KdV charges in $T\bar{T}$ theories and new models with super-Hagedorn behavior
Bruno Le Floch, M\'ark Mezei
TL;DR
This paper investigates how KdV charges in 2D CFTs and integrable models evolve under $T\bar{T}$ deformations, revealing their behavior as probes of energy flow and discovering models with super-Hagedorn density growth.
Contribution
It derives flow equations for KdV charges under $T\bar{T}$ deformation and introduces a Lorentz-breaking deformation leading to super-Hagedorn behavior.
Findings
KdV charges obey Burgers flow equations under deformation.
$T\bar{T}$ deformation preserves KdV charge conservation.
Lorentz-breaking deformation results in super-Hagedorn density of states.
Abstract
Two-dimensional CFTs and integrable models have an infinite set of conserved KdV higher spin currents. These currents can be argued to remain conserved under the deformation and its generalizations. We determine the flow equations the KdV charges obey under the deformation: they behave as probes "riding the Burgers flow" of the energy eigenvalues. We also study a Lorentz-breaking deformation built from a KdV current and the stress tensor, and find a super-Hagedorn growth of the density of states.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
**KdV charges in theories and
new models with super-Hagedorn behavior**
B. Le Floch1*, M. Mezei2
1 Philippe Meyer Institute, Physics Department, École Normale Supérieure, PSL Research University, Paris, France
2 Simons Center for Geometry and Physics, SUNY, Stony Brook, USA
July 2019
Abstract
**Two-dimensional CFTs and integrable models have an infinite set of conserved KdV higher spin currents. These currents can be argued to remain conserved under the deformation and its generalizations. We determine the flow equations the KdV charges obey under the deformation: they behave as probes “riding the Burgers flow” of the energy eigenvalues. We also study a Lorentz-breaking deformation built from a KdV current and the stress tensor, and find a super-Hagedorn growth of the density of states. **
Contents
1 Introduction and summary
The deformation of two-dimensional field theories has attracted significant attention recently due to its connection to disparate directions of research. It is a universal (and often leading) irrelevant operator near the infrared fixed point of renormalization group flows [1, 2, 3]. The deformation greatly increases the space of known integrable theories [4, 5, 6]. A novel deformation of S-matrices [7, 8, 9] was understood to be equivalent to the deformation of the Lagrangian [10, 11, 12], and also led to an alternative description as matter coupled to flat space Jackiw-Teitelboim gravity. See also [13]. Its relationship to the holographic renormalization group was explored in [14, 15, 16, 17, 18, 19, 20, 21, 22, 23]. -deformed theories and their generalizations share features with little string theories that are holographically dual to asymptotically linear dilaton backgrounds. This connection was explored in [24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34].
Partition functions in -deformed theories have been computed with a multitude of methods. The torus partition function was determined by a path integral over random metrics in [35, 11] and it has been proven to be a unique modular covariant partition function satisfying certain conditions [36, 37, 38]. The partition function was computed using large- factorization in [39, 40] and in the -deformed two-dimensional Yang-Mills theory in [41], see also [42] for analysis of the theory put on . Entanglement entropies were computed using the replica trick in [43, 39, 44, 45, 46, 47, 48, 49, 50].
Other solvable irrelevant deformations were considered in [51, 52, 53, 54, 55, 56, 57, 58, 59, 60]. Closed form Lagrangians often provide important insight into these deformed theories, and many have been constructed in [5, 61, 62, 63, 57, 58, 64]. Correlation functions were investigated in [2, 27, 16, 65, 19, 55]. The interplay between the deformation and supersymmetry was explored in [66, 67, 68, 69, 64]. The S-matrix of various worldsheet theories has been connected to the deformation in [70, 71, 72, 73]. For a pedagoical introduction, see [74].
In this paper, we continue the quest of finding solvable examples of spectra of quantum field theories deformed by irrelevant operators. The first such example was provided by the pioneering papers [4, 5] for -deformed theories and a very simple extension was solved in [22]. The spectrum of the -deformed CFTs was obtained in [28], completing the work of [51]. In [57], we used background fields to determine the spectrum of CFTs deformed by irrelevant operators built from , where the former are the (anti)holomorphic currents of the theory and the latter is the stress tensor. Some steps in the derivation of [57] (and also in the determination of the -deformed spectrum in [28]) were conjectural and only backed up by various checks. In contrast, in this paper we derive rigorously the flow of the quantum KdV charges [75] under the deformation, and determine the energy spectrum, KdV charges, and asymptotic density of states in the zero momentum sector under a deformation starting from a CFT. We often refer henceforth to the theory which we start deforming as the seed theory.
It was shown in [4, 5] that the energy spectrum of -deformed relativistic theories on the cylinder is governed by the equation
[TABLE]
where and are the energy and momentum eigenvalues, is the circumference of the circle, and is the deformation parameter. In this paper we derive that the quantum KdV charges of the eigenstate , if present in the seed theory, obey
[TABLE]
The allowed values of are . This equation was also obtained using integrability techniques of [5, 58]. Our field theory derivation applies more broadly, to the deformation of any Lorentz-invariant theory that contains at least one higher spin conserved charge, and hence rules out the possibility that the evolution equation (1.2) is a miracle of some special models.
There is a beautiful analogy with hydrodynamics. Equation (1.1) is the forced inviscid Burgers equation
[TABLE]
where the right-hand side is the forcing term, and we made the identifications
[TABLE]
and used that . Then (1.2) is translated to
[TABLE]
which has the interpretation of particles probing the Burgers flow (but not backreacting on it): the left hand side is the material derivative of and the right hand side is a forcing term. This equation is referred to as a passive scalar equation in the fluid dynamics literature, see the elegant review [76]. Admittedly, we have not encountered the particular forcing term in (1.5) in the fluid dynamics literature. For a CFT seed theory, we also solve these equations.
As the second major result of the paper, we obtain the evolution of the spectrum for -deformed relativistic theories, where is the current of the KdV charge , in the zero momentum sector:
[TABLE]
We are not able to derive a closed set of equations for sectors with that would generalize the equation above. Solving these equations for a CFT seed theory, we find that the eigenstates that start their lives as primaries in the CFT exhibit super-Hagedorn asymptotic density of states
[TABLE]
where is a number that we determine and is the central charge of the seed CFT.
Let us indicate the major steps in our derivation of the two main results by giving the outline of the paper. Section 2 is largely a review of [4]. We introduce the higher spin KdV currents and their charges, operators that have factorizing expectation values, and show that deforming a theory by quadratic composites of KdV currents preserve these symmetries. New results presented in this section are: the proof of factorization without the non-degeneracy assumption on the energy spectrum (which is important for CFTs, where there are many states degenerate in energy in a Virasoro module); the proof that the (possibly non-abelian) algebra of charges does not get deformed, and hence the KdV charges continue to commute in the deformed theory proving a conjecture made in [4]; and the generalization of the factorization property to new composite operators that are products of arbitrarily many factors. In Section 3 we use these results to derive an evolution equation for the KdV charges. An important step in the derivation is a novel formula for the expectation value of the space component of a KdV current as a length-derivative of the KdV charge in Lorentz-invariant theories. In Section 4 we apply results of Section 2 to a deformation. Superficially similar deformations were analyzed in [58] and our results partially agree despite fundamental differences in the two deformations, which we explain in Appendix F. Other appendices discuss various technical points used in the main text.
