A simple solution for marginal deformations in open string field theory

We derive a new open string field theory solution for boundary marginal deformations generated by chiral currents with singular self-OPE. The solution is algebraically identical to the Kiermaier-Okawa-Soler solution and it is gauge equivalent to the Takahashi-Tanimoto identity-based solution. It is wedge-based and we can analytically evaluate the Ellwood invariant and the action, reproducing the expected results from BCFT. By studying the isomorphism between the states of the initial and final background a dual derivation of the Ellwood invariant is also obtained.


Introduction and Conclusion
A major question in Open String Field Theory (OSFT) is how the different sets of conformal boundary conditions, in a given closed string background, can be described by the gauge orbits of classical solutions. Hidden in this correspondence there is the mechanism by which OSFT is supposed to tame contact-term singularities. In the sigma-model approach one can formally move in the space of two-dimensional boundary field theories by means of boundary interactions. However, when interpreted as operator insertions in the world-sheet path integral of the starting background, such interactions have notorious contact-term problems. The advantage of Witten's cubic open string field theory, in this regard, is that contact-singularities can be naturally avoided by expanding the string field in the Fock-space basis (level truncation), thanks to the explicit "security strips" that every Fock-space state has. However, the level expansion is not well fit for analytic computations. On the other hand, with the standard wedge-based analytic methods we have today, essentially stemming from Schnabl's original work, [1], it is not known how to systematically deal with contact term divergences.
Notable progress has been achieved in the case of boundary marginal deformations, [2], in [3,4,5,6], where consistent ways have been devised to regularize and renormalize the contact divergences of boundary marginal operators, order by order in a perturbative With the assumption that the g-function doesn't change, this gives a dual derivation of the Ellwood conjecture.
Despite the very simple algebraic structure, however, the behaviour of the solution towards the identity is, still, potentially problematic since we encounter a new, previously un-noticed, singularity which occurs when negative weight fields (such as the c-ghost) are placed off the boundary on a vanishing width wedge state. We devote an appendix to a preliminary presentation of these new kind of singularities which would deserve, by themselves, further study and whose presence, if not properly tamed, can be quite dramatic. Luckily, it is possible to avoid these singularities by deforming the original solution into a one-parameter gauge orbit which is safe by construction and which reduces to our original solution in a limit. Quite remarkably, the observables of the regularized solution can be exactly shown to reduce to the difference in observables of tachyon vacuum solutions, where the regulator can be safely removed.
The solution we are proposing is quite handy (essentially as easy-to-handle as the original KOS solution) and at least for chiral marginal deformations is hopefully more advantageous than the standard approaches for singular OPE's such as the counter-terms generalizations of B-gauge solutions [3,4] or the general method of [5,6], which are perturbative approaches in the marginal parameter. Our construction is based on the TT solution and hence on marginal deformations, but the algebraic structure we describe is completely general. We thus hope our results can be a useful step towards the analytic construction of more general backgrounds in open string field theory, whose numerical landscape has been recently shown to be vaster than what is known analytically, [13,14].

From TT to KOS
In this section we first review the needed ingredients from the Takahashi-Tanimoto (TT) solution, [8], formulated in the sliver frame. Then we show that, after a gauge transformation, the TT solution is mapped to a new solution which is algebraically identical to a KOS solution [11].

TT Solution
We start with a chiral current algebra  a (z) b (0) = g ab z 2 − c abc z  c (0) + (reg.), (2.1) and its antiholomorphic counterpart  a (z) b (0) = g ab z 2 − c abc z c (0) + (reg.), (2.2) with totally antisymmetric structure constant c abc . Our reference BCFT 0 is chosen to preserve a linear combination of the two isomorphic chiral algebras, from which it is possible to define a single chiral current, defined on the whole complex plane (doubling trick) j a (z) =  a (z), Im z > 0 (2.3) j a (z) = Ω ab b (z), Im z < 0, (2.4) where Ω ab is gluing map which is part of the data which define the starting background BCFT 0 .
