Arxiv:1907.03525V3 [Math.QA] 20 Oct 2020 .Teuiesladtemeromorphic the and Abelian Universal Meromorphic the the and Universal 7

Home , A Separation

arXiv:1907.03525v3 [math.QA] 20 Oct 2020 .Teuiesladtemeromorphic the and abelian universal meromorphic The the and universal 7. The element 6. The element 5. coproducts The Drinfeld and standard 4. The Yangian 3. The 2. pedxB nqeeso h universal the of Uniqueness References points of B. algebra Separation Appendix loop quantum the A. to Appendix Relation structures tensor 9. Meromorphic 8. .Introduction 1. AHNGUA,VLROTLDN AEO N UTSWENDLAN CURTIS AND LAREDO, TOLEDANO VALERIO GAUTAM, SACHIN H MEROMORPHIC THE es npriua,w ieacntutv ro fteexis the of proof constructive a give we particular, In as ness. properties formal same the osrcinrle nteGusdecomposition Gauss the on relies construction ntrt odto.Ec osse naypoi expansi asymptotic an possesses Each which condition. constraints, unitarity commutativity meromorphic two instead cs qiaety h esrctgr ffinite–dimensio w of compatible category and tensor natural of the be Equivalently, to solutions ucts. such finite-dimensi requires arbitrary on geom one hold if from to arising ceases rationality representations that and algebras, Kac–Moody h iegn bla term abelian divergent The a ouin fteQB nirdcbe finite–dimensiona irreducible, on Y QYBE the of solutions nal htte netietesadr ordc of coproduct standard the intertwine they that ettosb h rttoatosi [ in authors two first the by sentations Abstract. edcpoutitoue n[ in introduced coproduct feld R rnedpoe htteuniversal the that proved Drinfeld ~ ± ( Y g ( ~ .Ti eutwsrcnl xeddb alkOono os to Maulik–Okounkov by extended recently was result This ). s ( ,poeta hyaertoa nfiiedmninlrepres finite–dimensional on rational are they that prove ), g osntamtrtoa omttvt osrit.W con We constraints. commutativity rational admit not does ) Let R R Y ~ − − g oKlaRseihn nhs6t birthday. 60th his on Reshetikhin, Kolya To ( ( ( eacmlxsmsml i ler and algebra Lie semisimple complex a be g s s ) for ) ) sl 2 R 12 R 0 ( ]. Contents R s a eumdo nt–iesoa repre- finite–dimensional on resummed was ,adteeoecicdswt tb unique- by it with coincides therefore and ), MTI FTEYANGIAN THE OF –MATRIX R –matrix 1 12 .I h rsn ae,w construct we paper, present the In ]. R mtie of –matrices U R R q –matrix ( ( R Y L s ~ of ) + g ( R ) ( g s n h eomdDrin- deformed the and ) mtie of –matrices ) Y R · ~ ( g nlrepresentations, onal 0 ie iet ratio- to rise gives ) ( a representations nal ec of tence ersnain of representations l Y s nin on Y ) ~ ~ r eae ya by related are R · ( t esrprod- tensor ith ( ty eshow We etry. g g nain,and entations, ) t Yangian. its ) − s ( s R hc has which of ) Y ymmetric ( ~ s ( .Our ). struct g R ) ( s ). DT 48 46 43 38 36 32 23 20 14 11 7 2 2 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

1. Introduction 1.1. Let g be a complex, semisimple Lie algebra with an invariant inner product (·, ·), and Y~(g) the corresponding Yangian, which is a Hopf algebra deforming the current algebra U(g[z]) introduced by Drinfeld [4]. We assume that ~ ∈ C× is ﬁxed throughout. Drinfeld proved that Y~(g) possesses a unique universal R– matrix. Speciﬁcally, let ∆ : Y~(g) → Y~(g) ⊗ Y~(g) be the coproduct of Y~(g), and τs : Y~(g) → Y~(g) the one–parameter group of automorphisms which quantizes the shift automorphism z 7→ z + s of U(g[z]). Note that, if s is considered as a variable, τs may be regarded as a homomorphism Y~(g) → Y~(g)[s]. Then, the following holds. Theorem. [4, Thm. 3] (1) There is a unique formal series ∞ −k ⊗2 −1 R(s)=1+ Rks ∈ Y~(g) [[s ]] Xk=1 1 such that the following holds in Y~(g)⊗2[s; s−1]]

op −1 τs ⊗ 1 ◦ ∆ (a)= R(s) · τs ⊗ 1 ◦ ∆(a) · R(s) (1.1) for any a ∈ Y~(g), and

∆ ⊗ 1(R(s)) = R13(s) · R23(s) (1.2)

1 ⊗ ∆(R(s)) = R13(s) · R12(s) (1.3) (2) R satisﬁes the following identities −1 −2 • 1–jet: R(s)=1+ ~s Ωg + O(s ) −1 • Unitarity: R(s) = R21(−s) • Translation: τa ⊗ τb(R(s)) = R(s + a − b)

where Ωg ∈ g ⊗ g is the Casimir tensor corresponding to (·, ·). (3) R is a solution of the quantum Yang–Baxter equation (QYBE)2

R12(s1)R13(s1 + s2)R23(s2)= R23(s2)R13(s1 + s2)R12(s1) (1.4) 1.2. One of the main goals of this paper is to clarify the analytic nature of the formal power series R(s), and of the solutions of the QYBE obtained from it.

Let V1, V2 be two ﬁnite–dimensional representations of Y~(g), and RV1,V2 (s) ∈ −1 End(V1 ⊗ V2)[[s ]] the corresponding evaluation of R(s). Drinfeld proved that

RV1,V2 (s) has a zero radius of convergence in general [4, Examples 1,2], but never- theless gives rise to a rational solution of the QYBE as follows.

Theorem. [4, Thm. 4] Assume that V1 and V2 are irreducible with highest weight −1 −1 vectors v1, v2, and let ρV1,V2 (s) ∈ 1+ s C[[s ]] be the matrix element of RV1,V2 (s) given by

RV1,V2 (s) v1 ⊗ v2 = ρV1,V2 (s) v1 ⊗ v2

1Our conventions diﬀer slightly from those of [4], where the intertwining equation (1.1) is op −1 −1 written as 1 ⊗ τs ◦ ∆ (a) = R(s) · 1 ⊗ τs ◦ ∆(a) · R(s). Thus, our R(s) is Drinfeld’s R(−s) . 2 ⊗3 −1 −1 The QYBE may be viewed as an equation in Y~(g) [s1; s1 ]][[s2 ]] by expanding it as −1 k k −k−1 if |s2| ≫ |s1|, that is setting (s1 + s2) = Pk≥0(−1) s1 s2 , or as an equation in 3 1 −1 1 −k−1 Y~ ⊗ s s − s s s − ksks (g) [ 2; 2 ]][[ 1 ]], by setting ( 1 + 2) = Pk≥0(−1) 2 1 . The precise statement of (3) above is that (1.4) holds in either of these cases. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 3

Then, −1 (1) RV1,V2 (s)= RV1,V2 (s) · ρV1,V2 (s) is a rational function of s. (2) If V1 = V = V2, (1.4) together with the factorisation

RV1,V2 (s)= RV1,V2 (s) · ρV1,V2 (s) (1.5)

imply that RV,V (s) is a rational solution of the QYBE. More recently, a geometric construction of R–matrices corresponding to the (extended) Yangian of a symmetric Kac–Moody algebra was given by Maulik– Okounkov [20], which provides in particular an alternative construction of rational solutions of the QYBE on the equivariant cohomology of Nakajima quiver varieties. 1.3. One of the byproducts of this paper is to extend the factorisation (1.5) to an arbitrary pair of (not necessarily irreducible) ﬁnite–dimensional representations. −1 In this case, the divergent factor ρV1,V2 (s) takes values in End(V1 ⊗ V2)[[s ]], and intertwines the action of Y~(g) given by ∆s = τs ⊗1◦∆, whereas the rational factor R op op V1,V2 (s) intertwines ∆s and ∆s = τs ⊗ 1 ◦ ∆ . However, since ρV1,V2 (s) is not scalar–valued in general, it is not clear whether RV1,V2 (s) satisﬁes the QYBE when V1 = V2.

We prove in fact that, even for g = sl2, no rational intertwiner RV1,V2 (s) ∈ End(V1 ⊗ V2) exists which is defined for any V1, V2 ∈ Repfd(Y~(g)), is natural in V1 and V2, and satisfies the cabling identities (1.2)–(1.3). Equivalently, the tensor category of finite–dimensional representations of Y~(g) does not admit rational commutativity constraints. In particular, this raises the question of whether one can consistently define rational solutions of the QYBE on all finite–dimensional 3 representations of Y~(g). 1.4. In the present paper, we propose an alternative solution to this issue, by constructing meromorphic commutativity constraints on Repfd(Y~(g)), and in particular consistent meromorphic solutions of the QYBE on all V ∈ Repfd(Y~(g)). Namely, we prove that the universal R–matrix of Y~(g), while generally divergent on a pair of finite–dimensional representations V1, V2, can be canonically resummed, in two distinct ways. This yields a pair of meromorphic functions R↑ (s), R↓ (s): C → End(V ⊗ V ) V1,V2 V1,V2 1 2 which are natural with respect to V1, V2, satisfy the intertwining relation (1.1), the cabling identities (1.2)–(1.3), as well as the translation property. The function R↑ (s) (resp. R↓ (s)) is asymptotic to R (s) as s → ∞ with Re(s/~) > 0 V1,V2 V1,V2 V1,V2 (resp. Re(s/~) < 0), and is related to R↓ (s) by the unitarity relation V1,V2 R↑ (s)−1 = R↓ (−s)21 V1,V2 V2,V1 The situation is somewhat analogous to the case of the quantum loop algebra Uq(Lg). In that case, if R ∈ Uq(Lg)⊗Uq(Lg) is the universal R–matrix, then ∞ ⊗2 −1 0 ⊗2 R (z)= τz ⊗ 1(R) ∈ Uq(Lg) [[z ]] and R (z)= 1 ⊗ τz(R) ∈ Uq(Lg) [[z]] b converge, near z = ∞ and z = 0 respectively, to meromorphic functions of z ∈ C× on the tensor product V1 ⊗V2 of any two finite–dimensional representations [7, 16],

3The Maulik–Okounkov construction mentioned in 1.2 provides a partial solution to this question, since an arbitrary representation of Y~(g) may not have a geometric realisation. 4 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

R∞ −1 R0 −1 21 g and are related by V1,V2 (z) = V2,V1 (z ) . In the case of Uq(L ), however, R∞(z) and R0(z) are convergent as is, and do not need to be resummed. 1.5. Our approach does not rely on Drinfeld’s cohomological construction of R(s) to carry out the resummation. It produces the functions R↑/↓ (s) through a direct, V1,V2 explicit construction, which shows in particular that they have an asymptotic expansion as s → ∞. The fact that the latter coincides with RV1,V2 (s), and therefore a posteriori that RV1,V2 (s) can be resummed, follows from the fact that the asymptotic expansion can be lifted to Y~(g)⊗2[[s−1]], and shown to have the properties which uniquely determine R(s) by Theorem 1.1. In particular, our construction yields an independent, and constructive proof of the existence of R(s). 1.6. Our construction can be motivated by the following considerations. The R– matrix of Y~(g) is expected to arise as the canonical element in DY~(g)⊗DY~(g), where DY~(g) ⊃ Y~(g) is the double Yangian of g, which is a quantisation of the graded Drinfeld double b g[z±1], g[z],z−1g[z−1] of g[z]. Although a detailed understanding of DY~(g) is still lacking at present (see, however, [17] and [24]), this suggests that, given a triangular decomposition g = n+ ⊕ h ⊕ n− of g, R(s) should have a corresponding Gauss decomposition R(s)= R+(s) · R0(s) · R−(s) (1.6) where R0(s) quantises the canonical element in h[z]⊗z−1h[z−1], and R±(s) those −1 −1 in n±[z]⊗z n∓[z ] respectively. Moreover, the unitarity of R(s) suggests that R0(s)−1 = R0(−s)21 and R+(s)b−1 = R−(−s)21 b Accordingly, we construct each factor R0(s), R−(s), R+(s) = (R−(−s)21)−1, and their resummation on finite–dimensional representations separately. 1.7. Khoroshkin–Tolstoy gave a heuristic formula for R0 [17], as the exponential of an infinite sum in the abelian subalgebra of DY~(g) which quantises h[z,z−1]. In [12], the first two named authors gave a precise version of this formula, where the 0 exponent takes values in the abelian subalgebra Y~ (g) of Y~(g) which quantises h[z]. We showed moreover that this expression can be resummed on a tensor product V1 ⊗ V2 of finite–dimensional representations in two different ways. This yields two meromorphic functions R0,↑ (s), R0,↓ (s), which have the same asymptotic V1,V2 V1,V2 expansion on ± Re(s/~) > 0, and are related by R0,↑ (s)−1 = R0,↓ (−s)21. V1,V2 V2,V1 1.8. An important discovery of [12] is that these abelian R–matrices play a similar role to that of the full R–matrix of Y~(g), but with respect to the deformed Drinfeld tensor product. The latter was introduced in [12] by degenerating the Drinfeld tensor product of the quantum loop algebra introduced by Hernandez [15]. It gives rise to a family of actions of Y~(g) on the vector space V1 ⊗ V2, which is denoted by V1 ⊗ V2 and is a rational function of a parameter s ∈ C. The tensor product ⊗ is D,s D,s associative, in that the identification of vector spaces

(V1 ⊗ V2) ⊗ V3 = V1 ⊗ (V2 ⊗ V3) D,s1 D,s2 D,s1+s2 D,s2 intertwines the action of Y~(g) for any s1,s2 ∈ C, and endows Repfd(Y~(g)) with the structure of a meromorphic tensor category in the sense of [22, 23]. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 5

The endomorphisms R0,↑/↓ are meromorphic commutativity constraints with re- V1,V2 spect to ⊗ , that is they satisfy the representation theoretic version of the identities D,s (1) of Theorem 1.1. In the present paper, we complement the results of [12] by lifting ⊗ to a deformed Drinfeld coproduct D,s −1 ∆ : Y~(g) → (Y~(g) ⊗ Y~(g))[s; s ]] D,s and the common asymptotic expansion of R0,↑/↓ (s) to an element V1,V2 0 −1 R (s) ∈ (Y~(g) ⊗ Y~(g))[[s ]] which satisfy the identities (1) of Theorem 1.1, with τs ⊗ 1 ◦ ∆ replaced by ∆, and D,s R(s) by R0(s). 1.9. The central ingredient of the present paper is the construction of R±(s), which is based on the following. The fact that R(s) (resp. R0(s)) conjugates the standard coproduct ∆s = τs⊗1◦∆ (resp. the deformed Drinfeld coproduct ∆) to its D,s opposite, together with the Gauss decomposition (1.6), suggest that R−(s) should conjugate the standard coproduct ∆s to the deformed Drinfeld coproduct ∆. This D,s is consistent with the fact that an analogous statement holds for the quantum loop algebra, and the related construction of twists conjugating quantum coproducts corresponding to different polarisations of a Manin triple given in [6]. In this case, the standard (resp. Drinfeld) coproducts on Uq(Lg) correspond, respectively, to the polarisations −1 −1 ±1 ±1 −1 −1 ±1 g[z] ⊕ z g[z ]= g[z ]= n−[z ] ⊕ h[z] ⊕ z h[z ] ⊕ n+[z ] 1.10. We prove that this intertwining property uniquely determines an element − − + −1 R (s), provided it is required to lie in Y~ (g) ⊗ Y~ (g) [[s ]] and have constant ± term 1, where Y (g) ⊂ Y~(g) is the subalgebra deforming U(n [z]). We show in ~ ± fact that, under this triangularity assumption, R−(s) is uniquely determined by the requirement that it intertwines the standard and Drinfeld coproducts of the loop generators ti,0,ti,1 of Y~(g) which deform h ⊕ h ⊗ z ⊂ h[z]. − We then show that, for any V , V ∈ Rep (Y~(g)), R (s) is a rational function 1 2 fd V1,V2 of s. Moreover, the following cocycle identity holds for any V1, V2, V3 ∈ Repfd(Y~(g)) R− (s ) · R− (s )= R− (s + s ) · R− (s ) (1.7) V1 ⊗ V2,V3 2 V1,V2 1 V1,V2 ⊗ V3 1 2 V2,V3 2 D,s1 D,s2 Together with the identities satisfied by R0(s), this guarantees that the product R(s)= R+(s)·R0(s)·R−(s), where R+(s) = (R−(−s)21)−1, satisfies the identities (1.1)–(1.3) on any pair of finite–dimensional representations. A separation of points argument then implies that R(s) satisfies Drinfeld’s uniqueness criterion of the 4 universal R–matrix of Y~(g), and in particular coincides with it. Finally, since R− (s) is a rational function of s, the product V1,V2 R↑/↓ (s)= R+ (s) · R0,↑/↓ (s) · R− (s) (1.8) V1,V2 V1,V2 V1,V2 V1,V2

4The passage to ﬁnite–dimensional representations is dictated by the fact that the cocycle identity (1.7) does not appear to have a natural lift to Y~(g). Indeed, when lifted to Y~(g), the left– ⊗3 −1 −1 ⊗3 −1 −1 hand side lies in Y~(g) [s1; s1 ]][[s2 ]], while the right–hand side lies in Y~(g) [s2; s2 ]][[s1 ]]. 6 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

is a resummation of RV1,V2 (s), as well a meromorphic commutativity constraint on Repfd(Y~(g)) with respect to the standard tensor product.

