BCFT and OSFT moduli: an exact perturbative comparison

Starting from the pseudo-${\cal B}_0$ gauge solution for marginal deformations in OSFT, we analytically compute the relation between the perturbative deformation parameter $\tilde\lambda$ in the solution and the BCFT marginal parameter $\lambda$, up to fifth order, by evaluating the Ellwood invariants. We observe that the microscopic reason why $\tilde\lambda$ and $\lambda$ are different is that the OSFT propagator renormalizes contact term divergences differently from the contour deformation used in BCFT

A classical [15] yet not fully understood problem in this correspondence is how the D-branes moduli space is described in OSFT. Given an exactly marginal boundary field j, there is a corresponding family of OSFT solutions, which can be generically found in powers of a deformation parameterλ Ψλ =λ cj(0)|0 + Physically we expect that the deformation parameterλ which we used to construct the solution should be related to the natural parameter λ in boundary conformal field theory (BCFT), given by the coefficient in front of the boundary interaction which deforms the original worldsheet action On general grounds,λ does not have a gauge invariant meaning, but nonetheless it is useful to understand how λ andλ are related for a given solution, because this can shed light on the different mechanisms by which a classical solution changes the worldsheet boundary conditions. Analytic solutions for marginal deformations with nonsingular OPE (jj ∼ reg) have been computed to all orders in [2] and [3]. A different perturbative analytic solution for marginal currents with singular OPE has been constructed in [4] and generalized in [5] 4 . An analytic solution for any self local (hence exact [18]) marginal deformation has been constructed nonperturbatively in [9]. Conveniently, this solution is directly expressed in terms of the deformation parameter of the underlying BCFT, λ. In [19] this has been used to explicitly find the relation between the BCFT modulus λ and the coefficient of the marginal field in the solution c∂cj|Ψ(λ) . It has been observed that this function of λ starts linearly, then it has a local maximum and finally it approaches zero for large values of λ. Nontrivial evidence that this behaviour may also be present in Siegel gauge 5 has been given in [20] in level truncation, but it has not been possible there to establish the validity of the full equations of motion for large BCFT moduli.
In this note we would like to study this problem in another analytic wedge-based example which is quite close to Siegel gauge. We will analyze the observables of the solution proposed by Schnabl in [2], in the so-called pseudo-B 0 gauge for a chiral marginal current j(z) with OPE j(z) j(0) = 1 z 2 + regular. (1.6) Computing the Ellwood invariants and matching them with the BCFT expected answers, gives the following relatioñ λ =λ(λ) = λ − 3 log 2 λ 3 + 2.38996(7) λ 5 + O(λ 7 ) . (1.7) Let us comment on the found relation. Perhaps the most interesting fact about (1.7) is the origin itself of the found coefficients of λ 2n+1 . These coefficients are obtained by comparing the Ellwood invariants computed from the solution in powers ofλ, with the coefficients of the Ishibashi states obtained from the marginally deformed boundary state expressed in powers of λ, see eqs (4.11)- (4.16). Naively these two quantities reduce to the same worldsheet calculation and therefore one would expect to find perfect match between λ andλ, which is evidently not true. This is explained as follows. At orderλ k , the encountered Ellwood invariants have the structure of OSFT tree-level amplitudes between an on-shell closed string and k on-shell open strings given by the marginal field cj, with λ playing the role of the open string coupling constant. These amplitudes are naively affected by infrared divergences due to the collisions of the marginal fields at zero momentum, which correspond to the the propagation of the zero momentum tachyon. The propagator B 0 L 0 gives a uniquely defined prescription to renormalize these singularities, see section 3. On the other 5 In Siegel gauge the perturbative coefficientλ (1.1) and the coefficient of the marginal field in the solution c∂cj|Ψ coincide. This is not generically true for other perturbative solutions, see for example [5,17]. This is also not true for the solution [2] analysed in this paper and the relation can be computed, if needed, by the same methods of section 4. hand, in BCFT, the same contact term divergences are renormalized by contour deformation [18], so that the renormalized boundary interaction e −λ dsj(s) acquires a topological nature. This difference in the renormalization procedure of contact term divergences is the ultimate reason why λ andλ are different. Had the self-OPE between the currents been regular, we would have found no difference between the two quantities.
We also observe that the growing of the coefficients in (1.7) is in agreement with the findings from other non-perturbative approaches (although in different gauges) such as [19] and [20], and it suggests that the power series inλ may have a finite radius of convergence. It would be desirable to improve our calculation to be able to estimate the growing of the higher order coefficients and the nature of the singularity in the complexλ space. This would be a complementary (perturbative) way of understanding the reason why (in Siegel gauge) the marginal solution breaks down at a critical value ofλ. Indeed, it turns out that our computations in pseudo B 0 -gauge can be related to the analogous computations in Siegel gauge, whose direct evaluation is notoriuosly very complicated. Work in this direction is in progress [21].
The paper is organized as follows. In section 2 we review the needed material for constructing the boundary state in BCFT [18] and in OSFT [23]. Then we review the construction of the marginal solution in the pseudo-B 0 gauge [2], and we explicitly write it down up to the fifth order. Section 3 describes the regularization procedure implemented by the propagator B 0 L 0 . In section 4 we write down the coefficients of the Ishibashi states in the boundary state in terms of the deformation parameter λ using the standard BCFT prescription by Recknagel and Schomerus [18]. Then we compute the same quantities for the OSFT solution in the pseudo-B 0 gauge. Finally we compare the coefficients of the Ishibashi states in OSFT and BCFT and we obtain the functionλ =λ(λ) up to fifth order. An appendix contains useful formulas for the encountered correlators.

