Bulk flows in Virasoro minimal models with boundaries
Stefan Fredenhagen, Matthias R. Gaberdiel, Cornelius Schmidt-Colinet
aa r X i v : . [ h e p - t h ] O c t AEI-2009-063
Bulk flows in Virasoro minimal models with boundaries
Stefan Fredenhagen , Matthias R. Gaberdiel and Cornelius Schmidt-Colinet Max-Planck-Institut f¨ur Gravitationsphysik, Albert-Einstein-InstitutD-14424 Golm, Germany , Institut f¨ur Theoretische Physik, ETH Z¨urichCH-8093 Z¨urich, Switzerland
Abstract
The behaviour of boundary conditions under relevant bulk perturbations is studied for the Virasorominimal models. In particular, we consider the bulk deformation by the least relevant bulk fieldwhich interpolates between the m th and ( m − st unitary minimal model. In the presence of aboundary this bulk flow induces an RG flow on the boundary, which ensures that the resultingboundary condition is conformal in the ( m − st model. By combining perturbative RG techniqueswith insights from defects and results about non-perturbative boundary flows, we determine theendpoint of the flow, i.e. the boundary condition to which an arbitrary boundary condition of the m th theory flows to. E-mail: [email protected] E-mail: [email protected] E-mail: [email protected]
Introduction
Perturbations of conformal field theories by marginal or relevant operators play an important role invarious contexts, for example in string theory where they describe string moduli or time dependentprocesses. Many aspects of perturbed conformal field theories have been studied over the years,starting from the seminal work of Zamolodchikov [1, 2], and a number of structural results areknown, in particular the c -theorem for bulk perturbations [3], as well as the g -theorem for boundaryperturbations [4, 5].Most of the work so far has been done on bulk perturbations of bulk conformal field theories,or boundary perturbations of boundary conformal field theories. However, it is clear that a bulkperturbation will also affect the boundary condition since the boundary condition of the originaltheory will typically not be conformal with respect to the new bulk conformal fixed point. Thiscombined problem has only recently been addressed from the point of view of perturbed conformalfield theory [6, 7, 8] (see also [9, 10]), although there has been earlier work in the context of integrablemodels starting from [11] and further developed in [12, 13, 14]. In particular, these flows have beenstudied using a version of the thermodynamic Bethe ansatz (see for example [15, 16, 17, 18, 19]),the truncated conformal space approach (see for example [17, 18, 20]) and a form factor expansion[21, 22]. From the point of view of perturbed conformal field theory, a bulk perturbation genericallyinduces a boundary renormalisation group (RG) flow that will ensure that at the endpoint of theflow both bulk and boundary are again conformal [6, 7, 8].The resulting coupled RG equations have so far only been worked out for a few simple examples.In all of them, the bulk perturbation was actually exactly marginal in the bulk. As a consequencethe bulk RG equation was trivial, and one only had to solve the boundary RG equation with a(bulk) source term. In this paper we shall demonstrate that these techniques also work nicely for agenuinely coupled bulk boundary problem, where neither of the perturbations is marginal.The archetypal examples for which these kinds of problems can be studied are the Virasorominimal models. Indeed, the original analysis of Zamolodchikov [1] was performed in this context:he established that the perturbation of the m th minimal model by the least relevant field, the bulkfield φ (1 , , induces an RG flow whose endpoint is the ( m − st minimal model. The analogousanalysis for the boundary perturbation — the perturbation of a Cardy boundary condition [23] ofthe m th minimal model by the ψ (1 , boundary field — was done in [24]. They showed that theendpoint of this boundary flow is in general a superposition of fundamental boundary conditions.For the combined problem, the bulk perturbation by φ (1 , in the presence of a boundary, onlya few numerical studies have been performed so far [25, 26], and some conjectural TBA resultsexist [16, 27]. In this paper we will fill in this gap, and show how this problem can be analysedanalytically.Let us briefly sketch our argument. The bulk RG equation is unaffected by the presence ofthe boundary, and thus the old fixed point analysis of Zamolodchikov applies. The boundary RGequation, on the other hand, is of the form˙ µ = (1 − h ) µ + B λ + D µ + E λµ + F λ + O ( µ , λµ , λ µ, λ ) . (1.1)Here the first and third term are the usual boundary RG equation terms for the boundary coupling µ , while the other three terms involve also the bulk coupling constant λ . The Bλ term is the sourceterm that was studied in [6], whereas the Eλµ term describes how the bulk deformation modifiesthe conformal weight of the boundary field [8]. In some sense this term only appears at higher orderin perturbation theory, and it was only recently understood how to calculate it as an integral of a2 €€€€€ m A €€€€€€€€€ Μ Λ * Λ A €€€€€ m A €€€€€€€€€ Μ Λ * Λ H I LH II LH III L to (IV) αm α mαm α m Figure 1: The combined flow diagram for ( a , a ) = (2 ,
3) and m = 100 (for which α = −
4; seesection 2.2 for details). We have magnified the vectors ( ˙ µ, ˙ λ ) by a factor 2 .
5. The horizontal arrowindicates the pure boundary flow to the perturbative fixed-point (I) in the m th minimal model, thevertical arrow describes the flow of the boundary condition (I) to the boundary condition (II) inthe ( m − st minimal model. The three other flows that are depicted are generic bulk-boundaryRG-flows that at the end tend towards the fixed point (IV) where µ = + ∞ in the ( m − st minimalmodel ( λ = λ ∗ ).chiral four-point function [8]. The F λ term has not so far been studied in detail, but the coefficient F itself is parametrically small — it is of order 1 /m — and thus the F λ term is subleading. Infact, our analysis will be performed for large m , for which we shall find perturbative fixed pointsfor λ and µ that are of order 1 /m . The Eλµ term is then of the same order as the standard Dµ term, while the F λ will be subleading and can be ignored to leading order.The resulting RG equations have generically three perturbative fixed points: the pure boundaryperturbation fixed point of [24], as well as two perturbative fixed points in the ( m − st theory ( i.e. for non-trivial λ ). The first fixed point (I) can be identified as in [24], namely by computing theperturbed g -function and identifying it with the g -function of the fixed point boundary condition.However, the analysis for the other two fixed points (II & III) is not so straightforward since they livein a different bulk theory than the one we started with, and it is therefore not clear to which extentwe can compare the g -functions directly. However, it is reasonable to assume that it makes sense tocompare ratios of g -functions [18], and it is furthermore plausible that the boundary condition withthe overall smallest g -function — this is the boundary condition that corresponds to the identityrepresentation — should flow to the corresponding boundary condition in the ( m − st theory.This tells us how the overall scale of the g -functions changes, and thus allows us to make a definiteprediction for the g -function of the fixed point in the ( m − st minimal model. Progressing in thismanner, we can then identify the two perturbative fixed points in the ( m − st theory. As it turnsout, one of the fixed points (III) is actually the end-point of a pure ψ (1 , boundary perturbation of3he other (II) [28, 29], in agreement with the general structure of our RG flow diagram (see figure 1).As is clear from this diagram neither of these perturbative fixed points can be reached by ageneric RG flow: we can only get to the unstable fixed point (II) if we first perform the pureboundary flow to (I), followed by a pure bulk flow, and we can only get to (III) via (II). However,we can read off the actual end-point of a generic flow from this picture: it is the end-point of thepure boundary flow starting from (II), but flowing in the opposite direction to (III). The resultingfixed point is therefore the non-perturbative fixed point of a certain boundary perturbation of (II).At least in some specific cases this fixed point has been identified before using TCSA and TBAtechniques [30, 16, 27]; applying these results we can therefore make a prediction for the actualfixed point (IV) of our RG flow at least for some restricted set of initial boundary conditions.In order to determine the corresponding fixed point for an arbitrary initial boundary conditionwe finally use techniques from the perturbation theory of topological defects. In particular, we canidentify the RG fixed point of a certain topological defect from the above boundary flow results(that we know for a restricted set of initial conditions). We can then use the RG flow behaviour ofthis topological defect in order to make a prediction for the ultimate (IV) fixed point of an arbitraryinitial boundary condition. The resulting prediction — see eq. (4.20) — is the main result of ourpaper. As we shall demonstrate it satisfies a number of consistency conditions; in particular, theflows for different initial boundary conditions actually organise themselves into long chains (seefigure 2), some of whose individual flows were known before. In addition, we are able to confirmthis prediction, for a particular class of boundary conditions (namely those near the middle of theKac table), by a direct perturbative calculation — see section 4.4. As a final check of our prediction,we extrapolate it to small values of m , and show that our results are consistent with the numericalresults of [25, 26].The paper is organised as follows. The coupled RG equations are worked out in section 2.1.In section 2.2 we identify the various perturbative fixed points and study the structure of theRG flow diagram. To support the identification of the fixed points we then perform a detailedanalysis of the perturbed g -function in section 3. In section 4 we combine these results withinsights from non-perturbative flows and constraints coming from the action of defects to give acomplete picture of the flow diagram, including also the actual (non-perturbative) fixed point ofthe original flow (for a generic initial condition). We also discuss there various consistency checkswhich our analysis satisfies. Finally, we compare our findings with the numerical study of [25, 26] insection 5. There are three appendices, where we collect the leading behaviour of the various OPEcoefficients (appendix A), give explicit formulae for the bulk and boundary correlation functionson the upper half-plane and the disc (appendix B), and analyse the explicit solutions to the RGequations (appendix C). Let us begin by reviewing