation on a primary ﬁeld. this means that a spin j primary ﬁeld has dimension h = j(j + 1)/(k + 2).. = 2 z k + CA (15. . .163). z 2 k + CA Cr 1 + . .167). . The correlation functions of primary ﬁelds in a theory with Kac-Moody symmetry can be calculated using the Sugawara construction of T (z). . k + CA −1 0 (15. together with (15. a primary ﬁeld transforming under G as representation r has dimension T (z) ϕ(r) (0) = a a 1 T(r) T(r) /2 + . there is a null state given by L−1 − 1 J a T a ϕ(r) = 0 k + CA −1 (r) (15.13. the n = −1 term is: L−1 = 1 a a J a J a + J−2 J1 + .. using the fact that Jn for n > 0 annihilates primary ﬁelds.173) These ﬁrst-order diﬀerential equations can be solved to determine the correlation functions of primary ﬁelds. Expanding both sides of (15. The fact that the T given by the Sugawara construction satisﬁes the Virasoro algebra does not necessarily imply that it is the full energy-momentum tensor of the theory. + T (w) + 4 2 (z − w) (z − w) z−w (15.. According to (15.15.. .171) For the case of SU (2)k .167) in modes. ϕ(r ) (zn ) = 0 (15. KAC-MOODY ALGEBRAS 285 where |G| is the dimension of the Lie algebra. One can check that the next term in the OPE is as expected for the energy-momentum tensor.170) This immediately enables us to compute the dimensions of primary ﬁelds.174) (k + CA ) ∂ + n 1 ∂zk zj − zk j=k . It may only be the energy-momentum tensor of one This null state condition translates into the following diﬀerential equation – the Knizhnik-Zamolodchikov equation – for correlation functions of primary ﬁelds: a a T(rj ) T(rk ) ϕ(r ) (z1 ) . Hence. so that T (z) T (w) = 2 1 c/2 ∂T (w) + .

The WZW action is: k tr ∂µ U −1 ∂ µ U + SW ZW = 4π S 1 ×R . 2. a charge-1 primary ﬁeld under U (1) has dimension 1/4.286 CHAPTER 15. Meanwhile. For instance. Thus. Similarly. We now discuss them. In these theories. Thus. Earlier. consider a theory of kN Dirac fermions. with N = 2 and k channels. Since the group G is some curved space. we can think of our kN fermions as k sets of N fermions. cSU (kN )1 = kN − 1. There is not necessarily a unique way of decomposing T into Kac-Moody algebras. In the multichannel Kondo problem. The central charge associated with the SU (2) part of the theory is cSU (2) = 1. There is an SU (N )k symmetry among the N fermions (rotating all sets together) and also an SU (k)N symmetry 2 −1) 2 among the k sets. with T SU (2) = 1/2 : Ja Ja : k+2 (15. this proves to be a particularly useful decomposition. U . from (15. each of which obeys the Virasoro algebra. cSU (N )k = k(N −1) while cSU (k)N = N (k+k . takes values in some group G. a = 1. we alluded to the bosonic theories which posess Kac-Moody symmetry at level k > 1. cU (1) = 1. This theory has both U (1) and SU (2)1 Kac-Moody symmetries. T U (1) = 1 :JJ : 2 (15. The central charges add. with c = kN . 3. as expected. it has dimension 1/2. similar in spirit but diﬀerent in detail to. This can be decomposed into cU (1) = 1 and.176) The normalization is chosen so that the second term in the OPE of T U (1) with itself is 2T /z 2 . Since the level of the SU (2) Kac-Moody algebra is 1. A Dirac fermion carries charge-1 and spin-1/2. Consider our theory of two chiral Dirac fermions. a WZW model is an interacting ﬁeld theory. k+N N cU (1) + cSU (N )k + cSU (k)N = kN is an equally good decomposition of the theory into Kac-Moody algebras. say. RUDIMENTS OF CONFORMAL FIELD THEORY sector of the theory. the full energy-momentum tensor is then the sum of several tensors. a spin-1/2 primary ﬁeld has dimension 1/4. These theories are nonlinear σ-models of Wess-Zumino-Witten (WZW) type.175) since CA = 1(1 + 1) = 2. As a result. The total central charge cSU (2) + cU (1) = 2 is the central charge of two Dirac fermions.169) with |SU (N )| = N 2 − 1 and CA = N . let us consider such a theory on the cylinder S 1 × R. the basic ﬁeld. Alternatively. the O(3) non-linear σ model.

15.180) (15. z: ∂ U −1 ∂U = 0 This automatically implies that ∂ ∂U U −1 = 0 since ∂ U −1 ∂U = U −1 ∂∂U − U −1 ∂U U −1 ∂U = U −1 ∂ ∂U U −1 U −1 Thus. Note the asymmetry a between the deﬁnitions of J a and J .179) D 2 ×R where the T a s are the generators of the Lie algebra. The equation of motion which follows from the WZW model is ∂µ U −1 ∂ µ U − ǫµν U −1 ∂ν U = 0 (15.and left-handed currents J a = tr T a U −1 ∂U a J = tr T a ∂U U −1 (15. The equation of motion is only for the restriction of U to the boundary. This is completely analogous to the Berry phase term which we have in the action of a single spin. we have right. it only depends on the boundary values of U . J by ﬁrst deriving the canonical commutation relations from the action: they are those of a Kac-Moody algebra at level k.182) (15. but.16 Coulomb Gas Interacting Fermions Fusion and Braiding . It is useful to rewrite the equation of motion in terms of z. a We can compute the commutators of the modes of J a . D 2 × R.178) The second term in the parentheses results from the Wess-Zumino term in the action.177) The second term – often called a Wess-Zumino term – appears to be an integral over the 3D manifold given by the solid cylinder. 15.181) (15. in fact. S 1 × R.14. COULOMB GAS k 12π ǫµνλ tr ∂µ U U −1 ∂ν U U −1 ∂λ U U −1 287 (15. which is one way of seeing that the action is independent of the continuation from S 1 × R to D 2 × R.14 15.15 15.

RUDIMENTS OF CONFORMAL FIELD THEORY .288 CHAPTER 15.

Part V Symmetry-Breaking In Fermion Systems 289 .

.

the broken-symmetry state ψ(x) = 0. Ψc (x) satisﬁes diﬀerential equations (the saddle-point condition) which can be solved in diﬀerent experimental geometries or with diﬀerent boundary conditions.CHAPTER 16 Mean-Field Theory 16. the system becomes rather classical. the integrand contains only terms with an odd number of Grassman ﬁelds. at least in principle) symmetry-breaking ﬁelds cannot alter this conclusion. With fermions we don’t have such an option. it must vanish.e. Therefore. In this state. 291 . can’t have an expectation value. ψ e−S (16. just like a classical ﬁeld. Introducing small bosonic (i.1 The Classical Limit of Fermions At low-temperatures. ‘classical’ ground state of the system. there is a canonical guess for the ground state of a system of bosons ψ. This makes fermion problems intrinsically a little more diﬃcult: there is no stable phase which is a natural. One can deﬁne a classical ﬁeld Ψc (x) = ψ(x) which describes the properties of the system at the level of the saddle-point approximation to the functional integral. Hence. physical. χ. . Consider the functional integral for ψ ψ = dψdψ . . A fermion ﬁeld. we obtain only terms with an even number of Grassman ﬁelds.1) Taylor expanding exp(−S).

The broken symmetry state is a state in which the ‘bound state’ condenses. perturbation theory about the Fermi liquid critical point is. In strongly-interacting fermions systems. In such a regime.. one is better oﬀ starting deep within one of these phases. Let us ﬁrst consider superﬂuid/superconducting order parameters. In this section. It is sometimes useful to think of these fermion bilinears as annihilation operators for ‘bound states’ formed by two electrons or by an electron and a hole. In this part of the book. hopeless. can almost disguise itself as a stable phase. however. and producing a ‘classical ﬁeld’ describing the ground state. algorithms will converge very slowly because they must decide between many possible competing ground states. These stable phases fall into several categories. of course.2 Order Parameters.. the Fermi liquid can be unstable to several diﬀerent phases. To understand one or another of the stable phases. MEAN-FIELD THEORY This fact is somewhat obscured by the remarkable stability of the Fermi liquid critical point which.3) Focussing for the moment on the spin structure of the order parameter. we will discuss some of the possible order parameters in a fermionic system. t) ψβ (−k. A brute force numerical solution is almost certain to run into diﬃculties of the same origin: in many problems. . the symmetry ψα → eiθ ψα (16. we focus on phases in which the fermions organize themselves into a bosonic degree of freedom which condenses. In later parts of this book. The reason for the quotation marks is that there need not actually be a stable bound state. As we discussed in chapter . Symmetries The simplest bosonic degrees of freedom which can arise in fermionic systems are composed of fermion bilinears. This occurs when one of the following bilinears acquires an expectation value: ψα (k. In other words. we will consider two others: localized phases in dirty systems and topologically-ordered phases. 16.. breaking a symmetry. though it is a critical point.2) must be broken. Broken symmetry states occur when one of these bilinears acquires a non-zero expectation value. t) (16.292 CHAPTER 16. a superconductor is is characterized by the breaking of the global U (1) symmetry associated with charge conservation (if we ignore for the moment the electromagnetic ﬁeld and issues of gauge invariance in the superconducting case).

t) ψβ (−k. The cases ℓ = 0.2. and g-wave superconductors. For the case of density-wave order parameters.6) where ℓ must be even by Fermi statistics. 2. It is also possible. In the continuum. Once we have determined the L2 eigenvalue. which would require ﬁne-tuning. we can focus on angular momentum eigenstates of the form: ℓ ψα (k. Lz (let us ignore spin-orbit coupling). ORDER PARAMETERS. The former occur when ψα (k. Ψ(k) must be odd in k. 10-. It is possible for two diﬀerent pairing states with diﬀerent orbital and/or spin angular momenta to condense. The simplest way for this to happen is if a metal has several bands. Since the dm ’s are arbitrary complex numbers. Thus. d-wave. we will focus on the 2D square lattice. . Ψ(k). in principle. spin-singlet and spin-triplet superﬂuids/superconductors. A singlet superﬂuid can be thought of as a condensate of spin-singlet fermion pairs which have orbital wavefunction Ψ(k). t) = Ψ(k) ǫαβ while the latter are characterized by the expectation value ψα (k. In order to discuss the k dependence of the functions Ψ(k). t) ψβ (−k. as well as the case of a system on a 2D square lattice. while in a triplet superﬂuid/superconductor. we will consider a translationally. for this to occur within a single band. 4 are called s-wave. the order parameters for these three cases have 2-.4) Fermi statistics requires that in a singlet superﬂuids/superconductor Ψ(k) must be even in k. t) = Ψ(k) · σαγ ǫγβ (16. since this case applies to 3 He. however?). SYMMETRIES 293 we see that there are two categories. t) = ϕ(k) m=−ℓ ℓ ˆ dm Ym (k) ǫαβ (16. and the order parameters in diﬀerent bands have diﬀerent symmetries. t) ψβ (−k. to be introduced later in this section. we must make some further assumptions about the system.5) (16. there is still freedom in the choice of Lz eigenvalue m. thanks to rotational invariance. so we will not discuss this further here (see problems. ℓ(ℓ + 1). For the superﬂuid/superconducting case.and rotationallyinvariant system. Let us ﬁrst consider superﬂuids in the continuum. However. an energy eigenstate of such a pair will have well-deﬁned L2 . Other lattices will be considered in the problems at the end of the chapter.16. two diﬀerent angular momentum eigenstates would somehow have to become degenerate.

dm . needs to be ﬁnished In a triplet superﬂuid in the continuum. t) = ϕ(k) (kx + iky ) d · σαγ ǫγβ (16.e. we should allow the fermion ﬁelds to be at arbitrary momenta. the total spin of each pair has vanishing component along the d-direction. not necessarily k and −k. an ℓ = 1 triplet superﬂuid in the continuum has an 18-component order parameter dm . ϕ(k) is the radial part of the pair wavefunction. t) = ǫαβ ϕ(|k − k′ |) × d3 R iR·(k+k′ ) e (2π)3 ℓ ℓ ˆ dm (R)Ym (k) (16. The order parameter is composed of a product of two fermion operators. T The symmetry group of the system is U (1) × O(3) × SU (2) × Z2 . which would break independent orbital . t) ψβ (−k. without loss of generality. t) = ϕ(k) m=−ℓ ℓ Ym (k)dm · σαγ ǫγβ (16. Zℓ ∈ O(3) is the ℓ ˆ discrete set of rotations which leave Ym (k) invariant. t) ψβ (−k′ . The symmetry-breaking pattern depends on the dm s. U (1) ∈ O(3) is the subgroup of rotations about the direction of the angular momentum vector of the pair. any other direction is equally good). for some real vector d: ψα (k. MEAN-FIELD THEORY and 18-component order parameters. T The underlying symmetry of the system is U (1) × O(3) × Z2 . t) ψβ (−k..7) m=−ℓ Here.9) The angular momentum of the pair thus points in the z-direction (which we have arbitrarily chosen. while d+ = d. where T is the Z symmetry of time-reversal (there is also an SU (2) spin symZ2 2 metry which is unaﬀected by singlet ordering). . (We have ignored spin-orbit coupling. Finally. d0 = d− = 0. so the Z2 ∈ U (1) transformation ψ → −ψ leaves the order parameter unchanged. The pairs have spins ↑↓ + ↓↑ in the d-direction. This is broken down to T Z2 × U (1) × Zℓ × Z2 . respectively.294 CHAPTER 16. we have ℓ ψα (k. In order to allow for spatially inhomogeneous situations. ψα (k. such as those resulting from boundary conditions of the presence of a magnetic ﬁeld. In the A phase of 3 He..8) Thus. we have Fourier transformed the center-of-mass momentum of the pair in order to reveal the possible spatial dependence of dm (R). i.

t) ψβ (−k. 3π/2 – and 4 reﬂections – through the x-axis. then the order is called odd-gap superconductivity.13) Note that the order parameter vanishes along the direction kx = 0. t) = Ψ0 (sin kx a + i sin ky a) d · σαγ ǫγβ (16. The spin-orbit interaction of 3 He favors alignment of d with the angular momentum. simply. ORDER PARAMETERS. the remaining symmetry is U (1) × U (1). where the two U (1) factors correspond to spin rotations about the d-direction and rotations about the z-axis combined with gauge transformations (the former yield a phase factor which is cancelled by the latter). Time-reversal is broken by the selection of m = +1. where the SO(3) is the group of simultaneous rotations of both space and spin. the B phase is invariant under SO(3) × Z2 . If it is an odd function of t−t′ . d± = (ˆ ∓ iˆ)/ 2 or. The symmetry group of the square lattice is D4 . d0 = ˆ.2. which transforms into m = −1 under T. we see that the analog of the A phase is: ψα (k. t′ and allow the right-hand-side to have non-trivial dependence on t−t′ . t) ψβ (−k.11) Turning to p-wave superconductors. SYMMETRIES 295 angular momentum and spin conservation down to total angular momentum conservation. let us consider a system of electrons in 2D on a square lattice of side a. Now. π. A py superconductors has sin kx a replaced withsin ky a . t) = Ψ0 (sin kx a) d · σαγ ǫγβ (16. √ In the B phase. so that the correlation function actually vanishes for t = t′ . t) ψβ (−k. t) = ϕ(k) k · σαγ ǫγβ (16.) In this phase. The forms which we have chosen are actually too restrictive. An s-wave superconductor has an order parameter of the form: ψα (k. and the lines x = ±y.10) The pairs have orbital angular momentum ℓ = 1 and spin angular momentum S = 1 which add together to give total angular momentum J = 0. t) = Ψ0 ǫαβ (16. To be completely general. T Thus. π/2.16. z x y ˆ ψα (k.12) A px superconductor with vanishing spin component along the d-direction is even simpler: ψα (k. including the identity – by 0. we should allow the fermion operators to be at diﬀerent times t. t) ψβ (−k. the y-axis. with 8 elements: 4 rotations.

A dx2 −y2 + idxy superconductor breaks T with the order parameter: ψα (k. we must distinguish commmensurate and incommensurate ordering. A dx2 −y2 superconductor has ψα (k. t) ψβ (−k. A singlet s-wave density wave is simply a charge-density-wave: S(Q) = α ψ α† (k + Q. t) ψβ (k.19) Let us consider these order parameters in more detail on the square lattice. t) = ΦQ δβ d2 k β ψ α† (k + Q.15) We can deﬁne analogous neutral order parameters which do not break U (1). t) ψβ (k.18) If ΦQ (k) = 1 or if its integral over k is non-zero. f (k) is an element of some representation of the space group of the vector Q in the square lattice.20) In the higher angular momentum cases. t) = ΦQ (k) · σβ (16. Let us ﬁrst consider the singlet orderings: α ψ α† (k + Q. t) ψβ (−k. t) ψβ (k.14) while a dxy superconductor has cos kx a − cos ky a replaced by sin kx a sin ky a. t) = (2π)2 d2 k ΦQ (k) = 0 (16.17) (2π)2 The triplet orderings are of the form α ψ α† (k + Q. t) ψα (k.16) If ΦQ (k) = 1 or if its integral over k is non-zero. t) = ∆0 (cos kx a − cos ky a) ǫαβ (16. the state is a spin-densitywave (SDW): d2 k ΦQ (k) = 0 (2π)2 (16. t) = ΦQ (k) δβ (16. t) ψβ (k. t) σα = (2π)2 (16. MEAN-FIELD THEORY A d-wave superconductor must be a spin-singlet superconductor. the state is a chargedensity-wave (CDW): ρ(Q) = d2 k ψ α† (k + Q. t) = ∆0 (cos kx a − cos ky a + i sin kx a sin ky a) ǫαβ (16. However.296 CHAPTER 16. For commensurate ordering such that 2Q is . the spin structures will no longer be determined by Fermi statistics.

. if f (k+Q) = −f ∗ (k). e. π/a). t) = Φ∗ f ∗ (k) δβ Q α α ΦQ f (k + Q) δβ = Φ∗ f ∗ (k) δβ Q (16. t) ψα (k + Q. . . t) − ψ †α (x. π/a). the Q = (0. π/a). For singlet px ordering. A commensurate singlet px density-wave state has ordering α ψ α† (k + Q.16.24) Further insight into these states is obtained by considering their realspace forms. t) ψβ (k.26) As a result of the i. ΦQ must be real. − i |ΦQ | eiQ·x δβ (16. For singlet dx2 −y2 ordering. ψ †α (x. However. + |ΦQ | eiQ·x δβ (16. t) = ΦQ (sin kx a + i sin ky a) δβ (16. t) ψα (k + Q.25) The singlet state of this type breaks no other symmetries. the combination of T and translation by an odd number of . . π/a).g. ORDER PARAMETERS. SYMMETRIES 297 a reciprocal lattice vector. we can take the hermitian conjugate of the order parameter: α ψ †β (k. it is usually called the Peierls state or bond order wave. t) ψβ (x − aˆ. If Q = (0. t) ψβ (x + aˆ. this will be the case if Q = (π/a. t) = x x α . this will be the case if Q = (π/a. t) = ΦQ sin kx a δβ (16. t) ψβ (x + aˆ. ΦQ must be real. π/a). ΦQ must be imaginary. If f (k + Q) = f ∗ (k). For singlet px ordering.23) The commensurate singlet px + ipy density-wave states are deﬁned by: α ψ α† (k + Q. 0) or Q = (π/a. this will be the case if Q = (0. 0) must have imaginary ΦQ : ψ †α (x.2. 0) or Q = (π/a. this will be the case if Q = (π/a. π/a). t) = x x α .22). π/a) singlet px density-wave states break T . t) ψβ (x − aˆ.22) Hence. Q = (π/a. For singlet dxy ordering. From (16. a commensurate singlet px density-wave state with Q = (π/a. for Q commensurate Φ∗ f (k + Q) Q = ∗ (k) f ΦQ (16. t) ψβ (k. t) − ψ †α (x.21) Therefore. t) = Φ∗ f ∗ (k) δβ Q α ψ β† (k + Q + Q.

t) + ψ †α (x. the phase of the above bond correlation function – and. The incommensurate cases will be considered in the problems at the end of the chapter. Examples of singlet dx2 −y2 . in contrast to the superconducting case. t) ψβ (k. + |ΦQ | eiQ·x δβ 2 (16. dxy . t) + ψ †α (x. t) ψβ (x − aˆ. t) ψβ (x + aˆ. the singlet dx2 −y2 density-wave breaks T as well as translational and rotational invariance.298 CHAPTER 16. therefore. therefore. The commensurate Q = (π/a. π/a) singlet dxy must have real ΦQ . the commensurate singlet dx2 −y2 density-wave states have α ψ α† (k + Q. the singlet dx2 −y2 + idxy state does break T . It is also possible to break spin-rotational invariance without breaking translational symmetry or U (1) gauge symmetry. t) ψβ (x − aˆ. MEAN-FIELD THEORY lattice spacings remains unbroken. t) = ΦQ (cos kx a − cos ky a) δβ (16. it does not break T . . the ﬂux through each plaquette – is alternating. . it takes the form: ψ †α (x. In real space.22). Note that the nodeless commensurate singlet dx2 −y2 + idxy density-wave state does not break more symmetries than the commensurate singlet dx2 −y2 density-wave state. and dx2 −y2 + idxy density-wave states are depicted in ﬁgure ??.29) As a result of the i. The commensurate Q = (π/a. Similarly. π/a) singlet dx2 −y2 density-wave states must have imaginary ΦQ . The combination of time-reversal and a translation by one lattice spacing is preserved by this ordering. according to (16. π/a) singlet dx2 −y2 density-wave state is often called the staggered ﬂux state. t) = α ΦQ (cos kx a − cos ky a + i sin kx a sin ky a) δβ (16. Examples of commensurate and incommensurate singlet px and px + ipy density-wave states are depicted in ﬁgure ??. t) ψβ (x + aˆ. This is accomplished in a . The same is true of the commensurate singlet px + ipy density-wave states. t) ψβ (k. .). t) − x x ψ †α (x. On the other hand. as a result. There is also a contribution to this correlation function coming from ψ † (k)ψ(k) which is uniform in space (the .28) The commensurate Q = (π/a. t) = y y i α .27) while the commensurate singlet dx2 −y2 + idxy density-wave states have ψ α† (k + Q. .

16. when it condenses and the others remain gapped. let us consider the following integral over 2n Grassmann variables χa . This approximation can often become exact in some kind of large-N limit. the Hubbard-Stratonovich transformation. t) ψβ (k. χa : I= We can rewrite this integral as I=N dχa dχa dϕeχa χa −ϕχa χa − 2 ϕ n 2 dχa dχa eχa χa + 2n (χa χa ) 1 2 (16. such as the saddle-point approximation. THE HUBBARD-STRATONOVICH TRANSFORMATION ferromagnet: α ψ α† (k.16. t) ψβ (k. t) = δβ f (k) (16.33) (16. it is sometimes possible to detect such a tendency. in particular.3 The Hubbard-Stratonovich Transformation There is a formal transformation. By studying the instabilities of the RG ﬂows of a Fermi liquid in a periodic potential.3.31) It is also possible to break spatial rotational symmetries (either in the continuum or on the lattice) without breaking translational symmetry or spin rotational symmetry: α ψ α† (k. To illustrate this transformation. which allows us to introduce a bosonic degree of freedom to replace one of the possible bilinear combinations of the fermions. the fermions can be integrated out and an eﬀective theory for the bosonic variable can be derived. When the occurrence of such an order parameter causes a gap in the fermionic spectrum. This theory can then be solved by the usual methods applied to bosonic systems. This approach has a good chance of succeeding when one of the possible bosonic degrees of freedom dominates.30) Integrating over k. we have: S =M dd k g(k) (2π)d (16. t) = M · σβ g(k) 299 (16.32) where f (k) transforms non-trivially under rotations.34) .

35) and (16.37) The two integrals (16. diﬀerent approximations are suggested by the forms of these integrals. = (1 ± 5)/2. is the HubbardStratonovich transformation. The Grassmann integrals are now Gaussian. and they may be performed: I =N =N dϕ (1 − ϕ)n e− 2 ϕ dϕ en ln(1−ϕ)− 2 ϕ n n 2 2 (16.37) are equal so long as they are performed exactly.38) The two saddle-point values √ ϕ are the golden number and the negative of of its inverse. ϕs. the underlying physics will dictate which one is a better starting point. we have exchanged a non-Gaussian Grassmann integral for a nonGaussian ordinary integral. however. note that we could have decoupled the quartic term in another way: We can rewrite this integral as e 2n (χa χa ) = e 2n (χa χa ) (χb χb ) n 2 = N dϕab eχa ϕab χb − 2 ϕab 1 2 1 (16. This. MEAN-FIELD THEORY where N is a normalization constant. This may not seem like such a big success.36) Now the Grassmann integral is transformed into an integral over the matrix ϕab : I=N dϕab en ln(δab −ϕab )− 2 ϕab n 2 (16. Adding the contributions from both saddle-points. The saddle-point condition for (16. in a nutshell.33). However. However.35) Thus. Before doing this. we recover the integral (16. we can now use techniques such as the saddle-point approximation to evaluate the ordinary integral. 16.p.300 CHAPTER 16. By performing the Gaussian integral over ϕ.4 The Hartree and Fock Approximations Let us now consider the saddle-point evaluation of these integrals. In the analogous ﬁeld-theoretic context. we have I = N{ √ ( 5 + 1)/2 (−1)n n/2 e “ “√ ” √ ” 5−1 −n 1+ 1+2 5 +n ln 2 .35) is: −ϕ − 1 1−ϕ (16. we will be applying it to functional integrals.

For instance. 16.41) k with where the uk ’s and vk ’s are variational parameters with respect to which Ψ0 | H |Ψ0 is minimized.5 The Variational Approach There is an equivalent approach within the framework of canonical quantization. the interaction can be decoupled so that the bosonic ﬁeld is equal to the product of a creation operator from one density factor and an annihilation operator from the other density factor.42) .39) n/2 In the physical context in which we will be using these ideas. This approach can also be used to decouple an interaction which does not lead to the development of a non-trivial order parameter. a density-density interaction can be decoupled so that the electrons interact with a bosonic ﬁeld.40) The condition for minimizing H is the same as the saddle-point condition in the path integral approach. we will be considering Grassmann functional integrals which can be rewritten as bosonic functional integrals – or ‘decoupled’. This is the Hartree approximation. The diﬀerent bosonic ﬁelds which we introduce will be the diﬀerent possible order parameters of the system. |0 . THE VARIATIONAL APPROACH (−i)n + √ ( 5 − 1)/2 e “√ ” “ √ ” 5+1 −n 1+ 1− 5 +n ln 2 2 301 } (16. The size of the order parameter is the variational parameter which is tuned to minimize H = 0|H|0 0|0 (16. One introduces a trial ground state.16. This is the Fock approximation. The basic form of the trial wavefunction is. The wavefunction is normalized by taking 2 u2 + vk = 1 k (16. Alternatively. At the saddle-point level. The saddle-point condition is then a ‘gap equation’ (a name whose aptness will become clear when we look at examples). since the remaining Grassman integral is Gaussian – in a variety of ways. Mean-ﬁeld-theory – or the Hartree approximation – for any of these order parameters is simply the saddle-point approximation for the bosonic functional integral. which is based on the anticipated order parameter. in the superconducting case: |Ψ0 = † † uk + vk ψk↑ ψ−k↓ |0 (16. the bosonic ﬁeld is equal to the electron density.5.

we see that it is given by uk vk .α |0 (16. In the case of a density-wave at wavevector Q.41).43) k . MEAN-FIELD THEORY Thus. the variational method selects a ground state by minimizing the energy with respect to the order parameter. there is only one free variational parameter for each k. Calculating the order parameter in (16.302 CHAPTER 16. while the saddle-point approximation of the previous section ﬁnds a ground state by minimizing the action with respect to the order parameter. The combination uk vk is a convenient way of parametrizing it.α + vk ψk+Q. Thus. the trial wavefunction takes the form: |Ψ0 = † † uk ψk.

the variational method. k) (iǫ − vF k) ψσ (ǫ. S= d2 k dǫ † ψ (ǫ. k) (2π)2 2π σ 303 . there is only one instability. mean ﬁeld theory). If the Fermi surface is rotationally symmetric. there are competing instabilities in which case it is very diﬃcult to ﬁnd a new saddle-point or an appropriate variational ansatz.) In the case of electrons in a solid. and spin. the general strategy is to use the RG to go to low energies so that the irrelevant variables have all become small and the relevant variable is dominant.e. We saw a trivial example of this with a shift of the chemical potential of a Fermi liquid.and charge-density wave instabilities are possible when the Fermi surface satisﬁes certain special conditions (‘nesting’). the Fermi surface need not be rotationally symmetric. Consider the action of electrons in D = 2 with F = 0 but non-zero V . This occurs in the case of a 1D system of fermions. This approach has proven very successful in the study of ordering in condensed matter physics. When the instability is due to interaction terms. as we found earlier: the Cooper pairing instability. The problem with a single relevant interaction must then be solved by ﬁnding a new saddle-point (i. this perturbation generally leads to a fundamental reorganization of the ground state.1 Instabilities of the Fermi Liquid When a ﬁxed point has a relevant perturbation. (Sometimes. or some other non-perturbative method.CHAPTER 17 Superconductivity 17.

SUPERCONDUCTIVITY d2 k d2 k′ dǫ1 dǫ2 dǫ3 † † ψ (ǫ4 . k′ )∆(ǫ1 + ǫ2 . these are relevant. k′ )ψ↓ (ǫ2 .4) Then. k) = (17. k) + ∆† (ǫ1 + ǫ2 . −k′ ) V (k. 17. k′′ ) = δ(k − k′′ ) (2π)2 where V −1 (k. we now consider spin 1/2 electrons. −k)ψ↓ (ǫ1 .6) . k′ )ψ↓ (ǫ2 . k′ ) (2π)2 We now make the change of variables: ∆(ǫ1 + ǫ2 . k′ )] (17. k′ ) Ψ† (ǫ1 + ǫ2 .5) d2 k V −1 (k. k′ ) ψ↑ (ǫ2 . k) (2π)2 (2π)2 2π 2π +ψ↑ (ǫ1 .1) (2π)2 (2π)2 2π 2π 2π ↑ − where V (k. we will ﬁnd the new saddle point which is appropriate for the case in which V is relevant. k) (2π)2 2π σ d2 k d2 k′ dǫ1 dǫ2 † † − [ψ (ǫ1 . In chapter 14. we showed that the Fourier modes of V (θ1 − θ2 ) satisfy the RG equation: 1 dVm =− V2 (17. k′ )ψ↓ (ǫ3 . Unlike in previous chapters. −k′ ) Ψ(ǫ1 + ǫ2 .3) d2 k′ V (k. k) (iǫ − vF k) ψσ (ǫ. k′ )ψ↓ (ǫ2 . k) + Ψ† (ǫ1 + ǫ2 . −k′ ) ∆† (ǫ1 + ǫ2 . k) (17.2) dℓ 2πvF m When negative. k′ ) ≡ V (θ1 − θ2 ) is a function of the angles only. the action can be rewritten: S= d2 k dǫ † ψ (ǫ. ǫ2 . −k′ ) Ψ† (ǫ1 .2 Saddle-Point Approximation We introduce a Hubbard-Stratonovich ﬁeld Ψ(k.304 CHAPTER 17. k′ ) is the inverse of V (k. k′ ) V (k′ . In the next section. ω) to decouple the BCS interaction: S= d2 k dǫ † ψ (ǫ. k′ ) [ψ↑ (ǫ1 . k)V −1 (k. k′ )ψ↓ (ǫ2 . k′ ): (17. −k′ ) ∆(ǫ1 + ǫ2 . We will also mention brieﬂy the equivalent variational ansatz (which was the historical method of solution). k) (2π)2 2π σ d2 k d2 k′ dǫ1 dǫ2 † † − V (k. k)Ψ(ǫ1 + ǫ2 . where we dealt with spinless fermions. k′ )] (17. k) (2π)2 (2π)2 2π 2π ↑ +ψ↑ (ǫ1 . k) (iǫ − vF k) ψσ (ǫ.

k′ )∆(k′ ) (2π)2 (17. the ǫ integral in the ﬁrst term can be done (at ﬁnitetemperature. V > 0. For V attractive. k′ )∆(k′ ) (2π)2 (17.2. SADDLE-POINT APPROXIMATION 305 Since the action is quadratic in the fermion ﬁelds ψσ . ∆(ǫ.7). we must do a Matsubara sum instead). k′ ) (2π)2 (2π)2 2π (17. k′ )∆(ǫ. k′ )∆(ǫ.17.e. V (k.7) We look for a frequency-independent solution. k) = ∆(k) of the saddle point equations. k)V −1 (k. k′ )∆(k′ ) (vF k′ )2 + |∆(k′ )|2 = ∆(k) (17. giving: d2 k (2π)2 or 1 (vF k)2 + |∆(k)|2 d2 k′ (2π)2 = d2 k′ −1 V (k. Consider the simplest case. Then the gap equation reads: V∆ d2 k′ =∆ (17. It determines the order parameter ∆ which breaks the U (1) symmetry ∆ → eiθ ∆ of the action (17. δS =0 (17. k′ ) (2π)2 (2π)2 2π d2 k dǫ =− ln (iǫ)2 − (vF k)2 − |∆(k)|2 + (2π)2 2π d2 k d2 k′ dǫ1 † ∆ (ǫ.12) (2π)2 (vF k′ )2 + |∆|2 d2 k′ (2π)2 1 (vF k′ )2 + |∆|2 = 1 V (17. of an s-wave attraction.7).10) V (k. we have the saddle-point equations: 1 d2 k dǫ = 2 2π (iǫ)2 − (v k)2 − |∆(k)|2 (2π) F d2 k′ −1 V (k. The is the BCS gap equation. this equation always has a solution.13) .11) or. k′ ) = V . i. we can integrate out the fermions to get an eﬀective action S[∆]: S[∆] = − T r ln (iǫ)2 − (vF k)2 − |∆(k)|2 + d2 k d2 k′ dǫ † ∆ (ǫ. k)V −1 (k.9) At zero-temperature.8) δ∆ From (17.

14) Λ 2π sinh m∗ V 2π (17. the ﬁnite-temperature gap equation is: βEk′ V (k.20) 17. k′ )∆(k′ ) d2 k′ tanh = ∆(k) (17.3 BCS Variational Wavefunction For purposes of comparison. Cooper.16) Note that the gap is not analytic in V .18) For an s-wave attraction. there is always a non-trivial saddle-point solution when V > 0. and Schrieﬀer. this gap equation has solution ∆ = when: dξ tanh (17. SUPERCONDUCTIVITY Since the left-hand-side is logarithmically divergent at the Fermi surface if ∆ = 0.21) k with where the uk ’s and vk ’s are variational parameters with respect to which Ψ0 | H |Ψ0 is minimized. They wrote down the wavefunction |Ψ0 = † † uk + vk ψk↑ ψ−k↓ |0 (17. m∗ 2π or ∆= dξ 1 ξ2 + ∆2 = 1 V (17. m∗ V /2π ≪ 1. then ∆ = 2Λe− m∗ V (17.17) 2 ′ )2 + |∆(k ′ )|2 (2π) 2 (vF k with Ek = m∗ 2π (vF k)2 + |∆(k)|2 1 ξ2 1 βξ = 2 V (17. The wavefunction is normalized by taking 2 u2 + vk = 1 k (17.306 CHAPTER 17. As you will show in the problem set. it could never be discovered in perturbation theory.19) So the critical temperature for the onset of superconductivity is: Tc = 1.22) .14 Λe− m∗ V 2π (17. consider the route taken by Bardeen.15) If the attraction is weak.

k′ )uk vk uk′ vk′ k. ξk 1 uk = √ 1 + Ek 2 1 ξk vk = √ 1 − Ek 2 with Ek = 2 ξk + |∆(k)|2 1 2 ∆(k) 2 2 ξk + |∆(k)|2 (17.k ′ (17.30) .25) = 4vk ξk − 2 V (k.27) 1 2 (17.28) (17.k ′ † † V (k.3.24) ∂uk vk uk′ vk′ ∂vk (17. equivalently. It is the Hamiltonian which only contains the relevant interaction. we assume that the k’s are discrete (as they are in a ﬁnite-size system). k′ ) k ∆(k′ ) 2 ξk + |∆(k′ )|2 = ∆(k) (17. The expectation value of the Hamiltonian is: Ψ0 | H |Ψ0 = Hence. k′ ) uk uk′ vk′ + 2 k′ k 2 2vk ξk − V (k.17. k′ ) u2 − vk uk′ vk′ k (17.29) Then we can rewrite the minimization condition as the BCS gap equation: V (k. k′ ) k′ 2 u2 − vk k uk uk′ vk′ The minimum of Ψ0 | H |Ψ0 occurs when 2ξk uk vk = k′ 2 V (k. ∂ Ψ0 | H |Ψ0 = 4vk ξk − ∂vk V (k.26) If we deﬁne ∆(k) by uk vk = or. The irrelevant and marginal interactions have been dropped. k′ )ψk↑ ψ−k↓ ψk′ ↑ ψ−k′ ↓ (17. BCS VARIATIONAL WAVEFUNCTION 307 For notational simplicity.23) This Hamiltonian is called the BCS reduced Hamiltonian.1) is: H= k † ξk ψkσ ψkσ − k. The Hamiltonian which follows from (17.

1 Condensate fraction and superﬂuid density** Single-Particle Properties of a Superconductor Green Functions When ∆ takes a non-zero. with an inverse which gives † Gσσ′ (iǫ. −k′ ) ∆(k) − ψ↑ (iǫ. −k′ ) ∆†(17. the action for the fermions is: S= d2 k dǫ † [ψ (iǫ. k)ψσ′ (iǫ. k)ψσ′ (−iǫ. −k) = ǫσσ′ iǫ + ξk 2 − ξk − |∆(k)|2 ∆(k) (17. 17. k′ )ψ↓ (−iǫ. the propagator is obtained by inverting the quadratic part of the action. k′ )ψ↓ (−iǫ. We denote F(iǫ. k) by a line with two arrows pointing away from each other † † and F † (iǫ.5. The electron spectral function is given by A(k.5 17. This is now a matrix.1: The graphical representation of (a) G (b) F and (c) F † . k) = δσσ′ Fσσ′ (iǫ. k) (2π)2 2π σ † † − ψ↑ (iǫ. k) = ψσ (iǫ. ǫ) = Im ǫ + ξk 2 (ǫ + − ξk − |∆(k)|2 2 2 = uk δ(ǫ − Ek ) + vk δ(ǫ + Ek ) iδ)2 (17. −k) by a line with two arrows pointing towards each other.32) 2 (iǫ)2 − ξk − |∆(k)|2 (iǫ)2 We denote G(iǫ. frequency-independent value.308 (a) CHAPTER 17. k) = ψσ (iǫ. k) by a line with two arrows pointing in the same direction.33) . k)ψσ′ (−iǫ. k) = ψσ (iǫ. k) (iǫ − vF k) ψσ (iǫ.4 17.31) (k)] As usual. SUPERCONDUCTIVITY (b) (c) Figure 17.

so γσ (k) creates a hole below the Fermi surface. Another way of understanding the single-particle properties of a superconductor is to diagonalize the action. k)γσ′ (iǫ. ǫ) (iǫ − Ek ) γσ (k.17. costing positive energy. ǫ) = uk ψ↑ (k. ǫ) − vk ψ↓ (−k. vk → 0. When ξk ≫ ∆.5. For some purposes – such as the Hebel-Slichter peak in NMR – we can ignore the fact that they are a superposition of an electron and a hole and treat the superconductor as a semiconductor with energy bands ±Ek . since they are a superposition of an electron and a hole. they are superpositions of fermions and holes. Since the density of single quasiparticle states. ǫ) = uk ψ↓ (k. . uk → 1. The coherence factors can cancel the density of states divergence at |E| → ∆. for other properties – such as the acoustic attenuation – the mixing between between electron and hole state (‘coherence factors’) is important. However.34) S= d2 k dǫ † γ (k.36) The γ(k)’s are the basic single-particle excitations – ‘Bogoliubov-DeGennes quasiparticles’ – of a superconductor. vk → 1. also costing positive energy. ǫ) † γ↓ (k. ǫ) (2π)2 2π σ (17. the basic excitations have indeﬁnite charge. In the case of electrons. they are spin eigenstates.37) is divergent for |E| → ∆ and vanishing for |E| < ∆. |E| m∗ dk √ = θ (|E| − ∆) dE 2π E 2 − ∆2 (17. ǫ) (17. The action is diagonalized by the γ(k)’s † γ↑ (k.35) The γ(k)’s have propagator: † γσ (iǫ. k) = δσσ′ iǫ − Ek (17. so γσ (k) creates a fermion above the Fermi surface. the semiconductor model predicts sharp increases in these quantities for T ∼ ∆ and exponential decay for T ≪ ∆. When ξk ≪ −∆. † uk → 0. ǫ) + vk ψ↑ (−k. † Note that Ek > 0. Although they are not charge eigenstates. SINGLE-PARTICLE PROPERTIES OF A SUPERCONDUCTOR 309 which shows that the electron has spectral weight u2 at Ek and spectral k 2 weight vk at −Ek . and there is no enhancement for T ∼ ∆.