The boundary state and the marginal solution
Let us consider a deformation of a BCFT by a boundary primary operator j(x) of conformal weight one From the OSFT point of view the new theory can be described by a classical solution, a state in the original BCFT The leading term inλ satisfies the linearized equations of motion If j is exactly marginal higher orders inλ should exist and they can be found by solving the recursive equations of motion with the initial condition (2.3). Notice that while in BCFT the perturbation is unique, the OSFT solution is not unique because it can be changed by gauge transformations. We can get rid of this gauge redundancy by computing observables. In particular the information on the marginal deformation can be effectively cast in the boundary state.
Boundary states in bosonic string theory can be written as a superposition of Ishibashi states where |B gh is the universal ghost part. When we deform a given worldsheet theory with an exactly marginal boundary deformation, the boundary state will be deformed to where [. . .] R means that a regularization is needed (and it will be reviewed later on), and |B 0 is the boundary state of the starting BCFT. On the other hand, given an OSFT solution Ψλ, the boundary state will depend onλ The two boundary states should be the same by the Ellwood conjecture [25] and this induces a functional relationλ =λ(λ) . where V m | is the BPZ conjugate of the matter Virasoro primary |V n so that V m |V n = V m |V n = δ n m where we used the fact that Ishibashi states have the generic form |V n = |V n + Virasoro descendants. (2.13) The series expansion of the exponential in (2.12) gives rise to contact divergences and one needs to renormalize them properly. In the next section we will review the standard procedure of [18]. The way to compute the n m 's from OSFT was given in [23] by appropriately generalising the Ellwood invariant where Ψ TV is a tachyon vacuum solution and V (0,0) m (i, −i) is a weight (0, 0) bulk field of the form As explained in detail in [23], the auxiliary bulk field V (1−hm,1−hm) aux lives in an auxiliary BCFT aux of c = 0 and has unit one-point function on the disk In a similar way the open string fields entering in (2.14) are lifted to the extended BCFT For the solution we will be dealing with this lifting procedure is trivial and amounts to the substitution L n → L n + L (aux) n in the equations that will follow. For this reason we will not distinguish between normal and lifted string fields in the sequel.
As far as the solution itself is concerned, we search for it in the convenient pseudo-B 0 gauge [2], making the following ansatz with the gauge condition B 0Ψr |0 = 0 , (2.19) where B 0 is the zero mode of the b ghost in the sliver frame, obtained from the UHP by the conformal transformation z = 2 π arctan w 20) and the operators U r are the common exponentials of total Virasoro operators creating the wedge states [27] in the well known way Solving order by order inλ (2.6) we find The rhs is explicitly given by where the cj insertions are written in the sliver frame. The solution is therefore where L 0 is the zero mode of the energy-momentum tensor in the sliver frame, Note that inverting Q B using B 0 /L 0 is only meaningful if the OPE of cj with itself does not produce weight zero terms, otherwise we would find a vanishing eigenvalue of L 0 . As is well known this is the first nontrivial condition for j to generate an exactly marginal deformation.
At the third order the solution Ψ 3 is written in terms of Ψ 2 . We write the state [cj, where in the second step we explicitly write the width of the wedge states using the U r operators. The inside insertions have to be placed according to (A.2): to lighten a bit the notation we have defined the graded commutator-like symbol [[. . .]] as Finally we can write the solution at the third order as We find the fifth order as This procedure can be continued to higher order 7 . Although higher orders can be easily written down, their Ellwood invariants become more and more complicated because they involve a large number of multiple integrals which by themselves need to be properly renormalized, as we will see in the next section.

