Introduction
Hereafter the ground field is C \mathbb{C} C , and ⟨ ⋯ ⟩ \langle\cdots\rangle ⟨ ⋯ ⟩ denotes the linear span; a subalgebra generated by a set is called so in words. For preliminaries, see [14 ] ; here, we correct several statements there and in [5 ] .
Feasible classification problems of invariant differential operators
The unary invariant operators. The projectively invariant differential operators between spaces of tensor fields on a (super)manifold are called Bol operators [Bol] or just Bols . The problem of their description was completely solved on superstrings — supermanifolds of dimension 1 ∣ n 1|n 1∣ n — endowed with a contact structure, and only for the spaces of weighted densities, not for general tensor fields. Let g \mathfrak{g} g be the Lie superalgebra of all vector fields, or its subalgebra preserving a contact structure, and let g = ⨁ g i \mathfrak{g} = \bigoplus\mathfrak{g}_i g = ⨁ g i be its standard Z \mathbb{Z} Z -grading. The Bols of [14 ] are invariant with respect to a simple maximal subalgebra s ⊂ g \mathfrak{s} \subset\mathfrak{g} s ⊂ g containing the negative part g − = ⨁ i < 0 g i \mathfrak{g}_{-} = \bigoplus_{i<0} \mathfrak{g}_i g − = ⨁ i < 0 g i .
Observe that on contact supermanifolds in dimensions 1 ∣ n 1|n 1∣ n for n = 0 , 1 , 2 n = 0, 1, 2 n = 0 , 1 , 2 there are no other irreducible tensor fields except for weighted densities, since o ( n ) \mathfrak{o}(n) o ( n ) is commutative for these n n n , so its irreducible finite-dimensional modules are 1-dimensional (without a contact structure this fails already for n = 1 n = 1 n = 1 ), and the amount of possible analogs of generalized Bols between spaces tensor fields on supermanifolds of other superdimensions remained unclear.
In [15 ] , we showed that only cases of spaces of tensor fields on supermanifolds of “key superdimensions” ( 2 ∣ 0 , 0 ∣ n (2|0, 0|n ( 2∣0 , 0∣ n and 1 ∣ 1 ) 1|1) 1∣1 ) are of practical interest. For the supermanifolds of other superdimensions a ∣ b a|b a ∣ b , the number of invariant operators is indescribably voluminous and depends on a + b − 1 a+b-1 a + b − 1 parameters even for the 1st order operators. In contrast, on supermanifolds of these other superdimensions a ∣ b a|b a ∣ b , there are no non-scalar invariant differential operators between spaces of weighted densities.
The binary invariant operators. Consider binary operators invariant with respect to the maximal subalgebra containing all negative elements of the ambient superalgebra (of all vector fields, or of its subalgebra preserving a contact structure). For them, the answer is richer than for unary operators. (Without the requirement of invariance relative a maximal subalgebra, the binary operators invariant relative the whole of v e c t ( n ) \mathfrak{vect}(n) vect ( n ) , resp. h ( 2 n ) \mathfrak{h}(2n) h ( 2 n ) , were classified in [33 ] , resp. [34 ] , for supermanifolds, see [35 ] ; on the line, the anti-symmetric m m m -ary ones, for every m m m , were described in [25 ] . Here we consider the binary operators called the Gordan transvectants a.k.a. Rankin-Cohen brackets — briefly: GRC operators in what follows. In this paper, the ambient superalgebra is always a contact superalgebra: k ( 1 ∣ n ) \mathfrak{k}(1|n) k ( 1∣ n ) for n ≤ 6 n \le6 n ≤ 6 , and the pericontact m ( 1 ) \mathfrak{m}(1) m ( 1 ) ; we make no claim about v e c t ( m ∣ n ) \mathfrak{vect}(m|n) vect ( m ∣ n ) , cf. § 1.3 . For the answer on distinguished superstrings, see [LShRC].
The ternary invariant operators. On manifolds and between the spaces of weighted densities, they were classified in [11 ] .
Overview of the literature
The GRC operators were introduced by Gordan [32 ] , see the review by Gières [30 ] . They were rediscovered by Rankin [45 ] and Cohen [17 ] and gained a great deal of currency as Rankin-Cohen operators , see [Z] and the review [24 ] . For their representation-theoretic counterparts and relatives, see [50 , 42 BCK] .
The GRC operators between spaces of weighted densities when there are no other types of tensor fields.
As noted in [13 ] , the situation when all types of tensor fields are only weighted densities is only possible on superstrings of superdimension 1 ∣ n 1|n 1∣ n for n = 0 , 1 , 2 n = 0, 1, 2 n = 0 , 1 , 2 . Moreover, a contact structure must be present, given by an odd contact or an even pericontact form.
The first to describe super GRC operators — bilinear operators invariant under the image of o s p ( 1 ∣ 2 ) \mathfrak{osp}(1|2) osp ( 1∣2 ) embedded into k ( 1 ∣ 1 ) \mathfrak{k}(1|1) k ( 1∣1 ) , i.e., on the ( 1 ∣ 1 ) (1|1) ( 1∣1 ) -dimensional superstring with a contact structure — were Gières and Theisen, see [31 ] , see also [36 , 28 ] . Next, a slightly modified expression was given by Gargoubi and Ovsienko [29 ] who interpreted the super versions of the GRC operators they found in terms of the ( 1 ∣ 2 ) (1|2) ( 1∣2 ) -dimensional superstring C 1 ∣ 2 \mathbb{C}^{1|2} C 1∣2 endowed with a contact form.
In [43 ] , Omri described the GRC operators invariant under o s p ( 2 ∣ 2 ) \mathfrak{osp}(2|2) osp ( 2∣2 ) embedded into k ( 1 ∣ 2 ) \mathfrak{k}(1|2) k ( 1∣2 ) , but he considered weighted densities with only 1 parameter, whereas there are 2 parameters, see [14 ] , so the result is incomplete.
The paper [5 ] is similar to [43 ] , but deals with o s p ( 3 ∣ 2 ) \mathfrak{osp}(3|2) osp ( 3∣2 ) embedded into k ( 1 ∣ 3 ) \mathfrak{k}(1|3) k ( 1∣3 ) ; in this case spaces of weighted densities depend on 1 parameter; this is a complete result. For the unary o s p ( n ∣ 2 ) \mathfrak{osp}(n|2) osp ( n ∣2 ) -invariant operators between spaces of weighted densities, see [2 ] .
The GRC operators between spaces of weighted densities when there are other tensor fields.
For the list of maximal subalgebras m ⊂ g \mathfrak{m} \subset\mathfrak{g} m ⊂ g such that m − = g − \mathfrak{m}_{-} = \mathfrak{g}_{-} m − = g − in the Z \mathbb{Z} Z -graded Lie (super)algebra of vector fields g = v e c t ( m ∣ n ) \mathfrak{g} = \mathfrak{vect}(m|n) g = vect ( m ∣ n ) or its contact subalgebra, see [14 Table (4)] and [LSch]. In the non-super case, there are two series of such subalgebras: s l ( n + 1 ) \mathfrak{sl}(n+1) sl ( n + 1 ) and o ( n + 2 ) \mathfrak{o}(n+2) o ( n + 2 ) in v e c t ( n ) \mathfrak{vect}(n) vect ( n ) .
For the maximal subalgebra s l ( n + 1 ) ⊂ v e c t ( n ) \mathfrak{sl}(n+1) \subset\mathfrak{vect}(n) sl ( n + 1 ) ⊂ vect ( n ) , Lecomte and Ovsienko [37 ] considered the projectively equivariant symbol calculus on the symmetric contravariant tensor fields, see also [22 ] ; their result was extended to these tensor fields tensored with a weighted density in [9 ] . The analogous problem for the conformal subalgebra o ( a + 1 , b + 1 ) ⊂ v e c t R ( n ) \mathfrak{o}(a+1,b+1) \subset\mathfrak{vect}_{\mathbb{R}}(n) o ( a + 1 , b + 1 ) ⊂ vect R ( n ) and these tensor fields is open.
Ovsienko and Redou ([44 ] , see also [8 , 21 , 23 ] ) considered analogs of GRC brackets between spaces of weighted densities on the n n n -dimensional real manifold with a non-degenerate metric of signature ( a , b ) (a,b) ( a , b ) . They classified GRC brackets over R \mathbb{R} R conformally invariant with respect to the maximal subalgebra o ( a + 1 , b + 1 ) ⊂ v e c t R ( n ) \mathfrak{o}(a+1,b+1) \subset\mathfrak{vect}_{\mathbb{R}}(n) o ( a + 1 , b + 1 ) ⊂ vect R ( n ) .
Some applications of the super GRC brackets.
For the first appearances, see [30 , 31 , 36 , 18 ] .
These brackets are used in [4 ] to compute H 2 ( t ( 1 ∣ 2 ) , o s p ( 2 ∣ 2 ) ; D 2 ( F λ , F μ ) ) H^{2}(\mathfrak{t}(1|2),\mathfrak{osp}(2|2);\mathcal{D}^{2}(\mathcal{F}_{\lambda},\mathcal{F}_{\mu})) H 2 ( t ( 1∣2 ) , osp ( 2∣2 ) ; D 2 ( F λ , F μ )) . Similar computations were performed by Omri, who computed H 1 ( t ( 1 ∣ 2 ) , o s p ( 2 ∣ 2 ) ; D 2 ( F λ , F μ ) ) H^{1}(\mathfrak{t}(1|2),\mathfrak{osp}(2|2);\mathcal{D}^{2}(\mathcal{F}_{\lambda},\mathcal{F}_{\mu})) H 1 ( t ( 1∣2 ) , osp ( 2∣2 ) ; D 2 ( F λ , F μ )) , see [43 Thm. 4.1] . Here, D 2 ( F λ , F μ ) \mathcal{D}^{2}(\mathcal{F}_{\lambda},\mathcal{F}_{\mu}) D 2 ( F λ , F μ ) is the space of all linear differential operators acting between the spaces of weighted densities on R 1 ∣ 2 \mathbb{R}^{1|2} R 1∣2 . The superscript 2 is the number of odd indeterminates, not the order. (The goal was to superize the result of [16 ] where the cohomology H 1 ( v e c t ( 1 ) , s l ( 2 ) ; D ( F λ , F μ ) ) H^{1}(\mathop{\rm vect}(1),\mathfrak{sl}(2);\mathcal{D}(\mathcal{F}_{\lambda},\mathcal{F}_{\mu})) H 1 ( vect ( 1 ) , sl ( 2 ) ; D ( F λ , F μ )) , where D \mathcal{D} D is the space of (all) differential operators acting on the spaces of weighted densities was computed. This cohomology appears in the context of equivariant quantization, cf. [41 ] ; for the cohomology of the Lie algebras of vector fields on R P 1 \mathbb{R}P^{1} R P 1 with coefficients in the bilinear differential operators, see [10 ] .)
Next, Omri described the deformation of the space of symbols; this deformation becomes trivial once restricted to o s p ( 2 ∣ 2 ) \mathfrak{osp}(2|2) osp ( 2∣2 ) . To describe it, he used a kind of Massey product of cocycles he had found previously. He also gave a necessary and sufficient condition for an infinitesimal deformation to be a global deformation.
Ben Ammar with his student Mtaouaa computed the deformation that does not necessarily become trivial once restricted on o s p ( 2 ∣ 2 ) \mathfrak{osp}(2|2) osp ( 2∣2 ) , see [BeM]. They also obtained some necessary and sufficient conditions for that similar to what Omri did in [43 ] .
What is done in this paper
We consider six cases, cf. [14 Table (4)] ; the first five are contact, the sixth is pericontact:
o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2) \subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) for n = 1 , 2 n=1,2 n = 1 , 2 , see §2. For n = 2 n=2 n = 2 we classify all binary i ( o s p ( 2 ∣ 2 ) ) i(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) -invariant operators, for all weights of the densities — both parameters of the weight being free, which is what [43 ] lacked, see Theorem 2.9 . There are exactly two operators of each order n ≥ 1 n \ge1 n ≥ 1 , so the lists in [29 , 43 ] are incomplete: for every order there is one more even operator, and there are odd operators of every odd order, not only of order 1.
o s p ( 3 ∣ 2 ) ⊂ k ( 1 ∣ 3 ) \mathfrak{osp}(3|2) \subset\mathfrak{k}(1|3) osp ( 3∣2 ) ⊂ k ( 1∣3 ) , see §3. This case is treated in [5 ] ; we verify their answer and correct three statements of their Theorem 4.1 — the signs in the even supertransvectants, formula (4.2) of [5 ] , the missing Levi-Civita sign in the odd ones, formula (4.1) of [5 ] , and the description of the exceptional weights.
o s p a ( 4 ∣ 2 ) ⊂ k ( 1 ∣ 4 ) \mathfrak{ospa}(4|2) \subset\mathfrak{k}(1|4) ospa ( 4∣2 ) ⊂ k ( 1∣4 ) , see §4. Here, the subalgebra is not one but a one-parameter family. We describe the GRC brackets of every member in orders ≤ 4 \le4 ≤ 4 (there are none in orders 1 and 3). We describe its Bol operators of orders ≤ 8 \le8 ≤ 8 , thereby confirming [14 Thm 1.6] in these orders by an independent computation. We also answer the question asked in [14 §3.2.3a] about isomorphic algebras with different answers.
o s p ( 5 ∣ 2 ) ⊂ k ( 1 ∣ 5 ) \mathfrak{osp}(5|2) \subset\mathfrak{k}(1|5) osp ( 5∣2 ) ⊂ k ( 1∣5 ) , see §5. Here, on the contrary, the subalgebra o s p ( 5 ∣ 2 ) \mathfrak{osp}(5|2) osp ( 5∣2 ) is rigid (the family occurs for n = 4 n=4 n = 4 and for no other n n n ). The odd n n n shows itself in the parity: the Bol operators are odd , of order 5 + 2 k 5+2k 5 + 2 k for every k ∈ Z + k \in\mathbb{Z}_{+} k ∈ Z + , namely T 5 ∘ ∂ t k T_{5}\circ\partial_{t}^{k} T 5 ∘ ∂ t k , see Theorem 5.3.1, and so are the GRC brackets of odd order. The first of these brackets is of order 5, not of order 3 as for n = 3 n=3 n = 3 . For generic weights, there is one bracket in each even order and one in each odd order from 5 on.
q ⊂ a s ⊂ k ( 1 ∣ 6 ) \mathfrak{q} \subset\mathfrak{as} \subset\mathfrak{k}(1|6) q ⊂ as ⊂ k ( 1∣6 ) , see §6. This is a case [14 §1.3.1] left “for the future”. It is unlike the others: g \mathfrak{g} g consists of the contact fields K f K_{f} K f with ∂ t ( f ) = 0 \partial_{t}(f)=0 ∂ t ( f ) = 0 , so the weight of the densities is invisible to it. There are no exceptional weights and no analogs of Bol operators: the only invariant unary operators are the powers of ∂ t \partial_{t} ∂ t . Every invariant binary operator of order ≤ 4 \le4 ≤ 4 is a linear combination of the operators ∂ t a ⊗ ∂ t b \partial_{t}^{a}\otimes\partial_{t}^{b} ∂ t a ⊗ ∂ t b and of the single non-trivial operator Ω \Omega Ω of order 2, precomposed with ∂ t \partial_{t} ∂ t in one of the arguments, see Theorem 6.5.1. Since the densities thus carry nothing, we go on to the tensor fields of the other types, see § 6.6 . There, the weights, and the
twists by a power of the determinant, are invisible to a s \mathfrak{as} as as well. For the half-spinors, the two chiralities behave differently : exactly one of the two Dirac operators is a s \mathfrak{as} as -invariant.
p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2) \subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 ) , see §7. Here, the operators J k , l \mathcal{J}_{k,l} J k , l and J k , r \mathcal{J}_{k,r} J k , r are the Bol operators of [14 Thm 5.1] acting in one of the two arguments. The weights on which they are invariant are computed in Theorem 7.3. For k = 2 k=2 k = 2 , there is a third, independent, invariant operator, see Statement 7.4. As a by-product, we correct [14 Thm 5.1] itself at k = 2 k=2 k = 2 : there, the order-2 Bol operator is not isolated but a member of the 1-parameter family D ξ D q + 2 a ∂ τ : F a , b ⟶ F a − 1 , b \mathcal{D}_{\xi}\mathcal{D}_{q}+2a\partial_{\tau}:\mathcal{F}_{a,b}\longrightarrow\mathcal{F}_{a-1,b} D ξ D q + 2 a ∂ τ : F a , b ⟶ F a − 1 , b , invariant for all weights. This family exhausts the invariant operators of its degree, charge and parity, whereas only the member a = 1 2 a=\frac{1}{2} a = 2 1 is printed in [14 ] . For k = 1 , 3 , 4 k=1,3,4 k = 1 , 3 , 4 , the theorem is confirmed as printed, see Statement 7.5.1. The complete classification of the GRC brackets over p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2)\subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 ) , for every order and every weight, is given in §8.
We make no claim about the general vectorial series v e c t ( m ∣ n ) \mathfrak{vect}(m|n) vect ( m ∣ n ) : nothing in this text concerns them. One further contact case, s l ( 1 ∣ m ∣ n ∣ 1 ) ⊂ k ( 2 m + 1 ∣ 2 n ) \mathfrak{sl}(1|m|n|1)\subset\mathfrak{k}(2m+1|2n) sl ( 1∣ m ∣ n ∣1 ) ⊂ k ( 2 m + 1∣2 n ) , is postponed.
“Statements” are proved by computations. Where the computation covers a whole finite list, the Statement is final. Where it covers several first cases, these cases give us reasons to formulate a Conjecture for the general case, stated right after the Statement that supports it. All the classification statements below were obtained, and all the formulas printed below were checked, by computation with all the generators of the relevant subalgebra.
Super GRC operators: the o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2)\subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) case
In this Section, F \mathcal{F} F is the space of functions on the superstring, i.e., the coefficients of the weighted densities, and f ( a ) : = ( ∂ ∂ t ) a ( f ) f^{(a)}:=\left(\frac{\partial}{\partial t}\right)^{a}(f) f ( a ) := ( ∂ t ∂ ) a ( f ) for every f ∈ F f\in\mathcal{F} f ∈ F . Theorem 2.1 considerably simplifies calculations; it needs the following notation.
Recall the definition of K f K_f K f and introduce the operators D f D_f D f , see [14 ] , for deg K f < 0 \deg K_f<0 deg K f < 0 :
D 1 = ∂ t , K 1 = 2 ∂ t ; D η i : = ξ i ∂ t + ∂ η i , D ξ i : = η i ∂ t + ∂ ξ i , D θ : = θ ∂ t + ∂ θ ; K ξ i = ξ i ∂ t − ∂ η i , K η i = η i ∂ t − ∂ ξ i , K θ = θ ∂ t − ∂ θ . (1) \begin{aligned}
D_1&=\partial_t,\quad K_1=2\partial_t;\\
D_{\eta_i}&:=\xi_i\partial_t+\partial_{\eta_i},\quad D_{\xi_i}:=\eta_i\partial_t+\partial_{\xi_i},\quad D_\theta:=\theta\partial_t+\partial_\theta;\\
K_{\xi_i}&=\xi_i\partial_t-\partial_{\eta_i},\quad K_{\eta_i}=\eta_i\partial_t-\partial_{\xi_i},\quad K_\theta=\theta\partial_t-\partial_\theta. \tag*{(1)}
\end{aligned} D 1 D η i K ξ i = ∂ t , K 1 = 2 ∂ t ; := ξ i ∂ t + ∂ η i , D ξ i := η i ∂ t + ∂ ξ i , D θ := θ ∂ t + ∂ θ ; = ξ i ∂ t − ∂ η i , K η i = η i ∂ t − ∂ ξ i , K θ = θ ∂ t − ∂ θ . (1)
Recall the definition of M f M_f M f and introduce the operators D f D_f D f for deg M f < 0 \deg M_f<0 deg M f < 0 :
D 1 = ∂ τ , M 1 = 2 ∂ τ ; M ξ i : = ξ i ∂ τ + ∂ q i , M q i : = q i ∂ τ − ∂ ξ i ; D ξ i : = ξ i ∂ τ − ∂ q i , D q i : = q i ∂ τ + ∂ ξ i . (2) \begin{aligned}
D_1&=\partial_\tau,\quad M_1=2\partial_\tau;\\
M_{\xi_i}&:=\xi_i\partial_\tau+\partial_{q_i},\quad M_{q_i}:=q_i\partial_\tau-\partial_{\xi_i};\\
D_{\xi_i}&:=\xi_i\partial_\tau-\partial_{q_i},\quad D_{q_i}:=q_i\partial_\tau+\partial_{\xi_i}. \tag*{(2)}
\end{aligned} D 1 M ξ i D ξ i = ∂ τ , M 1 = 2 ∂ τ ; := ξ i ∂ τ + ∂ q i , M q i := q i ∂ τ − ∂ ξ i ; := ξ i ∂ τ − ∂ q i , D q i := q i ∂ τ + ∂ ξ i . (2)
In the description of GRC operators, we will need, for odd operators X X X and Y Y Y , the expression denoted in [14 eq. (42)] by [ X , Y ] + [X,Y]_+ [ X , Y ] + (note that the supercommutator of odd operators is X Y + Y X XY+YX X Y + Y X , not X Y − Y X XY-YX X Y − Y X ):
[ X , Y ] + : = X Y − Y X . (3) [X,Y]_+:=XY-YX. \tag*{(3)} [ X , Y ] + := X Y − Y X . (3)
2.1. Theorem (All invariant differential operators have constant coefficients in an invariant basis Sh). Let h \mathfrak{h} h be a subalgebra in a vectorial Lie superalgebra g \mathfrak{g} g , and h − = g − \mathfrak{h}_{-}=\mathfrak{g}_{-} h − = g − .
Let g : = k ( 2 m + 1 ∣ n ) \mathfrak{g}:=\mathfrak{k}(2m+1|n) g := k ( 2 m + 1∣ n ) . Any h − \mathfrak{h}_{-} h − -invariant differential operator is a polynomial with constant coefficients in the operators D f D_f D f , see eq. (1).
If g = m ( n ) \mathfrak{g}=\mathfrak{m}(n) g = m ( n ) , then any h − \mathfrak{h}_{-} h − -invariant differential operator is a polynomial with constant coefficients in the operators D f D_f D f , see eq. (2).
If g = vect ( m ∣ n ) \mathfrak{g}=\operatorname{vect}(m|n) g = vect ( m ∣ n ) , then any h − \mathfrak{h}_{-} h − -invariant differential operator is a polynomial with constant coefficients in partial derivatives.
Warning. The expressions below look different from those given in [29 ] and [43 ] because we use different normal shapes of the contact form and bracket.
The same on superstrings, with two exceptions occurring for n = 4 n=4 n = 4 .
Theorem 2.1 is stated for polynomial coefficients, whereas on a superstring — a supermanifold of superdimension ( 1 ∣ n ) (1|n) ( 1∣ n ) — the coefficients are Laurent polynomials. There are two superstrings to consider. The first superstring is associated with the trivial bundle of rank n n n ; we denote its contact Lie superalgebra by k L ( 1 ∣ n ) \mathfrak{k}^{L}(1|n) k L ( 1∣ n ) .1 The second superstring is associated with the Whitney sum of the trivial bundle of rank n − 1 n-1 n − 1 and the Möbius bundle; we denote its contact Lie superalgebra by k M ( 1 ∣ n ) \mathfrak{k}^{M}(1|n) k M ( 1∣ n ) . For the definitions, see [LShRC]. In the standard grading, k L ( 1 ∣ n ) 0 = o ( n ) ⊕ C z \mathfrak{k}^{L}(1|n)_0=\mathfrak{o}(n)\oplus\mathbb{C}z k L ( 1∣ n ) 0 = o ( n ) ⊕ C z , whereas k M ( 1 ∣ n ) 0 = o ( n − 1 ) ⊕ C z \mathfrak{k}^{M}(1|n)_0=\mathfrak{o}(n-1)\oplus\mathbb{C}z k M ( 1∣ n ) 0 = o ( n − 1 ) ⊕ C z for n ≤ 4 n\leq4 n ≤ 4 . Since Theorem 2.1 is the tool on which every ansatz of this paper rests, let us record what it becomes there.
Here, as in [14 ] , an operator B : F a ⟶ F b B:\mathcal{F}_a\longrightarrow\mathcal{F}_b B : F a ⟶ F b between spaces of weighted densities is described by the operator B ‾ \overline{B} B acting on the functions:
B ( f α a ) = B ‾ ( f ) α b , where f ∈ F 0 , (4) B(f\alpha^a)=\overline{B}(f)\alpha^b,\qquad\text{where }f\in\mathcal{F}_0, \tag*{(4)} B ( f α a ) = B ( f ) α b , where f ∈ F 0 , (4)
where α \alpha α designates the generator of a rank-1 module of weighted densities over the algebra of functions; recall that the weight might be a pair of numbers, see [14 ] .
Passing to Laurent coefficients changes nothing. Consider the stringy Lie superalgebra preserving a structure — a tensor or a distribution — on the superstring considered. All the operators B ‾ \overline{B} B invariant with respect to it are polynomials with constant coefficients in an invariant basis, that of Y Y Y -type vectors of [49 ] , precisely as in Theorem 2.1 . There are exactly two exceptions, and both occur for n = 4 n=4 n = 4 : for k M ( 1 ∣ 4 ) \mathfrak{k}^{M}(1|4) k M ( 1∣4 ) and for the derived algebra k L ′ ( 1 ∣ 4 ) \mathfrak{k}^{L\prime}(1|4) k L ′ ( 1∣4 ) , one extra generator is needed. It is written as a product of a negative power of a Y Y Y -vector and a product of Y Y Y -vectors, see formulas (6) and (7). This product is not the composition of its factors, and it is not a differential operator. This is essential: it has to be simplified before it is evaluated, i.e., treated as one whole.
For k M ( 1 ∣ n ) \mathfrak{k}^{M}(1|n) k M ( 1∣ n ) , let ζ \zeta ζ be the odd coordinate of the Möbius summand and θ 1 , … , θ n − 1 \theta_1,\ldots,\theta_{n-1} θ 1 , … , θ n − 1 those of the trivial summand. Introduce the operators
D : = ∂ ∂ t − ζ 2 t ∂ ∂ ζ = 1 2 K 1 M , K ζ M : = ζ ∂ t − 1 2 t ∂ ζ , D ζ M : = ζ ∂ t + 1 2 t ∂ ζ for n ≥ 1 ; K θ i M : = θ i ∂ t − ∂ θ i − θ i ζ 2 t ∂ ζ , D θ i M : = θ i ∂ t + ∂ θ i − θ i ζ 2 t ∂ ζ for n ≥ 2. (5) \begin{aligned}
\mathcal{D}&:=\frac{\partial}{\partial t}-\frac{\zeta}{2t}\frac{\partial}{\partial\zeta}=\frac{1}{2}K^{M}_{1},\\
K^{M}_{\zeta}&:=\zeta\partial_t-\frac{1}{2t}\partial_\zeta,\qquad D^{M}_{\zeta}:=\zeta\partial_t+\frac{1}{2t}\partial_\zeta&& \text{for }n\geq1;\\
K^{M}_{\theta_i}&:=\theta_i\partial_t-\partial_{\theta_i}-\frac{\theta_i\zeta}{2t}\partial_\zeta,\qquad D^{M}_{\theta_i}:=\theta_i\partial_t+\partial_{\theta_i}-\frac{\theta_i\zeta}{2t}\partial_\zeta&& \text{for }n\geq2. \tag*{(5)}
\end{aligned} D K ζ M K θ i M := ∂ t ∂ − 2 t ζ ∂ ζ ∂ = 2 1 K 1 M , := ζ ∂ t − 2 t 1 ∂ ζ , D ζ M := ζ ∂ t + 2 t 1 ∂ ζ := θ i ∂ t − ∂ θ i − 2 t θ i ζ ∂ ζ , D θ i M := θ i ∂ t + ∂ θ i − 2 t θ i ζ ∂ ζ for n ≥ 1 ; for n ≥ 2. (5)
The two extra generators, introduced in [LShRC] in order to write the cocycles of these two exceptional stringy superalgebras, are the following ones. For k L ′ ( 1 ∣ 4 ) \mathfrak{k}^{L\prime}(1|4) k L ′ ( 1∣4 ) , it is the operator Δ I S \Delta_{IS} Δ I S defined by the table in formula (6), where D 1 − 1 = ∂ t − 1 D_1^{-1}=\partial_t^{-1} D 1 − 1 = ∂ t − 1 is the formal inverse of D 1 = ∂ t D_1=\partial_t D 1 = ∂ t , i.e.,
t a ↦ 1 a + 1 t a + 1 for a ≠ − 1 t^a \mapsto\frac{1}{a+1}t^{a+1}\quad\text{for }a\ne-1 t a ↦ a + 1 1 t a + 1 for a = − 1
Δ I S : = D 1 − 1 D θ 1 D θ 2 D θ 3 D θ 4 : t a ↦ a ( a − 1 ) ( a − 2 ) ( a − 3 ) t a − 3 θ 1 θ 2 θ 3 θ 4 , t a θ 1 ↦ − a ( a − 1 ) t a − 2 θ 2 θ 3 θ 4 , t a θ 1 θ 2 ↦ − a t a − 1 θ 3 θ 4 , t a θ 1 θ 2 θ 3 ↦ t a θ 4 , t a θ 1 θ 2 θ 3 θ 4 ↦ 1 a + 1 t a + 1 (6) \Delta_{IS}:=D_1^{-1}D_{\theta_1}D_{\theta_2}D_{\theta_3}D_{\theta_4}:
\qquad
\begin{aligned}
t^a&\mapsto a(a-1)(a-2)(a-3)t^{a-3}\theta_1\theta_2\theta_3\theta_4,\\
t^a\theta_1&\mapsto-a(a-1)t^{a-2}\theta_2\theta_3\theta_4,\\
t^a\theta_1\theta_2&\mapsto-at^{a-1}\theta_3\theta_4,\\
t^a\theta_1\theta_2\theta_3&\mapsto t^a\theta_4,\\
t^a\theta_1\theta_2\theta_3\theta_4&\mapsto\frac{1}{a+1}t^{a+1}
\end{aligned} \tag*{(6)} Δ I S := D 1 − 1 D θ 1 D θ 2 D θ 3 D θ 4 : t a t a θ 1 t a θ 1 θ 2 t a θ 1 θ 2 θ 3 t a θ 1 θ 2 θ 3 θ 4 ↦ a ( a − 1 ) ( a − 2 ) ( a − 3 ) t a − 3 θ 1 θ 2 θ 3 θ 4 , ↦ − a ( a − 1 ) t a − 2 θ 2 θ 3 θ 4 , ↦ − a t a − 1 θ 3 θ 4 , ↦ t a θ 4 , ↦ a + 1 1 t a + 1 (6)
(and similarly for the other θ \theta θ -monomials), whose only singularity — the case a = − 1 a=-1 a = − 1 in the last line — is precisely the element deleted when passing from k L ( 1 ∣ 4 ) \mathfrak{k}^{L}(1|4) k L ( 1∣4 ) to k L ′ ( 1 ∣ 4 ) \mathfrak{k}^{L'}(1|4) k L ′ ( 1∣4 ) . For k M ( 1 ∣ 4 ) \mathfrak{k}^{M}(1|4) k M ( 1∣4 ) , it is the operator Δ \Delta Δ defined, for every φ ∈ Λ ( θ 1 , θ 2 , θ 3 ) \varphi\in\Lambda(\theta_1,\theta_2,\theta_3) φ ∈ Λ ( θ 1 , θ 2 , θ 3 ) and every n n n , by the two formulas
Δ ( t n φ ( θ ) ) = t n ζ φ ( θ ) , Δ ( t n ζ φ ( θ ) ) = t n n + 1 2 φ ( θ ) ; (7) \Delta(t^n\varphi(\theta))=t^n\zeta\varphi(\theta),\qquad
\Delta(t^n\zeta\varphi(\theta))=\frac{t^n}{n+\frac{1}{2}}\varphi(\theta); \tag*{(7)} Δ ( t n φ ( θ )) = t n ζ φ ( θ ) , Δ ( t n ζ φ ( θ )) = n + 2 1 t n φ ( θ ) ; (7)
we write Δ = D ζ M D − 1 \Delta=D^M_\zeta D^{-1} Δ = D ζ M D − 1 for short, an operator which is defined everywhere and coincides with the composition D ζ M ∘ D − 1 D^M_\zeta\circ D^{-1} D ζ M ∘ D − 1 on every function on which the composition D ζ M D − 1 D^M_\zeta D^{-1} D ζ M D − 1 is defined — the pole of D − 1 D^{-1} D − 1 at n = − 1 n=-1 n = − 1 being canceled by the zero of D ζ M D^M_\zeta D ζ M . In both cases, the literal reading of the composition — perform one factor, then the other — is not the operator meant: it loses the cancellation. For Δ I S \Delta_{IS} Δ I S and g = t a g=t^a g = t a , for instance, the four D θ i D_{\theta_i} D θ i produce a ( a − 1 ) ( a − 2 ) ( a − 3 ) t a − 4 θ 1 θ 2 θ 3 θ 4 a(a-1)(a-2)(a-3)t^{a-4}\theta_1\theta_2\theta_3\theta_4 a ( a − 1 ) ( a − 2 ) ( a − 3 ) t a − 4 θ 1 θ 2 θ 3 θ 4 and D 1 − 1 D_1^{-1} D 1 − 1 divides by a − 3 a-3 a − 3 , so the literal reading returns 0 0 0 at a = 3 a=3 a = 3 , whereas eq. (6) returns 6 θ 1 θ 2 θ 3 θ 4 6\theta_1\theta_2\theta_3\theta_4 6 θ 1 θ 2 θ 3 θ 4 , which is what the cocycle identity requires. For Δ \Delta Δ and g = t − 1 θ 1 θ 2 θ 3 g=t^{-1}\theta_1\theta_2\theta_3 g = t − 1 θ 1 θ 2 θ 3 , an element of the algebra, the literal reading is not defined at all.
Theorem. Extending Theorem 2.1 to superstrings Sh, BLS1, LShRC
Theorem (Extending Theorem 2.1 to superstrings; [49 , 14 LShRC] ). The description of the h − \mathfrak{h}_{-} h − -invariant operators B ‾ \overline{B} B given in Theorem 2.1 holds verbatim for the stringy versions of the Lie superalgebras listed there, i.e., for Laurent coefficients.
If g = k M ( 1 ∣ n ) \mathfrak{g}=\mathfrak{k}^{M}(1|n) g = k M ( 1∣ n ) , where n ≤ 3 n\le3 n ≤ 3 , then any g \mathfrak{g} g -invariant differential operator B ‾ \overline{B} B is a polynomial with constant coefficients in the operators D f D_f D f of eq. (5).
If g = k M ( 1 ∣ 4 ) \mathfrak{g}=\mathfrak{k}^{M}(1|4) g = k M ( 1∣4 ) , then any g \mathfrak{g} g -invariant operator B ‾ \overline{B} B is a polynomial with constant coefficients in the operators D f D_f D f of eq. (5) and the composition Δ D θ 3 M D θ 2 M D θ 1 M \Delta D^M_{\theta_3}D^M_{\theta_2}D^M_{\theta_1} Δ D θ 3 M D θ 2 M D θ 1 M , see eq. (7).
If g = k L ′ ( 1 ∣ 4 ) \mathfrak{g}=\mathfrak{k}^{L'}(1|4) g = k L ′ ( 1∣4 ) , then any g \mathfrak{g} g -invariant operator B ‾ \overline{B} B is a polynomial with constant coefficients in the operators D f D_f D f of eq. (1) and the operator Δ I S \Delta_{IS} Δ I S , see eq. (6).
In the last three claims, g \mathfrak{g} g -invariance, and not h − \mathfrak{h}_{-} h − -invariance, is meant: for k M ( 1 ∣ n ) \mathfrak{k}^{M}(1|n) k M ( 1∣ n ) , the operators D f M D^M_f D f M do not supercommute with the whole of the analog of h − \mathfrak{h}_{-} h − ; e.g., [ D , D ζ M ] = − 1 2 t D ζ M [D,D^M_\zeta]=-\frac{1}{2t}D^M_\zeta [ D , D ζ M ] = − 2 t 1 D ζ M . So the claims describe the shape of an operator invariant with respect to the whole of g \mathfrak{g} g ; they say nothing about the operators invariant with respect to a subalgebra.
Proof. All the claims, except for the cases k M ( 1 ∣ n ) \mathfrak{k}^{M}(1|n) k M ( 1∣ n ) and k L ′ ( 1 ∣ 4 ) \mathfrak{k}^{L'}(1|4) k L ′ ( 1∣4 ) , are proved in [49 , 14 ] . The exceptional cases are proved by the same arguments applied to the differential part of B ‾ \overline{B} B . It remains to observe that each of the two extra operators, Δ \Delta Δ and Δ I S \Delta_{IS} Δ I S , is the only — up to a factor — invariant extension of that differential part to the whole algebra. Note that the arguments of [49 , 14 ] concern differential operators, whereas Δ \Delta Δ and Δ I S \Delta_{IS} Δ I S are not differential ones. □
What this means for the ansatz.
For the cases treated in this paper, n ≤ 3 n\le3 n ≤ 3 , Theorem 2.2 says that nothing has to be added. The ansatz “a polynomial with constant coefficients in the D f D_f D f ”, used from Subsection 2.3 on, is complete on the superstring as well as on the supercircle. In the case n = 4 n=4 n = 4 , the algebra k L ( 1 ∣ 4 ) \mathfrak{k}^{L}(1|4) k L ( 1∣4 ) is not simple, and the Bol and GRC operators are not described. For n = 4 n=4 n = 4 , the ansatz has to be enlarged by the corresponding operator (6) or (7). Since these are not differential operators, the very notion of the order of the operator has to be revised there.
The order and the degree of an operator
Both words are used in the literature we quote, and they denote two different integers. Here is the convention of this paper, applied from now on without further comment.
By Theorem 2.1 , every operator considered below is a polynomial with constant coefficients in the operators D f D_f D f of eq. (1) (resp. D f D_f D f of eq. (2) ). For such a polynomial:
Its order ord \operatorname{ord} ord is its degree in the D f D_f D f , where each D ξ i D_{\xi_i} D ξ i , D η i D_{\eta_i} D η i , D θ i D_{\theta_i} D θ i (resp. D ξ i D_{\xi_i} D ξ i , D q i D_{q_i} D q i ) counts for 1, whereas D 1 = ∂ t D_1 = \partial_t D 1 = ∂ t (resp. D 1 = ∂ τ D_1 = \partial_\tau D 1 = ∂ τ ) counts for 2; thus ord ( D ξ 1 D ξ 2 ∂ t ) = 4 \operatorname{ord}(D_{\xi_1}D_{\xi_2}\partial_t) = 4 ord ( D ξ 1 D ξ 2 ∂ t ) = 4 .
Its degree s s s is its weight relative the grading deg t = 2 \deg t = 2 deg t = 2 , deg ξ i = deg η i = deg θ i = 1 \deg\xi_i = \deg\eta_i = \deg\theta_i = 1 deg ξ i = deg η i = deg θ i = 1 , the grading for which k ( 1 ∣ n ) 0 = c o ( n ) \mathfrak{k}(1|n)_0 = \mathfrak{co}(n) k ( 1∣ n ) 0 = co ( n ) .
Since deg ∂ t = − 2 \deg\partial_t = -2 deg ∂ t = − 2 whereas deg D f = − 1 \deg D_f = -1 deg D f = − 1 for odd f f f , these are one and the same integer read with opposite signs:
s = − ord . (8) s=-\operatorname{ord}. \tag*{(8)} s = − ord . (8)
We speak of the order everywhere, this being the parameter by which the ansatz is graded. We use the degree only where the convention of [14 ] is needed literally. There, an operator of degree s s s shifts the weight λ \lambda λ of the density by − s -s − s , and it is the degree that enters the weight of the target together with the charge, see eq. (11) .
A warning concerning the index k k k .
The index of a bracket is a subscript, not a superscript: in J 2 \mathcal{J}_2 J 2 , J 4 \mathcal{J}_4 J 4 , J 1 / 2 , ξ \mathcal{J}_{1/2,\xi} J 1/2 , ξ , J k , l \mathcal{J}_{k,l} J k , l , the 2 and the 4 are labels, not exponents; the second subscript, l l l , r r r , ξ \xi ξ or η \eta η , tells apart brackets of one order. The weights λ 1 , λ 2 ∈ C \lambda_1,\lambda_2 \in\mathbb{C} λ 1 , λ 2 ∈ C (and ν 1 , ν 2 \nu_1,\nu_2 ν 1 , ν 2 , if any) are those of the arguments: J k : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{J}_k : \mathcal{F}_{\lambda_1} \otimes\mathcal{F}_{\lambda_2} \longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+\operatorname{ord}} J k : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord . In the brackets J k \mathcal{J}_k J k and in the Bol operators Bol k \operatorname{Bol}_k Bol k , the letter k k k is not always the order, and it does not have the same meaning in all Sections, since in each case we keep the notation of the source:
In §2, after [29 , 43 ] , the bracket J k \mathcal{J}_k J k is of order 2 k 2k 2 k , where k ∈ 1 2 N k \in\frac{1}{2}\mathbb{N} k ∈ 2 1 N , and it shifts the weight by 2 k 2k 2 k .
In §3, after [5 ] , the supertransvectant of index k k k is likewise of order 2 k 2k 2 k , but in the normalization (23) of [5 ] it shifts the weight by k k k .
In §7, after [14 ] , the Bol operator Bol k \operatorname{Bol}_k Bol k and the brackets J k , l \mathcal{J}_{k,l} J k , l and J k , r \mathcal{J}_{k,r} J k , r built from it are of order k k k .
Wherever the order is meant, it is written ord \operatorname{ord} ord , or spelled out in words.
Conventions for k ( 1 ∣ 2 ) \mathfrak{k}(1|2) k ( 1∣2 ) , and the two realizations of the odd coordinates
We follow [14 ] throughout; we recall the formulas because everything below depends on the normalization. The contact form is α 1 = d t + ξ d η + η d ξ \alpha_1 = dt+\xi d\eta+\eta d\xi α 1 = d t + ξ d η + η d ξ , see [14 (18)] , and, with E = ξ ∂ ξ + η ∂ η E=\xi\partial_\xi+\eta\partial_\eta E = ξ ∂ ξ + η ∂ η ,
K f = ( 2 − E ) ( f ) ∂ t − H f + ∂ t ( f ) E , D ξ = η ∂ t + ∂ ξ , D η = ξ ∂ t + ∂ η , (9) K_f=(2-E)(f)\partial_t-H_f+\partial_t(f)E,\qquad D_\xi=\eta\partial_t+\partial_\xi,\qquad D_\eta=\xi\partial_t+\partial_\eta, \tag*{(9)} K f = ( 2 − E ) ( f ) ∂ t − H f + ∂ t ( f ) E , D ξ = η ∂ t + ∂ ξ , D η = ξ ∂ t + ∂ η , (9)
where H f : = − ( − 1 ) p ( f ) ( ∂ ξ ( f ) ∂ η + ∂ η ( f ) ∂ ξ ) H_f := -(-1)^{p(f)}\left(\partial_\xi(f)\partial_\eta+\partial_\eta(f)\partial_\xi\right) H f := − ( − 1 ) p ( f ) ( ∂ ξ ( f ) ∂ η + ∂ η ( f ) ∂ ξ ) is the Hamiltonian vector field of f f f , see [14 (24), (39)] , in accordance with eq. (1) . The spaces of weighted densities are F λ , ν = F α 1 λ / 2 β ν \mathcal{F}_{\lambda,\nu}=\mathcal{F}\alpha_1^{\lambda/2}\beta^\nu F λ , ν = F α 1 λ /2 β ν , where β = [ d ξ ( d η ) − 1 ] \beta=\sqrt{[d\xi(d\eta)^{-1}]} β = [ d ξ ( d η ) − 1 ] , with
L K f ( φ ) = K f ( φ ) + ( λ ∂ t ( f ) + ( − 1 ) p ( f ) ν div 2 ( f ) ) φ , div 2 = K ~ η K ~ ξ − ∂ t , (10) L_{K_f}(\varphi)=K_f(\varphi)+\left(\lambda\partial_t(f)+(-1)^{p(f)}\nu\operatorname{div}_2(f)\right)\varphi,\qquad\operatorname{div}_2=\widetilde{K}_\eta\widetilde{K}_\xi-\partial_t, \tag*{(10)} L K f ( φ ) = K f ( φ ) + ( λ ∂ t ( f ) + ( − 1 ) p ( f ) ν div 2 ( f ) ) φ , div 2 = K η K ξ − ∂ t , (10)
where
K ~ ξ ( f ) = ( − 1 ) p ( f ) ( ∂ η − ξ ∂ t ) ( f ) , K ~ η ( f ) = ( − 1 ) p ( f ) ( ∂ ξ − η ∂ t ) ( f ) , \widetilde{K}_\xi(f)=(-1)^{p(f)}(\partial_\eta-\xi\partial_t)(f),\qquad\widetilde{K}_\eta(f)=(-1)^{p(f)}(\partial_\xi-\eta\partial_t)(f), K ξ ( f ) = ( − 1 ) p ( f ) ( ∂ η − ξ ∂ t ) ( f ) , K η ( f ) = ( − 1 ) p ( f ) ( ∂ ξ − η ∂ t ) ( f ) ,
see [14 (33), (35), 2.2.1a] . With this normalization, an operator of degree s s s shifts λ \lambda λ by − s -s − s , see §2.3, so the bracket J k \mathcal{J}_k J k in sdim = 1 ∣ 1 \operatorname{sdim}=1|1 sdim = 1∣1 , which is of order ord = 2 k \operatorname{ord}=2k ord = 2 k , maps F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 k \mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+2k} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 k , as in Theorem 2.5 . Set deg t = 2 \deg t = 2 deg t = 2 and deg ξ = deg η = 1 \deg\xi= \deg\eta= 1 deg ξ = deg η = 1 , so that deg ∂ t = − 2 \deg\partial_t = -2 deg ∂ t = − 2 and deg D ξ = deg D η = − 1 \deg D_\xi= \deg D_\eta= -1 deg D ξ = deg D η = − 1 . Since k ( 1 ∣ 2 ) 0 = C K t ⊕ C K ξ η \mathfrak{k}(1|2)_0 = \mathbb{C}K_t \oplus\mathbb{C}K_{\xi\eta} k ( 1∣2 ) 0 = C K t ⊕ C K ξ η is 2-dimensional, the space of invariant operators is bigraded: besides the degree s = − ord s=-\operatorname{ord} s = − ord of §2.3, which is its weight relative K t K_t K t , every operator has a charge , its weight relative K ξ η = − ξ ∂ ξ + η ∂ η K_{\xi\eta}=-\xi\partial_\xi+\eta\partial_\eta K ξ η = − ξ ∂ ξ + η ∂ η , so that
ch D ξ = + 1 , ch D η = − 1 , ch ∂ t = 0. \operatorname{ch}D_\xi=+1,\qquad\operatorname{ch}D_\eta=-1,\qquad\operatorname{ch}\partial_t=0. ch D ξ = + 1 , ch D η = − 1 , ch ∂ t = 0.
For an operator of degree s s s and charge w w w the weight of the target is forced:
λ = λ 1 + λ 2 − s , ν = ν 1 + ν 2 − w . (11) \lambda=\lambda_1+\lambda_2-s,\qquad\nu=\nu_1+\nu_2-w. \tag*{(11)} λ = λ 1 + λ 2 − s , ν = ν 1 + ν 2 − w . (11)
We write P ⊗ Q P\otimes Q P ⊗ Q for the bilinear operator ( P ⊗ Q ) ( φ , ψ ) = ( − 1 ) p ( Q ) p ( φ ) P ( φ ) Q ( ψ ) (P\otimes Q)(\varphi,\psi)=(-1)^{p(Q)p(\varphi)}P(\varphi)Q(\psi) ( P ⊗ Q ) ( φ , ψ ) = ( − 1 ) p ( Q ) p ( φ ) P ( φ ) Q ( ψ ) ; the signs ( − 1 ) p ( φ ) (-1)^{p(\varphi)} ( − 1 ) p ( φ ) written out in the formulas below are exactly this Koszul rule. Put
P : = D ξ D η − ∂ t = 1 2 ( D ξ D η − D η D ξ ) . (12) P:=D_\xi D_\eta-\partial_t=\frac{1}{2}(D_\xi D_\eta-D_\eta D_\xi). \tag*{(12)} P := D ξ D η − ∂ t = 2 1 ( D ξ D η − D η D ξ ) . (12)
The dictionary θ ↔ ξ η \theta\leftrightarrow\xi\eta θ ↔ ξ η .
Some of the formulas below are quoted from [29 , 43 ] in the realization in which the Gram matrix of the bilinear form on the odd coordinates is diagonal, i.e., in terms of θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 . The rest are given in the realization (9), in which this matrix is anti-diagonal. Since θ 1 d θ 1 + θ 2 d θ 2 = ξ d η + η d ξ \theta_1d\theta_1+\theta_2d\theta_2=\xi d\eta+\eta d\xi θ 1 d θ 1 + θ 2 d θ 2 = ξ d η + η d ξ for
ξ = θ 1 − i θ 2 2 , η = θ 1 + i θ 2 2 , i = − 1 , (13) \xi=\frac{\theta_1-i\theta_2}{\sqrt{2}},\qquad\eta=\frac{\theta_1+i\theta_2}{\sqrt{2}},\qquad i=\sqrt{-1}, \tag*{(13)} ξ = 2 θ 1 − i θ 2 , η = 2 θ 1 + i θ 2 , i = − 1 , (13)
we have, with D θ j = ∂ θ j + θ j ∂ t D_{\theta_j}=\partial_{\theta_j}+\theta_j\partial_t D θ j = ∂ θ j + θ j ∂ t and with ν ξ η \nu^{\xi\eta} ν ξ η , resp. ν θ \nu^\theta ν θ , denoting the second weight ν \nu ν of the densities F λ , ν \mathcal{F}_{\lambda,\nu} F λ , ν in the ξ η \xi\eta ξ η , resp. θ \theta θ , coordinates,
D ξ = D θ 1 + i D θ 2 2 , D η = D θ 1 − i D θ 2 2 , ξ η = i θ 1 θ 2 , hence ν ξ η = i ν θ , (14) D_\xi=\frac{D_{\theta_1}+iD_{\theta_2}}{\sqrt{2}},\qquad D_\eta=\frac{D_{\theta_1}-iD_{\theta_2}}{\sqrt{2}},\qquad\xi\eta=i\theta_1\theta_2,\qquad\text{hence }\nu^{\xi\eta}=i\nu^\theta, \tag*{(14)} D ξ = 2 D θ 1 + i D θ 2 , D η = 2 D θ 1 − i D θ 2 , ξ η = i θ 1 θ 2 , hence ν ξ η = i ν θ , (14)
and, as an identity with no stray factor,
D ξ ⊗ D η + D η ⊗ D ξ = D θ 1 ⊗ D θ 1 + D θ 2 ⊗ D θ 2 , D ξ ⊗ D η − D η ⊗ D ξ = − i ( D θ 1 ⊗ D θ 2 − D θ 2 ⊗ D θ 1 ) , (15) \begin{aligned}
D_\xi\otimes D_\eta+D_\eta\otimes D_\xi&=D_{\theta_1}\otimes D_{\theta_1}+D_{\theta_2}\otimes D_{\theta_2},\\
D_\xi\otimes D_\eta-D_\eta\otimes D_\xi&=-i(D_{\theta_1}\otimes D_{\theta_2}-D_{\theta_2}\otimes D_{\theta_1}), \tag*{(15)}
\end{aligned} D ξ ⊗ D η + D η ⊗ D ξ D ξ ⊗ D η − D η ⊗ D ξ = D θ 1 ⊗ D θ 1 + D θ 2 ⊗ D θ 2 , = − i ( D θ 1 ⊗ D θ 2 − D θ 2 ⊗ D θ 1 ) , (15)
whereas P = − i D θ 1 D θ 2 P=-iD_{\theta_1}D_{\theta_2} P = − i D θ 1 D θ 2 for P P P as in eq. (12). Thus, the answer in the ξ η \xi\eta ξ η realization is the answer in the θ \theta θ realization written in a basis in which not only ad K t \operatorname{ad}K_t ad K t but also ad K ξ η \operatorname{ad}K_{\xi\eta} ad K ξ η acts diagonally, i.e., a basis consisting of eigenvectors of both operators. In that basis, the weights of the densities are free complex parameters and no − 1 \sqrt{-1} − 1 occurs; this is why we state the classification, Theorem 2.9, in the ξ η \xi\eta ξ η realization.
Theorem. The GRC brackets in sdim=1|1; GO
Theorem 2.5. (The GRC brackets in sdim = 1 ∣ 1 =1|1 = 1∣1 ; [29 ] ). Let dim M = 1 ∣ 1 \dim\mathcal{M}=1|1 dim M = 1∣1 with coordinates t t t (even) and θ \theta θ (odd). There are o s p ( 1 ∣ 2 ) \mathfrak{osp}(1|2) osp ( 1∣2 ) -invariant bilinear differential operators labeled by k ∈ 1 2 N k\in\frac{1}{2}\mathbb{N} k ∈ 2 1 N and a , b ∈ Z + : = { 0 , 1 , 2 , … } a,b\in\mathbb{Z}_+:=\{0,1,2,\ldots\} a , b ∈ Z + := { 0 , 1 , 2 , … }
J k ( φ , ψ ) : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 k , \mathcal{J}^k(\varphi,\psi):\mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+2k}, J k ( φ , ψ ) : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 k , more precisely:
(i) If k ∈ N k\in\mathbb{N} k ∈ N , then J k \mathcal{J}^k J k is even; it is given by the formula
J k ( φ , ψ ) = ∑ a + b = k − 1 ( − 1 ) p ( φ ) γ a , b , k − 1 D θ ( φ ( a ) ) D θ ( ψ ( b ) ) − ∑ a + b = k γ a , b , k − 1 φ ( a ) ψ ( b ) , \mathcal{J}^k(\varphi,\psi)=\sum_{a+b=k-1}(-1)^{p(\varphi)}\gamma_{a,b,k-1}D_\theta(\varphi^{(a)})D_\theta(\psi^{(b)})-\sum_{a+b=k}\gamma_{a,b,k-1}\varphi^{(a)}\psi^{(b)}, J k ( φ , ψ ) = a + b = k − 1 ∑ ( − 1 ) p ( φ ) γ a , b , k − 1 D θ ( φ ( a ) ) D θ ( ψ ( b ) ) − a + b = k ∑ γ a , b , k − 1 φ ( a ) ψ ( b ) , where D θ = ∂ θ + θ ∂ t D_\theta=\partial_\theta+\theta\partial_t D θ = ∂ θ + θ ∂ t , and (this formula will be needed for half-integer values of k k k as well in (ii))
γ a , b , k : = ( − 1 ) a + 1 ( λ 1 + ⌊ k ⌋ b ) ( λ 2 + ⌊ k ⌋ a ) . (16) \gamma_{a,b,k}:=(-1)^{a+1}\binom{\lambda_1+\lfloor k\rfloor}{b}\binom{\lambda_2+\lfloor k\rfloor}{a}. \tag*{(16)} γ a , b , k := ( − 1 ) a + 1 ( b λ 1 + ⌊ k ⌋ ) ( a λ 2 + ⌊ k ⌋ ) . (16) (ii) If k ∈ N − 1 2 k\in\mathbb{N}-\frac{1}{2} k ∈ N − 2 1 , then J k \mathcal{J}^k J k is odd; it is given by the formula
J k ( φ , ψ ) = ∑ a + b = ⌊ k ⌋ γ a , b , k ( ( − 1 ) p ( φ ) ( λ 1 + a ) φ ( a ) D θ ( ψ ( b ) ) − ( λ 2 + b ) D θ ( φ ( a ) ) ψ ( b ) ) . \mathcal{J}^{k}(\varphi,\psi)=\sum_{a+b=\lfloor k\rfloor}\gamma_{a,b,k}\left((-1)^{p(\varphi)}(\lambda_1+a)\varphi^{(a)}D_{\theta}(\psi^{(b)})-(\lambda_2+b)D_{\theta}(\varphi^{(a)})\psi^{(b)}\right). J k ( φ , ψ ) = a + b = ⌊ k ⌋ ∑ γ a , b , k ( ( − 1 ) p ( φ ) ( λ 1 + a ) φ ( a ) D θ ( ψ ( b ) ) − ( λ 2 + b ) D θ ( φ ( a ) ) ψ ( b ) ) .
There are no other non-zero non-scalar bilinear i ( o s p ( 1 ∣ 2 ) ) i(\mathfrak{osp}(1|2)) i ( osp ( 1∣2 )) -invariant differential operators.
Theorem. The GRC brackets in sdim = 1|2 found so far
Let dim M = 1 ∣ 2 \dim\mathcal{M}=1|2 dim M = 1∣2 . There are the following i ( o s p ( 2 ∣ 2 ) ) i(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) -invariant bilinear differential operators
J k ( φ , ψ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 2 k , ω . \mathcal{J}^{k}(\varphi,\psi):\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+2k,\omega}. J k ( φ , ψ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 2 k , ω .
([43 Thm. 3.1] ) If k ∈ N k\in\mathbb{N} k ∈ N and a , b ∈ Z + a,b\in\mathbb{Z}_{+} a , b ∈ Z + , then J k \mathcal{J}^{k} J k is even; it is given by the formula
J k ( φ , ψ ) = ∑ a + b = k − 1 ( − 1 ) p ( φ ) γ a , b , k − 1 ( D θ 1 ( φ ( a ) ) D θ 1 ( ψ ( b ) ) + D θ 2 ( φ ( a ) ) D θ 2 ( ψ ( b ) ) ) − ∑ a + b = k γ a , b , k − 1 φ ( a ) ψ ( b ) + ∑ a + b = k − 2 γ a , b , k − 1 ( D θ 1 D θ 2 ( φ ( a ) ) D θ 1 D θ 2 ( ψ ( b ) ) ) . \begin{aligned}
\mathcal{J}^{k}(\varphi,\psi)
={}&\sum_{a+b=k-1}(-1)^{p(\varphi)}\gamma_{a,b,k-1}\left(D_{\theta_1}(\varphi^{(a)})D_{\theta_1}(\psi^{(b)})+D_{\theta_2}(\varphi^{(a)})D_{\theta_2}(\psi^{(b)})\right)\\
&-\sum_{a+b=k}\gamma_{a,b,k-1}\varphi^{(a)}\psi^{(b)}
+\sum_{a+b=k-2}\gamma_{a,b,k-1}\left(D_{\theta_1}D_{\theta_2}(\varphi^{(a)})D_{\theta_1}D_{\theta_2}(\psi^{(b)})\right).
\end{aligned} J k ( φ , ψ ) = a + b = k − 1 ∑ ( − 1 ) p ( φ ) γ a , b , k − 1 ( D θ 1 ( φ ( a ) ) D θ 1 ( ψ ( b ) ) + D θ 2 ( φ ( a ) ) D θ 2 ( ψ ( b ) ) ) − a + b = k ∑ γ a , b , k − 1 φ ( a ) ψ ( b ) + a + b = k − 2 ∑ γ a , b , k − 1 ( D θ 1 D θ 2 ( φ ( a ) ) D θ 1 D θ 2 ( ψ ( b ) ) ) .
where D θ s = ∂ θ s + θ s ∂ t D_{\theta_s}=\partial_{\theta_s}+\theta_s\partial_t D θ s = ∂ θ s + θ s ∂ t for s = 1 , 2 s=1,2 s = 1 , 2 and where ν 1 = ν 2 = ω = 0 \nu_1=\nu_2=\omega=0 ν 1 = ν 2 = ω = 0 . (For the reason why ν 1 = ν 2 = ω = 0 \nu_1=\nu_2=\omega=0 ν 1 = ν 2 = ω = 0 has to be assumed here, and for what happens when ν 1 , ν 2 , ω ≠ 0 \nu_1,\nu_2,\omega\ne0 ν 1 , ν 2 , ω = 0 , see Corollary 2.11 : it is a property of the ansatz used, not of the operators.)
There are two odd operators of order 1; each of them is given by the formula
J k ( φ , ψ ) = ρ D θ 1 ( φ ) ψ + γ D θ 2 ( φ ) ψ + ( − 1 ) p ( φ ) ( σ φ D θ 1 ( ψ ) + χ φ D θ 2 ( ψ ) ) . \mathcal{J}^{k}(\varphi,\psi)=\rho D_{\theta_1}(\varphi)\psi+\gamma D_{\theta_2}(\varphi)\psi+(-1)^{p(\varphi)}\left(\sigma\varphi D_{\theta_1}(\psi)+\chi\varphi D_{\theta_2}(\psi)\right). J k ( φ , ψ ) = ρ D θ 1 ( φ ) ψ + γ D θ 2 ( φ ) ψ + ( − 1 ) p ( φ ) ( σ φ D θ 1 ( ψ ) + χ φ D θ 2 ( ψ ) ) .
where only the following two cases are possible and where i = − 1 i=\sqrt{-1} i = − 1 (for the same two operators written in the ξ η \xi\eta ξ η realization, where no − 1 \sqrt{-1} − 1 occurs, see Statement 2.7):
ω = − i + ν 1 + ν 2 , σ = λ 1 − i ν 1 , ρ = − λ 2 + i ν 2 , χ = − i λ 1 − ν 1 , γ = i λ 2 + ν 2 \omega=-i+\nu_1+\nu_2,\ \sigma=\lambda_1-i\nu_1,\ \rho=-\lambda_2+i\nu_2,\ \chi=-i\lambda_1-\nu_1,\ \gamma=i\lambda_2+\nu_2 ω = − i + ν 1 + ν 2 , σ = λ 1 − i ν 1 , ρ = − λ 2 + i ν 2 , χ = − i λ 1 − ν 1 , γ = i λ 2 + ν 2 ,
ω = i + ν 1 + ν 2 , σ = i λ 1 − ν 1 , ρ = − i λ 2 + ν 2 , χ = − λ 1 − i ν 1 , γ = λ 2 + i ν 2 \omega=i+\nu_1+\nu_2,\ \sigma=i\lambda_1-\nu_1,\ \rho=-i\lambda_2+\nu_2,\ \chi=-\lambda_1-i\nu_1,\ \gamma=\lambda_2+i\nu_2 ω = i + ν 1 + ν 2 , σ = i λ 1 − ν 1 , ρ = − i λ 2 + ν 2 , χ = − λ 1 − i ν 1 , γ = λ 2 + i ν 2 .
These are not all the operators: for every order there is exactly one more, see Theorem 2.9 .
Part (i) is Omri’s Theorem 3.1 in [43 ] . There the densities have one parameter, i.e., ν 1 = ν 2 = ω = 0 \nu_1=\nu_2=\omega=0 ν 1 = ν 2 = ω = 0 , and for half-integer k k k Omri proves that every o s p ( 2 ∣ 2 ) \mathfrak{osp}(2|2) osp ( 2∣2 ) -invariant bilinear operator vanishes. This agrees with part (ii): the odd operators exist only for ω = ν 1 + ν 2 ± i \omega=\nu_1+\nu_2\pm i ω = ν 1 + ν 2 ± i , so never with ν 1 = ν 2 = ω = 0 \nu_1=\nu_2=\omega=0 ν 1 = ν 2 = ω = 0 . They appear as soon as the second parameter of the weight is allowed.
This is the first time we encounter weights of densities belonging to C ∖ R \mathbb{C}\setminus\mathbb{R} C ∖ R . This happens because the passage from the normal shape of the bilinear form with the diagonal Gram matrix to the shape with an anti-diagonal Gram matrix involves − 1 \sqrt{-1} − 1 . This is nothing spectacular, actually: in the ξ η \xi\eta ξ η realization, see §2.4.1, the weights are free complex parameters and nothing is imaginary. In that realization, the two cases 1) and 2) of part (ii) become the two chiralities of one formula.
Statement. The two odd operators of order 1 in the realization
For all λ 1 , ν 1 , λ 2 , ν 2 \lambda_1,\nu_1,\lambda_2,\nu_2 λ 1 , ν 1 , λ 2 , ν 2 , the odd i ( o s p ( 2 ∣ 2 ) ) i(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) -invariant bilinear operators of order 1 form a 2-dimensional space, spanned by
J 1 / 2 , ξ = μ 2 + ( D ξ ⊗ 1 ) − μ 1 + ( 1 ⊗ D ξ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 1 , ν 1 + ν 2 − 1 , (17) \mathcal{J}_{1/2,\xi}=\mu_2^{+}(D_{\xi}\otimes1)-\mu_1^{+}(1\otimes D_{\xi}):\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+1,\nu_1+\nu_2-1}, \tag*{(17)} J 1/2 , ξ = μ 2 + ( D ξ ⊗ 1 ) − μ 1 + ( 1 ⊗ D ξ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 1 , ν 1 + ν 2 − 1 , (17)
J 1 / 2 , η = μ 2 − ( D η ⊗ 1 ) − μ 1 − ( 1 ⊗ D η ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 1 , ν 1 + ν 2 + 1 . \mathcal{J}_{1/2,\eta}=\mu_2^{-}(D_{\eta}\otimes1)-\mu_1^{-}(1\otimes D_{\eta}):\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+1,\nu_1+\nu_2+1}. J 1/2 , η = μ 2 − ( D η ⊗ 1 ) − μ 1 − ( 1 ⊗ D η ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 1 , ν 1 + ν 2 + 1 .
where μ i ± : = λ i ± ν i \mu_i^{\pm}:=\lambda_i\pm\nu_i μ i ± := λ i ± ν i ; in longhand, J 1 / 2 , ξ ( φ , ψ ) = μ 2 + D ξ ( φ ) ψ − ( − 1 ) p ( φ ) μ 1 + φ D ξ ( ψ ) \mathcal{J}_{1/2,\xi}(\varphi,\psi)=\mu_2^+D_\xi(\varphi)\psi-(-1)^{p(\varphi)}\mu_1^+\varphi D_\xi(\psi) J 1/2 , ξ ( φ , ψ ) = μ 2 + D ξ ( φ ) ψ − ( − 1 ) p ( φ ) μ 1 + φ D ξ ( ψ ) , and similarly for η \eta η . The operator J 1 / 2 , ξ \mathcal{J}_{1/2,\xi} J 1/2 , ξ vanishes identically if and only if μ 1 + = μ 2 + = 0 \mu_1^+=\mu_2^+=0 μ 1 + = μ 2 + = 0 . If exactly one of the μ i + \mu_i^+ μ i + vanishes, it degenerates into the composition of the multiplication with the odd Bol operator B ξ B_\xi B ξ of [14 Thm 1.5(ii)] acting in the other argument.