−iωm ) = 1 β d3 k G↓↓ (iǫn . k) G↑↑ (iǫn + iωm .38) The spin-spin correlation function S+ (q. k + q) (2π)3 (17.2. −iωm ) is given by the sum of the two diagrams of ﬁgure 17. SUPERCONDUCTIVITY Figure 17.40) n n If we replace the sums over Matsubara frequencies by contour integrals which avoid z = (2n + 1)πi/β. −iωm ) = 1 β 1 + β iǫn + iωm + ξk+q d3 k iǫn + ξk 3 (iǫ )2 − ξ 2 − |∆|2 (iǫ + iω )2 − ξ 2 2 (2π) n n m k k+q − |∆| d3 k ∆ ∆ 3 (iǫ )2 − ξ 2 − |∆|2 (iǫ + iω )2 − ξ 2 2 (2π) n n m k k+q − |∆| (17. iωm ) S− (−q.5.93). this is: S+ (q. k) F↓↑ (iǫn + iωm . iωm ) S− (−q. S+ (q. k + q) (2π)3 d3 k † F↑↓ (iǫn .39) n 1 + β n or. 17. Assuming that ∆(k) = ∆. the NMR relaxation rate is given by: 1 = T1 T d2 q 1 A(q) lim χ′′ (q.2: The two diagrams which contribute to the spin-spin correlation function of a superconductor.2 NMR Relaxation Rate According to (6. dz nF (z) C 2πi dz nF (z) + 2πi C z + iωm + ξk+q z + ξk d3 k 3 (z)2 − ξ 2 − |∆|2 (z + iω )2 − ξ 2 2 (2π) m k k+q − |∆| ∆ ∆ d3 k 2 2 (2π)3 (z)2 − ξk − |∆|2 (z + iωm )2 − ξk+q − |∆|2 . ω) +− 2 ω→0 ω (2π) (17.310 CHAPTER 17. iωm ) S− (−q.

17. −iωm ) = 2 ξk+q + |∆|2 (17.5. SINGLE-PARTICLE PROPERTIES OF A SUPERCONDUCTOR 311 (17. S+ (q. dropping terms such as δ(ω − Ek − Ek+q ) which vanish for ω < 2∆). thereby. iωm ) S− (−q.42) d3 k ξk + Ek Ek + iωm + ξk+q n (Ek ) 2 3 F (2π) 2Ek (Ek + iωm )2 − Ek+q ξk − Ek −Ek + iωm + ξk+q d3 k nF (−Ek ) + 2 (2π)3 −2Ek (−Ek + iωm )2 − Ek+q 3k ξk+q + Ek+q Ek+q − iωm + ξk d n (Ek+q ) + 2 3 F (2π) 2Ek+q (Ek+q − iωm )2 − Ek 3k ξk+q − Ek+q −Ek+q − iωm + ξk d n (−Ek+q ) + 2 3 F (2π) −2Ek+q (Ek+q + iωm )2 − Ek ∆ ∆ d3 k nF (Ek ) + 2 (2π)3 2Ek (Ek + iωm )2 − Ek+q ∆ ∆ d3 k n (−Ek ) + 2 3 F (2π) −2Ek (−Ek + iωm )2 − Ek+q ∆ ∆ d3 k nF (Ek+q ) + 2 (2π)3 2Ek+q (Ek+q − iωm )2 − Ek 3k d ∆ ∆ (17.41) these integrals receive contributions only from the poles at z=± 2 ξk + |∆|2 z = −iωn ± Hence. ω) = +− (ξk + Ek ) (Ek + ξk+q ) d3 k n (Ek ) δ(ω + Ek − Ek+q ) 3 F (2π) 2Ek Ek+q (ξk − Ek ) (−Ek + ξk+q ) d3 k − δ(ω + Ek − Ek+q ) n (−Ek ) 3 F (2π) 2Ek Ek+q (ξk+q + Ek+q ) (Ek+q + ξk ) d3 k + n (Ek+q ) δ(ω + Ek − Ek+q ) 3 F (2π) 2Ek+q Ek (ξk+q − Ek+q ) (−Ek+q + ξk ) d3 k n (−Ek+q ) δ(ω + Ek+q − Ek ) − 3 F (2π) 2Ek+q Ek d3 k ∆2 + δ(ω + Ek+q − Ek ) nF (Ek ) (2π)3 2Ek Ek+q .43) + n (−Ek+q ) 2 3 F (2π) −2Ek+q (Ek+q + iωm )2 − Ek If we now take iωm → ω + iδ. we obtain: χ′′ (q. and ω < 2∆ (and.

SUPERCONDUCTIVITY d3 k ∆2 δ(ω + Ek − Ek+q ) nF (−Ek ) (2π)3 2Ek Ek+q d3 k ∆2 nF (Ek+q ) δ(ω + Ek − Ek+q ) (2π)3 2Ek+q Ek d3 k ∆2 nF (−Ek+q ) δ(ω + Ek+q −(17. and using nF (−Ek ) = 1 − nF (Ek ).48) E+ω 1+ dE √ E E 2 − ∆2 For ω → 0. we have: χ′′ (q.50) . ω) = +− d3 k (nF (Ek ) − nF (Ek+q )) (2π)3 1+ ξk ξk+q + ∆2 2Ek Ek+q δ(ω + Ek − Ek+q ) (17.45) Let us assume that A(q) = A. the right-hand-side is exponentially suppressed as a result of the ∂nF (E)/∂E.46) − Ek′ ) or. we can write this as: 1 =A T1 T m∗ 2π 2 Λ ∆ dE √ E 2 − ∆2 E E+ω (E + ω)2 − ∆2 1+ ∆2 2E(E + ω) ∂ nF (E) ∂E (17. A 1 = T1 T ω or A 1 = T1 T ω m∗ 2π 2 Λ ∆ m∗ 2π 2 Λ Λ dE ∆ ∆ dE ′ √ E E′ ∆2 √ 1+ 2EE ′ E 2 − ∆2 E ′2 − ∆2 ′ × nF (E) − nF (E ) δ(ω + E (17. and −∆ 1 ∼e T T1 T (17.47) − E′) ∆2 2E(E + ω) (E + ω)2 − ∆2 × (nF (E) − nF (E + ω)) (17.44) Ek ) (2π)3 2Ek+q Ek − + − dropping terms which are odd in ξk or ξk+q . using the single-particle density of states to re-write the momentum integrals as energy integrals. dropping the term linear in ξk and ξk ′ A 1 = T1 T ω d3 k′ d3 k (nF (Ek ) − nF (Ek′ )) (2π)3 (2π)3 1+ ∆2 2Ek Ek′ δ(ω + Ek (17.312 CHAPTER 17. Then.49) For T → 0.

5. From our calculation of 1/T1 . there is a moderate. ω) ρρ (2π)3 (17.51) ∂E which is a divergent integral at E = ∆. SINGLE-PARTICLE PROPERTIES OF A SUPERCONDUCTOR 313 For T ∼ ∆. there is a crucial minus sign arising from the ordering of the electron operators: ρ(q. 17. For realistic values of ω. α. we can take the ω → 0 limit: α=A m∗ 2π 2 Λ dE ∆ E2 E 2 − ∆2 1− ∆2 E2 ∂ nF (E) ∂E . the exponential suppression is not very strong so the density of states divergence is important.53) where g(q) is the electron-phonon coupling.54) As a result of the −sign. but clearly observable increase of 1/T1 for T < Tc with a maximum which is called the Hebel-Slichter peak. k) F↓↑ (iǫn + iωm . which is essentially the phonon lifetime. of a phonon of frequency ω is essentially given by α= d3 q g(q) χ′′ (q. k + q) (2π)3 (17. since there are density operators rather than spin operators at the vertices of these diagrams.52) n 1 − β n The acoustic attenuation rate. is given by the diagrams of ﬁgure 17. −iωm ) = 1 β d3 k G↓↓ (iǫn . Phonons are coupled to the electron density.5. too. k) G↑↑ (iǫn + iωm . we see that this is (assuming constant g): α=g m∗ 2π 2 Λ ∆ dE √ E 2 − ∆2 E E+ω (E + ω)2 − ∆2 1− ∆2 E(E + ω) ∂ nF (E) ∂E (17. iωm ) ρ(−q. for ω = 0 1 =A ω→0 T1 T lim m∗ 2π 2 Λ dE ∆ E2 E2 − ∆2 1+ ∆2 2E 2 ∂ nF (E) (17. In fact. k + q) (2π)3 d3 k † F↑↓ (iǫn .3 Acoustic Attenuation Rate Suppose we compute the acoustic attenuation rate. This. so the phonon lifetime is determined by a density-density correlation function. However.17.2.

this is exponentially decreasing at low. the density of states divergence has been cancelled by the quantum interference between particles and holes.55) As in the case of 1/T1 . T .58) . SUPERCONDUCTIVITY ∂ nF (E) ∂E =A dE ∆ (17. their contribution to the density-density correlation function is suppressed. Since the underlying quasiparticles are a superposition of electrons and holes such that their charge vanishes as the Fermi surface is approached. On the other hand. between the superconductor and the other system. 17. so there is no enhancement for T ∼ ∆.57) † where ψσ (k) is the creation operator for an electron in the superconductor and χ† (k) is the creation operator for an electron in the other system. the quasiparticles carry spin 1/2 (since they are a mixture of an up-spin electron and a down-spin hole) so their contribution to the spin-spin correlation function is unsuppressed. k′ ) ψσ (k) χσ (k) + t∗ (k. HV = V = V NL d3 k † ψ (k) ψσ (k) (2π)3 σ (17. α∼e −∆ T (17. This suppression cancels the divergence in the density of states. V . σ t(k. hence the density of states divergence has dramatic consequences leading to the Hebel-Slichter peak. k′ ) is the tunneling matrix element for an electron of momentum k in the superconductor to tunnelin into a momentum k′ state in the other system. Tunneling occurs when there is a voltage diﬀerence.5.314 m∗ 2π 2 Λ CHAPTER 17.4 Tunneling Tunneling is a classic probe of the single-particle properties of an electron system. An approximate description of the coupling between the superconductor and the other system is given by the tunneling Hamiltonian: HT = d3 k d3 k′ † † t(k. Let us suppose we connect a superconductor on the left with another system – which may or may not be a superconductor – on the right.56) However. k′ ) χσ (k) ψσ (k) (2π)3 (2π)3 ≡ B + B† (17.

62) Converting the Matsubara sum in the ﬁrst term to an integral. B(t′ ) B † (t). B † (t′ ) B(t).5. −∞ HT eiV tNL (17. we have: I(t) = + ∞ −∞ ∞ dt′ θ(t − t′ ) eieV (t −t) i dt′ θ(t − t′ ) ′ ′ B(t). ǫ) [nF (ǫ) − nF (ǫ + eV )] −∞ 2π d3 k d3 k′ † FL (k. iǫn )GR (k. k′ ) χσ (k) ψσ (k) (2π)3 (2π)3 (17. ǫ + eV ) AR (k′ .59) ≡ i B − B† Following the steps by which we derived the conductivity and other response functions in chapter 7. B † (t′ ) (17. iǫn − iω) n iω→eV +iδ + t2 Im e2ieV t d3 k d3 k′ (2π)3 (2π)3 n † FL (k. I(t) computed to linear order in HT is given by: I(t) = T eiV tNL +i t Rt −∞ HT I(t) T e−iV tNL −i Rt −∞ HT =i I(t). iǫn )FR (k. we have: I = t2 d3 k d3 k′ (2π)3 (2π)3 dǫ AL (k.60) Substituting the above expressions for I and HT . k′ ) = t. B(t′ ) − eieV (t−t ) i − eieV (t+t ) i ′ ′ B † (t).17. SINGLE-PARTICLE PROPERTIES OF A SUPERCONDUCTOR 315 The current ﬂowing between the superconductor and the other system is I=i d3 k d3 k′ † † t(k.61) e−ieV (t+t ) i −∞ Suppose that t(k. iǫn )FR (k. we see that the current. analytically continuing. iǫn − iω) (2π)3 (2π)3 n iω→iδ (17. iǫn − iω) iω→iδ (17. Then the real part of the current is I = t2 Im d3 k d3 k′ (2π)3 (2π)3 GL (k.63) ∞ + t2 Im e2ieV t . k′ ) ψσ (k) χkσ − t∗ (k. and taking the imaginary part (as we have done so often before).

It vanishes unless the system on the right is a superconductor. It can be rewritten as: IE = t2 m∗ kF m∗ kF R L 2π 2 2π 2 dξk dξk′ dǫ 2 2 u δ(ǫ + eV − Ek ) + vk δ(ǫ + eV + Ek ) 2π k × AR (k′ . this vanishes for eV < ∆ and asymptotes I ∝ V for V large. we ﬁnd: IE = t2 m∗ kF m∗ kF R L 2π 2 2π 2 ∞ dE∆ E E 2 − ∆2 L (V − E)2 − ∆2 R V −E [nF (E) − nF (E (17. Let us assume that |∆L | = |∆R | = ∆. For T ﬁnite. However. I is exponentially small for V < ∆. ǫ) [nF (ǫ) − nF (ǫ + eV )] (17. ∆L = ∆R eiφ and m∗ = m∗ . ǫ) = δ(ǫ − ξk′ ) Then. SUPERCONDUCTIVITY Let us ﬁrst focus on the ﬁrst term. If the system on the right is also a superocnductor. iǫn )FR (k.316 CHAPTER 17.65) E dE √ [nF (E) − nF (E − eV )] (17. IE = t2 m∗ kF m∗ kF R L 2π 2 2π 2 ∞ ∆ (17. with AR (k′ . We call this current IJ .68) This is one of the few cases in which it is advantageous to do the momentum integrals ﬁrst. The current IE resulting from the tunneling of electrons can be understood in terms of the semiconductor.66) 2 − ∆2 E For T = 0. R L IJ = t 2 m∗ kF 2π 2 m∗ kF 2π 2 2 Im e2ieV t n 2 dξk π∆L n dξk′ ∆∗ ∆L R 2 2 (iǫn )2 − ξk − |∆|2 (iǫn − iω)2 − ξk − |∆|2 π∆∗ R (ǫn − ω)2 + ∆2 iω→iδ iω→iδ = t2 Im e2ieV t ǫ2 + ∆ 2 n .63) cannot.67) − eV )] This is exponentially small (vanishing at T = 0) for eV < ∆L + ∆R .64) Suppose the system on the left is a Fermi liquid. We will call this current IE since it results from the tunneling of electrons. the current described by the second term in (17. iǫn − iω) n iω→iδ (17. IJ = t2 Im e2ieV t d3 k d3 k′ (2π)3 (2π)3 † FL (k. since it was ﬁrst discovered by Josephson.

This action is very similar to our eﬀective action for 4 He: the U (1) symmetry Ψ → eiθ Ψ is broken when Ψ has an expectation value.7) in powers of Ψ and its gradients. (The ﬂow of this DC current will feed back into the electrostatics of the problem and.) The Josephson current cannot be understood with the semiconductor model since it is due to the tunneling of pairs. A DC voltage V leads to an AC Josephson current at frequency 2eV . there will be a DC Josephson current if the superconducting order parameters onthe left and right have diﬀerent phases. turn oﬀ this current. The principal diﬀerence is the electromagnetic ﬁeld. for the sake of simplicity. and include the action of the electromagnetic ﬁeld. It can be understood as an oscillation of the phase diﬀerence between two superconductors.72) . it is an example of a collective mode in a superocnductor.17. eventually. Even if the voltage diﬀerence is zero. Following our analysis of 4 He. then we have: S= dt d3 x (Ψ† (i∂t − A0 ) Ψ + 1 1 |(i∂i − Ai ) Ψ|2 + V (|Ψ|) + E2 − B2 ) (17.6 Collective Modes of a Superconductor If we expand the eﬀective action (17. 17. but let us.71) for some constants a and ρs .6. approximate it by: V (|Ψ|) = a |Ψ|2 − ρs 2 (17. we write: ψ= We can rewrite the action as: S= 1 dτ dd x ( ∂τ δρ + ρs (∂t θ + A0 ) + δρ (∂t θ + A0 ) 2 (ρs + δρ) eiθ (17. As such.69) Im e2ieV t+φ (π∆)2 2 |∆| tanh β∆ 2 sin (2eV t + φ) The Josephson current results from the tunneling of pairs between two superconductors.70) ∗ 2m 8π V (|Ψ|) is actually a complicated function. COLLECTIVE MODES OF A SUPERCONDUCTOR m∗ kF 2π 2 m∗ kF 2π 2 m∗ kF 2π 2 317 = t2 = t2 =t 2 Im e2ieV t+φ 2 n (π∆)2 ǫ2 + ∆ 2 n 1 β∆ tanh |∆| 2 (17.

77) (17. there is no Goldstone boson.75) also implies the London equation: j≡ δS δA = ρs A (17. This is the Meissner eﬀect: the magnetic ﬁeld vanishes in the interior of a superconductor. The action (17. Then we have: S= 1 (∇δρ)2 + aδρ2 2ρs + ∇ · E − δρ A0 2 1 + ρs A2 − ∇×A ) 8π dτ dd x ( + (17. . the would-be Goldstone boson is at the plasma frequency. Although the U (1) symmetry has been broken.78) from which the inﬁnite conductivity of a superconductor follows. Its equation of motion is: ∇ 2 A = ρs A which has solutions: A(x) = A(0) e−λx (17. However. as we saw in chapter 15. The would-be Goldstone boson.75) From the third line. θ.318 + CHAPTER 17. SUPERCONDUCTIVITY 1 (∇δρ)2 + ρs ∇θ − A 2 (δρ + ρs ) 1 E2 − B2 ) + 8π 2 + δρ ∇θ − A 2 + aδρ2 (17. we see that the transverse electromagnetic ﬁeld now aquires a gap. the plasma frequency. Hence. so the free part of this action is S= + dτ dd x (δρ (∂t θ + A0 ) + 1 E2 − B2 ) 8π 1 (∇δρ)2 + ρs ∇θ − A 2ρs 2 + aδρ2 (17. the Coulomb interaction pushes the density oscllation up to a high frequency.73) The ﬁrst two terms can (naively) be neglected since they are total derivatives.76) where λ2 = 1/ρs . has been gauged away.74) Let us take the gauge θ = 0. the Goldstone mode would be an oscillation of the density. To put this more physically.

k2 (A0 )−k = δρk we have: (A0 )−k = 1 δρk k2 (17.79) 2ρs 8π H= From the constraint (the A0 equation of motion).17.82) Since δρ and θ are canonical conjugates. the mode is gapped. k → 0. we see that δρ and θ are canonical conjugates and the Hamiltonian is: dd k 1 1 2 k + a |δρk |2 + (A0 )−k δρk + ρs k2 |θk |2 + E2 + B2 (17.74).83) with ωk = 4 (ρs k2 ) 1 2 1 +a+ k 2 k 2ρs (17. COLLECTIVE MODES OF A SUPERCONDUCTOR 319 From the ﬁrst term in (17. we have: dd k 1 2 1 k +a+ 2 2ρs k |δρk |2 + ρs k2 |θk |2 H= (17.81) (17. since all velocities are much smaller than the speed of light.6. √ ρs (17.84) In the long-wavelength limit.80) Neglecting the magnetic ﬁeld.e.85) . ωk = i. this is of the harmonic osciallator form H= dd k 1 1 2 |Pk |2 + m ωk |Xk |2 2m 2 (17.

SUPERCONDUCTIVITY 17.9 17. k′ ) = V if |ξk | > ωD or |ξk′ | > ωD — (V − Va ) if |ξk | . so one might wonder how superconductivity can occur at all. as we saw earlier. |ξk′ | < ωD with V > 0 and V − Va > 0 so that the interaction is repulsive everywhere. but less repulsive near the Fermi surface – i. there is a large repulsive force due to Coulomb intractions.86) + ∆2 2 If we assume that Λ ≫ ωD and ωD ≫ ∆2 then we have: ∆1 = −V ∆1 m∗ ln 2π Λ ωD − V ∆2 m∗ ln 2π ωD ∆2 .10 17.8 17. the repulsion is much weaker. This repulsion is much stronger than the weak attraction due to the exchange of phonons. At the low energies at which superconductivity occurs. The answer is that the repulsive interaction occurs at short time scales and high-energies. Since a repulsive interaction in the BCS channel is marginally irrelevant. Consider the following illustrative model: V (k.11 The Higgs Boson Broken gauge symmetry** The Josephson Eﬀect-xxx Response Functions of a Superconductorxxx Repulsive Interactions In any real metal.e. it will be logarithmically suppressed. Let ∆(k) = The gap equation is: m∗ 2π m∗ ∆1 = −V ∆1 2π ∆1 = −V ∆1 Λ ∆1 if ωD < |ξk | < Λ or |ξk′ | > ωD — ∆2 if |ξk | < ωD m∗ 2π ωD dξ ωD Λ 1 ξ2 + 1 ξ2 + ∆2 1 ∆2 1 − V ∆2 dξ 0 1 ωD 0 dξ ωD − (V − Va ) ∆2 m∗ 2π ξ 2 + ∆2 2 dξ 1 ξ2 (17.7 17. −Va is the weak attraction on top of the repulsion V .320 CHAPTER 17.

17.16 Experimental techniques*** .12 17.2 Phonon-Mediated Superconductivity-xxx The Vortex State*** Fluctuation eﬀects*** Condensation in a non-zero angular momentum state*** Liquid 3 He*** Cuprate superconductors*** 17.15.87) From the ﬁrst equation.89) ∗ 2π ∆2 1 + m V ln Λ 2π ωD V 1+ 2π V ln Λ ωD >0 (17. ∆1 and ∆2 must have opposite signs. Substituting into the second equation.90) even if Va − V < 0. PHONON-MEDIATED SUPERCONDUCTIVITY-XXX ∆1 = −V ∆1 m∗ ln 2π Λ ωD − (V − Va ) ∆2 m∗ ln 2π ωD ∆2 321 (17.15 17.88) This equation will have a solution if Va − m∗ Hence. we have: ∆1 = − V 1+ m∗ 2π V ln Λ ωD ∆2 m∗ ln 2π ωD ∆2 (17. In other words.1 17. we ﬁnd: ∗ V Va − m ln ωD = 1 (17.15.12.14 17. 17. the bare interaction may be repulsive. but the eﬀective pairing interaction can be attractive because the repulsive part will be logarithmically suppressed.13 17.

SUPERCONDUCTIVITY .322 CHAPTER 17.

3 18. 18.CHAPTER 18 Density waves in solids 18.1 Spin density wave Much of the formalism which we used in the previous chapter can be adapted to the case of density-waves in fermion systems with nested or nearly nested Fermi surfaces.2 18.4 Charge density wave*** Density waves with non-trivial angular momentumxxx Incommensurate density waves*** 323 .

DENSITY WAVES IN SOLIDS .324 CHAPTER 18.

Part VI Gauge Fields and Fractionalization 325 .

.

and that interesting physics occurs only when they are gapless. As we will see.1 The Aharonov-Bohm eﬀect As we have discussed. systems of many particles tend to form energy gaps as a way of lowering their energy. unlike the average American family. and Gauge Fields 19. such properties can be described by gauge ﬁelds. an electron. However. One might be tempted to conclude that their low-energy properties are. It is the fact that fractionalization is necessarily characterized by integers which guarantees that it is stable if it occurs. Braiding Statistics. These topological properties are concomitant with the phenomenon of fractionalization. thin solenoid at the origin which is threaded 327 . presumably electrons. Suppose an inﬁnitely-long. whereby the quantum numbers of the low-energy excitations of a system can be fractions of the quantum numbers of its basic microscopic constituents. Phases which are characterized by fractionalization are stable against small perturbations: if the electron breaks into n pieces. trivial.4 children. which do not require low-energy local degrees of freedom. cannot have 2. as a result.7 The basic idea can be understood by considering the Aharonov-Bohm effect. a small perturbation cannot change this continuously. non-trivial low-energy physics can occur even when a system is fully gapped. A fully gapped system can have non-trivial topological properties.CHAPTER 19 Topology. either because they are tuned to a critical point or because their ground state spontaneously breaks a symmetry.

is not itself a measurable quantity because all physically measurable quantities are invariant under the gauge transformation: Aµ (x) → Aµ (x) − ∂µ χ(x) ψ(x) → eiχ(x) ψ(x) (19. A(x) = ˆ Φ z×x 2π |x|2 (19. B alone cannot be local.1) In other words. the scattering cross-section is non-trivial and depends on Φ even though the electron never enters the region in which B = 0 (the interior of the solenoid). but not globally. Any description of the physics of this system in terms of the electric and magnetic ﬁelds E. ∇ × A = 0 does not always mean that an f (x) exists such that A = ∇f .CHAPTER 19. the cross-section for an electron of momentum p to scatter oﬀ the ﬂux tube is: 1 Φ dσ = sin2 2 dθ 2 2πpsin (θ/2) (19.3) The electromagnetic potential is an example of a gauge ﬁeld. the electromagnetic potential is equivalent under a gauge transformation to zero. a local description can be given in terms of the vector potential.2) is an example of such a topologically non-trivial vector ﬁeld. As result. This redundancy is the price which must be paid in order to retain a local description. when Aµ (x) = ∂µ f for some f . E which satisfy ∇ · B = 0. AND GAUGE 328 FIELDS by ﬂux Φ (in units in which = e = c = 1. electrons are prevented from entering the solenoid. As Aharonov and Bohm showed. Hψ = 1 (p − A)2 ψ 2m (19. By this. . It is locally equivalent to zero. as a result of the singularity at the origin. ∇ · E = 4πρ. However. Aµ .4) The gauge ﬁeld Aµ (x) is a redundant way of parametrizing B. The potential (19. one ﬂux quantum is Φ = 2π) is surrounded by an inﬁnitely-high potential barrier. TOPOLOGY. and they move in a region in which the magnetic ﬁeld is zero. BRAIDING STATISTICS. However. In particular. It must involve some kind of action-at-a-distance so that the magnetic ﬁeld inside the solenoid can aﬀect the electron outside the solenoid.2) by simply including this vector potential in the Hamiltonian. we mean that the vector potential.

The eﬀective ﬁeld theories of these ground states and of the ground states with a ﬁxed number of quasiparticles are called topological quantum ﬁeld theories.1. As we discuss below. as the electron encircles the origin.7) which distinguishes the vector potential from a trivial one. A local description of the physics of these excitations must incorporate gauge ﬁelds for the reason which we saw above. As we will see. The ground state is degenerate. The phase aquired by the electron is independent of how close the electron comes to the solenoid or how fast it moves and depends only on the topology of the electron’s path. these braiding properties are manifested in the ground state on manifolds of non-trivial topology. These phases result from the electronelectron correlations which are encoded in the ground-state wavefunction. . Such a system can be fully gapped.6) This is because. However. the gauge ﬁeld (19. as the above example shows. THE AHARONOV-BOHM EFFECT 329 If we were to try to gauge away the vector potential (19. θ) (19.e.19.5) the wavefunction would no longer be single-valued: ψ(r. have non-trivial braiding properties – when they encircle each other. which is felt by the electron even if it is inﬁnitely far away from the solenoid. the low-energy excitations of a strongly-interacting electron system can aquire similar phases – i.2) gives rise to a ‘topological interaction’. these gauge ﬁelds will be a dynamically generated feature of the low-energy properties of the system. in which case the non-trivial braiding properties of the excitations come into play at the ﬁnite energies at which these excitations are created. it aquires a gaugeinvariant ‘Aharonov-Bohm phase’ ei H A·dl = eiΦ (19. we can work with a vanishing vector potential at the cost of having a multi-valued wavefunction. However. θ) → eiΦθ/2π ψ(r. namely how many times it winds about the origin. a reﬂection of the braiding properties of the quasiparticles. Unlike the electromagnetic ﬁeld. Hence. even at low-energies.2) by taking the singular function f= y Φ Φ tan−1 = θ 2π x 2π (19. it is customary in certain circumstances to separate such topological interactions from ordinary ones which do depend on distance and lump them into particle ‘statistics’.

observed. then we must allow exchanges as well. to weight each of these diﬀerent components by a diﬀerent phase factor. whose elements are the diﬀerent components of trajectory space with a composition operation obtained by simply following one trajectory by another. which we can label by half-integer windings. x2 . If the particles are identical. These phase factors realize an Abelian representation of the braid group. AND GAUGE 330 FIELDS x 1 x 2 x 3 x 1 x 2 x 3 Figure 19. This particular assignment of weights gives consistency with the classical limit. Since these components are not continuously deformable into each other. the quantum-mechanical amplitude for hard-core particles which are at x1 . at the quantum mechanical level. Let us consider the case of two identical particles. . BRAIDING STATISTICS. 19. The braid group is simply the group of integers.1). TOPOLOGY. Consequently. . we have the freedom.CHAPTER 19. as Leinaas and Myrrheim. Wilczek. According to Feynman. . However. The diﬀerent representations of the braid group of two identical particles are la- . . . xn at an initial time t0 to return to these coordinates at a later time t is given by a sum over all trajectories. Each trajectory is summed with weight eiS . as may be seen in ﬁgure (19. with integer n corresponding to the number of times that one particle winds counter-clockwise about the other (negative integers are clockwise windings). the stationary phase condition associated with the classical limit does not constrain these phases. .1: Diﬀerent trajectories of hard-core particles in 2 + 1 dimensions which are not adiabatically deformable into each other. .2 Exotic Braiding Statistics Let us consider the braiding properties of particle trajectories in 2 + 1dimensions (2 spatial and 1 time dimension). a peculiarity of two spatial dimensions is that the space of trajectories of hard-core particles is disconnected.

we use the term ‘statistics’ to describe these representations of the braid group. if α = π. In reality.2: An assignment of phases to diﬀerent disonnected components of the space of trajectories of two particles. we will discuss its ﬁeld-theoretical implementation. Figure 19. they cannot be exchanged). the particles are bosons. The braid group of N particles has more complicated representations which can be non-abelian. For intermediate values of α. the particles are called anyons. it is more like a topological interaction since it is not limited to identical particles. Diﬀerent particle species can have ‘mutual statistics’ when they wind about each other (since they are not identical. First. but a class of its representations is just an extension of the two-particle case: whenever any of N identical particles is exchanged counter-clockwise n times with another. EXOTIC BRAIDING STATISTICS 331 + e 2πια + e 4πια + . To emphasize the distinction between this notion of statistics and the usual one. the phase associated with this is einα . . belled by a phase α.19. where there is no braid group.2. it is realized in all of its glory in the quantum Hall eﬀect. the particles are fermions. This is quite diﬀerent from the case in higher dimensions. In a slight abuse of terminology. and we only have the permutation group – which acts only on identical particles – whose only abelian representations are bosonic and fermionic. If α = 0... so that a trajectory in which one particle is exchanged counter-clockwise with the other n times receives the phase einα . As we will see in the next chapter. however. we will use the term ‘braiding statistics’. this expanded notion of statistics is more than a mathematical curiosity.

2).9) The action (19. Consider the addition of such a term to the action for a system of free fermions: S= ψ † (i∂t − a0 ) ψ + 1 1 † ψ (i∇ − a)2 ψ + 2m 2Φ dt d2 x ǫµνρ aµ ∂ν aρ (19.9) is invariant under the gauge transformation aµ (x) → aµ (x) − ∂µ χ(x) ψ(x) → eiχ(x) ψ(x) up to the boundary term δS = 1 2Φ d2 xχǫij ∂i aj ∂R (19. the gauge ﬁeld aµ has no independent dynamics of its own. as we saw at the beginning of this chapter. up to gauge transformations. When one particle winds around another. as we will discuss in the context of the quantum Hall eﬀect. Such a gauge ﬁeld is produced automatically if we add a Chern-Simons term to the action. TOPOLOGY.8) i where xi is the position of the ith particle. in the spirit of (19. so half of this phase is aquired during an exchange. BRAIDING STATISTICS.10) (19. it aquires a phase. However. If we vary it. AND GAUGE 332 FIELDS 19.CHAPTER 19. An exchange is half of a wind.3 Chern-Simons Theory Non-trivial braiding statistics can be implemented by taking wavefunctions which are multi-valued so that a phase is aquired whenever one particle is exchanged with another. When we consider a bounded region R of the plane. we can make these wavefunctions single-valued by introducing a gauge ﬁeld a (distinct from the electromagnetic ﬁeld A) which gives rise to a vanishing magnetic ﬁeld but is not gauge-equivalent to zero. we can ignore this boundary term. a(x) = Φ 2π ˆ z × (x − xi ) |x − xi |2 (19.12) This constraint completely ﬁxes aµ . Since no time derivative of a0 appears in the Lagrange multiplier. this term will be important. it is completely . it is a Lagrange multiplier. we obtain the constraint: ∇ × a = Φ ψ† ψ (19.11) In an inﬁnite system or on a compact manifold. Hence.

Let us further assume that the Chern-Simons coeﬃcient is an integer m divided by 4π. We will return to this assumption below.14) This theory would appear to be completely trivial.15) and the Hamiltonian vanishes. the eﬀective action at low energies is simply S= m 4π ǫµνρ aµ ∂ν aρ (19. Note. Furthermore. suppose that the theory is deﬁned on the torus. it is . Aγ does not depend on the precise curve γ but only on how many times it winds around the generators of the torus. that the Hamiltonian must be supplemented by the constraint (19. implement anyonic braiding statistics – and it does nothing else. the ground state is a unique. without any other ﬁelds to it) has no other states. According to the constraint. According to the constraint (19. however. Then.12). However.e. Pure Chern-Simons theory (i.13) Thus. the Hamiltonian of the Chern-Simons gauge ﬁeld vanishes. Thus. so that we can integrate them out. The action is S= ψ † (i∂t − a0 ) ψ + 1 1 † ψ (i∇ − a)2 ψ + 2m 2Φ (a1 ∂0 a2 − a2 ∂0 a1 ) (19. Then ∇ × a = 0 can still give rise to non-trivial eiAγ = e H γ a·dl (19.4 Ground States on Higher-Genus Manifolds Let us now imagine that the particles are all gapped. The gauge ﬁeld is ﬁxed by the constraint ∇×a=0 (19.19. 19. the eﬀective action only describes the ground state – or states. GROUND STATES ON HIGHER-GENUS MANIFOLDS 333 determined by ψ(x). nondegenerate state.e.16) if γ winds around one of the non-trivial cycles of the torus. the Chern-Simons term does what we want – i.12) a ﬂux Φ is attached to each fermion. On the inﬁnite plane or the sphere.4. Hence. Let us consider the Chern-Simons action in the gauge a0 = 0.

19) 2π (2) δ (x − x′ ) m (19.18) Since A1 . clear that Aγ is additive in the sense that its value for a curve γ which winds twice around one of the generators of the torus is twice its value for a curve γ which winds once. we have the following equal-time commutation relations: a1 (x). We must work with the gauge invariant quantities eiAi . x for a single particle.3: The basic operators A1 and A2 are constructed from the line integrals of a around γ1 and γ2 . A1 .17) From (19. Since eiA1 eiA2 = e[A1 . A2 are not themselves gauge-invariant.CHAPTER 19.13). 2π] on the torus. Hence.21) (19.A2 ]/2 eiA1 +iA2 we have the commutation relation eiA1 eiA2 = e2πi/m eiA2 eiA1 (19. a2 (x′ ) = i from which it follows that [A1 . A2 ] = 2πi m (19. A2 associated with the two generators of the torus. BRAIDING STATISTICS. If we take coordinates θ1 . TOPOLOGY. we have only two independent variables. They have more complicated comutation relations. AND GAUGE 334 FIELDS γ1 γ2 Figure 19. we cannot simply use the analogy between their commutation relations and those of p. then 2π Ai = 0 ai dθi (19.20) . θ2 ∈ [0.

taking them around one of the two non-trivial loops on the torus and annihilating them. T2 . This is precisely the same algebra (19.22) i. Hence. This has an interpretation in terms of the quasiparticle spectrum of the theory – about which we thought that we had lost all information by going to low energies below the quasiparticle energy gap. . taking them around either of the generators of the torus and annihilating them leads yields two non-commuting operations which encode quasiparticle statistics in the ground state degeneracy.19. On a genus g manifold.21) which we found above. (m − 1)π/m. . then T1 T2 = e2πi/m T2 T1 (19. as depicted on the right of ﬁgure 19. If the quasiparticles have statistics π/m.4. if we know that the ground state degeneracy of a system on a genus-g manifold is mg . This algebra can be implemented on a space of minimum dimension m: eiA1 |n = e2πni/m |n eiA2 |n = |n + 1 (19. GROUND STATES ON HIGHER-GENUS MANIFOLDS 335 Figure 19. equivalently. when we require invariance under large gauge transformations. then one explanation of this degeneracy is that it has non-trivial quasiparticles of statistics 0.4: Creating a quasiparticle-quasihole pair.e.4. . with a minimal representation of dimension m. this generalizes to mg . . π/m. Imagine creating a quasihole-quasiparticle pair.23) because the particles wind around each other during such a process. Call the corresponding operators T1 . the Chern-Simons action transforms under a gauge . Why did we take the Chern-Simons coeﬃcient to be an integer? This is required when we deﬁne Chern-Simons theory on compact manifolds or. On a compact manifold. the ground state is m-fold degenerate.

invariance of the functional integral mandates that we take m to be an integer. AND GAUGE 336 FIELDS transformation deﬁned by a function χ(x) as: m 4π ǫµνρ aµ ∂ν aρ → m 4π ǫµνρ aµ ∂ν aρ + 2πm N (19. . BRAIDING STATISTICS.CHAPTER 19.24) where N is the winding number of the map from x to ei χ(x) ∈ U (1). TOPOLOGY. Hence.