Contact term divergences and the propagator
The computation of the Ellwood invariants for the solution we have just presented involves in general contact divergences due to the definition of the propagator B 0 /L 0 . As usual, we start by defining the inverse of L 0 via the Schwinger representation which is well defined for eigenvalues of L 0 with a strictly positive real part. The operator s L 0 is the generator of dilatations z → sz in the sliver frame. Its action on a primary field with conformal weight h in the sliver frame is The integral representation (3.1) is only valid for fields with a positive scaling dimension h > 0, if we apply the above integral representation to a state |ϕ −|h| with negative weight −|h| we find a divergence, as s approaches zero. But this is just the reflection that the integral representation (3.1) has been used outside its domain of validity. This can be easily remedied in the following way This prescription amounts to computing the Schwinger integral in its region of convergence by assuming Re(ǫ) > |h|, and then analytically continuing to ǫ = 0 8 This analytic continuation allows to define L −1 0 on every state we encounter during our computations except on weight zero states which remain as an obstruction, as it should be. 9 Pragmatically, this procedure is equivalent to add and remove the tachyon contribution from the OPE, for example and to define 1/L 0 on the tachyon as 4 Comparing λ andλ In this section we perturbatively compute the coefficients of the series expansion ofλ =λ(λ) up to fifth order. On general grounds we expect that b 0 = 0 and b 1 = 1, and this will be verified in the next subsections. The b k 's are computed by equating the coefficients of the Ishibashi states in the boundary state in BCFT and OSFT [22,23] n BCFT 8 An equivalent prescription is the Hadamard regularization, we thank M. Frau for discussions on this. See also [2,26]. 9 Notice that one could in principle define L −1 0 on negative weight states as however this integral representation does not work for positive weight states. Since the star product generates both positive and negative weight states at the same time, we need a representation of L −1 0 that works on the whole set of fields (except, of course, the weight zero fields).
In both cases one can expand the above coefficients in power series of the corresponding deformation parameter The B BCFT k,m coefficients can be found expanding the exponential in (2.12) where the conformal factor 2 2hm comes from the transformation of V m under the map from the disk to the UHP. These integrals need a renormalization, discussed by Recknagel and Schomerus [18]: thanks to the self locality property of the current j, one can modify the path of each integral to be parallel to the real axis but with a positive immaginary part ǫ, with 0 < ǫ << 1, In such a way all the contact divergences between the currents are avoided and only the contraction of the currents with the closed string will give contribution. Thanks to this renormalization the loop operator e −λ j(s)ds R becomes a topological defect. For the sake of simplicity we consider an exactly marginal deformation produced by the operator on an initial Neumann boundary condition of a free boson compactified at the self-dual radius R = 1(α ′ = 1) 10 . This deformation switches on a Wilson line in the compactified direction which can be detected by a closed string vertex operator carrying winding charge where m is the winding number (which specifies the closed string index) andX(z,z) = X(z) − X(z) the T-dual field of X(z,z) 11 . This closed string state has conformal weight ( m 2 4 , m 2 4 ). Performing the renormalized integral (4.4), one obtain with this choice of the current and closed string state (see appendix A for conventions and basic correlators) Since we are considering a compactified theory at the self-dual radius, there are other marginal operators in the enlarged SU (2) chiral algebra. Our choice (4.6) is equivalent to the chiral marginal operators i √ 2 sin (2X(z)) and i √ 2 cos (2X(z)), which have been studied in a similar context in [20,22] 11 If R is not self dual, our computation goes on unaffected by replacing the self-dual winding mode m with the winding mode at generic radius, mR.
which can be easily resummed to (4.9) In the OSFT framework, the analytic computation of coefficients of the Ishibashi state involves the Ellwood invariants and we compute them order by order inλ, starting from (2.14) which gives (4.10) The relation between λ andλ must be universal, in the sense that it cannot depend on the particular choice of the closed string. In our specific computation we will see that this is the case by verifying that the relation is independent of the winding charge m.

