Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study
We construct explicit geometric models for moduli spaces of semi-stable strongly parabolic Higgs bundles over the Riemann sphere, in the case of rank two, four marked points, arbitrary degree, and arbitrary weights. The mechanism of construction relies on elementary geometric and combinatorial techn...
Saved in:
| Published in: | Symmetry, Integrability and Geometry: Methods and Applications |
|---|---|
| Date: | 2022 |
| Main Author: | |
| Format: | Article |
| Language: | English |
| Published: |
Інститут математики НАН України
2022
|
| Online Access: | https://nasplib.isofts.kiev.ua/handle/123456789/211725 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Cite this: | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study. Claudio Meneses. SIGMA 18 (2022), 062, 41 pages |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| _version_ | 1859863904743587840 |
|---|---|
| author | Meneses, Claudio |
| author_facet | Meneses, Claudio |
| citation_txt | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study. Claudio Meneses. SIGMA 18 (2022), 062, 41 pages |
| collection | DSpace DC |
| container_title | Symmetry, Integrability and Geometry: Methods and Applications |
| description | We construct explicit geometric models for moduli spaces of semi-stable strongly parabolic Higgs bundles over the Riemann sphere, in the case of rank two, four marked points, arbitrary degree, and arbitrary weights. The mechanism of construction relies on elementary geometric and combinatorial techniques, based on a detailed study of orbit stability of (in general non-reductive) bundle automorphism groups on certain carefully crafted spaces. The aforementioned techniques are not exclusive to the case we examine, and this work elucidates a general approach to construct arbitrary moduli spaces of semi-stable parabolic Higgs bundles in genus 0, which is encoded into the combinatorics of weight polytopes. We also present a comprehensive analysis of the geometric models' behavior under variation of parabolic weights and wall-crossing, which is concentrated on their nilpotent cones.
|
| first_indexed | 2026-03-16T23:40:37Z |
| format | Article |
| fulltext |
Symmetry, Integrability and Geometry: Methods and Applications SIGMA 18 (2022), 062, 41 pages
Geometric Models and Variation of Weights on Moduli
of Parabolic Higgs Bundles over the Riemann Sphere:
a Case Study
Claudio MENESES
Mathematisches Seminar, Christian-Albrechts Universität zu Kiel,
Heinrich-Hecht-Platz 6, 24118 Kiel, Germany
E-mail: meneses@math.uni-kiel.de
URL: http://www.math.uni-kiel.de/geometrie/de/claudio-meneses/
Received September 21, 2021, in final form July 28, 2022; Published online August 13, 2022
https://doi.org/10.3842/SIGMA.2022.062
Abstract. We construct explicit geometric models for moduli spaces of semi-stable strongly
parabolic Higgs bundles over the Riemann sphere, in the case of rank two, four marked
points, arbitrary degree, and arbitrary weights. The mechanism of construction relies
on elementary geometric and combinatorial techniques, based on a detailed study of or-
bit stability of (in general non-reductive) bundle automorphism groups on certain carefully
crafted spaces. The aforementioned techniques are not exclusive to the case we examine, and
this work elucidates a general approach to construct arbitrary moduli spaces of semi-stable
parabolic Higgs bundles in genus 0, which is encoded into the combinatorics of weight poly-
topes. We also present a comprehensive analysis of the geometric models’ behavior under
variation of parabolic weights and wall-crossing, which is concentrated on their nilpotent
cones.
Key words: parabolic Higgs bundle; Hitchin fibration; nilpotent cone
2020 Mathematics Subject Classification: 14H60; 14D22; 32G13; 22E25
To Professor Leon A. Takhtajan on his 70th birthday,
with profound respect and admiration.
1 Introduction
The moduli spaces of parabolic Higgs bundles on a Riemann surface of lowest possible dimen-
sion are elliptic surfaces in virtue of their Hitchin fibrations [16]. These examples occur in low
genus and encode at once, in its simplest possible form, the characteristic features of the rich
and intricate geometry occuring in arbitrary dimensions. From the general theory of elliptic
surfaces [25], they are necessarily biholomorphic to one of the fibrations in Kodaira’s list. Fol-
lowing the seminal work of Gaiotto–Moore–Neitzke [8], their hyperkähler geometry has recently
attracted renewed interest in relation to the study of gravitational instantons of ALG type [7].
