LU TP 17-24 May 2017
Solitons in High Energy Physics
Edvin Olofsson
Department of Astronomy and Theoretical Physics, Lund University
Bachelor thesis supervised by Johan Bijnens Abstract
Solitons are solutions to classical field equations that are particle like, meaning they are localized in space, have definite energy and can be stable against deformations into the vacuum. In this thesis we investigate under which circumstances such solutions may exist and what effects their existence could have. It will turn out that topological considerations will play a large role in the question of existence and stability, and that results depend on the number of spatial dimensions in which the theory is defined. Mixed in with the more general results are several examples of solitons, like sine-Gordon kinks, vortices and magnetic monopoles, that serve the purpose of illustrating the ideas behind the arguments as well as showing what effects the existence of solitons may have. A large part of the thesis is dedicated to showing the equivalence of the sine-Gordon model and the massive Thirring model. The equivalence of the two theories will make it possible to connect the strong coupling regime of one theory to the weak coupling regime of the other, and vice versa. It will also be possible to connect the fermion number of the Thirring model to the topological charge of the sine-Gordon model. Popul¨arvetenskaplig beskrivning
Partikelfysik ¨arden del av fysiken som beskriver de allra minsta best˚andsdelarnai v˚ar v¨arld. De ekvationer som anv¨andsf¨oratt beskriva dessa sm˚apartiklar kan vara mycket sv˚araatt l¨osa,det finns faktiskt ett pris p˚aen miljon dollar till den som lyckas l¨osaett visst problem inom partikelfysik. F¨oratt underl¨attaalla ber¨akningar som ska utf¨oras, s˚ahar fysiker tagit fram n˚agraregler som man kan f¨olja. Till varje h¨andelsef¨orloppsom ska unders¨okas s˚aritar man upp diagram som ska representera vad som h¨ander.Fr˚anett s˚adant diagram g˚ardet sedan att r¨aknap˚asannolikheten f¨oratt just en s˚adan reaktion ska ske. Detta ¨arvad som kallas st¨orningsteori. Till en och samma reaktion s˚akan det finnas v¨aldigtm˚anga diagram att h˚allareda p˚a. Som tur ¨arbrukar det inte vara n¨odv¨andigt att bry sig om alla diagram, eftersom de allra flesta representerar h¨andelsersom ¨arv¨aldigt osannolika. Det ¨ard¨arf¨orofta okej att bara bry sig om de f˚adiagram som bidrar med mest sannolikhet. Med hj¨alpav dessa verktyg har fysiker idag lyckats beskriva det mesta av den materia som vi st¨oterp˚ah¨arp˚a jorden. Problemet med st¨orningsteori¨ardock att det bara fungerar om krafterna mellan partiklarna ¨ars˚asvaga att man kan anse att kraften endast utg¨oren liten st¨orningf¨or partiklarna. Det ¨ard¨arf¨orn¨odv¨andigtatt ibland f¨ors¨oka l¨osasina ekvationer utan att anv¨andast¨orningsteori,f¨oratt se om det dyker upp n˚agotsom man annars skulle missa. Det h¨arprojektet har unders¨okten viss typ av partiklar som kallas solitoner som det normalt inte tas h¨ansyntill inom st¨orningsteori.En intressant f¨oruts¨agelsesom h¨orihop med solitoner ¨aratt de kan ge upphov till partiklar med magnetisk laddning ist¨alletf¨or endast elektrisk laddning.
2 Contents
1 Introduction 4
2 The Sine-Gordon and Massive Thirring Model Equivalence 4 2.1 Calculations in classical field theory ...... 5 2.2 Normal ordering of the SG Hamiltonian ...... 7 2.3 Variational computation ...... 11 2.4 Equivalence of sine-Gordon and Thirring model ...... 12 2.5 Consequences of the SG/Thirring equivalence ...... 14
3 Solitons in Higher Dimensions 15 3.1 The global U(1) vortex and Derrick’s theorem ...... 16 3.2 U(1) vortices with gauge symmetry ...... 20 3.3 Adding Fermions ...... 23 3.4 Solitons from the perspective of homotopy ...... 26
4 Magnetic Monopoles 28
5 Conclusions 31
A Vortex and monopole equations near origin 31
B Calculation of fermion equations 32
C Some definitions related to homotopy 33
D Calculating the energy density for the monopole 33
3 1 Introduction
When working with quantum field theories one often identifies the classical wave solutions with the elementary particles of the theory. However, some classical field theories already have particle like solutions in the form of solitons, which are solutions to the field equations that are localized in space and whose energy remains the same as time passes. It might therefore be interesting to see what effects solitons can have on a theory. We will find that the existence of solitons is very dependent on the number of spatial dimensions, D, in which the theory is defined. In this thesis we will almost exclusively consider the cases of one, two or three spatial dimensions, with higher dimensions being discussed briefly in sections 3.1 and 3.4. The thesis is divided into three main sections where the results are presented and discussed, after which follows a concluding section where the thesis is summarized and some outlook is provided. Section 2 starts with a discussion about solitons in one spatial dimension, with the sine-Gordon model as an example, and then moves on to prove the equivalence of the sine-Gordon model and the massive Thirring model. After the proof, we consider some of the physical consequences of the equivalence, such as how charges and couplings are related between the two models. Section 2 is based on chapter 2 in [1] and work done by Coleman [2]. In section 3, we discuss some criteria that need to be fulfilled in order for solitons to exist in more than one spatial dimension. Specifically we consider Derrick’s theorem in section 3.1 which states that no solitons should exist for a certain class of theories, while section 3.4 is dedicated to the connection between the topology of the space of vacua for a scalar potential V (φ) and the topology of spatial infinity. Throughout section 3 we will consider the example of vortices in D = 2, with an excursion to D = 3 in section 3.3. The topics discussed in this section are covered in chapters 3 and 4 of [1]. Section 4 will deal with magnetic monopoles in D = 3 and is based on the material in chapter 5 of [1]. The appendix contains some derivations and definitions that were left out of the main thesis. As is standard practice in particle physics, Greek letters like µ, ν will denote the Lorentz indices 0,1,2,..., while Roman letters like i, j denote Euclidean indices 1,2,... . ijk... is the completely antisymmetric tensor on n indices defined such that 123...n = 1. We will make use of the Einstein summation convention where repeated indices indicate summation. Natural unitsh ¯ = c = 1 are employed throughout the text.