Zeroth order
As a starting consistency check, the zeroth order of the expansion of the coefficients of the Ishibashi states in OSFT is where as explained in [23] the tachyon vacuum contribution can be replaced with Ψ TV → 2 π c(0)|0 . The amplitude then becomes 12 where we used the conformal map defining the identity string field f I (z) = 2z 1−z 2 . Then

First order
As an extra starting check, let us look at the first order, where we have to compute where in the last step we wrote the correlator on a cylinder of width one C 1 , without any conformal factor because the conformal weight of all the insertions is zero. Acting with the map z → e 2πiz , (4.23) this two point function on the cylinder becomes the two point function on the disk D, which equals the amplitude computed from BCFT (4.8), and so the corresponding coefficient in theλ/λ relation is b 1 = 1 . 12 From now on we will write Vm instead of V (0,0) m to denote the lifted closed string state associated to the spinless matter primary Vm.

Second order
At second order the Ellwood invariant contains one B 0 /L 0 propagator inside Ψ 2 , This amplitude is depicted in figure 1. The action of the propagator on the double insertion of cj follows the regularization (3.3), so this state can be written as The B 0 ghost acts on c(w) as the contour integral (4.30) Therefore (renaming s/2 → s)Ψ 2 simplifies where we have used the obvious rotational invariance of the bc CFT on the cylinder. Using Wick theorem (which is reviewed in Appendix A) and in particular (A.17) we obtain where the ǫ prescription acts only on the first term because it is the only one which is divergent. This gives here we have used our analytic continuation which, as explained in section 3, amounts to computing the integral in the region of the ǫ-complex plane where it converges (Re ǫ > 1) and then analytically continue to ǫ → 0. In doing this we have also took the freedom of ignoring convergent terms proportional to ǫ since we are only interested in the ǫ → 0 limit.
Computing also the other convergent integral, we obtain again perfect match with the BCFT results B SFT 2,m = − m 2 π 2 = B BCFT

Third order
At this level the amplitude we have to compute is where Ψ 3 is defined in (2.28). This amplitude is depicted in Figure 2. Explicitly we find where ǫ 1 is the regulator for the most internal propagator (the one inside the lower order con-tributionΨ 2 (2.23)) and ǫ 2 is the regulator for the external propagator. From the perturbative construction of the solution, it is clear that ǫ 1 should be analytically continued to zero before ǫ 2 . Using the symmetries of the correlator in the matter and ghost sector and renaming t/2 → t, the whole Ellwood invariant reduces to It is useful to change variable with x = 3 2 y and s = ty = 2 3 xt so as to rewrite the integral as (4.43) The first integral contains a divergence in j(s)j(−s) when s approaches zero. Explicitly using (A.13) this part of the amplitude is given by  Notice that the integral in x is convergent which tells us that we could have avoided the ǫ 2 regulator. This is because the external propagator acts on a state which is in the fusion of three marginal operators and therefore it cannot contain the tachyon in its level expansion. Summarizing: from the BCFT side we found that the third order is proportional to m 3 (4.8) and there are no other terms. Instead, in the OSFT computation, at the third order we still get the same BCFT number proportional to m 3 but in addition to it there is another contribution coming from the peculiar renormalization implicitily defined by the propagator B 0 /L 0 . This is the first time in which there appears a discrepancy between the two approaches. As a consequence the third order coefficient in theλ(λ) relation (4.1) is

Fourth order
The fourth order Ellwood invariant is given by there are two contributions coming from the Ψλ solution (2.29) |0 . (4.48) The Ellwood invariant at this order is given by