2 Change of KdV currents under irrelevant deformations
An important property of the class of irrelevant deformations built from an antisymmetric product of currents considered in [4] is that they preserve many symmetries of the undeformed theory: any current whose charge commutes with the charges of the currents building the deformation can be adjusted so that it remains conserved in the new theory. In the case of these are the currents that do not involve the coordinates explicitly. See Appendix G for a derivation of these facts.
2.1 KdV currents and the operators
Let us consider the deformation of a CFT first. Since the dilation current of a CFT, depends on the coordinates explicitly, dilation is not a symmetry of the deformed theory. Similarly the currents whose charges are the Virasoro generators with cannot be adjusted to remain conserved, thus most of the conformal group is lost. There is still a remnant of the infinite symmetry algebra in the deformed theory, and the maximal commuting set is formed by the KdV currents and charges, which in a CFT take the form
[TABLE]
where the stand for terms that involve derivatives and lower powers of the stress tensor. These can be adjusted to remain conserved after deformation, namely
[TABLE]
To show that this is indeed possible we have to review the methods of [4]. We work in the Hamiltonian formalism.
In a CFT the algebra of conserved charges is the universal enveloping algebra of the Virasoro algebra , which is formed by the sums of products of ’s.111There is an antiholomorphic copy as well. These charges are integrals of local holomorphic currents such as . These charges in general do not commute with the Hamiltonian , they are conserved because their noncommutativity with is compensated by their explicit time dependence. The maximal commuting set of these charges are the KdV charges. In fact, a noncommutative subalgebra of charges generated by , where is also preserved (with undeformed structure constants) by the deformation. We develop this direction in Appendix G. In the main text, we focus on the KdV charges only. These charges are also preserved in integrable massive deformations of minimal models.
To avoid needlessly duplicating later equations, we denote by , , the charges and currents denoted by , , in [4]. For , we get the left- and right-moving combinations of and that act as derivatives on local operators with no explicit coordinate dependence:
[TABLE]
Many derivations below are simplified by using these commutators instead of derivatives. Next, we use the fact that the commutativity of charges is equivalent to the integral of vanishing on any cycle, hence it is an exact one-form [4], which in commutator language can be written as:222In the notations of [4] our operators are equal to for , for , for , and for .
[TABLE]
i.e. the commutators on the LHS give derivatives of a local operator that is only defined up to addition of the identity. The unnatural position of indices in simplifies notations for antisymmetrization later on. From the definitions it follows that
[TABLE]
In Appendix A.1 we analyze some further basic identities obeyed by . In Appendix B we show in Lorentz invariant theories that
[TABLE]
2.2 Factorizing operators
Let us introduce the bilinear operators of [4]:
[TABLE]
which are spin- operators. The regulator terms are total derivatives and do not play an important role in any of the subsequent results. As we show in Appendix C improvement transformations of the currents only change the regulator terms, and hence drop out from subsequent results. A special case is the bilinear operator
[TABLE]
that we can informally call “” following the precedent set by the usage of in the literature. These composite operators defined by point splitting have remarkable properties. They obey factorization, in joint eigenstates of all KdV charges on the cylinder ,
[TABLE]
where we omitted writing the arguments of operators, as one point functions in eigenstates do not depend on the position of the operator. In Appendix A.3 we present an algebraic proof of (2.9), which relaxes the assumption of non-degenerate energy spectrum that was needed in [4], and also allows for the generalization of factorization to the operator:
[TABLE]
Note that defined in (2.7). The point-splitting regularization is detailed in Appendix A.2.
2.3 Special deformations preserve the KdV charges
As was stated at the beginning of this section, deforming the theory by preserves the symmetries (2.2). The proof proceeds by constructing the deformation of conserved currents under the irrelevant deformation. Let us assume that we deform the Hamiltonian by and ask what conditions need to be satisfied so that the current with charge remains conserved.
First we linearize in the coupling of to obtain:
[TABLE]
This equation can only be satisfied if is a total derivative, i.e. there exist such that
[TABLE]
and then
[TABLE]
obeys (2.11) thanks to the fact that the integral of a derivative vanishes, which we use in the form . In Appendix D we show that the condition (2.12) (with local) is enough to ensure that remains the integral of a local conserved current.
We just saw that the currents remain conserved if satisfies the condition (2.12). Let us check that (2.12) is obeyed for . We remind ourselves that according to (2.10)
[TABLE]
For details on the regulator terms see Appendix A.2. Using (A.6) we find
[TABLE]
where we used the definition (2.10). From (2.15) we conclude that (2.12) is obeyed with and .
In summary, we have that under deformation:
[TABLE]
Because we can add , the time component of a conserved current (with commuting charge) to the integrand in (2.16), there is some ambiguity in (2.16). Ambiguities are discussed and partially resolved in Appendix C. It can be verified that (2.16) leads to as assumed. (If it did not, we would have had to shift found in (2.15) by some conserved current to make the story consistent.)
A generalization of (2.15), namely (A.14), states in particular that , which implies
[TABLE]
where we used that integrates to zero. This proves Smirnov and Zamolodchikov’s conjecture in [4] that the deformation leaves the (adjusted) KdV charges commuting. We explain in Appendix G which parts of the story presented here generalize to nonabelian charges that do not commute with the KdV charges and to internal symmetry charges.
We remark that while the operators defined in (2.10) retain many of the nice properties of , deforming by them does not preserve the KdV charges, as the condition (2.12) is not satisfied for them. There is one other, somewhat trivial deformation that preserves the KdV charges, the deformation by or . Note that these operators can also be written as . These deformations by conserved current components correspond to turning on background gauge fields. The condition (2.12) is satisfied with (for the deformation) and (for the deformation), which is verified from the definitions (2.4). This choice is not good enough however, as it leads to . This problem can be taken care of by using the ambiguity discussed below (2.16) of adding conserved currents to . We work out the example of , as the case can be treated in complete analogy. We shift
[TABLE]
which then gives as required.
3 Evolution of the spectrum of KdV charges
3.1 Evolution under generic deformations
In Section 2 we understood how the KdV charges change under irrelevant deformations. Let us now choose joint eigenstates of the commuting charges , and denote their eigenvalues by . We can use the Hellman-Feynman theorem for the infinitesimal deformation to write:
[TABLE]
where we introduced as the coupling constant of , in the first line we used (2.16) and the spacetime independence of one point functions in energy eigenstates to evaluate the space integral, in the second line we used factorization (2.10), and in the third we used (2.5). We obtained an evolution equation for the change in the spectrum of conserved charges under irrelevant deformations.333For momentum () the second line of (3.1) gives , as required by momentum quantization. That such an equation can be derived is already remarkable, but to make the equation useful, we have to be able to determine the matrix elements .
While in the main text we focus our attention on KdV charges, we have not used any of their particular properties, and (3.1) applies to other conserved charges with the appropriate modifications. E.g. flavor symmetry charges (namely ) remain fixed: the equation gives because . A more general framework for charges is worked out in Appendix G. However, we leave for future work the incorporation of supersymmetry into the algebraic framework used in this paper.
A note of caution is in order: we have yet to fix the ambiguities corresponding to the mixing of conserved charges discussed below (2.16). Without this, (3.1) is just valid for one choice of KdV charges. In Lorentz invariant theories () we can use spin to prevent the KdV charges from mixing with each other. It can be checked that (2.16) and hence (3.1) respects spin. In fact, when the seed theory is Lorentz-invariant, we can use Lorentz-invariance even for , by assigning a spin to the coupling . This spurion analysis is performed in Appendix C.