2 The Sine-Gordon and Massive Thirring Model Equiv- alence
The purpose of this section is to first find solitons in the sine-Gordon model and then derive the equivalence of the sine-Gordon and the massive Thirring models, which will be done by comparing the perturbation series of the two theories to all orders and showing that they are equal, provided certain identifications are made between the theories. The sine-Gordon Model, SG, is a two dimensional scalar field theory defined by the
4 Lagrangian density 1 α L = ∂ φ∂µφ + 0 cos(βφ) + γ , (2.1) SG 2 µ β2 0
where α0, β and γ0 are real, positive parameters. The massless Thirring model is described by the Lagrangian 1 L = ψiγ ∂µψ − gjµj , (2.2) Th µ 2 µ with the current jµ = ψγµψ. The massive Thirring model is obtained by adding the term 0 −m σ to LTh, where σ = Zψψ. When we do perturbation theory for the Thirring model, we will consider the mass term as a perturbation to the massless theory.
2.1 Calculations in classical field theory In classical field theory, the Euler-Lagrange equation for a Lagrangian with a potential V (φ) is ∂2φ ∂2φ ∂V (φ) − + = 0, (2.3) ∂t2 ∂x2 ∂φ where the static case is d2φ dV (φ) = . (2.4) dx2 dφ dφ It is possible to rewrite this equation, by multiplying with dx , in the following form " # d2φ dφ dV (φ) dφ d 1 dφ2 0 = − = − V (φ) . (2.5) dx2 dx dφ dx dx 2 dx
An integration then yields dφ2 = 2V (φ) + C, (2.6) dx where C is some constant. If V = 0 at its global minima and φ approaches a constant vacuum value at x = ±∞, then C = 0 and dφ = ±p2V (φ). (2.7) dx Solving for x, we get Z φ dφ0 x = ± . (2.8) p 0 φ0 2V (φ ) If x should vary continuously between ±∞, φ must connect two adjacent minima of V in a monotone fashion, since there the integral diverges. The sine-Gordon potential has N2π minima at φ = β = Nv, where N is an integer. Substituting the sine-Gordon potential
5 into (2.8), integrating and then solving for φ, we find the solution that connects the Nth minimum to the one above: 2v √ φ(x) = Nv + arctan(exp( α (x − x ))), (2.9) π 0 0
where x0 determines where the soliton is localized. A solution of this type is called a kink. By exchanging the sign in front of the arctan, solutions called anti-kinks, connecting the the Nth minima to the N − 1th are obtained. From equation (2.7), it is possible to calculate the energy of the kink. ! Z ∞ 1 dφ2 Z ∞ E = dx + V (φ) = 2 dxV (φ) = −∞ 2 dx −∞ (2.10) Z φ(∞) dφ−1 Z φ(∞) = dφ V (φ) = dφp2V (φ) φ(−∞) dx φ(−∞)
Plugging in the solution (2.9) and the expression for V (φ) into equation (2.10), and using the periodicity of V , we arrive at s Z v 2α0 2π Ekink = dφ 2 1 − cos φ . (2.11) 0 β v
√ 8 α0 Evaluating the integral results in Ekink = β2 . It should be noted here that equation (2.4) is also a condition for φ to be a stationary point of the energy, since in the static case the integrand in (2.10) is just −LSG. This will be used to derive the field equations of the 2-D vortices in section 3. The topological current is defined by 1 jµ = µν∂ φ, (2.12) top 2v ν
µν where is the two index Levi-Civita tensor. Defining the topological charge, Qtop as the spatial integral of the timelike component of jµ results in Z 1 Z 1 Q = dxj0 = dx∂ φ = (φ(∞) − φ(−∞)) . (2.13) top 2v 1 2v
The topological charge of the kink in (2.9) therefore is +1 while the anti-kinks have topolog- ical charge -1. The topological current is different from the currents defined in Noether’s theorem in the sense that it does not correspond to a continuous symmetry of the La- grangian. Since changing the values of φ at ±∞ would require both infinite time and energy, (2.13) shows that topological charge is conserved. Therefore, we should expect the sine-Gordon kink to be a stable solution.