our conventions. We shall consider the unitary minimal models withcentral charge c m = 1 − m ( m + 1) , m = 3 , , , . . . . (2.2)More specifically, we shall always work with the diagonal (charge conjugation) modular invarianttheory, for which the left- and the right-moving representations are identical. For c = c m , the4llowed irreducible highest weight representations of the Virasoro algebra have highest weight h ( r,s ) = (cid:0) ( m + 1) r − ms (cid:1) − m ( m + 1) , (2.3)where 1 ≤ r ≤ m − ≤ s ≤ m , and we have the identifications ( r, s ) ∼ = ( m − r, m + 1 − s ).For the charge conjugation theory, the conformal primary fields φ i are labelled by i ≡ ( r, s ), andthe conformal dimension of φ i is ∆ i = 2 h i .We shall be interested in the conformal field theory defined on a Riemann surface with boundary,more specifically the upper half plane (or equivalently the disc). In order to characterise the theoryon a surface with boundary we also have to specify the boundary conditions for the various fields.For the diagonal modular invariant theories, the possible conformal boundary conditions are alsoparameterised by the highest weight representations of the Virasoro algebra. Thus the most generalconformal boundary condition is a superposition of boundary conditions associated to the irreduciblerepresentations labelled by ( a , a ), where a and a have the same ranges and identifications as r and s in (2.3). It is sometimes convenient to describe the boundary condition in terms of theassociated boundary state; for the boundary condition a = ( a , a ) the boundary state is given bythe Cardy formula [23] k a ii ≡ k a , a ii = X ( r,s ) S ( r,s )( a ,a ) q S ( r,s )(1 , | r, s ii . (2.4)Here | r, s ii denotes the Ishibashi state [31] in the ( r, s ) sector, and S ( r,s )( a ,a ) is the modular S -matrix,whose entries are explicitly given as S ( r,s )( a ,a ) = s m ( m + 1) ( − a s + a r sin (cid:0) m +1 m πa r (cid:1) sin (cid:0) mm +1 πa s (cid:1) . (2.5)In the presence of a boundary there are also excitations located at the boundary. These are describedby boundary fields ψ i , and they are again characterised by unitary representations of the Virasoroalgebra, i ≡ ( r, s ). The conformal dimension of the boundary field ψ i is given by the weight h i asin (2.3). On the boundary condition a , the possible boundary fields are those that appear in thefusion rules of a with itself.In the following section we shall work on the upper half-plane, where we denote the operatorproduct expansions (OPEs) of the bulk and boundary fields as φ i ( z , ¯ z ) φ j ( z , ¯ z ) = X k C kij φ k ( z , ¯ z ) | z − z | ∆ k − ∆ i − ∆ j + · · · ,φ i ( z = x + iy, ¯ z ) = X k B ik ψ k ( x )(2 y ) h k − ∆ i + · · · , (2.6) ψ i ( x ) ψ j ( y ) = X k D kij ψ k ( y )( x − y ) h k − h i − h j + · · · ( x > y ) . We are interested in the perturbation of the conformal field theory by the least relevant bulk field φ (1 , of conformal weight h ≡ h (1 , = m − m +1 . As is well known, this perturbation induces an RG flowthat drives the minimal model c m to the one corresponding to c m − [1, 32]. In this paper we want5o study what happens if we consider this perturbation in the presence of a boundary. It is clearthat the presence of the boundary will not affect the flow in the bulk, i.e. that we are still endingup with the minimal model corresponding to c m − . However, it is not so clear what happens to theboundary condition ( a , a ) of the c m theory under the RG flow; this is the question we want toaddress in the following.In the presence of a boundary, the RG flow will also switch on boundary fields, and we shouldtherefore consider the general perturbation δS = X k λ k ǫ ∆ k − Z d z φ k ( z, ¯ z ) + X l µ l ǫ h l − Z dx ψ l ( x ) . (2.7)Here the λ k and the µ l are (small) dimensionless coupling constants, and ǫ is an ultraviolet cut-off.As has been studied before, the combined RG equations are [6, 8]˙ λ k = (2 − ∆ k ) λ k + X ij πC kij λ i λ j + O ( λ ) , (2.8)˙ µ l = (1 − h l ) µ l + X i B il λ i + X ij E lij λ i µ j + X ij D lij µ i µ j + X ij F lij λ i λ j + O ( µ , λ µ, λµ ) , where ˙ λ stands for dλ/d log ǫ , and similarly for ˙ µ . In the OPE scheme that we shall consider inthe following, the coefficients C kij , B il and D lij are those from the OPEs (2.6). Furthermore, thecoefficient E lij can be calculated in terms of an integral of a four point function [8] (see below formore details), while the coefficient F lij comes from the correlation function of two bulk fields andone boundary field. As will be explained below, this contribution is only subleading.Note that we have written out explicitly all terms that are of degree less or equal to two in λ and µ . We are interested in a perturbative fixed point for λ and µ , for which both are of order1 /m . The terms we have spelled out are therefore all the terms that contribute up to order 1 /m .In particular, it is important that we also include the λµ term for the µ RG equation since it is ofthe same order as the µ term. (In fact, if one leaves it out, one does not find any perturbativefixed point for µ .)The perturbation by the (1 ,
3) bulk field is particularly tractable since the successive OPEs of φ (1 , with itself only contain one relevant field apart from the identity, namely φ (1 , itself. As aconsequence we can restrict the RG equation to λ ≡ λ (1 , , and the bulk equation is therefore˙ λ = (2 − h ) λ + πC (1 , , , λ + · · · . (2.9)This leads to the well-known fixed point [1, 32] λ ∗ = − πm + O ( m − ), where we have used theleading behaviour of the OPE coefficients as given in appendix A.In the presence of a boundary a = ( a , a ), the bulk perturbation also induces boundary pertur-bations, and the only relevant boundary field (apart from the identity) that is switched on is theboundary ψ (1 , field. Furthermore, successive OPEs of ψ (1 , generate, apart from the identity field,only one relevant field, namely ψ (1 , itself. It is therefore again consistent to restrict our attentionto this field. Writing µ = µ (1 , , the RG equations are˙ µ = (1 − h ) µ + B (1 , , λ + E (1 , , , λµ + D (1 , , , µ + F λ + higher order . (2.10) To leading order in 1 /m , the calculation is actually the same as in a minimal subtraction scheme that is moreconvenient for the calculation of the perturbed g -function. Obviously, this only makes sense if ψ (1 , appears in the boundary spectrum of the boundary condition a =( a , a ); this is the case provided that 1 < a < m . B and D [33] are given in appendix A.1, while E can becalculated following [8] as E (1 , , , = lim ǫ → Z dx θ ( ǫ − | x | ) (cid:20) h i D (1 , , , (cid:10) φ (1 , ( x + i , x − i ) ψ (1 , (0) ψ (1 , ( ∞ ) (cid:11) − B (1 , , D (1 , , , − B (1 , , D (1 , , , θ ( | x | − | x | h (cid:21) . (2.11)The correlator in the first line of this expression is a chiral four-point function which is in principledetermined by the differential equation that comes from the null vector descendant of the highestweight state with h = h (1 , . Unfortunately, this differential equation does not seem to have a simplesolution. However, we are only interested in the chiral four point function to leading order in 1 /m ,and in this limit we find1 h i D (1 , , , (cid:10) φ (1 , ( z, ¯ z ) ψ (1 , (0) ψ (1 , ( ∞ ) (cid:11) = ( z − ¯ z ) − ( z + ¯ z ) | z | ( z − ¯ z ) + O ( m − ) . (2.12)Some remarks on the calculation can be found in appendix B. Note that this function has the correctasymptotic behaviour as z approaches the boundary away from the origin, because in this limit itgoes as ∼ y , which is the expected behaviour since (see appendix A.1) B (1 , , D (1 , , , = 3 + O ( m − ) . (2.13)The first term in the second line in (2.11) subtracts precisely this leading term. On the other hand,the other channels do not contribute to leading order in 1 /m since (see again appendix A.1) B (1 , , D (1 , , , = O ( m − ) , B (1 , , D (1 , , , = O ( m − ) . (2.14)In particular, the integral thus converges, and we find explicitly the rather simple result E (1 , , , = 12 Z ∞−∞ dx (cid:20) x − x − x ) − (cid:21) + O ( m − ) = − π + O ( m − ) . (2.15)Note that the leading order result (in 1 /m ) is finite, and apparently scheme independent. This tiesin with the general observation of [8] that E is universal provided that the resonance condition issatisfied, which is here the case to leading order in 1 /m .Finally, the term proportional to F is subleading relative to these terms since it arises from thecorrelation function of two bulk fields and one boundary field. This correlation function is of theasymptotic form1 h i D (1 , , , (cid:10) φ (1 , ( z , ¯ z ) φ (1 , ( z , ¯ z ) ψ (1 , ( ∞ ) (cid:11) ∼ X i , j B (1 , i B (1 , j D ij (1 , f ij ( z , ¯ z , z , ¯ z ) , (2.16)where the f ij ( z , ¯ z , z , ¯ z ) are some functions that give the asymptotic dependence on the insertionpoints, and i , j = (1 , , (1 , , (1 , /m . The total contribution of the F λ term is therefore subleading relative to the other terms.7 .2 Analysis of fixed points Putting everything together, we thus have the coupled RG equations˙ λ = 4 m λ + 4 πλ + O ( λ )˙ µ = 2 m µ + 2 παm λ − πλµ − α µ + O ( λ µ, λµ , µ ) , (2.17)where α = ( a − a )( a + 1) a > a ( a − m a = a ( a − a )( a − a > a > . (2.18)These equations hold provided that a >
1. Otherwise, the boundary theory does not contain therelevant ψ (1 , field, and there is therefore no equation for ˙ µ .Apart from the trivial fixed point ( λ = µ = 0), these equations have the following three fixedpoints (for a > λ ∗ = 0 , µ ∗ = α m . (2.19)This is simply the perturbative fixed point in the pure boundary analysis of [24] (see also[28]). As was explained there, it describes the flow( a , a ) m −→ min( a ,a ) M l =1 ( a + a + 1 − l, m . (2.20)The end-point of the flow is a boundary condition in the m th theory.(II) The fixed point at λ ∗ = − πm , µ ∗ = α m . (2.21)The interpretation of this fixed point will be determined in detail in section 3, where we willshow that it describes the superposition of boundary conditions in the ( m − st theory( a , a ) m −→ min( a ,a ) M l =1 (1 , a + a + 1 − l ) m − . (2.22)(III) The fixed point at λ ∗ = − πm , µ ∗ = αm . (2.23)As we will also show in section 3, this fixed point describes the end-point of a perturbativeboundary flow in the ( m − st theory, starting from the boundary condition at (II). Theend-point describes the boundary condition( a , a ) m via (II) −→ ( a , a ) m − , (2.24)8n agreement with a boundary flow of [28], where it was observed that by turning on all ψ (1 , boundary condition changing fields one can find the following two perturbative flows, min( a ,a ) M l =1 (1 , a + a + 1 − l ) m − −→ (cid:26) ( a , a ) m − ( a , a ) m − (2.25)(see eqs (5.29) and (5.30) in [28]).The above analysis applies to the case when both labels a and a are small, and a >
