Symmetry, Integrability and Geometry: Methods and Applications SIGMA 11 (2015), 091, 41 pages
Populations of Solutions
to Cyclotomic Bethe Equations
Alexander VARCHENKO † and Charles A.S. YOUNG ‡
† Department of Mathematics, University of North Carolina at Chapel Hill,
Chapel Hill, NC 27599-3250, USA
E-mail: anv@email.unc.edu
‡ School of Physics, Astronomy and Mathematics, University of Hertfordshire,
College Lane, Hatfield AL10 9AB, UK
E-mail: charlesyoung@cantab.net
Received June 17, 2014, in final form November 05, 2015; Published online November 14, 2015
http://dx.doi.org/10.3842/SIGMA.2015.091
Abstract. We study solutions of the Bethe Ansatz equations for the cyclotomic Gaudin
model of [Vicedo B., Young C.A.S., arXiv:1409.6937]. We give two interpretations of such
solutions: as critical points of a cyclotomic master function, and as critical points with
cyclotomic symmetry of a certain “extended” master function. In finite types, this yields
a correspondence between the Bethe eigenvectors and eigenvalues of the cyclotomic Gaudin
model and those of an “extended” non-cyclotomic Gaudin model. We proceed to define
populations of solutions to the cyclotomic Bethe equations, in the sense of [Mukhin E.,
Varchenko A., Commun. Contemp. Math. 6 (2004), 111–163, math.QA/0209017], for dia-
gram automorphisms of Kac–Moody Lie algebras. In the case of type A with the diagram
automorphism, we associate to each population a vector space of quasi-polynomials with
specified ramification conditions. This vector space is equipped with a Z2-gradation and
a non-degenerate bilinear form which is (skew-)symmetric on the even (resp. odd) graded
subspace. We show that the population of cyclotomic critical points is isomorphic to the
variety of isotropic full flags in this space.
Key words: Bethe equations; cyclotomic symmetry
2010 Mathematics Subject Classification: 82B23; 32S22; 17B81; 81R12
1 Introduction
Let g be a complex Kac–Moody Lie algebra and σ : g→ g an automorphism of order M ∈ Z≥1.
Let ω ∈ C× be a primitive Mth root of unity. We may choose a Cartan subalgebra h ⊂ g such
that σ(h) = h. We have the canonical pairing 〈·, ·〉 : h∗ ⊗ h → C, and the simple roots αi ∈ h∗
and coroots α∨i ∈ h, where i runs over the set I of nodes of the Dynkin diagram.
Consider the following system of equations in m ∈ Z≥0 variables t = (t1, . . . , tm) ∈ Cm and
labels c = (c(1), . . . , c(m)) ∈ Im:
0 =
M−1∑
k=0
N∑
i=1
〈
σkΛi, α∨c(j)
〉
tj − ωkzi
−
M−1∑
k=0
m∑
i=1
i 6=j
〈
σkαc(i), α
∨
c(j)
〉
tj − ωrti
+
1
tj
(
〈
Λ0, α
∨
c(j)
〉
−
M−1∑
k=1
〈
σkαc(j), α
∨
c(j)
〉
1− ωk
)
, j = 1, . . . ,m, (1.1)
where Λ0,Λ1, . . . ,ΛN ∈ h∗ are weights (with σΛ0 = Λ0) and z1, . . . , zN are non-zero points
2 A. Varchenko and C.A.S. Young
in the complex plane whose orbits, under the action of the cyclic group ωZ, are pairwise dis-
joint.
When σ = id, ω = 1 and Λ0 = 0, these equations reduce to the following well-known set of
equations in mathematical physics:
0 =
N∑
i=0
〈
Λi, α∨c(j)
〉
tj − zi
−
m∑
i=1
i 6=j
〈
αc(i), α
∨
c(j)
〉
tj − ti
, j = 1, . . . ,m. (1.2)
These are the equations for critical points of the master functions [19] which appear in the inte-
gral expressions for hypergeometric solutions to the Knizhnik–Zamolodchikov (KZ) equations.
They are also the Bethe equations of the quantum Gaudin model [1, 5, 18].
The equations (1.1) were introduced (for simple g) in the study of cyclotomic generalizations
of the Gaudin model [26, 27] – see also [3, 21, 22] – as we recall in Section 3 below. Let us
call them the cyclotomic Bethe equations. (Cyclotomic generalizations of the KZ equations were
studied in [2, 4], and appear in, in particular, the representation theory of cyclotomic Hecke
algebras [23].)
It is natural to ask whether the cyclotomic Bethe equations (1.1) can be interpreted as the
equations for critical points of some master function. In the present paper we begin by giving two
different such interpretations. First, they are indeed the critical point equations for a cyclotomic
master function, which we write down in (2.2). But they are also the equations for critical points
with cyclotomic – more precisely Smn(Z/MZ)m – symmetry of what we call an extended master
function, (2.8).
Recall that a master function is specified by a weighted arrangement of hyperplanes: that
is, by a finite collection C of affine hyperplanes in a complex affine space of finite dimension,
together with an assignment of a number a(H) ∈ C to each hyperplane H ∈ C. Indeed, for each
H ∈ C, let `H = 0 be an affine equation for H; then the master function is Φ =
∑
H∈C
a(H) log `H .
The cyclotomic master function corresponds to a hyperplane arrangement in Cm whose hy-
perplanes include ti = ωktj , 1 ≤ i < j ≤ m, for each k ∈ Z/MZ. By contrast, the extended
master function corresponds to a hyperplane arrangement in CmM , but has only those hyper-
planes corresponding to the type A root system, i.e., ti = tj , 1 ≤ i < j ≤ mM , etc. Because the
extended master function is a master function of this standard form, its critical point equations
are the Bethe equations for a certain standard (i.e., non-cyclotomic) Gaudin model, which we
call the extended Gaudin model. This observation leads to our first result: a correspondence
between the spectrum of the cyclotomic Gaudin model and a “cyclotomic” part of the spectrum
of the extended Gaudin model. See Theorem 3.5.
Solutions to the Bethe equations (1.2) form families called populations. Populations were
first introduced in [15, 20], where a generation procedure was given which produces families
of new solutions to the Bethe equations starting from a given solution. A population is then
defined to be the Zariski closure of the set of all solutions to the Bethe equations obtained by
repeated application of this generation procedure, starting from a given solution. It is known
that if g is simple then every population is isomorphic to the flag variety of the Langlands dual
Lie algebra Lg. This was shown in [15] for types A, B, C and in all finite types in [6, 16].
(A population can also be understood as the variety of Miura opers with a given underlying
oper; see [6, 16].)
In the present work our main goal is to initiate the study of cyclotomic populations: popula-
tions of solutions to the equations (1.1).
We formulate in Section 4 a definition of cyclotomic populations for g a general Kac–Moody
Lie algebra and σ any diagram automorphism of g satisfying the linking condition. (We also
place certain restrictions on the weight Λ0; see Section 4.1.) The linking condition [7] states
Populations of Solutions to Cyclotomic Bethe Equations 3
that, for every node i ∈ I, the restriction of the Dynkin diagram to the orbit σZ(i) consists
either of disconnected nodes (in which case i has linking number Li = 1), or of a number of
disconnected copies of the A2 Dynkin diagram (in which case i has linking number Li = 2).
What the linking condition ensures is that it is possible to “fold” the Dynkin diagram by the
automorphism σ. See Section 2.3 and [7].
In Section 4 we define the cyclotomic population to be the Zariski closure of the set of all
cyclotomic critical points obtained by repeated application of a certain “cyclotomic generation
procedure”, starting from a given cyclotomic critical point. So the key ingredient is this gene-
ration procedure. Let us describe it, in outline. There is an “elementary cyclotomic generation”
step associated to each orbit σZ(i). There are two cases: Li = 1 and Li = 2.
First, suppose i ∈ I is a node with linking number Li = 1. A critical point (t, c) is represented
by a tuple of polynomials, y = (yi(x))i∈I , where the roots of the polynomial yi(x), i ∈ I, are
the Bethe variables ts of “colour” i, i.e., those such that c(s) = i. Following [15], one defines
a function of x,
y(i)i (x; c) := yi(x)
∫ x
ξ〈Λ0,α
∨
i 〉Ti(ξ)
∏
j∈I
yj(ξ)
−〈αj ,α∨i 〉dξ + cyi(x), (1.3)
depending on a parameter c ∈ C . Here Ti(x), i ∈ I, are certain functions encoding the
“frame” data, i.e., the points z1, . . . , zN and the weights Λ1, . . . ,ΛN ; see (4.5). The Bethe
equations ensure that y(i)i (x; c) is in fact a polynomial, and moreover that if we consider the new
tuple y(i)(c) in which yi(x) is replaced by y
(i)
i (x; c), then for almost all values of c this new tuple
again represents a solution to the Bethe equations. Call the replacement y 7→ y(i)(c) elementary
generation in direction i. Now suppose the initial tuple y represents a cyclotomic point. That
means
yσj(ωx) ' yj(x), j ∈ I;
see Lemma 4.5. Since the orbit σZ(i) consists of disconnected nodes of the Dynkin diagram, the
operations of elementary generation in the directions σZ(i) commute. By performing each of
them once, in any order, we can arrange to arrive at a new cyclotomic point. See Theorem 4.6.
Next, suppose i ∈ I is a node with linking number Li = 2. Then for every copy of the A2
diagram, with nodes say j and ¯, one must perform the sequence of generation steps j, ¯, j. Doing
this for each copy of A2 in turn, in any order, we can arrange to arrive at a new cyclotomic
point. See Theorem 4.20.
When Li = 2 there is a subtlety coming from our assumptions about the weight at the
origin, Λ0. Throughout Section 4, motivated by [26], we assume that 〈Λ0, α∨i 〉 is non-integral
when Li = 2. That means that the expression (1.3) develops a branch point at the origin. The
upshot is that at certain intermediate steps, the weight at the origin is shifted to si · Λ0, before
eventually being shifted back to Λ0. See Proposition 4.10 and compare [17].
In either case, Li = 1 or Li = 2, we write y(i,σ)(c) for the tuple of polynomials representing the
new cyclotomic critical point. It depends on a single parameter c. The replacement y 7→ y(i,σ)(c)
is the elementary cyclotomic generation, in the direction of the orbit σZ(i).
To a critical point (t, c) represented by a tuple of polynomials y one can associate a weight Λ∞.
See (2.4) and (4.10). For fixed Λ0,Λ1, . . . ,ΛN , we may regard Λ∞ as encoding the number of
roots ts of each “colour” i ∈ I, i.e., the degrees of the polynomials yi(x). It is known that
Λ∞(y(i)(c)) is equal either to Λ∞(y) or to si · Λ∞(y), where si · denotes the shifted action of
the Weyl reflection in root αi. See [15]. We have an analogous statement in the cyclotomic
case. Namely, there is a “folded” Weyl group W σ with generators sσi . See Section 2.3. And we
show that Λ∞(y(i,σ)(c)) is equal either to Λ∞(y) or to sσi ·Λ∞(y). For the precise statement see
Theorems 4.6 and 4.20.
4 A. Varchenko and C.A.S. Young
We proceed in Section 5 to treat in detail the case of type A with the diagram automorphism.
Recall first from [15] the structure of populations in type AR, R ∈ Z≥1, for the master
functions associated to marked points z1, . . . , zN and integral dominant weights Λ1, . . . ,ΛN .
In that setting, every population of critical points is isomorphic to a variety of full flags in
a certain (R + 1)-dimensional vector space K of polynomials. The ramification points of K
are z1, . . . , zN and ∞, and the ramification data at these points are specified by the weights
Λ1, . . . ,ΛN and an integral dominant weight Λ˜∞. Given a full flag F = {0 = F0 ⊂ F1 ⊂ F2 ⊂
· · · ⊂ FR+1 = K} in K, pick any basis (ui(x))
R+1
i=1 of polynomials adjusted to this flag, i.e., such
that Fk = spanC(u1(x), . . . , uk(x)). Then define a tuple of functions y
F = (yFk (x))
R
k=1 by
yFk (x) = Wr(u1(x), . . . , uk(x))/
(
T k−11 (x)T
k−2
2 (x) . . . Tk−1(x)
)
,
where – as in (1.3) above – the (Ti(x))Ri=1 are functions encoding the “frame” data z1, . . . , zN
and Λ1, . . . ,ΛN , and where Wr(u1(x), . . . , uk(x)) denotes the Wronskian determinant. The
ramification properties of K ensure that the yFk (x) are in fact polynomials. Moreover the map
F 7→ yF is an isomorphism of varieties from the variety of full flags in K to the population
associated with K. The space K is the kernel of a certain linear differential operator D of order
R + 1 (essentially a type A oper). This operator D can be defined in terms of the (Ti(x))Ri=1
together with the polynomials (yi(x))Ri=1 of (any) point in the population. (See Section 5.4.)
Now let us discuss how the picture changes in our present setting. For us, the weight at the
origin Λ0 need not be integer dominant. We assume it satisfies weaker assumptions given in (5.1).
These assumptions mean that we are led to consider vector spaces K of quasi-polynomials:
that is, polynomials in x
1
2 . The local behaviour of these quasi-polynomials near the origin is
encoded in Λ0. The remaining ramification points are z1, . . . , zN , −z1, . . . ,−zN , and ∞. See
Definition 5.2.
The space of quasi-polynomials K admits a natural Z2 gradation K = KO ⊕ KSp. We call
flags which respect this gradation decomposable. Decomposable full flags are classified by their
type; see Section 5.3. In particular the flags F ∈ FLS(K) of a certain preferred type S, (5.9), are
sent to polynomials under the map F 7→ yF . This map of varieties FLS(K) → P(C[x])R is an
isomorphism onto its image. The cyclotomic population is then the set of cyclotomic tuples in
this image, i.e., the set of tuples yF , F ∈ FLS(K), such that yi(x) ' yR+1−i(−x), i = 1, . . . , R.
The question is: which flags in FLS(K) map to cyclotomic tuples?
To answer this question we introduce the notion of a cyclotomically self-dual space of quasi-
polynomials. The space K has a natural dual space K† of quasi-polynomials – see Section 5.5 –
and we say K is cyclotomically self-dual if for all v(x) ∈ K, v(−x) ∈ K†. (Compare the very
similar notion of a self-dual space of polynomials in [15].) We show that a sufficient condition
for K to be cyclotomically self-dual is that there exists at least one full flag F in K such that yF
is cyclotomic (Theorem 5.14). If K is cyclotomically self-dual then it admits a canonical non-
degenerate bilinear form B. We show that, for all full flags F in K, the tuple yF is cyclotomic
if and only if F is isotropic with respect to B (Theorem 5.17).
Therefore the cyclotomic population is isomorphic to the variety FL⊥S (K) of isotropic flags
of type S in K. The bilinear form B is symmetric on KO and skew-symmetric on KSp, and
these subspaces are mutually orthogonal with respect to B (Theorem 5.23). Hence this variety
FL⊥S (K) is isomorphic to the direct product of spaces of isotropic flags FL
⊥(KSp)× FL⊥(KO).
2 Master functions and cyclotomic symmetry
2.1 Kac–Moody algebras
Let I be a finite set of indices and A = (ai,j)i,j∈I a generalized Cartan matrix, i.e., ai,i = 2
and ai,j ∈ Z≤0 whenever i 6= j, with ai,j = 0 if and only if aj,i = 0. Let g := g(A) be the
Populations of Solutions to Cyclotomic Bethe Equations 5
corresponding complex Kac–Moody Lie algebra [11, Section 1], h ⊂ g a Cartan subalgebra, and
g = n− ⊕ h⊕ n+
a triangular decomposition. Let αi ∈ h∗, α∨i ∈ h, i ∈ I be collections of simple roots and coroots
respectively. We have dim h = |I|+ dim kerA = 2|I| − rankA. By definition,
〈αi, α
∨
j 〉 = aj,i,
where 〈·, ·〉 : h∗ ⊗ h→ C is the canonical pairing.
We assume that A is symmetrizable, i.e., there exists a diagonal matrix D = diag(di)i∈I ,
whose entries are coprime positive integers, such that the matrix B = DA is symmetric. Let
(·, ·) be the associated symmetric bilinear form on h∗. We have (αi, αj) = diai,j and
〈λ, α∨i 〉 = 2(λ, αi)/(αi, αi) for all λ ∈ h
∗.
The form (·, ·) is non-degenerate. Therefore it gives an identification h ∼=C h∗ and hence a
non-degenerate symmetric bilinear form on h which we also write as (·, ·).
Let P := {λ ∈ h∗ : 〈λ, α∨i 〉 ∈ Z} be the integral weight lattice and P+ := {λ ∈ h
∗ : 〈λ, α∨i 〉 ∈
Z≥0} the set of dominant integral weights.