6 2.2 Normal ordering of the SG Hamiltonian Here the sine-Gordon model will be quantized, in order for us to compare its perturbation expansion with that of the Thirring model. Most of the work done in the rest of section 2 will be based on the paper by Coleman [2]. The quantization amounts to expanding the fields in terms of creation and annihilation operators that satisfy canonical commutation relations. For a free scalar field of mass m we have in the Heisenberg picture, where operators are time dependent,
Z dp 1 1/2 φ(x) = [a e−ipx + a†eipx]. (2.14) 2π 2E(p, m) p p
† Here, the operator ap annihilates a particle with x-momentum p while ap creates one. E(p, m) = pp2 + m2 is just the regular relativistic expression for the energy of a particle. In the exponents of the exponentials, the product px is to be understood as the product between the vectors pµ = (E, p) and xµ = (t, x), µ = 0, 1. The creation and annihilation operators satisfy the following commutation relations
h † i 0 h † † i ap, ap0 = 2πδ(p − p ), [ap, ap0 ] = ap, ap0 = 0, (2.15)
where δ(p) is the Dirac delta function. In order to be able to find the relationship between the sine-Gordon model and the Thirring model, it will be necessary to work with the Hamiltonian density, instead of the Lagrangian. First we define the momentum density as ∂L π = . (2.16) ∂φ˙ In terms of a creation and annihilation operators, the momentum density for a free field will be Z dp E(p, m)1/2 π(x) = ∂ φ = −i [a e−ipx − a†eipx]. (2.17) 0 2π 2 p p The Hamiltonian (density) is defined as
H = φπ˙ − L (2.18) and the sine-Gordon Hamiltonian is therefore 1 α 1 1 α H = φπ˙ − ∂ φ∂µφ − 0 cos(βφ) − γ = (∂ φ)2 − (∂ φ)2 + (∂ φ)2 − 0 cos(βφ) − γ = 2 µ β2 0 0 2 0 2 1 β2 0 1 1 α 1 1 α = (∂ φ)2 + (∂ φ)2 − 0 cos(βφ) − γ = π2 + (∂ φ)2 − 0 cos(βφ) − γ . 2 0 2 1 β2 0 2 2 1 β2 0 (2.19)
Before we can do any perturbative calculations, it is necessary to normal order the Hamiltonian, which amounts to rearranging creation and annihilation operators so that in
7 any product containing such operators, all annihilation operators are to the right of the † † creation operators. For example if Napap0 means the normal ordered version of apap0 , then † † Napap0 = ap0 ap. Throughout the text, a subscript on N will be used to indicate which mass is used to define φ during normal ordering, since it will be necessary to consider two different masses, m and µ, in our calculations. Starting with the momentum density, the square of (2.17) becomes the following integral
Z 0 0 1/2 2 dpdp E(p, m)E(p , m) −ipx † ipx −ip0x † ip0x π = − [a e − a e ][a 0 e − a 0 e ] = (2π)2 4 p p p p Z 0 0 1/2 dpdp E(p, m)E(p , m) −i(p+p0)x † −i(p−p0)x = − [a a 0 e − a a 0 e − (2π)2 4 p p p p Z 0 0 1/2 † i(p−p0)x † † i(p+p0)x dpdp E(p, m)E(p , m) −i(p+p0)x − a a 0 e + a a 0 e ] = − [a a 0 e − p p p p (2π)2 4 p p † 0 −i(p−p0)x † i(p−p0)x † † i(p+p0)x − (apap0 + 2πδ(p − p ))e − apap0 e + apap0 e ] (2.20) where the commutation relations (2.15) have been used in the last equality. Doing the same for ∂1φ results in
Z 0 1/2 2 dpdp 0 1 −i(p+p0)x (∂ φ) = − pp [a a 0 e − 1 (2π)2 4E(p, m)E(p0, m) p p (2.21) † 0 −i(p−p0)x † i(p−p0)x † † i(p+p0)x − (apap0 + 2πδ(p − p ))e − apap0 e + apap0 e ]. If we define 1 1 H = π2 + (∂ φ)2, (2.22) 0 2 2 1 then (2.20) and (2.21) imply
Z 0 0 0 dpdp E(p, m)E(p , m) + pp 0 −i(p−p0)x H0 = NmH0 + δ(p − p )e = 8π (E(p, m)E(p0, m))1/2 (2.23) Z dp E(p, m)2 + p2 = N H + = N H + E (m). m 0 8π E(p, m) m 0 0
The integral Z dp E(p, m)2 + p2 Z dp 2p2 + m2 E0(m) = = (2.24) 8π E(p, m) 8π pp2 + m2 diverges unless a cutoff is introduced, since the integrand does not approach zero for large p. Choosing a symmetric cutoff Λ we get,
Z Λ 2 2 √ dp 2p + m 2 E0(m, Λ) = = 2Λ Λ + m . (2.25) p 2 2 −Λ 8π p + m
8 The divergence can be handled by a redefinition of γ0, so that the quantity γ = γ0 −E0(m) is finite in the limit where Λ → ∞. If the normal ordering was done with respect to a different mass µ, the result would be
H0 = NµH0 + E0(µ). (2.26)
The two different orderings are related by
NmH0 = NµH0 + E0(µ) − E0(m). (2.27)
Computing the difference E0(µ) − E0(m) shows that it is finite in the limit Λ → ∞,
√ 2 2 2 2 2Λ p 2 2 2 2 2Λ µ − m µ − m E0(µ) − E0(m) = Λ + µ − Λ + m = √ → . 8π 8π pΛ2 + µ2 + Λ2 + m2 8π (2.28) The parameter µ is introduced to handle divergences that arise from the fact that there is no clear mass term in the SG Lagrangian, which means for instance that the definitions of the field operators diverge. In the perturbative calculations of section 2.4, µ will take the role of a mass that will later be sent to zero. It now remains to normal order the cosine term of the Hamiltonian, and for that purpose we will need the formula Z Z 1 Z exp i J(x)φ(x)d2x = N exp i J(x)φ(x)d2x exp − J(x)∆(x − y)J(y)d2xd2y , m 2 (2.29) where J(x) is some function of the spacetime coordinates and ∆(x − y) is the commutator
∆(x − y) = φ+(x), φ−(y) , (2.30)
with φ+ containing all the annihilation operators of φ and φ− all the creation operators. This can be done using the general expression
A B A+B+ 1 [A,B] e e = e 2 , (2.31)