The current algebra structure (2.2) guarantees that each j a (z), when placed at the boundary of the worldsheet, generates an exactly marginal boundary deformation of BCFT 0 , [2]. The TT identity-based solution can then be written as a state in BCFT 0 as The f a (z) are functions defined on the imaginary axis, whose properties will be derived shortly. Here we are employing the rather formal but quite useful notation [15,7] φ(z) ≡ e zK φe −zK , which allows to manipulate string fields as if they were local operators on the world sheet 2 . For generic z, φ(z) is a formal string field which only makes sense if it is multiplied (from the correct side) by a wedge state of minimum width |Rez|. When Rez = 0, φ(z) is an identity based string field which can be given a Fock space expansion and which can be multiplied by wedge based states. The identity-like string field φ is defined as whereφ(w) is a local vertex operator in the 2 π arctan-sliver frame, and I is the identity string field.
For concreteness we will specialize to a single polarization inside the current algebra (2.2), by choosing one single current for constant t a , with OPE j(z)j(w) = 1 (z − w) 2 + reg, (2.8) although most of our results readily apply to the fully non abelian case (2.6).
With this understanding we explicitly write Given a generic vertex operator φ(z) in the sliver frame, the Fock space definition of the identity-based string field Φ is given by computing a correlator on a cylinder C L of width L = 1 In order for Φ to have well-defined Fock space coefficients (2.10, the function f (z) must vanish fast enough at the midpoint ±i∞, so that the dz integral will be finite. The finiteness of the first term involving cj(z) gives the generic condition where is the contraction between cj on the imaginary axis and the test state at z = 1/2. This condition essentially states that f (z) should be integrable towards ±i∞. The finiteness of the second term involving c(z) gives a much stronger constraint since the negative weight field c must be damped as it approaches the midpoint. For example, by contracting with c∂c(0)|0 , we get the condition Other contractions with ghost number two Fock states similarly imply that f (z) must separately vanish at ±i∞ at least exponentially, faster than e −π|z| , to make the integral convergent. We will see in the appendix that the requirement of finite contractions with generic wedge based states will further damp the behaviour of f at the midpoint. Let's see how the equation of motion works in the sliver frame. In order to consider QΦ + Φ 2 as a concrete thing, we need some worldsheet, since this is not provided by the solution itself. Let us then consider 14) The kinetic term readily gives The interaction term gives three possible contributions We now demand that f (z) is analytic in an infinitesimal strip containing the imaginary axis. 3 Then, since f is also suppressed at the midpoint, we can slightly shift the dz integrals on the left and on the right of the imaginary axis, respectively for the first and second terms in the parentheses (while staying on the surface thanks to the added strips of worldsheet). Then the two terms in the parentheses are equivalent to a contour integral around w 4 Only the cj-cj OPE can give a simple pole and therefore a non vanishing result Since the solution is identity-based, it is not possible to directly compute its observables, because they would correspond to correlators on cylinders of vanishing width. To appreciate this, let's compute a possible (naive) regularization of the kinetic term by simply inserting small regulating strips, for a choice of function f (z) = e z 2 , which is well suppressed at the midpoint. We get (2.20) Not only the limit (ǫ 1 , ǫ 2 ) → 0 does not exist, but it also wildly oscillates from −∞ to ∞.