If we suppose that these particles have charge e (with a sign to be determined) and mass m. following Hall. jx . Hence. B = Bˆ are: z dpx = eEx − ωc py − px /τ dt dpy = eEy + ωc px − py /τ dt (20.1) where ωc = eB/m and τ is a relaxation rate determined by collisions with impurities. other electrons. and Φ0 = is the ﬂux quantum.2). Let us. ˆ ˆ and a magnetic ﬁeld. the classical equations of motion of charged particles in an electric ﬁeld. the sign of the charge carriers can be determined from a measurement of the transverse voltage in a magnetic ﬁeld. Hall performed an experiment designed to determine the sign of the current-carrying particles in metals. Furthermore.1 Introduction In 1879. E. according to (20. dpx /dt = dpy /dt = jy = 0. through it. 337 . we must have Ex = ne2 τ jx and −e h Φ/Φ0 B jx (20.2) Ey = − jx = ne |e| e2 N where n and N are the density and number of electrons in the wire. E = Ex x + Ey y. place a wire along the x direction in the above magnetic ﬁelds and run a current.CHAPTER 20 Introduction to the Quantum Hall Eﬀect 20. Φ is the magnetic ﬂux penetrating the wire. ˆ m In the steady state. etc.H.

and. this is roughly what is observed. B. Some insight into this phenomenon can be gained by considering the quantum mechanics of a single electron in a magnetic ﬁeld. ?]. as may be seen in Figure 20.1: ρxx and ρxy vs. the density of charge carriers – i.1 (taken from [?]). one of the highest precision experiments of any kind. in fact. in the quantum Hall regime. ρxy = ν e2 . where ν is a rational number. 2 α = e c . ρxy passes through a series of plateaus. At high temperatures. electrons – can be determined from the slope of the ρxy = Ey /jx vs B. This data was taken at Princeton on a GaAs-AlGaAs heterostructure. In the quantum Hall regime. INTRODUCTION TO THE QUANTUM HALL EFFECT Figure 20. The quantization is accurate to a few parts in 108 . very diﬀerent behavior is found in two-dimensional electron sys1 h tems. magnetic ﬁeld. Let us sup- . namely at low-temperatures and high magnetic ﬁelds.338 CHAPTER 20. making this one of the most precise measurements of the ﬁne structure constant.e. A number of integer and fractional plateaus can be clearly seen. at which ρxx vanishes [?.

called Landau levels.m is concentrated on a narrow ring about the origin at radius rm = ℓ0 2(m + 1). where Lm (z.m (z.7) are analytic functions of z multiplied by a Gaussian factor. 2 e 1 −i ∇ + A (20.8) The state ψn=0. To analyze this degeneracy (and in most of what follows) it will be more convenient to work in symmetric 1 gauge. z )e 4ℓ0 ¯ n |z|2 (20. A = 2 B × r Writing z = x + iy.5) with (unnormalized) energy eigenfunctions: ¯ ψn.m (z. The wavefunctions in the lowest Landau level. INTRODUCTION 339 pose that the electron’s motion is planar and that the magnetic ﬁeld is perpendicular to the plane. y) = eikx φ(y). are highly degenerate because the energy is independent of k. the charge of the electron is −e. and ℓ0 = Let’s concentrate on the lowest Landau level. The general lowest Landau level wavefunction can be written: ψn=0. z ) = z m − 2 Lm (z. n = 0. For now. and in what follows.) If we write the wavefunction ψ(x. ψn=0. z ) = f (z) e ¯ − |z|2 4ℓ2 0 (20.m (z.6) ¯ at energies En = (n + 1 ) ωc . we have: 2 H= m −2 ∂ − z ¯ 4ℓ2 0 z ¯ ∂+ 2 4ℓ0 + 1 2ℓ2 0 (20. we will assume that the electron is spin-polarized by the magnetic ﬁeld and ignore the spin degree of freedom. z ) = z e ¯ m − |z|2 4ℓ0 2 (20. Then the highest m for which ψn=0.3) H= 2m c takes the form of a harmonic oscillator Hamiltonian in the gauge Ax = −By.m lies within the disc is . I will take e = |e|. The Hamiltonian.20. z ) are the Laguerre polynomials n 2 /(eB) is the magnetic length. (Here. Suppose the electron is conﬁned to a disc in the plane of area A. Ay = 0. then: Hψ = 1 2m eB c 2 y+ k + 1 (−i ∂y )2 φ(y) eikx 2m (20.1.4) 1 The energy levels En = (n + 2 ) ωc .

and we expect to ﬁnd the quantized Hall conductance. These simple considerations neglected two factors which are crucial to the observation of the quantum Hall eﬀect. Inverting the semi-classical resistivity matrix. though h intuitive.9) for p ﬁlled Landau levels. such a potential will break the degeneracy of the diﬀerent states in a Landau level.9). Hence. we see that in the thermodynamic limit. Clearly. namely the eﬀects of impurities and inter-electron interactions. and substituting this electron number. More worrisome. The possible eﬀects of impurities are summarized in the hypothetical density of states depicted in Figure 20. where NΦ = Φ/Φ0 . we need Ne = p(Φ/Φ0 ) electrons. in the fractional quantum Hall eﬀect. To completely ﬁll p Landau levels. µ. or. so long as the chemical potential lies between two Landau levels (see ﬁgure 20. this conclusion. The higher Landau levels have the same degeneracy. 9 20. In fact. an integer number of Landau levels will be ﬁlled.8 The integer quantum Hall eﬀect occurs in the regime in which impurities dominate.2. (20. there are Φ/Φ0 degenerate single-electron states in the lowest Landau level of a twodimensional electron system penetrated by a uniform magnetic ﬂux Φ. still. we would be led to naively expect that the Hall conductance is 2 less than e p when p Landau levels are ﬁlled. electrons and let us ignore the interactions between these electrons. is completely wrong. the energies of the Landau levels will shift relative to the chemical potential. however. where Φ = BA is the total ﬂux. However. be thought of as copies of the lowest Landau level. As the magnetic ﬁeld is varied. mmax + 1 = Φ/Φ0 . but many. Suppose that we ﬁx the chemical potential. Let us now imagine that we have not one. we ﬁnd: σxy = e2 e2 Ne p = h NΦ h (20. In a very instructive calculation (at least from a pedagogical standpoint).340 CHAPTER 20. Prange [?] analyzed the exactly solvable model . Higher Landau levels can. simply. at a qualitative level. interactions dominate. INTRODUCTION TO THE QUANTUM HALL EFFECT given by A = π rmmax .2). Hence. is the possibility that some of the states might be localized by the random potential and therefore unable to carry any current at all.2 The Integer Quantum Hall Eﬀect Let us model the eﬀects of impurities by a random potential in which noninteracting electrons move. The detailed structure of states in higher Landau levels is diﬀerent.

µ E E b. So long as the chemical potential lies between Landau levels. . numerical studies indicate that the extended state(s) occur only at the center of the band.20.2: (a) The density of states in a pure system. (b) Hypothetical density of states in a system with impurities. THE INTEGER QUANTUM HALL EFFECT 341 ρ ρ a. The Landau levels are broadened into bands and some of the states are localized. ρ c.2. E Figure 20. (c) As we mention later. a quantized conductance is observed. The shaded regions denote extended states.

which relate it to gauge invariance. In this case. and the conductance remains 2 at the quantized value. INTRODUCTION TO THE QUANTUM HALL EFFECT Φ E r R µ a. This one. a single localized state. there will simply be a Landau level. States at radius r will have energies similar to to those depicted in Figure 20. of electrons in the lowest Landau level interacting with a single δ-function impurity. is formed. we will turn to the beautiful arguments of Laughlin (and their reﬁnement by Halperin [?]). This calculation gives an important hint h of the robustness of the quantization. but cannot be easily generalized to the physically relevant situation in which there is a random distribution of impurities. σxy = e .342 CHAPTER 20. b. the quantum Hall eﬀect is independent of the shape of the sample. as an experimental fact. which carries no current. To understand the quantization of the Hall conductance in this more general setting. (b) Hypothetical distribution of energy levels as a function of radial distance. is particularly convenient. we can choose any geometry that we like. with energies that are pushed up at the edges of the sample by the walls (or a . Since. Outside the impurity region.3: (a) The Corbino annular geometry. r impurity region R Figure 20. the Corbino geometry. Let us consider a two-dimensional electron gas conﬁned to an annulus such that all of the impurities are conﬁned to a smaller annulus. The current carried by each of the extended states is increased so as to exactly compensate for the localized state.3. as shown in Figure 20.3.

µ. is above the lowest Landau level. is that there are gapless states at the edges of the system. Then the only states at the chemical potential are at the inner and outer edges of the annulus and. I dt = e and V dt = dΦ = h/e. in the impurity region. by construction. Let us suppose that the chemical potential. In other words. the ﬂux can be gauged away everywhere in the annulus. Then. Locally. Hence. THE INTEGER QUANTUM HALL EFFECT 343 smooth conﬁning potential). Hence. these states are unaﬀected by impurities. the Landau level will broaden into a band. possibly. Let us further assume that the states at the chemical potential in the impurity region – if there are any – are all localized. In the impurity region. If there were – and there probably are in samples that are too dirty to exhibit the quantum Hall eﬀect – then the above arguments break down.10) Clearly. localized states. Then. Localized states are unaﬀected by the ﬂux. µ > ωc /2. the associated vector potential is pure gauge.20. Only extended states can be aﬀected by the ﬂux. according to the adiabatic theorem.2. so: dt I= e2 V h (20. so long as the strength of the impurity potential is small compared to ωc . As a result. states far from the chemical potential will be unable to make transitions to unoccupied states because the excitation energies associated with a slowly-varying ﬂux will be too small. The only possible diﬀerence will be in the occupancies of the extended states near the chemical potential. The other crucial assumption. emphasized by Halperin [?]. Φ(t) = pΦ0 . let us slowly thread a time-dependent ﬂux Φ(t) through the center of the annulus. we know how they are aﬀected by the ﬂux: each ﬂux quantum removes an electron from the inner edge and adds an electron to the outer edge. Hence. which do not wind around the annulus. if the chemical potential is above the center of the band. Now. the system will be in some eigenstate of the Φ(t) = 0 Hamiltonian. Numerical studies [?] indicate that. the key assumption is that there are no extended states at the chemical potential in the impurity region. the only states that will be aﬀected are the gapless states at the inner and outer edges. this was guaranteed because there were no impurities at the . the conditions of our discussion are satisﬁed. Since. When an integer number of ﬂux quanta thread the annulus. In the special setup which we assumed. are completely unaﬀected by the ﬂux.2). extended states exist only at the center of the Landau band (see Figure 20. the single-electron states will be unchanged. the Hamiltonian in the annulus is gauge equivalent to the zero-ﬂux Hamiltonian.

As we will see below. once again. we would have σxy = p e . we can argue that for an annular setup similar to the above there are no extended states at the chemical potential except at the edge. Impurities are not expected to aﬀect this because there can be no backscattering in a totally chiral system. some of the Hall current will be carried by the edge states at the two cuts (i. be left with the uneasy feeling that these gauge invariance arguments are somehow too ‘slick. there is a non-zero gap between the ground state and the lowest excited state. the fractional quantum Hall eﬀect will be destroyed. relate the existence of gapless edge excitations to gauge invariance. ωc ≫ e0 . consider the annulus with a wedge cut out. Laughlin’s argument for exact quantization will apply to the fractional quantum Hall eﬀect if we can show that the clean system has a gap. while ωc = e0 at 6T . In what follows. Hence. One might. if threading q ﬂux quanta removes p electrons from the inner edge and adds p to the 2 outer edge. However. higher Lan3 ℓ dau levels are probably unimportant qualitatively. Then. these gapless states are a onedimensional chiral Fermi liquid. In a sample with density 1. including Coulomb repulsion.25 × 10−11 cm−2 . the impurity potential will be a weak perturbation which pins the quasiparticles but does not drastically aﬀect the physics of the ground state. This degeneracy is broken by fairly generic interactions. which is topologically equivalent to a rectangle (see the article by D. This assumption will be valid in the limit that the cyclotron energy is much larger than the Coulomb 2 interaction energy. We will further assume that we can ignore all higher Landau levels.J. Then. More general arguments. we will try to understand the physics of a partially ﬁlled Landau level of interacting electrons in a clean system.’ To allay these worries. as we would expect at ν = p/q. but could lead to some A partially ﬁlled Landau level of non-interacting electrons has a highly degenerate ground state in the absence of impurities.344 CHAPTER 20. INTRODUCTION TO THE QUANTUM HALL EFFECT edges. to annular topology. ℓ 2 . at special ﬁlling fractions.e. the edges which run radially at ﬁxed azimuthal angle). at ﬁrst. q h 20. however.3 The Fractional Quantum Hall Eﬀect: The Laughlin States the ν = 1 state is seen at 15T . Thouless in the ﬁrst reference in [?]). which we will mention in the context of the fractional quantum Hall eﬀect. In the integer quantum Hall eﬀect. probes which measure the Hall voltage between the two cuts will eﬀectively couple these two edges leading. In such a case. If the sample is too dirty. In very clean samples.

Hence. z2 . Vm = m|V |m . m ≥ k remain degenerate at E = 0. Let us. zk . k = i. . Dropping the constant ωc . z) (we will be sloppy and drop the Gaussian factors because they are unimportant for this analysis). (zi − zj )3 is a factor of ψ for all i = j. and z1 and z2 are the coordinates of the two electrons. . then Hψ = 0. zN ) can be written: ij Vm Pm (20.12) m odd The restriction to odd m is due to Fermi statistics. . zN ) (20. j) (20.3. which are given in position space by z|m = z m . . Then the states ψ(z) = z m are pushed up to energies Em = Vm for m < k but the states ψ(z) = z m . We separate the center-of-mass and relative motions.13) ψ= m odd (zi − zj )m Fm (zi + zj . To do this. following Haldane [?]. where Z = z1 + z2 . The Hamiltonian has no kinetic part in the lowest Landau level. In this case. m|V |m′ vanishes for m = m′ if V is rotationally invariant. The Hamiltonian for the N -electron problem is just: H= i>j m odd ij where Pm projects the i − j pair onto a state of relative angular momentum m. but unrealistic interaction V1 = 0.14) If we take Fm = 0 for m = 1. d dz (20.20. Then. Vm = 0 for m > 1. Let us consider the simple. ψ = P (Z. Any wavefunction in the lowest Landau level. z = z1 − z2 .15) i>j . THE FRACTIONAL QUANTUM HALL EFFECT: THE LAUGHLIN STATES 345 quantitative corrections. ψ = P (z1 . V . which depends only on the relative motion H=V z. let us ﬁrst look at the two-electron problem in the lowest Landau level. Lz |m = m|m . z2 .11) Let us now switch to a basis of relative (canonical) angular momentum eigenstates. . consider the special interactions for which the Laughlin states are the exact ground states. . . we can write: H= Vm Pm (20. Suppose we take Vm > 0 for m < k and Vm = 0 for m ≥ k. Pm is the projection operator onto states of relative angular momentum m. . the following wavefunctions all have zero energy ψ= (zi − zj )3 S(z1 . it is given simply by the interaction.

In a conﬁning potential. it has remarkably large overlap with the exact .19) L at ﬁxed angular momentum L (A is the area of the system). INTRODUCTION TO THE QUANTUM HALL EFFECT where S(z1 . A more precise and general way of stating this result involves calculating the compressibility. . N The Laughlin wavefunction (20. the Laughlin state of N electrons occupies area A = 2πℓ2 (3(N − 0 0 1) + 1) = 2π /(eB)(3(N − 1) + 1). The discontinuity in the chemical potential N implies incompressibility according to (20. zN ) is a symmetric polynomial. Hence. this will be the ground state. Had we chosen Vm > 0 for m < 2k + 1 and Vm = 0 for m ≥ 2k + 1. the Laughlin wavefunction [?].19). Since the single-electron state with canonical angular momentum m encloses area 2πℓ2 (m+1). κ κ= A N2 ∂N ∂µ (20.17) was initially proposed as a trial variational wavefunction for electrons interacting with Coulomb interactions. ν is ν= N 1 N = → Φ/Φ0 3(N − 1) + 1 3 (20. ψ3 = (zi − zj )3 (20. The total ﬂux piercing this area is Φ = BA = Φ0 (3(N − 1) + 1). but the discontinuity will persist.16) with the power 3 replaced by 2k + 1: ψ2k+1 = i>j (zi − zj )2k+1 (20.16) i>j is the most spatially compact droplet. at least one pair of particles will have relative angular momentum m = 1. that is.18) in the thermodynamic limit.346 CHAPTER 20. µ− = E0 (N ) − E0 (N − 1) = 0 while N µ+ = E0 (N + 1) − E0 (N ) = 0. The other symmetric polynomials correspond to quasiholes and edge excitations. it may no longer be true that µ− = 0. For small numbers of electrons. to get ν < 1/3. . . the ﬁlling fraction. Hence.17) The maximum power of any zi in the Laughlin state is 3(N − 1). we would have found the wavefunction (20. To compress this state. For more realistic potentials. E0 (N ) = E0 (N − 1) = 0 but E0 (N + 1) > 0 for ﬁxed total angular momentum 3N (N − 1). Of these wavefunctions. z2 . These states describe droplets of electrons. . which costs a ﬁnite amount of energy. For our choice of interaction.

say the Coulomb interaction. by arguments very similar to those used in the annulus geometry. In particular. THE FRACTIONAL QUANTUM HALL EFFECT: THE LAUGHLIN STATES 347 ground state (see. A state at ν = 1/(2k + 1) is speciﬁed by the positions of the other zeroes. The quantum Hall eﬀect breaks down precisely when the zeroes dissociate from the electrons. there is a 2k + 1-fold zero at the positions of the other electrons. then. there are still (2k + 1) zeroes bound to each electron. |ψ|2 = e 2k+1 (2(2k+1) 1 2 P ln |zi −zj |−(2k+1) P |zi |2 /4ℓ2 ) 0 = e−βHplasma (20.D. will have (2k + 1) zeroes for each of the N − 1 electrons. Since the electrons are fermions.10 Combining this result with our earlier discussion of the compressibility. the article by F.3. zN −1 of N − 1 of the electrons. the article by R. . there must be at least one zero at the positions of the other electrons. leaving an eigenstate of the original Hamiltonian with a deﬁcit of 1/(2k + 1) of an electron at the origin [?]. since the temperature T = 2k + 1 is above the melting temperature for the plasma.B. The ﬂux tube can be gauged away. an excitation with an excess charge of −e/(2k + 1) at the origin would be created. we can say that the Laughlin states describe incompressible quantum liquids. considered as a function of the remaining electron. zN .20. To establish the quantum Hall eﬀect in these states.’ . we expect charge e/(2k + 1) to be transported from the insertion point to the outer edge of the system. . for instance. .M. At ﬁlling fraction ν = 1/(2k + 1). all of the zeroes are at the electron locations. Haldane in the ﬁrst reference in [?]).’ If the inserted ﬂux were oppositely directed. but they are slightly displaced from the electron. In other words. if we ﬁx the coordinates z1 . In the Laughlin state. the wavefunction must be a homogeneous polynomial of degree (2k +1)N (N −1)/2.11 Such an excitation is called a ‘quasihole. for instance. we must show that there is a ﬁnite energy gap separating the ground state from excited states. For instance. z2 .20) the expectation value of many operators in the ground state is just given by the corresponding expectation values in the plasma. we need to understand the excitation spectrum. If we imagine adiabatically inserting a ﬂux tube at the origin in a Laughlin state at ν = 1/(2k + 1). In the exact ground state of electrons with some other kind of interaction. then the wavefunction. A particularly useful technique for obtaining many properties of the Laughlin states is the plasma analogy (see. or a ‘quasiparticle. Since |ψ|2 is of the form of the Boltzmann weight for a classical ﬁnite-temperature plasma of charge 2k + 1 particles in a neutralizing background. Laughlin in the ﬁrst reference in [?]). . we can conclude that the correlation functions of the density do not exhibit long-range positional order.

this is not the smallest gap in the system. The state: qh ψ2k+1 = (zi − η)2k+1 ψ2k+1 (20.13) has not been found. INTRODUCTION TO THE QUANTUM HALL EFFECT Laughlin suggested the following quasihole state. A quasiparticle wavefunction which is an exact eigenstate of the Hamiltonian (20. the state (20.348 CHAPTER 20. we can think of the quasihole and quasiparticle – which 0 have core sizes on the order of a few magnetic lengths – as point charges. A quasiholequasiparticle pair at ﬁnite separation will have lower energy as a result of the Coulomb interaction between them. the angular momentum of each electron is increased by one and the net ﬂux penetrating the electron droplet is increased by one ﬂux quantum. the gap between the ground state and a state with a widely-separated quasihole-quasiparticle pair is just (µ+ − µ− )/(2k + 1). Hence. As a general rule.13) and has a large overlap with the exact quasihole state of a system with a small number of electrons interacting through the Coulomb interactions. The quasiparticle has fractional charge −e/(2k + 1). However.23) has reasonably good overlap with the exact quasihole state in systems with a small number of electrons. The pair will move in a straight line with velocity vk = ∂E(k) ∂k 0 perpendicular to the line connecting . The trial wavefunction: qh ψ2k+1 = i ∂ − η ψ2k+1 ∂zi (20. In this state. ∆ = (µ+ − µ− )/(2k + 1) is the gap which is measured in transport experiments – for instance from ρxx ∼ e∆/T – since a widely separated pair must be created to carry a longitudinal current. Most of the properties of quasiparticles can be inferred from those of quasiholes. qh ψ2k+1 = i (zi − η) ψ2k+1 (20. so we will primarily discuss quasiholes.21) which is an exact zero-energy eigenstate of the Hamiltonian (20. At ν = 1/(2k + 1). This follows from the deﬁnition of µ± and the fact that a widely separated pair will have no interaction energy.21) has charge e/(2k + 1) at η. Suppose the distance between the quasihole and quasiparticle is parametrized by k so that the distance is kℓ2 ≫ ℓ0 .22) i looks like the ground state of N + 1 electrons. Then. but with a deﬁcit of one electron at the position η. exact quasiparticle eigenstates are more diﬃcult to come by than quasihole states. e2 the energy of the pair will be E(k) = ∆ − kℓ2 [?].

This is remarkable. since m quasiholes is equal to a deﬁcit of one electron. the quasihole-quasiparticle pair evolves continuously into a collective mode.21). • It is an excellent approximation to the ground state of electrons in a magnetic ﬁeld interacting through the Coulomb potential. 2 20. The name magneto-roton stems from the fact that this collective excitation is obtained in the single-mode approximation just as the roton was in Feynman’s analysis of superﬂuid 4 He. From the wavefunction (20. we see that ei(2k+1)φ = e2πi and . we don’t want too much disorder since this might simply pin the electrons and prevent them from forming a correlated state (20.e. the Laughlin state has the following properties: • It is a wavefunction describing electrons in a strong magnetic ﬁeld. to which we now turn. The electrons are assumed to be in the lowest Landau level. then the phase for taking one electron around another 2 is ei(2k+1) φ . • It is the non-degenerate ground state of a model repulsive Hamiltonian (20. but has a precedent in polyacetylene. i. FRACTIONAL CHARGE AND STATISTICS OF QUASIPARTICLES 349 them since the Coulomb force between them will exactly balance the Lorentz force due to the magnetic ﬁeld. it is a liquid. If we suppose that the phase acquired by the wavefunction when one quasihole moves around another is eiφ . • The state does not break translational symmetry.13). the analogy between the quantum Hall eﬀect and superﬂuidity can be further exploited. We found earlier that these excitations carry fractional electric charge. the magneto-roton [?]. At low k. h σxx = 0. However. and the phase associated with taking an electron around a quasihole is ei(2k+1)φ . 1 In order to observe a fractional quantum Hall plateau with σxy = 2k+1 e . is perhaps even more exotic.16). we also need a small amount of impurities as well. the statistics.20.4. • The state is incompressible. To summarize. in order to pin any quasiparticles which are produced by small changes of the magnetic ﬁeld or electron density.4 Fractional Charge and Statistics of Quasiparticles Let us return to a discussion of the quantum numbers of the quasiholes and quasiparticles. As we will see later.

these quantum numbers can be determined directly. parameterized by {λ1 . and a Berry’s phase will accumulate. γn remains ﬁnite in T the adiabatic limit. INTRODUCTION TO THE QUANTUM HALL EFFECT ei(2k+1) φ = e2π(2k+1)i . Schrieﬀer. then. λk }. 0 E(t) dt.24) Suppose we vary the λi ’s slowly with time 13 . by o γn = i n({λi (t)})| d |n({λi (t)}) dt (20. Recall that the adiabatic theorem deals with a family of Hamiltonians.27) is reparameterization invariant. λi = λi (t). in particular. so Berry’s phase depends only on the path in parameter space. Often. where M is a phase. but this is unnecessary. . H({λi }).28) i . but we need not be concerned with the details of the Hamiltonian to do this calculation. o H({λi (t)})|ψ(t) = so that |ψ(T ) = M eiγn e i d |ψ(t) dt |ψ(0) (20. .25) RT 0 E(t) dt (20. γn . such that H(λi (0)) = H(λi (T )). this eigenstate evolves. with non-degenerate eigenstates |n({λi }) : H({λi }) |n({λi }) = En ({λi })|n({λi }) (20. As the location of the pinning potential is moved.12 These heuristic arguments for the charge and statistics of the quasiholes and quasiparticles are inadequate even though they give the correct answers. say. and Wilczek [?]. we will be interested in the Berry’s phase acquired by quasihole wavefunctions as the quasiholes are moved around.27) only depends on the Hamiltonian implicitly. according to Schr¨dinger’s equation (20. . Similar arguments would lead us to conclude that quasiparticles have the same statistics. λ2 .350 2 CHAPTER 20.27) The integral (20. One other point worth mentioning is that Berry’s phase (20.25). we require M = 1. This would lead us to conclude that eiφ = e2πi/(2k+1) . unlike the dynamical phase.26) Berry’s phase. is given. the Laughlin quasihole |ψ(t) = (η(t) − zi ) ψ2k+1 (20. Following Arovas. We consider. We will implicitly assume that there is some Hamiltonian with a pinning potential. A state |ψ(t) satisfying |ψ(0) = |n({λi (0)}) will evolve subject to Schr¨dinger’s equation. Fortunately. we will calculate the Berry’s phase [?] acquired when quasiholes are moved around loops. then |n({λi (0)}) = M |n({λi (T )}) . In what follows. for which the state with a quasihole at η is a non-degenerate eigenstate. .

32) 2 up to corrections of order rc /AR . now. Then. we would ﬁnd: γn = i = −2π =− d2 z dη d2 z ρ(z) ρ(z) η−z (20. ρ(z) would no longer be given 1 1 by ρ0 = AR 2k+1 Φ/Φ0 . There would be a charge deﬁcit at the position of the second quasihole. In other words.29) i d2 z ψ(t)| ρ(z) |ψ(t) η(t) − z i where ρ(z) is the density. the Berry’s phase acquired in a circuit. bounding a region R of area AR is: γn = i dη d2 z ρ(z) η−z 1 2k+1 (20. where rc is the size of the quasihole. This is just the phase that we would expect for a particle of charge νe‘ = e/(2k + 1) in a magnetic ﬁeld. If the loop C had enclosed another quasihole. Since d dη |ψ(t) = dt dt we can rewrite ψ(t)| d |ψ(t) = dt = d2 z 1 ψ(t)| η(t) − z δ(z − zi ) |ψ(t) (20.30) 1 |ψ(t) η(t) − zi (20. ρ(z) = ρ0 except in the core of the quasihole. γn = i = −2π =− d2 z dη ρ(z) η−z d2 z ρ(z) 2π (ΦR /Φ0 ) 2k + 1 (20. that we had considered a multi-quasihole wavefunction. C. there is an additional phase 2π/(2k + 1) acquired when one quasihole winds around another. Suppose.4. quasiholes in the Laughlin state at . Then. Since ρ0 AR = N = where ΦR is the ﬂux in the region R.31) ΦR /Φ0 .33) 2π 2π (ΦR /Φ0 ) + 2k + 1 2k + 1 Hence.20. FRACTIONAL CHARGE AND STATISTICS OF QUASIPARTICLES 351 and take η(t) to move slowly around some loop as a function of t.

. . {λi (t)} dt (20. The Berry phase matrix. Vm−1 = 0. . By the arguments given there. which are functions deﬁned on the torus. In such a case. which can occur when there is a set of degenerate states. INTRODUCTION TO THE QUANTUM HALL EFFECT ν = 1/(2k + 1) have fractional statistics given by the statistics parameter α = 1/(2k + 1). {λi (t)} at t = 0 and t = T can diﬀer by an overall rotation.36) . {λi (0)} evolves into: |ψ(T ) = eiγab e i RT 0 E(t) dt M |b. . . α = 1. where bosons have α = 0 and fermions. {λi } . g. Then the ϑ-functions are deﬁned by: ϑ1 (z|τ ) = and 1 ϑ2 (z|τ ) = ϑ1 z + |τ 2 1 ϑ3 (z|τ ) = eiπτ /4 eiπz ϑ1 z + (1 + τ )|τ 2 ∞ n=−∞ eπi(n+ 2 ) τ e2πi(n+ 2 )(z+ 2 ) 1 2 1 1 (20. {λi (t)}| d |b.352 CHAPTER 20. . . The fractional charge and statistics of the quasiholes are the characteristic features of fractional quantum Hall states. where τ is some complex number which is called the modular parameter of the torus.5 Fractional Quantum Hall States on the Torus As we discussed in the last chapter the existence of anyonic quasiparticles in a system is reﬂected in its ground state degeneracy on higher-genus surfaces. 2. = 0. M is the matrix which implements this rotation. . we will make use of the Cauchy ϑ-functions. we expect the Laughlin state for ν = 1/m to be m-fold degenerate on a torus. Vm = Vm+1 = . a = 1. we will be interested in non-Abelian statistics.34) The degenerate subspaces must be equivalent at t = 0 and t = T since the Hamiltonians coincide. . {λi (0)} (20. Let us assume that z is a complex coordinate on the torus and that the torus is deﬁned by z ≡ z + 1. In order to do so. . In this section.35) 20. is given by: γab = i a. we will construct the m wavefunctions on a torus which are annihilated by the Hamiltonian with V1 . In chapter 4. but the states |a. |a. z ≡ z + τ . γab . a state |ψ(t) satisfying |ψ(0) = |a.

The only remaining requireent is that it be periodic under z → z + 1. These will be satisﬁed if −e iπτ mN (−1)mN eiK = (−1)Nφ eiK = 1 P eiKτ e2π a Za = (−1)Nφ eiK = 1 (20.6. (Haldane. From Figure (20. Since they are charged particles in a magnetic ﬁeld.1). ?]. observe that the ratio between any two wavefunctions associated with two choices of K. To see this.38) Armed with these functions. we have only explained the existence of the quantized Hall plateaus at ν = 1/(2k + 1). we can see that there are plateaus at several other odd-denominator fractions. we might expect that the quasiholes or quasiparticles themselves would be in a primary state (e.39) At short distances.6 The Hierarchy of Fractional Quantum Hall States Thus far.37) The following properties of ϑ1 will be useful: ϑ1 (z + 1|τ ) = −ϑ1 (z|τ ) ϑ1 (z + τ |τ ) = −e−iπτ e−2πiz ϑ1 (z|τ ) ϑ1 (−z|τ ) = −ϑ1 (z|τ ) to: m (20. THE HIERARCHY OF FRACTIONAL QUANTUM HALL STATES 1 ϑ4 (z|τ ) = ϑ1 z + τ |τ 2 353 (20. These other states can be thought of as descending from the Laughlin states [?. Za . By a special case of the Riemann-Roch theorem. we can generalize the Laughlin wavefunction ψ= i>j [ϑ1 (zi − zj |τ )]m eiKZ a=1 ϑ1 (Z − Za |τ ) (20. z → z + τ .20. so this wavefunction is annihilated by the Hamiltonian which annihilates the Laughlin wavefunction on the plane. Za is a meromorphic function of Z on the torus with m simple poles. a Laughlin state) at certain preferred . there are m linearly independent such functions. let us consider a ‘primary’ state at ν = 1/(2k + 1) with a ﬁnite density of quasiholes or quasiparticles. 1984) 20. however.g. Following Halperin.40) There are m diﬀerent choices of K. ϑ1 (zi − zj |τ ) → zi − zj .

the quasiholes or quasiparticles could form Wigner crystal states at some ﬁlling fractions rather than quantum Hall states. fractional quantum Hall states are not observed at all of these fractions. INTRODUCTION TO THE QUANTUM HALL EFFECT quasihole densities.42) Every odd-denominator fraction less than 1 can be obtained in this way. respectively. z ¯ Hence. 20. (20.41) If we now imagine the quasiholes or quasiparticles of this state forming a Laughlin state. At what densities would we expect this? Electrons form Laughlin states only at ν = 1/(2k + 1) because these are the only ﬁlling fractions at which (zi − zi )1/ν is an acceptable fermionic wavefunction. This can be accomplished by introducing an auxiliary Chern-Simons . 2p2 ± . Hence. However. Following this reasoning.354 CHAPTER 20. As we descend through this hierarchy of states. Furthermore. Of course. even in a pure system. a ‘descendent’ of the ν = 1/(2k + 1) primary state which has quasiholes or quasiparticles in. the preferred ﬁlling fractions for quasiparticles and quasiholes are 1 1 1/(2p− 2k+1 ) and 1/(2p+ 2k+1 ).7 Flux Exchange and ‘Composite Fermions’ Another perspective on the hierarchy of fractional quantum Hall states involves mapping a fractional quantum Hall state to an integer quantum Hall state. and so on. the energy gaps become smaller and hence more easily destroyed by impurities. ﬁlling fraction 1/(2p ± 2k+1 ) states has electron ﬁlling fraction: ν= 1 1 ∓ 2k + 1 2k + 1 1 = 1 (2k + 1) ± 2p 1 2k+1 2p ± 1 2k+1 (20. a Laughlin state of quasiparticles of statistics −1/(2k + 1) 1 would be of the form (zi − zi )2p− 2k+1 . we will get the continued fractions: ν= 2k + 1 ± 1 1 2p1 ± 1 . . so their Landau levels will have (Φ/(2k + 1))/Φ0 states rather than Φ/Φ0 . 1 respectively. while a quasihole state would be 1 of the form (¯i − zi )2p+ 2k+1 since quasiholes have the opposite charge. we should remember that these particles are fractionally charged as well. A Laughlin state of bosonic particles would form at ν = 1/(2k).

If we now replace this ﬁeld by its spatial average. However. . take θ = (±2k + 1)π. 3 3 2 1 1 1 . n. then this mean ﬁeld theory is just the problem of fermions in an eﬀective magnetic ﬁeld eBeﬀ = e ∇ × (a + A) = eB + 2(θ − π)ρ (20. this state will have a gap. e ∇ × a = 2(θ − π)ρ. 2n+1 . . However. FLUX EXCHANGE AND ‘COMPOSITE FERMIONS’ 355 gauge ﬁeld which attaches an even number of ﬂux tubes to each electron. The attachment of an even number of ﬂux tubes has no physical eﬀect since it will change the phase acquired under braiding or exchange by a multiple of 2π. In such a case. instead. 3 . = = .7. we have ν = 1 . then the ﬂuctuations about mean-ﬁeld theory can probably be ignored. n. . . eBeﬀ = eB ± 2π(2k)ρ. 3.. this is just the Laughlin sequence. or 1/νeﬀ = 1/ν ± 2k.47) ν= . Then. Suppose our anyons are actually fermions.46) For n = 1. By exchanging real magnetic ﬂux for the ﬁctitious statistical ﬂux of an auxiliary Chern-Simons gauge ﬁeld. 2. 2 . we have represented the anyons as fermions interacting with a ChernSimons gauge ﬁeld. ν= n 2kn ± 1 (20. (20. are ﬁlled. we have related the Laughlin states to a single ﬁlled Landau level. we can take θ = π and eBeﬀ = eB. Let us consider the more general problem of anyons with statistics parameter θ in a magnetic ﬁeld: H= where e ∇ × a = 2(θ − π) i 1 (p − e (a + A))2 + Hint 2m δ(r − ri ) (20. we could. Let us choose Beﬀ so that an integral number of Landau levels. Since νeﬀ = n. 3 5 7 These are the ﬁlling fractions of the hierarchical sequence descending from ν = 1. . approximations that would have seemed unnatural without the auxiliary gauge ﬁeld appear quite sensible in its presence. n If we ﬁx k = 1 and consider νeﬀ = 1.20.43) (20... . . 3 5 7 1 1 3− 3− 2 1 2− 2 .44) Here. since this will give fermionic statistics as well. .45) If there is a state of fermions in Beﬀ with a gap.

hence. the lowest Landau level projection is accomplished by moving all of the factors of zi to the left and ¯ ∂ making the replacement zi → ∂zi . That they do can be shown by demonstrating that both constructions lead to states with quasiparticles of the same charge and statistics and.356 CHAPTER 20. At an operational level. the same ground state degeneracy on a torus. so it has vanishing projection into the lowest Landau level. How¯ 2k will multiply this by many more powers of ever. As we have seen. we have only shown that there are quantum Hall states obtained by the ‘composite fermion’ construction at the same ﬁlling fractions at which there are hierarchical states. INTRODUCTION TO THE QUANTUM HALL EFFECT Successive levels of the hierarchy are thereby related to states with additional ﬁlled Landau levels. This object ﬁlls an integer number of Landau levels of the remaining. At this point. This mapping has even proven to be useful starting point for a quantitative (zi − zj )2k Ψn (zk ) (20. The term ‘composite fermion’ refers to a composite object formed by an electron and an even number of ﬂux quanta. the fractional quantum Hall states of electrons are integer quantum Hall states of ‘composite fermions’ [?]. The wavefunction Ψn (zk ) is the wavefunction of n ﬁlled Landau levels. It may be shown that the resulting expression has large projection into the lowest Landau level. Jain [?] used the ‘composite fermion’ construction to motivate the foln lowing trial states for the ﬁlling fractions ν = 2kn+1 : n Ψ 2kn+1 (zk ) = PLLL i>j PLLL indicates projection into the lowest Landau level. and will contain powers of zi . but ubiquitous. weakly interacting) fermion system at a diﬀerent ﬁlling fraction has shed considerable light on the fractional quantum Hall eﬀect.48) . uncanceled magnetic ﬁeld. jargon. Similar constructions are also available for spin-unpolarized and multi-layer systems. These wavefunctions have large overlaps ¯ with the exact ground states of systems with small numbers of particles. It is not clear that the two diﬀerent constructions yield states in the same universality class. Here we have considered only the simplest ‘composite fermion’ states. We will show this in the next chapter using the ﬁeld-theoretic descriptions of these states. In somewhat misleading. More complicated states can be constructed by introducing Chern-Simons gauge ﬁelds which only interact with electrons in particular Landau levels. the mapping of an electron system at one ﬁlling fraction to a (presumably. the factor i>j (zi − zj ) zi .

zN ) i>j (zi − zj )m (20. we saw that there were necessarily gapless excitations at the edge of the system. in such a way that 2ρ∆θ = e∆B or. with properties qualitatively unchanged along the diagonals [?]. In the next chapter. ∆ θ π =∆ 1 ν (20. The Laughlin state itself. we will see that the mapping from a fractional quantum Hall state to a Bose superﬂuid is the starting point for eﬀective ﬁeld theories of the quantum Hall eﬀect. In this way. This mapping is a special case of the ﬂux exchange process [?]: if we change the braiding statistics of the particles in a system and. with S = 1 is then the ground state. to Bose systems. To see this.49) then the properties of the system will not change.53) . .50) (20.4 holds. z2 .51) are also annihilated by H. there will be a conﬁning potential V (r) which favors states of lower total angular momentum. at the same time. change the magnetic ﬁeld. The same is true in the fractional quantum Hall eﬀect. If we assume that the ﬂuctuations about mean-ﬁeld theory are small. at the mean ﬁeld level. and the other states are edge excitations. we can map electron systems to other fermion systems. let us consider again our simple Hamiltonian which annihilates the Laughlin state. 20.8.52) where sn = i n zi (20. . They are spanned by: ψ= n (sn )pn i>j (zi − zj )m (20. In a more realistic model.8 Edge Excitations In our discussion of the integer quantum Hall eﬀect. All of the states ψ = S(z1 . EDGE EXCITATIONS 357 analysis. . then the phase diagram of Figure 20.20. or even to systems whose basic constituents are anyons. . equivalently.