Let W ⊂ End(h∗) be the Weyl group. It is generated by the reflections si, i ∈ I, given by
si(λ) := λ− 〈λ, α∨i 〉αi, λ ∈ h
∗.
Let ρ ∈ h∗ be a vector such that 〈ρ, α∨i 〉 = 1 for i ∈ I. We use · to denote the shifted action
of the Weyl group, i.e.,
s · λ := w(λ+ ρ)− ρ, s ∈W, λ ∈ h∗.
2.2 Diagram automorphism
Suppose σ is an automorphism of the Dynkin diagram [11, Section 4.7] of A. That is, σ is
a permutation of the index set I such that
aσi,σj = ai,j .
Let M be the order of σ and let ω ∈ C× be a primitive Mth root of unity.
To such a permutation is associated a diagram automorphism g → g of the Kac–Moody Lie
algebra [7], which we shall also write as σ. We have
σEi = Eσi, σFi = Fσi, σα
∨
i = α
∨
σi, i ∈ I,
where Ei ∈ n, Fi ∈ n−, i ∈ I, are a set of Chevalley generators of [g, g]. This defines σ on
the derived subalgebra [g, g] of g. For the action of σ on the derivations, i.e., on a complement
of [g, g] in g, see [7, Section 3.2]. This action may be chosen to ensure that σ : g→ g has order M
and respects the bilinear form (·, ·) on h:
(σX, σY ) = (X,Y ) for all X,Y ∈ h.
The action of σ on h∗ is defined by σλ := λ ◦ σ−1 so that 〈σλ, σX〉 = 〈λ,X〉 for all λ ∈ h∗,
X ∈ h. Note that then σαi = ασi for all i ∈ I.
Let gσ ⊂ g be the Lie subalgebra of elements invariant under σ. We have
gσ = nσ− ⊕ h
σ ⊕ nσ+
with nσ± = g
σ ∩ n± and hσ = gσ ∩ h.
6 A. Varchenko and C.A.S. Young
2.3 The linking condition and the folded diagram
For any i ∈ I let
Mi :=
∣
∣
{
i, σi, σ2i, . . . , σM−1i
}∣
∣
be the length of the orbit of the node i under the automorphism σ of the Dynkin diagram A.
Define
Li := 1−
Mi−1∑
k=1
aσki,i.
Note that Li ≥ 1. Following [7], we say that σ obeys the linking condition if and only if
Li ≤ 2 for all i ∈ I. (2.1)
To understand the meaning of this condition, consider the restriction of the Dynkin diagram
to the orbit of the node i. If Li = 1 then this induced subgraph has no edges at all. If Li = 2
then it consists of Mi/2 disconnected copies of the type A2 Dynkin diagram.
Remark 2.1. If A is of finite type, then all diagram automorphisms obey the linking condition.
Moreover, in all finite types except A2n, n ∈ Z≥1, we in fact have Li = 1 for every node i: that
is, no two distinct nodes in the same σ-orbit are ever linked by an edge of the Dynkin diagram.
In type A2n the non-trivial diagram automorphism gives Li = 2 for i ∈ {n, n + 1} and Li = 1
otherwise:
n− 1 n n+ 1 n+ 2 2n1
Remark 2.2. If A is of affine type then all diagram automorphisms obey the linking condition
with the following exception. In type A(1)n , n ∈ Z≥2, let R be a generator of the cyclic sub-
group Cn+1 of the full automorphism group of the Dynkin diagram (which is the dihedral
group Dn+1). Then R does not obey the linking condition. Indeed, the R-orbit of any node i is
the whole diagram, and Li = 1 + n.
Given any diagram automorphism satisfying the linking condition it is possible to define
a folded Dynkin diagram. Let us make a choice of subset
Iσ ⊆ I
consisting of exactly one representative of each σ-orbit. Then the Cartan matrix Aσ = (aσi,j)i,j∈Iσ
of the folded diagram is given by
aσi,j = Li
Mi−1∑
k=0
aσki,j .
Remark 2.3. Compare Section 3.3 of [7], noting that our convention aj,i = 〈αi, α∨j 〉 differs from
that of [7].
Lemma 2.4 ([7]). If σ obeys the linking condition then Aσ (and its transpose) is a symmetrizable
Cartan matrix whose type (finite, affine, or indefinite) is the same as that of A.
Populations of Solutions to Cyclotomic Bethe Equations 7
For each i ∈ Iσ let us define also
α∨,σi := Li
Mi−1∑
k=0
α∨σki and E
σ
i :=
Mi−1∑
k=0
Eσi, F
σ
i := Li
Mi−1∑
k=0
Fσi.
Then we have
[
Eσi , F
σ
j
]
= δi,jα
∨,σ
i ,
[
α∨,σi , E
σ
j
]
= Eσj a
σ
j,i,
[
α∨,σi , F
σ
j
]
= −F σj a
σ
j,i i, j ∈ Iσ.
Thus α∨,σi , E
σ
i , F
σ
i , i ∈ Iσ generate a copy of (the derived subalgebra of) the Kac–Moody Lie
algebra g(Aσ) inside gσ := {X ∈ g : σX = X}. Next, for all i ∈ Iσ, if we let
ασi :=
Li
Mi
Mi−1∑
k=0
ασki ∈ h
∗
then 〈ασi , α
∨,σ
j 〉 = a
σ
j,i. Define W
σ to be the group generated by the elements sσi ∈ End(h
∗) given
by
sσi (λ) := λ− 〈λ, α
∨,σ
i 〉α
σ
i , i ∈ Iσ.
Lemma 2.5. W σ is a subgroup of W . Indeed, we have
sσi =



Mi−1∏
k=0
sσki, Li = 1,


Mi/2−1∏
k=0
sσki




Mi/2−1∏
k=0
sσk+Mi/2i




Mi/2−1∏
k=0
sσki

 , Li = 2.
2.4 The cyclotomic master function
Let Λ = (Λi)Ni=1 be a collection of N ∈ Z≥0 integral dominant weights Λi ∈ P+. Let z = (zi)
N
i=1
be a collection of nonzero points zi ∈ C× such that ωZzi ∩ ωZzj = ∅ whenever i 6= j. We shall
call Λi the weight at zi.
In addition, we pick a weight Λ0 ∈ hσ,∗. We call Λ0 the weight at the origin.
Let c = (c(j))mj=1 be an m-tuple of elements of I, and introduce variables t = (tj)
m
j=1. We
shall say that tj is a variable of colour c(j).
We define the cyclotomic master function Φ = Φg,σ(t; c; z; Λ,Λ0) associated to these data to
be
Φ :=
N∑
i=1
(
1
2
M−1∑
k=1
(
Λi, σ
kΛi
)
+ (Λi,Λ0)
)
log zi +
M−1∑
k=0
∑
1≤i<j≤n
(
Λi, σ
kΛj
)
log
(
zi − ω
kzj
)
−
M−1∑
k=0
N∑
i=1
m∑
j=1
(
αc(j), σ
kΛi
)
log
(
tj − ω
kzi
)
+
M−1∑
k=0
∑
1≤i<j≤m
(
αc(i), σ
kαc(j)
)
log
(
ti − ω
ktj
)
+
m∑
i=1
(
1
2
M−1∑
k=1
(
αc(i), σ
kαc(i)
)
− (αc(i),Λ0)
)
log ti. (2.2)
8 A. Varchenko and C.A.S. Young
A point t with complex coordinates is called a critical point of the cyclotomic master function
if
∂Φ
∂ti
= 0, i = 1, . . . ,m,
or equivalently (in view of Lemma 2.6 below) if the following equations are satisfied:
0 =
M−1∑
k=0
N∑
i=1
(
αc(j), σ
kΛi
)
tj − ωkzi
−
M−1∑
k=0
m∑
i=1
i 6=j
(
αc(j), σ
kαc(i)
)
tj − ωrti
+
1
tj
(
−
M−1∑
k=1
(
αc(j), σ
kαc(j)
)
1− ωk
+ (αc(j),Λ0)
)
(2.3)
for j = 1, . . . ,m. Call this system of equations the cyclotomic Bethe equations.
Lemma 2.6. For any λ ∈ h∗,
M−1∑
k=1
(λ, σkλ)
1− ωk
=
1
2
M−1∑
k=1
(
(λ, σkλ)
1− ωk
+
(λ, σkλ)
1− ω−k
)
=
1
2
M−1∑
k=1
(λ, σkλ).
Define Λ∞, the weight at infinity, to be
Λ∞ := Λ0 +
M−1∑
k=0
N∑
i=1
Λσk(i) −
M−1∑
k=0
m∑
i=1
ασkc(i). (2.4)
The group Sm acts on pairs of m-tuples (t, c) by permuting indices:
ρ.(t, c) =
((
tρ−1(1), . . . , tρ−1(m)
)
,
(
c
(
ρ−1(1)
)
, . . . , c
(
ρ−1(m)
)))
.
The group Z/MZ acts on pairs (t, c) ∈ C×I by k.(t, c) = (ωkti, σkc). This gives rise to an action
of the wreath product Sm o (Z/MZ) := Sm n (Z/MZ)m on pairs of tuples (t, c) ∈ Cm × Im.
Lemma 2.7. Up to an additive constant, the cyclotomic master function Φ is invariant under
the pull-back of this action of Sm o (Z/MZ). In particular, if t is a critical point of Φ(t; c)
then X.t is a critical point of Φ(X.t;X.c), for all X ∈ Sm o (Z/MZ).
2.5 The extended master function
The equations (2.3) admit another, closely related, interpretation. Recall the definition of the
(usual) master function [19]. Namely, let Λ˜ = (Λ˜i)N˜i=0 be a collection of N˜ + 1 ∈ Z≥0 weights
Λ˜i ∈ h∗, and let z˜ = (z˜i)N˜i=0 be a collection of nonzero points z˜i ∈ C
×. Pick m˜ ∈ Z≥0, let
c = (c(j))m˜j=1 be an m˜-tuple of elements of I and introduce variables t = (tj)
m˜
j=1. The master
function associated to these data is
Φ˜ :=
∑
0≤i<j≤N˜
(
Λ˜i, Λ˜j
)
log(z˜i − z˜j)−
N˜∑
i=0
m˜∑
j=1
(
αc(j), Λ˜i
)
log(tj − z˜i)
+
∑
1≤i<j≤m˜
(
αc(i), αc(j)
)
log(ti − tj). (2.5)
Populations of Solutions to Cyclotomic Bethe Equations 9
It is a function of the variables t, depending on the parameters c, z˜ and Λ˜. The critical points
of the master function are those points t with complex coordinates such that ∂Φ˜/∂tj = 0 for
j = 1, . . . , m˜, i.e., those points such that the following equations are satisfied:
0 =
N˜∑
i=0
(αc(j), Λ˜i)
tj − z˜i
−
m˜∑
i=1
i 6=j
(αc(j), αc(i))
tj − ti
, j = 1, . . . , m˜. (2.6)
In this paper we are concerned with the following special case. Let N˜ = NM , choose (z˜i)NMi=0 to
be
z˜0 = 0, z˜k+Mi = ω
kzi, k = 0, 1, . . . ,M − 1, i = 1, . . . , N, (2.7a)
and choose the weights at these points to be
Λ˜0 = Λ0, Λ˜k+Mi = σ
kΛi, (2.7b)
where zi, Λi, i = 1, . . . , N , and Λ0 are as in Section 2.4. We call the master function in this case
the extended master function, Φ̂ = Φ̂g,σ(t; c; z; Λ; Λ0). It is given by
Φ̂ :=
M−1∑
k=0
N∑
i=1
(
Λ0, σ
kΛi
)
log
(
−ωkzi
)
+
M−1∑
k, l=0
∑
1≤i<j≤n
(
σkΛi, σ
lΛj
)
log
(
ωkzi − ω
lzj
)
+
∑
0≤k<l≤T−1
N∑
i=1
(
σkΛi, σ
lΛi
)
log
(
ωk − ωl
)
zi −
m˜∑
j=1
(αc(j),Λ0) log(tj)
−
M−1∑
k=0
N∑
i=1
m˜∑
j=1
(
αc(j), ω
kΛi
)
log
(
tj − ω
kzi
)
+
∑
1≤i<j≤m˜
(αc(i), αc(j)) log(ti − tj), (2.8)
and the critical point equations (2.6) take the form
0 =
M−1∑
k=0
N∑
i=1
(
αc(j), σ
kΛi
)
tj − ωkzi
+
(αc(j),Λ0)
tj
−
m˜∑
i=1
i 6=j
(αc(j), αc(i))
tj − ti
, j = 1, . . . , m˜. (2.9)
The group Sm˜ acts on pairs of m˜-tuples (t, c) by permuting indices:
ρ.(t, c) =
((
tρ−1(1), . . . , tρ−1(m˜)
)
,
(
c
(
ρ−1(1)
)
, . . . , c
(
ρ−1(m˜)
)))
. (2.10)
Lemma 2.8. Any master function of the form (2.5) is invariant under the pull-back of this
action of Sm˜. In particular the extended master function (2.8) is invariant.
Let us call a point (t, c) ∈ Cm˜×Im˜ a cyclotomic point if we have m˜ = Mm for some m ∈ Z≥0
and, by acting with some permutation in Sm˜, we can arrange that
ti+mk = ω
kti c(i+mk) = σkc(i), i = 1, . . . ,m, k = 0, . . . ,M − 1. (2.11)
Lemma 2.9. This point (ti)m˜i=1 is a critical point of the extended master function if and only
if (ti)mi=1 is a critical point of the cyclotomic master function, i.e., (ti)
m
i=1 obeys (2.3).
Proof. Given (2.11), the equation (2.9) for tj is nothing but the corresponding equation in (2.3)
and the equation for tj+km, k = 1, . . . ,M − 1, is actually the same equation up to an overall
factor of ω−k. (To see this one must use the compatibility of σ with the inner product: (σx, y) =
(x, σ−1y).) 
Thus, the cyclotomic Bethe equations (2.3) are also the equations for cyclotomic critical
points of the extended master function.
10 A. Varchenko and C.A.S. Young
3 Gaudin models and the Bethe ansatz equations
Our first result, Theorem 3.5, concerns the relationship between critical points and the eigenval-
ues of Gaudin Hamiltonians. Suppose, for this section only, that the Cartan matrix is of finite
type, i.e., that g is semisimple, and that σ is an automorphism of g of order M > 1. Recall [8, 9]
that the quadratic Gaudin Hamiltonians are the following N˜ + 1 elements of U(g)⊗(N˜+1):
H˜(i) :=
N˜∑
j=0
j 6=i
dim g∑
a=1
Ia(i)I(j)a
z˜i − z˜j
, i = 0, 1, . . . , N˜ ,
where Ia, a = 1, . . . ,dim g, is a basis of g, Ia is the dual basis with respect to the non-degenerate
invariant bilinear form (·, ·) : g× g→ C, and we write X(i) for X acting in the ith tensor factor.
(For convenience we number these factors starting from 0.)
For Λ ∈ h∗, let MΛ denote the Verma module over g with highest weight Λ, MΛ :=
Indgh⊕n+ CvΛ. Let us represent the H˜
(i) as linear maps in End
(⊗N˜
i=0MΛ˜i
)
. Then the following
can be shown using the techniques of the Bethe Ansatz.
Theorem 3.1 ([1, 18]). To any critical point t of the master function Φ˜, i.e., to any solution
to the equations (2.6), there corresponds a simultaneous eigenvector ψ˜t of the linear operators
H˜(i) ∈ End
(⊗N˜
i=0MΛ˜i
)
. For each i = 0, . . . , N˜ the eigenvalue of H˜(i) on ψ˜t is
E˜(i) :=
∂Φ˜
∂z˜i
=
N˜∑
j=0
j 6=i
(Λ˜i, Λ˜j)
z˜i − z˜j
−
m˜∑
j=1
(Λi, αc(j))
z˜i − tj
. (3.1)
The eigenvector ψ˜t is given explicitly by
ψ˜t =
∑
n∈Pm˜,N˜+1
N˜⊗
i=0
Fc(ni1)Fc(ni2) · · ·Fc(nipi−1)
Fc(nipi )
vΛ˜i
(
wni1 − wni2
)
· · ·
(
wnipi−1
− wnipi
)(
wnipi
− zi
) , (3.2)
where the sum n ∈ Pm˜,N˜+1 is over ordered partitions of the labels {1, . . . , m˜} into N˜ + 1 parts.
(The fact that this simultaneous eigenvector is nonzero is proved for g = sln nondegenerate
critical points in [14], for g = sln isolated critical points in [13], and for semisimple g and isolated
critical points in [25]. See also [24].)
In [26]1, B. Vicedo and one of the present authors defined cyclotomic Gaudin Hamiltonians.