Despite the failure of a naive direct evaluation of the action, following the discussion in [8,18], the solution is expected to describe a marginal deformation with marginal parameter given by the reparametrization invariant (see appendix A for the relation between f (z) and F (w)) This quantity is real if the reality condition (A.10) is obeyed. As discussed in [7], it is useful to define the matter string field 5 22) and the deformed worldsheet hamiltonian generating horizontal translations on the cylinder C L whose BRST variation is given by 6 (2.24) The string field K + J is exact in the cohomology of the shifted BRST operator where we have used the notation of [19] for the kinetic operator between two backgrounds A and B Generic functions of K ′ are thus killed by Q ΦΦ The string field F (K ′ ), if analytic for Re K ′ ≥ 0, can be geometrically understood as a superposition of wedge-states with a path-ordered exponential integration of the chiral current, [7], in much the same way as [11,20,21] Notice however that the exponential interaction integrates the marginal current j(z) over the whole bulk. This bulk (rather than boundary) integration is what naturally regularizes the contact term divergences between the j's. The more common BCFT intuition of a renormalized boundary interaction, [2], can be achieved by studying the phantom term of the solution [22], along the lines of [19,23], essentially observing that very large deformed wedges can be reparametrized to finite size while localizing the function f (z) to the boundary. Indeed, considering the scaling derivation [1] and we can easily show For t → ∞ (which is the needed rescaling to bring the sliver to finite width) the support of the function tf (tz) gets localized to Im z = 0 and the bulk interaction (2.29) localizes to the boundary, see also [10] for an almost equivalent mechanism. In [7] it was also proven that (appropriately normalizing the space time volume) This correlator is a regularized expression for the marginally deformed disk partition function which should therefore coincide with the undeformed one, as it is the case.

KOS-like solution
Using the ingredients discussed in the previous subsection, we can write down the solution 7 Although not self-evident, this solution falls in the class of solutions studied by Kiermaier Okawa and Soler (KOS), [11]. To see this we formally write where the string fields σ L,R 's obey the algebraic properties The expression (2.36) is precisely the pure gauge form of the TT solution, advocated in [8]. If we assume the existence of a logarithmic chiral field χ(z) which is a 'primitive' for j(z),

41)
7 This is obtained via the "Zeze map", [24], and, in our case, F (K) = 1 1+K ). Because the map is a gauge transformation it maps solutions to solutions and it can be useful for turning identity-based solutions into more regular ones. It is not guaranteed, however, that the "identity-ness" can always be removed by gauge transformations, the residual solutions of [25] being a counter-example.
then we can write 8 One can explicitly verify (2.36) by appropriately differentiating the operator/star exponentials defining the σ's. As elaborated in [17], we can try to trivialize the solution Φ by making χ f an allowed state, integrating by part where, with no loss of generality, we choose h(z) as However, since χ(z) is logarithmic, the boundary term only vanishes if The parameter defined in (2.21) is thus zero if and only if the solution Φ can be trivialized. Otherwise, if Φ is non trivial, the σ's are formal objects which do not belong to the state space of BCFT 0 (very much like bcc operators). The use of the σ's is nevertheless quite useful to rewrite some of the objects we previously defined. In particular we have and (2.48) 8 As an explicit example one can take j = i √ 2∂X and χ = √ 2X, for a free boson. Notice that the exponentials defining the σ's are not normal ordered (the contact singularities are spread in the bulk).
Notice that, differently from the original paper by KOS, the formal string fields σ L,R don't correspond to local boundary insertions of weight zero matter primaries, and their worldsheet realization is only meaningful when a pair of them appears where the non-local operator J(s) is defined in (2.30). In the following, whenever possible, we will avoid using explicitly σ L,R and instead use the more general expression (2.35). At will, one can easily switch between the two notations, having (2.36, 2.49) in mind. In subsection 3.3 we will elaborate more on the σ's in presence of generic vertex operators. Notice also that the auxiliary derivation B − ≡ 1 2 (B 0 − B * 0 ) doesn't annihilate Φ and therefore, contrary to the original KOS construction, the solution is not in a dressed B-gauge, [16,11]. This matches with the expectation that a solution for marginal deformations cannot be found in a dressed B-gauge when, as is generically the case here, the marginal field has singular OPE with itself.