. INTRODUCTION TO THE QUANTUM HALL EFFECT 1/ν = eB/ρ ∆(1/ν) = ∆(θ/π) primary Laughlin states incompressible anyon states 3 filled Landau level additional FQHE states 2 HLR state anyon superfluids 1 anyon metals boson superfluid 1/2 1/3 1 Fermi liquid bosons 2 fermions 3 4 fermions ∆(θ/π) (quantum statistics) 1 θ Figure 20.4: Systems at diﬀerent points along the diagonals ∆ π = ∆ ν in the magnetic ﬁeld-statistics plane have the same properties at the meanﬁeld level.358 CHAPTER 20.

Hence.8. . rather than a disk. They will exist in any incompressible chiral ﬂuid. since only pi > 0 are allowed. . . We can now create edge excitations generated by sn = i n zi (20. We will take t ∼ N so that both the inner and outer radii of the annulus are macroscopic. This theory has free bosonic excitations which are divided into m sectors. . relative to the ground state: Hdisk |p1 . so that there are inner and outer edges. Then. . EDGE EXCITATIONS 359 Suppose that Hdisk = H + Vconf . so long as n < t. Essentially. p2 .20. . If we take a quasiparticle in the bulk and move it along a trajectory encircling the origin. . . . . corresponding to states which are built by acting with the sn ’s on ψ= i zi tm+r i>j (zi − zj )m (20. we have carved out the inner hole of the annulus by removing t electrons from the center of the disk. the combined theory of both edges is a non-chiral boson. The Laughlin state on an annulus can be described by: ψ= i zi tm i>j (zi − zj )m (20. to lowest order in the angular momentum. These bosonic excitations are simply the edge waves of an incompressible liquid. = λ pn n |p1 . p2 . This is the spectrum of a free bosonic ﬁeld. p2 . . . then for reasonable values of n. consider a quantum Hall state on an annulus. p2 .56) . = f (M ) |p1 . . we will have excitations of both chiralities. but a chiral one. p2 . M . . ≈ (const. To see how the edge excitations of a given quantum Hall state depend on the particular state. where v = 2πλ/L.54) where t is a large integer so that the inner radius of the annulus is ℓ0 2(tm + 1) while the outer radius is ℓ0 2m(N + t). it will not aquire a phase e2πit = 1. .) M |p1 .55) for both positive and negative n. n =v n pn 2πn L |p1 .

The sectors of the edge theory correspond to the m-fold degenerate ground states of the theory on a torus. . INTRODUCTION TO THE QUANTUM HALL EFFECT where r = 0.. . sectors which diﬀer by the transference of an ineteger number of electrons from on edge to the other are equivalent. m − 1. These diﬀerent sectors correspond to transferring a quasiparticle from the inner edge to the outer edge. as may be seen by gluing the inner and outer edges of the annulus to form a torus. 1. r = 0 corresponds to ‘twisted’ boundary conditions for quasiparticles. .360 CHAPTER 20. . they aquire a phase e2πir/m upon encircling the origin. r = 0 corresponds to periodic boundary conditions for quasiparticles. The diﬀerent sectors may be distinguished by the phases which are aquired when quasiparticles encircle the origin. . Note that r ≡ r + m by shifting t by one.

namely ‘topological ordering.CHAPTER 21 Eﬀective Field Theories of the Quantum Hall Eﬀect 21. What has been missing to this point is a precise sense of which properties of these wavefunctions deﬁne the universality class.’ Recall that in a superﬂuid or a superconductor the oﬀ-diagonal entries 361 . and it is. One important diﬀerence. all of which exhibit the fractional quantum Hall eﬀect. One formulation of these eﬀective ﬁeld theories is in the form of a Landau-Ginzburg theory which is strongly reminiscent of superﬂuidity or superconductivity. However. Their usefulness lies in the fact that they are representatives of a universality class of states. however. long-wavelength eﬀective ﬁeld theories for the fractional quantum Hall eﬀect. This is not a trivial distinction.1 Chern-Simons Theories of the Quantum Hall Eﬀect The preceding discussion has been heavily dependent on Laughlin’s wavefunctions. related to the conclusion that a novel type of ordering is present in the quantum Hall states. these wavefunctions are not the exact ground states of any real experimental system. is that the order parameter is not a local function of the electron variables. ultimately. We alluded earlier to the binding of zeroes to electrons. and which ones are irrelevant perturbations. We will formalize this notion and use it to ﬁnd low-energy.

d2 z2 i |zj |2 /2ℓ2 0 (z − zi )2k+1 (¯′ − zi )2k+1 z ¯ k>l |zk − zl |2(2k+1) e− P (21. . we can choose states of indeﬁnite particle number such that ψ(r) = φ(r). On the other hand. . The drawback of this order parameter is that it is not an analytic function of the z’s. or quasi-long-range order.2). More importantly. As a result of (21.2) in a Laughlin state state at ν = 1/(2k + 1)? ρ(r. but also remains strictly within the lowest Landau level [?]. is not a lowest Landau level operator. so it can be chosen everywhere real and positive. . oﬀ-diagonal long-range order is the hallmark of superﬂuidity. r′ ) → φ∗ (r)φ(r′ ) (21. r2 .5) as may be shown using the plasma analogy [?]. drN ψ ∗ (r. take: φ† (z) = e−im R d2 z ′ ln(z−z ′ ) ψ † (z) (21.1) exhibit oﬀ-diagonal long-range order.4) this phase is removed and we ﬁnd algebraic falloﬀ of correlation functions. . We could. (21. φ† (z ′ )φ(z) ∼ 1 |z − z ′ |(2k+1)/2 (21. EFFECTIVE FIELD THEORIES OF THE QUANTUM 362 HALL EFFECT of the density matrix: ρ(r.2) will hold. . . rN ) (21. if we consider correlation functions of: φ† (z) = e−i(2k+1) R d2 z ′ Im ln(z−z ′ ) ψ † (z) (21.3) This correlation function does not show any signs of long-range order. . φ(r) can be treated as a classical ﬁeld and used to analyze interference phenomena such as the Josephson eﬀect. In the absence of phase variations. . the ﬁeld theory of this operator is more complicated. However.4). rN ) ψ(r′ . . and.6) which not only has true long-range order. . r′ ) = ψ † (r)ψ(r′ ) = dr2 . so we will use (21. The ﬂuctuating phases of the ﬁrst two terms in the integral lead to exponential falloﬀ. . hence. What happens if we calculate (21.2) for some non-zero function φ(r). . ρ(r. . r′ ) = ψ † (r)ψ(r′ ) = d2 z2 .CHAPTER 21. instead. r2 . Feynman argued that the ground state wavefunction of a Bose ﬂuid would have no zeroes.

8) Note that the coeﬃcient of the Chern-Simons term is 2k + 1. interacting with a gauge ﬁeld. excitations about this mean-ﬁeld solution are gapped. |φ|2 = ρ – breaks a U (1) symmetry. vortices are created.8) if |φ|2 = ρ. Furthermore. The Meissner eﬀect results. so that its only role is to transform the bosons φ into fermionic electrons. what we expect for a quantum Hall state: there is a gap to all excitations.9) The Chern-Simons gauge ﬁeld equation attaches 2k + 1 ﬂux tubes to each φ boson. i.7) We now rewrite this in terms of a bosonic ﬁeld.e.21. as in the case of a superconductor. φ(x). or ν = 1/(2k + 1). As in the Landau-Ginzburg theory of a superconductor.4) [?] in the following way. This is. CHERN-SIMONS THEORIES OF THE QUANTUM HALL EFFECT 363 A Landau-Ginzburg theory may be derived for the order parameter (21. of course. We could have chosen any odd integer in order to obtain the correct statistics. To see that the correct statistics are obtained.1. as in a type II superconductor: φ(r. The gauge ﬁeld is given a Chern-Simons action of the type discussed two chapters ago. As one boson is exchanged with another. This implies that B = −∇ × a = 2π(2k + 1)ρ. long-range order in the bosonic ﬁeld φ – i. a + A = 0. θ) = |φ(r)| eiθ . If B is increased or decreased from this value.e. it acquires an AharonovBohm phase of (−1)2k+1 = −1 as a result of these ﬂux tubes. Begin with the Lagrangian for interacting electrons in a magnetic ﬁeld such that ν = 1/(2k + 1): Leﬀ = ψ ∗ (i∂0 − A0 ) ψ + ψ ∗ (i∇ − A)2 ψ − µψ † ψ 2m∗ + V (x − x′ )ψ † (x)ψ(x)ψ † (x′ )ψ(x′ ) 2 (21. Leﬀ = φ∗ (i∂0 − (a0 + A0 )) φ + φ∗ (i∇ − (a + A))2 φ 2m∗ −µφ† φ + V (x − x′ )φ† (x)φ(x)φ† (x′ )φ(x′ ) 1 µνρ 1 + ǫ aµ ∂ν aρ 2k + 1 4π 2 (21. note that the a0 equation of motion is: ∇ × a(r) = 2π(2k + 1) ρ(r) (21. the coeﬃcient 2k + 1 is chosen for reasons which will become clear momentarily. as a result of the Anderson-Higgs eﬀect. since a non-zero constant eﬀective magnetic ﬁeld ∇ × (a + A) would lead to badly divergent energy in (21.

These vortices are the Laughlin quasiholes and quasiparticles.CHAPTER 21. In a superconductor. with a density ∼ e−∆/T . Essentially. destabilize the Bose condensed state. on the other hand.9). they have charge 1/(2k + 1). the electrons have become bound to 2k+1 ﬂux tubes – or 2k+ 1 zeroes as we put it earlier – thereby transmuting them into bosons in zero ﬁeld. the deviation from the zero-temperature behavior is small for small T . the fractional quantum Hall liquids are these Bose condensed states. |φ(∞)| = ρ0 . which satisﬁes ∇×a(r) = (2k+1)ρ(r). Consider a Landau-Ginzburg theory for a U (1) symmetry: L = ψ ∗ i∂0 ψ + 2 2m∗ ψ ∗ ∇2 ψ + V |ψ|2 (21. The ﬂuctuations of a around its average value could. they are thereby diﬀerentiated. has been replaced by its spatial average. this can be made even more explicit. The Chern-Simons term attaches ﬂux to charges. f (∞) = 0. according to (21. They have one ﬂux quantum of the a gauge ﬁeld and very little real magnetic ﬂux. the ChernSimons gauge ﬁeld.2 Duality in 2 + 1 Dimensions The Landau-Ginzburg theory which we have just discussed has a dual formulation which will prove useful in much of the following discussion. so both particles (electrons) and vortices (quasiparticles) carry charge and ﬂux. We will ﬁrst consider duality more generally for a U (1) theory in 2 + 1 dimensions and then consider the particular case of the quantum Hall eﬀect.10) . they are very much on the same footing. As a result of the ﬂux quantum of a which they carry. where ∆ is the (ﬁnite) energy cost of a quasiparticle. Note that the Chern-Simons term results in an important diﬀerence between the Landau-Ginzburg theory of the quantum Hall eﬀect and the Landau-Ginzburg theory of a superconductor. 21. there will always be some thermally excited quasiparticles. The presence of these quasiparticles means that the quantum Hall eﬀect is destroyed at any ﬁnite temperature: ρxx ∼ e−∆/T . particles (Cooper pairs) carry charge while vortices carry ﬂux. but they do not because there is an energy gap. ∇ × a(r) = (2k + 1) ρ . The average ﬁeld cancels the magnetic ﬁeld so that the bosons can condense. in principle. As a result. At ﬁnite temperature. Howver. EFFECTIVE FIELD THEORIES OF THE QUANTUM 364 HALL EFFECT b + B = f (r) √ with |φ(0)| = 0. as we will see later. Said slightly diﬀerently. These bosons undergo Bose condensation.

but is subject to the gauge transformation aλ → aλ + ∂µ χ.13): v ∂ν fµν = ρs jµ (21.12) can be automatically satisﬁed if we take jµ = ǫµνλ ∂ν aλ (21.18) The ﬁnal term in the dual Lagrangian is a Maxwell term for the gauge ﬁeld. Then the vortex current takes the form: v jµ = ǫµνλ ∂ν ∂λ θ (21. we showed that such a theory could.14) Note that aλ is not uniquely deﬁned.17) where κ is a vortex stiﬀness and VΦ (|Φv |) is the vortex-vortex interaction. be simpliﬁed by writing ψ = ρs eiθ and integrating out the gapped ﬂuctuations of ρs : ρs L= (∂µ θ)2 (21. then (21. fµν fµν = e2 − b2 ei = ∂0 ai − ∂i a0 . in its broken symmetry √ phase.13) The conservation law (21.16) If we introduce a vortex annihilation operator. namely vortices. DUALITY IN 2 + 1 DIMENSIONS 365 In chapter 11.2. The vortex density and current are given by v jµ = Im {κΦ∗ (∂µ − iaµ )Φv } v (21. we can allow one type of amplitude ﬂuctuations.14) can be used to solve for θ and substituted into equation (21.11) 2m This Lagrangian has a conserved current. Equation (21. However. if we allow θ to have singularities.21.15) is the equation of motion of the dual Lagrangian: LDual = 1 1 κ |(∂µ − iaµ )Φv |2 + VΦ (|Φv |) + fµν fµν 2 2ρs (21.15) where fµν is the ﬁeld strength associated with the gauge ﬁeld aλ : fµν = ∂ν aµ − ∂µ aν (21. ∂µ jµ = 0 given by jµ = ρs ∂µ θ (21.12) We have assumed that there are no ﬂuctuations in the amplitude. Φv .

topological identity in the dual representation. one gauge-invariant degree of freedom per point – i. the same number of degrees of freedom as a scalar ﬁeld. it is possible. carry one quantum of aµ ﬂux. Fµν Fµν = E 2 − B 2 . Φ. Suppose we consider this theory Leﬀ = φ∗ (i∂0 − (a0 + A0 )) φ + 2 2m∗ φ∗ (i∇ − (a + A))2 φ . Φv = Φ0 = 0. Hence. The deﬁnition (21. Hence. Topological defects in the vortex order parameter.19) which is of the same form as the action for the electromagnetic gauge ﬁeld. EFFECTIVE FIELD THEORIES OF THE QUANTUM 366 HALL EFFECT b = ǫij ∂i aj (21.CHAPTER 21. In 2 + 1 dimensions.11). Notice that the conservation law (21. The broken symmetry phase of our original theory (21. the fundamental quanta are vortices while the topological defects are Cooper pairs. by a duality transformation. In the dual representations. Hence.g.20) The gauge ﬁeld aµ is the dual formulation of the Goldstone boson. In the original representation (21. Vortices are gapped. when the symmetry is restored by the proliferation and condensation of vortices.15) in the dual representation.21) and the gauge ﬁeld aµ becomes massive. the duality operation exchanges particles and vortices.12) which followed from the equations of motion in the original representation is a trivial. Cooper pairs if the LandauGinzburg theory describes a superconductor. The degrees of freedom of the scalar ﬁeld θ have a dual representation in terms of a gauge ﬁeld aµ . to pass between an XY theory and a U (1) Higgs theory.e. a guage ﬁeld has no local degrees of freedom. In 1 + 1 dimensions. However. following from (21.14). Let us now extend this to transformation to the Chern-Simons theory of the quantum Hall eﬀect. the low-energy eﬀective action in the dual language is simply LDual = 1 fµν fµν 2ρs (21.13) of the vortex current in the original representation is the equation of motion (21. a gauge ﬁeld has one transverse component – i.11) is the phase in which Φv = 0. the Cooper pairs are the fundamental quanta while vortices are topological defects.e. they are simply charges – e. as we will see later. Duality in 1 + 1 dimensions connects two scalar ﬁeld theories. the dual action is in its Higgs phase: LDual = κ 1 |Φ0 |2 aµ aµ + fµν fµν 2 2ρs (21.

22) ρeiθ and integrate in its fractional quantized Hall phase.2. they are simply Laughlin quasiparticles and quasiholes. We write φ = out the gapped ﬂuctuations of ρ: 1 1 µνρ 1 Leﬀ = ρ (∂µ θ − aµ − Aµ )2 + ǫ aµ ∂ν aρ 2 2k + 1 4π (21. which appears quadratically. Let us .28) + 2π 2k + 1 4π (21. DUALITY IN 2 + 1 DIMENSIONS + V |φ|2 + 1 µνρ 1 ǫ aµ ∂ν aρ 2k + 1 4π √ 367 (21.21.26) Using the dual expression for the current to eliminate ∂µ θ from the righthand-side of this equation. As we saw at the beginning of this chapter. Integrating out aµ .25) As in the derivation above.23) This theory has a conserved current which is simply the electrical current: jµ = ρ (∂µ θ − aµ − Aµ ) (21. Their current is given by: qp jµ = ǫµνλ ∂ν ∂λ θ (21.24) We construct the dual representation of this current with a gauge ﬁeld α: jµ = ǫµνλ ∂ν αλ (21. it is irrelevant in the long-wavelength limit. we ﬁnd: LDual = 1 κ |(∂µ − iαµ )Φqp |2 + VΦ (|Φqp |) + 2 1 1 2k + 1 µνρ ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ ∂ν Aµ (21.29) 4π 2ρ 2π Since the Maxwell term for αµ has one extra derivative compared to the Chern-Simons term. we have: 1 v ∂ν fµν = jµ − ǫµνλ ∂ν (aµ + Aµ ) ρs This is the equation of motion of the dual Lagrangian: LDual = 1 1 κ |(∂µ − iαµ )Φqp |2 + VΦ (|Φqp |) + fµν fµν 2 2ρ 1 µνρ 1 1 αµ ǫµνλ ∂ν (aλ + Aλ ) + ǫ aµ ∂ν aρ (21.27) where Φ is the vortex annihilation operator. we consider vortices in the order parameter.

When Φqp = Φqp = 0. In the phase in which Φqp = 0. |Φqp (∞)| = ρ0 . the external electromagnetic potential A0 is coupled to the ﬁctitious ﬂux ǫij ∂i αj . Hence. According to the last term in (21. it has charge e/(2k + 1). f (∞) = 0.29). To do this. which can now be integrated out. we must show that the basic soliton in (21. In doing so. The resulting eﬀective action for the electromagnetic gauge ﬁeld is then Leﬀ = 1 κ0 Fµν Fµν 2 (21. We can then integrate out αµ .31) ∂ν Aµ 2 0 4π 2ρ The ﬁrst term gives a Higgs mass to the gauge ﬁeld αµ . thereby renormalizing the Maxwell term. When Φqp = Φqp = 0. Since each quasiparticle has ﬂux 1/(2k + 1) attached to it.30) = 2k + 1 which leaves us with an eﬀective action for the electromagnetic ﬁeld which incorporates the Hall conductance 1/(2k + 1). we have the eﬀective action: 0 LDual = 2k + 1 µνρ 1 κ qp 2 |Φ | αµ αµ + ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ(21. we can neglect the Chern-Simons term. quasiparticles have statistics π/(2k + 1). Since the quasiparticle annihilation operator is coupled to the Chern-Simons gauge ﬁeld. They carry one ﬂux quantum of the gauge ﬁeld. αµ .32) . statistics π/(2k+1) quasiparticle into an action (21.CHAPTER 21.22) in which the basic ﬁeld φ represents a charge e fermionic electron and the basic soliton is a charge e/(2k+1). each quasiparticle has ﬂux 1/(2k + 1) attached to it. this duality transformation has transformed an action (21. electrons are the solitonic excitations of the state in which Φqp condense. Hence. we must consider the state in which Φqp condenses. statistics π/(2k + 1) quasiparticle. θ) = |Φqp (r)| eiθ β + B = f (r) √ with |Φqp (0)| = 0. charge e particles – i.29) in which the basic ﬁeld Φqp represents a charge e/(2k + 1).e. Leﬀ = 1 1 2k + 1 µνρ ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ ∂ν Aµ 4π 2ρ 2π 1 Aµ ǫµνλ ∂ν Aµ (21. we can integrate out the quasiparticles.29) is a charge e fermionic electron. EFFECTIVE FIELD THEORIES OF THE QUANTUM 368 HALL EFFECT drop the Maxwell term and consider the eﬀect of the Chern-Simons term. there are solitons in this state 0 Φqp (r. To complete the correspondence. so they are fermionic. which is irrelevant compared to the Higgs mass.

the system is an insulator. As we saw earlier in our construction of the hierarchy. 21. In this section.34) 4π 2p ˜ Let’s suppose that Φ condenses. the hierarchy construction is simply a variant of duality in 2 + 1 dimensions. we have considered the ‘usual’ case. p = 0. The quantum Hall state with σxy = νe2 /h is dual to the insulating state with σxy = 0 which is formed when quasiparticles condense. Then the corresponding Meissner eﬀect requires that: ∇×α+∇×f =0 (21. f . LDual = Let us now introduce another gauge ﬁeld. therefore.37) (21. which attaches an even number of ﬂux tubes to the quasiparticles and. thereby leading to σxy = (2k + 1 ± 1/p)−1 e2 /h. the α0 and f0 equations of motion are: (2k + 1)∇ × α + ∇ × A = 2π ρqp ∇ × f = −2π(2p)ρqp . as we discuss in the next section.29) is equivalent to the eﬀective action LDual = 1 ˜ ˜ κ |(∂µ − iαµ − ifµ )Φqp |2 + VΦ |Φqp | + ˜ 2 1 1 2k + 1 µνρ ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ ∂ν Aµ 4π 2ρ 2π 1 1 µνρ − ǫ fµ ∂ν fρ (21. THE HIERARCHY AND THE JAIN SEQUENCE 369 In other words. Combining these three equations. quasiparticles can also condense into fractional quantum Hall states. let us begin with the dual theory (21.35) Meanwhile. has no eﬀect. Note that this insulating state is not the only state into which quasiparticles can condense. (21.33) 4π 2ρ 2π To construct the hierarchy.36) .3 The Hierarchy and the Jain Sequence 1 κ |(∂µ − iαµ )Φqp |2 + VΦ (|Φqp |) + 2 2k + 1 µνρ 1 1 ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ ∂ν Aµ (21.29).3.21. In other words. 2k + 1 − 1 2p ∇×α+∇×A=0 (21. Hence.

CHAPTER 21. These charges are not. it is not useful to introduce a single Chern-Simons gauge ﬁeld. conserved: only the total charge is conserved. namely the charge in each Landau level. The problem with such an approach is that we can only introduce n gauge ﬁelds if there are n diﬀerent conserved quantities. we ﬁnd ν = 1/(2k + 1 − 2p ). Instead. it is more useful introduce n gauge ﬁelds. and assume for now that this will not make a diﬀerence. Let’s now consider the ‘composite fermion’ construction for the ‘Jain sequence’ ν = n/(2pn ± 1). we introduced a ChernSimons gauge ﬁeld so that we could represent each electron as a boson attached to m ﬂux quanta. . The Lagrangian can then be written: 2 Leﬀ = ψ ∗ (i∂0 − c0 − A0 ) ψ + ψ ∗ (i∇ − c − A)2 ψ 2m∗ −µψ † ψ + V (x − x′ )ψ † (x)ψ(x)ψ † (x′ )ψ(x′ ) 1 1 µνρ − ǫ cµ ∂ν cρ 2p 4π (21.39) ρ 2p 4π 4π 2 2 where φI annihilates a boson corresponding to an electron in the I th Landau level. we can ﬁnd the Landau-Ginzburg theories for all of the hierarchy states. Continuing in this way. which is such that the ψ’s – the composite fermions – ﬁll n Landau levels in the eﬀective magnetic ﬁeld. To derive the eﬀective theory for this state. At ν = 1/m. We represent the electrons as fermions interacting with a Chern-Simons gauge ﬁeld cµ which attaches 2p ﬂux tubes to each fermion. we must now construct the eﬀective theory for n Landau levels. We will come back to this point later. We can then represent each electron as a boson attached to one ﬂux quantum. in general. Then. At ν = n. each of which is coupled to the electrons in one of the Landau levels. EFFECTIVE FIELD THEORIES OF THE QUANTUM 370 HALL EFFECT 1 Since ∇ × αequiv2πρ.38) The ﬂux of cµ is anti-aligned with the magnetic ﬁeld so the eﬀective magnetic ﬁeld seen by the fermions is ∇ × (c + A). the Lagrangian takes the form: Leﬀ = φ∗ i∂0 − aI − c0 − A0 φI + I 0 φ∗ i∇ − aI − c − A φI 2m∗∗ I ′ −µφ∗ φI + V (x − x′ )φ∗ (x)φI (x)φJ (x )φJ (x′ ) I I 1 1 µνρ 1 µνρ I − ǫ cµ ∂ν cρ + ǫ aµ ∂ν aI (21. which would allow us to represent each electron as an anyon attached to 1/n ﬂux quanta.

simply Ldual = 2k + 1 µνρ µ ǫ αµ ∂ν αρ + Aµ ǫµνρ ∂ν αρ + αµ jvortex + Lvortex 4π (21.e.29).42) where Lvortex is the quasiparticle eﬀective Lagrangian. Consider the Landau-Ginzburg theory for ν = 1/(2k + 1): Leﬀ = φ∗ (i∂0 − (a0 + A0 )) φ+ 1 µνρ 1 ǫ aµ ∂ν aρ 2k + 1 4π (21. which appear quadratically: σH = I.4 K-matrices A compact summary of the information in the Landau-Ginzburg theory is given by the dual theory. we can drop the last term.46) .J tI tJ (K −1 )IJ (21.44) The charge of a vortex (i.41) or. Let’s assume that vortices are gapped.45) and the braiding statistics between vortices of types i and j is: θIJ = (K −1 )IJ . K-MATRICES 371 21.21.40) Let’s apply 2 + 1-dimensional duality to this Lagrangian. Lvortex . (21. We ﬁnd the dual theory: 2m∗ φ∗ (i∇ − (a + A))2 φ+u|φ|4 + LDual = 1 κ |(∂µ − iαµ )Φqp |2 + VΦ (|Φqp |) + 2 2k + 1 µνρ 1 ǫ αµ ∂ν αρ + fµν fµν + αµ ǫµνλ ∂ν Aµ 4π 2ρ 2 (21.4.43) The Hall conductance of such a state can be obtained by integrating out the Chern-Simons gauge ﬁelds. but allow for the possibility that the magnetic ﬁeld is not quite commensurate with the density so that there is some ﬁxed number of pinned vortices. a quasiparticle) of type i is: qI = J tJ (K −1 )IJ (21. following (21. Then. keeping only the most relevant terms. This generalizes to an arbitrary abelian Chern-Simons theory: Ldual = 1 µ KIJ ǫµνρ αI ∂ν αJ + tI Aµ ǫµνρ ∂ν αI + αI jvortex I µ ρ ρ µ 4π (21.22)(21.

52) 0 0 1 p3 .CHAPTER 21. . 1 Leﬀ = κ |(∂µ − ic1 − if µ )Φqp |2 + V (Φqp ) µ 2 2k + 1 µνρ 1 1 µνρ 1 1 1 µνρ + ǫ cµ ∂ν c1 + ǫ cµ ∂ν Aρ − ǫ fµ ∂ν fρ(21. First.34) of a hierarchy state. . .29) to (21. Distinct quantum Hall states are therefore represented by equivalence classes of (K.42): 2k + 1 1 0 0 . We write Φqp = |Φqp |eiϕqp . 1 1 µνρ 1 µνρ 1 ǫ cµ + fµ ∂ν c2 − ǫ fµ ∂ν fρ ρ 2π 4π 2p 1 µνρ 1 2k + 1 µνρ 1 ǫ cµ ∂ν c1 + ǫ cµ ∂ν Aρ + ρ 4π 2π Finally. Continuing in this fashion.47) ρ 4π 2π 4π 2p ˜ ˜ where c1 ≡ αµ . 1 p1 1 0 1 p2 1 Kh = 0 (21.50) (21. . Z) basis changes where κ is the rank of the K-matrix. EFFECTIVE FIELD THEORIES OF THE QUANTUM 372 HALL EFFECT Implicit in the normalizations is the assumption that the charges associated I with the jµ vortex are quantized in integers. c2 : µ µ ǫµνλ ∂ν c2 = Jqp ≡ |φqp |2 (∂ µ ϕqp − cµ − f µ ) λ (21. consider the Landau-Ginzburg theory (21. and apply steps (21. we integrate out fµ : Leﬀ = 2k + 1 µνρ 1 2p µνρ 2 ǫ cµ ∂ν c1 + ǫ cµ ∂ν c2 ρ ρ 4π 4π 1 µνρ 1 1 µνρ 1 + ǫ cµ ∂ν c2 + ǫ cµ ∂ν Aρ ρ 2π 2π Hence.22)-(21. . integrate out the gapped ﬂuctuµ ations of |φqp |.34). . Let’s now construct the K-matrices associated with the hierarchy and the Jain sequence. we ﬁnd the K-matrix of an arbitrary hierarchy state (20. a state at the ﬁrst level of the hierarchy has Leﬀ = K= 2k + 1 1 1 p1 (21.49) (21. 0).51) and tI = (1.34) by introducing a gauge ﬁeld.. t) pairs under SL(κ.48) We use the 2 + 1-dimensional duality transformation to substitute this into (21. .

. .53) Using 2 + 1-dimensional duality. Hence. .. tJain is transformed into th . Jain sequence is sum. . 1).. Meanwhile. .54) ρ 4π 4π 2p Integrating out aI and cµ . and apply steps (21. If we make the change of basis K h = W T K Jain W and th = W −1 tJain . 1. αI : µ µ ǫµνλ ∂ν αI = JI ≡ |φI |2 ∂ µ ϕI − aI µ − cµ − Aµ λ (21. the ﬂux exchange construction of the marized by the K-matrix: 2p + 1 2p 2p 2p 2p 2p + 1 2p 2p 2p Jain 2p 2p + 1 2p K = 2p 2p 2p 2p + 1 . . Starting with (21. and dropping the subleading Maxwell terms we µ ﬁnd Leﬀ = 2p µνρ 1 µνρ I ǫ αµ ∂ν αI + ǫ ρ 4π 2π + 1 µνρ ǫ 2π αI ∂ν Aρ µ I αI µ I ∂ν J αJ µ (21. Let’s now consider the ﬂux-exchange construction of the Jain sequence. = 2.29) to (21.4.21.39) by introducing gauge ﬁelds. then (21.J .56) . The K-matrix formalism also applies to some quantum Hall states which we have not yet discussed.52) with 2k+1 = 2p+1 and p1 = p2 = .55) and tI = (1. K-MATRICES 373 I and tI = δ1 . .. we write φI = |φI |eϕI . These include double-layer quantum Hall states – in which there are two parallel layers of electrons – and spin-unpolarized In other words.56) is transformed into (21. with W = δIJ −δI+1. (21. integrate out the gapped ﬂuctuations of |φI |. we re-write this as Leﬀ = 1 µνρ I 1 µνρ 1 I 2 αI ∂ν (cρ + Aρ ) + ǫ ǫ αµ ∂ν aI fµν + µ ρ 2ρ 2π 2π I 1 µνρ I 1 1 µνρ + ǫ aµ ∂ν aI − ǫ cµ ∂ν cρ (21.39). the two constructions are identical for the corresponding ﬁlling fractions. . .22)-(21. .

45) and statistics (21.43) vanishes. On a surface of genus g. by dimensional analysis.CHAPTER 21. EFFECTIVE FIELD THEORIES OF THE QUANTUM 374 HALL EFFECT systems. The leading irrelevant 2 corrections to (21.57) is sloppy. m.46) are the essential physics of the ground state which is encapsulated in this theory. even the state with no quasiparticles is degenerate. Two chapters ago. An example of a wavefunction which can describe spin-polarized electrons in a double-layer system or spin-unpolarized electrons in a single-layer system is the (m. since they are the lowest energy states at a given ﬁlling fraction. This includes states with – essentially non-dynamical – quasiholes and quasiparticles at ﬁxed positions.and down-spin electrons. so that it may be necessary to include both spins when describing electrons in moderately strong magnetic ﬁelds. This theory is only valid at energies much smaller than the gap since it ignores all of the physics above the gap.and lower-layer electrons and the ﬁlling fraction is 2 ν = m+n . the Chern-Simons theory is a theory of the ground state(s).57) The wi ’s and zj ’s are. This is not all.m. What exactly do we mean when we say that a Chern-Simons theory such as (21. (21.n) (wi . must have a coeﬃcient suppressed by the inverse of the gap. The notation of (21.j (wi − zj )n . In other words. band mass and g factor corrections make the ratio of Zeeman to cyclotron energies ∼ 7/400. respectively. Although. so every state in the theory has vanishing energy. however. rather than in some bounded region of the plane as it would be in a real experiment. The quasiparticle charges (21. the coordinates of the up and down spin electrons. n) states or states of unpolarized electrons in multi-layer systems.57) should be multiplied by the spin part of the wavefunction and antisymmetrized with respect to exchanges of up.58) and t1 = t2 = 1. we saw that a Chern-Simons . The Hamiltonian of (21. (21. zj ) = i<j (wi − wj )m i<j (zi − zj )m i. or upper. we can imagine a cornucopia of fractional quantum Hall states speciﬁed by K matrices.43) is the low-energy eﬀective ﬁeld theory of a quantum Hall state? Let us ﬁrst imagine that our quantum Hall liquid is on a compact surface such as a sphere or a torus. n) wavefunction: Ψ(m. even when the ﬁlling fraction is less than unity [?].43) are Maxwell terms of the form ∂ ν αλ − ∂ λ αν which. The K-matrix for this state is K= m n n m (21. we have thus far assumed that the electrons are spinpolarized by the magnetic ﬁeld. By considering hierarchies built on the (m. m.

the degeneracy is an important means of distinguishing distinct quantum Hall states with diﬀerent K-matrices at the same ﬁlling fraction. we ﬁnd that all of the action is at the edge: S= 1 2π dt dx KIJ ∂t φI ∂x φJ − VIJ ∂x φI ∂x φJ + Aµ ǫµν ∂ν φI tI (21.60) The Chern-Simons theory of the bulk has been reduced to a theory of (chiral) bosons at the edge of the system. the equation of motion following from the variation of KIJ αI + VIJ αI in (21.5. the degeneracy is simply detK. the most general such boundary condition is KIJ αI + VIJ αI = 0. in other words.5 Field Theories of Edge Excitations in the Quantum Hall Eﬀect If. On a genus-g surface. This can be generalized to an arbitrary quantum Hall state by diagonalizing its K-matrix and multipying the degeneracies of the resulting decoupled Chern-Simons terms or. The action is extremized if we take ǫµνρ ∂ν αJ = 0 subρ ject to boundary conditions such that KIJ αI ǫµνρ δαJ = 0 at the boundary. instead. Clearly. then the variation of the action S = L is: δS = 1 2π + d3 x KIJ δαI ǫµνρ ∂ν αJ µ ρ boundary 1 2π dt dx nν KIJ ǫµνρ αI δαJ + tJ ǫµνρ Aµ δαJ (21.21. In this gauge. This is precisely the ground state degeneracy which we obtained in the previous chapter by adapting the Laughlin wavefunctions to the torus. ρ µ Let us suppose that x and y are the coordinates along and perpendicular to the boundary. we look at the Chern-Simons theory (21. These are precisely the excitations which .43) is a constraint. FIELD THEORIES OF EDGE EXCITATIONS IN THE QUANTUM HALL EFFECT 375 with coeﬃcient 2k + 1 has a 2k + 1-fold degenerate ground state on the torus. Substituting this i into the Lagrangian and integrating by parts. x 0 which can be satisﬁed if aI = ∂i φI for some scalar ﬁeld φ. 21. Since numerical studies can be – and usually are – done on the sphere or torus. this becomes (detK)g [?].59) µ ρ ρ if we set Aµ = jµ = 0. VIJ is a symmetric matrix which will depend on the x 0 details of the boundary such as the steepness of the conﬁning potential.43) on a bounded region of the plane [?]. Then. KIJ αI + VIJ αI = 0 would be a sensible gauge choice since it is x 0 compatible with the boundary condition.

φ(x′ ) = i m By varying the electromagnetic ﬁeld. Let’s consider the simplest case. The equal-time commutations relations which follow from this action are: 2π δ(x − x′ ) (21. we derived by multiplying the Laughlin state by symmetric polynomials in the previous chapter.CHAPTER 21. ν = 1/m S= m 2π dt dx ∂t φ ∂x φ − v ∂x φ ∂x φ + 1 2π dt dx Aµ ǫµν ∂ν φI tI (21. t) = φ(x + vt).62) are satisﬁed if the ﬁeld is chiral.61) This is the action for a free chiral boson. we derive the charge and current operators: ρ= 1 ∂x φ 2π . EFFECTIVE FIELD THEORIES OF THE QUANTUM 376 HALL EFFECT Chern-Simons Theory of the Bulk Conformal Field Theory of the Edge Figure 21.1: The Chern-Simons theory which describes the braiding of quasiparticles in the bulk is associated with a Conformal Field Theory which describes the gapless excitations at the edge.63) ∂x φ(x). φ(x. The equations of motion (for Aµ = 0 for simplicity) (∂t − v∂x ) ∂x φ = 0 (21.