1.11. Our results may be rephrased as follows. As proved in [12], and mentioned above, ﬁnite–dimensional representations of Y~(g), together with the deformed Drinfeld tensor product ⊗ and one of the resummed abelian R–matrices D,s R0,↑/↓(s) is a meromorphic braided tensor category. Similarly, Repfd(Y~(g)) endowed with the deformed standard tensor product ∗ ⊗s = ⊗◦ (τs ⊗ 1) is a meromorphic (in fact, polynomial) tensor category. Our construction of the resummed R–matrices R↑/↓(s) endows this category with a meromorphic braiding.5 Moreover, the element R−(s) is a rational braided tensor structure on the identity functor

0,↑/↓ ↑/↓ (Repfd(Y~(g)), ⊗ , R ) → (Repfd(Y~(g)), ⊗s, R ) D,s

That is, R−(s) gives rise to a system of natural isomorphisms of Y~(g)–modules R− (s): V ⊗ V → V ⊗ V , which is compatible with the (trivial) associativity V1,V2 1 s 2 1 2 D,s constraints and the meromorphic braidings, i.e., such that the following diagrams commute for any V1, V2, V3 ∈ Repfd(Y~(g))

(V1 ⊗s1 V2) ⊗s2 V3 V1 ⊗s1+s2 (V2 ⊗s2 V3)

− − R (s1)⊗1V 1V ⊗R (s2) V1,V2 3 1 V2,V3

(V1 ⊗ V2) ⊗s2 V3 V1 ⊗s1+s2 (V2 ⊗ V3) D,s1 D,s2

− − R (s2) R (s1+s2) V1 ⊗ V2,V3 V1,V2 ⊗ V3 D,s1 D,s2 (V1 ⊗ V2) ⊗ V3 V1 ⊗ (V2 ⊗ V3) D,s1 D,s2 D,s1+s2 D,s2 as dictated by the cocycle equation (1.7), and

(1 2)◦R↑/↓ (s) V1,V2 / V1(s) ⊗ V2 V2 ⊗ V1(s)

R− (s) R− (−s) V1,V2 V2,V1 V (s) ⊗ V / V ⊗ V (s) 1 2 0,↑/↓ 2 1 D,0 (1 2)◦R (s) D,0 V1,V2 which follows from the Gauss decomposition (1.8), together with the fact that (R−(−s)21)−1 = R+(s).

5An analogous statement was proved by Kazhdan–Soibelman for the quantum loop algebra in [16]. As pointed out in 1.4, however, in the case of Uq(Lg) no resummation of the universal R–matrix of Uq(Lg) is needed. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 7

1.12. Outline of the paper. We review the definition of Y~(g) in Section 2, and that of the standard and Drinfeld coproducts in Section 3. In Section 4, we prove the existence and uniqueness of R−(s), and establish its various properties. In − Section 5, we give an explicit expressions for R (s) when g = sl2. In Section 6, we review the construction of R0,↑/↓(s) given in [12]. We then explicitly lift its asymp- 0 ⊗2 −1 totic expansion to Y~ (g) [[s ]], and prove that it satisfies properties analogous to those of Drinfeld’s R–matrix, but with respect to the deformed Drinfeld coproduct. We also prove that there is no rational commutativity constraint on Repfd(Y~(g)). Combined with the results of Section 4, we obtain the same assertions for the standard tensor product in Section 7. We give a proof of the uniqueness of the universal R–matrix of the Yangian in Appendix B, thus completing the proof that our construction gives rise to Drinfeld’s R–matrix. In Section 8, we restate our results in the language of meromorphic tensor categories. In the final Section 9, we discuss the analogous case of the quantum loop algebra, and relate the two by means of the meromorphic tensor functor constructed in [12]. Appendix A contains a proof due to Drinfeld that finite–dimensional representations separate points of Y~(g). 1.13. Acknowledgments. We would like to thank Pavel Etingof for several helpful discussions about q KZ equations and R–matrices. We are also grateful to Maria Angelica Cueto for helping us with the combinatorial aspects of the paper. The first author was supported through the Simons foundation collaboration grant 526947. The second author was supported through the NSF grant DMS–1802412. The third author was supported by an NSERC CGS D graduate award and an NSERC PDF postdoctoral fellowship.

2. The Yangian Y~(g) 2.1. Let g be a complex, semisimple Lie algebra and (·, ·) an invariant, symmetric, non–degenerate bilinear form on g. Let h ⊂ g be a Cartan subalgebra of g, {αi}i∈I ⊂ ∗ h a basis of simple roots of g relative to h and aij = 2(αi, αj )/(αi, αi) the entries ∗ of the corresponding Cartan matrix A. Let Φ+ ⊂ h be the corresponding set of Q ∗ positive roots, and = ZΦ+ = i∈I Zαi ⊂ h the root lattice. We assume that (·, ·) is normalised so that the square length of short roots is 2. Set di = (αi, αi)/2 ∈ L −1 {1, 2, 3}, so that diaij = dj aji for any i, j ∈ I. In addition, we set hi = ν (αi)/di ± + − ∗ and choose root vectors xi ∈ g±αi such that [xi , xi ] = dihi, where ν : h → h is the isomorphism determined by (·, ·).

2.2. The Yangian Y~(g)[5]. Let ~ ∈ C. The Yangian Y~(g) is the C–algebra ± I Z generated by elements {xi,r, ξi,r}i∈ ,r∈ ≥0 , subject to the following relations. (Y1) For any i, j ∈ I, r, s ∈ Z≥0:[ξi,r, ξj,s]=0

± ± (Y2) For i, j ∈ I and s ∈ Z≥0:[ξi,0, xj,s]= ±diaij xj,s

(Y3) For i, j ∈ I and r, s ∈ Z≥0: d a [ξ , x± ] − [ξ , x± ]= ±~ i ij (ξ x± + x± ξ ) i,r+1 j,s i,r j,s+1 2 i,r j,s j,s i,r

(Y4) For i, j ∈ I and r, s ∈ Z≥0: d a [x± , x± ] − [x± , x± ]= ±~ i ij (x± x± + x± x± ) i,r+1 j,s i,r j,s+1 2 i,r j,s j,s i,r 8 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

+ − (Y5) For i, j ∈ I and r, s ∈ Z≥0:[xi,r, xj,s]= δij ξi,r+s

(Y6) Let i 6= j ∈ I and set m =1 − aij . For any r1, ··· , rm ∈ Z≥0 and s ∈ Z≥0

x± , x± , ··· , x± , x± ··· =0 i,rπ(1) i,rπ(2) i,rπ(m) j,s S πX∈ m h h h h i ii 0 ± We denote by Y~ (g) and Y~ (g) the unital subalgebras of Y~(g) generated by ± I Z I Z {ξi,r}i∈ ,r∈ ≥0 and {xi,r}i∈ ,r∈ ≥0 , respectively.

± −1 2.3. Assume henceforth that ~ 6= 0, and deﬁne ξi(u), xi (u) ∈ Y~(g)[[u ]] by −r−1 ± ± −r−1 ξi(u)=1+ ~ ξi,ru and xi (u)= ~ xi,ru Xr≥0 Xr≥0 Proposition. [11, Prop. 2.3] The relations (Y1),(Y2)–(Y3),(Y4),(Y5) and (Y6) are respectively equivalent to the following identities in Y~(g)[u, v; u−1, v−1]].

(Y1) For any i, j ∈ I, [ξi(u), ξj (v)]=0

± ± (Y2) For any i, j ∈ I, [ξi,0, xj (u)] = ±diaij xj (u)

(Y3) For any i, j ∈ I, and a = ~diaij /2: ± ± ± (u − v ∓ a)ξi(u)xj (v) = (u − v ± a)xj (v)ξi(u) ∓ 2axj (u ∓ a)ξi(u)

(Y4) For any i, j ∈ I, and a = ~diaij /2:

± ± (u − v ∓ a)xi (u)xj (v) ± ± ± ± ± ± = (u − v ± a)xj (v)xi (u)+ ~ [xi,0, xj (v)] − [xi (u), xj,0] (Y5) For any i, j ∈ I: + − (u − v)[xi (u), xj (v)] = −δij ~ (ξi(u) − ξi(v))

(Y6) For any i 6= j ∈ I, m =1 − aij , r1, ··· , rm ∈ Z≥0, and s ∈ Z≥0:

± ± ± ± xi (uπ1 ), xi (uπ(2)), ··· , xi (uπ(m)), xj (v) ··· =0 π∈S Xm ± ± ± Remark. When g = sl2, we will write ξr, xr , ξ(u) and x (u) in place of ξi,r, xi,r, ± ξi(u) and xi (u), respectively.

0 0 2.4. Alternative generators of Y~ (g). Let {ti,r}i∈I,r∈Z≥0 ⊂ Y~ (g) be the generators deﬁned by −r−1 ti(u)= ~ ti,ru := log(ξi(u)) Xr≥0 In particular, ti,0 = ξi,0 and ~ t = ξ − ξ2 (2.1) i,1 i,1 2 i,0 The relations (Y2)–(Y3) of Y~(g) imply that for any i, j ∈ I and r ∈ Z≥0, ± ± [ti,1, xj,r]= ±diaij xj,r+1 (2.2) ± so that ti,1 act as shift operators on the generators xj,r. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 9

± The relation (2.2) also implies that {ξi,0, xi,0,ti,1}i∈I generate Y~(g) as an algebra. We refer the reader to [18, Thm. 1.2] for a presentation of Y~(g) given in terms of these generators, and to [13, Thm. 2.13] for a reﬁnement of this result.

2.5. Shift automorphism. The group of translations of the complex plane acts on Y~(g) by r r τ (y )= ar−sy a r s s s=0 X ± where a ∈ C and y is one of ξi,ti, xi . In terms of the generating series introduced in 2.3 and 2.4, we have

τa(y(u)) = y(u − a) ∗ Given a representation V of Y~(g) and a ∈ C, set V (a)= τa (V ).

2.6. PBW theorem. Consider the loop filtration F•(Y~(g)) on Y~(g) defined by ± deg(yr)= r for each of the generators y = ξi, xi . Note that deg(ti,r)= r. The Hopf algebra structure on Y~(g) preserves this filtration, and endows gr(Y~(g)) with the structure of a graded Hopf algebra. The PBW Theorem for Y~(g)[19] (see also [8, Thm. B.6] and [14, Prop. 2.2]) is equivalent to the assertion that the assignments ± r ± r ¯ xi .z 7→ x¯i,r and dihi.z 7→ ξi,r uniquely extend to an isomorphism of graded Hopf algebras ∼ U(g[z]) → gr(Y~(g)), (2.3) where, for any fixed k ∈ Z≥0 and element yk ∈Fk(Y~(g)),

y¯k ∈ grk(Y~(g)) := Fk(Y~(g))/Fk−1(Y~(g)) is defined to be the image of yk in the k-th graded component grk(Y~(g)) of the associated graded algebra gr(Y~(g)). Henceforth, we shall freely make use of the above identification without further comment. Similarly, we will exploit the fact that it allows us to identify U(g[z]) ⊗ U(g[w]) with gr(Y~(g) ⊗ Y~(g)) =∼ gr(Y~(g)) ⊗ gr(Y~(g)), the associated graded algebra of Y~(g) ⊗ Y~(g) with respect to the tensor product filtration F•(Y~(g) ⊗ Y~(g)) induced by F•(Y~(g)).

2.7. The embedding U(g) ⊂ Y~(g). Since gr0(Y~(g)) = F0(Y~(g)) ⊂ Y~(g), the isomorphism (2.3) restricts to an embedding of U(g) into Y~(g), given by ± ± xi 7→ xi,0 and dihi 7→ ξi,0

We shall henceforth identify U(g) ⊂ Y~(g), with the above embedding implicitly understood. Viewed as a module over h ⊂ Y~(g), we then have Y~(g)= β∈Q Y~(g)β A second embedding T : h → Y~(g) is given by setting T(d h )= t for all i ∈ I, i i i,1L where ti,1 is deﬁned by (2.1). The relation (2.2) then reads ± ± [T(h), xi,r]= ±αi(h)xi,r+1 (2.4) ⊥ and implies in particular that, for any h ∈ αi and r ≥ 0, ± [T(h), xi,r] = 0 (2.5) 10 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

2.8. Formal series ﬁltration. For any k ∈ Z, set

⊗2 −1 k ⊗2 −n Fk(Y~(g) [s; s ]])) = s Fn(Y~(g) )s nY≥0 (2.6) m ⊗2 −1 = y s ∈ Y~(g) [s; s ]] deg(y ) ≤ k − m  m m  m≤M  X 

For k ≤ 0, we shall write this as F (Y~(g)⊗2[[s−1]])), for obvious reasons.  k  The above spaces generate a Z-ﬁltered algebra

⊗2 −1 ⊗2 −1 Fk(Y~(g) [s; s ]]) ⊂ Y~(g) [s; s ]] Z k[∈ with associated graded algebra that can (and will) be identiﬁed with the C[s±1]– submodule of (U(g[z]) ⊗ U(g[w])[s; s−1]] generated by

−n (U(g[z]) ⊗ U(g[w]))ns nY≥0 where (U(g[z]) ⊗ U(g[w]))n is the n-th graded component of U(g[z]) ⊗ U(g[w]). ⊗2 −1 If X(s) ∈Fk(Y~(g) [s; s ]]), we denote by ⊗2 −1 X(s)= X(s) mod Fk−1(Y~(g) [s; s ]]) k −n ∈ s (U(g[z]) ⊗ U(g[w]))ns nY≥0 ⊂ (U(g[z]) ⊗ U(g[w])[s; s−1]] the image of X(s) in the k–th graded component of the associated graded algebra. ⊗2 −1 Note in passing that if X(s) ∈Fk(Y~(g) [s; s ]]), with k< 0, then exp(X(s))−1 ∈ ⊗2 −1 Fk(Y~(g) [s; s ]]), and exp(X(s)) − 1= X(s). (2.7)

2.9. Rationality. The following rationality property is due to Beck–Kac [1] and Hernandez [15] for the analogous case of the quantum loop algebra, and to the ﬁrst two authors for Y~(g). In the form below, the result appears in [11, Prop. 3.6].

Proposition. Let V be a Y~(g)–module on which {ξi,0}i∈I act semisimply with ﬁnite–dimensional weight spaces. Then, for every weight µ of V , the generating series −1 ± −1 ξi(u) ∈ End(Vµ)[[u ]] and xi (u) ∈ Hom(Vµ, Vµ±αi )[[u ]] are the Laurent expansions at ∞ of rational functions of u. Speciﬁcally, ad(t ) −1 x±(u)= ~u−1 1 ∓ i,1 x± i 2d u i,0 i and + − ξi(u) = 1 + [xi (u), xi,0]

If V is a finite–dimensional Y~(g)–module, we define σ(V ) ⊂ C to be the (finite) ± set of poles of the rational End(V )–valued functions {ξi(u), xi (u)}i∈I. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 11

3. The standard and Drinfeld coproducts

We review the definition of the standard coproduct on Y~(g) following [13], and the deformed Drinfeld tensor product on its finite–dimensional representations introduced in [12]. We then lift the latter to a deformed Drinfeld coproduct ∆ : Y~(g) → Y~(g)⊗2[s; s−1]]. D,s 3.1. Standard coproduct. Set r − + = xβ,0 ⊗ xβ,0 (3.1) βX∈Φ+ ± − + where xβ,0 ∈ g±β ⊂ Y~(g) are root vectors such that (xβ,0, xβ,0) = 1. The coproduct ∆ : Y~(g) → Y~(g) ⊗ Y~(g) is defined by the following formulae

∆(ξi,0)= ξi,0 ⊗ 1 +1 ⊗ ξi,0

± ± ± ∆(xi,0)= xi,0 ⊗ 1+1 ⊗ xi,0

∆(ti,1)= ti,1 ⊗ 1 +1 ⊗ ti,1 + ~ ad(ξi,0 ⊗ 1)r − + = ti,1 ⊗ 1 +1 ⊗ ti,1 − ~ (β, αi)xβ,0 ⊗ xβ,0 βX∈Φ+ We refer the reader to [13, §4.2] for a proof that ∆ is an algebra homomorphism. It is immediate that ∆ is coassociative (see [13, §4.5]). 3.2. Deformed Drinfeld tensor product. We review below the deﬁnition of the deformed Drinfeld tensor product introduced in [12, Section 4.4]. Let V, W ∈ Repfd(Y~(g)), and σ(V ), σ(W ) ⊂ C their sets of poles. Let s ∈ C be such that σ(V )+ s and σ(W ) are disjoint, and deﬁne an action of the generators of Y~(g) on V ⊗ W by

∆(ξi(u)) = ξi(u − s) ⊗ ξi(u) D,s

+ + 1 + ∆(xi (u)) = xi (u − s) ⊗ 1+ ξi(v − s) ⊗ xi (v)dv D,s u − v IC2 − 1 − − ∆(xi (u)) = xi (v − s) ⊗ ξi(v)dv +1 ⊗ xi (u) D,s u − v IC1 where

• C1 encloses σ(V )+ s and none of the points in σ(W ). • C2 encloses σ(W ) and none of the points in σ(V )+ s. • The integral (resp. ) is understood to mean the holomorphic func- C1 C2 tion of u it deﬁnes for u outside of C (resp. C ). H H 1 2 0 Note that in terms of the generators ti,r of Y~ (g),

∆(ti(u)) = ti(u − s) ⊗ 1+1 ⊗ ti(u) D,s Theorem. [12, Thm. 4.6] (1) The formulae above deﬁne an action of Y~(g) on V ⊗W . The corresponding representation is denoted by V ⊗ W . D,s (2) The action of Y~(g) on V ⊗ W is a rational function of s, with poles D,s contained in σ(W ) − σ(V ). 12 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

(3) The identiﬁcation of vector spaces

(V1 ⊗ V2) ⊗ V3 = V1 ⊗ (V2 ⊗ V3) D,s1 D,s2 D,s1+s2 D,s2

intertwines the action of Y~(g). (4) If V =∼ C is the trivial representation of Y~(g), then V ⊗ W = W and W ⊗ V = W (s) D,s D,s (5) The following holds for any s,t ∈ C, V (t) ⊗ W (t) = (V ⊗ W )(t) and V (t) ⊗ W = V ⊗ W D,s D,s D,s D,s+t In particular, V ⊗ W (t) = (V ⊗ W )(t). D,s D,s−t (6) The following holds for any s ∈ C, σ(V ⊗ W ) ⊂ (s + σ(V )) ∪ σ(W ) D,s 3.3. Laurent expansion of the deformed Drinfeld tensor product. Proposition. The Laurent expansion at s = ∞ of the formulae of 3.2 is given by

∆(ti,r)= τs(ti,r) ⊗ 1+1 ⊗ ti,r D,s and + + + ∆(xi,r)= τs(xi,r) ⊗ 1+1 ⊗ xi,r D,s N N + ~ (−1)n+1 ξ ⊗ x+ s−N−1 n i,n i,r+N−n n=0 ! NX≥0 X − − − ∆(xi,r)= τs(xi,r) ⊗ 1+1 ⊗ xi,r D,s N N + ~ (−1)n+1 x− ⊗ ξ s−N−1 n i,r+n i,N−n n=0 ! NX≥0 X Proof. The expansion of ∆(ti,r) follows from ∆(ti(u)) = τs(ti(u)) ⊗ 1+1 ⊗ ti(u). D,s D,s Expanding in u−1 yields

+ + 1 r + ∆(xi,r)= τs(xi,r) ⊗ 1+ v ξi(v − s) ⊗ xi (v) dv D,s ~ IC2 p + m Expanding now ξ (v−s) with respect to s−1 by using (1−x)−p−1 = xm i m≥0 p yields: P −p−1 ξi(v − s)=1+ ~ ξi,p(v − s) Xp≥0 p + q =1+ ~ ξ (−s)−p−1 vqs−q i,p p Xp≥0 Xq≥0 m =1+ ~ s−m−1 (−1)p+1 vqξ p i,p m≥0 p,q≥0 X p+Xq=m THE MEROMORPHIC R–MATRIX OF THE YANGIAN 13

Substituting gives

+ + 1 r + ∆(xi,r)= τs(xi,r) ⊗ 1+ v 1 ⊗ xi (v) dv D,s ~ IC2 m + s−m−1 (−1)p+1 vr+qξ ⊗ x+(v) dv p i,p i m≥0 p,q≥0 IC2 X p+Xq=m which is the claimed result since, for any a ≥ 0, vax+(v) dv = ~x+ . The C2 i i,a − expansion of ∆(xi,r) is obtained in the same way. D,s H

3.4. Deformed Drinfeld coproduct. We now lift the deformed Drinfeld tensor product to an algebra homomorphism −1 ∆ : Y~(g) → (Y~(g) ⊗ Y~(g))[s; s ]]. D,s Theorem. The Laurent expansions of Section 3.3 give rise to an algebra homomorphism −1 ∆ : Y~(g) → (Y~(g) ⊗ Y~(g))[s; s ]] D,s The deformed Drinfeld coproduct ∆ has the following properties. D,s (1) It is compatible with the counit ǫ, that is

ǫ ⊗ 1 ◦ ∆ = 1 and 1 ⊗ ǫ ◦ ∆ = τs D,s D,s

(2) For every x ∈ Y~(g), the following holds in (Y~(g) ⊗ Y~(g)[a])[s; s−1]]

τa ⊗ τa ◦ ∆(x) = ∆ ◦ τa(x) and τa ⊗ 1 ◦ ∆(x) = ∆ (x) D,s D,s D,s D,s+a (3) ∆ is a filtered homomorphism, that is D,s ⊗2 −1 ∆(Fk(Y~(g))) ⊂Fk(Y~(g) [s; s ]]) D,s ⊗2 −1 for each k ≥ 0, where F•(Y~(g) [s; s ]]) is the filtration defined in (2.6).

