Statistical Field Theory

Description

Statistical Field theoryThierry GiamarchiNovember 8, 2003Contents1 Introduction 52 Equilibrium statistical mechanics 72.1 Basics; correlation functions . . ............................... 72.2 Externalfield;linearresponse. 82.3 Continuous systems: functional integral . . . ....................... 112.4 Examples:elasticsystems.................................. 173 Calculation of Functional Integrals 293.1 Perturbationtheory ..................................... 293.2 Variationalmethod...................................... 313.3 Saddlepointmethod. 324 Quantum problems; functional integral 354.1 Singleparticle......................................... 354.2 Correlation functions; Matsubara frequencies ....................... 394.3 Many degrees of freedom: example of the elastic system . . ............... 424.4 Linkwithclasicalproblems................................. 445 Perturbation; Linear response 475.1 Method............................................ 475.2 Analyticalcontinuation ................................... 505.3 Fluctuationdissipationtheorem............................... 525.4 Kuboformula......................................... 535.5 Scattering, Memory function . 576 Examples 596.1 Compressibility . ....................................... 596.2 Conductivity of quantum crystals; Hall effect ....................... 616.3 Commensuratesystems ................................... 667 Out of equilibrium classical systems 697.1 ...

Informations

• Publié par: Chion
• Nombre de lectures: 34
• Langue: English

Extrait

Statistical

Thierry

Field

theory

Giamarchi

November

8,

2003

Contents

1 Introduction 2 Equilibrium statistical mechanics 2.1 Basics; correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 External field; linear response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Continuous systems: functional integral . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Examples: elastic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Calculation of Functional Integrals 3.1 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Variational method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Saddle point method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Quantum problems; functional integral 4.1 Single particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Correlation functions; Matsubara frequencies . . . . . . . . . . . . . . . . . . . . . . . 4.3 Many degrees of freedom: example of the elastic system . . . . . . . . . . . . . . . . . 4.4 Link with classical problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Perturbation; Linear response 5.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Analytical continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Fluctuation dissipation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Kubo formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Scattering, Memory function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Examples 6.1 Compressibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Conductivity of quantum crystals; Hall effect . . . . . . . . . . . . . . . . . . . . . . . 6.3 Commensurate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Out of equilibrium classical systems 7.1 Overdamped dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Link with thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 MSR Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

5 7 7 8 11 17 29 29 31 32 35 35 39 42 44 47 47 50 52 53 57 59 59 61 66 69 69 70 72

Chapter 1

Introduction

The material discussed in these notes can also be found in more specialized books. Among them

•Statistical mechanics and functional integrals [1, 2, 3, 4] •Many body physics [5, 3]

These notes are still in a larval stage. This is specially true for chapters 3 and 7. Despite the at- tempts to unify the notations and eliminates mistakes, factors of 2 and other errors of signs, it is clear that many of those remain. I’ll try to update the notes from time to time. If you find these notes useful and feel like helping to improve them, do not hesitate to email to me ([e-mail]) if you have found errors in those notes or simply to let me know if there are parts that you would like to see improved or expanded.

5

Chapter 2

Equilibrium statistical mechanics

2.1 Basics; correlation functions In statistical mechanics one is interested in getting basic quantities such as the partition function. Space is meshed with a network on which some variable describe the state of the system. To be more specific let us consider an Ising model, on a square lattice. On each siteiexists a spinσ i =±. The state of the system is describe by all spin variablesσ 1 , σ 2 , . . . σ N , whereNgrows as the volume of the system. The set of all these variables is a configuration that we denote in the following as{σ}. To compute physical quantities one needs in addition an Hamiltonian describing the energy of the system for a given configuration. To be specific let us again consider a Ising type Hamiltonian H 0 =−  J i,j σ i σ j (2.1) i,j A physical quantity such as the partition function results from the sum over all configurations of the system Z=  e −βH 0 =   · · ·  e −βH 0 [{σ}] (2.2) {σ} σ 1 =±σ 1 =±σ N =± The difficulty if of course to make the sum over the humongous (thermodynamics) number of variables. FromZorF=−Tlog(Z) many simple physical quantities can be directly extracted (specific heat, etc.). In addition to these rather global quantities, one is often interested in correlation functions, mea- suring thermodynamics averages of local observables or products of local observables. Some examples are

