Partially-massless higher-spin algebras and their finite-dimensional truncations

The global symmetry algebras of partially-massless (PM) higher-spin (HS) fields in (A)dSd+1 are studied. The algebras involving PM generators up to depth 2 (ℓ − 1) are defined as the maximal symmetries of free conformal scalar field with 2 ℓ order wave equation in d dimensions. We review the construction of these algebras by quotienting certain ideals in the universal enveloping algebra of (A)dSd+1 isometries. We discuss another description in terms of Howe duality and derive the formula for computing trace in these algebras. This enables us to explicitly calculate the bilinear form for this one-parameter family of algebras. In particular, the bilinear form shows the appearance of additional ideal for any non-negative integer values of ℓ − d/2 , which coincides with the annihilator of the one-row ℓ-box Young diagram representation of sod+2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathfrak{s}{\mathfrak{o}}_{d+2} $$\end{document}. Hence, the corresponding finite-dimensional coset algebra spanned by massless and PM generators is equivalent to the symmetries of this representation.


Introduction
Higher spin (HS) symmetries play a central role in the consistency of HS theories [1][2][3][4][5], and serve as important guideline in the understanding of different AdS/CFT models.In the recent work [6] the authors of the current article have studied several aspectssuch as minimal coadjoint orbit, minimal representation and Joseph ideal -related to symmetries of massless HS fields of symmetric index type, and found a convenient formula for computing trace in HS algebra.In this article, we continue our previous work and clarify analogous aspects of a more general type of HS symmetries -global symmetries of a theory involving partially-massless (PM) HS fields [7].
PM fields are special spectra which exist (or stay irreducible) only in the backgrounds with non-vanishing constant curvature.More precisely, they are unitary irreducible in dS background, whereas in AdS they are irreducible but not unitary.For a given spin s, there are s different PM fields labelled by their depth t = 0, 1, . . ., s − 1 , where t = 0 corresponds to the massless field.They are described by gauge potentials that transform with gauge parameters of lower rank compared to massless fields of the same spin: where ϕ (t) is a PM field of depth t, ∇ denotes (A)dS covariant derivative and Λ is the cosmological constant.The propagating degrees of freedom of the spin s PM field with depth t > 0 are more than those of the massless spin s but less than those of the massive one.In the flat space limit, PM field of depth t is decomposed into a collection of massless fields of spin s, s − 1, . . ., s − t [8].The Killing tensor of a PM field of spin s and depth t are given by O(d + 2) Young diagram with two rows of respective length s − 1 and s − 1 − t [10].Therefore, algebras involving such generators can be interpreted as global symmetry of a theory involving PM fields.
In order to construct a physically consistent algebra based on these generators, we need to require following two conditions: • The isometry algebra so d+2 1 is a subalgebra.This implies that the corresponding HS theory contains gravity sector.
• All the generators of different spins transform covariantly under isometries.This is to say that in the corresponding theory, HS fields couple to gravity in a diffeomorphism invariant manner, which is tantamount to the requirement of the equivalence principle to hold.
A priori, finding out an algebra satisfying the above conditions is highly non-trivial task in a bottom-up approach, as we have to solve for the Lie brackets among different PM generators (labelled by infinitely many spins and depths) satisfying Jacobi identity 2 .Fortunately, a series of algebras involving PM HS generators and satisfying the above conditions have been already found in the literature.The first discussions are from mathematics literature, the simplest case [11] and the generalizations [12,13], where certain PM HS algebras are defined as the maximal symmetries of higher order Laplace operators generalizing the definition of the massless HS algebra (ℓ = 1) .In physics, the unfolded formulation of the corresponding PM HS spectra has been proposed in [5] and explored in [14,15].In [16][17][18], the Flato-Fronsdal theorem was generalized to higher-order singletons and arguments were provided for a conjecture of AdS/CFT duality, relating a Vasiliev-type theory with PM symmetric tensor fields to the O(N ) model at a multicritical isotropic Lifshitz point.
In this paper, we revisit the PM algebras including the aforementioned ones from various points of view.We first define the PM algebras as particular cosets of universal enveloping algebra (UEA) of the isometry algebra.This requires to identify proper ideals of the UEA and such ideals depend on a parameter λ .For generic value of λ , the algebra is spanned by PM generators of any even depths.When λ takes an integer value ℓ , it truncates into the algebra associated with (1.2) allowing only the generators of depth smaller than 2 ℓ .We also show how the same algebras can be constructed making use of oscillators and Howe duality.The oscillator description allows us to perform more concrete calculations and we derive the explicit form of trace and bilinear form for these PM algebras.The explicit expression of bilinear form reveals a new structure of PM algebras.For (half-)integer values of λ greater than d/2 , with (odd)even d, the algebras truncate into finite dimensional ones.
The paper is organized as follows.In Section 2, we provide generalities on the global symmetries of free PM HS fields and the vector space of PM HS algebra.In Section 3, we define PM HS algebras as cosets of the UEA of (A)dS isometry algebra.In Section 4, we present Howe duality and oscillator realization of PM HS algebras.In Section 5, the trace for the PM HS algebras are defined and the bilinear forms are calculated.In Section 6, we show the finite-dimensional truncation of algebras for special values of λ.In Section 7, the relations of the PM HS algebras to conformal higher spin theory are discussed.Appendix A contains computational details.