Proof. Using Theorem 4.1, and the fact that ∆s is an algebra homomorphism, we conclude the same for ∆. Properties (1) and (2) above are easy to verify directly D,s from the deﬁnition. Property (3) follows immediately from the explicit formulas given in Proposition 3.3.

Remark. The coassociativity property of the Drinfeld tensor product ⊗ does not D,s appear to have a natural lift to ∆. The candidate identity D,s ∆ ⊗ 1 ◦ ∆ (x)= 1 ⊗ ∆ ◦ ∆ (x) (3.2) D,s1 D,s2 D,s2 D,s1+s2 holds if x is one of the commuting generators ti,r. However, if x is an arbitrary element of Y~(g), the left–hand side and right–hand side lie, respectively, in ⊗3 −1 −1 ⊗3 −1 −1 Y~(g) [s1; s1 ]][s2; s2 ]] and Y~(g) [s2; s2 ]][s1; s1 ]] and cannot therefore be directly compared. The coassociativity of ⊗ implies, how- D,s ever, that the evaluation of the left– and right–hand sides of (3.2) on a tensor product V1 ⊗ V2 ⊗ V3 of ﬁnite–dimensional representations are, respectively, the expansions at |s2| ≫ |s1| and |s1| ≫ |s2| of the same rational function. 14 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

4. The element R−(s) ∗ 4.1. Set Q+ = i∈I Z≥0 αi ⊂ h . The following is one of the main result of this paper. L Theorem. (1) There is a unique element

− −1 R (s) ∈ (Y~(g) ⊗ Y~(g))[[s ]] which is • strictly lower triangular, that is R−(s)= R−(s) , with β,γ∈Q+ β,γ

− −1 − R (s)β,γ ∈ (Y~(g)−β ⊗ Y~(g)γ )[[s ]] and PR (s)0 =1 ⊗ 1 • of weight zero • such that, for any i ∈ I,

− − R (s) · ∆s (ti,1) = ∆ (ti,1) · R (s) (4.1) D,s

− − + −1 (2) The element R (s) lies in (Y~ (g) ⊗ Y~ (g))[[s ]], and has the following additional properties. • For any i ∈ I,

− ± ± − R (s) · ∆s xi,0 = ∆ xi,0 · R (s) (4.2) D,s • For any a,b ∈ C,

− − τa ⊗ τb R (s) = R (s + a − b)

− ⊗2 −1 • R (s) − 1 ∈F−1(Y~(g) [[s ]]), with semiclassical limit given by ~r R−(s) − 1= ∈ (U(g[z]) ⊗ U(g[w]))[[s−1]] z + s − w In particular, R−(s)=1+ ~r/s + O(s−2). − −1 (3) Let V , V ∈ Rep (Y~(g)), and R (s) ∈ End(V ⊗ V )[[s ]] the corre- 1 2 fd V1,V2 1 2 sponding evaluation of R−(s). Then, • R− (s) is the Taylor expansion at s = ∞ of a rational function. V1,V2 • The following cocycle equation holds for any V1, V2, V3 ∈ Repfd(Y~(g)): R− (s ) · R− (s )= R− (s + s ) · R− (s ) (4.3) V1 ⊗ V2,V3 2 V1,V2 1 V1,V2 ⊗ V3 1 2 V2,V3 2 D,s1 D,s2 Remark. Analogously to Remark 3.4, the cocycle equation (4.3) only holds as an identity of rational functions with values in End(V1 ⊗ V2 ⊗ V3), and does not seem to possess a natural lift to Y~(g). Indeed, the candidate identity

− − − − ∆ ⊗ 1(R (s2)) · R12(s1)= 1 ⊗ ∆ R (s1 + s2) · R23(s2) D,s1 D,s2

⊗3 −1 −1 does not make sense since the left–hand side lies in Y~(g) [s1; s1 ]][[s2 ]], while ⊗3 −1 −1 the right–hand side lies in Y~(g) [s2; s2 ]][[s1 ]]. The rest of the section is devoted to the proof of Theorem 4.1. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 15

4.2. Existence and uniqueness of R−(s). The triangularity and zero–weight assumptions are equivalent to the requirement that R−(s) have the form

− − − −1 R (s)= R (s)β with R (s)β ∈ (Y~(g)−β ⊗ Y~(g)β )[[s ]] (4.4) Q βX∈ + − − and R (s)0 = 1 ⊗ 1. We shall construct R (s)β recursively in β, and prove that the sum over β converges in the s–adic topology. In fact, deﬁne ν : Q+ → Z≥0 by

(1) (k) (1) (k) ν(β) = min k ∈ Z≥0|β = α + ··· + α , α ,...,α ∈ Φ+

n − −ν(β) ⊗2 −1 o with ν(0) = 0. Then, we shall prove that R (s)β ∈ s Y~(g) [[s ]]. Let r be given by (3.1) and, for any h ∈ h, set − + r(h) = ad(h ⊗ 1)(r)= − γ(h) xγ,0 ⊗ xγ,0. γX∈Φ+ By 3.1 and 3.3, 2 ∆(T(h)) = T(h)(a) + sh(1) D,s a=1 X 2 (a) (1) ∆s(T(h)) = T(h) + sh + ~r(h), a=1 X where we use the standard notation: X(1) = X ⊗ 1 and X(2) = 1 ⊗ X. The intertwining equation (4.1) therefore reads as A(h)R−(s) = 0 where, for any h ∈ h,

(1) (2) (1) A(h)= λ(∆(T(h))) − ρ (∆s(T(h))) = ad T(h) + T(h) + sh − ~ρ(r(h)) D,s and λ, ρ denote left and right multiplication, respectively. In components, this reads: − − − + (T (h) − sβ(h)) R (s)β = −~ α(h)R (s)β−α xα,0 ⊗ xα,0 αX∈Φ+ (1) (2) − where T (h) = ad(T(h) + T(h) ) and R (s)γ = 0 if γ∈ / Q+. If h ∈ h is such that β(h) 6= 0 for any nonzero β ∈ Q+, this yields ~ T (h) −1 R−(s) = 1 − α(h)R−(s) x− ⊗ x+ (4.5) β sβ(h) sβ(h) β−α α,0 α,0 αX∈Φ+ T (h)k = ~ α(h)R−(s) x− ⊗ x+ (4.6) (sβ(h))k+1 β−α α,0 α,0 Xk≥0 αX∈Φ+ − − This shows that R (s) is uniquely determined by R (s)0, and that it lies in − + −1 − − + (Y~ (g) ⊗ Y~ (g))[[s ]] if R (s)0 does, since Y~ (g), Y~ (g) are invariant under adT(h). Fix now h ∈ h \ Ker β. The above equations can be used to define β∈(Q+\{0}) elements R−(s) recursively on the height of β, starting from R−(s) = 1, which β S 0 lie in s−ν(β)Y~(g)⊗2[[s−1]]. The corresponding sum R−(s) = R−(s) is β∈Q+ β therefore well–defined, and satisfies A(h)R−(s) = 0. We claim that it also satisfies P A(h′)R−(s) =0 for any h′ ∈ h. Note that ′ ′ ′ [A(h), A(h )] = λ[∆(T(h)), ∆(T(h ))] − ρ[∆s(T(h)), ∆s(T(h ))] D,s D,s 16 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

′ which vanishes because ∆s is an algebra homomorphism, and ∆(T(h)), ∆(T(h )) ∈ D,s D,s 0 ⊗2 ′ − Y~ (g) . Thus, A(h )R (s) satisﬁes

A(h)A(h′)R−(s)= A(h′)A(h)R−(s)=0

Since A(h′)R−(s) is also triangular, with

′ − ′ (1) ′ (2) ′(1) − A(h )R (s) 0 = ad T(h ) + T(h ) + sh R (s)0 =0 it follows by uniqueness that A(h′)R−(s) = 0 as claimed.

− − 4.3. Translation invariance. The identity τa ⊗τb (R (s)) = R (s+a−b) follows by uniqueness, since both sides are strictly lower triangular, intertwine

τa ⊗ τb ◦ ∆s(ti,m)= τa−b ⊗ 1 ◦ ∆s(τb(ti,m)) = ∆s+a−b(τb(ti,m)) and τa ⊗ τb ◦ ∆(ti,m) = ∆ (τb(ti,m)) for i ∈ I and m = 0, 1, and the span of D,s D,s+a−b {ti,m}i∈I,m=0,1 is invariant under τb.

4.4. Semiclassical limit. Since T (h) is a ﬁltered operator of degree 1, the re- − ⊗2 −1 cursion (4.6) shows that R (s)β ∈ F−ν(β)(Y~(g) [[s ]]) for any β ∈ Q+. In − ⊗2 −1 ⊗2 −1 particular, R (s) − 1 ∈F−1(Y~(g) [[s ]]) and, mod F−2(Y~(g) [[s ]]),

~ T (h)k R−(s) − 1= R−(s) = x− ⊗ x+ β s (sα(h))k α,0 α,0 β:νX(β)=1 αX∈Φ+ kX≥0 ⊗2 −1 ⊗2 −1 whose image in F−1(Y~(g) [[s ]])/F−2(Y~(g) [[s ]]) is ~r/(z + s − w).

4.5. Rationality. The argument given in 4.2 can be carried out in End(V1 ⊗ V2) rather than Y~(g)⊗2, and shows the existence and uniqueness of an element

R− (s)= R− (s) ∈ End(V ⊗ V )[[s−1]] V1,V2 V1,V2 β 1 2 Q βX∈ + with

[h ⊗ 1, R− (s) ]= −β(h) R− (s) [1 ⊗ h, R− (s) ]= β(h) R− (s) V1,V2 β V1 ,V2 β V1,V2 β V1,V2 β for any h ∈ h, R− (s) = 1, and V1,V2 0

− − RV ,V (s) · ∆s (ti,1) = ∆ (ti,1) · RV ,V (s) 1 2 D,s 1 2

The recursive construction of R− (s) given by (4.5) shows that each R− (s) V1,V2 V1,V2 β is a rational function of s, regular at s = ∞. Since only ﬁnitely many R− (s) V1,V2 β are non–zero by ﬁnite–dimensionality, R− (s) is therefore rational. It follows by V1,V2 uniqueness that the Taylor expansion of R− (s) is the evaluation of R−(s) on V1,V2 V1 ⊗ V2. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 17

4.6. Cocycle equation. Let V1, V2, V3 ∈ Repfd(Y~(g)). We obtain below an alternative version of the cocycle equation, namely R− (s ) · R− (s )= R− (s ) · R− (s + s ) (4.7) V1,V2 1 V1 ⊗V2,V3 2 V2,V3 2 V1,V2 ⊗V3 1 2 s1 s2 Note that (4.7) and (4.3) are equivalent, provided the intertwining equation (4.2) is established. The latter will be proved in 4.8, by relying in part on (4.7).6 The intertwining property of R−(s) implies that R− (s ) · R− (s ) · π (t ) V1,V2 1 V1 ⊗V2,V3 2 (V1 ⊗V2)⊗V3 i,1 s s s1 1 2 = R− (s ) · π (t ) · R− (s ) V1,V2 1 (V1 ⊗V2) ⊗ V3 i,1 V1 ⊗ V2,V3 2 s D,s 1 2 D,s1 = π (t ) · R− (s ) · R− (s ), (V1 ⊗ V2) ⊗ V3 i,1 V1,V2 1 V1 ⊗V2,V3 2 D,s1 D,s2 s1 where the second equality stems from the fact that

π(V1 ⊗V2) ⊗ V3 (ti,1)= π(V1 ⊗V2)(τs2 (ti,1)) + πV3 (ti,1) s1 D,s2 s1 Similarly, R− (s ) · R− (s + s ) · π (t ) V2,V3 2 V1,V2 ⊗V3 1 2 V1 ⊗ (V ⊗V3) i,1 s s2 s1+s2 2 = R− (s ) · π (t ) · R− (s + s ) V2,V3 2 V1 ⊗ (V ⊗V3) i,1 V1,V2 ⊗V3 1 2 s D,s1+s2 2 s2 = π (t ) · R− (s ) · R− (s + s ) V1 ⊗ (V ⊗ V3) i,1 V2,V3 2 V1,V2 ⊗V3 1 2 D,s1+s2 D,s2 s2 Since ⊗, ⊗ are coassociative, both sides of the cocycle equation (4.7) are therefore s D,s solutions of

X · ∆s1 ⊗ 1 ◦ ∆s2 (ti,1) = ∆ ⊗ 1 ◦ ∆ (ti,1) · X (4.8) D,s1 D,s2 By 3.1 and 3.3, 3 (a) (1) (2) ∆ ⊗ 1 ◦ ∆ (T(h)) = T(h) + (s1 + s2) h + s2 h D,s1 D,s2 a=1 X 3 (a) (1) (2) ∆s1 ⊗ 1 ◦ ∆s2 (T(h)) = T(h) + (s1 + s2) h + s2 h + ~ (r(h)13 + r(h)23) a=1 X The conclusion now follows by noting that, analogously to 4.2, (4.8) admits at most one rational solution X(s1,s2) with values in End(V1 ⊗ V2 ⊗ V3), provided it is strictly lower triangular, that is of the form X = X , where β,γ∈Q+ β,γ

Xβ,γ ∈ End(V1)[−β] ⊗ End(V2)[β − γP] ⊗ End(V3)[γ] and X0,0 = 1.

6The cocycle equation (4.3) arises from the deﬁnition of a tensor structure on a functor F : C→D as a system of natural isomorphisms JU,V : F (U) ⊗D F (V ) → F (U ⊗C V ), by interpreting − R (s) as a tensor structure on the identity functor (Repfd(Y~(g)), ⊗ ) → (Repfd(Y~(g)), ⊗). D,s s Similarly, (4.7) arises by adopting the opposite convention of a tensor structure as a system of − isomorphisms KU,V : F (U ⊗C V ) → F (U) ⊗D F (V ), and interpreting R (s) as a tensor structure on (Repfd(Y~(g)), ⊗) → (Repfd(Y~(g)), ⊗ ). s D,s 18 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

4.7. Rank 1 reduction. Consider the intertwining identity

− ± ± − R (s) · ∆s(xi,0) = ∆(xi,0) · R (s) (4.9) D,s

We claim that it holds for any g and i ∈ I if, and only if it holds for g = sl2. Set (i) ∗ ⊥ ∗ (i) (i) Q = Q/Zαi ⊂ h /Cαi = (αi ) , and let Q+ be the image of Q+ in Q . Both sides of (4.9) are lower triangular and of weight zero with respect to the adjoint ⊥ (i) (i) −1 Q(i) action of αi ⊂ h i.e., lie in Q(i) (Y~(g)−β ⊗ Y~(g)β )[[s ]] where, for γ ∈ , β∈ +

(i) L ⊥ Y~(g)γ = {x ∈ Y~(g)| [h, x]= γ(h)x, h ∈ αi }

⊥ Since both intertwine ∆s(T(h)) and ∆(T(h)) for h ∈ αi , it follows by uniqueness D,s (i) (i) −1 that they are equal if, and only if, their projections on (Y~(g)0 ⊗ Y~(g)0 )[[s ]] coincide. (i) ~ ~ ~ Let πi ∈ End(Y (g)) be the projection onto Y (g)0 = m∈Z Y (g)mαi . Then, − ± − ± πi ⊗ πi(R (s) · ∆s(xi,0)) = πi ⊗ πi(R (s))L· ∆s(xi,0)

± − ± − πi ⊗ πi(∆(xi,0) · R (s)) = ∆(xi,0) · πi ⊗ πi(R (s)) D,s D,s

−1/2 ± ~ ~ ± Let ϕi : Ydi (sl2) → Y (g) be the algebra homomorphism given by xr 7→ di xi,r −1 and ξr 7→ di ξi,r. Then, ± ⊗2 ± ± ⊗2 ± ∆s(xi,0)= di · ϕi ◦ ∆s(x0 ) and ∆(xi,0)= di · ϕi ◦ ∆(x0 ). D,s D,s p p We claim that π ⊗ π (R−(s)) = ϕ ⊗ ϕ (R− (s)) so that, by the foregoing, i i i i (sl2) (4.9) holds for any g and i ∈ I if, and only if it holds for g = sl2. The claim follows by uniqueness, since both π ⊗ π (R−(s)) and ϕ ⊗ ϕ (R− (s)) are strictly lower i i i i (sl2) triangular and of weight zero with respect to the action of ξi,0, and both intertwine ⊗2 −1 ⊗2 ⊗2 −1 −1 ⊗2 ϕi ◦∆s(ti)= di πi ◦∆s(ti,1) and ϕi ◦∆(ti)= di ∆(ti,1)= di πi ◦∆(ti,1). D,s D,s D,s

4.8. Rank 1 intertwining relations. Assume g = sl2, and consider the identity − − − − R (s) · ∆s(x0 ) = ∆(x0 ) · R (s) (4.10) D,s The latter can be proved by a lengthy, direct calculation7. We give below an alternative proof, which relies on the cocycle identity (4.3) satisfied by R−(s) to reduce it to the case when R−(s) is acting on C2 ⊗ V , where V is an arbitrary finite–dimensional representation. By Appendix A, it is sufficient to prove that (4.10) holds on the tensor product V1 ⊗ V2, where V1 (resp. V2) is chosen from a collection V1 (resp. V2) of finite–dimensional representations of Y~(g) which is stable under tensor product, contains the trivial representation, and a representation whose restriction to g is faithful. We choose V2 to consist of all finite–dimensional representations of Y~(g), 2 2 while V1 consists of arbitrary tensor products C (a1) ⊗···⊗ C (am) of evaluation representations.