which holds when both A and B commute with the commutator, see chapter 19 of [3]. Normal ordering of exp i R J(x)φ(x)d2x corresponds to taking A = i R dx2J(x)φ−(x) and R 2 + 1 [B,A] B = i dx J(x)φ (x) in (2.31). If both sides are multiplied by e 2 , (2.29) is obtained. For our purposes, it will be necessary to evaluate the commutator explicitly for spacelike x. Writing the fields in terms of ladder operators and using the commutation relations
9 results in
Z 0 1/2 + − dpdp 1 † i(p0y−px) † i(p0y−px) φ (x), φ (y) = a a 0 e − a 0 a e = 8π2 E(p, m)E(p0, m) p p p p Z 0 1/2 dpdp 1 h † i i(p0y−px) = a , a 0 e = 8π2 E(p, m)E(p0, m) p p 0 1/2 Z dpdp 1 0 = 2πδ(p − p0)ei(p y−px) = 8π2 E(p, m)E(p0, m) Z dp eip(y−x) = . 4π pp2 + m2 (2.32)
The commutator is therefore a function of z = y − x. For spacelike z, it is always possible to make a Lorentz-transformation so that the timelike component is zero, which means that the exponent depends only on the spacelike components of z and p. We also see that the integral diverges for z = 0, which will be dealt with later when we consider small z. It is allowed to make a Lorentz transformations since dp/E is Lorentz invariant. Rewriting the exponential in terms of sine and cosine, we find that the sine integral vanishes since the rest of the integrand is an even function. The final result is 1 Z ∞ cos pz 2 Z ∞ cos pz 1 ∆(z) = dp = dp = 2K0(m|z|) (2.33) p 2 2 p 2 2 4π −∞ p + m 4π 0 p + m 4π where Kν are the modified Bessel functions of the second kind[4][5].The first few terms in the series representation of K0 are, section 11.5 in [6],
2 K0(x) = − ln x − γ + ln 2 + O(x ), (2.34) where γ is the Euler-Mascheroni constant. ∆(z) can therefore be written as 1 ∆(z) = − ln cm2z2, (2.35) 4π for small z. The constants in (2.34) have been collected into ln c. To bypass the divergence of ∆, we make the replacement
1 1 1 m2 ∆(z) → ∆(z, Λ) = − ln cm2z2 + ln cΛ2z2 = − ln . (2.36) 4π 4π 4π Λ2
Like with the normal ordering of H0, Λ is used as a cutoff in this context. If we take J = βδ(y − x), then (2.29) gives
2 m2 β /8π eiβφ = N eiβφ. (2.37) Λ2 m
10 Defining α as 2 m2 β /8π α = α (2.38) 0 Λ2 results in α0 cos βφ = αNm cos βφ. (2.39)
α will be a finite parameter after renormalization of α0 to remove the dependence on Λ. The result of a normal ordering with respect to the parameter µ would then be
2 µ2 β /8π α cos βφ = α N cos βφ (2.40) 0 0 Λ2 µ
leading to the identification
2 2 2 m2 β /8π µ2 β /8π µ2 β /8π N cos βφ = N cos βφ ⇔ N cos βφ = N cos βφ, Λ2 m Λ2 µ m m2 µ (2.41) which will be used in the following two sections. Using (2.23) and (2.39), the normal ordered version of (2.19) is
H = NmH0 − αNm cos βφ − γ. (2.42)
2.3 Variational computation It is possible to draw some conclusions about the allowed values of parameters in the model by doing variational calculations with respect to µ. Define trial vacuum states such that a(p, µ) |0, µi = 0. The result of normal ordering (2.19) with respect to µ, by using (2.27) and (2.41), is
2 ! α µ2 β /8π 1 H = N H − cos βφ − γ + µ2 − m2 . (2.43) µ 0 β2 m2 8π
The vacuum expectation value is therefore
2 ! α µ2 β /8π 1 h0, µ| H |0, µi = h0, µ| N H − cos βφ − γ + µ2 − m2 |0, µi = µ 0 β2 m2 8π
2 α µ2 β /8π 1 1 = − h0, µ| N 1 − β2φ2 + ... |0, µi − γ + µ2 − m2 = β2 m2 µ 2 8π 2 α µ2 β /8π 1 = − − γ + µ2 − m2 . β2 m2 8π (2.44)
11 Equation (2.44) implies that, for β2 > 8π, the vacuum expectation value approaches −∞ when µ → ∞, since the terms involving µ can be rewritten as 1 α 2 µ2 − µ2β /8π−2 → −∞. (2.45) 8π β2m2 Therefore, by the variational method, the energy density of the ground state is unbounded below which is unphysical.
2.4 Equivalence of sine-Gordon and Thirring model Having normal ordered the Hamiltonian in section 2.2 and performed some variational com- putations in section 2.3, it is time to establish the correspondence between the sine-Gordon model and the massive Thirring model. The equivalence between the two models is found by calculating the vacuum expectation value of products of the interaction Hamiltonians HI, which contains the interaction terms of H, evaluated at different spacetime points,
h0 |T ΠiHI (xi)| 0i. (2.46)
The T operator denotes time ordering, which is similar to normal ordering, but here opera- tors evaluated at earlier times are to the right of operators evaluated at later times. In the case of the SG model, HI corresponds to the cosine term of H. These expectation values are called Green’s functions and they determine the amplitude for a particle to propagate from one spacetime point to another. Before the Green’s functions for the SG model are calculated, it is useful to calculate more general Green’s functions. The SG Green’s functions are then special cases of the more general ones. These more general objects are of the form
iβiφ(xi) h0, µ T ΠiNme 0, µi, (2.47) where µ is a small mass added to the theory to handle divergences and that will be sent to zero at the end of the calculations. The index on the βi:s indicates that the β:s can depend on xi. In (2.47), the process of time ordering only amounts to a permutation of the xi which 0 0 0 means that there is no loss in generality in assuming that x1 ≥ x2 ≥ x3 .... If the xi:s are not already in time order, then after time ordering of the operators we can always relabel the points so that the above inequality holds. The operator product is therefore