1. If a = 1, on the other hand, there is just the RG flow for λ , whose fixed point is λ ∗ = − πm . As willalso be explained in section 3, it describes the flow( a , m −→ (1 , a ) m − . (2.26)It is easy to see from the flow diagram in the introduction (see figure 1) that for a > λ = µ = 0we do not get to the fixed point (II), unless we first perform the pure boundary flow leading to(I), followed by the pure bulk flow (2.26). To reach the fixed point (III) we first have to go to (II)via (I), and then have to switch on a pure boundary perturbation at (II). Indeed, the fixed point(II) is again unstable since the boundary condition (2.22) has at least one relevant ψ (1 , field inits spectrum. This can also be seen by expanding the RG equation around the fixed point (II) bysetting λ = λ ∗ and µ = α/ m + ˜ µ , ˙˜ µ (cid:12)(cid:12)(cid:12) λ = λ ∗ = 2 m ˜ µ − α ˜ µ + · · · . (2.27)The coefficient of the term linear in ˜ µ allows us to read off the conformal weight of the boundaryfield to which ˜ µ couples, and one finds indeed h ˜ µ = h (1 , = 1 − m + · · · .Given the various different kinds of flows we can consider, our resulting picture will have tosatisfy a number of consistency conditions. These will be discussed in section 4, where we shallalso analyse the actual non-perturbative fixed point (IV) to which a generic initial configurationwill flow. Before we discuss these issues, let us first analyse the perturbed g -function in order toidentify the different perturbative fixed points. g -function In order to corroborate our above claims about the perturbative fixed points, we shall now calculatethe perturbed boundary entropy, as was done for the case of the pure boundary perturbation in[24]. In order to be able to compare with their results, we shall now work on the disc.Recall that the boundary entropy g ( a ) of a boundary condition a is defined to be the correctlynormalised one-point function of the identity operator in the presence of the boundary condition a [4] g ( m ) ( a ) = S a p S = (cid:18) m ( m + 1) (cid:19) sin πa m sin πa m +1 (sin πm sin πm +1 ) . (3.1)It was conjectured in [4] and perturbatively verified in [43, 24] that this quantity decreases underpure boundary RG flows. Obviously, the same need not be true in our context, since our perturbation9lso changes the bulk theory, and we are therefore comparing one-point functions in different bulkmodels [18] (see also [44]).We should thus not expect to be able to say much about the overall g -functions. On the otherhand, the relative g -functions should continue to have a well-defined meaning (see also the discussionin [18]). Furthermore, we expect that the boundary condition a = (1 ,
1) should flow to itself, sincethe boundary spectrum of the (1 ,
1) boundary does not contain any relevant fields (apart from theidentity), and since (1 ,
1) is the boundary condition with the smallest g -function. Thus it is naturalto consider the relative boundary entropy with respect to a = (1 , g ( m ) ( a ) = g ( m ) ( a ) g ( m ) (1 ,
1) = sin πa m sin πa m +1 sin πm sin πm +1 . (3.2)For the following it will be important to consider the asymptotic expansion of the g -function forlarge m with fixed boundary labels a , a ,ˆ g ( m ) ( a ) = a a (cid:18) − π (cid:18) a + a − (cid:19) m + π (cid:18) a − (cid:19) m + O ( m − ) (cid:19) . (3.3)Our aim is therefore to calculate the perturbed g -function of the boundary condition a up to order1 /m , and to deduce from it the perturbed value of the relative g -function (up to this order). Thisshould then be identified with the relative g -function (in the c m − theory) of the boundary conditionto which a flows to. Let us first deal with the case where a >
1, so that the original boundarycondition has a relevant boundary field in its spectrum. a > Since we are interested in ratios of g -functions, it is convenient to consider the logarithm of theperturbed g -functionlog (cid:10) e δS (cid:11) a = log h i a + 1 h i a (cid:18) µ ∗ ǫ h − Z dw dw h ψ ( w ) ψ ( w ) i c a (3.4)+ 16 µ ∗ ǫ h − Z dw dw dw h ψ ( w ) ψ ( w ) ψ ( w ) i c a + λ ∗ ǫ h − Z d u h φ ( u, ¯ u ) i c a + λ ∗ µ ∗ ǫ h − Z d u dw h φ ( u, ¯ u ) ψ ( w ) i c a + 12 λ ∗ µ ∗ ǫ h − Z d u dw dw h φ ( u, ¯ u ) ψ ( w ) ψ ( w ) i c a + · · · (cid:19) . Here the suffix c at the correlators indicates that we are only considering the connected compo-nents; for the terms that we have written explicitly above, this only makes a difference for the lastcontribution.In order to identify the fixed points we need to evaluate the perturbed g -function up to order1 /m . Since both λ ∗ and µ ∗ are of order 1 /m , we only need to consider terms that are at mostof cubic order in these coupling constants. We have written out explicitly all such terms exceptthose that are proportional to λ . The reason for this is that to order 1 /m , they turn out not to To leading order in 1 /m the bare and the renormalised coupling constants agree, and we we can therefore directlyuse the above fixed points (for the renormalised coupling constants) here. a , a ), and therefore will not contribute to the relative entropyat order 1 /m . This can be seen from considering the possible asymptotics of the respective cor-relators. For the correlator of two bulk fields in the λ term there are the asymptotic channels i , j = (1 , , (1 , , (1 , h φ ( u , ¯ u ) φ ( u , ¯ u ) i a ∼ X i , j B (1 , i B (1 , j D (1 , ij h i a f ij ( u , ¯ u , u , ¯ u ) , (3.5)where the f ij are some functions that give the asymptotic dependence on the insertion points,compare (2.16). With the OPE constants from appendix A.1 one can see that all channels contributeboundary-dependent terms only at order 1 /m , so that the whole contribution of the λ term willbe of order 1 /m . For the terms with coupling constants λ µ and λ , a similar analysis shows thattheir contributions will only affect the order 1 /m as well.The above integrals are not well defined, and we need to introduce a regularisation scheme tomake sense of them. In each case the leading term in 1 /m that is dependent on the boundarylabels ( a , a ) turns out to be of order 1 /m , and thus we are effectively working to leading order in1 /m . In particular, we can therefore take the large m expansion of the integrand before we do theintegral. This integral will be regularised by some cut-off scheme; in particular, we shall introducea cut-off ǫ to separate the boundary fields from one another, and a cut-off ξ to restrict the radialbulk integration from 0 ≤ r ≤ − ξ . We shall then discard the terms proportional to ǫ − , asthey describe non-universal terms (that have the wrong scaling behaviour). We shall also impose asimilar procedure for the terms that are singular in ξ → /m thequantities we calculate should be universal, and thus this distinction should not play a role; thisexpectation will be borne out by our results. We have also checked this explicitly for some of theterms; the advantage of the cut-off scheme we are using here is that the calculations are muchsimpler since we do not need to know the integrand for arbitrary m (but only in the m → ∞ limit).We shall now discuss the various terms in turn. The first two terms are the pure boundaryintegrals that were already considered in [24]. The boundary contribution proportional to µ ∗ involvesthe integral I = Z dθ dθ (cid:12)(cid:12)(cid:12)(cid:12) θ − θ (cid:12)(cid:12)(cid:12)(cid:12) − h = π Z π − ǫǫ dθ sin − θ (cid:0) θ (cid:1) sin θ m + O ( m − ) ! = 2 π (cid:18) cot ǫ m (cid:16) ǫ − π + cot ǫ (cid:16) ǫ (cid:17)(cid:17) (cid:19) + O ( m − )= 4 πǫ + (cid:18) π (1 + log ǫ ) ǫ − π (cid:19) m + O ( ǫ, m − ) . (3.6)The first nontrivial contribution of the integral is hence of order 1 /m , and we obtain the contribution(dropping the non-universal term proportional to 1 /ǫ )12 µ ∗ ǫ h − Z dw dw h ψ ( w ) ψ ( w ) i c a = − π m D (1 , , , h i a µ ∗ + O ( m − )= − π m (cid:16) µ ∗ α (cid:17) h i a ( a −
1) + O ( m − ) , (3.7)11here the correction term is of order m − for all fixed points of interest. The boundary contributiondepending on µ ∗ involves the integral I = Z dθ dθ dθ (cid:12)(cid:12)(cid:12)(cid:12) θ θ θ (cid:12)(cid:12)(cid:12)(cid:12) − h = 4 π Z θ >θ dθ dθ (cid:18) θ θ θ (cid:19) − + O ( m − ) . (3.8)If we introduce cut-offs only where necessary, so that the integral does not diverge, we find I = 8 π cot ǫ − π + O ( m − ) = 16 πǫ − π + O ( ǫ, m − ) . (3.9)Again dropping the non-universal 1 /ǫ term, we get the contribution16 µ ∗ ǫ h − Z dw dw dw h ψ ( w ) ψ ( w ) ψ ( w ) i c a = 2 π D (1 , , , D (1 , , , h i a µ ∗ + O ( m − )= 16 π (cid:16) µ ∗ α (cid:17) h i a ( a −
1) + O ( m − ) . (3.10)The contribution which depends linearly on λ ∗ and is independent of µ ∗ involves the integral I = Z dr r (1 − r ) − h = Z − ξ dr r (1 − r ) − + O ( m − )= (1 − ξ ) ξ (2 − ξ ) + O ( m − ) = 18 2 − ξξ + O ( ξ, m − ) . (3.11)This yields λ ∗ ǫ h − Z d u h φ ( u, ¯ u ) i c a = λ ∗ h i a B (1 , , (cid:18) π − ξξ + O ( ξ, m − ) (cid:19) = λ ∗ h i a (cid:18) − π ( a − m − ξξ + f ( m, ξ ) + O ( ξ, m − ) (cid:19) , (3.12)where f ( m, ξ ) is some function which does not depend on the boundary labels and has the large- m asymptotic behaviour f ( m, ξ ) = 3 π − ξξ + O ( ξ, m − ) . (3.13)The contribution that depends linearly on λ ∗ and µ ∗ involves the integral I = Z drdθ dθ r ((1 − r )(1 − r cos( θ − θ ) + r )) − h = Z − ξ dr π r (1 − r ) + O ( m − ) = π − ξξ + O ( ξ, m − ) , (3.14)leading to λ ∗ µ ∗ ǫ h − Z d u dw h φ ( u, ¯ u ) ψ ( w ) i c a = λ ∗ µ ∗ h i a B (1 , , D (1 , , , (cid:18) π − ξξ + O ( ξ, m − ) (cid:19) = λ ∗ (cid:16) µ ∗ α (cid:17) h i a π m ( a − (cid:18) − ξξ + O ( ξ, m − ) (cid:19) (3.15)12inally, the integrand for the term of order λ ∗ µ ∗ is the same as in the calculation of the coefficient E in the RG equation of section 2. Deferring the details of this calculation to appendix B theresulting contribution turns out to be12 λ ∗ µ ∗ ǫ h − Z d u dw dw h φ ( u, ¯ u ) ψ ( w ) ψ ( w ) i c a = π λ ∗ µ ∗ h i a B (1 , , D (1 , , , D (1 , , , (cid:18) − π − ξξ + O ( ξ ) + · · · (cid:19) = − λ ∗ (cid:16) µ ∗ α (cid:17) h i a π ( a −
1) 2 − ξξ + O ( ξ, m − ) , (3.16)where the ellipses in the second line refer to terms of order O ( ǫ, m − ). Adding all the relevant termstogether we then arrive atlog (cid:10) e δS (cid:11) a = log h i a + ( a − h − π m (cid:16) µ ∗ α (cid:17) + 16 π (cid:16) µ ∗ α (cid:17) + λ ∗ − ξξ (cid:16) − π m + 4 π m (cid:16) µ ∗ α (cid:17) − π (cid:16) µ ∗ α (cid:17) (cid:17)i + · · · , (3.17)where the ellipses either denote terms that are independent of the boundary labels, or terms thatare of order O ( m − ) for the fixed points of interest. The first two coefficients (that are independentof λ ∗ reproduce exactly what was found in [24]. Using their results it thus follows that the fixedpoint (I) is indeed the perturbative fixed point of the pure boundary perturbation. At the fixed point (II), we have µ ∗ α = m , and it follows from (3.17) that the last three terms (thatare all proportional to λ ∗ ) cancel identically. The perturbed g function (3.17) is then given bylog g ( m ) λ ∗ ,µ ∗ ( a ) = log g ( m ) ( a ) − π m ( a −