(4.49)
First term V mÂ2,4 In the first term, as before, we need to compute the commutator of the insertions and apply the propagators,Â |0 . (4.50) The corresponding amplitude is depicted in Figure 3. Applying the two propagators, the amplitude takes the form . (4.51) Using the symmetries of the problem translating the correlators ξ → ξ + z 2 and changing variables w = sz and x = yz, the amplitude simplifies (4.52) The second term is the Ellwood invariant ofÂ 3,3 , and it is depicted in Figure 4. Explicitly we have to compute the star product of two Ψ 2 and then act with a propagator B 0 /L 0 : . Again the four different insertions of the ghosts contribute in the same way, and therefore (4.55) With the change of variable x = zt and y = zs, V mÂ3,3

4,m
The complete term at this order is given by summing the two integrals (4.52) and (4.56),

Fifth order
At the fifth order the solution is composed of three terms, |0 . (4.64) Then the Ellwood invariant we have to compute is (4.65) First term V mÂ2,5 The first term involves the statê The amplitude to compute is depicted in Figure 5 and after some manipulations involving changes of variables and conformal transformations we get 2πi V mÂ2,5 where the Schwinger parameters t 1 , t 2 , t 3 , t 4 are related to the integration variables as (4.68) Second term V mÂ3,4 The state here isÂ where the integration variables are related to the Schwinger parameters as (4.71) Third term V mÂ 3, 3 2,5 The last term involves the statê where the Schwinger parameters are related to the integration variables as (4.75) The a 5 coefficient is easily computed which gives the usual exact match with the BCFT results. As far as a 3 is concerned the computation follows closely the fourth order (with one more integral) and everything can be analitically done giving the result a 3 = 9 log 2 . This is precisely the needed number to ensure that b 5 is m-independent (4.16) so this is a consistency check.
Let us now address the a 1 coefficient, which is determined by the O(m) winding number contribution in (4.75). This is generated by the term from the Wick theorem with the maximal number of contractions between the j's and computing the four dimensional integrals (coming from the three diagrams) analitically is not possible. Therefore we procede analitically as far as we can and then we resort to numerics. The Y and Z integrals can be analitically computed in all of the three diagrams, including the subtraction of the tachyon divergence. Rescaling the X variable in the first diagram X → 2X, the O(m) contribution E i from each diagram is reduced to an expression of the form where the function f i are known analitically. To renormalize the tachyon divergence in the X integration we explicitly subtract the second order pole in X from the function f i in the following way .
(4.79) It turns out that the coefficient of the 1/X 2 pole, which in the above formula is indicated as This treatment leaves us with three numerical functions of T , which have to be integrated in the inteval 0, 5 2 . Surprisingly each of these functions shows a nonvanishing 1/T pole, as shown in Figure 8, . (4.81) These poles are potentially problematic, and it is reassuring that the sum of them vanishes see also Figure 9. This is an important consistency check, because a 1/T pole would be an obstruction to the existence of the solution at the fifth order. Numerically computing the integral over T finally gives a 1 = 10.58226 (7) . To conclude the total contribution to fifth order is given by

A Conventions and correlators
Witten's star product Witten star product is best understood in the sliver frame (with coordinate z), which is related to the UHP (with coordinatew) by the map In this frame the star product of wedge states with insertions is given by where the coordinates are in the" 2/π" sliver frame (A.1), which implies some rescaling wrt to (2.24) of [1], where the 2/π factor in (A.1) was omitted.

Upper Half Plane
In the BCFT of a free boson X at the self-dual radius R = 1, with Neumann boundary conditions, consider the bulk winding mode (A.5) The current j(x) has the two point function j(x) j(y) UHP = 1 (x − y) 2 , (A. 6) and it has the following OPE with the bulk winding mode V m (z,z) = e imX(z,z) We also have X (i, −i) ∂X(x) and, using Wick theorem (

Wick theorem
In the main text we deal with correlators of the following form (im) k k! : (X(z,z)) k : This correlator significantly simplifies on a cylinder C n , when the bulk operator (properly dressed with the ghosts and the auxiliary sector to acquire total weight zero, (2.15)) is placed at the midpoint i∞. In particular , thanks to the rotational invariance of C n we have The term containing the maximal number of contractions with the closed string is given by : (X(i∞)) N : where the only non-trivial correlator involvingX is given by (A. 18)