2 P 2 2 βi /8π 2 i βi /8π iβ φ(x ) iβ φ(x ) µ iβ φ(x ) µ iβ φ− iβ φ+ T Π N e i i = Π N e i i = Π N e i i = Π e i i e i i , i m i m i m2 µ m2 i (2.48) where in the second step (2.41) was used and φi is shorthand for φ(xi). Using the formula
eAeB = eBeAe[A,B], (2.49)
12 iβ φ+ it is possible to normal order the entire product. A given e j j will need to be commuted iβ φ− with all the e i i for which j > i. The result is therefore
P 2 2 i βi /8π µ −β β φ+,φ− T Π N eiβi(xi) = Π N eiβiφi · Π e i j [ i j ] = i m m2 i µ i>j (2.50) P β2/8π µ2 i i βiβj = Π N eiβiφi · Π cµ2(x − x )2 4π . m2 i µ i>j i j
Since the exponentials have been normal ordered, they will just contribute a factor of 1, which comes from the first term of the series expansion, to the matrix elements (2.47). This leads to
P β2/8π µ2 i i βiβj iβi(xi) 2 2 4π h0, µ T ΠiNme 0, µi = Πi>j cµ (xi − xj) (2.51) m2
2 P β /4π The final expression is proportional to µ( i i) , which can be seen by using the relation P 2 P 2 P ( i pi) = i pi + i>j 2pipj. This means that, in the limit µ → 0, the matrix elements will be zero unless the βi sum to zero. It is now time to carry out the calculations for the actual SG Green’s functions. As stated in the beginning of this section, the interaction terms of this model are contained in α α H = −N cos βφ = − (A − A ), (2.52) I m β2 2β2 + − ±iβφ where A± = Nme . As a consequence of the arguments below eq. (2.51), only Green’s functions involving an even equal number of A+ and A− can be nonzero. It is therefore enough to calculate matrix elements of the form
n h0 |T Πi A+(xi)A−(yi)| 0i, (2.53) where the limit µ → 0 has already been taken. Using (2.51) we arrive at
2 2 4 2 2 β /4π n Πi>j [c m (xi − xj) (yi − yj) ] h0 |T Πi A+(xi)A−(yi)| 0i = 2 . (2.54) 2 2 β /4π Πij [cm (xi − yj) ] Klaiber was able to show, [7], that in the massless Thirring Model the following relation holds for the Green’s functions
2n 4 2 2 1+b/π n 1 Πi>j [M (xi − xj) (yi − yj) ] h0 |T Πi σ+(xi)σ−(yi)| 0i = , (2.55) 2 2 1+b/π 2 Πij [M (xi − yj) ]
g where M is an arbitrary mass and b = − 1+g/π . The σ± are defined as 1 σ = Zψ(1 ± γ )ψ. (2.56) ± 2 5 13 Comparing (2.54) and (2.55) shows that the Green’s functions of the two theories are equal provided the following identifications are made
σ± = −A± M 2 = cm2 (2.57) 1 β2 = . 1 + g/π 4π
2.5 Consequences of the SG/Thirring equivalence By considering two particular commutators,
µ [j (x), σ±(y)] and [∂νφ(x),A±(y)] , (2.58) and using the identifications made in (2.57), it is possible to connect the topological current (2.12) with jµ. Klaiber showed, [7], that in the massless Thirring model the following commutation relation holds
µ µν −1 µν [j (x), ψ(y)] = − g + (1 + g/π) γ5 ψ(y)Dν(x − y), (2.59) where Dµ(x − y) = ∂µ[φ(x), φ(y)]. Only the commutators involving σ+ and A+ will be calculated explicitly, since the other two are calculated in the same manner.
µ 1 µ [j (x), σ+(y)] = Z j , ψ(1 + γ5)ψ = 2 (2.60) 1 = Z jµ, ψ ψ + ψ [jµ, ψ] + jµ, ψ γ ψ + ψ [jµ, γ ψ] 2 5 5 Where the operator identity [A, BC] = [A, B]C +B[A, C] has been used. The commutators that involve ψ can be calculated directly from (2.59), while those involving ψ require the application of the identity µ µ † j , ψ = − [j , γ0ψ] . (2.61) Another application of (2.59), and using that gµν is symmetric and µν is antisymmetric, results in µ µν −1 µν j , ψ = g ψ − (1 + g/π) ψγ5 Dν(x − y). (2.62) The end result is
µ −1 µν [j (x), σ±(y)] = ∓2(1 + g/π) σ±Dν(x − y). (2.63)
Moving on to the commutator involving scalar fields, we can immediately write
[∂νφ(x),A+(y)] = ∂ν [φ(x),A+(y)] , (2.64)
14 since the gradient is with respect to x and not y. Remembering the definitions of φ+ and φ− from section 2.2 we can calculate the commutator above as
iβφ− iβφ+ iβφ− iβφ+ iβφ− iβφ+ [φ(x),A+(y)] = φ(x), e e = φ(x), e e + e φ(x), e =
iβφ− iβφ+ iβφ− iβφ+ = iβ [φ+(x), φ−(y)] e e + iβ [φ−(x), φ+(y)] e e = (2.65)
iβφ− iβφ+ = iβ [φ(x), φ(y)] e e = iβ [φ(x), φ(y)] A+(y), where the second line follows from expanding the exponentials as power series and using the formula [A, Bn] = nBn−1 [A, B] . (2.66) from chapter 19 in [3]. Plugging this into (2.64) yields
[∂νφ(x),A±(y)] = ±iβDνA±. (2.67)
It is now possible to make the identification 2 β jµ = − µν∂ φ = − µν∂ φ = −2jµ , (2.68) β(1 + g/π) ν 2π ν top
µ with jtop defined as in (2.12). This result for jµ combined with the identifications (2.57) makes it possible to connect a theory with only bosons, the sine-Gordon model, with a theory with only fermions, the massive Thirring model. From (2.57), we see that strong coupling in one theory corresponds to weak coupling in the other, which can be exploited when doing calculations by, for instance, recasting a theory about fermions into bosonic form when dealing with the strong coupling regime of the fermionic theory. The relations between the currents also make it possible to connect the fermion number, with the topological charge.