1) + f ( m ) + O (cid:0) m − (cid:1) , (3.18)where f ( m ) is a function which is at least of order 1 /m and does not depend on the boundarylabels. Subtracting the corresponding expression for the boundary entropy of the (1 ,
1) boundarycondition, we thus find that the perturbed relative entropy ˆ g equalsˆ g ( m ) λ ∗ ,µ ∗ ( a ) = ˆ g ( m ) ( a ) (cid:18) − π a −
1) 1 m + O ( m − ) (cid:19) = a a (cid:18) − π a + a −
2) 1 m + O ( m − ) (cid:19) , (3.19)where in the final line we have used the asymptotic expansion (3.3) for ˆ g ( m ) ( a ).This is now to be compared with the relative entropy of the boundary condition b = ( b , b ) inthe c m − theory,ˆ g ( m − ( b ) = b b (cid:18) − π b + b −
2) 1 m − π b −
1) 1 m + O ( m − ) (cid:19) . (3.20)If a = 1, (3.20) equals (3.19) for a single fundamental boundary condition ba = (1 , a ) m (II) −→ b = (1 , a ) m − . (3.21)13n the general case, we cannot solve for (3.19) = (3.20) with a single boundary condition b = ( b , b ). As in [24] we therefore consider superpositions of fundamental boundary conditions B = N M l =1 b l , b l = ( b l , b l ) . (3.22)The entropy of the superposition is just the sum of the individual entropies, and we thus get theequations X l b l b l = a a , X l b l b l (cid:16) ( b l ) + ( b l ) − (cid:17) = a a (cid:16) a + a − (cid:17) , (3.23) X l b l b l (cid:16) ( b l ) − (cid:17) = 0 . The last equation implies that b l = 1 for all l , and the equations are generically solved by N = min( a , a ) , b l = 1 ∀ l , b l = a + a + 1 − l . (3.24)This suggests flows of the form a = ( a , a ) m (II) −→ B = min( a ,a ) X l =1 (1 , a + a + 1 − l ) m − . (3.25)These flows have to be understood as sequences of flows (first a pure boundary flow to (I) followedby the flow to (II)), as was discussed at the end of section 2 (see also figure 1). Note that the resultof (3.25) is similar to what happened in the case of a pure boundary perturbation [24], except thatthere the end-points of the boundary flow were superpositions of boundary conditions ( b l , , b l ).We should mention that the g function does not allow us to determine the resulting boundaryconditions uniquely, since ˆ g ( m − ( b , b ) = ˆ g ( m − ( b , m − b ) . (3.26)One can partially fix this ambiguity as in [24]. To zeroth order in perturbation theory, we knowthat the ( r, s ) bulk field of the m th theory becomes the ( s, r ) bulk field of the ( m − st theory [1].Thus to leading order in 1 /m we need to have that (cid:16) a B (1 , r,s ) (cid:17) ( m ) = (cid:16) B B (1 , s,r ) (cid:17) ( m − . (3.27)Since (cid:16) ( a ,a ) B (1 , r,s ) (cid:17) ( m ) = ( − ( r + s )( a + a ) sin (cid:0) ra πm (cid:1) sin (cid:0) sa πm +1 (cid:1) sin (cid:0) a πm (cid:1) sin (cid:0) a πm +1 (cid:1) , (3.28)this requires, in particular, that ( − ( r + s )( a + a ) = ( − ( r + s )( b l + b l ) (3.29)for all ( r, s ) and all l . This is evidently satisfied by our ansatz, but would not in general be true ifwe replaced some ( b l , b l ) by ( b l , m − b l ). In [44] a similar calculation was done to lower order in 1 /m , from which the authors concluded that the flow issimply ( a , a ) → ( a , a ). This is compatible with the analysis to order 1 /m , but not to order 1 /m . .1.2 Fixed point (III) The analysis for the fixed point (III) is more complicated, since now the λ ∗ term in the last lineof (3.17) does contribute. We therefore need to understand how to deal with the singular part as ξ →
0. We do not have a fundamental understanding of how this must be done, but we shall nowpropose a procedure that will lead to consistent results. In fact, the procedure is already determinedby considering the consistency of the flow of the ( a ,
1) boundary condition (see section 4, as wellas the remarks below in section 3.2), and the fact that all the other consistency conditions thenalso work out (that will be explained in detail in section 4) makes us confident that this is indeedthe correct prescription. Since the cut-off ξ bounds the radial bulk integral to the region 0 ≤ r ≤ − ξ , it is actually ameasure of an area; in fact, the area of the missing annulus is simply π (1 − (1 − ξ ) ) = 2 πξ (1 − ξ/ i.e. π (1 − ξ ). We thus propose that thegood cut-off parameter is η = ξ (1 − ξ/ − ξ ) = ξ (1 + ξ/
2) + O ( ξ ) . (3.30)The expression of interest can therefore be written as2 − ξξ = 2 η − O ( η ) . (3.31)Our prescription is now that we should discard the non-universal 1 /η pole, but keep the constantterm. Using that at (III) µ ∗ α = m as well as λ ∗ = − πm then leads tolog g ( m ) λ ∗ ,µ ∗ ( a ) = log g ( m ) ( a ) − π m ( a −
1) + f ( m ) + O (cid:0) m − (cid:1) . (3.32)The perturbed relative entropy ˆ g then equalsˆ g ( m ) λ ∗ ,µ ∗ ( a ) = ˆ g ( m ) ( a ) (cid:18) − π a −
1) 1 m + O ( m − ) (cid:19) = a a (cid:18) − π a + a −
2) 1 m − π m ( a −
1) + O ( m − ) (cid:19) , (3.33)where in the final line we have used the asymptotic expansion (3.3) for ˆ g ( m ) ( a ). Comparing thiswith (3.20) we then find that the end-point should be given by( a , a ) m (III) −→ ( a , a ) m − . (3.34)This flow has to be understood as a sequence of asymptotic flows, similar to (3.25). a = 1 For a = 1 the analysis is simpler since the boundary condition does not have the relevant boundaryfield ψ (1 , in its spectrum, and thus we have no RG equation for µ . Of the terms in (3.4) hence only The prescription we propose actually has a very natural interpretation if we consider the theory on the semi-infinite cylinder (see appendix B.2). We thank Anatoly Konechny for discussions on this point. a , and since we are only interestedin the ratio of the perturbed g -function relative to a = (1 ,
1) it does not contribute to the rescaled g function in the ( m − st theory. Thus we conclude that the (rescaled) g -function does not changeat all. Using eq. (3.1) we find thatˆ g ( m ) ( a ,
1) = sin πa m sin πm = ˆ g ( m − (1 , a ) . (3.35)This then establishes (2.26). Note that this argument holds for arbitrary values of a , not necessarilysmall relative to m .Actually, since in this case no combined flow takes place, one may suspect that not only the ratioof g -functions is correctly reproduced by this analysis, but also the overall value of the g -function.As in the pure boundary case, let us consider the logarithmic change of the g function. The onlyterm that contributes in this case is the term (3.12) that contains a contribution of order 1 /m thatis independent of the boundary labels. Using (3.13) we thus have to first orderlog g ( m ) λ ∗ ,µ ∗ ( a , g ( m ) ( a ,
1) = − m − ξξ + O ( ξ, m − ) , (3.36)which we must compare with the expansion of the resulting boundary condition,log g ( m − (1 , a ) g ( m ) ( a ,
1) = 32 m + O ( m − ) . (3.37)These two expressions then agree precisely to this order if we use the same prescription as in (3.31). So far we have identified the perturbative fixed points by studying the perturbed g -function. Theactual fixed point (IV) of the RG flow for a generic initial condition however appears at λ ∗ = − πm and µ ∗ = + ∞ (for α < α > µ ∗ = −∞ ), as is clear from the flowdiagram (see figure 1). We now want to combine what we have found so far with results aboutnon-perturbative boundary flows in order to identify this actual (non-perturbative) fixed point. Inorder to tie it down completely, we shall also use some constraints that arise upon using defects.First of all, one way to get to (IV) is to turn on a ψ (1 , field on the boundary condition (II).Indeed, as we have shown above in (2.27) the coupling constant ˜ µ couples to a ψ (1 , field at the fixedpoint (II). Furthermore, the flow from (II) to (III) is the usual perturbative ψ (1 , flow, and thusthe flow from (II) to (IV) must be the corresponding non-perturbative flow (where the field ψ (1 , isswitched on with the opposite sign). The situation is particularly simple if the boundary spectrumof (II) only contains a single ψ (1 , field — then it is clear that this is the field to which ˜ µ must couple.This will be the case provided that we begin with a boundary condition of type (1 , a ) m , since thenthe perturbative fixed point (II) consists of a single boundary condition (1 , a ) m − that has a single ψ (1 , boundary field in its spectrum. The fixed point (III) is then the end-point ( a , m − of theperturbative boundary flow (provided that a is not too large) [24].In this case we can determine the end-point of the non-perturbative ψ (1 , flow, using the resultsof [30, 16, 27] that have been obtained using TCSA and TBA techniques (see also [34, 35]). Forsmall label a , the result is simply(1 , a ) m − −→ ( a − , m − . (4.1)16ummarising our results so far, we predict the end-points of the RG flows to be( a , m −→ (1 , a ) m − (4.2)(1 , a ) m −→ ( a − , m − for 1 < a ≪ m . (4.3)If we now want to consider the case of a general boundary condition ( a , a ) m , we face the problemthat we do not know a priori how to identify the boundary field that couples to ˜ µ . One way toproceed would be to argue that the flow in question must be the non-perturbative flow correspondingto the perturbative flow from (II) to (III) that we have identified above; this will, as we shall see,lead to the correct result. However, in the following we shall follow a different route by studyingthe action of topological defect lines. So far we have discussed how boundary conditions adjust themselves under bulk deformations. Wecan also use similar methods to analyse what happens to defect lines. Recall that defect linesinterpolate between in general different conformal field theories. In the following we shall onlyconsider defects where the theories on both sides are the same (namely some minimal model). Morespecifically, we shall consider topological defects that are characterised by the property that thecorresponding defect operator commutes with the action of both left- and right-moving Virasorogenerators. In particular, this implies that the defect only depends on the homotopy class of thedefect line [36]. By moving the defect line to the boundary ( i.e. by ‘fusion’), such topological defectsdefine then an action on the (conformal) boundary conditions of the theory.For the charge conjugation theories we are considering here, the topological defects are labelledby the same labels as the Cardy boundary states. We can therefore denote them as D ( d , d ), where d = ( d , d ) has the same range and identification rules as the boundary labels a = ( a , a ). Theaction of the defect D ( d ) on the boundary condition a is then described by the usual fusion rules D ( d ) × a = M b N dab b . (4.4)In addition to the identity defect D (1 , D ( m − , ∼ = D (1 , m )that generates a Z symmetry of the charge conjugation minimal model (such a defect is called a’group-like defect’ in [37] and a ’symmetry defect’ in [38]). Its action on a bulk field φ ( r,s ) is givenby D (1 , m ) : φ ( r,s ) ( − ( m +1) r − ms +1 φ ( r,s ) , (4.5)which is indeed a symmetry of the bulk theory. Note that the field φ (1 , that we use to perturb thetheory is invariant under this Z action. The effect of the Z symmetry on a boundary condition is( a , a )