0) −m2 φ(0. Let us suppose that a tunnel junction is created between a quantum Hall ﬂuid and a Fermi liquid. 0) − φ(0. The tunneling of electrons from the edge of . we must ﬁrst compute the φ − φ correlation function.t) φ(0. 0) φ(0.68) Hence.0) φ(0. Hence.t) φ(0. eiφ(x ) = ρ(x). this gives: φ(x.5. To compute their correlation functions.0) 1 (x + ivτ )1/m (21. FIELD THEORIES OF EDGE EXCITATIONS IN THE QUANTUM HALL EFFECT j= 1 ∂t φ 2π 377 (21.64) Hence.65) These operators create quasiparticles and electrons respectively.0) − φ(0. the quasiparticle correlation function is given by: eiφ(x.21.0) 1 (x + ivτ )m (21. 0) = 1 dk dω 2π eiωτ −ikx − 1 2π 2π m k(iω − vk) dk 1 −ik(x+ivτ ) 2π e −1 = m 2π k 1 Λ dk 1 =− 1 m 2π k x+ivτ 1 ln [(x + ivτ )/a] (21. the quasiparticle creation operator has dimension 2/m while the electron creation operator has dimension m/2.τ ) eimφ(0.66) =− m where a = 1/Λ is a short-distance cutoﬀ.0) = e = φ(x.τ ) eiφ(0.0) φ(0. t) φ(0. the operators eiφ and eimφ create excitations of charge 1/m: ρ(x).67) while the electron correlation function is: eimφ(x. eimφ(x ) ′ ′ 1 iφ(x′ ) e δ(x − x′ ) m ′ = eiφ(x ) δ(x − x′ ) (21. In realspace. This is most simply obtained from the imaginary-time functional integral by inverting the quadratic part of the action.0) = em = 2 φ(x.

c. (21.τ ) ψ(0.c. EFFECTIVE FIELD THEORIES OF THE QUANTUM 378 HALL EFFECT the quantum Hall ﬂuid to the Fermi liquid can be described by adding the following term to the action: Stun = t dτ eimφ(0. This term is irrelevant for m > 1: 1 dt = (1 − m) t dℓ 2 (21. x = 0 is the point at which the junction is located and ψ(x. Then quasiparticles can tunnel across the interior of the Hall ﬂuid. (21.τ ) + c.73) Suppose we put a constriction in a Hall bar so that tunneling is possible from the top edge of the bar to the bottom edge.τ ) e−imφ2 (0.CHAPTER 21.71) A tunnel junction between two identical quantum Hall ﬂuids has tunneling action: Stun = t Hence. (21. τ ) is the electron annihilation operator in the Fermi liquid.70) Hence. As usual.75) (21.76) (21. it is a dimension 1/2 operator. dt = (1 − m) t dℓ and the tunneling conductance varies with temperature as: Gt ∼ t2 T 2m−2 while the current at zero-temperature varies as: It ∼ t2 V 2m−1 (21.74) dτ eimφ1 (0.69) Here.72) (21.c.77) . The ﬁnitetemperature tunneling conductance varies with temperature as: Gt ∼ t2 T m−1 while the current at zero-temperature varies as: It ∼ t2 V m (21.τ ) e−iφ2 (0. The tunneling Hamiltonian is: Stun = v dτ eiφ1 (0. τ ) + c. it can be handled perturbatively at low-temperature.τ ) + c.

82) In other words.78) where φ1 and φ2 are the edge operators of the two edges. In the quantum Hall eﬀect. DUALITY IN 1 + 1 DIMENSIONS The tunneling of quasiparticles is relevant dt = dℓ 1− 1 m v 379 (21. 21.83) with Y (x) − 1 ∼ −x for x → 0 and Y (x) ∼ x−m for x → ∞. but they can still be analyzed by the basic methods of free ﬁeld theory. the tunneling conductance varies with temperature as: 2 (21. we saw that a problem which could.21. in the strong coupling limit. Details can be found in . the bar is eﬀectively split in two so that all that remains is the tunneling of electrons from the left side to the right side: G ∼ t2 T 2m−2 2 1 e2 Y v 2 T m −2 m h (21.)v 2 T m −2 (21. it can be treated perturbatively only at high-temperatures or large voltages. If we measure the Hall conductance of the bar by running current from the left to the right.6 Duality in 1 + 1 Dimensions At the end of the previous section. This is an example of a situation in which there are two dual descriptions of the same problem.81) m h When T becomes low enough.80) so long as V is not too small.6.79) Gt ∼ v 2 T m −2 while the current at zero-temperature varies as: It ∼ v 2 V 2 −1 m (21. in the weak-coupling limit. Hence. At low voltage and high temperature. described by the tunneling of electrons. the edge structure is more complicated since there will be several bosonic ﬁelds. there is one description in which . For a general K-matrix. then it will be reduced by the tunneling current: 2 1 e2 G= − (const. be described by the tunneling of quasiparticles was. the conductance is given by a scaling function of the form: G= (21.

85) (21. Let us rescale ϕ → ϕ/ g so that the action is of the form: S0 = 1 8π dx dτ (∂τ ϕ)2 + v 2 (∂x ϕ)2 (21. It also has a current D jµ = ∂µ ϕ ˜ (21. ϕ now satisﬁes the identiﬁcation ϕ ≡ ϕ + 2π g. In the next section. We have already discussed this duality in the 2 + 1-dimensional Chern-Simons Landau-Ginzburg theory which describes the bulk. It can be expressed in terms of chrial ﬁelds φL and φR : ϕ = φL + φR ϕ = φL − φR ˜ Here. EFFECTIVE FIELD THEORIES OF THE QUANTUM 380 HALL EFFECT electrons are the fundamental objects and quasiparticles appear as vortices in the electron ﬂuid and another description in which quasiparticles are the fundamental objects and electrons appear as aggregates of three quasiparticles. ∂µ jµ = 0 jµ = ∂µ ϕ which is conserved by the equation of motion. Observe that: ˜ ∂µ ϕ = ǫµν ∂ν ϕ ˜ The free action takes the form: S0 = g 8π dx dτ (∂τ ϕ)2 + v 2 (∂x ϕ)2 (21.87) As a result of the rescaling of ϕ which we performed in going from (21. we will look at the analogous structure in the bulk ﬁeld theory. Note that this theory has a conserved current. Let us consider a free non-chiral boson ϕ.86) (21.84) √ where ϕ is an agular variable: ϕ ≡ ϕ + 2π. As we will see.CHAPTER 21. In this section.88) (21. we will examine more carefully the implementation of this duality in the edge ﬁeld theories.89) which is trivially conserved. we have deﬁned the dual ﬁeld ϕ. .86) √ to (21. it essentially the same as the duality which we used in our analysis of the Kosterlitz-Thouless transition.87).

physical operators of the theory must respect the periodicity condition of ϕ. The periodicity condition satisﬁed by ϕ0 imposes the following quantization condition on (pL + pR ): M pL + pR = √ . ϕ(x. from (21. we have: √ M pL = √ + N g 2 g M √ pR = √ − N g 2 g (21. From the canonical commutation relations for ϕ. and rescale √ the length so that x ∈ [0. Let us assume that our edges are closed loops of ﬁnite extent. τ ) = 1 R x + pR (iτ − x) + i 2 0 1 φL (x. π] with ϕ(τ.91) n n where ϕ0 = xL + xR and ϕ0 = xL − xR . Hence.90) n Hence. τ ) = ϕ0 + i (pL − pR ) τ + (pL + pR ) x − i ˜ ˜ 1 αn e−n(τ +ix) + αn e−n(τ −ix) ˜ n 1 αn e−n(τ +ix) − αn e−n(τ −ix) ˜ n (21. N ∈ Z (21. τ ) = ϕ0 + i (pL + pR ) τ + (pL − pR ) x + i ϕ(x.τ ) {M.93) These degrees of freedom are called ‘winding modes’.92) and they have dimension M 2 /g.94) Furthermore. From the identiﬁcation ϕ ≡ ˜ 0 0 0 √ 0 √ ϕ+ 2π g. it follows that ϕ0 and (pL + pR ) /2 are canonical conjugates.21. x + π) + 2πN g for some integer N . τ ) = xL + pL (iτ + x) + i 2 0 1 αn e−n(τ +ix) n 1 αn e−n(τ −ix) ˜ n 381 n (21. x) ≡ ϕ(τ. M ∈ Z g (21. 0} ≡ e g (21. DUALITY IN 1 + 1 DIMENSIONS Let us consider the Fourier decomposition of φR.91). we see that we must have √ (pL − pR ) = 2N g .95) . Then. The allowed exponential operators involving ϕ are of the form: M i √ ϕ(x. it follows that ϕ0 ≡ ϕ0 + 2π g.L : φR (x.6.

CHAPTER 21. EFFECTIVE FIELD THEORIES OF THE QUANTUM 382 HALL EFFECT Note that this is reversed when we consider ϕ. Momentum modes are ˜ replaced by winding modes and vice-versa. Following our earlier steps, but taking this reversal into account, the allowed exponentials of ϕ are of the ˜ form: √ ˜ (21.96) {0, N } ≡ ei2N g ϕ(x,τ ) Hence, the most general exponential operator is of the form: {M, N } ≡ e

“ √ ” √ ˜ i 2Mg ϕ+N g ϕ h“ “ √ ” √ ” i √ √ i 2Mg +N g φL + 2Mg −N g φR

=e

(21.97)

**with scaling dimension: dim(M, N ) = 1 M √ √ +N g 2 2 g M2 + N 2g = 4g (M/2)2 = + N 2g g
**

2

+

√ 1 M √ −N g 2 2 g

2

(21.98)

These dimensions are invariant under the transformation g ↔ 1/4g, M ↔ N . In fact the entire theory is invariant under this transformation. It is simply the transformation which exchanges ϕ and ϕ. ˜ When we couple two identical √ non-chiral bosons, ϕ1 and ϕ2 , we form √ ϕ± = (ϕ1 ± ϕ1 )/ 2. The factor of 2 is included so that both ϕ± have the same coeﬃcent, 1/8π, in front of their actions. However, this now means that ϕ± ≡ ϕ± + 2π g/2. When we couple two bosons though exponential tunneling operators, only ϕ− is aﬀected. Hence, the appropriate duality is that for ϕ− : (g/2) ↔ 1/[4(g/2)] or, simply g ↔ 1/g. This duality exchanges cos ϕ− / g/2 and cos ϕ− g/2, which transfer, repectively, a pair of solitons ˜ (i.e. electrons) and a particle-hole pair from system 1 to system 2. Let us now apply these considerations to quantum Hall edges. In order to apply the above duality – which applies to non-chiral bosons – to a quantum Hall edge, which is chiral, we must ‘fold’ the edge in order to deﬁne a nonchiral ﬁeld, as depicted in ﬁgure 21.2. √ If we fold the edge at x = 0, we can deﬁne ϕ = (φ(x) + φ(−x))/ 2 and √ ϕ = (φ(x) − φ(−x))/ 2. The latter vanishes at the origin; only the former ˜ is important for edge tunneling. The allowed operators are: eiN ϕ/

√ m

=

N2 2m

(21.99)

21.6. DUALITY IN 1 + 1 DIMENSIONS

383

Figure 21.2: An inﬁnite chiral edge mode with a tunnel junction at one point can be folded into a semi-inﬁnite nonchiral mode with a tunnel junction at its endpoint. √ The factor of 1/2 on the right-hand-side comes from the 2 in the deﬁnition of ϕ. If we couple two edges, we can now deﬁne ϕ− , which has allowed operators √ N2 eiN ϕ− / m/2 = (21.100) m and dual operators √ ˜ (21.101) e2iM ϕ− m/2 = M 2 m which are dual under M ↔ N , m ↔ 1/m. The description in terms of ϕ is equivalent to the description in terms of ϕ. However, as we saw in the ˜ previous section, the tunneling of quasiparticles between the two edges of a quantum Hall droplet is most easily discussed in terms of ϕ when the tunneling is weak, i.e. in the ultraviolet, when the tunneling operator can √ be written eiϕ/ m . However, when the tunneling becomes strong, in the infrared, the dual description in terms of ϕ is preferable, since the corresonding ˜ √ ˜ tunneling operator is eiϕ m .

CHAPTER 21. EFFECTIVE FIELD THEORIES OF THE QUANTUM 384 HALL EFFECT

CHAPTER

22

Frontiers in Electron Fractionalization

22.1

Introduction

At present, the only topological phases which are know to occur in nature are those which are observed in the fractional quantum Hall regime. As described in the previous chapter, a number of experimentally-observed phases are associated with various Abelian topological states. There is some reason for believing that there are also non-Abelian topological phases lurking in some relatively weak quantum Hall plateaus in the second Landau level. Furthermore, there is nothing about topological phases which is intrinsic to the quantum Hall regime. In principle, they can occur in a number of different physical contexts and, in fact, an even wider variety of phases (such as those which are P, T -invariant) might appear as we explore the full freedom of the phase diagram of electrons in solids. This is a frontier topic, so our discussion will necessarily be tentative. Since very little is known on the experimental side, our discussion will be rather speculative, but we will try to stick to topics where at least the mathematical and theoretical issues are clear. In particular, we will focus on the eﬀective ﬁeld theories which describe various topological phases. We will also try to brieﬂy address the question of which particular Hamiltonians of electrons in solids will actually give rise, but this is still very much an open question. 385

386 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION

22.2

A Simple Model of a Topological Phase in P, T -Invariant Systems

Consider the following simple model of spins on a honeycomb lattice. Each s = 1/2 spin lies on a link of the lattice. The spins interact through the Hamiltonian: (22.1) H = J1 Ai − J2 Fp

i p

where

z Ai ≡ Πk∈N (i) σk , x Fp ≡ Πk∈p σk

(22.2)

z and σk = ±. These operators all commute,

Fp , Fp′ = [Ai , Aj ] = [Fp , Aj ] = 0

(22.3)

so the model can be solved exactly by diagonalizing each term in the Hamiltonian: the ground state |0 satisﬁes Ai |0 = −|0 , Fp |0 = |0 . If we represent σz = 1 by colored bonds and σz = −1 by uncolored bonds, then Ai |0 = −|0 requires chains of bonds to never end, while Fp |0 = |0 requires the ground state to contain an equal superposition of any conﬁguration with one obtained from it by ﬂipping all of the spins on any plaquette (i.e. switching colored and uncolored bonds). To understand this Hamiltonian better, it is useful to consider the J1 → ∞ limit and to introduce the following representation for the low-energy states of this model. We consider wavefunctions Ψ[α] which assign a complex amplitude for any ‘multi-loop’ α on the honeycomb lattice. By ‘multi-loop’, we mean simply the disjoint union of a number of closed loops which do not share any links of the lattice. The multi-loop α simply represents the z z locations of the σk = +1 spins, so this representation is just the σk basis. The ground state condition, Fp |0 = |0 is the statement that Ψ[α] is invariant under various geometrical manipulations of the multi-loop α. If we draw the multi-loops as if they were in the continuum, but remember that they are really on the lattice, then we can draw the relations imposed by Fp |0 = |0 in the following pictorial form as three distinct relations. Depending on whether the plaquette is empty The ﬁrst follows from Fp acting on a empty plaquette. It ﬂips this plaquette into one which contains a small, contractible loop. Hence, the ground state wavefunction is invariant under such an operation, as depicted in ﬁgure 22.1. Suppose now that a loop passes through a single link of a

22.2. A SIMPLE MODEL OF A TOPOLOGICAL PHASE IN P, T -INVARIANT SYSTEMS

387

Ψ

= Ψ

Figure 22.1: The ground state wavefunction is invariant under the removal of a small contractible loop. plaquette. When Fp acts on this plaquette, it deforms the loop so that it passes through the other 5 links of the plaquette and now encloses it. The action of Fp causes similar deformation of a loop which passes through a plaquette along 2, 3, . . . , 5 consecutive links. Hence, the ground state must be invariant under such a deformation of any loop, as depicted in ﬁgure 22.2. If two loops touch a plaquette, then Fp cuts the loops and rejoins them so

Ψ

= Ψ

Figure 22.2: The ground state wavefunction is invariant under smooth deformation of any loop. In the ﬁgure, the loop in the upper left has been deformed. that they form one big loop. Conversely, if the same loop passes through a plaquette twice, then Fp breaks it into two loops. The ground state must be invariant under such surgery operations, as depicted in ﬁgure 22.3. If three

Ψ

= Ψ

Figure 22.3: The ground state wavefunction is invariant under a surgery procedure which cuts and rejoins two loops which come near each other. loops touch a plaquette, then Fp performs a surgery which is equivalent to two pairwise surgeries, which can be performed in any order. Now consider this model on an annulus. It is now possible for loops

388 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION to wind around the annulus. However, the surgery relation implies that ground state wavefunctions must have the same value on conﬁgurations with winding numbers 0, 2, 4, . . .. Similarly, they must have the same value on conﬁgurations with winding numbers 1, 3, 5, . . .. Thus, there are two ground states, corresponding to even and odd winding numbers of the loops around the annulus. On the torus, there are four ground states, corresponding to even/odd winding numbers around the two generators of the torus. Thus, the ground state degeneracy depends on the topology of the manifold on which the system is deﬁned – on a genus g surface, it is 4g . Note, further, that the diﬀerent ground states are locally indistinguishable. They all have Ai = −1, Fp = 1 at every vertex and plaquette, so correlation functions of local operators are all the same (and vanish unless the two operators share a plaquette). The only way to distinguish the various ground states is to do a measurement which is sensitive to the global topology of the system. Thus, it is in a topological phase. Consider now the excited states of the system. They are collections of localized excitations which come in two varieties: vertices at which Ai = 1 and plaquettes at which Fp = −1. In other words, we have excitations which are endpoints of broken loops, which we can think of as ‘electric’ particles. Clearly, they can only be created in pairs. We also have excitations which are frustrated plaquettes: the state acquires a minus sign whenever a loop moves through this plaquette. We can think of these excitations as vortices or ‘magnetic’ particles.

= −

Figure 22.4: When an Av = −1 electric excitation is taken around an Fp = −1 magnetic excitation, the wavefunction acquires a −1. Now, observe that when an electric particle is taken around a magnetic particle, we must move the curve attached to it though the excited plaquette, so the wavefunction acquires a phase π, as depicted in ﬁgure 22.4. Hence, electric and magnetic particles have non-trivial mutual braiding statistics. On the other hand, electric particles have bosonic statistics with themselves, as do magnetic particles. A composite formed by an electric and a magnetic particle is fermionic.

22.3. EFFECTIVE FIELD THEORIES

389

22.3

Eﬀective Field Theories

The basic physics of the Hamiltonian (22.1) is the relative statistics between the ‘electric’ and ‘magnetic’ particles. This is encapsulated in the following topological ﬁeld theory: SBF = 1 2π

electric magnetic d2 x dτ ǫµνλ eµ fνλ − jµ aµ − jµ eµ

(22.4)

where fνλ = ∂ν aλ − ∂λ aν , as usual. This theory is commonly known as ‘BF theory’ because the ﬁeld eµ is usually called bµ . However, eµ is better notation because it is the canonical conjugate of aµ , as may be seen from (22.4): ∂L = ǫij ej ⇒ [ai (x), ǫkj ej (0)] = i δik δ(x) (22.5) ∂ai The time derivatives of the ﬁelds e0 and a0 do not appear in this action. They are Lagrange multipliers which enforce the constraints: f12 = π ρmagnetic ∂1 e2 − ∂2 e1 = π ρelectric (22.6)

Thus, each magnetic particle is accompanied by π-ﬂux of the aµ gauge ﬁeld. Electric particles ‘see’ the gauge ﬁeld aµ according to their coupling electric a , so when an electric particle goes around a magnetic particle, the jµ µ wavefunction changes sign. Conversely, each electric particle is accompanied by π-ﬂux of the eµ ﬁeld, so when a magnetic particle goes around an electric particle, the wavefunction changes sign. We can quantize this theory along the lines of our quantization of ChernSimons theory, to which it is related according to 2SBF [e, a] = SCS [a + e] − SCS [a − e] (22.7)

By introducing Wilson loop operators for aµ , eµ and quantizing their algebra, we can see that the theory (22.4) has ground state degeneracy 4 on the torus. Rather than recapitulating this, we will see how this structure arises in another version of the theory below. The transition into this phase can be understood as the deconﬁnement transition of Z2 gauge theory, whose deconﬁned phase is described at low energies by the U(1) BF theory above. There are two diﬀerent ways of realizing such a theory. We could begin with a U (1) gauge theory with Maxwell action which is coupled to a charge-2 matter ﬁeld. When this matter ﬁeld condenses, the U (1) symmetry is broken to Z2 . This construction can be

390 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION done directly in the continuum. Alternatively, one can work with Z2 gauge ﬁelds from the beginning. However, one must, in such a case, work on a lattice. Let us follow the latter avenue. We consider a 2 + 1 dimensional space-time lattice on which there is an Ising gauge ﬁeld degree of freedom σz = ±1 on each link of the lattice. We will label them by a lattice site, x, and a direction i = x, y, τ so that there are three links associated with each site. The action is the sum over all plaquettes of the product of σz s around a plaquette: σz σz σz σz (22.8) S = −K plaq. To quantize this theory, it is useful to choose temporal gauge, σz (x, τ ) = 1 for all x. In this gauge, the Hamiltonian takes the form: H=− σx (x, i) − K σz σz σz σz x,i spatial plaq. (22.9)

In temporal gauge, there are residual global symmetries generated by the operators ˆ ˆ G(x) = σx (x, x)σx (x, y)σx (x − x, x)σx (x − y, y). (22.10)

The extreme low-energy limit, in which this theory becomes topological, is the K → ∞ limit. In this limit, σz σz σz σz = 1 for every spatial plaquette. It is useful to deﬁne operators W [γ] associated with closed curves γ on the lattice: σz (x, i) (22.11) L[γ] =

x,i∈γ

We also need operators Y [α] associated with closed curves on the dual lattice, i.e. closed curves which pass through the centers of a sequence of adjacent plaquettes. σx (x, i) (22.12) Y [α] =

x,i⊥α

The product is over all links which α intersects. L[γ] is analogous to a Wilson loop operator while Y [γ] creates a Dirac string. Let us consider the space of states which are annihilated by the Hamiltonian; this is the Hilbert space of the K → ∞ limit. When restricted to states within this Hilbert space, L[γ] and Y [α] satisfy the operator algebra L[γ] Y [α] = (−1)I(γ,α) Y [α] L[γ] [L[γ], L[α]] = [Y [γ], Y [α]] = 0 (22.13)

22.4. OTHER P, T -INVARIANT TOPOLOGICAL PHASES

391

Now, it is clear that such an operator algebra can be represented on a vector space of the form derived in the previous section: L[γ] Ψ[{α}] = (−1)I(γ,α) Ψ[{α}] Y [γ] Ψ[{α}] = Ψ[{α∪γ}]

(22.14)

The notable diﬀerence is that the allowed states must now satisfy the constraints Ψ[{α}] = Ψ[{α ∪ Ψ[{α}] = Ψ[{˜ }] α }]

(22.15)

⌣

Again, α is obtained from α by performing the surgery operation )(→ ⌢ on ˜ any part of α. If α is contractible, then Y [α] commutes with all other operators in the theory, so its eﬀect on any wavefunction should be multiplication by a scalar. If we take this scalar to be 1, then we have the ﬁrst constraint above. The second constraint is necessary in order to realize the operator algebra (22.14) and is also required by consistency with the ﬁrst. As a result of the second line of (22.15), we can resolve crossings of α and γ in either way since they are equivalent in the low-energy Hilbert space.

22.4

Other P, T -Invariant Topological Phases

It is clear that the structure which we have described above is rather general. Any system whose intermediate-scale degrees of freedom are ﬂuctuating loops can give rise to such a phase. This includes, for instance, domain walls between Ising spins. If the Ising spins lie on the vertices of a triangular lattice, the domain walls will lie on the honeycomb lattice, as depicted in ﬁgure 22.5. Suppose we wish to generalize this structure. We can modify the relations and/or change the degrees of freedom to e.g. directed or colored loops. Consider the ﬁrst approach. The relation Ψ[α] = Ψ[dα] where dα is any continuous deformation of the multi-curve α must presumably be satisﬁed by in topological phase. However, the other two relations can be modiﬁed. Suppose Ψ[α] = Ψ[α ∪ ] is modiﬁed to Ψ[α] = d Ψ[α ∪ ]. Then the surgery relation must be modiﬁed as well since the surgery relation of ﬁgure 22.3 is in conﬂict with d = 1, as may be seen in ﬁgure 22.6. Thus, for any d = 1, a new surgery relation which is consistent with it must be found. It turns out that for most d, such a relation cannot be found. Only for the

392 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION

Figure 22.5: A set of basis states of a system of Ising spins on a triangular lattice is equivalent to the possible loop conﬁgurations of a honeycomb lattice.

Ψ[

] = Ψ[

] = Ψ[

] = d Ψ[

]

Figure 22.6: If d = 1, the surgery relation of ﬁgure 22.3 cannot hold since the above contradiction would follow. special values d = cos π k+2 (22.16)

do such relations exist. The k = 1 case is the d = 1 phase √ which we discussed above. The next phase is the k = 2 phase, which has d = 2 and the surgery relation The surgery relations become more complex with increasing k.

Ψ[

] − 2 Ψ[ ] − 2 Ψ[ + Ψ[ ] + Ψ[ ] = 0

Figure 22.7: The surgery relation for d = √ 2.

]

22.5. NON-ABELIAN STATISTICS

393

22.5

Non-Abelian Statistics

The basic feature of these generalizations is that their quasiparticles exhibit non-Abelian braiding statistics. This is a possibility which we neglected in our discussion of exotic statistics in chapter ??. Suppose we have g degenerate states, ψa , a = 1, 2, . . . , g of particles at x1 , x2 , . . . , xn . Exchanging particles 1 and 2 might do more than just change the phase of the wavefunction. It might rotate it into a diﬀerent one in the space spanned by the ψa s, so that: 12 (22.17) ψa → Mab ψb On the other hand, exchanging particles 2 and 3 leads to:

23 ψa → Mab ψb

(22.18)

12 23 12 23 23 12 If Mab and Mab do not not commute, Mab Mbc = Mab Mbc , the particles obey non-Abelian braiding statistics. To see how such a phenomenon might occur, consider one of the topological phases mentioned at the end of the previous section. Because such a phase will not have a surgery relation on two strands, there are multiple linearly-independent states with four particles, as may be seen in ﬁgure 22.8

Figure 22.8: These two states are linearly independent, except for k = 1. In the k = 1 case, the surgery relation of ﬁgure 22.3 implies that these two states are the same. Thus, we have satisﬁed the ﬁrst condition necessary for non-Abelian statistics: a degenerate set of states of quasiparticles at ﬁxed positions. We further observe that braiding particles rotates states in this degenerate subspace into each other. For instance, taking particle 2 around 3 and 4 transforms the ﬁrst state in ﬁgure 22.8 into the second. The transformations enacted by braiding operations are such that the order is important, as may be seen in ﬁgure 22.9. The number of states with n-quasiparticles grows very rapidly, in fact π exponentially ∼ dn in the states with d = 2 cos k+2 alluded to in the previous section.

394 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION

Figure 22.9: By switching the order of two exchanges, we obtain a diﬀerent state. Starting from the state in the upper left, we can exchange quasiparticles 3 and 4 ﬁrst and then 2 and 3 (depicted on the right). Alternatively, we can exchange 2 and 3 ﬁrst and then 3 and 4 (depicted on the left). Non-Abelian braiding statistics can also occur in the quantum Hall regime. The likeliest candidate for such a state is the quantized Hall plateau observed at ν = 5 . From the perspective of our earlier discussion of the hi2 erarchy of Abelian states, this plateau is strange because ν = 2 + 1 has a 2 fractional part which does not have an odd denominator. Thus, it cannot arise in the hierarchy. Thus, we must consider states outside of the Abelian hierarchy, such as the ‘Pfaﬃan state’, which we discuss below. There is some numerical evidence that this particular state is a good description of the ground state at ν = 5 . 2 The Pfaﬃan (ground) state takes the form Ψ(zj ) =

j<k

(zj − zk )2

e−|zj |

j

2 /4

· Pf (

1 ). zj − zk

(22.19)

In this equation the last factor is the Pfaﬃan: one chooses a speciﬁc ordering z1 , z2 , ... of the electrons, chooses a pairing, takes the product of the indicated factor for all pairs in the chosen pairing, and ﬁnally takes the sum over all pairings, with the overall sign determined by the evenness or oddness of the order in which the zs appear. The result is a totally antisymmetric function. For example for four electrons the Pfaﬃan takes the form 1 1 1 1 1 1 + + . z1 − z2 z3 − z4 z1 − z3 z4 − z2 z1 − z4 z2 − z3 (22.20)

When the wavefunction is expanded in this way, in each term the electrons are grouped in pairs. Indeed, the Pfaﬃan state is reminiscent of the real-space form of the BCS pairing wavefunction; it is the quantum Hall incarnation of a p-wave superconducting state.

22.5. NON-ABELIAN STATISTICS

395

This wavefunction may be considered as a variational ansatz for electrons at ﬁlling fraction ν = 1 in the ﬁrst excited Landau level (with both spins 2 of the lowest Landau level ﬁlled) interacting through Coulomb interactions. This is the approach used in determining the relevance of this wavefunction to experiments, but for a theoretical study of quasiparticle statistics, it is more useful to consider this wavefunction (and quasihole excitations in it) as the exact zero-energy states of the three-body Hamiltonian below. Working with this Hamiltonian has the great advantage of making the entire discussion quite explicit and tractable. H=

i j=i k=i,j

δ′ (zi − zj )δ′ (zi − zk )

(22.21)

(22.22) has half-ﬂux quantum quasiholes at η1 and η2 . These excitations have charge e/4. One includes 2n quasiholes at points ηα by modifying the Pfaﬃan in the manner (zj − ηα )(zj − ηβ )...(zk − ηρ )(zk − ησ )... + (j ↔ k) 1 ) → Pf ( ). Pf ( zj − zk zj − zk (22.23) In this expression, the 2n quasiholes have been divided into two groups of n each (i. e. here α, β, ... and ρ, σ, ...), such that the quasiholes within each group always act on the same electron coordinates within an electron pair. There are apparently (2n)! (22.24) 2 n!n!

This Hamiltonian annihilates wavefunctions for which every triplet of electrons i, j, k satisﬁes the condition that if i and j have relative angular momentum 1 then i and k must have relative angular momentum ≥ 2. The Pfaﬃan state (22.19) satisﬁes this condition since i and j have relative angular momentum 1 only when they are paired, but if i is paired with j, then it cannot be paired with k. Since the distance between particles is proportional to their relative angular momentum, this roughly translates into the following: by pairing up and and getting near particle j, particle i is able to stay further away from all of the other particles, thereby minimizing its interaction energy. As in a superconductor, there are half-ﬂux quantum excitations. The state (zj − η1 )(zk − η2 ) + (zj − η2 )(zk − η1 ) 2 ). Ψ2 qh = e−|zj | /4 ·Pf ( (zj −zk )2 zj − zk

j<k j

396 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION ways of making such a division; the factor 1/2 arising from the possibility to swap the two groups of n as wholes. In fact, not all of these wavefunctions are linearly independent: the true dimension of this space of wave functions is actually 2n−1 . Consider ﬁrst the case of four quasiholes. The basic identity that has to be taken into account is, in its most primitive form, (z1 − η1 )(z1 − η2 )(z2 − η3 )(z2 − η4 ) − (z1 − η1 )(z1 − η3 )(z2 − η4 )(z2 − η2 ) + (z1 ↔ z2 ) = (z1 − z2 )2 (η1 − η4 )(η2 − η3 ) . (22.25) It will be convenient to abbreviate the left-hand side to (12)(34) − (13)(24). Then we have as an immediate consequence of (22.25) the relation (η1 − η4 )(η2 − η3 ) (12)(34) − (13)(24) = . (12)(34) − (14)(23) (η1 − η3 )(η2 − η4 ) (22.26)

It is interesting that on the right-hand side the basic projective invariant of four complex numbers, the cross-ratio, appears. For present purposes, however, the important point simply that it is independent of the zs. An immediate consequence is that for two electrons and four quasiholes the three apparently diﬀerent ways of constructing quasihole states are reduced to two through the relation (12)(34)(η1 −η2 )(η3 −η4 )+(13)(42)(η1 −η3 )(η4 −η2 )+(14)(23)(η1 −η4 )(η2 −η3 ) = 0 . (22.27) Now we want to argue that (22.26) and (22.27) still hold good for any even number of electrons, Ne . To see this we insert (22.25) into the Pfaﬃan of (22.23): Pf (13)(24) = A ( (13)(24) (13)(24) (13)(24) . . .) z1 − z2 z3 − z4 z5 − z6 (12)(34) − (z1 − z2 )2 η14 η23 (12)(34) − (z3 − z4 )2 η14 η23 = A( . . (22.28) .) z1 − z2 z3 − z4

where ηij ≡ ηi − ηj and A denotes the instruction to antisymmetrize on the zs. If we expand, A( (12)(34) − (z1 − z2 )2 η14 η23 (12)(34) − (z3 − z4 )2 η14 η23 . . .) z1 − z2 z3 − z4 (12)(34) (12)(34) . . .) = A( z1 − z2 z3 − z4 (12)(34) . . .) − A ((z1 − z2 )η14 η23 z3 − z4

22.5. NON-ABELIAN STATISTICS + A ((z1 − z2 )η14 η23 × (z3 − z4 )η14 η23 ×

397 (12)(34) . . .) + . (22.29) .. z5 − z6

there will be terms on the right hand side of (22.29) with zero, one, two, . . . , Ne factors of ( zi − zj ). Upon antisymmetrization, however, a term with k factors of (zi − zj ) would have to antisymmetrize 2k variables with a polynomial that is linear in each. Since this is impossible for k > 1, such terms vanish. Hence A( (12)(34) − (z1 − z2 )2 η14 η23 (12)(34) − (z3 − z4 )2 η14 η23 . . .) z1 − z2 z3 − z4 (12)(34) (12)(34) . . .) = A( z1 − z2 z3 − z4 (12)(34) . . .) . (22.30) − A ((z1 − z2 )η14 η23 z3 − z4 (12)(34) (12)(34) . . .) z1 − z2 z3 − z4 (12)(34) . . .) . + A ((z1 − z2 )η13 η24 z3 − z4 η14 η23 (Pf (12)(34) − Pf (13)(24) ) . η13 η24

Similarly, one has Pf (14)(23) =A (

(22.31)

From these we deduce the many-electron generalization of (22.25): Pf (12)(34) − Pf (14)(23) = (22.32)

This is a linear relation among the three pairing possibilities for two quasiholes. It depends on their coordinates but – remarkably – takes the same form for any number of electrons. Thus, we have shown that there are two four-quasihole states. In the same way, it can be shown that there are 2n−1 2n-quasihole states. Braiding operations cause these states to be rotated into linear combinations of each other, as further analysis shows [].

398 CHAPTER 22. FRONTIERS IN ELECTRON FRACTIONALIZATION

Part VII

Localized and Extended Excitations in Dirty Systems

399

.

To see the signiﬁcance of this.CHAPTER 23 Impurities in Solids 23. it will have low-energy excitations even without any tuning. and the phosphorus impurities get stuck (at least on time scales which are relevant for experiments) at random positions. this equilibrium is never reached when the alloy is made: it is cooled down before equilibrium is reached. By ‘dirt’ or ‘disorder’. we discuss the low-energy excitations which result from the presence of ‘dirt’ or ‘disorder’ in a solid. These random static spatial ﬂuctuations have interesting eﬀects on the electronic states of the system: they can engender low-lying excitations and they can dramatically change the nature of such excitations. Consider phosphorous impurities in silicon. suppose that. the true ground state of such a mixture is one in which the phosphorus atoms form a superlattice within the silicon lattice. we have discussed the low-energy excitations which result from broken symmetry. in the absence of disorder. criticality. a system will generically form a gap between the ground state and all excited states. or fractionalization. However. By tuning to a critical state – such as a Fermi liquid – we can arrange for a system to have low-energy excitations. To get a sense of why this should be so. in the presence of disorder.1 Impurity States In the previous parts of this book. we mean impurities which are frozen into the solid in some random way. in the absence of a broken continuous symmetry. recall that. Presumably. However. In this ﬁnal part of the book. there were a gap 401 .

we expect a distribution of bound state energies so that the gap will be ﬁlled in. If there is a random distribution of phosphorus impurities.402 CHAPTER 23. there A are states within the band gap. in the presence of impurities. and ǫ is the dielectric constant of silicon.1) where Hlattice is the Hamiltonian of a perfect lattice of silicon ions and Himpurity is the potential due to the extra proton at the phosphorus site. The phosphorus takes the place of one of the silicon atoms. as in the case of a Fermi liquid. aB = m∗ e2 ≈ 20˚). there will be some place in the system where the potential energy oﬀsets the energy required to create a qasiparticle.2) where uk0 (r) eik0 ·r is the eigenstate of Hlattice at the conduction band minimum. then disorder can extend the critical state into a stable phase and it can cause the low-lying exciations of the system to become spatially localized.3) where m∗ is the eﬀective mass in the conduction band. Then χ(r) satisﬁes the equation: 2 − 2m ∇2 − ∗ e2 ǫr χ(r) = Eb χ(r) (23. If the binding energy e2 is much less than the band gap. then our neglect of electronelectron interactions is justiﬁed because the interaction will be too weak to excite electrons across the gap. if the system were gapless in the absence of disorder. Let us neglect electron-electron interactions and write Schr¨dinger’s o equation for the extra electron in the form: (Hlattice + Himpurity ) ψ(r) = E ψ(r) (23. Phosphorus has an extra electron compared to silicon and an extra positive charge in its nucleus. there will generically be states at . Note that such an excitation. silicon is an insulator with a band gap Eg . Hence. Consider the simple example of a single phosphorus impurity in silicon. the potential varies from place to place. though low in energy. Eb is measured from the bottom of the conduction band. Let us write ψ(r) in the form ψ(r) = χ(r) uk0 (r) eik0 ·r (23. 2ǫaB ≪ Eg . By the same token. (Note that aB is the eﬀective Bohr radius ǫ 2 in silicon. Hence. so long as we can neglect electron-electron interactions. may not have large spatial extent. in the thermodynamic limit. IMPURITIES IN SOLIDS to the creation of a quasiparticle. In this region. In other words. and. the electron will be trapped in a bound state at the impurity. it won’t cost any energy to create a quasiparticle. In the presence of disorder. Without the phosphorus.

this naive expectation is not always correct.2 23.c. This would lead us to expect the system to be metallic. unlike in a pure system. Anderson simpliﬁed the problem by arranging the . One could imagine such a model arising as an eﬀective description of the many-impurity problem. i ij (23. as we saw in the previous section. Since the impurities are located at random positions. n. How about the low-density limit? In the case of a single impurity. ǫi is the energy of an electron at impurity i.2. as we will see in this section.23. One would expect the kinetic energy to increase with the density. the electron is trapped in a hydrogenic bound state. These are random variables which are determined by the positions of the various impurities. the case of a high density of electrons and impurities. but ﬁnite overlaps between the hydrogenic bound states so that a metal with very small bandwidth forms. as n2/d while the potential energy should increase as n1/d . we ignore electron-electron interactions. ﬁnite density of impurities? One might expect exponentially small. 23. One might also suppose that their diﬀerent locations will lead to diﬀerent eﬀective environments and therefore diﬀerent bound state energies ǫi . When the kinetic energy dominates. there tunneling matrix elements between their respective bound states will be random tij . What happens when there is a small. where it is possible for the chemical potential to lie within an energy gap. and tij is the hopping matrix element from impurity i to impurity j.2. we might expect some kind of one-electron band theory to be valid. so that the kinetic energy should dominate for large n.4) In this model. LOCALIZATION 403 the chemical potential (although there will be a small density of states when the impurity density is small). ﬁrst. Consider the Anderson model: H= i ǫi c† ci − i tij c† cj + h.1 Localization Anderson Model What is the nature of these electronic states when there are several impurities? Presumably. This is not the case in the low-density limit. so we take the electrons to be spinless. As we will see. Spin is an inessential complication. there is some mixing between the states at diﬀerent impurities. One can imagine that this mixing leads to the formation of states which are a superposition of many impurity states so that they extend across the system. Consider.

by which we mean that the DC conductivity vanishes at T = 0. Where is the transition between the t/W = ∞ and the t/W = 0 limits? Is it at ﬁnite t/W = ∞? The answer is yes. we expect corrections to be small – and. which are taken to be independent random variables which are equally likely to take any value ǫi ∈ [−W/2. it can be shown (Anderson. Hence. For W = 0.5) j Perturbation theory will be valid so long as the second term is small. the Bloch waves are weakly scattered by the random potential. and the system is insulating. the electronic states are localized. The system is metallic. The typical value of ǫi − ǫj is W/2. The perturbed eigenstates will. This is a drastic simpliﬁcation. then all states will be localized. where z is the coordination number of the lattice. this is not a foolproof argument because there is always some probability for ǫi − ǫj small. 2tz/W ≫ 1. it is not surprising that there are localized states in a disordered system. In other words. If the disorder strength is weaker than this limit. but it already contains rich physics. one can only make probabilistic statements. as we will see. Now consider the opposite limit. as a result. The eﬀect of the randomness is encapsulated in the ǫi ’s. Frohlich and Spencer) that perturbation theory converges with a probability which approaches 1 in the thermodynamic limit. so the electronic states are simply Bloch waves. the states will be Bloch waves. weakly-scattered by impurities as . for t/W small. i. there is no randomness. the entire band will consist of localized states. with eigenenergies ǫi . (We say that a single-particle state is localized if it falls oﬀ as e−r/ξ . perturbation theory to be valid – if 2tz/W < 1. Since the ǫi ’s are random. For W ≪ t. In perturbation theory. What is surprising is that if the disorder strength is suﬃciently strong. W/2]. IMPURITIES IN SOLIDS sites i on a lattice and setting tij = t for nearest neighbor sites and zero otherwise. The system is insulating. to lowest order. On the other hand. all of the eigenstates are localized at individual sites.e. be of the form: |i + t |j ǫi − ǫj (23. From our discussion of the hydrogenic bound state of an electron at an impurity in a semiconductor. Nevertheless. hence. there is a regime at low density (small t) where the electronic states are localized and the system is insulating.404 CHAPTER 23. then we can do perturbation theory in the random potential. consider perturbation theory in t/W . The typical smallest value for any given i is W/2z. the eigenstates are |i . and. For t = 0. Hence. ξ is called the localization length) To see why this is true.