As a side-comment 9 , notice that given the objects, (σ L,R , K) one can also construct a Kiermaier-Okawa-like solution, [5], via the substitution of the building block where the λ dependence in the σ's, (2.42)(or equivalently in J) is realized by choosing In this case everything is already finite and directly applies to the case of a marginal field with singular self-OPE (assuming it is local wrt all the fields in the theory, which is true if it belongs to the chiral algebra).

Ellwood invariant
To compute the Ellwood invariant [9], and thus the boundary state [13], we use a simple but powerful trick. Writing the solution as, [12] the Ellwood invariant is easily evaluated by inserting the KBc-identity as 10 where, in going from the second to the third line, we have used the identity What we have obtained is precisely the difference of the invariants of the Erler-Schnabl solutions in the original background and in the background expanded around Φ. (3.8) The first observable has been computed in [16], while the second has been computed in [7] and shown to reproduce the closed string tadpoles of a marginally deformed BCFT at deformation parameter λ BCFT given by (2.21). We will present an alternative derivation of this result in section 4. Notice that all traces in the game involve computation of correlators on cylinders of generic finite width, by the usual Schwinger parametrization of 1 1+K = ∞ 0 e −t(1+K) . In addition, our algebraic derivation is also applicable to the regularized solution (C.6) discussed in the appendix, which has the advantage of having support on wedge based states with strictly positive width, thus avoiding the potentially problematic t → 0 limit in the overall Schwinger integral 10 The notation is as follows where I| is the bpz of the identity string field and V is a weight zero bulk operator V = ccV matter .

Action
Using a similar trick we can evaluate the action. Dropping the trivial BRST exact pieces in (3.1) and appropriately rotating the trace we have This quantity can be in principle computed as the partition function of a wedge state with insertions and deformed/undeformed regions, with four Schwinger parameters to integrate over. This doesn't look simple at all. But let us insert [B, c] = 1 rightmost in the trace and, as we did for the Ellwood invariant, pull out the adjoint action of B on the other string fields in the trace (3.10) Using (3.5) three times, we get some cancellations and we end up with Now recognize the BRST-exact quantities which allow to integrate by part Tr where, in the third line, we have defined and in the last two lines we have used the cyclicity of the trace, the algebraic property (3.5) as well as (3.16) Using (2.26) we can therefore write The last line vanishes on account of the generic property Therefore the action evaluated on the solution equals where in the last line we have specialized to our precise case where [Φ, c] = 0. Same as the Ellwood invariant, this is precisely the difference between the action of the Erler-Schnabl solutions (3.7,3.8) in the original and Φ-background. Using (2.33), we see that the two actions equal each other, [7], as it is expected since the solution Ψ describe a continuos family of marginal deformations of the perturbative vacuum and must have therefore a vanishing action. It should be noted, however, that vanishing of the action is not an algebraic consequence of our derivation. There is a reason for this: given any solution Φ, one can always construct a gauge equivalent KOS-like solution and follow the computation of the energy we have just presented, to reduce it to the shift in the tachyon vacuum's energy. Since Φ can be a generic solution, there is no reason to expect to find a vanishing action. Therefore the algebraic form of the KOS-type solution we are discussing, can be useful for generic backgrounds, not just marginal deformations.