The quadratic cyclotomic Gaudin Hamiltonians are the elements of U(g)⊗N given by
Hi :=
M−1∑
p=0
N∑
j=1
j 6=i
dim g∑
a=1
Ia(i)σpI(j)a
zi − ω−pzj
+
1
zi
M−1∑
p=1
dim g∑
a=1
Ia(i)σpI(i)a
(1− ωp)
, i = 1, . . . , n. (3.3)
Remark 3.2. These Hamiltonians can be understood in a number of ways. Physically, one
thinks of them as describing the dynamics of a “long-range spin chain” in which the “spin”
at zi interacts not only directly with the other spins at the points zj , j 6= i, but also with their
images under rotations of the spectral plane [3]. At the level of the Lax matrix, this corresponds
1In [26] σ : g → g is allowed to be any automorphism commuting with the Cartan involution, not necessarily
a diagram involution. A posteriori the Bethe equations and energy eigenvalues depend on the inner part of σ only
through the definition of Λ0, (3.5).
Populations of Solutions to Cyclotomic Bethe Equations 11
to replacing the usual rational skew-symmetric solution to the classical Yang–Baxter equation,
r(u, v) = Ia⊗ Ia/(u−v), by a certain non-skew-symmetric solution – see [21, 22] and discussion
in [26]. The motivation for such models comes in part from physics, where in certain important
cases the Lax matrix has cyclotomic symmetry in the spectral variable [12, 28].
Let us assign to the point zi the Verma module MΛi , Λi ∈ h
∗. In other words, let us represent
the Hamiltonians (3.3) as linear maps
H(i) ∈ End
(
N⊗
i=1
MΛi
)
, i = 1, . . . , N. (3.4)
Let (in this section, Section 3) Λ0 ∈ hσ,∗ be the weight given by
Λ0(h) :=
M−1∑
r=1
trn(σ−r adh)
1− ωr
. (3.5)
Theorem 3.3 ([26]). To any critical point of the cyclotomic master function, i.e., to any so-
lution t to the equations (2.3), there corresponds a simultaneous eigenvector ψt of the linear
operators H(i), i = 1, . . . , N . The eigenvalue of H(i) on ψt is
E(i) :=
∂Φ
∂zi
=
N∑
j=1
j 6=i
M−1∑
s=0
(Λi, σsΛj)
zi − ωszj
−
m∑
j=1
M−1∑
s=0
(Λi, σsαc(j))
zi − ωstj
+
1
zi
(
(Λi,Λ0) +
M−1∑
s=1
(Λi, σsΛi)
1− ωs
)
. (3.6)
The explicit form of the eigenvector ψt is
ψt = (3.7)
=
∑
n∈Pm,N
(k1,...,km)∈ZmM
N⊗
i=1
σˇ
kni1
(
Fc(ni1)
)
σˇ
kni2
(
Fc(ni2)
)
· · · σˇ
knipi−1
(
Fc(nipi−1)
)
σˇ
knipi
(
Fc(nipi )
)
vΛi
(
ω
kni1wni1 − ω
kni2wni2
)
· · ·
(
ω
knipi−1wnipi−1
− ω
knipiwnipi
)(
ω
knipiwnipi
− zi
)
,
where σˇ(X) := ωσ(X).
(It is an interesting open problem to determine under what circumstances the vector ψt is
non-zero.)
On the other hand, consider the (usual) quadratic Gaudin Hamiltonians in the special
case (2.7). We refer to this situation as the extended Gaudin model, and write H˜(i) as H(i)ext.
Note that
H(i)ext ∈ End
(
MΛ0 ⊗
M−1⊗
k=0
N⊗
i=1
MσkΛi
)
, i = 0, 1, . . . , NM. (3.8)
The following is then a corollary of Theorem 3.1.
Corollary 3.4. To any critical point of the cyclotomic master function, i.e., to any solution t
to the equations (2.3), there corresponds a simultaneous eigenvector of the linear operators H(i)ext,
i = 0, 1, . . . , nM , such that H(0)ext has eigenvalue zero and, for each k = 0, . . . ,M − 1 and
i = 1, . . . , N , the eigenvalue of H(k+Mi)ext is given by ω
−kEi with Ei as in (3.6).
12 A. Varchenko and C.A.S. Young
Proof. Let t be the corresponding (by Lemma 2.9) cyclotomic critical point of the extended
master function Φ̂. Then the result is a special case of Theorem 3.1, by substituting (2.7)
and (2.11) into (3.1). (To see that H(0)ext has eigenvalue zero note that
N∑
i=1
M−1∑
s=0
(Λ0, σsΛi)
0− ωszi
−
m∑
j=1
M−1∑
s=0
(Λ0, σsαc(j))
0− ωstj
= 0,
because
M−1∑
s=0
ω−sσ−sΛ0 = Λ0
M−1∑
s=0
ω−s = 0 since σΛ0 = Λ0 and M > 1.) 
In summary, we have the following observation.
Theorem 3.5. To any critical point of the cyclotomic master function there corresponds both
a simultaneous eigenvector (3.7) of the Hamiltonians H(i) of the cyclotomic Gaudin model and
a simultaneous eigenvector (3.2) of the Hamiltonians H(i)ext of the extended Gaudin model, i =
1, . . . , n, with the corresponding eigenvalues equal and in both cases being given by (3.6).
Remark 3.6. The operators H(i) and H(i)ext are acting in different spaces, (3.4) and (3.8) re-
spectively. It would be interesting to relate these operators by some means independent of the
Bethe ansatz.
4 Cyclotomic generation procedure
In [15, 20] a procedure was introduced which generates new critical points of master functions
starting from a given initial critical point. There is an “elementary generation” step associated
to each i ∈ I. The Zariski closure of the collection of all critical points obtained by recursively
applying elementary generations in all possible ways is called the “population” to which the
initial critical point belongs.
The extended master functions, (2.8) above, are master functions of the standard form (un-
like the cyclotomic master functions (2.2)). Modulo subtleties coming from the fact that the
weight Λ0 at the origin need not be dominant integral, that means the generation procedure can
be applied.
In this section we describe this generation procedure and go on to show how, given a cyclo-
tomic critical point, one can obtain new cyclotomic critical points by applying the elementary
generation steps in certain carefully chosen combinations. The resulting collections of cyclotomic
critical points will be called “cyclotomic populations”.
4.1 Conditions on Λ0
In the remainder of the paper we assume that σ is a diagram automorphism obeying the linking
condition (2.1). That means for each i ∈ I, either Li = 1 or Li = 2.
In addition, in this section, Section 4, we place the following conditions on the weight
Λ0 ∈ hσ,∗.
For each i ∈ I such that Li = 1, we suppose that
〈Λ0, α
∨
i 〉 ∈ Z≥0 (4.1)
and
〈Λ0, α
∨
i 〉+ 1 ≡ 0 mod M/Mi. (4.2)
For each i ∈ I such that Li = 2, we suppose that
2〈Λ0, α
∨
i 〉+ 1 ∈ Z≥0. (4.3)
Populations of Solutions to Cyclotomic Bethe Equations 13
Remark 4.1. One can verify that these conditions are satisfied by the weight Λ0 of (3.5) in
the case of diagram automorphisms of finite-type Dynkin diagrams. Our assumptions on Λ0 in
the treatment of type AR in Section 5 below are weaker.
4.2 Tuples of polynomials
To any pair (t; c) with t ∈ Cm˜ and c ∈ Im˜, we may associate a tuple of polynomials y =
(y1(x), . . . , yr(x)), given by
yi(x) :=
m˜∏
j=1
c(j)=i
(x− tj), i ∈ I. (4.4)
We say that this tuple y represents the pair (t; c). We consider each coordinate yi(x) only up
to multiplication by a non-zero complex number, since we are only concerned with their zeros.
So the tuple y defines a point in the direct product P(C[x])|I| of |I| copies of the projective
space P(C[x]), where C[x] is the vector space of complex polynomials in x.
Conversely, given any y ∈ P(C[x])|I| we may extract the pair (t; c) ∈ Cm˜× Im˜ such that (4.4)
holds. This pair is unique up to permutation by an element of Sm˜; see (2.10).
Define Ti(x), i ∈ I, to be
Ti(x) :=
N∏
s=1
M−1∏
k=0
(
x− ωkzs
)〈σkΛs,α∨i 〉. (4.5)
We say that a tuple of polynomials y = (yi(x))i∈I ∈ P(C[x])|I| is generic (with respect
to (Ti(x))i∈I) if for each i ∈ I, yi(x) has no root in common with Ti(x), or with any yj(x),
j ∈ I\{i}, such that 〈αj , α∨i 〉 6= 0.
Note that if y represents a critical point of the extended master function Φ̂(t; c; z; Λ), (2.8),
i.e., its roots obey (2.9), then the tuple y must be generic. (Indeed, if (2.9) holds then in
particular each summand on the left hand side of (2.9) must have non-zero denominator. By
definition that implies that the corresponding tuple is generic.)
4.3 Elementary generation: the Li = 1 case
Throughout this subsection, we suppose i ∈ I is such that Li = 1. That means that the simple
roots ασki, i = 1, . . . ,Mi, are mutually orthogonal. Equivalently it means that the reflections
sσki ∈W , i = 1, . . . ,Mi, are mutually commuting.
Let y(i)i (x) be of the form
y(i)i (x) = yi(x)
∫ x
ξ〈Λ0,α
∨
i 〉Ti(ξ)
∏
j∈I
yj(ξ)
−〈αj ,α∨i 〉dξ, (4.6)
so that y(i)i (x) is a solution to the equation
Wr(yi(x), y
(i)
i (x)) = x
〈Λ0,α∨i 〉Ti(x)
∏
j∈I\{i}
yj(x)
−〈αj ,α∨i 〉, (4.7)
where Wr(f(x), g(x)) := f(x)g′(x)− f ′(x)g(x) denotes the Wronskian determinant.
Proposition 4.2. If y represents a critical point then y(i)i (x) is a polynomial.
14 A. Varchenko and C.A.S. Young
Proof. We have 〈Λ0, α∨i 〉 ∈ Z≥0 as in (4.1), and for each s ∈ {1, . . . , N}, Λs is integral dominant
so 〈Λs, α∨i 〉 ∈ Z≥0. So the integrand is a rational function with poles at most at the points tp,
p ∈ {1, . . . ,m}, for which c(p) = i. Consider such a point tp. Note that
∂
∂x
log x〈Λ0,α
∨
i 〉Ti(x)(x− tp)
2
∏
j∈I
yj(x)
−〈αj ,α∨i 〉
=
M−1∑
k=0
N∑
i=1
〈σkΛi, α∨c(p)〉
x− ωkzi
+
〈Λ0, α∨c(p)〉
x
−
m˜∑
i=1
i 6=p
〈αc(i), αc(p)〉
x− ti
. (4.8)
This vanishes at x = tp by virtue of the critical point equations (2.9). It follows that the residue
of the integrand at tp vanishes: indeed, this residue is


∂
∂x
x〈Λ0,α
∨
i 〉Ti(x)(x− tp)
2
∏
j∈I
yj(x)
−〈αj ,α∨i 〉


x=tp
,
which vanishes if (4.8) vanishes. This shows that y(i)i (x) is an entire function. It is of polynomial
growth for large x. Therefore it is a polynomial. 
If y(i)i (x) is any solution to (4.7) then so too is y
(i)
i (x) + cyi(x) for any c ∈ C.
Thus, given any tuple y representing a critical point we have, for each value of a parameter
c ∈ C, a new tuple of polynomials y(i), obtained from the tuple y by replacing yi(x) with
y(i)i (x) + cyi(x). We say y
(i) is obtained from y by generation in the ith direction, and we
call y(i) the immediate descendant of y in the ith direction.
Proposition 4.3 ([15]). The tuple of polynomials y(i) is generic for almost all c. If y(i) is
generic then it represents a critical point.
The tuples y(i) describe a projective line in P(C[x])|I|. It will be useful to have the following
specific parameterization of this line. There exists a unique solution y(i)i (x) to the equation (4.7),
call it y(i)i (x; 0), such that the coefficient of x
deg yi in y(i)i (x; 0) is zero. Let us define
y(i)i (x; c) := y
(i)
i (x; 0) + cyi(x), (4.9)
and define y(i)(c) ∈ P(C[x])|I| to be the tuple obtained from the tuple y by replacing yi(x) with
y(i)i (x; 0) + cyi(x).
We say generation in the ith direction is degree-increasing if deg y(i)i > deg yi for almost all c.
Recall that there is a weight at infinity, Λ∞, associated to any critical point. For the critical
point represented by y this weight is, cf. (2.4),
Λ∞(y) = Λ0 +
N∑
s=1
M−1∑
k=0
σkΛs −
∑
j∈I
αj deg yj . (4.10)
For fixed Λ0,Λ1, . . . ,ΛN we can think of Λ∞ as encoding the degrees of the polynomials yj .
Note that deg y(i)i (x; 0) = deg yi+ 〈Λ∞, α
∨
i 〉+1. It follows that the weight at infinity of y
(i)(0) is
Λ∞ − αi
(
〈Λ∞, α
∨
i 〉+ 1
)
= Λ∞ − 〈Λ∞ + ρ, α
∨
i 〉αi = si · Λ∞.
This establishes the following lemma.
Lemma 4.4. Generation in the ith direction (with Li = 1) is degree-increasing if and only if Λ∞
is i-dominant, i.e., 〈Λ∞, α∨i 〉 ∈ Z≥0.
If generation in the ith direction is degree-increasing, then the weight at infinity associated
with the critical point represented by y(i)i (c) is si · Λ∞. Otherwise it is Λ∞ for all c 6= 0 (and
si · Λ∞ for c = 0).
Populations of Solutions to Cyclotomic Bethe Equations 15
4.4 Cyclotomic generation: the Li = 1 case
We continue to suppose that i is such that Li = 1.
If y represents a cyclotomic point then its immediate descendant y(i) in the ith direction
generically does not. However if, starting from a cyclotomic critical point, we successively
generate in each of the directions σki, k = 1, . . . ,Mi, in turn, in any order, then we can arrange
to arrive at a (new) cyclotomic critical point. This is the content of Theorem 4.6 below.
Let ' denote equality up to a constant (independent of x) nonzero factor. Recall the
definition (2.11) of a cyclotomic point.
Lemma 4.5. A tuple of polynomials y represents a cyclotomic point if and only if
yσj(ωx) ' yj(x)
for all j ∈ I. If yj(x) and yσj(x) share the same leading coefficient for all j ∈ I, then the tuple y
represents a cyclotomic point if and only if
yσj(ωx) = ω
deg yjyj(x)
for all j ∈ I.
For the rest of this subsection, we suppose y represents a cyclotomic critical point. Hence in
particular σΛ∞ = Λ∞. Let y
(i)
i (x; c) = y
(i)
i,0(x) + cyi(x) be as in (4.9). (So y
(i)
i (x; c) is a para-
meterization of the space of solutions to (4.7).) Define y(i,σ)(c) to be the tuple of polynomials
given by
y(i,σ)σki
(
ωkx; c
)
:= ωk deg y
(i)
i y(i)i (x; c), k = 0, 1, . . . ,Mi − 1,
and y(i,σ)j (x; c) = yj(x) for j ∈ I\σ
Zi. Recall sσi from Lemma 2.5.
Theorem 4.6. For almost all c ∈ C, the tuple y(i,σ)(c) represents a cyclotomic critical point.
The exceptional values of c form a finite subset of C.
The weight at infinity of y(i,σ)(c) is sσi · Λ∞ if 〈Λ∞, α
∨,σ
i 〉 ∈ Z≥0. Otherwise it is Λ∞ for all
c 6= 0, and sσi · Λ∞ for c = 0.
Proof. First let us show that y(i,σ) represents a cyclotomic point for all c ∈ C. Comparing our
definition of y(i,σ) with the criterion in Lemma 4.5, one sees that it is enough to check that
y(i)i
(
ωMix; c
)
= ωMi deg y
(i)
i y(i)i (x; c).
Inspecting (4.6), we see that this equality holds for all c ∈ C if and only if
ωMi〈Λ∞+ρ,α
∨
i 〉 = 1. (4.11)
But now, given (4.10) and the assumption that Λs, s = 1, . . . , n are integral, the following lemma
implies that (4.11) holds if and only if we impose the condition (4.2) on Λ0.
Lemma 4.7. Suppose Λ ∈ h∗ is an integral weight. Then, for any j ∈ I,
M−1∑
k=0
〈σkΛ, α∨j 〉Mj ≡ 0 mod M.