W/2] and the hopping parameter is t then there are no states below the band edge. the states near the band edge are localized.2 Lifschitz Tails As noted above. The asymptotic density of such states can be obtained by the following argument due to Lifschitz. This form for the conductance can be motivated as follows. the potential couples to the phase of the wavefunction.7) . The probability of having such a conﬁguration of neighboring sites is ∼ (I/W )N . −W/2 + I].2. g.6) ∆E δE is the disorder-avergaged energy change of eigenstates at EF when the boundary conditions are changed from periodic to antiperiodic. When it lies in the region of extended states. Near the band edges. (In order to derive such a relation analytically. it is metallic. and the conductance will vanish in the thermodynamic limit. there are extended states near the band center and localized states near the band edges. When the chemical potential lies in the region of localized states. If the on-site disorder lies in the range [−W/2. The kinetic energy of such a state is δE ∼ t N −2/d .2. To see this. Such states are called extended states.) If the states at EF are localized with localization length ξ. Emin = −2dt − W/2 on the d-dimensional hypercubic lattice. The perturbative analysis is correct for states in the center of the band. one must consider disorder-averaged quantities.23. for weak disorder. then δE ∼ e−L/ξ . Hence. g(L) = 23. Consider a state near Emin which is spread over N sites. is the energy relative to the band edge. we write the DC conductance. so a state which is at energy Emin + δE is spread over N sites with N ∼ (δE/t)−d/2 . Thus. so the conductance can be viewed as the sensitivity of the wavefunction to changes of the boundary conditions. N (Emin + δE) ∼ e−c(t/δE) d/2 (23. against which the disorder strength should be measured. A systematic derivation proceeds from the Kubo formula. the system is insulating. however. the perturbative analysis breaks down and states are localized: the correct measure of the electron kinetic energy. of s system of non-interacting fermions of size L in the following form: δE (23. The conductance is the response of the current to a change in the electromagnetic potential (which can be applied entirely at the boundary). LOCALIZATION 405 we discuss in the next section. However. ∆E is the mean level spacing of eigenstates. Then each of these sites has on-site energy in the range [−W/2.

electron-electron interactions tend to stabilize the metallic state by screening the random potential due to impurities.U iα <i. electron-electron interactions should. As we will discuss in a later section. so that they are ac2 tually of the form e e−λr . Hence. IMPURITIES IN SOLIDS where c is a constant. unscreened electron-electron Coulomb interactions do strongly inﬂuence the properties of the insulating state. of course. The implications is that it can be more or less continuously deformed into a non-interacting fermion system which is insulating as a result of Anderson localization. At low density.j> i ni↑ ni↑ (23. λ → 0. when the density is suﬃciently low. A ‘Mott insulator’ is an insulator which would be metallic in the absence of electronelectron interactions. At low densities where Coulomb interactions dominate over the kinetic energy. dE dN dE 23. The term ‘Anderson insulator’ is usually used for an insulator which would be metallic in the absence of impurities. enhance metallic behavior at high density.c.406 CHAPTER 23. They can lead to ordered states such as superconductivity or ferromagnetism. electrons will form a Wigner crystal.8) . naively. Thus. require a re-examination of the above analysis. we expect the ǫr impurities to be extremely well-screened so that there won’t be any impurity bound states. Of course. The Wigner crystal is an example of a Mott insulator. What about the eﬀect of electron-electron interactions? At high density. there isn’t a van Hove singularity at the band edge at which diverges. In fact. Another canonical example is given by the Hubbard model: H=− tc† cjα + h.3 Anderson Insulators vs. this screening eﬀect is too weak to prevent localization. To summarize. it would be more accurate to say that electron-electron interactions enhance metallic behavior in the high-density limit but cause insulating behavior in the low-density limit. so the impurities are not eﬀectively screened. we expect the electrons to screen the impurity potentials.2. λ → ∞. which would. and we expect the preceeding analysis to hold. However. Hence. thereby making it insulating. but rather an essential singularity at which dN vanishes. A tiny amount of disorder will pin this crystal. In the high-density limit. electron-electron interactions can do other things besides screen the impurities. suﬃciently strong interactions can also cause insulating behavior even in an extremely clean system. Mott Insulators The preceding analysis shows that a disorder-induced insulating state of non-interacting fermions is stable over a ﬁnite range of hopping strengths.

their insulating behavior cannot be entirely ascribed to one or the other. consider a non-interacting Fermi system whose single-particle states at the Fermi energy are localized with localization length ξ. we would still expect a sharp crossover between highly-conducting and poorly-conducting regimes at. Then the conductivity of the system will be σ ∼ L2−d e−L/ξ in a system of size L. Therefore. momentarily. where λ is ǫr large. as far as its electronic properties are concerned. it is as far from a perfect crystalline lattice as possible.3 Physics of the Insulating State At T > 0. While they might be metallic in the absence of both disorder and electron-electron interactions. Doped semiconductors and many other real materials do not fall so neatly into one class or the other. however. respectively. Thus. the system is insulating. it would be metallic at half-ﬁlling.) Anderson localization of the impurity bound states will make the crossover from a metal to a poor metal into a sharp metal-insulator transition.3. However. we discuss the properties of a disorder-driven insulating state. but physical properties are dominated by localized excitations so that it is diﬃcult to construct even an averaged description which is translationally . Because the energy cost for having two electrons on any site is prohibitive. as the dopant density is decreased. For instance. 2 V (r) ∼ e e−λr has no bound states in the high-density limit. albeit with a very narrow band. it is more accurate to say that insulating behavior is due to the interplay between localization and interactions. This would lead to small conductivity. for suﬃciently large U there is precisely one electron. When a solid is driven into an insulating phase as a result of impurities. (But not zero yet because the exponentially-small overlap between these bound states would still be expected to lead to metallic behavior. 23. If we were to neglect localization. Thus. high and low densities. the formation of hydrogenic bound states in a doped semiconductor.23. we deﬁne an insulator as any state in which the DC conductivity vanishes at T = 0 in the thermodynamic limit. were it not for localization. Let us reconsider. The screened Coulomb interaction. bound states will begin to appear when λ falls below a certain value ∼ aB . In this section. The conductivity of a ﬁnite-sized system will also always be non-vanishing. Not only is translational symmetry absent. any system will have ﬁnite conductivity since there will always be some probability that thermally-excited carriers can transport charge across the system. PHYSICS OF THE INSULATING STATE 407 For U = 0.

which we take to be exponentially decaying ∼ e−2αR (where R = 3R/4 is the average hopping distance) and on the probability that the electron receives the activation energy needed to hop. a phonon so that it hops to another nearby localized state. However. The number of available states per unit 4 energy within such a distance is 3 πR3 NF . which follows the Boltzmann distribution ∼ e−βW (R) Hence.1 Variable Range Hopping One of the most celebrated results in the study of disordered insulators is Mott’s formula for the DC conductivity due to variable range hopping. 23.9) The conductivity associated with such a hopping process depends on the matrix element (squared) between the two states. One might imagine that electrons will only hop to the nearest neighboring states since the matrix element to hop to a more distant state will be exponentially (in distance) smaller. The competition between these competing eﬀects leads to variable range hopping. However. the conductivity due to such a hopping process varies as σR ∼ e−2αR−βW (R) (23.3. Hence.408 CHAPTER 23. and speciﬁc heat – can be deduced by simple arguments. a state which is further away might be closer in energy. However. say. Through a sequence of such hops. The electron could hop a short distance with relatively large matrix . and therefore exponentially (in temperature) easier to reach. Suppose that an electron which is in a localized electronic state ψ hops to another one a distance R away.10) The prefactor of the exponential contains various comparatively weak dependences on R. which we present in this section. at ﬁnite-temperature. IMPURITIES IN SOLIDS invariant. for instance. where NF is the density of states at the Fermi energy. We imagine that the electrons are all in single-particle localized states. There are many such hopping processes available to our electron in state ψ. an electron can be excited by. some of these properties – the DC and AC conductivities. magnetic response. there should be a factor of R corresponding to the fact that a longer hop constitutes a larger contribution to the current. the typical smallest energy diﬀerence between the state ψ and a state within a distance R is the inverse of this: W (R) ≡ typical smallest energy diﬀerence between states separated by a distance R = 3 4πNF R3 (23. Let us consider an insulator at low temperatures. it can conduct electricity.

The resulting temperature dependence of the conductivity is: σ(T ) = σ0 (T ) e−(T0 /T ) 1/4 (23. the current matrix element is ∼ ωR e−αR . so that the energy eigenstates are actually linear combinations of the hydrogenic bound states. and we turn to the second possibility. the electron makes longer and longer hops. Since the matrix element will be exponentially suppressed. At frequency ω. power-law. the current matrix element is ∼ ωR.11) (23.3. Then the dipole matrix element is ∼ R e−αR .3. (2) The distance R is not very large. (1) R is large. but the activation energy would probably be high in such a case.13) where σ0 (T ) is relatively weakly-dependent on T . e. PHYSICS OF THE INSULATING STATE 409 element. the contribution of such resonant pairs will be small. but the matrix element would be very small. as the temperature is decreased.23. In d dimensions. There is no exponential . Then the energy diﬀerence between the eigenstates is essentially the matrix element of the Hamiltonian between the hydrogenic bound states ∼ I0 e−αR . The conductivity of an insulator does not fall within this class. The AC conductivity is entirely diﬀerent in form. an electron will oscillate between states separated in energy by ω.10): −2α − β or Ropt = 3 2παNF T dW =0 dR 1/4 (23. so that they are essentially individual.14) 23. the preceeding analysis immediately generalizes to σ(T ) = σ0 (T ) e−(T0 /T ) 1 d+1 (23. independent bound states separated in energy by the diﬀerence between their respective potential wells. There are two possibilities. It could instead hop a long distance and ﬁnd a state with low activation energy. Let us think about these resonant states in terms of hydrogenic bound states.12) Hence. The optimal route is for the electron to hop a distance which minimizes the exponential in (23.2 AC Conductivity There are many situations in which the ω ≫ kB T frequency dependence of a quantity can be obtained from its DC temperature dependence by simply replacing kB T with ω. The dipole matrix element is ∼ R.g.

However. If the distance between the states is small.410 CHAPTER 23. their eﬀects are particularly strong. this Coulomb energy will be large and will overcompensate the single-particle energy.15) The available phase space is the density of states per unit volume multiplied by the volume in which the states can lie.18) (23. so long as ω ≫ T . R ∼ ln(I0 /ω) Thus. The ‘single-particle energy cost’ associated with promoting an electron from the lower one to the higher one would be at least partially oﬀset by the negative Coulomb energy associated with the resulting particle-hole pair. we have the following formula for the conductivity: σ(ω) ∼ ω 2 ln4 (I0 /ω) (23. More quantitatively. the states must be far apart and.16) This formula also holds at ﬁnite temperatures.3. when the frequency is decreased. by the arguments of the preceding paragraph. Hence. one just below the Fermi energy and one just above. One eﬀect. suppose the single-particle energy separation between an un-occupied state above the Fermi energy and an occupied one below the Fermi energy is ∆Eunocc−occ while the distance between the states . 23. Coulomb interactions are not screened at long-wavelengths and low-frequencies because there aren’t mobile charges. the conductivity will cross over to the variable-range hopping form.3 Eﬀect of Coulomb Interactions In an insulator. therefore. the single-particle density of states (per unit volume) must be small. The basic physics is as follows. pointed out by Efros and Shklovskii is a suppression of the single-particle density-of-states at the Fermi energy. IMPURITIES IN SOLIDS suppression because the two eigenstates are diﬀerent linear combinations of the same two bound states. ∆R is the size ξ of a localized state. Thus.17) (23. Then the conductivity due to this resonant pair is: σ(ω) ∼ (dipole matrix element)2 · (available phase space) (23. Hence. given that they are a distance R 2 apart: ∼ NF R2 ∆R. Suppose that there were two (strongly-localized) single-particle states. σ(ω) ∼ (ωR)2 R2 ∼ ω 2 R4 However.

the Fermi energy moves with it. The right-hand-side is the Coulomb energy between the electron and the hole which results when such a transfer occurs. Hence. contradicting our assumption. when we change the applied voltage on our disordered insulator.20) from which we see that the density of states is N (∆E) ∼ (∆E)2 (23.19) or else the energy would be lowered by transferring an electron from the occupied state to the unoccupied one. dI/dV increases with voltage as V 2 . Then it must be the case that ∆Eunocc−occ > e2 r (23. as shown in ﬁgure 23. The ‘single-particle energies’ include the Coulomb interaction energy between the two states under consideration and all of the other occupied states in the system. The diﬀerential conductance dI/dV will be proportional to the density of states. and cool it down again. PHYSICS OF THE INSULATING STATE 411 is r. to which the density√ states returns when the distance from of the Fermi surface exceeds ∆ = e3 N0 . the number of states per unit volume which are within energy ∆E of the Fermi surface is n(∆E) ∼ 1 < (∆E)3 r3 (23. As we change the voltage. r∆ by e2 =∆ r∆ 3 N0 r∆ ∆ = 1 (23. dI/dV will be suppressed at zero voltage and will increase with voltage as V 2 . Shouldn’t the diﬀerential conductance remain zero? The answer is no. One could measure the single-particle density of states by performing a tunneling experiment in which current tunnels into a disordered insulator from a metallic lead to which it is contacted. change the applied voltage to V0 . the single-particle density of states vanishes at the Fermi energy. and is suppressed near it.21) Thus.22) where N0 is the density of states in the absence of long-range Coulomb interactions.23. Suppose we deﬁne energy.1. as if the chemical potential had not moved. According to the above arguments. we ﬁnd that dI/dV varies . However.3. if we raise the temperature of the insulator. length scales ∆. This eﬀect is called the Coulomb gap. However.

if an experiment were done slowly enough to allow such re-equilibration.23) 23. This will occur in such a way as to allow a state near the Fermi surface.3. as expected if the energy is determined by Coulomb interactions. The suppression of the density-of states near the Fermi energy has consequences for variable-range hopping. The non-equilibrium nature of the Coulomb gap is related to the assumption above that the only Coulomb interaction energy which must be accounted for is that between the excited electron and the hole left behind.3. when an electron is excited from an occupied state to an unoccupied one.4 Magnetic Properties In the absence of electron-electron interactions. the suppression calculated above would not occur. Thus. all of the other electrons can and will rearrange themselves. −e2 /r. However.1. Essentially.412 CHAPTER 23. Recall that we estimated that within a region of size R. However. the typical smallest energy separation ∆E is ∆ E ∼ 1/(N0 R3 ). there will be two electrons in each localized single-particle state. we ﬁnd that the variable-range-hopping formula in the presence of Coulomb interactions (in arbitrary dimension) is: σ ∼ e−(T0 /T ) 1/2 (23. Following the minimization procedure of subsection 23. as (V − V0 )2 . so that the Fermi energy has moved to V0 . or ∆E ∼ 1/R. For |EEF | > ∆. N (∆E) ∼ (∆E)2 . IMPURITIES IN SOLIDS N(E) ∆ EF N0 E Figure 23.1: The single-particle density-of-states is suppressed at the Fermi energy. and the ground state will be para- . the single-particle density-of-states returns to N0 . the suppression of the density of states remains wherever it was when the system last equilibrated.

24) If we have many such spins but ignore the interactions between them. χ(T ) will exhibit diﬀerent behavior. TSG . the susceptibility will have this form at high-temperatures. in which they all point in the same direction. we will have a susceptibility of the same form. Thus. Cu:Mn and Au:Fe alloys is spin glass behavior. but once T becomes comparable to or less than the typical spin-spin interaction strength J.). we must consider the spin-spin interactions between the electrons in diﬀerent localized states. At the freezing temperature. . in which the susceptibility has a cusp at a temperature TSG . however. Spin glasses are rather incompletely understood. but unlike in a ferromagnet. An isolated impurity state. for instance. However. as shown in ﬁgure 23. These issues are discussed in (refs. the basic caricature is that the spins order. However. and many of their properties are still the subject of controversy. the spins point in a random (but ﬁxed) direction which varies from point to point. An alternate possibility which can occur in a system which has predominantly antiferromagnetic interactions (but of randomly-varying strengths) is that the spins do not order but rather form a random singlet phase.2. There are several possibilities. The non-linear susceptibility actually diverges at TSG . the system essentially breaks up into a set of pairs of spins which interact with strong antiferromagnetic exchange couplings Ji . the system undergoes a transition into such a state in which S(x) = s0 (x) where s0 (x) is a randomly-varying function of x. which occurs in. We expect them to prevent two electrons from occupying the same state. In such a phase. proportional to the density of spins. Spin-glass order is believed to require spin-spin interactions of both ferromagnetic and antiferromagnetic signs which are suﬃciently disordered as to frustrate either of these ordering tendencies of clean systems. the spins order in a novel type of ordered state which can occur only in a dirty system. we include the interactions between spins. One. it is clearly unrealistic to ignore electron-electron interactions. occupied by a single electron will have a Curie form for the magnetic susceptibility: χc (T ) ∼ 1 T (23.23.3. As a result. PHYSICS OF THE INSULATING STATE 413 magnetic. There is some disagreement about whether or not such a state has inﬁnitely-many distinct degenerate ground states (over and above the degeneracy due to broken rotational symmetry) and the related issue of whether or not there is a sharp phase transition in non-zero magnetic ﬁeld. If.

414 CHAPTER 23.25) The susceptibility of the system must be summed over all pairs with some probability distribution.28) . P (J). Deﬁne p(R) as the probability that the nearest neighbor is a distance R or less.27) where ρ(T ) is the number of unpaired spins at temperature T . Then P (R) = dp dR (23. and suppose that the coupling between spins separated by a distance r is J(r) = J0 e−2r/a for some constant a which is roughly the size of the localized states. IMPURITIES IN SOLIDS χ TSG T Figure 23. we know P (J) if we know P (R) such that P (R)dR is the probability that a spin has its nearest neighbor lying in a thin shell between R and R + dR. With this functional form for J. We can compute P (R) in the following way. P (J).2: The susceptibility of a spin-glass has a kink at TSG . T . there will be some spins whose J is smaller than T . which will be the dominant contribution: χunpaired (T ) ∼ ρ(T ) T (23. Such spins will not have formed singlets yet. so they will give a Curie-like contribution to the susceptibility. The function ρ(T ) depends on the distribution of exchange couplings. Each of these pairs has susceptibility χith pair (T ) = 2 1 T 3 + eβJi (23.26) At a given temperature. for the values of J: χ(T ) = 1 T dJ P (J) 2 3 + eβJ (23. Let us suppose that there is a small density n of randomly-situated localized spins.

χ(T ) = 1 T dR 4πnR2 e− 3 πnR 4 3 2 3 + eβJ(R) (23.32) In the low-temperature limit. simply. the integral can be evaluated by saddle-point approximation: χ(T ) ∼ 1 − 1 πna3 ln3 (J0 /T ) e 6 T (23.34) The physical interpretation of this result is simple. 1 − p(R).26).23. dropping ln ln T terms 2 which are subleading at low T .31) We substitute this expression into (23. The approximate form which we used for P (R) underestimates the number of unpaired spins.29) (23.35) 2 The density of such spins should therefore be: ρ(T ) ∼ e− 6 πna 1 3 ln3 (J0 /T ) (23. Hence. 4πnR2 dR.36) as we found above. 1 − p(R) P (R) = 4πnR2 (1 − p(R)) or.3. . multiplied by the probability that there is no closer neighbor. Unpaired spins at temperature T are those spins whose largest coupling to the other spins is less than T : a J(R) ∼ T ⇒ R ∼ ln (J0 /T ) (23. namely 4πnR2 . multiplied by the probability that there were no closer neighbors.33) The integrand has a saddle-point at R ≈ a ln(J0 /T ). we have p(R) = 1 − e−4πnR 3 (23. this is simply χ(T ) ≈ 1 T dR 8πnR2 e− 3 πnR e−βJ0 4 3 exp(−2R/a) (23. dp = 4πnR2 (1 − p) dR Solving this diﬀerential equation. We took P (R) to be the probability of having a neighboring spin at R. thereby underestimating the susceptibility. PHYSICS OF THE INSULATING STATE 415 The probability that the nearest neighbor is between R and R + dR is given by the number of impurities in a thin shell at R.30) ⇒ P (R) = 4πnR2 e−4πnR 3 (23.

that the susceptibility diverges at T → 0.4 23.) Hence.416 CHAPTER 23.37) (23. In order to do so. which is roughly 1 − p(R). we should multiply by the probability that the neighbor at distance R doesn’t have a closer neighbor of its own. the asymptotic low-temperature behavior can be found exactly (refs. In 1D. ρ(T ) ∼ 1 ln (J0 /T ) 3 (23. we will need to handle . P (R) = 4πnR2 (1 − p(R))2 or P (R) = 4πnR2 4 1 + 3 πnR3 2 (23.1 Physics of the Metallic State Disorder-Averaged Perturbation Theory A metallic state in a disordered system can be approached perturbatively from the clean metallic state.26).38) Substituting this into (23. (It should really be a fraction of this because we should exclude the region which is less than a distance R from both spins since this region has already been accounted for by the ﬁrst factor of 1 − p(R).40) Hence.41) goes to zero very slowly. 2 the saddle-point approximation for the integral is: χ(T ) ∼ 1 T ln (J0 /T ) 3 (23. up to ln ln T terms. Hence.4. in fact. the neighboring spin at R might itself have already paired up with a third spin. 23. the number of unpaired spins. IMPURITIES IN SOLIDS However. Thus. so slowly.). we ﬁnd the low-temperature form χ(T ) ≈ 1 T dR 8πnR2 1+ 4 3 2 3 πnR e−βJ0 exp(−2R/a) (23. in which case our spin must look for an even further neighbor with which to form a singlet pair. Exercise: Compute the speciﬁc heat at low T in a random singlet phase.39) The saddle-point is again at R ≈ a ln(J0 /T ).

the free energy in a disordered system can be computed by computing the free energy for a ﬁxed realization of the disorder and then averaging over all realizations of the disorder (with respect to some probability distribution for the disorder). Let us imagine breaking our system into a large number N cells. we will average . qn ) (23. ψ † ] = Sclean [ψ. Since each cell will have a diﬀerent impurity conﬁguration. τ )ψ(x. All higher moments are taken to be factorizable into products of V V .. for simplicity. . suppose that the action is S[ψ. We can therefore organize the free energy in powers of V (q): F [V (q)] = n 1 n! . a distribution for V (q) of the form V (q) = 0 V (q)V (q′ ) = ni v 2 δ(q + q′ ) (23.. then the free energy will only contain connected diagrams. V (qn )f (q1 . If we compute an extensive quantity. In other words. PHYSICS OF THE METALLIC STATE 417 the perturbation caused by impurity scattering. Let us consider. Hence. namely that Sclean is a free theory).44) where ni is the impurity concentration and v is a measure of the strength of the scattering potential due to each impurity.23. the sum over cells will be an average over impurity conﬁgurations. For the sake of concreteness. τ ) (23. we describe the simplest way to do so. . Then the conﬁguration of impurities in each cell will be diﬀerent. we assign a factor of the Fourier transform of the potential. Let us consider a perturbative computation of the free energy. . Such an approximation (which we take to be exact in the thermodynamic limit) for the free energy is called the quenched approximation and such an average is called a quenched average.3a.4. In this section. Now suppose that we average over V (q). . . such as the free energy. To each such vertex. each of which is macroscopic. then it will be essentially (up to boundary terms) the sum of the free energies of each of the cells.43) If we think about computing the free energy perturbatively in some coupling constant in Sclean (which includes the simplest case. ψ † ]] + dτ d3 x V (x)ψ † (x. V (q) and integrate over q. V (q1 ) .42) Then there is a vertex in the theory of the form depicted in ﬁgure 23.

4. then the disorder-averaged perturbation theory can have no closed electron loops. For example.3: (a) The impurity vertex for a ﬁxed V (q). F [V (q)] over the distribution: F = DV F [V ] e− R dd xV 2 (x)/2ni v2 (23. if the clean theory is a non-interacting theory. There are two important diﬀerences with most of the Feynman rules which we have considered thus far. (b) The induced ‘interaction’ which results from averaging over the propability distribution for V (q). only momentum. as shown in ﬁgure 23.45) Integrating over V ties together diﬀerent impurity vertices with impurity lines in all possible ways. ﬁgure 23. IMPURITIES IN SOLIDS V(q) k k −q n iv2 k (a) k+q (b) k k+q Figure 23.4a is allowed.2 Lifetime.418 CHAPTER 23. but 23. according to Wick’s theorem. In particular. . we attach a factor of ni v 2 δ(q + q′ ). 23. To each such line. Correlation functions are obtained from functional derivatives of the free energy with respect to sources.3b. (2) The diagrams must be connected even if all of the impurity lines are cut. we have a new Feynman rule associated with impurity lines connecting impurity vertices. In essence. (1) No energy ﬂows along these lines. so the same rules apply to them. Therefore.4b is not. Mean-Free-Path The properties of a dirty metal can be derived neatly with the use of an eﬀective ﬁeld theory which we will discuss in the next chapter.

Consider the electron Green function. Many of the properties of a dirty metal are qualitatively the same as those of a clean metal: the compressibility is ﬁnite and approximately temperature-independent at low-temperatures. for reasons which will be clear shortly.4: (a) An allowed diagram in disorder-averaged perturbation theory. it approaches a constant value at T = 0. we will just summarize these properties here and describe their qualitative features. as is the magnetic susceptibility. The speciﬁc heat is linear in T . The principal new feature in a dirty metal is that the DC conductivity does not diverge at low-temperatures.23. V (q) V (q′ ) = ni v 2 δ(3)(q + q′ ) (23.47) The disorder-averaged value of this diagram is equal to 1 sgn(ǫn ) ≡ ImΣ(ǫ. If the interaction between the electrons and the impurities is Himp = d3 k d3 k′ dǫ V (k′ − k) ψ † (k. This diagram is the lowest-order contribution to the self-energy. ǫ) (2π)3 (2π)3 2π (23. The lowest-order contribution to the self-energy comes from the diagram depicted in ﬁgure 23. This can be seen with a perturbative calculation. PHYSICS OF THE METALLIC STATE 419 (a) (b) Figure 23.4a. We will call the imaginary part of this diagram (i/2τ ) sgn(ǫn ).4. we take V (q) = 0 . (b) A disallowed diargam in disorder-averaged perturbation theory. k) 2τ . for simplicity. ǫ) ψ(k′ .46) where.

adv (k.4. Fourier transforming the fermion Green function at the Fermi energy into real space.52) .49) The corresponding retarded and advanced Green functions are: Gret. we can perturbatively compute the conductivity. we see that the mean-freepath ℓ = vF τ is the momentum uncertainty of an energy eigenstate. ωn ) j(0. Consider the basic ‘bubble’ diagram.5. where it has power-law behavior. The inverse-lifetime 1/τ is the energy-uncertainty of a momentum eigenstate. ﬁgure 23.420 CHAPTER 23. ǫn ) = 1 i iǫn − ξk + 2τ sgn(ǫn ) (23.48) The real part merely renormalizes the chemical potential. unlike in a clean system. momentum eignenstates are no longer energy eigenstates. −ωn ) = 1 1 2 vF NF dξk 3 β n 1 iǫn + iωm − ξk + i 2τn+m 1 iǫn − ξk + i 2τn (23. j(0.51) Thus. ǫ) = 1 ǫ − ξk ± i 2τ (23. It is essentially the average distance between scatterings of an electron since the momentum is the inverse of the spatial rate of change of the wavefunction and the latter is scrambled at collisions. ǫ) ∼ e−|x|/ℓ (23.49) rather than the bare Green function. we have a Matsubara Green function of the form: G(k. at q = 0 but with each electron line representing the Green function (23.3 Conductivity With this single-particle Green function. 23. Writing ξk = vF |k − kF |.50) In the presence of impurities. IMPURITIES IN SOLIDS d3 k′ V (k − k′ ) V (k′ − k) (2π)3 iǫn − ξk = Im = 2πNF ni v 2 sgn(ǫn ) (23. we ﬁnd Gret (x. If we add this self-energy to the inverse Green function. the single-particle Green function decays exponentially.

4.5: The basic conductivity bubble. −ωn ) = dξk dξk 1 dǫ nF (ǫ) ǫ + iωn − ξk + −∞ 2πi dǫ nF (ǫ) −∞ 2πi ∞ ∞ 1 i 2τ ǫ − ξk + 1 i 2τ i 2τ − 1 ǫ − ξk − i 2τ 1 ǫ − ξk + i 2τ − ǫ − ξk − 1 i ǫ − iωn − ξk − 2τ (23.54) Only the second term in parenthesis in the ﬁrst line of (23. ωn ) j(0. we convert the sum over Matsubara frequencies to an integral j(0. ωn ) j(0. −ωn ) = − 2 vF NF 3 dξk dz 1 nF (z) z + iωn − ξk + c 2πi 1 i 2τz+m z − ξk + i 2τz (23. where we have introduced the shorthand τn = τ sgn(ǫn ). The factor of 1/3 comes from the angular average over the Fermi surface. PHYSICS OF THE METALLIC STATE 421 j(q.54) contributes to the imaginary part of the integral and only the ﬁrst term in parenthesis in the second line of (23.53) The integrand is non-analytic if z or z + iωm is real so the integral is equal to the contributions from the contour as is passes just above and just below these lines. Let us assume that ωn > 0.ω ) j(−q. As usual.23. Hence. − − 2 vF NF 3 2 vF NF 3 j(0.54) contributes to the imaginary part of the integral because the other terms have poles on the same side of the axis.−ω) Figure 23. we . The electron lines represent lines dressed by the one-loop self-energy correction computed in the previous section.

57) Taking ω → 0 and noting that (nF (ǫ) − nF (ǫ + ω)) /ω → δ(ǫ). In the ﬁnal line. −ωn ) = 2 vF NF 3 CHAPTER 23. In general. we have σ(ω) = 2 1 vF NF ω 3 dξk dξk − 2 1 vF NF ω 3 dǫ 1 nF (ǫ) 2π ǫ + ω − ξk + −∞ dǫ 1 nF (ǫ) 2π ǫ − ξk + −∞ ∞ i 2τ ∞ 1 i 2τ 1 ǫ − ω − ξk − ǫ − ξk − i 2τ i 2τ (23. we have: σ(ω) = 2 1 vF NF ω 3 dξk 1 dǫ (nF (ǫ) − nF (ǫ + ω)) 2π ǫ + ω − ξk + −∞ ∞ 1 i 2τ ǫ − ξk − i 2τ (23. we have written vF NF = kF /π 2 .55) Taking iωn → ω and dividing by ω. we have σDC = 2 vF 1 NF dξk 2+ 1 6π ξk 2τ 1 2 = vF τ NF 3 1 2 ≡ 2 kF ℓ 3π 2 (23. (This is the d−1 3D expression.58) As expected. IMPURITIES IN SOLIDS dξk dξk − 2 vF NF 3 1 dǫ nF (ǫ) ǫ + iωn − ξk + −∞ 2πi 1 dǫ nF (ǫ) 2πi ǫ − ξk + −∞ ∞ i 2τ ∞ 1 i 2τ 1 ǫ − iωn − ξk − ǫ − ξk − i 2τ i 2τ (23.) .56) Shifting ǫ → ǫ + ω in the second integral and combining the two terms. ωn ) j(0. it is a constant determined by the lifetime due to impurity 2 scattering.422 have j(0. it is proportional to kF .

and done the momentum integral without a cutoﬀ (e. If we deﬁne a diﬀusion constant D = 1 2 3 vF τ . Note that we have not included vertex corrections in our calculation. can be obtained explicitly by computing the densitydensity correlation function. i = 1. so there are important vertex corrections to the density vertex. as a result of the cutoﬀ.59) As usual. However. This is not a problem because there are no vertex corrections to the current vertex to this order. we have used the notation pµ = (ω. which is the Einstein relation. would be missing. by contour integration) then we would have actually obtained the same result for the real part of the current-current correlation function. 0) = ∂Σ(p) ∂pµ (23. 3. which relates vertex corrections to the derivative of the self-energy. The order in which we did the integrals is safest since it can be done with a momentum cutoﬀ. Further insight into this can be gained by computing the density-density correlation function. it is safest to introduce a momentum cutoﬀ. D. 23. then σ = NF D. When they do not converge. 2. this correlation function has a very diﬀerent form in a disordered electron system. pi ). the imaginary part. If we had ignored this subtlety and done the integrals in the opposite order. Γ0 . so the most natural way to proceed is to perform the energy integral ﬁrst. to which we turn in the next section. which we found indirectly in the DC conductivity above.4. As we will see. e.23. which includes a term which cancels the diamagnetic term. On the other hand. The resulting momentum integral is then convergent in most cases. as in the present case.4 Diﬀusion The ﬁnite diﬀusion constant of electrons. as may be seen from the Ward identity. the self-energy is energy-dependent.g. .g. there are no vertex corrections to the current vertices Γ0 . p. Since the selfenergy is momentum-independent. When this is done. A contour integration over ξk cannot be done. Γµ (p. PHYSICS OF THE METALLIC STATE 423 The ﬁnite conductivity of electrons reﬂects the fact that electrons move diﬀusively rather than ballistically. it doesn’t matter in what order they are done. when the integrands are formally divergent by power-counting. however. Note that we computed this integral by performing the energy integrals ﬁrst and then the momentum integrals. When these integrations do not commute.4. the order of integrations does not matter. the order does matter. When the integrals are convergent.

As noted in the previous subsection. ǫn + ωm ))2 + . IMPURITIES IN SOLIDS (a) + k+q k +q + k+q k +q + k +q + k k k + . ǫn + ωm ) + ni v 2 (I(q. ǫn .424 CHAPTER 23.. the lifetime). there are no vertex corrections to the conductivity (i. ǫn . ǫn . ǫn ) G(k + q. ǫn +ωm ) (2π)3 n (23. i. The higher-order diagrams constitute vertex corrections to the ﬁrst diagram in the series.60) where Γ(q. ǫn + ωm ) (2π)3 (23.61) and the integral I(q. (b) 1 + k k + . ǫn + ωm ) = d3 k G(k. such vertex corrections are important for the density-density correlation function.e. ǫn ) G(k+q.. . ǫn ..6: (a) A geometric series of diagrams which contribute to the density-density correlation function. again. −ωm ) = 1 β Γ(q. the corresponding diagrams for the current-current correlation function do not correct the vertex in the DC limit. ǫn +ωm ) G(k. We will show that the resulting correlation function is consistent with the Ward identity. ωm ) ρ(−q. . (b) The quantity Γ(q. However. the electron Green function dressed with the one-loop self-energy.62) 2 .e. ǫn + ωm ) = 1 + ni v 2 I(q. ǫn + ωm ) is the inﬁnite series of ladder diagrams in (23. Consider the set of diagrams in ﬁgure (23. ǫn . which is why we computed only a single diagram). (23.6)a (with. They form a geometric series which we can sum once we have obtained the value of the ﬁrst. All of these diagrams contribute to the density-density correlation function..6)b ρ(q. The sum of these diagrams is d3 k Γ(q. ǫn . ǫn +ωm) which is the ladder sum which sits inside the density-density bubble. ǫn + ωm ) is given by I(q. ǫn . Figure 23. ǫn .

we have I(q. ǫn . Integrating ξk .63) where we have used the shorthand τn = τ sgn(ǫn ). Γ(q. ǫn . ǫn + ωm ) = iπNF [θ(ǫn ) − θ(ǫn + ωm )] ln vF q = iπNF [θ(ǫn ) − θ(ǫn + ωm )] ln vF q iωm + vF q + iωm − vF q + i τ i τ d(cos θ) θ(ǫn ) − θ(ǫn + ωm ) i iωm − vF q cos θ + τ [θ(ǫn + ωm ) − θ(ǫn )] (23. I(q. ǫn . ǫn+m have opposite signs. ǫn . ǫn + ωm ) = 1 1 − ni v 2 I(q. ǫ n . ǫn + ωm ) = iπNF Integrating cos θ. ǫn .61).4. I(q. in which case: I(q. Summing the geometric series (23.66) 2 2ivF τ q − 2ivF τ q ωm τ [θ(ǫn + ωm ) − θ(ǫn )] + (ivF τ q)3 3 This expression vanishes unless ǫn . PHYSICS OF THE METALLIC STATE Written explicitly.64) [θ(ǫn + ωm ) − θ(ǫn )] [θ(ǫn + ωm ) − θ(ǫn )] 1 + (ωm τ + ivF τ q) [θ(ǫn + ωm ) − θ(ǫn )] 1 + (ωm τ − ivF τ q) [θ(ǫn + ωm ) − θ(ǫn )] (23. ǫn + ωm ) . ǫn (ǫn + ωm ) < 0) = 2πNF τ 1 2 1 − |ωm | τ − vF τ 2 q 2 3 = 2πNF τ 1 − |ωm | τ − Dτ q 2 (23. ǫn + ωm ) = 1 NF dξk d(cos θ) 2 iǫn + iωm − ξk − vF q cos θ + 425 i 2τn+m 1 i iǫn − ξk + 2τn (23. ǫn + ωm ) = − iπNF [θ(ǫn + ωm ) − θ(ǫn )]2 × vF q (23.67) 2 where the diﬀusion constant D is given by D = vF τ /3.65) Expanding to lowest non-trivial order in ωm and q.23. we have: I(q.