7The calculation is carried out in §5 of the earlier version of this paper, arXiv:1907.03525 v1. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 19

For any V1, V2, V3 ∈ Repfd(Y~(g)), the cocycle identity (4.7) implies that (4.10) holds on (V1(s1) ⊗ V2) ⊗ V3 if it holds on V1 ⊗ V2, V2 ⊗ V3, and V1 ⊗ (V2(s2) ⊗ V3). Indeed, R− (s ) · π (x−) V1 ⊗V2,V3 2 (V1 ⊗V2)⊗V3 0 s s s1 1 2 = R− (s )−1 · R− (s ) · R− (s + s ) · π (x−) V1,V2 1 V2,V3 2 V1,V2 ⊗V3 1 2 V1 ⊗ (V2 ⊗V3) 0 s s2 s1+s2 2 = R− (s )−1 · R− (s ) · π (x−) · R− (s + s ) V1,V2 1 V2,V3 2 V1 ⊗ (V2 ⊗V3) 0 V1,V2 ⊗V3 1 2 s D,s1+s2 2 s2 = R− (s )−1 · π (x−) · R− (s ) · R− (s + s ) V1,V2 1 V1 ⊗ (V2 ⊗ V3) 0 V2,V3 2 V1,V2 ⊗V3 1 2 D,s1+s2 D,s2 s2 = R− (s )−1 · π (x−) · R− (s ) · R− (s + s ) V1,V2 1 (V1 ⊗ V2) ⊗ V3 0 V2,V3 2 V1,V2 ⊗V3 1 2 D,s1 D,s2 s2 = π (x−) · R− (s )−1 · R− (s ) · R− (s + s ) (V1 ⊗V2) ⊗ V3 0 V1,V2 1 V2,V3 2 V1,V2 ⊗V3 1 2 s 1 D,s2 s2 = π (x−) · R− (s ) (V1 ⊗V2) ⊗ V3 0 V1 ⊗V2,V3 2 s 1 D,s2 s1 It is therefore sufficient to check (4.10) on C2(a)⊗V , where V is an arbitrary finite– dimensional representation of Y~(g), and a ∈ C. Moreover, by using the translation invariance of R−(s), ⊗, and ⊗ , it is sufficient to consider the case a = 0. s D,s ± Let now x , ξ be the standard nilpotent and semisimple generators of sl2 acting 2 2 on C . Then, the following define an action of Y~(sl2) on C : x± ξ x±(u)= ~ and ξ(u)=1+ ~ u u − − − By 3.1 and 3.2, πC2⊗V (x0 )= x ⊗ 1+1 ⊗ x0 , and s − − x − − − πC2 ⊗ V (x0 )= ⊗ ξi(v) dv +1 ⊗ x0 = x ⊗ ξ(s)+1 ⊗ x0 D,s v − s IC1 2 where C1 encloses σ(C )+ s = {s} and none of the points in σ(V ). On the other hand, since x−(u)x−(v)=0on C2, 5.4 and Theorem 5.5 yield x− x− R− (s)=1 − ⊗ x+(u) du =1+ ⊗ x+(u) du =1+ x− ⊗ x+(s) C2,V u − s u − s IC2 IC1 2 where C2 encloses σ(V ) and none of the points in σ(C )+ s = s, and the second equality follows because the residue of the integrand at infinity is 0. Therefore, − − − − − + − RC2,V (s) · πC2⊗V (x0 )= x ⊗ 1+1 ⊗ x0 + x ⊗ x (s)x0 s − − − − − − + πC2 ⊗ V (x0 ) · RC2,V (s)= x ⊗ ξ(s)+1 ⊗ x0 + x ⊗ x0 x (s) D,s so that − − − − − + − RC2,V (s) · πC2⊗V (x0 ) − πC2 ⊗ V (x0 ) · RC2,V (s)= x ⊗ 1 + [x (s), x0 ] − ξ(s) s D,s + − which is equal to zero since [x (s), x0 ]= ξ(s) − 1. − + + − The identity R (s) · ∆s(x0 ) = ∆(x0 ) · R (s) is proved in a similar way, by D,s taking V1 to consist of all finite–dimensional representations, V2 of tensor products 2 2 C (a1) ⊗···⊗ C (am), and using the cocycle identity to reduce this to a check on 2 V ⊗ C , for an arbitrary V ∈ Repfd(Y~(g)). 20 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

− 5. The element R (s) for sl2 − In this section, we give an explicit formula for the element R (s) when g = sl2.

5.1. It will be convenient to consider the generating series

− − n −1 R (s,z)= Rnα(s)z ∈ (Y~(sl2) ⊗ Y~(sl2))[z][[s ]] nX≥0 where α is the positive root of g. Let G = PSL2(C) be the complex Lie group of adjoint type corresponding to g, and H ⊂ G its maximal torus with Lie algebra h. We identify H with C× via the character corresponding to −α. In particular, R−(s,z) = Ad(z(1))R−(s). Moreover, R−(s,z) satisﬁes

− z z − R (s,z) · ∆s (t1) = ∆ (t1) · R (s,z) D,s

z (1) z (1) where ∆s = Ad(z ) ◦ ∆s and ∆ = Ad(z ) ◦ ∆, so that D,s D,s

2 2 z (a) (1) z (a) (1) ∆ (t1)= t1 + sh and ∆s(t1)= t1 + sh + ~zr D,s a=1 a=1 X X 5.2. The intertwining equation for R−(s,z) may be written as the following ODE, together with the initial condition R−(s, 0)=1 ⊗ 1 1 sz∂ − ad(t ⊗ 1+1 ⊗ t ) R−(s,z)= R−(s,z) · ~zr (5.1) z 2 1 1 − + where r = x0 ⊗ x0 . − −1 − Lemma. Set ω(s,z)= R (s,z) · z∂zR (s,z). Then, ω(s,z) satisﬁes

(sz∂z − T + ~z ad(r)) ω(s,z)= ~rz (5.2) 1 where T = ad(t ⊗ 1+1 ⊗ t ). 2 1 1

Proof. Since (sz∂z − T ) is a derivation, we have − −1 − (sz∂z − T ) ω(s,z)= −~zrω(s,z)+ R (s,z) z∂z(R (s,z)~zr) = −~z[r,ω(s,z)] + ~zr

5.3. Formula for ω(s,z). Given a vector space H, let

− du : H[u; u−1]] → H I be the formal residue, given by taking the coeﬃcient of u−1. Proposition. The series ω(s,z) is given by (−1)k ω(s,z)= zk x−(u − s)k ⊗ x+(u)k du (5.3) k~ kX≥1 I THE MEROMORPHIC R–MATRIX OF THE YANGIAN 21

k Proof. Write ω(s,z)= k≥1 ωk(s)z . For k =1, (5.2) yields P −1 −m−1 m ω1(s) = (s − T ) ~r = ~ s T r mX≥0 m m = ~ s−m−1 (−1)n x− ⊗ x+ n n m−n n=0 mX≥0 X ± ± where the last equality follows from [t1/2, xk ]= ±xk+1. On the other hand,

− − −n−1 x (u − s)= ~ xn (u − s) nX≥0 n + a = −~ (−1)nx−s−n−1 uas−a n n nX≥0 Xa≥0 m m = −~ s−m−1 (−1)n x−um−n n n n=0 mX≥0 X which shows that ω1 is given by (5.3). (−1)k Let now k ≥ 2, and set I (v,s)= x−(v − s)k ⊗ x+(v)k. We claim that k k~

(sk − T )Ik(v,s)= −[~r, Ik−1(v,s)] which proves in particular that (5.3) satisﬁes (5.2). We shall need the commu- ± ± ± 2 tation relation [x0 , x (u)] = ∓x (u) . The latter follows from relation (Y4) of Proposition 2.3, namely

± ± ± ± ± ± ± ± (u − v ∓ ~)x (u)x (v) − (u − v ± ~)x (u)x (v)= ~([x0 , x (v)] − [x (u), x0 ]),

± ± k ± k+1 by taking u = v. Thus, for every k ≥ 1, ad(x0 ) · x (u) = ∓kx (u) . t Using this, and [ 1 , x±(u)] = ±(ux±(u) − ~x±), yields 2 0

t k 1 , x±(u)k = ± x±(u)j−1(ux±(u) − ~x±)x±(u)k−j 2 0 j=1 X k ± k ± j−1 ± ± k−j = ±kux (u) ∓ ~ x (u) x0 x (u) j=1 X k(k − 1) = ±kux±(u)k ∓ ~rx±(u)k−1x± + ~ x±(u)k 0 2

Thus,

(sr − T ) x−(v − s)k ⊗ x+(v)k = −~k(k − 1)x−(v − s)k ⊗ x+(v)k − k−1 − + k − k + k−1 + − ~kx (v − s) x0 ⊗ x (v) + ~kx (v − s) ⊗ x (v) x0 . 22 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Together with [A ⊗ B, C ⊗ D] = [A, C] ⊗ [B,D] + [A, C] ⊗ DB + CA ⊗ [B,D], this yields − + − k + k − − k + + k − − k + k + [x0 ⊗ x0 , x (u) ⊗ x (u) ] = [x0 , x (u) ] ⊗ [x0 , x (u) ] + [x0 , x (u) ] ⊗ x (u) x0 − k − + + k + x (u) x0 ⊗ [x0 , x (u) ] 2 − k+1 + k+1 − k+1 + k + = −k x (u) ⊗ x (u) + kx (u) ⊗ x (u) x0 − k − + k+1 − kx (u) x0 ⊗ x (u) as claimed.

5.4. Formula for ωV1,V2 (s,z). If V1, V2 are ﬁnite–dimensional representations of

Y~(g), Theorem 4.1 (3) implies that ωV1,V2 (s,z) = πV1 ⊗ πV2 (ω(s,z)) is a rational function of s. It follows from the lemma below that (−1)kzk ω (s,z)= x−(u − s)k ⊗ x+(u)k du V1,V2 ~ k C2 kX≥1 I where C2 encloses σ(V2) and none of the points in σ(V1)+ s. Note that in this case ± the sum over k is ﬁnite since x (u) are nilpotent on V1, V2. Lemma. Let A be a ﬁnite–dimensional dimensional algebra over C, and f,g : C → A rational functions which are regular at ∞. Consider the integral

I(s)= f(u − s)g(u) du IC2 where s ∈ C, and C2 is a contour enclosing all poles of g(u) and none of those of f(u − s). Then, the Taylor expansion I(s) of I(s) at s = ∞ is equal to

I(s)= fb(u − s)g(u) du I where − du is the formal residueb deﬁnedb in 5.3b , f, g are the Taylor series of f,g I at ∞, and f(u − s) is expanded in A[u][[s−1]]. b b −1 Proof. I(sb) is equal to f(u − s)g(u)du. Since f(u − s) ∈ A[u][[s ]], it suﬃces IC2 to proveb that p(u)g(u)dub= p(u)g(u)du for anyb p ∈ A[u], which follows by C2 deformation ofI contour. I b − − − 5.5. Formula for R (s). Integrating z∂zR (s,z)= R (s,z) · ω(s,z) and setting z = 1 yields the following corollary of Proposition 5.3. − ⊗2 −1 Theorem. The element R (s) ∈ Y~(sl2) [[s ]] is given by

− ⊗2 1 R (s)=1 + ω(s)k1 ··· ω(s)kr k1(k1 + k2) ··· (k1 + ··· + kr) n≥1 k1,...,kr∈Z≥1 X k1+···X+kr =n where (−1)k ω(s) = x−(u − s)k ⊗ x+(u)k du k k~ I Remark. It is an interesting problem to give an explicit formula for R−(s) for g ≇ sl2. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 23

6. The universal and the meromorphic abelian R–matrices of Y~(g) In this section, we review the construction of the meromorphic abelian R–matrix of Y~(g) given in [12]. We then show that it gives rise to rational intertwiners V1(s) ⊗ V2 → V2 ⊗ V1(s) for any ﬁnite–dimensional representations V1, V2. We D,0 D,0 also prove that these cannot be chosen to be both natural and compatible with the Drinfeld tensor product. Finally, we lift the meromorphic abelian R–matrix to obtain a universal abelian R–matrix for the deformed Drinfeld coproduct.

6.1. The endomorphism AV1,V2 (s)[12]. Let V ∈ Repfd(Y~(g)). For any i ∈ I, ±1 let σV (ξi) be the set of poles of ξi(u) acting on V , and

XV (ξi)= [0,a] a∈σ[V (ξi) the union of straight line segments joining 0 to points in σV (ξi). Then, the gen- −r−1 erating series ti(v) = ~ r≥0 ti,rv introduced in 2.4 converges to a holomorphic function t (v): C \ X (ξ ) → EndC(V ), which is uniquely determined by i P V i exp(ti(v)) = ξi(v) and ti(∞) = 0 [12, §5.4].

Let V1, V2 ∈ Repfd(Y~(g)), s ∈ C, and deﬁne AV1,V2 (s) ∈ EndC(V1 ⊗ V2) by

dt (v) (ℓ + r)~ A (s) = exp − c(r) i ⊗ t v + s + dv V1,V2  ij dv j 2  i,j∈I IC1  Xr∈Z    where

• C1 is a contour enclosing σV1 (ξi). ∨ ∨ (θ,θ) • ℓ = mh , with h the dual Coxeter number of g and m = 2 for θ ∈ Φ+ the highest root. (r) • The non–negative integers cij are the entries of the following matrix [12, Appendix A]8

(r) r −1 cij (q)= cij q = [ℓ]q · ([diaij ]q) Z ! Xr∈ • s is large enough so that tj (v + s + ~(ℓ + r)/2) is an analytic function of v (r) within C1, for every j ∈ I and r ∈ Z such that cij 6= 0 for some i ∈ I.

Then, AV1,V2 (s) is a rational function of s, regular at ∞, with an expansion of the ℓ~ form 1 − Ω + O(s−3), where Ω ∈ h⊗h is the Cartan part of the Casimir tensor s2 h h ′ ′ [12, Thm. 5.5]. Moreover, [AV1,V2 (s), AV1,V2 (s )] = 0 for any s,s ∈ C.

8 −1 It was proved in [12, Appendix A] that cij (q) ∈ Z≥0[q,q ]. It is clear from the deﬁnition −1 that cij (q)= cji(q)= cij (q ), and the matrix identity can be expanded as I X cik(q)[dkakj ]q = δij [ℓ]q ∀i,j ∈ (6.1) k∈I 24 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

6.2. The meromorphic abelian R–matrix of Y~(g)[12]. Consider the additive diﬀerence equation determined by AV1,V2 (s) 0 ~ 0 RV1,V2 (s + ℓ )= AV1,V2 (s) · RV1,V2 (s) (6.2) It admits two meromorphic solutions R0,↑/↓ (s), which are uniquely determined by V1,V2 the requirement that R0,↑ (s) (resp. R0,↓ (s)) is holomorphic and invertible for V1,V2 V1,V2 Re(s/ℓ~) ≫ 0 (resp. Re(s/ℓ~) ≪ 0), and possesses an asymptotic expansion of the form 1 + O(s−1) as s → ∞ in any halfplane of the form Re(s/ℓ~) > m (resp. Re(s/ℓ~) 0), where Z (resp. P) is the set of poles of A(s)−1 (resp. A(s)). Then, the following holds [12, Thm 5.9].