16 A. Varchenko and C.A.S. Young
Proof. We have
〈
M−1∑
k=0
σkΛ, α∨j
〉
Mj =
〈
Λ,
M−1∑
k=0
σ−kα∨j
〉
Mj
=
〈
Λ,
M
Mj
Mj−1∑
k=0
α∨j
〉
Mj = M
〈
Λ,
Mj−1∑
k=0
α∨j
〉
∈MZ. 
Now we show that y(i,σ) represents a critical point for all but finitely many c ∈ C. Note that
from definition (4.5) we have
Tσj(ωx) = ω
〈 N∑
s=1
M−1∑
k=0
σkΛs,α∨j
〉
Tj(x), j ∈ I. (4.12)
Hence, in view of (4.10),
x〈Λ0,α
∨
σi〉Tσi(ωx)
∏
j∈I
yj(ωx)
−〈αj ,α∨σi〉 = x〈Λ0,α
∨
i 〉Tσi(ωx)
∏
j∈I
yσj(ωx)
−〈ασj ,α∨σi〉
= ω〈Λ∞,α
∨
i 〉

x〈Λ0,α
∨
i 〉Ti(x)
∏
j∈I
yj(x)
−〈αj ,α∨i 〉

 .
Note also that since Li = 1, no node j in the orbit of i is linked by an edge of the Dynkin
diagram to i. That is, no yj for j in the orbit of i appears on the right of (4.7). Hence, for
k = 1, . . . ,Mi − 1, y
(i,σ)
σki (x; c) obeys the equation
Wr(yσki(x), y
(i,σ)
σki (x; c)) = x
〈Λ0,α∨
σki
〉Tσki(x)
∏
j∈I\{σki}
yj(x)
−〈αj ,α∨σki
〉.
and the tuple y(i,σ) is indeed the result of generating in each of the directions i, σi, . . . , σMi−1i (in
any order). It follows from Proposition 4.3 that y(i,σ) is generic for almost all c, and represents
a critical point whenever it is generic.
The statements about the weight at infinity follow from Lemma 4.4 and Section 2.3. This
completes the proof of Theorem 4.6. 
4.5 Elementary generation: the Li = 2 case
For this subsection we suppose that i ∈ I is such that Li = 2. That implies Mi is even and the
restriction of the Dynkin diagram to the nodes σZi consists of Mi2 ∈ Z≥1 disconnected copies of
the Dynkin diagram of type A2, as sketched below:
σ0i
σ0 ı¯
σ1i
σ1 ı¯
σ2i
σ2 ı¯
σMi/2−1i
σMi/2−1 ı¯
Here, for brevity, we write ı¯ := σMi/2i.
Populations of Solutions to Cyclotomic Bethe Equations 17
Remark 4.8. Among finite and affine types, only the case Mi/2 = 1 occurs.
We define y(i)i (x) by
y(i)i (x) := yi(x)x
−〈Λ0,α∨i 〉−1
∫ x
0
ξ〈Λ0,α
∨
i 〉Ti(ξ)
∏
j∈I
yj(ξ)
−〈αj ,α∨i 〉dξ.
Here the limits
∫ x
0 mean that y
(i)
i (x) is holomorphic at x = 0. This condition defines the integral
uniquely, since 〈Λ0, α∨i 〉 /∈ Z by our assumption (4.3).
Proposition 4.9. If y represents a critical point then y(i)i (x) is a polynomial. It has degree
deg y(i)i = deg yi + 〈Λ∞ − Λ0, α
∨
i 〉.
Proof. The proof is as for Proposition 4.2. 
Let y(i) = (y(i)j (x))j∈I be the tuple of polynomials whose ith component y
(i)
i (x) is as above,
and whose remaining components are the same as those of y, i.e.,
y(i)j (x) = yj(x) for all j ∈ I\{i}.
Let (t(i); c(i)) denote the pair represented by this tuple in the sense of Section 4.2. It turns out
that t(i) is not in general a critical point of the extended master function Φ̂(t(i); c(i); z; Λ), i.e.,
it does not in general obey the equations (2.9). Instead, the following result gives the analogous
collection of equations that it does obey, provided y(i) is generic.
Proposition 4.10. If y represents a critical point and y(i) is generic, then
〈
si · Λ0, α∨c(i)(p)
〉
t(i)p
+
N∑
s=1
M−1∑
k=0
〈
σkΛs, α∨c(i)(p)
〉
t(i)p − ωkzs
−
∑
r : r 6=p
〈
αc(i)(r), α
∨
c(i)(p)
〉
t(i)p − t
(i)
r
= 0
for each p.
Proof. By (2.9) for each root tp in the tuple t we have
〈
Λ0, α∨c(p)
〉
tp
+
N∑
s=1
M−1∑
k=0
〈
σkΛs, α∨c(p)
〉
tp − ωkzs
−
∑
r : r 6=p
〈
αc(r), α
∨
c(p)
〉
tp − tr
= 0. (4.13)
For all roots of colours j ∈ I such that 〈αj , α∨i 〉 = 0 this is immediately equivalent to the
required equation. So we must consider roots of colour i, and roots of colours j ∈ I such that
〈αj , α∨i 〉 < 0.
By definition of y(i)i (x) we have
Wr(yi(x), x
〈Λ0,α∨i 〉+1y(i)i (x)) = x
〈Λ0,α∨i 〉Ti(x)
∏
j 6=i
yj(x)
−〈αj ,α∨i 〉 (4.14)
or equivalently
y′i(x)
yi(x)
−
y(i)i
′(x)
y(i)i (x)
−
1 + 〈Λ0, α∨i 〉
x
=
Ti(x)
∏
j 6=i
yj(x)−〈αj ,α
∨
i 〉
xyi(x)y
(i)
i (x)
. (4.15)
18 A. Varchenko and C.A.S. Young
By definition of (t(i), c(i)), the left-hand side of (4.15) is
∑
r : c(r)=i
1
x− tr
−
∑
r : c(i)(r)=i
1
x− t(i)r
−
1 + 〈Λ0, α∨i 〉
x
. (4.16)
Now suppose j ∈ I is such that 〈αj , α∨i 〉 ∈ Z<0. By definition y
(i)
j (x) = yj(x). Suppose tp is
a root of yj(x), i.e., suppose c(p) = j. Since y represents a critical point, y must be generic, and
hence tp is not a root of yi(x). By our assumption that y(i) is generic, tp is not a root of y
(i)
i (x)
either. Hence the right-hand side of (4.15) is zero at x = tp and so, in view of (4.16), we have
∑
r : c(r)=i
1
tp − tr
−
∑
r : c(i)(r)=i
1
tp − t
(i)
r
−
1 + 〈Λ0, α∨i 〉
tp
= 0.
On adding this equation multiplied by 〈αi, α∨j 〉 to the equation (4.13), we arrive at
〈Λ0, α∨j 〉 − 〈αi, α
∨
j 〉〈Λ0 + ρ, α
∨
i 〉
t(i)p
+
N∑
s=1
M−1∑
k=0
〈σkΛs, α∨j 〉
t(i)p − ωkzs
−
∑
r : r 6=p
〈αc(i)(r), α
∨
j 〉
t(i)p − t
(i)
r
= 0,
which is the required equality (since si · Λ0 = Λ0 − 〈Λ0 + ρ, α∨i 〉αi).
It remains to consider roots of colour i. First note that yi(x) and y
(i)
i (x) have no common
roots. Indeed, if t were a common root of yi(x) and y
(i)
i (x) then the right-hand side of (4.14)
would have to vanish at x = t. In other words yi(x) would have a root in common with the
right-hand side of (4.14). But by our definition of what it means for y to be generic, Section 4.2,
this is impossible.
Suppose t(i)p is any root of y
(i)
i (x). By our assumption that y
(i) is generic, it follows from (4.14)
and Lemmas 4.12 and 4.13 below that
2(1 + 〈Λ0, α∨i 〉)
t(i)p
−
〈Λ0, α∨i 〉
t(i)p
−
N∑
s=1
M−1∑
k=0
〈σkΛs, α∨i 〉
t(i)p − ωkzs
+
∑
r:r 6=p
〈αc(i)(r), α
∨
i 〉
t(i)p − t
(i)
r
= 0,
which is the required equality. 
Remark 4.11. Propositions 4.9 and 4.10 also follow from Theorem 3.5 in [17].
Lemma 4.12. For any α ∈ C, if g(x) = xα
J∏
j=1
(x − sj), where (sj)Jj=1 are all distinct and
non-zero, then
g′′(x)
g′(x)
∣
∣
∣
∣
x=sk
=
2α
sk
+
J∑
j=1
j 6=k
2
sk − sj
.
Lemma 4.13. If Wr(f(x), g(x)) = W (x) then
g′′(x)
g′(x)
−
W ′(x)
W (x)
=
g(x)
(
W (x)f ′′(x)−W ′(x)f ′(x)
)
f(x)g′(x)W (x)
.
Proof. We have W Wr(f, g)′ = W ′Wr(f, g). Hence
W (x)f(x)g′′(x)−W ′(x)f(x)g′(x) = W (x)f ′′(x)g(x)−W ′(x)f ′(x)g(x)
and hence the result. 
Populations of Solutions to Cyclotomic Bethe Equations 19
To deal with the case in which y(i) fails to be generic, we shall also need the following
observation, which follows from (4.14).
Lemma 4.14. For any j ∈ I such that 〈αj , α∨i 〉 < 0, if t is a root of both yj(x) and y
(i)
i (x) then
it is a root of y(i)i (x) with multiplicity 2. In particular, if t is a root of both yı¯(x) and y
(i)
i (x)
then it is a root of y(i)i (x) with multiplicity 2.
Now we define
y(ı¯,i)ı¯ (x) := yı¯(x)
∫
ξ〈Λ0,α
∨
i 〉Tı¯(ξ)
ξ1+〈Λ0,α
∨
i 〉y(i)i (ξ)
∏
j 6=i,¯ı
yj(ξ)−〈αj ,α
∨
ı¯ 〉
yı¯(ξ)2
dξ
= yı¯(x)
∫
ξ1+2〈Λ0,α
∨
i 〉Tı¯(ξ)
y(i)i (ξ)
∏
j 6=i,¯ı
yj(ξ)−〈αj ,α
∨
ı¯ 〉
yı¯(ξ)2
dξ. (4.17)
Proposition 4.15. If y represents a critical point then y(ı¯,i)ı¯ (x) is a polynomial.
Proof. By our assumption (4.3) that 2〈Λ0, α∨i 〉 + 1 ∈ Z≥0, the integrand is regular at x = 0.
Hence, by Lemma 4.14, it is a rational function with poles at most at those roots of yı¯(x) that
are not also roots of y(i)i (x). Let tp be any such root. The residue of the integrand at ξ = tp is
∂
∂x
(x− t)2x〈Λ0,α
∨
i 〉Ti(x)
y(i)i (x)
∏
j 6=i,¯ı
y(i)j (x)
−〈αj ,α∨ı¯ 〉
yı¯(x)2
∣
∣
∣
∣
∣
∣
∣
∣
x=tp
,
which must vanish, because according to Proposition 4.10 the following vanishes:
∂
∂x
log(x− t)2x〈Λ0,α
∨
i 〉Ti(x)
y(i)i (x)
∏
j 6=i,¯ı
y(i)j (x)
−〈αj ,α∨ı¯ 〉
yı¯(x)2
∣
∣
∣
∣
∣
∣
∣
∣
x=tp
=
1 + 2〈Λ0, α∨i 〉
tp
+
N∑
s=1
M−1∑
k=0
〈σkΛs, α∨ı¯ 〉
tp − ωkzs
−
∑
r : r 6=p
〈αc(i)(r), α
∨
ı¯ 〉
tp − t
(i)
r
.
(Note 〈Λ0, α∨i 〉 = 〈Λ0, α
∨
ı¯ 〉 since σΛ0 = Λ0.) 
The polynomial y(ı¯,i)ı¯ (x) is defined up a to the addition of a constant multiple of yı¯(x), coming
from the constant of integration in (4.17).
We say generation in the ith direction from y is degree-increasing if deg y(ı¯,i)ı¯ (x) > deg yı¯(x).
Generation in the ith direction is degree-increasing if and only if
〈Λ∞ + ρ, α
∨
i + α
∨
ı¯ 〉 > 0. (4.18)
Indeed, if (4.18) holds then
deg y(ı¯,i)ı¯ (x) = deg yı¯(x) + 〈Λ∞ + ρ, α
∨
i + α
∨
ı¯ 〉 > deg yı¯(x) (4.19)
for all values of the constant of integration. If (4.18) does not hold then deg y(ı¯,i)ı¯ (x) ≤ deg yı¯(x),
with equality for all but one value of the constant of integration in (4.17).
20 A. Varchenko and C.A.S. Young
Let y(ı¯,i)ı¯ (x; 0) be the unique solution to (4.17) whose coefficient of x
deg yı¯ is zero. The degree
of y(ı¯,i)ı¯ (x; 0) is always given by
deg y(ı¯,i)ı¯ (x; 0) = deg yı¯(x) + 〈Λ∞ + ρ, α
∨
i + α
∨
ı¯ 〉,
whether or not generation is degree-increasing. (Note that 〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉 is odd, by our
assumption (4.3), and in particular not zero.)
Let then y(ı¯,i)(c) = (y(ı¯,i)j (x; c))j∈I be the tuple of polynomials whose ı¯th component is
y(ı¯,i)ı¯ (x; c) := y
(ı¯,i)
ı¯ (x; 0) + cyı¯(x)
and whose remaining components are the same as those of y(i), i.e.,
y(ı¯,i)i (x; c) = y
(i)
i (x; c), and y
(ı¯,i)
j (x) = y
(i)
j (x) = yj(x) for all j ∈ I\{i, ı¯}.
Let (t(ı¯.i); c(ı¯,i)) denote the pair represented by this tuple in the sense of Section 4.2.
The following result says that whenever y(ı¯,i)(c) is generic, this new pair (t(ı¯,i)(c), c(ı¯,i)) obeys
the same form of equations as did (t(i), c(i)).
Proposition 4.16. If y represents a critical point then, for all c ∈ C such that y(ı¯,i)(c) is
generic, we have
〈
si · Λ0, α∨c(ı¯,i)(p)
〉
t(ı¯,i)p (c)
+
N∑
s=1
M−1∑
k=0
〈
σkΛs, α∨c(ı¯,i)(p)
〉
t(ı¯,i)p (c)− ωkzs
−
∑
r : r 6=p
〈
αc(ı¯,i)(r), α
∨
c(ı¯,i)(p)
〉
t(ı¯,i)p (c)− t
(ı¯,i)
r (c)
= 0
for each p.
Proof. The proof is analogous to that of Proposition 4.10. 
Finally, we define y(i,¯ı,i)i (x; c) by
y(i,¯ı,i)i (x; c) = y
(i)
i (x)x
〈Λ0,α∨i 〉+1
∫ x
0
ξ〈Λ0,α
∨
i 〉Ti(ξ)
y(ı¯,i)ı¯ (ξ; c)
∏
j∈I\{i,¯ı}
yj(ξ)−〈αj ,α
∨
i 〉
(
ξ〈Λ0,α
∨
i 〉+1y(i)i (ξ)
)2 dξ
= y(i)i (x)x
〈Λ0,α∨i 〉+1
∫ x
0
ξ−〈Λ0,α
∨
i 〉−2Ti(ξ)
y(ı¯,i)ı¯ (ξ; c)
∏
j∈I\{i,¯ı}
yj(ξ)−〈αj ,α
∨
i 〉
y(i)i (ξ)
2
dξ.
Here the limits
∫ x
0 mean that y
(i,¯ı,i)
i (x; c) is holomorphic at x = 0. This condition defines the
integral uniquely.
Proposition 4.17. For all c ∈ C, if y represents a critical point then y(i,¯ı,i)i (x; c) is a polynomial.
Proof. Pick any root t(ı¯,i)p of y
(i)
i (x) = y
(ı¯,i)
i (x). The residue of the integrand at ξ = t
(ı¯,i) is
zero. Indeed, we have
∂
∂x
log
(
x− t(i)p
)2
x−〈Λ0,α
∨
i 〉−2Ti(x)
y(ı¯,i)ı¯ (x; c)
∏
j∈I\{i,¯ı}
yj(x)−〈αj ,α
∨
i 〉
y(i)i (x)
2
∣
∣
∣
∣
∣
∣
∣
∣
x=t(ı¯,i)p
=
−〈Λ0, α∨i 〉 − 2
t(ı¯,i)p
+
N∑
s=1
M−1∑
k=0
〈σkΛs, α∨ı¯ 〉
t(ı¯,i)p − ωkzs
−
∑
r : r 6=p
〈αc(ı¯,i)(r), α
∨
i 〉
t(ı¯,i)p − t
(ı¯,i)
r
,
and this vanishes by Proposition 4.16. 