Partially-Massless Fields and Their Global Symmetries
The global-symmetry generators, or Killing tensors, corresponding to PM fields can be identified by analyzing the relevant Killing equations.They were first identified in the frame-like formulation in [10].Below, we shall rederive the PM Killing tensors in the metric-like ambient formulation.As it has been shown in [15,19], the Fierz system of the PM field of spin s and depth t has the following gauge equivalence relation, where PM gauge parameters E (t) are traceless tensors: As usual, being ambient-space fields, Φ (t) and E (t) should satisfy the tangentiality and homogeneity conditions: in order to be reduced to (A)dS Fierz system.The global symmetries are given by parameters, that satisfy the Killing equation, δ E Φ (t) = 0 , which combined with the tangentiality and homogeneity conditions defines the following system, 3 It is obvious to see from here that the PM Killing tensors of depth t correspond to the (t + 1)-dimensional representation of sp 2 : We shall also see the relevance of this sp 2 representation later in the other construction.The conditions (2.6) and (2.2) are solved by with the tensor M (r,t) taking values in the {r, r − t} Young diagram of O(d + 2) : (2.9) Therefore, any theory involving a PM field of spin s and depth t should admit global symmetry containing generator M (s−1,t) .

Coset Construction from Universal Enveloping Algebra
Similarly to the massless HS algebra, the algebras involving PM HS generators can be approached from the universal enveloping algebra (UEA) of so d+2 .One of the advantages of UEA construction is that the two physical consistency conditions mentioned in Introduction are automatically satisfied.Let us remind that the massless HS algebra can be obtained as the coset, 3 One can equally describe PM fields in the Stueckelberg formulation where the gauge transformation has the standard form, δE Φ whereas the tangentiality condition is modified to Therefore, in the Stueckelberg formulation, we get conditions equivalent to (2.6) but X and U exchanged.
where U denotes the UEA of so d+2 , while (-) denotes the Joseph ideal, generated by the following elements in so d+2 ⊙ so d+2 : This procedure determines the eigenvalues of all Casimir operators of so d+2 .In particular, the quadratic Casimir is given by Let us now consider deformations of the above construction where we take a quotient with the ideal generated by only one of J ab and J abcd .In such cases, the quadratic Casimir C 2 remains arbitrary while the other Casimirs are fixed as functions of C 2 as shown in [20,21].Hence, one can take a further quotient with C 2 −ν .In this way, we have two ideals labelled by a continuous parameter ν as Since these ideals contain fewer elements than the Joseph ideal, the corresponding coset algebras contain more generators than the original one.These one-parameter families of algebras have already been investigated in [13] and [21], respectively.In the case of I(ν) , we get the global symmetry algebra of a theory involving PM HS fields, whereas for J (ν) , we get the symmetry of a theory involving mixed symmetry HS fields.Leaving aside the discussion about the mixed symmetry case, in the current article we shall consider only the ideal I(ν) and corresponding PM HS algebras.
Let us resume the discussion with the coset algebra, where we have parametrized ν in terms of λ for later convenience.The ideal generated by J abcd contains all GL(d + 2)-tensors in U having more than two rows, hence the algebra A λ is spanned by the generators which have the symmetry of two-row GL(d + 2) Young diagrams.When decomposed into the traceless O(d + 2) tensors, it is given by where K p are the vector space of the Killing tensors associated with the PM field of any spin and depth t = 2p : Hence, the generators of A λ are not only the massless Killing tensors K 0 but also the PM ones K p≥1 corresponding to the depth 2p [14] (see also [22] for related discussions).
As has been shown in [13] , when λ takes integer values ℓ = 1, 2, . . ., the algebra A ℓ develops the ideals q ℓ which consist of PM Killing tensors of even depth not smaller than 2 ℓ : As a result, the algebra is decomposed into where p ℓ is the coset algebra: containing generators of even depth smaller than 2 ℓ .Therefore, we obtain two algebras q ℓ and p ℓ for λ = ℓ : the former algebra contains generators of even depths not smaller than 2 ℓ4 , whereas the latter one is associated with the theory involving PM fields of even depths smaller than 2 ℓ .The usual massless HS algebra corresponds to p 1 .