7→ D (1 , m ) × ( a , a ) = ( a , m + 1 − a ) . (4.6)The concept of defect lines has proven very useful in the discussion of boundary flows [39, 40] andcombined bulk-boundary flows [38].After these preparations we now want to study what happens to these topological defects underthe bulk deformation where we perturb the bulk theory on both sides of the defect by the φ (1 , bulkfield. In principle, this can also be analysed by the above methods since we can think of the defectas a boundary condition in the doubled theory [41]. Typically, the bulk deformation will switch on17efect fields — the analogue of the boundary fields for boundary conditions — that will ensure thatat the end of the flow the defect is again conformal. In general, however, a topological defect willnot flow to a topological defect again, and thus the identification of the end-point will be difficult.There is, however, one particularly simple class of defects for which the analysis is essentiallytrivial. These are the defects D ( d , φ (1 , perturbation. In fact, the spectrum of a topological defect D ( d ) is given by H D ( d ) = M a , b , c N dcd N abc H a ⊗ H b . (4.7)The defect spectrum can be decomposed under the action of the two Virasoro algebras. Thus eachdefect field ψ ab carries two labels a = ( a , a ) and b = ( b , b ). For d = ( d , b = ( a + 2 n, a ) for some integer n . As the bulk field φ (1 , canonly turn on defect fields with a = b = 1 and a , b odd, we can only turn on the defect fields ψ (1 , p )(1 , p ) (for some integer p ) on D ( d , ψ (1 , , whose conformal weight is h = 2 h (1 , = 1 + m − m +1 . This is greater than1 for m >
3, and thus the field is irrelevant.This argument shows that no relevant defect fields are turned on in our case. This should thenimply that the end-point of the flow is again a topological defect. Under this assumption it is theneasy to determine to which topological defect D ( d , m flows in the ( m − st model. To see howthis goes let us consider a configuration with a defect line D ( d , m and the simplest boundarycondition (1 , m . Now we can either first act with the defect line on the boundary and then do thebulk perturbation, or we can first analyse the bulk perturbation when the defect and the boundaryare far apart, and then let the perturbed defect act on the perturbed boundary condition. If wefirst let the defect act on the boundary condition in the m th theory, we get the boundary condition( d , m , which under the bulk perturbation flows to (1 , d ) m − (see (4.2)). If we do the bulkperturbation first, we flow to a topological defect ˜ D m − and the boundary condition (1 , m − . Tohave compatibility with the above discussion, the action of the (topological) defect ˜ D m − on (1 , m − must give (1 , d ) m − ; the only topological defect that has this property is ˜ D m − = D (1 , d ) m − . Thuswe conclude that D ( d , m −→ D (1 , d ) m − . (4.8)Note that this rule should be true for all values of d since (4.2) holds for all values of a . Inparticular, it also applies to the defect D ( m − , m that generates the Z symmetry. As the bulkperturbation by φ (1 , is invariant under the Z symmetry, the defect D ( m − , m should flow tothe Z generating defect D ( m − , m − ∼ = D (1 , m − m − in the ( m − st theory, which is preciselyconsistent with (4.8).As another consistency check of this proposal we can consider the action on bulk fields. Fromthe analysis of [1] we know that the ( r, s ) bulk field in the m th theory flows to a linear combination φ ′ of bulk fields φ ( m )( r,s ) −→ φ ′ = X s ′ − s even c s ′ (cid:2) φ ( m − s ′ ,r ) (cid:3) , (4.9)where c s ′ are some constants and we indicated by the square brackets that also descendants of theprimary bulk fields can appear. On the other hand, we know the action of a topological defect An independent argument for this defect to remain topological under the renormalisation group flow was alsovery recently given in [42]. ( d , d ) m on a bulk field in the m th theory (see e.g. [36]), D ( d , d ) m φ ( m )( r,s ) = ( − ( d − d )( r − s ) sin (cid:16) πd rm (cid:17) sin (cid:16) πrm (cid:17) sin (cid:16) πd sm +1 (cid:17) sin (cid:16) πsm +1 (cid:17) φ ( m )( r,s ) . (4.10)If we first apply the defect D ( d , m in the m th theory, and then flow to the ( m − st theory weobtain D ( d , m φ ( m )( r,s ) = ( − ( d − r − s ) sin (cid:16) πd rm (cid:17) sin (cid:16) πrm (cid:17) φ ( m )( r,s ) −→ ( − ( d − r − s ) sin (cid:16) πd rm (cid:17) sin (cid:16) πrm (cid:17) φ ′ . (4.11)On the other hand, if we first let both the defect and the bulk field flow, we have to evaluate D (1 , d ) m − φ ′ = X s ′ − s even c s ′ ( − ( d − r − s ′ ) sin (cid:16) πd rm (cid:17) sin (cid:16) πrm (cid:17) (cid:2) φ ( m − s ′ ,r ) (cid:3) = ( − ( d − r − s ) sin (cid:16) πd rm (cid:17) sin (cid:16) πrm (cid:17) φ ′ , (4.12)which precisely equals (4.11), thus giving further support to the proposal (4.8). Putting everything together we can now determine what happens to an arbitrary boundary condition( a , a ) m (1 < a ≪ m ) under the bulk flow. To this end we write it as the fusion of an appropriatetopological defect on an elementary boundary condition and then perform the flow under the φ (1 , bulk perturbation, ( a , a ) m = D ( a , m × (1 , a ) m −→ D (1 , a ) m − × ( a − , m − = ( a − , a ) m − . (4.13)Here we have assumed 1 < a ≪ m since we have used (4.3). On the other hand a is arbitrarybecause the defect argument did not depend on a being small or not. If a = 1 there is no relevantboundary field in the boundary spectrum, and we have instead of (4.13) simply the flow (4.2).Note that we can also use the defect argument to check our identifications for the perturbativefixed points (I), (II) and (III): if we apply the defects D ( a , m → D (1 , a ) m − to the flow sequencefor a = 1 (1 , a ) m (I) −→ ( a , m (II) −→ (1 , a ) m − −→ ( a , m − for 1 < a ≪ m , (4.14)we find the sequence( a , a ) m (I) −→ M l ( b ( l ) , m (II) −→ M l (1 , b ( l )) m − −→ ( a , a ) m − for 1 < a ≪ m , (4.15)where b ( l ) = | a − a |− l and l = 1 , . . . , min( a , a , m − a , m − a ). This reproduces in particularour results from section 2 for small values of a and a . However, the current analysis is also truefor arbitrary values of a since the rule for the defect flow is not restricted to small values of a .19owever, we still have the restriction that a should be small. By using the identification rules( a , a ) m ∼ = ( m − a , m + 1 − a ) m we can also get a result for labels a close to m . Firstly, we noticethat the flow (4.2) translates into ( a , m ) m −→ (1 , m − a ) m − . (4.16)For non-trivial label a , the sequence (4.15) of the perturbative fixed-points (I), (II), (III) is mappedto ( a , a ) m (I) −→ M l ( b ′ ( l ) , m (II) −→ M l (1 , b ′ ( l )) m − −→ ( a − , a ) m − for 1 ≪ a < m , (4.17)where now b ′ ( l ) = | a − a + 1 | − l and l = 1 , . . . , min( a , a − , m − a , m + 1 − a ). If weextrapolate this answer to small values of a , its form is different from that of the sequence (4.15).The reason for this is that the notion of which fixed points are perturbative and which are non-perturbative changes as we extrapolate a from small to large values. In particular, instead of theperturbative pure boundary fixed point (2.20) the first flow (to fixed point (I)) is now of the sameform as the non-perturbative boundary flow (that generalises (4.1))( a , a ) −→ min( a ,a − M l =1 ( a + a − l, , (4.18)see [30, 16, 27, 24]. Similarly, the third flow (from (II) to (III)) has now the same structure as thenon-perturbative analogue of (2.24), see [28, 29, 34, 35].Remarkably, however, the form of the actual fixed-point (IV) does not change for large valuesof a , and we find ( a , a ) m −→ ( a − , a ) m − for 1 ≪ a < m − , (4.19)which is the same formula as (4.13). This suggests that the result for the end point — the fixedpoint (IV) — of the actual flow can be interpolated to intermediate values of a ; a rather non-trivialtest for this conjecture will be presented in section 4.4. Thus we are led to the final answer for thebulk-boundary flow ( a , a ) m −→ (1 , a ) m − if a = 1( a − , a ) m − if 1 < a < m (1 , m − a ) m − if a = m (4.20)under the φ (1 , bulk perturbation. We should mention that (4.20) reproduces in particular theconjectured flow of [16, 27], (1 , a ) m −→ ( a − , m − (4.21)for 1 < a < m .Finally, let us comment on the behaviour of the flows under the Z symmetry of the minimalmodels (see (4.6)). The bulk field φ (1 , is invariant under this symmetry (see (4.5)), and thus weexpect the flows and fixed-points to respect the symmetry. In fact, it is not difficult to see thatunder the map ( a , a ) m ( a , m + 1 − a ) m the two sequences (4.15) and (4.17) are mapped intoone another, and that the results (4.13) and (4.19) on the actual fixed-point (IV) are left invariant. One can explicitly check that the coupled RG equations remain invariant under the symmetry. There is a slightsubtlety here, namely that the normalisation of the boundary fields in [33] — this has an impact on the OPEcoefficients that enter this calculation, see appendix A — depend on m . Once this has been taken into account, wefind the same structures for the case of small a and a and for the Z -image of this case. .3 Consistency constraints Finally, there are a number of consistency constraints that we can check. They are all a variant ofthe following observation. The flows from our initial boundary condition to the fixed-point (IV)come in a one-parameter family labelled by a parameter χ that measures the initial strength of theboundary perturbation by ψ (1 , relative to the bulk perturbation. (More precisely χ is proportionalto the initial value of the ratio µ/ p | λ | when the perturbation is turned on, see (C.9) in appendix C.)For large negative values of χ , the flow first follows closely the pure boundary flow to (I), thenthe bulk flow down to (II), and finally the pure boundary flow from (II) to (IV) (see figure 1). In thelimit χ → −∞ , the flow completely decomposes into three separate flows. If our analysis is correct,the separate flows should involve the same fixed-points that we identified in our analysis. Thefixed point (I) that is reached by the pure boundary perturbation is described by a superpositionof boundary conditions ( b, i.e. (2.20) and (2.26) then leads indeed to (2.22) with the same configuration at (II).Similarly we can consider the limit χ → + ∞ . Then we first follow a pure boundary flow thatis the non-perturbative counterpart of the perturbative flow to (I), and we reach a fixed point (I’)that corresponds to the superposition (4.18). For the subsequent bulk flow we can again use (2.26)and we arrive at a superposition (II’) of boundary conditions of the form (1 , b i ). From there wecan follow a pure boundary flow. Although we do not know a priori which boundary (1 , a = 1where there is only one perturbative boundary (1,3)-flow from (II’), and so it is easily identified.To get the flow for the general case, one then again uses the fusion of a defect (4.8).This last boundary flow in the c m − theory is actually the perturbative flow from (II’) to (III’)for another initial boundary condition a ′ = ( a , a − a , · ) into a chain of flows; figure 2 shows a generic segment of thechain. (II)(I) (IV) (I’)(II’)(III) (III’) min( a ,a − M l =1 ( a + a − l, min( a ,a − M l =1 (1 , a + a − l )( a , a ) min( a ,a ) M l =1 ( a + a + 1 − l,