Proof. Direct computation with all eight generators of i ( o s p ( 2 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) . That formula (17) gives the same pair of operators as in part (ii) of Theorem 2.6 is seen from formulas (13)–(14): with A j : = λ j + i ν j θ = μ j + A_j:=\lambda_j+i\nu_j^\theta=\mu_j^+ A j := λ j + i ν j θ = μ j + , case 2) reads ρ = − i A 2 \rho=-iA_2 ρ = − i A 2 , γ = A 2 \gamma=A_2 γ = A 2 , σ = i A 1 \sigma=iA_1 σ = i A 1 , χ = − A 1 \chi=-A_1 χ = − A 1 , that is,
A 2 ( D θ 2 − i D θ 1 ) ( φ ) ψ − ( − 1 ) p ( φ ) A 1 φ ( D θ 2 − i D θ 1 ) ( ψ ) , A_2(D_{\theta_2}-iD_{\theta_1})(\varphi)\psi-(-1)^{p(\varphi)}A_1\varphi(D_{\theta_2}-iD_{\theta_1})(\psi), A 2 ( D θ 2 − i D θ 1 ) ( φ ) ψ − ( − 1 ) p ( φ ) A 1 φ ( D θ 2 − i D θ 1 ) ( ψ ) ,
which is J 1 / 2 , ξ \mathcal{J}_{1/2,\xi} J 1/2 , ξ up to the factor − i 2 -i\sqrt{2} − i 2 , since D θ 2 − i D θ 1 = − i 2 D ξ D_{\theta_2}-iD_{\theta_1}=-i\sqrt{2}D_\xi D θ 2 − i D θ 1 = − i 2 D ξ ; and ω θ = i + ν 1 θ + ν 2 θ \omega^\theta=i+\nu_1^\theta+\nu_2^\theta ω θ = i + ν 1 θ + ν 2 θ becomes ν = ν 1 + ν 2 − 1 \nu=\nu_1+\nu_2-1 ν = ν 1 + ν 2 − 1 after multiplication by i i i . Case 1) is J 1 / 2 , η \mathcal{J}_{1/2,\eta} J 1/2 , η . Thus, the two “cases” are the eigenvalues ∓ 1 \mp1 ∓ 1 of the charge, and the weights λ ± ν \lambda\pm\nu λ ± ν replace λ ± i ν \lambda\pm i\nu λ ± i ν .
Part (i) of Theorem 2.6 for k = 1 k=1 k = 1 likewise takes a transparent shape: by the identities (15) and γ 1 , 0 , 0 = λ 2 \gamma_{1,0,0}=\lambda_2 γ 1 , 0 , 0 = λ 2 , γ 0 , 1 , 0 = − λ 1 \gamma_{0,1,0}=-\lambda_1 γ 0 , 1 , 0 = − λ 1 , γ 0 , 0 , 0 = − 1 \gamma_{0,0,0}=-1 γ 0 , 0 , 0 = − 1 ,
− J 1 = λ 2 ( ∂ t ⊗ 1 ) − λ 1 ( 1 ⊗ ∂ t ) + D ξ ⊗ D η + D η ⊗ D ξ , (18) -\mathcal{J}_1=\lambda_2(\partial_t\otimes1)-\lambda_1(1\otimes\partial_t)+D_\xi\otimes D_\eta+D_\eta\otimes D_\xi, \tag*{(18)} − J 1 = λ 2 ( ∂ t ⊗ 1 ) − λ 1 ( 1 ⊗ ∂ t ) + D ξ ⊗ D η + D η ⊗ D ξ , (18)
and this operator is indeed invariant — for ν 1 = ν 2 = ω = 0 \nu_1=\nu_2=\omega=0 ν 1 = ν 2 = ω = 0 and all λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 .
The classification for osp(2|2)k(1|2)
The lists in [29 , 43 ] and in Theorem 2.6 are not complete. The complete answer is as follows; observe that no restriction on the weights is imposed.
Theorem. The classification for osp(2|2)k(1|2)
Theorem 2.9. The classification for o s p ( 2 ∣ 2 ) ⊂ k ( 1 ∣ 2 ) \mathfrak{osp}(2|2)\subset\mathfrak{k}(1|2) osp ( 2∣2 ) ⊂ k ( 1∣2 ) : two operators of every order, for all weights. Let n ≥ 1 n\ge1 n ≥ 1 . For all weights ( λ 1 , ν 1 ) (\lambda_1,\nu_1) ( λ 1 , ν 1 ) and ( λ 2 , ν 2 ) (\lambda_2,\nu_2) ( λ 2 , ν 2 ) , the space of i ( o s p ( 2 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) -invariant bilinear differential operators of degree s = − n s=-n s = − n , i.e., of order n n n ,
F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ , ν \mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda,\nu} F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ , ν is 2-dimensional; the two operators are interchanged by the chirality flip ξ ↔ η \xi\leftrightarrow\eta ξ ↔ η , i.e., ν i ↦ − ν i \nu_i\mapsto-\nu_i ν i ↦ − ν i , D ξ ↔ D η D_\xi\leftrightarrow D_\eta D ξ ↔ D η , P ↦ − P P\mapsto-P P ↦ − P . For n n n odd, their charges are w = + 1 w=+1 w = + 1 and w = − 1 w=-1 w = − 1 , and both operators are odd; for n n n even both have charge w = 0 w=0 w = 0 , both operators are even, and there is nothing of charge ± 2 \pm2 ± 2 . The weight of the target is given by formula (11). For n = 0 n=0 n = 0 , the only invariant operator is the multiplication.
The operators of Theorem 2.9 can be written explicitly. Set
T : = ∂ t T:=\partial_t T := ∂ t , E j : = T j E_j:=T^j E j := T j , F j : = P T j − 1 F_j:=PT^{j-1} F j := P T j − 1 ( j ≥ 1 ) (j\ge1) ( j ≥ 1 ) , X j : = D ξ T j X_j:=D_\xi T^j X j := D ξ T j , Y j : = D η T j Y_j:=D_\eta T^j Y j := D η T j ,
with P P P as in formula (12), and Λ i ± : = λ i ± ν i \Lambda_i^\pm:=\lambda_i\pm\nu_i Λ i ± := λ i ± ν i for i = 1 , 2 i=1,2 i = 1 , 2 .
Statement. The operators of Theorem 2.9, explicitly
Let the weights be generic, q ∈ Z + q\in\mathbb{Z}_+ q ∈ Z + , and let J \mathcal{J} J be an i ( o s p ( 2 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) -invariant bilinear operator F λ 1 , ν 1 ⊗ F λ 2 , ν 2 → F λ , ν \mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\to\mathcal{F}_{\lambda,\nu} F λ 1 , ν 1 ⊗ F λ 2 , ν 2 → F λ , ν .
(i) Let the order be ord = 2 q + 1 \operatorname{ord}=2q+1 ord = 2 q + 1 . For the charge w = + 1 w=+1 w = + 1 , the operator is odd and, up to a factor, it is
J q + 1 / 2 , ξ = ∑ 0 ≤ j ≤ q π j ( ( Λ 2 + + q − j ) X j ⊗ E q − j − ( Λ 1 + + j ) E j ⊗ X q − j − ( q − j ) X j ⊗ F q − j + j F j ⊗ X q − j ) , (19) \mathcal{J}_{q+1/2,\xi}=\sum_{0\le j\le q}\pi_j\left((\Lambda_2^++q-j)X_j\otimes E_{q-j}-(\Lambda_1^++j)E_j\otimes X_{q-j}-(q-j)X_j\otimes F_{q-j}+jF_j\otimes X_{q-j}\right), \tag*{(19)} J q + 1/2 , ξ = 0 ≤ j ≤ q ∑ π j ( ( Λ 2 + + q − j ) X j ⊗ E q − j − ( Λ 1 + + j ) E j ⊗ X q − j − ( q − j ) X j ⊗ F q − j + j F j ⊗ X q − j ) , (19)
where the terms with F 0 F_0 F 0 are absent, the target is λ = λ 1 + λ 2 + 2 q + 1 \lambda=\lambda_1+\lambda_2+2q+1 λ = λ 1 + λ 2 + 2 q + 1 , ν = ν 1 + ν 2 − 1 \nu=\nu_1+\nu_2-1 ν = ν 1 + ν 2 − 1 , and
π j : = ( − 1 ) j ( q j ) ∏ 0 ≤ r ≤ j − 1 ( λ 2 + q − r ) ∏ j + 1 ≤ r ≤ q ( λ 1 + r ) . (20) \pi_j:=(-1)^j\binom{q}{j}\prod_{0\le r\le j-1}(\lambda_2+q-r)\prod_{j+1\le r\le q}(\lambda_1+r). \tag*{(20)} π j := ( − 1 ) j ( j q ) 0 ≤ r ≤ j − 1 ∏ ( λ 2 + q − r ) j + 1 ≤ r ≤ q ∏ ( λ 1 + r ) . (20)
The operator J q + 1 / 2 , η \mathcal{J}_{q+1/2,\eta} J q + 1/2 , η of charge w = − 1 w=-1 w = − 1 is obtained from formula (19) by the chirality flip X j ↦ Y j X_j\mapsto Y_j X j ↦ Y j , Λ j + ↦ Λ j − \Lambda_j^+\mapsto\Lambda_j^- Λ j + ↦ Λ j − , F j ↦ − F j F_j\mapsto-F_j F j ↦ − F j ; its target is ν = ν 1 + ν 2 + 1 \nu=\nu_1+\nu_2+1 ν = ν 1 + ν 2 + 1 .
(ii) Let the order be ord = 2 q \operatorname{ord}=2q ord = 2 q , where q ≥ 1 q\ge1 q ≥ 1 . The charge is w = 0 w=0 w = 0 , the target is λ = λ 1 + λ 2 + 2 q \lambda=\lambda_1+\lambda_2+2q λ = λ 1 + λ 2 + 2 q , ν = ν 1 + ν 2 \nu=\nu_1+\nu_2 ν = ν 1 + ν 2 , and the operator is even:
J q = ∑ 0 ≤ j ≤ q A j E j ⊗ E q − j + ∑ 0 ≤ j ≤ q − 1 B j E j ⊗ F q − j + ∑ 1 ≤ j ≤ q C j F j ⊗ E q − j + ∑ 1 ≤ j ≤ q − 1 D j F j ⊗ F q − j + ∑ 0 ≤ j ≤ q − 1 ( U j X j ⊗ Y q − j − 1 + V j Y j ⊗ X q − j − 1 ) . (21) \begin{aligned}
\mathcal{J}_q={}&\sum_{0\le j\le q}A_jE_j\otimes E_{q-j}+\sum_{0\le j\le q-1}B_jE_j\otimes F_{q-j}+\sum_{1\le j\le q}C_jF_j\otimes E_{q-j}\\
&+\sum_{1\le j\le q-1}D_jF_j\otimes F_{q-j}+\sum_{0\le j\le q-1}\left(U_jX_j\otimes Y_{q-j-1}+V_jY_j\otimes X_{q-j-1}\right). \tag*{(21)}
\end{aligned} J q = 0 ≤ j ≤ q ∑ A j E j ⊗ E q − j + 0 ≤ j ≤ q − 1 ∑ B j E j ⊗ F q − j + 1 ≤ j ≤ q ∑ C j F j ⊗ E q − j + 1 ≤ j ≤ q − 1 ∑ D j F j ⊗ F q − j + 0 ≤ j ≤ q − 1 ∑ ( U j X j ⊗ Y q − j − 1 + V j Y j ⊗ X q − j − 1 ) . (21)
where U 0 U_0 U 0 and V 0 V_0 V 0 are arbitrary, which is the 2-dimensionality of Theorem 2.9, and
U j = ρ j U 0 , V j = ρ j V 0 , ρ j : = ( − 1 ) j ( q − 1 j ) ∏ 1 ≤ r ≤ j λ 2 + q − r λ 1 + r . U_j=\rho_jU_0,\qquad V_j=\rho_jV_0,\qquad\rho_j:=(-1)^j\binom{q-1}{j}\prod_{1\le r\le j}\frac{\lambda_2+q-r}{\lambda_1+r}. U j = ρ j U 0 , V j = ρ j V 0 , ρ j := ( − 1 ) j ( j q − 1 ) 1 ≤ r ≤ j ∏ λ 1 + r λ 2 + q − r .
For 0 ≤ j ≤ q − 1 0\le j\le q-1 0 ≤ j ≤ q − 1 , the remaining coefficients are
B j = ( Λ 1 − + j ) V j − ( Λ 1 + + j ) U j 2 ( λ 2 + q − j ) , B_j=\frac{(\Lambda_1^-+j)V_j-(\Lambda_1^++j)U_j}{2(\lambda_2+q-j)}, B j = 2 ( λ 2 + q − j ) ( Λ 1 − + j ) V j − ( Λ 1 + + j ) U j ,
A j = − ( Λ 1 + + j ) ( Λ 2 − + q − j ) U j + ( Λ 1 − + j ) ( Λ 2 + + q − j ) V j 2 ( q − j ) ( λ 2 + q − j ) . A_j=-\frac{(\Lambda_1^++j)(\Lambda_2^-+q-j)U_j+(\Lambda_1^-+j)(\Lambda_2^++q-j)V_j}{2(q-j)(\lambda_2+q-j)}. A j = − 2 ( q − j ) ( λ 2 + q − j ) ( Λ 1 + + j ) ( Λ 2 − + q − j ) U j + ( Λ 1 − + j ) ( Λ 2 + + q − j ) V j .
for 1 ≤ j ≤ q − 1 1\le j\le q-1 1 ≤ j ≤ q − 1 they are
D j = j ( U j + V j ) 2 ( λ 2 + q − j ) , C j = j ( ( Λ 2 − + q − j ) U j − ( Λ 2 + + q − j ) V j ) 2 ( q − j ) ( λ 2 + q − j ) , D_j=\frac{j(U_j+V_j)}{2(\lambda_2+q-j)},\qquad C_j=\frac{j\left((\Lambda_2^-+q-j)U_j-(\Lambda_2^++q-j)V_j\right)}{2(q-j)(\lambda_2+q-j)}, D j = 2 ( λ 2 + q − j ) j ( U j + V j ) , C j = 2 ( q − j ) ( λ 2 + q − j ) j ( ( Λ 2 − + q − j ) U j − ( Λ 2 + + q − j ) V j ) ,
and at the right end
C q = Λ 2 + V q − 1 − Λ 2 − U q − 1 2 ( λ 1 + q ) , A q = ( Λ 1 + + q ) Λ 2 − U q − 1 + ( Λ 1 − + q ) Λ 2 + V q − 1 2 q ( λ 1 + q ) . C_q=\frac{\Lambda_2^+V_{q-1}-\Lambda_2^-U_{q-1}}{2(\lambda_1+q)},\qquad A_q=\frac{(\Lambda_1^++q)\Lambda_2^-U_{q-1}+(\Lambda_1^-+q)\Lambda_2^+V_{q-1}}{2q(\lambda_1+q)}. C q = 2 ( λ 1 + q ) Λ 2 + V q − 1 − Λ 2 − U q − 1 , A q = 2 q ( λ 1 + q ) ( Λ 1 + + q ) Λ 2 − U q − 1 + ( Λ 1 − + q ) Λ 2 + V q − 1 .
For q = 0 q=0 q = 0 , formula (19) is J 1 / 2 , ξ \mathcal{J}_{1/2,\xi} J 1/2 , ξ of formula (17) . Verified for n ≤ 6 n\le6 n ≤ 6 by direct computation, with all four weights symbolic; for n ≤ 4 n\le4 n ≤ 4 the computation was repeated with all eight generators of i ( o s p ( 2 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(2|2)) i ( osp ( 2∣2 )) and with a much larger set of test monomials.
The operators of order 1, 2 and 3.
We write them in the basis { 1 , D ξ , D η , P } \{1,D_\xi,D_\eta,P\} { 1 , D ξ , D η , P } , with P P P as in formula (12) , for each of the two arguments. The operators of order 1 are given by formula (17) . Order 2:
J 1 , ξ = 2 ( λ 1 + 1 ) ( λ 2 + 1 ) D ξ ⊗ D η + ( λ 2 + 1 ) μ 2 − ( μ 1 + + 1 ) ∂ t ⊗ 1 − ( λ 1 + 1 ) μ 1 − ( μ 2 + + 1 ) 1 ⊗ ∂ t − ( λ 2 + 1 ) μ 2 − P ⊗ 1 − ( λ 1 + 1 ) μ 1 + 1 ⊗ P . (22) \begin{aligned}
\mathcal{J}_{1,\xi}={}&2(\lambda_1+1)(\lambda_2+1)D_\xi\otimes D_\eta+(\lambda_2+1)\mu_2^-(\mu_1^++1)\partial_t\otimes1\\
&-(\lambda_1+1)\mu_1^-(\mu_2^++1)1\otimes\partial_t-(\lambda_2+1)\mu_2^-P\otimes1-(\lambda_1+1)\mu_1^+1\otimes P. \tag*{(22)}
\end{aligned} J 1 , ξ = 2 ( λ 1 + 1 ) ( λ 2 + 1 ) D ξ ⊗ D η + ( λ 2 + 1 ) μ 2 − ( μ 1 + + 1 ) ∂ t ⊗ 1 − ( λ 1 + 1 ) μ 1 − ( μ 2 + + 1 ) 1 ⊗ ∂ t − ( λ 2 + 1 ) μ 2 − P ⊗ 1 − ( λ 1 + 1 ) μ 1 + 1 ⊗ P . (22)
and J 1 , η \mathcal{J}_{1,\eta} J 1 , η is its mirror image (D ξ ⊗ D η ↦ D η ⊗ D ξ D_\xi\otimes D_\eta\mapsto D_\eta\otimes D_\xi D ξ ⊗ D η ↦ D η ⊗ D ξ , μ ± ↦ μ ∓ \mu^\pm\mapsto\mu^\mp μ ± ↦ μ ∓ , P ↦ − P P\mapsto-P P ↦ − P ); both map F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 2 , ν 1 + ν 2 \mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+2,\nu_1+\nu_2} F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 + 2 , ν 1 + ν 2 . Order 3, charge + 1 +1 + 1 :
J 3 / 2 , ξ = ( λ 2 + 1 ) μ 2 + ∂ t D ξ ⊗ 1 + ( λ 1 + 1 ) μ 1 + 1 ⊗ ∂ t D ξ − ( λ 2 + 1 ) ( μ 1 + + 1 ) ∂ t ⊗ D ξ − ( λ 1 + 1 ) ( μ 2 + + 1 ) D ξ ⊗ ∂ t + ( λ 2 + 1 ) P ⊗ D ξ + ( λ 1 + 1 ) D ξ ⊗ P . (23) \begin{aligned}
\mathcal{J}_{3/2,\xi}={}&(\lambda_2+1)\mu_2^+\partial_tD_\xi\otimes1+(\lambda_1+1)\mu_1^+1\otimes\partial_tD_\xi-(\lambda_2+1)(\mu_1^++1)\partial_t\otimes D_\xi\\
&-(\lambda_1+1)(\mu_2^++1)D_\xi\otimes\partial_t+(\lambda_2+1)P\otimes D_\xi+(\lambda_1+1)D_\xi\otimes P. \tag*{(23)}
\end{aligned} J 3/2 , ξ = ( λ 2 + 1 ) μ 2 + ∂ t D ξ ⊗ 1 + ( λ 1 + 1 ) μ 1 + 1 ⊗ ∂ t D ξ − ( λ 2 + 1 ) ( μ 1 + + 1 ) ∂ t ⊗ D ξ − ( λ 1 + 1 ) ( μ 2 + + 1 ) D ξ ⊗ ∂ t + ( λ 2 + 1 ) P ⊗ D ξ + ( λ 1 + 1 ) D ξ ⊗ P . (23)
with target F λ 1 + λ 2 + 3 , ν 1 + ν 2 − 1 \mathcal{F}_{\lambda_1+\lambda_2+3,\nu_1+\nu_2-1} F λ 1 + λ 2 + 3 , ν 1 + ν 2 − 1 , and J 3 / 2 , η \mathcal{J}_{3/2,\eta} J 3/2 , η is its mirror image. The operators of orders 4 and 5 have the same shape, with powers of ∂ t \partial_t ∂ t inserted and with coefficients that are products of the same linear forms λ i + c \lambda_i+c λ i + c and μ i ± + c \mu_i^\pm+c μ i ± + c .
Corollary 2.11 (What happens when ν 1 ν 2 ω ≠ 0 \nu_{1}\nu_{2}\omega\ne0 ν 1 ν 2 ω = 0 ). There is something when ν 1 ν 2 ω ≠ 0 \nu_{1}\nu_{2}\omega\ne0 ν 1 ν 2 ω = 0 : the two operators of every order exist for all weights, with ω = ν 1 + ν 2 − w \omega= \nu_{1} + \nu_{2} - w ω = ν 1 + ν 2 − w , where w w w is the charge. What is special about ν 1 = ν 2 = 0 \nu_{1} = \nu_{2} = 0 ν 1 = ν 2 = 0 is the shape of the answer, not its existence. The ansatz used in part (i) of Theorem 2.6 — linear combinations of φ ( a ) ψ ( b ) \varphi^{(a)}\psi^{(b)} φ ( a ) ψ ( b ) , ∑ s D θ s ( φ ( a ) ) D θ s ( ψ ( b ) ) \sum_{s} D_{\theta_s}(\varphi^{(a)})D_{\theta_s}(\psi^{(b)}) ∑ s D θ s ( φ ( a ) ) D θ s ( ψ ( b ) ) and D θ 1 D θ 2 ( φ ( a ) ) D θ 1 D θ 2 ( ψ ( b ) ) D_{\theta_1}D_{\theta_2}(\varphi^{(a)})D_{\theta_1}D_{\theta_2}(\psi^{(b)}) D θ 1 D θ 2 ( φ ( a ) ) D θ 1 D θ 2 ( ψ ( b ) ) — contains a non-zero invariant operator if and only if ν 1 = ν 2 = 0 \nu_{1} = \nu_{2} = 0 ν 1 = ν 2 = 0 . In that case, it contains exactly the line spanned by J 1 , ξ + J 1 , η \mathcal{J}_{1,\xi} + \mathcal{J}_{1,\eta} J 1 , ξ + J 1 , η , which is the operator (18).
Proof (for order 2; the general case is the same computation). Due to the identities (15), the ansatz in the ξ η \xi\eta ξ η realization is the span of ∂ t ⊗ 1 \partial_t \otimes1 ∂ t ⊗ 1 , 1 ⊗ ∂ t 1 \otimes\partial_t 1 ⊗ ∂ t and D ξ ⊗ D η + D η ⊗ D ξ D_{\xi} \otimes D_{\eta} + D_{\eta} \otimes D_{\xi} D ξ ⊗ D η + D η ⊗ D ξ ; it contains neither P ⊗ 1 P \otimes1 P ⊗ 1 nor 1 ⊗ P 1 \otimes P 1 ⊗ P , and the two D D D -terms enter with equal coefficients. Imposing these three conditions on a J 1 , ξ + b J 1 , η a\mathcal{J}_{1,\xi} + b\mathcal{J}_{1,\eta} a J 1 , ξ + b J 1 , η gives, by formulas (21), a = b a = b a = b (the coefficients of D ξ ⊗ D η D_{\xi} \otimes D_{\eta} D ξ ⊗ D η and D η ⊗ D ξ D_{\eta} \otimes D_{\xi} D η ⊗ D ξ ), a μ 2 − = b μ 2 + a\mu_{2}^{-} = b\mu_{2}^{+} a μ 2 − = b μ 2 + (no P ⊗ 1 P \otimes1 P ⊗ 1 ) and a μ 1 + = b μ 1 − a\mu_{1}^{+} = b\mu_{1}^{-} a μ 1 + = b μ 1 − (no 1 ⊗ P 1 \otimes P 1 ⊗ P ), whence ν 1 = ν 2 = 0 \nu_{1} = \nu_{2} = 0 ν 1 = ν 2 = 0 , for λ 1 , λ 2 ≠ − 1 \lambda_{1}, \lambda_{2} \ne-1 λ 1 , λ 2 = − 1 . □
What exactly is missing from the ansatz of [Om].
To catch everything one must allow, besides the three families printed in part (i) of Theorem 2.6, also
( D θ 1 D θ 2 φ ( a ) ) ψ ( b ) , φ ( a ) ( D θ 1 D θ 2 ψ ( b ) ) , (D_{\theta_1}D_{\theta_2}\varphi^{(a)})\psi^{(b)}, \qquad\varphi^{(a)}(D_{\theta_1}D_{\theta_2}\psi^{(b)}), ( D θ 1 D θ 2 φ ( a ) ) ψ ( b ) , φ ( a ) ( D θ 1 D θ 2 ψ ( b ) ) ,
D θ 1 ( φ ( a ) ) D θ 2 ( ψ ( b ) ) − D θ 2 ( φ ( a ) ) D θ 1 ( ψ ( b ) ) . D_{\theta_1}(\varphi^{(a)})D_{\theta_2}(\psi^{(b)}) - D_{\theta_2}(\varphi^{(a)})D_{\theta_1}(\psi^{(b)}). D θ 1 ( φ ( a ) ) D θ 2 ( ψ ( b ) ) − D θ 2 ( φ ( a ) ) D θ 1 ( ψ ( b ) ) .
i.e., in the ξ η \xi\eta ξ η realization, the operators P ⊗ 1 P \otimes1 P ⊗ 1 , 1 ⊗ P 1 \otimes P 1 ⊗ P and D ξ ⊗ D η − D η ⊗ D ξ D_{\xi} \otimes D_{\eta} - D_{\eta} \otimes D_{\xi} D ξ ⊗ D η − D η ⊗ D ξ . Since k ( 1 ∣ 2 ) 0 \mathfrak{k}(1|2)_0 k ( 1∣2 ) 0 is 2-dimensional, the charge is a second grading of the space of operators, and any ansatz must be closed under it; the printed one is not. Consequently, already for ν 1 = ν 2 = 0 \nu_{1} = \nu_{2} = 0 ν 1 = ν 2 = 0 , the even invariant operators of each order form a pencil, not a line. For order 2, the missing operator is J 1 , ξ − J 1 , η \mathcal{J}_{1,\xi} - \mathcal{J}_{1,\eta} J 1 , ξ − J 1 , η , the printed J 1 \mathcal{J}_{1} J 1 being the sum J 1 , ξ + J 1 , η \mathcal{J}_{1,\xi} + \mathcal{J}_{1,\eta} J 1 , ξ + J 1 , η . Moreover, there are odd invariant operators of every odd order ord = 2 k \operatorname{ord} = 2k ord = 2 k , k ∈ 1 2 + Z + k \in\frac{1}{2} + \mathbb{Z}_{+} k ∈ 2 1 + Z + , whereas only ord = 1 \operatorname{ord} = 1 ord = 1 (i.e., k = 1 2 k = \frac{1}{2} k = 2 1 ) is listed.
On the title of this section
In [14 (4)] , the pair o s p ( n ∣ 2 m + 2 ) ⊂ k ( 2 m + 1 ∣ n ) \mathfrak{osp}(n|2m + 2) \subset\mathfrak{k}(2m + 1|n) osp ( n ∣2 m + 2 ) ⊂ k ( 2 m + 1∣ n ) is considered; for m = 0 m = 0 m = 0 this is indeed o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2) \subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) , but only n = 1 , 2 n = 1, 2 n = 1 , 2 are treated here. The case n = 3 n = 3 n = 3 is the subject of [5 ] ; we verify it and correct it in §3. For n = 4 n = 4 n = 4 , the pair is a one-parameter family o s p a ( 4 ∣ 2 ) ⊂ k ( 1 ∣ 4 ) \mathfrak{osp}_{a}(4|2) \subset\mathfrak{k}(1|4) osp a ( 4∣2 ) ⊂ k ( 1∣4 ) , see §4; the case n = 5 n = 5 n = 5 is §5, and for n = 6 n = 6 n = 6 we consider not i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) but a s ⊂ k a s \mathfrak{as} \subset\mathfrak{k}_{\mathrm{as}} as ⊂ k as , see §6.
Super GRC brackets: the o s p ( 3 ∣ 2 ) ⊂ k ( 1 ∣ 3 ) \mathfrak{osp}(3|2) \subset\mathfrak{k}(1|3) osp ( 3∣2 ) ⊂ k ( 1∣3 ) case; BLO verified and corrected
Summary of this Section
The case n = 3 n = 3 n = 3 of the pair o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2) \subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) is treated in [5 ] . We have checked it against an independent computation carried out from scratch in the conventions of [5 ] , with symbolic weights. Formulas (3.2)–(3.3) of [5 ] for N = 1 N = 1 N = 1 and formulas (3.11)–(3.12) of [5 ] for N = 2 N = 2 N = 2 are invariant as printed. The o s p ( 3 ∣ 2 ) \mathfrak{osp}(3|2) osp ( 3∣2 ) part of [5 ] , however, contains three errors, all of them in its Theorem 4.1:
(a) In the even supertransvectants, formula (4.2) of [5 ] , three of the four groups of terms carry the wrong sign;
(b) In the odd supertransvectants, formula (4.1) of [5 ] , the two sums over 1 ≤ p < ℓ ≤ 3 1 \le p < \ell\le3 1 ≤ p < ℓ ≤ 3 have lost the Levi-Civita sign;
(c) The description of the exceptional weights in items (ii)–(v) of Theorem 4.1 of [5 ] is wrong. The true set of weights at which the space of invariant operators is 2-dimensional is different from theirs. For the indices k = 1 k = 1 k = 1 and k = 2 k = 2 k = 2 , i.e., in the orders ord = 2 \operatorname{ord} = 2 ord = 2 and ord = 4 \operatorname{ord} = 4 ord = 4 , there are no exceptional weights at all. In cases (a) and (b), the corrections are stated explicitly and verified, see Statements 3.3.2 and 3.4.2; in case (c) the true answer for k ≤ 2 k \le2 k ≤ 2 is given in Statement 3.5.2. What Theorem 4.1 of [5 ] gets right is listed in §3.6.
The conventions of [5 ] , and the ansatz.
Conventions, taken from Section II of [5 ] .
Everything in this Section is computed in the conventions of [5 ] , not in ours: on R [ x , θ 1 , … , θ N ] \mathbb{R}[x,\theta_1,\ldots,\theta_N] R [ x , θ 1 , … , θ N ] ,
η ˉ i = ∂ ∂ θ i − θ i ∂ ∂ x , X F = F ∂ x − 1 2 ( − 1 ) p ( F ) ∑ 1 ≤ i ≤ N η ˉ i ( F ) η ˉ i , L X F λ = X F + λ F ′ , (24) \bar{\eta}_i=\frac{\partial}{\partial\theta_i}-\theta_i\frac{\partial}{\partial x},\qquad X_F=F\partial_x-\frac{1}{2}(-1)^{p(F)}\sum_{1\leq i\leq N}\bar{\eta}_i(F)\bar{\eta}_i,\qquad\mathcal{L}_{X_F}^{\lambda}=X_F+\lambda F', \tag*{(24)} η ˉ i = ∂ θ i ∂ − θ i ∂ x ∂ , X F = F ∂ x − 2 1 ( − 1 ) p ( F ) 1 ≤ i ≤ N ∑ η ˉ i ( F ) η ˉ i , L X F λ = X F + λ F ′ , (24)
with { F , G } \{F,G\} { F , G } their formula (2.2), with o s p ( N ∣ 2 ) = Span ( X 1 , X x , X x 2 , X θ i , X x θ i , X θ i θ j ) \mathfrak{osp}(N|2)=\operatorname{Span}(X_1,X_x,X_{x^2},X_{\theta_i},X_{x\theta_i},X_{\theta_i\theta_j}) osp ( N ∣2 ) = Span ( X 1 , X x , X x 2 , X θ i , X x θ i , X θ i θ j ) , and with
Γ i , j , k τ , λ = ( − 1 ) j ( 2 τ + [ k ] j ) ( 2 λ + [ k ] i ) (25) \Gamma_{i,j,k}^{\tau,\lambda}=(-1)^j\binom{2\tau+[k]}{j}\binom{2\lambda+[k]}{i} \tag*{(25)} Γ i , j , k τ , λ = ( − 1 ) j ( j 2 τ + [ k ] ) ( i 2 λ + [ k ] ) (25)
their formula (3.4). Their η ˉ i \bar{\eta}_i η ˉ i is thus not our D θ i D_{\theta_i} D θ i of eq. (1) , and their F τ N \mathcal{F}_\tau^N F τ N is a 1-parameter family of weighted densities; K ( N ) \mathcal{K}(N) K ( N ) denotes the k ( 1 ∣ N ) \mathfrak{k}(1|N) k ( 1∣ N ) -module of contact vector fields. A bilinear operator T : F τ N ⊗ F λ N ⟶ F τ + λ + k N T:\mathcal{F}_\tau^N\otimes\mathcal{F}_\lambda^N\longrightarrow\mathcal{F}_{\tau+\lambda+k}^N T : F τ N ⊗ F λ N ⟶ F τ + λ + k N is called invariant if
L X F μ ∘ T = T ∘ ( L X F τ ⊗ 1 + 1 ⊗ L X F λ ) \mathcal{L}_{X_F}^{\mu}\circ T=T\circ\left(\mathcal{L}_{X_F}^{\tau}\otimes1+1\otimes\mathcal{L}_{X_F}^{\lambda}\right) L X F μ ∘ T = T ∘ ( L X F τ ⊗ 1 + 1 ⊗ L X F λ )
with the sign rule taken into account, i.e.,
L X F μ ( T ( G , H ) ) = ( − 1 ) p ( F ) p ( T ) ( T ( L X F τ G , H ) + ( − 1 ) p ( F ) p ( G ) T ( G , L X F λ H ) ) \mathcal{L}_{X_F}^{\mu}(T(G,H))=(-1)^{p(F)p(T)}\left(T(\mathcal{L}_{X_F}^{\tau}G,H)+(-1)^{p(F)p(G)}T(G,\mathcal{L}_{X_F}^{\lambda}H)\right) L X F μ ( T ( G , H )) = ( − 1 ) p ( F ) p ( T ) ( T ( L X F τ G , H ) + ( − 1 ) p ( F ) p ( G ) T ( G , L X F λ H ) )
for every generating function F F F of o s p ( N ∣ 2 ) \mathfrak{osp}(N|2) osp ( N ∣2 ) ; here μ = τ + λ + k \mu=\tau+\lambda+k μ = τ + λ + k .
A dictionary between the notation of [5 ] and ours.
Put x = − t x=-t x = − t . Then, their contact form is d x + ∑ 1 ≤ i ≤ N θ i d θ i = − ( d t − ∑ 1 ≤ i ≤ N θ i d θ i ) dx+\sum_{1\leq i\leq N}\theta_i\,d\theta_i=-\left(dt-\sum_{1\leq i\leq N}\theta_i\,d\theta_i\right) d x + ∑ 1 ≤ i ≤ N θ i d θ i = − ( d t − ∑ 1 ≤ i ≤ N θ i d θ i ) , and their operators η ˉ i \bar{\eta}_i η ˉ i become our operators D θ i D_{\theta_i} D θ i of eq. (1) . Their weighted density of weight τ \tau τ is f α τ f\alpha^\tau f α τ , whereas ours of weight λ \lambda λ is f α λ / 2 f\alpha^{\lambda/2} f α λ /2 . So
x , θ i , η ˉ i of [BLO] ⟷ − t , θ i , D θ i , F τ N of [BLO] ⟷ F 2 τ , K ( N ) ≃ F − 1 N ⟷ F − 2 , supertransvectant of index k ⟷ bracket of order ord = 2 k . (26) \begin{aligned}
x,\ \theta_i,\ \bar{\eta}_i\ \text{of [BLO]}&\longleftrightarrow-t,\ \theta_i,\ D_{\theta_i},\\
\mathcal{F}_\tau^N\ \text{of [BLO]}&\longleftrightarrow\mathcal{F}_{2\tau},\qquad\mathcal{K}(N)\simeq\mathcal{F}_{-1}^N\longleftrightarrow\mathcal{F}_{-2},\\
\text{supertransvectant of index }k&\longleftrightarrow\text{ bracket of order }\operatorname{ord}=2k. \tag*{(26)}
\end{aligned} x , θ i , η ˉ i of [BLO] F τ N of [BLO] supertransvectant of index k ⟷ − t , θ i , D θ i , ⟷ F 2 τ , K ( N ) ≃ F − 1 N ⟷ F − 2 , ⟷ bracket of order ord = 2 k . (26)
For example, the exceptional pairs ( τ , λ ) (\tau,\lambda) ( τ , λ ) of Statement 3.5.2 are the pairs ( λ 1 , λ 2 ) = ( 2 τ , 2 λ ) (\lambda_1,\lambda_2)=(2\tau,2\lambda) ( λ 1 , λ 2 ) = ( 2 τ , 2 λ ) in our notation, and the involution of Remark 3.5.3 is λ i ↦ − 2 − λ i \lambda_i\mapsto-2-\lambda_i λ i ↦ − 2 − λ i .
The classification itself is done by an ansatz T = ∑ c a , A , b , B ∂ x a η ˉ A ⊗ ∂ x b η ˉ B T=\sum c_{a,A,b,B}\partial_x^a\bar{\eta}_A\otimes\partial_x^b\bar{\eta}_B T = ∑ c a , A , b , B ∂ x a η ˉ A ⊗ ∂ x b η ˉ B with constant coefficients (legitimate: invariance under X 1 X_1 X 1 and X θ i X_{\theta_i} X θ i forces it, cf. Theorem 2.1 ), followed by solving the resulting linear system over Q \mathbb{Q} Q or over Q ( τ , λ ) \mathbb{Q}(\tau,\lambda) Q ( τ , λ ) .
The even supertransvectants (4.2) of [5 ] : a sign misprint in three of the four groups.
Written out, formula (4.2) of [5 ] is the sum over the subsets A ⊆ { 1 , 2 , 3 } A\subseteq\{1,2,3\} A ⊆ { 1 , 2 , 3 }
J k τ , λ , 3 ( F , G ) = ∑ 0 ≤ r ≤ 3 ∑ ∣ A ∣ = r ∑ i + j = k − r ε r ( − 1 ) r p ( F ) Γ i , j , k − r τ , λ η ˉ A ( F ( i ) ) η ˉ A ( G ( j ) ) , (27) J_k^{\tau,\lambda,3}(F,G)=\sum_{0\leq r\leq3}\sum_{|A|=r}\sum_{i+j=k-r}\varepsilon_r(-1)^{rp(F)}\Gamma_{i,j,k-r}^{\tau,\lambda}\bar{\eta}_A(F^{(i)})\bar{\eta}_A(G^{(j)}), \tag*{(27)} J k τ , λ , 3 ( F , G ) = 0 ≤ r ≤ 3 ∑ ∣ A ∣ = r ∑ i + j = k − r ∑ ε r ( − 1 ) r p ( F ) Γ i , j , k − r τ , λ η ˉ A ( F ( i ) ) η ˉ A ( G ( j ) ) , (27)
where η ˉ A = η ˉ p 1 ⋯ η ˉ p r \bar{\eta}_A=\bar{\eta}_{p_1}\cdots\bar{\eta}_{p_r} η ˉ A = η ˉ p 1 ⋯ η ˉ p r for A = { p 1 < ⋯ < p r } A=\{p_1<\cdots<p_r\} A = { p 1 < ⋯ < p r } , and where the signs as printed are
( ε 0 , ε 1 , ε 2 , ε 3 ) = ( − , − , − , + ) . (\varepsilon_0,\varepsilon_1,\varepsilon_2,\varepsilon_3)=(-,-,-,+). ( ε 0 , ε 1 , ε 2 , ε 3 ) = ( − , − , − , + ) .
This operator is not invariant for generic ( τ , λ ) (\tau,\lambda) ( τ , λ ) and k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 .
Statement. The correct signs in eq. (4.2) of [5 ]
Statement (The correct signs in eq. (4.2) of [5 ] ). Formula (26) becomes invariant — and then, being nonzero, is the supertransvectant — precisely for
( ε 0 , ε 1 , ε 2 , ε 3 ) = ( − , + , + , − ) , i.e. ε r = − ( − 1 ) r ( r + 1 ) / 2 , (\varepsilon_0,\varepsilon_1,\varepsilon_2,\varepsilon_3)=(-,+,+,-),\qquad\text{i.e.}\qquad\varepsilon_r=-(-1)^{r(r+1)/2}, ( ε 0 , ε 1 , ε 2 , ε 3 ) = ( − , + , + , − ) , i.e. ε r = − ( − 1 ) r ( r + 1 ) /2 ,
up to an overall factor. All 16 sign patterns were tried: for k = 3 k=3 k = 3 , where all four sums over ∣ A ∣ = r |A|=r ∣ A ∣ = r , r = 0 , 1 , 2 , 3 r=0,1,2,3 r = 0 , 1 , 2 , 3 , in formula (26) are nonempty, only this pattern and its negative survive; the corrected operator was then verified invariant and nonzero for k = 1 , 2 , 3 , 4 k=1,2,3,4 k = 1 , 2 , 3 , 4 .
Two remarks make the misprint evident. First, only ϵ 0 \epsilon_0 ϵ 0 is right as printed; the three sums that carry η ˉ \bar{\eta} η ˉ ’s are all wrong. Second, the correct pattern is the pattern [5 ] themselves use in the cases N = 1 N=1 N = 1 and N = 2 N=2 N = 2 : their formula (3.3) has ( ϵ 0 , ϵ 1 ) = ( − , + ) (\epsilon_0,\epsilon_1)=(-,+) ( ϵ 0 , ϵ 1 ) = ( − , + ) and their formula (3.11) has ( ϵ 0 , ϵ 1 , ϵ 2 ) = ( − , + , + ) (\epsilon_0,\epsilon_1,\epsilon_2)=(-,+,+) ( ϵ 0 , ϵ 1 , ϵ 2 ) = ( − , + , + ) , both in agreement with ϵ r = − ( − 1 ) r ( r + 1 ) / 2 \epsilon_r= -(-1)^{r(r+1)/2} ϵ r = − ( − 1 ) r ( r + 1 ) /2 . So formula (4.2) of [5 ] is the only place where the pattern is broken.
The smallest case makes it concrete. For the index k = 1 k=1 k = 1 , i.e., in order ord = 2 \operatorname{ord}=2 ord = 2 , the unique invariant operator is, with symbolic weights,
J 1 τ , λ , 3 ( F , G ) = 2 τ F G ′ − 2 λ F ′ G + ( − 1 ) p ( F ) ∑ 1 ≤ p ≤ 3 η ˉ p ( F ) η ˉ p ( G ) , (28) J_{1}^{\tau,\lambda,3}(F,G)=2\tau FG'-2\lambda F'G+(-1)^{p(F)}\sum_{1\le p\le3}\bar{\eta}_p(F)\bar{\eta}_p(G), \tag*{(28)} J 1 τ , λ , 3 ( F , G ) = 2 τ F G ′ − 2 λ F ′ G + ( − 1 ) p ( F ) 1 ≤ p ≤ 3 ∑ η ˉ p ( F ) η ˉ p ( G ) , (28)
whereas formula (4.2) of [5 ] as printed gives the same first two terms and the term
− ( − 1 ) p ( F ) ∑ 1 ≤ p ≤ 3 η ˉ p ( F ) η ˉ p ( G ) . -(-1)^{p(F)}\sum_{1\le p\le3}\bar{\eta}_p(F)\bar{\eta}_p(G). − ( − 1 ) p ( F ) 1 ≤ p ≤ 3 ∑ η ˉ p ( F ) η ˉ p ( G ) .
Note that the r = 0 r=0 r = 0 part of formula (4.2) of [5 ] reproduces the terms 2 τ F G ′ − 2 λ F ′ G 2\tau FG'-2\lambda F'G 2 τ F G ′ − 2 λ F ′ G exactly, so the Γ \Gamma Γ ’s are right.
The odd supertransvectants (4.1) of BLO: a missing Levi-Civita sign
The correction.
Here, the situation is better and worse at the same time. On the one hand, all the coefficients of formula (4.1) of [5 ] have the right absolute value : the Γ \Gamma Γ ’s, the Λ j i = ( j + 2 ) ( j + 1 ) ( i + 1 ) \Lambda^{i}_{j}=(j+2)(j+1)(i+1) Λ j i = ( j + 2 ) ( j + 1 ) ( i + 1 ) and the index shifts i + 1 i+1 i + 1 , j + 2 j+2 j + 2 etc. are all correct. On the other hand, the two sums over 1 ≤ p < ℓ ≤ 3 1\le p<\ell\le3 1 ≤ p < ℓ ≤ 3 give one and the same coefficient to all three pairs ( p , ℓ ) (p,\ell) ( p , ℓ ) , whereas the invariant operator requires the three to alternate.
Statement. The missing Levi-Civita sign in eq. (4.1) of BLO
Statement. In the first two sums of formula (4.1) of [5 ] , the terms must be multiplied by the Levi-Civita sign, see [40 ] , i.e., by the sign of the permutation ( 1 , 2 , 3 ) ↦ ( p , ℓ , q ) (1,2,3)\mapsto(p,\ell,q) ( 1 , 2 , 3 ) ↦ ( p , ℓ , q ) of the three distinct indices:
ϵ p ℓ q , q = 6 − ℓ − p , i.e. + 1 for ( p , ℓ ) = ( 1 , 2 ) , ( 2 , 3 ) , − 1 for ( p , ℓ ) = ( 1 , 3 ) , \epsilon_{p\ell q},\qquad q=6-\ell-p,\qquad\text{i.e.}\quad+1\ \text{for }(p,\ell)=(1,2),(2,3),\qquad-1\ \text{for }(p,\ell)=(1,3), ϵ p ℓ q , q = 6 − ℓ − p , i.e. + 1 for ( p , ℓ ) = ( 1 , 2 ) , ( 2 , 3 ) , − 1 for ( p , ℓ ) = ( 1 , 3 ) ,
after which formula (4.1) — with an overall change of sign, which is only a normalization — is invariant.
The comparison was made term by term: for k = 3 2 k=\frac{3}{2} k = 2 3 and k = 5 2 k=\frac{5}{2} k = 2 5 the unique invariant operator was computed and matched against formula (4.1) of [5 ] in the same basis. At k = 3 2 k=\frac{3}{2} k = 2 3 , four of the eight coefficients differ from those of [5 ] , at k = 5 2 k=\frac{5}{2} k = 2 5 eight of the sixteen; in every case the moduli agree and only the sign is off, and exactly on the pairs ( p , ℓ ) = ( 1 , 2 ) (p,\ell)=(1,2) ( p , ℓ ) = ( 1 , 2 ) and ( 2 , 3 ) (2,3) ( 2 , 3 ) .
Explicitly, for the index k = 3 2 k=\frac{3}{2} k = 2 3 , i.e., in order ord = 3 \operatorname{ord}=3 ord = 3 , the invariant operator is, after clearing denominators,
J 3 2 τ , λ , 3 ( F , G ) = − ( − 1 ) p ( F ) τ ( τ + 1 ) ( 2 τ + 1 ) F η ˉ 1 η ˉ 2 η ˉ 3 ( G ) + ( τ + 1 ) ( 2 τ + 1 ) ( λ + 1 ) S 1 ( F , G ) + ( − 1 ) p ( F ) ( τ + 1 ) ( λ + 1 ) ( 2 λ + 1 ) S 2 ( F , G ) + λ ( λ + 1 ) ( 2 λ + 1 ) η ˉ 1 η ˉ 2 η ˉ 3 ( F ) G , (29) \begin{aligned}
J_{\frac{3}{2}}^{\tau,\lambda,3}(F,G)
={}&-(-1)^{p(F)}\tau(\tau+1)(2\tau+1)F\bar{\eta}_1\bar{\eta}_2\bar{\eta}_3(G)\\
&+(\tau+1)(2\tau+1)(\lambda+1)S_1(F,G)\\
&+(-1)^{p(F)}(\tau+1)(\lambda+1)(2\lambda+1)S_2(F,G)\\
&+\lambda(\lambda+1)(2\lambda+1)\bar{\eta}_1\bar{\eta}_2\bar{\eta}_3(F)G, \tag*{(29)}
\end{aligned} J 2 3 τ , λ , 3 ( F , G ) = − ( − 1 ) p ( F ) τ ( τ + 1 ) ( 2 τ + 1 ) F η ˉ 1 η ˉ 2 η ˉ 3 ( G ) + ( τ + 1 ) ( 2 τ + 1 ) ( λ + 1 ) S 1 ( F , G ) + ( − 1 ) p ( F ) ( τ + 1 ) ( λ + 1 ) ( 2 λ + 1 ) S 2 ( F , G ) + λ ( λ + 1 ) ( 2 λ + 1 ) η ˉ 1 η ˉ 2 η ˉ 3 ( F ) G , (29)
where
S 1 ( F , G ) : = η ˉ 1 ( F ) η ˉ 2 η ˉ 3 ( G ) − η ˉ 2 ( F ) η ˉ 1 η ˉ 3 ( G ) + η ˉ 3 ( F ) η ˉ 1 η ˉ 2 ( G ) , S_1(F,G):=\bar{\eta}_1(F)\bar{\eta}_2\bar{\eta}_3(G)-\bar{\eta}_2(F)\bar{\eta}_1\bar{\eta}_3(G)+\bar{\eta}_3(F)\bar{\eta}_1\bar{\eta}_2(G), S 1 ( F , G ) := η ˉ 1 ( F ) η ˉ 2 η ˉ 3 ( G ) − η ˉ 2 ( F ) η ˉ 1 η ˉ 3 ( G ) + η ˉ 3 ( F ) η ˉ 1 η ˉ 2 ( G ) ,
S 2 ( F , G ) : = − η ˉ 1 η ˉ 2 ( F ) η ˉ 3 ( G ) + η ˉ 1 η ˉ 3 ( F ) η ˉ 2 ( G ) − η ˉ 2 η ˉ 3 ( F ) η ˉ 1 ( G ) . S_2(F,G):=-\bar{\eta}_1\bar{\eta}_2(F)\bar{\eta}_3(G)+\bar{\eta}_1\bar{\eta}_3(F)\bar{\eta}_2(G)-\bar{\eta}_2\bar{\eta}_3(F)\bar{\eta}_1(G). S 2 ( F , G ) := − η ˉ 1 η ˉ 2 ( F ) η ˉ 3 ( G ) + η ˉ 1 η ˉ 3 ( F ) η ˉ 2 ( G ) − η ˉ 2 η ˉ 3 ( F ) η ˉ 1 ( G ) .
The signs + + + , − - − , + + + in S 1 S_1 S 1 and − - − , + + + , − - − in S 2 S_2 S 2 are ϵ p ℓ q \epsilon_{p\ell q} ϵ p ℓ q and − ϵ p ℓ q -\epsilon_{p\ell q} − ϵ p ℓ q , respectively, with ϵ p ℓ q \epsilon_{p\ell q} ϵ p ℓ q the Levi-Civita sign of Statement 3.4.2; it is absent from formula (4.1) of [5 ] .
The exceptional weights: Theorem 4.1(ii)–(v) of BLO is wrong
The exceptional weights for osp(3|2), recomputed.
Let d ( τ , λ , k ) d(\tau,\lambda,k) d ( τ , λ , k ) denote the dimension of the space of o s p ( 3 ∣ 2 ) \mathfrak{osp}(3|2) osp ( 3∣2 ) -invariant bilinear operators F τ 3 ⊗ F λ 3 ⟶ F τ + λ + k 3 \mathcal{F}^3_\tau\otimes\mathcal{F}^3_\lambda\longrightarrow\mathcal{F}^3_{\tau+\lambda+k} F τ 3 ⊗ F λ 3 ⟶ F τ + λ + k 3 , the superscript being the number N = 3 N=3 N = 3 of odd coordinates, as in F N \mathcal{F}^N F N of [5 ] ; in our notation, it is F 2 τ \mathcal{F}_{2\tau} F 2 τ , see the dictionary (26) . It was computed on the grid τ , λ ∈ { − 3 , − 5 2 , − 2 , − 3 2 , − 1 , − 1 2 , 0 , 1 2 , 1 } \tau,\lambda\in\{-3,-\frac{5}{2},-2,-\frac{3}{2},-1,-\frac{1}{2},0,\frac{1}{2},1\} τ , λ ∈ { − 3 , − 2 5 , − 2 , − 2 3 , − 1 , − 2 1 , 0 , 2 1 , 1 } together with two generic values, for k = 1 2 , 1 , 3 2 , 2 k=\frac{1}{2},1,\frac{3}{2},2 k = 2 1 , 1 , 2 3 , 2 .
Statement. The exceptional weights for osp(3|2)k(1|3)
The dimension is d = 1 d=1 d = 1 for all these weights and all k ∈ { 1 , 3 2 , 2 } k\in\{1,\frac{3}{2},2\} k ∈ { 1 , 2 3 , 2 } , except that for k = 3 2 k=\frac{3}{2} k = 2 3
d ( τ , λ , 3 2 ) = 2 exactly for ( τ , λ ) ∈ { ( − 1 , − 1 ) , ( − 1 , − 1 2 ) , ( − 1 , 0 ) , ( − 1 2 , − 1 2 ) } d\left(\tau,\lambda,\frac{3}{2}\right)=2\quad\text{exactly for}\quad(\tau,\lambda)\in\left\{(-1,-1),\left(-1,-\frac{1}{2}\right),(-1,0),\left(-\frac{1}{2},-\frac{1}{2}\right)\right\} d ( τ , λ , 2 3 ) = 2 exactly for ( τ , λ ) ∈ { ( − 1 , − 1 ) , ( − 1 , − 2 1 ) , ( − 1 , 0 ) , ( − 2 1 , − 2 1 ) }
and at the pairs ( λ , τ ) (\lambda,\tau) ( λ , τ ) obtained from these by interchanging the arguments; and d = 0 d=0 d = 0 for k = 1 2 k=\frac{1}{2} k = 2 1 . In particular, for k = 1 k=1 k = 1 and k = 2 k=2 k = 2 there are no exceptional weights at all on this grid.
Interchanging the arguments, B ↦ B ′ B\mapsto B' B ↦ B ′ with B ′ ( G , F ) : = ( − 1 ) p ( F ) p ( G ) B ( F , G ) B'(G,F):=(-1)^{p(F)p(G)}B(F,G) B ′ ( G , F ) := ( − 1 ) p ( F ) p ( G ) B ( F , G ) , is a bijection between the invariant operators F τ 3 ⊗ F λ 3 ⟶ F τ + λ + k 3 \mathcal{F}^3_\tau\otimes\mathcal{F}^3_\lambda\longrightarrow\mathcal{F}^3_{\tau+\lambda+k} F τ 3 ⊗ F λ 3 ⟶ F τ + λ + k 3 and F λ 3 ⊗ F τ 3 ⟶ F τ + λ + k 3 \mathcal{F}^3_\lambda\otimes\mathcal{F}^3_\tau\longrightarrow\mathcal{F}^3_{\tau+\lambda+k} F λ 3 ⊗ F τ 3 ⟶ F τ + λ + k 3 , so d ( τ , λ , k ) = d ( λ , τ , k ) d(\tau,\lambda,k)=d(\lambda,\tau,k) d ( τ , λ , k ) = d ( λ , τ , k ) .
Theorem 4.1(v) of [5 ] says instead that d = 2 d=2 d = 2 whenever τ = 1 − m 4 \tau=\frac{1-m}{4} τ = 4 1 − m , λ = 1 − ℓ 4 \lambda=\frac{1-\ell}{4} λ = 4 1 − ℓ with m , ℓ m,\ell m , ℓ odd and ℓ + m 2 ≤ k + 2 \frac{\ell+m}{2}\leq k+2 2 ℓ + m ≤ k + 2 . For k = 3 2 k=\frac{3}{2} k = 2 3 , this predicts d = 2 d=2 d = 2 exactly at
( τ , λ ) ∈ { ( 0 , 0 ) , ( 0 , − 1 2 ) , ( 0 , − 1 ) , ( − 1 2 , − 1 2 ) } (\tau,\lambda)\in\left\{(0,0),\left(0,-\frac{1}{2}\right),(0,-1),\left(-\frac{1}{2},-\frac{1}{2}\right)\right\} ( τ , λ ) ∈ { ( 0 , 0 ) , ( 0 , − 2 1 ) , ( 0 , − 1 ) , ( − 2 1 , − 2 1 ) }
and the interchanged pairs, and for k = 1 k=1 k = 1 and k = 2 k=2 k = 2 it predicts a nonempty list as well. So their Theorem 4.1(iv)–(v) is wrong in both directions: it claims d = 2 d=2 d = 2 at ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( 0 , − 1 2 ) \left(0,-\frac{1}{2}\right) ( 0 , − 2 1 ) , where in fact d = 1 d=1 d = 1 . It also claims uniqueness at ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) and ( − 1 , − 1 2 ) \left(-1,-\frac{1}{2}\right) ( − 1 , − 2 1 ) — these fall under their (iv), since ℓ + m 2 > k + 2 \frac{\ell+m}{2}>k+2 2 ℓ + m > k + 2 there — where in fact d = 2 d=2 d = 2 .
Both halves of this are certain, for different reasons:
A computed value d = 1 d=1 d = 1 is not an underestimate. The classifier solves a linear system built from test monomials of x x x -degree ≤ t max \leq t_{\max} ≤ t m a x (the degree of the monomial, not of the operator). Taking t max t_{\max} t m a x too small can only inflate the answer. Moreover, d ≥ 1 d\geq1 d ≥ 1 for all weights: the coefficients of the generic operator are rational functions of ( τ , λ ) (\tau,\lambda) ( τ , λ ) ; clearing the denominators and dividing by the highest common power of τ − τ 0 \tau-\tau_0 τ − τ 0 , λ − λ 0 \lambda-\lambda_0 λ − λ 0 , one gets a nonzero invariant operator at every point ( τ 0 , λ 0 ) (\tau_0,\lambda_0) ( τ 0 , λ 0 ) (equivalently: the rank of the linear system can only drop at special weights, so the dimension of its solution space can only grow). Getting 1 1 1 therefore proves d = 1 d=1 d = 1 . This kills the claim of 2-dimensionality at ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( 0 , − 1 2 ) \left(0,-\frac{1}{2}\right) ( 0 , − 2 1 ) .
The values d = 2 d=2 d = 2 were confirmed independently: for t max = 3 , 4 , 5 t_{\max}=3,4,5 t m a x = 3 , 4 , 5 , the dimension is stable, and both basis operators produced at each of the four points were fed back into the invariance test and are invariant and nonzero.
Remark. Where the slip might have entered
The set predicted by Theorem 4.1(v) of [5 ] and the true set are exchanged by the involution
τ ⟼ − 1 − τ , λ ⟼ − 1 − λ \tau\longmapsto-1-\tau,\qquad\lambda\longmapsto-1-\lambda τ ⟼ − 1 − τ , λ ⟼ − 1 − λ
which fixes ( − 1 2 , − 1 2 ) \left(-\frac{1}{2},-\frac{1}{2}\right) ( − 2 1 , − 2 1 ) and swaps ( 0 , − 1 ) ↔ ( − 1 , 0 ) (0,-1)\leftrightarrow(-1,0) ( 0 , − 1 ) ↔ ( − 1 , 0 ) — the two points where the two answers happen to agree. In the normalization (24) , one has K ( 3 ) ≅ F − 1 3 \mathcal{K}(3)\cong\mathcal{F}^3_{-1} K ( 3 ) ≅ F − 1 3 , so this involution sends F τ 3 \mathcal{F}^3_\tau F τ 3 to the weight complementary to τ \tau τ with respect to K ( 3 ) \mathcal{K}(3) K ( 3 ) . It is not the duality of the densities: the volume forms are Vol ≅ F ( 2 − N ) / 2 N \operatorname{Vol}\cong\mathcal{F}^{N}_{(2-N)/2} Vol ≅ F ( 2 − N ) /2 N (our F 2 − n \mathcal{F}_{2-n} F 2 − n , cf. §6), i.e., F − 1 / 2 3 \mathcal{F}^3_{-1/2} F − 1/2 3 for N = 3 N=3 N = 3 , so the pairing F τ 3 ⊗ F μ 3 ⟶ Vol \mathcal{F}^3_\tau\otimes\mathcal{F}^3_\mu\longrightarrow\operatorname{Vol} F τ 3 ⊗ F μ 3 ⟶ Vol exists for μ = − 1 2 − τ \mu=-\frac{1}{2}-\tau μ = − 2 1 − τ . It looks as though the parametrization τ = 1 − m 4 \tau=\frac{1-m}{4} τ = 4 1 − m , natural for N = 1 N=1 N = 1 , had been used for N = 3 N=3 N = 3 without applying the above involution τ ↦ − 1 − τ \tau\mapsto-1-\tau τ ↦ − 1 − τ , λ ↦ − 1 − λ \lambda\mapsto-1-\lambda λ ↦ − 1 − λ . Note also that the operator η ‾ 1 m − 4 η ‾ 2 η ‾ 3 \overline{\eta}_{1}^{m-4}\overline{\eta}_{2}\overline{\eta}_{3} η 1 m − 4 η 2 η 3 of their formula (4.3) requires m ≥ 5 m\geq5 m ≥ 5 , i.e. τ ≤ − 1 \tau\leq-1 τ ≤ − 1 , whereas the set I k + 2 : = { 0 , − 1 2 , − 1 , … , − ⌊ k ⌋ + 2 2 } I_{k+2}:=\left\{0,-\frac{1}{2},-1,\ldots,-\frac{\lfloor k\rfloor+2}{2}\right\} I k + 2 := { 0 , − 2 1 , − 1 , … , − 2 ⌊ k ⌋ + 2 } of their Theorem 4.1(ii) starts at τ = 0 \tau=0 τ = 0 . The classification of the exceptional weights for all k k k is an open problem.
What Theorem 4.1 of BLO gets right
For the record:
There is no operator of order ord = 1 \operatorname{ord}=1 ord = 1 , i.e., for the index k = 1 2 k=\frac{1}{2} k = 2 1 : the computation gives d = 0 d=0 d = 0 for all weights tested. The restriction 2 k ∈ N ∖ { 0 , 1 } 2k\in\mathbb{N}\setminus\{0,1\} 2 k ∈ N ∖ { 0 , 1 } in their Theorem 4.1(i) is thus correct; there is no analogue of J 1 / 2 \mathcal{J}_{1/2} J 1/2 here.
Generic uniqueness holds: d = 1 d=1 d = 1 for every k ≥ 1 k\geq1 k ≥ 1 and generic weights, in agreement with their (i)–(ii). (Unlike the case N = 2 N=2 N = 2 , where the generic dimension is 2 2 2 , see §2.8.)
The coefficients Γ i , j , k τ , λ \Gamma_{i,j,k}^{\tau,\lambda} Γ i , j , k τ , λ and Λ i j \Lambda_i^j Λ i j , and all the index shifts, are correct: the errors in formulas (4.1) and (4.2) of [5 ] are purely errors of sign.
Their N = 1 N=1 N = 1 and N = 2 N=2 N = 2 formulas (3.2)–(3.3) and (3.11)–(3.12) are correct as printed.
Super GRC and the Bol operators for the family o s p a ( 4 ∣ 2 ) ⊂ k ( 1 ∣ 4 ) \mathfrak{osp}_{a}(4|2)\subset\mathfrak{k}(1|4) osp a ( 4∣2 ) ⊂ k ( 1∣4 )
Summary of this Section
For n = 4 n=4 n = 4 , there is not one subalgebra to consider, but a one-parameter family of them. The Z \mathbb{Z} Z -graded subalgebras h ⊂ k ( 1 ∣ 4 ) \mathfrak{h}\subset\mathfrak{k}(1|4) h ⊂ k ( 1∣4 ) with h − = k ( 1 ∣ 4 ) − \mathfrak{h}_{-}=\mathfrak{k}(1|4)_{-} h − = k ( 1∣4 ) − and dim h = 17 \dim\mathfrak{h}=17 dim h = 17 form a family h ( a ) \mathfrak{h}(a) h ( a ) , where a ∈ C ∪ { ∞ } a\in\mathbb{C}\cup\{\infty\} a ∈ C ∪ { ∞ } , and h ( a ) ≃ o s p a ( 4 ∣ 2 ) \mathfrak{h}(a)\simeq\mathfrak{osp}_{a}(4|2) h ( a ) ≃ osp a ( 4∣2 ) . This is the family described by eq. (65) of [14 ] . We answer four questions about it.
The GRC brackets. For generic weights, the space of h ( a ) \mathfrak{h}(a) h ( a ) -invariant binary operators F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+\operatorname{ord}} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord is of dimension 0 , 1 , 0 , 2 0,1,0,2 0 , 1 , 0 , 2 for ord = 1 , 2 , 3 , 4 \operatorname{ord}=1,2,3,4 ord = 1 , 2 , 3 , 4 : in odd orders there is nothing at all , in contrast with n = 1 , 2 , 3 n=1,2,3 n = 1 , 2 , 3 , see Theorem 4.3.1.
Where the parameter is felt. The order-2 bracket is the same for every member of the family. It is the contact bracket, invariant under the whole of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , see Statement 4.3.2, so it says nothing about the subalgebra. The order-4 space, on the contrary, does depend on a a a : the planes obtained for a = 0 , 2 , 3 a=0,2,3 a = 0 , 2 , 3 span a 5-dimensional space, not a 2-dimensional space.
The Bol operators. They see the parameter too, and in a transparent way: for every a a a and every k ≥ 0 k\geq0 k ≥ 0 there is exactly one Bol operator, of order ord = 4 + 2 k \operatorname{ord}=4+2k ord = 4 + 2 k , namely
T 4 ∘ ∂ t k + a k + 2 ∂ t k + 2 : F − 3 − k ⟶ F 1 + k , T_4\circ\partial_t^k+\frac{a}{k+2}\partial_t^{k+2}:\mathcal{F}_{-3-k}\longrightarrow\mathcal{F}_{1+k}, T 4 ∘ ∂ t k + k + 2 a ∂ t k + 2 : F − 3 − k ⟶ F 1 + k ,
where T 4 T_4 T 4 is the operator (32) and ∘ \circ ∘ is the composition of operators: ( T 4 ∘ ∂ t k ) ( φ ) = T 4 ( ∂ t k ( φ ) ) (T_4\circ\partial_t^k)(\varphi)=T_4(\partial_t^k(\varphi)) ( T 4 ∘ ∂ t k ) ( φ ) = T 4 ( ∂ t k ( φ )) . Since ∂ t \partial_t ∂ t commutes with every D ξ i D_{\xi_i} D ξ i and D η i D_{\eta_i} D η i , the order of the two factors does not matter. There is nothing else at all, see Statement 4.4.1. This confirms [14 Thm 1.6] —obtained there by a different method—by an independent computation in the conventions of the present paper. The dictionary between the two parametrizations is a ′ = 1 − a 1 + a a'=\frac{1-a}{1+a} a ′ = 1 + a 1 − a , where a ′ a' a ′ is the parameter of eq. (65) of [14 ] , see §4.2.2. This dictionary identifies the two answers, including both degenerate members.
Isomorphic algebras, non-equivalent embeddings. In [14 §3.2.3a] , the following is asked: since o s p 1 ( 4 ∣ 2 ) ≃ o s p − 2 ( 4 ∣ 2 ) ≃ o s p − 1 / 2 ( 4 ∣ 2 ) \mathfrak{osp}_{1}(4|2)\simeq\mathfrak{osp}_{-2}(4|2)\simeq\mathfrak{osp}_{-1/2}(4|2) osp 1 ( 4∣2 ) ≃ osp − 2 ( 4∣2 ) ≃ osp − 1/2 ( 4∣2 ) , why are the answers for these values different—“Is there a mistake?”. There is no mistake, and the reason can be made precise: the group acting on the parameter of the embedding is the group Z / 2 \mathbb{Z}/2 Z /2 generated by the involution a ↦ − a a\mapsto-a a ↦ − a , which is induced by an automorphism of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , whereas the S 3 S_3 S 3 of the abstract classification is not realized by automorphisms of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) . The invariant of the embedding is a 2 a^2 a 2 , i.e., α + α − 1 \alpha+\alpha^{-1} α + α − 1 , not the invariant j ( α ) = ( α 2 + α + 1 ) 3 α 2 ( α + 1 ) 2 j(\alpha)=\frac{(\alpha^2+\alpha+1)^3}{\alpha^2(\alpha+1)^2} j ( α ) = α 2 ( α + 1 ) 2 ( α 2 + α + 1 ) 3 of the abstract algebra o s p α ( 4 ∣ 2 ) \mathfrak{osp}_{\alpha}(4|2) osp α ( 4∣2 ) , see Statement 4.2.1. In particular, o s p ( 4 ∣ 2 ) \mathfrak{osp}(4|2) osp ( 4∣2 ) itself sits in k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) in exactly two non-equivalent ways,
a = 0 a = 0 a = 0 and a = ± 3 a = \pm3 a = ± 3 , with different Bol operators, see §4.5. Thus, the Bol operators and the GRC brackets depend not only on the isomorphism class of the subalgebra but on its embedding.