Let us focus now on the coset algebra p ℓ which contains the usual HS symmetry in the ℓ = 1 case.In fact, it can be obtained directly from U as the coset, where the ideal I ℓ is given by [12] (3.12) The vector space (3.10) of the p ℓ can be straightforwardly identified using this ideal, or equivalently from the corresponding coadjoint orbit defined by Any polynomials of W ab subject to the above conditions are dual to the generators of p ℓ so give rise to the vector space (3.10) .The values of the quadratic Casimir (3.5) can be also calculated from the ideal I ℓ by inquiring the consistency of the ideal.The algebra p ℓ corresponds in fact to the symmetry algebra of the ℓ-th power of Laplacian operator, or in other words the symmetry of the action, This symmetry has been studied for ℓ = 1, 2 in [11,23] and for arbitrary positive integers ℓ in [12].The aspect of holographic correspondence has been also explored recently in [17].
For generic values of λ , the algebra A λ is generated by the Killing tensors of all even depth PM HS fields.Hence, if there exists a theory based on this field content, then A λ may well correspond to its global symmetry algebra.However, the viability of such a theory is not obvious: since the value of the quadratic Casimir used for the quotienting does not correspond any more to that of a free scalar field on the boundary, it is not clear what might be the CFT dual.The ideal q ℓ appearing for the integer value of λ = ℓ corresponds to an (higher-order) on-shell condition for the boundary scalar, but A λ does not admit such an ideal, hence the boundary scalar seems to be an off-shell one with a fixed conformal weight.It will be interesting to examine the relevant Flato-Fronsdal theorem to see whether this CFT might be dual to the putative PM HS theory, but it is beyond the scope of the current paper. 5

Howe Duality and Oscillator Realization
The coset construction from UEA provides a concise definition of the PM symmetries, but for explicit calculations we may consider another description of PM algebra.Similarly to massless case, PM algebra A λ can be defined using a reductive dual pair correspondence (introduced in HS context in [4]), aka Howe duality, as The starting point is again the metaplectic representation of sp 2(d+2) which can be realized by the star product algebra with the product,6 The mutual stabilizers sp 2 and so d+2 correspond to and the off-shell HS algebra hs(so d+2 ) is defined as the centralizer of sp 2 : any element By solving this condition, one can show that hs(so d+2 ) is spanned by polynomials of M ab .The usual (on-shell) massless HS algebra is the coset of hs(so d+2 ) by the sp 2 -triviality relation, K αβ ∼ 0 .This relation actually defines the trivial representation of sp 2 , which is dual to the singleton representation of so d+2 .In the following, we shall consider generalization of the last step and define the PM algebras A λ and p ℓ .
PM algebra A λ In order to obtain A λ from hs(so d+2 ) , we can quotient the latter with the sp 2 Casimir relation, which also fixes the quadratic Casimir of so d+2 in terms of λ to (3.5) since in hs(so d+2 ) the two Casimirs are related by In the case of λ = ℓ , the relation (4.5) gives an indecomposable representation consisting of the (ℓ + 1)-dimensional representation and the infinite-dimensional representation with lowest (or heighest) weight ℓ + 1 (or −(ℓ + 1)) .Hence, the algebra A ℓ becomes the semidirect-sum of the coset algebra p ℓ and the ideal algebra q ℓ as in (3.9).
PM algebra p ℓ In order to obtain directly p ℓ from hs(so d+2 ) , we can quotient the latter with the following equivalence relations which define the (ℓ + 1)-dimensional representation of sp 2 .This relations have asymmetric form with respect to sp 2 indices, but one can show in fact that any other relations defining the (ℓ + 1)-dimensional representation such as, give the same algebra p ℓ .Instead of (4.8), one may also use the equivalence relation7 which generalizes the relation K αβ ∼ 0 of the massless HS algebra to PM cases in a manifestly sp 2 covariant manner.However, this relation is weaker than (4.8) since the relations ∼ 0 with n = 0, 1, . . ., ℓ − 1 also imply (4.9).Therefore, only complemented by (4.5), the relation (4.9) becomes equivalent to (4.8) so defines the PM algebra p ℓ .The condition (4.5) implies multiplicity one for each generator of given spin and depth.In the following we shall use the relations (4.5) and (4.9) to derive the trace of PM algebras.Instead, in Appendix A, we make use of relations (4.7) for alternative derivation of the trace.