Populations of Solutions to Cyclotomic Bethe Equations 21
Let y(i,¯ı,i)(c)=(y(i,¯ı,i)j (x; c))j∈I be the tuple of polynomials whose ith component is y
(i,¯ı,i)(x; c)
as above and whose remaining components are those of y(ı¯,i)(c), i.e.,
y(i,¯ı,i)ı¯ (x; c) = y
(ı¯,i)
ı¯ (x; c), and y
(i,¯ı,i)
j (x) = yj(x) for all j ∈ I\{i, ı¯}.
Let (t(i,¯ı.i); c(i,¯ı,i)) denote the pair represented by this tuple in the sense of Section 4.2.
Proposition 4.18. If y represents a critical point and y(i,¯ı,i)(c) is generic, then y(i,¯ı,i)(c) rep-
resents a critical point. That is, the pair (ti,¯ı,i(c), ci,¯ı,i) obeys the equations
〈
Λ0, α∨c(p)
〉
t(i,¯ı,i)p (c)
+
N∑
s=1
M−1∑
k=0
〈
σkΛs, α∨c(p)
〉
t(i,¯ı,i)p (c)− ωkzs
−
∑
r : r 6=p
〈
αc(r), α
∨
c(p)
〉
t(i,¯ı,i)p (c)− t
(i,¯ı,i)
r (c)
= 0
for each p.
Proof. The proof is analogous to that of Proposition 4.10. 
We say y(i,¯ı,i)(c) is obtained from y by generation in the ith direction, and we call y(i,¯ı,i)(c)
the immediate descendant of y in the ith direction. We have the following; cf. Lemma 4.4.
Lemma 4.19. Generation in the ith direction (with Li = 2) is degree-increasing if and only if
〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉 ∈ Z>0.
If generation in the ith direction is degree-increasing, then the weight at inf inity associated
with the critical point represented by y(i,¯ı,i)i (c) is (sisı¯si) · Λ∞. Otherwise it is Λ∞ for all c 6= 0
(and (sisı¯si) · Λ∞ for c = 0).
Proof. Recall that (4.19) holds if and only if (4.18) holds. Note also that
deg y(i,¯ı,i)i = deg yi + 〈Λ∞ + ρ, α
∨
i + α
∨
ı¯ 〉.
By direct calculation, one verifies that
(sisı¯si) · Λ∞ = Λ∞ − (αı¯ + αi)〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉,
so we have the result. 
4.6 Cyclotomic generation: the Li = 2 case
We continue to suppose that i ∈ I is such that Li = 2.
Suppose for the rest of this subsection that y represents a cyclotomic critical point. Define
y(i,σ)(c) to be the tuple of polynomials given by
y(i,σ)σki (ω
kx; c) := y(i,¯ı,i)i (x; c),
y(i,σ)σk ı¯ (ω
kx; c) := y(i,¯ı,i)ı¯ (x; c), k = 0, 1, . . . ,Mi/2− 1, (4.20)
and y(i,σ)j (x; c) = yj(x) for j ∈ I\σ
Zi.
Theorem 4.20. For almost all c ∈ C, the tuple y(i,σ)(x; c) represents a cyclotomic critical point.
The exceptional values of c form a f inite subset of C.
The weight at inf inity of y(i,σ)(x; c) is sσi ·Λ∞ if 〈Λ∞+ρ, α
σ
i +α
σ
ı¯ 〉 ∈ Z≥1. Otherwise it is Λ∞
for all c 6= 0, and sσi · Λ∞ for c = 0.
22 A. Varchenko and C.A.S. Young
Proof. First let us show that y(i,σ)(x; c) represents a critical point for all but finitely many
c ∈ C. As in the proof of Theorem 4.6, we first observe that y(i,σ) is indeed the result of
generating in each of the directions i, σi, . . . , σMi/2−1i (in any order). By Proposition 4.18 it is
enough to check that y(i,¯ı,i)(c) is generic for all but finitely many c ∈ C. This follows from (4.21)
and Lemma 4.22, below.
The statements about the weight at infinity follow from Lemma 4.19 and Section 2.3.
Finally we must check that y(i,σ)(x; c) represents a cyclotomic point. Given Lemma 4.5 and
the definition (4.20), it is enough to check that
y(i,¯ı,i)ı¯ (−x; c) = (−1)
deg y(i,ı¯,i)i y(i,¯ı,i)i (x; c). (4.21)
This is effectively a statement about the case of type A2 and we are in the setting of Section 5
below, with R = 2n, n = 1, p = 1. The statement (4.21) follows from Theorem 5.34 and
Lemma 5.36. 
Lemma 4.21. We have
y(ı¯,i,¯ı)j (−x; c) = (−1)
deg y(i,ı¯,i)¯ y(i,¯ı,i)¯ (x;−c)
for all j ∈ I.
Proof. Note first that from (4.12) we have
T¯(−x) = (−1)
〈M−1∑
k=0
N∑
s=1
ωkΛs,α∨j
〉
Tj(x)
for all j ∈ I. It follows that
y(ı¯)ı¯ (−x) = (−1)
deg yi+〈Λ∞−Λ0,α∨i 〉y(i)i (x).
Then, from the definition of y(ı¯,i)ı¯ (x; c) and (4.19) we have that
y(i,¯ı)i (−x; c) = (−1)
deg y(i,ı¯)ı¯ y(ı¯,i)ı¯ (x; c¯)
if and only if c and c¯ are related by c = (−1)2+〈Λ∞,α
∨
ı¯ +α
∨
i 〉c¯. Since the Λs, s = 1, . . . , N , are
integral, we have
(−1)〈Λ∞,α
∨
i +α
∨
ı¯ 〉 = (−1)〈Λ0,α
∨
i +α
∨
ı¯ 〉 = (−1)2〈Λ0,α
∨
i 〉 = −1,
using σΛ0 = Λ0 and the property (4.1). 
Lemma 4.22. For all but finitely many c ∈ C, y(i,¯ı,i)ı¯ (x; c) and y
(i,¯ı,i)
ı¯ (−x; c) have no root in
common.
Proof. Recall y(i,¯ı,i)ı¯ (x; c) = y
(ı¯,i)
ı¯ (x; c). Consider the leading behaviour in small c. As c → 0,
deg yı¯ of the roots of y
(ı¯,i)
ı¯ (x; c) tend to the deg yı¯ roots of yı¯(x). By the assumption that y was
generic and cyclotomic, none of these are roots of yı¯(−x) ' yi(x).
Recall (4.19) and the fact that 〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉 is odd, by the assumption (4.3).
If 〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉 < 0, then these are all the roots of y
(ı¯,i)
ı¯ (x; c).
If 〈Λ∞ + ρ, α∨i + α
∨
ı¯ 〉 6< 0 then the remaining 〈Λ∞ + ρ, α
∨
i + α
∨
ı¯ 〉 > 0 roots of y
(ı¯,i)
ı¯ (x; c) tend
to the roots of the equation cx〈Λ∞+ρ,α
∨
i +α
∨
ı¯ 〉+ 1 = 0. This limiting set of roots multiplied by −1
does not intersect itself. This implies the lemma. 
Populations of Solutions to Cyclotomic Bethe Equations 23
4.7 Definition of the cyclotomic population
Suppose y ∈ P(C[x])|I| is a tuple of polynomials representing a cyclotomic critical point.
Recall the definition of y(i,σ)(c), from Section 4.5 when Li = 1 and from Section 4.6 when
Li = 2. We say y(i,σ)(c) is obtained from y by cyclotomic generation in the direction i.
Let us define the cyclotomic population originated at y to be the Zariski closure of the set
of all tuples of polynomials obtained from y by repeated cyclotomic generation, in all directions
i ∈ I.
5 The case of type AR: vector spaces of quasi-polynomials
5.1 Type A data
Throughout this section we specialise to g = slR+1. We shall treat in parallel the cases where
R = 2n− 1 and R = 2n, n ∈ Z≥0. We have the usual identification of h ∼= h∗ with a subspace of
(R+ 1)-dimensional Euclidean space, given by αi = α∨i = i+1 − i, i = 1, . . . , R, where (i)
R+1
i=1
is the standard orthonormal basis.
Let σ : g→ g be the unique non-trivial diagram automorphism, whose order is 2. The nodes I
of the Dynkin diagram, and the action of σ on these nodes, are as shown below:
n− 1 n n+ 1 2n− 11
n− 1 n n+ 1 n+ 2 2n1
When R = 2n − 1, then Li = 1 for all i ∈ I, and Mi =
{
1, i = n,
2, i 6= n.
When R = 2n then
Li =
{
2, i = n, n+ 1
1, otherwise
and Mi = 2 for all i ∈ I.
Let (zi)Ni=1 be nonzero points zi ∈ C
× such that zi ± zj 6= 0 whenever i 6= j. Let Λ1, . . . ,ΛN
be dominant integral weights.
We suppose the weight at the origin, Λ0 ∈ h∗, obeys σΛ0 = Λ0 (as always). That is,
〈Λ0, α
∨
i 〉 = 〈Λ0, α
∨
R+1−i〉, i = 1, . . . , R.
In addition, we pick and fix an integer p ∈ {0, 1, . . . , n}, and suppose that
〈Λ0, α
∨
i 〉 ∈ 2Z≥0/Mi if i /∈ {p,R+ 1− p} (5.1a)
and
〈Λ0, α
∨
p 〉 ∈
{
1
2(2Z≥0 − 1) = {−
1
2 ,
1
2 ,
3
2 , . . . } if p ≤ R/2,
2Z≥0 + 1 = {1, 3, . . . } if p = n and R = 2n− 1.
(5.1b)
Note the following particular cases:
• If R = 2n is even and p = 0 then (5.1) just says that Λ0 is dominant integral.
24 A. Varchenko and C.A.S. Young
• If R = 2n− 1 is odd and p = 0 then Λ0 is dominant integral and 〈Λ0, α∨n〉 is even.
• If R = 2n− 1 is odd and p = n then Λ0 is dominant integral and 〈Λ0, α∨n〉 is odd.
In the case p = n (and any R) our choice of Λ0 obeys the assumptions set out in Section 4.1.
5.2 Vector spaces of quasi-polynomials
Let
T˜i(x) = x
〈Λ0,α∨i 〉
N∏
s=1
(x− zs)
〈Λs,α∨i 〉(x+ zs)
〈ΛR+1−s,α∨i 〉, i ∈ I.
Thus T˜i(x) = x〈Λ0,α
∨
i 〉Ti(x) with Ti(x) as in (4.5).
In view of (5.1), T˜i(x) ∈ C[x] for all i /∈ {p,R+ 1− p}. If 0 < p < R+ 1− p < R then T˜p(x)
and T˜R+1−p(x) belong to x−
1
2C[x]. If p = R+ 1− p then T˜p(x) ∈ C[x].
We define the degree, deg p, of a Laurent polynomial p(x) ∈ C[x±
1
2 ] to be the leading power
of x (for large x) that appears in p(x) with non-zero coefficient.
We will call any polynomial in x
1
2 a quasi-polynomial.
A vector space V ⊂ C
[
x
1
2
]
of quasi-polynomials is decomposable if
V = V ∩ C[x]⊕ V ∩ x
1
2C[x].
A tuple of quasi-polynomials is decomposable if each element lies in either C[x] or x
1
2C[x]. In
particular, a decomposable basis of a decomposable vector space V ⊂ C
[
x
1
2
]
is one in which each
basis vector lies in either C[x] or x
1
2C[x].
Define the divided Wronksian determinant of quasi-polynomials u1, . . . , uk ∈ C
[
x
1
2
]
by
Wr†(u1, . . . , uk) :=
Wr(u1, . . . , uk)
T˜ k−11 T˜
k−2
2 · · · T˜k−1
, Wr(u1, . . . , uk) := det
(
dj−1ui
dxj−1
)k
i,j=1
,
for k = 1, . . . , R+ 1.
Define
Λ := Λ0 +
N∑
s=1
(Λs + σΛs), (5.2)
and suppose Λ˜∞ ∈ h∗ is a dominant weight such that Λ − Λ˜∞ =
∑
i∈I
kiαi for some ki ∈ Z≥0.
Such a weight defines numbers d1, . . . , dR+1 ∈ Z/2, 0 ≤ d1 < · · · < dR+1, by
d1 := 〈Λ− Λ˜∞, 1〉, dk := 〈Λ− (s1 · · · sk−1) · Λ˜∞, 1〉, k = 2, . . . , R+ 1. (5.3)
Lemma 5.1. We have
dk = d1 + 〈Λ˜∞ + ρ, α
∨
1 + · · ·+ α
∨
k−1〉, k = 2, . . . , R+ 1. (5.4)
Hence, for all p > 0, d1, . . . , dp and dR+2−p, . . . , dR+1 are integers while dp+1, . . . , dR+1−p are
half odd integers, i.e., have the form m+ 12 for m ∈ Z. If p = 0 then d1, . . . , dR+1 are all integers.
Proof. We have dk−d1 = 〈Λ˜∞+ρ− (s1 · · · sk−1)(Λ˜∞+ρ), 1〉 = 〈Λ˜∞+ρ, 1− (sk−1 · · · s1)1〉 =
〈Λ˜∞ + ρ, 1 − k〉 and hence (5.4). 
Populations of Solutions to Cyclotomic Bethe Equations 25
Definition 5.2. We say a vector space of quasi-polynomials K⊂C
[
x
1
2
]
has frame T˜1, . . . , T˜R; Λ˜∞
if the following conditions hold:
(i) There is a basis (uk(x))
R+1
k=1 of K such that deg uk = dk for each k = 1, . . . , R+ 1.
(ii) For any z ∈ C\{0} and v1, . . . , vk ∈ K, k = 1, . . . , R + 1, the divided Wronskian Wr†(v1,
. . . , vk) is regular at z, and moreover, Wr
†(v1, . . . , vk) is nonzero at z for suitable v1, . . . , vk.
(iii) For all v1, . . . , vk ∈ K, k = 1, . . . , R + 1, the divided Wronskian Wr
†(v1, . . . , vk) has at
x = 0 an expansion of the form
∑
m∈Z≥0/2
amxm and moreover this expansion has nonzero a0
for suitable v1, . . . , vk.
In the remainder of this section, K will denote a decomposable vector space of quasi-polyno-
mials with frame T˜1, . . . , T˜R; Λ˜∞.
Conditions (ii) and (iii) specify the ramification conditions of K at every point z ∈ C.
Condition (i) specifies the ramification conditions at ∞. See [15, Section 5.5]. The degrees
0 ≤ d1 < d2 < · · · < dR+1 will be called the exponents of K at infinity.
Note that conditions (ii) and (iii) together imply in particular that K has no base points.
That is, there is no z ∈ C such that u(z) = 0 for all u ∈ K. They also imply the following
important lemma.
Lemma 5.3. For all v1, . . . , vk ∈ K, k = 1, . . . , R+ 1, the divided Wronskian Wr†(v1, . . . , vk) is
a quasi-polynomial.
Since K is decomposable it follows from condition (i) that K admits a decomposable basis
(uk)
R+1
k=1 such that deg uk = dk for each k. We call any such basis a special basis.
Lemma 5.4. Any two special bases (uk)
R+1
k=1 and (u
′
k)
R+1
k=1 are related by a triangular change of
basis, u′k =
∑
j≤k
akjuj, such that akj = 0 whenever dk − dj /∈ Z.
Lemma 5.5. Let m ∈ Z≥1. Let n1, . . . , nm be non-negative integers. Then
Wr(xn1 , . . . , xnm) = x
m∑
i=1
ni−
m(m−1)
2
∏
1≤j<i≤m
(ni − nj).
Lemma 5.6. Let (ui(x))
R+1
i=1 be a special basis of K. Then
Wr†(u1, . . . , uR+1) =
∏
1≤j<i≤R+1
(di − dj).
Proof. By Lemma 5.3, Wr†(u1, . . . , uR+1) ∈ C
[
x
1
2
]
. We must show that it has degree zero
and compute the constant term. From the condition that Λ − Λ˜∞ ∈ Z≥0[αi]i∈I it follows that
0 = 〈Λ− Λ˜∞,−(R+ 1)1 +Rα∨1 + (R− 1)α
∨
2 + · · ·+ 2α
∨
R−1 + α
∨
R〉 and therefore
(R+ 1)d1 = 〈Λ− Λ˜∞, Rα
∨
1 + (R− 1)α
∨
2 + · · ·+ 2α
∨
R−1 + α
∨
R〉. (5.5)
Then (5.4) implies
R+1∑
i=1
di −
(R+ 1)R
2
= 〈Λ, Rα∨1 + (R− 1)α
∨
2 + · · ·+ 2α
∨
R−1 + α
∨
R〉. (5.6)
The result follows by Lemma 5.5. 