The toy model [13] for the Hitchin fibration and C∗-action of the moduli spaces of rank 2
parabolic Higgs bundles over CP1 with four marked points, is the elliptic surface Mtoy associated
to a choice (z1, z2, z3, z4) ∈ Conf4(CP1) as follows. Let ΣD → CP1 be the elliptic curve of
D = z1+ z2+ z3+ z4, with involution τ and ramification points {w1, w2, w3, w4}. The extension
of τ to an action in ΣD × C in terms of the character generated by ρ(τ) = −1 determines
a 2-dimensional orbifold (τ × ρ)\(ΣD ×C) which is smooth away of the points [(wi, 0)]. Mtoy is
This paper is a contribution to the Special Issue on Mathematics of Integrable Systems: Classical and Quan-
tum in honor of Leon Takhtajan.
The full collection is available at https://www.emis.de/journals/SIGMA/Takhtajan.html
mailto:meneses@math.uni-kiel.de
http://www.math.uni-kiel.de/geometrie/de/claudio-meneses/
https://doi.org/10.3842/SIGMA.2022.062
https://www.emis.de/journals/SIGMA/Takhtajan.html
2 C. Meneses
defined as the desingularization of (τ × ρ)\(ΣD × C) by blow-up at those points. The elliptic
fibration π2 : Mtoy → C is given by extending the map [(w, z)] 7→ z2 trivially along exceptional
divisors, while its C∗-action is induced by the standard C∗-action on C. The loci where the
(holomorphic) involution extending τ × ρ acts freely is a Zariski open subset U ∼= T ∗CP1\CP1,
equipping Mtoy with an additional projection π1 : U → CP1. Intuitively, a biholomorphism
between Mtoy and the aforementioned moduli spaces of semi-stable parabolic Higgs bundles
could be constructed by relating π1 to the forgetful map to parabolic structures, and π2 to the
determinant map on parabolic Higgs fields.
There are some caveats when trying to carry out the previous biholomorphisms, though. First
of all, they are not expected to hold for other ranks and number of marked points. A nontrivial
feature of parabolic Higgs bundles is the dependence of their moduli problem on a choice of
parabolic weights. While several classical results on this dependence [27, 32] are linked to the
study of the birational geometry of the moduli problem, not only they are trivial when the moduli
space is a complex surface, but also they shed no light on other subtle phenomena. For instance,
it is unclear how holomorphic invariants such as Harder–Narasimhan stratifications may depend
on parabolic weights, or what happens to the moduli space when semi-stable parabolic bundles
don’t exist [1, 5].1 Moreover, the natural hyperkähler structures on these moduli spaces depend
nontrivially on parabolic weights [17, 24], and the hyperkähler moduli problem motivates the
search for suitable complex-analytic structures capturing this dependence. The will of attacking
these problems is the main motivation to carry out this work. More specifically, this work sprung
as a natural advance towards the parabolic Higgs bundle generalization of the results in [24],
dealing with the study of Kähler forms on moduli spaces of semi-stable parabolic bundles, where
the understanding of refined moduli space stratifications is crucial.
We propose a construction of moduli spaces of rank 2 stable parabolic Higgs bundles on
CP1 with four marked points, on which the dependence on parabolic weights and its wall-
crossing phenomena becomes as transparent as possible. The construction is comprised of two
steps. The first step is based on the observation that the peculiarities of genus 0 (namely, the
infinitesimal rigidity of holomorphic bundles dictated by the Birkhoff–Grothendieck theorem)
lead to the construction of Harder–Narasimhan strata for underlying holomorphic bundles E ∼=
O(m1) ⊕ O(m2) as semi-stable orbit spaces under the action of bundle automorphisms, since
any isomorphism class of quasi-parabolic Higgs bundles on E can be modeled as an orbit in
an explicit complex manifold QPH(E) with respect to an action of the group P(Aut(E)) of
projective automorphisms of E.2 The mechanism of construction is entirely self-contained, and
follows the general localization philosophy to study invariants of bundle automorphism actions
proposed in [22]. Once this is done, the remaining problem is to understand the way these strata
can be subsequently glued into a smooth complex manifold modeling the moduli space.
An important property of our construction is that it admits a generalization to arbitrary
rank and number of marked points. This way, this is the first installment (or “toy model”) of
a project focused on the study of complex-analytic structures of moduli spaces of parabolic Higgs
bundles (which could be understood as an instance of the so-called Gelfand principle), where
we exhaustively explore the simplest possible nontrivial example, and we rigorously exhibit the
independence of the biholomorphism type of these moduli spaces under variation of parabolic
weights and wall-crossing. Since the latter is exclusive of the case of four marked points, this is
yet another reason to present it separately.