The relation between 1) and 3) is the relation already met in §7. At the weights at which a Bol operator exists, it enters the binary picture as “a Bol operator acting in one of the two arguments”, i.e., as the binary operators ( φ , ψ ) ↦ B ( φ ) ψ (\varphi,\psi) \mapsto B(\varphi)\psi ( φ , ψ ) ↦ B ( φ ) ψ and ( φ , ψ ) ↦ φ B ( ψ ) (\varphi,\psi) \mapsto\varphi B(\psi) ( φ , ψ ) ↦ φB ( ψ ) , see §4.6.
Where the parameter comes from
Grade k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) by deg t = 2 \deg t = 2 deg t = 2 , deg ξ i = deg η i = 1 \deg\xi_i = \deg\eta_i = 1 deg ξ i = deg η i = 1 for all i i i , so that deg K f = deg f − 2 \deg K_f = \deg f - 2 deg K f = deg f − 2 . Then,
g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ∣ all i ⟩ , \mathfrak{g}_{-2} = \langle K_1\rangle,\qquad\mathfrak{g}_{-1} = \langle K_{y_i}\mid\text{all }i\rangle, g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ∣ all i ⟩ ,
g 0 = ⟨ K t , K y i y j ∣ all i , j ⟩ = C ⊕ o ( 4 ) , g 1 = ⟨ K t y i ⟩ ⊕ ⟨ K y i y j y k ∣ all i , j , k ⟩ . \mathfrak{g}_0 = \langle K_t, K_{y_i y_j}\mid\text{all }i,j\rangle= \mathbb{C} \oplus\mathfrak{o}(4),\qquad\mathfrak{g}_1 = \langle K_{t y_i}\rangle\oplus\langle K_{y_i y_j y_k}\mid\text{all }i,j,k\rangle. g 0 = ⟨ K t , K y i y j ∣ all i , j ⟩ = C ⊕ o ( 4 ) , g 1 = ⟨ K t y i ⟩ ⊕ ⟨ K y i y j y k ∣ all i , j , k ⟩ .
where y 1 , … , y 4 y_1,\ldots,y_4 y 1 , … , y 4 stands for ξ 1 , η 1 , ξ 2 , η 2 \xi_1,\eta_1,\xi_2,\eta_2 ξ 1 , η 1 , ξ 2 , η 2 . As o ( 4 ) \mathfrak{o}(4) o ( 4 ) -modules, g 1 = V ⊕ Λ 3 V \mathfrak{g}_1 = V \oplus\Lambda^3 V g 1 = V ⊕ Λ 3 V with V ≃ Λ 3 V V \simeq\Lambda^3 V V ≃ Λ 3 V irreducible. Therefore, the g 0 \mathfrak{g}_0 g 0 -stable complements to Λ 3 V \Lambda^3 V Λ 3 V in g 1 \mathfrak{g}_1 g 1 form a one-parameter family. They are the graphs of a σ a\sigma aσ , where σ : V ⟶ Λ 3 V \sigma: V \longrightarrow\Lambda^3 V σ : V ⟶ Λ 3 V is the equivariant map, unique up to a factor by Schur’s lemma. This is where the parameter of o s p a ( 4 ∣ 2 ) \mathfrak{osp}_a(4|2) osp a ( 4∣2 ) comes from.
Statement. The family h ( a ) ≃ o s p a ( 4 ∣ 2 ) ⊂ k ( 1 ∣ 4 ) \mathfrak{h}(a) \simeq\mathfrak{osp}_a(4|2) \subset\mathfrak{k}(1|4) h ( a ) ≃ osp a ( 4∣2 ) ⊂ k ( 1∣4 )
Put, for every a ∈ C a \in\mathbb{C} a ∈ C ,
w ξ 1 : = t ξ 1 − a ξ 1 ξ 2 η 2 , w η 1 : = t η 1 + a η 1 ξ 2 η 2 , w_{\xi_1}:=t\xi_1-a\xi_1\xi_2\eta_2,\qquad w_{\eta_1}:=t\eta_1+a\eta_1\xi_2\eta_2, w ξ 1 := t ξ 1 − a ξ 1 ξ 2 η 2 , w η 1 := t η 1 + a η 1 ξ 2 η 2 ,
w ξ 2 : = t ξ 2 − a ξ 1 η 1 ξ 2 , w η 2 : = t η 2 + a ξ 1 η 1 η 2 , w_{\xi_2}:=t\xi_2-a\xi_1\eta_1\xi_2,\qquad w_{\eta_2}:=t\eta_2+a\xi_1\eta_1\eta_2, w ξ 2 := t ξ 2 − a ξ 1 η 1 ξ 2 , w η 2 := t η 2 + a ξ 1 η 1 η 2 ,
and let h ( a ) \mathfrak{h}(a) h ( a ) be the subalgebra generated by k ( 1 ∣ 4 ) ≤ 0 \mathfrak{k}(1|4)_{\leq0} k ( 1∣4 ) ≤ 0 and by K w y i K_{w_{y_i}} K w y i for all i i i . Then:
(i) { w ξ 1 , w η 1 } = { w ξ 2 , w η 2 } = − t 2 + 2 a ξ 1 η 1 ξ 2 η 2 \{w_{\xi_1},w_{\eta_1}\}=\{w_{\xi_2},w_{\eta_2}\}=-t^2+2a\xi_1\eta_1\xi_2\eta_2 { w ξ 1 , w η 1 } = { w ξ 2 , w η 2 } = − t 2 + 2 a ξ 1 η 1 ξ 2 η 2 , all the other brackets of the w w w ’s vanish, and { w y i , { w y j , w y k } } = 0 \{w_{y_i},\{w_{y_j},w_{y_k}\}\}=0 { w y i , { w y j , w y k }} = 0 . Hence,
h ( a ) = g − 2 ⊕ g − 1 ⊕ g 0 ⊕ ⟨ K w y i ∣ for all i ⟩ ⊕ ⟨ K t 2 − 2 a ξ 1 η 1 ξ 2 η 2 ⟩ \mathfrak{h}(a)=\mathfrak{g}_{-2}\oplus\mathfrak{g}_{-1}\oplus\mathfrak{g}_0\oplus\langle K_{w_{y_i}}\mid\text{for all }i\rangle\oplus\langle K_{t^2-2a\xi_1\eta_1\xi_2\eta_2}\rangle h ( a ) = g − 2 ⊕ g − 1 ⊕ g 0 ⊕ ⟨ K w y i ∣ for all i ⟩ ⊕ ⟨ K t 2 − 2 a ξ 1 η 1 ξ 2 η 2 ⟩
is a graded subalgebra of dimension 17 for every a a a , and h ( 0 ) = i ( o s p ( 4 ∣ 2 ) ) \mathfrak{h}(0)=\mathfrak{i}(\mathfrak{osp}(4|2)) h ( 0 ) = i ( osp ( 4∣2 )) . Moreover,
[ k ( 1 ∣ 4 ) − 1 , ⟨ K w y i ⟩ ] = g 0 if and only if a ≠ ± 1. [\mathfrak{k}(1|4)_{-1},\langle K_{w_{y_i}}\rangle]=\mathfrak{g}_0\quad\text{if and only if}\quad a\ne\pm1. [ k ( 1∣4 ) − 1 , ⟨ K w y i ⟩] = g 0 if and only if a = ± 1.
(ii) h ( a ) ≃ o s p α ( 4 ∣ 2 ) \mathfrak{h}(a)\simeq\mathfrak{osp}_{\alpha}(4|2) h ( a ) ≃ osp α ( 4∣2 ) with
α = 1 + a 1 − a , j ( α ) : = ( α 2 + α + 1 ) 3 α 2 ( α + 1 ) 2 = ( a 2 + 3 ) 3 4 ( a 2 − 1 ) 2 , \alpha=\frac{1+a}{1-a},\qquad j(\alpha):=\frac{(\alpha^2+\alpha+1)^3}{\alpha^2(\alpha+1)^2}=\frac{(a^2+3)^3}{4(a^2-1)^2}, α = 1 − a 1 + a , j ( α ) := α 2 ( α + 1 ) 2 ( α 2 + α + 1 ) 3 = 4 ( a 2 − 1 ) 2 ( a 2 + 3 ) 3 ,
j j j being the invariant of the S 3 S_3 S 3 -action (29) on the parameter. In particular, h ( a ) ≃ o s p ( 4 ∣ 2 ) \mathfrak{h}(a)\simeq\mathfrak{osp}(4|2) h ( a ) ≃ osp ( 4∣2 ) (i.e., j = 27 4 j=\frac{27}{4} j = 4 27 ) if and only if a ∈ { 0 , 3 , − 3 } a\in\{0,3,-3\} a ∈ { 0 , 3 , − 3 } , i.e., if and only if α ∈ { 1 , − 2 , − 1 2 } \alpha\in\{1,-2,-\frac{1}{2}\} α ∈ { 1 , − 2 , − 2 1 } .
(iii) For a = ± 1 a=\pm1 a = ± 1 , i.e., for α = ∞ \alpha=\infty α = ∞ and α = 0 \alpha=0 α = 0 , the algebra h ( a ) \mathfrak{h}(a) h ( a ) is not simple; the subalgebra generated by g − \mathfrak{g}_{-} g − and ⟨ K w y i ⟩ \langle K_{w_{y_i}}\rangle ⟨ K w y i ⟩ is then 14-dimensional and isomorphic to p s l ( 2 ∣ 2 ) \mathfrak{psl}(2|2) psl ( 2∣2 ) , the “small N = 4 N=4 N = 4 ” subalgebra of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) . The remaining degenerate member of the family is a = ∞ a=\infty a = ∞ , i.e., h ( ∞ ) = ⟨ k ( 1 ∣ 4 ) ≤ 0 , K y i y j y k , K ξ 1 η 1 ξ 2 η 2 ⟩ \mathfrak{h}(\infty)=\langle\mathfrak{k}(1|4)_{\leq0},K_{y_i y_j y_k},K_{\xi_1\eta_1\xi_2\eta_2}\rangle h ( ∞ ) = ⟨ k ( 1∣4 ) ≤ 0 , K y i y j y k , K ξ 1 η 1 ξ 2 η 2 ⟩ , for which α = − 1 \alpha=-1 α = − 1 .
Computation. Here, o s p α ( 4 ∣ 2 ) \mathfrak{osp}_{\alpha}(4|2) osp α ( 4∣2 ) is defined by its Cartan matrix, as in [14 §3.2.3] . σ \sigma σ is found by intertwining the o ( 4 ) \mathfrak{o}(4) o ( 4 ) -action on ⟨ t y i ⟩ \langle t y_i\rangle ⟨ t y i ⟩ and on the cubics; the solution space is 1-dimensional, as Schur’s lemma predicts. Part (i) is a direct check of the contact brackets. For (ii), fix the maximal toral subalgebra ⟨ K ξ 1 η 1 , K ξ 2 η 2 ⟩ \langle K_{\xi_1\eta_1},K_{\xi_2\eta_2}\rangle ⟨ K ξ 1 η 1 , K ξ 2 η 2 ⟩ , normalize the three s l ( 2 ) \mathfrak{sl}(2) sl ( 2 ) -triples by β i ( H i ) = 2 \beta_i(H_i)=2 β i ( H i ) = 2 , and take an odd root vector X X X of weight δ = ε 1 + ε 2 + ε 3 \delta=\varepsilon_1+\varepsilon_2+\varepsilon_3 δ = ε 1 + ε 2 + ε 3 together with the opposite vector Y Y Y . Then, [ X , Y ] = ∑ c i H i [X,Y]=\sum c_iH_i [ X , Y ] = ∑ c i H i with ( c 1 , c 2 , c 3 ) = ( 1 , 1 − a 2 , 1 + a 2 ) (c_1,c_2,c_3)=(1,\frac{1-a}{2},\frac{1+a}{2}) ( c 1 , c 2 , c 3 ) = ( 1 , 2 1 − a , 2 1 + a ) , and ( ε i , ε i ) ∝ c i (\varepsilon_i,\varepsilon_i)\propto c_i ( ε i , ε i ) ∝ c i for the invariant form. Comparing this with the standard normalization − ( 1 + α ) : 1 : α -(1+\alpha):1:\alpha − ( 1 + α ) : 1 : α , we get the value of α \alpha α . The identity c 1 = c 2 + c 3 c_1=c_2+c_3 c 1 = c 2 + c 3 is the isotropy of the odd roots; at a = 0 a=0 a = 0 the recipe returns α = 1 \alpha=1 α = 1 , i.e., o s p ( 4 ∣ 2 ) \mathfrak{osp}(4|2) osp ( 4∣2 ) , as it must.
The two parametrizations.
Recall from §3.2.3 of [14 ] that the abstract algebra o s p α ( 4 ∣ 2 ) \mathfrak{osp}_{\alpha}(4|2) osp α ( 4∣2 ) is determined by the coefficients σ 1 , σ 2 , σ 3 \sigma_1,\sigma_2,\sigma_3 σ 1 , σ 2 , σ 3 with ∑ σ i = 0 \sum\sigma_i=0 ∑ σ i = 0 , with which the three projections S 2 ( g 1 ˉ ) ⟶ s l i ( 2 ) S^2(\mathfrak{g}_{\bar{1}})\longrightarrow\mathfrak{sl}_i(2) S 2 ( g 1 ˉ ) ⟶ sl i ( 2 ) enter the bracket, and that
α ⟼ − 1 − α , α ⟼ α − 1 (30) \alpha\longmapsto-1-\alpha,\qquad\alpha\longmapsto\alpha^{-1} \tag*{(30)} α ⟼ − 1 − α , α ⟼ α − 1 (30)
generate the S 3 S_3 S 3 under which the isomorphism class does not change. The parameter a a a of Statement 4.2.1 is a coordinate on the family of embeddings . The parameter of eq. (65) of [14 ] , which we denote here by a ′ a' a ′ in order not to confuse it with ours, is another coordinate, and
a = 1 − a ′ 1 + a ′ , a ′ = 1 − a 1 + a , α = 1 a ′ . (31) a=\frac{1-a'}{1+a'},\qquad a'=\frac{1-a}{1+a},\qquad\alpha=\frac{1}{a'}. \tag*{(31)} a = 1 + a ′ 1 − a ′ , a ′ = 1 + a 1 − a , α = a ′ 1 . (31)
The three degenerate members are a = ± 1 a=\pm1 a = ± 1 and a = ∞ a=\infty a = ∞ , i.e., α ∈ { ∞ , 0 , − 1 } \alpha\in\{\infty,0,-1\} α ∈ { ∞ , 0 , − 1 } : the values at which o s p α ( 4 ∣ 2 ) \mathfrak{osp}_{\alpha}(4|2) osp α ( 4∣2 ) is not simple.
The GRC brackets of the family
Everything below is computed in the realization (1) of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , with α 1 = d t + ∑ i = 1 , 2 ( ξ i d η i + η i d ξ i ) \alpha_1=dt+\sum_{i=1,2}(\xi_i d\eta_i+\eta_i d\xi_i) α 1 = d t + ∑ i = 1 , 2 ( ξ i d η i + η i d ξ i ) and with the 1-parameter weighted densities F λ = F α λ / 2 \mathcal{F}_{\lambda}=\mathcal{F}^{\lambda/2}_{\alpha} F λ = F α λ /2 , on which L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ \mathcal{L}^{\lambda}_{K_f}(\varphi)=K_f(\varphi)+\lambda\,\partial_t(f)\varphi L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ . By Theorem 2.1, an invariant operator has constant coefficients in the D f D_f D f , so the ansatz for the binary operators of order ord \operatorname{ord} ord is ∑ c ∂ t p D A ⊗ ∂ t q D B \sum c\,\partial_t^pD_A\otimes\partial_t^qD_B ∑ c ∂ t p D A ⊗ ∂ t q D B with 2 p + ∣ A ∣ + 2 q + ∣ B ∣ = ord 2p+|A|+2q+|B|=\operatorname{ord} 2 p + ∣ A ∣ + 2 q + ∣ B ∣ = ord .
Theorem. The GRC brackets for ospa(4|2)k(1|4)
Theorem 4.3.1. (The GRC brackets for o s p a ( 4 ∣ 2 ) ⊂ k ( 1 ∣ 4 ) ) . L e t \mathfrak{osp}_a(4|2)\subset\mathfrak{k}(1|4)). Let osp a ( 4∣2 ) ⊂ k ( 1∣4 )) . L e t \operatorname{ord}\leq4$. For generic weights , the space of h ( a ) \mathfrak{h}(a) h ( a ) -invariant binary differential operators F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+\operatorname{ord}} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord is of dimension
0 0 0 (ord = 1 =1 = 1 ), 1 \qquad1 1 (ord = 2 =2 = 2 ), 0 \qquad0 0 (ord = 3 =3 = 3 ), 2 \qquad2 2 (ord = 4 =4 = 4 )
for every a ∈ C a\in\mathbb{C} a ∈ C , the degenerate members a = ± 1 a=\pm1 a = ± 1 included. The order-2 operator is the same for all these members of the family, namely
J 2 = λ 2 ∂ t ⊗ 1 − λ 1 1 ⊗ ∂ t + ∑ i = 1 , 2 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) . (32) J_2=\lambda_2\partial_t\otimes1-\lambda_1 1\otimes\partial_t+\sum_{i=1,2}(D_{\xi_i}\otimes D_{\eta_i}+D_{\eta_i}\otimes D_{\xi_i}). \tag*{(32)} J 2 = λ 2 ∂ t ⊗ 1 − λ 1 1 ⊗ ∂ t + i = 1 , 2 ∑ ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) . (32)
the order-4 space is not: the three planes obtained for a = 0 , 2 , 3 a=0,2,3 a = 0 , 2 , 3 at ( λ 1 , λ 2 ) = ( 3 7 , − 5 11 ) (\lambda_1,\lambda_2)=\left(\frac{3}{7},-\frac{5}{11}\right) ( λ 1 , λ 2 ) = ( 7 3 , − 11 5 ) span a 5-dimensional space.
For the remaining degenerate member a = ∞ a=\infty a = ∞ , the dimensions are 0 , 3 , 0 , 5 0,3,0,5 0 , 3 , 0 , 5 for ord = 1 , 2 , 3 , 4 \operatorname{ord}=1,2,3,4 ord = 1 , 2 , 3 , 4 : the algebra h ( ∞ ) \mathfrak{h}(\infty) h ( ∞ ) of Statement 4.2.1(iii) contains no K t 2 K_{t^2} K t 2 , so each of the three summands ∂ t ⊗ 1 \partial_t\otimes1 ∂ t ⊗ 1 , 1 ⊗ ∂ t 1\otimes\partial_t 1 ⊗ ∂ t and ∑ i = 1 , 2 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) \sum_{i=1,2}(D_{\xi_i}\otimes D_{\eta_i}+D_{\eta_i}\otimes D_{\xi_i}) ∑ i = 1 , 2 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) of formula (31) is h ( ∞ ) \mathfrak{h}(\infty) h ( ∞ ) -invariant by itself.
Statement 4.3.2. (The first GRC bracket is the contact bracket, and it is invariant under all of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) ). The operator (31) is invariant not only under h ( a ) \mathfrak{h}(a) h ( a ) but under the whole of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , and at λ 1 = λ 2 = − 2 \lambda_1=\lambda_2=-2 λ 1 = λ 2 = − 2 it is the contact bracket. (The same holds for every k ( 1 ∣ n ) \mathfrak{k}(1|n) k ( 1∣ n ) , cf. Statement 5.4.2 for $n=5.) This is why the first GRC bracket says nothing about the subalgebra.
Remark. The two subalgebras at a=1
Remark 4.3.3. (The two subalgebras at a = ± 1 a=\pm1 a = ± 1 ). At a = ± 1 a=\pm1 a = ± 1 one should distinguish h ( ± 1 ) \mathfrak{h}(\pm1) h ( ± 1 ) , which is 17-dimensional and not simple, from the subalgebra p s l ( 2 ∣ 2 ) \mathfrak{psl}(2|2) psl ( 2∣2 ) generated by g − \mathfrak{g}_{-} g − and h 1 ( ± 1 ) \mathfrak{h}_1(\pm1) h 1 ( ± 1 ) , which is 14-dimensional. Theorem 4.3.1 concerns h ( ± 1 ) \mathfrak{h}(\pm1) h ( ± 1 ) . For the subalgebra p s l ( 2 ∣ 2 ) \mathfrak{psl}(2|2) psl ( 2∣2 ) , the dimensions are larger, as they should be for a smaller subalgebra: 0 , 4 , 0 , 5 0,4,0,5 0 , 4 , 0 , 5 for ord = 1 , 2 , 3 , 4 \operatorname{ord}=1,2,3,4 ord = 1 , 2 , 3 , 4 , and there are three unary invariant operators of order 2 at λ = − 2 \lambda=-2 λ = − 2 , whereas h ( ± 1 ) \mathfrak{h}(\pm1) h ( ± 1 ) has none.
The Bol operators of the family
Set
T 4 : = ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) = D ξ 1 D η 1 D ξ 2 D η 2 − ∂ t ( D ξ 1 D η 1 + D ξ 2 D η 2 ) + ∂ t 2 (33) T_4 := (D_{\xi_1}D_{\eta_1}-\partial_t)(D_{\xi_2}D_{\eta_2}-\partial_t)=D_{\xi_1}D_{\eta_1}D_{\xi_2}D_{\eta_2}-\partial_t(D_{\xi_1}D_{\eta_1}+D_{\xi_2}D_{\eta_2})+\partial_t^2 \tag*{(33)} T 4 := ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) = D ξ 1 D η 1 D ξ 2 D η 2 − ∂ t ( D ξ 1 D η 1 + D ξ 2 D η 2 ) + ∂ t 2 (33)
the product of the commutators D ξ i D η i − ∂ t D_{\xi_i}D_{\eta_i}-\partial_t D ξ i D η i − ∂ t over both pairs; in the algebra generated by the D ξ i D_{\xi_i} D ξ i and D η i D_{\eta_i} D η i , which is a Clifford algebra over C [ ∂ t ] \mathbb{C}[\partial_t] C [ ∂ t ] , this is the normally ordered monomial of top degree; recall that D ξ i D η i + D η i D ξ i = 2 ∂ t D_{\xi_i}D_{\eta_i}+D_{\eta_i}D_{\xi_i}=2\partial_t D ξ i D η i + D η i D ξ i = 2 ∂ t , so that D ξ i D η i − ∂ t = 1 2 [ D ξ i , D η i ] D_{\xi_i}D_{\eta_i}-\partial_t=\frac{1}{2}[D_{\xi_i},D_{\eta_i}] D ξ i D η i − ∂ t = 2 1 [ D ξ i , D η i ] is the commutator, not the anticommutator.
Statement. Bol operators for ospa(4|2)k(1|4)
For every a a a and 0 ≤ k ≤ 2 0\leq k\leq2 0 ≤ k ≤ 2 , the operator
Bol k ( a ) : = T 4 ∘ ∂ t k + a k + 2 ∂ t k + 2 : F − 3 − k ⟶ F 1 + k (34) \operatorname{Bol}^{(a)}_k:=T_4\circ\partial_t^k+\frac{a}{k+2}\partial_t^{k+2}:\mathcal{F}_{-3-k}\longrightarrow\mathcal{F}_{1+k} \tag*{(34)} Bol k ( a ) := T 4 ∘ ∂ t k + k + 2 a ∂ t k + 2 : F − 3 − k ⟶ F 1 + k (34)
is h ( a ) \mathfrak{h}(a) h ( a ) -invariant. Up to a factor, these are the only non-scalar h ( a ) \mathfrak{h}(a) h ( a ) -invariant unary operators F λ ⟶ F λ + ord \mathcal{F}_{\lambda}\longrightarrow\mathcal{F}_{\lambda+\operatorname{ord}} F λ ⟶ F λ + ord with ord ≤ 8 \operatorname{ord}\leq8 ord ≤ 8 : there are none of odd order, none of order 2, and for ord = 4 + 2 k \operatorname{ord}=4+2k ord = 4 + 2 k the space is 1-dimensional at λ = − 3 − k \lambda=-3-k λ = − 3 − k and zero at every other weight. For the degenerate member a = ∞ a=\infty a = ∞ , i.e., h ( ∞ ) \mathfrak{h}(\infty) h ( ∞ ) of Statement 4.2.1(iii), the answer is different in kind: there the invariant operators are ∂ t k : F λ ⟶ F λ + 2 k \partial_t^k:\mathcal{F}_{\lambda}\longrightarrow\mathcal{F}_{\lambda+2k} ∂ t k : F λ ⟶ F λ + 2 k for every weight λ \lambda λ , and nothing else.
Conjecture. Bol operators for ospa(4|2)k(1|4) of every order
For every k ≥ 0 k\geq0 k ≥ 0 , the operator (34) is, up to a factor, the only non-scalar h ( a ) \mathfrak{h}(a) h ( a ) -invariant unary operator of order 4 + 2 k 4+2k 4 + 2 k , and F − 3 − k \mathcal{F}_{-3-k} F − 3 − k is the only weight on which it acts. (That it is invariant for every k k k is [14 Thm 1.6] , see Subsec. 4.4.3.)
The coefficient in formula (34) was obtained with a symbolic.
Comparison with [Thm 1.6]BLS1.
What Statement 4.4.1 says is Theorem 1.6 of [14 ] , rewritten in the parametrization (30). Indeed, by eq. (42) of [14 ] the symbol [ X , Y ] + [X,Y]_+ [ X , Y ] + there means X Y − Y X XY-YX X Y − Y X , so
[ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) , and hence [ D η 1 , D ξ 1 ] + ∘ [ D η 2 , D ξ 2 ] + = 4 T 4 . [D_{\eta_i},D_{\xi_i}]_+=-2(D_{\xi_i}D_{\eta_i}-\partial_t),
\qquad\text{and hence}\qquad
[D_{\eta_1},D_{\xi_1}]_+\circ[D_{\eta_2},D_{\xi_2}]_+=4T_4. [ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) , and hence [ D η 1 , D ξ 1 ] + ∘ [ D η 2 , D ξ 2 ] + = 4 T 4 .
Dividing eq. (11) of [14 ] by 4 we get T 4 ∘ ∂ t k + 1 − a ′ ( k + 2 ) ( 1 + a ′ ) ∂ t k + 2 T_4\circ\partial_t^k+\frac{1-a'}{(k+2)(1+a')}\partial_t^{k+2} T 4 ∘ ∂ t k + ( k + 2 ) ( 1 + a ′ ) 1 − a ′ ∂ t k + 2 , which is the operator (34) precisely when a = 1 − a ′ 1 + a ′ a=\frac{1-a'}{1+a'} a = 1 + a ′ 1 − a ′ , as claimed in formula (30). Both degenerate cases agree as well: part (ii) of [14 Thm 1.6] (a ′ = − 1 a'=-1 a ′ = − 1 , i.e., a = ∞ a=\infty a = ∞ for us) is the last sentence of Statement 4.4.1. Part (iii) (a ′ = ∞ a'=\infty a ′ = ∞ , i.e., a = − 1 a=-1 a = − 1 for us) is the operator (34) with a = − 1 a=-1 a = − 1 , whose coefficient − 1 k + 2 -\frac{1}{k+2} − k + 2 1 is the limit given there. Thus, [14 Thm 1.6] is confirmed — and, on the range of orders and weights scanned, it is complete: nothing else is invariant.
Isomorphic algebras, non-equivalent embeddings
Now we can answer the question asked in [14 §3.2.3a] : since o s p 1 ( 4 ∣ 2 ) ≃ o s p − 2 ( 4 ∣ 2 ) ≃ o s p − 1 / 2 ( 4 ∣ 2 ) \mathfrak{osp}_1(4|2)\simeq\mathfrak{osp}_{-2}(4|2)\simeq\mathfrak{osp}_{-1/2}(4|2) osp 1 ( 4∣2 ) ≃ osp − 2 ( 4∣2 ) ≃ osp − 1/2 ( 4∣2 ) while the Bol operators listed for these values differ, “Is there a mistake?”.
Statement. The Z/2 acting on the family
The substitution ξ 1 ↔ η 1 \xi_1\leftrightarrow\eta_1 ξ 1 ↔ η 1 preserves the contact form α 1 \alpha_1 α 1 , hence induces an automorphism Φ \Phi Φ of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) and of every F λ \mathcal{F}_{\lambda} F λ . It maps h ( a ) \mathfrak{h}(a) h ( a ) onto h ( − a ) \mathfrak{h}(-a) h ( − a ) and Bol k ( a ) \operatorname{Bol}^{(a)}_k Bol k ( a ) onto − Bol k ( − a ) -\operatorname{Bol}^{(-a)}_k − Bol k ( − a ) . Consequently:
the family of embeddings is acted upon by the group Z / 2 \mathbb{Z}/2 Z /2 generated by a ↦ − a a\mapsto-a a ↦ − a , i.e., by α ↦ α − 1 \alpha\mapsto\alpha^{-1} α ↦ α − 1 , and only by it: the invariant of the embedding, i.e., a function of a a a taking equal values on equivalent embeddings and different ones on non-equivalent ones, is a 2 a^2 a 2 , equivalently a + a − 1 a+a^{-1} a + a − 1 , and not the invariant j ( a ) j(a) j ( a ) of the abstract algebra. In other words, h ( a ) \mathfrak{h}(a) h ( a ) is carried onto h ( b ) \mathfrak{h}(b) h ( b ) by an automorphism of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) if and only if a 2 = b 2 a^2=b^2 a 2 = b 2 , whereas h ( a ) ≃ h ( b ) \mathfrak{h}(a)\simeq\mathfrak{h}(b) h ( a ) ≃ h ( b ) as abstract
Lie superalgebras if and only if j j j takes equal values at the corresponding α \alpha α ’s, which happens for more pairs, e.g., for a = 0 a=0 a = 0 and b = 3 b=3 b = 3 ;
(ii) the three values α = 1 , − 2 , − 1 2 \alpha=1,-2,-\frac{1}{2} α = 1 , − 2 , − 2 1 of Statement 4.2.1(ii), for which the algebra is one and the same o s p ( 4 ∣ 2 ) \mathfrak{osp}(4|2) osp ( 4∣2 ) , correspond to a = 0 , 3 , − 3 a=0,3,-3 a = 0 , 3 , − 3 and hence to two non-equivalent embeddings: h ( 0 ) \mathfrak{h}(0) h ( 0 ) , which Φ \Phi Φ maps onto itself since − 0 = 0 -0=0 − 0 = 0 , and the pair h ( ± 3 ) \mathfrak{h}(\pm3) h ( ± 3 ) interchanged by Φ \Phi Φ .
So there is no mistake in [14 Thm 1.6] : the answers for α = − 2 \alpha=-2 α = − 2 and α = − 1 2 \alpha=-\frac{1}{2} α = − 2 1 are carried into each other by the automorphism Φ \Phi Φ , so they are the same answer written in two coordinate systems. In contrast, the answer for α = 1 \alpha=1 α = 1 (our a = 0 a=0 a = 0 , where the ∂ t k + 2 \partial_t^{k+2} ∂ t k + 2 -term of the operator (34) is absent) genuinely differs from both, because the embedding is genuinely different. This is the precise form of the explanation given in [14 §3.2.3a] , that “the corresponding spaces of parameters α \alpha α and a a a have non-isomorphic symmetry groups”. The group acting on the parameter of the embedding is Z / 2 \mathbb{Z}/2 Z /2 , generated by an automorphism of the ambient k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) . In contrast, the S 3 S_3 S 3 of eq. (30) permutes the three s l ( 2 ) \mathfrak{sl}(2) sl ( 2 ) ’s of o s p α ( 4 ∣ 2 ) \mathfrak{osp}_{\alpha}(4|2) osp α ( 4∣2 ) , and only two of them may be permuted here. The third s l ( 2 ) \mathfrak{sl}(2) sl ( 2 ) , ⟨ K 1 , K t , K t 2 ⟩ \langle K_1,K_t,K_{t^2}\rangle ⟨ K 1 , K t , K t 2 ⟩ , is distinguished by the grading of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , and no automorphism of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) moves it.
How the Bol operators enter the binary picture
If P : F λ 1 ⟶ F λ 1 + ord P:\mathcal{F}_{\lambda_1}\longrightarrow\mathcal{F}_{\lambda_1+\operatorname{ord}} P : F λ 1 ⟶ F λ 1 + ord is invariant, then so is the binary operator ( φ , ψ ) ↦ ( P φ ) ψ (\varphi,\psi)\mapsto(P\varphi)\psi ( φ , ψ ) ↦ ( P φ ) ψ , because the product F μ ⊗ F λ 2 ⟶ F μ + λ 2 \mathcal{F}_{\mu}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\mu+\lambda_2} F μ ⊗ F λ 2 ⟶ F μ + λ 2 is invariant; same for the second argument. This is how the Bol operators of [14 Thm 5.1] appear in §7, and it is how the parameter a a a enters the binary answer for n = 4 n=4 n = 4 .
Statement. How a Bol operator enters the binary picture
For λ 1 = − 3 − k \lambda_1=-3-k λ 1 = − 3 − k and generic λ 2 \lambda_2 λ 2 , the space of h ( a ) \mathfrak{h}(a) h ( a ) -invariant binary operators of order ord = 4 + 2 k \operatorname{ord}=4+2k ord = 4 + 2 k contains Bol k ( a ) ⊗ 1 \operatorname{Bol}^{(a)}_k\otimes1 Bol k ( a ) ⊗ 1 , and for λ 2 = − 3 − k \lambda_2=-3-k λ 2 = − 3 − k it contains 1 ⊗ Bol k ( a ) 1\otimes\operatorname{Bol}^{(a)}_k 1 ⊗ Bol k ( a ) . For k = 0 k=0 k = 0 , this was verified directly: at λ 1 = − 3 \lambda_1=-3 λ 1 = − 3 the space is still 2-dimensional, and Bol 0 ( a ) ⊗ 1 \operatorname{Bol}^{(a)}_0\otimes1 Bol 0 ( a ) ⊗ 1 lies in it, for each of a = 0 , 2 , 3 , − 3 a=0,2,3,-3 a = 0 , 2 , 3 , − 3 .
The operator Bol 0 ( a ) ⊗ 1 \operatorname{Bol}^{(a)}_0\otimes1 Bol 0 ( a ) ⊗ 1 depends on a a a , whereas ∂ t 2 ⊗ 1 \partial_t^2\otimes1 ∂ t 2 ⊗ 1 and T 4 ⊗ 1 T_4\otimes1 T 4 ⊗ 1 are separately not invariant for any a a a (Statement 4.4.1 gives no operator of order 2, and T 4 T_4 T 4 alone is invariant only for a = 0 a=0 a = 0 ). Hence, the plane at λ 1 = − 3 \lambda_1=-3 λ 1 = − 3 must depend on a a a , in agreement with the last sentence of Theorem 4.3.1. In order 2, there is no such phenomenon: the first Bol operator has order 4, and the order-2 bracket (32) is invariant under all of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) .
Open problems for n = 4 n=4 n = 4
A closed formula, in the shape of the Gordan transvectants, for the operator of an arbitrary even order ord \operatorname{ord} ord ; the exceptional weights of the binary problem for n = 4 n=4 n = 4 ; and the orders ord ≥ 5 \operatorname{ord}\ge5 ord ≥ 5 for the binary problem. The case k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) is done in §5, and the case k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) , i.e., the pair a s ⊂ e a s \mathfrak{as}\subset\mathfrak{eas} as ⊂ eas , in §6.
Super GRC brackets: the o s p ( 5 ∣ 2 ) ⊂ k ( 1 ∣ 5 ) \mathfrak{osp}(5|2)\subset\mathfrak{k}(1|5) osp ( 5∣2 ) ⊂ k ( 1∣5 ) case
Summary of this Section
This is the only case of the pair o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2)\subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) with n ≤ 6 n\le6 n ≤ 6 which was not settled above. It is the last odd n n n in our range, and this is what makes it different from its neighbors. An odd number of odd indeterminates leaves one of them, θ \theta θ , outside the hyperbolic pairs. Everything below follows from what D θ D_\theta D θ does.
The embedding is rigid. Unlike the case n = 4 n=4 n = 4 of §4, where the subalgebra is a one-parameter family, the algebra i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) sits in k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) in exactly one way. In the standard grading, g 1 = V ⊕ Λ 3 V \mathfrak{g}_1=V\oplus\Lambda^3V g 1 = V ⊕ Λ 3 V with dim V = 5 \dim V=5 dim V = 5 and dim Λ 3 V = 10 \dim\Lambda^3V=10 dim Λ 3 V = 10 , the two summands are irreducible and non-isomorphic. Hence, there is no equivariant map V ⟶ Λ 3 V V\longrightarrow\Lambda^3V V ⟶ Λ 3 V to take the graph of, see Statement 5.2.1. Among all n n n , the family occurs for n = 4 n=4 n = 4 and for no other n n n , because Λ 3 V ≃ V \Lambda^3V\simeq V Λ 3 V ≃ V requires ( n 3 ) = n \binom{n}{3}=n ( 3 n ) = n .
The Bol operators are odd. For every k ≥ 0 k \ge0 k ≥ 0 , there is exactly one invariant unary operator of order ord = 5 + 2 k \operatorname{ord} = 5 + 2k ord = 5 + 2 k , namely
T 5 ∘ ∂ θ k : F − 4 − k ⟶ F 1 + k , T 5 : = D θ ∘ T 4 = D θ ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) , T_{5} \circ\partial_{\mathord{\theta}}^{k}: \mathcal{F}_{-4-k} \longrightarrow\mathcal{F}_{1+k}, \qquad T_{5} := D_{\theta} \circ T_{4} = D_{\theta}(D_{\xi_{1}}D_{\eta_{1}} - \partial_{t})(D_{\xi_{2}}D_{\eta_{2}} - \partial_{t}), T 5 ∘ ∂ θ k : F − 4 − k ⟶ F 1 + k , T 5 := D θ ∘ T 4 = D θ ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) ,
and there is nothing at any other weight and nothing of any other order, see Theorem 5.3.1. This is [14 Thm 1.4.2] read at n = 5 n = 5 n = 5 , confirmed by an independent computation in the conventions of the present paper. What the odd n n n adds is that the operator is odd , which for even n n n cannot happen at all, see Lemma 6.3.2.
The GRC brackets. For generic weights, the space of i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) -invariant binary operators F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{F}_{\lambda_{1}} \otimes\mathcal{F}_{\lambda_{2}} \longrightarrow\mathcal{F}_{\lambda_{1}+\lambda_{2}+\operatorname{ord}} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord is of dimension 1 , 0 , 1 , 0 , 1 , 1 , 1 , 1 1, 0, 1, 0, 1, 1, 1, 1 1 , 0 , 1 , 0 , 1 , 1 , 1 , 1 for ord = 0 , … , 7 \operatorname{ord} = 0,\ldots,7 ord = 0 , … , 7 , see Theorem 5.4.1 . So the even orders carry one bracket each, unlike n = 4 n = 4 n = 4 , where order 4 4 4 carries two. The odd orders, however, are of a third kind. Unlike n = 4 n = 4 n = 4 , where they are empty, and unlike n = 3 n = 3 n = 3 , where the first of them is ord = 3 \operatorname{ord} = 3 ord = 3 , here they are empty up to ord = 3 \operatorname{ord} = 3 ord = 3 and carry one operator from ord = 5 \operatorname{ord} = 5 ord = 5 on. The dividing line is the order of T 5 T_{5} T 5 .
Where the subalgebra is felt. The order-2 2 2 bracket is the contact bracket, invariant under the whole of k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) , see Statement 5.4.2, so — exactly as for n = 4 n = 4 n = 4 — it says nothing about the subalgebra. The order-4 4 4 bracket (37) exists for all weights, its coefficients being polynomials in λ 1 \lambda_{1} λ 1 and λ 2 \lambda_{2} λ 2 , and in the orders ord ≤ 3 \operatorname{ord} \le3 ord ≤ 3 there are no exceptional weights. The order-5 5 5 bracket, on the contrary, is governed by the Bol operator: its two extreme terms are T 5 ⊗ 1 T_{5} \otimes1 T 5 ⊗ 1 and 1 ⊗ T 5 1 \otimes T_{5} 1 ⊗ T 5 , and their ratio is
− ∏ 0 ≤ i ≤ 4 ( λ 1 + i ) ∏ 0 ≤ i ≤ 4 ( λ 2 + i ) . -\frac{\prod_{0 \le i \le4}(\lambda_{1}+i)}{\prod_{0 \le i \le4}(\lambda_{2}+i)}. − ∏ 0 ≤ i ≤ 4 ( λ 2 + i ) ∏ 0 ≤ i ≤ 4 ( λ 1 + i ) .
so at the five weights λ 1 = 0 , − 1 , − 2 , − 3 , − 4 \lambda_{1} = 0,-1,-2,-3,-4 λ 1 = 0 , − 1 , − 2 , − 3 , − 4 the second block disappears, and at λ 1 = − 4 \lambda_{1} = -4 λ 1 = − 4 — the weight at which T 5 T_{5} T 5 itself is invariant — what is left is T 5 ⊗ 1 T_{5} \otimes1 T 5 ⊗ 1 , the Bol operator acting in the first argument, see §5.4.4. This is the phenomenon of §4.6 and §7, in the shape it takes for odd n n n .
The embedding is rigid: there is no family here
Grade k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) by deg t = 2 \deg t = 2 deg t = 2 , deg y i = 1 \deg y_i = 1 deg y i = 1 for all i i i , so that deg K f = deg f − 2 \deg K_f = \deg f - 2 deg K f = deg f − 2 , where y 1 , … , y 5 y_1,\ldots,y_5 y 1 , … , y 5 stands for θ , ξ 1 , η 1 , ξ 2 , η 2 \theta,\xi_1,\eta_1,\xi_2,\eta_2 θ , ξ 1 , η 1 , ξ 2 , η 2 . Then,
g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ∣ all i ⟩ , \mathfrak{g}_{-2} = \langle K_1\rangle,\qquad\mathfrak{g}_{-1} = \langle K_{y_i}\mid\text{all }i\rangle, g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ∣ all i ⟩ ,
g 0 = ⟨ K t , K y i y j ∣ all i , j ⟩ = C ⊕ o ( 5 ) , g 1 = ⟨ K t y i ⟩ ⊕ ⟨ K y i y j y k ∣ all i , j , k ⟩ . \mathfrak{g}_{0} = \langle K_t,K_{y_i y_j}\mid\text{all }i,j\rangle= \mathbb{C} \oplus\mathfrak{o}(5),\qquad\mathfrak{g}_{1} = \langle K_{t y_i}\rangle\oplus\langle K_{y_i y_j y_k}\mid\text{all }i,j,k\rangle. g 0 = ⟨ K t , K y i y j ∣ all i , j ⟩ = C ⊕ o ( 5 ) , g 1 = ⟨ K t y i ⟩ ⊕ ⟨ K y i y j y k ∣ all i , j , k ⟩ .
As o ( 5 ) \mathfrak{o}(5) o ( 5 ) -modules, g 1 = V ⊕ Λ 3 V \mathfrak{g}_{1} = V \oplus\Lambda^{3}V g 1 = V ⊕ Λ 3 V , where V V V is the tautological module. Recall that for n = 4 n = 4 n = 4 one has Λ 3 V ≃ V \Lambda^{3}V \simeq V Λ 3 V ≃ V , and that this is the whole source of the family o s p a ( 4 ∣ 2 ) \mathfrak{osp}_{a}(4|2) osp a ( 4∣2 ) of Statement 4.2.1. Namely, the g 0 \mathfrak{g}_{0} g 0 -stable complements to Λ 3 V \Lambda^{3}V Λ 3 V in g 1 \mathfrak{g}_{1} g 1 are the graphs of a σ a\sigma aσ with σ : V ⟶ Λ 3 V \sigma: V \longrightarrow\Lambda^{3}V σ : V ⟶ Λ 3 V the equivariant map, see §4.
Statement. i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) is rigid in k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 )
Statement. i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) is rigid in k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) : no family, and n = 4 n = 4 n = 4 is the only n n n that has one). For n = 5 n = 5 n = 5 , the o ( 5 ) \mathfrak{o}(5) o ( 5 ) -modules V V V and Λ 3 V \Lambda^{3}V Λ 3 V are irreducible , of dimensions 5 5 5 and 10 10 10 , and non-isomorphic ; equivalently, dim End o ( 5 ) ( g 1 ) = 2 \dim\operatorname{End}_{\mathfrak{o}(5)}(\mathfrak{g}_{1}) = 2 dim End o ( 5 ) ( g 1 ) = 2 . Hence, Hom o ( 5 ) ( V , Λ 3 V ) = 0 \operatorname{Hom}_{\mathfrak{o}(5)}(V,\Lambda^{3}V) = 0 Hom o ( 5 ) ( V , Λ 3 V ) = 0 , the module g 1 \mathfrak{g}_{1} g 1 has exactly two proper non-zero g 0 \mathfrak{g}_{0} g 0 -submodules, and the graded subalgebra h ⊂ k ( 1 ∣ 5 ) \mathfrak{h} \subset\mathfrak{k}(1|5) h ⊂ k ( 1∣5 ) with h − = g − \mathfrak{h}_{-} = \mathfrak{g}_{-} h − = g − , h 0 = g 0 \mathfrak{h}_{0} = \mathfrak{g}_{0} h 0 = g 0 and h 1 ≃ V \mathfrak{h}_{1} \simeq V h 1 ≃ V is unique: it is i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) .
:::
More generally,
dim Hom o ( n ) ( V , Λ 3 V ) = 0 , 1 , 0 , 0 and dim End o ( n ) ( g 1 ) = 2 , 4 , 2 , 3 for n = 3 , 4 , 5 , 6. \dim\operatorname{Hom}_{\mathfrak{o}(n)}(V,\Lambda^{3}V) = 0,1,0,0 \quad\text{and}\quad\dim\operatorname{End}_{\mathfrak{o}(n)}(\mathfrak{g}_{1}) = 2,4,2,3 \quad\text{for } n = 3,4,5,6. dim Hom o ( n ) ( V , Λ 3 V ) = 0 , 1 , 0 , 0 and dim End o ( n ) ( g 1 ) = 2 , 4 , 2 , 3 for n = 3 , 4 , 5 , 6.
Computation. The spaces Hom o ( n ) ( V , Λ 3 V ) \operatorname{Hom}_{\mathfrak{o}(n)}(V,\Lambda^{3}V) Hom o ( n ) ( V , Λ 3 V ) and End o ( n ) ( g 1 ) \operatorname{End}_{\mathfrak{o}(n)}(\mathfrak{g}_{1}) End o ( n ) ( g 1 ) are found by intertwining ad K y i y j \operatorname{ad}_{K_{y_i y_j}} ad K y i y j on the bases { K t y i } \{K_{t y_i}\} { K t y i } and { K y i y j y k } \{K_{y_i y_j y_k}\} { K y i y j y k } ; the answer is the dimension of the kernel of an explicit rational matrix.
The general statement needs no computation: Λ 3 V ≃ V \Lambda^{3}V \simeq V Λ 3 V ≃ V forces ( n 3 ) = n \binom{n}{3}=n ( 3 n ) = n , which for n ≥ 3 n \ge3 n ≥ 3 holds only for n = 4 n=4 n = 4 . For n = 6 n=6 n = 6 , the module Λ 3 V \Lambda^{3}V Λ 3 V is not irreducible — it splits as Λ ξ 3 ⊕ Λ η 3 \Lambda^{3}_{\xi} \oplus\Lambda^{3}_{\eta} Λ ξ 3 ⊕ Λ η 3 , which is what k a s \mathfrak{kas} kas is made of, see Statement 6.2.2 — but each summand is of dimension 10, so no family arises there either. The value dim End o ( n ) ( g 1 ) = 4 \dim\operatorname{End}_{\mathfrak{o}(n)}(\mathfrak{g}_{1})=4 dim End o ( n ) ( g 1 ) = 4 at n = 4 n=4 n = 4 is the signature of two isomorphic irreducible summands, i.e., of the family; the values 2 and 3 say that the summands are pairwise non-isomorphic. □
The unary operators: the Bol operators for n = 5 n=5 n = 5
Everything below is computed in the realization (1) of k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) , with
α 1 = d t + θ d θ + ∑ 1 ≤ i ≤ 2 ( ξ i d η i + η i d ξ i ) , F λ = F α 1 λ / 2 , L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ . \alpha_{1}=dt+\theta\,d\theta+\sum_{1\leq i\leq2}(\xi_{i}\,d\eta_{i}+\eta_{i}\,d\xi_{i}),\qquad\mathcal{F}_{\lambda}=\mathcal{F}_{\alpha_{1}^{\lambda/2}},\qquad\mathcal{L}_{K_{f}}^{\lambda}(\varphi)=K_{f}(\varphi)+\lambda\,\partial_{t}(f)\,\varphi. α 1 = d t + θ d θ + 1 ≤ i ≤ 2 ∑ ( ξ i d η i + η i d ξ i ) , F λ = F α 1 λ /2 , L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ .
Let D i : = D y i D_{i}:=D_{y_{i}} D i := D y i , where y 1 , … , y 5 y_{1},\ldots,y_{5} y 1 , … , y 5 stands for θ , ξ 1 , η 1 , ξ 2 , η 2 \theta,\xi_{1},\eta_{1},\xi_{2},\eta_{2} θ , ξ 1 , η 1 , ξ 2 , η 2 , and let B = ( B i j ) B=(B_{ij}) B = ( B ij ) be the Gram matrix of the odd part of α 1 \alpha_{1} α 1 in this basis, i.e., B θ θ = 1 B_{\theta\theta}=1 B θ θ = 1 , B ξ i η i = B η i ξ i = 1 B_{\xi_{i}\eta_{i}}=B_{\eta_{i}\xi_{i}}=1 B ξ i η i = B η i ξ i = 1 , and all other entries are 0. Then, D i D j + D j D i = 2 B i j ∂ t D_{i}D_{j}+D_{j}D_{i}=2B_{ij}\partial_{t} D i D j + D j D i = 2 B ij ∂ t , so D θ 2 = ∂ t D_{\theta}^{2}=\partial_{t} D θ 2 = ∂ t and D ξ i D η i + D η i D ξ i = 2 ∂ t D_{\xi_{i}}D_{\eta_{i}}+D_{\eta_{i}}D_{\xi_{i}}=2\partial_{t} D ξ i D η i + D η i D ξ i = 2 ∂ t . Put
T 5 : = D θ ∘ T 4 = D θ ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) , (35) T_{5}:=D_{\theta}\circ T_{4}=D_{\theta}(D_{\xi_{1}}D_{\eta_{1}}-\partial_{t})(D_{\xi_{2}}D_{\eta_{2}}-\partial_{t}), \tag*{(35)} T 5 := D θ ∘ T 4 = D θ ( D ξ 1 D η 1 − ∂ t ) ( D ξ 2 D η 2 − ∂ t ) , (35)
where T 4 T_{4} T 4 is the operator (32). The operator T 5 T_{5} T 5 is odd, and its order is 5.
Theorem. The unary i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) -invariant operators
Theorem (The unary i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) -invariant operators on the densities are the T 5 ∘ ∂ t k T_{5}\circ\partial_{t}^{k} T 5 ∘ ∂ t k , and they are odd). For every k ≥ 0 k\geq0 k ≥ 0 , the composition
T 5 ∘ ∂ t k : F − 4 − k ⟶ F 1 + k , ord = 5 + 2 k , T_{5}\circ\partial_{t}^{k}:\mathcal{F}_{-4-k}\longrightarrow\mathcal{F}_{1+k},\qquad\operatorname{ord}=5+2k, T 5 ∘ ∂ t k : F − 4 − k ⟶ F 1 + k , ord = 5 + 2 k ,
of ∂ t k \partial_{t}^{k} ∂ t k and the operator (34) is i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) -invariant, and it is invariant at no other weight. There is nothing else: no non-scalar invariant unary operator of any other order, and none at any other weight. Every such operator is odd.
In the order ord = 5 \operatorname{ord}=5 ord = 5 , the operator is
∂ t 2 D θ − ∂ t D θ D ξ 1 D η 1 − ∂ t D θ D ξ 2 D η 2 + D θ D ξ 1 D η 1 D ξ 2 D η 2 , \partial_{t}^{2}D_{\theta}-\partial_{t}D_{\theta}D_{\xi_{1}}D_{\eta_{1}}-\partial_{t}D_{\theta}D_{\xi_{2}}D_{\eta_{2}}+D_{\theta}D_{\xi_{1}}D_{\eta_{1}}D_{\xi_{2}}D_{\eta_{2}}, ∂ t 2 D θ − ∂ t D θ D ξ 1 D η 1 − ∂ t D θ D ξ 2 D η 2 + D θ D ξ 1 D η 1 D ξ 2 D η 2 ,
which is the operator (34) expanded.
Comparison with [Thm 1.4.2]BLS1.
The weights there are λ = 1 − k − n \lambda=1-k-n λ = 1 − k − n and μ = k + 1 \mu=k+1 μ = k + 1 , which at n = 5 n=5 n = 5 are ours. The parity is the point that the odd n n n contributes: by eq. (42) of [14 ] the operator is built of the operators [ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) [D_{\eta_{i}},D_{\xi_{i}}]_{+}=-2(D_{\xi_{i}}D_{\eta_{i}}-\partial_{t}) [ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) of formula (3), one for each hyperbolic pair. For odd n n n , there is one factor left over, D θ D_{\theta} D θ , of odd parity. So, for odd n n n , in particular for n = 5 n=5 n = 5 , the Bol operators are odd and of odd order, whereas the same count gives even operators of even order for even n n n , cf. Lemma 6.3.2 for n = 6 n=6 n = 6 .
The GRC brackets
By Theorem 2.1, an invariant operator has constant coefficients in the D f D_{f} D f , so the ansatz for the binary operators of order ord \operatorname{ord} ord is ∑ c ∂ t p D A ⊗ ∂ t q D B \sum c\,\partial_{t}^{p}D_{A}\otimes\partial_{t}^{q}D_{B} ∑ c ∂ t p D A ⊗ ∂ t q D B with 2 p + ∣ A ∣ + 2 q + ∣ B ∣ = ord 2p+|A|+2q+|B|=\operatorname{ord} 2 p + ∣ A ∣ + 2 q + ∣ B ∣ = ord , exactly as in §4.3. Write
Π : = D θ ⊗ D θ + ∑ 1 ≤ i ≤ 2 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) (36) \Pi:=D_{\theta}\otimes D_{\theta}+\sum_{1\leq i\leq2}(D_{\xi_{i}}\otimes D_{\eta_{i}}+D_{\eta_{i}}\otimes D_{\xi_{i}}) \tag*{(36)} Π := D θ ⊗ D θ + 1 ≤ i ≤ 2 ∑ ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) (36)
for the contraction with the inverse Gram matrix, and let Π l \Pi_{l} Π l , resp. Π r \Pi_{r} Π r , be (35) with ∂ t \partial_{t} ∂ t inserted in the left, resp. right, argument.
Theorem (The GRC brackets for o s p ( 5 ∣ 2 ) ⊂ k ( 1 ∣ 5 ) \mathfrak{osp}(5|2) \subset\mathfrak{k}(1|5) osp ( 5∣2 ) ⊂ k ( 1∣5 ) : one in each even order, and one in each odd order from 5 on). Let ord ≤ 7 \operatorname{ord} \le7 ord ≤ 7 . For generic weights , the space of i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) -invariant binary differential operators F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{F}_{\lambda_1} \otimes\mathcal{F}_{\lambda_2} \longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+\operatorname{ord}} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord is of dimension
For ord = 0 \operatorname{ord}=0 ord = 0 this is the multiplication. For ord even the operator is even, for ord odd it is odd. In the orders ord = 2 \operatorname{ord}=2 ord = 2 and ord = 4 \operatorname{ord}=4 ord = 4 , the space is non-zero for every pair of weights, not only for the generic ones. Indeed, the coefficients of the brackets (36) and (37) are polynomials in λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 that do not all vanish for any pair.
Computation. The maximal toral subalgebra, spanned by K ξ 1 η 1 K_{\xi_1\eta_1} K ξ 1 η 1 and K ξ 2 η 2 K_{\xi_2\eta_2} K ξ 2 η 2 , lies in i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) and acts on D ξ i D_{\xi_i} D ξ i , D η i D_{\eta_i} D η i , D θ D_\theta D θ , ∂ t \partial_t ∂ t with the weights 1 , − 1 , 0 , 0 1,-1,0,0 1 , − 1 , 0 , 0 . Hence, as in the proof of Lemma 6.3.2, an invariant operator lies in the weight-0 part of the ansatz. In the order ord = 4 \operatorname{ord}=4 ord = 4 , the answer with both weights symbolic is the bracket (37). In the orders ord = 2 , 3 \operatorname{ord}=2,3 ord = 2 , 3 , there are no exceptional weights.
The parity claim is the same weight argument: a term of weight 0 carries an even number of the D ξ i , D η i D_{\xi_i},D_{\eta_i} D ξ i , D η i taken together. Hence, its order and the number of its D θ D_\theta D θ ’s have the same parity. The parity of the term is that of the number of its D θ D_\theta D θ ’s. Unlike the case of even n n n , where the same argument leaves no odd order at all, here nothing is excluded a priori — and the orders ord = 1 \operatorname{ord}=1 ord = 1 and ord = 3 \operatorname{ord}=3 ord = 3 come out empty all the same. □
Statement (The order-2 bracket is the contact bracket, so it says nothing about the subalgebra). The order-2 bracket J 2 : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 J_2:\mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+2} J 2 : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 2 , where λ 1 , λ 2 ∈ C \lambda_1,\lambda_2\in\mathbb{C} λ 1 , λ 2 ∈ C are the weights of the two arguments, is
J 2 = Π − λ 1 1 ⊗ ∂ t + λ 2 ∂ t ⊗ 1 , (37) J_2=\Pi-\lambda_1\,1\otimes\partial_t+\lambda_2\,\partial_t\otimes1, \tag*{(37)} J 2 = Π − λ 1 1 ⊗ ∂ t + λ 2 ∂ t ⊗ 1 , (37)
and it is invariant not merely under i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) but under the whole of k ( 1 ∣ 5 ) \mathfrak{k}(1|5) k ( 1∣5 ) .
The order-4 bracket.
The order-4 bracket J 4 : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 4 J_4:\mathcal{F}_{\lambda_1}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+4} J 4 : F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + 4 exists for all weights λ 1 , λ 2 ∈ C \lambda_1,\lambda_2\in\mathbb{C} λ 1 , λ 2 ∈ C , and its coefficients are polynomials in λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 :
J 4 = 1 2 λ 1 ( λ 1 + 1 ) 1 ⊗ ∂ t 2 + ( 1 − λ 1 − λ 2 − λ 1 λ 2 ) ∂ t ⊗ ∂ t + 1 2 λ 2 ( λ 2 + 1 ) ∂ t 2 ⊗ 1 − ( λ 1 + 1 ) Π r + ( λ 2 + 1 ) Π l − ∑ 1 ≤ i ≤ 2 ( ∂ t ⊗ D ξ i D η i + D ξ i D η i ⊗ ∂ t ) + Π ( 2 ) . (38) \begin{aligned}
J_4={}&\frac{1}{2}\lambda_1(\lambda_1+1)\,1\otimes\partial_t^2+(1-\lambda_1-\lambda_2-\lambda_1\lambda_2)\,\partial_t\otimes\partial_t+\frac{1}{2}\lambda_2(\lambda_2+1)\,\partial_t^2\otimes1\\
&-(\lambda_1+1)\Pi_r+(\lambda_2+1)\Pi_l-\sum_{1\leq i\leq2}\left(\partial_t\otimes D_{\xi_i}D_{\eta_i}+D_{\xi_i}D_{\eta_i}\otimes\partial_t\right)+\Pi^{(2)}. \tag*{(38)}
\end{aligned} J 4 = 2 1 λ 1 ( λ 1 + 1 ) 1 ⊗ ∂ t 2 + ( 1 − λ 1 − λ 2 − λ 1 λ 2 ) ∂ t ⊗ ∂ t + 2 1 λ 2 ( λ 2 + 1 ) ∂ t 2 ⊗ 1 − ( λ 1 + 1 ) Π r + ( λ 2 + 1 ) Π l − 1 ≤ i ≤ 2 ∑ ( ∂ t ⊗ D ξ i D η i + D ξ i D η i ⊗ ∂ t ) + Π ( 2 ) . (38)
where Π ( 2 ) \Pi^{(2)} Π ( 2 ) , the second contraction, is
Π ( 2 ) = ∑ 1 ≤ i ≤ 2 ( D ξ i D η i ⊗ D ξ i D η i − D θ D ξ i ⊗ D θ D η i − D θ D η i ⊗ D θ D ξ i ) − D ξ 1 D ξ 2 ⊗ D η 1 D η 2 − D η 1 D η 2 ⊗ D ξ 1 D ξ 2 − D ξ 1 D η 2 ⊗ D η 1 D ξ 2 − D η 1 D ξ 2 ⊗ D ξ 1 D η 2 . (39) \begin{aligned}
\Pi^{(2)}={}&\sum_{1\leq i\leq2}\left(D_{\xi_i}D_{\eta_i}\otimes D_{\xi_i}D_{\eta_i}-D_\theta D_{\xi_i}\otimes D_\theta D_{\eta_i}-D_\theta D_{\eta_i}\otimes D_\theta D_{\xi_i}\right)\\
&-D_{\xi_1}D_{\xi_2}\otimes D_{\eta_1}D_{\eta_2}-D_{\eta_1}D_{\eta_2}\otimes D_{\xi_1}D_{\xi_2}-D_{\xi_1}D_{\eta_2}\otimes D_{\eta_1}D_{\xi_2}-D_{\eta_1}D_{\xi_2}\otimes D_{\xi_1}D_{\eta_2}. \tag*{(39)}
\end{aligned} Π ( 2 ) = 1 ≤ i ≤ 2 ∑ ( D ξ i D η i ⊗ D ξ i D η i − D θ D ξ i ⊗ D θ D η i − D θ D η i ⊗ D θ D ξ i ) − D ξ 1 D ξ 2 ⊗ D η 1 D η 2 − D η 1 D η 2 ⊗ D ξ 1 D ξ 2 − D ξ 1 D η 2 ⊗ D η 1 D ξ 2 − D η 1 D ξ 2 ⊗ D ξ 1 D η 2 . (39)
Both brackets (36) and (37) were re-entered by hand, from the formulas as they are printed here. They were then fed to the independent invariance test against all 23 generating functions of i ( o s p ( 5 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(5|2)) i ( osp ( 5∣2 )) at the weight pairs ( 3 7 , − 5 11 ) \left(\frac{3}{7},-\frac{5}{11}\right) ( 7 3 , − 11 5 ) , ( 1 3 , 2 5 ) \left(\frac{1}{3},\frac{2}{5}\right) ( 3 1 , 5 2 ) , ( − 7 3 , 5 2 ) \left(-\frac{7}{3},\frac{5}{2}\right) ( − 3 7 , 2 5 ) , ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( − 4 , 1 2 ) \left(-4,\frac{1}{2}\right) ( − 4 , 2 1 ) . Compare the bracket (37) with the bracket (36): the shape is the same, one contraction higher, with λ i \lambda_i λ i replaced by λ i + 1 \lambda_i+1 λ i + 1 in the middle line — which is the shape of the classical Gordan transvectants.
The order-5 bracket, and the Bol operator inside it.
In the order ord = 5 \operatorname{ord}=5 ord = 5 , the answer is 1-dimensional and odd. Normalize it so that the coefficient of T 5 ⊗ 1 T_5 \otimes1 T 5 ⊗ 1 is 1. Then, the four terms of T 5 T_5 T 5 occur, in each of the two extreme blocks, with exactly the relative coefficients they have in formula (35) , and
Statement 5.4.5 (The ratio of the two extreme blocks of the order-5 bracket). In J 5 \mathcal{J}_5 J 5 , normalized by coeff ( T 5 ⊗ 1 ) = 1 \operatorname{coeff}(T_5 \otimes1)=1 coeff ( T 5 ⊗ 1 ) = 1 ,
coeff ( 1 ⊗ T 5 ) = − ∏ 0 ≤ i ≤ 4 ( λ 1 + i ) ∏ 0 ≤ i ≤ 4 ( λ 2 + i ) . (40) \operatorname{coeff}(1 \otimes T_5)=-\frac{\prod_{0\leq i\leq4}(\lambda_1+i)}{\prod_{0\leq i\leq4}(\lambda_2+i)}. \tag*{(40)} coeff ( 1 ⊗ T 5 ) = − ∏ 0 ≤ i ≤ 4 ( λ 2 + i ) ∏ 0 ≤ i ≤ 4 ( λ 1 + i ) . (40) In particular, coeff ( 1 ⊗ T 5 ) = − 1 \operatorname{coeff}(1 \otimes T_5)=-1 coeff ( 1 ⊗ T 5 ) = − 1 for λ 1 = λ 2 \lambda_1=\lambda_2 λ 1 = λ 2 , and coeff ( 1 ⊗ T 5 ) = 0 \operatorname{coeff}(1 \otimes T_5)=0 coeff ( 1 ⊗ T 5 ) = 0 if and only if λ 1 ∈ { 0 , − 1 , − 2 , − 3 , − 4 } \lambda_1\in\{0,-1,-2,-3,-4\} λ 1 ∈ { 0 , − 1 , − 2 , − 3 , − 4 } .
For n = 3 n=3 n = 3 , the same computation gives the ratio
∏ 0 ≤ i ≤ 2 ( λ 1 + i ) ∏ 0 ≤ i ≤ 2 ( λ 2 + i ) , \frac{\prod_{0\leq i\leq2}(\lambda_1+i)}{\prod_{0\leq i\leq2}(\lambda_2+i)}, ∏ 0 ≤ i ≤ 2 ( λ 2 + i ) ∏ 0 ≤ i ≤ 2 ( λ 1 + i ) ,
so formula (40) is the instance n = 5 n=5 n = 5 of a pattern with n n n factors upstairs and n n n downstairs.