26 A. Varchenko and C.A.S. Young
Corollary 5.7. Wr†(v1, . . . , vR+1) is a constant (independent of x) for all v1, . . . , vR+1 ∈ K.
Let (uk)
R+1
k=1 be a special basis of K. Introduce the subspaces
KSp := spanC(u1, . . . , up)⊕ spanC(uR+2−p, . . . , uR+1),
KO := spanC(up+1, . . . , uR+1−p),
so that
K = KSp ⊕KO.
By Lemma 5.4, these definitions of do not depend on the choice of special basis (uk)
R+1
k=1 . By
Lemma 5.1 we have that, whenever p > 0, then
KSp = K ∩ C[x], KO = K ∩ x
1
2C[x]. (5.7)
Exceptionally, when p = 0, we have KSp = {0}, KO = K ⊂ C[x].
Given a decomposable subspace V , we write sdimV for the pair of numbers
sdimV := (dimV ∩ KSp|dimV ∩ KO).
5.3 Flags in K
Let FL(K) denote the space of full (i.e., R+ 1-step) flags in K.
We say an r-step flag F = {0 = F0 ⊂ F1 ⊂ F2 ⊂ · · · ⊂ Fr = K} in K is decomposable if
each Fk is decomposable.
The space of decomposable full flags in K has
(R+1
2p
)
connected components. These connected
components are labeled by 2p-element subsets Q ⊂ {1, . . . , R + 1}. Define FLQ(K) to be the
subset consisting of the flags F = {0 = F0 ⊂ F1 ⊂ · · · ⊂ FR+1 = K} such that for each k,
sdimFk − sdimFk−1 =
{
(1|0) if k ∈ Q,
(0|1) if k /∈ Q.
We call elements of FLQ(K) flags of type Q.
For each Q the variety FLQ(K) is isomorphic to the direct product of full flag spaces
FL(KSp)× FL(KO). The isomorphism
ηQ : FL(KSp)× FL(KO)→ FLQ(K) (5.8)
sends a pair of flags F1,+ ⊂ · · · ⊂ F2p,+, F1,− ⊂ · · · ⊂ FR+1−2p,− to the flag F1 ⊂ · · · ⊂ FR,
where Fk = Fk1,+ ⊕ Fk2,−, k1 = |Q ∩ {1, . . . , k}|, k2 = k − k1.
Call a 2p-element subset Q ⊂ {1, . . . , R + 1} symmetric if Q is invariant with respect to the
involution k 7→ R+ 2− k. In particular, the following subset S is symmetric
S := {1, . . . , p, R+ 2− p, . . . , R+ 1}. (5.9)
If (uk)
R+1
k=1 is a special basis of K then the full flag F = {0 = F0 ⊂ F1 ⊂ F2 ⊂ · · · ⊂ FR+1 = K}
defined by
Fk = spanC(u1, . . . , uk), k = 1, . . . , R+ 1, (5.10)
belongs to FLS(K). By Lemma 5.4 this flag is independent of the choice of special basis.
Populations of Solutions to Cyclotomic Bethe Equations 27
To any full flag F = {0 ⊂ F1 ⊂ F2 ⊂ · · · ⊂ FR+1 = K} in FL(K) one can associate a tuple
yF = (yi(x))Ri=1 ∈ P
(
C
[
x
1
2
])R
. Namely, let (uFk (x))
R+1
k=1 be any basis of K such that
Fk = spanC
(
uF1 , . . . , u
F
k
)
, k = 1, . . . , R+ 1.
(we say such a basis is adjusted to F) and then let
yFk := Wr
† (uF1 , . . . , u
F
k
)
, k = 1, . . . , R. (5.11)
By Lemma 5.3, these are quasi-polynomials.
We have the shifted action of the Weyl group of type AR on weights as in Section 2.1. The
weight at infinity Λ∞(yF ), as in (4.10), belongs to the shifted Weyl orbit of Λ˜∞ [15, Section 3.6].
It is equal to Λ˜∞ if and only if F is the flag given in (5.10).
The map F 7→ yF defines a morphism of varieties,
β : FL(K)→ P
(
C
[
x1/2
])R
.
This morphism β defines an isomorphism of FL(K) onto its image, as in Lemmas 5.14–5.16
of [15].
Lemma 5.8. The image β(FLS(K)) of the variety of flags of type S lies in P(C[x])R, i.e.,
consists of tuples of polynomials.
Proof. In the exceptional case p = 0 no fractional powers are present at all and the result is
clear. Suppose p > 0. Let F ∈ FLS(K) and let (uFk )
R+1
k=1 be a basis of K adjusted to F . By
inspection one sees that because F ∈ FLS(K), Wr(uF1 , . . . , u
F
k ) lies in C[x] (resp. x
1
2C[x]) for
precisely those k such that the product T˜ k−11 · · · T˜k−1 lies in C[x
±1] (resp. x
1
2C[x±1]). For each k,
Lemma 5.3 guarantees that yFk ∈ C
[
x
1
2
]
. Hence in fact yFk ∈ C[x]. 
Lemma 5.9. The tuple β(F) = yF is decomposable if and only if F is a decomposable flag.
If F is a decomposable flag of type Q then
yFk ∈
{
C[x] if |S4Q ∩ {1, . . . , k}| ∈ 2Z,
x
1
2C[x] if |S4Q ∩ {1, . . . , k}| ∈ 2Z+ 1,
where S4Q := (S\Q) ∪ (Q\S) denotes the symmetric difference of S and Q. In particular yF
is a tuple of polynomials if and only if Q = S.
5.4 Fundamental differential operator and the recovery theorem
To any given a tuple y = (yi(x))Ri=1 ∈ P
(
C
[
x
1
2
])R
of quasi-polynomials, we may associate
a differential operator D(y), defined by
D(y) :=
(
∂ − log′
T˜1T˜2 · · · T˜R
yR
)(
∂ − log′
yRT˜1T˜2 · · · T˜R−1
yR−1
)
· · ·
×
(
∂ − log′
y2T˜1
y1
)
(
∂ − log′ y1
)
=
−→
R∏
i=0





∂ − log′
yR+1−i
R−i∏
j=1
T˜j
yR−i





,
with the understanding that y0 = yR+1 = 1. Here ∂ := ∂/∂x and log
′ f := f ′(x)/f(x).
Theorem 5.10 ([15, Lemma 5.6]). Let y ∈ β(FL(K)). Then K = kerD.
28 A. Varchenko and C.A.S. Young
5.5 The dual space K†
Let K† be the complex vector space
K† := spanC
{
Wr†(v1, . . . , vR) : v1, . . . , vR ∈ K
}
⊂ C
[
x
1
2
]
.
The space K† is a space of quasi-polynomials by Lemma 5.3. The spaces K† and K are dual
with respect to the pairing
(·, ·) : K† ×K → C
defined by
(
v1,Wr
†(v2, . . . , vR+1)
)
:= Wr†(v1, v2, . . . , vR+1).
Given any basis (ui(x))
R+1
i=1 of K there is a basis (Wi(x))
R+1
i=1 of K
† defined by
Wi := Wr
†(u1, . . . , ûi, . . . , uR+1) ∈ K
†, i = 1, . . . , R+ 1,
where ûi denotes omission. We have
(ui,Wj) = 0 if i 6= j, (ui,Wi) 6= 0.
Let d†1 > · · · > d
†
R+1 be the numbers given by
d†R+1 := −〈Λ− Λ˜∞, R+1〉, d
†
k := −〈Λ− (sR · · · sk) · Λ˜∞, R+1〉, k = 1, . . . , R,
cf. (5.3). We have
d†k = d
†
R+1 + 〈Λ˜∞ + ρ, α
∨
k + · · ·+ α
∨
R〉, k = 1, . . . , R, (5.12)
by an argument as for Lemma 5.1.
Lemma 5.11. Let (ui(x))
R+1
i=1 be a special basis of K. Then degWk = d
†
k, k = 1, . . . , R+ 1, and
the basis (Wk)
R+1
k=1 is decomposable.
Proof. From Λ− Λ˜∞ ∈ Z≥0[αi]i∈I we have 0 = 〈Λ− Λ˜∞, (R+ 1)R+1 + α∨1 + 2α
∨
2 + · · ·+ (R−
1)α∨R−1 +Rα
∨
R〉, and hence
(R+ 1)d†R+1 = 〈Λ− Λ˜∞, α
∨
1 + 2α
∨
2 + · · ·+Rα
∨
R〉. (5.13)
Now
degWR+1 = deg Wr
†(u1, . . . , uR) =
R∑
i=1
di −
R(R− 1)
2
− 〈Λ, (R− 1)α∨1 + · · ·+ α
∨
R−1〉
= Rd1 + 〈Λ˜∞ + ρ, (R− 1)α
∨
1 + · · ·+ α
∨
R−1〉 −
R(R− 1)
2
− 〈Λ, (R− 1)α∨1 + · · ·+ α
∨
R−1〉
= Rd1 + 〈Λ˜∞ − Λ, (R− 1)α
∨
1 + · · ·+ α
∨
R−1〉,
where we used (5.4). Hence, using (5.5), we have
(R+ 1) degWR+1 = R〈Λ− Λ˜∞, Rα
∨
1 + (R− 1)α
∨
2 + · · ·+ 2α
∨
R−1 + α
∨
R〉
− (R+ 1)〈Λ− Λ˜∞, (R− 1)α
∨
1 + · · ·+ α
∨
R−1〉
= 〈Λ− Λ˜∞, α
∨
1 + 2α
∨
2 + · · ·+Rα
∨
R〉
Populations of Solutions to Cyclotomic Bethe Equations 29
since R(R+1−k)−(R+1)(R−k) = k. Comparing this with (5.13) we see that d†R+1 = degWR+1.
Then for the remaining Wk, we note that degWR+1 − degWk = dk − dR+1 for k = 1, . . . , R.
And by (5.4) and (5.12),
dk − dR+1 = −〈Λ˜∞ + ρ, α
∨
k + · · ·+ α
∨
R〉 = d
†
R+1 − d
†
k.
Thus d†k = degWk for k = 1, . . . , R + 1. Finally, since the basis (uk)
R+1
k=1 is decomposable and
each T˜k lies in either C[x±1] or x
1
2C[x±1], it follows that (Wk)
R+1
k=1 is decomposable. 
5.6 Cyclotomic points and cyclotomic self-duality
Let us fix (−1)m := empii for m ∈ Z/2. Then given a monomial q(x) = xm, m ∈ Z/2, we define
q(−x) := (−1)mxm. We extend the transformation q(x) 7→ q(−x) to Laurent polynomials in x
1
2
by linearity.
We say that K is cyclotomically self-dual if
u(x) ∈ K ⇔ u(−x) ∈ K†.
Lemma 5.12. If K is cyclotomically self-dual then
dk + dR+2−k = R+ 〈Λ, α
∨
1 + · · ·+ α
∨
R〉, k = 1, . . . , R+ 1.
Proof. If K is cyclotomically self-dual then we must have dk = d
†
R+2−k, k = 1, . . . , R + 1.
Comparing (5.4) and (5.12) we see that this implies that
〈Λ˜∞ + ρ, α
∨
1 + . . . α
∨
k 〉 = dk+1 − d1 = d
†
R+1−k − d
†
R+1
= 〈Λ˜∞ + ρ, α
∨
R+1−k + · · ·+ α
∨
R〉, k = 1, . . . , R,
and hence
〈Λ˜∞, α
∨
k 〉 = 〈Λ˜∞, α
∨
R+1−k〉, k = 1, . . . , R.
Therefore
dk + dR+2−k = 2d1 + 〈Λ˜∞ + ρ, α
∨
1 + · · ·+ α
∨
R〉, k = 1, . . . , R+ 1,
and so, because the right-hand side here does not depend on k,
dk + dR+2−k =
2
R+ 1
R+1∑
j=1
dj , k = 1, . . . , R+ 1. (5.14)
Recall (5.6) and the definition (5.2) of Λ. Using now the fact that 〈Λ, α∨i 〉 = 〈Λ, α
∨
R+1−i〉,
i = 1, . . . , R, we have
R+1∑
j=1
dj −
(R+ 1)R
2
=
R+ 1
2
〈Λ, α∨1 + · · ·+ α
∨
R〉.
Thus, given (5.14), we have the result. 
30 A. Varchenko and C.A.S. Young
If K is cyclotomically self-dual then there is a non-degenerate bilinear form B on K defined
by
B(u(x), v(x)) := (u(x), v(−x)),
i.e.,
B(u, v) = Wr†(u, v1, . . . , vR), where v(−x) = Wr
†(v1, . . . , vR).
Let us call a tuple of quasi-polynomials y ∈ P
(
C
[
x
1
2
])R
cyclotomic if
yk(−x) ' yR+1−k(x), k = 1, . . . , R.
Proposition 5.13. Let F ∈ FL(K). If the tuple β(F) ∈ P
(
C
[
x
1
2
])R
is cyclotomic then F is
a decomposable flag.
Proof. Let yF = β(F). To prove that F is decomposable it is enough to show that each
entry yFk of this tuple lies in C[x] or in x
1/2C[x]. For each k = 1, . . . , R we have yFk (x) =
x
1
2ak(x)+bk(x) for some polynomials ak(x) and bk(x) in x. If yF is cyclotomic then yR+1−k(x) '
yFk (−x) = (−1)
1
2x
1
2ak(−x) + bk(−x) for each k. That is, aR+1−k(x) = (−1)
1
2 ckak(−x) and
bR+1−k(x) = ckbk(−x) for some non-zero constants ck. But that means
ak(x) = (−1)
1
2 cR+1−kaR+1−k(−x) = −cR+1−kckak(x),
bk(x) = cR+1−kbR+1−k(−x) = +cR+1−kckbk(x),
from which we conclude that for each k at least one of ak(x) and bk(x) must vanish. 
Theorem 5.14. Suppose β(FL(K)) contains a cyclotomic tuple. Then K is cyclotomically self-
dual.
Proof. We shall need the following identity among Wronskian determinants.
Lemma 5.15 ([15]). Given integers 0 ≤ k ≤ s and functions f1, . . . , fs+1, we have
Wr
(
Wr(f1, . . . , fs−k, . . . , fs, f̂s+1),Wr(f1, . . . , fs−k, . . . , f̂s, fs+1), . . . ,
Wr(f1, . . . , fs−k, f̂s−k+1, . . . , fs+1)
)
= Wr(f1, . . . , fs−k)
(
Wr(f1, . . . , fs+1)
)k
,
where f̂ denotes omission.
To prove Theorem 5.14 we argue as for Theorem 6.8 in [15]. Let F ∈ FL(K) be a full flag
in K and (ui(x))
R+1
i=1 a basis of K adjusted to this flag. Let y = y
F be the corresponding tuple of
quasi-polynomials as in (5.11), and (Wi(x))
R+1
i=1 the corresponding basis of K
† as in (5.5). Then
Theorem 5.14 follows from the case k = R+ 1 of the following lemma.
Lemma 5.16. If y is cyclotomic then
spanC(u1(−x), . . . , uk(−x)) = spanC(WR+1,WR, . . . ,WR+2−k), k = 1, . . . , R+ 1.
Proof. Let us prove the lemma by induction on k. For k = 1 we have
u1(−x) = y1(−x) ' yR(x) = Wr
†(u1, . . . , uR) = WR+1
as required. Assume the statement holds for all values up to some k. For the inductive step it
is enough to show that
Wr(u1(−x), . . . , uk(−x),WR+1−k) 'Wr(u1(−x), . . . , uk(−x), uk+1(−x)). (5.15)
Populations of Solutions to Cyclotomic Bethe Equations 31
Indeed, (5.15) is an inhomogeneous differential equation in WR+1−k(x) and if it holds then it
must be that WR+1−k(x) is proportional to uk+1(−x) modulo spanC(u1(−x), . . . , uk(−x)), which
is sufficient given the inductive assumption.
By the inductive assumption, we have
Wr(u1(−x), . . . , uk(−x),WR+1−k)
'Wr(WR+1,WR, . . . ,WR+1−k+1,WR+1−k)
= Wr
(
Wr(u1, . . . , uR+1−k−1, . . . , uR, ûR+1),
Wr(u1, . . . , uR+1−k−1, . . . , ûR, uR+1), . . . ,
Wr(u1, . . . , uR+1−k−1, ûR+1−k, . . . , uR+1)
)/(
T˜R−11 T˜
R+1−3
2 · · · T˜
1
R−1
)k+1
= Wr(u1, . . . , uR−k)(Wr(u1, . . . , uR+1))
k/(T˜R−11 T˜
R+1−3
2 · · · T˜
1
R−1
)k+1
,
the final equality by Lemma 5.15. Since Wr†(u1, . . . , uR+1) = Wr(u1, . . . , uR+1)/T˜R1 T˜
R−1
2 · · · T˜
1
R
is a nonzero constant by Lemma 5.6 we therefore have
Wr(u1(−x), . . . , uk(−x),WR+1−k) 'Wr(u1, . . . , uR−k)
(
T˜R1 · · · T˜
1
R
)k
(
T˜R−11 · · · T˜R−1
)k+1
=
Wr(u1, . . . , uR−k)
T˜R+1−k−21 · · · T˜
1
R+1−k−2
T˜ kR · · · T˜
1
R+1−k.