1Incidentally, in [28] Rayan introduced the notion of “quiver at the bottom of the nilpotent cone”, in the
context of twisted Higgs bundles on CP1. The idea that for parabolic Higgs bundles, the subspace parametrized
by these objects may not correspond to a moduli space of parabolic bundles, seems to be standard folklore [6],
and resurfaces in our work in an explicit way.
2We remark that bundle automorphism groups are in general non-reductive. It should be possible to reinterpret
our constructions as explicit examples of variations of non-reductive GIT quotients [3, 4] (cf. [12]).
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 3
1.1 Structure of the paper
Even though the techniques employed in this work are all elementary, in the end the total
number of steps involved in the construction of geometric models is substantial. This requires
an adequate signposting strategy to guide the reader through it, which we now provide. The
starting point are the model spaces QPH(E) of quasi-parabolic Higgs bundles on E, together
with a pair of projections
QPH(E)
par
ww
det
((
QP(E) :=
4∏
i=1
P(E|zi) H0
(
CP1,K2
CP1
( 4∑
i=1
zi
))
par is the P(Aut(E))-equivariant forgetful map to quasi-parabolic structures, while det is the
P(Aut(E))-invariant determinant map on parabolic Higgs fields.
The construction and classification of all different Harder–Narasimhan strata as spaces of
semi-stable P(Aut(E))-orbits in the complex manifolds QPH(E) follows from a systematic break
down of the semi-stability condition for parabolic Higgs bundles into structural units, recollected
in Sections 3–6. This process consists of four parts which are built progressively:
Section 3: QP(E) is stratified in terms of interpolation loci for line sub-bundles of E. The
resulting stratification is then related to the stratification by pointwise P(Aut(E))-stabilizer
subgroups.
Section 4: det stratifies QPH(E) into a (possibly empty) bulk QPHC∗(E) and its nilpotent
locus QPH0(E). Potentially unstable points can only occur in QPH0(E) (Lemma 4.2). This
leads to a stratification of QPH0(E) in terms of the interpolation properties of potentially
destabilizing line sub-bundles, which arise as kernels of parabolic Higgs fields on E when the
latter are nonzero. The stratifications of QPH(E) and QP(E) are then related by the projection
par (Proposition 4.6).
Section 5: A combinatorial classification of semi-stability walls H and open chambers C in the
even and odd weight polytopes W• is explained. Open chambers are defined as the connected
components in the complement of the union of all semi-stability walls in W•, and parametrize the
different stable loci QPHs
C(E) in QPH(E). Open chambers are split into interior and exterior
(denoted as CI and CI respectively; this notation is carefully explained), according to whether
stable parabolic bundles exist or not.
Section 6: In order to determine stable loci QPHs
C(E), the stratifications of QPH0(E) and
the combinatorics of W• are combined to classify conditionally stable loci (Definition 6.3) as
well as their (in general non-Hausdorff) orbit spaces. The latter are organized in terms of the
combinatorics of some classical projective plane configurations, and then dissected into basic
building blocks (Definition 6.6). Finally, the structure of basic building blocks is classified (they
are biholomorphic to CP1 or C), and a nilpotent cone assembly kit (Definition 6.8) is conformed
for every choice of open chamber C ⊂ W•.
Once all Harder–Narasimhan strata are constructed, their glueing is brought in, and the
interpretation of semi-stability via the geometry of P(Aut(E))-actions leads to an explicit classi-
fication of wall-crossing as nilpotent cone transformations, in terms of the combinatorics of W•,
as well as an additional characterization of parabolic S-equivalence along semi-stability walls.
We list our main results, whose proof is presented in Section 7 together with a series of corollaries.
Theorem 1.1 (orbit space models for Harder–Narasimhan strata). For any open chamber
C ⊂ W• and vector bundle E, if QPHs
C(E) ̸= ∅, then P(Aut(E)) acts freely and properly on it.
4 C. Meneses
There is an isomorphism
MC(E) ∼= QPHs
C(E)/P(Aut(E)),
where MC(E) is the Harder–Narasimhan stratum associated to E. MC(E) is smooth if it is
2-dimensional, and has at most two nodal singularities otherwise.
Theorem 1.2 (glueing of Harder–Narasimhan strata).
(i) For d even, the two orbit spaces
QPHC∗(E)/P(Aut(E)), m1 −m2 = 0, 2,
glue into a 2 : 1-branched covering over CP1 × C∗, ramified over {z1, 0, 1,∞}× C∗.
(ii) For every choice of open chamber C ⊂ W•, the components in the nilpotent cone assembly
kit are glued into a D4-configuration (Figure 1). Moreover, there is an isomorphism
MC ∼= Mtoy,
where MC is the smooth complex surface resulting from the glueing of geometric models of
Harder–Narasimhan strata of corresponding degree.
z1 0 1 ∞ CP1
Figure 1. The D4-configuration.