Decomposition of traceful generators into PM ones
For concreteness, let us look at the generators of A λ (and p ℓ ) more closely and show that they indeed correspond to PM Killing tensors given in (2.9).We begin with hs(so d+2 ) spanned by M (r) which are traceful tensors corresponding to GL(d + 2) {r, r} Young diagrams.The traceful tensor M (r) (4.10) can be decomposed into traceless tensors M (r,2t) with t = 0, . . ., [r/2] , which can be interpreted as PM generators of spin r + 1 and depth 2t .These tensors can be conveniently handled by contracting them with Then, the traceless generators of hs(so d+2 ) given by can be expressed solely through traces of M ab 's and W ab 's, as they do not have any free index.Using the following identity based on (4.3) and (4.12), one can show that any trace of type (4.14) will be broken down to a function of Here again, one can show that the last object with p ≥ 2 is not independent and can be expressed in terms of the rest.Hence, M (r,2t) ( W ) in (4.13) are polynomials of M 2 W 2 , M W and M 2 W 2 .Rewriting them as ⋆ polynomials, we can replace M 2 by 2(1 − λ 2 ) in A λ .In this way, any element of A λ , namely PM generators, can be written as ⋆ polynomials of M W and M 2 W 2 with W 2 -dependent coefficients: where a and b are some constants.Here, one can notice that the ⋆ polynomial corresponding to M (r,2t) has maximal orders r − 2t and t in M W and M 2 W 2 , respectively.In A λ , any ⋆ powers of M 2 W 2 can appear so PM generators of any even depth are present there.However in p ℓ , one can show using (4.9) that the ℓ-th ⋆ power of M 2 W 2 reduces to lower order ones: with some constant a .Therefore, the generators of depth not smaller than 2 ℓ become linear combinations of generators of lower depth.In order to remove this redundancy, one can impose on W ab the orbit condition (3.13), which is equivalent to Here, W ab denotes W ab satisfying the above condition.With this, any higher than ℓ − 1 ⋆-powers of M 2 W 2 vanish in p ℓ : It will be convenient to require another condition on W ab , to hold.Note that this requirement does not project any component of generators in (4.13) so we do not loose any generality by imposing it.To summarize, with (4.19) and (4.21), any elements of hs(so d+2 ) are projected to the generators of A λ (or p ℓ ) in a simple manner as where the second case only holds for p ℓ .

Trace and Bilinear Form
Analogously to the case of massless HS algebra, it will be useful to define a trace in PM HS algebras, from which the bi-/tri-linear forms and the structure constants can be derived.We define the trace as the coefficient of identity, similarly to the massless HS algebra, where T a denotes a generator of PM HS algebra.The expression of trace for any element in the algebra can be obtained in principle by only using the definition (5.1) as in the case of massless HS algebra [6], and the trace formula for PM HS algebra p ℓ is obtained in that way for a few lower ℓ's in Appendix A.
In the following, we will derive the bilinear form of A λ , that should manifest the appearance of the ideal q ℓ and the corresponding coset p ℓ when λ = ℓ .