Now, recall Theorem 5.3.1: T 5 : F − 4 ⟶ F 1 T_5:\mathcal{F}_{-4}\longrightarrow\mathcal{F}_1 T 5 : F − 4 ⟶ F 1 is invariant, and only there. The weight λ 1 = − 4 \lambda_1=-4 λ 1 = − 4 is one of the five at which the ratio (40) vanishes, and it is the weight at which the surviving block is invariant on its own:
Corollary 5.4.6 (How the Bol operator enters the binary picture). If P : F λ 1 ⟶ F λ 1 + ord P:\mathcal{F}_{\lambda_1}\longrightarrow\mathcal{F}_{\lambda_1+\operatorname{ord}} P : F λ 1 ⟶ F λ 1 + ord is invariant, then so is ( φ , ψ ) ↦ ( P φ ) ψ (\varphi,\psi)\mapsto(P\varphi)\psi ( φ , ψ ) ↦ ( P φ ) ψ , because the multiplication F μ ⊗ F λ 2 ⟶ F μ + λ 2 \mathcal{F}_{\mu}\otimes\mathcal{F}_{\lambda_2}\longrightarrow\mathcal{F}_{\mu+\lambda_2} F μ ⊗ F λ 2 ⟶ F μ + λ 2 is invariant; same for the second argument. Hence, by Theorem 5.3.1, the operator T 5 ⊗ 1 T_5\otimes1 T 5 ⊗ 1 is invariant at λ 1 = − 4 \lambda_1=-4 λ 1 = − 4 , whatever λ 2 \lambda_2 λ 2 is, and 1 ⊗ T 5 1\otimes T_5 1 ⊗ T 5 is invariant at λ 2 = − 4 \lambda_2=-4 λ 2 = − 4 . At λ 1 = − 4 \lambda_1=-4 λ 1 = − 4 and generic λ 2 \lambda_2 λ 2 , the space of order-5 brackets is still 1-dimensional. Moreover, J 5 \mathcal{J}_5 J 5 is exactly T 5 ⊗ 1 T_5\otimes1 T 5 ⊗ 1 : not only does the block 1 ⊗ T 5 1\otimes T_5 1 ⊗ T 5 drop out, as formula (40) says, but so does every term between the two.
The last claim is a computation: at ( λ 1 , λ 2 ) = ( − 4 , 1 2 ) (\lambda_1,\lambda_2)=(-4,\frac{1}{2}) ( λ 1 , λ 2 ) = ( − 4 , 2 1 ) and ( − 4 , 3 7 ) (-4,\frac{3}{7}) ( − 4 , 7 3 ) the order-5 space is 1-dimensional and its only solution is T 5 ⊗ 1 T_5\otimes1 T 5 ⊗ 1 on the nose, the other 70 terms of the ansatz having zero coefficients.
This is the mechanism of § 4.6 and of §7. What is new for odd n n n is that the operator it produces is odd. Hence, it lives in an odd order, where for even n n n there is nothing at all.
Open problems for n = 5 n=5 n = 5
The orders ord ≥ 8 \operatorname{ord}\geq8 ord ≥ 8 , i.e., whether the pattern of Theorem 5.4.1 — one bracket in every even order, and one in every odd order from 5 on — continues. A closed formula, in the shape of the Gordan transvectants, for the bracket of an arbitrary order; formulas (36), (37) and (40) are its first three instances. The exceptional weights in the orders ord ≥ 4 \operatorname{ord}\geq4 ord ≥ 4 , for which formula (40) is the first piece of evidence. The remaining case of the pair o s p ( n ∣ 2 ) ⊂ k ( 1 ∣ n ) \mathfrak{osp}(n|2)\subset\mathfrak{k}(1|n) osp ( n ∣2 ) ⊂ k ( 1∣ n ) in our range, n = 6 n=6 n = 6 , is §6, where the subalgebra considered is not i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) but a s ⊂ k a s \mathfrak{as}\subset\mathfrak{kas} as ⊂ kas .
Super Gordan (Rankin-Cohen) operators: the a s ⊂ k a s ⊂ k ( 1 ∣ 6 ) \mathfrak{as}\subset\mathfrak{kas}\subset\mathfrak{k}(1|6) as ⊂ kas ⊂ k ( 1∣6 ) case
Summary of this Section
This is the exceptional case which [14 §1.3.1] left “for the future”. The answer turns out to be of a different kind from the ones in §§2–5, and the reason for that is structural, not computational.
The embedding. In the standard grading of k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) we have g 1 = Λ 3 ⊕ t Λ 1 \mathfrak{g}_1=\Lambda^3\oplus t\Lambda^1 g 1 = Λ 3 ⊕ t Λ 1 , and Λ 3 = Λ ξ 3 ⊕ Λ η 3 \Lambda^3=\Lambda^3_{\xi}\oplus\Lambda^3_{\eta} Λ 3 = Λ ξ 3 ⊕ Λ η 3 splits into two irreducible g 0 \mathfrak{g}_0 g 0 -modules — this happens for n = 6 n=6 n = 6 only, and
it is what k a s \mathfrak{kas} kas is made of: k a s \mathfrak{kas} kas is the Cartan prolong of g − ⊕ g 0 \mathfrak{g}_{-} \oplus\mathfrak{g}_{0} g − ⊕ g 0 in which g 1 \mathfrak{g}_{1} g 1 is replaced with Λ ξ 3 ⊕ Λ 1 \Lambda^{3}_{\xi} \oplus\Lambda^{1} Λ ξ 3 ⊕ Λ 1 , i.e., one of the two summands of Λ 3 \Lambda^{3} Λ 3 is dropped, see §6.2 and [48 ] . Sergeev’s algebra sits inside as
a s = ⟨ K f ∣ f ∈ Λ 0 ⊕ Λ 1 ⊕ Λ 2 ⊕ Λ ξ 3 ⟩ , sdim a s = ( 16 ∣ 16 ) , \mathfrak{as} = \langle K_f \mid f \in\Lambda^{0} \oplus\Lambda^{1} \oplus\Lambda^{2} \oplus\Lambda^{3}_{\xi} \rangle,\qquad\operatorname{sdim}\mathfrak{as} = (16|16), as = ⟨ K f ∣ f ∈ Λ 0 ⊕ Λ 1 ⊕ Λ 2 ⊕ Λ ξ 3 ⟩ , sdim as = ( 16∣16 ) ,
see Statement 6.2.1; the center is ⟨ K 1 ⟩ \langle K_1\rangle ⟨ K 1 ⟩ and a s / ⟨ K 1 ⟩ ≃ s p e ( 4 ) \mathfrak{as}/\langle K_1\rangle\simeq\mathfrak{spe}(4) as / ⟨ K 1 ⟩ ≃ spe ( 4 ) . Unlike the case n = 4 n=4 n = 4 of §4, this embedding sits in no family: the three irreducible summands of g 1 \mathfrak{g}_{1} g 1 are pairwise non-isomorphic, see Statement 6.2.2.
The weights are invisible to a s \mathfrak{as} as . Every element of a s \mathfrak{as} as is K f K_f K f with ∂ t ( f ) = 0 \partial_t(f)=0 ∂ t ( f ) = 0 , i.e., a s ⊂ p o ( 0 ∣ 6 ) ⊂ k ( 1 ∣ 6 ) \mathfrak{as} \subset\mathfrak{po}(0|6) \subset\mathfrak{k}(1|6) as ⊂ po ( 0∣6 ) ⊂ k ( 1∣6 ) . For such fields L K f λ = K f \mathcal{L}^{\lambda}_{K_f}=K_f L K f λ = K f whatever λ \lambda λ is, so all the F λ \mathcal{F}_{\lambda} F λ are one and the same a s \mathfrak{as} as -module, see Lemma 6.3.1. Hence, for the pair a s ⊂ k a s \mathfrak{as} \subset\mathfrak{kas} as ⊂ kas , there are no exceptional weights, and no operator can be tied to a distinguished weight: the mechanism which produces the Bol operators is absent.
The unary operators. The only non-scalar a s \mathfrak{as} as -invariant ones are the powers ∂ t k \partial_t^k ∂ t k of order 2 k 2k 2 k , and they are invariant as operators F λ ⟶ F μ \mathcal{F}_{\lambda} \longrightarrow\mathcal{F}_{\mu} F λ ⟶ F μ for every pair of weights, see Theorem 6.4.1. In particular, the pair a s ⊂ k a s \mathfrak{as} \subset\mathfrak{kas} as ⊂ kas has no analogs of Bol operators. The ambient k a s \mathfrak{kas} kas has no invariant unary operators at all, whereas for the other maximal finite-dimensional subalgebra of k a s \mathfrak{kas} kas , namely i ( o s p ( 6 ∣ 2 ) ) i(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) , the answer is the classical Bol operator, T 6 ∘ ∂ t k : F − 5 − k ⟶ F 1 + k T_6 \circ\partial_t^k:\mathcal{F}_{-5-k}\longrightarrow\mathcal{F}_{1+k} T 6 ∘ ∂ t k : F − 5 − k ⟶ F 1 + k , which is [14 Thm 1.4.2] for n = 6 n=6 n = 6 , see Statement 6.4.2. (This last computation is our validation of the machinery in the case n = 6 n=6 n = 6 .)
The GRC brackets. In odd orders, there is nothing — for a reason valid for every even n n n , see Lemma 6.3.2, which also proves the corresponding assertion of Theorem 4.3.1. In even orders, the dimensions are
Table 3.
see Theorem 6.5.1. All the a s \mathfrak{as} as -invariant binary operators of order ≤ 4 \le4 ≤ 4 are obtained from ∂ t \partial_t ∂ t acting in one of the arguments and from the single binary operator
Ω = ∑ 1 ≤ i ≤ 3 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) , \Omega= \sum_{1\leq i\leq3}\left(D_{\xi_i}\otimes D_{\eta_i}+D_{\eta_i}\otimes D_{\xi_i}\right), Ω = 1 ≤ i ≤ 3 ∑ ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) ,
which is of order 2. The bracket (42) is Ω \Omega Ω plus terms with ∂ t \partial_t ∂ t , and it coincides with Ω \Omega Ω only at λ 1 = λ 2 = 0 \lambda_1=\lambda_2=0 λ 1 = λ 2 = 0 .
The tensor fields other than densities, see § 6.6 . Since the densities carry nothing, this is where the Veblen problem (to describe the invariant differential operators between spaces of tensor fields) for the pair a s ⊂ k a s \mathfrak{as} \subset\mathfrak{kas} as ⊂ kas has content. The weights stay invisible there as well, and so does the twist by a power of the determinant, see Lemma 6.6.2: for a s \mathfrak{as} as a module of tensor fields depends on the underlying o ( 6 ) \mathfrak{o}(6) o ( 6 ) -module V V V and on nothing else. For V = Λ 2 ( id ) V=\Lambda^{2}(\operatorname{id}) V = Λ 2 ( id ) , everything is generated by a gradient and a divergence, both of order 1, see Theorem 6.6.5. For V = S 2 ( id ) V=S^{2}(\operatorname{id}) V = S 2 ( id ) , there is one new operator, of order 3, see Theorem 6.6.7. For the half-spinors V = id , id ∗ V=\operatorname{id},\operatorname{id}^{*} V = id , id ∗ , the two chiralities behave differently : exactly one of the two Dirac operators is a s \mathfrak{as} as -invariant, see Theorem 6.6.6. This is the only place in this paper where the two halves of Λ 3 \Lambda^{3} Λ 3 are told apart, and Statement 6.6.8 shows that they, and nothing else, are the cause.
The comparison with [35 §10] , see §6.7, where the unary k a s \mathfrak{kas} kas -invariant operators are described in the language of singular vectors. Only four of the entries of that list have both fibers finite-dimensional. Each of the four is reproduced here, with its weights, see Theorem 6.7.3. Among them is the Dirac operator, which exists already over k a s \mathfrak{kas} kas , at one weight, and is not
k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant, see Statement 6.7.4. One operator we find is not on that list, namely the divergence, which is the formal adjoint of the gradient, see Statement 6.7.5.
The pair a s \mathfrak{as} as
Everything below is computed in the realization (1) of k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) , with
α 1 = d t + ∑ 1 ≤ i ≤ 3 ( ξ i d η i + η i d ξ i ) , F λ = F α 1 λ / 2 , L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ , \alpha_1=dt+\sum_{1\leq i\leq3}(\xi_i d\eta_i+\eta_i d\xi_i),\qquad\mathcal{F}_{\lambda}=\mathcal{F}\alpha_1^{\lambda/2},\qquad\mathcal{L}^{\lambda}_{K_f}(\varphi)=K_f(\varphi)+\lambda\partial_t(f)\varphi, α 1 = d t + 1 ≤ i ≤ 3 ∑ ( ξ i d η i + η i d ξ i ) , F λ = F α 1 λ /2 , L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ ,
and with the standard grading deg t = 2 \deg t=2 deg t = 2 , deg ξ i = deg η i = 1 \deg\xi_i=\deg\eta_i=1 deg ξ i = deg η i = 1 , deg K f = deg f − 2 \deg K_f=\deg f-2 deg K f = deg f − 2 , so that
g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ⟩ , g 0 = ⟨ K t ⟩ ⊕ Λ 2 = c o ( 6 ) ≃ g l ( 4 ) , g 1 = Λ 3 ⊕ t Λ 1 , \mathfrak{g}_{-2}=\langle K_1\rangle,\qquad\mathfrak{g}_{-1}=\langle K_{y_i}\rangle,\qquad\mathfrak{g}_0=\langle K_t\rangle\oplus\Lambda^2=\mathfrak{co}(6)\simeq\mathfrak{gl}(4),\qquad\mathfrak{g}_1=\Lambda^3\oplus t\Lambda^1, g − 2 = ⟨ K 1 ⟩ , g − 1 = ⟨ K y i ⟩ , g 0 = ⟨ K t ⟩ ⊕ Λ 2 = co ( 6 ) ≃ gl ( 4 ) , g 1 = Λ 3 ⊕ t Λ 1 ,
where y 1 , … , y 6 y_1,\ldots,y_6 y 1 , … , y 6 stands for ξ 1 , η 1 , ξ 2 , η 2 , ξ 3 , η 3 \xi_1,\eta_1,\xi_2,\eta_2,\xi_3,\eta_3 ξ 1 , η 1 , ξ 2 , η 2 , ξ 3 , η 3 and where we write Λ k \Lambda^k Λ k for ⟨ K f ∣ f ∈ Λ k ( ξ , η ) ⟩ \langle K_f\mid f\in\Lambda^k(\xi,\eta)\rangle ⟨ K f ∣ f ∈ Λ k ( ξ , η )⟩ . Only for n = 6 n=6 n = 6 does Λ 3 \Lambda^3 Λ 3 split into two irreducible g 0 \mathfrak{g}_0 g 0 -modules: Λ ξ 3 \Lambda^3_{\xi} Λ ξ 3 , the one containing ξ 1 ξ 2 ξ 3 \xi_1\xi_2\xi_3 ξ 1 ξ 2 ξ 3 , and Λ η 3 \Lambda^3_{\eta} Λ η 3 , the one containing η 1 η 2 η 3 \eta_1\eta_2\eta_3 η 1 η 2 η 3 ; as g l ( 4 ) \mathfrak{gl}(4) gl ( 4 ) -modules they are S 2 ( i d ) S^2(\mathrm{id}) S 2 ( id ) and S 2 ( i d ∗ ) S^2(\mathrm{id}^*) S 2 ( id ∗ ) . Recall from [48 ] that
k a s = k a s ξ = ( g − ⊕ g 0 , Λ ξ 3 ⊕ t Λ 1 ) m k \mathfrak{kas}=\mathfrak{kas}^{\xi}=(\mathfrak{g}_{-}\oplus\mathfrak{g}_0,\Lambda^3_{\xi}\oplus t\Lambda^1)^{mk} kas = kas ξ = ( g − ⊕ g 0 , Λ ξ 3 ⊕ t Λ 1 ) mk
is simple and infinite-dimensional, and that k a s ξ ≃ k a s η \mathfrak{kas}^{\xi}\simeq\mathfrak{kas}^{\eta} kas ξ ≃ kas η . The next two Statements are due to Shchepochkina, see [47 , 48 ] .
Statement. The structure of a s \mathfrak{as} as
Statement 6.2.1 (The structure of a s \mathfrak{as} as : Sergeev’s central extension of s p e ( 4 ) ) . P u t \mathfrak{spe}(4)). Put spe ( 4 )) . P u t \mathfrak{as}:=\langle K_f\mid f\in\Lambda^0\oplus\Lambda^1\oplus\Lambda^2\oplus\Lambda^3_{\xi}\rangle$. Then:
(i) { Λ ξ 3 , Λ ξ 3 } = 0 \{\Lambda^3_{\xi},\Lambda^3_{\xi}\}=0 { Λ ξ 3 , Λ ξ 3 } = 0 (and likewise for Λ η 3 \Lambda^3_{\eta} Λ η 3 ), so a s \mathfrak{as} as is a subalgebra of k a s \mathfrak{kas} kas , of superdimension ( 16 ∣ 16 ) (16|16) ( 16∣16 ) , and a s ⊃ g − \mathfrak{as}\supset\mathfrak{g}_{-} as ⊃ g − , so Theorem 2.1 applies to it.
(ii) K 1 K_1 K 1 spans the center of a s \mathfrak{as} as , the quotient a s / ⟨ K 1 ⟩ \mathfrak{as}/\langle K_1\rangle as / ⟨ K 1 ⟩ is s p e ( 4 ) \mathfrak{spe}(4) spe ( 4 ) — its even part is Λ 2 ≃ o ( 6 ) ≃ s l ( 4 ) \Lambda^2\simeq\mathfrak{o}(6)\simeq\mathfrak{sl}(4) Λ 2 ≃ o ( 6 ) ≃ sl ( 4 ) , its odd part is Λ 1 ⊕ Λ ξ 3 ≃ Λ 2 ( i d ∗ ) ⊕ S 2 ( i d ) \Lambda^1\oplus\Lambda^3_{\xi}\simeq\Lambda^2(\mathrm{id}^*)\oplus S^2(\mathrm{id}) Λ 1 ⊕ Λ ξ 3 ≃ Λ 2 ( id ∗ ) ⊕ S 2 ( id ) — and the extension is non-split, since { y i , y j } ∈ C ⋅ K 1 \{y_i,y_j\}\in\mathbb{C}\cdot K_1 { y i , y j } ∈ C ⋅ K 1 is not identically zero. Thus, a s \mathfrak{as} as is Sergeev’s central extension of s p e ( 4 ) \mathfrak{spe}(4) spe ( 4 ) , cf. [48 , 12 ] .
(iii) K t ∉ a s K_t\notin\mathfrak{as} K t ∈ / as , and a s ⊕ ⟨ K t ⟩ \mathfrak{as}\oplus\langle K_t\rangle as ⊕ ⟨ K t ⟩ is again a subalgebra (we denote it by c a s \mathfrak{cas} cas ); it is a s \mathfrak{as} as that is listed in [14 Table (4)] .
Computation. All three parts are a direct check of the contact brackets; for (i) note also that { Λ ξ 3 , Λ ξ 3 } \{\Lambda^3_{\xi},\Lambda^3_{\xi}\} { Λ ξ 3 , Λ ξ 3 } would have to lie in Λ 4 ≃ Λ 2 \Lambda^4\simeq\Lambda^2 Λ 4 ≃ Λ 2 , whereas S 2 ( S 2 ( i d ) ) = S 4 ( i d ) ⊕ S ( 2 , 2 ) ( i d ) S^2(S^2(\mathrm{id}))=S^4(\mathrm{id})\oplus S^{(2,2)}(\mathrm{id}) S 2 ( S 2 ( id )) = S 4 ( id ) ⊕ S ( 2 , 2 ) ( id ) , where S ( 2 , 2 ) S^{(2,2)} S ( 2 , 2 ) is the Schur functor corresponding to the partition ( 2 , 2 ) (2,2) ( 2 , 2 ) , see [27 Lecture 6] , so that sdim S ( 2 , 2 ) ( i d ) = ( 20 ∣ 0 ) \operatorname{sdim}S^{(2,2)}(\mathrm{id})=(20|0) sdim S ( 2 , 2 ) ( id ) = ( 20∣0 ) , contains no summand ≃ s l ( 4 ) \simeq\mathfrak{sl}(4) ≃ sl ( 4 ) . □
Statement. No family, in contrast with n = 4 n=4 n = 4
Statement 6.2.2 (No family, in contrast with n = 4 n=4 n = 4 ). As o ( 6 ) \mathfrak{o}(6) o ( 6 ) -modules, g 1 = Λ ξ 3 ⊕ Λ η 3 ⊕ t Λ 1 \mathfrak{g}_1=\Lambda^3_{\xi}\oplus\Lambda^3_{\eta}\oplus t\Lambda^1 g 1 = Λ ξ 3 ⊕ Λ η 3 ⊕ t Λ 1 is a sum of three pairwise non-isomorphic irreducibles, of dimensions 10, 10 and 6. Hence, every o ( 6 ) \mathfrak{o}(6) o ( 6 ) -stable subspace of g 1 \mathfrak{g}_1 g 1 is a sum of some of them, and the family of Statement 4.2.1 has no counterpart here: the embedding a s ⊂ k ( 1 ∣ 6 ) \mathfrak{as}\subset\mathfrak{k}(1|6) as ⊂ k ( 1∣6 ) is rigid, and so is i ( o s p ( 6 ∣ 2 ) ) ⊂ k ( 1 ∣ 6 ) \mathfrak{i}(\mathfrak{osp}(6|2))\subset\mathfrak{k}(1|6) i ( osp ( 6∣2 )) ⊂ k ( 1∣6 ) .
Computation. Each of the three summands is generated, as an o ( 6 ) \mathfrak{o}(6) o ( 6 ) -module, by any of its nonzero vectors, and Hom o ( 6 ) ( t Λ 1 , Λ ξ 3 ) = 0 \operatorname{Hom}_{\mathfrak{o}(6)}(t\Lambda^1,\Lambda^3_{\xi})=0 Hom o ( 6 ) ( t Λ 1 , Λ ξ 3 ) = 0 ; the map σ \sigma σ which produced the parameter a a a in §4.2 is what is missing here. □
Why the weights are invisible to a s \mathfrak{as} as
Lemma. a s ⊂ p o ( 0 ∣ 6 ) \mathfrak{as}\subset\mathfrak{po}(0|6) as ⊂ po ( 0∣6 ) , so the weight is invisible to it
Lemma 6.3.1 (a s ⊂ p o ( 0 ∣ 6 ) \mathfrak{as}\subset\mathfrak{po}(0|6) as ⊂ po ( 0∣6 ) , so the weight is invisible to it). Every element of a s \mathfrak{as} as is of the form K f K_f K f with ∂ t ( f ) = 0 \partial_t(f)=0 ∂ t ( f ) = 0 , i.e., a s ⊂ p o ( 0 ∣ 6 ) \mathfrak{as}\subset\mathfrak{po}(0|6) as ⊂ po ( 0∣6 ) , where p o ( 0 ∣ 6 ) = ⟨ K f ∣ f ∈ Λ ( ξ , η ) ⟩ \mathfrak{po}(0|6)=\langle K_f\mid f\in\Lambda(\xi,\eta)\rangle po ( 0∣6 ) = ⟨ K f ∣ f ∈ Λ ( ξ , η )⟩ . Consequently:
(i) L K f λ = K f \mathcal{L}^{\lambda}_{K_f}=K_f L K f λ = K f on F λ \mathcal{F}_{\lambda} F λ for every λ \lambda λ and every K f ∈ a s K_f\in\mathfrak{as} K f ∈ as , and the map φ α 1 λ / 2 ⟼ φ α 1 μ / 2 \varphi\alpha_1^{\lambda/2}\longmapsto\varphi\alpha_1^{\mu/2} φ α 1 λ /2 ⟼ φ α 1 μ /2 is an isomorphism of a s \mathfrak{as} as -modules F λ ⟶ F μ \mathcal{F}_{\lambda}\longrightarrow\mathcal{F}_{\mu} F λ ⟶ F μ for all λ \lambda λ and μ \mu μ ;
(ii) [ K f , ∂ t ] = 0 [K_f,\partial_t]=0 [ K f , ∂ t ] = 0 for every K f ∈ a s K_f\in\mathfrak{as} K f ∈ as , so ∂ t k \partial_t^k ∂ t k is a s \mathfrak{as} as -invariant as an operator F λ ⟶ F μ \mathcal{F}_{\lambda}\longrightarrow\mathcal{F}_{\mu} F λ ⟶ F μ for any λ \lambda λ and $\mu; (iii) an operator (unary or binary) is a s \mathfrak{as} as -invariant for one choice of the weights if and only if it is a s \mathfrak{as} as -invariant for all of them. In particular, for the pair a s ⊂ k a s \mathfrak{as} \subset\mathfrak{kas} as ⊂ kas there are no exceptional weights.
Proof. a s \mathfrak{as} as is spanned by the K f K_f K f with f ∈ Λ ( ξ , η ) f \in\Lambda(\xi,\eta) f ∈ Λ ( ξ , η ) , and for such an f f f we have ∂ t ( f ) = 0 \partial_t(f)=0 ∂ t ( f ) = 0 , so the summand λ ∂ t ( f ) φ \lambda\partial_t(f)\varphi λ ∂ t ( f ) φ of L K f λ \mathcal{L}^{\lambda}_{K_f} L K f λ vanishes and K f = ( 2 − E ) ( f ) ∂ t − H f K_f=(2-E)(f)\partial_t-H_f K f = ( 2 − E ) ( f ) ∂ t − H f has coefficients free of t t t ; therefore [ K f , ∂ t ] = 0 [K_f,\partial_t]=0 [ K f , ∂ t ] = 0 . Claims (i)–(iii) follow.
Lemma 6.3.1 is the reason why this case is not like the ones treated above. There, the subalgebra contained K t y i K_{ty_i} K t y i , and it is precisely the term λ ∂ t ( t y i ) = λ y i \lambda\partial_t(ty_i)=\lambda y_i λ ∂ t ( t y i ) = λ y i that makes the invariance equations depend on the weight and singles out the weights at which a Bol operator lives. Here that term is identically zero.
No invariant operators of odd order
Lemma 6.3.2. Let n = 2 m n=2m n = 2 m and let h ⊂ k ( 1 ∣ n ) \mathfrak{h}\subset\mathfrak{k}(1|n) h ⊂ k ( 1∣ n ) contain the negative part k ( 1 ∣ n ) − \mathfrak{k}(1|n)_{-} k ( 1∣ n ) − and the torus ⟨ K ξ i η i ∣ 1 ≤ i ≤ m ⟩ \langle K_{\xi_i\eta_i}\mid1\le i\le m\rangle ⟨ K ξ i η i ∣ 1 ≤ i ≤ m ⟩ . Then, every h \mathfrak{h} h -invariant differential operator, unary or binary, between the spaces F λ \mathcal{F}_{\lambda} F λ of one-parameter weighted densities is of even order.
Proof. By Theorem 2.1 , the operator is a sum of terms ∂ t p D A ⊗ ∂ t q D B \partial_t^pD_A\otimes\partial_t^qD_B ∂ t p D A ⊗ ∂ t q D B with constant coefficients. Since ∂ t ( ξ i η i ) = 0 \partial_t(\xi_i\eta_i)=0 ∂ t ( ξ i η i ) = 0 , the field K ξ i η i K_{\xi_i\eta_i} K ξ i η i acts on every F λ \mathcal{F}_{\lambda} F λ by one and the same operator, and it acts on D ξ i D_{\xi_i} D ξ i , D η i D_{\eta_i} D η i , ∂ t \partial_t ∂ t with the weights 1 1 1 , − 1 -1 − 1 , 0 0 0 . Hence, an invariant operator is of weight 0 0 0 , and a term of weight 0 0 0 has ∣ A ∣ + ∣ B ∣ |A|+|B| ∣ A ∣ + ∣ B ∣ even, so its order 2 p + ∣ A ∣ + 2 q + ∣ B ∣ 2p+|A|+2q+|B| 2 p + ∣ A ∣ + 2 q + ∣ B ∣ is even.
For n = 2 n=2 n = 2 , the Lemma does not apply: there the densities have a second parameter, see §2.4, and the torus acts on β ν \beta^\nu β ν non-trivially. For n n n odd, there is the Z \mathbb{Z} Z -weight-zero derivative D θ D_\theta D θ . This is why odd orders do occur in §§2–3 and, from the order 5 on, in §5, but not in §4 and here.
The unary operators: no analogs of Bol operators
The unary as-invariant operators on the densities
Theorem 6.4.1. The non-zero non-scalar a s \mathfrak{as} as -invariant unary differential operators of the form F λ ⟶ F μ \mathcal{F}_{\lambda}\longrightarrow\mathcal{F}_{\mu} F λ ⟶ F μ are the polynomials in ∂ t \partial_t ∂ t of positive degree, and each of them is invariant for every pair ( λ , μ ) (\lambda,\mu) ( λ , μ ) ; the homogeneous ones are the multiples of ∂ t k \partial_t^k ∂ t k , where k ≥ 1 k\ge1 k ≥ 1 , of order 2 k 2k 2 k . For c a s = a s ⊕ ⟨ K t ⟩ \mathfrak{cas}=\mathfrak{as}\oplus\langle K_t\rangle cas = as ⊕ ⟨ K t ⟩ , only the homogeneous ones survive, and then μ = λ + 2 k \mu=\lambda+2k μ = λ + 2 k .
That nothing of odd order occurs is Lemma 6.3.2; moreover, in even orders nothing but ∂ t k \partial_t^k ∂ t k survives — in particular the “top” Clifford monomial
T 6 : = ∏ 1 ≤ i ≤ 3 ( D ξ i D η i − ∂ t ) (41) T_6:=\prod_{1\le i\le3}(D_{\xi_i}D_{\eta_i}-\partial_t) \tag*{(41)} T 6 := 1 ≤ i ≤ 3 ∏ ( D ξ i D η i − ∂ t ) (41)
is not a s \mathfrak{as} as -invariant, at any weight, although it is the o s p ( 6 ∣ 2 ) \mathfrak{osp}(6|2) osp ( 6∣2 ) -invariant Bol operator, see Statement 6.4.2.
What kas and i(osp(6|2)) see on the densities
Statement 6.4.2. (i) There are no non-zero non-scalar k a s \mathfrak{kas} kas -invariant unary operators at all, whatever the weights (orders ord ≤ 8 \operatorname{ord}\le8 ord ≤ 8 , the same grid of weights as above).
(ii) For i ( o s p ( 6 ∣ 2 ) ) ⊂ k a s i(\mathfrak{osp}(6|2))\subset\mathfrak{kas} i ( osp ( 6∣2 )) ⊂ kas the invariant unary operators are, for every k ≥ 0 k\ge0 k ≥ 0 , exactly
T 6 ∘ ∂ t k : F − 5 − k ⟶ F 1 + k , ord = 6 + 2 k , T_6\circ\partial_t^k:\quad\mathcal{F}_{-5-k}\longrightarrow\mathcal{F}_{1+k},\qquad\operatorname{ord}=6+2k, T 6 ∘ ∂ t k : F − 5 − k ⟶ F 1 + k , ord = 6 + 2 k ,
and there is nothing at any other weight, and nothing of order < 6 <6 < 6 . This is [14 Thm 1.4.2] for n = 6 n=6 n = 6 : by eq. (42) of [14 ] we have [ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) [D_{\eta_i},D_{\xi_i}]_+=-2(D_{\xi_i}D_{\eta_i}-\partial_t) [ D η i , D ξ i ] + = − 2 ( D ξ i D η i − ∂ t ) , so the operator (8) of [14 ] is − 8 T 6 ∘ ∂ t k -8T_6\circ\partial_t^k − 8 T 6 ∘ ∂ t k , and its weights, λ = 1 − k − n \lambda=1-k-n λ = 1 − k − n and μ = k + 1 \mu=k+1 μ = k + 1 , are ours for n = 6 n=6 n = 6 .
The GRC brackets
Theorem. The binary operators on the densities over a s \mathfrak{as} as , k a s \mathfrak{kas} kas and i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) . Let ord ≤ 4 \operatorname{ord} \le4 ord ≤ 4 . The space of invariant binary operators F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord \mathcal{F}_{\lambda_1} \otimes\mathcal{F}_{\lambda_2} \longrightarrow\mathcal{F}_{\lambda_1+\lambda_2+\operatorname{ord}} F λ 1 ⊗ F λ 2 ⟶ F λ 1 + λ 2 + ord is of dimension
(for a s \mathfrak{as} as the answer does not depend on λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 at all, by Lemma 6.3.1*).* Explicitly:
(i) the order-2 space of a s \mathfrak{as} as is spanned by
∂ t ⊗ 1 , 1 ⊗ ∂ t , Ω : = ∑ 1 ≤ i ≤ 3 ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) ; (42) \partial_t \otimes1,\qquad1 \otimes\partial_t,\qquad\Omega:= \sum_{1 \le i \le3}\left(D_{\xi_i} \otimes D_{\eta_i} + D_{\eta_i} \otimes D_{\xi_i}\right); \tag*{(42)} ∂ t ⊗ 1 , 1 ⊗ ∂ t , Ω := 1 ≤ i ≤ 3 ∑ ( D ξ i ⊗ D η i + D η i ⊗ D ξ i ) ; (42) (ii) the order-2 space of k a s \mathfrak{kas} kas , and that of i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) , is spanned by the single operator
J 2 = λ 2 ∂ t ⊗ 1 − λ 1 1 ⊗ ∂ t + Ω , (43) J_2 = \lambda_2 \partial_t \otimes1 - \lambda_1 1 \otimes\partial_t + \Omega, \tag*{(43)} J 2 = λ 2 ∂ t ⊗ 1 − λ 1 1 ⊗ ∂ t + Ω , (43) which is invariant under the whole of k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) and, for λ 1 = λ 2 = − 2 \lambda_1=\lambda_2=-2 λ 1 = λ 2 = − 2 , is the contact bracket. Thus, Statement 4.3.2 holds for n = 6 n=6 n = 6 as well, and again the first GRC bracket says nothing about the subalgebra;
(iii) the order-4 space of a s \mathfrak{as} as is spanned by
∂ t 2 ⊗ 1 , ∂ t ⊗ ∂ t , 1 ⊗ ∂ t 2 , Ω ∘ ( ∂ t ⊗ 1 ) , Ω ∘ ( 1 ⊗ ∂ t ) , \partial_t^2 \otimes1,\qquad\partial_t \otimes\partial_t,\qquad1 \otimes\partial_t^2,\qquad\Omega\circ(\partial_t \otimes1),\qquad\Omega\circ(1 \otimes\partial_t), ∂ t 2 ⊗ 1 , ∂ t ⊗ ∂ t , 1 ⊗ ∂ t 2 , Ω ∘ ( ∂ t ⊗ 1 ) , Ω ∘ ( 1 ⊗ ∂ t ) , where
Ω ∘ ( ∂ t ⊗ 1 ) : = ∑ 1 ≤ i ≤ 3 ( ∂ t D ξ i ⊗ D η i + ∂ t D η i ⊗ D ξ i ) , \Omega\circ(\partial_t \otimes1) := \sum_{1 \le i \le3}\left(\partial_tD_{\xi_i} \otimes D_{\eta_i} + \partial_tD_{\eta_i} \otimes D_{\xi_i}\right), Ω ∘ ( ∂ t ⊗ 1 ) := 1 ≤ i ≤ 3 ∑ ( ∂ t D ξ i ⊗ D η i + ∂ t D η i ⊗ D ξ i ) , and similarly for the last composition. In particular, no new bracket appears in order 4: everything is Ω \Omega Ω and ∂ t \partial_t ∂ t , the operator ∂ t \partial_t ∂ t being invariant by Lemma 6.3.1(ii);
(iv) the order-4 space of i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) is spanned by one operator, the second GRC bracket, namely
J 4 = ∑ 1 ≤ i ≤ 3 D ξ i D η i ⊗ D ξ i D η i − ∑ 1 ≤ i < j ≤ 3 ( D ξ i D ξ j ⊗ D η i D η j + D ξ i D η j ⊗ D η i D ξ j + D η i D ξ j ⊗ D ξ i D η j + D η i D η j ⊗ D ξ i D ξ j ) − ∑ 1 ≤ i ≤ 3 ( ∂ t ⊗ D ξ i D η i + D ξ i D η i ⊗ ∂ t ) + ( λ 2 + 1 ) Ω ∘ ( ∂ t ⊗ 1 ) − ( λ 1 + 1 ) Ω ∘ ( 1 ⊗ ∂ t ) + 1 2 λ 2 ( λ 2 + 1 ) ∂ t 2 ⊗ 1 + ( 2 − λ 1 − λ 2 − λ 1 λ 2 ) ∂ t ⊗ ∂ t + 1 2 λ 1 ( λ 1 + 1 ) 1 ⊗ ∂ t 2 . (44) \begin{aligned}
J_4={}&\sum_{1 \le i \le3}D_{\xi_i}D_{\eta_i}\otimes D_{\xi_i}D_{\eta_i}
-\sum_{1 \le i<j \le3}\left(D_{\xi_i}D_{\xi_j}\otimes D_{\eta_i}D_{\eta_j}+D_{\xi_i}D_{\eta_j}\otimes D_{\eta_i}D_{\xi_j}\right.\\
&\left.+D_{\eta_i}D_{\xi_j}\otimes D_{\xi_i}D_{\eta_j}+D_{\eta_i}D_{\eta_j}\otimes D_{\xi_i}D_{\xi_j}\right)
-\sum_{1 \le i \le3}\left(\partial_t\otimes D_{\xi_i}D_{\eta_i}+D_{\xi_i}D_{\eta_i}\otimes\partial_t\right)\\
&+(\lambda_2+1)\Omega\circ(\partial_t\otimes1)-(\lambda_1+1)\Omega\circ(1\otimes\partial_t)\\
&+\frac{1}{2}\lambda_2(\lambda_2+1)\partial_t^2\otimes1+(2-\lambda_1-\lambda_2-\lambda_1\lambda_2)\partial_t\otimes\partial_t+\frac{1}{2}\lambda_1(\lambda_1+1)1\otimes\partial_t^2. \tag*{(44)}
\end{aligned} J 4 = 1 ≤ i ≤ 3 ∑ D ξ i D η i ⊗ D ξ i D η i − 1 ≤ i < j ≤ 3 ∑ ( D ξ i D ξ j ⊗ D η i D η j + D ξ i D η j ⊗ D η i D ξ j + D η i D ξ j ⊗ D ξ i D η j + D η i D η j ⊗ D ξ i D ξ j ) − 1 ≤ i ≤ 3 ∑ ( ∂ t ⊗ D ξ i D η i + D ξ i D η i ⊗ ∂ t ) + ( λ 2 + 1 ) Ω ∘ ( ∂ t ⊗ 1 ) − ( λ 1 + 1 ) Ω ∘ ( 1 ⊗ ∂ t ) + 2 1 λ 2 ( λ 2 + 1 ) ∂ t 2 ⊗ 1 + ( 2 − λ 1 − λ 2 − λ 1 λ 2 ) ∂ t ⊗ ∂ t + 2 1 λ 1 ( λ 1 + 1 ) 1 ⊗ ∂ t 2 . (44) Its part free of the weights is the “square” of Ω \Omega Ω ; the weights enter only through Ω \Omega Ω and ∂ t \partial_t ∂ t . Note that J 4 J_4 J 4 is not k a s \mathfrak{kas} kas -invariant — in order 4 the ambient exceptional algebra has nothing.
The three operators (41) are separately invariant precisely because the weights are invisible to a s \mathfrak{as} as : the k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant combination (42) is a s \mathfrak{as} as -invariant for every λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 , hence so is each of its three parts.
Tensor fields other than densities
By Theorem 6.4.1, the weighted densities carry almost nothing: the only a s \mathfrak{as} as -invariant unary operators between them are the powers of ∂ t \partial_t ∂ t . Moreover, by Lemma 6.3.1, all the F λ \mathcal{F}_\lambda F λ are one and the same a s \mathfrak{as} as -module. So for the pair a s ⊂ k a s \mathfrak{as} \subset\mathfrak{kas} as ⊂ kas the Veblen problem is really a question about the other k a s 0 \mathfrak{kas}_0 kas 0 -modules, and this is what we do here. It turns out that the answer is again completely explicit, and that it distinguishes the two chiralities — something no other case of this paper does.
What the tensor fields of type V are.
In the standard grading of §6.2, k a s 0 = k ( 1 ∣ 6 ) 0 = c o ( 6 ) ≃ g l ( 4 ) \mathfrak{k}\mathfrak{as}_0=\mathfrak{k}(1|6)_0=\mathfrak{co}(6)\simeq\mathfrak{gl}(4) k as 0 = k ( 1∣6 ) 0 = co ( 6 ) ≃ gl ( 4 ) . The tensor fields of type V V V are the fields coinduced from an irreducible k a s 0 \mathfrak{k}\mathfrak{as}_0 k as 0 -module V V V , with k a s > 0 \mathfrak{k}\mathfrak{as}_{>0} k as > 0 acting trivially; they are realized as V V V -valued functions. The weighted densities are precisely the 1-dimensional V V V ’s: there o ( 6 ) \mathfrak{o}(6) o ( 6 ) acts trivially and K t K_t K t acts by the weight.
The action is written out as follows. Contact fields preserve the contact distribution spanned by the D i D_i D i of eq. (1) , so for every function f f f there is a 6 × 6 6\times6 6 × 6 matrix a ( f ) a(f) a ( f ) , with entries functions of parity p ( f ) p(f) p ( f ) , with
[ K f , D i ] = ∑ 1 ≤ j ≤ 6 a ( f ) i j D j (45) [K_f,D_i]=\sum_{1\leq j\leq6}a(f)_{ij}D_j \tag*{(45)} [ K f , D i ] = 1 ≤ j ≤ 6 ∑ a ( f ) ij D j (45)
and a ( f ) a(f) a ( f ) takes its values in c o ( 6 ) \mathfrak{co}(6) co ( 6 ) pointwise. The tensor fields of type V V V and weight λ \lambda λ are then F ⊗ V \mathcal{F}\otimes V F ⊗ V with
( L K f V , λ φ ) b = K f ( φ b ) + λ ∂ t ( f ) φ b + ∑ a ρ ( a ( f ) t ) b a φ a , (46) (\mathcal{L}^{V,\lambda}_{K_f}\varphi)_b=K_f(\varphi_b)+\lambda\partial_t(f)\varphi_b+\sum_a\rho\bigl(a(f)^t\bigr)_{ba}\varphi_a, \tag*{(46)} ( L K f V , λ φ ) b = K f ( φ b ) + λ ∂ t ( f ) φ b + a ∑ ρ ( a ( f ) t ) ba φ a , (46)
where ρ \rho ρ is the representation of c o ( 6 ) \mathfrak{co}(6) co ( 6 ) on V V V and a t a^t a t is the transposed matrix. For V V V the vector representation and ρ \rho ρ the tautological representation of o ( 6 ) \mathfrak{o}(6) o ( 6 ) , this is nothing but the module of vector fields along the contact distribution, X = ∑ 1 ≤ i ≤ 6 φ i D i X=\sum_{1\leq i\leq6}\varphi_iD_i X = ∑ 1 ≤ i ≤ 6 φ i D i with L K f X = [ K f , X ] \mathcal{L}_{K_f}X=[K_f,X] L K f X = [ K f , X ] . Expanding that bracket is the derivation of formulas (44) and (45). For dim V = 1 \dim V=1 dim V = 1 , eq. (46) is the familiar L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ \mathcal{L}^{\lambda}_{K_f}(\varphi)=K_f(\varphi)+\lambda\partial_t(f)\varphi L K f λ ( φ ) = K f ( φ ) + λ ∂ t ( f ) φ .
Lemma. The weights are invisible to as on every tensor field
Lemma. The weights are invisible to a s \mathfrak{as} as on every tensor field. For every function f f f ,
tr a ( f ) = − 6 ∂ t ( f ) . (47) \operatorname{tr}a(f)=-6\partial_t(f). \tag*{(47)} tr a ( f ) = − 6 ∂ t ( f ) . (47) Hence, a ( f ) a(f) a ( f ) is o ( 6 ) \mathfrak{o}(6) o ( 6 ) -valued if and only if ∂ t ( f ) = 0 \partial_t(f)=0 ∂ t ( f ) = 0 , i.e., if and only if K f ∈ p o ( 0 ∣ 6 ) K_f\in\mathfrak{po}(0|6) K f ∈ po ( 0∣6 ) , and in particular for every K f ∈ a s K_f\in\mathfrak{as} K f ∈ as . Consequently, on the tensor fields of type V V V , the algebra a s \mathfrak{as} as feels neither the weight λ \lambda λ nor the twist of V V V by a power of the determinant. As an a s \mathfrak{as} as -module, the space of tensor fields of type V V V depends on the o ( 6 ) \mathfrak{o}(6) o ( 6 ) -module underlying V V V and on nothing else.
Proof. Since K t = 2 t ∂ t + ∑ 1 ≤ i ≤ 6 y i ∂ y i K_t=2t\partial_t+\sum_{1\leq i\leq6}y_i\partial_{y_i} K t = 2 t ∂ t + ∑ 1 ≤ i ≤ 6 y i ∂ y i , where y i y_i y i are the odd coordinates, one has L K f α 1 = 2 ∂ t ( f ) α 1 \mathcal{L}_{K_f}\alpha_1=2\partial_t(f)\alpha_1 L K f α 1 = 2 ∂ t ( f ) α 1 , hence L K f ( d α 1 ) = 2 ∂ t ( f ) d α 1 \mathcal{L}_{K_f}(d\alpha_1)=2\partial_t(f)d\alpha_1 L K f ( d α 1 ) = 2 ∂ t ( f ) d α 1 on the contact distribution. The values d α 1 ( D i , D j ) d\alpha_1(D_i,D_j) d α 1 ( D i , D j ) are constants, the entries of a nondegenerate matrix B B B . Apply L K f \mathcal{L}_{K_f} L K f to them and use formula (44): a ( f ) B + B a ( f ) t = − 2 ∂ t ( f ) B a(f)B+Ba(f)^t=-2\partial_t(f)B a ( f ) B + B a ( f ) t = − 2 ∂ t ( f ) B (the signs for odd f f f being given by the sign rule). Multiply by B − 1 B^{-1} B − 1 and take the trace: 2 tr a ( f ) = − 12 ∂ t ( f ) 2\operatorname{tr}a(f)=-12\partial_t(f) 2 tr a ( f ) = − 12 ∂ t ( f ) , which is formula (46). So the scalar part of a ( f ) ∈ c o ( 6 ) = C ⊕ o ( 6 ) a(f)\in\mathfrak{co}(6)=\mathbb{C}\oplus\mathfrak{o}(6) a ( f ) ∈ co ( 6 ) = C ⊕ o ( 6 ) is 1 6 tr a ( f ) = − ∂ t ( f ) \frac{1}{6}\operatorname{tr}a(f)=-\partial_t(f) 6 1 tr a ( f ) = − ∂ t ( f ) . This is the direction K t K_t K t of c o ( 6 ) \mathfrak{co}(6) co ( 6 ) by which the determinant twist acts, and it is also the coefficient of λ \lambda λ in formula (45). (47) was verified for all the generating functions of k a s \mathfrak{k}\mathfrak{as} k as and for K t 2 K_{t^2} K t 2 , hence for k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) . □
This is Lemma 6.3.1 extended from the densities to all tensor fields. It is what makes the classification below a finite problem: for a s \mathfrak{as} as there is one module of tensor fields per o ( 6 ) \mathfrak{o}(6) o ( 6 ) -module V V V , with no continuous parameter attached to it.
The modules V we consider.
Let Λ k \Lambda^k Λ k be the k k k -th exterior power of the vector representation of o ( 6 ) \mathfrak{o}(6) o ( 6 ) , and let i d \mathrm{id} id be the tautological module of g l ( 4 ) \mathfrak{gl}(4) gl ( 4 ) , of dimension 4. A power of i d \mathrm{id} id always carries its argument, so the module Λ 2 \Lambda^2 Λ 2 , of dimension 15, is not the module Λ 2 ( i d ) \Lambda^2(\mathrm{id}) Λ 2 ( id ) , of dimension 6. Under the isomorphism o ( 6 ) ≃ s l ( 4 ) \mathfrak{o}(6)\simeq\mathfrak{sl}(4) o ( 6 ) ≃ sl ( 4 ) , the o ( 6 ) \mathfrak{o}(6) o ( 6 ) -modules on the left are the g l ( 4 ) \mathfrak{gl}(4) gl ( 4 ) -modules on the right:
Λ 1 ≃ Λ 2 ( i d ) ( dim 6 ) , Λ 2 ≃ s l ( 4 ) ( dim 15 ) , Λ 3 = Λ + 3 ⊕ Λ − 3 , \Lambda^1\simeq\Lambda^2(\mathrm{id})\ (\dim6),\qquad
\Lambda^2\simeq\mathfrak{sl}(4)\ (\dim15),\qquad
\Lambda^3=\Lambda^3_+\oplus\Lambda^3_-, Λ 1 ≃ Λ 2 ( id ) ( dim 6 ) , Λ 2 ≃ sl ( 4 ) ( dim 15 ) , Λ 3 = Λ + 3 ⊕ Λ − 3 ,
where Λ ± 3 \Lambda^{3}_{\pm} Λ ± 3 are of dimension 10 and, as g l ( 4 ) \mathfrak{gl}(4) gl ( 4 ) -modules, are S 2 ( i d ) S^{2}(\mathrm{id}) S 2 ( id ) and S 2 ( i d ∗ ) S^{2}(\mathrm{id}^{*}) S 2 ( id ∗ ) ; and there are the two half-spinors i d \mathrm{id} id and i d ∗ \mathrm{id}^{*} id ∗ , of dimension 4, which do not occur in Λ ∙ \Lambda^{\bullet} Λ ∙ . We label the two members of each of these pairs by the extreme weight relative the maximal toral subalgebra ⟨ K ξ i η i ⟩ \langle K_{\xi_i\eta_i}\rangle ⟨ K ξ i η i ⟩ : it is ( 1 , 1 , 1 ) (1,1,1) ( 1 , 1 , 1 ) for Λ + 3 \Lambda^{3}_{+} Λ + 3 and ( 1 2 , 1 2 , 1 2 ) \left(\frac{1}{2},\frac{1}{2},\frac{1}{2}\right) ( 2 1 , 2 1 , 2 1 ) for the half-spinor called S + S^{+} S + , with the opposite signs for Λ − 3 \Lambda^{3}_{-} Λ − 3 and S − S^{-} S − . Thus, S − S^{-} S − is the half of the Clifford module spanned by the states with an even number of the θ \theta θ ’s, its extreme weight being ( 1 2 , 1 2 , − 1 2 ) \left(\frac{1}{2},\frac{1}{2},-\frac{1}{2}\right) ( 2 1 , 2 1 , − 2 1 ) . The chirality flip ξ i ↔ η i \xi_i \leftrightarrow\eta_i ξ i ↔ η i interchanges the members of each pair, and it interchanges Λ ξ 3 \Lambda^{3}_{\xi} Λ ξ 3 and Λ η 3 \Lambda^{3}_{\eta} Λ η 3 , of which a s \mathfrak{as} as contains the first one only, see Statement 6.2.1. This is the source of everything asymmetric below, see Statement 6.6.8.
The gradient and the divergence.
Two operators built out of the contact distribution will describe case V = Λ 1 V=\Lambda^{1} V = Λ 1 completely. Order the odd coordinates as ( y 1 , … , y 6 ) = ( ξ 1 , ξ 2 , ξ 3 , η 1 , η 2 , η 3 ) (y_1,\ldots,y_6)=(\xi_1,\xi_2,\xi_3,\eta_1,\eta_2,\eta_3) ( y 1 , … , y 6 ) = ( ξ 1 , ξ 2 , ξ 3 , η 1 , η 2 , η 3 ) , let B B B be the Gram matrix of the form ∑ 1 ≤ i ≤ 3 ( ξ i d η i + η i d ξ i ) \sum_{1\leq i\leq3}(\xi_i\,d\eta_i+\eta_i\,d\xi_i) ∑ 1 ≤ i ≤ 3 ( ξ i d η i + η i d ξ i ) in this basis, i.e., B = ( 0 1 3 1 3 0 ) B=\begin{pmatrix}0&1_3\\1_3&0\end{pmatrix} B = ( 0 1 3 1 3 0 ) , and let C = B − 1 C=B^{-1} C = B − 1 (here C = B C=B C = B ). Put
grad : F ⟶ T ( Λ 1 ) , grad φ : = ∑ 1 ≤ i , j ≤ 6 C i j D j ( φ ) D i = ∑ 1 ≤ i ≤ 3 ( D η i ( φ ) D ξ i + D ξ i ( φ ) D η i ) ; div : T ( Λ 1 ) ⟶ F , div ( φ ) : = ∑ 1 ≤ i ≤ 6 D i ( φ i ) = ∑ 1 ≤ i ≤ 3 ( D ξ i ( φ ξ i ) + D η i ( φ η i ) ) , (48) \begin{aligned}
\operatorname{grad}: \mathcal{F} &\longrightarrow\mathcal{T}(\Lambda^{1}), \qquad\operatorname{grad}\varphi:= \sum_{1\leq i,j\leq6} C_{ij}D_j(\varphi)D_i = \sum_{1\leq i\leq3}\left(D_{\eta_i}(\varphi)D_{\xi_i}+D_{\xi_i}(\varphi)D_{\eta_i}\right);\\
\operatorname{div}: \mathcal{T}(\Lambda^{1}) &\longrightarrow\mathcal{F}, \qquad\operatorname{div}(\varphi):=\sum_{1\leq i\leq6}D_i(\varphi_i)=\sum_{1\leq i\leq3}\left(D_{\xi_i}(\varphi_{\xi_i})+D_{\eta_i}(\varphi_{\eta_i})\right), \tag*{(48)}
\end{aligned} grad : F div : T ( Λ 1 ) ⟶ T ( Λ 1 ) , grad φ := 1 ≤ i , j ≤ 6 ∑ C ij D j ( φ ) D i = 1 ≤ i ≤ 3 ∑ ( D η i ( φ ) D ξ i + D ξ i ( φ ) D η i ) ; ⟶ F , div ( φ ) := 1 ≤ i ≤ 6 ∑ D i ( φ i ) = 1 ≤ i ≤ 3 ∑ ( D ξ i ( φ ξ i ) + D η i ( φ η i ) ) , (48)
where T ( V ) \mathcal{T}(V) T ( V ) denotes the tensor fields of type V V V and φ i \varphi_i φ i are the components of φ = ∑ 1 ≤ i ≤ 6 φ i D i \varphi=\sum_{1\leq i\leq6}\varphi_iD_i φ = ∑ 1 ≤ i ≤ 6 φ i D i . Both are of order 1; both are a s \mathfrak{as} as -invariant for every weight, by Lemma 6.6.2 .
Theorem (V = Λ 2 ( i d ) \Lambda^{2}(\mathrm{id}) Λ 2 ( id ) , the vector representation). Let V = Λ 1 = Λ 2 ( i d ) V=\Lambda^{1}=\Lambda^{2}(\mathrm{id}) V = Λ 1 = Λ 2 ( id ) , of dimension 6. For the orders ord ≤ 6 \operatorname{ord}\leq6 ord ≤ 6 , the spaces of a s \mathfrak{as} as -invariant unary operators are of the following dimensions, for all weights:
Table 4.
and, explicitly, the spaces are spanned by
∂ t k ∘ grad , ∂ t k ∘ div ( ord = 2 k + 1 ) , \partial_t^k\circ\operatorname{grad},\qquad\partial_t^k\circ\operatorname{div}\qquad(\operatorname{ord}=2k+1), ∂ t k ∘ grad , ∂ t k ∘ div ( ord = 2 k + 1 ) , respectively by
∂ t k ⋅ 1 and ∂ t k − 1 ∘ grad ∘ div ( ord = 2 k ) . \partial_t^k\cdot1\qquad\text{and}\qquad\partial_t^{k-1}\circ\operatorname{grad}\circ\operatorname{div}\qquad(\operatorname{ord}=2k). ∂ t k ⋅ 1 and ∂ t k − 1 ∘ grad ∘ div ( ord = 2 k ) . In particular, grad ∘ div \operatorname{grad}\circ\operatorname{div} grad ∘ div is not a multiple of ∂ t \partial_t ∂ t , and up to order 6 everything is generated by grad \operatorname{grad} grad , div \operatorname{div} div and ∂ t \partial_t ∂ t .
Besides, in order 0 the space of invariant pairings T ( Λ 1 ) ⊗ T ( Λ 1 ) ⟶ F \mathcal{T}(\Lambda^{1})\otimes\mathcal{T}(\Lambda^{1})\longrightarrow\mathcal{F} T ( Λ 1 ) ⊗ T ( Λ 1 ) ⟶ F is 1-dimensional, spanned by ( φ , ψ ) ↦ ∑ 1 ≤ i , j ≤ 6 C i j φ i ψ j (\varphi,\psi)\mapsto\sum_{1\leq i,j\leq6}C_{ij}\varphi_i\psi_j ( φ , ψ ) ↦ ∑ 1 ≤ i , j ≤ 6 C ij φ i ψ j ; in order 1 there is nothing.
So, in contrast with the densities, for V = Λ 2 ( i d ) V=\Lambda^{2}(\mathrm{id}) V = Λ 2 ( id ) the pair a s ⊂ k a s \mathfrak{as}\subset\mathfrak{kas} as ⊂ kas does have invariant operators of odd order — Lemma 6.3.2 applies to the densities only, since its proof uses that the index carries no torus weight.
Theorem (V = i d \mathrm{id} id and V = i d ∗ V=\mathrm{id}^{*} V = id ∗ , the half-spinors). Let V V V be one of the two 4-dimensional half-spinors S ± S^{\pm} S ± .
(i) There are no non-zero a s \mathfrak{as} as -invariant operators between densities and tensor fields of type S ± S^{\pm} S ± in any order, in either direction. The reason is stronger than invariance: the ansatz itself is empty, because the torus weights of a half-spinor index are half-integers, whereas those of the words ∂ t k D i 1 ⋯ D i r \partial_t^kD_{i_1}\cdots D_{i_r} ∂ t k D i 1 ⋯ D i r and of the densities are integers.
(ii) Hom ( T ( S ± ) , T ( S ± ) ) \operatorname{Hom}(\mathcal{T}(S^{\pm}),\mathcal{T}(S^{\pm})) Hom ( T ( S ± ) , T ( S ± )) is 1-dimensional in the even orders ord = 2 , 4 \operatorname{ord}=2,4 ord = 2 , 4 , spanned by ∂ t k \partial_t^k ∂ t k , and zero in the odd orders ord ≤ 5 \operatorname{ord}\leq5 ord ≤ 5 .
(iii) (The asymmetry. ) For ord ≤ 5 \operatorname{ord}\leq5 ord ≤ 5 ,
The operator of order 1 is the Dirac operator2 {}^{2} 2 φ ↦ ∑ 1 ≤ i , j ≤ 6 C i j γ i D j ( φ ) \varphi\mapsto\sum_{1\leq i,j\leq6}C_{ij}\gamma_iD_j(\varphi) φ ↦ ∑ 1 ≤ i , j ≤ 6 C ij γ i D j ( φ ) , where the γ \gamma γ ’s are the Clifford matrices of o ( 6 ) \mathfrak{o}(6) o ( 6 ) (verified as such term by term). The order-3 operator is ∂ t \partial_t ∂ t times it. In order 5, a second, independent operator appears, which we have not identified. In the opposite direction there is nothing at all, although the ansatz is of exactly the same size (12 terms of torus weight 0 in order 1, 40 in order 3, 52 in order 5). Thus, exactly one of the two Dirac operators is a s \mathfrak{as} as -invariant.
(iv) The binary picture is asymmetric as well: in order 1 the space T ( S − ) ⊗ T ( S − ) ⟶ F \mathcal{T}(S^{-})\otimes\mathcal{T}(S^{-})\longrightarrow\mathcal{F} T ( S − ) ⊗ T ( S − ) ⟶ F is 2-dimensional, whereas T ( S + ) ⊗ T ( S + ) ⟶ F \mathcal{T}(S^{+})\otimes\mathcal{T}(S^{+})\longrightarrow\mathcal{F} T ( S + ) ⊗ T ( S + ) ⟶ F is 1-dimensional. The mixed operators T ( S ± ) ⊗ T ( S ∓ ) ⟶ F \mathcal{T}(S^{\pm})\otimes\mathcal{T}(S^{\mp})\longrightarrow\mathcal{F} T ( S ± ) ⊗ T ( S ∓ ) ⟶ F have an empty ansatz in order 1 by the parity of the torus weight. In order 0, however, they carry the invariant spinor pairing, whose space is 1-dimensional (whereas T ( S ± ) ⊗ T ( S ± ) ⟶ F \mathcal{T}(S^{\pm})\otimes\mathcal{T}(S^{\pm})\longrightarrow\mathcal{F} T ( S ± ) ⊗ T ( S ± ) ⟶ F has an empty ansatz in order 0).
Theorem. V=S2(id)
Theorem (V = S 2 ( id ) S^2(\operatorname{id}) S 2 ( id ) ). Let V = Λ ± 3 ≃ S 2 ( id ) V=\Lambda^3_{\pm}\simeq S^2(\operatorname{id}) V = Λ ± 3 ≃ S 2 ( id ) or S 2 ( id ∗ ) S^2(\operatorname{id}^*) S 2 ( id ∗ ) , of dimension 10. For the orders ord ≤ 5 \operatorname{ord}\leq5 ord ≤ 5 ,
The entries equal to 1 in the fifth row are the powers ∂ t k \partial_t^k ∂ t k and nothing else. The two entries equal to 1 in order 3 are the projection to Λ ± 3 \Lambda^3_{\pm} Λ ± 3 of the third covariant derivative, corrected by a term ∂ t D \partial_tD ∂ t D . They exist for one chirality each, and in opposite directions. The entries in order 5 are ∂ t \partial_t ∂ t times them. So here, too, the two chiralities are told apart, and in the same way.
Statement. The control experiment
Statement (The control experiment). Let a s op = ⟨ K 1 , Λ 1 , Λ 2 , Λ η 3 ⟩ \mathfrak{as}^{\operatorname{op}}=\langle K_1,\Lambda^1,\Lambda^2,\Lambda^3_{\eta}\rangle as op = ⟨ K 1 , Λ 1 , Λ 2 , Λ η 3 ⟩ be the image of a s \mathfrak{as} as under the chirality flip ξ i ↔ η i \xi_i\leftrightarrow\eta_i ξ i ↔ η i . Then, every asymmetric answer of Theorems 6.6.6 and 6.6.7 is interchanged: for example, the space Hom ( T ( S + ) , T ( S − ) ) \operatorname{Hom}(\mathcal{T}(S^{+}),\mathcal{T}(S^{-})) Hom ( T ( S + ) , T ( S − )) becomes nonzero in the odd orders, of dimension 1, 1, 2 for ord = 1 , 3 , 5 \operatorname{ord}=1,3,5 ord = 1 , 3 , 5 , whereas Hom ( T ( S − ) , T ( S + ) ) \operatorname{Hom}(\mathcal{T}(S^{-}),\mathcal{T}(S^{+})) Hom ( T ( S − ) , T ( S + )) becomes zero. Hence, the asymmetry is caused by the choice of the half of Λ 3 \Lambda^3 Λ 3 sitting inside a s \mathfrak{as} as , and by nothing else.
This is the phenomenon promised in §6.1. On the densities, the two chiralities are indistinguishable, because the whole of o ( 6 ) \mathfrak{o}(6) o ( 6 ) acts trivially there. On every V V V with dim V > 1 \dim V>1 dim V > 1 , the o ( 6 ) \mathfrak{o}(6) o ( 6 ) -index is visible, and then Λ ξ 3 ⊂ a s ⊅ Λ η 3 \Lambda^{3}_{\xi} \subset\mathfrak{as} \not\supset\Lambda^{3}_{\eta} Λ ξ 3 ⊂ as ⊃ Λ η 3 separates them. Note that i ( o s p ( 6 ∣ 2 ) ) \mathfrak{i}(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) , which contains no cubics at all, cannot see the difference.
How the operators were found.
As in §6.5, the ansatz is cut to its part of torus weight 0, the equations are imposed for a Lie-generating subset only, and every solution is then checked against the full list of the 32 generating functions of a s \mathfrak{as} as .
Comparison with the description of the unary k a s \mathfrak{kas} kas -invariant operators in GLS
In [35 ] , §10, the unary operators invariant relative k a s \mathfrak{kas} kas and relative k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) in the standard grading are described, in the language of singular vectors and for arbitrary — not necessarily finite-dimensional — fibers. Since a s ⊂ k a s ⊂ k ( 1 ∣ 6 ) \mathfrak{as} \subset\mathfrak{kas} \subset\mathfrak{k}(1|6) as ⊂ kas ⊂ k ( 1∣6 ) , that description is a test of everything computed in §§6.4–6.6. Each operator of [35 §10] whose two fibers are among the modules of §6.6.3 must occur in our tables. By Lemma 6.6.2, it must occur there at every weight . The test gives the following.
The dictionary.
Recall the setting of Bernstein and Rudakov, see [35 §1.7] and [Ru]: for a Z \mathbb{Z} Z -graded Lie superalgebra g = ⨁ g i \mathfrak{g} = \bigoplus\mathfrak{g}_{i} g = ⨁ g i of vector fields and a g 0 \mathfrak{g}_{0} g 0 -module V V V with g > 0 V = 0 \mathfrak{g}_{>0}V = 0 g > 0 V = 0 , one puts I ( V ) = U ( g ) ⊗ U ( g ≥ 0 ) V I(V) = U(\mathfrak{g}) \otimes_{U(\mathfrak{g}_{\geq0})} V I ( V ) = U ( g ) ⊗ U ( g ≥ 0 ) V , and then I ( V ) ∗ = T ( V ∗ ) I(V)^{*} = T(V^{*}) I ( V ) ∗ = T ( V ∗ ) . Hence, a singular vector of weight μ \mu μ in I ( V λ ) I(V_{\lambda}) I ( V λ ) — a g 0 \mathfrak{g}_{0} g 0 -highest weight vector killed by g > 0 \mathfrak{g}_{>0} g > 0 — is the same thing as a non-zero g \mathfrak{g} g -invariant differential operator between the two corresponding spaces of tensor fields. The order of the operator is equal to the degree of the singular vector in U ( g − ) U(\mathfrak{g}_{-}) U ( g − ) , where deg ∂ t = 2 \deg\partial_{t} = 2 deg ∂ t = 2 as in §2.3. The module in which the singular vector lives is the source .
In [35 §10] , the weights are taken relative the basis ( K t , K ξ 1 η 1 , K ξ 2 η 2 , K ξ 3 η 3 ) (K_{t}, K_{\xi_{1}\eta_{1}}, K_{\xi_{2}\eta_{2}}, K_{\xi_{3}\eta_{3}}) ( K t , K ξ 1 η 1 , K ξ 2 η 2 , K ξ 3 η 3 ) of k a s 0 = c o ( 6 ) \mathfrak{kas}_{0} = \mathfrak{co}(6) kas 0 = co ( 6 ) . A weight λ = ( λ 0 ; λ 1 , λ 2 , λ 3 ) \lambda= (\lambda_{0}; \lambda_{1}, \lambda_{2}, \lambda_{3}) λ = ( λ 0 ; λ 1 , λ 2 , λ 3 ) of [35 §10] corresponds to the space of tensor fields of §6.6.1 whose fiber and weight are given by
( λ 1 , λ 2 , λ 3 ) = the highest weight of the fiber V , λ 0 = − ε (49) (\lambda_{1}, \lambda_{2}, \lambda_{3}) = \text{the highest weight of the fiber } V,\qquad\lambda_{0} = -\varepsilon \tag*{(49)} ( λ 1 , λ 2 , λ 3 ) = the highest weight of the fiber V , λ 0 = − ε (49)
where ε \varepsilon ε is the scalar by which L K t V , λ \mathcal{L}^{V,\lambda}_{K_{t}} L K t V , λ acts on the fiber. In our normalization L K t V , λ = K t + λ + c V \mathcal{L}^{V,\lambda}_{K_{t}} = K_{t} + \lambda+ c_{V} L K t V , λ = K t + λ + c V , where c V c_{V} c V is the scalar by which ρ ( a ( t ) t ) \rho(a(t)^{t}) ρ ( a ( t ) t ) acts, see formula (46) ; since a ( t ) = − 1 a(t) = -1 a ( t ) = − 1 by formula (47) ,
c F = 0 , c Λ 1 = − 1 , c Λ 2 = − 2 , c Λ ± 3 = − 3 , c S ± = 0 , (50) c_{F} = 0,\qquad c_{\Lambda^{1}} = -1,\qquad c_{\Lambda^{2}} = -2,\qquad c_{\Lambda^{3}_{\pm}} = -3,\qquad c_{S_{\pm}} = 0, \tag*{(50)} c F = 0 , c Λ 1 = − 1 , c Λ 2 = − 2 , c Λ ± 3 = − 3 , c S ± = 0 , (50)
the last value being a normalization of ρ \rho ρ on the half-spinors — a normalization which is invisible to a s \mathfrak{as} as by Lemma 6.6.2, and which just fixes what we call the weight of a spinor field.