3 Solitons in Higher Dimensions
What made the existence of solitons in one spatial dimension possible, was the fact that the scalar field could take on different vacuum values at spatial infinity, making it impossible to deform the field configuration to a uniform vacuum solution. In one dimension, spatial infinity is represented by the two points ±∞ so only two distinct vacua are required in order for solitons to exist. In higher dimensions this requirement must be modified to reflect the topological characteristics of spatial infinity. For instance, with two spatial dimensions spatial infinity is described by a circle with radius r = ∞. If we still want the field to approach different vacuum values at different parts of spatial infinity and be continuous, we arrive at the conclusion that there must be a continuous family of vacua for solitons to exist. As will be discussed in more detail, it will also be necessary to consider gauge theories, since the coupling terms between the scalar field and the gauge field that come from the covariant derivative will be needed to balance the spatial gradient terms of the scalar field,
15 which would otherwise lead to infinite energy. The coupling between the gauge field and the scalar field will mean that the resulting field equations will not be possible to solve analytically, meaning that a lot of the information about the nature of solitons in higher dimensions comes from numerical studies, analysis of asymptotic behavior and qualitative reasoning. The derivations and discussions in this section follow chapters 3 and 4 in [1].
3.1 The global U(1) vortex and Derrick’s theorem As a first attempt to find a soliton in two spatial dimensions, we can consider a complex scalar field iα(x) φ(x) = φ1(x) + iφ2(x) = ρ(x)e , (3.1) with the Lagrangian 1 λ µ2 2 L = (∂ φ)† (∂µφ) − |φ|2 − . (3.2) 2 µ 4 λ µ and λ are both real, positive parameters. The theory posses a global U(1) symmetry, since the Lagrangian is invariant under the transformation φ → eiξφ, where ξ is a real number. The potential λ µ2 2 V (φ) = |φ|2 − (3.3) 4 λ has a continuous set of minima at r µ2 |φ| = v = . (3.4) λ The discussion earlier implies that it might therefore be possible for solitons to exist within this theory. In two spatial dimensions, we can define the vorticity of a scalar field as 1 I 1 I N[C] = dl∇(argφ) = dl∇α (3.5) 2π C 2π C where the contour C is the circle at r = ∞. N[C] counts how many rotations the phase of φ goes through during one loop of C. If φ is to be single-valued, then N[C] must be an integer since the line integral must be an integer multiple of 2π. The quantization of N means that a configuration with one value of N cannot be continuously deformed into a configuration with another value. If we consider C to be the circle at infinity, then the fact that φ is single-valued again means that N must be an integer. The exception to this is when φ has a zero on C, because then α is not well defined. Since the result of going around a loop enclosing a zero of φ must be an integer, we can see that if a loop not enclosing a particular zero is deformed to do so, the difference in N between the two loops must be the N that is found when encircling just the zero in question. If we call N[L]i the result of going around an infinitesimal loop L encircling the i:th zero of φ, then starting with N[C] = N for C at infinity and contracting C to a point
16 P that is not a zero of φ shows that i N[L]i = N. Thus, any configuration with nonzero vorticity must have at least one zero. Since a uniform vacuum solution must clearly have N = 0, it is reasonable to assume that a stable soliton in two dimensions should have N 6= 0 to ensure that it cannot be continuously deformed to the vacuum. The Euler Lagrange equations for φ1 and φ2 with the Lagrangian (3.2) are ∂L ∂L ∂ − = ∂µ∂ φ + λφ φ2 + φ2 − v2 = 0 µ ∂(∂ φ ) ∂φ µ 1 1 1 2 µ 1 1 (3.6) ∂L ∂L µ 2 2 2 ∂µ − = ∂ ∂µφ2 + λφ2 φ1 + φ2 − v = 0, ∂(∂µφ2) ∂φ2 which can be combined into the single equation ∂2 − ∇2 φ + λφ |φ|2 − v2 = 0. (3.7) ∂t2 The static version becomes ∇2φ − λφ(|φ|2 − v2) = 0. (3.8) To look for solutions to (3.8), we can try to make an ansatz that possesses symmetries that simplify the equation. A completely rotationally invariant ansatz, where φ is a function of only r = px2 + y2, would not lead to solitons since the field would have the same value everywhere at infinity. Another possible ansatz is
φ = f(r)eiθ, (3.9)
where θ is the polar angle. This is rotationally invariant in the sense that a rotation can be completely compensated by a U(1) transformation. If we also demand that φ should be invariant under a reflection through the x-axis combined with a complex conjugation we find φ(r, θ) = φ∗(r, −θ) = f ∗(r)(e−iθ)∗ = f ∗(r)eiθ, (3.10) which means that f must be a real function. Inserting the ansatz into (3.8) gives the following equation for f d2f 1 df f + − − λf(f 2 − v2) = 0. (3.11) dr2 r dr r2 This equation has the form of Bessel’s equation with an added nonlinear term. The non- linearity of the equation makes finding a solution difficult, so trying to find numerical solutions is a better option. In order to ensure that φ is well defined at the origin, we require f(0) = 0 since there eiθ is not defined. A more detailed discussion about the be- havior of f near r = 0 is provided in appendix A. Since we are trying to find solutions that approach a vacuum value at r = ∞, the second boundary condition for f becomes f(∞) = v. With these boundary conditions, we can see from (3.5) that N = 1. A nu- merical solution to (3.11) for λ = v = 1, generated using standard MATLAB functions, is presented in figure 1.
17 Figure 1: Numerical solution for the vortex with global U(1) symmetry.
The solution presented here has a problem in that the energy is in fact infinite. To see why this is the case we examine the form of the energy Z 1 E = d2x (∇φ)2 + V (φ) . (3.12) 2 Inserting our ansatz and changing to polar coordinates results in ! ! Z 1 ∂f 2 f 2 λ E = dθdrr + + f 2 − v22 . (3.13) 2 ∂r r2 4
As r becomes large, the term f 2/r2, which comes from the angular part of the gradient, will approach v2/r2 meaning that the extra factor of r that comes from the integration element will make the energy diverge. This is in fact a consequence of a result known as Derrick’s theorem [8], which is presented below. At first we will consider a theory in D spatial dimensions involving one or more scalar fields φa with a Lagrangian of the form 1 L = G (φ)∂ φ ∂µφ − V (φ) (3.14) 2 ab µ a b where G(φ) is a positive definite matrix for all φ and the potential has a global minimum at V (φmin) = 0. The global U(1) theory discussed in this section is an example of such a theory with D = 2. When considering static solutions the energy is given by Z D E = − d xL = Ik[φ] + IV [φ], (3.15) where 1 Z I [φ] = dDxG (φ)∂ φ ∂ φ k 2 ab i a i b Z (3.16) D IV [φ] = d xV (φ).