Deformed background
We can easily describe the states and the cohomology representatives in the new open string background described by the solution Ψ. In order to do so let us first address, in our formalism, the construction of the fluctuations around the TT-solution Φ itself, which was discussed in part in [8,17]. This will allow us to make some interesting connection with the standard BCFT description of a marginal deformation [2] and to perform an alternative, simpler, computation of the Ellwood invariant. Let Ξ be a Fock state around the starting background Ψ = 0 where V ≡Ṽ (1/2)I is an identity-like insertion. Since the TT solution can be written as this implies an isomorphism with the deformed stateŝ Notice that if the σ's would have been allowed fields, this would just be a gauge transformation. Explicitly, using the appropriate generalization of the Leibniz rule, [19], we see that the cohomology problem at the TT-background is mapped to the cohomology problem at the perturbative vacuum 12 It appears that the dressed vertex operatorsV = σ L V σ R are the only objects where a concrete definition of the σ's is needed, (2.42) However, the * -commutator [χ f , ·] can be rewritten using only local fields (while this is not true for left or right multiplication alone). Explicitly we can write (Rez "time ordering" is understood between the string fields χ and V ) The OPE between χ and V can be a pole or, in case V is a j-primary (meaning that j-V is a single pole), a logarithm. Other cases are excluded because j belongs to the chiral algebra and it is thus local wrt all bulk and boundary fields. When χ-V is a pole, we can close the two vertical contours and shrink around 0 Consider now a primitive for f (z), integrating by part the closed contour we get Notice that, under the assumption we are temporarily holding (j-V is not a simple pole) the integration constant in g doesn't play any role. The constant part of g enters the game only when we transform a j-primary, so that j-V is a single pole and χ-V is a logarithm. In this case we can assume we have already diagonalized the j-primaries V in such a way that they are eigenstates under the action of j and we can write 13 This left vertical integral (which is finite because of the integrable singularity of the log and the fall-off of f (z) at i∞) precisely cancels (together with the regular parts) against the right vertical integral in (4.6). Therefore we have We can conveniently summarize the result as This can be exponentiated to givê Suppose now we want to displace V (0) off the boundary . 13 We are also assuming that f (ix) = f (−ix), which allows to easily deal with the unphysical cuts in the logarithm (which are an artifact of the presence of χ). This condition was also implicitly used in the first of the papers [8]. Notice that a violation of f (ix) = f (−ix) would not change λ BCFT as defined in (2.21).
To compute the marginal transformation, we follow the above derivation and, again, we have to pay attention when V is a j-primary. In this case we have When we add the contribution from the right vertical path as in (4.6), the part proportional to log z cancels exactly as before, but now there is in addition the term where the cut has been chosen so that the overall contribution vanishes when x → 0. Therefore, also for holomorphic bulk insertions we find Notice that when the pole between j and V is at least triple, the transformation will start evaluating the derivatives of f (z). This is another reason to require that f is analytic around the imaginary axis. Assuming f (z) can be holomorphically extended beyond the imaginary axis (which is typically the case), we can also writê As an example, we can derive how the energy momentum tensor T (z) is deformed by the marginal flow induced by the solution. We havê T (w) = σ L T (w)σ R = e i w dz 2πi g(z)j(z) T (w). (4.21) Using the j-T OPE j(z)T (w) = j(w) (z − w) 2 + (reg.) (4.22) we get, using ig ′ (w) = f (w) which agrees with [17]. As a consistency check we can also computeT (w) by taking the deformed BRST variation of the antighost b(w) Another simple universal example is given bŷ (4.25) and one can easily check that, just as Q(j(z)) = ∂(cj(z)), we have Q ΦΦĵ (z) = ∂ z (cĵ(z)) = ∂ ′ (cĵ(z)). (4.26) This example is also teaching us that the star algebra operator ∂ ′ = ad K+J acts on a deformed vertex operatorV (z) precisely as ∂ z . An important property of theV 's is that, as suggested by the notation, they obey the same operator algebra as the original V ′ s The constant g ′ g is the ratio of the traces of the deformed and undeformed wedges, which, as proven in [7], is equal to 1.