We do not deduce the dictionary (48) from the conventions of [35 ] : passing from V V V to V ∗ V^{*} V ∗ interchanges, for o ( 6 ) \mathfrak{o}(6) o ( 6 ) , the two half-spinors and the two halves of Λ 3 \Lambda^{3} Λ 3 . Hence, a convention read off the text would be a guess. We calibrate the dictionary instead: the dictionary (48) is the only reading under which all four operators of Theorem 6.7.3 match [35 §10] . Moreover, they match with their weights, two of which are half-integers, and with the three vanishing mirror cases as well.
Which entries of [§10]GLS have finite-dimensional fibers.
The fiber attached to λ \lambda λ is the irreducible o ( 6 ) \mathfrak{o}(6) o ( 6 ) -module with highest weight ( λ 1 , λ 2 , λ 3 ) (\lambda_{1},\lambda_{2},\lambda_{3}) ( λ 1 , λ 2 , λ 3 ) ; it is finite-dimensional if and only if λ 1 ≥ λ 2 ≥ ∣ λ 3 ∣ \lambda_{1} \geq\lambda_{2} \geq|\lambda_{3}| λ 1 ≥ λ 2 ≥ ∣ λ 3 ∣ with all the λ i \lambda_{i} λ i integers, or all of them half-integers. Under this restriction, the list of [35 §10] shrinks to very little:
1a) λ = ( k , − k , l , l ) ⟶ λ + ( − 1 , 1 , 0 , 0 ) \lambda= (k,-k,l,l) \longrightarrow\lambda+ (-1,1,0,0) λ = ( k , − k , l , l ) ⟶ λ + ( − 1 , 1 , 0 , 0 ) : the fibers are ( − k , l , l ) (-k,l,l) ( − k , l , l ) and ( − k + 1 , l , l ) (-k+1,l,l) ( − k + 1 , l , l ) , so both lie in the list of §6.6.3 only for k = l = 0 k = l = 0 k = l = 0 : the pair F , Λ 1 \mathcal{F}, \Lambda^{1} F , Λ 1 .
1a*) λ = ( k , l , 1 − k , l + 1 ) \lambda= (k,l,1-k,l+1) λ = ( k , l , 1 − k , l + 1 ) : the fiber ( l , 1 − k , l + 1 ) (l,1-k,l+1) ( l , 1 − k , l + 1 ) is never dominant — l ≥ ∣ l + 1 ∣ l \geq|l+1| l ≥ ∣ l + 1∣ is impossible — so this family has no finite-dimensional specialization at all. 1b) λ = ( k , l , l , 2 − k ) ⟶ λ + ( − 1 , 0 , 0 , 1 ) \lambda=(k,l,l,2-k)\longrightarrow\lambda+(-1,0,0,1) λ = ( k , l , l , 2 − k ) ⟶ λ + ( − 1 , 0 , 0 , 1 ) : the fibers are ( l , l , 2 − k ) (l,l,2-k) ( l , l , 2 − k ) and ( l , l , 3 − k ) (l,l,3-k) ( l , l , 3 − k ) , dominant for l ≥ max ( ∣ 2 − k ∣ , ∣ 3 − k ∣ ) l\ge\max(\lvert2-k\rvert,\lvert3-k\rvert) l ≥ max (∣ 2 − k ∣ , ∣ 3 − k ∣) ; three specializations have both fibers in §6.6.3, namely ( k , l ) = ( 5 2 , 1 2 ) (k,l)=\left(\frac{5}{2},\frac{1}{2}\right) ( k , l ) = ( 2 5 , 2 1 ) , ( 3 , 1 ) (3,1) ( 3 , 1 ) and ( 2 , 1 ) (2,1) ( 2 , 1 ) .
1c) λ = ( k , l , − l − 2 , k − 2 ) \lambda=(k,l,-l-2,k-2) λ = ( k , l , − l − 2 , k − 2 ) : the fiber ( l , − l − 2 , k − 2 ) (l,-l-2,k-2) ( l , − l − 2 , k − 2 ) is never dominant, for the same reason as in 1a*).
Degree 2 has three singular vectors, 2a), 2a*) and 2b): their fibers are ( − 1 , − 1 , − 1 ) (-1,-1,-1) ( − 1 , − 1 , − 1 ) , ( − 1 , 0 , 0 ) (-1,0,0) ( − 1 , 0 , 0 ) and ( − 1 , − 1 , 1 ) (-1,-1,1) ( − 1 , − 1 , 1 ) , none of them dominant. So in order 2 there is nothing to compare : every operator of [35 §10] of order 2 acts in tensor fields with infinite-dimensional fibers.
Theorem. The four operators, and the three that are absent
Theorem (The four operators, and the three that are absent). Each of the four operators below spans a 1-dimensional space of k a s \mathfrak{kas} kas -invariant operators of order 1, for the stated pair of weights and for no other; and each is the entry of [35 §10] indicated in the last column, under the dictionary (48):
Table 5.
In the three mirror directions — T ( S + ) ⟶ T ( S − ) \mathcal{T}(S^+)\longrightarrow\mathcal{T}(S^-) T ( S + ) ⟶ T ( S − ) , T ( Λ 2 ) ⟶ T ( Λ − 3 ) \mathcal{T}(\Lambda^2)\longrightarrow\mathcal{T}(\Lambda^3_-) T ( Λ 2 ) ⟶ T ( Λ − 3 ) and T ( Λ + 3 ) ⟶ T ( Λ 2 ) \mathcal{T}(\Lambda^3_+)\longrightarrow\mathcal{T}(\Lambda^2) T ( Λ + 3 ) ⟶ T ( Λ 2 ) — there is nothing, for any weight, although the ansätze are of the same size (12, 54 and 54 terms). This is what [35 §10] requires as well: the mirror weights, e.g. ( 3 , 1 , 1 , 1 ) (3,1,1,1) ( 3 , 1 , 1 , 1 ) for T ( Λ + 3 ) ⟶ T ( Λ 2 ) \mathcal{T}(\Lambda^3_+)\longrightarrow\mathcal{T}(\Lambda^2) T ( Λ + 3 ) ⟶ T ( Λ 2 ) , are of none of the forms 1a)–1c).
The weights were found by a scan: for each pair of fibers, K t K_t K t -invariance alone confines the pair ( λ 1 , λ 2 ) (\lambda_1,\lambda_2) ( λ 1 , λ 2 ) to a line λ 2 = λ 1 + c \lambda_2=\lambda_1+c λ 2 = λ 1 + c , and λ 1 \lambda_1 λ 1 was run through the half-integers of [ − 8 , 8 ] [-8,8] [ − 8 , 8 ] on that line. Every operator found was then verified, by a routine sharing nothing with the solver, against all the 39 generating functions of k a s \mathfrak{kas} kas and on 192 test monomials. It was also verified to fail at the neighboring weight λ 1 ± 1 2 \lambda_1\pm\frac{1}{2} λ 1 ± 2 1 , so the weights above are exactly what they are.
Statement. What k(1|6) sees
Statement (What k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) sees). The gradient, at ( λ 1 , λ 2 ) = ( 0 , 2 ) (\lambda_1,\lambda_2)=(0,2) ( λ 1 , λ 2 ) = ( 0 , 2 ) , is invariant relative the whole of k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) . The Dirac operator, at ( − 5 2 , − 3 2 ) \left(-\frac{5}{2},-\frac{3}{2}\right) ( − 2 5 , − 2 3 ) , is not: it is k a s \mathfrak{kas} kas -invariant, and it fails precisely on the generating functions spanning Λ η 3 \Lambda^3_\eta Λ η 3 , i.e., on k ( 1 ∣ 6 ) ∖ k a s \mathfrak{k}(1|6)\setminus\mathfrak{kas} k ( 1∣6 ) ∖ kas . This is the distinction drawn in [35 §10] between the two columns there: 1a) holds for k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) only if l = 0 l=0 l = 0 , which is our case, whereas 1b) holds for k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) only if l = − 1 l=-1 l = − 1 , whereas l = 1 2 l=\frac{1}{2} l = 2 1 for the Dirac operator.
(50) So the Dirac operator of Theorem 6.6.6 is not an accident of a s \mathfrak{as} as : it already exists over the exceptional k a s \mathfrak{kas} kas , at one weight, and it is one of the operators that distinguish k a s \mathfrak{kas} kas from k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) . What a s \mathfrak{as} as adds is that it exists for every weight.
Statement. The divergence, and a duality
Statement (The divergence, and a duality). The divergence div \operatorname{div} div of formula (47) is k a s \mathfrak{kas} kas -invariant — and, moreover, k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant — for λ 1 = λ 2 = − 4 \lambda_1=\lambda_2=-4 λ 1 = λ 2 = − 4 , and for no other weight:
div : T ( Λ 1 ) − 4 ⟶ F − 4 = Vol , ord = 1 , \operatorname{div}:\mathcal{T}(\Lambda^1)_{-4}\longrightarrow\mathcal{F}_{-4}=\operatorname{Vol},\qquad\operatorname{ord}=1, div : T ( Λ 1 ) − 4 ⟶ F − 4 = Vol , ord = 1 ,
F − 4 \mathcal{F}_{-4} F − 4 being the volume forms, cf. Statement 6.4.2(ii), whose family F − 5 − k ⟶ F 1 + k \mathcal{F}_{-5-k}\longrightarrow\mathcal{F}_{1+k} F − 5 − k ⟶ F 1 + k is symmetric with respect to λ ↦ − 4 − λ \lambda\mapsto-4-\lambda λ ↦ − 4 − λ . In the weights of [35 ] , this operator is
( 5 , 1 , 0 , 0 ) ⟶ ( 4 , 0 , 0 , 0 ) , (5,1,0,0)\longrightarrow(4,0,0,0), ( 5 , 1 , 0 , 0 ) ⟶ ( 4 , 0 , 0 , 0 ) ,
of degree 1, and it is of none of the forms 1a)–1c) of [35 §10] .
It is the formal adjoint of the gradient. Indeed, the Berezin integral pairs T ( V , λ ) T(V,\lambda) T ( V , λ ) with T ( V ∗ , μ ) T(V^*,\mu) T ( V ∗ , μ ) into Vol = F − 4 \operatorname{Vol}=\mathcal{F}_{-4} Vol = F − 4 precisely when the two K t K_t K t -eigenvalues add up to − 4 -4 − 4 , so on the weights of [35 ] the formal adjoint acts by
λ = ( λ 0 , λ 1 , λ 2 , λ 3 ) ⟼ λ † : = ( 4 − λ 0 , λ 1 , λ 2 , − λ 3 ) , (51) \lambda=(\lambda_0,\lambda_1,\lambda_2,\lambda_3)\longmapsto\lambda^\dagger:=(4-\lambda_0,\lambda_1,\lambda_2,-\lambda_3), \tag*{(51)} λ = ( λ 0 , λ 1 , λ 2 , λ 3 ) ⟼ λ † := ( 4 − λ 0 , λ 1 , λ 2 , − λ 3 ) , (51)
and it turns the operator λ ⟶ μ \lambda\longrightarrow\mu λ ⟶ μ into the operator μ † ⟶ λ † \mu^\dagger\longrightarrow\lambda^\dagger μ † ⟶ λ † . Since the adjoint of an invariant operator is invariant, the list of all singular vectors must be stable under † \dagger † . The families 1b) and 1c) are indeed stable: † \dagger † sends 1b) with the parameters ( k , l ) (k,l) ( k , l ) to 1b) with ( 5 − k , l ) (5-k,l) ( 5 − k , l ) , and 1c) with ( k , l ) (k,l) ( k , l ) to 1c) with ( 5 − k , l ) (5-k,l) ( 5 − k , l ) . The families 1a) and 1a*) are not: † \dagger † sends them to ( 5 − k , 1 − k , l , − l ) ⟶ ( 4 − k , − k , l , − l ) (5-k,1-k,l,-l)\longrightarrow(4-k,-k,l,-l) ( 5 − k , 1 − k , l , − l ) ⟶ ( 4 − k , − k , l , − l ) and ( 5 − k , l , 2 − k , − l − 1 ) ⟶ ( 4 − k , l , 1 − k , − l − 1 ) (5-k,l,2-k,-l-1)\longrightarrow(4-k,l,1-k,-l-1) ( 5 − k , l , 2 − k , − l − 1 ) ⟶ ( 4 − k , l , 1 − k , − l − 1 ) , with the shifts − ε 1 -\varepsilon_1 − ε 1 and − ε 2 -\varepsilon_2 − ε 2 , and no family of [35 §10] has either of these two shifts. The divergence is the member k = l = 0 k=l=0 k = l = 0 of the first of the two, i.e., the adjoint of the gradient.
What exactly is missing from [§10]GLS.
This is the one point at which our computation does not reproduce [35 §10] . The operator itself is not in doubt: it is div, written out in formula (48) . It is the first operator anybody writes down after the gradient, and it is k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant at λ 1 = λ 2 = − 4 \lambda_1=\lambda_2=-4 λ 1 = λ 2 = − 4 . What is in doubt is only where it should stand in the table of [35 §10] . The next three subsections show that the table, and not the operator, is incomplete.
The involution is a symmetry of the whole of [§10]GLS, with two exceptions.
The involution † \dagger † is a symmetry of the whole of [35 §10] , with two exceptions. The involution (51) makes sense for each of the three theorems of [35 §10] — for k ( 1 ∣ 3 ) \mathfrak{k}(1|3) k ( 1∣3 ) , for k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) and for k a s , k ( 1 ∣ 6 ) \mathfrak{kas},\mathfrak{k}(1|6) kas , k ( 1∣6 ) — in the form
λ = ( λ 0 ; λ 1 , … , λ s ) ⟼ λ † = ( n − 2 − λ 0 ; − w 0 ( λ 1 , … , λ s ) ) , s = ⌊ n 2 ⌋ , (52) \lambda=(\lambda_0;\lambda_1,\ldots,\lambda_s)\longmapsto\lambda^\dagger=(n-2-\lambda_0;-w_0(\lambda_1,\ldots,\lambda_s)),\qquad s=\left\lfloor\frac{n}{2}\right\rfloor, \tag*{(52)} λ = ( λ 0 ; λ 1 , … , λ s ) ⟼ λ † = ( n − 2 − λ 0 ; − w 0 ( λ 1 , … , λ s )) , s = ⌊ 2 n ⌋ , (52)
where n − 2 n-2 n − 2 is the K t K_t K t -weight of Vol and − w 0 -w_0 − w 0 is the duality of o ( n ) \mathfrak{o}(n) o ( n ) -weights: the identity for o ( 3 ) = B 1 \mathfrak{o}(3)=B_1 o ( 3 ) = B 1 and o ( 4 ) = D 2 \mathfrak{o}(4)=D_2 o ( 4 ) = D 2 , the flip λ s ↦ − λ s \lambda_s\mapsto-\lambda_s λ s ↦ − λ s for o ( 6 ) = D 3 \mathfrak{o}(6)=D_3 o ( 6 ) = D 3 . Applied to every entry of the three theorems, it gives
Table 1.
The parameters match exactly, not just the shapes: e.g. for k ( 1 ∣ 3 ) \mathfrak{k}(1|3) k ( 1∣3 ) the family 1a) with the parameter k k k goes to 1a*) with the parameter − k -k − k , and for k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) the family 1a) with k k k goes to 1c) with 1 − k 1-k 1 − k . In the two lower lists, every single entry has its adjoint in the list — the only exceptions in all of [35 §10] being the degree-2 entries of k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , where nothing is forced, see the next paragraph. In the list for k a s \mathfrak{kas} kas and k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) , every entry has its adjoint in the list except 1a) and 1a*). Their images have the shifts − ε 1 -\varepsilon_1 − ε 1 and − ε 2 -\varepsilon_2 − ε 2 , which occur nowhere in that list, although the corresponding shifts do occur in the lists for k ( 1 ∣ 3 ) \mathfrak{k}(1|3) k ( 1∣3 ) and k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) , whose degree-1 parts carry all the ± ε i \pm\varepsilon_i ± ε i at once.
Where the closure under is forced, and where it is not.
The † \dagger † -argument is an argument about operators, and it applies verbatim only when both fibers are finite-dimensional. For an infinite-dimensional irreducible fiber, the dual module is not a highest weight module at all, and the involution (51) says nothing. This is exactly why the degree-2 entries of the k ( 1 ∣ 4 ) \mathfrak{k}(1|4) k ( 1∣4 ) list may fail to be closed under † \dagger † without any harm: every one of them has a non-dominant, hence infinite-dimensional, fiber. For k a s \mathfrak{kas} kas , the situation is the opposite. The family 1a) has finite-dimensional fibers ( − k , l , l ) (-k,l,l) ( − k , l , l ) and ( 1 − k , l , l ) (1-k,l,l) ( 1 − k , l , l ) for every l ≥ 0 l \ge0 l ≥ 0 and k ≤ − l k \le-l k ≤ − l with − k , l -k,l − k , l both integers or both half-integers. This is a two-parameter family of honest operators between tensor fields with finite-dimensional fibers. Its member k = l = 0 k=l=0 k = l = 0 is the gradient, its member ( k , l ) = ( − 1 , 0 ) (k,l)=(-1,0) ( k , l ) = ( − 1 , 0 ) is an operator T ( Λ 1 ) ⟶ T ( S 0 2 ) \mathcal{T}(\Lambda^1) \longrightarrow\mathcal{T}(S^2_0) T ( Λ 1 ) ⟶ T ( S 0 2 ) , and so on. Not one of their adjoints is in the list.
The two families that are missing from the theorem of [GLS, §10] for kas and k(1|6), and where they should be inserted in it
They are
1 a † ) λ = ( k , k − 4 , l , − l ) ⟶ λ + ( − 1 , − 1 , 0 , 0 ) , 1 a ∗ † ) λ = ( k , − l − 1 , k − 3 , l ) ⟶ λ + ( − 1 , 0 , − 1 , 0 ) . (53) \begin{aligned}
1a^\dagger)\quad&\lambda=(k,k-4,l,-l)\longrightarrow\lambda+(-1,-1,0,0),\\
1a^{*\dagger})\quad&\lambda=(k,-l-1,k-3,l)\longrightarrow\lambda+(-1,0,-1,0). \tag*{(53)}
\end{aligned} 1 a † ) 1 a ∗† ) λ = ( k , k − 4 , l , − l ) ⟶ λ + ( − 1 , − 1 , 0 , 0 ) , λ = ( k , − l − 1 , k − 3 , l ) ⟶ λ + ( − 1 , 0 , − 1 , 0 ) . (53)
i.e., the † \dagger † -images of 1a) and 1a*) reparametrized. Where they should be inserted is determined by what they are. They belong in the very list from which they are absent, namely in the theorem of [35 §10] that describes the singular vectors for kas and k(1|6). There, they belong in its degree-1 part, as two further entries standing next to 1a)–1c), of the same degree 1, with the same two parameters k k k and l l l , and read in the same conventions. We call them 1a†) and 1a*†) because they are the adjoints of 1a) and 1a*); nothing else in that theorem changes, its other entries being closed under † already. The divergence is the member ( k , l ) = ( 5 , 0 ) (k,l)=(5,0) ( k , l ) = ( 5 , 0 ) of the first of them: ( 5 , 1 , 0 , 0 ) ⟶ ( 4 , 0 , 0 , 0 ) (5,1,0,0)\longrightarrow(4,0,0,0) ( 5 , 1 , 0 , 0 ) ⟶ ( 4 , 0 , 0 , 0 ) . The restriction that [35 §10] puts on 1a) — for k(1|6) only l = 0 l=0 l = 0 , for kas no restriction — is consistent with what we compute, since div is k(1|6)-invariant and sits at l = 0 l=0 l = 0 .
The discrepancy with [GLS, §10] is not due to our reading of its conventions.
One would rather suspect the dictionary (48) than a theorem, so we tried to break it. The reading in which the fiber attached to λ \lambda λ is dualized but the arrow is read as we read it is excluded outright. It makes [35 §10] predict operators in the three directions T ( S + ) ⟶ T ( S − ) \mathcal{T}(S^+) \longrightarrow\mathcal{T}(S^-) T ( S + ) ⟶ T ( S − ) , T ( Λ 2 ) ⟶ T ( Λ − 3 ) \mathcal{T}(\Lambda^2) \longrightarrow\mathcal{T}(\Lambda^3_-) T ( Λ 2 ) ⟶ T ( Λ − 3 ) and T ( Λ + 3 ) ⟶ T ( Λ 2 ) \mathcal{T}(\Lambda^3_+) \longrightarrow\mathcal{T}(\Lambda^2) T ( Λ + 3 ) ⟶ T ( Λ 2 ) . In those directions, we find nothing, for any weight, see Theorem 6.7.3. The only other reading is the reading obtained from ours by composing with † \dagger † itself, i.e., by reading the arrow the other way round. Nothing excludes that reading: the family 1b) is † \dagger † -stable, so it still matches the same three operators, while 1a) at k = l = 0 k=l=0 k = l = 0 then matches the divergence instead of the gradient. In other words, the ambiguity of the convention is precisely † \dagger † , and under either reading exactly one of the two operators
grad : F ⟶ T ( Λ 1 ) and div : T ( Λ 1 ) ⟶ F \operatorname{grad}:\mathcal{F}\longrightarrow\mathcal{T}(\Lambda^1)
\qquad\text{and}\qquad
\operatorname{div}:\mathcal{T}(\Lambda^1)\longrightarrow\mathcal{F} grad : F ⟶ T ( Λ 1 ) and div : T ( Λ 1 ) ⟶ F
is described by [35 §10] and the other is not. The discrepancy therefore survives every reading of the conventions we can think of.
Statement. (The two families are missing from the table, not from the mathematics).
(i) Every member of the family 1a†) of the list (53) with l = 0 l=0 l = 0 is a non-zero kas-invariant operator of order 1 between tensor fields with finite-dimensional fibers. In the shape (54), it is the divergence of a symmetric tensor field of rank m m m . Its members m = 1 m=1 m = 1 and m = 2 m=2 m = 2 are written out in formula (47) and in Statement 6.7.13(ii).
(ii) In the reading (48), none of these operators is described by [35 §10] . Indeed, neither family of the list (53) is of any of the forms 1a)–1c), 2a), 2a ), 2b) of that theorem, for any values of the parameters.*
(iii) The only other reading the dictionary admits is the reading that reverses the arrows; the third conceivable reading is excluded by Theorem 6.7.3. The reversing reading interchanges the members of each adjoint pair, so that there 1a) is the family that is not described. Hence, under either reading, exactly one family of each of the two pairs { 1 a ) , 1 a † ) } \{1a),1a^\dagger)\} { 1 a ) , 1 a † )} and { 1 a ∗ ) , 1 a ∗ † ) } \{1a*),1a^{*\dagger})\} { 1 a ∗ ) , 1 a ∗† )} is absent from that theorem.
So the answer is that the table is incomplete, and by two families of degree 1, whichever reading is taken; in the reading (49) they are the two (53) . Nothing else in that theorem changes, its other entries being closed under † \dagger † already.
A recomputation, independent of the conventions.
Statement 6.7.11 is established without deciding the convention. The computations behind [35 §10] are no longer available to be re-inspected. Therefore, we computed again, in a direction in which the outcome cannot depend on the dictionary (49) : we do not compare a weight with a table, we produce an operator.
Let l = 0 l=0 l = 0 ; this is the case to which [35 §10] restricts 1a) when the algebra is k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) , and the case of grad and div. Then, the fibers of 1a) and of 1a† ^\dagger † ) are the traceless symmetric powers S 0 m = S 0 m ( V 6 ) S^{m}_{0}=S^{m}_{0}(V_{6}) S 0 m = S 0 m ( V 6 ) of the vector representation of o ( 6 ) \mathfrak{o}(6) o ( 6 ) , those of highest weight ( m , 0 , 0 ) (m,0,0) ( m , 0 , 0 ) , of dimensions
dim S 0 m = 1 , 6 , 20 , 50 , 105 , … for m = 0 , 1 , 2 , 3 , 4 , … , \dim S^{m}_{0}=1,\ 6,\ 20,\ 50,\ 105,\ \ldots\qquad\text{for }m=0,\ 1,\ 2,\ 3,\ 4,\ \ldots, dim S 0 m = 1 , 6 , 20 , 50 , 105 , … for m = 0 , 1 , 2 , 3 , 4 , … ,
with S 0 0 = F S^{0}_{0}=\mathcal{F} S 0 0 = F and S 0 1 = Λ 1 S^{1}_{0}=\Lambda^{1} S 0 1 = Λ 1 . For m ≥ 2 m\ge2 m ≥ 2 these fibers do not occur in k a s \mathfrak{kas} kas itself, so they are new here; we construct them as the submodules of S m ( V 6 ) S^{m}(V_{6}) S m ( V 6 ) generated by the highest weight vectors.
In these terms, the two families read
1a) of [GLS, §10] : T ( S 0 m ) 2 m ⟶ T ( S 0 m + 1 ) 2 m + 2 , m ≥ 0 , 1a † ) of the list (53) : T ( S 0 m ) − 4 ⟶ T ( S 0 m − 1 ) − 4 , m ≥ 1. (54) \begin{aligned}
\text{1a) of [GLS, §10]}&:\quad\mathcal{T}(S^{m}_{0})_{2m}\longrightarrow\mathcal{T}(S^{m+1}_{0})_{2m+2},\qquad m\ge0,\\
\text{1a}^{\dagger}\text{) of the list (53)}&:\quad\mathcal{T}(S^{m}_{0})_{-4}\longrightarrow\mathcal{T}(S^{m-1}_{0})_{-4},\qquad m\ge1. \tag*{(54)}
\end{aligned} 1a) of [GLS, §10] 1a † ) of the list (53) : T ( S 0 m ) 2 m ⟶ T ( S 0 m + 1 ) 2 m + 2 , m ≥ 0 , : T ( S 0 m ) − 4 ⟶ T ( S 0 m − 1 ) − 4 , m ≥ 1. (54)
both of order 1: the symmetrized gradient and the divergence of a symmetric tensor field, superized. The members m = 0 m=0 m = 0 and m = 1 m=1 m = 1 are grad and div. The whole of the second family sits at the single weight λ = − 4 \lambda=-4 λ = − 4 , the weight of Vol \operatorname{Vol} Vol — as it must, the pairing into Vol \operatorname{Vol} Vol being what produces it.
Statement. The next member of either family
Take m = 2 m=2 m = 2 , i.e., the 20-dimensional fiber S 0 2 S^{2}_{0} S 0 2 . In order 1, and for k a s \mathfrak{kas} kas :
(i) the space of invariant operators T ( Λ 1 ) λ 1 ⟶ T ( S 0 2 ) λ 2 \mathcal{T}(\Lambda^{1})_{\lambda_{1}}\longrightarrow\mathcal{T}(S^{2}_{0})_{\lambda_{2}} T ( Λ 1 ) λ 1 ⟶ T ( S 0 2 ) λ 2 is non-zero for exactly one weight, namely ( λ 1 , λ 2 ) = ( 2 , 4 ) (\lambda_{1},\lambda_{2})=(2,4) ( λ 1 , λ 2 ) = ( 2 , 4 ) , where it is 1-dimensional. The operator is k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant. In the weights of [35 ] it is ( − 1 , 1 , 0 , 0 ) ⟶ ( − 2 , 2 , 0 , 0 ) (-1,1,0,0)\longrightarrow(-2,2,0,0) ( − 1 , 1 , 0 , 0 ) ⟶ ( − 2 , 2 , 0 , 0 ) , the member k = − 1 k=-1 k = − 1 , l = 0 l=0 l = 0 of 1a);
(ii) the space of invariant operators T ( S 0 2 ) λ 1 ⟶ T ( Λ 1 ) λ 2 \mathcal{T}(S^{2}_{0})_{\lambda_{1}}\longrightarrow\mathcal{T}(\Lambda^{1})_{\lambda_{2}} T ( S 0 2 ) λ 1 ⟶ T ( Λ 1 ) λ 2 is non-zero for exactly one weight, namely ( λ 1 , λ 2 ) = ( − 4 , − 4 ) (\lambda_{1},\lambda_{2})=(-4,-4) ( λ 1 , λ 2 ) = ( − 4 , − 4 ) , where it is 1-dimensional. The operator is k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) -invariant. In the weights of [35 ] , it is ( 6 , 2 , 0 , 0 ) ⟶ ( 5 , 1 , 0 , 0 ) (6,2,0,0)\longrightarrow(5,1,0,0) ( 6 , 2 , 0 , 0 ) ⟶ ( 5 , 1 , 0 , 0 ) , the member k = 6 k=6 k = 6 , l = 0 l=0 l = 0 of 1a† ^\dagger † ). It is of none of the forms 1a)–1c), 2a), 2a∗ ^* ∗ ), 2b) of [35 §10] , for any values of the parameters.
The weights were found the same way as in Theorem 6.7.3: K t K_t K t -invariance confines ( λ 1 , λ 2 ) (\lambda_{1},\lambda_{2}) ( λ 1 , λ 2 ) to a line, here λ 2 = λ 1 + 2 \lambda_{2}=\lambda_{1}+2 λ 2 = λ 1 + 2 in case (i) and λ 2 = λ 1 \lambda_{2}=\lambda_{1} λ 2 = λ 1 in case (ii). Then, λ 1 \lambda_{1} λ 1 was run through the half-integers of [ − 8 , 8 ] [-8,8] [ − 8 , 8 ] on that line. The ansatz has 42 terms in either case. Each of the two operators was then re-verified against all the generating functions of k a s \mathfrak{kas} kas and of k ( 1 ∣ 6 ) \mathfrak{k}(1|6) k ( 1∣6 ) , and each was checked to fail at the weights shifted by ± 1 2 \pm\frac{1}{2} ± 2 1 and by 1.
So the discrepancy is not a feature of the pair (grad, div): it repeats, with new fibers, new weights and a new linear system. Of the two operators of Statement 6.7.13, which are formally adjoint to each other, the first is in [35 §10] and the second is not, exactly as formula (54) predicts. The reading of the conventions in which the arrows are reversed exchanges the two. Hence, under either reading precisely one half of formula (54) is described by [35 §10] and the other half is not.
Remark. The densities carry nothing over kas either
Statement 6.4.2(i) says that there are no k a s \mathfrak{kas} kas -invariant non-scalar unary operators between the F λ \mathcal{F}_{\lambda} F λ , for $\operatorname{ord}\le 8. This agrees with [35 §10] , and independently of any computation: an operator between weighted densities is an operator whose two fibers are 1-dimensional, i.e., whose two weights have ( λ 1 , λ 2 , λ 3 ) = ( 0 , 0 , 0 ) (\lambda_1,\lambda_2,\lambda_3)=(0,0,0) ( λ 1 , λ 2 , λ 3 ) = ( 0 , 0 , 0 ) . And no entry of [35 §10] has this property. Indeed, 1a) would need ( − k , l , l ) = 0 = ( − k + 1 , l , l ) (-k,l,l)=0=(-k+1,l,l) ( − k , l , l ) = 0 = ( − k + 1 , l , l ) ; 1a*) would need l = 0 = l + 1 l=0=l+1 l = 0 = l + 1 ; 1b) would need ( l , l , 2 − k ) = 0 = ( l , l , 3 − k ) (l,l,2-k)=0=(l,l,3-k) ( l , l , 2 − k ) = 0 = ( l , l , 3 − k ) ; 1c) would need l = 0 l=0 l = 0 and − l − 2 = 0 -l-2=0 − l − 2 = 0 ; and the fibers of the three vectors of degree 2 are ( − 1 , − 1 , − 1 ) (-1,-1,-1) ( − 1 , − 1 , − 1 ) , ( − 1 , 0 , 0 ) (-1,0,0) ( − 1 , 0 , 0 ) , ( − 1 , − 1 , 1 ) (-1,-1,1) ( − 1 , − 1 , 1 ) .
Remark. What the passage to as does
Each of the four operators of Theorem 6.7.3 lives, over k a s \mathfrak{kas} kas , at exactly one pair of weights. These are the degeneracy conditions of [35 §10] , the super analog of the conditions in Rudakov’s description of the vect ( n ) \operatorname{vect}(n) vect ( n ) -invariant operators, see [Ru]. Over a s \mathfrak{as} as , all of them evaporate. By Lemma 6.6.2 , neither the weight nor the twist by a power of the determinant is visible. Hence, each of the four is invariant for every pair of weights, cf. Theorems 6.6.5 –6.6.7 . And a s \mathfrak{as} as has, besides, invariant operators of a shape no entry of [35 §10] can have: the powers ∂ t k \partial_t^k ∂ t k on every T ( V ) \mathcal{T}(V) T ( V ) , the composition grad ∘ div \operatorname{grad}\circ\operatorname{div} grad ∘ div , which is not a multiple of ∂ t \partial_t ∂ t , and the operators of orders 3 and 5 of Theorems 6.6.6 and 6.6.7 . So the pair a s ⊂ k a s \mathfrak{as}\subset\mathfrak{kas} as ⊂ kas trades a discrete set of degeneracy conditions for no condition at all — which is another way of saying what §6.3 says.
The other gradings: [§§11, 12]GLS.
Sections 11 and 12 of [35 ] describe the singular vectors for k a s ( 1 ∣ 6 ; 3 ξ ) \mathfrak{kas}(1|6;3\xi) kas ( 1∣6 ; 3 ξ ) and k a s ( 1 ∣ 6 ; 3 η ) \mathfrak{kas}(1|6;3\eta) kas ( 1∣6 ; 3 η ) , i.e., for the two other W-regradings of the same exceptional algebra. In these regradings, the algebra is realized not on the supercircle but on a superspace of superdimension 4 ∣ 4 4|4 4∣4 , resp. 4 ∣ 3 4|3 4∣3 . The tensor fields there are those of another g 0 \mathfrak{g}_0 g 0 , so those lists compare with nothing in §§6.4–6.6. However, they do compare with a computation of their own. Since the operators in different W-regradings of one algebra are the same operators written in other terms, that computation is an independent probe of the same tables. Here, it is.
The two regradings, rebuilt.
By [35 §2] , the regradings are
r = 3 ξ : deg ξ i = 0 , deg η i = deg t = 1 ; r = 3 η : deg η i = 0 , deg ξ i = deg t = 1. (55) \begin{aligned}
r=3\xi&: \quad\deg\xi_i=0,\ \deg\eta_i=\deg t=1;\\
r=3\eta&: \quad\deg\eta_i=0,\ \deg\xi_i=\deg t=1. \tag*{(55)}
\end{aligned} r = 3 ξ r = 3 η : deg ξ i = 0 , deg η i = deg t = 1 ; : deg η i = 0 , deg ξ i = deg t = 1. (55)
The algebra does not change, only the degree does. Hence, nothing has to be reconstructed: the same generating functions of k a s ⊂ k ( 1 ∣ 6 ) \mathfrak{kas}\subset\mathfrak{k}(1|6) kas ⊂ k ( 1∣6 ) that §6.6.9 uses, sorted by the new degree, give the new components at once. The outcome reproduces the table of [35 §2] exactly: both regradings are of depth 1, and
r = 3 ξ : g − 1 = Λ ( 3 ) ( sdim = ( 4 ∣ 4 ) ) , g 0 ( sdim 13 ∣ 10 ) = Λ ( 3 ) ⊕ s l ( 1 ∣ 3 ) , r = 3 η : g − 1 = Vol 0 ( 0 ∣ 3 ) ( sdim 4 ∣ 3 ) , g 0 ( sdim 13 ∣ 12 ) = c ( v e c t ( 0 ∣ 3 ) ) . (56) \begin{aligned}
r=3\xi&: \quad\mathfrak{g}_{-1}=\Lambda(3)\ (\operatorname{sdim}=(4|4)),\quad\mathfrak{g}_0\ (\operatorname{sdim}13|10)=\Lambda(3)\oplus\mathfrak{sl}(1|3),\\
r=3\eta&: \quad\mathfrak{g}_{-1}=\operatorname{Vol}_0(0|3)\ (\operatorname{sdim}4|3),\quad\mathfrak{g}_0\ (\operatorname{sdim}13|12)=\mathfrak{c}(\mathfrak{vect}(0|3)). \tag*{(56)}
\end{aligned} r = 3 ξ r = 3 η : g − 1 = Λ ( 3 ) ( sdim = ( 4∣4 )) , g 0 ( sdim 13∣10 ) = Λ ( 3 ) ⊕ sl ( 1∣3 ) , : g − 1 = Vol 0 ( 0∣3 ) ( sdim 4∣3 ) , g 0 ( sdim 13∣12 ) = c ( vect ( 0∣3 )) . (56)
In both cases g − 1 \mathfrak{g}_{-1} g − 1 is commutative, so U ( g − 1 ) = S ( g − 1 ) U(\mathfrak{g}_{-1})=S(\mathfrak{g}_{-1}) U ( g − 1 ) = S ( g − 1 ) and a singular vector of degree ord is an element of S ( g − 1 ) ⊗ V S(\mathfrak{g}_{-1})\otimes V S ( g − 1 ) ⊗ V killed by g 1 \mathfrak{g}_1 g 1 (which generates g + \mathfrak{g}_+ g + ) and highest for g 0 \mathfrak{g}_0 g 0 .
The fibers that are beyond dispute.
In §6.7.2, the entries of [35 §10] had to be sieved by dominance, since a weight names a fiber that is usually infinite-dimensional. Here, there is a class of fibers for which no such question arises at all: the 1-dimensional ones, i.e., the characters of g 0 \mathfrak{g}_0 g 0 . And (56) decides how many there are, without any appeal to [35 ] :
For r = 3 ξ r=3\xi r = 3 ξ , g 0 \mathfrak{g}_0 g 0 is perfect — we compute [ g 0 , g 0 ] = g 0 [\mathfrak{g}_0,\mathfrak{g}_0]=\mathfrak{g}_0 [ g 0 , g 0 ] = g 0 , of dimension 23 — so the only 1-dimensional g 0 \mathfrak{g}_0 g 0 -module is the trivial module.
For r = 3 η r=3\eta r = 3 η , [ g 0 , g 0 ] [\mathfrak{g}_0,\mathfrak{g}_0] [ g 0 , g 0 ] has dimension 24 and is v e c t ( 0 ∣ 3 ) \mathfrak{vect}(0|3) vect ( 0∣3 ) , so the characters form a line: in the basis ( K t , K ξ i η i ) (K_t,K_{\xi_i\eta_i}) ( K t , K ξ i η i ) of the maximal toral subalgebra t \mathfrak{t} t they are λ ⋅ ( 1 , 1 , 1 , 1 ) \lambda\cdot(1,1,1,1) λ ⋅ ( 1 , 1 , 1 , 1 ) , and Vol \operatorname{Vol} Vol is the member λ = − 1 \lambda=-1 λ = − 1 .
Over such a fiber, every m i m_i m i with i > 1 i > 1 i > 1 of [35 §§11, 12] is zero , those being obtained from m 1 m_1 m 1 by lowering operators of g 0 \mathfrak{g}_0 g 0 . Hence, an entry of [35 ] survives there if and only if its source weight meets the line, the coefficient of its pure-m 1 m_1 m 1 part does not vanish at that point, and its own side conditions hold.
The dictionary between the two labellings is not guessed either. It is read off the weight shifts , which we can compute: the shift is the t \mathfrak{t} t -weight of the g − 1 \mathfrak{g}_{-1} g − 1 -part of the printed singular vector. It comes out over-determined: the six entries 1a), 1b), 1c), 1d), 1g), 2a) of [35 §12] all give the same answer, namely that the λ \lambda λ of [35 ] is
( λ 1 , λ 2 , λ 3 , λ 4 ) = ( μ 0 , − μ 2 , − μ 1 , − μ 3 ) , μ = the weight relative ( K t , K ξ 1 η 1 , K ξ 2 η 2 , K ξ 3 η 3 ) (57) (\lambda_1,\lambda_2,\lambda_3,\lambda_4)=(\mu_0,-\mu_2,-\mu_1,-\mu_3),\qquad\mu=\text{the weight relative }(K_t,K_{\xi_1\eta_1},K_{\xi_2\eta_2},K_{\xi_3\eta_3}) \tag*{(57)} ( λ 1 , λ 2 , λ 3 , λ 4 ) = ( μ 0 , − μ 2 , − μ 1 , − μ 3 ) , μ = the weight relative ( K t , K ξ 1 η 1 , K ξ 2 η 2 , K ξ 3 η 3 ) (57)
(One by-product: the first coordinate of [35 §12, 1a)] reads k ⟶ l k \longrightarrow l k ⟶ l , which no constant shift can do, and which every other entry of that section has as k ⟶ k k \longrightarrow k k ⟶ k ; we take it for a misprint for k ⟶ k k \longrightarrow k k ⟶ k .)
Theorem. The 1-dimensional fibers in the other gradings
Theorem 6.7.19. Over the 1-dimensional fibers, the singular vectors are these.
(i) For r = 3 ξ r=3\xi r = 3 ξ , over the trivial fiber — the only one — there is exactly one singular vector of degree 1, namely ξ 1 ξ 2 ξ 3 m 1 \xi_1\xi_2\xi_3m_1 ξ 1 ξ 2 ξ 3 m 1 , none at all of degree 2, exactly one of degree 3, and none of degree 4.
(ii) For r = 3 η r=3\eta r = 3 η , over the character λ ⋅ ( 1 , 1 , 1 , 1 ) \lambda\cdot(1,1,1,1) λ ⋅ ( 1 , 1 , 1 , 1 ) there is, for each 1 ≤ m ≤ 4 1\leq m\leq4 1 ≤ m ≤ 4 , exactly one singular vector of degree m m m , namely
( η 1 η 3 ) m m 1 at λ = − m − 1 2 , i.e., at λ 3 + λ 4 = m − 1 in the weights of [GLS], (58) (\eta_1\eta_3)^m m_1\quad\text{at}\quad\lambda=-\frac{m-1}{2},\qquad\text{i.e., at }\lambda_3+\lambda_4=m-1\text{ in the weights of [GLS],} \tag*{(58)} ( η 1 η 3 ) m m 1 at λ = − 2 m − 1 , i.e., at λ 3 + λ 4 = m − 1 in the weights of [GLS], (58) and no singular vector of degree m m m at any other value of λ \lambda λ .
What the comparison gives.
Running the sieve of §6.7.18 over the two lists, and comparing with Theorem 6.7.19 :
Section 11 is confirmed on everything that can be tested. Of its fourteen entries, exactly one survives at the trivial fiber — 1a), ξ 1 ξ 2 ξ 3 m 1 \xi_1\xi_2\xi_3m_1 ξ 1 ξ 2 ξ 3 m 1 , which [35 ] give for any λ \lambda λ — and it is what we find. Of its nine entries of degree 2, six never meet λ = 0 \lambda=0 λ = 0 at all, and the remaining three — 2c), 2d), 2f) — meet it with a vanishing m 1 m_1 m 1 -coefficient. Hence, [35 §11] predicts nothing in degree 2, and we find nothing. The degree-3 vector of Theorem 6.7.19 (i) is outside the range of that section, which lists degrees 1 and 2 only.
Section 12 is confirmed in degree 1 and is off by one in degree 2. Its entry 1a) survives at λ = 0 \lambda=0 λ = 0 with the vector η 1 η 3 m 1 \eta_1\eta_3m_1 η 1 η 3 m 1 , and that is exactly the m = 1 m=1 m = 1 member of the family (58). In degree 2 the sieve leaves exactly one entry, 2a), whose vector is ( η 1 η 3 ) 2 m 1 (\eta_1\eta_3)^2m_1 ( η 1 η 3 ) 2 m 1 — the m = 2 m=2 m = 2 member of the family (58) — and whose shift, λ ⟶ λ + ( 0 , 0 , − 2 , − 2 ) \lambda\longrightarrow\lambda+(0,0,-2,-2) λ ⟶ λ + ( 0 , 0 , − 2 , − 2 ) , is also exactly ours. But [35 §12, 2a)] put it at λ 4 = 2 − λ 3 \lambda_4=2-\lambda_3 λ 4 = 2 − λ 3 , whereas our computation puts it at λ 4 = 1 − λ 3 \lambda_4=1-\lambda_3 λ 4 = 1 − λ 3 ; and at λ 4 = 2 − λ 3 \lambda_4=2-\lambda_3 λ 4 = 2 − λ 3 there is indeed a singular vector, but it is ( η 1 η 3 ) 3 m 1 (\eta_1\eta_3)^3m_1 ( η 1 η 3 ) 3 m 1 , of degree 3. So the entry has the right shape, the right degree and the right shift, and the source condition of the next member of its own family.
Remark. The other gradings
Remark 6.7.21. Two things this says about §6.7.6. First, the probe is genuinely independent: it uses another g 0 \mathfrak{g}_0 g 0 , another g − 1 \mathfrak{g}_{-1} g − 1 , another induced module. It also tests the tables at fibers of a kind — the 1-dimensional ones — that plays no role in §6.7.6, where the pair at issue is ( grad , div ) (\operatorname{grad},\operatorname{div}) ( grad , div ) with fibers F \mathcal{F} F and Λ 1 \Lambda^1 Λ 1 . Second, it comes back with a defect of its own, and of a quite different species. Indeed, §6.7.6 exhibits a whole family absent from [35 §10] , whereas [35 §12, 2a)] is an entry that is present and carries the parameter condition of its neighbor. Neither is evidence against the other. Together, they say that the three tables of [35 ] are not a closed system that checks itself, which is a reason to read the omission of
§6.7.6 as an omission rather than as our misreading. None of this is a reproach: the tables are a remarkable computation, and it is only because they are as detailed as they are that they can be tested at all.
What is not settled by this. Our sieve reaches the 1-dimensional fibers only, so we have not tested the two lists for completeness at larger fibers — the analogue, in these gradings, of the missing family (53). And (58) is verified along the line of the characters; that the general member of the family is λ 4 = ( m − 1 ) − λ 3 \lambda_4=(m-1)-\lambda_3 λ 4 = ( m − 1 ) − λ 3 for arbitrary λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 , as the shape of 1a) and 2a) suggests, we conjecture but have not computed.
Open problems for n = 6 n=6 n = 6
The binary orders ord ≥ 6 \operatorname{ord}\geq6 ord ≥ 6 for a s \mathfrak{as} as between densities.
For the tensor fields of § 6.6 : the orders ord ≥ 7 \operatorname{ord}\geq7 ord ≥ 7 ; the binary operators of order ≥ 2 \geq2 ≥ 2 ; the remaining module V = s l ( 4 ) = Λ 2 V=\mathfrak{sl}(4)=\Lambda^2 V = sl ( 4 ) = Λ 2 , the adjoint; and the same questions for k a s \mathfrak{kas} kas and for i ( o s p ( 6 ∣ 2 ) ) i(\mathfrak{osp}(6|2)) i ( osp ( 6∣2 )) , where — unlike for a s \mathfrak{as} as — the weight and the determinant twist are not invisible , so that each V V V carries a two-parameter family of modules.
And the remaining case s l ( 1 ∣ m ∣ n ∣ 1 ) ⊂ k ( 2 m + 1 ∣ 2 n ) \mathfrak{sl}(1|m|n|1)\subset\mathfrak{k}(2m+1|2n) sl ( 1∣ m ∣ n ∣1 ) ⊂ k ( 2 m + 1∣2 n ) of [14 Table (4)] .
Super GRC operators: the p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2)\subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 ) case
The periplectic Lie superalgebra
The periplectic Lie superalgebra p e ( n ) \mathfrak{pe}(n) pe ( n ) is the Lie superalgebra that preserves the odd antisymmetric bilinear form with Gram matrix
( 0 1 n 1 n 0 ) \begin{pmatrix} 0 & 1_n \\ 1_n & 0 \end{pmatrix} ( 0 1 n 1 n 0 )
see [12 ] .
Realization of the elements of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) by vector fields in m ( 1 ) \mathfrak{m}(1) m ( 1 ) .
By applying Shchepochkina’s algorithm [49 ] , we see that
p e ( 2 ) ‾ 0 = Span { M ξ = ξ ∂ τ + ∂ q , M q τ = q τ ∂ τ − τ ∂ ξ + q E , M τ + q ξ = 2 τ ∂ τ + 2 q ∂ q , M τ − q ξ = 2 τ ∂ τ + 2 ξ ∂ ξ } ≃ g l ( 2 ) \overline{\mathfrak{pe}(2)}_0=\operatorname{Span}\{M_\xi=\xi\partial_\tau+\partial_q,\ M_{q\tau}=q\tau\partial_\tau-\tau\partial_\xi+qE,\ M_{\tau+q\xi}=2\tau\partial_\tau+2q\partial_q,\ M_{\tau-q\xi}=2\tau\partial_\tau+2\xi\partial_\xi\}\simeq\mathfrak{gl}(2) pe ( 2 ) 0 = Span { M ξ = ξ ∂ τ + ∂ q , M q τ = q τ ∂ τ − τ ∂ ξ + q E , M τ + q ξ = 2 τ ∂ τ + 2 q ∂ q , M τ − q ξ = 2 τ ∂ τ + 2 ξ ∂ ξ } ≃ gl ( 2 )
p e ( 2 ) ‾ 1 = Span { M 1 = 2 ∂ τ , M q = q ∂ τ − ∂ ξ , M q 2 = − 2 q ∂ ξ , M τ ξ = ( τ − q ξ ) ∂ q + τ ξ ∂ τ } , \overline{\mathfrak{pe}(2)}_1=\operatorname{Span}\{M_1=2\partial_\tau,\ M_q=q\partial_\tau-\partial_\xi,\ M_{q^2}=-2q\partial_\xi,\ M_{\tau\xi}=(\tau-q\xi)\partial_q+\tau\xi\partial_\tau\}, pe ( 2 ) 1 = Span { M 1 = 2 ∂ τ , M q = q ∂ τ − ∂ ξ , M q 2 = − 2 q ∂ ξ , M τ ξ = ( τ − q ξ ) ∂ q + τ ξ ∂ τ } ,
where E = q ∂ q + ξ ∂ ξ E=q\partial_q+\xi\partial_\xi E = q ∂ q + ξ ∂ ξ and the M f M_f M f are the vector fields (59) of §7.2.
The Lie superalgebra p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) is not simple; its second derived is given by
p e ( 2 ) ( 2 ) ‾ 0 = Span { M ξ = ξ ∂ τ + ∂ q , M τ + q ξ = 2 τ ∂ τ + 2 q ∂ q , M q τ = q τ ∂ τ − τ ∂ ξ + q E } ≃ s l ( 2 ) , \overline{\mathfrak{pe}^{(2)}(2)}_0=\operatorname{Span}\{M_\xi=\xi\partial_\tau+\partial_q,\ M_{\tau+q\xi}=2\tau\partial_\tau+2q\partial_q,\ M_{q\tau}=q\tau\partial_\tau-\tau\partial_\xi+qE\}\simeq\mathfrak{sl}(2), pe ( 2 ) ( 2 ) 0 = Span { M ξ = ξ ∂ τ + ∂ q , M τ + q ξ = 2 τ ∂ τ + 2 q ∂ q , M q τ = q τ ∂ τ − τ ∂ ξ + q E } ≃ sl ( 2 ) ,
p e ( 2 ) ( 2 ) ‾ 1 = Span { M 1 = 2 ∂ τ , M q = q ∂ τ − ∂ ξ , M q 2 = − 2 q ∂ ξ } . \overline{\mathfrak{pe}^{(2)}(2)}_1=\operatorname{Span}\{M_1=2\partial_\tau,\ M_q=q\partial_\tau-\partial_\xi,\ M_{q^2}=-2q\partial_\xi\}. pe ( 2 ) ( 2 ) 1 = Span { M 1 = 2 ∂ τ , M q = q ∂ τ − ∂ ξ , M q 2 = − 2 q ∂ ξ } .
The span of the eight generating functions above is closed under the bracket, and p e ( 2 ) ( 2 ) \mathfrak{pe}^{(2)}(2) pe ( 2 ) ( 2 ) is as displayed. Observe that M τ ξ = ( τ − q ξ ) ∂ q + τ ξ ∂ τ M_{\tau\xi}=(\tau-q\xi)\partial_q+\tau\xi\partial_\tau M τ ξ = ( τ − q ξ ) ∂ q + τ ξ ∂ τ and τ ∂ q − ξ ( q ∂ q + τ ∂ τ ) \tau\partial_q-\xi(q\partial_q+\tau\partial_\tau) τ ∂ q − ξ ( q ∂ q + τ ∂ τ ) , the shape in which M τ ξ M_{\tau\xi} M τ ξ is used in the proof of Theorem 7.3 below, are one and the same vector field.
Conventions for m ( 1 ) \mathfrak{m}(1) m ( 1 )
Again we follow [14 ] ; recall the description of the 2-parameter family of weighted densities over m ( n ) \mathfrak{m}(n) m ( n ) , see [14 §2.2.2] . The pericontact form is α 0 = d τ + ξ d q + q d ξ \alpha_0=d\tau+\xi\,dq+q\,d\xi α 0 = d τ + ξ d q + q d ξ and, with E = q ∂ q + ξ ∂ ξ E=q\partial_q+\xi\partial_\xi E = q ∂ q + ξ ∂ ξ and L e f = ∂ q ( f ) ∂ ξ + ( − 1 ) p ( f ) ∂ ξ ( f ) ∂ q L\!e_f=\partial_q(f)\partial_\xi+(-1)^{p(f)}\partial_\xi(f)\partial_q L e f = ∂ q ( f ) ∂ ξ + ( − 1 ) p ( f ) ∂ ξ ( f ) ∂ q ,
M f = ( 2 − E ) ( f ) ∂ τ − L e f − ( − 1 ) p ( f ) ∂ τ ( f ) E , (59) M_f=(2-E)(f)\partial_\tau-L\!e_f-(-1)^{p(f)}\partial_\tau(f)E, \tag*{(59)} M f = ( 2 − E ) ( f ) ∂ τ − L e f − ( − 1 ) p ( f ) ∂ τ ( f ) E , (59)
see [14 (26), (27)] , while D 1 = ∂ τ D_1=\partial_\tau D 1 = ∂ τ , D ξ = ξ ∂ τ − ∂ q D_\xi=\xi\partial_\tau-\partial_q D ξ = ξ ∂ τ − ∂ q and D q = q ∂ τ + ∂ ξ D_q=q\partial_\tau+\partial_\xi D q = q ∂ τ + ∂ ξ are the operators (2) . Recall that p ( D f ) = p ( f ) + 1 p(D_f)=p(f)+1 p ( D f ) = p ( f ) + 1 , so that D ξ D_\xi D ξ is even whereas D q D_q D q and D 1 D_1 D 1 are odd. The weighted densities are F a , b = F α 0 a vol b \mathcal{F}_{a,b}=\mathcal{F}\alpha_0^a\operatorname{vol}^b F a , b = F α 0 a vol b , where vol = vol ( τ , q ∣ ξ ) \operatorname{vol}=\operatorname{vol}(\tau,q|\xi) vol = vol ( τ , q ∣ ξ ) is the volume element, with
L M f ( φ ) = M f ( φ ) + 2 ( − 1 ) p ( f ) ( a ∂ τ ( f ) + b Δ m ( f ) ) φ , Δ m = ( 1 − E ) ∂ τ − ∂ 2 ∂ ξ ∂ q , (60) L_{M_f}(\varphi)=M_f(\varphi)+2(-1)^{p(f)}(a\partial_\tau(f)+b\Delta_m(f))\varphi,\qquad\Delta_m=(1-E)\partial_\tau-\frac{\partial^2}{\partial\xi\,\partial q}, \tag*{(60)} L M f ( φ ) = M f ( φ ) + 2 ( − 1 ) p ( f ) ( a ∂ τ ( f ) + b Δ m ( f )) φ , Δ m = ( 1 − E ) ∂ τ − ∂ ξ ∂ q ∂ 2 , (60)
see [14 (37), (38)] . Here, deg D ξ = deg D q = − 1 \deg D_{\xi}=\deg D_q=-1 deg D ξ = deg D q = − 1 and deg D 1 = − 2 \deg D_1=-2 deg D 1 = − 2 , while the charge, the weight relative M q ξ = q ∂ q − ξ ∂ ξ M_{q\xi}=q\partial_q-\xi\partial_\xi M q ξ = q ∂ q − ξ ∂ ξ , is ch D ξ = − 1 \operatorname{ch}D_{\xi}=-1 ch D ξ = − 1 , ch D q = + 1 \operatorname{ch}D_q=+1 ch D q = + 1 , ch D 1 = 0 \operatorname{ch}D_1=0 ch D 1 = 0 . Hence, for an operator of degree s s s and charge w w w ,
a = a 1 + a 2 + s + w 2 , b = b 1 + b 2 − w 2 . (61) a=a_1+a_2+\frac{s+w}{2},\qquad b=b_1+b_2-\frac{w}{2}. \tag*{(61)} a = a 1 + a 2 + 2 s + w , b = b 1 + b 2 − 2 w . (61)
Who is the Bol operator here.
It is the operator of [14 Thm 5.1] , which is a polynomial in the D f D_f D f of eq. (2), as Theorem 2.1 requires:
Bol k = D q ( − D ξ ) k − 1 + ( k − 1 ) ( − D ξ ) k − 2 ∂ τ , \operatorname{Bol}_k=D_q(-D_{\xi})^{k-1}+(k-1)(-D_{\xi})^{k-2}\partial_{\tau}, Bol k = D q ( − D ξ ) k − 1 + ( k − 1 ) ( − D ξ ) k − 2 ∂ τ ,
Bol 1 = D q , Bol 2 = − ( D ξ D q + ∂ τ ) , Bol 3 = D ξ 2 D q + 2 D ξ ∂ τ . (62) \operatorname{Bol}_1=D_q,\quad\operatorname{Bol}_2=-(D_{\xi}D_q+\partial_{\tau}),\quad\operatorname{Bol}_3=D_{\xi}^{2}D_q+2D_{\xi}\partial_{\tau}. \tag*{(62)} Bol 1 = D q , Bol 2 = − ( D ξ D q + ∂ τ ) , Bol 3 = D ξ 2 D q + 2 D ξ ∂ τ . (62)
Each Bol k \operatorname{Bol}_k Bol k is odd, of order ord = k \operatorname{ord}=k ord = k , hence of degree s = − k s=-k s = − k , and of charge 2 − k 2-k 2 − k . It should not be confused with ∂ q k \partial_q^k ∂ q k : the operator ∂ q k \partial_q^k ∂ q k is not i ( p e ( 2 ) ) i(\mathfrak{pe}(2)) i ( pe ( 2 )) -invariant F a , b ⟶ F a ′ , b ′ \mathcal{F}_{a,b}\longrightarrow\mathcal{F}_{a',b'} F a , b ⟶ F a ′ , b ′ for any weights whatsoever — indeed, ∂ q \partial_q ∂ q is not even p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − -invariant, since it is not a polynomial in D 1 , D ξ , D q D_1,D_{\xi},D_q D 1 , D ξ , D q , cf. Theorem 2.1. (Note that D ξ = ξ ∂ τ − ∂ q D_{\xi}=\xi\partial_{\tau}-\partial_q D ξ = ξ ∂ τ − ∂ q , so ∂ q k \partial_q^k ∂ q k is only the “leading term” of ( − D ξ ) k (-D_{\xi})^k ( − D ξ ) k .)
In formula (62), as everywhere below, a product of operators is their composition in the usual order, the right factor acting first. This is not a pedantry: the invariant derivatives do not commute,
[ D ξ , D q ] = − 2 ∂ τ , D q 2 = 0 , ∂ τ 2 = 0 , ∂ τ supercommutes with D ξ and with D q , (63) [D_{\xi},D_q]=-2\partial_{\tau},\qquad D_q^2=0,\qquad\partial_{\tau}^2=0,\qquad\partial_{\tau}\text{ supercommutes with }D_{\xi}\text{ and with }D_q, \tag*{(63)} [ D ξ , D q ] = − 2 ∂ τ , D q 2 = 0 , ∂ τ 2 = 0 , ∂ τ supercommutes with D ξ and with D q , (63)
so that reading the first summand of formula (62) as ( − D ξ ) k − 1 D q (-D_{\xi})^{k-1}D_q ( − D ξ ) k − 1 D q , i.e., letting D q D_q D q act first, yields another operator altogether: one invariant at a = − 1 2 a=-\frac{1}{2} a = − 2 1 instead of 1 2 \frac{1}{2} 2 1 for k = 2 k=2 k = 2 , and at no weight whatsoever for $k\geq 3.
The complete classification, for every order and every weight, is given in §8; Theorem 7.3 and Statement 7.4 below describe the lowest orders and the way the Bol operators of [14 ] enter.
Theorem. The GRC brackets over pe(2)m(1) are the Bol operators acting in one argument
Theorem 7.3. (The GRC brackets over p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2)\subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 ) are the Bol operators acting in one argument.) Let sdim M = ( 1 ∣ 2 ) \operatorname{sdim}\mathcal{M}=(1|2) sdim M = ( 1∣2 ) and M \mathcal{M} M be endowed with a pericontact structure. Let φ \varphi φ (resp. ψ \psi ψ ) be the coefficient of α 0 λ 1 vol ν 1 ∈ F λ 1 , ν 1 \alpha_0^{\lambda_1}\operatorname{vol}^{\nu_1}\in\mathcal{F}_{\lambda_1,\nu_1} α 0 λ 1 vol ν 1 ∈ F λ 1 , ν 1 (resp. of α 0 λ 2 vol ν 2 ∈ F λ 2 , ν 2 \alpha_0^{\lambda_2}\operatorname{vol}^{\nu_2}\in\mathcal{F}_{\lambda_2,\nu_2} α 0 λ 2 vol ν 2 ∈ F λ 2 , ν 2 ).
(i) For every integer k ≥ 1 k\geq1 k ≥ 1 , there are two odd i ( p e ( 2 ) ) i(\mathfrak{pe}(2)) i ( pe ( 2 )) -invariant non-zero non-scalar bilinear differential operators of order ord = k \operatorname{ord}=k ord = k
J l k ( φ , ψ ) \mathcal{J}_l^k(\varphi,\psi) J l k ( φ , ψ ) and J r k ( φ , ψ ) : F k − 1 2 , ν 1 ⊗ F k − 1 2 , ν 2 ⟶ F 0 , ω \mathcal{J}_r^k(\varphi,\psi):\mathcal{F}_{\frac{k-1}{2},\nu_1}\otimes\mathcal{F}_{\frac{k-1}{2},\nu_2}\longrightarrow\mathcal{F}_{0,\omega} J r k ( φ , ψ ) : F 2 k − 1 , ν 1 ⊗ F 2 k − 1 , ν 2 ⟶ F 0 , ω , where ω = ν 1 + ν 2 + k 2 − 1 \omega=\nu_1+\nu_2+\frac{k}{2}-1 ω = ν 1 + ν 2 + 2 k − 1 ,
namely
J l k ( φ , ψ ) = Bol k ( φ ) ψ , J r k ( φ , ψ ) = ( − 1 ) p ( φ ) φ Bol k ( ψ ) , \mathcal{J}_l^k(\varphi,\psi)=\operatorname{Bol}_k(\varphi)\psi,\quad\mathcal{J}_r^k(\varphi,\psi)=(-1)^{p(\varphi)}\varphi\operatorname{Bol}_k(\psi), J l k ( φ , ψ ) = Bol k ( φ ) ψ , J r k ( φ , ψ ) = ( − 1 ) p ( φ ) φ Bol k ( ψ ) ,
where Bol k \operatorname{Bol}_k Bol k is the Bol operator (62). More precisely, J k , l \mathcal{J}_{k,l} J k , l alone is invariant as soon as λ 1 = k − 1 2 \lambda_1=\frac{k-1}{2} λ 1 = 2 k − 1 , the weights λ 2 , ν 1 , ν 2 \lambda_2,\nu_1,\nu_2 λ 2 , ν 1 , ν 2 being arbitrary, the target being then F λ 2 + 1 − k 2 , ν 1 + ν 2 + k 2 − 1 \mathcal{F}_{\lambda_2+\frac{1-k}{2},\nu_1+\nu_2+\frac{k}{2}-1} F λ 2 + 2 1 − k , ν 1 + ν 2 + 2 k − 1 ; symmetrically for J k , r \mathcal{J}_{k,r} J k , r . Both are invariant simultaneously precisely at λ 1 = λ 2 = k − 1 2 \lambda_1=\lambda_2=\frac{k-1}{2} λ 1 = λ 2 = 2 k − 1 .
(ii) These are not all the operators, see Statement 7.4.
Proof. Since vol is the volume element, the weights ( λ i , ν i ) (\lambda_i,\nu_i) ( λ i , ν i ) used in the statement are literally the weights ( a i , b i ) (a_i,b_i) ( a i , b i ) of formula (60): a i = λ i a_i=\lambda_i a i = λ i and b i = ν i b_i=\nu_i b i = ν i . By [14 Thm 5.1] , the operator Bol k : F a , b ⟶ F a + 1 − k , b − 1 + k / 2 \operatorname{Bol}_k:\mathcal{F}_{a,b}\longrightarrow\mathcal{F}_{a+1-k,b-1+k/2} Bol k : F a , b ⟶ F a + 1 − k , b − 1 + k /2 — the target being the space prescribed by formula (61), since deg Bol k = − k \deg\operatorname{Bol}_k=-k deg Bol k = − k and ch Bol k = 2 − k \operatorname{ch}\operatorname{Bol}_k=2-k ch Bol k = 2 − k — is invariant if and only if a = k − 1 2 a=\frac{k-1}{2} a = 2 k − 1 , the parameter b b b remaining free. The multiplication F a , b ⊗ F a ′ , b ′ ⟶ F a + a ′ , b + b ′ \mathcal{F}_{a,b}\otimes\mathcal{F}_{a',b'}\longrightarrow\mathcal{F}_{a+a',b+b'} F a , b ⊗ F a ′ , b ′ ⟶ F a + a ′ , b + b ′ is invariant for all weights. Composing the two gives the statement; the first weight of the target of J k , l \mathcal{J}_{k,l} J k , l and J k , r \mathcal{J}_{k,r} J k , r is k − 1 2 + k − 1 2 + 1 − k = 0 \frac{k-1}{2}+\frac{k-1}{2}+1-k=0 2 k − 1 + 2 k − 1 + 1 − k = 0 . The sign ( − 1 ) p ( φ ) (-1)^{p(\varphi)} ( − 1 ) p ( φ ) in J k , r \mathcal{J}_{k,r} J k , r is the Koszul sign of moving the odd
operator Bol k \operatorname{Bol}_k Bol k past φ \varphi φ , and it is indispensable: without it, φ Bol k ( ψ ) \varphi\operatorname{Bol}_k(\psi) φ Bol k ( ψ ) is invariant for no weights whatsoever. □
Statement. The complete list over p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2)\subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 )
(i) At the weights of Theorem 7.3 , i.e., at λ 1 = λ 2 = k − 1 2 \lambda_1=\lambda_2=\frac{k-1}{2} λ 1 = λ 2 = 2 k − 1 with ν 1 , ν 2 \nu_1,\nu_2 ν 1 , ν 2 free, the space of i ( p e ( 2 ) ) \mathfrak{i}(\mathfrak{pe}(2)) i ( pe ( 2 )) -invariant bilinear operators of order ord = k \operatorname{ord}=k ord = k and charge 2 − k 2-k 2 − k is of dimension
2 for k = 1 , 3 for k = 2 , 2 for k = 3 ; 2 \text{ for } k=1,\qquad3 \text{ for } k=2,\qquad2 \text{ for } k=3; 2 for k = 1 , 3 for k = 2 , 2 for k = 3 ;
for k = 1 k=1 k = 1 and k = 3 k=3 k = 3 it is spanned by J k , l \mathcal{J}_{k,l} J k , l and J k , r \mathcal{J}_{k,r} J k , r , as stated in Theorem 7.3 . For k = 2 k=2 k = 2 there is a third, independent, invariant operator, namely
∂ τ ⊗ 1 + 1 ⊗ ∂ τ + D ξ ⊗ D q + D q ⊗ D ξ . \partial_\tau\otimes1+1\otimes\partial_\tau+D_\xi\otimes D_q+D_q\otimes D_\xi. ∂ τ ⊗ 1 + 1 ⊗ ∂ τ + D ξ ⊗ D q + D q ⊗ D ξ .