Theorem 1.3 (wall-crossing classification; see Sections 5 and 6 for notations). For each bound-
ary wall HI,|I|−3 between the exterior and interior chambers CI and CI(I), there is an isomorphism
MCI
∼= MCI(I)
exchanging the nilpotent cone’s central spheres SI and SI(I), and equal to the
identity otherwise. For any pair of neighboring interior chambers CI(I) and CI(I′), there is an
isomorphism MCI(I)
∼= MCI(I′) exchanging the compactified I- and I ′-basic building blocks, and
equal to the identity otherwise.
After this, the Hitchin elliptic fibrations, their degeneration into their nilpotent cones, as well
as a C∗-action and a collection of Hitchin sections, are explicitly identified as by-products of
the method of construction. In Appendix A we also introduce the alternative residue models
M ev
C (E) for Harder–Narasimhan strata.3 The generalization of this work to an arbitrary number
of punctures requires a more careful analysis of automorphism group actions, including a suitable
axiomatization of the intrinsic structures on display here, and will be treated elsewhere [23].
3Cf. Komyo–Saito [18], who provide explicit descriptions of Zariski open sets on moduli spaces of logarithmic
connections and parabolic Higgs bundles in odd degree, by apparent singularities and their dual parameters as
coordinates.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 5
1.2 Relation to other work
The literature on moduli spaces emerging from parabolic structures on CP1 is vast and comprises
different perspectives, strategies, and objectives. We compiled works with ideas overlapping
substantially with ours. Further details can be found in those works and references therein.
Related models were used by Loray–Saito–Simpson [20] in their proof of the non-abelian
Hodge foliation conjecture for moduli of rank 2 logarithmic connections, in odd degree and n = 4
marked points, in terms of transversal foliations arising as Lagrangian fibrations. There, a bundle
splitting type is necessarily generic, and their analysis of moduli dependence on parabolic weights
leads to an analogous characterization of exterior chambers. These fibrations are studied for
arbitrary n in [19] (cf. [18]), and in [15] the study of moduli of holomorphic connections in
genus 2 is reduced to the genus 0 case and n = 6. The overall framework of these works is
that of birational geometry. While similar in spirit, our proposal is considerably different in its
methods and results.
Godinho–Mandini [9, 10] defined an isomorphism between hyperpolygon spaces and the
Zariski open subset of rank 2 parabolic Higgs bundles of degree 0 that are holomorphically
trivial in terms of the residue data of the latter (a detailed description of the correspondence in
the case n = 4, in terms of the Nakajima quiver variety for the affine Dynkin diagram D̃4, is
discussed by Rayan–Schaposnik in [29, Section 4.3]). Similarly, Blaavand [6] provides a geomet-
ric construction of the nilpotent cone in degree 1 and n = 4 under a choice of parabolic weights
constrained to lie inside a specific open chamber (see Remark 7.7). Heller–Heller [14] applied
the abelianization of logarithmic connections in degree 0 when n = 4 to compute symplectic
volumes for moduli spaces of stable parabolic bundles.
2 Conventions and index of notation
We will fix a point z1 ∈ CP1\{z2, z3, z4}, where z2 = [0 : 1] = 0, z3 = [1 : 1] = 1, z4 = [1 : 0] = ∞.
When necessary, we will consider the choice of cross-ratio such that (z1, z2; z3, z4) = z1. Similarly,
we will denote D =
∑4
i=1 zi. A parabolic structure supported on D on a rank 2 holomorphic
vector bundle E → CP1 of degree d consists of flags
Fi ⊂ E|zi , i = 1, 2, 3, 4
weighted by real numbers 0 ≤ βi < 1 such that Fi = E|zi if and only if βi = 0. A quasi-parabolic
structure omits the choice of weights. Hereafter E∗ will denote a rank 2 parabolic bundle, i.e.,
a choice of parabolic structure over D on E, and E· the underlying quasi-parabolic bundle.
Given E∗, to any subset I ⊂ {1, 2, 3, 4} we associate the weighted sum
βI =
∑
i∈I
βi −
∑
j∈Ic
βj .
For any E·, a line sub-bundle L ⊂ E determines a subset IL,E· ⊂ {1, 2, 3, 4} defined as
IL,E· =
{
i ∈ {1, 2, 3, 4} : L|zi ⊂ Fi
}
.