Theorem. Fix ε ∈ {↑, ↓}. Then, the following holds for any V1, V2, V3 ∈ Repfd(Y~(g)) (1) The map (1 2) ◦ R0,ε (s): V (s) ⊗ V → V ⊗ V (s) V1,V2 1 2 2 1 D,0 D,0

is a morphism of Y~(g)–modules, which is natural in V1 and V2. (2) The following cabling identities hold: R0,ε (s )= R0,ε (s + s ) · R0,ε (s ) V1 ⊗ V2,V3 2 V1,V3 1 2 V2,V3 2 D,s1 R0,ε (s + s )= R0,ε (s + s ) · R0,ε (s ) V1,V2 ⊗ V3 1 2 V1,V3 1 2 V1,V2 1 D,s2 (3) For any a,b ∈ C, R0,ε (s)= R0,ε (s + a − b) V1(a),V2(b) V1,V2 (4) The following unitary condition holds: (1 2) ◦ R0,↑ (−s) ◦ (1 2)−1 = R0,↓ (s)−1 V1,V2 V2,V1 (5) R0,↑/↓ (s) have the same asymptotic expansion, as s → ∞ in any halfplane V1,V2 of the form ± Re(s/ℓ~) >m, which is of the form R0,↑/↓ (s) ∼ 1+ ~Ω s−1 + O(s−2) V1,V2 h

6.3. Existence of a rational intertwiner. Let V1, V2 ∈ Repfd(Y~(g)).

R0 C C Theorem. There is a rational map V1,V2 : → Aut (V1 ⊗ V2), which is nor- R0 malised by V1,V2 (∞)= 1 and such that R0 (1 2) ◦ V1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s) D,0 D,0 intertwines the action of Y~(g). In particular, V1(s) ⊗ V2 and V2 ⊗ V1(s) are D,0 D,0 isomorphic as Y~(g)–modules for all but ﬁnitely many values of s. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 25

R0 Proof. The map V1,V2 (s) will be obtained from the meromorphic intertwiners R0,↑/↓ (s) as follows. Consider the diﬀerence equation (6.2) satisﬁed by R0,↑/↓ (s). V1,V2 V1,V2 Its monodromy is given by

η0 (s)= R0,↑ (s)−1 · R0,↓ (s) V1,V2 V1,V2 V1,V2 0 ~ By construction, ηV1,V2 is an ℓ –periodic function of s, and in fact a rational 2πιs function of z = exp ℓ~ which takes the value 1 at z =0, ∞ [11, §4.8]. Moreover, 0 ~ g by Theorem 6.2 (1), ηV1,V2 (s) commutes with the action of Y ( ) on V1 ⊗ V2. In D,s 0 0 fact, the periodicity of ηV1,V2 implies that ηV1,V2 (s) commutes with the action of ′ Y~(g) on V1 ⊗ V2 for any s of the form s+ℓ~m, m ∈ Z. Since that action is rational D,s′ ′ 0 C in s , it follows that ηV1,V2 takes values in the subalgebra Z ⊂ End (V1 ⊗ V2) which ′ consists of elements commuting with the action of Y~(g) on V1 ⊗ V2 for any s ∈ C. D,s′ Consider now the following factorisation problem. Find two meromorphic functions X0,↑/↓ (s): C → Z such that V1,V2 (1) X0,↑/↓ (s) is holomorphic and invertible for Re(±s/~) ≫ 0. V1,V2 (2) X0,↑/↓ (s) possesses an asymptotic expansion of the form 1 + O(s−1), valid V1,V2 in any half–plane of the form Re(s/ℓ~) ≷ m. (3) η0 (s)= X0,↑ (s)−1 · X0,↓ (s). V1,V2 V1,V2 V1,V2 0 0 ′ ′ Since [ηV1,V2 (s), ηV1,V2 (s )] = 0 for any s,s , such a factorisation exists, and can be obtained explicitly, once a choice of representatives of poles of η0(s) modulo translations by ℓ~Z is made [11, §4.14]. We remark that the consistency equation, required upon such a choice in [11, §4.14], is vacuous in our case, since AV1,V2 (s)= 1+ O(s−2). 0 C Summarising, ηV1,V2 (s) admits two factorisations in End (V1 ⊗ V2), namely

X0,↑ (s)−1 · X0,↓ (s)= η0 (s)= R0,↑ (s)−1 · R0,↓ (s) V1,V2 V1,V2 V1,V2 V1,V2 V1,V2 Set now

R0 (s)= R0,↑ (s) · X0,↑ (s)−1 = R0,↓ (s) · X0,↓ (s)−1 V1,V2 V1,V2 V1,V2 V1,V2 V1,V2

R0 ~ g Then, (1 2) ◦ V1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s) intertwines the action of Y ( ), D,0 D,0 R0 § and V1,V2 is a rational function of s equal to 1 at s = ∞ [11, 4.11].

6.4. Non–existence of rational commutativity constraints. Theorem 6.3 raises R0 the question of whether the rational factor V1,V2 (s) may be chosen consistently for any pair of representations V1, V2 so as to satisfy the cabling identities of Theorem 6.2. The following shows that not to be the case.

R0 C C Theorem. There is no function V1,V2 : → Aut (V1 ⊗ V2) which is rational, deﬁned for any V1, V2 ∈ Repfd(Y~(g)), and such that the following conditions hold. 0 R ~ g (1) (1 2) ◦ V1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s) is Y ( )–linear, and natural in D,0 D,0 V1 and V2. 26 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

(2) For any V1, V2, V3 ∈ Repfd(Y~(g)) R0 R0 R0 V1 ⊗ V2,V3 (s2)= V1,V3 (s1 + s2) · V2,V3 (s2) (6.3) D,s1 R0 R0 R0 V1,V2 ⊗ V3 (s1 + s2)= V1,V3 (s1 + s2) · V1,V2 (s2) (6.4) D,s2

Proof. Note first if V1 is the trivial one–dimensional representation 1 of Y~(g), the R0 first cabling identity (6.3) and part (4) of Theorem 3.2 imply that 1,V3 (s)=IdV3 . Setting V = 1 in (6.3) then yields R0 (s) = R0 (s + a). Similarly, upon 2 V1(a),V3 V1,V3 setting V2 = 1 first, and then V3 = 1, the second cabling identity (6.4) implies that 0 0 0 R 1(s)=Id , and that R (s)= R (s − b). V1, V V1,V2(b) V1,V2 We now take g = sl2 and proceed by contradiction, assuming the existence of a R0 rational V1,V2 (s) with the stated properties. We will use the following facts about 2 Y~(sl2). There is a two–dimensional representation C of Y~(sl2) (see Section 4.8. Explicitly, in a fixed basis v+, v−, the action of Y~(sl2) is given by the following 2 × 2 matrices. ~ 1 0 ~ 0 1 ξ(u)= 1 + and x+(u)= = x−(u)T u 0 −1 u 0 0 2 2 Further, there is a Y~(sl2)–linear map 1 → C ⊗ C given by 1 7→ v+ ⊗ v−. D,~ R0 2 Given V ∈ Repfd(Y~(g)), we consider C2,V (s) ∈ EndC(C ⊗V )asa2×2–matrix with entries in End(V ) given by

0 α(s) β(s) R 2 (s)= C ,V γ(s) δ(s) R0 The intertwining property of C2,V (s) reads 0 21 0 RC2,V (s) · πC2(s) ⊗ V (x) = (πV ⊗ C2(s)(x)) · RC2,V (s) (6.5) D,0 D,0 for any x ∈ Y~(g). For x = ξ0,

21 1+ ξ0 0 πC2 (ξ ) = (π C2 (ξ )) = ξ ⊗ 1+1 ⊗ ξ = (s) ⊗ V 0 V ⊗ (s) 0 0 0 0 −1+ ξ D,0 D,0 0 the intertwining relation (6.5) yields

[ξ0, α] = 0 [ξ0,δ] = 0 [ξ0,β]= −2β [ξ0,γ]=2γ

For x = t1, πV1 ⊗ V2 (t1)= πV1 (t1)+ πV2 (t1), so that D,0

21 ~ 2 πC2(s) ⊗ V (t1) = (πV ⊗ C2(s)(t1)) = (t1+sξ0)⊗1+1⊗t1 = (ξ1− ξ0 +sξ0)⊗1+1⊗t1 D,0 D,0 2 2 Since ξ1 acts by 0 on C , πC2(s) ⊗ V (t1) acts by the matrix D,0

−~/2+ s + t1 0 0 ~/2 − s + t 1 The relation (6.5) then implies that [t1, α] = 0, [t1,δ] = 0, [t1,β] = −2sβ and [t1,γ]=2sγ. The last two relations imply in turn that β = γ = 0 since ad(t1) ± s is invertible on End(V ) for all but ﬁnitely many values of s. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 27

− − For x = x0 , 3.2 implies that the −+ coefficient of πV ⊗ C2(s)(x0 ) is 1V , while that D,0 − ξ(v) 2 of πC2 (x ) is equal to = ξ(s), where C1 encloses σ(C (s)) = {s} and (s) ⊗ V 0 C1 v−s D,0 none of the points in σ(V ). TakingH −+ coefficients in (6.5) then yields the relation α(s)= δ(s) · ξ(s). Similarly, taking the +− coefficients in the intertwining relation + (6.5) with x = x0 yields α = ξ · δ. Summarising, (6.5) implies that

0 α(s) 0 R 2 (s)= C ,V 0 δ(s) where α, δ ∈ End(V )(s) commute with ξ0,t1, and satisfy α = δ · ξ = ξ · δ. We now use the cabling relation (6.3) 0 0 0 RC2 C2 (s)= RC2 (s + ~) · RC2 (s) (1) ⊗ (2),V (1),V (2),V D,~ in EndC(C2 ⊗ C2 ⊗ V ), where the subscripts are added to emphasize the order of the tensors. The right–hand side applied to v+ ⊗ v− ⊗ v yields α(s + ~) · δ(s) v. On the other hand, by naturality, 0 0 RC2 C2 (s) v+ ⊗ v− ⊗ v = RC,V (s)1 ⊗ v = v (1) ⊗ (2),V D,~ Combining these relations yields α(s + ~) · α(s) = ξ(s). Taking now V = C2, this equation implies that the coefficient a(s) of v+ in α(s)v+ satisfies the additive difference equation a(s +2~) c(s + ~) s(s +2~) = = a(s) c(s) (s + ~)2 where c(s) = (s + ~)/s is the matrix coefficient of ξ(s) corresponding to v+. This equation has a unique solution ϕ(s) which is holomorphic and non–zero for Re(s/2~) ≫ 0 and is asymptotic to 1 + O(s−1) in that domain (see, e.g., [11, §4]). Clearly, Γ s · Γ s+2~ ϕ(s)= 2~ 2~ s+~ 2 Γ 2~ which is not a rational function.

6.5. Taylor expansion of AV1,V2 (s). Let V1, V2 ∈ Repfd(Y~(g)). We determine in 6.6 the asymptotic expansion of R0,↑/↓ (s) as s → ∞. As a preliminary step, we V1,V2 compute the Taylor series of the endomorphism AV1,V2 (s) introduced in 6.2. Let dt (v) (ℓ + r)~ L (s)= − c(r) i ⊗ t v + s + dv V1,V2 ij dv j 2 i,j∈I IC1 Xr∈Z be the logarithm of AV1,V2 .

Lemma. The Taylor series of LV1,V2 at s = ∞ is given by the action on V1 ⊗ V2 −2 0 0 −1 of the element L(s) ∈ s (Y~ (g) ⊗ Y~ (g))[[s ]] deﬁned by

2 (r) n −m−2 ti,n tj,m−n L(s)= −~ cij T ℓ+r (−1) (m + 1)!s ⊗ 2 n! (m − n)! i,j∈I m≥n≥0 Xr∈Z X where Tmf(s)= f(s + m~). 28 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Proof. Given that −m−1 tj (v + s)= ~ tj,m(v + s) mX≥0 m + n = ~ s−m−1t (−1)n vns−n j,m n mX≥0 nX≥0 m m = ~ s−m−1 (−1)n t vn n j,n−m n=0 mX≥0 X ′ −a−2 and that ti(v) = −~ a≥0(a+1)ti,av , the expansion of the summand of LV1,V2 corresponding to a triple i, j, r is equal to P m 2 (r) −m−1 n m ~ cij T ℓ+r s (−1) nti,n−1 ⊗ tj,n−m 2 n n=1 mX≥0 X

−m−2 m m −2 Remark. Since (m + 1)!s = (−1) ∂s s , the above reads (r) −2 L(s)= c T ℓ+r Bi(∂s) ⊗ Bj (−∂s) (−s ) (6.6) ij 2 i,j∈I Xr∈Z

tk,n n where Bk(z)= ~ n≥0 n! z is the inverse Borel transform of tk(v). 6.6. AsymptoticP expansion of R0,↑/↓ (s). Let9 g(x) ∈ x−1C[[x−1]] be the unique V1,V2 solution of the diﬀerence equation g(x +1)= g(x) − x−2. 0 −1 0 0 −1 Deﬁne R (s) ∈ 1+ s (Y~ (g) ⊗ Y~ (g))[[s ]] by

0 1 (r) s log(R (s)) = cij T ℓ+r Bi(∂s) ⊗ Bj (−∂s) g (6.7) ℓ2~2 2 ℓ~ i,j∈I Xr∈Z 0,↑/↓ Let V , V ∈ Rep (Y~(g)). By Theorem 6.2, R (s) has an asymptotic expansion 1 2 fd V1,V2 as s → ∞ in any halfplane Re(s/ℓ~) ≷ m, which is of the form 1 + O(s−1).

Proposition. The asymptotic expansion of R0,↑/↓ (s) as s → ∞ is given by V1,V2 0 0 RV1,V2 (s)= πV1 ⊗ πV2 (R (s)) Proof. By deﬁnition of log(R0(s)) and (6.6), we have

0 1 (r) s s (Tℓ − 1) log(R (s)) = cij T ℓ+r Bi(∂s) ⊗ Bj (−∂s) g +1 − g ℓ2~2 2 ℓ~ ℓ~ i,j∈I Xr∈Z

(r) 1 = cij T ℓ+r Bi(∂s) ⊗ Bj (−∂s) − 2 s2 i,j∈I Xr∈Z = L(s)

9 −k−1 g(x) is equal to Pk≥0 Bkx , and is the asymptotic expansion at x = ∞ of the trigamma d2 −2 function Ψ1(x) = ln Γ(x)= (x + n) . dx2 Pn≥0 THE MEROMORPHIC R–MATRIX OF THE YANGIAN 29

0 Thus, RV1,V2 (s) is the unique formal solution of 0 ~ 0 RV1,V2 (s + ℓ )= AV1,V2 (s) · RV1,V2 (s) and therefore equals the asymptotic expansion of R0,↑/↓ (s). V1,V2

6.7. Properties of R0(s). The following is the universal analogue of Theorem 6.2. 0 0 0 −1 Theorem. R (s) ∈ (Y~ (g) ⊗ Y~ (g))[[s ]] has the following properties. (1) For every x ∈ Y~(g), the following holds in Y~(g)⊗2[s; s−1]] 0 (21) 0 R (s) · ∆(x) = ∆ (τs(x)) · R (s) D,s D,−s (2) The cabling identities 0 0 0 ∆ ⊗ 1 R (s2) = R13(s1 + s2) · R23(s2) D,s1 0 0 0 1 ⊗ ∆ R (s1 + s2) = R13(s1 + s2) · R12(s1) D,s2 0 ⊗3 −1 0 ⊗3 −1 hold in Y~ (g) [s1][[s2 ]] and Y~ (g) [s2][[s1 ]] respectively. (3) R0(s) is unitary 0 −1 0 R (s) = R21(−s) (4) For any a,b ∈ C 0 0 (τa ⊗ τb)(R (s)) = R (s + a − b) 0 ⊗2 −1 (5) R (s) − 1 ∈F−1(Y~(g) [[s ]]), with semiclassical limit given by ~Ω R0(s) − 1= h ∈ (U(g[z]) ⊗ U(g[w]))[[s−1]] z + s − w 0 Statements (1)–(4) follow from that fact that RV1,V2 (s) is the asymptotic expansion of R0,↑/↓ (s) and Theorem 6.2, since ﬁnite–dimensional representations sepa- V1,V2 rate points in Y~(g) (Proposition A.1). For completeness, we give a direct proof of Theorem 6.7 below, which does not rely on this fact. 6.8. Direct proof of Theorem 6.7.

0 0 Proof of (2). Let P ⊂ Y~ (g) be the C–linear span of {ti,r}i∈I,r∈Z≥0 , so that log(R0(s)) ∈ (P 0 ⊗ P 0)[[s−1]]. The cabling identities follow from the fact that each ti,r is primitive with respect to the Drinfeld coproduct, that is satisﬁes

∆(ti,r)= τs(ti,r) ⊗ 1+1 ⊗ ti,r D,s 0 and the fact that Y~ (g) is a commutative subalgebra.

Proof of (3). Write

0 1 (r) s 1 log(R (s)) = c T r Bi(∂s) ⊗ Bj (−∂s) g + ℓ2~2 ij 2 ℓ~ 2 i,j∈I Xr∈Z 1 ~ s 1 2 ∂s = 2 2 cij (e ) Bi(∂s) ⊗ Bj (−∂s) g + ℓ ~ I ℓ~ 2 i,jX∈ 1 s 1 = Ω(∂ ) g + ℓ2~2 s ℓ~ 2 30 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT where Ω(z)= i,j∈I cij (z)Bi(z) ⊗ Bj (−z). The unitary condition follows from P 1 1 Ω21(z)=Ω(−z) and g + x = −g − x 2 2 −1 The ﬁrst identity holds because cij (q)= cij (q )= cji(q), and the second because g(x)= −g(1 − x), since both sides are solutions of the same diﬀerence equation.

az Proof of (4). Since τaBi(z)= e Bi(z),

0 1 (r) (a−b)∂s s τa ⊗ τb log(R (s)) = cij T ℓ+r Bi(∂s) ⊗ Bj (−∂s)e g ℓ2~2 2 ℓ~ i,j∈I Xr∈Z

1 (r) s + a − b = cij T ℓ+r Bi(∂s) ⊗ Bj (−∂s) g ℓ2~2 2 ℓ~ i,j∈I Xr∈Z = log(R0(s + a − b))

0 ⊗2 −1 Proof of (5). By (2.7), it suffices to prove that log(R (s)) ∈F−1(Y~(g) [[s ]]), and that Ω log(R0(s)) = ~ h z + s − w Note first that s ℓ~ · m! t ⊗ t (−1)m∂mg = t ⊗ t + O(s−m−2) i,n j,m s ℓ~ i,n j,m sm+1 ⊗2 −1 n m −m−1 lies in F−1(Y~(g) [[s ]]), and has symbol ℓ~ · m! dihiz ⊗ dj hjw s . Since ⊗2 −1 the shifts Tx preserve the filtration F•(Y~(g) [[s ]]) and act as the identity on its 0 ⊗2 −1 associated graded space, it follows that log(R (s)) ∈F−1(Y~(g) [[s ]]), and that ~ m log(R0(s)) = c (1) (−1)n d h zn ⊗ d h wm−n s−m−1 ij n i i j j ℓ I i,jX∈ mX≥n≥0 Ω = ~ h z + s − w where the last equality follows from the fact that

−1 ∨ cij (1)dj hj = ℓ (B )ij dj hj = ℓ̟i I I Xj∈ Xj∈ ∨ with B = (diaij ), and ̟i ∈ h the fundamental coweights.