(2.3) (2.4)

σ i1  σ i1 σ i2  where the thermodynamics averagestands for O O[α, . . . , β]=  {σ} [  σ {σ α , } e .. − . , β σ H 0 β ]e −βH 0 (2.5) WhereO Noteis any operator. that the correlation functions depends on the spatial positionsα, . . . , β, the average having been made over the microscopic variablesσ. 7

In fact the correlation functions can also be obtained from a partition function. If one adds to the Hamiltonian a source term −βH s =  h i O i (2.6) i one introduces the partition function, that now depends on a thermodynamics number of external sources Z[h 1 , h 2 , . . . , h N ] =  e −β[H 0 +H s ] (2.7) {σ} and the associated free energyF[h 1 , h 2 , . . . , h N ]. It is then easy to see that [O α . . . O β ]= Z[h 1 =]0 ∂Z[ ∂h h 1 , α h . 2 h , ∂ .,.. .. β h N ]  {h}=0 (2.8) Of course this is a rather formal relation since computingZ[h 1 , h 2 , . . . , h N ] is in general a formidable task. Similar formulas can be derived from the free energy. In particular one can obtain from the free energy O α =− β ∂F[h 1 ∂ ,h 2 h , α . . . , h N ] 9.)2(  {h}=0 It is easy to see that differentiating the free energy leads to the so-called connected correlations − β ∂F[h 1 h∂ ,h 2 h∂ .., β . , h N ]  {h}= = [O α O β  − O α O β ] (2.10) α 0 Each time one differentiate an average one gets two terms one coming from the numerator one from the denominator. Thus ∂ h γ . . .= [. . . O γ  − . . .O γ ] (2.11) It is thus easy to obtain the higher derivatives. The ingredients that we have illustrated on this example of the Ising model are totally general. It is important to understand that what is generally needed is to be able to perform the sums over the microsocpic degrees of freedom. 2.2 External field; linear response Among the various correlation functions some are of special importance. Let us assume that a physical variable of the system can be coupled to an external field, in a way similar to (2.6). In our Ising example this is the case if one puts the system in a magnetic field. The natural variable associated with such a perturbation is the magnetization of the system. One would have two way to compute it. Adding the magnetic field to the HamiltonianH 0 leads to H h =−h  σ i (2.12) i FromH=H 0 +H h one can compute the free energyF[h]. Standard thermodynamics tells us that the total magnetizationMof the system is simply given by M=− d hd F[h 3).12( ]

This relation comes immediately from (2.8) usingO=−  i σ i .Mdescribes the response of the system to the external fieldh for this simple case (. Evenhis space independent) is is quite complicated to get the magnetization. To generalize this notion of response, let us consider an operatorO i (that can a priori depend on the position) that couples to an external fieldh i (that can also depend on space) trough the Hamiltonian H p =−  h i O i (2.14) i Let us furthermore assume thatO i is such that in the absence of perturbationH p one hasO i  0 = 0, where 0 denotes averages performed withH 0  it is not the case one can simply subtract thealone. If average value fromO i . In that case one can measure the response of the system, that is given by O i [{h}] =O i (2.15) The notationO i [{h}] means that the valueO i at pointidepends in fact on the value of the external field at all points, i.e. of the whole set of values ofh even a simpleat each spatial point. ForH 0 it is hopeless to expect to compute the full responseO i . A simpler case, but of considerable interest, is to compute thelinear Indeedresponse of the system. if the external field is small one can expand (2.15), in powers of the external field. The lowest order term is linear in the external field (hence the term linear response). One has O i [{h}] = 0 +a d  χ ij h j (2.16) j which is the most general linear term that one can write. The factora d is here for convenience (see below). At that stage this merely definesχ. If we assume in addition thatH 0 is invariant by translation, thenχ ij can only depend on the distance betweeniandj(that we denote symbolically byi−j). A more transparent  usway to write the convolution (2.16) is to go to Fourier space. Let recall the convention that we use for fourier transform on the lattice. h q =a d  −iq·r i h i (2.17) e i h i = 1 Ω  e iq·r i h q (2.18) q=2πn/L where we have introduced a lattice spacinga(acan be set to 1 at that stage since it is purely formal). The volume of the system is Ω =N a d and one introduces a distancer=ai sum over. Theqruns fromq=−π/atoq= +π/a, i.e. overNvalues ofn corresponds to the standard Brillouin zone.. This In that case  e iq·r i =N δ q,0 (2.19) i whereδ i,j is the discreteδ. For a continuous systemr i →r, one has the continuous Fourier transform − h(q) =  Ω iq·r h(r) (2.20) d d re h(r Ω ) 1 =  e iq·r h(q) (2.21) q=2πn/L