For any z ∈ CP1, n(E|z) will denote the cone of nilpotent endomorphisms of E|z, i.e., the singular
quadric in sl(E|z) defined as the zero locus of its determinant map, and for any F ∈ P(E|z),
n(F ) is the line of nilpotent endomorphisms ϕ such that ker(ϕ) ⊃ F . Let End(E) := E∨ ⊗ E.
An element Φ ∈ H0(CP1,End(E) ⊗ KCP1(D)) is a (strongly) parabolic Higgs field of E∗ if for
each i = 1, 2, 3, 4, the residue Reszi Φ satisfies
Reszi Φ ∈ n(Fi).
6 C. Meneses
A parabolic Higgs bundle is a pair of the form (E∗,Φ). The pair (E∗,Φ) is called stable
(resp. semi-stable) if for every Φ-invariant line sub-bundle L ⊂ E we have that
βIL,E· < d− 2 deg(L) (resp. ≤). (2.1)
The (semi-)stability of a parabolic bundle E∗ is defined by letting Φ = 0. (E∗,Φ) is called
strictly semi-stable if it semi-stable but not stable.
Given a vector bundle E, a line bundle L, and their tensor product E′ = E ⊗ L, the bundle
isomorphism End(E′) ∼= End(E) induces a bijection between the sets of quasi-parabolic Higgs
bundles on E and E′. Since L′ is a line sub-bundle of E′ if and only if L−1 ⊗ L′ is a line sub-
bundle of E, it follows that the stability of a parabolic Higgs bundle is preserved under tensor
product by a line sub-bundle.
Remark 2.1. In the case of rank 2 holomorphic vector bundles, the standard definition of
parabolic stability [21, 30] involves parabolic weights 0 ≤ αi1 ≤ αi2 < 1, i = 1, 2, 3, 4. In that
case, the parabolic degree and slope of E∗ are defined as
par deg(E∗) := d+
4∑
i=1
(αi1 + αi2), parµ(E∗) = par deg(E∗)/2.
In turn, a line sub-bundle L ⊂ E acquires a set of parabolic weights α′
i defined as αi2 whenever
i ∈ I and αi1 otherwise. The induced (semi-)stability condition then reads
parµ(L∗) := deg(L) +
4∑
i=1
α′
i < parµ(E∗) (resp. ≤).
It can be verified that the last inequality only depends on the effective parameters
βi := αi2 − αi1 ∈ [0, 1), i = 1, 2, 3, 4,
and reduces to (2.1). A set of parabolic weights will be called admissible whenever the equation
par deg(E∗) = 0 is satisfied. The admissibility of parabolic weights, together with the stability
condition for parabolic Higgs bundles, are necessary for the non-abelian Hodge correspondence
to hold (see [30]), even though the moduli problem is meaningful without the former; e.g.,
in [26] Mukai considers parabolic weights corresponding to the choices αi1 = 0 and αi2 = βi, for
n marked points on CP1. On the other hand, Bauer [1] considers the SU(2)-constraints
par deg(E∗) = 0, 0 < αi1 = αi < 1/2, αi2 = 1− αi, i = 1, . . . , n,
fixing the value of d. When d ≡ n (mod 2), up to the tensor product of E by a suitable line
bundle, every set {βi} can be lifted to a unique set of parabolic weights satisfying the SU(2)-
constraints. When d ̸≡ n (mod 2), a lift to a set of admissible parabolic weights is only possible
under further numerical constraints, but a weaker lift satisfying the SU(2)-constraints can still
be achieved in terms of an extra marked point z0 with parabolic weights α01 = α02 = 1/2
(so that β0 = 0). This operation leaves the moduli problem unchanged, since it forces any
flag over E|z0 to be trivial and the corresponding residue Resz0 Φ to vanish.4 The alternative
convention −1/2 < −α′
i < α′
i < 1/2 for the SU(2)-constraints is related to the former in terms
of the transformations β′
i = 1− βi, i = 1, 2, 3, 4.
4An operation on parabolic bundles discussed in the literature is the parabolic tensor product. We don’t
consider it here as it does not preserve the underlying associated bundle End(E).
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 7
3 Geometry of P(Aut(E))-actions and stratifications in QP(E)
Let E → CP1 be a holomorphic vector bundle of rank 2 and degree d. A basic holomorphic invari-
ant associated to E is its Harder–Narasimhan filtration. In genus 0, the Birkhoff–Grothendieck
theorem postulates the existence of an isomorphism of vector bundles
E ∼= O(m1)⊕O(m2), m1 ≥ m2,
in such a way that d = m1 +m2. Every rank 2 bundle E is either semi-stable (i.e., it splitting
coefficients satisfy m1 = m2) or otherwise admits a unique filtration of the form
E1 ⊂ E2 = E, deg(E1) > deg(E2)/2.