1) ( a − , a ) min( a ,a ) M l =1 (1 , a + a + 1 − l )( a , a ) ( a , a − Figure 2: A segment of the chain of flows for small values of a , a . In the upper line we haveboundary conditions of the m th and in the lower line of the ( m − st minimal model.At the end of the chain the situation degenerates: the boundary condition ( a , m flows to(1 , a ), so the fixed-point (IV) coincides with the fixed-point (II’) that is reached from the purebulk flow from (I’), see figure 3.The chains that we get for different values of a look all similar. Indeed, the chain for general a can be obtained starting from the chain with a = 1 by fusion with a topological defect. The We thank Patrick Dorey for drawing our attention to this point. II) (II’)(I’)(IV)(I) ( a ,
2) ( a , , a )(2 , a )( a , , a − ⊕ (1 , a ) ⊕ (1 , a + 2)( a − , ⊕ ( a , ⊕ ( a + 2 ,
1) (1 , a − ⊕ (1 , a + 1)( a − , ⊕ ( a + 1 , Figure 3: The right end of the chain of flows. Starting from the boundary condition ( a , m , inone direction the flow decomposes into only two separate flows ( → ( a , m → (1 , a ) m − ) insteadof three separate flows in the generic case.topological defect to consider is the one that corresponds to D ( a , m in the m th model. Along theupper sequence ( i.e. in the m th theory) this topological defect maps indeed the upper part of the a = 1 chain to the upper part of the chain for a . As we have argued before, the defect D ( a , m flows under the bulk flow to D (1 , a ) m − , and it is therefore this topological defect that acts on thelower sequence. This then produces the lower sequence of the diagram for a . When we extrapolate our chain from the right to larger values of a , the perturbative boundaryflows become longer, and the non-perturbative flows become shorter. In the middle of the chain,they are of the same length, more precisely, for the boundary condition ( a , m +12 ) m ( m odd), thetwo boundary flows triggered by ψ (1 , are mapped to each other by the Z symmetry and are thusequally long (see figure 4). The diagram suggests that for this boundary condition the pure bulkperturbation does not switch on the boundary field ψ (1 , . Indeed one finds that the bulk boundarycoefficient B (1 , , = 0, so the fixed-point is in this case given by λ = λ ∗ and µ = 0. We can thencompute the perturbed g -function, which to leading order only gets a contribution from the bulkone-point function. We findlog g ( m ) λ ∗ , ( a , m +12 ) g ( m ) ( a , m +12 ) = 14 m − ξξ + O ( ξ, m − ) . (4.22)With our prescription for the cutoff ξ (see (3.31)), we obtain the expected resultlog g ( m − ( m − , a ) g ( m ) ( a , m +12 ) = − m + O ( m − ) . (4.23)This check gives support to the idea that our results can be extrapolated to any value of a . We cango even further and look at boundary conditions close to ( a , m +12 ) m . Figure 4 suggests that herethe bulk boundary flow to the true fixed point (IV) is perturbative whereas the pure boundary flowsare nonperturbative. To check this explicitly, let us consider the RG equations for the boundarycoupling µ of the ψ (1 , field for a boundary condition of the form a = ( a , m +12 − a ),˙ µ = (1 − h ) µ + 12 Bλ + Eλµ + · · · = 2 m µ − πa λ + 4 πλµ + · · · , (4.24)22 ( a , m +12 )( m − , a ) ( a , m − )( m − , a ) a M l =1 ( m +3+2 a − l , a M l =1 (1 , m +3+2 a − l ) a M l =1 ( m +1+2 a − l , a M l =1 (1 , m +1+2 a − l ) Figure 4: The middle of the chain of flows for m odd. The fixed point starting from the exactmiddle ( a , m +12 ) m is reached for λ = λ ∗ , µ = 0 (flow 1 in the figure). When we go slightly awayfrom the middle (flow 2), the flow can still be treated perturbatively.where a and a are now small. Here we have used the expressions for the coupling constants fromappendix A.2. Note that we neglected the term proportional to µ , because its coefficient D (1 , , , is suppressed by m − . For the calculation of E we need to determine the correlator of the bulk fieldand two boundary fields to leading order in 1 /m , which now takes the form1 h i D (1 , , , (cid:10) φ (1 , ( z, ¯ z ) ψ (1 , (0) ψ (1 , ( ∞ ) (cid:11) = − ( z − ¯ z ) + ( z + ¯ z ) | z | ( z − ¯ z ) + O ( m − ) . (4.25)In particular, this correlator has the correct asymptotics in the various channels: for the (1 ,
1) and(1 ,
5) channels the relevant coefficients are now B (1 , , D (1 , , , = − O ( m ) , B (1 , , D (1 , , , = + O ( m ) , (4.26)whereas the coefficient in the (1 ,
3) channel still vanishes to leading order B (1 , , D (1 , , , = O ( m − ) . (4.27)After integration as in (2.11) this leads to E = 4 π . This is the expected value since at the fixed point λ = λ ∗ = − πm the conformal dimension of the field coupling to µ should be h (3 , = 1 + m + · · · .For λ = λ ∗ the RG equation (4.24) has a unique fixed point at µ ∗ = a . This solution describesa perturbative fixed point. In fact the order one value of µ ∗ is only an artefact of our normalisationsince D (1 , , , is of order m − , and thus in the ‘natural normalisation’ we should rescale the field ψ by a factor of m , which would replace µ µm . Note that after this rescaling the boundary couplingconstant D (1 , , , is only of order m − , and hence can still be ignored in the RG equation. It isalso clear from the flow diagram (see figure 5) that this fixed point is actually reached by a genericflow.Using the same techniques as above in section 3, we can now determine the change in the g -factor, log g ( m ) λ ∗ ,µ ∗ ( a ) = log g ( m ) ( a ) − π m µ ∗ − π a m λ ∗ (cid:18) − a µ ∗ + 16 a µ ∗ (cid:19) + · · · = log g ( m ) ( a ) − π a m + · · · , (4.28)where the dots denote contributions that are either of higher order in 1 /m or do not depend on theboundary labels. To remove the terms that are independent of the boundary labels, we again look23 €€€€€€ a Λ * €€€€€€ a Λ * ΜΛ Μ Λ
Figure 5: The flow diagram for (4.24) for a = 1 and m = 101. The vectors ( ˙ µ, ˙ λ ) have beenmagnified by a factor 2 .
5. A generic flow (described by the solution (C.12) of (4.24)) reaches theperturbative fixed-point.at relative g -functions with respect to some reference boundary condition, which we choose here tobe (1 , m +12 ) m (notice that we have to use a boundary condition which is in the regime where theperturbative analysis applies). The relative g -function in the m th model is g ( m ( a ) g ( m ) (1 , m +12 ) = sin πa m sin πm (cid:18) − π a m + π a m + O ( m − ) (cid:19) . (4.29)After the perturbation by φ (1 , the relative g -function becomeslog g ( m ) λ ∗ ,µ ∗ ( a ) g ( m ) λ ∗ , (1 , m +12 ) = log sin πa m sin πm − π a m − π a m + O ( m − ) (4.30)= log g ( m − ( m − − a , a ) g ( m − ( m − ,
1) + O ( m − ) . (4.31)This perturbative calculation thus predicts that the end-point of the actual flow is (cid:18) a , m + 12 − a (cid:19) m (IV) −→ (cid:18) m − − a , a (cid:19) m − . (4.32)It is remarkable that in this case the actual end-point of the flow ( i.e. the fixed point (IV)) canbe directly computed perturbatively. The result (4.32) is in beautiful agreement with our generalanswer (4.20), and thus gives strong support for the claim that (4.20) is also true for intermediatevalues of a . 24 Comparison with numerical results
Finally let us analyse how our perturbative calculation compares with numerical calculations thathave been performed before. In particular, in [25, 26] the flow of the tricritical Ising ( m = 4) tothe Ising model ( m = 3) was considered in the presence of boundaries. They considered a cylinderdiagram with two boundaries for which they imposed the boundary conditions ( r,
1) and (1 , s ),respectively. According to the fusion rules, the relative open string spectrum between the twoboundary conditions transforms then in the ( r, s ) representation, and they found that under the φ (1 , flow, the open string character χ ( r,s ) flowed as m = 4 m = 3 χ (1 , −→ χ (1 , χ (2 , −→ χ (1 , χ (3 , −→ χ (1 , χ (1 , −→ χ (1 , χ (1 , −→ χ (2 , χ (2 , −→ χ (1 , . (5.1)In particular, the first flow implies that the (1 ,
1) boundary condition of the m = 4 theory flowsto the (1 ,
1) boundary condition of the m = 3 theory. In order to compare with our calculations(where we consider a single boundary condition) we can take either r = 1 or s = 1 since then oneof the two boundary conditions does not flow. In this way we deduce from (5.1) that the boundaryconditions flow as (1 , −→ (1 , (2 , −→ (1 , (3 , −→ (1 , (1 , −→ (1 , (1 , −→ (2 , ∼ = (1 , . (5.2)Unfortunately, we cannot determine the flow of the (2 ,
2) boundary condition in this manner, sincenone of the cylinder diagrams considered in [25, 26] involves this boundary condition. However,one may suspect that the behaviour of the cylinder diagram between (1 ,
1) and (2 ,
2) should be thesame as that between (1 ,
2) and (2 , , → (1 , , which is indeed in agreement with our general rule (4.20).The flows (5.2) agree perfectly with our general rule (4.20). A schematic view of the flows inthe space of couplings λ, µ is as follows: m =4(3,1) (2,1) (1,1)(1,3) (1,2) (1,1)(1,3) (1,2)=(2,1) (II)(I) m =3 Notice that there is a typographical error in Table 1 of [26]. As explained in the accompanying text of [26], theflow of the characters χ (1 , and χ (1 , in the m = 4 model is rather as indicated in (5.1). We thank Leung Chim forpointing this out to us. m = 4 and m = 3 model, respectively,which are driven by the ψ (1 , boundary field. The short horizontal arrows in the m = 4 modelindicate that these flows become the perturbative ψ (1 , flows when m is sent to large values. Finally, the flows indicated by the curved arrows agree precisely with the last two flows of (5.2).This final check of our general prediction (4.20) at this low value of m makes us confident thatour results hold for the whole (A-type) series of unitary minimal models. Acknowledgements
We are indebted to Patrick Dorey, Anatoly Konechny and Ingo Runkel for detailed explanationsand discussions. We also thank Albion Lawrence and Volker Schomerus for initial collaboration onthis project, and Andreas Recknagel for useful comments. The work of MRG and CSC is supportedby the Swiss National Science Foundation. MRG thanks the GGI in Florence, the Chinese Academyof Sciences in Beijing, and the IPMU in Tokyo for hospitality during the final stages of this work.