where Ω =L d is the volume of the system. The relation between the lattice variables and the continuous ones is thus h i →h(r) (2.22) h q →h(q) (2.23) For the continuous case  dre iq·r = Ωδ q,0 (2.24) still with a discreteδ the limit where the volume of the system goes to infinity, one can use. In 1  → 1 (2π) d  d d q(2.25) Ω q Ωδ q,0 →(2π) d δ(q) (2.26) whereδ(q) is the Diracδ. Using these definitions for the Fourier transform one can rewrite (2.16) as O q =a d N 1 i  j   e −iqr i +iq  r j χ i−j h q  (2.27) q usingr i =r j +r l one has O q =a d N 1    e i(q  −q)r j e −iqr l χ r l h q  (2.28) jl q =a d  e −iqr l χ r l h q (2.29) l =χ q h q (2.30) Thus in Fourier space the relation between the response and the external field is truly linear, each q the invariance by translation of the (i) is a direct consequence of: Thismode being independent. system which makes the Fourier base diagonal; (ii) the linear approximation for the response. The susceptibilityχ q (or in real spaceχ ij ) thus contains all the information we need to get the linear response of the system to any external perturbation. The good point is thatχcan be computed in the absence of the perturbationH p with the much simpler HamiltonianH 0 and thus we have better chance to obtain it. Let us get its general expression. To linear order in the perturbationH p we get from (2.15)  j O j h j )e −βH 0 O i [{h}]   {σ} {σ} O ( i 1+ +(1 β β  j O j h j )e −βH 0  O i  0 +  β[O i O j  0 − O i  0 O j  0 ]h j (2.31) j thus we can identify a d χ ij =β[O i O j  0 − O i  0 O j  0 ] (2.32) which is the so called connected correlation function, i.e. the correlation where the average values have been subtracted. Indeed O i O j  − O i O j =(O i − O i )(O j − O j )(2.33)

R 0 i

u i

Figure 2.1: Vibrations of a classical lattice. On each site one defines a displacementu i .

With our hypothesisO 0 function is equal to its connected part.= 0 the correlation Equ. (2.32) is a truly amazing equation. It shows that thelinear responseof the system is totally determined by thefluctuations Thisof the same system in the absence of the external perturbation. result is totally general and its only strong hypothesis is the fact that the system has reached thermal equilibrium. Since the imaginary part of the susceptibility of the system also determine dissipation that can take place in the system (we will come back to this point later), this result is also known as the fluctuation dissipation theorem. It has a crucial importance since it allows to compute properties of the system out of the unperturbed state alone. As an example, let us apply it to the relatively trivial case of an isolated Ising spin. In that case H 0  response to an external magnetic field (2.12) can be computed exactly for this trivial= 0. The case e βh − −βh m= e βh + e e −βh = tanh(βh) (2.34) leading to a susceptibilityχ=β Sincecan be readily obtained from (2.32). . ThisH 0 = 0 σσ 0 = 1 (2.35) Note that the uniform susceptibility (usually denoted simplyχ), i.e. response to a uniform the external fieldh i =his of course theq= 0 Fourier component of the susceptibility. is simply the This integrated correlation function χ=  βO i O j  0 (2.36) j Thus the longer range the correlations, the stronger the response of the system will be. This can be physically understood since a correlated region behaves as a gigantic spin that can respond coherently to the external field.

2.3 Continuous systems: functional integral So far we have delt with systems defined on a lattice. Although all systems in condensed matter exists on some kind of network, it is often very useful to be able to take a continuous limit, remembering the underlying lattice only as an ultraviolet cutoff. How to deal with such cases ? The first step is to have a variable that varies smoothly enough at the scale of the microscopic lattice so that a continuous variable can be defined. Let us take the very simple example of the vibrations of a classical lattice as shown on Fig. 2.1 On each site one defines a displacementu i . The energy is given by H=  k (u i,z 2 i −u i+z ) 2 (2.37)