In other words, every isomorphism E ∼= O(m1) ⊕ O(m2) is a refinement of the Harder–Nara-
simhan filtration of E. When E is not semi-stable, we have that E1
∼= O(m1), and a choice
of Birkhoff–Grothendieck splitting for E is equivalent to a choice of complementary sub-bundle
O(m2) ⊂ E. The integers m1 ≤ m2 are holomorphic invariants for E. E is called evenly-spit if
m1−m2 ≤ 1, or equivalently if H1(CP1,End(E)) = 0, where End(E) := E⊗E∨. Consequently,
on any holomorphic family F → B × CP1 of rank 2 and degree d holomorphic vector bundles
E → CP1, the existence of evenly-split bundles is an open condition.
Pointwise multiplication endows the space H0(CP1,End(E)) with an associative algebra
structure. The group of global bundle automorphisms of E is the Lie group of invertible elements
Aut(E) ⊂ H0
(
CP1,End(E)
)
.
Its projectivization is defined as P(Aut(E)) := Aut(E)/Z(Aut(E)), where its center Z(Aut(E))
consists of all nonzero multiples of the identity. Aut(E) ∼= GL(2,C) in the special case when E is
semi-stable. Otherwise, when m1 > m2, Aut(E) is (m1−m2+3)-dimensional and preserves the
Harder–Narasimhan filtration of E. Its unipotent radical is the normal subgroup of unipotent
automorphisms of E, and will be denoted by R(Aut(E)). R(Aut(E)) is maximally abelian, has
codimension 2 in Aut(E), and acts freely and transitively on the space Lm2(E). In particular,
the subgroup R(Aut(E)) ∩ Z(Aut(E)) is trivial.
Consider a point z ∈ CP1. For any F ∈ P(E|z), let P(F ) ⊂ GL(E|z) be the parabolic
subgroup stabilizing the flag F ⊂ E|z, and R(F ) its unipotent radical. The evaluation maps
Aut(E) 7→ Aut(E)|z ⊂ gl(E|z)
determine a subgroup of GL(E|z). This subgroup coincides with GL(E|z) when m1 = m2, in
which case the evaluation map is an isomorphism. Otherwise, whenm1 > m2, Aut(E)|z coincides
with the parabolic subgroup P(E1|z) that is induced by the Harder–Narasimhan filtration of E.
In this case we define the affine line
Vz(E) := P(E|z)\{E1|z}.
Since there is an isomorphism Aut(E)|z ∼= P(E1|z), and P(R(E1|z)) ∼= C acts freely and transi-
tively on Vz(E), the cell decomposition
P(E|z) = {E1|z} ⊔ Vz(E)
is the partition into two orbits determined by the action of P(Aut(E)|z) on P(E|z).
8 C. Meneses
3.1 Line sub-bundles and interpolation of quasi-parabolic structures
For any E, we will denote the space of all line sub-bundles L ⊂ E of degree j by Lj(E). Given
an isomorphism E ∼= O(m1)⊕O(m2), Lj(E) is modeled by the Zariski open set
Lj(O(m1)⊕O(m2)) ⊂ P
(
H0
(
CP1,O(m1 − j)
)
⊕H0
(
CP1,O(m2 − j)
))
generated by pairs of holomorphic sections (s1, s2) that are not simultaneously zero and have
disjoint zero sets. In the particular case when m1 = m2 = m, we have the isomorphisms
Lm(E) ∼= CP1, Lm−1(E) ∼= CP3\Segre
(
CP1 × CP1
)
.
On the other hand, when m1 > m2 there are no line sub-bundles L ⊂ E of degree j for
m1 > j > m2, and
Lm2(E) ∼= Am1−m2+1 ⊂ CPm1−m2+1.