Integrating in the opposite order will give us an answer which vanishes in the ωm = 0 limit. ωm : d3 k θ(ǫn ǫn+m ) G(k.70) In order to do this integral correctly. To make this a little more concrete. ǫn ) G(k+q.e. −ωm ) = + 1 β d3 k n 1 β n (2π)3 1 1 θ(−ǫn ǫn+m ) G(k. we must integrate frequency ﬁrst or else work with a cutoﬀ in momentum space.60): d3 k θ(ǫn ǫn+m ) G(k. ωm ) ρ(−q. we can do the integrals in the opposite order in evaluating that term. ǫn ) G(k + q. ωm ) (2π)3 iǫn − ξk − vF q cos θ iǫn − ξk = −NF + O(1/τ.68) = τ |ωm | + Dq 2 where we have used 2πNF τ ni v 2 = 1. we will miss the static part. The same form is obtained in any 2 dimension.69) The ﬁrst term is essentially the static response of free fermions with τ = ∞ up to corrections of order 1/τ . This form indicates that the electron density diﬀuses. ǫn ) G(k + q. 1 β n n . with minor diﬀerences. ωm ) (23. to which we now turn. ǫn +ωm) (2π)3 ρ(q.69) – we are left with a dynamic piece which is strongly convergent because the Matusbara sum is over a ﬁnite range. such as D = vF τ /d in d dimensions. ǫn + ωm ) τ |ωm | + Dq 2 (23. consider the density-density correlation function (23. IMPURITIES IN SOLIDS 1 Dτ q 2 ) [θ(ǫ 2 1 − (1 − |ωm | τ − n + ωm ) − θ(ǫn )] 1 1 θ(−ǫn ǫn+m ) + θ(ǫn ǫn+m ) (23.426 = CHAPTER 23. ǫn + ωm ) = (2π)3 1 β d3 k 1 1 + O(1/τ. By separating the density-density correlation function into static and dynamic pieces – the two terms of (23. Hence. i.

ωm ) ρ(−q.72) This correlation function has a pole at |ωm | = −Dq 2 . we will ﬁnd in the next section that such an argument is too quick.4. but these are clearly not the only possible diagrams. . then we have f (ω + iδ) = −iω. since the density-density correlation function which we obtained vanishes at q = 0. In order to continue to real time. (Exercise: check this. as required by the Ward identity (see below for a recapitulation of this fact). our neglect of other diagrams is at least consistent with charge conservation. ω) ρ(−q. t). However.71) Combining the two terms. −ωm ) = −NF Dq 2 |ωm | + Dq 2 (23. ǫn + ωm ) = (2π)3 τ |ωm | + Dq 2 n 1 1 [θ(ǫn+m ) − θ(ǫn )] i 2 τ |ωm | + Dq iωm − vF q cos θ + τ [θ(ǫn+m ) − θ(ǫn )] 2πi NF β 2 n ≈ θ(−ǫn ǫn+m ) 1 1 i 2 τ |ωm | + Dq τ = NF |ωm | |ωm | + Dq 2 (23. −ω) = −NF Dq 2 −iω + Dq 2 (23.73) The existence of a pole at iω = −Dq 2 means that the Fourier transform has the form: x2 1 (23. the electron density moves diﬀusively since a small.23. 0)] θ(t) ∼ d/2+1 e− 2Dt θ(t) t Hence. we have ρ(q. ǫn ) G(k + q. PHYSICS OF THE METALLIC STATE The second term can be evaluated by doing a ξk contour integral: 1 β 427 n 2πi NF β 2 d3 k 1 1 θ(−ǫn ǫn+m ) G(k. so that the retarded density-density correlation function becomes: ρ(q. There is a further justiﬁcation for neglecting other diagrams: they are suppressed by a power of 1/kF ℓ.74) [ρ(x. ρ(0. highly localized wavepacket prepared at t = 0 will spread out over a region of size x ∼ t1/2 at time t. observe that f (z) = −iz sgn (Im(z))) = |ωm | if z = iωm . We obtained a diﬀusive form for the propagation of electron density by summing a particular class of diagrams.) However. If we continue to z = ω + iδ.

−ω) = −NF Dq 2 −iω + Dq 2 (23. IMPURITIES IN SOLIDS With the density-density correlation function (23. S= d−1 kF (2π)d dl dd Ω dǫ † ψ (iǫ − vF l) ψ 2π d3 k d3 k′ dǫ V (k − k′ ) ψ † (k.78) In the presence of impurities. we have the following relation between correlation functions: q 2 j(q. ǫ) ψ(k′ . This can also be easily seen by direct calculation of S+ S− : the electron line will have spin +1/2 while the hole line will have spin −1/2. we have iω D 1 j(q. Consider. The spin-spin correlation function also has a diﬀusive form. ω) S− (−q. for instance. the Sz − Sz correlation function. from the previous expression for the density-density correlation function.73) in hand. electrons no longer move ballistically at long length scales.75) Fourier transforming this equation. They move diﬀusively because they undergo many collisions with impurities. the other spin-spin correlation functions must be identical.428 CHAPTER 23.) Hence. −ω) = NF (23. Charge conservation tells us that ∂t ρ + ∇ · j = 0 (23. −ω) = ω 2 ρ(q. ω) j(−q.79) + (2π)3 (2π)3 2π .76) (The vanishing of the right-hand-side for q = 0 conﬁrms that our expression for the density-density correlation function satisﬁes the Ward identity. we can re-visit the conductivity. Consider the eﬀective Lagrangian for a Fermi liquid in the presence of impurities. We can think of this in the following RG language. we have the DC conductivity. −ω) (23. By rotational symmetry. and then ω → 0. ǫ) (23. it is precisely the same as the density-density correlation function because Sz = ρ↑ − ρ↓ implies that Sz Sz = ρ↑ ρ↑ + ρ↑ ρ↑ = ρρ (the cross-terms vanish for non-interacting electrons).77) iω iω + Dq 2 Taking q = 0. but the diagrams will be unchanged from the density-density calculation because the fermion propagators are spin-independent: S+ (q. ω) ρ(−q. ω) j(−q. For non-interacting electrons.

In the next chapter. However. . some consolation that we can conﬁrm our intuition by direct calculation.) The resulting impurity-scattering term is strongly-relevant.4. When we integrate over diﬀerent impurity conﬁgurations according to (23. Consider. At this ﬁxed point. the density-density and spin-spin correlation functions takes the diﬀusive form (23.5 Weak Localization In the previous two subsections. perhaps. One hardly needs sophisticated ﬁeld-theoretic techniques to discover this.47) we generate the eﬀective action d−1 kF dǫ dl dd Ω ψ † (iǫ − vF l) ψ d (2π) 2π 3 k d3 k′ dǫ d3 p d3 p′ dǫ′ d ψ † (k. (23. we will construct an eﬀective ﬁeld theory for the long-wavelength.49) is short-ranged as a result of the i/τ in the denominator.23. ǫ′ ) (2π)3 (2π)3 2π (2π)3 (2π)3 2π (23.4. we expended substantial eﬀort to compute the electron Green function. (It is similar to a four-Fermi interaction but with one fewer energy integration. 23. scaling as s−1 . although it is. This is what would be expected from a single-impurity. This lifetime also enters the diﬀusive form of various cor2 relation functions through the diﬀusion constant D = vF τ /d. we suppress the marginal four-Fermi interactions parametrized by Landau parameters which are inessential to this discussion. ǫ)ψ † (p. this term is precisely marginal. ǫ′ ) ψ(p′ . and the density-density and spin-spin correlation functions.78) while the single-fermion Green function (23.) Thus the clean Fermi liquid is unstable to impurity-scattering. The RG ﬂow is towards the diﬀusive Fermi liquid. a random V (k). For a ﬁxed V (k). ǫ) ψ(k′ . PHYSICS OF THE METALLIC STATE 429 To avoid clutter. (We will introduce a formal way of implementing this in the next chapter.73).80)) were cut. low frequency modes of the diﬀusive ﬁxed point. the result of all of these diagrammatic calculations was no more than simple qualitative considerations and naive Drude theory would have given us: electrons have a ﬁnite conductivity which is determined by the lifetime τ .80) Seﬀ = + ni v 2 with the proviso that we should only allow diagrams which would remain connected even if all of the impurity lines (connecting the two pairs of fermions in the second line in equation (23. now.

ǫn+m ). the middle of the diagram is a particle-particle diagram rather than a particle-hole diagram.. (b) A re-drawing of the maximally-crossed diagrams which emphasizes the role of the particle-particle diagrams shown in (c). Because particles and holes .. we will see why these diagrams give a signiﬁcant contribution to the condutivity. the ‘subleading’ correction is divergent at low-temperatures and.7b are very similar to those of the diﬀusion propagator.. By time-reversal symmetry.7c W (k. the impurity lines are maximally-crossed.430 CHAPTER 23. In these diagrams. The diagrams of 23. IMPURITIES IN SOLIDS (a) + + + + . They are due to quantum interference eﬀects.. time-reversal symmetry relates these two. The main diﬀerence is that the arrows are in the same direction. Consider the diagrams of ﬁgure 23. One might naively think that. k k −q k −p k Figure 23. this is all that there is to the conductivity of a dirty metal.7: (a) Diagrams contributing to the conductivity with impurity lines maximally-crossed. This is a particular class of diagrams contributing to the conductivity. In this subsection. non-interacting electrons are always insulating in d = 2..e.7a. ǫn . Field-theoretic techniques are invaluable in obtaining and understanding these corrections. k′ . However. In two dimensions. the subleading (in 1/kF ℓ) impurity contributions to the conductivity are also interesting. k k+q k + k k+q k+p k + (c) k k −q k .. (b) + + + + . we can reverse the direction of one of the arrows if we also reverse the sign of the momentum. However. Observe that they can be re-drawn in the manner shown in ﬁgure 23. Call the sum of particle-particle diagrams in ﬁgure 23. in fact. rather than opposite directions.7b. i. in the absence of interactions.

ǫn ) G(k.82) From (23. we take k = −k′ . ǫn . a hole line running through a diagram can be replaced with a particle line at the same energies but opposite momenta without changing the value of the diagram. ǫn ) G(k′ . but carry opposite momentum. as in ﬁgure 23. we see that: W (k.8: A particle-particle diagram is equal to a particle-hole diagram with particle momenta reversed to become hole momenta. This is depicted in ﬁgure 23. ǫn . we obtain δσ(ωm ) = 2 1 vF ωm d dd k dd k′ 1 (2π)d (2π)d β n G(k. k′ . ǫn+m ) = ni v 2 I(k + k′ . ǫn+m ) 3 2 + ni v 2 I(k + k′ .23.4. both cost positive energy. ǫn . k′ . 2 ni v 2 I(k + k′ . Hence. ǫn + ωm )× W (k. we write q = k + k′ and change the variables of integration to k. PHYSICS OF THE METALLIC STATE k k+q k+p k k k+q k+p 431 k = k k −q k −p k −k −k+q −k+p −k Figure 23.81) τ |ωm | + D(k + k′ )2 2 If we substitute this into a conductivity bubble to compute its contribution to the conductivity.8. q. we see that W (k. . ǫn . ǫn . ǫn+m ) ni v 2 1 = θ(−ǫn ǫn+m ) (23. ǫn . In the second pair of Green functions. neglecting the weak . From this observation. . ǫn + ωm ) (23. ǫn+m ) + . ǫn . ǫn+m ) = 1 − (ni v 2 )I(k + k′ . ǫn+m ) G(k′ .7b.81). k′ . ǫn+m ) is sharply peaked at k ≈ −k′ .

82)-(23. so this is the appropriate cutoﬀ. D 1 ln 2 4π |ωm | ℓ2 δσ(ωm ) D 1 ln =− σ0 πkF ℓ |ωm | ℓ2 δσ(ωm ) = − (23. it is divergent in this limit for d < 2. in . Hence. in d ≤ 2.83) We have taken the ultraviolet cutoﬀ to be 1/ℓ.84) which is also divergent.84). For shorter wavelengths. one should use ballistic. which is formally smaller than the (semiclassical) Boltzmann result by a power of kF ℓ ∼ D is. While (23. it is. IMPURITIES IN SOLIDS q dependence of those factors. In d = 2. We now have: δσ(ωm ) = v2 1 NF F ωm d dξk dd q 1 (2π)d β 2 i 2τn dd q n ni v 2 τ θ(−ǫn ǫn+m ) |ωm | + Dq 2 2 i 2τn+m × = v2 4πi NF F ωm d 1 iǫn − ξk + 1 iǫn + iωm − ξk + n 1 d β (2π) ni v 2 τ θ(ǫn+m ) − θ(ǫn ) |ωm | + Dq 2 3 × 1 v2 4πi NF F ωm d i iωm + τ (θ(ǫn+m ) − θ(ǫn )) ≈ dd q 1 (2π)d β n ni v 2 τ θ(ǫn+m ) − θ(ǫn ) |ωm | + Dq 2 3 × =− v2 4π NF F τ 3 ωm d 1 i τ (θ(ǫn+m ) − θ(ǫn )) dd q 1 (2π)d β dd q n ni v 2 τ θ(−ǫn+m ǫn ) |ωm | + Dq 2 1 =− D π 1 2π d/2 1 =− d d d−2 π Γ( 2 ) (2π) 1 (2π)d |ωm | + Dq 2 1 ℓd−2 − |ωm | D d−2 2 (23. the correction (23. rather than diﬀusive propagators.432 CHAPTER 23.82) is a small correction in the low frequency limit for d > 2.

a trajectory from r back to r itself will always interfere constructively with the time-reversed trajectory which is scattered by the same impurities. it is called weak localization. Diﬀerent trajectories from r to r′ will therefore interfere with each other – sometimes constructively. Consider the probability for an electron at r to return to r within time tL : tL P (tL ) = τ dt (Dt)−d/2 (23. Isn’t it possible that if we computed some more diagrams. The diﬀusive mode is a slow mode because it is related to the conservation of charge. it drives the system towards localization. we will discuss the result of this instability in more detail but.85) where tL = D/L2 is the time by which the electron has diﬀused across the system. temperatures) and growing only logarithmically. a complicated function of the trajectory because of the possibility of repeated impurity scatterings along the way. Thus. the starting point of the diﬀusive ﬁxed point is unstable. In d = 2.4. Thus. However. when this correction is still weak (suﬃciently high frequencies or. The divergences which we ﬁnd are due to the existence of slow modes in the system. Thus. This enhanced backscattering is the source of weak localization. although these same slow modes can lead to subleading divergences. The amplitude for an electron to go from r to r′ is. A simple physical picture for weak-localization is given by the notion of coherent backscattering. In the next section. compared to the amplitude to go to any other r′ . sometimes destructively. as we will describe below. there will be an enhanced backscattering amplitude from r back to r. divergent as ω → 0. One might wonder whether is is valid to stop with the diagrams which we have calculated. so we do not need to worry about other entirely new divergences. The correction to the conductivity is δσ ∼ P (tL ) ∼ τ 1−d/2 − tL 1−d/2 σ (23. There is no other mode which is guaranteed to be slow by symmetry.86) Any eﬀect which inhibits constructive interference. We will discuss this point in the next chapter using an eﬀective ﬁeld theory for the slow modes of the system. for now. we observe that it suppresses the conductivity. we would ﬁnd other divergent corrections to the conductivity which might change our conclusions entirely? The answer is that this will not happen and we can stop here. such as inelastic scat- . PHYSICS OF THE METALLIC STATE 433 fact.23. but in reverse order. The particle-particle mode (called a ‘Cooperon’) is slow because it is related to the diﬀusion mode by timereversal symmetry.

If such sources of inelastic scattering are considered.88) δσ(ωm ) = − 2 ln(ℓThouless /ℓ) 2π If τin ∼ T −p . Thus far. from which we can obtain the single-electron density of states: N (E) = dd k G(E. the magnetoresistance is negative.89) 2π the conductivity increases as the magnetic ﬁeld is increased.4. spoil weak localization.6 23. they cut oﬀ the above divergences. IMPURITIES IN SOLIDS tering or a magnetic ﬁeld will. not T . Thus. A weak magnetic ﬁeld will also cut oﬀ localization at length scales longer √ than the magnetic length ℓH = 1 H. diﬀusive motion is so slow that it enables electron-electron interactions to cause divergent corrections at the diﬀusive ﬁxed point.7 Weak Magnetic Fields and Spin-Orbit Interactions: the Unitary and Symplectic Ensembles Electron-Electron Interactions in the Diﬀusive Fermi Liquid How are the conclusions of the previous subsections altered by the inﬂuence of electron-electron interactions? How does the diﬀusive motion of electrons aﬀect the relevance or irrelevance of interactions at the diﬀusive ﬁxed point.90) .e. as compared to their marginality at the clean Fermi liquid ﬁxed point? As we will see in this subsection. For example.87) inelastic processes will thwart localization and cause ohmic conduction.) At these length scales. 23. The simplest example of such a divergence is in the single-electron Green function. c/eH.4. k) (2π)d (23. the 2D weak-localization correction is: 1 (23. we have considered non-interacting electrons and neglected phonons.434 CHAPTER 23. as one might naively assume. i. therefore. then the frequency is replaced by ∼ T p . At length scales longer than the Thouless length ℓThouless = Dτin (23. the time-reversed trajectory it is ℓH = diﬀers in phase by ∼ π as a result of the magnetic ﬂux which it encloses and no longer interferes constructively. etc. (Restoring the fundamental constants. in the weak localization regime. where 1 δσ(ωm ) = − 2 ln(ℓH /ℓ) (23.

.4. ǫn .68). k) ≈ m dd q (2π)d 1 θ(−ǫn ǫn+m ) τ |ωm | + Dq 2 2 1 θ(−ǫn ǫn+m ) 2 (23. The basic density vertex is renormalized by repeated impurity scatterings according to (23.10. ǫn + ωn ). Figure 23. ǫn +ωn ) is given in (23. Its leading contribution to the self-energy is: Σex (ǫn . one for each particle-hole pair which is coupled by V . PHYSICS OF THE METALLIC STATE 435 + + . the interaction vertex V (q) is renormalized in a corresponding way.9: The renormalization of the density vertex by repeated impurity scatterings.91) where Γ(q. interact more strongly. k) = 1 β dd q 1 1 V (q) θ(−ǫn ǫn+m ) d (2π) τ |ωm | + Dq 2 1 iǫn + iωm − ξk+q + dd q (2π)d 2 m × (23. ǫn . There are two factors of Γ(q. Σex (ǫn .. Physically. q ≈ 0 so we write Σex (ǫn . therefore.10. as depicted in ﬁgure 23. The unarrowed solid line on the left is the interaction V . Then the main eﬀect of disorder is to modify the nature of the interaction vertex. Consider. ǫn + ωn ))2 (23. the integral is dominated by ωm ≈ 0. Consider the self-energy diagrams of ﬁgure 23. Hence.92) i 2τn+m Clearly.9: V (q) → V (q)(Γ(q. they spend more time in close proximity and. the ﬁrst diagram (exchange) in ﬁgure 23.23. then. k) ≈ V (0) iǫn − ξk + i 2τn 1 β m 1 ≈ −2iV (0)τ sgn(n) β Consider the case d = 2. ǫn .93) τ |ωm | + Dq 2 2 1 −iV (0) sgn(n) 2πτ β dq 2 m θ(−ǫn ǫn+m ) |ωm | + Dq 2 .68). Let us assume that the electrons interact through some eﬀective density-density interaction V (q). since the electrons move more slowly when the diﬀuse.

ǫ))2 Σ(ǫ. and we ﬁnd: V (0) sgn(n) ln(|ǫn |τ ) (2π)2 Dτ Σex (ǫn . k) (2π)2 V (0)NF ln(|ǫ|τ ) = (2π)2 D 2 V (0)NF = ln(|ǫ|τ ) 2πǫF τ δNex (ǫ) = (23. the vertices are dressed by repeated impurity scatterings so that they take a diﬀusive form. IMPURITIES IN SOLIDS Figure 23. For low T . that ǫn < 0.10 has opposite sign as a result of the closed fermion loop and an extra factor of two from the spin sum in .94) m Let us suppose. The second diagram (Hartree) in ﬁgure 23. without loss of generality. Then the Matsubara frequency sum runs over ωm > −ǫn .10: The exchange (Fock) and Hartree diagrams contributing to the electron self-energy. the sum can be approximated by an integral. Where particle and hole lines are connected by dotted lines.96) Thus. The unarrowed solid line on the left of the vertex is the interaction V . since for higher frequencies the diﬀusive form is not correct. The ultraviolet cutoﬀ is 1/τ . k) ≈ −i (23. the correction is singular.436 CHAPTER 23.95) The correction to the density-of-states is d2 k Im (G0 (k. 2π −iV (0) sgn(n) 2 Dτ (2π) β θ(−ǫn ǫn+m ) |ωm | ≈ (23.

Hence.97) As before.100) where F = 1 V (0) dθ V (2pF cos θ) 2π is an average of the interaction over all momenta connecting two points on the Fermi surface. so we have ImΣHartree (ǫn . k) ≈ 2τ 2 1 β m d2 q (2π)2 1 θ(−ǫn ǫn+m ) τ |ωm | + Dq 2 = 2 dθ V (2kF cos θ) 2π 2F V (0) sgn(n) ln(|ǫn |τ ) (23.4.101) Since F < 1 (because V (q) decreases with increasing q). we set ωm = q = 0 in the electron propagators.98) The electron Green functions are all peaked around the Fermi surface.99) (2π)2 Dτ (23. ΣHartree (ǫn . however. adding the exchange and Hartree terms we have V (0)NF δN (ǫ) = (1 − 2F ) ln(|ǫ|τ ) NF 2πǫF τ (23. notice that the interaction is not at zero momentum. It has leading contribution: ΣHartree (ǫn . k) = −2 1 β d2 p d2 p′ d2 q V (p−p′ ) (2π)2 (2π)2 (2π)2 1 i 2τn 437 m 1 iǫn − ξp + 1 θ(−ǫn ǫn+m ) τ |ωm | + Dq 2 2 × 1 iǫn − ξp′ + iǫn+m − ξp+q + i 2τn i 2τn+m × i 2τn+m 1 iǫn+m − ξp′ +q + (23.23. the exchange terms wins. k) ≈ − 2 β d2 p d2 p′ d2 q V (p−p′ ) (2π)2 (2π)2 (2π)2 i 2τn m 1 iǫn − ξp + 1 × i iǫn − ξp + 2τn 1 1 i iǫn − ξp′ + 2τn iǫn − ξp′ + 1 θ(−ǫn ǫn+m ) τ |ωm | + Dq 2 2 × i 2τn (23. and the net eﬀect of interactions at lowest-order is to suppress the . PHYSICS OF THE METALLIC STATE the loop.

103) The eﬀects of interactions bear the signature of this correction.104) If the electrons interact through the screened Coulomb interaction. the conductivity receives singular corrections even in the non-interacting system. 23. V (q → 0. interactions lead to singular corrections to the density of states in d = 2 (and worse singularities in d = 1 (exercise)). |ǫ|τ ≪ 1.102) V (0) → V (0) (1 + c ln(|ǫ|τ )) (23. A similar correction is expected for the speciﬁc heat.105) and Π(q. Thus. In d = 2. (ω) = 1 (1 − F ) ln(|ω|τ ) 2π 2 (23. Coul. They give a contribution δσ(ω) = V (0)NF (1 − F ) ln(|ω|τ ) 2π 2 (23. ω = 0) = −1 (Π(q → 0. How do interactions aﬀect this? Similar. in which the density-of-states is perfectly smooth. it is no longer valid to use lowest-order perturbation theory. On the other hand. that the conductivity correction is of the form δσscr. 0))−1 = NF . but one can speculate that this suppression eventually leads to a vanishing of the density-of-states at the Fermi energy. V (q) = 4πe2 q2 4πe2 Π(q. but more complicated diagrams than those of ﬁg.106) which is nearly identical in form and magnitude as the weak-localization correction. This implies.438 CHAPTER 23. analogous to the Coulomb gap discussed in the context of insulators. When this correction becomes large. ω) is the density-density correlation function. ω) q2 1+ (23. It is diﬃcult to disentangle the two eﬀects. but it can be done . this is: dt P (t) τ (23. Hence. IMPURITIES IN SOLIDS conductivity. for instance. The basic physics behind this correction is that the interaction between two electrons is renormalized by processes in which they diﬀuse back to each other and interact again: 1/ǫ V (0) → V (0) 1 + where P (t) ∼ 1/(Dt)d/2 .10 determine the lowest-order interaction correction to the conductivity of a diﬀusive metal. This is in contrast to a non-interacting disordered system.

then in the next chapter we will introduce an eﬀective ﬁeld theory for the diﬀusive degrees of freedom.23. Thus. seem as if we have chosen an arbitrary set of diagrams. If this seems a little unsatisfying. Let us call these ‘insulating’ regions. in fact. One way to think about this is to consider the potential V (x). .5. while it may. the conducting regions will grow larger and larger and will begin to merge. while the former is through the destruction of interference). In a system of non-interacting fermions. The regions in which V (x) < ǫF are classically-allowed for electrons at the Fermi energy. there will be some isolated regions which are ‘conducting’. such regions would have single-particle states at the Fermi energy which extend across them but do not leak substantially into the surrounding insulating regions. through the Zeeman coupling. Although it is insulating. but the diﬀusing particle-hole pair would not be near ω = q = 0. it will become possible to go from one end of the system to the other within one long conducting region which spans the system. Then the conducting regions will grow in size.5. we have chosen precisely the ones which have the most singular diﬀusive corrections. so they can be obtained by simply turning the crank. at ﬁrst glance. in which these corrections come from the only diagrams present at lowest order. whose spatial variation is random as a result of impurities. THE METAL-INSULATOR TRANSITION 439 by measuring the magnetoresistance since the weak-localization and interaction corrections are aﬀected very diﬀerently by a magnetic ﬁeld (the latter. Eventually.1 The Metal-Insulator Transition Percolation Consider a disordered solid which is insulating as a result of impurities. The regions in which V (x) > ǫF are classicallyforbidden for electrons at (or below) the Fermi energy. As we continue to decrease the impurity concentration.5 23. Let us consider these to be the ‘conducting’ regions. One could imagine dressing the diagrams of ﬁgure 23. 23.10 with impurity lines in other ways. The diagrams which we have computed in this subsection are the lowestorder diagrams which have diﬀusively-dressed interaction vertices such that the particle-hole lines which are diﬀfusing have low momentum/energy. this impurity concentration – which is called the percolation threshold – would be the transition point between metallic and insulating behavior. If electrical conduction were purely classical and ohmic. Suppose we decrease the impurity concentration.

For the sake of concreteness. the fraction. electrons can tunnel quantum mechanically from one conducting region to another. but not to other sites. site percolation and bond percolation. One site in a cluster can be reached from any other site in the cluster by a sequence of nearest-neighbor hops entirely within the cluster. It can hop directly from a red site to its nearest neighbors if they are also red. Even below the percolation threshold. IMPURITIES IN SOLIDS However. one randomly paints black some of the sites of a lattice and considers the properties of the remaining sites. At this point. and the quantum mechanical amplitudes for these processes might interfere destructively. electrical conduction is not classical. When the average bond density. one cluster extends across the length of the system. in a sense. is reached. An electron can hop from one lattice site to a nearest-neighbor site only if the bond between them is present. is low. For p < pc . the classical limit of the metal-insulator transition. As the bond density is increased. Sites which are red can be grouped into clusters of sites which are continguous with each other. if there is an isolated red site which is surrounded by black sites. the average distance between two sites on the same cluster. Since percolation is. on the other hand. all systems have ﬁnite conductivity) is comparable to that of the conducting ones. we will discuss it brieﬂy. at ﬁnite temperatures. In bond percolation. which are colored red. an electron can take two diﬀerent paths from point A to point B. We imagine that an electron can only be on a red site. Consequently. P . it is called the percolating cluster.440 CHAPTER 23. of the system which belongs to the percolating . Conversely. scales as ξ ∼ (pc − p)−ν (23. There are two basic lattice models of percolation. we expect the conductivity and other physical properties to show behavior characteristic of percolation processes. In site percolation. clusters grow until the percolation threshold. at temperatures which are higher than the characteristic phase coherence temperatures of the system – so that coherent quantum tunneling cannot occur – but not so high that the conductivity of the ‘insulating’ regions (insulating and metallic are precise distinctions only at zero temperature. Nevertheless. which we will call p. then an electron on that site is stuck there.107) For p > pc . a system might be insulating even above the percolation threshold. ξ. one randomly removes bonds from a lattice. let us consider bond percolation. Sites can again be grouped into clusters which are connected by bonds so that any site in a cluster can be reached from any other site in the cluster. pc . Thus. never on a black one. clusters tend to be small.

5. Near the bottom of a band.23. Instead. One might imagine that the conductivity (assuming that bonds which are present are considered to be conducting and those which are absent are insulating) would be proportional to the fractions of the system belonging to the largest cluster. the exponents ν. we discuss alternate approaches which take these eﬀects into account. Minimum Metallic Conductivity Let us now consider a metal-insulator transition in a system of non-interacting fermions. etc. as we pointed out earlier. Near the top of the band. However.108) (For p < pc . 23. THE METAL-INSULATOR TRANSITION cluster is P (p) ∼ (p − pc )β 441 (23. If the system is weakly-disordered.e. However. However.5.) Notice the analogy between these exponents and the correlation length and order-parameter exponents associated with critical phenomena.2 Mobility Edge. The precise value of pc depends on the particular lattice and whether it is bond or site percolation. It is useful . β.109) At the percolation point itself. and depend only on the dimension. µ. the conductivity is associated with a diﬀerent exponent. Such a transition can occur as a result of varying either the chemical potential or the disorder strength. (sometimes called t) which describes the ‘backbone’ of the percolating cluster – i. P = 0 because the largest cluster is ﬁnite even in a system of inﬁnite size. ν = 4/3 in two dimensions. more recently with the use of the stochastic Loewner equation. are universal. Many of the properties of the percolation point in two dimensions have been revealed using conformal ﬁeld theory and. In the remainder of this section. although there may not be a true gap between the bands because there will be exponentially-small tails in the density of states. then the energy spectrum will still approximately break up into bands. the cluster with dead-ends removed (dead-ends are those bonds which do not have two completely independent routes to the two ends of the cluster). the states will all be localized because the kinetic energies of these states will be low. this is not the case because many of the bonds in the cluster lead to dead-ends which do not contribute to the conductivity. the same will be true. For instance. the system will have power-law correlations. percolation does not correctly describe metal-insulator transitions because it misses quantum-mechanical effects. σ(p) ∼ (p − pc )µ (23.

Ec . the hole kinetic energies near the top of the band are low. When a localized state mixes with extended states.) As the chemical potential is moved through the mobility edge. Hence. on the other hand. to think of a band with chemical potential near the top as a nearly empty hole band. For EF < Ec . A small change in the particular realization of disorder (e. the locations of the impurities) would cause mixing between the localized states and the extended ones. Such an energy is called a mobility edge. the states will be extended if the disorder strength is suﬃciently small because the kinetic energies will be large in comparison. but this would not be generic and would require some special circumstances. IMPURITIES IN SOLIDS ρ(E) localized states Ec mobility edge extended states E Figure 23. As depicted in ﬁgure 23. there will be sharp energies which separate localized states from extended ones. (One can imagine there being more than a single pair of such energies. one should ﬁrst note that there are no energies at which there are both extended and localized states because localization is not stable in the presence of extended states at the same energy. the mobility edge separates the strongly-localized states with low kinetic energy in the band tails from the extended states with high kinetic energy in the center of the band. it becomes extended.11: The density-of-states in a disordered system. then even at the middle of the band the kinetic energy will not be large enough to overcome it. with extended and localized states and mobility edges shown. in such a case. (If the disorder strength is too large.442 CHAPTER 23. the zero- . the system undergoes a metal-insulator transition.g.11.) What happens as the Fermi energy is swept from the bottom of the band to the middle? To answer this question. Near the middle of the band.

The conductivity is given in Boltzmann transport theory by σ = Ad e2 d−1 kF ℓ (23. σmin = Ad e2 / . the preceding logic fails in the case of non-interacting electrons.5.6): δE (23.) g(L) = .23. (EF − Ec or any other control parameter will act as a proxy for p − pc near the transition. Since kF ∼ π/a and ℓ > a. For EF > Ec . Consider a system of non-interacting electrons of size L. one must consider disorder-averaged quantities.3 Scaling Theory for Non-Interacting Electrons Thouless had the important insight that one should study the scaling behavior of the conductance. THE METAL-INSULATOR TRANSITION 443 temperature DC conductivity σ vanishes. the conductivity vanishes precisely at the transition.) If scattering is due entirely to impurities. it is ﬁnite. which were set to one earlier. 23. as we will see in the next subsection. Thouless wrote the conductance of the system in the following form. the following heuristic argument suggests that it is ﬁnite. Hence.5. we conclude that σmin = Ad e2 d−2 kF (23. ∆E is the mean level spacing of eigenstates. (In order to derive such a relation analytically from the Kubo formula.110) where ℓ is the mean-free-path and Ad is a dimension-dependent constant.111) In particular. > kF ℓ ∼ π. σ ∼ |EF − Ec |µ . we have restored the factors of e. Indeed.110) – which is semiclassical and valid for large kF ℓ – in the regime kF ℓ ∼ 1. the mean-free path cannot be shorter than the inter-atomic spacing a in the solid. otherwise. one should be cautious in applying the Boltzmann transport formula (23. one could probably consider such a relation for a ﬁxed realization of disorder. which we encountered earlier (23. (Here. the minimum metallic conductivity can be written entirely in terms of fundamental constants. What happens at EF = Ec ? Is it equal to some minimum non-zero value σmin .) On the other hand. However.112) ∆E δE is the disorder-avergaged energy change of eigenstates at EF when the boundary conditions are changed from periodic to antiperiodic. in d = 2. the minimum metallic conductivity? According to the percolation model described in the preceding section.

We now study how g(L) scales as the system size is increased. Thus. The time required for an electron to diﬀuse across the system is the Thouless time: tL = L2 D(L) (23.114) (23. g(L)) (23. However. The key assumption of Abrahams.117) 1 D(L) ∼ tL L2 1 NF L d (23.113) According to the uncertainty principle. the sensitivity of the eigenfunctions to their boundary conditions is precisely what is encapsulated in g(L). and. and Ramakrishnan (‘the Gang of Four’) is that: g(bL) = f (b.116) ∆E The second equality follows from the Einstein relation between the diﬀusion constant and the conductivity. it is very reasonable to assume that δE(bL) depends only on b and g(L). the Ohmic scaling form is not exactly right microscopically. . However. such a relation is obvious.444 CHAPTER 23. suppose that D(L) is the diﬀusion constant for an electron at the Fermi energy. therefore. we need to ﬁnd a relation for δE(bL). L does not appear explicitly on the right-hand-side.117) holds throughout the entire region between these limits. The scaling assumption of the gang of four was that the form (23. We could increase the system size from L to bL by stacking b subsystems of size L. Anderson. δE = NF D(L) Ld−2 = σ Ld−2 = g(L) (23. As noted above. where g(L) = g0 e−L/ξ : g(bL) = g0 (g(L)/g0 )b . This relation might be complicated since we have to match the eigenfunctions from each size L block at their boudaries in order to construct the eigenfunctions of the size bL system. g(bL) = bd−2 g(L). It also holds in the localized regime. that g(bL) does. IMPURITIES IN SOLIDS To understand this formula. this assumption holds in the ohmic regime. ∆E = Hence. While ∆E(bL) = b−d ∆E(L) should hold generally. Licciardello.115) In particular. In a macroscopic Ohmic system. δE ∼ Meanwhile.

2π 2 g (23. β(g) > 0 and the conductance ﬂows to large values. which we drop. the metalinsulator transition occurs at g ∼ 1/ǫ. represents O 1/g2 terms and higher-order in ǫ terms have been dropped. Diﬀerentiating. we can compute in powers of ǫ. (2) There is a metal-insulator transition with no minimum metallic conductivity in d > 2. However. or large g. For g > gc . There are several consequences which follow immediately. (1) There is no metallic state of non-interacting fermions in d ≤ 2 (and. g(L) = g0 1 − 1 L ln 2g 2π 0 ℓ (23.) to inﬁnity. limit. The second statement follows from the positivity of β(g) for large g. For d = 2 + ǫ. β(g) < 0 and the conductance ﬂows to small values. Since g = σ in d = 2. For g < gc .12: since β(g) < 0 for all g. the conductance remains constant.23. therefore. the conductance always ﬂows to the localized regime of small g. In the clean. In the previous section.. This is interpolated with the small g limit in ﬁgure 23. no metalinsulator transition. this is the metallic state. g is small. then we can deﬁne d ln g 1 = d ln L g df db = β(g) b=1 445 (23. of course). we expect Ohmic scaling. Thouless length Dτin . which is in the perturbative regime. equivalently. and all states are localized. magnetic length. . For g = gc . β(g) must pass through zero at some gc . β(g) = ǫ − 1 + . we expect β(g) = ln(g/g0 ). Then 1/L is an infrared momentum cutoﬀ. THE METAL-INSULATOR TRANSITION If (23. then ..118) In strongly-disordered systems. if the conductance remains constant as the system size L goes to inﬁnity. this is the insulating state.. Suppose we reconsider this calculation in a system of ﬁnite-size L. we obtain β(g). and β(g) = d − 2.5.117) holds.. we can compute β(g) in powers of 1/g or.120) where . this is the transition point. We thus obtain σ(L) in d = 2 for large kF ℓ.12. where it is Ohmic. and we can take the other cutoﬀs used in √ the previous section (frequency 1/ ω/D. How do we connect the two limits? For g large. The ﬁrst statement is clear from ﬁgure 23. corrections to this formula for σ are O(ǫ).119) For d = 2 + ǫ. (3) For d = 2 + ǫ.. In this limit. . 1/kF ℓ. we have already computed the ﬁrst correction in 1/kF ℓ to the conductivity. In order to connect to the negative β(g) localized regime at small g.

i. electrical conduction is Ohmic and. ξ is a quantum-classical crossover length. . therefore. .446 CHAPTER 23. On the insulating side. On the conducting side as well. For length scales longer than ξ. ξ is the localization length. in (23. Since the . there is no minimum metallic conductivity. et al. We can say more about the critical region by expanding the β-function about gc : β(g) = 1 ν g − gc gc (23. we can trust our gc = 1/4π 2 ǫ conclusion only for ǫ small. the conductivity σc = gc L2−d → 0.122) This is an important length scale because the system will cross over from critical behavior at short-distances to Ohmic metallic (g > gc ) or localized insulating (g < gc ) behavior at longer length scales. which is obtained by smoothly connecting the Ohmic limit with the localized one. only near two dimensions.12: The β-function of Abrahams. Let us integrate until |g − gc |/gc ∼ 1.120).121) From our β-function (23. Hence. . IMPURITIES IN SOLIDS β gc d=2 d=3 g Figure 23.120). Suppose |g − gc | is small at the microscopic scale ℓ. The third statement follows immediately from (23. where ǫ = d − 2. We deﬁne the length scale at which this is reached as ξ: ξ=ℓ g − gc gc −ν (23.e.120) can be neglected only for g large. The linearized approximation to the β-function is applicable. we see that 1/ν = ǫ and gc = 1/4π 2 ǫ.