0 Proof of (1). The intertwining relation is obvious for x ∈ Y~ (g), since the latter is the commutative algebra generated by the elements ti,r, which satisfy

(21) ∆(ti,r)= τs(ti,r) ⊗ 1+1 ⊗ ti,r = ∆ (τsti,r) D,s D,−s Thus, it suﬃces to prove that, for any k ∈ I

R0(s)∆(x± ) = ∆ (21)(x± )R0(s) D,s k,0 D,−s k,0 THE MEROMORPHIC R–MATRIX OF THE YANGIAN 31

+ We verify this identity for the + case only. By Proposition 3.3, ∆(xk,0) is equal to D,s + + xk,0 ⊗ 1+1 ⊗ xk,0 + Xk(s), where N N X (s)= ~ s−N−1 (−1)n+1 ξ ⊗ x+ k n k,n k,N−n n=0 NX≥0 X We therefore have to prove that 0 + + + + (21) 0 Ad(R (s))· xk,0 ⊗ 1+1 ⊗ xk,0 = xk,0 ⊗1+1⊗xk,0+Xk (−s)−Ad(R (s))·Xk(s)

0 + + (21) We claim that Ad(R (s)) · (xk,0 ⊗ 1) = xk,0 ⊗ 1+ Xk (−s). Given this, the 0 −1 + + unitary condition (3) then implies that Ad(R (s)) · (1 ⊗ xk,0)=1 ⊗ xk,0 + Xk(s) + which, combined with the claim yields the required intertwining equation for xk,0. To prove the claim, we rely on the following commutation relation, which was obtained in [10, §2.9]

~ ~ diaik z − diaik z p e 2 − e 2 z [B (z), x+ ]= · x+ i k,n z  k,n+p p!  Xp≥0 Combining with the deﬁnition of Ω(z) given above, we can carry out the following 0 computation, for each k ∈ I, n ∈ Z≥0, and y ∈ Y~ (g). ~ ~ diaik z − diaik z + 1 ~ z e 2 − e 2 [Ω(z), x ⊗ y]= c e 2 k,n 2~2 ij ℓ I I z !! Xj∈ Xi∈ (−z)p · x+ ⊗ B (z)y  k,n+p p!  j Xp≥0  ~ ~  ℓ z − ℓ z p 1 e 2 − e 2 (−z) = · x+ ⊗ B (z)y ℓ2~2 z  k,n+p p!  k Xp≥0   Note that we used the equation (6.1) satisﬁed by (cij (q)) above. This calculation, combined with ℓ~ ∂ − ℓ~ ∂ 2 2 e 2 s − e 2 s s 1 ℓ ~ · g + = ∂ ℓ~ 2 s s yields the commutation relation

r + p [log(R0(s)), x+ ⊗ y]= (−1)px+ ⊗ ~ t s−r−p−1 y k,n k,n+p  r k,r  Xp≥0 Xr≥0 The claim now follows from   Ad(R0(s)) = exp(ad(log(R0(s)))) + 0 + where both sides are acting on Vk ⊗ Y~ (g), where Vk is the C–linear span of {xk,n}n≥0. 32 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

7. The universal and the meromorphic R–matrices of Y~(g)

In this section, we construct the meromorphic and universal R–matrices of Y~(g).

7.1. The meromorphic R–matrix. Given V1, V2 ∈ Repfd(Y~(g)) and ε ∈ {↑, ↓}, ε C deﬁne RV1,V2 : → End(V1 ⊗ V2) by

Rε (s)= R+ (s) · R0,ε (s) · R− (s), V1,V2 V1,V2 V1,V2 V1,V2 where R+ (s) = (1 2) ◦ R− (−s)−1 ◦ (1 2). V1,V2 V2,V1

ε Theorem. The meromorphic function RV1,V2 (s) has the following properties. (1) The map

ε (1 2) ◦ RV1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s)

is a morphism of Y~(g)–modules, which is natural in V1, V2. (2) For any V1, V2, V3 ∈ Repfd(Y~(g)), ε ε ε RV1 ⊗V2,V3 (s2)= RV1,V3 (s1 + s2) · RV2,V3 (s2) s1 ε ε ε RV1,V2 ⊗V3 (s1 + s2)= RV1,V3 (s1 + s2) · RV1,V2 (s1). s2

In particular, the QYBE holds on V1 ⊗ V2 ⊗ V3:

ε ε ε ε ε ε RV1,V2 (s1)RV1,V3 (s1 + s2)RV2,V3 (s2)= RV2,V3 (s2)RV1,V3 (s1 + s2)RV1,V2 (s1). (3) For any a,b ∈ C,

ε ε RV1(a),V2(b)(s)= RV1,V2 (s + a − b).

(4) R↑ (s) and R↓ (s) are related by the unitarity relation: V1,V2 V2,V1

(1 2) ◦ R↑ (−s) ◦ (12) = R↓ (s)−1. V1,V2 V2,V1

(5) R↑/↓ (s) have the same asymptotic expansion, which is of the form V1,V2

R↑/↓ (s) ∼ 1+ ~Ω s−1 + O(s−2) V1,V2 g as s → ∞ in any halfplane of the form Re(s/~) ≷ m.

Proof. (1) By deﬁnition,

(1 2) ◦ Rε (s)= R− (−s)−1 · (1 2) ◦ R0,ε (s) · R− (s). V1,V2 V2,V1 V1,V2 V1,V2

− The result therefore follows from the fact that R (s) is a morphism of Y~(g)– V1,V2 modules V1 ⊗ V2 → V1 ⊗ V2 (Theorem 4.1), and Theorem 6.2 (1). s D,s (2) We will prove the following equivalent version of the ﬁrst cabling identity

ε ε ε (1 2 3) ◦ RV1 ⊗V2,V3 (s2) = (1 2) RV1,V3 (s1 + s2) ⊗ 1 (2 3) 1 ⊗ RV2,V3 (s2) (7.1) s1 THE MEROMORPHIC R–MATRIX OF THE YANGIAN 33

By deﬁnition, the left–hand side is equal to

R− (−s )−1 · (1 2 3) ◦ R0,ε (s ) · R− (s ) V3,V1 ⊗V2 2 V1 ⊗V2,V3 2 V1 ⊗V2,V3 2 s1 s1 s1 =R− (−s )−1 · 1 ⊗ R− (s )−1 · (1 2 3) ◦ R0,ε (s ) · V3,V1 ⊗V2 2 V1,V2 1 V1 ⊗ V2,V3 2 s1 D,s1 ! R− (s ) ⊗ 1 · R− (s ) V1,V2 1 V1 ⊗V2,V3 2 s1 =R− (−s )−1 · 1 ⊗ R− (s )−1 · (1 2) R0,ε (s + s ) ⊗ 1 · V3,V1 ⊗V2 2 V1,V2 1 V1,V3 1 2 s1 (2 3) 1 ⊗ R0,ε (s ) · R− (s ) ⊗ 1 · R− (s ) V2,V3 2 V1,V2 1 V1 ⊗V2,V3 2 s1

In the ﬁrst equality, we used Theorem 4.1 in order to change V1 ⊗ V2 to V1 ⊗ V2, s1 D,s1 while the second equality follows from the cabling identity satisﬁed by R0,ε(s) (Theorem 6.2 (2)). Note that we have the following identity, which follows from the cocycle equation (4.7) after renaming variables

R− (s + s ) ⊗ 1 · R− (−s )= 1 ⊗ R− (−s ) · R− (s ) V1,V3 1 2 V1 ⊗ V3,V2 2 V3,V2 2 V1,V3 ⊗ V3 1 s1+s2 −s2

Inserting this operator and its inverse in the last line of the computation above allows us to write the left–hand side of (7.1) as A(s1,s2) · B(s1,s2), where

A(s ,s )= R− (−s )−1 · 1 ⊗ R− (s )−1 · (1 2) R0,ε (s + s ) ⊗ 1 · 1 2 V3,V1 ⊗V2 2 V1,V2 1 V1,V3 1 2 s1 R− (s + s ) ⊗ 1 · R− (−s ) V1,V3 1 2 V1 ⊗ V3,V2 2 s1+s2 −1 B(s ,s )= 1 ⊗ R− (−s ) · R− (s ) · 1 2 V3,V2 2 V1,V3 ⊗ V3 1 −s2 (2 3) 1 ⊗ R0,ε (s ) · R− (s ) ⊗ 1 · R− (s ) V2,V3 2 V1,V2 1 V1 ⊗V2,V3 2 s1

Thus, in order to prove (7.1), it is enough to show that

ε ε A(s1,s2) = (1 2) ◦ RV1,V3 (s1 + s2) ⊗ 1 and B(s1,s2) = (2 3) ◦ 1 ⊗ RV2,V3 (s2)

We verify the latter below, the proof of the former, being entirely analogous, is omitted. 34 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Using the cocycle equation (4.7) again, we have:

B(s ,s )= R− (s )−1 · 1 ⊗ R− (−s )−1 · (2 3) ◦ 1 ⊗ R0,ε (s ) · 1 2 V1,V3 ⊗ V2 1 V3,V2 2 V2,V3 2 −s2 1 ⊗ R− (s ) · R− (s + s ) V2,V3 2 V1,V2 ⊗V3 1 2 s2 = R− (s )−1 · (2 3) ◦ 1 ⊗ Rε (s ) · R− (s + s ) V1,V3 ⊗ V2 1 V2,V3 2 V1,V2 ⊗V3 1 2 −s2 s2 = R− (s )−1 · R − (s ) · (2 3)◦ 1 ⊗ Rε (s ) V1,V3 ⊗ V2 1 V1,V3 ⊗ V2 1 V2,V3 2 −s2 −s2 ε = (2 3) ◦ 1 ⊗ RV2,V3 (s2) In the second line, we have used the deﬁnition of Rε(s), and in the third, Part (1) of this theorem.

(3) follows from Theorem 4.1 (2), and Theorem 6.2 (3).

(4) By deﬁnition,

(1 2) ◦ R↑ (−s) ◦ (12) = R− (s)−1 · (1 2) ◦ R0,↑ (−s) ◦ (1 2) · R+ (s)−1 V1,V2 V2,V1 V1,V2 V2,V1 = R− (s)−1 · R0,↓ (s)−1 · R+ (s)−1 V2,V1 V2,V1 V2,V1 = R↓ (s)−1 V2,V1 where the second equality uses Theorem 6.2 (4).

(5) follows from Theorem 6.2 (5) and the Taylor series expansion of R− (s)= V1,V2 1+ ~r/s + O(s−2) given in Theorem 4.1 (3).

7.2. Existence of a rational intertwiner. The following extends to an arbitrary pair of representations V1, V2 ∈ Repfd(Y~(g)) a result due to Drinfeld, which is valid when V1, V2 are irreducible [4, Thm. 4], and Maulik–Okounkov, which is valid when g is simply–laced, and V1, V2 arise from geometry [20].

Theorem. There is a rational map RV1,V2 (s): C → EndC(V1 ⊗ V2), which is normalised by RV1,V2 (∞)= 1 and such that

(1 2) ◦ RV1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s) intertwines the action of Y~(g). In particular, V1(s) ⊗ V2 and V2 ⊗ V1(s) are isomorphic as Y~(g)–modules for all but ﬁnitely many values of s. R0 R0 Proof. Let V1,V2 (s) be the rational operator such that (1 2)◦ V1,V2 (s) intertwines the action on V1(s) ⊗ V2 and V2 ⊗ V1(s), as obtained in Theorem 6.3. Then, D,0 D,0

R (s)= R+ (s) · R0 (s) · R− (s) V1,V2 V1,V2 V1,V2 V1,V2 yields the required map. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 35

7.3. Non existence of rational commutativity constraints.

Theorem. There is no function RV1,V2 : C → AutC(V1 ⊗ V2) which is rational, deﬁned for any V1, V2 ∈ Repfd(Y~(g)), and such that the following holds

(1) (1 2) ◦ RV1,V2 (s): V1(s) ⊗ V2 → V2 ⊗ V1(s) intertwines the action of Y~(g),

and is natural in V1 and V2. (2) For any V1, V2, V3 ∈ Repfd(Y~(g)),

RV1 ⊗V2,V3 (s2)= RV1,V3 (s1 + s2) · RV2,V3 (s2) s1

RV1,V2 ⊗V3 (s1 + s2)= RV1,V3 (s1 + s2) · RV1,V2 (s2) s2

Proof. Assume that such a rational RV1,V2 (s) exists. Set R0 (s)= R+ (s)−1 · R (s) · R− (s)−1 V1,V2 V1,V2 V1,V2 V1,V2 R0 Then, the rational operator V1,V2 (s) contradicts Theorem 6.4, and the claim follows.

7.4. The universal R-matrix. We now turn our attention to obtaining a formal, universal analogue of Theorem 7.1. Consider the formal power series + 0 − −1 R(s)= R (s) · R (s) · R (s) ∈ (Y~(g) ⊗ Y~(g))[[s ]] + − −1 where R (s)= R21(−s) . This series admits an expansion ∞ −k R(s)=1+ Rks , Rk ∈Fk−1(Y~(g) ⊗ Y~(g)). kX=1 Theorem. The formal power series R(s) has the following properties. (1) For every x ∈ Y~(g), the following holds in Y~(g)⊗2[s; s−1]]

op −1 τs ⊗ 1 ◦ ∆ (x)= R(s) · τs ⊗ 1 ◦ ∆(x) · R(s) (2) R(s) satisﬁes the cabling identities

∆ ⊗ 1(R(s)) = R13(s)R23(s)

1 ⊗ ∆(R(s)) = R13(s)R12(s) (3) R(s) is unitary −1 R(s) = R21(−s). (4) For any a,b ∈ C, we have

(τa ⊗ τb)R(s)= R(s + a − b) ⊗2 −1 (5) R(s) − 1 ∈F−1(Y~(g) [[s ]]), with semiclassical limit given by ~Ω R(s) − 1= g ∈ (U(g[z]) ⊗ U(g[w]))[[s−1]] s + z − w −1 −2 In particular, R(s)=1+ ~s Ωg + O(s ). (6) For any V1, V2 ∈ Repfd(Y~(g)) and ε ∈ {↑, ↓}, we have ε RV1,V2 (s) ∼ RV1,V2 (s)

as s → ∞ in any halfplane of the form Re(s/~) ≷ m. Here RV1,V2 (s) =

πV1 ⊗ πV2 (R(s)). That is, RV1,V2 (s) is equal to the asymptotic expansion of ε RV1,V2 (s) from (5) of Theorem 7.1. 36 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Proof. Parts (1) and (3)–(6) are deduced directly from the deﬁnition of R(s) using the properties of R−(s) and R0(s) established in Theorems 4.1 and 6.7, respectively. We note, however, that the cabling identities (2) do not follow directly from Theorems 4.1 and 6.7, as we do not have access to a formal, universal version of the cocycle equation (4.3) of Theorem 4.1. To remedy this, we shall instead make use of the fact that Repfd(Y~(g)) is suﬃciently large to distinguish elements of Y~(g). More precisely, by Proposition A.1 (with n = 3), it is enough to prove that the following identities hold on V1 ⊗ V2 ⊗ V3, for every V1, V2, V3 ∈ Repfd(Y~(g)):

RV ⊗V ,V (s)= RV ,V (s)RV ,V (s) 1 2 3 1 3 2 3 (7.2) RV1,V2⊗V3 (s)= RV1,V3 (s)RV1 ,V2 (s) Fix ε ∈ {↑, ↓} and consider the first equality. By (6), the left-hand side (resp. right- ε hand side) is equal to the uniquely determined asymptotic expansion of RV1⊗V2,V3 (s) ε ε ~ (resp. RV1,V3 (s)RV2 ,V3 (s)) as s → ∞ in any halfplane of the form Re(s/ ) ≷ m. Moreover, by the first equality in (2) of Theorem 7.1, taken with s1 = 0 and s2 = s, we have ε ε ε RV1⊗V2,V3 (s)= RV1,V3 (s)RV2,V3 (s). Thus, we can conclude that the first cabling identity of (7.2) necessarily holds. An identical argument establishes the second identity. As an immediate consequence of the above theorem and the uniqueness assertion of Theorem 1.1 (see also Appendix B), we obtain the following corollary. Corollary. R(s) is equal to Drinfeld’s universal R–matrix. In particular, Theorem 7.4 provides an independent, and constructive proof of the existence of Drinfeld’s universal R–matrix.

8. Meromorphic tensor structures In this section, we reinterpret our results in the language of meromorphic tensor categories. We refer to [22, 23] for a more abstract and general treatment of meromorphic tensor categories. We caution the reader, however, that the framework developed in [22, 23] does not include examples where the tensor product depends non–trivially on a parameter, as is the case for the deformed Drinfeld tensor product. The setup of [22, 23] is also more general than needed for our purposes, in that it is adapted to pseudo–tensor categories, where the tensor product need not be deﬁned for all pairs of representations, or be representable.

8.1. Drinfeld tensor product. Proposition.

(1) The category (Repfd(Y~(g)), ⊗ ) is a meromorphic (in fact, rational) tensor D,s category over (C, +). (2) Each of the resummed abelian R–matrices R0,↑/↓(s) is a meromorphic braiding on (Repfd(Y~(g)), ⊗ ). D,s (3) (Repfd(Y~(g)), ⊗ ) does not admit a rational braiding. D,s THE MEROMORPHIC R–MATRIX OF THE YANGIAN 37

Proof. (1) Repfd(Y~(g)) admits an action of the additive group (C, +) given by V 7→ V (s). As recalled in 3.2, for every V, W ∈ Repfd(Y~(g)), there is a rational action of Y~(g) on V ⊗ W given by the deformed Drinfeld tensor product. The properties (1)–(5) of Theorem 3.2 mean exactly that (Repfd(Y~(g)), ⊗ ) is a rational D,s tensor category over (C, +). (2) is the content of Theorem 6.2 (1)–(3). (3) is a rephrasing of Theorem 6.4.

8.2. Standard tensor product. Proposition.

(1) The category (Repfd(Y~(g)), ⊗) is a meromorphic (in fact, polynomial) ten- s sor category over (C, +). (2) Each of the resummed R–matrices R↑/↓(s) is a meromorphic braiding on (Repfd(Y~(g)), ⊗). s (3) (Repfd(Y~(g)), ⊗) does not admit a rational braiding. s

Proof. (1) This is a consequence of the fact that V1 ⊗ V2 arises from the algebra s homomorphism ∆s : Y~(g) → (Y~(g) ⊗ Y~(g))[s]. This tensor product satisﬁes the properties analogous to (3)–(5) of Theorem 3.2. (2) is the content of (1)–(3) of Theorem 7.1. (3) is a rephrasing of Theorem 7.3.