Given I ⊂ {1, 2, 3, 4} and j ≤ m2, we define the O(j)-interpolation locus on QP(E), relative
to I, as
BI,j(E) :=
{
(F1, F2, F3, F4) ∈ QP(E) : ∃L ∈ Lj(E), Fk = L|zk ∀k ∈ I
}
, (3.1)
which satisfies
BI′,j(E) ⊂ BI,j(E) whenever I ⊂ I ′. (3.2)
For any vector bundle E, there is a natural action of P(Aut(E)) on its spaces of line sub-bund-
les Lj(E). In the particular case when m1 = m2 = m, this reduces to the standard action on
the projective line Lm(E) by Möbius transformations, and moreover, Lm−1(E) is a principal
homogeneous space for P(Aut(E)). Otherwise, in the case when m1 > m2, Lm2(E) is a principal
homogeneous space for R(Aut(E)). Since the action of Aut(E) on the spaces Lj(E) commutes
with the operations of evaluation, every BI,j(E) is invariant under the induced P(Aut(E))-
action. In the particular case when I = {1, 2, 3, 4}, it follows from the general properties of
polynomial interpolation under an arbitrary choice of bundle isomorphism E ∼= O(m1)⊕O(m2)
that if dimLj(E) < 4, then the induced line bundle evaluation map
evLj(E) : Lj(E) → QP(E), L ⊂ E 7→
(
L|z1 , L|z2 , L|z3 , L|z4
)
is a holomorphic embedding, and
evLj(E)(Lj(E)) = B{1,2,3,4},j(E).
In addition, when m1 > m2, O(m1)-interpolation relative to a subset I endows QP(E) with the
following P(Aut(E))-invariant stratification
QP(E) =
⊔
I⊂{1,2,3,4}
BI,m1(E), BI,m1(E) :=
{∏
i∈I
{E1|zi}
}
×
{ ∏
j∈Ic
Vzi(E)
}
.
3.2 Combinatorial stratification of orbit spaces in QP(E)
We will now describe the action of P(Aut(E)) on the spaces QP(E). Due to the structural
differences of the groups P(Aut(E)) depending on whether E is semi-simple or not, the treatment
is split into the two cases m1 = m2 and m1 > m2. In the first case, we define the configuration
space locus to be the following Zariski open subset of QP(E)
C(E) :=
⋂
|I|=2
Bc
I,m(E) ∼= Conf4
(
CP1
)
.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 9
Proposition 3.1. If m1 −m2 ≤ 2, then P(Aut(E)) acts freely on the Zariski open subset
U(E) := QP(E)
∖ ⋃
I⊔I′={1,2,3,4}
{
BI,m1(E) ∩ BI′,m2(E)
}
,
which is a P(Aut(E))-principal homogeneous space if m1 −m2 = 2, and empty if m1 −m2 > 2.
There is a P(Aut(E))-invariant “cross-ratio” map cr : U(E) → CP1 identifying the U(E)-orbit
space with a projective line with three (resp. four) double points if m1 = m2 (resp. m1−m2 = 1),
and such that cr(U′(E)) = CP1\{z1, 0, 1,∞} for
U′(E) :=
C(E)
∖
B{1,2,3,4},m−1(E), m1 = m2 = m,
B∅,m1(E)
∖ ⊔
|I′|≥3
BI′,m2(E), m1 −m2 = 1.
Proof. Case m1 = m2 = m. All nonempty loci BI,m(E) provide an invariant characterization
of the trivialization-dependent identities
Fi = Fj , ∀ i, j ∈ I,
since for any |I| ≥ 2
BI,m(E) =
⋂
I′⊂I, |I′|=2
BI′,m(E).
Therefore, C(E) is the locus in U(E), where P(Aut(E)) acts properly, in such a way that
C(E)/P(Aut(E)) ∼= CP1\{0, 1,∞}. Moreover, the six connected components in
U(E)\C(E) =
⊔
|I|=2
{
BI,m(E) ∩U(E)
}
are principal homogeneous spaces for P(Aut(E)), and it follows that there is a unique P(Aut(E))-
invariant map cr : U(E) → CP1 normalized in the following manner
cr
(
B{1,2},m(E) ∩U(E)
)
= cr
(
B{3,4},m(E) ∩U(E)
)
= 0,
cr
(
B{1,3},m(E) ∩U(E)
)
= cr
(
B{2,4},m(E) ∩U(E)
)
= 1,
cr
(
B{1,4},m(E) ∩U(E)
)
= cr
(
B{2,3},m(E) ∩U(E)
)
= ∞.
In turn, since B{1,2,3,4},m−1(E) is also a principal homogeneous space for P(Aut(E)), we have
that cr(U′(E)) = CP1\{z1, 0, 1,∞}, since under the previous normalization
cr
(
B{1,2,3,4},m−1(E)
)
= (z1, z2; z3, z4) = z1.
Case m1 > m2. In general, we have that for any partition I ⊔ I ′ = {1, 2, 3, 4} and |I| >
2−m1 +m2,
BI,m1(E) ∩ BI′,m2(E) = BI,m1(E).