Now we may use again the fact that y is cyclotomic, so yk(−x) ' yR+1−k(x). In view of (5.11),
that implies
Wr(u1, . . . , uR−k)
T˜R+1−k−21 · · · T˜
1
R+1−k−2
'
Wr(u1(−x), . . . , uk+1(−x))
T˜ k1 (−x) · · · T˜
1
k (−x)
. (5.16)
Recall that T˜R+1−k(x) ' T˜k(−x). Hence we have indeed that
Wr(u1(−x), . . . , uk(−x),WR+1−k) 'Wr(u1(−x), . . . , uk+1(−x)),
as required. 
This completes the proof of Theorem 5.14. 
Given a subspace U ⊂ K, let
U⊥ := {v ∈ K : B(u, v) = 0 for all u ∈ U}
denote its orthogonal complement in K with respect to the bilinear form B. Recall that a full
flag F = {0 = F0 ⊂ F1 ⊂ F2 ⊂ · · · ⊂ FR ⊂ FR+1 = K} ∈ FL(K) is called isotropic with respect
to B if Fk = F⊥R+1−k for k = 1, . . . , R.
Theorem 5.17. Suppose K is cyclotomically self-dual. A full flag F ∈ FL(K) is isotropic if
and only if the associated tuple yF is cyclotomic.
Proof. Let (ui(x))
R+1
i=1 be a basis of K adjusted to F , so that we have (5.11).
For the “only if” direction, suppose yF is cyclotomic. By Lemma 5.16,
Fk = spanC(u1, . . . , uk) = spanC(WR+1(−x), . . . ,WR+2−k(−x)).
We also have F⊥R+1−k = spanC(u1, . . . , uR+1−k)
⊥ = spanC(WR+1(−x), . . . ,WR+2−k(−x))
by (5.5). Therefore Fk = F⊥R+1−k.
For the “if” direction, suppose F = {Fk} is isotropic. Since Fk = F⊥R+1−k, and given (5.5),
we have two bases for Fk, namely (u1, . . . , uk) and (WR+1(−x), . . . ,WR+2−k(−x)). So to prove
that y is cyclotomic it suffices to establish the following lemma, which is the converse of
Lemma 5.16.
32 A. Varchenko and C.A.S. Young
Lemma 5.18. If
spanC(u1(−x), . . . , uk(−x)) = spanC(WR+1,WR, . . . ,WR+2−k), k = 1, . . . , R+ 1,
then y is cyclotomic.
Proof. Examining the induction in the proof of Lemma 5.16, one sees that we also have, by
a similar induction, that if spanC(u1(−x), . . . , uk(−x)) = spanC(WR+1,WR, . . . ,WR+2−k) for
each k then (5.16) must hold for each k, which says that y is cyclotomic. 
This completes the proof of Theorem 5.17. 
In view of Proposition 5.13 we have the following corollary.
Corollary 5.19. If F ∈ FL(K) is isotropic then F is decomposable.
5.7 Witt bases and the symmetries of the bilinear form B
We say that (rk)
R+1
k=1 is a Witt basis of the cyclotomically self-dual space K if
Wr†(r1, . . . , r̂k, . . . , rR+1) ' rR+2−k(−x), k = 1, . . . , R+ 1. (5.17)
The following lemma gives a useful alternative characterization of Witt bases.
Lemma 5.20. The basis (rk)
R+1
k=1 is a Witt basis if and only if
B(ri, rj) = 0 whenever i+ j 6= R+ 2. (5.18)
Proof. Suppose (uk)
R+1
k=1 is a basis of K and let (Wk)
R+1
k=1 be as in (5.5). Then (Wi(x))
R+1
i=1 and
(ui(−x))
R+1
i=1 are two bases of K
† and so ui(−x) =
R+1∑
j=1
CijWj(x), for some invertible matrix Cij .
We have B(ui, uj) =
R+1∑
k=1
Cjk Wr
†(ui, u1, u2, . . . , ûk, . . . , uR+1) = (−1)i−1Cji Wr
†(u1, . . . , uR+1).
Hence (5.17) is equivalent to (5.18). 
Theorem 5.21. Every cyclotomically self-dual space K has a special basis (rk)
R+1
k=1 which is also
Witt basis, and in which in fact
Wr†(r1, . . . , r̂k, . . . , rR+1) = (−1)
− deg rR+2−krR+2−k(−x), k = 1, . . . , R+ 1. (5.19)
Proof. Let (uk(x))
R+1
k=1 be a special basis of K. We may suppose that the uk(x) all have leading
coefficient 1. Let (Wk(x))
R+1
k=1 be the basis of K
† as in (5.5). By Lemma 5.11, degWk = d
†
k. By
Lemma 5.5, we have
Wk = Wr
†(u1, . . . , ûk, . . . , uR+1) = Dkx
d†k + · · · ,
where the ellipsis indicates terms of lower degree in x and where
Dk :=
∏
1≤j<i≤R+1
i 6=k, j 6=k
(di − dj), k = 1, . . . , R+ 1.
Since K is cyclotomically self-dual we must have
dk = d
†
R+2−k, k = 1, . . . , R+ 1,
Populations of Solutions to Cyclotomic Bethe Equations 33
and
Wk = Wr
†(u1, . . . , uˆk, . . . , uR+1) = Dk(−1)
−dR+2−kuR+2−k(−x) + · · · .
Now from (5.4) we have
dk − dl = dR+2−l − dR+2−k, 1 ≤ k < l ≤ R+ 1,
using which one verifies that
Dk = DR+2−k, k = 1, . . . , R+ 1.
Given this equality, if we set
qk := ukD
1
2
k
R+1∏
j=1
D
− 12R−2
j
then we have
Wr†(q1, . . . , q̂k, . . . , qR+1) = (−1)
−dR+2−kqR+2−k(−x) + · · · .
In this way, we arrive at
Wr†(q1, . . . , q̂k, . . . , qR+1)
= (−1)−dR+2−kqR+2−k(−x) +
R+1∑
j=2
qR+2−j(−x)c
j
k, k = 1, . . . , R+ 1, (5.20)
for some constants cjk. That is, we have
Wr†(q1, . . . , qR−1, qR, q̂R+1) = (−1)
−d1q1(−x),
Wr†(q1, . . . , qR−1, q̂R, qR+1) = (−1)
−d2q2(−x) + c
1
Rq1(−x),
Wr†(q1, . . . , q̂R−1, qR, qR+1) = (−1)
−d3q3(−x) + c
2
R−1q2(−x) + c
1
R−1q1(−x),
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · .
We define
(−1)−d1r1 := (−1)
−d1q1, (−1)
−d2r2 := (−1)
−d2q2 + c
1
Rr1,
so that
Wr†(r1, r2, q3, . . . , qR−1, qR, q̂R+1) = (−1)
−d1r1(−x),
Wr†(r1, r2, q3, . . . , qR−1, q̂R, qR+1) = (−1)
−d2r2(−x),
Wr†(r1, r2, q3, . . . , q̂R−1, qR, qR+1) = (−1)
−d3q3(−x) + c˜
2
R−1r2(−x) + c˜
1
R−1r1(−x),
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
for some new constants c˜jk, and we then define
(−1)−d3r3 := (−1)
−d3q3 + c˜
1
R−1r2 + c˜
1
R−1r1,
and so on. By an obvious induction, we arrive at a Witt basis (rk)
R+1
k=1 . By construction
deg rk = dk. Finally, note in (5.20) that c
j
k can be non-zero only when dk − dj ∈ Z since both
sides lie in either C[x] or x
1
2C[x]. Therefore this Witt basis (rk)
R+1
k=1 is special. 
34 A. Varchenko and C.A.S. Young
Lemma 5.22. Let (rk)
R+1
k=1 be the Witt basis of Theorem 5.21. Then Wr
†(r1, . . . , rR+1) = 1.
Proof. We have
Wr†(q1, . . . , qR+1) = Wr
†(u1, . . . , uR+1)
R+1∏
k=1

D
1
2
k
R+1∏
j=1
D
− 12R−2
j


= Wr†(u1, . . . , uR+1)
R+1∏
k=1
D
− 1R−1
k .
But then noting that
R+1∏
k=1
Dk =
∏
1≤j<i≤R+1
(di − dj)
R−1
and recalling Lemma 5.6, one finds
Wr†(q1, . . . , qR+1) = 1
and hence the result. 
Theorem 5.23. The subspaces KSp and KO are mutually orthogonal with respect to B. The
bilinear form B is skew-symmetric on KSp and symmetric on KO.
Proof. Let (rk)
R+1
k=1 be the special Witt basis constructed in Theorem 5.21. From (5.19) and
Lemma 5.22, we have
B(rk, rR+2−k) = (−1)
dR+2−k+k+1, k = 1, . . . , R+ 1, (5.21)
and B(ri, rj) = 0 if i + j 6= R + 2. This implies in particular that KSp and KO are mutually
orthogonal. By Lemma 5.12 it also gives
B(rk, rR+2−k)B(rR+2−k, rk) = (−1)
〈Λ,α∨1 +···+α
∨
R〉, k = 1, . . . , R+ 1.
Recall the definition of Λ, (5.2). Now 〈Λs + σΛs, α∨1 + · · ·+ α
∨
R〉 ∈ 2Z for each s = 1, . . . , N ,
since Λs is integral. Therefore it follows from (5.1) that
〈Λ, α∨1 + · · ·+ α
∨
R〉 ∈
{
2Z+ 1, p > 0,
2Z, p = 0.
Consider the case p > 0. Then we have
B(rk, rR+2−k)B(rR+2−k, rk) = −1, k = 1, . . . , R+ 1. (5.22)
Recall from (5.7) that deg rk and deg rR+2−k are both half odd integers if k = p + 1, . . . , R +
1 − p, and are integers otherwise. Hence, by (5.21), B(rk, rR+2−k) and B(rR+2−k, rk) lie in
{(−1)
1
2 , (−1)−
1
2 } if k = p+ 1, . . . , R+ 1−p and in {1,−1} otherwise. Combining this statement
with (5.22) we find
B(rk, rR+2−k) =
{
−B(rR+2−k, rk), k = 1, . . . , p, R+ 2− p, . . . , R+ 1,
+B(rR+2−k, rk), k = p+ 1, . . . , R+ 1− p,
which is the required result.
Populations of Solutions to Cyclotomic Bethe Equations 35
Finally, consider the case p = 0. Then
B(rk, rR+2−k)B(rR+2−k, rk) = 1, k = 1, . . . , R+ 1,
and since in this case deg rk is integral for all k, this implies
B(rk, rR+2−k) = B(rR+2−k, rk), k = 1, . . . , R+ 1
as required. 
The following are corollaries of Theorem 5.23 together with Lemma 5.20.
Corollary 5.24. Every Witt basis (rk)
R+1
k=1 of K is decomposable.
A basis (rk)
R+1
k=1 of K such that
Bij := B(ri, rj) = δR+2−i,jbi
with
bk :=



(−1)k, k = 1, . . . , p,
+1, k = p+ 1, . . . , R+ 1− p,
(−1)R+1−k, k = R+ 2− p, . . . , R+ 1
is called a reduced Witt basis. By Lemma 5.20, reduced Witt bases are Witt bases.
Corollary 5.25. Any Witt basis can be transformed to a reduced Witt basis by a suitable diagonal
transformation followed by a suitable permutation of the basis vectors.
Corollary 5.26. For any Witt basis (rk)
R+1
k=1 of K, the full flag F = {F1⊂ F2⊂· · ·⊂ FR+1 = K}
given by Fk = spanC(r1, . . . , rk), k = 1, . . . , R + 1, is isotropic (and hence the corresponding
tuple yF is cyclotomic by Theorem 5.17).
Conversely, given any isotropic full flag F = {F1 ⊂ F2 ⊂ · · · ⊂ FR+1 = K} there is a Witt
basis (rk)
R+1
k=1 such that Fk = spanC(r1, . . . , rk), k = 1, . . . , R + 1. If in addition F is of type S
then this basis can be chosen to be a reduced Witt basis.
Lemma 5.27. The full flag F given in (5.10) is isotropic and hence the corresponding tuple yF
is cyclotomic.
Proof. We can choose the special basis (uk)
R+1
k=1 defining F to be the Witt basis of Theorem 5.21.
Then the result follows from Corollary 5.26. 
5.8 Isotropic flags
Recall from Section 5.3 the notion of a symmetric subset of {1, . . . , R+ 1}.
Lemma 5.28. Let Q ⊂ {1, . . . , R + 1} be a 2p-element subset. The variety FLQ(K) contains
an isotropic flag if and only if Q is symmetric.
Lemma 5.29. If Q is symmetric then the variety FL⊥Q(K) of isotropic flags is isomorphic to the
direct product of spaces of isotropic flags FL⊥(KSp) × FL⊥(KO) and the isomorphism of these
varieties is given by the map ηQ defined in (5.8).
In view of these lemmas and Theorem 5.17, we have the following description of the subspace
of all cyclotomic tuples within the image β(FL(K)) ⊂ P
(
C
[
x
1
2
])R
.
Theorem 5.30. The irreducible components of the space β(FL⊥(K)) of all cyclotomic tuples
are labeled by symmetric subsets Q ⊂ {1, . . . , R+ 1}. The components do not intersect and each
is isomorphic to FL⊥(KSp)× FL⊥(KO).
36 A. Varchenko and C.A.S. Young
5.9 Infinitesimal deformation of isotropic flags of type S
The connected Lie group of endomorphisms of K preserving B acts transitively on the variety
of isotropic full flags of type Q, FL⊥Q(K), for each symmetric subset Q ⊂ {1, . . . , R + 1}. In
particular it acts transitively on FL⊥S (K), and hence on the cyclotomic tuples of polynomials
in the image β(FL⊥S (K)) ⊂ P(C[x])
R. We shall describe the infinitesimal action of this group
on β(FL⊥S (K)).
The connected Lie group of endomorphisms of K preserving B preserves each of the subspa-
ces KSp and KO. Thus this group is the product Sp(KSp) × SO(KO) of the group of special
symplectic transformations in End(KSp) and the group of special orthogonal transformations in
End(KO). Its Lie algebra sp(KSp)⊕ so(KO) consists of all traceless endomorphisms X of K such
that
B(Xu, v) +B(u,Xv) = 0
for all u, v ∈ K.
Pick any isotropic full flag F = {F1 ⊂ F2 ⊂ · · · ⊂ FR+1 = K} of type S. Then β(F) = yF
is a cyclotomic tuple of polynomials by Lemma 5.8. Let (rk)
R+1
k=1 be a reduced Witt basis such
that Fk = spanC(r1, . . . , rk), k = 1, . . . , R+ 1. Such a basis exists by Corollary 5.26.
This choice of basis gives identifications KSp ∼= C2p and KO ∼= CR+1−2p and hence sp(KSp) ∼=
sp2p and so(KO) ∼= soR+1−2p. The Lie algebra sp2p has root system of type Cp. The Lie algebra
soR+1−2p has root system of type Dn−p if R = 2n− 1 is odd and of type Bn−p if R = 2n is even.
Let (Ei,j)
R+1
i,j=1 be the basis of End(K) defined by
Ei,jrk = δikrj .
The lower-triangular subalgebra of sp(KSp) ∼= sp2p is generated by
Xk := Ek+1,k + ER+2−k,R+1−k, k = 1, . . . , p− 1,
and
Xp := ER+2−p,p.
When R = 2n− 1, the lower-triangular subalgebra of so(KO) ∼= so2n−2p is generated by
Yk := Ek+p,k+p−1 − E2n−p−k+1,2n−p−k, k = 1, . . . , n− p− 1,
and
Y˜n−p−1 := Ek+p+1,k+p−1 − E2n−p−k+1,2n−p−k−1.
When R = 2n, the lower-triangular subalgebra of so(KO) ∼= so2n−2p+1 is generated by
Zk := Ek+p,k+p−1 − E2n−p−k+2,2n−p−k+1, k = 1, . . . , n− p.
These generators define linear transformations belonging to End(KSp)⊕ End(KO).
Remark 5.31. The Lie algebra so(KO) ⊕ sp(KSp) is contained in the simple Lie superalgeb-
ra osp(K) of all orthosymplectic transformations of the space K. See [10] for the definition. It
would be interesting to understand whether this superalgebra plays a role here.