Non-Gaussian projector
In the massless case, it turns out [6] that the trace obtained from the definition (5.1) admits a simple representation, where ∆(y) is the projector introduced in [24, 25]8 .The function ∆(y) giving the trace is not unique as shown in [6] even though the definition (5.1) is unambiguous.Generalizing this massless result to PM one, we assume that the trace of the PM algebra A λ admits also an analogous expression, with the conditions (5.4) The above conditions are satisfied by the ansatz, (5.5) In the massless case, we also impose the condition that K αβ ⋆ ∆ 1 (z) = 0 .In the case of A λ , we impose instead In order to determine such ∆ λ , we assume with Π n (z) satisfying We shall show below that (5.6) fixes uniquely the coefficients d n (λ) and determines the projector.Since the condition (4.9) should hold for λ = ℓ , the coefficients in (5.7) will satisfy d n≥ℓ (ℓ) = 0 .The identification of Π n (z) is in parallel with ∆ 1 in [24], and the equation (5.8) for Π n takes the following form, Clearly Π m (z) with m < n also solves the equation for Π n but since we are considering a linear combination of them, without a loss of generality we can take a particular solution of (5.9), Π n (z) = up to an overall factor c which can in turn be fixed from the normalization condition Tr(1) = 1 (or, equivalently ∆ λ (0) = 1).Notice that the coefficient d n (λ) is ill-defined for d = 2n .This problem is in fact an artifact of non-Gaussian projector.In terms of Gaussian projector, this factor will disappear and in the end all formulas are well-defined.Details of computation for (5.11) can be found in Appendix A .Remark also that the dependence of λ and d in (5.11) is such that there exist special values of λ where the series (5.7) acquire special properties.These special cases deserve more careful analysis since they may imply the appearance of an ideal in the algebra A λ .For concrete treatment, we shall explicitly calculate the bilinear form using the trace formula.However, as in the massless case, the non-Gaussian dependence on oscillators in the projector complicates the computations.Fortunately, we can find an alternative projector which gives the same trace but admits a treatable expression.

Gaussian projector
The integral form (5.10) of Π n is simple but not very useful for actual computations as it involves (5.12) However, using the same method as the one described in the Appendix of [6], one can show that the expression is equivalent to Π n in the following sense: Here, one can see that the problematic factor d − 2n cancels out when working with the Gaussian projector.Since ∆ λ will be used only within the setting of ( 5.3), we can replace Π n by (d − 2n)P n in the formula of the projector.Evaluating the sum in (5.7), we obtain new projector as with Let us emphasize again that this projector is equivalent to ∆ λ since (5.17) The apparent advantage of this projector is that it involves a Gaussian function of oscillators and therefore is convenient for actual calculations.Hence, we will use the expression (5.15) in the computation of invariant bilinear form of the PM HS algebra.The trace defined with (5.15) can be equally obtained starting from the definition (5.1) as we have demonstrated for p ℓ in Appendix A.

Bilinear form
Analogously to massless case [6], we consider now the invariant bilinear form of the algebra given as a trace from star-product of generating functions of all PM generators: We use (5.17) for the trace, hence consider first which can be evaluated using the star product composition formula derived in [27] and used in [6] as This expression involves determinants of N × N matrices, which can be simplified to Further simplification can be made by defining the dual matrices V ij as and this enables us to rewrite (using Sylvester's determinant theorem) the N × N determinant as 4 × 4 one as Using the parameterization (4.19) and the condition (4.21), the above reduces to which can be straightforwardly evaluated to give (5.25) Finally, using (5.15), (5.19) and (5.25), the invariant bilinear form can be expressed as Let us make a few observations on the bilinear form (5.26).First of all, it is manifestly symmetric in λ → −λ, which is a straightforward consequence of the symmetry of the quadratic Casimir.Moreover, when λ = ℓ the bilinear form become degenerate for the generators M (r,2t) with t ≥ ℓ .This degeneracy is less manifest since does not exactly correspond to the contribution of M (r,2t) -in other words, the bilinear form is not diagonal.Taking the simplest example of λ = ℓ = 1 case, the W 2 -order term of the bilinear form is given by with some constant a .By decomposing the LHS of the equation into traceless tensors, one can verify that the PM generators of the second depth have vanishing bilinear from: hence form an ideal.Likewise, the appearance of the ideal q ℓ and the corresponding coset algebra p ℓ in A ℓ can be checked with diagonalization.And as we discussed before, the ideal part -having vanishing bilinear form -can be conveniently discarded by contracting with W ab satisfying the orbit condition (3.13).We shall keep this condition in the following section.