(ii) At generic weights, the space of i ( p e ( 2 ) ) \mathfrak{i}(\mathfrak{pe}(2)) i ( pe ( 2 )) -invariant bilinear operators is of dimension
1 in order 1 , 3 in order 2 , 1 in order 3 for ch = − 1 , 2 in order 3 for ch = + 1. 1\text{ in order }1,\qquad3\text{ in order }2,\qquad1\text{ in order }3\text{ for ch}=-1,\qquad2\text{ in order }3\text{ for ch}=+1. 1 in order 1 , 3 in order 2 , 1 in order 3 for ch = − 1 , 2 in order 3 for ch = + 1.
In order 1 1 1 , the unique operator is
λ 2 D q ( φ ) ψ − ( − 1 ) p ( φ ) λ 1 φ D q ( ψ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 , ν 1 + ν 2 − 1 2 (64) \lambda_2D_q(\varphi)\psi-(-1)^{p(\varphi)}\lambda_1\varphi D_q(\psi):\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2,\nu_1+\nu_2-\frac{1}{2}} \tag*{(64)} λ 2 D q ( φ ) ψ − ( − 1 ) p ( φ ) λ 1 φ D q ( ψ ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 , ν 1 + ν 2 − 2 1 (64)
it degenerates into J 1 , l = D q ⊗ 1 \mathcal{J}_{1,l}=D_q\otimes1 J 1 , l = D q ⊗ 1 exactly when λ 1 = 0 \lambda_1=0 λ 1 = 0 and into J 1 , r \mathcal{J}_{1,r} J 1 , r exactly when λ 2 = 0 \lambda_2=0 λ 2 = 0 , and both J 1 , l \mathcal{J}_{1,l} J 1 , l and J 1 , r \mathcal{J}_{1,r} J 1 , r are invariant only at λ 1 = λ 2 = 0 \lambda_1=\lambda_2=0 λ 1 = λ 2 = 0 , which is the locus of Theorem 7.3 for k = 1 k=1 k = 1 .
Both parts are obtained by computation with all eight generating functions of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) .
A correction to [Thm 5.1]BLS1 itself
[14 ] lists, for each order k k k , the Bol operator Bol k \operatorname{Bol}_k Bol k over m ( 1 ) \mathfrak{m}(1) m ( 1 ) together with the weights on which it is invariant, namely a = k − 1 2 a=\frac{k-1}{2} a = 2 k − 1 with b b b free . At* k = 2 k=2 k = 2 , that list is incomplete: there, the invariant operators form a 1-parameter family, of which the operator printed in [14 ] is one member. As far as we know, this gap is filled nowhere else — [14 ] has appeared in print and has not been corrected since — so we record the answer here.
Statement. A correction to Theorem 5.1 of BLS1 at k=2
(i) The order-2 Bol operator over m ( 1 ) \mathfrak{m}(1) m ( 1 ) is not isolated:
Bol ( a ) : = D ξ D q + 2 a ∂ τ : F a , b ⟶ F a − 1 , b is invariant for every a and b , (65) \operatorname{Bol}^{(a)}:=D_\xi D_q+2a\,\partial_\tau:\mathcal{F}_{a,b}\longrightarrow\mathcal{F}_{a-1,b}\text{ is invariant for every }a\text{ and }b, \tag*{(65)} Bol ( a ) := D ξ D q + 2 a ∂ τ : F a , b ⟶ F a − 1 , b is invariant for every a and b , (65)
whereas [14 ] lists only its member a = 1 2 a=\frac{1}{2} a = 2 1 , which is − Bol 2 -\operatorname{Bol}_2 − Bol 2 . Moreover, eq. (65) exhausts the invariant operators of that degree, charge and parity: their space is 1-dimensional at every weight ( a , b ) (a,b) ( a , b ) .
(ii) For k = 1 , 3 , 4 k=1,3,4 k = 1 , 3 , 4 , the condition a = k − 1 2 a=\frac{k-1}{2} a = 2 k − 1 of [14 ] is confirmed as printed. For these k k k , the space of invariant operators of the degree, charge and parity of Bol k \operatorname{Bol}_k Bol k is of dimension 0 at a generic weight. It is of dimension 1 exactly for a = k − 1 2 a=\frac{k-1}{2} a = 2 k − 1 , the weight b b b remaining free.
Both parts are obtained by computation with all the eight generating functions of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) , the weights a a a and b b b being symbolic.
(65) is the exact analogue, for m ( 1 ) \mathfrak{m}(1) m ( 1 ) , of the 1-parameter family in [14 Thm 1.5(i)] for t ( 1 ∣ 2 ) \mathfrak{t}(1|2) t ( 1∣2 ) . It explains the dimension 3 in order 2 of Statement 7.4(ii): two of the three operators are Bol ( λ 1 ) ⊗ 1 \operatorname{Bol}^{(\lambda_1)}\otimes1 Bol ( λ 1 ) ⊗ 1 and 1 ⊗ Bol ( λ 2 ) 1\otimes\operatorname{Bol}^{(\lambda_2)} 1 ⊗ Bol ( λ 2 ) , invariant at all weights, which at λ 1 = λ 2 = 1 2 \lambda_1=\lambda_2=\frac{1}{2} λ 1 = λ 2 = 2 1 are J 2 , l \mathcal{J}_{2,l} J 2 , l and J 2 , r \mathcal{J}_{2,r} J 2 , r up to sign; the third is the operator quoted in Statement 7.4(i).
The invariant derivatives ∂ τ \partial_{\tau} ∂ τ , D q D_q D q , D ξ D_{\xi} D ξ
In the sequel, differential operators are expressed in terms of the invariant derivatives (2) already used above, see §7.2:
∂ τ = D 1 , D q = q ∂ τ + ∂ ξ , D ξ = ξ ∂ τ − ∂ q . \partial_{\tau}=D_1,\qquad D_q=q\partial_{\tau}+\partial_{\xi},\qquad D_{\xi}=\xi\partial_{\tau}-\partial_q. ∂ τ = D 1 , D q = q ∂ τ + ∂ ξ , D ξ = ξ ∂ τ − ∂ q .
of parities p ( D ξ ) = 0 ˉ p(D_{\xi})=\bar{0} p ( D ξ ) = 0 ˉ and p ( D q ) = p ( ∂ τ ) = 1 ˉ p(D_q)=p(\partial_{\tau})=\bar{1} p ( D q ) = p ( ∂ τ ) = 1 ˉ , and of super-brackets (63) . In this Section and the next Section, it is convenient to count their degrees positively,
deg D ξ = deg D q = 1 , deg ∂ τ = 2 , \deg D_{\xi}=\deg D_q=1,\qquad\deg\partial_{\tau}=2, deg D ξ = deg D q = 1 , deg ∂ τ = 2 ,
i.e., with the sign opposite to the one of §7.2. The following lemma shows that no generality is lost by using D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ instead of the coordinate derivatives.
Lemma. Every differential operator is a sum of monomials in D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ
Lemma. Every differential operator on C [ q ∣ ξ , τ ] \mathbb{C}[q\mid\xi,\tau] C [ q ∣ ξ , τ ] can be written as a finite sum
∑ a , ε , δ F a , ε , δ ( q , ξ , τ ) D ξ a D q ε ∂ τ δ , where a ∈ Z ≥ 0 , ε , δ ∈ { 0 , 1 } . \sum_{a,\varepsilon,\delta}F_{a,\varepsilon,\delta}(q,\xi,\tau)D_{\xi}^{a}D_q^{\varepsilon}\partial_{\tau}^{\delta},\qquad\text{where }a\in\mathbb{Z}_{\geq0},\quad\varepsilon,\delta\in\{0,1\}. a , ε , δ ∑ F a , ε , δ ( q , ξ , τ ) D ξ a D q ε ∂ τ δ , where a ∈ Z ≥ 0 , ε , δ ∈ { 0 , 1 } .
Proof. The change from the coordinate derivatives to D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ is invertible:
∂ τ = ∂ τ , ∂ q = ξ ∂ τ − D ξ , ∂ ξ = D q − q ∂ τ . \partial_{\tau}=\partial_{\tau},\qquad\partial_q=\xi\partial_{\tau}-D_{\xi},\qquad\partial_{\xi}=D_q-q\partial_{\tau}. ∂ τ = ∂ τ , ∂ q = ξ ∂ τ − D ξ , ∂ ξ = D q − q ∂ τ .
Hence every differential operator written in terms of ∂ q , ∂ ξ , ∂ τ \partial_q,\partial_{\xi},\partial_{\tau} ∂ q , ∂ ξ , ∂ τ can be rewritten in terms of D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ , with coefficients in C [ q ∣ ξ , τ ] \mathbb{C}[q\mid\xi,\tau] C [ q ∣ ξ , τ ] . Reordering the resulting monomials gives the stated normal form. □
Accordingly, we shall use the differential monomials
P = D ξ a D q ε ∂ τ δ , a ∈ Z ≥ 0 , ε , δ ∈ { 0 , 1 } , P=D_{\xi}^{a}D_q^{\varepsilon}\partial_{\tau}^{\delta},\qquad a\in\mathbb{Z}_{\geq0},\qquad\varepsilon,\delta\in\{0,1\}, P = D ξ a D q ε ∂ τ δ , a ∈ Z ≥ 0 , ε , δ ∈ { 0 , 1 } ,
with
deg P = a + ε + 2 δ . \deg P=a+\varepsilon+2\delta. deg P = a + ε + 2 δ .
This basis is particularly convenient for imposing invariance under p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − .
Binary differential operators and the negative part
Let
J : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω J:\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\mu,\omega} J : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω
be a binary differential operator. By Lemma 7.7, J J J can be expanded in binary monomials constructed from D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ .
For homogeneous differential operators P P P and R R R , we use the graded tensor-product notation
( P ⊠ R ) ( φ , ψ ) : = ( − 1 ) p ( R ) p ( φ ) P ( φ ) R ( ψ ) , (P\boxtimes R)(\varphi,\psi):=(-1)^{p(R)p(\varphi)}P(\varphi)R(\psi), ( P ⊠ R ) ( φ , ψ ) := ( − 1 ) p ( R ) p ( φ ) P ( φ ) R ( ψ ) ,
extended by linearity. Thus, a general binary differential operator can be written as
J = ∑ α , β F α , β ( q , ξ , τ ) P α ⊠ P β , J=\sum_{\alpha,\beta}F_{\alpha,\beta}(q,\xi,\tau)P_{\alpha}\boxtimes P_{\beta}, J = α , β ∑ F α , β ( q , ξ , τ ) P α ⊠ P β ,
where P α , P β P_{\alpha},P_{\beta} P α , P β are monomials in D ξ D_{\xi} D ξ , D q D_q D q , ∂ τ \partial_{\tau} ∂ τ . The following elementary identity allows us to reduce the action on binary monomials to the corresponding unary actions.
Lemma. The Leibniz rule for P ⊠ R P\boxtimes R P ⊠ R
Lemma. The Leibniz rule for P ⊠ R P\boxtimes R P ⊠ R . For every homogeneous Y Y Y , P P P , R R R ,
Y ⋅ ( P ⊠ R ) = ( Y ⋅ P ) ⊠ R + ( − 1 ) p ( Y ) p ( P ) P ⊠ ( Y ⋅ R ) . Y\cdot(P\boxtimes R)=(Y\cdot P)\boxtimes R+(-1)^{p(Y)p(P)}P\boxtimes(Y\cdot R). Y ⋅ ( P ⊠ R ) = ( Y ⋅ P ) ⊠ R + ( − 1 ) p ( Y ) p ( P ) P ⊠ ( Y ⋅ R ) .
The proof is in the Appendix, §9.
We now impose invariance under the negative part
p e ( 2 ) − = ⟨ M 1 , M q , M ξ ⟩ . \mathfrak{pe}(2)_{-}=\langle M_{1},M_{q},M_{\xi}\rangle. pe ( 2 ) − = ⟨ M 1 , M q , M ξ ⟩ .
The choice of the D ξ , D q , ∂ τ D_{\xi},D_{q},\partial_{\tau} D ξ , D q , ∂ τ -basis is particularly useful here: these operators are adapted to the negative generators, and invariance forces the coefficient functions in the preceding expansion to be constant.
Lemma. Invariance under p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − forces constant coefficients
Lemma. (Invariance under p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − forces constant coefficients). Let
J = ∑ α , β F α , β ( q , ξ , τ ) P α ⊠ P β . J=\sum_{\alpha,\beta}F_{\alpha,\beta}(q,\xi,\tau)\,P_{\alpha}\boxtimes P_{\beta}. J = α , β ∑ F α , β ( q , ξ , τ ) P α ⊠ P β .
be a binary differential operator in the D ξ , D q , ∂ τ D_{\xi},D_{q},\partial_{\tau} D ξ , D q , ∂ τ -basis. If J J J is p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − -invariant, then F α , β ∈ C F_{\alpha,\beta}\in\mathbb{C} F α , β ∈ C for every pair ( α , β ) (\alpha,\beta) ( α , β ) .
The proof is in the Appendix, §9.
Consequently, every p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − -invariant binary differential operator is a constant-coefficient linear combination of the monomials
D ξ α D q ε ∂ τ δ ⊠ D ξ α ′ D q ε ′ ∂ τ δ ′ . D_{\xi}^{\alpha}D_{q}^{\varepsilon}\partial_{\tau}^{\delta}\boxtimes D_{\xi}^{\alpha'}D_{q}^{\varepsilon'}\partial_{\tau}^{\delta'}. D ξ α D q ε ∂ τ δ ⊠ D ξ α ′ D q ε ′ ∂ τ δ ′ .
Thus the classification problem is reduced from arbitrary polynomial-coefficient binary differential operators to a finite linear combination of tensor monomials of prescribed degree. We next impose invariance under the maximal toral subalgebra of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) .
p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators
The aim of this section is to classify the bilinear differential operators
J : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω J:\mathcal{F}_{\lambda_{1},\nu_{1}}\otimes\mathcal{F}_{\lambda_{2},\nu_{2}}\longrightarrow\mathcal{F}_{\mu,\omega} J : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω
that are invariant under the action of the periplectic Lie superalgebra p e ( 2 ) ⊂ m ( 1 ) \mathfrak{pe}(2)\subset\mathfrak{m}(1) pe ( 2 ) ⊂ m ( 1 ) . More precisely, we determine the values of the weights for which such operators exist and describe all p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant operators explicitly. The proofs in this Section are computations, and are collected in the Appendix, §9. Since M σ τ = 1 2 [ M q 2 , M τ ξ ] M_{\sigma\tau}=\frac{1}{2}[M_{q^{2}},M_{\tau\xi}] M σ τ = 2 1 [ M q 2 , M τ ξ ] , invariance under p e ( 2 ) − \mathfrak{pe}(2)_{-} pe ( 2 ) − , H 1 H_{1} H 1 , H 2 H_{2} H 2 , M τ ξ M_{\tau\xi} M τ ξ and M q 2 M_{q^{2}} M q 2 is invariance under the whole of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) .
Decomposition by the weights of the maximal torus
We now impose invariance under the maximal toral subalgebra of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) , generated by
H 1 : = 1 2 M τ + q ξ , H 2 : = 1 2 M τ − q ξ . H_{1}:=\frac{1}{2}M_{\tau+q\xi},\qquad H_{2}:=\frac{1}{2}M_{\tau-q\xi}. H 1 := 2 1 M τ + q ξ , H 2 := 2 1 M τ − q ξ .
To determine the corresponding weight conditions, consider differential monomials
P i = D ξ a i D q ε i ∂ τ δ i , i = 1 , 2 , P_{i}=D_{\xi}^{a_{i}}D_{q}^{\varepsilon_{i}}\partial_{\tau}^{\delta_{i}},\qquad i=1,2, P i = D ξ a i D q ε i ∂ τ δ i , i = 1 , 2 ,
where
a i ∈ Z ≥ 0 , ε i , δ i ∈ { 0 , 1 } . a_{i}\in\mathbb{Z}_{\geq0},\qquad\varepsilon_{i},\delta_{i}\in\{0,1\}. a i ∈ Z ≥ 0 , ε i , δ i ∈ { 0 , 1 } .
They determine the binary differential monomial
P 1 ⊠ P 2 : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω . P_{1}\boxtimes P_{2}:\mathcal{F}_{\lambda_{1},\nu_{1}}\otimes\mathcal{F}_{\lambda_{2},\nu_{2}}\longrightarrow\mathcal{F}_{\mu,\omega}. P 1 ⊠ P 2 : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F μ , ω .
For later convenience, set
u : = a 1 + a 2 + δ 1 + δ 2 , v : = ε 1 + ε 2 + δ 1 + δ 2 . u:=a_{1}+a_{2}+\delta_{1}+\delta_{2},\qquad v:=\varepsilon_{1}+\varepsilon_{2}+\delta_{1}+\delta_{2}. u := a 1 + a 2 + δ 1 + δ 2 , v := ε 1 + ε 2 + δ 1 + δ 2 .
Lemma 8.2 (The toral weights fix the target weight ( μ , ω ) (\mu,\omega) ( μ , ω ) ). If P 1 ⊠ P 2 P_1 \boxtimes P_2 P 1 ⊠ P 2 intertwines the action of the maximal toral subalgebra generated by H 1 H_1 H 1 and H 2 H_2 H 2 , then necessarily
μ = λ 1 + λ 2 − u , ω = ν 1 + ν 2 + u − v 2 . \mu= \lambda_1+\lambda_2-u,\qquad\omega=\nu_1+\nu_2+\frac{u-v}{2}. μ = λ 1 + λ 2 − u , ω = ν 1 + ν 2 + 2 u − v . Since deg D ξ = deg D q = 1 \deg\mathcal{D}_{\xi}=\deg\mathcal{D}_{q}=1 deg D ξ = deg D q = 1 , deg ∂ τ = 2 \deg\partial_{\tau}=2 deg ∂ τ = 2 , the total degree of P 1 ⊠ P 2 P_1\boxtimes P_2 P 1 ⊠ P 2 is
k = a 1 + a 2 + ϵ 1 + ϵ 2 + 2 δ 1 + 2 δ 2 = ( a 1 + a 2 + δ 1 + δ 2 ) + ( ϵ 1 + ϵ 2 + δ 1 + δ 2 ) . k=a_1+a_2+\epsilon_1+\epsilon_2+2\delta_1+2\delta_2=(a_1+a_2+\delta_1+\delta_2)+(\epsilon_1+\epsilon_2+\delta_1+\delta_2). k = a 1 + a 2 + ϵ 1 + ϵ 2 + 2 δ 1 + 2 δ 2 = ( a 1 + a 2 + δ 1 + δ 2 ) + ( ϵ 1 + ϵ 2 + δ 1 + δ 2 ) . The integer v v v also determines the parity of the monomial. Indeed,
p ( P 1 ⊠ P 2 ) ≡ ϵ 1 + ϵ 2 + δ 1 + δ 2 ≡ v ( m o d 2 ) . p(P_1\boxtimes P_2)\equiv\epsilon_1+\epsilon_2+\delta_1+\delta_2\equiv v\pmod{2}. p ( P 1 ⊠ P 2 ) ≡ ϵ 1 + ϵ 2 + δ 1 + δ 2 ≡ v ( mod 2 ) . Since each monomial factor contains at most one D q \mathcal{D}_q D q and at most one ∂ τ \partial_{\tau} ∂ τ , we have
0 ≤ v ≤ 4. 0\leq v\leq4. 0 ≤ v ≤ 4. Consequently, for an odd binary operator the only possible Cartan weights are
v = 1 , v = 3 , v=1,\qquad v=3, v = 1 , v = 3 , whereas for an even binary operator the only possibilities are
v = 0 , v = 2 , v = 4. v=0,\qquad v=2,\qquad v=4. v = 0 , v = 2 , v = 4. This gives the following decomposition by the toral weights.
For odd operators:
Table 1 (PDF p. 48).
For even operators:
Table 2.
The classification problem splits into five independent sectors: two odd sectors, corresponding to v = 1 v=1 v = 1 and v = 3 v=3 v = 3 , and three even sectors, corresponding to v = 0 , 2 , 4 v=0,2,4 v = 0 , 2 , 4 .
Odd operators of Type A: the generic case
We first consider odd operators of Type A and restrict ourselves to generic values of the weights. The exceptional values, for which the coefficient system loses rank, will be studied separately.
For Type A, one has v = 1 v=1 v = 1 and u = k − 1 u=k-1 u = k − 1 . Hence the Cartan equations determine the target weights as
μ = λ 1 + λ 2 − k + 1 , ω = ν 1 + ν 2 + k − 2 2 . \mu=\lambda_1+\lambda_2-k+1,\qquad\omega=\nu_1+\nu_2+\frac{k-2}{2}. μ = λ 1 + λ 2 − k + 1 , ω = ν 1 + ν 2 + 2 k − 2 .
Since v = 1 v=1 v = 1 , the monomials that may occur contain exactly one contribution to the second Cartan degree. There are four possibilities: D q \mathcal{D}_q D q or ∂ τ \partial_{\tau} ∂ τ may occur in either the first or the second tensor factor.
Odd Type A operators in low degrees
Before considering arbitrary degree, we record the first three cases explicitly. Besides, providing the low-degree classification, these examples show that k = 2 k=2 k = 2 is exceptional, whereas the general recurrence already applies from k = 3 k=3 k = 3 .
Degree k = 1 k=1 k = 1 . The target weights are μ = λ 1 + λ 2 \mu=\lambda_1+\lambda_2 μ = λ 1 + λ 2 , and ω = ν 1 + ν 2 − 1 2 \omega=\nu_1+\nu_2-\frac{1}{2} ω = ν 1 + ν 2 − 2 1 . The most general Type A operator is
J ˉ 1 , A ( 1 ) = A 0 1 ⊠ D q + E 1 D q ⊠ 1. \bar{J}^{(1)}_{1,A}=A_0 1\boxtimes D_q+E_1D_q\boxtimes1. J ˉ 1 , A ( 1 ) = A 0 1 ⊠ D q + E 1 D q ⊠ 1.
There is no condition coming from M q 2 M_{q^2} M q 2 . Invariance under M τ ξ M_{\tau\xi} M τ ξ gives
λ 2 A 0 + λ 1 E 1 = 0. \lambda_2A_0+\lambda_1E_1=0. λ 2 A 0 + λ 1 E 1 = 0.
Hence, for generic weights, the space of Type A invariants is one-dimensional. A convenient generator is
J ˉ 1 , A ( 1 ) = λ 1 1 ⊠ D q − λ 2 D q ⊠ 1. \bar{J}^{(1)}_{1,A}=\lambda_1 1\boxtimes D_q-\lambda_2D_q\boxtimes1. J ˉ 1 , A ( 1 ) = λ 1 1 ⊠ D q − λ 2 D q ⊠ 1.
Thus dim Inv 1 ˉ , A ( 1 ) = 1 \dim\operatorname{Inv}^{(1)}_{\bar{1},A}=1 dim Inv 1 ˉ , A ( 1 ) = 1 at generic weights. The exceptional case λ 1 = λ 2 = 0 \lambda_1=\lambda_2=0 λ 1 = λ 2 = 0 will be considered together with the resonant cases.
Degree k = 2 k=2 k = 2 . The target weights are μ = λ 1 + λ 2 − 1 \mu=\lambda_1+\lambda_2-1 μ = λ 1 + λ 2 − 1 , and ω = ν 1 + ν 2 \omega=\nu_1+\nu_2 ω = ν 1 + ν 2 . The most general Type A operator is
J ˉ 1 , A ( 2 ) = A 0 1 ⊠ D ξ D q + A 1 D ξ ⊠ D q + B 0 1 ⊠ ∂ τ + E 1 D q ⊠ D ξ + E 2 D ξ D q ⊠ 1 + F 2 ∂ τ ⊠ 1. \bar{J}^{(2)}_{1,A}=A_0 1\boxtimes D_\xi D_q+A_1D_\xi\boxtimes D_q+B_0 1\boxtimes\partial_\tau+E_1D_q\boxtimes D_\xi+E_2D_\xi D_q\boxtimes1+F_2\partial_\tau\boxtimes1. J ˉ 1 , A ( 2 ) = A 0 1 ⊠ D ξ D q + A 1 D ξ ⊠ D q + B 0 1 ⊠ ∂ τ + E 1 D q ⊠ D ξ + E 2 D ξ D q ⊠ 1 + F 2 ∂ τ ⊠ 1.
Invariance under M q 2 M_{q^2} M q 2 gives E 1 = A 1 E_1=A_1 E 1 = A 1 . The M τ ξ M_{\tau\xi} M τ ξ -equations are
B 0 = 2 λ 2 A 0 + 2 λ 1 A 1 , F 2 = 2 λ 2 A 1 + 2 λ 1 E 2 . B_0=2\lambda_2A_0+2\lambda_1A_1,\qquad F_2=2\lambda_2A_1+2\lambda_1E_2. B 0 = 2 λ 2 A 0 + 2 λ 1 A 1 , F 2 = 2 λ 2 A 1 + 2 λ 1 E 2 .
Thus A 0 , A 1 , E 2 A_0,A_1,E_2 A 0 , A 1 , E 2 are independent parameters, and every Type A operator of degree 2 can be written as
J ˉ 1 , A ( 2 ) = a J 1 ( 2 ) + b J 2 ( 2 ) + c J 3 ( 2 ) . \bar{J}^{(2)}_{1,A}=aJ^{(2)}_1+bJ^{(2)}_2+cJ^{(2)}_3. J ˉ 1 , A ( 2 ) = a J 1 ( 2 ) + b J 2 ( 2 ) + c J 3 ( 2 ) .
where
J 1 ( 2 ) = 1 ⊠ D ξ D q + 2 λ 2 1 ⊠ ∂ τ , J^{(2)}_1=1\boxtimes D_\xi D_q+2\lambda_2 1\boxtimes\partial_\tau, J 1 ( 2 ) = 1 ⊠ D ξ D q + 2 λ 2 1 ⊠ ∂ τ ,
J 2 ( 2 ) = D ξ ⊠ D q + D q ⊠ D ξ + 2 λ 1 1 ⊠ ∂ τ + 2 λ 2 ∂ τ ⊠ 1 , J^{(2)}_2=D_\xi\boxtimes D_q+D_q\boxtimes D_\xi+2\lambda_1 1\boxtimes\partial_\tau+2\lambda_2\partial_\tau\boxtimes1, J 2 ( 2 ) = D ξ ⊠ D q + D q ⊠ D ξ + 2 λ 1 1 ⊠ ∂ τ + 2 λ 2 ∂ τ ⊠ 1 ,
and
J 3 ( 2 ) = D ξ D q ⊠ 1 + 2 λ 1 ∂ τ ⊠ 1. J^{(2)}_3=D_\xi D_q\boxtimes1+2\lambda_1\partial_\tau\boxtimes1. J 3 ( 2 ) = D ξ D q ⊠ 1 + 2 λ 1 ∂ τ ⊠ 1.
Consequently,
dim Inv 1 ˉ , A ( 2 ) = 3. \dim\operatorname{Inv}^{(2)}_{\bar{1},A}=3. dim Inv 1 ˉ , A ( 2 ) = 3.
Thus k = 2 k=2 k = 2 is exceptional and must be treated separately from the general Type A family.
Degree k = 3 k=3 k = 3 . The target weights are μ = λ 1 + λ 2 − 2 \mu=\lambda_1+\lambda_2-2 μ = λ 1 + λ 2 − 2 , and ω = ν 1 + ν 2 + 1 2 \omega=\nu_1+\nu_2+\frac{1}{2} ω = ν 1 + ν 2 + 2 1 . The general Type A operator is
J ˉ 1 , A ( 3 ) = A 0 1 ⊠ D ξ 2 D q + A 1 D ξ ⊠ D ξ D q + A 2 D ξ 2 ⊠ D q + B 0 1 ⊠ D ξ ∂ τ + B 1 D ξ ⊠ ∂ τ + E 1 D q ⊠ D ξ 2 + E 2 D ξ D q ⊠ D ξ + E 3 D ξ 2 D q ⊠ 1 + F 2 ∂ τ ⊠ D ξ + F 3 D ξ ∂ τ ⊠ 1. \bar{J}^{(3)}_{1,A}=A_0 1\boxtimes D_\xi^2D_q+A_1D_\xi\boxtimes D_\xi D_q+A_2D_\xi^2\boxtimes D_q+B_0 1\boxtimes D_\xi\partial_\tau+B_1D_\xi\boxtimes\partial_\tau
+E_1D_q\boxtimes D_\xi^2+E_2D_\xi D_q\boxtimes D_\xi+E_3D_\xi^2D_q\boxtimes1+F_2\partial_\tau\boxtimes D_\xi+F_3D_\xi\partial_\tau\boxtimes1. J ˉ 1 , A ( 3 ) = A 0 1 ⊠ D ξ 2 D q + A 1 D ξ ⊠ D ξ D q + A 2 D ξ 2 ⊠ D q + B 0 1 ⊠ D ξ ∂ τ + B 1 D ξ ⊠ ∂ τ + E 1 D q ⊠ D ξ 2 + E 2 D ξ D q ⊠ D ξ + E 3 D ξ 2 D q ⊠ 1 + F 2 ∂ τ ⊠ D ξ + F 3 D ξ ∂ τ ⊠ 1.
Invariance under M q 2 M_{q^2} M q 2 gives
B 0 = 2 A 0 , B 1 = A 1 , E 1 = 1 2 A 1 , E 2 = 2 A 2 , F 2 = E 2 , F 3 = 2 E 3 . B_0=2A_0,\qquad B_1=A_1,\qquad E_1=\frac{1}{2}A_1,\qquad E_2=2A_2,\qquad F_2=E_2,\qquad F_3=2E_3. B 0 = 2 A 0 , B 1 = A 1 , E 1 = 2 1 A 1 , E 2 = 2 A 2 , F 2 = E 2 , F 3 = 2 E 3 .
The remaining M τ ξ M_{\tau\xi} M τ ξ -equations reduce to
2 λ 1 A 1 + 2 ( 2 λ 2 − 2 ) A 0 = 0 , 2 ( 2 λ 1 − 1 ) A 2 + ( 2 λ 2 − 1 ) A 1 = 0 , 2\lambda_1A_1+2(2\lambda_2-2)A_0=0,\qquad2(2\lambda_1-1)A_2+(2\lambda_2-1)A_1=0, 2 λ 1 A 1 + 2 ( 2 λ 2 − 2 ) A 0 = 0 , 2 ( 2 λ 1 − 1 ) A 2 + ( 2 λ 2 − 1 ) A 1 = 0 ,
and the endpoint equation
( 2 − 2 λ 1 ) E 3 = 2 λ 2 A 2 . (2-2\lambda_1)E_3=2\lambda_2A_2. ( 2 − 2 λ 1 ) E 3 = 2 λ 2 A 2 .
At generic weights these equations determine all coefficients from A 0 A_0 A 0 . Explicitly,
A 1 = − 2 ( λ 2 − 1 ) λ 1 A 0 , A 2 = ( λ 2 − 1 ) ( 2 λ 2 − 1 ) λ 1 ( 2 λ 1 − 1 ) A 0 , and E 3 = λ 2 1 − λ 1 A 2 . A_1=-\frac{2(\lambda_2-1)}{\lambda_1}A_0,\qquad A_2=\frac{(\lambda_2-1)(2\lambda_2-1)}{\lambda_1(2\lambda_1-1)}A_0,\qquad\text{and}\qquad E_3=\frac{\lambda_2}{1-\lambda_1}A_2. A 1 = − λ 1 2 ( λ 2 − 1 ) A 0 , A 2 = λ 1 ( 2 λ 1 − 1 ) ( λ 2 − 1 ) ( 2 λ 2 − 1 ) A 0 , and E 3 = 1 − λ 1 λ 2 A 2 .
Therefore,
dim Inv 1 ˉ , A ( 3 ) = 1 \dim\operatorname{Inv}^{(3)}_{\bar{1},A}=1 dim Inv 1 ˉ , A ( 3 ) = 1
at generic weights.
The exceptional values at which one or more of the preceding relations degenerates will be included in the resonance analysis.
Odd Type A operators in degree k k k
The most general odd Type A operator of degree k k k is
J 1 ˉ , A ( k ) = ∑ 0 ≤ i ≤ k − 1 A i D ξ i ⊠ D ξ k − i − 1 D q + ∑ 0 ≤ i ≤ k − 2 B i D ξ i ⊠ D ξ k − i − 2 ∂ τ + ∑ 1 ≤ i ≤ k E i D ξ i − 1 D q ⊠ D ξ k − i + ∑ 2 ≤ i ≤ k F i D ξ i − 2 ∂ τ ⊠ D ξ k − i . (66) J^{(k)}_{\bar{1},A}=
\sum_{0\leq i\leq k-1} A_i \mathcal{D}_{\xi}^{i}\boxtimes\mathcal{D}_{\xi}^{k-i-1}\mathcal{D}_{q}
+\sum_{0\leq i\leq k-2} B_i\mathcal{D}_{\xi}^{i}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\partial_{\tau}
+\sum_{1\leq i\leq k} E_i\mathcal{D}_{\xi}^{i-1}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i}
+\sum_{2\leq i\leq k}F_i\mathcal{D}_{\xi}^{i-2}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i}. \tag*{(66)} J 1 ˉ , A ( k ) = 0 ≤ i ≤ k − 1 ∑ A i D ξ i ⊠ D ξ k − i − 1 D q + 0 ≤ i ≤ k − 2 ∑ B i D ξ i ⊠ D ξ k − i − 2 ∂ τ + 1 ≤ i ≤ k ∑ E i D ξ i − 1 D q ⊠ D ξ k − i + 2 ≤ i ≤ k ∑ F i D ξ i − 2 ∂ τ ⊠ D ξ k − i . (66)
Imposing invariance under M τ ξ M_{\tau\xi} M τ ξ and M q 2 M_{q^2} M q 2 turns (66) into a recurrence for the coefficients A i A_i A i , B i B_i B i , E i E_i E i , F i F_i F i ; solving that recurrence gives the following generic answer.
Theorem. Odd Type A operators for generic weights
Let k k k be in the stable range for the above coefficient system, and assume that the weights are generic. Then, the space of odd Type A p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators
J 1 ˉ , A ( k ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 − k + 1 , ν 1 + ν 2 + k − 2 2 J^{(k)}_{\bar{1},A}:\mathcal{F}_{\lambda_1,\nu_1}\otimes\mathcal{F}_{\lambda_2,\nu_2}\longrightarrow\mathcal{F}_{\lambda_1+\lambda_2-k+1,\nu_1+\nu_2+\frac{k-2}{2}} J 1 ˉ , A ( k ) : F λ 1 , ν 1 ⊗ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 − k + 1 , ν 1 + ν 2 + 2 k − 2
given as in the ansatz (66) is one-dimensional.
Up to an overall scalar, its coefficients are determined by
A i = ( − 1 ) i ( k − 1 i ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 1 − r ) A 0 , 0 ≤ i ≤ k − 1 , A_i=(-1)^i\binom{k-1}{i}
\frac{\prod_{0\leq r\leq i-1}(2\lambda_2-k+r+1)}
{\prod_{0\leq r\leq i-1}(2\lambda_1-r)}A_0,\qquad0\leq i\leq k-1, A i = ( − 1 ) i ( i k − 1 ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 1 − r ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) A 0 , 0 ≤ i ≤ k − 1 ,
B i = ( k − i − 1 ) A i , ( 0 ≤ i ≤ k − 2 ) E i = i k − i A i , ( 1 ≤ i ≤ k − 1 ) B_i=(k-i-1)A_i,\qquad(0\leq i\leq k-2)\qquad E_i=\frac{i}{k-i}A_i,\qquad(1\leq i\leq k-1) B i = ( k − i − 1 ) A i , ( 0 ≤ i ≤ k − 2 ) E i = k − i i A i , ( 1 ≤ i ≤ k − 1 )
and
F i = ( i − 1 ) E i , ( 2 ≤ i ≤ k ) E k = 2 λ 2 k − 1 − 2 λ 1 A k − 1 . F_i=(i-1)E_i,\qquad(2\leq i\leq k)\qquad E_k=\frac{2\lambda_2}{k-1-2\lambda_1}A_{k-1}. F i = ( i − 1 ) E i , ( 2 ≤ i ≤ k ) E k = k − 1 − 2 λ 1 2 λ 2 A k − 1 .
Odd operators of Type A: the resonant case
Resonant odd Type A operators in low degrees
We now examine the exceptional weights for k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 by studying the rank of the coefficient system directly, without dividing by weight-dependent factors, since dimension jumps may occur even when no individual equation vanishes identically.
Degree $k=1. Recall that
J 1 ˉ , A ( 1 ) = A 0 1 ⊠ D q + E 1 D q ⊠ 1. J^{(1)}_{\bar{1},A}=A_0\,1\boxtimes\mathcal{D}_{q}+E_1\mathcal{D}_{q}\boxtimes1. J 1 ˉ , A ( 1 ) = A 0 1 ⊠ D q + E 1 D q ⊠ 1.
The only invariance equation is
λ 2 A 0 + λ 1 E 1 = 0. \lambda_2A_0+\lambda_1E_1=0. λ 2 A 0 + λ 1 E 1 = 0.
At ( λ 1 , λ 2 ) = ( 0 , 0 ) (\lambda_1,\lambda_2)=(0,0) ( λ 1 , λ 2 ) = ( 0 , 0 ) , the equation vanishes identically. Both A 0 A_0 A 0 and E 1 E_1 E 1 are then arbitrary, and therefore
dim Inv 1 ˉ , A ( 1 ) = 2. \dim\operatorname{Inv}^{(1)}_{\bar{1},A}=2. dim Inv 1 ˉ , A ( 1 ) = 2.
A basis at this resonant point is
1 ⊠ D q , D q ⊠ 1. 1\boxtimes D_q,\qquad D_q\boxtimes1. 1 ⊠ D q , D q ⊠ 1.
Consequently,
dim Inv 1 ˉ , A ( 1 ) = { 2 , ( λ 1 , λ 2 ) = ( 0 , 0 ) , 1 , otherwise . \dim\operatorname{Inv}^{(1)}_{\bar{1},A}=
\begin{cases}
2, & (\lambda_1,\lambda_2)=(0,0),\\
1, & \text{otherwise}.
\end{cases} dim Inv 1 ˉ , A ( 1 ) = { 2 , 1 , ( λ 1 , λ 2 ) = ( 0 , 0 ) , otherwise .
Degree k = 2 k=2 k = 2 . The general Type A operator is
J 1 ˉ , A ( 2 ) = A 0 1 ⊠ D ξ D q + A 1 D ξ ⊠ D q + B 0 1 ⊠ ∂ τ + E 1 D q ⊠ D ξ + E 2 D ξ D q ⊠ 1 + F 2 ∂ τ ⊠ 1. J^{(2)}_{\bar{1},A}=A_0\,1\boxtimes D_\xi D_q+A_1D_\xi\boxtimes D_q+B_0\,1\boxtimes\partial_\tau+E_1D_q\boxtimes D_\xi+E_2D_\xi D_q\boxtimes1+F_2\partial_\tau\boxtimes1. J 1 ˉ , A ( 2 ) = A 0 1 ⊠ D ξ D q + A 1 D ξ ⊠ D q + B 0 1 ⊠ ∂ τ + E 1 D q ⊠ D ξ + E 2 D ξ D q ⊠ 1 + F 2 ∂ τ ⊠ 1.
The complete coefficient system is
E 1 = A 1 , B 0 = 2 λ 2 A 0 + 2 λ 1 A 1 , and F 2 = 2 λ 2 A 1 + 2 λ 1 E 2 . E_1=A_1,\qquad B_0=2\lambda_2A_0+2\lambda_1A_1,\qquad\text{and}\qquad F_2=2\lambda_2A_1+2\lambda_1E_2. E 1 = A 1 , B 0 = 2 λ 2 A 0 + 2 λ 1 A 1 , and F 2 = 2 λ 2 A 1 + 2 λ 1 E 2 .
The coefficients A 0 A_0 A 0 , A 1 A_1 A 1 , E 2 E_2 E 2 are arbitrary, while E 1 E_1 E 1 , B 0 B_0 B 0 , F 2 F_2 F 2 are uniquely determined by them. In particular, the rank of this system is independent of ( λ 1 , λ 2 ) (\lambda_1,\lambda_2) ( λ 1 , λ 2 ) . Therefore, no resonance occurs in degree k = 2 k=2 k = 2 , and
dim Inv 1 ˉ , A ( 2 ) = 3 \dim\operatorname{Inv}^{(2)}_{\bar{1},A}=3 dim Inv 1 ˉ , A ( 2 ) = 3
for every pair of weights ( λ 1 , λ 2 ) (\lambda_1,\lambda_2) ( λ 1 , λ 2 ) .
Degree $k=3. After using the relations obtained from M q 2 M_{q^2} M q 2 -invariance, the independent coefficients may be chosen to be A 0 A_0 A 0 , A 1 A_1 A 1 , A 2 A_2 A 2 , E 3 E_3 E 3 . The remaining invariance equations are
2 ( 2 λ 2 − 2 ) A 0 + 2 λ 1 A 1 = 0 , ( 2 λ 2 − 1 ) A 1 + 2 ( 2 λ 1 − 1 ) A 2 = 0 , − 2 λ 2 A 2 + ( 2 − 2 λ 1 ) E 3 = 0. 2(2\lambda_2-2)A_0+2\lambda_1A_1=0,\qquad(2\lambda_2-1)A_1+2(2\lambda_1-1)A_2=0,\qquad-2\lambda_2A_2+(2-2\lambda_1)E_3=0. 2 ( 2 λ 2 − 2 ) A 0 + 2 λ 1 A 1 = 0 , ( 2 λ 2 − 1 ) A 1 + 2 ( 2 λ 1 − 1 ) A 2 = 0 , − 2 λ 2 A 2 + ( 2 − 2 λ 1 ) E 3 = 0.
The system can be written as
( 4 ( λ 2 − 1 ) 2 λ 1 0 0 0 2 λ 2 − 1 2 ( 2 λ 1 − 1 ) 0 0 0 − 2 λ 2 2 ( 1 − λ 1 ) ) ( A 0 A 1 A 2 E 3 ) = 0. \begin{pmatrix}
4(\lambda_2-1) & 2\lambda_1 & 0 & 0\\
0 & 2\lambda_2-1 & 2(2\lambda_1-1) & 0\\
0 & 0 & -2\lambda_2 & 2(1-\lambda_1)
\end{pmatrix}
\begin{pmatrix}
A_0\\
A_1\\
A_2\\
E_3
\end{pmatrix}
=0. 4 ( λ 2 − 1 ) 0 0 2 λ 1 2 λ 2 − 1 0 0 2 ( 2 λ 1 − 1 ) − 2 λ 2 0 0 2 ( 1 − λ 1 ) A 0 A 1 A 2 E 3 = 0.
We now determine the values for which the rank drops. A direct analysis of the above matrix shows that this happens precisely at
( λ 1 , λ 2 ) ∈ { ( 0 , 1 ) , ( 1 2 , 1 2 ) , ( 1 , 0 ) , ( 1 2 , 1 ) , ( 1 , 1 2 ) , ( 1 , 1 ) } . (\lambda_1,\lambda_2)\in\left\{(0,1),\left(\frac{1}{2},\frac{1}{2}\right),(1,0),\left(\frac{1}{2},1\right),\left(1,\frac{1}{2}\right),(1,1)\right\}. ( λ 1 , λ 2 ) ∈ { ( 0 , 1 ) , ( 2 1 , 2 1 ) , ( 1 , 0 ) , ( 2 1 , 1 ) , ( 1 , 2 1 ) , ( 1 , 1 ) } .
At each of these six points the matrix has rank 2 2 2 , and hence
dim Inv 1 ˉ , A ( 3 ) = 2. \dim\operatorname{Inv}^{(3)}_{\bar{1},A}=2. dim Inv 1 ˉ , A ( 3 ) = 2.
Let us describe the degenerations explicitly.
At ( λ 1 , λ 2 ) = ( 0 , 1 ) (\lambda_1,\lambda_2)=(0,1) ( λ 1 , λ 2 ) = ( 0 , 1 ) , the system gives A 1 = 2 A 2 A_1=2A_2 A 1 = 2 A 2 , and E 3 = A 2 E_3=A_2 E 3 = A 2 . Thus A 0 A_0 A 0 and A 2 A_2 A 2 are independent, and the solution space is two-dimensional.
At ( λ 1 , λ 2 ) = ( 1 2 , 1 2 ) (\lambda_1,\lambda_2)=\left(\frac{1}{2},\frac{1}{2}\right) ( λ 1 , λ 2 ) = ( 2 1 , 2 1 ) , the system gives A 1 = 2 A 0 A_1=2A_0 A 1 = 2 A 0 , and E 3 = A 2 E_3=A_2 E 3 = A 2 . Again the solution space is two-dimensional.
At ( λ 1 , λ 2 ) = ( 1 , 0 ) (\lambda_1,\lambda_2)=(1,0) ( λ 1 , λ 2 ) = ( 1 , 0 ) , the system gives A 1 = 2 A 0 A_1=2A_0 A 1 = 2 A 0 , and A 2 = A 0 A_2=A_0 A 2 = A 0 . Thus A 0 A_0 A 0 and E 3 E_3 E 3 are independent, and the space is two-dimensional.
There are three further resonant points at which no complete row needs to vanish. Instead, the rank drops because adjacent equations cease to impose independent conditions.
At ( λ 1 , λ 2 ) = ( 1 2 , 1 ) (\lambda_1,\lambda_2)=\left(\frac{1}{2},1\right) ( λ 1 , λ 2 ) = ( 2 1 , 1 ) , the system gives A 1 = 0 A_1=0 A 1 = 0 , and E 3 = 2 A 2 E_3=2A_2 E 3 = 2 A 2 , while A 0 A_0 A 0 and A 2 A_2 A 2 are arbitrary.
At ( λ 1 , λ 2 ) = ( 1 , 1 2 ) (\lambda_1,\lambda_2)=\left(1,\frac{1}{2}\right) ( λ 1 , λ 2 ) = ( 1 , 2 1 ) , one obtains A 1 = A 0 A_1=A_0 A 1 = A 0 , and A 2 = 0 A_2=0 A 2 = 0 , while A 0 A_0 A 0 and E 3 E_3 E 3 are arbitrary. Finally, at ( λ 1 , λ 2 ) = ( 1 , 1 ) (\lambda_1,\lambda_2)=(1,1) ( λ 1 , λ 2 ) = ( 1 , 1 ) , the system gives A 1 = A 2 = 0 A_1=A_2=0 A 1 = A 2 = 0 , while A 0 A_0 A 0 and E 3 E_3 E 3 remain independent.
We therefore obtain
dim Inv 1 ˉ , A ( 3 ) = { 2 , ( λ 1 , λ 2 ) ∈ { ( 0 , 1 ) , ( 1 2 , 1 2 ) , ( 1 , 0 ) , ( 1 2 , 1 ) , ( 1 , 1 2 ) , ( 1 , 1 ) } , 1 , otherwise. \dim\operatorname{Inv}_{\bar{1},A}^{(3)}=
\begin{cases}
2, & (\lambda_1,\lambda_2)\in\left\{(0,1),\left(\frac{1}{2},\frac{1}{2}\right),(1,0),\left(\frac{1}{2},1\right),\left(1,\frac{1}{2}\right),(1,1)\right\},\\
1, & \text{otherwise.}
\end{cases} dim Inv 1 ˉ , A ( 3 ) = { 2 , 1 , ( λ 1 , λ 2 ) ∈ { ( 0 , 1 ) , ( 2 1 , 2 1 ) , ( 1 , 0 ) , ( 2 1 , 1 ) , ( 1 , 2 1 ) , ( 1 , 1 ) } , otherwise.
8.5.1a. Summary of the low-degree cases. The odd Type A invariant spaces in degrees k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 are therefore described by
Table 6.
Resonant odd Type A operators for k ≥ 3 k\geq3 k ≥ 3
We now determine all exceptional weights for odd Type A operators in the stable range k ≥ 3 k\geq3 k ≥ 3 . The resonance problem is treated as a rank problem for the complete coefficient system; in particular, we do not divide by any weight-dependent coefficient.
The resonant set is described as follows.
Proposition 8.6. (The resonant weights in the odd sector A.) Let k ≥ 3 k\geq3 k ≥ 3 . The odd Type A coefficient system is resonant precisely when
2 λ 1 , 2 λ 2 ∈ { 0 , 1 , … , k − 1 } and 2 λ 1 + 2 λ 2 ≥ k − 1. 2\lambda_1,2\lambda_2\in\{0,1,\ldots,k-1\}\qquad\text{and}\qquad2\lambda_1+2\lambda_2\geq k-1. 2 λ 1 , 2 λ 2 ∈ { 0 , 1 , … , k − 1 } and 2 λ 1 + 2 λ 2 ≥ k − 1. At every such point,
dim Inv 1 ˉ , A ( k ) = 2. \dim\operatorname{Inv}_{\bar{1},A}^{(k)}=2. dim Inv 1 ˉ , A ( k ) = 2. At all other weights,
dim Inv 1 ˉ , A ( k ) = 1. \dim\operatorname{Inv}_{\bar{1},A}^{(k)}=1. dim Inv 1 ˉ , A ( k ) = 1.
Odd operators of Type B: the generic case
Odd operators of Type B in low degrees
We now turn to odd operators of Type B. Recall that Type B corresponds to the Cartan degree v = 3 v=3 v = 3 , u = k − 3 u=k-3 u = k − 3 , and therefore the target weights are
μ = λ 1 + λ 2 − k + 3 , ω = ν 1 + ν 2 + k − 6 2 . \mu=\lambda_1+\lambda_2-k+3,\qquad\omega=\nu_1+\nu_2+\frac{k-6}{2}. μ = λ 1 + λ 2 − k + 3 , ω = ν 1 + ν 2 + 2 k − 6 .
The general odd Type B ansatz is
J 1 ˉ , B ( k ) = ∑ 3 ≤ i ≤ k − 1 C i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + ∑ 3 ≤ i ≤ k − 2 D i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ J_{\bar{1},B}^{(k)}
=
\sum_{3\leq i\leq k-1} C_i D_{\xi}^{i-3}D_q\partial_\tau\boxtimes D_{\xi}^{k-i-1}D_q
+
\sum_{3\leq i\leq k-2} D_i D_{\xi}^{i-3}D_q\partial_\tau\boxtimes D_{\xi}^{k-i-2}\partial_\tau J 1 ˉ , B ( k ) = 3 ≤ i ≤ k − 1 ∑ C i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 3 ≤ i ≤ k − 2 ∑ D i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ
+ ∑ 1 ≤ i ≤ k − 3 G i D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ + ∑ 2 ≤ i ≤ k − 3 H i D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ . (67) \qquad+
\sum_{1\leq i\leq k-3} G_i D_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i-3}D_q\partial_\tau
+
\sum_{2\leq i\leq k-3} H_i D_{\xi}^{i-2}\partial_\tau\boxtimes D_{\xi}^{k-i-3}D_q\partial_\tau. \tag*{(67)} + 1 ≤ i ≤ k − 3 ∑ G i D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ + 2 ≤ i ≤ k − 3 ∑ H i D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ . (67)
We first explain why no nonzero Type B operator exists in degrees k = 1 , 2 , 3 , 4 k=1,2,3,4 k = 1 , 2 , 3 , 4 .
Degree k = 1 , 2 , 3 , 4 k=1,2,3,4 k = 1 , 2 , 3 , 4 . In this case, the condition v = 3 v=3 v = 3 is incompatible with the existence of a monomial of total degree k k k . Equivalently, all the sums in the ansatz (67) are empty. Therefore,
J 1 ˉ , B ( k ) = 0 , k = 1 , 2 , 3. J_{\bar{1},B}^{(k)}=0,\qquad k=1,2,3. J 1 ˉ , B ( k ) = 0 , k = 1 , 2 , 3.
The case k = 4 k=4 k = 4 requires a separate argument. Here, the Type B ansatz contains exactly two terms:
J 1 ˉ , B ( 4 ) = C 3 D q ∂ τ ⊠ D q + G 1 D q ⊠ D q ∂ τ . J^{(4)}_{\bar{1},B}=C_{3}D_{q}\partial_{\tau}\boxtimes D_{q}+G_{1}D_{q}\boxtimes D_{q}\partial_{\tau}. J 1 ˉ , B ( 4 ) = C 3 D q ∂ τ ⊠ D q + G 1 D q ⊠ D q ∂ τ .
We impose invariance under M τ ξ M_{\tau\xi} M τ ξ and obtain
M τ ξ ⋅ ( D q ∂ τ ⊠ D q ) = D ξ D q ⊠ D q + 2 λ 1 ∂ τ ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 , M_{\tau\xi}\cdot(D_{q}\partial_{\tau}\boxtimes D_{q})=D_{\xi}D_{q}\boxtimes D_{q}+2\lambda_{1}\partial_{\tau}\boxtimes D_{q}+2\lambda_{2}D_{q}\partial_{\tau}\boxtimes1, M τ ξ ⋅ ( D q ∂ τ ⊠ D q ) = D ξ D q ⊠ D q + 2 λ 1 ∂ τ ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 ,
M τ ξ ⋅ ( D q ⊠ D q ∂ τ ) = 2 λ 1 1 ⊠ D q ∂ τ − D q ⊠ D ξ D q − 2 λ 2 D q ⊠ ∂ τ . M_{\tau\xi}\cdot(D_{q}\boxtimes D_{q}\partial_{\tau})=2\lambda_{1}1\boxtimes D_{q}\partial_{\tau}-D_{q}\boxtimes D_{\xi}D_{q}-2\lambda_{2}D_{q}\boxtimes\partial_{\tau}. M τ ξ ⋅ ( D q ⊠ D q ∂ τ ) = 2 λ 1 1 ⊠ D q ∂ τ − D q ⊠ D ξ D q − 2 λ 2 D q ⊠ ∂ τ .
Consequently,
M τ ξ ⋅ J 1 ˉ , B ( 4 ) = C 3 ( D ξ D q ⊠ D q + 2 λ 1 ∂ τ ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 ) + G 1 ( 2 λ 1 1 ⊠ D q ∂ τ − D q ⊠ D ξ D q − 2 λ 2 D q ⊠ ∂ τ ) . M_{\tau\xi}\cdot J^{(4)}_{\bar{1},B}=C_{3}(D_{\xi}D_{q}\boxtimes D_{q}+2\lambda_{1}\partial_{\tau}\boxtimes D_{q}+2\lambda_{2}D_{q}\partial_{\tau}\boxtimes1)
+G_{1}(2\lambda_{1}1\boxtimes D_{q}\partial_{\tau}-D_{q}\boxtimes D_{\xi}D_{q}-2\lambda_{2}D_{q}\boxtimes\partial_{\tau}). M τ ξ ⋅ J 1 ˉ , B ( 4 ) = C 3 ( D ξ D q ⊠ D q + 2 λ 1 ∂ τ ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 ) + G 1 ( 2 λ 1 1 ⊠ D q ∂ τ − D q ⊠ D ξ D q − 2 λ 2 D q ⊠ ∂ τ ) .
Comparing the coefficients of the distinct monomials D ξ D q ⊠ D q D_{\xi}D_{q}\boxtimes D_{q} D ξ D q ⊠ D q and D q ⊠ D ξ D q D_{q}\boxtimes D_{\xi}D_{q} D q ⊠ D ξ D q , we get C 3 = G 1 = 0 C_{3}=G_{1}=0 C 3 = G 1 = 0 . Thus,
J 1 ˉ , B ( 4 ) = 0. J^{(4)}_{\bar{1},B}=0. J 1 ˉ , B ( 4 ) = 0.
Hence the first degree in which a nonzero odd Type B operator may occur is k = 5 k=5 k = 5 . We shall therefore study explicitly the cases k = 5 k=5 k = 5 and k = 6 k=6 k = 6 . The degree k = 5 k=5 k = 5 gives the first nonzero Type B operator, while k = 6 k=6 k = 6 is the first degree in which the coefficients satisfy a nontrivial recurrence.
Degree k = 5 k=5 k = 5 . In this case, the odd Type B operator has the form
J 1 ˉ , B ( 5 ) = C 3 D q ∂ τ ⊠ D ξ D q + C 4 D ξ D q ∂ τ ⊠ D q + D 3 D q ∂ τ ⊠ ∂ τ + G 1 D q ⊠ D ξ D q ∂ τ + G 2 D ξ D q ⊠ D q ∂ τ + H 2 ∂ τ ⊠ D q ∂ τ . J^{(5)}_{\bar{1},B}=C_{3}D_{q}\partial_{\tau}\boxtimes D_{\xi}D_{q}+C_{4}D_{\xi}D_{q}\partial_{\tau}\boxtimes D_{q}+D_{3}D_{q}\partial_{\tau}\boxtimes\partial_{\tau}
+G_{1}D_{q}\boxtimes D_{\xi}D_{q}\partial_{\tau}+G_{2}D_{\xi}D_{q}\boxtimes D_{q}\partial_{\tau}+H_{2}\partial_{\tau}\boxtimes D_{q}\partial_{\tau}. J 1 ˉ , B ( 5 ) = C 3 D q ∂ τ ⊠ D ξ D q + C 4 D ξ D q ∂ τ ⊠ D q + D 3 D q ∂ τ ⊠ ∂ τ + G 1 D q ⊠ D ξ D q ∂ τ + G 2 D ξ D q ⊠ D q ∂ τ + H 2 ∂ τ ⊠ D q ∂ τ .
A direct computation of the M τ ξ M_{\tau\xi} M τ ξ -invariance condition gives
C 4 = G 1 = 0 , G 2 = C 3 , D 3 = 2 λ 2 C 3 , H 2 = 2 λ 1 C 3 . C_{4}=G_{1}=0,\qquad G_{2}=C_{3},\qquad D_{3}=2\lambda_{2}C_{3},\qquad H_{2}=2\lambda_{1}C_{3}. C 4 = G 1 = 0 , G 2 = C 3 , D 3 = 2 λ 2 C 3 , H 2 = 2 λ 1 C 3 .
Hence, up to an overall scalar,
J 1 ˉ , B ( 5 ) = D q ∂ τ ⊠ D ξ D q + 2 λ 2 D q ∂ τ ⊠ ∂ τ + D ξ D q ⊠ D q ∂ τ + 2 λ 1 ∂ τ ⊠ D q ∂ τ . J^{(5)}_{\bar{1},B}=D_{q}\partial_{\tau}\boxtimes D_{\xi}D_{q}+2\lambda_{2}D_{q}\partial_{\tau}\boxtimes\partial_{\tau}+D_{\xi}D_{q}\boxtimes D_{q}\partial_{\tau}+2\lambda_{1}\partial_{\tau}\boxtimes D_{q}\partial_{\tau}. J 1 ˉ , B ( 5 ) = D q ∂ τ ⊠ D ξ D q + 2 λ 2 D q ∂ τ ⊠ ∂ τ + D ξ D q ⊠ D q ∂ τ + 2 λ 1 ∂ τ ⊠ D q ∂ τ .
A direct computation shows that this operator is also invariant under M q 2 M_{q^{2}} M q 2 , and no further condition on the weights is obtained. Therefore,
dim Inv 1 ˉ , B ( 5 ) = 1 \dim\operatorname{Inv}^{(5)}_{\bar{1},B}=1 dim Inv 1 ˉ , B ( 5 ) = 1
for all ( λ 1 , λ 2 ) (\lambda_{1},\lambda_{2}) ( λ 1 , λ 2 ) .
Odd Type B operators: the generic case for k ≥ 6 k\geq6 k ≥ 6
The most general odd Type B operator is
J 1 ˉ , B ( k ) = ∑ 3 ≤ i ≤ k − 1 C i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + ∑ 3 ≤ i ≤ k − 2 D i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + ∑ 1 ≤ i ≤ k − 3 G i D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ + ∑ 2 ≤ i ≤ k − 3 H i D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ . (68) J^{(k)}_{\bar{1},B}=
\sum_{3\leq i\leq k-1}C_{i}D_{\xi}^{i-3}D_{q}\partial_{\tau}\boxtimes D_{\xi}^{k-i-1}D_{q}
+\sum_{3\leq i\leq k-2}D_{i}D_{\xi}^{i-3}D_{q}\partial_{\tau}\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}
+\sum_{1\leq i\leq k-3}G_{i}D_{\xi}^{i-1}D_{q}\boxtimes D_{\xi}^{k-i-3}D_{q}\partial_{\tau}
+\sum_{2\leq i\leq k-3}H_{i}D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i-3}D_{q}\partial_{\tau}. \tag*{(68)} J 1 ˉ , B ( k ) = 3 ≤ i ≤ k − 1 ∑ C i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 3 ≤ i ≤ k − 2 ∑ D i D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + 1 ≤ i ≤ k − 3 ∑ G i D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ + 2 ≤ i ≤ k − 3 ∑ H i D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ . (68)
Invariance under M τ ξ M_{\tau\xi} M τ ξ and M q 2 M_{q^{2}} M q 2 reduces (68) to a bidiagonal recurrence for the coefficients C i C_{i} C i ; its solution gives the following generic classification.
Theorem. Odd Type B operators for generic weights
Theorem 8.8. (Odd Type B operators for generic weights). Let k ≥ 6 k\geq6 k ≥ 6 , and assume that the weights are generic so that the coefficient system below does not lose rank. Then, the space of odd Type B p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators
J 1 ˉ , B ( k ) : F λ 1 , ν 1 ⊠ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 − k + 3 , ν 1 + ν 2 + k − 6 2 J^{(k)}_{\bar{1},B}:\mathcal{F}_{\lambda_{1},\nu_{1}}\boxtimes\mathcal{F}_{\lambda_{2},\nu_{2}}\longrightarrow\mathcal{F}_{\lambda_{1}+\lambda_{2}-k+3,\nu_{1}+\nu_{2}+\frac{k-6}{2}} J 1 ˉ , B ( k ) : F λ 1 , ν 1 ⊠ F λ 2 , ν 2 ⟶ F λ 1 + λ 2 − k + 3 , ν 1 + ν 2 + 2 k − 6
is one-dimensional. Up to an overall scalar, the operator is
J 1 , B ( k ) = ∑ 3 ≤ i ≤ k − 2 C i ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + D ξ i − 2 D q ⊠ D ξ k − i − 2 D q ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 D q ∂ τ ) , J^{(k)}_{1,B}=\sum_{3\leq i\leq k-2}C_i\left(D_{\xi}^{i-3}D_q\partial_\tau\boxtimes D_{\xi}^{k-i-1}D_q+2\lambda_2D_{\xi}^{i-3}D_q\partial_\tau\boxtimes D_{\xi}^{k-i-2}\partial_\tau+D_{\xi}^{i-2}D_q\boxtimes D_{\xi}^{k-i-2}D_q\partial_\tau+2\lambda_1D_{\xi}^{i-3}\partial_\tau\boxtimes D_{\xi}^{k-i-2}D_q\partial_\tau\right), J 1 , B ( k ) = 3 ≤ i ≤ k − 2 ∑ C i ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + D ξ i − 2 D q ⊠ D ξ k − i − 2 D q ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 D q ∂ τ ) ,
where the coefficients C i C_i C i satisfy
( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0. (i-2)(2\lambda_1-i+1)C_{i+1}+(k-i-2)(2\lambda_2-k+i+1)C_i=0. ( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0.
Equivalently,
C i = ( − 1 ) i − 3 ( k − 5 i − 3 ) ( ∏ 3 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) ) ( ∏ i ≤ r ≤ k − 3 ( 2 λ 1 − r + 1 ) ) , 3 ≤ i ≤ k − 2. C_i=(-1)^{i-3}\binom{k-5}{i-3}\left(\prod_{3\leq r\leq i-1}(2\lambda_2-k+r+1)\right)\left(\prod_{i\leq r\leq k-3}(2\lambda_1-r+1)\right),\qquad3\leq i\leq k-2. C i = ( − 1 ) i − 3 ( i − 3 k − 5 ) ( 3 ≤ r ≤ i − 1 ∏ ( 2 λ 2 − k + r + 1 ) ) ( i ≤ r ≤ k − 3 ∏ ( 2 λ 1 − r + 1 ) ) , 3 ≤ i ≤ k − 2.
Low-degree resonances for odd Type B operators
Before studying the resonant Type B system for arbitrary k k k , we consider the first two nonzero cases, namely k = 5 k=5 k = 5 and k = 6 k=6 k = 6 .
Degree k = 5 k=5 k = 5 . After imposing M τ ξ M_{\tau\xi} M τ ξ -invariance, every Type B operator of degree 5 is determined by a single coefficient C 3 C_3 C 3 and has the form
J 1 , B ( 5 ) = C 3 ( D q ∂ τ ⊠ D ξ D q + 2 λ 2 D q ∂ τ ⊠ ∂ τ + D ξ D q ⊠ D q ∂ τ + 2 λ 1 ∂ τ ⊠ D q ∂ τ ) . J^{(5)}_{1,B}=C_3\left(D_q\partial_\tau\boxtimes D_\xi D_q+2\lambda_2D_q\partial_\tau\boxtimes\partial_\tau+D_\xi D_q\boxtimes D_q\partial_\tau+2\lambda_1\partial_\tau\boxtimes D_q\partial_\tau\right). J 1 , B ( 5 ) = C 3 ( D q ∂ τ ⊠ D ξ D q + 2 λ 2 D q ∂ τ ⊠ ∂ τ + D ξ D q ⊠ D q ∂ τ + 2 λ 1 ∂ τ ⊠ D q ∂ τ ) .
The M q 2 M_{q^2} M q 2 -invariance equations are governed, in the stable Type B system, by the recurrence
( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , 3 ≤ i ≤ k − 3. (i-2)(2\lambda_1-i+1)C_{i+1}+(k-i-2)(2\lambda_2-k+i+1)C_i=0,\qquad3\leq i\leq k-3. ( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , 3 ≤ i ≤ k − 3.