18 A finite energy static solution should be a stationary point of E. If a solution is stationary when we consider all possible field configurations, it must also be stationary if we consider any subset of configurations to which it belongs. If we assume that φ is a solution, we can then consider variations of this solution that amount to a rescaling of coordinates
fλ(x) = φ(λx), (3.17)
where λ is a real number. The energy of such a configuration is 1 Z Z E(λ) = I [f ] + I [f ] = dDxG (φ(λx))∂ φ (λx)∂ φ (λx) + dDxV (φ(λx)). (3.18) k λ V λ 2 ab i a i b
Changing variables to y = λx, the volume element will contribute a factor of λ−D to both Ik and IV while each derivative factor in Ik will contribute a factor λ. We therefore have
2−D −D E(λ) = λ Ik φ + λ IV φ . (3.19)
The assumption that f1 = φ(x) is a solution requires that λ = 1 is a stationary point of E(λ). We should therefore have
dE 1−D −(D+1) 0 = = (2 − D)λ Ik − Dλ IV = (2 − D)Ik φ − DIV φ , (3.20) dλ λ=1 λ=1 or (D − 2)Ik φ + DIV φ = 0. (3.21) For D = 1, this can be seen as a generalization of the relation (2.6). Since both integrals Ik and IV are positive, D = 2 implies IV (φ) = 0 which means that any φ must be equal to the vacuum everywhere, and so no solutions with nonzero vorticity can exist since those require φ to have at least one zero. D ≥ 3 requires Ik = IV = 0, so φ must identically equal one of the vacuum values, which is uninteresting. Since the above theories with global symmetries do not have solitons in higher dimen- sions, we can try to see if theories with gauge symmetries are better alternatives. We consider a gauge theory, that can be Abelian or non-Abelian, with a Lagrangian 1 1 L = − trF F µν + (D φ)† (Dµφ) − V (φ) (3.22) 2 µν 2 µ where Fµν = ∂µAν − ∂νAν + ig [Aµ,Aν] (3.23) is the field strength. The covariant derivative is
Dµ = ∂µ + igAµ, (3.24)
g is the coupling strength, the gauge fields are
a a Aµ = AµT (3.25)
19 and T a are the generators of the symmetry. We will only consider static solutions in a gauge such that A0 = 0, i.e. F0j = 0 which means there is no electric field. This is what we will work with in this section and in the section about magnetic monopoles, section 4. The energy of such a solution is given by
E = IF [A] + Ik[φ, A] + IV [φ] (3.26)
with 1 Z I [A] = dDxtrF F F 2 ij ij 1 Z I [φ, A] = dDx (D φ)† (D φ) (3.27) k 2 i j Z D IV [φ] = d xV (φ).
Given a solution φ, Ai, we define a rescaling such that φ is transformed like in (3.17) and Ai is transformed like giλ(x) = λAi(λx). (3.28)
The reason for multiplying A by λ is to ensure that the result of the transformation on Fµν and Dµ is to just multiply them by powers of λ. By the same reasoning as in the global case, we arrive at an analog of (3.21), namely (D − 4)IF A + (D − 2)Ik φ, A + DIV φ = 0 (3.29)
This equation allows for the possibility of having both the scalar and gauge fields be nontrivial for D = 1, 2, 3. If D = 4, it might be possible to have nontrivial gauge fields, so long as φ = 0 and V (0) = 0. For D ≥ 5 we only expect trivial solutions since the integrals and their prefactors are all positive.
3.2 U(1) vortices with gauge symmetry In the previous section we saw that it might be possible for solitons to exist in two spatial dimensions, provided we are working with a local gauge theory. This section will argue for the existence of finite energy solitons with nonzero vorticity in a local U(1) theory. The Lagrangian under consideration is
1 1 λ µ2 2 L = − F F µν + (D φ)† (Dµφ) − |φ|2 − , (3.30) 4 µν 2 µ 4 λ
with F = ∂ A − ∂ A µν µ ν ν µ (3.31) Dµ = ∂µ + ieAµ.
20 Since the potential is just (3.3), this theory has the same family of possible vacuum ex- pectation values as the global vortex. If one were to expand√ φ around such a vacuum, one would find that the theory has a scalar with mass mφ = 2µ and a vector boson of mass mA = ev. These masses will later play a role in the long range behavior of φ and Aµ. From standard particle physics we know that L is invariant under
φ → eieΛ(x)φ (3.32) Aµ → Aµ − ∂µΛ(x).
A quick calculation also shows that under the same transformation 1 I 1 I e I N[C] → N 0[C] = dl∇(argφ) = dl∇(α + eΛ(x)) = N[C] + dl∇Λ(x) 2π C 2π C 2π C (3.33) i.e. that N[C] is also invariant, provided Λ is well behaved at infinity. Trying to find a N = 1 vortex, we make the ansatz
φ = veiθf(evr) a(evr) A = jkrˆk (3.34) j er A0 = 0, where f and a are real functions andr ˆi = xi/r. We see that due to the factor eiθ, N = 1 and f(0) must be zero in order for φ to be well defined at the origin. a(0) = 0 is necessary to avoid singularities in Aj. After calculating the necessary partial derivatives of (3.34), one finds that
2 da2 F F = ij ij e2r2 dr (3.35) df 2 v2(1 − a)2f 2 (D φ)† (D φ) = v2 + . i j dr r2
Plugging into (3.26), changing to polar coordinates and introducing u = evr, we find ! Z 1 da2 df 2 (1 − a)2f 2 λ E = πv2 duu + + + 1 − f 22 = πv2G(λ/2e2). u2 du du u2 2e2 (3.36) The Euler-Lagrange equations for a and f become
2 d a 1 da 2 0 = 2 − + (1 − a)f du u du (3.37) d2f 1 df (1 − a)2 λ 0 = + − f + f 1 − f 2 . du2 u du u2 e2
21 Figure 2: Numerical solutions of (3.37) for different values of λ/e2. On the left λ/e2 = 1/4, while on the right λ/e2 = 10. The solid and dashed lines represent f and a respectively.
We have already found the boundary conditions at u = 0. If IV should be finite, then f(∞) = 1 so that φ approaches vacuum values. Inserting this into the second equation shows that a(∞) = 1. Two numerical solutions for different values of λ/2e2 are provided in figure 2. The behavior of f and a near u = 0 is explored in appendix A. As figure 2 shows, the behavior of f seems to have a large dependence on λ/2e2 com- pared to a, which seems largely unchanged between the two plots. The reason for this can be understood by considering the asymptotic behavior of both functions, which is done by linearizing (3.37). For large u, we can expand both functions around 1
a = 1 − (3.38) f = 1 − δ,
and insert into the first equation of (3.37), keeping only linear terms. The resulting equa- tion is 1 00 − 0 − = 0. (3.39) u This equation has a solution in terms of Bessel functions [5]
AuJ1(iu) + BuY1(iu). (3.40)
Asymptotically these Bessel functions take the form
1/2 u−i3/4π −u+i3/4π 1/2 u−i3/4π −u+i3/4π J1(iu) ∝ u e + e ,Y1(iu) ∝ −iu e − e , (3.41)
where only the first term in the asymptotic expansion of each function has been kept. Since should approach 0 for large u, we require that A = iB which leads to √ √ ∼ ue−u ∼ re−mAr. (3.42)
22 Linearizing the second equation then leads to 1 2λ δ00 + δ0 − δ = 0. (3.43) u e2 Similar arguments as above result in √ √ e− 2λu/e e− 2λvr e−mφr δ ∼ √ ∼ √ = √ . (3.44) u r r
2 However, the last result is only valid if 2mA > mφ, λ/e < 2, since otherwise the nonlinear term 2/u2, that is neglected after linearizing, does not fall faster than any of the terms in (3.43). δ must now match the behavior of this term, which goes like e−2mAr/r. Using the guess that δ = Ce−2u/u, and neglecting any terms in the derivatives of δ that fall faster than this, shows that δ ∼ e−2u/u. In summary, we have for δ
e−mφr δ ∼ √ , λ/e2 < 2 r (3.45) e−2mAr δ ∼ , λ/e2 > 2. r This dependence on λ/e2 has an effect on the interactions between vortices. The force associated with Aµ is the B field, B = F12 since all other components of Fµν are zero, which will behave like 1 da e−mAr B = ∂ A − ∂ A = ∼ √ . (3.46) 1 2 2 1 er dr r
When mA < mφ this force will dominate over the interaction mediated by φ, while for mA > mφ the roles are reversed. If we consider the force between two vortices with the same charge, then φ will give an attractive force while B will give a repulsive force, the sign of the force depending on whether the mediating boson has even or odd spin, see section 3.4 [1].
3.3 Adding Fermions Having found a soliton in two dimensions we will now consider what happens when a fermion is included in the theory. First we can extend our vortex solution in two space dimensions into three and assume that the vortex has no z dependence. Such a solution will then be interpreted as a string of vortices along the z-axis. When D = 3, a fermion is described by a four component spinor ψ that is the sum of two chirality eigenspinors, ψR and ψL with,
5 γ ψR = ψR 5 (3.47) γ ψL = −ψL.
23 We add the Lagrangian
µ µ ∗ LF = iψRγ DµψR + iψLγ DµψL − gφψLψR − gφ ψRψL, (3.48)
5 to (3.30). Since ψR and ψL are eigenspinors of γ ,we have
iβγ5 iβ iβγ5 −iβ e ψR = e ψR, e ψL = e ψL. (3.49)
Our Lagrangian is therefore invariant under the chiral transformation
ψ → eiβγ5 ψ (3.50) φ → e−i2βφ.
Including mass terms, e.g. mψψ, breaks this symmetry, since something like mψLψR is not invariant. It should also be noted that in order for LF to be invariant under U(1), φ must have twice the charge of ψ The Dirac equations for ψR and ψL that follow from (3.48) are
∂LF ∂LF ∗ µ ∂µ − = gφ ψL − iγ DµψR = 0 ∂ ∂µψ ∂ψ R R (3.51) ∂LF ∂LF µ ∂µ − = gφψR − iγ DµψL = 0. ∂ ∂µψL ∂ψL In trying to look for solutions to these equations we first make the assumption that there should be no t or z dependence. We can also consider ψR and ψL to only depend on the distance r from the z-axis and for them to be eigenspinors of iγ1γ2
1 2 iγ γ ψL = ψL 1 2 (3.52) iγ γ ψR = −ψR.
With these assumptions and with φ and Aµ taking their local vortex form, (3.51) simplifies to, see appendix B, a ∂ + ψ − igvfγ1ψ = 0 r 2r R L a (3.53) ∂ + ψ − igvfγ1ψ = 0. r 2r L R We see that the equations decouple if
1 1 ψL = −iγ ψR, ψR = −iγ ψL, (3.54) leading to a ∂ + ψ + gvfψ = 0, (3.55) r 2r L L
24 and a similar equation for ψR. (3.55) is solved by
Z r 0 0 0 a(r ) ψL = A exp − dr gvf(r ) + 0 χ (3.56) 0 2r where A is a constant and χ is constant left handed spinor with iγ1γ2χ = χ. A simple way to introduce t and z dependence for the fermion field is to multiply our existing solutions ψR and ψL by a function b(z, t)
0 0 ψR = b(z, t)ψR, ψL = b(z, t)ψL. (3.57)
Inserting into the second equation of (3.51) and using that ψR solves the part that does not involve t, z leads to 0 3 i γ ∂0 + γ ∂3 b(z, t)ψL = 0. (3.58)
Since ψL satisfies (3.56), we have