8.3. Meromorphic tensor structures. Proposition. R−(s) is a rational braided tensor structure on the identity functor

0,↑/↓ ↑/↓ Repfd(Y~(g)), ⊗ , R (s) −→ Repfd(Y~(g)), ⊗, R (s) D,s s Proof. By deﬁnition of a tensor structure on a functor, the statement means that, for every V1, V2 ∈ Repfd(Y~(g)), there is a rational End(V1 ⊗ V2)–valued function of s, R− (s), which satisﬁes (1)–(3) of Theorem 4.1. Namely, V1,V2 R− (s): V ⊗ V → V ⊗ V V1,V2 1 2 1 2 s D,s − − is a Y~(g)–intertwiner such that R (s)= R (s+a−b), and the following V1(a),V2(b) V1,V2 diagram commutes, for every V1, V2, V3 ∈ Repfd(Y~(g))

(V1 ⊗s1 V2) ⊗s2 V3 V1 ⊗s1+s2 (V2 ⊗s2 V3)

− − R (s1)⊗1V 1V ⊗R (s1) V1,V2 3 1 V2,V3

(V1 ⊗ V2) ⊗s2 V3 V1 ⊗s1+s2 (V2 ⊗ V3) D,s1 D,s2

− − R (s2) R (s1+s2) V1 ⊗ V2,V3 V1,V2 ⊗ V3 D,s1 D,s2 (V1 ⊗ V2) ⊗ V3 V1 ⊗ (V2 ⊗ V3) D,s1 D,s2 D,s1+s2 D,s2 38 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Lastly, it is claimed in (3) that R−(s) is compatible with the braidings on the two categories, that is satisﬁes

(1 2)◦R↑/↓ (s) V1,V2 / V1(s) ⊗ V2 V2 ⊗ V1(s)

R− (s) R− (−s) V1,V2 V2,V1 V (s) ⊗ V / V ⊗ V (s) 1 2 0,↑/↓ 2 1 D,0 (1 2)◦R (s) D,0 V1,V2 The commutativity of the diagram follows from the fact that, by deﬁnition

R↑/↓ (s)= R+ (s) · R0,↑/↓ (s) · R− (s) V1,V2 V1,V2 V1,V2 V1,V2 where R+ (s) = (1 2) ◦ R− (−s)−1 ◦ (1 2). V1,V2 V2,V1

9. Relation to the quantum loop algebra Uq(Lg) In this section, we review the construction of the tensor functor between ﬁnite– dimensional representations of Y~(g) and the quantum loop algebra Uq(Lg) obtained in [11, 12]. We then disprove a conjecture stated in [12], and relate the meromorphic R–matrices of Y~(g) and Uq(Lg).

9.1. The functor Γ [11]. Set q = exp(πι~) and assume that ~ ∈ C \ Q, so that q is not a root of unity. Let Uq(Lg) be the quantum loop algebra of g. We refer to [2, Ch. 12] and references therein for the deﬁnition and basic properties of Uq(Lg). A ﬁnite–dimensional representation V ∈ Repfd(Y~(g)) is said to be non–congruent if, for every a,b ∈ σ(V ), a − b 6∈ Z6=0. The full subcategory of such representations NC is denoted by Repfd (Y~(g)) ⊂ Repfd(Y~(g)). In [11, §5], an exact, essentially surjective and faithful functor Γ NC : Repfd (Y~(g)) → Repfd(Uq(Lg))

NC is deﬁned. Γ is such that Γ(V )= V as vector spaces for any V ∈ Repfd (Y~(g)), and NC restricts to an isomorphism of categories on an explicit subcategory of Repfd (Y~(g)) determined by a choice of log. NC Given V ∈ Repfd (Y~(g)), and i ∈ I, consider the additive diﬀerence equation

φi(u +1)= ξi(u)φi(u) determined by the action of the commuting current ξi(u) of Y~(g) on V . The action of the commuting current ψi(z) of Uq(Lg) on V is given by the monodromy of this equation, that is by

ψi(z) = lim ξi(u + n) ··· ξi(u − n) n→∞ z=e2πιu

The action of the raising and lowering operators of Uq( Lg) on V requires the non– congruence hypothesis. It is not relevant for our current discussion, and we refer to [11, §5] for details. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 39

9.2. Meromorphic tensor structure on Γ [12]. Let V1, V2 ∈ Repfd(Y~(g)), and consider the abelian qKZ equation determined by R0,↑/↓ (s), that is the diﬀerence V1,V2 equation Φ(s +1)= R0,↑/↓ (s) · Φ(s) (9.1) V1,V2 ↑/↓ Let J (s): C → EndC(V ⊗ V ) be the left canonical solution of (9.1), which is V1,V2 1 2 uniquely determined by the requirement that it be holomorphic and invertible for Re(s) ≪ 0, and possess an asymptotic expansion of the form (−s)~Ω 1+ O(s−1) as s → ∞ in any halfplane of the form Re(s)

Γ ↑/↓ NC , J (s) : Repfd (Y~(g)), ⊗ −→ Repfd(Uq(Lg)), ⊗ D,s D,ζ where ζ = exp(2πιs).10 9.3. Tensor structure with respect to the standard coproducts. The Drin- feld coproduct of Uq(Lg) is known to be conjugated to the standard coproduct by − the lower triangular part R (ζ) of the universal R–matrix of Uq(Lg) (see, for exam- NC ple, [6]). Thus, for any V1, V2 ∈ Repfd (Y~(g)), we have the following isomorphisms of Uq(Lg)–modules

↑/↓ JV ,V (s) Γ(V1) ⊗ Γ(V2) 1 2 Γ/ ζ V1 ⊗ V2 s

R− (ζ) R− (s) Γ(V1),Γ(V2) V1,V2

Γ(V1) ⊗ Γ(V2) /Γ D,ζ V1 ⊗ V2 D,s J ↑/↓ (s) V1 ,V2 where V ⊗ V = τ ∗V ⊗V for any V , V ∈ Rep (U (Lg)), and J ↑/↓ (s) is deﬁned 1 2 ζ 1 2 1 2 fd q V1,V2 ζ as the composition J ↑/↓ (s)= R− (s)−1 · J ↑/↓ (s) · R− (ζ) (9.2) V1,V2 V1,V2 V1,V2 Γ(V1),Γ(V2) Theorem 9.2 and Proposition 8.3 therefore imply the following Corollary. J ↑/↓ (s) is a meromorphic tensor structure on the functor Γ, with V1,V2 respect to the standard tensor products

Γ ↑/↓ NC ,J (s) : Repfd (Y~(g)), ⊗ −→ Repfd(Uq(Lg)), ⊗ s ζ 0 10In [12], the right canonical solution of the equations φ(s +1) = R ,↑/↓ (s) · φ(s) is shown to V1,V2 give rise to a tensor structure on Γ. A similar computation shows that the left solution yields a tensor structure on a variant of Γ, which we denote Γ for simplicity. 40 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

9.4. Non regularity of J ↑/↓ (s). Since the tensor products ⊗ and ⊗ are polyno- V1,V2 s ζ mial, the first two authors conjectured in [12, §2.13] that J ↑/↓ (s) is regular and V1,V2 invertible at s = 0. If so, J ↑/↓ (0) would give rise to a (non–meromorphic) tensor V1,V2 structure on the functor Γ with respect to the standard (unshifted) tensor products NC on Repfd (Y~(g)) and Repfd(Uq(Lg)). The following shows that this is not the case. Proposition. The meromorphic tensor structure J ↑/↓(s) is either singular or not invertible at s =0. Proof. Since Γ(V (a)) = Γ(V )(e2πιa) and each of the factors in the definition (9.2) of J ↑/↓ is compatible with shifts, we have J ↑/↓ (s)= J ↑/↓ (s + a − b) V1(a),V2(b) V1,V2 Thus, if J ↑/↓ (s) were regular and invertible at s = 0 for every V , V , then for a V1,V2 1 2 fixed V1, V2 it would be holomorphic and invertible for all s. This cannot be true, as the following argument shows. Assume that V1, V2 are irreducible, with highest weight vectors v1, v2 of weights ∗ λ1, λ2 ∈ h , and that (λ1, λ2) 6= 0. Let W ⊂ V1 ⊗ V2 be the subspace spanned by − − v1 ⊗ v2. The triangularity of R (s) and R (ζ) implies that ↑ ↑ JV ,V (s) = JV ,V (s) 1 2 W 1 2 W

If the restriction of J ↑ (s) to W were holomorphic and invertible for every s ∈ C, V1,V2 the same would be true for R0,↑ (s), since by 9.1 V1,V2 J ↑ (s)= R0,↑ (s + 1) · J ↑ (s + 1). V1,V2 V1,V2 V1,V2

In turn, AV1,V2 (s)|W would also be holomorphic and invertible, since (see equation (6.2) above) R0,↑ (s + ℓ~)= A (s)R0,↑ (s). V1,V2 V1,V2 V1 ,V2

Since AV1,V2 (s) is a rational function of s such that AV1,V2 (∞) = 1, this implies ℓ~ that A (s)| ≡ 1. The expansion A(s)=1 − Ω + O(s−3), then yields V1,V2 W s2 h Ωh|W = 0, which contradicts the fact that Ωhv1 ⊗ v2 = (λ1, λ2)v1 ⊗ v2.

Remark. The above result leaves open the question of whether there exists a tensor functor between the (non meromorphic) tensor categories

(Repfd(Y~(g)), ⊗) → (Repfd(Uq(Lg)), ⊗)

9.5. The meromorphic abelian R–matrix of Uq(Lg). Assume now that |q| 6= 1. We review below the analogue of Theorem 6.2 for Uq(Lg) obtained in [12], based on the results of [3]. Let V1, V2 be two ﬁnite–dimensional representations of Uq(Lg). In [12, §8], a A 1 rational function V1,V2 (ζ): P → EndC(V1 ⊗V2) is deﬁned via a contour integral formula involving the commuting generators of Uq(Lg), which is analogous to the A one given in 6.1. V1,V2 (ζ) is regular at ζ =0, ∞, and such that A A V1,V2 (0) = 1 = V1,V2 (∞) A A ′ ′ Moreover, [ V1,V2 (ζ), V1,V2 (ζ )] = 0 for any ζ, ζ . THE MEROMORPHIC R–MATRIX OF THE YANGIAN 41

2ℓ A Consider the (regular) diﬀerence equation with step q determined by V1,V2 (ζ) R 2ℓ A R (q ζ)= V1,V2 (ζ) · (ζ) (9.3) ↑ ↓ This equation admits two meromorphic solutions R (ζ), R (ζ), which are uniquely ↑/↓ determined by the requirement that R (ζ) be holomorphic near z = q±∞ and ↑/↓ such that R (q±∞) = 1. These are explicitly given by −→ −→ R↑ A 2ℓn −1 R↓ A −2ℓn (ζ)= V1,V2 (q ζ) and (ζ)= V1,V2 (q ζ) nY≥0 nY≥1 Now set ↑/↓ q∓Ωh · R (ζ) if |q| < 1 R0,↑/↓ (ζ)= V1,V2 ↑/↓  ±Ωh R  q · (ζ) if |q| > 1

 R0,↑/↓ Theorem. [12, §8] The category Repfd(Uq(Lg)), ⊗ , (ζ) is a meromorphic D,ζ braided tensor category.

0 Remark. Let R be the abelian part of the universal R–matrix of Uq(Lg), and set 0 0 ⊗2 R (ζ)= τζ ⊗ 1(R ) ∈ Uq(Lg) [[ζ]] R0 § It is easy to see that V1,V2 (ζ) satisﬁes the diﬀerence equation (9.3)[12, 8]. It follows by uniqueness that R0 (ζ) is the Taylor expansion at ζ =0 of R0,↑ (ζ) V1,V2 V1,V2 if |q| < 1, and of R0,↓ (ζ) if |q| > 1. Similarly, if V1.V2 0,∨ 0 ⊗2 −1 R (ζ)= 1 ⊗ τζ (R ) ∈ Uq(Lg) [[ζ ]] then R0,∨ (ζ) is the Taylor expansion at ζ = ∞ of R0,↓ (ζ) if |q| < 1, and of V1,V2 V1,V2 R0,↑ (ζ) if |q| > 1. V1.V2

9.6. Abelian qDrinfeld–Kohno theorem. Let now V1, V2 ∈ Repfd(Y~(g)), and consider the abelian qKZ equations (9.1) determined by R0,↑/↓ (s). V1,V2 Theorem. [12, Thm. 9.3] The monodromy of the abelian qKZ equations (9.1) is equal to R0,↑/↓ (ζ). In other words, the following holds Γ(V1),Γ(V2)

R0,↑/↓ (ζ) = lim R0,↑/↓ (s + n) ···R0,↑/↓ (s) ···R0,↑/↓ (s − n) Γ(V ),Γ(V ) V1,V2 V1,V2 V1,V2 1 2 n→∞ ζ=exp(2πιs)

(9.4)

In [12, §9.6], this assertion is strengthened to include the abelian qKZ equations NC with values in V1 ⊗···⊗ Vn, where Vi ∈ Repfd (Y~(g)) and n ≥ 3. Thus, Theorem 9.6 is an analogue of the Drinfeld–Kohno theorem for the abelian qKZ equations determined by R0,↑/↓(s). As is the case for the Drinfeld–Kohno theorem, Theorems 9.2 and 9.6 can be understood as deﬁning a meromorphic braided tensor functor

NC 0,↑/↓ R0,↑/↓ Repfd (Y~(g)), ⊗ , R (s) −→ Repfd(Uq(Lg)), ⊗ , (ζ) D,s D,ζ 42 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Unlike its non–meromorphic counterpart, this notion involves an ordered pair (K, Kˇ) of meromorphic tensor structures on the functor Γ, such that the following holds [12, Rem. 9.3]11 0,↑/↓ −1 0,↑/↓ R (ζ)= KˇV ,V (−s) · R (s) · KV ,V (s) (9.5) Γ(V1),Γ(V2) 2 1 21 V1,V2 1 2

Comparing (9.5) with (9.4), and using the deﬁnition of JV1,V2 given in 9.1, we see that one can take K = J ↑/↓ , which is a regularisation of the product V1,V2 V1,V2 R0,↑/↓ (s − 1) · R0,↑/↓ (s − 2) ··· V1,V2 V1,V2 and Kˇ = J ↓/↑ , which is a regularisation of V1,V2 V1,V2 −1 −1 R0,↓/↑ (s − 1) · R0,↓/↑ (s − 2) ··· = R0,↑/↓ (−s + 1) · R0,↑/↓ (−s + 2) ··· V2,V1 V1,V2 V2,V1 21 V2,V1 21 −1 where we used the unitarity relation R0,↓/↑ (s)= R0,↑/↓ (s) . V1,V2 V2,V1 21 9.7. Meromorphic braided tensor equivalence for standard coproducts. + 0 − Let R be the universal R–matrix of Uq(Lg), R = R · R · R its Gauss de- ± ± composition, set R(ζ) = τζ ⊗ 1(R), and R (ζ) = τζ ⊗ 1(R ). Then, if V1, V2 ∈ Rep (U (Lg)), R± (ζ) are rational functions of ζ such that fd q V1,V2 −1 R+ (ζ)= R− (ζ−1) V1,V2 V2,V1 21

Deﬁne the meromorphic R–matrix of Uq(Lg) by R↑/↓ (ζ)= R+ (ζ) · R0,↑/↓ (ζ) · R− (ζ) (9.6) V1,V2 V1,V2 V1,V2 V1,V2 By Remark 9.5, R (ζ) is the Taylor expansion at ζ =0 of R↑ (ζ) if |q| < 1, V1,V2 V1,V2 and of R↓ (ζ) if |q| > 1. V1,V2 Proposition. The pair J ↑/↓ (s),J ↓/↑ (s) is a meromorphic braided tensor struc- V1,V2 V1,V2 ture on the functor Γ with respect to the standard tensor products and meromorphic R–matrices

NC ↑/↓ R↑/↓ Repfd (Y~(g)), ⊗, R (s) −→ Repfd(Uq(Lg)), ⊗, (ζ) s ζ Proof. By Corollary 9.3, and 9.6, we only need to check the compatibility of J ↑/↓ (s),J ↓/↑ (s) with the meromorphic braidings, that is the relation V1,V2 V1,V2 −1 R↑/↓ (ζ)= J ↓/↑ (−s) · R↑/↓ (s) · J ↑/↓ (s) (9.7) Γ(V1),Γ(V2) V2,V1 21 V1,V2 V1,V2 The Gauss decomposition (9.6) yields R↑/↓ (ζ)= R+ (ζ) · R0,↑/↓ (ζ) · R− (ζ) Γ(V1),Γ(V2) Γ(V1),Γ(V2) Γ(V1),Γ(V2) Γ(V1),Γ(V2) −1 = R+ (ζ) · J ↓/↑ (−s) · R0,↑/↓ (s) · J ↑/↓ (s) · R− (ζ) Γ(V1),Γ(V2) V2,V1 21 V1,V2 V1,V2 Γ(V1),Γ(V2) −1 = R+ (ζ) · J ↓/↑ (−s) · R+ (s)−1 · R↑/↓ (s) Γ(V1),Γ(V2) V2,V1 21 V1,V2 V1,V2 · R− (s)−1 · J ↑/↓ (s) · R− (ζ) V1,V2 V1,V2 Γ(V1),Γ(V2) −1 = J ↓/↑ (−s) · R↑/↓ (s) · J ↑/↓ (s) V2,V1 21 V1,V2 V1,V2

11For the meromorphic braided tensor structures discussed in 8.3, Kˇ = K. THE MEROMORPHIC R–MATRIX OF THE YANGIAN 43 where the second equality follows from the twist relation (9.5), the third from the deﬁnition of R↑/↓ (s) given in 7.1, and the fourth from the deﬁnition (9.2) V1,V2 −1 of J ↑/↓ (s), together with the unitarity relations R+ (ζ) = R− (ζ−1) and V1,V2 V1,V2 V2,V1 21 −1 R+ (s)= R− (−s) . V1,V2 V2,V1 21

Remark. Much like (9.4), the twist relation (9.7) can be regarded as a monodromy relation. Indeed, both

J ↑/↓ (s)= R− (s)−1 · J ↑/↓ (s) · R− (ζ) V1,V2 V1,V2 V1,V2 Γ(V1),Γ(V2) and

−1 −1 ↓/↑ ↑/↓ − −1 0,↑/↓ −1 ↓/↑ J (−s) · R (s) = R (s) · R (s) · J (−s)21 V2,V1 21 V1,V2 V1,V2 V1,V2 V2,V1 · R− (ζ−1) Γ(V2),Γ(V1) 21 = R− (s)−1 · R0,↓/↑ (−s) · J ↓/↑ (−s) V1,V2 V2,V1 21 V2,V1 21 − −1 · R (ζ )21 Γ(V2),Γ(V1) are solutions of the diﬀerence equation

Φ(s +1)= R−(s + 1)−1 · R0,↑/↓ (s) · R−(s) · Φ(s) V1,V2 V1,V2 V1,V2 though, due to the presence of the factors R− (ζ) and R− (ζ−1) , Γ(V1),Γ(V2) Γ(V2),Γ(V1) 21 neither is a canonical left or right solution. Unlike (9.4), however, the twist relation (9.7) should not be considered as a diﬀerence version of the (non–abelian) Drinfeld– Kohno Theorem on n = 2 points for several reasons. (1) As just pointed out, the diﬀerence equations underlying (9.7) are not the qKZ equations determined by R↑/↓ (s), but (a gauge transformation of) V1,V2 the abelian qKZ equations determined by R0,↑/↓ (s). V1,V2 (2) The qKZ equations of Frenkel–Reshetikhin [9] include a dynamical parameter λ ∈ h, and are given by

Φ(s +1)= eλ ⊗ 1 · R↑/↓ (s) · Φ(s) V1,V2 Since the form of the asymptotics of solutions of this equation depends on the regularity of eλ, its monodromy is a meromorphic function of λ, which is conjectured to be equivalent to Rε (ζ), via a λ–dependent gauge Γ(V1)Γ(V2) transformation.