For k = 5 k=5 k = 5 , this index range is empty. Consequently, M q 2 M_{q^2} M q 2 -invariance imposes no further condition on C 3 C_3 C 3 . Therefore, C 3 C_3 C 3 is arbitrary for every pair ( λ 1 , λ 2 ) (\lambda_1,\lambda_2) ( λ 1 , λ 2 ) , and
dim Inv 1 , B ( 5 ) = 1. \dim\operatorname{Inv}^{(5)}_{1,B}=1. dim Inv 1 , B ( 5 ) = 1.
Thus there are no resonant weights in degree 5: the dimension of the Type B invariant space is equal to 1 for all weights.
Degree k = 6 k=6 k = 6 . After imposing M τ ξ M_{\tau\xi} M τ ξ -invariance, the Type B operator depends on two coefficients C 3 C_3 C 3 and C 4 C_4 C 4 :
J 1 , B ( 6 ) = C 3 D q ∂ τ ⊠ D ξ 2 D q + C 4 D ξ D q ∂ τ ⊠ D ξ D q + 2 λ 2 C 3 D q ∂ τ ⊠ D ξ ∂ τ + 2 λ 2 C 4 D ξ D q ∂ τ ⊠ ∂ τ + C 3 D ξ D q ⊠ D ξ D q ∂ τ + C 4 D ξ 2 D q ⊠ D q ∂ τ + 2 λ 1 C 3 ∂ τ ⊠ D ξ D q ∂ τ + 2 λ 1 C 4 D ξ ∂ τ ⊠ D q ∂ τ . J^{(6)}_{1,B}=C_3D_q\partial_\tau\boxtimes D_\xi^2D_q+C_4D_\xi D_q\partial_\tau\boxtimes D_\xi D_q+2\lambda_2C_3D_q\partial_\tau\boxtimes D_\xi\partial_\tau+2\lambda_2C_4D_\xi D_q\partial_\tau\boxtimes\partial_\tau
+C_3D_\xi D_q\boxtimes D_\xi D_q\partial_\tau+C_4D_\xi^2D_q\boxtimes D_q\partial_\tau+2\lambda_1C_3\partial_\tau\boxtimes D_\xi D_q\partial_\tau+2\lambda_1C_4D_\xi\partial_\tau\boxtimes D_q\partial_\tau. J 1 , B ( 6 ) = C 3 D q ∂ τ ⊠ D ξ 2 D q + C 4 D ξ D q ∂ τ ⊠ D ξ D q + 2 λ 2 C 3 D q ∂ τ ⊠ D ξ ∂ τ + 2 λ 2 C 4 D ξ D q ∂ τ ⊠ ∂ τ + C 3 D ξ D q ⊠ D ξ D q ∂ τ + C 4 D ξ 2 D q ⊠ D q ∂ τ + 2 λ 1 C 3 ∂ τ ⊠ D ξ D q ∂ τ + 2 λ 1 C 4 D ξ ∂ τ ⊠ D q ∂ τ .
In this degree, the M q 2 M_{q^2} M q 2 -invariance condition consists of the single equation
( 2 λ 1 − 2 ) C 4 + ( 2 λ 2 − 2 ) C 3 = 0. (2\lambda_1-2)C_4+(2\lambda_2-2)C_3=0. ( 2 λ 1 − 2 ) C 4 + ( 2 λ 2 − 2 ) C 3 = 0.
If ( λ 1 , λ 2 ) ≠ ( 1 , 1 ) (\lambda_1,\lambda_2)\neq(1,1) ( λ 1 , λ 2 ) = ( 1 , 1 ) , this is a nontrivial linear relation between C 3 C_3 C 3 and C 4 C_4 C 4 . Hence,
dim Inv 1 , B ( 6 ) = 1. \dim\operatorname{Inv}^{(6)}_{1,B}=1. dim Inv 1 , B ( 6 ) = 1.
At the exceptional point ( λ 1 , λ 2 ) = ( 1 , 1 ) (\lambda_1,\lambda_2)=(1,1) ( λ 1 , λ 2 ) = ( 1 , 1 ) , both coefficients C 3 C_3 C 3 and C 4 C_4 C 4 become independent, and
dim Inv 1 , B ( 6 ) = 2. \dim\operatorname{Inv}^{(6)}_{1,B}=2. dim Inv 1 , B ( 6 ) = 2.
A convenient basis of the two-dimensional resonant space is obtained by taking successively
( C 3 , C 4 ) = ( 1 , 0 ) and ( C 3 , C 4 ) = ( 0 , 1 ) . (C_3,C_4)=(1,0)\qquad\text{and}\qquad(C_3,C_4)=(0,1). ( C 3 , C 4 ) = ( 1 , 0 ) and ( C 3 , C 4 ) = ( 0 , 1 ) .
This gives
J 1 , B ; 1 ( 6 ) = D q ∂ τ ⊠ D ξ 2 D q + 2 D q ∂ τ ⊠ D ξ ∂ τ + D ξ D q ⊠ D ξ D q ∂ τ + 2 ∂ τ ⊠ D ξ D q ∂ τ , J^{(6)}_{1,B;1}=D_q\partial_\tau\boxtimes D_\xi^2D_q+2D_q\partial_\tau\boxtimes D_\xi\partial_\tau+D_\xi D_q\boxtimes D_\xi D_q\partial_\tau+2\partial_\tau\boxtimes D_\xi D_q\partial_\tau, J 1 , B ; 1 ( 6 ) = D q ∂ τ ⊠ D ξ 2 D q + 2 D q ∂ τ ⊠ D ξ ∂ τ + D ξ D q ⊠ D ξ D q ∂ τ + 2 ∂ τ ⊠ D ξ D q ∂ τ ,
and
J 1 , B ; 2 ( 6 ) = D ξ D q ∂ τ ⊠ D ξ D q + 2 D ξ D q ∂ τ ⊠ ∂ τ + D ξ 2 D q ⊠ D q ∂ τ + 2 D ξ ∂ τ ⊠ D q ∂ τ . J^{(6)}_{1,B;2}=\mathcal{D}_{\xi}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}\mathcal{D}_{q}+2\mathcal{D}_{\xi}\mathcal{D}_{q}\partial_{\tau}\boxtimes\partial_{\tau}+\mathcal{D}_{\xi}^{2}\mathcal{D}_{q}\boxtimes\mathcal{D}_{q}\partial_{\tau}+2\mathcal{D}_{\xi}\partial_{\tau}\boxtimes\mathcal{D}_{q}\partial_{\tau}. J 1 , B ; 2 ( 6 ) = D ξ D q ∂ τ ⊠ D ξ D q + 2 D ξ D q ∂ τ ⊠ ∂ τ + D ξ 2 D q ⊠ D q ∂ τ + 2 D ξ ∂ τ ⊠ D q ∂ τ .
We therefore obtain
dim Inv 1 ˉ , B ( 6 ) = { 2 , ( λ 1 , λ 2 ) = ( 1 , 1 ) , 1 , otherwise . \dim\operatorname{Inv}^{(6)}_{\bar{1},B}=
\begin{cases}
2, & (\lambda_{1},\lambda_{2})=(1,1),\\
1, & \text{otherwise}.
\end{cases} dim Inv 1 ˉ , B ( 6 ) = { 2 , 1 , ( λ 1 , λ 2 ) = ( 1 , 1 ) , otherwise .
The cases k = 5 k=5 k = 5 and k = 6 k=6 k = 6 illustrate the beginning of the general pattern. In degree 5 5 5 , there is only one coefficient and hence no possible rank drop. Starting from degree 6 6 6 , the Type B coefficients form a bidiagonal system, and resonant weights are precisely those at which the rank of this system decreases. We now analyze this phenomenon for arbitrary k ≥ 6 k\ge6 k ≥ 6 .
Resonant odd Type B operators for k6
We now determine the exceptional weights for odd Type B operators in arbitrary degree k ≥ 6 k\ge6 k ≥ 6 .
The answer is the following.
Proposition 8.9. (The resonant weights of odd Type B operators). Let k ≥ 6 k\ge6 k ≥ 6 . The odd Type B coefficient system is resonant precisely when
2 λ 1 , 2 λ 2 ∈ { 2 , 3 , … , k − 4 } and 2 λ 1 + 2 λ 2 ≥ k − 2. 2\lambda_{1},2\lambda_{2}\in\{2,3,\ldots,k-4\}\quad\text{and}\quad2\lambda_{1}+2\lambda_{2}\ge k-2. 2 λ 1 , 2 λ 2 ∈ { 2 , 3 , … , k − 4 } and 2 λ 1 + 2 λ 2 ≥ k − 2. At every resonant point,
dim Inv 1 ˉ , B ( k ) = 2. \dim\operatorname{Inv}^{(k)}_{\bar{1},B}=2. dim Inv 1 ˉ , B ( k ) = 2. At all other weights,
dim Inv 1 ˉ , B ( k ) = 1. \dim\operatorname{Inv}^{(k)}_{\bar{1},B}=1. dim Inv 1 ˉ , B ( k ) = 1.
Remark. The only resonant point for k=6
For k = 6 k=6 k = 6 , the conditions reduce to 2 λ 1 = 2 λ 2 = 2 2\lambda_{1}=2\lambda_{2}=2 2 λ 1 = 2 λ 2 = 2 , so the unique resonant point is ( λ 1 , λ 2 ) = ( 1 , 1 ) (\lambda_{1},\lambda_{2})=(1,1) ( λ 1 , λ 2 ) = ( 1 , 1 ) , in agreement with the direct degree-6 computation.
For k = 7 k=7 k = 7 , we have
2 λ 1 , 2 λ 2 ∈ { 2 , 3 } , 2 λ 1 + 2 λ 2 ≥ 5. 2\lambda_{1},2\lambda_{2}\in\{2,3\},\qquad2\lambda_{1}+2\lambda_{2}\ge5. 2 λ 1 , 2 λ 2 ∈ { 2 , 3 } , 2 λ 1 + 2 λ 2 ≥ 5.
Hence the resonant points are
( λ 1 , λ 2 ) ∈ { ( 1 , 3 2 ) , ( 3 2 , 1 ) , ( 3 2 , 3 2 ) } . (\lambda_{1},\lambda_{2})\in\left\{\left(1,\frac{3}{2}\right),\left(\frac{3}{2},1\right),\left(\frac{3}{2},\frac{3}{2}\right)\right\}. ( λ 1 , λ 2 ) ∈ { ( 1 , 2 3 ) , ( 2 3 , 1 ) , ( 2 3 , 2 3 ) } .
At each of these points the rank drops by 1 1 1 , and therefore
dim Inv 1 ˉ , B ( 7 ) = 2. \dim\operatorname{Inv}^{(7)}_{\bar{1},B}=2. dim Inv 1 ˉ , B ( 7 ) = 2.
Even operators of type A
We first consider even bilinear differential operators of type A and show that the p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariance conditions rule out any nonzero operator of this type.
Proposition 8.12. (There are no even Type A operators). For every k ≥ 1 k\ge1 k ≥ 1 , there are no nonzero even Type A p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators.
Even operators of type B: the generic case
We now turn to even bilinear differential operators of type B in low degrees. We first treat the generic case, where the invariance conditions determine the operators without degeneracy. Then, we examine the resonant values of the weights separately; for them, the rank of the resulting coefficient system may drop and additional invariant operators may occur.
Even type B operators in low degrees
We begin by examining operators of low degree.
Proposition. Even Type B operators of degree k ≤ 5 k \le5 k ≤ 5
For even Type B operators, the generic invariant space satisfies
dim Inv 0 ˉ , B ( k ) = { 0 , k = 2 , 2 , k = 3 , 3 , k = 4 , 2 , k = 5. \dim\operatorname{Inv}^{(k)}_{\bar{0},B} =
\begin{cases}
0, & k=2,\\
2, & k=3,\\
3, & k=4,\\
2, & k=5.
\end{cases} dim Inv 0 ˉ , B ( k ) = ⎩ ⎨ ⎧ 0 , 2 , 3 , 2 , k = 2 , k = 3 , k = 4 , k = 5.
More precisely, for k = 2 k=2 k = 2 , the invariance conditions force all coefficients to vanish at generic weights. For k = 3 k=3 k = 3 , a basis is
D q ⊠ Bol ( λ 2 ) − 2 λ 1 1 ⊠ D q ∂ τ , Bol ( λ 1 ) ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 , \mathcal{D}_q \boxtimes\operatorname{Bol}^{(\lambda_2)}-2\lambda_1\,1\boxtimes\mathcal{D}_q\partial_\tau,\qquad
\operatorname{Bol}^{(\lambda_1)}\boxtimes\mathcal{D}_q+2\lambda_2\mathcal{D}_q\partial_\tau\boxtimes1, D q ⊠ Bol ( λ 2 ) − 2 λ 1 1 ⊠ D q ∂ τ , Bol ( λ 1 ) ⊠ D q + 2 λ 2 D q ∂ τ ⊠ 1 ,
where Bol ( a ) = D ξ D q + 2 a ∂ τ \operatorname{Bol}^{(a)}=\mathcal{D}_\xi\mathcal{D}_q+2a\partial_\tau Bol ( a ) = D ξ D q + 2 a ∂ τ is the operator (65); these are the two operators of order 3 and charge + 1 +1 + 1 of Statement 7.4(ii). For k = 4 k=4 k = 4 , one of the three basis operators is Bol ( λ 1 ) ⊠ Bol ( λ 2 ) \operatorname{Bol}^{(\lambda_1)}\boxtimes\operatorname{Bol}^{(\lambda_2)} Bol ( λ 1 ) ⊠ Bol ( λ 2 ) . For k = 5 k=5 k = 5 , the target weights are
μ = λ 1 + λ 2 − 3 , ω = ν 1 + ν 2 + 1 2 , \mu=\lambda_1+\lambda_2-3,\qquad\omega=\nu_1+\nu_2+\frac{1}{2}, μ = λ 1 + λ 2 − 3 , ω = ν 1 + ν 2 + 2 1 ,
and the full Type B ansatz is
J 0 ˉ , B ( 5 ) = ∑ 0 ≤ i ≤ 3 A i D ξ i D q ⊠ D ξ 3 − i D q + ∑ 0 ≤ i ≤ 2 B i D ξ i D q ⊠ D ξ 2 − i ∂ τ J_{\bar{0},B}^{(5)}
=
\sum_{0\leq i\leq3} A_i\mathcal{D}_\xi^i\mathcal{D}_q\boxtimes\mathcal{D}_\xi^{3-i}\mathcal{D}_q
+
\sum_{0\leq i\leq2} B_i\mathcal{D}_\xi^i\mathcal{D}_q\boxtimes\mathcal{D}_\xi^{2-i}\partial_\tau J 0 ˉ , B ( 5 ) = 0 ≤ i ≤ 3 ∑ A i D ξ i D q ⊠ D ξ 3 − i D q + 0 ≤ i ≤ 2 ∑ B i D ξ i D q ⊠ D ξ 2 − i ∂ τ
+ ∑ 0 ≤ i ≤ 2 C i D ξ i ∂ τ ⊠ D ξ 2 − i D q + ∑ 0 ≤ i ≤ 2 D i D ξ i D q ∂ τ ⊠ D ξ 2 − i + ∑ 0 ≤ i ≤ 2 E i D ξ i ⊠ D ξ 2 − i D q ∂ τ \qquad+
\sum_{0\leq i\leq2} C_i\mathcal{D}_\xi^i\partial_\tau\boxtimes\mathcal{D}_\xi^{2-i}\mathcal{D}_q
+
\sum_{0\leq i\leq2} D_i\mathcal{D}_\xi^i\mathcal{D}_q\partial_\tau\boxtimes\mathcal{D}_\xi^{2-i}
+
\sum_{0\leq i\leq2} E_i\mathcal{D}_\xi^i\boxtimes\mathcal{D}_\xi^{2-i}\mathcal{D}_q\partial_\tau + 0 ≤ i ≤ 2 ∑ C i D ξ i ∂ τ ⊠ D ξ 2 − i D q + 0 ≤ i ≤ 2 ∑ D i D ξ i D q ∂ τ ⊠ D ξ 2 − i + 0 ≤ i ≤ 2 ∑ E i D ξ i ⊠ D ξ 2 − i D q ∂ τ
+ ∑ 0 ≤ i ≤ 1 F i D ξ i ∂ τ ⊠ D ξ 1 − i ∂ τ . \qquad+
\sum_{0\leq i\leq1} F_i\mathcal{D}_\xi^i\partial_\tau\boxtimes\mathcal{D}_\xi^{1-i}\partial_\tau. + 0 ≤ i ≤ 1 ∑ F i D ξ i ∂ τ ⊠ D ξ 1 − i ∂ τ .
At generic weights, the equations
M τ ξ ⋅ J 0 ˉ , B ( 5 ) = 0 , M q 2 ⋅ J 0 ˉ , B ( 5 ) = 0 M_{\tau\xi}\cdot J_{\bar{0},B}^{(5)}=0,\qquad M_{q^2}\cdot J_{\bar{0},B}^{(5)}=0 M τ ξ ⋅ J 0 ˉ , B ( 5 ) = 0 , M q 2 ⋅ J 0 ˉ , B ( 5 ) = 0
have rank 16 on the 18 coefficients. Hence, the solution space is two-dimensional.
Even type B operators: the generic case for k ≥ 5 k \ge5 k ≥ 5
Recall that an even Type B operator has Cartan bidegree
( u , v ) = ( k − 2 , 2 ) , (u,v)=(k-2,2), ( u , v ) = ( k − 2 , 2 ) ,
and therefore
μ = λ 1 + λ 2 − k + 2 , ω = ν 1 + ν 2 + k − 4 2 . \mu=\lambda_1+\lambda_2-k+2,\qquad\omega=\nu_1+\nu_2+\frac{k-4}{2}. μ = λ 1 + λ 2 − k + 2 , ω = ν 1 + ν 2 + 2 k − 4 .
Theorem. The general even Type B operator, k ≥ 5 k \ge5 k ≥ 5
Let k ≥ 5 k\geq5 k ≥ 5 . An even type B operator has a general form:
J 0 ˉ , B ( k ) = ∑ 0 ≤ i ≤ k − 2 A i D ξ i D q ⊠ D ξ k − i − 2 D q + ∑ 0 ≤ i ≤ k − 3 B i D ξ i D q ⊠ D ξ k − i − 3 ∂ τ J_{\bar{0},B}^{(k)}
=
\sum_{0\leq i\leq k-2} A_i\mathcal{D}_\xi^i\mathcal{D}_q\boxtimes\mathcal{D}_\xi^{k-i-2}\mathcal{D}_q
+
\sum_{0\leq i\leq k-3} B_i\mathcal{D}_\xi^i\mathcal{D}_q\boxtimes\mathcal{D}_\xi^{k-i-3}\partial_\tau J 0 ˉ , B ( k ) = 0 ≤ i ≤ k − 2 ∑ A i D ξ i D q ⊠ D ξ k − i − 2 D q + 0 ≤ i ≤ k − 3 ∑ B i D ξ i D q ⊠ D ξ k − i − 3 ∂ τ
+ ∑ 0 ≤ i ≤ k − 3 C i D ξ i ∂ τ ⊠ D ξ k − i − 3 D q + ∑ 0 ≤ i ≤ k − 3 D i D ξ i D q ∂ τ ⊠ D ξ k − i − 3 \qquad+
\sum_{0\leq i\leq k-3} C_i\mathcal{D}_\xi^i\partial_\tau\boxtimes\mathcal{D}_\xi^{k-i-3}\mathcal{D}_q
+
\sum_{0\leq i\leq k-3} D_i\mathcal{D}_\xi^i\mathcal{D}_q\partial_\tau\boxtimes\mathcal{D}_\xi^{k-i-3} + 0 ≤ i ≤ k − 3 ∑ C i D ξ i ∂ τ ⊠ D ξ k − i − 3 D q + 0 ≤ i ≤ k − 3 ∑ D i D ξ i D q ∂ τ ⊠ D ξ k − i − 3
+ ∑ 0 ≤ i ≤ k − 3 E i D ξ i ⊠ D ξ k − i − 3 D q ∂ τ + ∑ 0 ≤ i ≤ k − 4 F i D ξ i ∂ τ ⊠ D ξ k − i − 4 ∂ τ . \qquad+
\sum_{0\leq i\leq k-3} E_i\mathcal{D}_\xi^i\boxtimes\mathcal{D}_\xi^{k-i-3}\mathcal{D}_q\partial_\tau
+
\sum_{0\leq i\leq k-4} F_i\mathcal{D}_\xi^i\partial_\tau\boxtimes\mathcal{D}_\xi^{k-i-4}\partial_\tau. + 0 ≤ i ≤ k − 3 ∑ E i D ξ i ⊠ D ξ k − i − 3 D q ∂ τ + 0 ≤ i ≤ k − 4 ∑ F i D ξ i ∂ τ ⊠ D ξ k − i − 4 ∂ τ .
At generic weights,
dim Inv 0 ‾ , B ( k ) = 2. \dim\operatorname{Inv}_{\overline{0},B}^{(k)}=2. dim Inv 0 , B ( k ) = 2.
Even operators of Type B: the resonant case
Resonant even Type B operators in low degrees
Statement. Exceptional weights of even Type B operators, k = 2 , 3 , 4 k=2,3,4 k = 2 , 3 , 4
For the even Type B sector in degrees k = 2 , 3 , 4 k=2,3,4 k = 2 , 3 , 4 , the exceptional weights and dimensions of the invariant spaces are as follows.
For k = 2 k=2 k = 2 ,
dim Inv 0 ‾ , B ( 2 ) = { 1 , ( λ 1 , λ 2 ) = ( 0 , 0 ) , 0 , otherwise . \dim\operatorname{Inv}_{\overline{0},B}^{(2)}=
\begin{cases}
1, & (\lambda_1,\lambda_2)=(0,0),\\
0, & \text{otherwise}.
\end{cases} dim Inv 0 , B ( 2 ) = { 1 , 0 , ( λ 1 , λ 2 ) = ( 0 , 0 ) , otherwise .
For k = 3 k=3 k = 3 and k = 4 k=4 k = 4 , there are no exceptional weights: the space is of dimension 2, resp. 3, at every pair ( λ 1 , λ 2 ) (\lambda_1,\lambda_2) ( λ 1 , λ 2 ) with 2 λ 1 , 2 λ 2 ∈ { − 2 , − 1 , … , 4 } 2\lambda_1,2\lambda_2\in\{-2,-1,\ldots,4\} 2 λ 1 , 2 λ 2 ∈ { − 2 , − 1 , … , 4 } , the generic value of Proposition 8.14.
Proposition. Exceptional weights of even Type B operators, k = 5 k=5 k = 5
For k = 5 k=5 k = 5 , the generic space of even Type B p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators is two-dimensional. Its dimension increases precisely at
R 5 = { ( 1 2 , 1 ) , ( 1 2 , 3 2 ) , ( 1 , 1 2 ) , ( 1 , 1 ) , ( 1 , 3 2 ) , ( 3 2 , 1 2 ) , ( 3 2 , 1 ) } . R_5=\left\{
\left(\frac{1}{2},1\right),
\left(\frac{1}{2},\frac{3}{2}\right),
\left(1,\frac{1}{2}\right),
(1,1),
\left(1,\frac{3}{2}\right),
\left(\frac{3}{2},\frac{1}{2}\right),
\left(\frac{3}{2},1\right)
\right\}. R 5 = { ( 2 1 , 1 ) , ( 2 1 , 2 3 ) , ( 1 , 2 1 ) , ( 1 , 1 ) , ( 1 , 2 3 ) , ( 2 3 , 2 1 ) , ( 2 3 , 1 ) } .
More precisely,
dim Inv 0 ‾ , B ( 5 ) = { 4 , ( λ 1 , λ 2 ) = ( 1 , 1 ) , 3 , ( λ 1 , λ 2 ) ∈ R 5 ∖ { ( 1 , 1 ) } , 2 , ( λ 1 , λ 2 ) ∉ R 5 . \dim\operatorname{Inv}_{\overline{0},B}^{(5)}=
\begin{cases}
4, & (\lambda_1,\lambda_2)=(1,1),\\
3, & (\lambda_1,\lambda_2)\in R_5\setminus\{(1,1)\},\\
2, & (\lambda_1,\lambda_2)\notin R_5.
\end{cases} dim Inv 0 , B ( 5 ) = ⎩ ⎨ ⎧ 4 , 3 , 2 , ( λ 1 , λ 2 ) = ( 1 , 1 ) , ( λ 1 , λ 2 ) ∈ R 5 ∖ {( 1 , 1 )} , ( λ 1 , λ 2 ) ∈ / R 5 .
Resonant even Type B operators for k 5 k5 k 5
Let us put
a : = 2 λ 1 , b : = 2 λ 2 . a:=2\lambda_1,\qquad b:=2\lambda_2. a := 2 λ 1 , b := 2 λ 2 .
For the even Type B sector, let
M k ( a , b ) \mathcal{M}_k(a,b) M k ( a , b )
denote the matrix of the full p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariance system. Its generic rank is
rank M k = 6 k − 14. \operatorname{rank}\mathcal{M}_k=6k-14. rank M k = 6 k − 14.
Define
R k = { ( a , b ) ∈ Z 2 ∣ 1 ≤ a , b ≤ k − 2 , k − 2 ≤ a + b ≤ 2 k − 5 } , R_k=\left\{(a,b)\in\mathbb{Z}^2\mid1\leq a,b\leq k-2,\quad k-2\leq a+b\leq2k-5\right\}, R k = { ( a , b ) ∈ Z 2 ∣ 1 ≤ a , b ≤ k − 2 , k − 2 ≤ a + b ≤ 2 k − 5 } ,
and
R k ( 2 ) = { ( a , b ) ∈ Z 2 ∣ 2 ≤ a , b ≤ k − 3 , k − 1 ≤ a + b ≤ 2 k − 6 } . R_k^{(2)}=\left\{(a,b)\in\mathbb{Z}^2\mid2\leq a,b\leq k-3,\quad k-1\leq a+b\leq2k-6\right\}. R k ( 2 ) = { ( a , b ) ∈ Z 2 ∣ 2 ≤ a , b ≤ k − 3 , k − 1 ≤ a + b ≤ 2 k − 6 } .
Then
R k ( 2 ) ⊂ R k . R_k^{(2)}\subset R_k. R k ( 2 ) ⊂ R k .
Theorem. The rank of the even Type B system at the resonant weights
For k ≥ 5 k\geq5 k ≥ 5 , we have
rank M k ( a , b ) = { 6 k − 16 , ( a , b ) ∈ R k ( 2 ) , 6 k − 15 , ( a , b ) ∈ R k ∖ R k ( 2 ) , 6 k − 14 , ( a , b ) ∉ R k . \operatorname{rank}\mathcal{M}_k(a,b)=
\begin{cases}
6k-16, & (a,b)\in R_k^{(2)},\\
6k-15, & (a,b)\in R_k\setminus R_k^{(2)},\\
6k-14, & (a,b)\notin R_k.
\end{cases} rank M k ( a , b ) = ⎩ ⎨ ⎧ 6 k − 16 , 6 k − 15 , 6 k − 14 , ( a , b ) ∈ R k ( 2 ) , ( a , b ) ∈ R k ∖ R k ( 2 ) , ( a , b ) ∈ / R k .
Equivalently,
dim Inv 0 ˉ , B ( k ) = { 4 , ( a , b ) ∈ R k ( 2 ) , 3 , ( a , b ) ∈ R k ∖ R k ( 2 ) , 2 , ( a , b ) ∉ R k . \dim\operatorname{Inv}^{(k)}_{\bar{0},B} =
\begin{cases}
4, & (a,b) \in\mathcal{R}^{(2)}_{k}, \\
3, & (a,b) \in\mathcal{R}_{k} \setminus\mathcal{R}^{(2)}_{k}, \\
2, & (a,b) \notin\mathcal{R}_{k}.
\end{cases} dim Inv 0 ˉ , B ( k ) = ⎩ ⎨ ⎧ 4 , 3 , 2 , ( a , b ) ∈ R k ( 2 ) , ( a , b ) ∈ R k ∖ R k ( 2 ) , ( a , b ) ∈ / R k .
Even operators of type C
We finally consider even bilinear differential operators of type C and show that the p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariance conditions rule out any nonzero operator of this type.
Theorem. (There are no even Type C operators). For every k ≥ 4 k \ge4 k ≥ 4 , there are no nonzero even Type C p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) -invariant binary differential operators. In other words,
J 0 ˉ , C ( k ) = 0 for all k ≥ 4 J^{(k)}_{\bar{0},C} = 0 \qquad\text{for all } k \ge4 J 0 ˉ , C ( k ) = 0 for all k ≥ 4 and for all values of the weights.
Appendix. Proofs
Collected here are the proofs of those statements of §§7 and 8 whose verification is a computation. The statements themselves, the operators they describe and the ansatz each computation starts from are in the body.
Proofs for §7
Proof of Lemma 7.9. For brevity, write
y : = p ( Y ) , p : = p ( P ) , r : = p ( R ) , x : = p ( φ ) . y := p(Y), \qquad p := p(P), \qquad r := p(R), \qquad x := p(\varphi). y := p ( Y ) , p := p ( P ) , r := p ( R ) , x := p ( φ ) .
By definition, the action of Y Y Y on a homogeneous binary operator J J J is
( Y ⋅ J ) ( φ , ψ ) = L Y ( J ( φ , ψ ) ) − ( − 1 ) y p ( J ) J ( L Y φ , ψ ) − ( − 1 ) y ( p ( J ) + x ) J ( φ , L Y ψ ) . (Y \cdot J)(\varphi,\psi) = L_Y(J(\varphi,\psi)) - (-1)^{yp(J)}J(L_Y\varphi,\psi) - (-1)^{y(p(J)+x)}J(\varphi,L_Y\psi). ( Y ⋅ J ) ( φ , ψ ) = L Y ( J ( φ , ψ )) − ( − 1 ) y p ( J ) J ( L Y φ , ψ ) − ( − 1 ) y ( p ( J ) + x ) J ( φ , L Y ψ ) .
Applying this formula to J = P ⊠ R J = P \boxtimes R J = P ⊠ R , using
( P ⊠ R ) ( φ , ψ ) = ( − 1 ) r x P ( φ ) R ( ψ ) , (P \boxtimes R)(\varphi,\psi) = (-1)^{rx}P(\varphi)R(\psi), ( P ⊠ R ) ( φ , ψ ) = ( − 1 ) r x P ( φ ) R ( ψ ) ,
and the graded Leibniz rule, we obtain
( Y ⋅ ( P ⊠ R ) ) ( φ , ψ ) = ( − 1 ) r x ( L Y ( P ( φ ) ) − ( − 1 ) y p P ( L Y φ ) ) R ( ψ ) + ( − 1 ) r x + y ( p + x ) P ( φ ) L Y ( R ( ψ ) ) − ( − 1 ) r x + y ( p + r + x ) P ( φ ) R ( L Y ψ ) . \begin{aligned}
(Y \cdot(P \boxtimes R))(\varphi,\psi) ={}& (-1)^{rx}\left(L_Y(P(\varphi)) - (-1)^{yp}P(L_Y\varphi)\right)R(\psi) \\
&+ (-1)^{rx+y(p+x)}P(\varphi)L_Y(R(\psi)) - (-1)^{rx+y(p+r+x)}P(\varphi)R(L_Y\psi).
\end{aligned} ( Y ⋅ ( P ⊠ R )) ( φ , ψ ) = ( − 1 ) r x ( L Y ( P ( φ )) − ( − 1 ) y p P ( L Y φ ) ) R ( ψ ) + ( − 1 ) r x + y ( p + x ) P ( φ ) L Y ( R ( ψ )) − ( − 1 ) r x + y ( p + r + x ) P ( φ ) R ( L Y ψ ) .
The first line is
( ( Y ⋅ P ) ⊠ R ) ( φ , ψ ) , ((Y \cdot P) \boxtimes R)(\varphi,\psi), (( Y ⋅ P ) ⊠ R ) ( φ , ψ ) ,
while the last two lines combine to
( − 1 ) y p ( P ⊠ ( Y ⋅ R ) ) ( φ , ψ ) . (-1)^{yp}(P \boxtimes(Y \cdot R))(\varphi,\psi). ( − 1 ) y p ( P ⊠ ( Y ⋅ R )) ( φ , ψ ) .
Hence
Y ⋅ ( P ⊠ R ) = ( Y ⋅ P ) ⊠ R + ( − 1 ) p ( Y ) p ( P ) P ⊠ ( Y ⋅ R ) . Y \cdot(P \boxtimes R) = (Y \cdot P) \boxtimes R + (-1)^{p(Y)p(P)}P \boxtimes(Y \cdot R). Y ⋅ ( P ⊠ R ) = ( Y ⋅ P ) ⊠ R + ( − 1 ) p ( Y ) p ( P ) P ⊠ ( Y ⋅ R ) .
Proof of Lemma 7.10. We begin with M 1 = 2 ∂ τ M_1 = 2\partial_\tau M 1 = 2 ∂ τ . Since D ξ D_\xi D ξ , D q D_q D q , ∂ τ \partial_\tau ∂ τ supercommute with M 1 M_1 M 1 , the action of M 1 M_1 M 1 falls entirely on the coefficient functions:
M 1 ⋅ J = ∑ α , β M 1 ( F α , β ) P α ⊠ P β . M_1 \cdot J = \sum_{\alpha,\beta} M_1(F_{\alpha,\beta}) P_\alpha\boxtimes P_\beta. M 1 ⋅ J = α , β ∑ M 1 ( F α , β ) P α ⊠ P β .
Thus M 1 ⋅ J = 0 M_{1}\cdot J=0 M 1 ⋅ J = 0 implies ∂ τ F α , β = 0 \partial_{\tau}F_{\alpha,\beta}=0 ∂ τ F α , β = 0 . Applying the same argument to M q M_{q} M q and M ξ M_{\xi} M ξ eliminates the dependence on the remaining variables q q q and ξ \xi ξ . Hence,
F α , β ∈ C . F_{\alpha,\beta}\in\mathbb{C}. F α , β ∈ C .
Proofs for §8
Proof of Lemma 8.2. Recall that the action of H 1 H_{1} H 1 and H 2 H_{2} H 2 on the two-parameter density module F λ , ν \mathcal{F}_{\lambda,\nu} F λ , ν are
L H 1 μ , ω = H 1 − λ , and L H 2 μ , ω = H 2 − λ − 2 ν . L_{H_{1}}^{\mu,\omega}=H_{1}-\lambda,\qquad\text{and}\qquad L_{H_{2}}^{\mu,\omega}=H_{2}-\lambda-2\nu. L H 1 μ , ω = H 1 − λ , and L H 2 μ , ω = H 2 − λ − 2 ν .
Recall, the basic commutation relations are
[ H 1 , D ξ ] = − D ξ , [ H 1 , D q ] = 0 , [ H 1 , ∂ τ ] = − ∂ τ , [H_{1},D_{\xi}]=-D_{\xi},\qquad[H_{1},D_{q}]=0,\qquad[H_{1},\partial_{\tau}]=-\partial_{\tau}, [ H 1 , D ξ ] = − D ξ , [ H 1 , D q ] = 0 , [ H 1 , ∂ τ ] = − ∂ τ ,
and
[ H 2 , D ξ ] = 0 , [ H 2 , D q ] = − D q , [ H 2 , ∂ τ ] = − ∂ τ . [H_{2},D_{\xi}]=0,\qquad[H_{2},D_{q}]=-D_{q},\qquad[H_{2},\partial_{\tau}]=-\partial_{\tau}. [ H 2 , D ξ ] = 0 , [ H 2 , D q ] = − D q , [ H 2 , ∂ τ ] = − ∂ τ .
Since H 1 H_{1} H 1 and H 2 H_{2} H 2 are even, it follows that
[ H 1 , P j ] = − ( a j + δ j ) P j , and [ H 2 , P j ] = − ( ϵ j + δ j ) P j , j = 1 , 2. [H_{1},P_{j}]=-(a_{j}+\delta_{j})P_{j},\qquad\text{and}\qquad[H_{2},P_{j}]=-(\epsilon_{j}+\delta_{j})P_{j},\qquad j=1,2. [ H 1 , P j ] = − ( a j + δ j ) P j , and [ H 2 , P j ] = − ( ϵ j + δ j ) P j , j = 1 , 2.
Therefore, for the binary monomial P 1 ⊠ P 2 P_{1}\boxtimes P_{2} P 1 ⊠ P 2 ,
[ H 1 , P 1 ⊠ P 2 ] = − u ( P 1 ⊠ P 2 ) , and [ H 2 , P 1 ⊠ P 2 ] = − v ( P 1 ⊠ P 2 ) . [H_{1},P_{1}\boxtimes P_{2}]=-u\left(P_{1}\boxtimes P_{2}\right),\qquad\text{and}\qquad[H_{2},P_{1}\boxtimes P_{2}]=-v\left(P_{1}\boxtimes P_{2}\right). [ H 1 , P 1 ⊠ P 2 ] = − u ( P 1 ⊠ P 2 ) , and [ H 2 , P 1 ⊠ P 2 ] = − v ( P 1 ⊠ P 2 ) .
We now impose the Cartan invariance equations. Since H 1 H_{1} H 1 is even,
L H 1 μ , ω ( ( P 1 ⊠ P 2 ) ( φ , ψ ) ) = ( P 1 ⊠ P 2 ) ( L H 1 λ 1 , ν 1 φ , ψ ) + ( P 1 ⊠ P 2 ) ( φ , L H 1 λ 2 , ν 2 ψ ) . L_{H_{1}}^{\mu,\omega}\left((P_{1}\boxtimes P_{2})(\varphi,\psi)\right)=(P_{1}\boxtimes P_{2})\left(L_{H_{1}}^{\lambda_{1},\nu_{1}}\varphi,\psi\right)+(P_{1}\boxtimes P_{2})\left(\varphi,L_{H_{1}}^{\lambda_{2},\nu_{2}}\psi\right). L H 1 μ , ω ( ( P 1 ⊠ P 2 ) ( φ , ψ ) ) = ( P 1 ⊠ P 2 ) ( L H 1 λ 1 , ν 1 φ , ψ ) + ( P 1 ⊠ P 2 ) ( φ , L H 1 λ 2 , ν 2 ψ ) .
Using
L H 1 μ , ω = H 1 − μ , L H 1 λ 1 , ν 1 = H 1 − λ 1 , and L H 1 λ 2 , ν 2 = H 1 − λ 2 , L_{H_{1}}^{\mu,\omega}=H_{1}-\mu,\qquad L_{H_{1}}^{\lambda_{1},\nu_{1}}=H_{1}-\lambda_{1},\qquad\text{and}\qquad L_{H_{1}}^{\lambda_{2},\nu_{2}}=H_{1}-\lambda_{2}, L H 1 μ , ω = H 1 − μ , L H 1 λ 1 , ν 1 = H 1 − λ 1 , and L H 1 λ 2 , ν 2 = H 1 − λ 2 ,
we obtain
[ H 1 , P 1 ⊠ P 2 ] + ( λ 1 + λ 2 − μ ) ( P 1 ⊠ P 2 ) = 0. [H_{1},P_{1}\boxtimes P_{2}]+(\lambda_{1}+\lambda_{2}-\mu)(P_{1}\boxtimes P_{2})=0. [ H 1 , P 1 ⊠ P 2 ] + ( λ 1 + λ 2 − μ ) ( P 1 ⊠ P 2 ) = 0.
It follows that
μ = λ 1 + λ 2 − u . \mu=\lambda_{1}+\lambda_{2}-u. μ = λ 1 + λ 2 − u .
Similarly, the H 2 H_{2} H 2 -invariance equation gives
[ H 2 , P 1 ⊠ P 2 ] + ( λ 1 + 2 ν 1 + λ 2 + 2 ν 2 − μ − 2 ω ) ( P 1 ⊠ P 2 ) = 0. [H_{2},P_{1}\boxtimes P_{2}]+(\lambda_{1}+2\nu_{1}+\lambda_{2}+2\nu_{2}-\mu-2\omega)(P_{1}\boxtimes P_{2})=0. [ H 2 , P 1 ⊠ P 2 ] + ( λ 1 + 2 ν 1 + λ 2 + 2 ν 2 − μ − 2 ω ) ( P 1 ⊠ P 2 ) = 0.
Using
[ H 2 , P 1 ⊠ P 2 ] = − v ( P 1 ⊠ P 2 ) , [H_{2},P_{1}\boxtimes P_{2}]=-v(P_{1}\boxtimes P_{2}), [ H 2 , P 1 ⊠ P 2 ] = − v ( P 1 ⊠ P 2 ) ,
we obtain
μ + 2 ω = λ 1 + λ 2 + 2 ν 1 + 2 ν 2 − v . \mu+2\omega=\lambda_{1}+\lambda_{2}+2\nu_{1}+2\nu_{2}-v. μ + 2 ω = λ 1 + λ 2 + 2 ν 1 + 2 ν 2 − v .
Substituting μ = λ 1 + λ 2 − u \mu=\lambda_{1}+\lambda_{2}-u μ = λ 1 + λ 2 − u , we get
ω = ν 1 + ν 2 + u − v 2 . \omega=\nu_{1}+\nu_{2}+\frac{u-v}{2}. ω = ν 1 + ν 2 + 2 u − v .
This proves the claim.
Proof of Theorem 8.4. The action of M τ ξ M_{\tau\xi} M τ ξ on the monomials occurring in the ansatz (66) is determined by
M τ ξ ⋅ ( D ξ i ⊠ D ξ k − i − 1 D q ) = 2 λ 2 D ξ i ⊠ D ξ k − i − 1 , M τ ξ ⋅ ( D ξ i ⊠ D ξ k − i − 2 ∂ τ ) = − D ξ i ⊠ D ξ k − i − 1 M_{\tau\xi}\cdot\left(D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-1}D_{q}\right)=2\lambda_{2}D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-1},\qquad M_{\tau\xi}\cdot\left(D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}\right)=-D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-1} M τ ξ ⋅ ( D ξ i ⊠ D ξ k − i − 1 D q ) = 2 λ 2 D ξ i ⊠ D ξ k − i − 1 , M τ ξ ⋅ ( D ξ i ⊠ D ξ k − i − 2 ∂ τ ) = − D ξ i ⊠ D ξ k − i − 1
M τ ξ ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i ) = 2 λ 1 D ξ i − 1 ⊠ D ξ k − i , M τ ξ ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i ) = − D ξ i − 1 ⊠ D ξ k − i . M_{\tau\xi}\cdot\left(D_{\xi}^{i-1}D_{q}\boxtimes D_{\xi}^{k-i}\right)=2\lambda_{1}D_{\xi}^{i-1}\boxtimes D_{\xi}^{k-i},\qquad M_{\tau\xi}\cdot\left(D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i}\right)=-D_{\xi}^{i-1}\boxtimes D_{\xi}^{k-i}. M τ ξ ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i ) = 2 λ 1 D ξ i − 1 ⊠ D ξ k − i , M τ ξ ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i ) = − D ξ i − 1 ⊠ D ξ k − i .
By comparing the coefficients of D ξ i ⊠ D ξ k − i − 1 D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-1} D ξ i ⊠ D ξ k − i − 1 , for 0 ≤ i ≤ k − 1 0\leq i\leq k-1 0 ≤ i ≤ k − 1 , gives
2 λ 2 A i − B i + 2 λ 1 E i + 1 − F i + 1 = 0 , 0 ≤ i ≤ k − 1. (69) 2\lambda_{2}A_{i}-B_{i}+2\lambda_{1}E_{i+1}-F_{i+1}=0,\qquad0\leq i\leq k-1. \tag*{(69)} 2 λ 2 A i − B i + 2 λ 1 E i + 1 − F i + 1 = 0 , 0 ≤ i ≤ k − 1. (69)
where, at the endpoints, we use the convention
B k − 1 = 0 , F 1 = 0. B_{k-1}=0,\qquad F_1=0. B k − 1 = 0 , F 1 = 0.
Now the invariance under M q 2 M_{q^2} M q 2 gives
M q 2 ⋅ ( D ξ i ⊠ D ξ k − i − 1 D q ) = − 2 i D ξ i − 1 D q ⊠ D ξ k − i − 1 D q − 2 i ( i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 1 D q + 2 ( k − i − 1 ) ( k − i − 2 ) D ξ i ⊠ D ξ k − i − 3 D q ∂ τ . M_{q^2}\cdot\left(D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-1}D_q\right)=-2iD_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i-1}D_q-2i(i-1)D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i-1}D_q+2(k-i-1)(k-i-2)D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-3}D_q\partial_{\tau}. M q 2 ⋅ ( D ξ i ⊠ D ξ k − i − 1 D q ) = − 2 i D ξ i − 1 D q ⊠ D ξ k − i − 1 D q − 2 i ( i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 1 D q + 2 ( k − i − 1 ) ( k − i − 2 ) D ξ i ⊠ D ξ k − i − 3 D q ∂ τ .
M q 2 ⋅ ( D ξ i ⊠ D ξ k − i − 2 ∂ τ ) = − 2 i D ξ i − 1 D q ⊠ D ξ k − i − 2 ∂ τ − 2 i ( i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 ∂ τ − 2 ( k − i − 2 ) D ξ i ⊠ D ξ k − i − 3 D q ∂ τ . M_{q^2}\cdot\left(D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}\right)=-2iD_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}-2i(i-1)D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}-2(k-i-2)D_{\xi}^{i}\boxtimes D_{\xi}^{k-i-3}D_q\partial_{\tau}. M q 2 ⋅ ( D ξ i ⊠ D ξ k − i − 2 ∂ τ ) = − 2 i D ξ i − 1 D q ⊠ D ξ k − i − 2 ∂ τ − 2 i ( i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 ∂ τ − 2 ( k − i − 2 ) D ξ i ⊠ D ξ k − i − 3 D q ∂ τ .
M q 2 ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i ) = 2 ( i − 1 ) ( i − 2 ) D ξ i − 3 D q ∂ τ ⊠ D ξ k − i + 2 ( k − i ) D ξ i − 1 D q ⊠ D ξ k − i − 1 D q + 2 ( k − i ) ( k − i − 1 ) D ξ i − 1 D q ⊠ D ξ k − i − 2 ∂ τ . M_{q^2}\cdot\left(D_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i}\right)=2(i-1)(i-2)D_{\xi}^{i-3}D_q\partial_{\tau}\boxtimes D_{\xi}^{k-i}+2(k-i)D_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i-1}D_q+2(k-i)(k-i-1)D_{\xi}^{i-1}D_q\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}. M q 2 ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i ) = 2 ( i − 1 ) ( i − 2 ) D ξ i − 3 D q ∂ τ ⊠ D ξ k − i + 2 ( k − i ) D ξ i − 1 D q ⊠ D ξ k − i − 1 D q + 2 ( k − i ) ( k − i − 1 ) D ξ i − 1 D q ⊠ D ξ k − i − 2 ∂ τ .
M q 2 ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i ) = − 2 ( i − 2 ) D ξ i − 3 D q ∂ τ ⊠ D ξ k − i + 2 ( k − i ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 1 D q + 2 ( k − i ) ( k − i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 ∂ τ . M_{q^2}\cdot\left(D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i}\right)=-2(i-2)D_{\xi}^{i-3}D_q\partial_{\tau}\boxtimes D_{\xi}^{k-i}+2(k-i)D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i-1}D_q+2(k-i)(k-i-1)D_{\xi}^{i-2}\partial_{\tau}\boxtimes D_{\xi}^{k-i-2}\partial_{\tau}. M q 2 ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i ) = − 2 ( i − 2 ) D ξ i − 3 D q ∂ τ ⊠ D ξ k − i + 2 ( k − i ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 1 D q + 2 ( k − i ) ( k − i − 1 ) D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 ∂ τ .
Substituting (66) into M q 2 ⋅ J 1 , A ( k ) = 0 M_{q^2}\cdot J^{(k)}_{1,A}=0 M q 2 ⋅ J 1 , A ( k ) = 0 and comparing independent monomials gives
i A i = ( k − i ) E i , 1 ≤ i ≤ k − 1. iA_i=(k-i)E_i,\qquad1\leq i\leq k-1. i A i = ( k − i ) E i , 1 ≤ i ≤ k − 1.
together with
B i = ( k − i − 1 ) A i , 0 ≤ i ≤ k − 2. (70) B_i=(k-i-1)A_i,\qquad0\leq i\leq k-2. \tag*{(70)} B i = ( k − i − 1 ) A i , 0 ≤ i ≤ k − 2. (70)
and
F i = ( i − 1 ) E i , 2 ≤ i ≤ k . (71) F_i=(i-1)E_i,\qquad2\leq i\leq k. \tag*{(71)} F i = ( i − 1 ) E i , 2 ≤ i ≤ k . (71)
In particular,
E i = i k − i A i , 1 ≤ i ≤ k − 1. (72) E_i=\frac{i}{k-i}A_i,\qquad1\leq i\leq k-1. \tag*{(72)} E i = k − i i A i , 1 ≤ i ≤ k − 1. (72)
For 0 ≤ i ≤ k − 2 0\leq i\leq k-2 0 ≤ i ≤ k − 2 , substitute the relations (70), (71), and (72) into the equations (69). We obtain
( k − i − 1 ) A i + i E i + 1 = 2 λ 2 A i + 2 λ 1 E i + 1 . (k-i-1)A_i+iE_{i+1}=2\lambda_2A_i+2\lambda_1E_{i+1}. ( k − i − 1 ) A i + i E i + 1 = 2 λ 2 A i + 2 λ 1 E i + 1 .
Since
E i + 1 = i + 1 k − i − 1 A i + 1 , E_{i+1}=\frac{i+1}{k-i-1}A_{i+1}, E i + 1 = k − i − 1 i + 1 A i + 1 ,
this becomes
( k − i − 1 ) ( k − i − 1 − 2 λ 2 ) A i + ( i + 1 ) ( i − 2 λ 1 ) A i + 1 = 0. (k-i-1)(k-i-1-2\lambda_2)A_i+(i+1)(i-2\lambda_1)A_{i+1}=0. ( k − i − 1 ) ( k − i − 1 − 2 λ 2 ) A i + ( i + 1 ) ( i − 2 λ 1 ) A i + 1 = 0.
Equivalently,
( i + 1 ) ( 2 λ 1 − i ) A i + 1 + ( k − i − 1 ) ( 2 λ 2 − k + i + 1 ) A i = 0 , for 0 ≤ i ≤ k − 2. (i+1)(2\lambda_1-i)A_{i+1}+(k-i-1)(2\lambda_2-k+i+1)A_i=0,\qquad\text{for }0\leq i\leq k-2. ( i + 1 ) ( 2 λ 1 − i ) A i + 1 + ( k − i − 1 ) ( 2 λ 2 − k + i + 1 ) A i = 0 , for 0 ≤ i ≤ k − 2.
At generic weights, we assume in particular that
2 λ 1 − i ≠ 0 , 0 ≤ i ≤ k − 2. 2\lambda_1-i\ne0,\qquad0\leq i\leq k-2. 2 λ 1 − i = 0 , 0 ≤ i ≤ k − 2.
The recurrence can then be solved successively:
A i + 1 = − ( k − i − 1 ) ( 2 λ 2 − k + i + 1 ) ( i + 1 ) ( 2 λ 1 − i ) A i . (73) A_{i+1}=-\frac{(k-i-1)(2\lambda_2-k+i+1)}{(i+1)(2\lambda_1-i)}A_i. \tag*{(73)} A i + 1 = − ( i + 1 ) ( 2 λ 1 − i ) ( k − i − 1 ) ( 2 λ 2 − k + i + 1 ) A i . (73)
Thus, all the coefficients A 1 , … , A k − 1 A_1,\ldots,A_{k-1} A 1 , … , A k − 1 are determined by A 0 A_0 A 0 . Iterating (73) yields
A i = ( − 1 ) i ( k − 1 i ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 1 − r ) A 0 , 0 ≤ i ≤ k − 1 , A_i = (-1)^i \binom{k-1}{i}\frac{\prod_{0\le r\le i-1}(2\lambda_2-k+r+1)}{\prod_{0\le r\le i-1}(2\lambda_1-r)}A_0,\qquad0\le i\le k-1, A i = ( − 1 ) i ( i k − 1 ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 1 − r ) ∏ 0 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) A 0 , 0 ≤ i ≤ k − 1 ,
where an empty product is understood to be equal to 1.
It remains to determine E k E_k E k . The equation (69) at i = k − 1 i=k-1 i = k − 1 reads
2 λ 2 A k − 1 + 2 λ 1 E k − F k = 0. 2\lambda_2 A_{k-1}+2\lambda_1 E_k-F_k=0. 2 λ 2 A k − 1 + 2 λ 1 E k − F k = 0.
Using F k = ( k − 1 ) E k F_k=(k-1)E_k F k = ( k − 1 ) E k , we obtain
( k − 1 − 2 λ 1 ) E k = 2 λ 2 A k − 1 . (k-1-2\lambda_1)E_k=2\lambda_2 A_{k-1}. ( k − 1 − 2 λ 1 ) E k = 2 λ 2 A k − 1 .
At generic weights we also assume k − 1 − 2 λ 1 ≠ 0 k-1-2\lambda_1\ne0 k − 1 − 2 λ 1 = 0 . Hence
E k = 2 λ 2 k − 1 − 2 λ 1 A k − 1 . E_k=\frac{2\lambda_2}{k-1-2\lambda_1}A_{k-1}. E k = k − 1 − 2 λ 1 2 λ 2 A k − 1 .
All the remaining coefficients are then recovered from
B i = ( k − i − 1 ) A i , E i = i k − i A i , 1 ≤ i ≤ k − 1 , B_i=(k-i-1)A_i,\qquad E_i=\frac{i}{k-i}A_i,\qquad1\le i\le k-1, B i = ( k − i − 1 ) A i , E i = k − i i A i , 1 ≤ i ≤ k − 1 ,
and
F i = ( i − 1 ) E i . F_i=(i-1)E_i. F i = ( i − 1 ) E i .
Proof of Proposition 8.6. Set a : = 2 λ 1 a:=2\lambda_1 a := 2 λ 1 , and b : = 2 λ 2 b:=2\lambda_2 b := 2 λ 2 . The coefficients A 0 , A 1 , … , A k − 1 , E k A_0,A_1,\ldots,A_{k-1},E_k A 0 , A 1 , … , A k − 1 , E k satisfy
( i + 1 ) ( a − i ) A i + 1 + ( k − i − 1 ) ( b − k + i + 1 ) A i = 0 , 0 ≤ i ≤ k − 2 , (i+1)(a-i)A_{i+1}+(k-i-1)(b-k+i+1)A_i=0,\qquad0\le i\le k-2, ( i + 1 ) ( a − i ) A i + 1 + ( k − i − 1 ) ( b − k + i + 1 ) A i = 0 , 0 ≤ i ≤ k − 2 ,
together with the endpoint equation
( k − 1 − a ) E k − b A k − 1 = 0. (k-1-a)E_k-bA_{k-1}=0. ( k − 1 − a ) E k − b A k − 1 = 0.
Introduce
α i : = ( i + 1 ) ( a − i ) , β i : = ( k − i − 1 ) ( b − k + i + 1 ) , 0 ≤ i ≤ k − 2. \alpha_i:=(i+1)(a-i),\qquad\beta_i:=(k-i-1)(b-k+i+1),\qquad0\le i\le k-2. α i := ( i + 1 ) ( a − i ) , β i := ( k − i − 1 ) ( b − k + i + 1 ) , 0 ≤ i ≤ k − 2.
Then the system takes the matrix form
M A ( a , b ) ( A 0 A 1 ⋮ A k − 1 E k ) = 0 , \mathcal{M}_A(a,b)
\begin{pmatrix}
A_0\\
A_1\\
\vdots\\
A_{k-1}\\
E_k
\end{pmatrix}
=0, M A ( a , b ) A 0 A 1 ⋮ A k − 1 E k = 0 ,
where
M A ( a , b ) = ( β 0 α 0 0 ⋯ 0 0 0 β 1 α 1 ⋱ ⋮ ⋮ ⋮ ⋱ ⋱ ⋱ 0 0 0 ⋯ 0 β k − 2 α k − 2 0 0 ⋯ 0 0 − b k − 1 − a ) . \mathcal{M}_A(a,b)=
\begin{pmatrix}
\beta_0 & \alpha_0 & 0 & \cdots& 0 & 0\\
0 & \beta_1 & \alpha_1 & \ddots& \vdots& \vdots\\
\vdots& \ddots& \ddots& \ddots& 0 & 0\\
0 & \cdots& 0 & \beta_{k-2} & \alpha_{k-2} & 0\\
0 & \cdots& 0 & 0 & -b & k-1-a
\end{pmatrix}. M A ( a , b ) = β 0 0 ⋮ 0 0 α 0 β 1 ⋱ ⋯ ⋯ 0 α 1 ⋱ 0 0 ⋯ ⋱ ⋱ β k − 2 0 0 ⋮ 0 α k − 2 − b 0 ⋮ 0 0 k − 1 − a .
This is a k × ( k + 1 ) k\times(k+1) k × ( k + 1 ) bidiagonal matrix.
We now determine precisely when the rows become linearly dependent. Suppose that
c 0 R 0 + c 1 R 1 + ⋯ + c k − 1 R k − 1 = 0 c_0R_0 + c_1R_1 + \cdots+ c_{k-1}R_{k-1} = 0 c 0 R 0 + c 1 R 1 + ⋯ + c k − 1 R k − 1 = 0
is a nonzero linear relation among the rows of M A ( a , b ) \mathcal{M}_A(a,b) M A ( a , b ) . Let
p : = min { i ∣ c i ≠ 0 } , q : = max { i ∣ c i ≠ 0 } . p := \min\{i \mid c_i \ne0\}, \qquad q := \max\{i \mid c_i \ne0\}. p := min { i ∣ c i = 0 } , q := max { i ∣ c i = 0 } .
We first determine the possible starting index p p p .
If p ≤ k − 2 p \le k - 2 p ≤ k − 2 , it follows that c p β p = 0 c_p\beta_p = 0 c p β p = 0 . Since c p ≠ 0 c_p \ne0 c p = 0 , then β p = 0 \beta_p = 0 β p = 0 , and hence b − k + p + 1 = 0 b - k + p + 1 = 0 b − k + p + 1 = 0 . Therefore,
p = k − b − 1. p = k - b - 1. p = k − b − 1.
If p = k − 1 p = k - 1 p = k − 1 , it follows that − b c k − 1 = 0 -bc_{k-1} = 0 − b c k − 1 = 0 , so b = 0 b = 0 b = 0 . This is again described by
p = k − b − 1. p = k - b - 1. p = k − b − 1.
Consequently, a nonzero row relation can exist only if
b ∈ { 0 , 1 , … , k − 1 } , b \in\{0,1,\ldots,k-1\}, b ∈ { 0 , 1 , … , k − 1 } ,
and in this case its first nonzero row must be
p = k − b − 1 (74) p = k - b - 1 \tag*{(74)} p = k − b − 1 (74)
We next determine the possible terminal index q q q .
If q ≤ k − 2 q \le k - 2 q ≤ k − 2 , it follows that c q α q = 0 c_q\alpha_q = 0 c q α q = 0 . Since c q ≠ 0 c_q \ne0 c q = 0 , then α q = 0 \alpha_q = 0 α q = 0 , and therefore a = q a = q a = q .
If q = k − 1 q = k - 1 q = k − 1 , it follows that ( k − 1 − a ) c k − 1 = 0 (k - 1 - a)c_{k-1} = 0 ( k − 1 − a ) c k − 1 = 0 , and hence a = k − 1 a = k - 1 a = k − 1 .
Thus, in all cases
Consequently, a nonzero row relation can exist only if
a ∈ { 0 , 1 , … , k − 1 } , a \in\{0,1,\ldots,k-1\}, a ∈ { 0 , 1 , … , k − 1 } ,
and its last nonzero row must be
q = a (75) q = a \tag*{(75)} q = a (75)
For a nonzero relation to exist, the starting row must occur no later than the terminal row: p ≤ q p \le q p ≤ q . Using formulas (74) and (75), this becomes k − b − 1 ≤ a k - b - 1 \le a k − b − 1 ≤ a , or equivalently,
a + b ≥ k − 1. a + b \ge k - 1. a + b ≥ k − 1.
Conversely, suppose that
a , b ∈ { 0 , 1 , … , k − 1 } , a + b ≥ k − 1. a,b \in\{0,1,\ldots,k-1\}, \qquad a + b \ge k - 1. a , b ∈ { 0 , 1 , … , k − 1 } , a + b ≥ k − 1.
Set p = k − b − 1 p = k - b - 1 p = k − b − 1 and q = a q = a q = a . Then p ≤ q p \le q p ≤ q , and
β p = 0 , α q = 0. \beta_p = 0, \qquad\alpha_q = 0. β p = 0 , α q = 0.
Moreover,
β j ≠ 0 for p < j ≤ q , \beta_j \ne0 \qquad\text{for } p < j \le q, β j = 0 for p < j ≤ q ,
because β j \beta_j β j has only one zero in the range 0 ≤ j ≤ k − 1 0 \le j \le k - 1 0 ≤ j ≤ k − 1 .
Choose c p ≠ 0 c_p \ne0 c p = 0 . The equations
c j − 1 α j − 1 + c j β j = 0 , p < j ≤ q , c_{j-1}\alpha_{j-1} + c_j\beta_j = 0, \qquad p < j \le q, c j − 1 α j − 1 + c j β j = 0 , p < j ≤ q ,
determine successively
c j = − α j − 1 β j c j − 1 . c_j = -\frac{\alpha_{j-1}}{\beta_j}c_{j-1}. c j = − β j α j − 1 c j − 1 .
Thus there exists a nonzero row relation supported on
R p , R p + 1 , … , R q . R_p,R_{p+1},\ldots,R_q. R p , R p + 1 , … , R q .
Since the starting and terminal indices are uniquely determined by a a a and b b b , this row relation is unique up to an overall scalar. Hence, the rank drops by exactly 1:
rank M A ( a , b ) = k − 1. \operatorname{rank}\mathcal{M}_{A}(a,b)=k-1. rank M A ( a , b ) = k − 1.
Therefore
dim ker M A ( a , b ) = 2. \dim\ker\mathcal{M}_{A}(a,b)=2. dim ker M A ( a , b ) = 2.
Returning to the original weights gives the statement. □
Proof of Theorem 8.8. Let us study invariance under M τ ξ M_{\tau\xi} M τ ξ . Using the transformation rules
M τ ξ ⋅ ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q ) = D ξ i − 2 D q ⊠ D ξ k − i − 1 D q + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 . M_{\tau\xi}\cdot\left(\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-1}\mathcal{D}_{q}\right)=\mathcal{D}_{\xi}^{i-2}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-1}\mathcal{D}_{q}+2\lambda_{1}\mathcal{D}_{\xi}^{i-3}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-1}\mathcal{D}_{q}
+2\lambda_{2}\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-1}. M τ ξ ⋅ ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q ) = D ξ i − 2 D q ⊠ D ξ k − i − 1 D q + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 .
M τ ξ ⋅ ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ ) = D ξ i − 2 D q ⊠ D ξ k − i − 2 ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 ∂ τ − D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 . M_{\tau\xi}\cdot\left(\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\partial_{\tau}\right)=\mathcal{D}_{\xi}^{i-2}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\partial_{\tau}+2\lambda_{1}\mathcal{D}_{\xi}^{i-3}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\partial_{\tau}
-\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-1}. M τ ξ ⋅ ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ ) = D ξ i − 2 D q ⊠ D ξ k − i − 2 ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 ∂ τ − D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 .
M τ ξ ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ ) = 2 λ 1 D ξ i − 1 ⊠ D ξ k − i − 3 D q ∂ τ − D ξ i − 1 D q ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i − 1 D q ⊠ D ξ k − i − 3 ∂ τ . M_{\tau\xi}\cdot\left(\mathcal{D}_{\xi}^{i-1}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\mathcal{D}_{q}\partial_{\tau}\right)=2\lambda_{1}\mathcal{D}_{\xi}^{i-1}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\mathcal{D}_{q}\partial_{\tau}-\mathcal{D}_{\xi}^{i-1}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\mathcal{D}_{q}
-2\lambda_{2}\mathcal{D}_{\xi}^{i-1}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\partial_{\tau}. M τ ξ ⋅ ( D ξ i − 1 D q ⊠ D ξ k − i − 3 D q ∂ τ ) = 2 λ 1 D ξ i − 1 ⊠ D ξ k − i − 3 D q ∂ τ − D ξ i − 1 D q ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i − 1 D q ⊠ D ξ k − i − 3 ∂ τ .
M τ ξ ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ ) = − D ξ i − 1 ⊠ D ξ k − i − 3 D q ∂ τ − D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 ∂ τ . M_{\tau\xi}\cdot\left(\mathcal{D}_{\xi}^{i-2}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\mathcal{D}_{q}\partial_{\tau}\right)=-\mathcal{D}_{\xi}^{i-1}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\mathcal{D}_{q}\partial_{\tau}-\mathcal{D}_{\xi}^{i-2}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\mathcal{D}_{q}
-2\lambda_{2}\mathcal{D}_{\xi}^{i-2}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-3}\partial_{\tau}. M τ ξ ⋅ ( D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 D q ∂ τ ) = − D ξ i − 1 ⊠ D ξ k − i − 3 D q ∂ τ − D ξ i − 2 ∂ τ ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i − 2 ∂ τ ⊠ D ξ k − i − 3 ∂ τ .
and comparing the independent tensor monomials, we obtain the endpoint conditions
G 1 = 0 , C k − 1 = 0. G_{1}=0,\qquad C_{k-1}=0. G 1 = 0 , C k − 1 = 0.
together with
(76)
G i = C i + 1 , 2 ≤ i ≤ k − 3. (76) G_{i}=C_{i+1},\qquad2\leq i\leq k-3. \tag*{(76)} G i = C i + 1 , 2 ≤ i ≤ k − 3. (76)
(77)
D i = 2 λ 2 C i , 3 ≤ i ≤ k − 2. (77) D_{i}=2\lambda_{2}C_{i},\qquad3\leq i\leq k-2. \tag*{(77)} D i = 2 λ 2 C i , 3 ≤ i ≤ k − 2. (77)
and
(78)
H i = 2 λ 1 C i + 1 , 2 ≤ i ≤ k − 3. (78) H_{i}=2\lambda_{1}C_{i+1},\qquad2\leq i\leq k-3. \tag*{(78)} H i = 2 λ 1 C i + 1 , 2 ≤ i ≤ k − 3. (78)
Thus, after imposing M τ ξ M_{\tau\xi} M τ ξ -invariance, the whole operator is determined by the coefficients
C 3 , C 4 , … , C k − 2 . C_{3},C_{4},\ldots,C_{k-2}. C 3 , C 4 , … , C k − 2 .
Substituting the relations (76)–(78) into the ansatz (68), we obtain
J 1 , B ( k ) = ∑ 3 ≤ i ≤ k − 2 C i ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + D ξ i − 2 D q ⊠ D ξ k − i − 2 D q ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 D q ∂ τ ) . J_{1,B}^{(k)}=\sum_{3\leq i\leq k-2}C_{i}\left(\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-1}\mathcal{D}_{q}+2\lambda_{2}\mathcal{D}_{\xi}^{i-3}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\partial_{\tau}
+\mathcal{D}_{\xi}^{i-2}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\mathcal{D}_{q}\partial_{\tau}+2\lambda_{1}\mathcal{D}_{\xi}^{i-3}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-2}\mathcal{D}_{q}\partial_{\tau}\right). J 1 , B ( k ) = 3 ≤ i ≤ k − 2 ∑ C i ( D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 1 D q + 2 λ 2 D ξ i − 3 D q ∂ τ ⊠ D ξ k − i − 2 ∂ τ + D ξ i − 2 D q ⊠ D ξ k − i − 2 D q ∂ τ + 2 λ 1 D ξ i − 3 ∂ τ ⊠ D ξ k − i − 2 D q ∂ τ ) .