We have analyzed and solved a problem similar to (3.1) for a family of deformations made out of an abelian current and the stress tensor in [57]. There we have also demonstrated that our current tools are inadequate to determine in general. In the rest of this section we focus on the case . In Section 4 we analyze a special case of (3.1) where we can make progress, while we discuss perturbative aspects in Section 4 and in Appendix F.
3.2 Evolution under the deformation
Let us determine the evolution of KdV charges under the deformation of a Lorentz invariant theory by plugging in into (3.1). Using (2.6), we obtain
[TABLE]
We still have to determine the expectation values . The sum of them gives
[TABLE]
but the difference, requires additional input. In general the expectation value of the spatial component of a current (which the quantity in question is), does not have a universal expression. In [57] we faced this problem for the case of an internal symmetry current and of , and we treated it by introducing background fields. For the case of KdV charges, we found another way to proceed.
Let us first set . From the interpretation of as pressure, we have
[TABLE]
Then transforming to complex coordinates,444We use the same conventions as in [57]. In (A.9) of that paper, we gave this result.
[TABLE]
and using (3.3) together with the valid in Lorentz invariant theories, we read off
[TABLE]
where in the second line we used (2.3) and that which follows from momentum quantization, . Similarly, we get .
This motivates us to compute . In Appendix E we show
[TABLE]
which reduces to the equations above for . Taking diagonal matrix elements and using that (valid in eigenstates of ), we find
[TABLE]
where in the second equality we used (2.6) valid in Lorentz invariant theories.
From (3.8) and (3.3) we can express and separately, and plug back their expression into (3.2) to find the flow equation:
[TABLE]
This is our main result. Setting and using (2.3) we recover the Burgers equation for , and the fact that remains undeformed:
[TABLE]
where in the second equation we used . The KdV charges obey linear equations, (3.9), which take as an input the energy eigenvalue that solves the nonlinear Burgers equation. We provided a hydrodynamical interpretation of these results in the Introduction. We find a similar set of evolution equations in Section 4, where we also show that (3.9) holds even in the absence of Lorentz invariance in the zero-momentum sector of the theory.
Let us start by solving the equations in two special case. We drop the expectation value symbols and the subscript to lighten the notation. If we set , the equation simply propagates the initial data along characteristics determined by :
[TABLE]
In the conformal case we can solve the equations for any eigenstate. The initial conditions are
[TABLE]
where are numbers only dependent on the state, but not on . The solution of the Burgers equation with this initial data is familiar from the literature:555The initial data are related to by (2.3), i.e. . In a CFT and .
[TABLE]
Once we know the Burgers flow, we can solve for the KdV charges that probe it. In fact we do not have to know the explicit form of the solution, (3.13), to verify that
[TABLE]
solves (3.9), if we use that satisfies (3.10).
3.3 A check from integrability and concluding comments
Integrable field theories provide a useful testing ground of our results. The deformation changes the two particle S-matrix of an integrable field theory by a simple CDD factor:
[TABLE]
where is the mass, and the rapidities. Plugging this result into the nonlinear integral equation that determines the spectrum gives the deformed spectrum in terms of the initial one. This computation was done for the energy in [5] and extended to KdV charges and other deformations in [58]. Instead of repeating their derivation, we simply copy their equations (4.47) and (4.50) in our notation in (F.1), and here we specialize to the case (corresponding to taking in (F.1)). The equation reads
[TABLE]
We recognize that the flow equation is identical to (3.9). This match is a strong check of our results. In Appendix F we discuss in detail their deformations with .
Let us comment on the regimes of validity of the different derivations of (3.9). The derivation by [5, 58] applies to the sine-Gordon model and minimal model CFTs. The derivation is expected to generalize straightforwardly to any massive integrable model. Our derivation applies to the deformation of any Lorentz-invariant theory that contains at least one higher spin conserved charge, and hence is more general than that of [5, 58]. Our result rules out the possibility that the evolution equation (3.9) is a miracle of some special (integrable) models.666Here we use the word integrable in a restrictive sense; by it we mean that the infinitely many conserved charges completely determine the dynamics of the theory. While any two-dimensional CFT has infinitely many KdV conserved charges, this does not make them integrable. CFTs with a semiclassical holographic dual are prime examples of non-integrable theories with infinitely many conserved charges. It is expected that a generic CFT is non-integrable, despite the scarcity of explicit constructions of such theories, see however [77].
A similar relationship holds between the two derivations of the Burgers equation for the -deformed spectrum: the one by [5] applies to the sine-Gordon model and minimal model CFTs, while the one by [4] applies to any Lorentz-invariant field theory.
4 Non-Lorentz-invariant deformations
We return to the analysis of (3.1): we study deformations such as (sometimes called for short) that break Lorentz invariance. Specifically, we write an evolution equation for the spectrum of zero-momentum states under the deformation. This incidentally implies that our main result (3.9) holds for zero-momentum states even without assuming Lorentz invariance. We then explain why our current methods do not allow writing an evolution equation for general states under these deformations. Finally, we solve the evolution of zero-momentum states and find the asymptotic density of states shows super-Hagedorn growth.
Without Lorentz invariance (for ) there is no preferred basis in the space of commuting conserved charges, as discussed around (3.1) and in Appendix C. Our results in this section apply to the choice of basis, specified by (2.16), for which (3.1) holds. Importantly, this choice is preferred if the seed theory is a CFT, as we show in Appendix C. This makes it nontrivial to compare our results with those of [58]. What we find in Appendix F is that the two papers describe different deformations, even after accounting for the possible change of basis. It would be interesting to parametrize the ambiguities in our results more completely.
4.1 Zero-momentum states
As observed by Cardy [53], Lorentz invariance is not needed to derive the inviscid Burgers equation for energy levels of zero-momentum states under the deformation. We generalize this to the evolution of all KdV charges of zero-momentum states under the deformation. This deformation reduces to the usual deformation both for and for , and in a CFT it reduces to for and for .
For the deformation by , (3.1) gives
[TABLE]
The relation (3.8) holds without assuming Lorentz invariance. As always, there is no general way to determine . For states with zero momentum this issue does not show up since , and one has
[TABLE]
For these states, the charge evolves according to the inviscid Burgers equation while all other charges describe probe particles riding the Burgers flow. Taking we find that our main result (3.9) on the deformation holds for zero-momentum states even without assuming Lorentz invariance.
We note that (4.2) also describes the deformation , which is a special case of the family of theories analyzed in [57], in the equation we have to make the replacement , with not evolving with due to its quantized nature.777The -independence of is consistent with (4.2): if we set , replace , and use that , we get a consistent equation.
It is also interesting to compare with the integrability result (3.16). As we explain in Appendix F the deformation described by integrability techniques is not a deformation by a local operator, hence is not in the class we consider. Nevertheless, equations (3.16) and (4.2) surprisingly agree for states that have zero momentum and (for instance states that are parity-invariant in the seed theory).
Finally, an easy calculation shows that the and deformations commute (in the zero-momentum sector), since the following result is symmetric in :
[TABLE]
In [57] we also studied whether deformations commute and we found some cases where they do not. It would be interesting to give a full description of the commutators of different deformations.
4.2 General states
The evolution equation (3.1) for KdV charges under an deformation involves expectation values of operators . Crucially, these cannot be determined from the KdV charges .