Finite dimensional truncations
The normalization factor N λ (5.16) of the bilinear form (5.26) contains which vanishes for λ − d 2 = 0, 1, . . . .Remind that N λ is fixed with the condition Tr(1) = 1 which means the norm of the identity is one.When N λ itself vanishes, the only way to keep the normalization of the identity is that the integral (5.26) should diverge for the identity part.In this case, if there exists generators which give finite integral, then their norm will vanish and the algebra develops a new ideal corresponding to such generators.
In order to study these new ideals and corresponding coset algebras, it is useful to use the transformation of hypergeometric function, to rewrite the bilinear form as Focusing on the M (r,2t) part, which corresponds to the integral is proportional, up to finite coefficient, to .5)This integral diverges when the first factor Γ( d 2 − λ + r − t) diverges while the summation part is always finite.Hence, for a fixed value of λ with vanishing N λ : the algebra develops an infinite-dimensional ideal corresponding to the generators M (r,2t) with r − t > k .The coset algebra, denoted henceforth by f k , is finite dimensional one with the generators M (r,2t) satisfying r − t ≤ k .(6.7) The Young diagram of M (r,2t) has r and r−2t boxes in the first and second row respectively, so the total number is 2(r − t) .Therefore, the algebra f k consists of the generators whose Young diagram contains no more than 2k boxes.This situation is very similar to the one with massless HS algebras in three dimensions 9 , and five dimensions [6,28] where the corresponding HS algebras are the symmetries of symmetric tensor representations of sl 4 or, equivalently, (anti-)self-dual (spin-)tensors of so 6 described by rectangular threerow Young diagrams.For example, f k with k = 3 have (we omit the identity generator, corresponding to Young diagram with no boxes) where we have organized the generators in terms of the number of boxes.Regrouping the same set according to spins, we The dimension of f 3 is (d+1)(d+2)(d+6) 6 2 . For generic integer value of k , the dimension of f k is given again by a perfect square M 2 k with .10)This suggests that the algebra f k is isomorphic to gl M k , an endomorphism of a so d+2 representation with dimension M k .It turns out that this representation corresponds to the finite-dimensional one-row so d+2 Young diagram with length k : .11)This can be verified by comparing the quadratic Casimir of (6.11) -given by k(k + d)with that of (6.6) : the latter gives hence they coincide as expected.This proves that the finite-dimensional coset algebra f k corresponds to the symmetries of the finite-dimensional representation (6.11).The latter representation can be also understood from the boundary point of view: they should correspond to the solution (sub)space of higher-order Laplace equation for scalar field d/2+k φ = 0 .Since any higher-order Laplace equation, ℓ φ = 0 , is equivalent to the set of equations in the (d + 2)-dimensional ambient space: one can study its solutions using the above system (see e.g.[17]) 10 .For ℓ = d 2 + k , the ambient field Φ has homogeneity degree k , any k-th order polynomials in X a dual to traceless tensors (6.11) become solutions of the system (6.13).
Let us conclude this section with a short summary.In this section, using the bilinear form (6.3), we have identified the finite dimensional PM HS algebras corresponding to the symmetry of the symmetric tensor representation (6.11) of (A)dS algebra.Retrospectively, these algebras could be identified with no reference to the bilinear form but only scanning possible finite dimensional representations of (A)dS algebra whose tensor square satisfy the condition of PM algebras.