This is completely analogous to the case of classical thermal phase transitions studied in chapter . this will be a little more obvious. 23. there is an important diﬀerence: a metal-insulator transition of non-interacting fermions is not a thermodynamic phase transition. et. as one may see by drawing diagrams which might renormalize the single-particle Green function. A simple way of seeing why it is smooth is to consider the disorder-averaged single-particle Green function. However. Within the eﬀective ﬁeld theory approach of the next chapter. the scaling hypothesis of Abrahams. It is short-ranged on both sides of the transition and does not show any qualitative diﬀerence between the metallic and insulating states. after we have developed the appropriate eﬀective ﬁeld theories. was proved by Thouless).123) The vanishing of the minimum metallic conductivity can be understood as the divergence of ξ at the metal-insulator transition. or quantum phase transitions studied in chapter . weak-localization eﬀects. however. Thus. ξ. quantum interference eﬀects can occur and classical ideas are inappropriate. In an interacting system. al is equivalent to the assumption that there is a single length scale. The latter follows from the smoothness of the density-of-states across the transition (which. are perfectly smooth across the transition. such as the ground state energy. For instance. a metal-insulator transition might be a true thermodynamic phase transition.6. Because interacting disordered systems are technically challenging. all thermodynamic properties. which becomes large near the metal-insulator transition. do not aﬀect the single-particle Green function... Since the conductivity is independent of length in the classical Ohmic regime. in turn. which suppress the conductivity. so it (and the density-of-states which is derived from it) varies smoothly across the transition.6 The Integer Quantum Hall Plateau Transition . we will postpone further discussion of this issue to the next chapter. For length scales shorter than ξ.23. THE INTEGER QUANTUM HALL PLATEAU TRANSITION 447 essentially classical. While the conductivity changes sharply at the transition. it is equal to σ= gc ξ d−2 (23.

IMPURITIES IN SOLIDS .448 CHAPTER 23.

G−1 = ǫ − ξk + i/2τ is imaginary. we would like an eﬀective ﬁeld theory which directly gives us this diﬀusion physics without 449 . however. we construct an eﬀective ﬁeld theory for the low-energy. The basic physical observation is that the only low-energy. The ‘mass term’ in the Green function. long-wavelength correlation functions (after disorder-averaging) are densitydensity.1 Introduction In this chapter. Thus. which reﬂects the fact that there is still a Fermi surface’s worth of single-particle states near the Fermi energy. non-zero density of states. it is worth remembering that there is still a constant. for instance.CHAPTER 24 Non-Linear σ-Models for Diﬀusing Electrons and Anderson Localization 24. spin-spin.1. only the diﬀusion of conserved currents remains at long-wavelengths. These states show up in. the speciﬁc heat. long-wavelength correlation functions at the diﬀusive Fermi gas ﬁxed point of ﬁgure 24. Hence. It’s not that there are no low-energy states. (However. etc. correlation functions which are diﬀusive in form as a result of a conservation law. Upon disorder-averaging. All other disorder-averaged correlation functions decay exponentially. Rather. as would be the case with a real mass term. it is that there are low-energy states. but they are not momentum eigenstates. we will present a formalism which will elucidate the results of the previous chapter and cast them in the form of a ﬁeld theory which brings out analogies between the metal-insulator transition and other critical phenomena.

also. there should be some coupling of the density to the electron operator. since it has long-ranged correlations.1: Renormalization group ﬂow diagram for a Fermi gas in the presence of disorder. presumably of the form ρψ † ψ.450 CHAPTER 24.1) However. the encumbrance of the low-energy single-particle states. When the disorder strength becomes suﬃciently strong. The clean ﬁxed point is highly unstable to disorder and ﬂows into the diﬀusive metal. the problem of interaction terms. the latter would induce singular corrections to the electron Green function.) One might be tempted. so we would expect it to have an action like S[ρ] = dd xdt ρ ∂t + ∇2 ρ (24. In order to endow our theory with the necessary properties. which shouldn’t be present in the absence of electron-electron interactions. so we can drop it from the action! Consider. at ﬁrst. to construct an eﬀective theory built on the density operator. but. a metal-insulator transition occurs. Let’s imagine trying to do this. Also. the general philosophy is correct. there are problems with such terms: for instance. However. the time-derivative term is a total derivative. Thus. . to which we turn in the next section. as we will see later. The ﬁrst problem which one must confront is that the density satisﬁes the diﬀusion equation. we see that a simple-minded approach to an eﬀective ﬁeld theory won’t work. The density should have self-interactions – perhaps of the form ρ4 – since these must generate the quantum interference eﬀects which we computed in the previous chapter. we will need some more formalism. beyond which the ﬂows are into the Anderson insulating ﬁxed point. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION Metal−Insulator Transition Clean Fermi Gas * * Diffusive Fermi Gas * Anderson Insulator * Figure 24.

we cannot write. little is known about the validity of taking the replica limit.2) While Z[V ] has a functional integral representation.V ] e− dd xV 2 (x)/2ni v2 (24. There is some sleight of hand involved here since we can do the calculation for integer n but we then take the limit n → 0. there is no diﬃculty in taking this limit. the n independent ψa s become coupled together.V ] e− R dd xV 2 (x)/2ni v2 (24. we have a disorder-averaged action: e−Savg [ψa . However.2 The Replica Method Let us return now to the problem of performing the disorder average of the free energy. an eﬀective action for the disorder-averaged system. THE REPLICA METHOD 451 24.V ] e− R R dd xV 2 (x)/2ni v2 = a=1 † Dψa Dψa DV e− Pn † a=1 S[ψa . . and the limit n → 0 is called the replica method. Thus. apart from a constant 1/n term.ψa .3) This is a useful expression because Z n has a simple functional integral representation if n is an integer: we just imagine that we have n independent copies of the system.ψa . n labels the n identical copies of the system.5) After V has been integrated out. the free energy is 1 F = lim n→0 n n n DV a=1 † Dψa Dψa e− Pn † a=1 S[ψa . Hence. say. .2. ln Z[V ] does not. we now observe that ln Z[V ] = lim 1 n (Z − 1) n→0 n (24. one usually proceeds without answering this question and hopes for the best.24. . it is simply a bookkeeping device which discards unwanted diagrams with closed electron loops. Beyond perturbation theory.ψa ] = † DV e− Pn † a=1 S[ψa . F = = DV F [V ] e− R dx V 2 (x)/2ni v2 R dd xV 2 (x)/2ni v2 DV ln Z[V ] e− (24. 2. within perturbation theory. This method for averaging the free energy is called the replica method. . Hence.4) The replica index a = 1.ψa . . However.

as we saw when we studied perturbative corrections in the previous chapter.ψa ] † (24. a system of non-interacting electrons moving in a random potential: S[ψ. The RG perspective on the diﬀusive Fermi liquid which we introduced in the previous chapter applies directly to this replicated action. electrons interact. ψ † ] = dτ dd x ψ † ∂τ + 1 2 ∇ +µ ψ 2m + † dτ dd x V (x)ψa (x.1 Non-Interacting Electrons Derivation of the σ-model We will focus on interacting electrons or. In the replica limit. and (2) in the real world. and these interactions are evidently important in twodimensions. ψa ] = † dτ dd x ψa ∂τ + 1 n n a=1 † Dψa Dψa e−Savg [ψa . τ )ψa (x. we focus on interacting electrons in the presence of .8) The second term. τ ′ )ψb (x. Hence. we ﬁnd that F = where † Savg [ψa .452 CHAPTER 24. τ ′ ) (24.4b so that this diagram is proportional to n. on methods which can be applied to both non-interacting and interacting electrons. τ ) (24.7) 1 2 ∇ + µ ψa 2m + ni v 2 † † dτ dτ ′ dd x ψa (x.3. which is beyond the scope of this book and is well-described in books and review articles. τ ) ψb (x. Mention Supersymmetry. at least. for example. This action has both ﬁgures 23.6) Following the replica procedure. is non-local in time (reﬂecting the time-independence of the disorder) and couples diﬀerent replicas. τ )ψa (x. Schwinger-Keldysh 24.b but with a free summation over a replica index in 23. There are tow reasons for this: (1) there is a great deal of formalism which has evolved to describe non-interacting electrons. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION Consider. this diagram vanishes.3 24.4a. which results from the disorder average.

E ′ (24. x)ψ(E + iδ.11) In order to perform the disorder average. unlike the electron Green function. Hence. Some of the analogies with other critical phenomena can be made more precise.9) ψ † (E + iδ. x) = (24. NON-INTERACTING ELECTRONS 453 disorder.24.E ′ = dd x † ψR E + iδ + † † † † DψR (x) DψR (x) DψA (x) DψA (x) ψR (x)ψR (x) ψA (x)ψA (x) e−SE. we introduce this theory brieﬂy and comment on its physical interpretation. the problem of non-interacting electrons in under much better control. so that impurity averaging does not make them short-ranged. We introduce n copies of the functional integral n Z n [V ] = a=1 † † DψaR (x) DψaR (x) DψaA (x) DψaA (x) e−Sn E. x) 2 = ψ † (E − iδ. x) (24. x)ψ(E − iδ.12) . x)ψ(E + iδ. Such products of an advanced and a retarded Green function have their phase cancelled.E † + ψA 1 2 † ∇ ψR + V (x)ψR (x)ψR (x) 2m 1 2 † ∇ ψA + V (x)ψA (x)ψA (x) E ′ − iδ + 2m (24. before moving on to the interacting case. x)ψ(E − iδ. we use the replica method. the advanced Green function is the complex conjugate of a retarded Green function at the same energy.10) where SE. x)ψ(E + iδ. x) These Green functions can be obtained from the following functional integral: ψ † (E + iδ.3. so ψ † (E + iδ. However. we need to compute the product of an advanced and a retarded electron Green function at the chemical potential. For instance. and the theory is on much safer footing. In order to compute the conductivity or density-density correlation function. x) ψ † (E − iδ.

However. as we noted earlier. ψaB ] = † dd x ψaB δBC E+ 1 2 ∇ 2m + ΛBC ω + iδ 2 ψaC (24. ψaR . we would recover the results of the previous chapter. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION Sn E.13) † + ψaA E ′ − iδ + 1 2 † ∇ ψaA + V (x)ψaA (x)ψaA (x) 2m We now average over disorder: Zn = = a=1 ⌈V e− n R dd xV 2 (x)/2ni v2 Z n [V ] rep −SE. so that this can be re-written † rep SE+ω/2. it is a function of only two.15) We introduce a capital Latin index B = R.E ′ † † DψaR (x) DψaR (x) DψaA (x) DψaA (x) e (24. Here. equivalently. but had to sum an inﬁnite set of diagrams in order to obtain physically-reasonable results. ψaA .E−ω/2 [ψaB .454 where CHAPTER 24. If we were to compute perturbatively with this eﬀective action. not three energies. A. Indeed. the ﬁnal term. ψaA ] = † dd x ψaR E + iδ + 1 2 ∇ ψaR 2m † + ψaA E ′ − iδ + 1 2 ∇ ψaA 2m † † ψbR ψbR + ψbA ψbA † † + ni v 2 ψaR ψaR + ψaA ψaA (24.E ′ [ψaR . It is no-local in time or. Thus. −1). is a strongly-relevant perturbation. we should not expect perturbation theory to be valid for this term. we will deal with this relevant perturbation by decoupling it with a Hubbard-Stratonovich transformation. we did not us low-order perturbation theory in the previous chapter.14) with rep † † SE.16) † † + ni v 2 ψbB ψbB ψcC ψcC where ΛBC = diag(1.E ′ = dd x † ψaR E + iδ + 1 2 † ∇ ψaR + V (x)ψaR (x)ψaR (x) 2m (24. .

is non-local in time (since it is a function of two energies.17) We introduce a Hubbard-Statonovich ﬁeld Qbc. In so doing. NON-INTERACTING ELECTRONS 455 There are two choices of Hubbard-Stratonovich decomposition.17) resulted in the i in this equation. have non-trivial structure in replica and advanced/retarded indices.21) 4τ DQ e−Seﬀ [Q. The latter. Thus.16) as: † † † † ψbB ψbB ψcC ψcC = −ψbB ψcC ψcC ψbB (24. ψaB . we re-write the ﬁnal term in (24.BC e−SHS [Q. it is just a shift of the chemical potential. The former is just the density of electrons at energy E + ω/2 plus the density of electrons at energy E − ω/2. ψaB ] = † dd x ψaB δBC ω 1 2 ∇ + ΛBC + iδ ψaC 2m 2 πNF i † Qbc. however.BC by rewriting the partition function as: n Zn where = a=1 † DψaR (x) DψaR (x)DQbc. Integrating out the electrons.BC Qcb.ψ] (24.24. ] is given by Seﬀ [Q] = dd x −tr ln δBC E+ i ω 1 2 ∇ + ΛBC + iδ + QBC 2m 2 2τ πNF + tr Q2 (24. we decouple the quartic term in this channel.20) .) We can either in† troduce a Hubbard-Stratonovich ﬁeld which is essentially equal to ψbB ψbB † or one which is equal to ψbB ψcC . E ± ω/2) and can.3. (We will make the simpliﬁcation of ignoring spin in this section. we will integrate out the electrons and look for a saddle-point for Q.CB + Qbc. To proceed. If it develops an expectation value.19) Note that the minus sign in equation (24.BC ψbB ψcC + 2τ 4τ E+ (24. we ﬁnd: Z= where Seﬀ [Q.ψ † .18) † SHS [Qbc. in principle.] (24.BC .

we have: Seﬀ [Q] = dd x − πNF 1 tr δQ2 tr (GδQ GδQ) + 8τ 2 4τ (24. Consider the .BC = ΛBC δbc (24. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION If we assume a translationally-invariant saddle-point.456 CHAPTER 24. the saddle-point has self-consistently computed corrections to the propagator and summed them to obtain the self-energy correction.22) πNF Q = i 2 1 i ˆ (2π) iǫˆ + 2m ∇2 + µ + 2τ Q n It has solution Qbc.26) The linear terms in δQ vanish by the saddle-point condition. We can now obtain the spectrum of ﬂuctuations about the saddle-point by expanding (24.25) 2τ 4τ E+ where GRR = GR E + ω .24) Thus. 2 2 Expanding the logarithm to second order.A (k.23) At this saddle-point. the saddle-point equation is: 1 d2 p ˆ (24.21) about Q = Λ: Seﬀ [Q] = dd x −tr ln δBC ω i 1 2 i ∇ + ΛBC + + δQBC 2m 2 2τ 2τ πNF πNF + tr (δQΛ) + tr δQ2 2τ 4τ i = dd x −tr ln GBC −1 + + δQBC 2τ πNF πNF + tr (δQΛ) + tr δQ2 (24. GAA = GA E − ω . and GRA = GAR = 0. the electron Green functions are of the expected form: GB=R. E) = = 1 E− E− k2 2m + ± 1 i 2τ QBB i 2τ k2 2m (24.

we see that any transformation of the form ψbB → Ubc.CB Ubd. Suppose we would like to compute Seﬀ [Q] to higher orders. We could continue to expand in powers of δQ. In † fact.EC . QAA are non-zero at the saddle-point. we computed the p integrals in imaginary time. these will constrain the form of Seﬀ [Q] completely. QRR and QAA are massive ﬁelds.28) 2 and similarly for the other two integrals. we have dd p GR (p + q. one might say that the longitudinal ﬁelds are massive. E + (2π)d dd p GR (p + q. NON-INTERACTING ELECTRONS ﬁrst term in equation (24.BC ψcC will leave (24. since they . Continuing to real-time. E + 2 ω ) = 2πNF τ 1 − iωτ − Dτ q 2 2 ω )=0 (24.27) In the previous chapter.BD = δcd δCD . QRA . Conﬁgurations in which Q varies slowly within this manifold must be gapless.BD Qde.16) for ω.26): tr (GδQ GδQ) = dd p dd q (GR (p + q)δQRA (q)GA (p)δQAR (−q) (2π)d (2π)d + GA (p + q)δQAR (q)GR (p)δQRA (−q) + GR (p + q)δQRR (q)GR (p)δQRR (−q) 457 + GA (p + q)δQAA (q)GA (p)δQAA (−q)) (24. E + (2π)d ω )GA (p.23) is not the only saddle-point of the action.BC → Ubd. Inspecting (24. As we will see momentarily.BD ΛDE Udc.16) invari† ant so long as Ucb. (24. this interpretation is correct. the entire manifold Ubd. QAR are massless. A much simpler way is to identify the symmetries of the action. Under such a transformation. as in a non-linear σ-model. † Qbc. the eﬀective action Seﬀ [Q] must vanish for any Q of this form. while the transverse ﬁelds. but this is tedious.3. Since such a transformation must leave the action invariant. Since QRR .EC extremizes the action. QAR are massless for ω = 0.24. Thus. while QRA .DE Uec. Thus.29) To this order. E − 2 ω )GR (p. we have to second order in δQ: Seﬀ [Q] = πNF 4 dd x Dtr(∇δQRA )2 + Dtr(∇δQAR )2 1 2 2 + iω tr δQ2 + δQ2 RA AR + tr δQRR + δQAA τ (24.

but also † † those which mix ψbB with ψcC so long as they leave ψcC ψcC invariant (and. Meanwhile. Small but non-zero ω + iδ gives these excitations a small gap because it favors Q = Λ.31) (24.33) or. we can rewrite † † ψbB ψbB ψcC ψcC = 1 1 χbBi Jij χbBj χcCk Jkl χcCl = − χbBi Jij χcCl χcCk Jkl χbBj 4 4 (24. The above paragraph is almost correct.CB = 2τ χ JQχ + 4τ tr Q 2τ 4τ (24.CD. all transformations χbBi → Mbc. simply.35) and decouple this term with Q so that it becomes πNF ij i T πNF i jl 2 Qbc.BD are not the only symmetries of the action (24. therefore.BC. One gap in these arguments is that the unitary transformations Ubd.30) 1 χbBi Jij χbBj 2 where Jij = ǫij .BC.16) for ω + iδ = 0 so long as T Mbc. These transformations can be parametrized by deﬁning χbB1 χbB2 Then we can write † ψbB ψbB = = ψbB † ψbB (24.BC χbBi Jij χcCl + Qbc. we should not only allow transformations which mix the diﬀerent ψbB s. the equations satisﬁed by ψ are the same as those satisﬁed by ψ † . Sp(2n) does not act in a simple way on Qbc. Equation (24.ij Jjk Mcd.34) is the deﬁning equation of the symplectic group Sp(2n). conﬁgurations in which Q moves out of this manifold should cost energy.BC Qji cb. M T JM = J (24. this is why we found that δQRR . δQAA are massive while δQRA .kl = Jil δbd δBD (24.16) invariant for ω + iδ = 0).36) . As a result of time-reversal symmetry. Hence.458 CHAPTER 24. δQAR are massless.32) (24. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION have only gradient energy.ij χcCj are symmetries of (24. However.16). In other words.34) where J ≡ Jil δbd δBD . leave (24.BC as we deﬁned it above.

24. we have used (24.34).E−ω/2 [ψaB . Hence.3. Q must transform so that these terms remain invariant.36) transforms as χT JQχ → χT M T JQ′ M χ = χT JM −1 Q′ M χ Q′ = M QM −1 which also leaves the second term in (24. then transformations which mix ψ and ψ † are no longer symmetries of the action. Any M which doesn’t mix retarded ﬁelds with advanced ones commutes with Λ. . In particular. the saddle-point manifold is given by Q = M Λ M −1 (24. ψaB ] = † dd x ψaB δBC E+ 1 (∇ − iA)2 2m + ΛBC ω + iδ 2 ψaC (24. Hence.CD ψdD because the ﬁrst derivative term A · ∇ + ∇ · A is not invariant under the the rest of the Sp(2n) group. However. The symmetries of the action † rep SE+ω/2. Therefore the corresponding subset of transformations Sp(n)×Sp(n) ⊂ Sp(2n) don’t lead to new conﬁgurations. we have a ﬁeld which takes values in a diﬀerent coset.38) This manifold is parametrized by Sp(2n) matrices M .39) (24. The eﬀective action Seﬀ [Q] describes a non-linear σ-model. NON-INTERACTING ELECTRONS 459 With this deﬁnition. For such Q. . (24.37) In the ﬁrst equality. refers to higher-derivative terms.36) invariant. In other words. namely Sp(2n)/Sp(n) × Sp(n). which is equivalent to the condition QRA = QAR = 0 which we had in the secondorder expansion of Seﬀ [Q]. This will remain invariant if (24. .40) where the . . the action is simply: Seﬀ [Q] = πNF 4 dd x Dtr(∇Q)2 + (iω − δ) tr(ΛQ) + . If we consider a system of electrons in a weak magnetic ﬁeld. The main complication is that instead of a ﬁeld which takes values on the sphere S 2 = O(3)/O(2) (as in an antiferromagnet). the ﬁrst term in (24. not every diﬀerent M leads to a diﬀerent Q. The second term has been written in the above manner because non-zero ω + iδ favors Q = Λ. . Q ∈ Sp(2n)/Sp(n)×Sp(n).41) † † + ni v 2 ψbB ψbB ψcC ψcC are U (2n) symmetries ψcC → Ucd.

in the latter case. and we would have found NLσMs for the cosets O(n. While . The reason for this confusing terminology is that the method which we have used in this section is called ‘fermionic replicas’ because we could have used bosonic ﬁelds to construct the functional integral (24.2 Interpretation of the σ-model. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION a system of electrons in a weak magnetic ﬁeld (‘weak’ means too weak for the quantum Hall eﬀect to occur) will also be described by a NLσM of the form (24. In the latter case (24. and symplectic cases. it is the metalinsulator transition. n)/O(n) × O(n). there is a disordered phase in which the symmetry is restored and the order parameter vanishes. Meanwhile.40). have symplectic symmetry group Sp(2n) and said to be in the ‘orthogonal ensemble’. etc. However the β-function for O(n. e the similarity ends here. Finally. 24. this is the metallic phase. For non-interacting electrons. Seﬀ [n] = πNF 4 dd x ρs tr(∇n)2 + H · n 2 (24. n)/Sp(n) × Sp(n) for the orthogonal. the symmetries of the bosonic action are diﬀerent. However. n)/O(n) × O(n) is the same as the β-function for Sp(2n)/Sp(n) × Sp(n). n)/U (n) × U (n). our NLσM Seﬀ [Q] = πNF 4 dd x Dtr(∇Q)2 + (iω − δ) tr(ΛQ) (24. it is the Curie or N´el point. there is a Goldstone phase.3.43).42) looks very similar to the NLσM which describes an classical ferromagnet or antiferromagnet near its lower critical dimension. with diﬀusion propagators (−iω + Dq 2 )−1 . so the same physics is obtained. in the latter case. Sp(n. In the former case.460 CHAPTER 24.43) In both cases. d = 2. U (n. Systems in this universality class – characterized by the absence of time-reversal symmetry – are said to be in the ‘unitary ensemble’ because of the unitary symmetry group U (2n). Analogies with Classical Critical Phenomena Superﬁcially.) and are said to be in the ‘symplectic ensemble’. systems with spin-orbit scattering have symmetry group O(2n) (see ref. with spin-wave propagators (H + ρs q 2 )−1 . but with Q taking values in U (2n)/U (n) × U (n). there is a critical point in d = 2 + ǫ which is near the Goldstone phase’s stable ﬁxed point. as we discussed above.10). However. thee two representations are equivalent. this is the magnetically-ordered phase. systems with time-reversal invariance. In both cases. unitary. In the former case. respectively.

QRA (x)QAR (y) = |G(x. therefore. Where the El are the energies of localized states. the density-of-states is non-zero in the insulating state as well and varies smoothly through the transition.3. who pointed this out. Hence. Hence.46) while the isolated poles due to localized states cannot be avoided.24.48) where L is the system size. NON-INTERACTING ELECTRONS 461 the same is true for any ﬁnite n in (24. then |G(x.44) However. GB=R. n (24. E + iδ) = En − E − iδ n and the fact that the cuts due to a continuum of extended states can be deformed away from the integration contour in (24. the spectral function and.42). the density-of-states would vanish. If E is in the regime of localized states. this does not quite happen in the replica limit. if they were precisely equal. This follows from ∗ ψn (x)ψn (y) (24. so ρ(E) ∼ 1 tr(ΛQ)δ→0+ ∼ |E − EC |β ∼ const.45) Thus. the correlation function is divergent . these Green functions would be precisely the same. n → 0. E) = 1 E− k2 2m + i 2τ QBB (24. E + iδ)|2 → π δ δ(E − El ) |ψl (x)|2 |ψl (y)|2 (24. in contrast to the magnetization at a magnetic transition. y.46) l for δ → 0. the density-of-states plays the role of an order parameter. the symmetry is always broken. However. explained this as a special feature of disordered systems. β = 0. If Q were to vanish and we were to take iδ = 0.A (k. the critical exponent for the order parameter vanishes. McKane and Stone. Therefore. y. The basic symmetry (replicated and enmeshed with particle-hole symmetry) is between advanced and retarded Green functions. This factor comes from the spatial density of localized states l in (24. As we will show in the next subsection.47) G(x. y. Diﬀusion modes are slowly-varying oscillations of this order parameter which remain within the low-energy manifold Q = M ΛM −1 . E + iδ)|2 ∼ 1 −2|x−y|/ξ e δ (24.46).

462 CHAPTER 24. The symmetry is broken in both the metallic and insulating phases. we see that it is divergent for all p: 1 (24. In the latter case. but not the restoration of symmetry or the dynamical generation of a mass. there can be no mass gap because the order parameter is susceptible to inﬁnitesimal perturbations. When a symmetry is broken.3 RG Equations for the NLσM We will now compute the one-loop RG equations for the NLσMs introduced in this section. This situation should be viewed as a generalization of Goldstone’s theorem.51) Q= 1/2 † V − 1 − V †V . there are no long-ranged correlations or long-wavelength excitations.50) Comparing these two expressions. if E is in the regime of extended states. There.49) QRA (p)QAR (−p) ∼ g(p) δ where g(p) is a smooth function of p which is ﬁnite at p = 0. the diﬀerence is that (a) occurs in the former while (b) occurs in the latter. the correlation function in the disordered state is of the form n(p)n(−p) ∼ 1/(Dp2 + m2 ). On the other hand. thereby recovering the Gang of Four’s scaling equation. Thus. There are two ways in which this can happen: (a) the conventional way – the gapless states are Goldstone modes – or (b) they can be localized excitations. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION for arbitrary separation as δ → 0. However. then we have diﬀusion: QRA (p)QAR (−p) ∼ Dp2 1 + 2δ (24. a mass term is generated and the spin stiﬀness D (the analog of the diﬀusion constant) is irrelevant compared to it.3. but divergent at all p in the insulating phase. Compare this to the O(3) NLσM which we studied in the context of antiferromagnetism. but neither is there a gap – the density of states at the chemical potential is ﬁnite. so that inﬁnitesimal δ leads to a ﬁnite diﬀerence between then. Therefore. Fourier transforming. the basic symmetry between advanced and retarded Green functions – which diﬀer only by the iδ in their deﬁnition – is broken. 24. One simple one is: 1/2 1−VV† V (24. the localized phase has D → 0. it is divergent only at p = 0 in the metallic phase. In order to do so. the QRA (p)QAR (−p) correlation function is divergent in both phases. we need a convenient parametrization for Q.

and we must go to two loops. In the absence of a clear symmetry justiﬁcation. Instead of Sp(2n)/Sp(n) × Sp(n). We can no longer consider individual frequencies in isolation. For n → 0. there is no completely controlled route to an eﬀective theory for diﬀusion modes. the one-loop β-function vanishes. Such a theory must be some form of NLσM in limit that the interaction strength vanishes. it is suﬃcient to expand this action to quartic order and.4 Interacting Electrons We now turn to the problem of interacting electrons. therefore. For U (2n)/U (n) × U (n). Let t = 1/D and let G be the symmetry group of the NLσM. Hence. (24. the elements of V are real numbers and † means the transpose. the elements of V are quaternions and † means take the transpose and perform the quaternion conjugate. this give us β(t) = ǫt − t2 + . . INTERACTING ELECTRONS 463 where V is an unconstrained n × n matrix. βU (n) (t) = ǫt − nt2 + . Then β(t) = ǫt − Ct2 + . the elements of V are complex numbers and † means the adjoint. . and we would obtain the same result. . The result is simple to state. while in the presence of a weak magnetic ﬁeld. For O(2n)/O(n) × O(n). . . In particular βSp(2n) (t) = ǫt − (2n + 1)t2 + . 24. which is equivalent to the β-function which we found in the previous chapter. In order to compute the one-loop β-function. suﬃcient to expand Q to quadratic order in V . (24.24. the β-function has sign reversed. we could just as easily take Sp(2n)/Sp(p) × Sp(2n − p). As a result. .52) where C is the Casimir (in the adjoint representation) of the symmetry group G. For Sp(2n)/Sp(n) × Sp(n). .4. βO(n) (t) = ǫt − (n − 2)t2 + . we must rely on physical arguments: clearly the low-energy degrees of freedom should be . . In the presence of spin-orbit coupling. there is no longer an exact symmetry of the theory which we can exploit.54) for electrons in zero ﬁeld. . (24. .53) Note that the β-function does not depend on the subgroup which is being modded out.

(p.α (p)ψl. then the associated bare couping constants are Γ1 = V (0). The imaginary-time action is S = S0 + Sdisorder + Sint where S0 + Sdisorder = T n † d2 x ψn. p′ ≈ p and exchange scattering. these are subject to ﬁnite Fermi liquid renormalization in the ballistic regime. where k′ ≈ k. even before they begin to ﬂow in the diﬀusive regime. in which electrons with momenta and spins (k.55) 1 2 ∇ + µ + V (x) ψn.α (k)ψl. k3 ≈ −k4 . It turns out that the most straightforward way to do this is to construct a ﬁeld theory with inﬁnitelymany terms in its Lagrangian. p′ ≈ k.α (x) (24. we can only hope that the relationship between these terms is preserved. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION diﬀusion modes (at least for weak interactions). is completely non-local in time) – and also terms which are local in time. Furthermore. α). this.56) 2m and Sint = T n. we have written the second term in momentum space and have split the interaction into nearly forward scattering. Γ2 = (dθ/2π) V (2kF cos θ/2). β) are scattered to (k′ .57) For later convenience. At one-loop order this holds. in which k′ ≈ p. Unlike in the non-interacting case.β (p) † † + Γ2 ψn. β). For simplicity. so that we can ignore Cooper scattering. we consider the unitary ensemble. At one-loop. (p′ . therefore.α (k + q)ψm. too. the theory contains terms which are completely non-local in time – such as the non-interacting electron NLσM (which is at a ﬁxed energy and. does not occur. α).β (k)] (24. . terms which are only somewhat non-local in time could be generated.α (k + q)ψm.m. for which k1 ≈ −k2 .l † † [Γ1 ψn.β (p − q)ψl+m−n. In principle.β (p − q)ψl+m−n. We begin with a system of interacting electrons moving in a quenched random potential V (x) in 2D.α (x) iǫn + (24. If the microscopic electron-electron interaction is V (q). all of whose coeﬃcients are related.464 CHAPTER 24. We assume that the system is in a weak magnetic ﬁeld. As Finkelstein noted. so we should write down an eﬀective ﬁeld theory which captures their physics.

60) 2 2 In this expression. because interactions do not mix diﬀerent replicas. N is a replica index.4.α (k + q)ψm. we replicate the theory and then average the replicated functional integral over 1 the probability distribution V (x)V (x′ ) = 2πNF τ δ(d) (x − x′ ). . .58) where a = 1. Note that Xαβ σβα and Y are essentially the spin and charge density. Matsubara indices are treated as ordinary matrix indices except that their summations come with factors of T . Integrating out the fermions. we have S rep = T n.β (p − q)ψl+m−n. Y ] = + [−tr ln iǫn + πNF tr Q2 4τ i 1 2 ∇ + µ + Q + iΓ1 Y + Γ2 X 2m 2τ 1 1 + Γ1 tr Y 2 + Γ2 tr X 2 ] (24. so they are local in time.24. NF is the single-particle density-of-states at the Fermi energy.α ψn. we use ‘tr’ to mean the trace over all of the indices which are not explicitly written.Y ] (24.α (x)+ 2m 4πNF τ a† a ψn. . We decouple these quartic terms with three Hubbard-Stratonovich ﬁelds.α 2 ] T n.α (p)ψl.59) where Seﬀ [Q. we must take the replica limit. we can write the partition function as Z= DQ DX DY e−Seﬀ [Q.αβ = Xαβ (n−m). It will be useful to think of X. Ynm = Y a (n−m). X.α (x) iǫn + 1 1 2 ∇ + µ ψn. The interaction terms depend on three Matsubara frequencies.β (p − q)ψl+m−n.l.m. a a a ab Qnm. INTERACTING ELECTRONS 465 In order to perform the average over the quenched random potential. X.α (k)ψl.αβ . eventually.β (k)] (24. N → 0. Xnm. they have only a single replica index. . Now.α (k + q)ψm.a a† a† a a [Γ1 ψn.X. and they do not couple diﬀerent replicas. .a † d2 x [ψn. however. Note that the disorder term is non-local in time since it only depends on two Matsubara frequencies. m and also as functions of a single frequency ωn−m.β (p) a† a† a a + Γ2 ψn. and it couples diﬀerent replicas. 2. The resulting eﬀective action now contains four quartic terms: one disorder term and two interaction terms. Y ] is given by Seﬀ [Q. Y both as matrices with indices n.

As we will see later. they have the solution n 1 2 2m ∇ Y (m) = iT +µ+ i ˆ 2τ Qn+m. Hence. These ﬂuctuations correspond to diﬀusion of charge and spin. m < 0 while the upper right block corresponds to n < 0. to n > 0.63) vanish for m = 0 because the saddle-point Q is diagonal in Matusubara frequency while they vanish for m = 0 because of cancellation between positive and negative frequencies. according to (24.n + 1 2 2m ∇ +µ+ 1 i ˆ 2τ Q ˆ ˆ + iΓ1 Y + Γ2 X (24. Y are gapped at the classical level (i.63) The momentum integrals on the right-hand-sides of (24. which can be parametrized by Vnm. tree-level).63) are simply iπNF Qn+m.62) and (24. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION The saddle-point equations are ˆ πNF Q = i d2 p (2π)2 iǫˆ + n 1 1 2 2m ∇ +µ+ i ˆ 2τ Q + iΓ1 Y + Γ2 X 1 (24.62). This is the Stoner instability.61) X(m) = T n d2 p (2π)2 iǫn δn+m. m > 0.αβ : 1−VV† † 1/2 Q= V − 1 − V †V 1/2 V (24. In order to derive an eﬀective ﬁeld theory for these diﬀusion modes.e.61).62) ˆ ˆ + iΓ1 Y + Γ2 X (24.n + d2 p (2π)2 iǫn δn+m.n . the lower left block. X = 0 becomes unstable and there is an X = 0 solution of (24. for Γ2 sufﬁciently large.m ab Qab mn.64) Note that the right-hand sides of the second and third equations of (24.62) and (24.65) The diagonal blocks of Qnm correspond to Matsubara indices n > 0. we . m < 0.466 CHAPTER 24. m > 0 and n < 0.αβ = sgn (ǫn ) δmn δαβ δ a Xαβ (m) = 0 Y a (m) = 0 (24. The low-energy ﬂuctuations about this saddle-point are transverse ﬂucab tuations of Q. The longitudinal ﬂuctuations of Q as well as ﬂuctuations of X.

Thus. the factors of τ (T m ) in the ﬁnal two terms are inﬁnite as a result of the unrestricted Matsubara sum. Γ2 . we integrate out X. .) about the diﬀusive saddle-point.) about the diﬀusive saddle-point and keep only terms up to second order in X and Y . . INTERACTING ELECTRONS shift Q → Q − 2τ Γ1 Y + 2τ iΓ2 X in (24. Finkelstein actually expanded the tr ln(. Following Finkelstein. However.66). dd x {D tr(∇Q)2 − 4iZtr (ˆQ) ǫ − Γ1 tr (QY ) + iΓ2 tr (QX) Γ1 1 tr Y 2 + (1 + NF Γ1 ) 2 πNF Γ2 1 tr X 2 } (24. X. .60) Seﬀ [Q. For small Γ1 . . upon integrating out the massive longitudinal modes of Q. which is simply 1/2π.65). We would then obtain the same expressions. Alternatively.4. We now expand the tr ln(. = τ T m / (1 + 2πτ T m ). The coeﬃcient of X 2 is now 1 − NF Γ2 . but the Stoner instabilty is missed. Y and obtain an eﬀective ac- Seﬀ [Q] = πNF .67) + (1 − NF Γ2 ) 2 πNF 2 The diﬀusion constant D is given by D = vF τ /2. we see that they are the ﬁrst terms in the series τ T m − 2π (τ T m )2 + . we replace the factors of τ (T m ) by 1/2π in (24. we require that Q is a constrained ﬁeld of the form given in (24. Thus. .24.66) Formally. . we could expand the tr ln(. namely (24. Y ] = + [−tr ln iǫn + i 1 2 ∇ +µ+ Q 2m 2τ 467 πNF tr Q2 − πNF Γ1 tr (QY ) + iπNF Γ2 tr (QX) 4τ 1 1 + 2π τ T NF Γ1 Γ1 tr Y 2 + 2 m + 1 2 1 − 2π τ T NF Γ2 m Γ2 tr X 2 ] (24. there is no diﬀerence. which becomes negative for Γ2 > 1/NF : the Stoner instability.66) with the factors of τ (T m ) replaced by 1/2π.) only to linear order in X and Y . . .

the coeﬃcient of the tr (ˆQ) term is 1. Seﬀ [Q] = πNF dd x {D tr(∇Q)2 − 4itr (ˆQ) ǫ ˜ n − πNF Γ1 Qaa n2 .5 24.2 ).2 small.2 = Γ1.468 CHAPTER 24. For Γ1.αα Qaan4 . as in (24.67). Γ1. These corrections to Γ1.2 = Γ1.2 follow from our retention of the X 2 and Y 2 terms which Finkel˜ stein drops. so we have followed Finkelstein in introducing the coupling Z. Initially. In this expression.2 /(1 ± NF Γ1. Γ1.ββ δn1 −n2 +n3 −n4 n3 1 ˜ 2 Qaa δn +n + πNF Γ Qaa n1 n2 αβ n3 n4 βα 1 2 −n3 −n4 } (24.2 . The Γ1. 24. but quanǫ tum corrections cause it to ﬂow. local ˜ in time. NON-LINEAR σ-MODELS FOR DIFFUSING ELECTRONS AND ANDERSON LOCALIZATION tion for the diﬀusion modes. and Finkelstein’s eﬀective action is recovered.2 interaction terms do not involve matrix multiplication and are. consequently.6 The Metal-Insulator Transition Mesoscopic ﬂuctuations*** .68) Note that ordinary matrix multiplication rules reﬂect the non-locality in ˜ time of the ﬁrst term (the ‘disorder term’) in this action.

- Quantum Field Theory of Many-body Systems-Wen
- Tinkham M. Introduction to Superconductivity
- Sakurai - Modern Quantum Mechanics
- Quantum Condensed Matter Physics-Lecture Notes
- braket
- Quantum Mechanics Review
- SE102_13
- Load Dependent Ritz Vector Algorithm and Error Ananlysis
- Standford Intro, Ab Math
- Appliid Math Text
- How Sss b Possible
- Paul Turner- Five Lectures on Khovanov Homology
- Untitled
- g Exp H Semi-Riemann Metric on Tangent Bundle and Its Index
- Vector and Matrix norm
- control(9).pdf
- MTH 210 Syllabus Fall 2011
- Session01 A
- mfk
- Manifolds Notes
- 1306.1157
- Fuzzy Subspace, Balanced Fuzzy Set and Absorbing Fuzzy Set in Fuzzy Vector Space
- mbk-76-prev
- SzeghyPeriodica
- Katzgraber - Topology in Physics
- mathgen-1764886742
- Linear Guest
- 1-s2.0-S0196885814000220-main
- 2.I. DefinitionOfVectorSpace
- Beltrami Short
- many_body

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd