Texts on physics, maths and programming (Несколько текстов для зачёта), страница 27

Consider therefore two concentric semicircles Γ₁ and Γ₂ in the upper half plane, centred on the origin, and of radii R₁ < R₂. For |r| < R₁ let α^μ be conformal, with α^z = α (z), while for |r| > R₂ take α^μ = 0. In between, α^μ is not conformal, but is differentiable, so that . This can be integrated by parts to give a term (which must vanish because α_ν is arbitrary in this region, implying that ∂_μT^μν = 0) and two surface terms. That on Γ₂ vanishes because α^μ = 0 there. We are left with

(43)

where dℓ^λ is the line element along Γ₁.

The fact that T^μν is conserved means, in complex coordinates, that , so that the correlations functions of T (z) ≡ T_zz are holomorphic functions of z, while those of are antiholomorphic. Eq. (43) may then be written

(44)

In any field theory with a boundary, it is necessary to impose some boundary condition. It can be argued that any translationally invariant boundary condition flows under the RG to conditions satisfying T_xy = 0, which in complex coordinates means that on the real axis. This means that the correlators of are those of T analytically continued into the lower half plane. The second term in (44) may then be dropped if the contour in the first term is around a complete circle.

The conclusion of all this is that the effect of an infinitesimal conformal transformation on any correlator of observables inside Γ₁ is the same as inserting a contour integral ∫ T (z)α (z) dz/2πi into the correlator.

Another important element of CFT is the operator product expansion (OPE) of the stress tensor with other local operators. Since T is holomorphic, this has the form

(45)

where the ⁽ⁿ⁾ are (possibly new) local operators. By taking α (z)   z (corresponding to a scale transformation) we see that ⁽⁰⁾ = h , where h is its scaling dimension. Similarly, by taking α = const., ⁽⁻¹⁾ = ∂_x . Local operators for which ⁽ⁿ⁾ vanishes for n   1 are called primary. T itself is not primary: its OPE with itself takes the form

T(z)·T(0)=c/2z4+(2/z2)T(0)+(1/z)∂zT(0)+ ,

(46)

where c is the conformal anomaly number, ubiquitous in CFT. For example, the partition function on a long cylinder of length L and circumference ℓ behaves as exp(πcL/ℓ), cf. (10).

5.2. Radial quantisation

This is the most important concept in understanding the link between SLE and CFT. We introduce it in the context of boundary CFT. As before, suppose there is some set of fundamental fields {ψ (r)}, with a Gibbs measure e^−S[ψ][dψ]. Let Γ be a semicircle in the upper half plane, centered on the origin. The Hilbert space of the BCFT is the function space (with a suitable norm) of field configurations {ψ_Γ} on Γ.

The vacuum state is given by weighting each state by the (normalised) path integral restricted to the interior of Γ and conditioned to take the specified values on the boundary

(47)

Note that because of scale invariance different choices of the radius of Γ are equivalent, up to a normalisation factor.

Similarly, inserting a local operator  (0) at the origin into the path integral defines a state | . This is called the operator-state correspondence of CFT. If we also insert (1/2πi)∫_C zⁿ⁺¹T (z) dz, where C lies inside Γ, we get a state L_n| . The L_n act linearly on the Hilbert space. From the OPE (45) we see that L_n|    | ⁽ⁿ⁾ , and that, in particular, L₀|  = h | . If is primary, L_n|  = 0 for n   1. From the OPE (46) of T with itself follow the commutation relations for the L_n

(48)

which are known as the Virasoro algebra. The state | together with all its descendants, formed by acting on | an arbitrary number of times with the L_n with n   −1, give a highest weight representation (where the weight is defined as the eigenvalue of −L₀).

There is another way of generating such a highest weight representation. Suppose the boundary conditions on the negative and positive real axes are both conformal, that is they satisfy , but they are different. The vacuum with these boundary conditions gives a highest weight state which it is sometimes useful to think of as corresponding to the insertion of a ‘boundary condition changing’ (bcc) operator at the origin. An example is the continuum limit of an Ising model in which the spins on the negative real axis are −1, and those on the positive axis are +1.

5.3. Curves and states

In this section, we describe a way of associating states in the Hilbert space of the BCFT with the growing curves of the Loewner process. This was first understood by Bauer and Bernard [10], but we shall present the arguments slightly differently.

The boundary conditions associated with a bcc operator guarantee the existence, on the lattice, of a domain wall connecting the origin to infinity. Given a particular realisation γ, we can condition the Ising spins on its existence. We would like to be able to assume that this property continues to hold in the continuum limit: that is, we can condition the fields {ψ} on the existence of a such a curve. However, this involves conditioning on an event with probability zero: it turns out that in general the probability that, with respect to the measure in the path integral, the probability that a domain wall hits the real axis somewhere in an interval of length vanishes like ^h. In what follows we shall regard as small but fixed, and assume that the usual properties of SLE are applicable to this more general case.

Any such curve may be generated by a Loewner process: denote as before the part of the curve up to time t by γ_t. The existence of this curve depends on only the field configurations ψ in the interior of Γ, as long as γ_t lies wholly inside this region. Then we can condition the fields contributing to the path integral on the existence of γ_t, thus defining a state

(49)

The path integral (over the whole of the upper half plane, not just the interior of Γ), when conditioned on γ_t, gives a measure dμ(γ_t) on these curves. The state

|h =|ht ≡∫dμ(γt)|γt

(50)

is in fact independent of t, since it is just given by the path integral conditioned on there being a curve connecting the origin to infinity, as guaranteed by the boundary conditions. In fact we see that |h is just the state corresponding to a boundary condition changing operator at the origin.

However, dμ (γ_t) is also given by the measure on a_t in Loewner evolution, through the iterated sequence of conformal mappings satisfying dgˆ_t = 2dt/gˆ_t − da_t. This corresponds to an infinitesimal conformal mapping of the upper half plane minus K_t. As explained in the previous section, dgˆ_t corresponds to inserting (1/2πi)∫_C (2dt/z − da_t)T (z) dz. In operator language, this corresponds to acting on |γ_t with 2L₋₂dt − L₋₁da_t where L_n = (1/2πi)∫_C zⁿ⁺¹T (z) dz. Thus, for any t₁ < t

(51)

where T denotes a time-ordered exponential.

The measure on γ_t is the product of the measure of γ_t γ_t1, conditioned on γ_t1, with the unconditioned measure on γ_t1. The first is the same as the unconditioned measure on g_t1(γ_t), and the second is given by the measure on a_t^′ for t′   [0, t₁]. Thus, we can rewrite both the measure and the state in (50) as

(52)

For SLE, a_t is proportional to a Brownian process. The integration over realisations of this for t′   [0, t₁] may be performed by breaking up the time interval into small segments of size δt, expanding out the exponential to O (δt), using (B_δt)² ≈ δt, and re-exponentiating. The result is

(53)

But, as we argued earlier, |h_t is independent of t, and therefore

(54)

This means that the descendant states L₋₂|h and are linearly dependent. We say that the Virasoro representation corresponding to |h has a null state at level 2. From this follow an number of important consequences. Acting on (54) with L₁ and L₂, and using the Virasoro algebra (48) and the fact that L₁|h  = L₂|h  = 0 while L₀|h  = h|h , leads to:

(55)

(56)

These are the fundamental relations between the parameter κ of SLE and the data of CFT. The conventional notation h_2,1 comes from the Kac formula in CFT which we do not need here. In fact this is appropriate to the case κ < 4: for κ > 4 it corresponds to h_1,2. (To further confuse the matter, some authors reverse the labels.) Note that the boundary exponent h parametrises the failure of locality in (23). From CFT we may also deduce that, with respect to the path integral measure, the probability that a curve connects small intervals of size about points r₁, r₂ on the real axis behaves like

(57)

Such a result, elementary in CFT, is difficult to obtain directly from SLE in which the curves are conditioned to begin and end at given points.

Note that the central charge c vanishes when either locality (κ = 6) or restriction (κ = 8/3) hold. These cases correspond to the continuum limit of percolation and self-avoiding walks, respectively, corresponding to formal limits Q → 1 in the Potts model and n → 0 in the O (n) model for which the unconditioned partition function is trivial.