PM Algebras and CHS Symmetries
Conformal Higher Spin (CHS) theory is an interacting theory of conformal higher-spin fields -that is, defined by Fradkin-Tseytlin free action [30] -of all spins [31,32] and its symmetry coincides with the symmetry of massless HS theory in one higher dimensions.It has been shown that (d + 1)-dimensional free conformal spin-s field can be decomposed, around (A)dS d+1 , into the set of spin-s PM fields with all depths [33][34][35]: where PM (s,t) with t ≥ s are massive fields.Therefore, CHS symmetry itself can be regarded as the symmetry of PM fields of any depths.Then a natural question arises: whether the even depth PM symmetries discussed in this paper (A λ , p ℓ or f k ) can be embedded as a subalgebra in CHS symmetry.First of all, it is rather straightforward to see that the even depth PM part of CHS symmetry form a subalgebra.CHS symmetry in (A)dS d+1 is isomorphic to the massless HS algebra in (A)dS d+2 , so can be realized by 10 In (6.13), we can consider the transformation to flip the sign of ℓ .Higher order singletons are usually described with the positive sign -that is, with homogeneity −(d + 2ℓ)/2 -together with (X 2 ) ℓ Φ(X) = 0 .Equivalent description is given by equation (∂ 2 X ) ℓ Φ(X) = 0 supplemented with condition (X 2 ) Φ(X) = 0 .These conditions can be solved to get higher order conformal wave operators (GJMS operators) in arbitrary background [29].
d + 2 sets of oscillators, Y αA = (y αa , z α ) .Now considering the map, ρ : ( y αa , z α ) → ( y αa , −z α ) , which is an automorphism for hs(so d+3 ) the ρ invariant space of hs(so d+3 ) forms a subalgebra and such space is generated by even depth PM generators.The next question is to what this symmetry corresponds.One of the simplest way to answer this question is by examining the Howe duality in the oscillator construction.The quotient relation of massless HS algebra in AdS d+2 can be translated in AdS d+1 to where K αβ is the sp 2 generators dual to so d+2 -not so d+3 .The above relation simply means that the dual sp 2 carries the representation realized by z α z β .By calculating the quadratic Casimir, we get and can recognize that it coincides with the C 2 of A 1

2
. This proves that A 1 2 is the even depth subalgebra of CHS symmetry.
The existence of even depth PM subalgebra A 1 2 inside of CHS symmetry has an interesting implication towards a unitary truncation of CHS theory.CHS theory is considered to be non-unitary since its linearized spectrum is described by a higher derivative action.The non-unitarity of the latter become clear when it is decomposed into the actions of PM fields around (A)dS d+1 : (−1) t S PM (s,t) , (7.5) where even depth PM fields and odd depth PM fields have relatively negative sign for kinetic terms.One may wonder whether the above CHS action can be truncated into a unitary one by simply selecting positive sign part.Although it is not certain whether this truncation might be consistent, one can immediately identify one necessary condition for the consistency: the symmetry of the truncated spectrum should form a subalgebra of the original symmetry.For example, the spectrum of Conformal Gravity (CG) decomposes into massless spin two and PM spin two, and their kinetic terms have relatively negative sign.The symmetry of CG -that is the conformal symmetry -contains the symmetry of Einstein Gravity -that is the isometry -as a subalgebra, and the action of CG written as (7.5) can be consistently truncated to Einstein Hilbert one [36][37][38].Now coming back to CHS theory, if CHS action can be truncated into a theory of even depth PM fields, we can escape from the non-unitarity problem at least at the linear level.The fact that A 1 2 corresponds to a subalgebra of CHS symmetry suggests that such a truncation might be viable.
Another interesting question is that whether other even depth PM symmetries we have encountered in this paper can be still considered as a truncation of (a variant of) CHS theory.Although we do not have a proof, we do not see how the other PM algebras can be embedded in CHS symmetry.Moreover, we do not see even the massless HS symmetry, that is p 1 , can be so: interpreted differently, we do not see how usual HS symmetry can be embedded inside of the same symmetry in one higher dimensions.Considering the embedding of p ℓ within conformal-like symmetry, there is a chance that p ℓ is supplemented with odd depth generators to form a new symmetry.Corresponding theories may coinside with Weyl-like theories of partially-massless fields of both even and odd depth lower than a given number that were considered in [33].Conversely, Weyl-like theories of [33] may have even-depth unitary truncations that make use of the algebras p ℓ .