It is interesting to extend the marginal transformation (4.20) to closed-string bulk operators. In our doubling-trick notation a bulk operator will be written as w * ≡w, Im w > 0, (4.30) where both V i and V j are holomorphic (but typically not chiral) fields. We thus havê with g(z) defined in (4.14). Let us now assume that both V i and V j are j-primaries (all boundary states obtained from BCFT 0 by deforming with j, will be written as a sum of Ishibashi states of j-primaries, defined with the appropriate deformation of the gluing map, [2] from which we easily get 15 Now imagine we want to compute the Ellwood invariant of the tachyon vacuum solution (3.8), as it was done in [7]. After standard string field manipulations, we end up with the following trace where V ≡ ccV . We can follow (and generalize to finite x, still assuming f (iy) = f (−iy)) the explicit computation of [7] to find (4.36) 14 As an example, in case of j = i √ 2∂X, with Neumann boundary conditions, we have that bulk momentum modes have a i = −b j while bulk winding modes have a i = b j . The situation is exactly opposite in case of Dirichlet boundary conditions. 15 With our assumption f (z) = f (−z) we have that g(ix) = −g(−ix). Thus, with this condition, a bulk operator with a i = −b j is not transformed by the marginal deformation. But in fact a bulk operator with a i = −b j has a vanishing tadpole in BCFT 0 , and this remains true by deforming with j, [2].
Or we can proceed differently, (4.29) The closed string stateV ij is the inverse of the transformation (4.33) Therefore we get which coincides with (4.35), remembering that we are taking f (ix) = f (−ix) and Notice that, in this 'dual' derivation, the Ellwood invariant is precisely reduced to a deformed closed string tadpole, in the sense of [2] (see e.g. eq (3.3) there), and Ellwood conjecture is transparent. In the BCFT description of [2], the countours, encircling the bulk operator, were originally at the boundary, while in this peculiar OSFT description they originate from vertical line integrals, (4.6).
Assuming that the contour integral dz 2πi g(z)j(z) is well defined on local vertex operators (which is true if j belongs to the chiral algebra, but generically false if j is only self-local, [2]), the isomorphism (4.2) can be performed on the whole Fock space of BCFT 0 and, being a similarity transformation, it is clearly compatible with both the star product and the BRST differential. All the above can be straightforwardly extended to the KOS-like solution Ψ (2.35, 2.48), where the previous isomorphism is dressed with the gauge parameters connecting the TT solution with the KOS solution Again, we can use generic states of BCFT 0 , to describe the off-shell degrees of freedom around the new backgorund Ψ, and the cohomology is again mapped in the cohomology.
Notice that, with this construction of the off-shell fluctuations, the action in the new background Ψ coincide with the action around the TT background Φ. Notice however that the states (4.39) are not real despite the solution Ψ essentially is Given Ψ real one can find real cohomology elements by using the right gauge transformation [19] 1 − F K and the (reality-conjugate) left gauge transformation both connecting Φ with Ψ real . This gives a construction of the cohomology which is essentially the one considered in [3,27] This is a map from cohomology to cohomology but, contrary to the non-real construction (4.39), it is not a star algebra homomorphism. It is also possible, at least formally, to connect Φ with Ψ real with a real gauge transformation W = U † 1 √ U U † obeying W † W = W W † = 1, [28] which gives a star algebra homomorphism compatible with reality. However, we do not see obvious problems in using the simpler non real deformed states (4.39).

A TT on the upper half plane, BPZ and reality
To make contact with the original form of the TT solution, we relate the function f (z) to the function F (w) appearing in the work of TT [8] by mapping the semi-infinite cylinder C L of circumference L = 2 (with coordinate z) to the upper half plane (with coordinate w) by w = tan πz 2 .
where C left is the semicircle in the complex plane connecting −i, 1, i, oriented towards i.