In a CFT, one checks for instance that
[TABLE]
cannot be written in terms of KdV charges. It is not a linear combination of
[TABLE]
and of other KdV currents. More stringently, its expectation value in low-level descendants of primary states is not expressible in terms of the eigenvalues of KdV charges , , (dimensional analysis restricts the set of charges to consider).
The only cases where our main evolution equation (3.1) can be solved with the tools at hand are when the dependence on completely drops out. In Section 3 this happened thanks to . In (4.1) this happened by restricting to the zero-momentum subsector. It is conceivable that for some seed theories there would be relations between with and some computable quantities. For instance in a massive free scalar one actually has up to total derivatives. However, one should check whether the relation holds after the deformation.
4.3 Evolution of zero-momentum states
The evolution equation (4.2) transports KdV charges along characteristics determined by (to avoid clutter we leave implicit the dependence on ), so we can simply adapt results (3.11) from the case and get
[TABLE]
As for the solution with CFT initial conditions is much more explicit. We use the same logic as around (3.12). First we set , and using the CFT initial conditions (3.12) we find the solution:
[TABLE]
where is the unique solution to the polynomial equation
[TABLE]
that obeys .888One can write a series solution and recast it as a hypergeometric function
\displaystyle=\frac{\left\lvert{u}\right\rvert}{x(\left\lvert{u}\right\rvert+1)}\biggl{(}{}_{\left\lvert{u}\right\rvert}F_{\left\lvert{u}\right\rvert-1}\biggl{(}\begin{smallmatrix}\frac{1}{\left\lvert{u}\right\rvert+1},\dots,\frac{\left\lvert{u}\right\rvert-1}{\left\lvert{u}\right\rvert+1},\frac{-1}{\left\lvert{u}\right\rvert+1}\\ \frac{1}{\left\lvert{u}\right\rvert},\frac{2}{\left\lvert{u}\right\rvert},\dots,\frac{\left\lvert{u}\right\rvert-1}{\left\lvert{u}\right\rvert}\end{smallmatrix}\biggm{|}-x\frac{(\left\lvert{u}\right\rvert+1)^{\left\lvert{u}\right\rvert+1}}{\left\lvert{u}\right\rvert^{\left\lvert{u}\right\rvert}}\biggr{)}-1\biggr{)}\,,
which takes real values for and has a branch point at . Another way to find this branch point is to compute the discriminant of (4.8), when seen as a polynomial of . The discriminant is , which vanishes at , indicating that two solutions collide for this value of . This is the analogue of the square-root singularity in the usual Burgers equation. The other KdV charges probe this flow, and they are given by
[TABLE]
The deformation () has to be treated separately, and the solution of (4.2) is
[TABLE]
Note that we get a divergence for , which is the analog of the branch point that we found for , see footnote 8. For the special case of this result agrees with what was found for the energy spectrum in [57] with very different methods (see also [34]).999To recover this result from the formulas (6.4) of [57], we take corresponding to the deformation, then , where we simply set , and and specialize to zero momentum. The very attentive reader will notice that we absorbed an in the definition of the deformation compared to [57] to make formulas real. In [57] the special case was not analyzed separately, this was first done in [34]. We take this agreement as a check of both the computations presented in this section and the methods of [57].
4.4 The density of states
It is particularly interesting to consider the asymptotic behavior of the spectrum. For that we need to solve (4.8) for ,101010The equation does not have a real solution for . where we get , which for gives
[TABLE]
For negative enough (see footnote 8) we formally get a complex solution, a familiar behavior from the study of .
In the CFT, high energy primary states in the zero momentum sector have
[TABLE]
where the inconvenient alternating sign ultimately follows from the sign in the decomposition . To have a real asymptotic spectrum, it follows from the condition that for and for .
Plugging (4.12) into (4.11) for we get (again for )
[TABLE]
In a CFT, we know that the density of primaries is asymptotically [78]
[TABLE]
where we used that in the CFT. Expressing with the energies of the deformed theory from (4.13), we obtain
[TABLE]
for the appropriate sign of that depends on the value of as discussed above. Note that the density of states is now independent of , in stark contrast to the extensive entropy expected in local field theories. For the above result is the Hagedorn growth of the density of states of the -deformed theory [7, 24]. We expect that the total density of states including spinning primaries and descendants would exhibit the same behavior, with only numerical factors modified.
A generalization is to deform a CFT by a linear combination . Similar calculations111111A convenient shortcut goes as follows. Charges are transported along characteristics, specifically P_{s}(\lambda,L)=P_{s}^{(0)}\bigl{(}L+2\pi^{2}\sum_{u}\lambda_{u}P_{u}\bigr{)} as in (4.6). High-energy primary states of the CFT obey (4.12) . This relation is transported along characteristics. Now use the definition valid for any combined with the transport equation to express the initial dimensionless energy in terms of the deformed energy : this gives e=\bigl{(}L+\sum_{u}2\pi^{2}(-1)^{(\left\lvert{u}\right\rvert+1)/2}\lambda_{u}(E/2)^{\left\lvert{u}\right\rvert}\bigr{)}\,E. Deleting the negligible term from this expression and plugging into the Cardy growth (4.14) for gives (4.16). lead to
[TABLE]
Different choices of appear to accomodate arbitrarily strong (e.g., doubly exponential) super-Hagedorn growth of the density of states.121212Even though (4.16) formally allows for depletion of the density of states if is fine tuned, the formula breaks down for those cases due to a Jacobian factor that we neglected, and we expect descendant states to ruin cancellations either way. However, since our results only concern zero-momentum states, they are not sufficient to determine when the deformation remains well-defined: there could be divergences in the sum over for some states.
In the case of the deformation the Cardy growth remains, but the central charge is replaced by a charge dependent expression:
[TABLE]
where we took the full density of states, hence the replacement . This behavior was understood in [34].
The super-Hagedorn growth of the density of states is a novel behavior exhibited by this system. The three systems known to us with such growth of density of states are flat space quantum gravity in dimensions, which is expected to have an asymptotic density of states from black holes; -branes, whose density of states was found to grow as (with ) in the semiclassical approximation in [79, 80, 81, 82]; and -dimensional -deformed theories, whose density of states was recently found to agree with this by a large- analysis [19]. We do not suggest that these theories to have much to do with each other. The result (4.15) however provides extra motivation to study the deformation, as these Lorentz invariant theories may give rise to exotic UV asymptotics, which would manifest itself in a density of states similar to (4.15). A natural guess based on the simple dependence of on of (4.15) and dimensional analysis for the density of states in these theories is
[TABLE]
New ideas will be needed to establish (or rule out) this guess.
Acknowledgements
We thank Zohar Komargodski, Alex Maloney, Stefano Negro, Roberto Tateo, and the participants of the “ and Other Solvable Deformations of Quantum Field Theories” workshop for discussions. MM is supported by the Simons Center for Geometry and Physics. BLF gratefully acknowledges support from the Simons Center for Geometry and Physics, Stony Brook University at which some of the research for this paper was performed.
Appendix A The ’s, their collisions and factorization
A.1 Manipulating ’s
Let us first derive two simple equations. Combining (2.4) and (2.5), we get:
[TABLE]
On the other hand, and the Jacobi identity imply that
[TABLE]
First, we can deduce a symmetry property. To make the derivation easier to parse, above the equal signs we write the relation we use. We repeatedly transpose neighboring subscripts to find
[TABLE]
This implies that and can at most differ by a constant. Taking the expectation value of both quantities in a joint eigenstate of gives zero, thus we conclude from (A.3) that
[TABLE]
This also shows more generally that is totally symmetric in the .
Second, we can establish that the condition (2.12) is obeyed for . We work here with the point-splitted version of and return later to the discussion of regulator terms. The key identity is a generalization of (A.4) involving operators at different points:
[TABLE]
in which the subscripts are totally antisymmetrized. Specializing to and we learn that
[TABLE]
is a total derivative: this is condition (2.12) for the point-splitted version of . Note that for any product other than on the left-hand side we would have gotten commutators on the right-hand side beyond just derivatives .
A.2 Collision limits
To go from the point-split equation (A.6) to an equation for itself we need to understand collision limits of operators. Let us define the point-splitted object (here stands for )
[TABLE]
We take a derivative with respect to one of the coordinates only (say, the first), keeping implicit the position dependence of each for brevity:
[TABLE]
The notation means that indices (but not ) are antisymmetrized in each term. The result is a sum of and we shall call it a -commutator. Similarly, derivatives of with respect to any of the or are -commutators. In fact, (A.8) also holds with replaced by any , but we will not use that observation.
We have just shown that all derivatives of are -commutators. Let us use the OPE
[TABLE]
written in a basis of functions that includes the constant function . (Typically one can use monomials .) Antisymmetrizing over indices , what we have shown above is that
[TABLE]
is a -commutator, where denotes the vector of all and derivatives. Since form a basis, we learn that each is a -commutator except for (constant coefficient). Hence,
[TABLE]
The -independent term defines a local operator , and the other terms are counterterms. Due to freedom in changing basis, which may add constants to the coefficients of the counterterms, the operators are only defined up to -commutators. Given our construction, the are totally antisymmetric in their indices, but might only be antisymmetric in their indices up to -commutators.131313Relatedly, when point-splitting we only showed that the OPE is regular (up to -commutators) when antisymmetrizing the : antisymmetrizing the instead may not give a well-defined operator. Alternatively we could have written the OPE in a basis of local operators that splits into the subspace spanned by for with local , and a complement of that subspace. This gives another definition of modulo -commutators.
Recall now that we are trying to take the collision limit in (A.5), namely in
[TABLE]
We get
[TABLE]
where if and otherwise so that the -th subscript of is . One could hope to use the symmetry to get terms with to cancel, but this would require the corresponding operators to be the same, which they are a priori not. Instead, we notice that the equation takes the form of a linear relation between (because the first term is -independent) and with . Since these form a basis by assumption, all of their coefficients in the linear relation vanish. In particular we have established the following symmetry for the collision limits:
[TABLE]
This then establishes that (A.5) holds in the coincident point limit with regulator terms included. As explained around (A.5), this establishes that the deformation preserves the KdV charges through the key condition (2.12).
A.3 Factorization of matrix elements
In this appendix we show the factorization property of the composite operators in diagonal matrix elements between energy eigensates. We work in a basis of states in which all charges are diagonal. We assume that the theory has a non-degenerate spectrum, namely that each joint eigenspace of all the charges is one-dimensional. This is a much weaker assumption than the assumption in [1] that the energy spectrum is non-degenerate. (For instance CFTs have a highly degenerate energy spectrum.)
Consider basis states and of equal for some . The matrix element of (A.4), namely , gives
[TABLE]
where denotes . From the nondegeneracy assumption we deduce that is diagonal in this sector. The argument applies likewise to the point splitted operator with a slight modification: is a sum of commutators so we need to restrict to a subspace of fixed for all . Altogether,
[TABLE]
Next, insert a complete set of states in a diagonal matrix element of :
[TABLE]
Then consider one of the off-diagonal terms () and let be one of the charges for which . Such a charge exists by our non-degeneracy assumption. Then we can perform a calculation very similar to (A.8) but using additionally that . We find
[TABLE]
Therefore the sum (A.17) above restricts to . An induction on shows a factorization property generalizing those for proven in [1, 4]:
[TABLE]
Note that we have omitted the positions of these operators because derivatives with respect to these positions vanish in diagonal matrix elements. We can now take the coincident point limit in : the regulator (and finite but ambiguous) terms of the form drop out in diagonal matrix elements. Hence
[TABLE]
which combined with (A.19) concludes the proof of factorization.
Zamolodchikov in [1] proved the factorization of diagonal matrix elements of only in states that have no energy and momentum degeneracy. An improvement here is that the equation holds for all states whose degeneracy can be lifted by any set of commuting local conserved charges.141414For example, consider a theory with flavor symmetry and consider an irreducible representation of inside the Hilbert space. Our reasoning shows the factorization property for eigenstates of , but also by symmetry for eigenstates of any other element of . How can the non-linear property of factorization hold for all these linearly-related states in at the same time? The key is that and and commute with hence are multiples of the identity when acting on .
Appendix B Proof of (2.6) in Lorentz-invariant theories
By construction, and . We show here that in Lorentz-invariant theories the operators are given by (2.6), namely
[TABLE]
provided one suitably improves the symmetry current by adding a total derivative \bigl{(}[P_{1},U^{s}],-[P_{-1},U^{s}]\bigr{)} with given later in (B.14). A surprising side-effect of (B.1) is that it fixes a preferred choice of improvements for all higher-spin currents (), because are not affected by improvements of . Improvements are discussed in detail in Appendix C.
Let us first derive a consequence of (B.1). These relations can be stated as . Combined with (A.4) we get
[TABLE]
so derivatives of the difference vanish. This difference is thus a multiple of the identity, hence must be zero unless its spin is the same as that of the identity operator. In that case () the multiple of the identity can be absorbed into the definition of , for instance by normalizing their ground state expectation value to zero. Altogether, we conclude
[TABLE]
It remains to prove (B.1). For , (B.1) is immediate. Our strategy for other spins is to show that and have vanishing derivative, as we state in (B.15) and (B.22). Then these local operators must be multiples of the identity, hence vanish because their spin is non-zero (). In the special case one determines through their derivatives , where we used that a flavor symmetry charge commutes with all stress tensor components . Henceforth we focus on spins .
Throughout our proof of (B.1) we write equal-time commutators of local operators as151515We sometimes denote without specifying the point when that point is clear from context.
[TABLE]
Notice for instance that and differ by derivatives since
[TABLE]
B.1 Computing some derivatives
First we work out
[TABLE]
and likewise
[TABLE]
All terms except the ones are manifestly -derivatives. Let us check the terms also are:
[TABLE]
Thus, when restricted to , the sums in (B.6) and (B.7) give and . From these we want to subtract derivatives of and respectively.
To make factors of spin appear, we consider commutators with the spin operator acting by rotations around the point . By definition,161616We left the point of origin implicit in our notation for . is the charge corresponding to the rotation current that is conserved by virtue of the symmetry of the stress tensor. Since the coordinates are not well-defined on the cylinder, the expression we gave for only makes sense locally, but our calculations are local so doing them on the plane would be equivalent.
[TABLE]
so we can express for any local operator in terms of the commutator :
[TABLE]
From the fact that and have spins we learn that
[TABLE]
We obtain that derivatives of and are quite complicated:
[TABLE]
B.2 Improvement
It is not immediately obvious how to absorb right-hand sides into an improvement of . Because , the sum of these equations simplifies and gives second derivatives and higher:
[TABLE]
This is precisely as expected because the time component of a current is shifted by a space derivative upon improvements. For the right-hand side of (B.13) is absorbed by using the following improved current (in the main text we drop the hats)171717For the left-hand side of (B.13) vanishes by construction, so the right-hand side must vanish. This is difficult to prove by direct calculations.
[TABLE]
Explicitly,
[TABLE]
B.3 Space component
Next we prove the analogous equation for the space components, namely with the signs of and flipped. We first compute the improvement term in , namely . It involves the operator , which, by definition of , is the -th term in the following commutator
[TABLE]
Applying the Jacobi identity and the conservation equations for and gives
[TABLE]
The space derivatives and can be pulled out of the commutators, which can then both be expanded as with appropriate arguments. Moving the derivatives back into the sum gives
[TABLE]
Equating coefficients of in (B.16) and (B.18) teaches us that for
[TABLE]
We conclude that
[TABLE]
Returning to (B.12) and using we work out
[TABLE]
namely
[TABLE]
Since if the space derivative of an operator with spin vanishes, it must be the zero operator, (B.15) and (B.22) conclude the proof of (2.6).
Appendix C Ambiguities
In this Appendix we collect results about ambiguities that we encountered in our derivation. First we present four ambiguities, the most problematic being the ambiguity in choosing the basis of conserved charges. For a Lorentz-invariant seed theory we use Lorentz invariance and a spurion analysis to partly resolve this basis ambiguity. For a CFT seed, dimensional analysis mostly eliminates the remaining basis ambiguity. In cases where we are eventually unable to resolve some of the ambiguity, our equations are only valid for the specific choice of basis that we prescribe.
C.1 Four ambiguities
Conserved currents are only defined up to improvement transformations. Under an improvement , we get using (2.4) that . Let us now take antisymmetric combinations of the ’s that define the operator modulo -commutators (see Appendix A.2). Under an improvement the point-splitted operator is shifted as
[TABLE]
where each term in the sum can be rewritten as using (A.5). The change in the collision due to improvements can thus be absorbed into the regulator terms (-commutators), as claimed below (2.7).
Note that the ambiguity in the choice of these regulator terms drops out from diagonal matrix elements in joint eigenstates of KdV charges, since . In fact, under an improvement none of the expectation values on either side of the factorization property (A.19)–(A.20) are affected:
[TABLE]
There is a trivial ambiguity in the definition of , the shift by multiples of the identity: . Because (2.5) fixes , the ambiguity does not affect . However, it changes by mixing it with combinations of of fewer indices. The only case relevant to us is : the variation (2.16) of under the deformation is constructed from it and we get
[TABLE]
This mixing of charges is a special case of the ambiguities discussed next.
Finally, we focus on an ambiguity that is not easily resolved. The algebra of local conserved charges is in general non-abelian (for instance in case of non-abelian flavor symmetry); for our purposes we need to choose a maximal commuting subalgebra that includes the Hamiltonian and momentum. Within this subalgebra, we still have to choose a basis. While any function of the charges is conserved, only their linear combinations plus shifts by the identity must derive from a local conserved current. Let us implement the change , with (so ) to respect momentum quantization and the fact that our deformations are always specified by how they act on the energy, with no ambiguity. It shifts local operators as follows:
[TABLE]
where the shift of is a particular choice that preserves (2.5). There are other satisfactory choices, as discussed around (C.3).
This basis ambiguity enters as follows in the story presented in the main text. The definition of in (2.12) is ambiguous by the addition of conserved currents, and this leads to a freedom of adding a linear combination of conserved currents to (2.16). We consider below various conditions on the seed theory or on the deformation and determine how much they reduce the ambiguity. This may be useful when comparing our results to other approaches, as such approaches may only respect some of the conditions that we use to uniquely characterize our choice of deformation.
C.2 Lorentz invariance, spurions and dimensional analysis
Consider first the Lorentz-preserving deformation of a relativistic theory. One may not add multiples of the identity to any charge: indeed, the identity could only be added to current components of spin [math], namely and , but these are fixed by . In addition, one may only linearly combine currents of the same spin, namely shift by for some coefficients (more generally a combination of all charges of the same spin). If the seed theory is a CFT, dimensional analysis eliminates the ambiguity because it only allows a singular . For a massive theory, can depend nontrivially on the dimensionless combination of the mass scale of the seed theory and the irrelevant coupling . In the absence of a nonabelian charge algebra, no physical principle forbids such rescaling, but there is a minimal choice (2.16) that we employ in this paper.181818For deformations other than this statement must be qualified: (2.16) does not fully define a choice of charges. The ambiguity resurfaces. Lorentz-invariance only allows , and because of (B.3) it requires . Plugging into (C.3) for the deformation we find and . This means that (2.16) does not fully define a choice of charges and : specifically one could rescale both of them (by the same factor because ). This caveat does not affect our results: for the deformation, because of (2.5), so (2.16) fully defines all . In the case it is also the natural definition of charges that emerges in the integrability context in [5, 58]. If we have a nonabelian algebra, as it is the case for a CFT seed theory, we cannot rescale the different generators arbitrarily as that would violate the commutation relations. The choice made in (2.16) is compatible with the preservation of the algebra as shown in Appendix G. In summary, equation (3.9) giving the evolution of KdV charges under the flow is unambiguous for a CFT seed, and otherwise its only ambiguity is to scale each KdV charges. This ambiguity is frozen by our choice (2.16).
It is still worth contemplating how easy would it be to recognize the evolution considered in this paper, if we were handed the spectrum of the theory with a different choice of rescaling. Since the rescaling acts the same way on each eigenvalue, the ratio of two eigenvalues is unambiguous, and it would readily lead to the identification of the deformation and the rescaling used.
Next, consider a relativistic seed theory, but deform it by an arbitrary . The key to using Lorentz-invariance of the original theory is to promote the coupling to a background field (also called a spurion) that has spin , so that the action is deformed by the Lorentz-invariant combination . To illustrate how the spurion helps, note that our minimal prescription for is an integral of operators of spin , consistent with the spins of the current components and . Using the same idea, the only mixing ambiguities in the deformation of a relativistic seed are
[TABLE]
for some coefficients (more generally one should allow in each term any charge of the same spin as ). Without further input these ambiguities cannot be eliminated. If the seed is a CFT then we use dimensional analysis: has dimension while has dimension . Only terms with are dimensionally consistent. This condition means have the same sign or are zero.
In particular, the deformations of a CFT with have no ambiguity.
For deformations of a CFT (say, ), vanishes because it is an antisymmetric combination of holomorphic currents. The deformation thus ought to be trivial, but our general prescription (2.16) turns out to mandate a change of basis among holomorphic currents. Indeed, it sets to an integral of operators . For this vanishes because and vanish, as is built from a different Virasoro algebra than and . For however, the operator may be non-zero: it is simply a holomorphic conserved current. We see that our general prescription is in this case not a “minimal” choice of how charges are deformed, as one could have taken simply . (This minimal choice cannot be generalized to non-CFTs.) The spurion and dimensional analysis above simply teaches us that for , is not ambiguous, while for the variation has the full ambiguity (C.5). That ambiguity is enough to relate the choice made in (2.16) to the minimal choice.
C.3 Ambiguities for Section 4
Our spurion analysis (for relativistic seeds) and dimensional analysis (for CFT seeds) extends to linear combinations of deformations by assigning separate spins and dimensions to all of the coupling constants. In particular let us discuss the deformation of Section 4, taking for definiteness (the case is ). For the case of a CFT seed we will eliminate the whole ambiguity.
Assume first that we start from a Lorentz-invariant theory. The couplings of have different spins . A charge can thus be mixed with for . We can now reduce to a single coupling and write the ambiguity as
[TABLE]
This ambiguity cannot be eliminated without further assumptions.
For a CFT seed we can eliminate these ambiguities completely. Among ambiguities (C.6) allowed by the spurion analysis, dimensional analysis (where has dimension ) only allows those with . Using the triangle inequality one has
[TABLE]
with equality if and only if and . Thus, (C.6) becomes
[TABLE]
Focus on states that start out as primary states in the CFT. Our evolution equation (4.2) preserves . In contrast, any shift (C.8) spoils this because the charges all have positive spins and their expectation values all have different scalings in terms of the state’s energy. The condition of preserving thus characterizes our deformation when the seed is a CFT.
Appendix D Existence of local currents generating the KdV charges
In this Appendix we show that if satisfies (2.12) then remains conserved and the integral of a local current. The conservation equation (2.2) in the canonical formalism takes the form
[TABLE]
which we linearize in the coupling of to obtain
[TABLE]
Quantization of the momentum implies , and using together with (2.3) implies that , reducing (D.2) to
[TABLE]
The commutator of two local operators can in general be written as
[TABLE]
Integrating this commutator over gives . In (D.3) we need the integral of this expression in :
[TABLE]
where we used (2.3). Plugging this result back into (D.3) we see that we can satisfy that equation only if the condition (2.12) is obeyed. Putting (D.5), (2.12), and (D.3) together we get that191919The corrections to the currents coming from in (D.6) were given in an explicit form in [4], while the terms coming from the were referred to as contact term corrections in a footnote.
[TABLE]
Appendix E Rescaling space
We show here how KdV charges respond to a rescaling of space. Specifically we show
[TABLE]
for some nonlocal operator , which however does not influence diagonal matrix elements in eigenstates, since . One way to reach this equation is to start from and apply the general machinery (2.12) with to determine how KdV charges can be adjusted to remain conserved. A minimal choice is (E.1). However, this approach leaves a lot of ambiguity because the KdV charges could be mixed under this deformation. We take a different, more direct, approach here to show (E.1) that avoids this mixing ambiguity. As in (B.4), we will use the notation
[TABLE]
Let us start with the left-hand side of (E.1). The action of a local spatial translation on a local operator is to shift it as
[TABLE]
Integrating with measure gives
[TABLE]
To rescale space we take . We compute the commutator using (E.2):
[TABLE]
Altogether,
[TABLE]
for . In particular, taking , whose integral is , we get
[TABLE]
Next we work out the right-hand side of (E.1). We compute
[TABLE]
The term is a derivative, like the other terms:
[TABLE]
so we get
[TABLE]
up to shifts by multiples of the identity (the only local operator whose derivative vanishes). Then
[TABLE]
We are done showing (E.1), because the right-hand sides of (E.7) and (E.11) agree up to .
Appendix F A comment on an integrability result
We show here that the evolution equation found in [58] using integrability describes some deformation that is outside the class of operator deformations that we study. Our results cannot be compared. Let us copy their equation for the -th deformation here in our notations:
[TABLE]
where we used the translation and kept their notation for . Because it reproduces results on the and deformations202020More precisely, for a CFT the limit has a four-parameter generalization, and a choice of these parameters gives the usual deformation. the authors naturally suggested that for general spin it might describe the (plus ) deformations.
We give a general argument based on translation invariance that shows that (F.1) cannot correspond to adding to the action the integral of any local operator and working with charges of local conserved currents. We then give a more restricted argument that the equation cannot describe and/or deformations, based on the observation that (F.1) does not involve the operators. This might help determine what the deformation described by (F.1) actually is in the operator language.
F.1 Nonlocality of the deformation or the charges
Assume that (F.1) described adding to the action the integral of a local operator and working with charges of local conserved currents. Then invariance under translation along the (compact) spatial direction would be preserved, so momentum would remain quantized, hence independent:
[TABLE]
by which we mean the momentum of the original CFT state . In the CFT, .
While in our framework we kept equal to momentum (the quantized charge of spatial translation), (F.1) leads to
[TABLE]
where we simplified a derivative by using that momentum depends on as . In an updated version of [58] another momentum is also defined, and it is found not to depend on and hence coincides with the momentum we are using in the main text. The relation between and in [58] is the same (to linear order) as what we find in (F.3); what we are showing below is that and defined in [58] cannot both be integrals of local currents.
We would thus have two conserved charges: , and momentum . Their difference would be a conserved charge as well, namely there would exist a conserved current such that (we divided by for later convenience)
[TABLE]
Notice in passing that for the right-hand side cancels out and one can simply have . For there is no cancellation and the right-hand side is the eigenvalue of in the state . Each is an integral of a local operator over the spatial circle, so this quadratic combination is an integrated two-point function of components of currents. There is no reason to expect such an integrated two-point function to reduce to the one-point function of a well-chosen operator.
Let us make the argument sharp when starting from a CFT, for instance a minimal model: after all, the integrability results apply equally well to these theories. In a CFT with no further symmetry the KdV charges have odd spins .
Consider first and focus on a primary state with conformal dimensions . In that state, and are polynomials of degree in and , respectively, so
[TABLE]
In a generic CFT, conserved charges split into a sum of a holomorphic and an antiholomorphic charges, and their one-point function in a primary state is of the form . For , (F.5) is not of this form, so the current cannot exist. This concludes our proof in that case.
For , the matrix element is a sum of terms and that each involve only one of the chiral Virasoro algebras. However there is no way to write these terms as the expectation value of a local conserved current. Let us see this explicitly for . Note that since is an eigenstate of ,
[TABLE]
which cannot be equal for all states to a linear combination of
[TABLE]
because the sum of cannot cancel in all states.
We conclude that (F.1) cannot describe in general for the evolution of local charges under a deformation that respects periodic translation invariance and locality. If (F.1) describes the effect of field-dependent changes of coordinates as proposed in [58], then it is perhaps not surprising that periodicity of the space coordinate is not preserved. It may be the case that the deformation only makes sense on the plane rather than the cylinder. Another possibility may be that the charges appearing in (F.1) are not integrals of local conserved currents.
F.2 Linear order around a CFT
While our proof above rules out deformations by arbitrary local operators it is instructive to look more carefully at why the integrability equation (F.1) does not correspond to a deformation by operators.
Let us consider deformations of CFT to linear order by a combination of and (for ), namely by for some coefficients . As we explained, our formalism expresses the variation of KdV charges in terms of operators . In a CFT, these operators vanish when signs of and differ, and furthermore they have the symmetry derived in (B.3). This allows us to write (3.1) as
[TABLE]
where the superscript denotes CFT quantities. At this point we must remember that (3.1) is only one choice of how to deform KdV charges in such a way as to keep them conserved: one can add to it other conserved charges of the CFT, as discussed in detail in Appendix C.212121In fact, dimensional analysis and spurion analysis together rule out such mixing for the deformation (). Since () vanish in a CFT, it is not possible to distinguish (at linear order around the CFT) the deformation from a sum of this deformation and of any (). While the couplings of are invisible in the Hamiltonian at this order, they weaken dimensional and spurion analysis because of their varied dimensions and spins. These couplings allow a large class of mixing ambiguities. We thus move on with the proof without using dimensional and spurion analysis.