Populations of Solutions to Cyclotomic Bethe Equations 37
For any k = 1, . . . , p and all c ∈ C, the basis ecXkr is again a Witt basis of K. Let ecXkF
denote the corresponding isotropic flag and β(ecXkF) the corresponding tuple representing a
cyclotomic point. Let us describe the dependence on c of this tuple.
For k = 1, . . . , p− 1, we have
ecXkr = (r1, . . . , rk−1, rk + crk+1, rk+1, . . . , rR−k, rR+1−k + crR+2−k, rR+2−k, . . . , rR+1)
and hence
β
(
ecXkF
)
=
(
yF1 , . . . , y
F
k−1, yk(x, c), y
F
k+1, . . . , y
F
R+1−k, yR+1−k(x, c), y
F
R+2−k, . . . , y
F
R+1
)
,
where
yk(x, c) := Wr
†(r1, . . . , rk−1, rk + crk+1) = y
F
k + cWr
†(r1, . . . , rk−1, rk+1) (5.23a)
and
yR+1−k(x, c) := Wr
†(r1, . . . , rR−k, rR+1−k + crR+2−k)
= yFR+1−k + cWr
†(r1, . . . , rR−k, rR+2−k). (5.23b)
Finally (for k = p) we have
ecXpr = (r1, . . . , rp−1, rp + crR+2−p, rp+1, . . . , rR+1)
and hence
β
(
ecXpF
)
=
(
yF1 , . . . , y
F
p−1, yp(x, c), y
F
p+1, . . . , . . . , y
F
R+1
)
,
yp(x, c) := Wr
†(r1, . . . , rp−1, rp + crR+2−p) = y
F
p + cWr
†(r1, . . . , rp−1, rR+2−p). (5.24)
The flows in P(C[x])R corresponding to the generators of so(KO) can be described similarly.
5.10 Populations of cyclotomic critical points in type A
Recall the definition of the extended master function Φ̂, (2.8). In the setting of the present
section (see Section 5.1) it has the explicit form
Φ̂(t; c; z; Λ; Λ0) =
N∑
i=1
(Λ0,Λi)(log(−zi) + log(zi)) +
N∑
i=1
(Λi,ΛR+1−i) log 2zi
+
∑
1≤i<j≤n
(Λi,Λj) log(zi − zj) +
∑
1≤i<j≤n
(ΛR+1−i,Λj) log(−zi − zj)
+
∑
1≤i<j≤n
(Λi,ΛR+1−j) log(zi + zj)
+
∑
1≤i<j≤n
(ΛR+1−i,ΛR+1−j) log(−zi + zj)−
m˜∑
j=1
(αc(j),Λ0) log(tj)
−
N∑
i=1
m˜∑
j=1
(αc(j),Λi) log(tj − zi)−
N∑
i=1
m˜∑
j=1
(αc(j),ΛR+1−i) log(tj + zi)
+
∑
1≤i<j≤m˜
(αc(i), αc(j)) log(ti − tj) (5.25)
38 A. Varchenko and C.A.S. Young
and the critical point equations (2.9) become
0 =
N∑
i=1
(αc(j),Λi)
tj − zi
+
N∑
i=1
(αc(j),ΛR+1−i)
tj + zi
+
(αc(j),Λ0)
tj
−
m˜∑
i=1
i 6=j
(αc(j), αc(i))
tj − ti
, j = 1, . . . , m˜. (5.26)
Given a tuple of polynomials y ∈ P(C[x])R, we have the pair (t, c) ∈ Cm˜ × Im˜ represented
by y in the sense of Section 4.2. We say the tuple y represents a critical point of Φ̂ if t is a
critical point of Φ̂(t; c; z; Λ; Λ0), i.e., if (t, c) satisfy the equations (5.26).
The following theorem says that we can go from cyclotomic critical points of the extended
master function Φ̂, (5.25), to decomposable cyclotomically self-dual vector spaces of quasi-
polynomials.
Theorem 5.32. Suppose y ∈ P(C[x])R represents a cyclotomic critical point of Φ̂, (5.25).
The kernel kerD(y) of the fundamental differential operator D(y), Section 5.4, is a decom-
posable cyclotomically self-dual vector space of quasi-polynomials with frame T˜1, . . . , T˜R; Λ˜∞,
where Λ˜∞ is the unique dominant weight in the orbit of Λ∞(y), (4.10), under the shifted action
of the Weyl group of type AR.
There exists an isotropic flag F ∈ FL⊥S (kerD(y)) such that y = β(F).
Proof. Arguing as in [15] – see especially Lemma 5.10 – we have that kerD(y) is a vector space
of quasi-polynomials with frame T˜1, . . . , T˜R; Λ˜∞, and that the flag F ∈ FL(ker(D(y))) such that
β(F) = y can be constructed as follows. Define quasi-polynomials y(i,i+1,...,k)i , 1 ≤ i ≤ k ≤ R,
recursively by
Wr(y(k)k , yk) = yk−1T˜kyk+1, Wr(y
(i,i+1,...,k)
i , yi) = yi−1T˜iy
(i+1,...,k)
i+1 , i < k
(recall we set y0 = yR+1 = 1 for convenience). Set u1 = y1 and uk = y
(1,...,k−1)
1 for k =
2, . . . , R+ 1. Then (uk)
R+1
k=1 is a basis of kerD(y). Moreover Wr
†(u1, . . . , uk) = yk, k = 1, . . . , R.
That is, β(F) = y for the flag F = {Fk} given by Fk = spanC(u1, . . . , uk), k = 1, . . . , R+1. Since
y is cyclotomic, Theorem 5.14 states that kerD(y) is cyclotomically self-dual. By Lemma 5.9,
F is a decomposable flag of type S, and by Theorem 5.17 it is isotropic. 
Conversely, we have the following, arguing as in Lemmas 3.1, 3.2 and 5.15 in [15] and using
Theorem 5.17.
Theorem 5.33. Let K be a decomposable cyclotomically self-dual vector space of quasi-polyno-
mials with frame T˜1, . . . , T˜R; Λ˜∞.
Suppose there exists an isotropic flag F ∈ FL⊥S (K) such that the tuple y
F is generic. Then yF
represents a cyclotomic critical point of Φ̂, (5.25).
Since being generic is an open condition, the set of generic tuples in the image β(FL⊥S (K)) is
either empty or it is open and dense in β(FL⊥S (K)).
Starting from an initial tuple y that represents a cyclotomic critical point of Φ̂, (5.25), we
may let K = kerD(y) as in Theorem 5.32. Then we have the variety
β(FL⊥S (K)) ∼= FL
⊥(KSp)× FL
⊥(KO), (5.27)
where the isomorphism is by Theorem 5.30. Almost all of the tuples in β(FL⊥S (K)) are generic
and hence represent cyclotomic critical points of Φ̂. Call this variety β(FL⊥S (K)) ⊂ P(C[x])
R the
cyclotomic population originated at y.
Populations of Solutions to Cyclotomic Bethe Equations 39
5.11 The case p = n
Consider the case p = n in (5.1). Namely, suppose that we are either in
• type A2n−1 with Λ0 integral and 〈Λ0, αn〉 odd, or
• type A2n with 〈Λ0, α∨i 〉 ∈ Z for all i < n and 〈Λ0, α
∨
n〉 ∈
1
2(2Z≥0 − 1) =
{
−12 ,
1
2 ,
3
2 , . . .
}
.
Then Λ0 obeys the assumptions from Section 4.1 and so we are in the setting of Section 4. That
means we have two notions of a cyclotomic population: the one in the previous subsection, and
the one in Section 4.7. Let us show that these two notions coincide.
Theorem 5.34. Let p = n in (5.1). Let y represent a cyclotomic critical point of the ex-
tended master function Φ̂ of (5.25). Then the variety β(FL⊥S (K)) is isomorphic to the variety of
isotropic full flags in a complex symplectic vector space of dimension 2n. The cyclotomic popula-
tion in P(C[x])R originated at y in the sense of Section 4.7 coincides with this variety β(FL⊥S (K)).
Proof. When p = n we have either KO = {0}, if R = 2n− 1, or KO ∼= C, if R = 2n. In either
case FL⊥(KO) is a point, and (5.27) reduces to
β
(
FL⊥S (K)
) ∼= FL⊥(KSp),
i.e., β(FL⊥S (K)) is isomorphic to the variety of isotropic full flags in the vector space KSp
∼= C2n
endowed with the symplectic form B|KSp .
Starting from any such isotropic full flag, F ∈ FL⊥S (K), we choose a reduced Witt basis
adapted to F (Corollary 5.26). Then every other flag in FL⊥S (K) can be reached by an element of
the lower-triangular (as in Section 5.9) unipotent subgroup of Sp2n. This subgroup is generated
by the one-parameter groups corresponding to negative simple root generators Xk of Section 5.9.
Lemmas 5.35 and 5.36 below show that the flows in β(FL⊥S (K)) generated by the Xk coincide
with notion of cyclotomic generation from Section 4. That shows that the set of all tuples of
polynomials obtained from y by repeated cyclotomic generation, in all directions i ∈ I, contains
a non-empty open subset of β(FL⊥S (K)). Therefore it is dense in β(FL
⊥
S (K)). Hence its Zariski
closure is β(FL⊥S (K)) itself. 
Lemma 5.35. The image β(ecXkF) ∈ P(C[x])R coincides with the tuple y(k,σ)(1/c) of Theo-
rem 4.6, for every k = 1, . . . , n− 1 (and also for k = n when we are in type AR = A2n−1).
Proof. It is enough to note that, in view of (5.23) and Lemma 5.15, we have
Wr(yFk , yk(x, c)) = cT˜ky
F
k−1y
F
k+1,
Wr(yFR+1−k, yR+1−k(x, c)) = cT˜R+1−ky
F
R−ky
F
R+2−k. 
It remains to consider the case k = n in type A2n.
Lemma 5.36. In type A2n, the image β(e−cXnF)∈P(C[x])2n coincides with the tuple y(n,σ)(1/c)
of Theorem 4.20.
Proof. We have (5.24) with p = n. Namely,
ecXnr = (r1, . . . , rn−1, rn + crn+2, rn+1, rn+2, . . . , r2n+1)
and hence
β
(
ecXnF
)
=
(
yF1 , . . . , y
F
n−1, yn(x, c), yn+1(x, c), y
F
n+2, . . . , y
F
2n+1
)
,
40 A. Varchenko and C.A.S. Young
where
yn(x, c) := Wr
†(r1, . . . , rn−1, rn + crn+2) = y
F
n + cWr
†(r1, . . . , rn−1, rn+2)
and
yn+1(x, c) := Wr
†(r1, . . . , rn−1, rn + crn+2, rn+1) = y
F
n+1 + cWr
†(r1, . . . , rn−1, rn+2, rn+1).
Now let
y(n)n := Wr
†(r1, . . . , rn−1, rn+1).
Then by Lemma 5.15 we have
Wr
(
yFn , y
(n)
n
)
= T˜ny
F
n−1y
F
n+1,
Wr
(
yFn+1, yn+1(x, c)
)
= −cT˜n+1y
(n)
n y
F
n+2,
Wr
(
y(n)n , yn(x, c)
)
= T˜nyn−1yn+1(x, c).
This establishes the lemma. 
Acknowledgments
The research of AV is supported in part by NSF grant DMS-1362924. CY is grateful to the
Department of Mathematics at UNC Chapel Hill for hospitality during a visit in October 2014
when this work was initiated. CY thanks Benoit Vicedo for valuable discussions.
References
[1] Babujian H.M., Flume R., Off-shell Bethe ansatz equation for Gaudin magnets and solutions of Knizhnik–
Zamolodchikov equations, Modern Phys. Lett. A 9 (1994), 2029–2039, hep-th/9310110.
[2] Brochier A., A Kohno–Drinfeld theorem for the monodromy of cyclotomic KZ connections, Comm. Math.
Phys. 311 (2012), 55–96, arXiv:1011.4285.
[3] Crampe´ N., Young C.A.S., Integrable models from twisted half-loop algebras, J. Phys. A: Math. Theor. 40
(2007), 5491–5509, math-ph/0609057.
[4] Enriquez B., Quasi-reflection algebras and cyclotomic associators, Selecta Math. (N.S.) 13 (2007), 391–463,
math.QA/0408035.
[5] Feigin B., Frenkel E., Reshetikhin N., Gaudin model, Bethe ansatz and critical level, Comm. Math. Phys.
166 (1994), 27–62, hep-th/9402022.
[6] Frenkel E., Gaudin model and opers, in Infinite dimensional algebras and quantum integrable systems,
Progr. Math., Vol. 237, Birkha¨user, Basel, 2005, 1–58, math.QA/0407524.
[7] Fuchs J., Schellekens B., Schweigert C., From Dynkin diagram symmetries to fixed point structures, Comm.
Math. Phys. 180 (1996), 39–97, hep-th/9506135.
[8] Gaudin M., Diagonalisation d’une classe d’Hamiltoniens de spin, J. Physique 37 (1976), 1089–1098.
[9] Gaudin M., La fonction d’onde de Bethe, Collection du Commissariat a` l’E´nergie Atomique: Se´rie Scien-
tifique, Masson, Paris, 1983.
[10] Kac V.G., A sketch of Lie superalgebra theory, Comm. Math. Phys. 53 (1977), 31–64.
[11] Kac V.G., Infinite-dimensional Lie algebras, 3rd ed., Cambridge University Press, Cambridge, 1990.
[12] Kagan D., Young C.A.S., Conformal sigma models on supercoset targets, Nuclear Phys. B 745 (2006),
109–122, hep-th/0512250.
[13] Mukhin E., Tarasov V., Varchenko A., Bethe eigenvectors of higher transfer matrices, J. Stat. Mech. Theory
Exp. 2006 (2006), P08002, 44 pages, math.QA/0605015.
Populations of Solutions to Cyclotomic Bethe Equations 41
[14] Mukhin E., Varchenko A., Remarks on critical points of phase functions and norms of Bethe vectors,
in Arrangements – Tokyo 1998, Adv. Stud. Pure Math., Vol. 27, Kinokuniya, Tokyo, 2000, 239–246,
math.RT/9810087.
[15] Mukhin E., Varchenko A., Critical points of master functions and flag varieties, Commun. Contemp. Math.
6 (2004), 111–163, math.QA/0209017.
[16] Mukhin E., Varchenko A., Miura opers and critical points of master functions, Cent. Eur. J. Math. 3 (2005),
155–182, math.QA/0312406.
[17] Mukhin E., Varchenko A., Quasi-polynomials and the Bethe ansatz, in Groups, Homotopy and Configuration
Spaces, Geom. Topol. Monogr., Vol. 13, Geom. Topol. Publ., Coventry, 2008, 385–420, math.QA/0604048.
[18] Reshetikhin N., Varchenko A., Quasiclassical asymptotics of solutions to the KZ equations, in Geometry,
Topology, & Physics, Conf. Proc. Lecture Notes Geom. Topology, Vol. 4, Int. Press, Cambridge, MA, 1995,
293–322, hep-th/9402126.
[19] Schechtman V.V., Varchenko A.N., Arrangements of hyperplanes and Lie algebra homology, Invent. Math.
106 (1991), 139–194.
[20] Scherbak I., Varchenko A., Critical points of functions, sl2 representations, and Fuchsian differential equa-
tions with only univalued solutions, Mosc. Math. J. 3 (2003), 621–645, math.QA/0112269.
[21] Skrypnyk T., Integrable quantum spin chains, non-skew symmetric r-matrices and quasigraded Lie algebras,
J. Geom. Phys. 57 (2006), 53–67.
[22] Skrypnyk T., “Z2-graded” Gaudin models and analytical Bethe ansatz, Nuclear Phys. B 870 (2013), 495–
529.
[23] Varagnolo M., Vasserot E., Cyclotomic double affine Hecke algebras and affine parabolic category O, Adv.
Math. 225 (2010), 1523–1588, arXiv:0810.5000.
[24] Varchenko A., Bethe ansatz for arrangements of hyperplanes and the Gaudin model, Mosc. Math. J. 6
(2006), 195–210, math.QA/0408001.
[25] Varchenko A., Quantum integrable model of an arrangement of hyperplanes, SIGMA 7 (2011), 032, 55 pages,
arXiv:1001.4553.
[26] Vicedo B., Young C.A.S., Cyclotomic Gaudin models: construction and Bethe ansatz, arXiv:1409.6937.
[27] Vicedo B., Young C.A.S., Vertex Lie algebras and cyclotomic coinvariants, arXiv:1410.7664.
[28] Young C.A.S., Flat currents, Zm gradings and coset space actions, Phys. Lett. B 632 (2006), 559–565.