In [8] and in the papers that followed, the authors also require that F (w) obey which, in the sliver frame, translates into We do not require this periodicity condition because (for example) a simple scale transformation in the sliver frame (a reparametrization generated by L − ≡ 1 2 (L 0 − L * 0 )) would not respect it. As explained in [8,17], this property ensures that so that we can write where the commutation between left and right charges holds if F (±i) = 0. However, to prove the equation of motion, as we saw in section 2, we only used that F (w) vanishes at the midpoint and that it is analytic in an infinitesimal neighborhood of C left . The corresponding right charge Φ R can be defined, if needed, by the same expression (A.4) but with which is a right-type charge which also vanishes at the midpoint and is analytic around C right . In this way (A.5) is still satisfied because we can use the generic properties which encode the gluing conditions for N-strings vertices. If we like, given F (w) defined on C left (Re w > 0) we can always extend F (w) on C right (Re w < 0) by F (w) ≡ 1/w 2 F (−1/w), for Rew < 0, but this isn't in general an analytic continuation. 16 Since, to define the solution, we only need to know F (w) on C L , we avoid talking about the value of F (w) on C R . The reality condition, on the other hand, gives a real constraint on F (w). The string field Φ is real (bpz=hc) if the function F (w) satisfies which in the sliver frame translates into the quite intuitive

B A new singularity towards the identity
The simple algebraic derivation of observables we have presented in section 3 is potentially endangered by a singularity towards the identity which has to do with the c-ghost, as we now briefly explain.
To start with, it is better to specialize a bit on the function f (z) which defines the solution. Because of the omnipresence of the quantity 1 1+K , a basic requirement is that, when we add Φ to K, B, c, its contraction is well defined against wedge-based states of arbitrarily small width. While this was essentially guaranteed in previous enlargements of the KBc algebra, [11,20], which dealt with boundary insertions, and even [7], where only matter operators were allowed to enter the bulk, here the story is more delicate. To appreciate the problem consider the simple overlap This integral is divergent for small enough t > 0 unless the function f (z) is suppressed at i∞ more than exponentially. For example, the standard choice by Takahashi and Tanimoto [8], does not respect this property. Indeed, although the TT-solution based on f TT is finite in the Fock space, a finite L − reparametrization of it, (2.31), appears to be singular 17 . In particular This is certainly un-welcome for the purpose of enlarging the K, B, c algebra with Φ, as we have been doing in the previous sections. Therefore we would like to limit the choice of f in such a way that generic contractions with wedge based states and finite L − reparametrizations give finite results. A simple example that does the job is the family of gaussians for which we have In the following we will specialize to the family of gauge equivalent solutions described by f t (z). These solutions are all related by L − -reparametrizations We can easily check that, for this choice of function, the TT solution Φ is finite in the Fock space, and also against generic wedge states with insertions whose width can be taken arbitrarily small. In particular, for example Notice however that, although the overlap is finite for s > 0, it nevertheless diverges super-exponentially in the identity limit s → 0. Sticking to this example, this means that the following overlap is badly divergent We may hope that further suppressing f at the midpoint could improve the situation, but in fact there is a more basic problem. When two c-ghosts have a separation with a tiny imaginary part on a cylinder of width t, the correlator always diverges in the limit t → 0. Notice that the negative scaling dimension of c would suppress the correlator, but this comes together with a rather violent exponential divergence, which only occurs when c is placed off the boundary. Therefore, even in the original K, B, c algebra we have the problem This is a new kind of identity-like singularity which would be worth studying by itself. Notice in particular that naive attempts to evaluate the action of the TT solution Φ are affected by this singularity, (2.20). Notice that the singularity is much more violent than previous identity-like singularities studied in the sliver frame, [25], whose behavior is typically power law. Since we know quite little about these singularities and how they effectively cancel in the algebraic computations we have been doing in the main text, our primary aim will be to show that these singularities can be avoided by an infinitesimal deformation of our solution.