In contrast, using that the -th KdV charge scales as in the CFT, the integrability result (F.1) gives
[TABLE]
In both of these lines we recognize one of the terms in (F.8) (with ) but not the term for (and its complex conjugate for ). As discussed in Section 4.2, cannot be determined from the integrals of motion . What is less immediate is whether the term could be fully absorbed by the freedom to shift by a conserved charge,222222In fact, this essentially happens in Section 4. To linear order around a CFT the deformation studied there is , corresponding to and here, and we focus there on the zero-momentum sector. In that sector we can check . The first term is a shift by the conserved charge of the holomorphic current . The second is expressed in terms of charges that we have control on. Away from the zero-momentum sector this switch to holomorphic quantities is not possible. possibly combined with a change of .
This can be ruled out tediously in an ad-hoc manner by considering the case where is a primary state of conformal dimensions , and working out the leading powers of and in each expectation value. The question then boils down to whether there could be some coefficient such that
[TABLE]
is for some conserved charge (here denote known coefficients). Since the leading monomials cannot cancel for , by the same logic as around (F.5), (F.10) does not have the form of the expectation value of a conserved charge.
Thus, (F.1) would need significant modifications involving to describe the or deformations.
Appendix G Nonabelian symmetries
In the main text we exclusively work with a chosen commuting subset of the conserved charges. Here we discuss what changes for charges that do not commute. Most prominenly this includes non-abelian flavor symmetries. Another example is the full set of monomials built from and its derivative in a CFT: this forms a non-abelian extension of the KdV charges.
We learn that it only makes sense to deform by bilinears combinations of currents when the corresponding charges and commute. Along the deformation, one can preserve the charges that commute with both of these, and the structure constants of these charges are not deformed. For instance, the deformation preserves the full charge algebra (non-abelian flavor symmetries and perhaps more surprisingly the non-abelian KdV charge algebra of a CFT) including its structure constants.
G.1 The operators
We denote structure constants as , so that .
Because , the integral of on any cycle vanishes, hence this one-form is exact. Namely,
[TABLE]
where are some (local) operators defined up to shifts by multiples of the identity. We also denote to emphasize that the operator depends on a choice of charge and a choice of current, which are two somewhat asymmetric inputs. The operators considered in the main text are special cases of . With this notation it is easy to check that
[TABLE]
up to the shift-by-identity freedom. Improving the currents affects as follows:
[TABLE]
In another appendix we showed a symmetry property (A.4) for the case of commuting charges. To show it the main point was to show the and derivatives vanished. Let us follow the same strategy when structure constants are non-zero. We work out
[TABLE]
where the first equality is the definition (G.1), the second equality uses the Jacobi identity and , and the last equality expresses each commutator before using a cancellation that is due to the Jacobi identity . Together with the analogous result for , this means that
[TABLE]
is a translationally-invariant but local operator, hence a multiple of the identity. This reduces to the definition of upon specializing to and using (G.2): this uses that structure constants vanish when , because any conserved charge commutes by definition with these charges.232323To be more precise this assumes that currents do not depend explicitly on coordinates; otherwise the conservation equation and the trivial equation do not translate to . Another interesting case is when , and commute. Then all structure constants drop out, so the operator is traceless242424In this infinite-dimensional setting the trace is ill-defined. One can consider instead the expectation value in any common eigenstate of and . hence vanishes. In other words,
[TABLE]
G.2 The operators and deformations
Consider a pair of conserved currents and . For the same reason as the usual collision, we can define by point-splitting, modulo total derivatives. Indeed, conservation leads to
[TABLE]
hence the collision is independent of the offset , modulo total derivatives. Amusingly we did not need to assume that the charges and commute.
Now deform the action by . The key question is which symmetries can be preserved. As we showed in (2.12), the condition is that needs to be a total derivative. One can compute
[TABLE]
A word of warning: the bilinears regulated by point splitting have significantly more ambiguities than those we discuss in Appendix A.2 for the case of commuting charges.
In order for the deformation to make sense beyond linear order, the symmetries and that define the deformation must themselves be preserved by the deformation. Setting and we see that the above commutator is only a total derivative if is both proportional to and to , hence is simply zero.
We learn that it only makes sense to deform by bilinears of commuting currents.
Then, apart from fine-tuned cases where somehow cancels, the charges that are preserved by the deformation are the charges that commute with and . An important special case is for the deformation: charges can be preserved if and only if they commute with , namely the corresponding currents do not depend explicitly on coordinates.
G.3 Structure constants are preserved
Under a deformation by (with ), consider two charges and that commute with and . In other words, these four charges commute pairwise except and , whose commutator we wish to study. Ignoring regulator terms (which work out in the same way as explained in Appendix A.2) we have
[TABLE]
where we simply expanded the commutators. Rewriting the commutators , and using (G.5) to rewrite (other structure constants vanish), we get
[TABLE]
The first two and last two terms combine into derivatives, while the middle two terms are simply . Altogether, \delta\bigl{(}[Q_{c},Q_{d}]-f_{cd}{}^{e}Q_{e}\bigr{)}=0, namely structure constants do not change. This is in harmony with the conjecture in [4] that the deformation leaves the KdV charges commuting, which we showed in (2.17) in a less abstract language.
In Appendix C we analyze ambiguities that affect the definition of currents, charges and appearing throughout the paper. In this appendix we worked with the specific fixing of ambiguities and saw that the symmetry algebra remains undeformed. If we were to reintroduce ambiguities, the nonabelian structure would get deformed. Hence, if a nonabelian algebra is preserved, requiring it to remain undeformed is an efficient principle to fix the ambiguities.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] A. B. Zamolodchikov, Expectation value of composite field T anti-T in two-dimensional quantum field theory (2004), hep-th/0401146 .
- 2[2] J. Cardy, Quantum Quenches to a Critical Point in One Dimension: some further results , J. Stat. Mech. 1602 (2), 023103 (2016), 10.1088/1742-5468/2016/02/023103 , 1507.07266 . · doi ↗
- 3[3] G. Turiaci and H. Verlinde, Towards a 2d QFT Analog of the SYK Model , JHEP 10 , 167 (2017), 10.1007/JHEP 10(2017)167 , 1701.00528 . · doi ↗
- 4[4] F. A. Smirnov and A. B. Zamolodchikov, On space of integrable quantum field theories , Nucl. Phys. B 915 , 363 (2017), 10.1016/j.nuclphysb.2016.12.014 , 1608.05499 . · doi ↗
- 5[5] A. Cavaglià, S. Negro, I. M. Szécsényi and R. Tateo, T T ¯ 𝑇 ¯ 𝑇 T\bar{T} -deformed 2D Quantum Field Theories , JHEP 10 , 112 (2016), 10.1007/JHEP 10(2016)112 , 1608.05534 . · doi ↗
- 6[6] M. Lashkevich and Y. Pugai, The complex sinh-Gordon model: form factors of descendant operators and current-current perturbations , JHEP 01 , 071 (2019), 10.1007/JHEP 01(2019)071 , 1811.02631 . · doi ↗
- 7[7] S. Dubovsky, R. Flauger and V. Gorbenko, Solving the Simplest Theory of Quantum Gravity , JHEP 09 , 133 (2012), 10.1007/JHEP 09(2012)133 , 1205.6805 . · doi ↗
- 8[8] S. Dubovsky, V. Gorbenko and M. Mirbabayi, Natural Tuning: Towards A Proof of Concept , JHEP 09 , 045 (2013), 10.1007/JHEP 09(2013)045 , 1305.6939 . · doi ↗