AppendixA OPE coefficients
For the charge conjugation minimal model (the A-series), the various OPE coefficients have beendetermined in [33] in terms of the F matrices. For large m the bulk coupling constant C (1 , , , hasthe following large m behaviour C (1 , , , = 4 − m −
83 (3 + π ) 1 m + O ( m − ) . (A.1)We also need the large m expansions of the boundary and bulk-boundary OPE coefficients. A.1 The expansion for small a , a For a boundary condition a = ( a , a ) for which a , a ≪ m , we have the following asymptoticexpansions of the OPE coefficients: D (1 , , , = 1 , (A.2) D (1 , , , = a − a − a ) ( a +1) + O ( m ) ( a > a ) m ( a − − ma − + a − + O ( m ) ( a = a ) a +1)( a − a ) ( a − + O ( m ) ( a < a ) , (A.3) D (1 , , , = − a − a )( a +1) + O ( m ) ( a > a ) − m ( a − + O (1) ( a = a ) − a − a )( a − + O ( m ) ( a < a ) , (A.4) The flow (1 , → (2 , becomes (1 , m → (2 , m for large m , while the flow (1 , → (2 , becomes(1 , m − m → ( m − , m . (1 , , , = − a − a +1)( a − a ) (( a − a ) − + O ( m ) ( a ≥ a + 2) ma +2 + a − a +2 + O ( m ) ( a = a + 1) m a − − ma − + a +77+12 π ) a − + O ( m ) ( a = a ) − ma − + a +34) a − + O ( m ) ( a = a − − a +2)( a − a − a ) (( a − a ) − + O ( m ) ( a ≤ a − , (A.5) D (1 , , , = a − a − a +1)( a +2)( a − a ) (( a − a ) − + O ( m ) ( a ≥ a + 2) a m ( a +3)( a +2)( a − + O ( m ) ( a = a + 1) m ( a +2)( a +1)( a − a − + O ( m ) ( a = a ) a m ( a +1)( a − a − + O ( m ) ( a = a − a +1)( a +2)( a − a − a − a ) (( a − a ) − + O ( m ) ( a ≤ a − , (A.6) D (1 , , , = − a − a − a ) − ( a − a ) ( a +1)( a +2) + O ( m ) ( a ≥ a + 2) m ( a +3)( a +2)( a − + O ( m ) ( a = a + 1) − m ( a +2)( a +1)( a − a − + O ( m ) ( a = a ) − m ( a − a − a +1) + O ( m ) ( a = a − − a +2)(( a − a ) − ( a − a ) ( a − a − + O ( m ) ( a ≤ a − , (A.7) B (1 , , = 3 − π m a + O ( 1 m ) , (A.8) B (1 , , = πm ( a − a )( a + 1) + O ( m ) ( a > a ) πm ( a −
1) + O ( m ) ( a = a ) πm ( a − a )( a −
1) + O ( m ) ( a < a ) , (A.9) B (1 , , = − π m ( a + 1)( a + 2)( a − a ) (( a − a ) −
1) + O ( m ) ( a ≥ a + 2) π m ( a − a + 2)( a + 3) + O ( m ) ( a = a + 1) π m ( a − a −
1) + O ( m ) ( a = a ) − π m ( a + 1)( a − a −
3) + O ( m ) ( a = a − − π m ( a − a − a − a ) (( a − a ) −
1) + O ( m ) ( a ≤ a − . (A.10) A.2 Asymptotic expansion for a small and a close to m +12 For boundary conditions of the form ( a , m +12 − a ) m where m is odd and a , a ≪ m we have onthe other hand D (1 , , , = 8 m + O ( m ) , (A.11) D (1 , , , = 8 π a m + O ( m ) , (A.12) D (1 , , , = 1 , (A.13)27 (1 , , , = − m + O ( log mm ) , (A.14) D (1 , , , = 9 · m + O ( log mm ) , (A.15) D (1 , , , = 2 · π a m + O ( log mm ) , (A.16) B (1 , , = − π a m + O ( m ) , (A.17) B (1 , , = − πa + O ( m ) , (A.18) B (1 , , = − m
288 + O (log( m ) m ) . (A.19) B Correlation functions and integrals
B.1 Disc and upper half plane
The upper half plane is defined by H + = { z ∈ C | Im z ≥ } , while the disc is defined by D = { w ∈ C | | w | ≤ } . On the disc we often use the polar coordinates w = re iθ , where 0 ≤ r ≤ − π < θ ≤ π . The transition function from H + to D is given by z ( w ) = i − w w . (B.1)Primary fields of the CFT are denoted φ in the bulk, with conformal dimension ∆ = 2 h , and ψ onthe boundary, with conformal dimension h . The identity field is denoted by 11, both in the bulkand on the boundary. The basic correlators on the upper half-plane are h ψ ( x ) i H + = δ ψ, h i H + , h ψ ( x ) ψ ( x ) i H + = D ψψ h i H + ( x − x ) − h ( x > x ) , h φ ( x + iy, x − iy ) i H + = B φ h i H + (2 y ) − ∆ ( y > , h φ ( x + iy , x − iy ) ψ ( x ) i H + = B φψ D ψψ h i H + (2 y ) − ∆+ h (cid:18) ( x − x ) + y (cid:19) − h ( y > , h ψ ( x ) ψ ( x ) ψ ( x ) i H + = D ψ ψ ψ D ψ ψ h i H + ( x − x ) h − h − h × ( x − x ) h − h − h ( x − x ) h − h − h ( x > x > x ) .
28n the disc, these are (cid:10) ψ ( re iθ ) (cid:11) D = δ ψ, h i D , (cid:10) ψ ( e iθ ) ψ ( e iθ ) (cid:11) D = D ψψ h i D (cid:12)(cid:12) θ − θ (cid:12)(cid:12) − h , (cid:10) φ ( re iθ , re − iθ ) (cid:11) D = B φ h i D (1 − r ) − ∆ , (cid:10) φ ( re iθ , re − iθ ) ψ ( e iθ ) (cid:11) D = B φψ D ψψ h i D (1 − r ) − ∆ (cid:0) − r − r cos( θ − θ )+ r (cid:1) h , (cid:10) ψ ( e iθ ) ψ ( e iθ ) ψ ( e iθ ) (cid:11) D = D ψ ψ ψ D ψ ψ h i D (cid:12)(cid:12) θ − θ (cid:12)(cid:12) h − h − h × (cid:12)(cid:12) θ − θ (cid:12)(cid:12) h − h − h (cid:12)(cid:12) θ − θ (cid:12)(cid:12) h − h − h . (B.2)In sections 2.1 and 3, we also need an expression for the correlator (cid:10) φ ( z, ¯ z ) ψ ( x ) ψ ( y ) (cid:11) a (B.3)in the limit m → ∞ . This is a chiral four-point function which can be computed from the differentialequation associated with the singular vector of the (1 ,
3) module at level 3, N = (cid:18) L − − h − L − L − + ( m + 1) m ( m −
1) ( L − ) (cid:19) φ . (B.4)The general solution to the corresponding third order differential equation in the cross-ratio η canbe obtained in the limit m → ∞ . One can fix the solution by the asymptotic behaviour when thebulk-field approaches the boundary, but as the dimensions of the fields in the asymptotic channelsdiffer by integers in the limit m → ∞ , this is not the easiest method. A better approach is to solvethe differential equation for finite m in an expansion around η = 0 where the channels can be clearlyseparated, and then match the two sets of solutions in the limit m → ∞ . Of the three conformalblocks associated to the channels (1 , , , m → ∞ . On the upper half-plane, we find h φ ( z, ¯ z ) ψ ( x ) ψ ( y ) i H + = η − η + η )6( η η η ) B (1 , , D (1 , , , D (1 , , , h i H + (1 + O ( m − )) , (B.5)where η = ( z − ¯ z )( x − y ) , η = ( z − x )(¯ z − y ) , η = η − η = ( z − y )(¯ z − x ) . (B.6)Multiplying with y , this yields (2.12) in the limit x → y → ∞ . To perform the integral thatleads to (3.16), we need an expression for the correlator on the disc, (cid:10) φ ( w, ¯ w ) ψ ( e iθ ) ψ ( e iθ ) (cid:11) D = η ′ − η ′ + η ′ )6( η ′ η ′ η ′ ) B (1 , , D (1 , , , D (1 , , , h i H + (1 + O ( m − )) , (B.7)where η ′ = (1 − | w | )( e i θ − θ − e − i θ − θ ) , η ′ = e i θ − θ ( we − iθ − − ¯ we iθ ) , (B.8) η ′ = η ′ − η ′ = e i θ − θ ( we − iθ − − ¯ we iθ ) . (B.9)29ntegrating this expression with our cut-offs ǫ on the boundary and ξ between bulk and boundary,we find 2 π B (1 , , D (1 , , , D (1 , , , h i D (1 + O ( m − )) × (cid:18)(cid:18) π (2 − ξ )2 ξ + O ( ξ ) (cid:19) ǫ + (cid:18) π (3 ξ − ξ + O ( ξ ) (cid:19) (B.10)+ (cid:18) πξ − π ξ + π O ( ξ ) (cid:19) ǫ + O ( ǫ ) (cid:19) . The term proportional to ǫ − is subtracted when we compute the connected correlation function.The ǫ -independent part, on the other hand, gives our result (3.16). B.2 Semi-infinite cylinder
Instead of considering the theory on the disc we now put it on a semi-infinite cylinder by theconformal map w = re iθ v = − log w = − log r − iθ =: v + iv . (B.11)The coordinate v runs from 0 to ∞ , whereas v is periodic with period 2 π . A bulk one-pointfunction in these coordinates reads h Φ( v, ¯ v ) i C = B φ h i (cid:0) v (cid:1) − ∆ . (B.12)Let us consider the integral of a one-point function of a marginal bulk field over the semi-infinitecylinder. To regularise the integral we have to introduce a cutoff η such that v cannot come tooclose to the boundary. This leads to I C = Z ∞ η dv (cid:0) v (cid:1) − = 14 (cid:0) − η (cid:1) = 14 (cid:18) η − O ( η ) (cid:19) . (B.13)Comparing this with the disc integral I in (3.11) that appeared in the computation of the one-point contribution to the g -factor, we see that we get coinciding results precisely when we relate thecutoffs ξ and η as in (3.31). We conclude that on the semi-infinite cylinder the natural cutoff η isalready the right one. One could have expected this because here bulk and boundary contributionsshould be disentangled as the boundary is not curved with respect to the bulk metric, in contrastto the disc. C Solutions of the RG equations
The set of RG equations in (2.17),˙ λ = 4 m λ + 4 πλ , (C.1)˙ µ = 2 m µ + 2 παm λ − πλµ − α µ (C.2) This observation and the preceding computation resulted from discussions with Anatoly Konechny. λ ( t ) = λ ∗ Ce m t Ce m t , λ ∗ = − πm , (C.3)where C is some constant. For the flows to connect λ = 0 at t → −∞ and λ = λ ∗ at t → + ∞ , wehave to take C >