C Regularization
The singular expressions we met in the previous section are structurally quite close to the expressions that appear in the computation of observables in the main text. For example, consider the kinetic term of the string field χ ≡ 1 1+K Φ, which is a part of our solution Ψ, (2.35). The explicit computation goes as follows If we consider the kernel we see that this is not a well defined quantity since, if we perform the dt integral first, we encounter a bad exponential singularity at t → 0 of the type e 2π |x−y| t . On the other hand, performing the q integral first, the dependence on (x − y) drops and everything is finite. In this case, it is not difficult to realize that the 'correct' prescription for computing the above integral would be to define With this regularization of the trace (cut-off in the overall Schwinger parameter of the string field whose trace we want to compute), the algebraic derivations of the Ellwood invariant and the kinetic term, presented in section 3, are rigorously justified, as it is easy to check. However it is very difficult to understand this regularization at the level of the individual string fields before * -multiplication and, importantly, to understand how the equation of motion in the action is violated and how (and if) it is restored when the regulator is removed. Our algebraic computation suggests there exists a prescription (which is consistent with the equation of motion) to correctly compute the observables. But to make this precise, we need a regularization which allows us to control the t → 0 limit in the overall Schwinger integral and to maintain, at the same time, the equation of motion. Perhaps the simplest and safest approach (but other strategies might be possible) is to realize that the solution we are dealing with can be obtained as a limit of a one parameter family of gauge equivalent solutions which have, generically, a minimum fixed width and therefore, by construction, cannot have any singularity related to the identity string field. To construct such a family is easy and amounts to choose a security strip in the Zeze map (2.34) given by (for example, other choices are of course possible ) where Ω ≡ e −K = |0 SL(2,R) . (C.5) The regularized solutions are given by These solutions span a gauge orbit interpolating from KOS (ǫ = 0) and the generalization of the Schnabl-KORZ solution, [3,4], (ǫ = 1) which, for completeness, reads The strategy is therefore to define the observables of the KOS solution as the ǫ → 0 + limit of the observables of the interpolating solutions (which, by gauge invariance, will be ǫ-independent). At finite ǫ it is guaranteed that no identity-singularity can affect the computation. However the price we pay for manifest regularity is that the generic solution in the orbit is much more complicated than the original KOS solution and it is not clear, at this stage, how one could compute the observables as we did in the main text. But in fact we can rewrite the regularized solution (C.6) again as a KOS solution, where the fields (K, B, c, Φ) have undergone the automorphism [26] c → c ǫ = c KB G ǫ (K) c (C.9) (C.13) (C.14) where G ǫ (K) is defined by 18 Notice that the TT solution remains invariant under the automorphism.
With the new variables, and some standard algebra, we can re-write (C.6) in few interesting ways The reader can explicitly verify that the first 'physical' term in the regularized solution (C.18) has support on wedge-based states of minimum width 2ǫ while the BRST exact one has minimum width ǫ. Notice that we have 19) which reveals that the automorphism mixes the objects in the game in a rather non trivial way. In particular which trivially descend from the automorphism, but which appear rather surprising in the original variables. Other notable quantities are given by notice that the automorphism doesn't increase the minimum width of the above quantities, which all continue to have a non vanishing support on the identity. Since the fields (K ǫ , B ǫ , c ǫ , Φ) have identical properties to (K, B, c, Φ) the computations for Ψ ǫ can be read-off from the main text by formally substituting (K, B, c, J) with (K ǫ , B ǫ , c ǫ , J ǫ ). By inspecting the involved correlators, we see than only well defined combinations of the deformed variables explicitly appear, so we have just to trace back how the simplifications in the deformed variables occur in the original variables. In doing this we encounter very non trivial simplifications between different structures, which would have been practically impossible to discover if not guided by the formal automorphism (C.9-C.13). For fixed ǫ = 0 we can then precisely show that the observables of the regularized solution Ψ ǫ reduce to the shift in the observables of the tachyon vacuum's When we take the ǫ → 0 limit these two solutions become the Erler-Schnabl solutions (3.7,3.8), which are manifestly safe from the identity singularities we encountered in the previous section, simply because no explicit Φ enters in their definition and therefore there is no c-field going off the boundary. For completeness, it would be interesting to have an analytic computation of the observables of Ψ (Φ) T V,ǫ , at finite ǫ, which we leave for the future.