Now, invariance under M q 2 M_{q}^{2} M q 2 gives
(79)
( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , for 3 ≤ i ≤ k − 3. (79) (i-2)(2\lambda_{1}-i+1)C_{i+1}+(k-i-2)(2\lambda_{2}-k+i+1)C_{i}=0,\qquad\text{for }3\leq i\leq k-3. \tag*{(79)} ( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , for 3 ≤ i ≤ k − 3. (79)
Equation (79) is the fundamental recurrence for odd Type B operators.
We now restrict to generic weights. In particular, assume that
2 λ 1 − i + 1 ≠ 0 , 3 ≤ i ≤ k − 3. 2\lambda_{1}-i+1\ne0,\qquad3\leq i\leq k-3. 2 λ 1 − i + 1 = 0 , 3 ≤ i ≤ k − 3.
Then (79) can be solved successively:
(80)
C i + 1 = − ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) ( i − 2 ) ( 2 λ 1 − i + 1 ) C i , 3 ≤ i ≤ k − 3. (80) C_{i+1}=-\frac{(k-i-2)(2\lambda_{2}-k+i+1)}{(i-2)(2\lambda_{1}-i+1)}C_{i},\qquad3\leq i\leq k-3. \tag*{(80)} C i + 1 = − ( i − 2 ) ( 2 λ 1 − i + 1 ) ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i , 3 ≤ i ≤ k − 3. (80)
Hence, every coefficient
C 4 , … , C k − 2 C_4,\ldots,C_{k-2} C 4 , … , C k − 2
is uniquely determined by C 3 C_3 C 3 . Thus, the generic solution space is one-dimensional. Iterating (80), and choose
C 3 = ∏ 3 ≤ r ≤ k − 3 ( 2 λ 1 − r + 1 ) , C_3=\prod_{3\le r\le k-3}(2\lambda_1-r+1), C 3 = 3 ≤ r ≤ k − 3 ∏ ( 2 λ 1 − r + 1 ) ,
we get
C i = ( − 1 ) i − 3 ( k − 5 i − 3 ) ( ∏ 3 ≤ r ≤ i − 1 ( 2 λ 2 − k + r + 1 ) ) ( ∏ i ≤ r ≤ k − 3 ( 2 λ 1 − r + 1 ) ) , for 3 ≤ i ≤ k − 2. C_i=(-1)^{i-3}\binom{k-5}{i-3}\left(\prod_{3\le r\le i-1}(2\lambda_2-k+r+1)\right)\left(\prod_{i\le r\le k-3}(2\lambda_1-r+1)\right),\quad\text{for }3\le i\le k-2. C i = ( − 1 ) i − 3 ( i − 3 k − 5 ) ( 3 ≤ r ≤ i − 1 ∏ ( 2 λ 2 − k + r + 1 ) ) ( i ≤ r ≤ k − 3 ∏ ( 2 λ 1 − r + 1 ) ) , for 3 ≤ i ≤ k − 2.
Proof of Proposition 8.9. After imposing invariance under M τ ξ M_{\tau\xi} M τ ξ , the operator is determined by C 3 , C 4 , … , C k − 2 C_3,C_4,\ldots,C_{k-2} C 3 , C 4 , … , C k − 2 . Invariance under M q 2 M_{q_2} M q 2 gives the recurrence
( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , for 3 ≤ i ≤ k − 3. (i-2)(2\lambda_1-i+1)C_{i+1}+(k-i-2)(2\lambda_2-k+i+1)C_i=0,\quad\text{for }3\le i\le k-3. ( i − 2 ) ( 2 λ 1 − i + 1 ) C i + 1 + ( k − i − 2 ) ( 2 λ 2 − k + i + 1 ) C i = 0 , for 3 ≤ i ≤ k − 3.
Set a : = 2 λ 1 a:=2\lambda_1 a := 2 λ 1 , b : = 2 λ 2 b:=2\lambda_2 b := 2 λ 2 , and introduce
α i : = ( i − 2 ) ( a − i + 1 ) , β i : = ( k − i − 2 ) ( b − k + i + 1 ) . \alpha_i:=(i-2)(a-i+1),\qquad\beta_i:=(k-i-2)(b-k+i+1). α i := ( i − 2 ) ( a − i + 1 ) , β i := ( k − i − 2 ) ( b − k + i + 1 ) .
the coefficient vector
( C 3 , C 4 , … , C k − 2 ) t (C_3,C_4,\ldots,C_{k-2})^t ( C 3 , C 4 , … , C k − 2 ) t
belongs to the kernel of the bidiagonal matrix
M B ( a , b ) = ( β 3 α 3 0 ⋯ 0 0 β 4 α 4 ⋱ ⋮ ⋮ ⋱ ⋱ ⋱ 0 0 ⋯ 0 β k − 3 α k − 3 ) . \mathcal{M}_B(a,b)=
\begin{pmatrix}
\beta_3 & \alpha_3 & 0 & \cdots& 0\\
0 & \beta_4 & \alpha_4 & \ddots& \vdots\\
\vdots& \ddots& \ddots& \ddots& 0\\
0 & \cdots& 0 & \beta_{k-3} & \alpha_{k-3}
\end{pmatrix}. M B ( a , b ) = β 3 0 ⋮ 0 α 3 β 4 ⋱ ⋯ 0 α 4 ⋱ 0 ⋯ ⋱ ⋱ β k − 3 0 ⋮ 0 α k − 3 .
The matrix has k − 5 k-5 k − 5 rows and k − 4 k-4 k − 4 columns. At generic weights,
rank M B ( a , b ) = k − 5 , \operatorname{rank}\mathcal{M}_B(a,b)=k-5, rank M B ( a , b ) = k − 5 ,
and therefore
dim ker M B ( a , b ) = 1. \dim\ker\mathcal{M}_B(a,b)=1. dim ker M B ( a , b ) = 1.
We now determine precisely when the rank drops. Suppose that there exists a nontrivial linear relation among the rows,
∑ 3 ≤ i ≤ k − 3 c i R i = 0. \sum_{3\le i\le k-3}c_iR_i=0. 3 ≤ i ≤ k − 3 ∑ c i R i = 0.
Let
p : = min { i ∣ c i ≠ 0 } , q : = max { i ∣ c i ≠ 0 } . p:=\min\{i\mid c_i\ne0\},\qquad q:=\max\{i\mid c_i\ne0\}. p := min { i ∣ c i = 0 } , q := max { i ∣ c i = 0 } .
It follows that c p β p = 0 c_p\beta_p=0 c p β p = 0 and since c p ≠ 0 c_p\ne0 c p = 0 , we must have β p = 0 \beta_p=0 β p = 0 . Because k − p − 2 ≠ 0 k-p-2\ne0 k − p − 2 = 0 for 3 ≤ p ≤ k − 3 3\le p\le k-3 3 ≤ p ≤ k − 3 , this is equivalent to
p = k − b − 1. (81) p=k-b-1. \tag*{(81)} p = k − b − 1. (81)
Since 3 ≤ p ≤ k − 3 3\le p\le k-3 3 ≤ p ≤ k − 3 , it follows that
b ∈ { 2 , 3 , … , k − 4 } . b\in\{2,3,\ldots,k-4\}. b ∈ { 2 , 3 , … , k − 4 } .
Similarly, the same reasoning implies that α q = 0 \alpha_q=0 α q = 0 . Since q − 2 ≠ 0 q-2\ne0 q − 2 = 0 , this gives
q = a + 1. (82) q=a+1. \tag*{(82)} q = a + 1. (82)
Because 3 ≤ q ≤ k − 3 3\le q\le k-3 3 ≤ q ≤ k − 3 , we obtain
a ∈ { 2 , 3 , … , k − 4 } . a\in\{2,3,\ldots,k-4\}. a ∈ { 2 , 3 , … , k − 4 } .
A nontrivial row relation can exist only if p ≤ q p\le q p ≤ q . Using formulas (81) and (82) , this condition becomes
k − b − 1 ≤ a + 1 , k-b-1\le a+1, k − b − 1 ≤ a + 1 ,
or equivalently
a + b ≥ k − 2. a+b\ge k-2. a + b ≥ k − 2.
Conversely, suppose that
a , b ∈ { 2 , … , k − 4 } a,b\in\{2,\ldots,k-4\} a , b ∈ { 2 , … , k − 4 }
and
a + b ≥ k − 2. a+b\ge k-2. a + b ≥ k − 2.
Define
p : = k − b − 1 , q : = a + 1. p:=k-b-1,\qquad q:=a+1. p := k − b − 1 , q := a + 1.
Then p ≤ q p\le q p ≤ q , and
β p = 0 , α q = 0. \beta_p=0,\qquad\alpha_q=0. β p = 0 , α q = 0.
Choose c p ≠ 0 c_p\ne0 c p = 0 . For p < j ≤ q p<j\le q p < j ≤ q , the coefficients c j c_j c j are determined successively by
c j − 1 α j − 1 + c j β j = 0. c_{j-1}\alpha_{j-1}+c_j\beta_j=0. c j − 1 α j − 1 + c j β j = 0.
Thus
c j = − α j − 1 β j c j − 1 . c_j=-\frac{\alpha_{j-1}}{\beta_j}c_{j-1}. c j = − β j α j − 1 c j − 1 .
This produces a nonzero linear relation supported on
R p , R p + 1 , … , R q . R_p,R_{p+1},\ldots,R_q. R p , R p + 1 , … , R q .
The indices p p p and q q q are uniquely determined by a a a and b b b , so this row relation is unique up to an overall scalar. Consequently, the rank drops by exactly 1:
rank M B ( a , b ) = k − 6. \operatorname{rank}\mathcal{M}_B(a,b)=k-6. rank M B ( a , b ) = k − 6.
Since M B \mathcal{M}_B M B has k − 4 k-4 k − 4 columns, we obtain
dim ker M B ( a , b ) = 2. \dim\ker\mathcal{M}_B(a,b)=2. dim ker M B ( a , b ) = 2.
Returning to the original density weights gives the statement. □
Proof of Proposition 8.12. The general Type A operator is
J 0 , A ( k ) = ∑ 0 ≤ i ≤ k A i D ξ i ⊠ D ξ k − i . J^{(k)}_{0,A}=\sum_{0\le i\le k}A_iD_\xi^i\boxtimes D_\xi^{k-i}. J 0 , A ( k ) = 0 ≤ i ≤ k ∑ A i D ξ i ⊠ D ξ k − i .
Using
M q 2 ⋅ D ξ m = − 2 m D ξ m − 1 D q − 2 m ( m − 1 ) D ξ m − 2 ∂ τ , M_{q^2}\cdot D_\xi^m=-2mD_\xi^{m-1}D_q-2m(m-1)D_\xi^{m-2}\partial_\tau, M q 2 ⋅ D ξ m = − 2 m D ξ m − 1 D q − 2 m ( m − 1 ) D ξ m − 2 ∂ τ ,
the coefficient of 1 ⊠ D ξ k − 1 D q 1\boxtimes D_\xi^{k-1}D_q 1 ⊠ D ξ k − 1 D q in M q 2 ⋅ J 0 , A ( k ) M_{q^2}\cdot J^{(k)}_{0,A} M q 2 ⋅ J 0 , A ( k ) gives A 0 = 0 A_0=0 A 0 = 0 . More generally, the coefficient of
D ξ i − 1 D q ⊠ D ξ k − i , 1 ≤ i ≤ k , D_\xi^{i-1}D_q\boxtimes D_\xi^{k-i},\qquad1\le i\le k, D ξ i − 1 D q ⊠ D ξ k − i , 1 ≤ i ≤ k ,
is − 2 i A i -2iA_i − 2 i A i . Hence,
A i = 0 , 0 ≤ i ≤ k , A_i=0,\qquad0\le i\le k, A i = 0 , 0 ≤ i ≤ k ,
and therefore J 0 , A ( k ) = 0 J^{(k)}_{0,A}=0 J 0 , A ( k ) = 0 . □
Proof of Proposition 8.14. For k = 2 k=2 k = 2 , the ansatz is A 0 D q ⊠ D q A_0D_q\boxtimes D_q A 0 D q ⊠ D q , and the invariance equations are λ 1 A 0 = λ 2 A 0 = 0 \lambda_1A_0=\lambda_2A_0=0 λ 1 A 0 = λ 2 A 0 = 0 . For k = 3 k=3 k = 3 and k = 4 k=4 k = 4 , the operators displayed are invariant under all eight generating functions of p e ( 2 ) \mathfrak{pe}(2) pe ( 2 ) , and the ranks of the systems are 4 on 6, resp. 9 on 12, coefficients; both are computations.
Let us study the case where k = 5 k=5 k = 5 .
The condition M T ξ ⋅ J 0 , B ( 5 ) = 0 M_{T\xi}\cdot J^{(5)}_{0,B}=0 M T ξ ⋅ J 0 , B ( 5 ) = 0 gives
E 0 = − 2 λ 1 A 0 , E i = C i − 1 − 2 λ 1 A i , i = 1 , 2 , C 2 = 2 λ 1 A 3 , B 0 = 2 λ 2 A 0 , B i = 2 λ 2 A i − D i − 1 , i = 1 , 2 , D 2 = 2 λ 2 A 3 . \begin{aligned}
E_0&=-2\lambda_1A_0,\qquad E_i=C_{i-1}-2\lambda_1A_i,\quad i=1,2,\qquad C_2=2\lambda_1A_3,\\
B_0&=2\lambda_2A_0,\qquad B_i=2\lambda_2A_i-D_{i-1},\quad i=1,2,\qquad D_2=2\lambda_2A_3.
\end{aligned} E 0 B 0 = − 2 λ 1 A 0 , E i = C i − 1 − 2 λ 1 A i , i = 1 , 2 , C 2 = 2 λ 1 A 3 , = 2 λ 2 A 0 , B i = 2 λ 2 A i − D i − 1 , i = 1 , 2 , D 2 = 2 λ 2 A 3 .
together with
2 λ 1 B 0 + 2 λ 2 E 0 = 0 , 2 λ 1 B 1 + 2 λ 2 E 1 − F 0 = 0 , 2 λ 1 B 2 + 2 λ 2 E 2 − F 1 = 0 , − 2 λ 2 C 0 + 2 λ 1 D 0 + F 0 = 0 , − 2 λ 2 C 1 + 2 λ 1 D 1 + F 1 = 0 , − 2 λ 2 C 2 + 2 λ 1 D 2 = 0. \begin{aligned}
2\lambda_1B_0+2\lambda_2E_0&=0,\qquad2\lambda_1B_1+2\lambda_2E_1-F_0=0,\qquad2\lambda_1B_2+2\lambda_2E_2-F_1=0,\\
-2\lambda_2C_0+2\lambda_1D_0+F_0&=0,\qquad-2\lambda_2C_1+2\lambda_1D_1+F_1=0,\qquad-2\lambda_2C_2+2\lambda_1D_2=0.
\end{aligned} 2 λ 1 B 0 + 2 λ 2 E 0 − 2 λ 2 C 0 + 2 λ 1 D 0 + F 0 = 0 , 2 λ 1 B 1 + 2 λ 2 E 1 − F 0 = 0 , 2 λ 1 B 2 + 2 λ 2 E 2 − F 1 = 0 , = 0 , − 2 λ 2 C 1 + 2 λ 1 D 1 + F 1 = 0 , − 2 λ 2 C 2 + 2 λ 1 D 2 = 0.
On the other hand, the condition M q 2 ⋅ J 0 , B ( 5 ) = 0 M_{q_2}\cdot J^{(5)}_{0,B}=0 M q 2 ⋅ J 0 , B ( 5 ) = 0 yields
2 A 2 − C 1 − 2 D 0 = 0 , 6 A 3 − 2 C 2 − D 1 = 0 , − 6 A 0 + 2 B 0 − E 1 = 0 , 2A_2-C_1-2D_0=0,\qquad6A_3-2C_2-D_1=0,\qquad-6A_0+2B_0-E_1=0, 2 A 2 − C 1 − 2 D 0 = 0 , 6 A 3 − 2 C 2 − D 1 = 0 , − 6 A 0 + 2 B 0 − E 1 = 0 ,
− 2 A 1 + B 1 − 2 E 2 = 0 , 2 B 2 − 2 D 0 − F 1 = 0 , − 2 C 0 − 2 E 2 + F 0 = 0. -2A_1+B_1-2E_2=0,\qquad2B_2-2D_0-F_1=0,\qquad-2C_0-2E_2+F_0=0. − 2 A 1 + B 1 − 2 E 2 = 0 , 2 B 2 − 2 D 0 − F 1 = 0 , − 2 C 0 − 2 E 2 + F 0 = 0.
At generic weights this system has rank 16 on the 18 coefficients. Consequently,
dim Inv 0 , B ( 5 ) = 2. \dim\operatorname{Inv}^{(5)}_{0,B}=2. dim Inv 0 , B ( 5 ) = 2.
For instance, E 2 E_2 E 2 and F 1 F_1 F 1 may be chosen as free parameters, and all the remaining coefficients are determined by them. □ \square □
Proof of Theorem 8.15. We impose successively the invariance conditions with respect to M T ξ M_{T\xi} M T ξ and M q 2 M_{q_2} M q 2 .
Set
Y : = M T ξ . Y:=M_{T\xi}. Y := M T ξ .
The action of Y Y Y on the six families of monomials is
Y ⋅ ( D ξ i D q ⊠ D ξ k − i − 2 D q ) = 2 λ 1 D ξ i ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i D q ⊠ D ξ k − i − 2 , Y\cdot\left(D_\xi^iD_q\boxtimes D_\xi^{k-i-2}D_q\right)=2\lambda_1D_\xi^i\boxtimes D_\xi^{k-i-2}D_q-2\lambda_2D_\xi^iD_q\boxtimes D_\xi^{k-i-2}, Y ⋅ ( D ξ i D q ⊠ D ξ k − i − 2 D q ) = 2 λ 1 D ξ i ⊠ D ξ k − i − 2 D q − 2 λ 2 D ξ i D q ⊠ D ξ k − i − 2 ,
Y ⋅ ( D ξ i D q ⊠ D ξ k − i − 3 ∂ τ ) = 2 λ 1 D ξ i ⊠ D ξ k − i − 3 ∂ τ + D ξ i D q ⊠ D ξ k − i − 2 , Y\cdot\left(D_\xi^iD_q\boxtimes D_\xi^{k-i-3}\partial_\tau\right)=2\lambda_1D_\xi^i\boxtimes D_\xi^{k-i-3}\partial_\tau+D_\xi^iD_q\boxtimes D_\xi^{k-i-2}, Y ⋅ ( D ξ i D q ⊠ D ξ k − i − 3 ∂ τ ) = 2 λ 1 D ξ i ⊠ D ξ k − i − 3 ∂ τ + D ξ i D q ⊠ D ξ k − i − 2 ,
Y ⋅ ( D ξ i ∂ τ ⊠ D ξ k − i − 3 D q ) = − D ξ i + 1 ⊠ D ξ k − i − 3 D q − 2 λ 2 D ξ i ∂ τ ⊠ D ξ k − i − 3 , Y\cdot\left(D_\xi^i\partial_\tau\boxtimes D_\xi^{k-i-3}D_q\right)=-D_\xi^{i+1}\boxtimes D_\xi^{k-i-3}D_q-2\lambda_2D_\xi^i\partial_\tau\boxtimes D_\xi^{k-i-3}, Y ⋅ ( D ξ i ∂ τ ⊠ D ξ k − i − 3 D q ) = − D ξ i + 1 ⊠ D ξ k − i − 3 D q − 2 λ 2 D ξ i ∂ τ ⊠ D ξ k − i − 3 ,
Y ⋅ ( D ξ i D q ∂ τ ⊠ D ξ k − i − 3 ) = D ξ i + 1 D q ⊠ D ξ k − i − 3 + 2 λ 1 D ξ i ∂ τ ⊠ D ξ k − i − 3 , Y\cdot\left(D_\xi^iD_q\partial_\tau\boxtimes D_\xi^{k-i-3}\right)=D_\xi^{i+1}D_q\boxtimes D_\xi^{k-i-3}+2\lambda_1D_\xi^i\partial_\tau\boxtimes D_\xi^{k-i-3}, Y ⋅ ( D ξ i D q ∂ τ ⊠ D ξ k − i − 3 ) = D ξ i + 1 D q ⊠ D ξ k − i − 3 + 2 λ 1 D ξ i ∂ τ ⊠ D ξ k − i − 3 ,
Y ⋅ ( D ξ i ⊠ D ξ k − i − 3 D q ∂ τ ) = D ξ i ⊠ D ξ k − i − 2 D q + 2 λ 2 D ξ i ⊠ D ξ k − i − 3 ∂ τ , Y\cdot\left(D_\xi^i\boxtimes D_\xi^{k-i-3}D_q\partial_\tau\right)=D_\xi^i\boxtimes D_\xi^{k-i-2}D_q+2\lambda_2D_\xi^i\boxtimes D_\xi^{k-i-3}\partial_\tau, Y ⋅ ( D ξ i ⊠ D ξ k − i − 3 D q ∂ τ ) = D ξ i ⊠ D ξ k − i − 2 D q + 2 λ 2 D ξ i ⊠ D ξ k − i − 3 ∂ τ ,
and
Y ⋅ ( D ξ i ∂ τ ⊠ D ξ k − i − 4 ∂ τ ) = − D ξ i + 1 ⊠ D ξ k − i − 4 ∂ τ + D ξ i ∂ τ ⊠ D ξ k − i − 3 . Y\cdot\left(D_\xi^i\partial_\tau\boxtimes D_\xi^{k-i-4}\partial_\tau\right)=-D_\xi^{i+1}\boxtimes D_\xi^{k-i-4}\partial_\tau+D_\xi^i\partial_\tau\boxtimes D_\xi^{k-i-3}. Y ⋅ ( D ξ i ∂ τ ⊠ D ξ k − i − 4 ∂ τ ) = − D ξ i + 1 ⊠ D ξ k − i − 4 ∂ τ + D ξ i ∂ τ ⊠ D ξ k − i − 3 .
Comparing equal tensor monomials yields
E 0 = − 2 λ 1 A 0 , B 0 = 2 λ 2 A 0 , E_0=-2\lambda_1A_0,\qquad B_0=2\lambda_2A_0, E 0 = − 2 λ 1 A 0 , B 0 = 2 λ 2 A 0 ,
and, for 1 ≤ i ≤ k − 3 1\le i\le k-3 1 ≤ i ≤ k − 3 ,
E i = C i − 1 − 2 λ 1 A i , B i = 2 λ 2 A i − D i − 1 , E_i=C_{i-1}-2\lambda_1A_i,\qquad B_i=2\lambda_2A_i-D_{i-1}, E i = C i − 1 − 2 λ 1 A i , B i = 2 λ 2 A i − D i − 1 ,
together with the endpoint relations
C k − 3 = 2 λ 1 A k − 2 , D k − 3 = 2 λ 2 A k − 2 . C_{k-3}=2\lambda_1A_{k-2},\qquad D_{k-3}=2\lambda_2A_{k-2}. C k − 3 = 2 λ 1 A k − 2 , D k − 3 = 2 λ 2 A k − 2 .
Moreover,
F i = 2 λ 2 C i − 2 λ 1 D i , 0 ≤ i ≤ k − 4. F_i=2\lambda_2C_i-2\lambda_1D_i,\qquad0\le i\le k-4. F i = 2 λ 2 C i − 2 λ 1 D i , 0 ≤ i ≤ k − 4.
Thus M T ξ M_{T\xi} M T ξ -invariance expresses the coefficients B i B_i B i , E i E_i E i , F i F_i F i in terms of A i A_i A i , C i C_i C i , $D_i. We next impose M q 2 M_{q^2} M q 2 -invariance. Put N : = − 1 2 M q 2 N := -\frac{1}{2}M_{q^2} N := − 2 1 M q 2 . We use
[ N , D ξ m ] = m D ξ m − 1 D q + m ( m − 1 ) D ξ m − 2 ∂ τ , [N,\mathcal{D}_{\xi}^{m}] = m\mathcal{D}_{\xi}^{m-1}\mathcal{D}_{q} + m(m-1)\mathcal{D}_{\xi}^{m-2}\partial_{\tau}, [ N , D ξ m ] = m D ξ m − 1 D q + m ( m − 1 ) D ξ m − 2 ∂ τ ,
[ N , D ξ m D q ] = − m ( m − 1 ) D ξ m − 2 D q ∂ τ , [N,\mathcal{D}_{\xi}^{m}\mathcal{D}_{q}] = -m(m-1)\mathcal{D}_{\xi}^{m-2}\mathcal{D}_{q}\partial_{\tau}, [ N , D ξ m D q ] = − m ( m − 1 ) D ξ m − 2 D q ∂ τ ,
and
[ N , D ξ m ∂ τ ] = m D ξ m − 1 D q ∂ τ , [ N , D ξ m D q ∂ τ ] = 0. [N,\mathcal{D}_{\xi}^{m}\partial_{\tau}] = m\mathcal{D}_{\xi}^{m-1}\mathcal{D}_{q}\partial_{\tau}, \qquad[N,\mathcal{D}_{\xi}^{m}\mathcal{D}_{q}\partial_{\tau}] = 0. [ N , D ξ m ∂ τ ] = m D ξ m − 1 D q ∂ τ , [ N , D ξ m D q ∂ τ ] = 0.
Comparison of the coefficients of
D ξ i D q ∂ τ ⊠ D ξ k − i − 4 D q \mathcal{D}_{\xi}^{i}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-4}\mathcal{D}_{q} D ξ i D q ∂ τ ⊠ D ξ k − i − 4 D q
gives
− ( i + 2 ) ( i + 1 ) A i + 2 + ( i + 1 ) C i + 1 + ( k − i − 3 ) D i = 0 , 0 ≤ i ≤ k − 4. (83) -(i+2)(i+1)A_{i+2}+(i+1)C_{i+1}+(k-i-3)D_i=0,\qquad0\leq i\leq k-4. \tag*{(83)} − ( i + 2 ) ( i + 1 ) A i + 2 + ( i + 1 ) C i + 1 + ( k − i − 3 ) D i = 0 , 0 ≤ i ≤ k − 4. (83)
Similarly, comparison of
D ξ i D q ⊠ D ξ k − i − 4 D q ∂ τ \mathcal{D}_{\xi}^{i}\mathcal{D}_{q}\boxtimes\mathcal{D}_{\xi}^{k-i-4}\mathcal{D}_{q}\partial_{\tau} D ξ i D q ⊠ D ξ k − i − 4 D q ∂ τ
gives
( k − i − 2 ) ( k − i − 3 ) A i − ( k − i − 3 ) B i + ( i + 1 ) E i + 1 = 0 , 0 ≤ i ≤ k − 4. (84) (k-i-2)(k-i-3)A_i-(k-i-3)B_i+(i+1)E_{i+1}=0,\qquad0\leq i\leq k-4. \tag*{(84)} ( k − i − 2 ) ( k − i − 3 ) A i − ( k − i − 3 ) B i + ( i + 1 ) E i + 1 = 0 , 0 ≤ i ≤ k − 4. (84)
The coefficients of
D ξ i D q ∂ τ ⊠ D ξ k − i − 5 ∂ τ \mathcal{D}_{\xi}^{i}\mathcal{D}_{q}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-5}\partial_{\tau} D ξ i D q ∂ τ ⊠ D ξ k − i − 5 ∂ τ
and
D ξ i ∂ τ ⊠ D ξ k − i − 5 D q ∂ τ \mathcal{D}_{\xi}^{i}\partial_{\tau}\boxtimes\mathcal{D}_{\xi}^{k-i-5}\mathcal{D}_{q}\partial_{\tau} D ξ i ∂ τ ⊠ D ξ k − i − 5 D q ∂ τ
give, respectively,
− ( i + 2 ) ( i + 1 ) B i + 2 + ( k − i − 3 ) ( k − i − 4 ) D i + ( i + 1 ) F i + 1 = 0 , (85) -(i+2)(i+1)B_{i+2}+(k-i-3)(k-i-4)D_i+(i+1)F_{i+1}=0, \tag*{(85)} − ( i + 2 ) ( i + 1 ) B i + 2 + ( k − i − 3 ) ( k − i − 4 ) D i + ( i + 1 ) F i + 1 = 0 , (85)
and
( k − i − 3 ) ( k − i − 4 ) C i + ( i + 2 ) ( i + 1 ) E i + 2 − ( k − i − 4 ) F i = 0 , (86) (k-i-3)(k-i-4)C_i+(i+2)(i+1)E_{i+2}-(k-i-4)F_i=0, \tag*{(86)} ( k − i − 3 ) ( k − i − 4 ) C i + ( i + 2 ) ( i + 1 ) E i + 2 − ( k − i − 4 ) F i = 0 , (86)
for 0 ≤ i ≤ k − 5 0\leq i\leq k-5 0 ≤ i ≤ k − 5 .
Substituting the M τ ξ M_{\tau\xi} M τ ξ -relations into the equation (84) and using the equation (83), we obtain the recurrence
( i + 1 ) ( i + 2 − 2 λ 1 ) D i + 1 + ( k − i − 3 ) ( k − i − 4 − 2 λ 2 ) D i = 0 , (i+1)(i+2-2\lambda_1)D_{i+1}+(k-i-3)(k-i-4-2\lambda_2)D_i=0, ( i + 1 ) ( i + 2 − 2 λ 1 ) D i + 1 + ( k − i − 3 ) ( k − i − 4 − 2 λ 2 ) D i = 0 ,
for 0 ≤ i ≤ k − 5 0\leq i\leq k-5 0 ≤ i ≤ k − 5 .
Likewise, substituting into the equation (86) and using the equation (84) gives
( i + 2 ) ( i + 1 − 2 λ 1 ) E i + 2 + ( k − i − 4 ) ( k − i − 3 − 2 λ 2 ) E i + 1 = 0. (i+2)(i+1-2\lambda_1)E_{i+2}+(k-i-4)(k-i-3-2\lambda_2)E_{i+1}=0. ( i + 2 ) ( i + 1 − 2 λ 1 ) E i + 2 + ( k − i − 4 ) ( k − i − 3 − 2 λ 2 ) E i + 1 = 0.
Equivalently, setting j = i + 1 j=i+1 j = i + 1 ,
( j + 1 ) ( j − 2 λ 1 ) E j + 1 + ( k − j − 3 ) ( k − j − 2 − 2 λ 2 ) E j = 0 , (j+1)(j-2\lambda_1)E_{j+1}+(k-j-3)(k-j-2-2\lambda_2)E_j=0, ( j + 1 ) ( j − 2 λ 1 ) E j + 1 + ( k − j − 3 ) ( k − j − 2 − 2 λ 2 ) E j = 0 ,
for 1 ≤ j ≤ k − 4 1\leq j\leq k-4 1 ≤ j ≤ k − 4 .
At generic weights, the coefficients occurring in these recurrences do not vanish. Hence,
D i = ( − 1 ) i ∏ 0 ≤ s ≤ i − 1 ( k − s − 3 ) ( k − s − 4 − 2 λ 2 ) ( s + 1 ) ( s + 2 − 2 λ 1 ) D 0 , 0 ≤ i ≤ k − 4 , D_i=(-1)^i\prod_{0\leq s\leq i-1}\frac{(k-s-3)(k-s-4-2\lambda_2)}{(s+1)(s+2-2\lambda_1)}D_0,\qquad0\leq i\leq k-4, D i = ( − 1 ) i 0 ≤ s ≤ i − 1 ∏ ( s + 1 ) ( s + 2 − 2 λ 1 ) ( k − s − 3 ) ( k − s − 4 − 2 λ 2 ) D 0 , 0 ≤ i ≤ k − 4 ,
and
E i = ( − 1 ) i − 1 ∏ 1 ≤ s ≤ i − 1 ( k − s − 3 ) ( k − s − 2 − 2 λ 2 ) ( s + 1 ) ( s − 2 λ 1 ) E 1 , 1 ≤ i ≤ k − 3. E_i=(-1)^{i-1}\prod_{1\leq s\leq i-1}\frac{(k-s-3)(k-s-2-2\lambda_2)}{(s+1)(s-2\lambda_1)}E_1,\qquad1\leq i\leq k-3. E i = ( − 1 ) i − 1 1 ≤ s ≤ i − 1 ∏ ( s + 1 ) ( s − 2 λ 1 ) ( k − s − 3 ) ( k − s − 2 − 2 λ 2 ) E 1 , 1 ≤ i ≤ k − 3.
The remaining coefficients are recovered from the equations (83)–(84). In particular,
A 0 = − E 1 ( k − 3 ) ( k − 2 − 2 λ 2 ) , A_0=-\frac{E_1}{(k-3)(k-2-2\lambda_2)}, A 0 = − ( k − 3 ) ( k − 2 − 2 λ 2 ) E 1 ,
and, for 1 ≤ i ≤ k − 4 1\leq i\leq k-4 1 ≤ i ≤ k − 4 ,
A i = − ( k − i − 3 ) D i − 1 + ( i + 1 ) E i + 1 ( k − i − 3 ) ( k − i − 2 − 2 λ 2 ) . A_i=-\frac{(k-i-3)D_{i-1}+(i+1)E_{i+1}}{(k-i-3)(k-i-2-2\lambda_2)}. A i = − ( k − i − 3 ) ( k − i − 2 − 2 λ 2 ) ( k − i − 3 ) D i − 1 + ( i + 1 ) E i + 1 .
The terminal coefficients are given by
A k − 3 = ( k − 4 ) E k − 3 + 2 D k − 5 ( k − 4 ) ( k − 3 − 2 λ 1 ) , and A k − 2 = D k − 4 ( k − 3 ) ( k − 2 − 2 λ 1 ) . A_{k-3}=\frac{(k-4)E_{k-3}+2D_{k-5}}{(k-4)(k-3-2\lambda_1)},
\qquad\text{and}\qquad
A_{k-2}=\frac{D_{k-4}}{(k-3)(k-2-2\lambda_1)}. A k − 3 = ( k − 4 ) ( k − 3 − 2 λ 1 ) ( k − 4 ) E k − 3 + 2 D k − 5 , and A k − 2 = ( k − 3 ) ( k − 2 − 2 λ 1 ) D k − 4 .
Finally,
C i = E i + 1 + 2 λ 1 A i + 1 , 0 ≤ i ≤ k − 4 , C_i=E_{i+1}+2\lambda_1A_{i+1},\qquad0\leq i\leq k-4, C i = E i + 1 + 2 λ 1 A i + 1 , 0 ≤ i ≤ k − 4 ,
C k − 3 = 2 λ 1 A k − 2 , B 0 = 2 λ 2 A 0 , E 0 = − 2 λ 1 A 0 , D k − 3 = 2 λ 2 A k − 2 , C_{k-3}=2\lambda_1A_{k-2},\qquad
B_0=2\lambda_2A_0,\qquad
E_0=-2\lambda_1A_0,\qquad
D_{k-3}=2\lambda_2A_{k-2}, C k − 3 = 2 λ 1 A k − 2 , B 0 = 2 λ 2 A 0 , E 0 = − 2 λ 1 A 0 , D k − 3 = 2 λ 2 A k − 2 ,
and
B i = 2 λ 2 A i − D i − 1 ( 1 ≤ i ≤ k − 3 ) , F i = 2 λ 2 C i − 2 λ 1 D i ( 0 ≤ i ≤ k − 4 ) . B_i=2\lambda_2A_i-D_{i-1}\quad(1\leq i\leq k-3),\qquad
F_i=2\lambda_2C_i-2\lambda_1D_i\quad(0\leq i\leq k-4). B i = 2 λ 2 A i − D i − 1 ( 1 ≤ i ≤ k − 3 ) , F i = 2 λ 2 C i − 2 λ 1 D i ( 0 ≤ i ≤ k − 4 ) .
Thus every coefficient of J 0 , B ( k ) J^{(k)}_{0,B} J 0 , B ( k ) is uniquely determined by the two independent parameters
D 0 , E 1 . D_0,\qquad E_1. D 0 , E 1 .
Therefore, at generic weights,
dim Inv 0 , B ( k ) = 2 , k ≥ 5. \dim\operatorname{Inv}^{(k)}_{0,B}=2,\qquad k\geq5. dim Inv 0 , B ( k ) = 2 , k ≥ 5.
Proof of Statement 8.17. For k = 2 k=2 k = 2 , the operator is
J 0 , B ( 2 ) = A 0 D q ⊠ D q , J^{(2)}_{0,B}=A_0D_q\boxtimes D_q, J 0 , B ( 2 ) = A 0 D q ⊠ D q ,
and the invariance equations are
λ 1 A 0 = 0 , λ 2 A 0 = 0. \lambda_1A_0=0,\qquad\lambda_2A_0=0. λ 1 A 0 = 0 , λ 2 A 0 = 0.
Hence a nonzero solution exists only at
( λ 1 , λ 2 ) = ( 0 , 0 ) . (\lambda_1,\lambda_2)=(0,0). ( λ 1 , λ 2 ) = ( 0 , 0 ) .
For k = 3 k=3 k = 3 and k = 4 k=4 k = 4 , the operators of Proposition 8.14 have polynomial coefficients and pairwise distinct leading monomials D q ⊠ D ξ D q D_q\boxtimes D_\xi D_q D q ⊠ D ξ D q , D ξ D q ⊠ D q D_\xi D_q\boxtimes D_q D ξ D q ⊠ D q , resp. D q ⊠ D ξ 2 D q D_q\boxtimes D_\xi^2D_q D q ⊠ D ξ 2 D q , D ξ D q ⊠ D ξ D q D_\xi D_q\boxtimes D_\xi D_q D ξ D q ⊠ D ξ D q , D ξ 2 D q ⊠ D q D_\xi^2D_q\boxtimes D_q D ξ 2 D q ⊠ D q after the denominators are cleared, so the dimension is at least the generic one at every weight; that it is not larger on the grid is a computation.
Proof of Proposition 8.18. The complete k = 5 k=5 k = 5 invariance equations obtained from
M τ ξ ⋅ J 0 , B ( 5 ) = 0 , M q 2 ⋅ J 0 , B ( 5 ) = 0 M_{\tau\xi}\cdot J^{(5)}_{0,B}=0,\qquad M_{q^2}\cdot J^{(5)}_{0,B}=0 M τ ξ ⋅ J 0 , B ( 5 ) = 0 , M q 2 ⋅ J 0 , B ( 5 ) = 0
form a homogeneous linear system in the 18 coefficients of J 0 , B ( 5 ) J^{(5)}_{0,B} J 0 , B ( 5 ) . At generic weights its rank is 16, and therefore
dim Inv 0 , B ( 5 ) = 18 − 16 = 2. \dim\operatorname{Inv}^{(5)}_{0,B}=18-16=2. dim Inv 0 , B ( 5 ) = 18 − 16 = 2.
The common zero locus of the maximal minors is contained in
2 λ 1 , 2 λ 2 ∈ { 0 , 1 , 2 , 3 } . 2\lambda_1,2\lambda_2\in\{0,1,2,3\}. 2 λ 1 , 2 λ 2 ∈ { 0 , 1 , 2 , 3 } .
Computing the rank of the full system at these points gives
λ 2 = 0 λ 2 = 1 2 λ 2 = 1 λ 2 = 3 2 λ 1 = 0 2 2 2 2 λ 1 = 1 2 2 2 3 3 λ 1 = 1 2 3 4 3 λ 1 = 3 2 2 3 3 2 \begin{array}{c|cccc}
& \lambda_{2}=0 & \lambda_{2}=\frac{1}{2} & \lambda_{2}=1 & \lambda_{2}=\frac{3}{2} \\
\hline
\lambda_{1}=0 & 2 & 2 & 2 & 2 \\
\lambda_{1}=\frac{1}{2} & 2 & 2 & 3 & 3 \\
\lambda_{1}=1 & 2 & 3 & 4 & 3 \\
\lambda_{1}=\frac{3}{2} & 2 & 3 & 3 & 2
\end{array} λ 1 = 0 λ 1 = 2 1 λ 1 = 1 λ 1 = 2 3 λ 2 = 0 2 2 2 2 λ 2 = 2 1 2 2 3 3 λ 2 = 1 2 3 4 3 λ 2 = 2 3 2 3 3 2
where each entry denotes dim Inv 0 ˉ , B ( 5 ) \dim\operatorname{Inv}_{\bar{0},B}^{(5)} dim Inv 0 ˉ , B ( 5 ) . This yields precisely the stated resonant set and dimensions. □ \square □
Proof of Theorem 8.19. The general even Type B operator contains 6 k − 12 6k-12 6 k − 12 coefficients. After imposing M τ ξ M_{\tau\xi} M τ ξ -invariance, the coefficients B i B_i B i , C i C_i C i , and F i F_i F i are expressed in terms of A i A_i A i , D i D_i D i , and E i E_i E i as follows:
B 0 = 2 λ 2 A 0 , B i = 2 λ 2 A i − D i − 1 , 1 ≤ i ≤ k − 3. B_0=2\lambda_2 A_0,\qquad B_i=2\lambda_2 A_i-D_{i-1},\qquad1\leq i\leq k-3. B 0 = 2 λ 2 A 0 , B i = 2 λ 2 A i − D i − 1 , 1 ≤ i ≤ k − 3.
C i = E i + 1 + 2 λ 1 A i + 1 , 0 ≤ i ≤ k − 4 , C k − 3 = 2 λ 1 A k − 2 , C_i=E_{i+1}+2\lambda_1 A_{i+1},\qquad0\leq i\leq k-4,\qquad C_{k-3}=2\lambda_1 A_{k-2}, C i = E i + 1 + 2 λ 1 A i + 1 , 0 ≤ i ≤ k − 4 , C k − 3 = 2 λ 1 A k − 2 ,
and
F i = 2 λ 2 C i − 2 λ 1 D i = 2 λ 2 E i + 1 + 4 λ 1 λ 2 A i + 1 − 2 λ 1 D i , 0 ≤ i ≤ k − 4. F_i=2\lambda_2 C_i-2\lambda_1 D_i=2\lambda_2 E_{i+1}+4\lambda_1\lambda_2 A_{i+1}-2\lambda_1 D_i,\qquad0\leq i\leq k-4. F i = 2 λ 2 C i − 2 λ 1 D i = 2 λ 2 E i + 1 + 4 λ 1 λ 2 A i + 1 − 2 λ 1 D i , 0 ≤ i ≤ k − 4.
The remaining endpoint relations are
E 0 = − 2 λ 1 A 0 , D k − 3 = 2 λ 2 A k − 2 . E_0=-2\lambda_1 A_0,\qquad D_{k-3}=2\lambda_2 A_{k-2}. E 0 = − 2 λ 1 A 0 , D k − 3 = 2 λ 2 A k − 2 .
Set
a : = 2 λ 1 , b : = 2 λ 2 . a:=2\lambda_1,\qquad b:=2\lambda_2. a := 2 λ 1 , b := 2 λ 2 .
After eliminating the coefficients A i A_i A i , the remaining M q 2 M_{q^2} M q 2 -invariance conditions reduce to two bidiagonal chains, together with the corresponding endpoint equations. The first chain is
( i + 1 ) ( i + 2 − a ) D i + 1 + ( k − i − 3 ) ( k − i − 4 − b ) D i = 0 , 0 ≤ i ≤ k − 5. (87) (i+1)(i+2-a)D_{i+1}+(k-i-3)(k-i-4-b)D_i=0,\qquad0\leq i\leq k-5. \tag*{(87)} ( i + 1 ) ( i + 2 − a ) D i + 1 + ( k − i − 3 ) ( k − i − 4 − b ) D i = 0 , 0 ≤ i ≤ k − 5. (87)
Set
α i : = ( i + 1 ) ( i + 2 − a ) , β i : = ( k − i − 3 ) ( k − i − 4 − b ) . \alpha_i:=(i+1)(i+2-a),\qquad\beta_i:=(k-i-3)(k-i-4-b). α i := ( i + 1 ) ( i + 2 − a ) , β i := ( k − i − 3 ) ( k − i − 4 − b ) .
The possible interior break points are therefore
s D = a − 2 , r D = k − b − 4. s_D=a-2,\qquad r_D=k-b-4. s D = a − 2 , r D = k − b − 4.
In addition, the right endpoint relation is
D k − 4 − ( k − 3 ) ( k − 2 − a ) A k − 2 = 0. (88) D_{k-4}-(k-3)(k-2-a)A_{k-2}=0. \tag*{(88)} D k − 4 − ( k − 3 ) ( k − 2 − a ) A k − 2 = 0. (88)
Away from r D r_D r D and s D s_D s D , both α i \alpha_i α i and β i \beta_i β i are nonzero, and (87) determines D i + 1 D_{i+1} D i + 1 uniquely from D i D_i D i . We distinguish three cases.
If r D < s D r_D<s_D r D < s D , then β r D = 0 \beta_{r_D}=0 β r D = 0 occurs first and α r D ≠ 0 \alpha_{r_D}\ne0 α r D = 0 . Hence, D r D + 1 = 0 D_{r_D+1}=0 D r D + 1 = 0 . The recurrence then forces D r D + 2 = ⋯ = D s D = 0 D_{r_D+2}=\cdots=D_{s_D}=0 D r D + 2 = ⋯ = D s D = 0 . At i = s D i=s_D i = s D , one has α s D = 0 \alpha_{s_D}=0 α s D = 0 ; since D s D = 0 D_{s_D}=0 D s D = 0 , the corresponding equation is identically satisfied. Therefore, D s D + 1 D_{s_D+1} D s D + 1 becomes free, and the chain splits into two independent branches.
If r D = s D r_D=s_D r D = s D , then α r D = β r D = 0 \alpha_{r_D}=\beta_{r_D}=0 α r D = β r D = 0 , so the corresponding equation is identically satisfied. Thus, D r D + 1 D_{r_D+1} D r D + 1 is independent of the coefficients to the left, and the chain again splits into two branches.
If r D > s D r_D>s_D r D > s D , then the zero of α i \alpha_i α i occurs first. At i = s D i=s_D i = s D , since α s D = 0 \alpha_{s_D}=0 α s D = 0 and β s D ≠ 0 \beta_{s_D}\ne0 β s D = 0 , one obtains D s D = 0 D_{s_D}=0 D s D = 0 . Since both coefficients are nonzero for i < s D i<s_D i < s D , this vanishing propagates backwards: D s D − 1 = ⋯ = D 0 = 0 D_{s_D-1}=\cdots=D_0=0 D s D − 1 = ⋯ = D 0 = 0 . The coefficient D s D + 1 D_{s_D+1} D s D + 1 is free and determines the part of the chain to its right. Hence, the generic free parameter D 0 D_0 D 0 is merely replaced with D s D + 1 D_{s_D+1} D s D + 1 , and no additional independent parameter is created.
Consequently, the interior D D D -chain acquires an additional independent branch if and only if r D ≤ s D r_D \le s_D r D ≤ s D , that is, if and only if a + b ≥ k − 2 a+b \ge k-2 a + b ≥ k − 2 . The second chain is
( j + 1 ) ( j − a ) E j + 1 + ( k − j − 3 ) ( k − j − 2 − b ) E j = 0 , 1 ≤ j ≤ k − 4. (89) (j+1)(j-a)E_{j+1}+(k-j-3)(k-j-2-b)E_j=0,\qquad1\le j\le k-4. \tag*{(89)} ( j + 1 ) ( j − a ) E j + 1 + ( k − j − 3 ) ( k − j − 2 − b ) E j = 0 , 1 ≤ j ≤ k − 4. (89)
Its possible interior break points are
s E = a , r E = k − b − 2. s_E=a,\qquad r_E=k-b-2. s E = a , r E = k − b − 2.
The left endpoint equation is
E 1 + ( k − 3 ) ( k − 2 − b ) A 0 = 0. (90) E_1+(k-3)(k-2-b)A_0=0. \tag*{(90)} E 1 + ( k − 3 ) ( k − 2 − b ) A 0 = 0. (90)
Again, away from the break points, the recurrence determines the successive coefficients uniquely.
If r E < s E r_E<s_E r E < s E , then the zero of the coefficient of E j E_j E j occurs first and forces E r E + 1 = 0 E_{r_E+1}=0 E r E + 1 = 0 . The recurrence propagates this vanishing up to E s E E_{s_E} E s E , while at j = s E j=s_E j = s E the coefficient of E s E + 1 E_{s_E+1} E s E + 1 vanishes. Hence, E s E + 1 E_{s_E+1} E s E + 1 becomes free and the chain splits into two independent branches.
If r E = s E r_E=s_E r E = s E , both coefficients vanish at the same index, so the corresponding equation is identically satisfied. Again, the right-hand part of the chain becomes independent of the left-hand part, producing one additional free parameter.
If r E > s E r_E>s_E r E > s E , then the zero of the coefficient of E j + 1 E_{j+1} E j + 1 occurs first. At j = s E j=s_E j = s E , the recurrence reduces to ( k − s E − 3 ) ( k − s E − 2 − b ) E s E = 0 (k-s_E-3)(k-s_E-2-b)E_{s_E}=0 ( k − s E − 3 ) ( k − s E − 2 − b ) E s E = 0 , and therefore E s E = 0 E_{s_E}=0 E s E = 0 . Since both coefficients are nonzero for j < s E j<s_E j < s E , this vanishing propagates backwards: E s E − 1 = ⋯ = E 1 = 0 E_{s_E-1}=\cdots=E_1=0 E s E − 1 = ⋯ = E 1 = 0 . The coefficient E s E + 1 E_{s_E+1} E s E + 1 is free and generates the remaining part of the chain. Thus, the generic parameter E 1 E_1 E 1 is replaced with E s E + 1 E_{s_E+1} E s E + 1 , and no additional independent parameter is created.
Hence, the interior E E E -chain acquires an additional branch precisely when r E ≤ s E r_E\le s_E r E ≤ s E , namely k − b − 2 ≤ a k-b-2\le a k − b − 2 ≤ a , which is again equivalent to a + b ≥ k − 2 a+b\ge k-2 a + b ≥ k − 2 .
We next include the endpoint resonances.
If a = k − 2 a=k-2 a = k − 2 , then (88) reduces to D k − 4 = 0 D_{k-4}=0 D k − 4 = 0 , and the coefficient A k − 2 A_{k-2} A k − 2 is no longer determined by this equation. For 1 ≤ b ≤ k − 3 1\le b\le k-3 1 ≤ b ≤ k − 3 , the remaining recurrence relations are compatible with D k − 4 = 0 D_{k-4}=0 D k − 4 = 0 , so A k − 2 A_{k-2} A k − 2 becomes an additional free parameter.
Similarly, if b = k − 2 b=k-2 b = k − 2 , then (90) reduces to E 1 = 0 E_1=0 E 1 = 0 , while A 0 A_0 A 0 is no longer determined. For 1 ≤ a ≤ k − 3 1\le a\le k-3 1 ≤ a ≤ k − 3 , the remaining equations are compatible with this condition, and A 0 A_0 A 0 provides one additional free parameter.
At the corner ( a , b ) = ( k − 2 , k − 2 ) (a,b)=(k-2,k-2) ( a , b ) = ( k − 2 , k − 2 ) , the endpoint equations give
D k − 4 = 0 , E 1 = 0. D_{k-4}=0,\qquad E_1=0. D k − 4 = 0 , E 1 = 0.
The two generic parameters are then replaced with the free endpoint coefficients A k − 2 A_{k-2} A k − 2 and A 0 A_0 A 0 . Thus, the number of independent parameters remains equal to two, and no rank drop occurs at this corner.
Combining the interior and endpoint cases gives the full resonance region
R k = { ( a , b ) ∈ Z 2 ∣ 1 ≤ a , b ≤ k − 2 , k − 2 ≤ a + b ≤ 2 k − 5 } . \mathcal{R}_k=\{(a,b)\in\mathbb{Z}^2\mid1\le a,b\le k-2,\qquad k-2\le a+b\le2k-5\}. R k = {( a , b ) ∈ Z 2 ∣ 1 ≤ a , b ≤ k − 2 , k − 2 ≤ a + b ≤ 2 k − 5 } .
Outside R k \mathcal{R}_k R k , neither chain contributes an additional independent parameter, and therefore dim M k = 2 \dim\mathcal{M}_k=2 dim M k = 2 . It remains to determine when the two chains contribute two additional independent parameters simultaneously. The boundary a + b = k − 2 a+b=k-2 a + b = k − 2 requires special attention. In this case,
r D = s D = a − 2 , r E = s E = a , r_D=s_D=a-2,\qquad r_E=s_E=a, r D = s D = a − 2 , r E = s E = a ,
so that, if the two chains are considered separately, they appear to produce the free parameters D a − 1 D_{a-1} D a − 1 and E a + 1 E_{a+1} E a + 1 .
These two parameters are not independent. Indeed, one of the original M q 2 M_{q^2} M q 2 -invariance equations is
( k − i − 2 ) ( k − i − 3 ) A i − ( k − i − 3 ) B i + ( i + 1 ) E i + 1 = 0. (k-i-2)(k-i-3)A_i-(k-i-3)B_i+(i+1)E_{i+1}=0. ( k − i − 2 ) ( k − i − 3 ) A i − ( k − i − 3 ) B i + ( i + 1 ) E i + 1 = 0.
Using B i = b A i − D i − 1 B_i=bA_i-D_{i-1} B i = b A i − D i − 1 , this becomes
( k − i − 3 ) ( k − i − 2 − b ) A i + ( k − i − 3 ) D i − 1 + ( i + 1 ) E i + 1 = 0. (k-i-3)(k-i-2-b)A_i+(k-i-3)D_{i-1}+(i+1)E_{i+1}=0. ( k − i − 3 ) ( k − i − 2 − b ) A i + ( k − i − 3 ) D i − 1 + ( i + 1 ) E i + 1 = 0.
Setting i = a i=a i = a and using a + b = k − 2 a+b=k-2 a + b = k − 2 , the coefficient of A a A_a A a vanishes and one obtains
( k − a − 3 ) D a − 1 + ( a + 1 ) E a + 1 = 0. (k-a-3)D_{a-1}+(a+1)E_{a+1}=0. ( k − a − 3 ) D a − 1 + ( a + 1 ) E a + 1 = 0.
Thus the two apparent new parameters satisfy one nontrivial linear relation. Hence, on the boundary a + b = k − 2 a+b=k-2 a + b = k − 2 , the kernel gains only one additional dimension.
For two additional independent branches one must therefore have a + b > k − 2 a+b>k-2 a + b > k − 2 , or, since a , b ∈ Z a,b\in\mathbb{Z} a , b ∈ Z , a + b ≥ k − 1 a+b\ge k-1 a + b ≥ k − 1 . For strictly interior simultaneous resonance, the break points of both chains must belong to their admissible ranges. For the D D D -chain, 0 ≤ s D , r D ≤ k − 5 0\le s_D,r_D\le k-5 0 ≤ s D , r D ≤ k − 5 gives
2 ≤ a ≤ k − 3 , 1 ≤ b ≤ k − 4. 2\le a\le k-3,\qquad1\le b\le k-4. 2 ≤ a ≤ k − 3 , 1 ≤ b ≤ k − 4.
whereas for the E E E -chain, 1 ≤ s E , r E ≤ k − 4 1\le s_E,r_E\le k-4 1 ≤ s E , r E ≤ k − 4 gives
1 ≤ a ≤ k − 4 , 2 ≤ b ≤ k − 3. 1\le a\le k-4,\qquad2\le b\le k-3. 1 ≤ a ≤ k − 4 , 2 ≤ b ≤ k − 3.
Their intersection yields 2 ≤ a , b ≤ k − 4 2\le a,b\le k-4 2 ≤ a , b ≤ k − 4 . The boundary values a = k − 3 a=k-3 a = k − 3 and b = k − 3 b=k-3 b = k − 3 must therefore be checked directly in the unreduced system.
Suppose first that
a = k − 3 , 2 ≤ b ≤ k − 3. a=k-3,\qquad2\le b\le k-3. a = k − 3 , 2 ≤ b ≤ k − 3.
Then
s D = k − 5 , r D = k − b − 4 < s D , s_D=k-5,\qquad r_D=k-b-4<s_D, s D = k − 5 , r D = k − b − 4 < s D ,
so the D D D -chain forces D r D + 1 = ⋯ = D k − 5 = 0 D_{r_D+1}=\cdots=D_{k-5}=0 D r D + 1 = ⋯ = D k − 5 = 0 . Likewise, the E E E -chain implies E r E + 1 = ⋯ = E k − 3 = 0 E_{r_E+1}=\cdots=E_{k-3}=0 E r E + 1 = ⋯ = E k − 3 = 0 . In particular,
D k − 5 = 0 , E k − 3 = 0. D_{k-5}=0,\qquad E_{k-3}=0. D k − 5 = 0 , E k − 3 = 0.
The unreduced equation involving A k − 3 A_{k-3} A k − 3 is
− ( k − 3 ) ( k − 4 ) A k − 3 + ( k − 4 ) C k − 4 + 2 D k − 5 = 0. -(k-3)(k-4)A_{k-3}+(k-4)C_{k-4}+2D_{k-5}=0. − ( k − 3 ) ( k − 4 ) A k − 3 + ( k − 4 ) C k − 4 + 2 D k − 5 = 0.
Since C k − 4 = E k − 3 + a A k − 3 C_{k-4}=E_{k-3}+aA_{k-3} C k − 4 = E k − 3 + a A k − 3 , this becomes
( k − 4 ) ( k − 3 − a ) A k − 3 = ( k − 4 ) E k − 3 + 2 D k − 5 . (k-4)(k-3-a)A_{k-3}=(k-4)E_{k-3}+2D_{k-5}. ( k − 4 ) ( k − 3 − a ) A k − 3 = ( k − 4 ) E k − 3 + 2 D k − 5 .
For a = k − 3 a=k-3 a = k − 3 , both sides vanish, so A k − 3 A_{k-3} A k − 3 is free. Hence, the boundary value a = k − 3 a=k-3 a = k − 3 contributes the second additional independent parameter.
The case
b = k − 3 , 2 ≤ a ≤ k − 3 , b=k-3,\qquad2\le a\le k-3, b = k − 3 , 2 ≤ a ≤ k − 3 ,
is analogous. The corresponding recurrences imply
D 0 = 0 , E 2 = 0. D_0=0,\qquad E_2=0. D 0 = 0 , E 2 = 0.
The unreduced equation involving A 1 A_1 A 1 is
( k − 4 ) ( k − 3 − b ) A 1 + ( k − 4 ) D 0 + 2 E 2 = 0. (k-4)(k-3-b)A_1+(k-4)D_0+2E_2=0. ( k − 4 ) ( k − 3 − b ) A 1 + ( k − 4 ) D 0 + 2 E 2 = 0.
When b = k − 3 b = k - 3 b = k − 3 , this equation becomes identically satisfied, so A 1 A_1 A 1 is free and provides the second additional independent parameter.
Thus, simultaneous resonance extends to 2 ≤ a , b ≤ k − 3 2 \le a,b \le k - 3 2 ≤ a , b ≤ k − 3 . Together with a + b ≥ k − 1 a+b \ge k - 1 a + b ≥ k − 1 , this gives
R k ( 2 ) = { ( a , b ) ∈ Z 2 ∣ 2 ≤ a , b ≤ k − 3 , k − 1 ≤ a + b ≤ 2 k − 6 } . \mathcal{R}^{(2)}_k = \{(a,b) \in\mathbb{Z}^2 \mid2 \le a,b \le k - 3,\quad k - 1 \le a+b \le2k - 6\}. R k ( 2 ) = {( a , b ) ∈ Z 2 ∣ 2 ≤ a , b ≤ k − 3 , k − 1 ≤ a + b ≤ 2 k − 6 } .
Consequently,
dim ker M k = { 4 , ( a , b ) ∈ R k ( 2 ) , 3 , ( a , b ) ∈ R k ∖ R k ( 2 ) , 2 , ( a , b ) ∉ R k . \dim\ker\mathcal{M}_k =
\begin{cases}
4, & (a,b) \in\mathcal{R}^{(2)}_k,\\
3, & (a,b) \in\mathcal{R}_k \setminus\mathcal{R}^{(2)}_k,\\
2, & (a,b) \notin\mathcal{R}_k.
\end{cases} dim ker M k = ⎩ ⎨ ⎧ 4 , 3 , 2 , ( a , b ) ∈ R k ( 2 ) , ( a , b ) ∈ R k ∖ R k ( 2 ) , ( a , b ) ∈ / R k .
Since M k \mathcal{M}_k M k has 6 k − 12 6k - 12 6 k − 12 columns,
rank M k = ( 6 k − 12 ) − dim ker M k . \operatorname{rank} \mathcal{M}_k = (6k - 12) - \dim\ker\mathcal{M}_k. rank M k = ( 6 k − 12 ) − dim ker M k .
Therefore,
rank M k ( a , b ) = { 6 k − 16 , ( a , b ) ∈ R k ( 2 ) , 6 k − 15 , ( a , b ) ∈ R k ∖ R k ( 2 ) , 6 k − 14 , ( a , b ) ∉ R k . \operatorname{rank} \mathcal{M}_k(a,b) =
\begin{cases}
6k - 16, & (a,b) \in\mathcal{R}^{(2)}_k,\\
6k - 15, & (a,b) \in\mathcal{R}_k \setminus\mathcal{R}^{(2)}_k,\\
6k - 14, & (a,b) \notin\mathcal{R}_k.
\end{cases} rank M k ( a , b ) = ⎩ ⎨ ⎧ 6 k − 16 , 6 k − 15 , 6 k − 14 , ( a , b ) ∈ R k ( 2 ) , ( a , b ) ∈ R k ∖ R k ( 2 ) , ( a , b ) ∈ / R k .
This proves the result. □
Proof of Theorem 8.21. For Type C one has
( u , v ) = ( k − 4 , 4 ) . (u,v) = (k - 4,4). ( u , v ) = ( k − 4 , 4 ) .
Since
v = ε + ε ′ + δ + δ ′ = 4 , v = \varepsilon+ \varepsilon' + \delta+ \delta' = 4, v = ε + ε ′ + δ + δ ′ = 4 ,
we must have
ε = ε ′ = δ = δ ′ = 1. \varepsilon= \varepsilon' = \delta= \delta' = 1. ε = ε ′ = δ = δ ′ = 1.
Hence the complete Type C ansatz is
J 0 , C ( k ) = ∑ 0 ≤ i ≤ k − 4 A i D ξ i D q ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ . J^{(k)}_{0,C} = \sum_{0 \le i \le k-4} A_i D_{\xi}^{i}D_q\partial_\tau\boxtimes D_{\xi}^{k-4-i}D_q\partial_\tau. J 0 , C ( k ) = 0 ≤ i ≤ k − 4 ∑ A i D ξ i D q ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ .
The Cartan equations determine the target weights as
μ = λ 1 + λ 2 − k + 4 , ω = ν 1 + ν 2 + k − 8 2 . \mu= \lambda_1 + \lambda_2 - k + 4,\qquad\omega= \nu_1 + \nu_2 + \frac{k-8}{2}. μ = λ 1 + λ 2 − k + 4 , ω = ν 1 + ν 2 + 2 k − 8 .
It remains to impose invariance under M τ ξ M_{\tau\xi} M τ ξ . For a factor of weight λ \lambda λ , one has
M τ ξ ⋅ ( D ξ i D q ∂ τ ) = D ξ i + 1 D q + 2 λ D ξ i ∂ τ . M_{\tau\xi} \cdot\left(D_{\xi}^{i}D_q\partial_\tau\right) = D_{\xi}^{i+1}D_q + 2\lambda D_{\xi}^{i}\partial_\tau. M τ ξ ⋅ ( D ξ i D q ∂ τ ) = D ξ i + 1 D q + 2 λ D ξ i ∂ τ .
Therefore
M τ ξ ⋅ ( D ξ i D q ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ ) = D ξ i + 1 D q ⊠ D ξ k − 4 − i D q ∂ τ + 2 λ 1 D ξ i ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ + D ξ i D q ∂ τ ⊠ D ξ k − 3 − i D q + 2 λ 2 D ξ i D q ∂ τ ⊠ D ξ k − 4 − i ∂ τ . \begin{aligned}
M_{\tau\xi} \cdot\left(D_{\xi}^{i}D_q\partial_\tau\boxtimes D_{\xi}^{k-4-i}D_q\partial_\tau\right)
={}&D_{\xi}^{i+1}D_q \boxtimes D_{\xi}^{k-4-i}D_q\partial_\tau
+2\lambda_1D_{\xi}^{i}\partial_\tau\boxtimes D_{\xi}^{k-4-i}D_q\partial_\tau\\
&+D_{\xi}^{i}D_q\partial_\tau\boxtimes D_{\xi}^{k-3-i}D_q
+2\lambda_2D_{\xi}^{i}D_q\partial_\tau\boxtimes D_{\xi}^{k-4-i}\partial_\tau.
\end{aligned} M τ ξ ⋅ ( D ξ i D q ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ ) = D ξ i + 1 D q ⊠ D ξ k − 4 − i D q ∂ τ + 2 λ 1 D ξ i ∂ τ ⊠ D ξ k − 4 − i D q ∂ τ + D ξ i D q ∂ τ ⊠ D ξ k − 3 − i D q + 2 λ 2 D ξ i D q ∂ τ ⊠ D ξ k − 4 − i ∂ τ .
Consequently,
M τ ξ ⋅ J 0 , C ( k ) M_{\tau\xi} \cdot J^{(k)}_{0,C} M τ ξ ⋅ J 0 , C ( k )
contains the monomials
D ξ i + 1 D q ⊠ D ξ k − 4 − i D q ∂ τ , 0 ≤ i ≤ k − 4 , D_{\xi}^{i+1}D_q \boxtimes D_{\xi}^{k-4-i}D_q\partial_\tau,\qquad0 \le i \le k - 4, D ξ i + 1 D q ⊠ D ξ k − 4 − i D q ∂ τ , 0 ≤ i ≤ k − 4 ,
with respective coefficients A i A_i A i . These tensor monomials are pairwise distinct, and none of them can arise from the action on the second tensor factor or from any other summand of J 0 ˉ , C ( k ) J^{(k)}_{\bar{0},C} J 0 ˉ , C ( k ) . Hence, M τ ξ M_{\tau\xi} M τ ξ -invariance forces
A i = 0 , 0 ≤ i ≤ k − 4. A_i=0,\qquad0\leq i\leq k-4. A i = 0 , 0 ≤ i ≤ k − 4.
Thus
J 0 ˉ , C ( k ) = 0. J^{(k)}_{\bar{0},C}=0. J 0 ˉ , C ( k ) = 0.
Since no division by the weights was used, the conclusion holds for all ( λ 1 , ν 1 ) (\lambda_1,\nu_1) ( λ 1 , ν 1 ) and ( λ 2 , ν 2 ) (\lambda_2,\nu_2) ( λ 2 , ν 2 ) , including all resonant values.
Acknowledgments. S.B. was supported by the grant NYUAD 065.
Disclosures. No conflict of interest. The data used to support the findings of this study are included within the article. The codes are available upon request.
*Corresponding author; aDivision of Science and Mathematics, New York University Abu Dhabi, Po Box 129188, United Arab Emirates; sofiane.bouarroudj@nyu.edu
bDepartment of Mathematics, Albanovägen 28, Stockholm University, Stockholm, SE-106 91 Sweden; dimleites@gmail.com
cFaculty of Mechanics and Mathematics, Taras Shevchenko National University of Kyiv, Hlushkova Avenue 4e, 03127 Kyiv, Ukraine; alozhechnik@gmail.com
dIndependent University of Moscow, B. Vlasievsky per., d. 11, RU-119 002 Moscow, Russia; irina@mccme.ru