0. In that case λ flows strictly monotonically, and we can parameterise µ ( t ) = παf ( λ ( t )) , (C.4)with some function f ( λ ). Differentiating (C.4) with respect to t and plugging in (C.1) and (C.2),we obtain an ordinary first order differential equation for f ( λ ),2 λ (1 + πmλ ) f ′ ( λ ) = f ( λ ) + λ − πmλf ( λ ) − πmf ( λ ) . (C.5)This equation has the solution f ( λ ) = λ ∗ χ q λ ∗ λ − λ ∗ λ − (cid:18) χ q λ ∗ λ − (cid:19) , (C.6)with an arbitrary constant χ . Let us discuss the different solutions corresponding to different valuesof χ and their properties in turn. • For χ → ∞ , the function f ( λ ) becomes constant, f ( λ ) ≡ − λ ∗ πm = ⇒ µ ( t ) ≡ α m . (C.7)This solution corresponds to the pure bulk flow from the fixed point (I) to (II). • For χ finite, let us expand the function f ( λ ) for small (negative) λ , f ( λ ) = λ ∗ (cid:16) χ p λ/λ ∗ + (1 + 2 χ )( λ/λ ∗ ) + · · · (cid:17) . (C.8)We see that for all finite χ we have f (0) = 0, so that the flow at t = −∞ starts at λ = µ = 0. • For χ = 0, µ grows at the same rate as λ for t → −∞ . This corresponds to the situationwhen µ itself is not turned on initially, but is just sourced by λ . • For finite χ = 0, µ grows as παλ ∗ χ √ Ce m t for t → −∞ which is the solution of the first orderapproximation ˙ µ = m µ + · · · to the boundary RG equation. χ thus determines how muchand with which sign we turn on the boundary field in the beginning, indeed we have χ = − πα p | λ ∗ | lim t →−∞ µ ( t ) p | λ ( t ) | . (C.9) • For any finite χ the function f ( λ ) has a pole at λ = λ ∗ (cid:18) − χ p χ (cid:19) , (C.10)at which the function f ( λ ) diverges to −∞ when λ/λ ∗ approaches λ /λ ∗ from below. Notethat the perturbative fixed points (I) and (II) at µ = α/ m correspond to f = − λ ∗ / >
0, sothe solution always runs away from the perturbative fixed point for large t .31 For finite χ <
0, the function f first develops towards positive values, then reaches its mostpositive value f max = f ( λ max ) = − λ ∗ (cid:18) − p χ (cid:19) at λ max = λ ∗ (cid:18) − p χ (cid:19) , (C.11)and then runs towards −∞ . The maximal value f max approaches − λ ∗ / χ → −∞ , so wecan come arbitrarily close to the perturbative fixed-point at µ = α/ m by tuning χ to largenegative values.Similarly, the solution to the RG equation (4.24) is of the form µ = πa f ( λ ( t )) , f ( λ ) = − mλ √− λ √ mπλ χ , (C.12)where χ is again an arbitrary real parameter. References [1] A.B. Zamolodchikov,
Renormalization group and perturbation theory near fixed points in two-dimensional field theory , Sov. J. Nucl. Phys. (1987) 1090 [Yad. Fiz. (1987) 1819].[2] A.B. Zamolodchikov, Integrable field theory from conformal field theory , Adv. Stud. Pure Math. (1989) 641.[3] A.B. Zamolodchikov, Irreversibility of the flux of the renormalization group in a 2D field theory ,JETP Lett. (1986) 730 [Pisma Zh. Eksp. Teor. Fiz. (1986) 565].[4] I. Affleck and A.W.W. Ludwig, Universal noninteger ’ground state degeneracy’ in critical quan-tum systems , Phys. Rev. Lett. (1991) 161.[5] D. Friedan and A. Konechny, On the boundary entropy of one-dimensional quantum systemsat low temperature , Phys. Rev. Lett. (2004) 030402 [arXiv:hep-th/0312197] .[6] S. Fredenhagen, M.R. Gaberdiel and C.A. Keller, Bulk induced boundary perturbations , J. Phys.A (2007) F17 [arXiv:hep-th/0609034] .[7] S. Fredenhagen, M.R. Gaberdiel and C.A. Keller, Symmetries of perturbed conformal fieldtheories , J. Phys. A (2007) 13685 [arXiv:0707.2511 [hep-th]] .[8] M.R. Gaberdiel, A. Konechny and C. Schmidt-Colinet, Conformal perturbation theory beyondthe leading order , J. Phys. A (2009) 105402 [arXiv:0811.3149 [hep-th]] .[9] M.R. Gaberdiel and A. Lawrence, Bulk perturbations of N = 2 branes , JHEP (2007) 087 [arXiv:hep-th/0702036] .[10] M.R. Gaberdiel and O. Schlotterer,
Bulk induced boundary perturbations for N=1 superconfor-mal field theories , J. Phys. A (2009) 115209 [arXiv:0810.4719 [hep-th]] .[11] I.V. Cherednik, Factorizing particles on a half line and root systems , Theor. Math. Phys. (1984) 977 [Teor. Mat. Fiz. (1984) 35]. 3212] E.K. Sklyanin, Boundary conditions for integrable quantum systems , J. Phys. A (1988) 2375.[13] A. Fring and R. K¨oberle, Factorized scattering in the presence of reflecting boundaries , Nucl.Phys. B (1994) 159 [arXiv:hep-th/9304141] .[14] S. Ghoshal and A.B. Zamolodchikov,
Boundary S matrix and boundary state in two-dimensionalintegrable quantum field theory , Int. J. Mod. Phys. A (1994) 3841 (Erratum-ibid. A (1994)4353) [arXiv:hep-th/9306002] .[15] A. LeClair, G. Mussardo, H. Saleur and S. Skorik, Boundary energy and boundary states inintegrable quantum field theories , Nucl. Phys. B (1995) 581 [arXiv:hep-th/9503227] .[16] F. Lesage, H. Saleur and P. Simonetti,
Boundary flows in minimal models , Phys. Lett. B (1998) 85 [arXiv:hep-th/9802061] .[17] P. Dorey, A. Pocklington, R. Tateo and G.M.T. Watts,
TBA and TCSA with boundaries andexcited states , Nucl. Phys. B (1998) 641 [arXiv:hep-th/9712197] .[18] P. Dorey, I. Runkel, R. Tateo and G.M.T. Watts, g-function flow in perturbed boundary con-formal field theories , Nucl. Phys. B (2000) 85 [arXiv:hep-th/9909216] .[19] P. Dorey, D. Fioravanti, C. Rim and R. Tateo,
Integrable quantum field theory with boundaries:the exact g-function , Nucl. Phys. B (2004) 445 [arXiv:hep-th/0404014] .[20] P. Dorey, M. Pillin, R. Tateo and G.M.T. Watts,
One-point functions in perturbed boundaryconformal field theories , Nucl. Phys. B (2001) 625 [arXiv:hep-th/0007077] .[21] Z. Bajnok, L. Palla and G. Takacs,
On the boundary form factor program , Nucl. Phys. B (2006) 179 [arXiv:hep-th/0603171] .[22] O.A. Castro-Alvaredo,
Boundary form factors of the sinh-Gordon model with Dirichlet bound-ary conditions at the self-dual point , J. Phys. A (2006) 11901 [arXiv:hep-th/0606269] .[23] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula , Nucl. Phys. B (1989) 581.[24] A. Recknagel, D. Roggenkamp and V. Schomerus,
On relevant boundary perturbations of uni-tary minimal models , Nucl. Phys. B (2000) 552 [arXiv:hep-th/0003110] .[25] P.A. Pearce, L. Chim and C. Ahn,
Excited TBA equations. I: Massive tricritical Ising model ,Nucl. Phys. B (2001) 539 [arXiv:hep-th/0012223] .[26] P.A. Pearce, L. Chim and C. Ahn,
Excited TBA equations II: Massless flow from tricritical tocritical Ising model , Nucl. Phys. B (2003) 579 [arXiv:hep-th/0302093] .[27] C. Ahn and C. Rim,
Boundary flows in the general coset theories , J. Phys. A (1999) 2509 [arXiv:hep-th/9805101] .[28] K. Graham, On perturbations of unitary minimal models by boundary condition changing op-erators , JHEP (2002) 028 [arXiv:hep-th/0111205] .[29] S. Fredenhagen and V. Schomerus,
D-branes in coset models , JHEP (2002) 005 [arXiv:hep-th/0111189] . 3330] L. Chim,
Boundary S-matrix for the tricritical Ising model , Int. J. Mod. Phys. A (1996)4491 [arXiv:hep-th/9510008] .[31] N. Ishibashi, The boundary and crosscap states in conformal field theories , Mod. Phys. Lett. A (1989) 251.[32] J.L. Cardy, Conformal invariance and statistical mechanics , Les Houches Summer School 1988.[33] I. Runkel,
Boundary structure constants for the A-series Virasoro minimal models , Nucl. Phys.B (1999) 563 [arXiv:hep-th/9811178] .[34] S. Fredenhagen and V. Schomerus,
On boundary RG-flows in coset conformal field theories ,Phys. Rev. D (2003) 085001 [arXiv:hep-th/0205011] .[35] S. Fredenhagen, Organizing boundary RG flows , Nucl. Phys. B (2003) 436 [arXiv:hep-th/0301229] .[36] V.B. Petkova and J.B. Zuber,
Generalised twisted partition functions , Phys. Lett. B (2001)157 [arXiv:hep-th/0011021] .[37] J. Fr¨ohlich, J. Fuchs, I. Runkel and C. Schweigert,
Duality and defects in rational conformalfield theory , Nucl. Phys. B (2007) 354 [arXiv:hep-th/0607247] .[38] I. Brunner and D. Roggenkamp,
Defects and bulk perturbations of boundary Landau-Ginzburgorbifolds , JHEP (2008) 001 [arXiv:0712.0188 [hep-th]] .[39] K. Graham and G.M.T. Watts,
Defect lines and boundary flows , JHEP (2004) 019 [arXiv:hep-th/0306167] .[40] C. Bachas and M.R. Gaberdiel,
Loop operators and the Kondo problem , JHEP (2004)065 [arXiv:hep-th/0411067] .[41] M. Oshikawa and I. Affleck,
Boundary conformal field theory approach to the criticaltwo-dimensional Ising model with a defect line , Nucl. Phys. B (1997) 533 [arXiv:cond-mat/9612187] .[42] M. Kormos, I. Runkel and G.M.T. Watts,
Defect flows in minimal models , arXiv:0907.1497[hep-th] .[43] I. Affleck and A.W.W. Ludwig, Exact conformal field theory results on the multichannel Kondoeffect: Single fermion Green’s function, selfenergy and resistivity , Phys. Rev. B (1993) 7297.[44] D.R. Green, M. Mulligan and D. Starr, Boundary entropy can increase under bulk RG flow ,Nucl. Phys. B (2008) 491 [arXiv:0710.4348 [hep-th]][arXiv:0710.4348 [hep-th]]