Appendix A. Separation of points

A.1. Let V be a collection of ﬁnite–dimensional representations of Y~(g) such that

(1) C ∈V, and V1 ⊗ V2 ∈V for all V1, V2 ∈V, (2) ∃ V ∈V such that Ker(πV |g)= {0}. The goal of this section is to prove that the elements of V separate points in Y~(g). More generally, the following holds. 44 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Proposition. Let V1,..., Vn be collections of ﬁnite–dimensional representations of ⊗n Y~(g) satisfying the conditions (1) and (2) above. Then, the ideal Jn ⊂ Y~(g) deﬁned by

Jn = Ker(πV1 ⊗···⊗ πVn ) Vi\∈Vi is the zero ideal.

Remark. The analogous statement for U(g[z]) fails. Indeed, take n = 1 and let ∗ Vg be any faithful, ﬁnite-dimensional g-module. Set V = ev (Vg), where ev is the evaluation morphism ev : g[z] → g, f(z) → f(0). ⊗n Then, V is a U(g[z])-module satisfying (2), and V = {V }n∈Z≥0 satisﬁes (1)–(2). However,

zg[z] ⊂ Ker(πV ⊗k ) k\≥0 The proof of the proposition is given in Sections A.2–A.5. In Section A.2, we reduce the task to proving that J1 = {0}. The reduction step is elementary, but is included for the sake of completeness. That the ideal J1 ⊂ Y~(g) vanishes is an unpublished result of V. Drinfeld’s, whose proof in the ~–formal setting has been reproduced in [10, Prop. 8.8]. After recalling relevant background material on co- Poisson Hopf algebras and Lie bialgebras in Sections A.3 and A.4, we explain in Section A.5 how to modify the argument given in [10] to deduce that J1 = {0}.

A.2. Reduction step. Let H1 and H2 be associative algebras over C. Assume in addition that, for each i ∈ {1, 2}, Hi is equipped with a family of representations Vi such that

JVi := Ker(πV )= {0} V\∈Vi The following general result, coupled with a simple induction on n, implies that Proposition A.1 holds, provided J1 = {0}. Lemma. We have

Ker(πV1 ⊗ πV2 )= {0} Vi\∈Vi Proof. The lemma follows easily from the the identity

−1 Ker(πV1 ⊗ πV2 ) = (1 ⊗ πV2 ) Ker(πV1 ) ⊗ End(V2) ,

the assumption JVi = {0}, and the fact that, for any vector spaces M, N and collection of subspaces {Mλ}λ∈Λ ⊂ M, we have the following equality in M ⊗ N:

(Mλ ⊗ N)= Mλ ⊗ N ! λ\∈Λ λ\∈Λ THE MEROMORPHIC R–MATRIX OF THE YANGIAN 45

A.3. We now pause to collect pertinent facts from the theories of co-Poisson Hopf algebras and Lie bialgebras. Fix a Lie algebra a over C, and recall that a co-Poisson algebra structure on U(a) is given by the additional data of a Poisson cobracket δ, i.e. a linear map δ : U(a) → U(a) ∧ U(a) ⊂ U(a)⊗2 satisfying the co-Jacobi and co-Leibniz identities: (1 +(123)+(132)) ◦ δ ⊗ 1 ◦ δ =0 1 ⊗ ∆ ◦ δ = δ ⊗ 1 ◦ ∆ + (1 2) ◦ 1 ⊗ δ ◦ ∆ If in addition δ satisfies the compatibility condition δ(xy)= δ(x)∆(y) + ∆(x)δ(y) then (U(a),δ) is said to be a co-Poisson Hopf algebra. In this case, the cobracket δ is uniquely determined by its restriction δ|a to a, which can be shown to define a Lie bialgebra structure on a. Conversely, every Lie bialgebra cocommutator on a uniquely extends to a co-Poisson Hopf cobracket on U(a): see [2, Prop. 6.2.3]. A perhaps less well-known correspondence, which we shall exploit, takes place at the level of ideals. Recall that a co-Poisson bialgebra ideal of (U(a),δ) is a two-sided ideal J of U(a) satisfying the coideal and co-Poisson conditions ∆(J) ⊂ J ⊗ U(a)+ U(a) ⊗ J, ǫ(J) = 0 δ(J) ⊂ J ⊗ U(a)+ U(a) ⊗ J where ǫ is the counit for U(a). Similarly, a Lie bialgebra ideal of (a,δ|a) is a Lie algebra ideal l of a satisfying δ(l) ⊂ l ⊗ a + a ⊗ l. Lemma. Fix a co-Poisson Hopf algebra structure on U(a). Then the assignment J ⊂ U(a) 7→ J ∩ a ⊂ a (A.1) determines a bijective correspondence between co-Poisson bialgebra ideals J of U(a) and Lie bialgebra ideals l ⊂ a, with inverse l ⊂ a 7→ hli⊂ U(a) (A.2) where hli is the two-sided ideal generated by l. When the underlying cobracket is taken to be trivial (that is, δ ≡ 0), this re- duces to the more familiar assertion (see [21, Prop. 4.8], for example) that (A.1) determines a bijective correspondence between bialgebra ideals of U(a) and Lie algebra ideals in a, with inverse (A.2). The lemma follows readily from this special case by a straightforward verification that (A.1) and (A.2) preserve any additional co-Poisson structure. A.4. We now narrow our focus to a = g[z]. Let us begin by recalling how one passes from Y~(g) to the standard Lie bialgebra structure on g[z], given by (A.4) below. Since Y~(g) is a filtered Hopf deformation of the cocommutative Hopf algebra U(g[z]) (see Section 2.6), the linear map

op ∆Alt := ∆ − ∆ : Y~(g) → Y~(g) ⊗ Y~(g) is a ﬁltered linear map of degree −1. That is, it satisﬁes ⊗2 ∆Alt(Fk(Y~(g))) ⊂Fk−1(Y~(g) ) ∀ k ∈ Z≥0 46 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

⊗2 where F−n(Y~(g) )= {0} for n> 0. We may therefore view it as a filtered map ⊗2 ⊗2 (Y~(g), F•(Y~(g))) → (Y~(g) , F•(Y~(g) )) ⊗2 ⊗2 ⊗2 where F•(Y~(g) ) is the vector space filtration on Y~(g) defined by Fk(Y~(g) ) := ⊗2 Fk−1(Y~(g) ) for all k ∈ Z≥0. By definition, the semiclassical limit of ∆ is the associated graded map

∆Alt δ := gr : U(g[z]) → gr (Y~(g) ⊗ Y~(g)) =∼ U(g[z]) ⊗ U(g[w]) (A.3) ~ F It is a Poisson cobracket which endows U(g[z]) with the structure of a co-Poisson Hopf algebra. Moreover, this co-Poisson structure induces the standard Lie bialgebra structure on g[z], with cocommutator ∼ δ|g[z] : g[z] → g[z] ⊗ g[w] = (g ⊗ g)[z, w] given on f(z) ∈ g[z] by the formula Ω δ(f(z)) = f(z) ⊗ 1+1 ⊗ f(w), g ∈ (g ⊗ g)[z, w] (A.4) z − w We may summarize the above discussion concisely by saying that Y~(g) is a ﬁltered quantization of the Lie bialgebra g[z], equipped with the above cocommutator.

The last ingredient we shall need is the following lemma, which is immediately obtained from Corollary 8.9 of [10] with the help of Lemma A.3.

Lemma. Let ǫU : U(g[z]) → C be the counit. (1) If l ⊂ g[z] is a Lie bialgebra ideal, then l = {0} or l = g[z]. (2) If J ⊂ U(g[z]) is a co-Poisson bialgebra ideal, then J= {0} or J = Ker(ǫU ).

A.5. Proof that J vanishes. Consider now the ﬁltration F•(J ) on J given by Fk(J )= Fk(Y~(g)) ∩ J for all k ∈ Z≥0, and the associated graded ideal

gr(J )= Fk(J )/Fk−1(J ) ⊂ gr(Y~(g)) = U(g[z]) Mk≥0 If x ∈ J is nonzero and k ∈ Z≥0 is minimal such that x ∈Fk(J ), then the image of x in grk(J ) is a nonzero element. Hence, our task is reduced to proving that gr(J )= {0}. Next, note that the condition (1) imposed on V guarantees that J is a bialgebra ideal in Y~(g). As ∆ and ǫ are ﬁltered, it follows that gr(J ) is necessarily a co- Poisson bialgebra ideal of U(g[z]). Using Lemma A.4, we deduce that gr(J )= {0} or gr(J ) = Ker(ǫU ). If gr(J ) = Ker(ǫU ), then g ⊂ gr(J ), and thus

g ⊂ gr0(J )= F0(J ) ⊂ J This contradicts the assumption (2) on V, which guarantees that J intersects g trivially. Hence, we may conclude that gr(J ), and thus J itself, vanishes.

Appendix B. Uniqueness of the universal R–matrix The aim of this section is to give a proof of the uniqueness part of Drinfeld’s theorem (Theorem 1.1). Namely, we assume that we are given two formal series (1) (2) −1 −1 R (s), R (s) ∈ 1+ s (Y~(g) ⊗ Y~(g))[[s ]] THE MEROMORPHIC R–MATRIX OF THE YANGIAN 47 satisfying the hypotheses of Theorem 1.1. That is, for i =1, 2, (i) (i) (i) ∆ ⊗ 1(R (s)) = R13 (s)R23 (s) (i) (i) (i) 1 ⊗ ∆(R (s)) = R13 (s)R12 (s) and, for every a ∈ Y~(g):

op (i) (i) −1 τs ⊗ 1 ◦ ∆ (a)= R (s) · τs ⊗ 1 ◦ ∆(a) · R (s) . Our main tool will be the following lemma. B.1. Lemma. The Lie algebra of primitive elements ∆ Prim (Y~(g)) = {y ∈ Y~(g) : ∆(y)= y ⊗ 1+1 ⊗ y} is equal to g. Proof. We shall again exploit the fact, recalled in Section A.4, that the Yangian Y~(g) provides a ﬁltered quantization of the Lie bialgebra structure (g[z],δ) on g[z] given by (A.4). By (A.3), this means that, for each x ∈ g and k ≥ 0, we have

k op ~ · δ(x.z ) = ∆(y) − ∆ (y) mod Fk−2(Y~(g) ⊗ Y~(g)) (B.1) k for any y ∈Fk(Y~(g)) whose imagey ¯ ∈ grk(Y~(g)) ⊂ U(g[z]) coincides with x.z .

Now let y ∈ Y~(g) be an arbitrary nonzero primitive element. Assume that k ≥ 0 is such that y ∈Fk(Y~(g))\Fk−1(Y~(g)). As ∆ is ﬁltered with gr(∆) recovering the standard coproduct on U(g[z]) (see Section 2.6), we can conclude that the imagey ¯ of y in grk(Y~(g)) ⊂ U(g[z]) is a nonzero, primitive degree k element, and thus of the formy ¯ = x.zk for some x ∈ g.

Using that ∆(y) = ∆op(y), we deduce from (B.1) that δ(x.zk) = 0. On the other hand, by deﬁnition of δ, we have zk − wk 0= δ(x.zk)= [x ⊗ 1, Ω ] z − w g Hence, k = 0 and y ∈ g ⊂ Y~(g).

B.2. Now let n ≥ 1 and X ∈ Y~(g) ⊗ Y~(g) be such that R(1)(s) − R(2)(s) = s−nX + O(s−n−1). We will prove that X = 0.

Comparing the coeﬃcients of s−n on both sides of the cabling identities, we obtain the following

∆ ⊗ 1(X)= X13 + X23

1 ⊗ ∆(X)= X13 + X12 ∆ In other words, X ∈ Prim (Y~(g))⊗2, that is, X ∈ g ⊗ g ⊂ Y~(g) ⊗ Y~(g), by the lemma above.

By the intertwining equation for a ∈ g, we conclude that X ∈ (g ⊗ g)g. Hence X is a scalar multiple of the Casimir tensor: X = cΩg, for some c ∈ C. 48 S.GAUTAM,V.TOLEDANOLAREDO,ANDC.WENDLANDT

Let us now consider the intertwining equation for a = T(h)

ad(T(h) ⊗ 1+1 ⊗ T(h)+ sh ⊗ 1) · R(i)(s)= ~r21(h)R(i)(s)+ ~R(i)r(h)

Take the diﬀerence of the two equations, for i = 1, 2, and compare the coeﬃcient −n+1 of s , to get c[h ⊗ 1, Ωg] = 0, for every h ∈ h. But that means c = 0 and hence X = 0, which is what we wanted to show.

References 1. J. Beck and V. G. Kac, Finite-dimensional representations of quantum affine algebras at roots of unity, J. Amer. Math. Soc. 9 (1996), no. 2, 391–423. MR 1317228 2. V. Chari and A. Pressley, A guide to quantum groups, Cambridge University Press, Cam- bridge, 1994. MR 1300632 3. I. Damiani, La R-matrice pour les algèbres quantiques de type affine non tordu, Ann. Sci. Ecole´ Norm. Sup. (4) 31 (1998), no. 4, 493–523. MR 1634087 4. V. G. Drinfeld, Hopf algebras and the quantum Yang-Baxter equation, Dokl. Akad. Nauk SSSR 283 (1985), no. 5, 1060–1064. MR 802128 5. , A new realization of Yangians and of quantum affine algebras, Dokl. Akad. Nauk SSSR 296 (1987), no. 1, 13–17. MR 914215 6. B. Enriquez, S. Khoroshkin, and S. Pakuliak, Weight functions and Drinfeld currents, Comm. Math. Phys. 276 (2007), no. 3, 691–725. MR 2350435 7. P. I. Etingof and A. A. Moura, On the quantum Kazhdan-Lusztig functor, Math. Res. Lett. 9 (2002), no. 4, 449–463. MR 1928865 8. Michael Finkelberg and Alexander Tsymbaliuk, Shifted quantum affine algebras: integral forms in type A, Arnold Math. J. 5 (2019), no. 2-3, 197–283. MR 4031357 9. I. B. Frenkel and N. Yu. Reshetikhin, Quantum affine algebras and holonomic difference equations, Comm. Math. Phys. 146 (1992), no. 1, 1–60. MR 1163666 10. S. Gautam and V. Toledano Laredo, Yangians and quantum loop algebras, Selecta Math. (N.S.) 19 (2013), no. 2, 271–336. MR 3090231 11. , Yangians, quantum loop algebras, and abelian difference equations, J. Amer. Math. Soc. 29 (2016), no. 3, 775–824. MR 3486172 12. , Meromorphic tensor equivalence for Yangians and quantum loop algebras, Publ. Math. Inst. Hautes Etudes´ Sci. 125 (2017), 267–337. MR 3668651 13. N. Guay, H. Nakajima, and C. Wendlandt, Coproduct for Yangians of affine Kac-Moody algebras, Adv. Math. 338 (2018), 865–911. MR 3861718 14. N. Guay, V. Regelskis, and C. Wendlandt, Equivalences between three presentations of orthog- onal and symplectic Yangians, Lett. Math. Phys. 109 (2019), no. 2, 327–379. MR 3917347 15. D. Hernandez, Drinfeld coproduct, quantum fusion tensor category and applications, Proc. Lond. Math. Soc. (3) 95 (2007), no. 3, 567–608. MR 2368277 16. D. Kazhdan and Y.S. Soibelman, Representations of quantum affine algebras, Selecta Math. (N.S.) 1 (1995), no. 3, 537–595. MR 1366624 17. S. M. Khoroshkin and V. N. Tolstoy, Yangian double, Lett. Math. Phys. 36 (1996), no. 4, 373–402. MR 1384643 18. S. Z. Levendorski˘ı, On generators and defining relations of Yangians, J. Geom. Phys. 12 (1993), no. 1, 1–11. MR 1226802 19. , On PBW bases for Yangians, Lett. Math. Phys. 27 (1993), no. 1, 37–42. MR 1212024 20. D. Maulik and A. Okounkov, Quantum groups and quantum cohomology, Astérisque (2019), no. 408, ix+209. MR 3951025 21. D. Passman, Elementary bialgebra properties of group rings and enveloping rings: an introduction to Hopf algebras, Comm. Algebra 42 (2014), no. 5, 2222–2253. MR 3169701 22. Y. Soibelman, Meromorphic tensor categories, quantum affine and chiral algebras. I, Recent developments in quantum affine algebras and related topics (Raleigh, NC, 1998), Contemp. Math., vol. 248, Amer. Math. Soc., Providence, RI, 1999, pp. 437–451. MR 1745272 23. Y. S. Soibelman, The meromorphic braided category arising in quantum affine algebras, In- ternat. Math. Res. Notices (1999), no. 19, 1067–1079. MR 1725484 24. C. Wendlandt, The formal shift operator on the Yangian double, arxiv:2008.10590 (39 pages). THE MEROMORPHIC R–MATRIX OF THE YANGIAN 49

Department of Mathematics, The Ohio State University, 231 W 18th St. Columbus, OH (USA) Email address: [email protected]

Department of Mathematics, Northeastern University, 360 Huntington Avenue, Boston, MA (USA) Email address: [email protected]

Department of Mathematics, The Ohio State University, 231 W 18th St. Columbus, OH (USA) Email address: [email protected]