Moreover, when |I| ≤ 2 − m1 + m2, the P(Aut(E))-action on BI,m1(E)\BIc,m2(E) is free and
proper, as it can be factored through the following intermediate affine quotients{
BI,m1(E)\BIc,m2(E)
}
/R(Aut(E)) ∼= C3−m1+m2−|I|\{0},
10 C. Meneses
on which P(Aut(E))/R(Aut(E)) ∼= C∗ acts in the standard way. Since we can re-express
U(E) =
⊔
|I|≤2−m1+m2
{
BI,m1(E)\BIc,m2(E)
}
,
it follows that U(E) is empty if m1 −m2 > 2, a principal homogeneous space for P(Aut(E)) if
m1 −m2 = 2, and moreover, that there exists a P(Aut(E))-invariant map U(E) → CP1 when
m1 −m2 = 1, which determines an isomorphism
U′(E)/P(Aut(E)) ∼= CP1\{z1, 0, 1,∞},
and the complement U(E)\U′(E) consists of eight connected components, namely
B∅,m1(E) ∩ B{j,k,l},m2
(E) and B{i},m1
(E)\B{j,k,l},m2
(E),
each of which is a principal homogeneous space for P(Aut(E)). These eight orbits determine
four different pairs of compactification points for U′(E)/P(Aut(E)). ■
Proposition 3.2.
(i) The elements of the sets
X(E) :=
⋃
|I|=2−m1+m2
{
BI,m1(E) ∩ BIc,m2(E)
}∖
B{1,2,3,4},m1
(E),
Y(E) :=
⋃
|I|=3−m1+m2,
|I|=1−m1+m2
{
BI,m1(E) ∩ BIc,m2(E)
}∖
B{1,2,3,4},m1
(E)
have pointwise stabilizer subgroups P(Aut(E))(F1,F2,F3,F4)
∼= C∗.
(ii) If m1 −m2 ≤ 2 then X(E) and Y(E) are degeneration loci for the orbits in U(E)\U′(E),
where U′(E) = ∅ if m1 −m2 = 2, i.e.,
X(E) ∪Y(E) ⊂ U(E)\U′(E).
Proof. The first claim is verified in analogy to the proof of Proposition 3.1. Notice that when
m1 > m2, {BI,m1(E) ∩ BIc,m2(E)} ∩ B{1,2,3,4},m1
(E) = ∅, X(E) = ∅ if m1 − m2 > 2, and
the stabilizer of any point in X(E) or Y(E) is biholomorphic to P(Aut(E))/R(Aut(E)) ∼= C∗.
The second claim is trivial when m1 −m2 = 2. When m1 = m2 = m, the result follows since
for any partition I ⊔ I ′ = {1, 2, 3, 4}, |I| = 2{
BI,m(E) ∩ BI′,m(E)
}
⊂ BI,m(E) ∩U(E) ∩ BI′,m(E) ∩U(E),
while when |I| = 3
BI,m(E) ⊂
⋂
I′⊂I, |I′|=2
BI′,m(E) ∩U(E).
The case m1 −m2 = 1, |I| = 1, follows since for each partition {i} ⊔ {j, k, l} = {1, 2, 3, 4}
B{i},m1
(E) ∩ B{j,k,l},m2
(E) ⊂ B∅,m1(E) ∩ B{j,k,l},m2
(E) ∩ B{i},m1
(E)\B{j,k,l},m2
(E),
and is trivial when |I| = 2 or |I| = 0 from the definition of U′(E). ■
The stratification of QP(E) can be completed by also grouping quasi-parabolic structures with
stabilizer subgroups of higher dimension. We don’t describe these additional strata explicitly
since they won’t appear in subsequent constructions.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 11
4 Structural results and stratification of QPH(E)
The construction of geometric models for Harder–Narasimhan strata will be based on the fol-
lowing series of results on the structure of spaces of quasi-parabolic Higgs bundles. For any fixed
holomorphic rank 2 bundle E → CP1, let QPH(E) denote the space of all quasi-parabolic Higgs
bundles (E·,Φ) with quasi-parabolic structure supported on D. We can identify QPH(E) with
an algebraic subvariety of the product
QP(E)×H0
(
CP1,End(E)⊗KCP1(D)
)
characterized as the locus of all points (F1, F2, F3, F4,Φ) satisfying the incidence constraints
Reszi Φ ∈ n(E|zi), Fi ⊂ ker(Reszi Φ), i = 1, 2, 3, 4.
Equivalently, if we denote by Int(E) the intersection of the four quadrics in the space{
Φ ∈ H0
(
CP1,End(E)⊗KCP1(D)
)
: tr(Φ)|zi = 0, i = 1, 2, 3, 4
}
defined by the equations
det(Φ)|zi = 0, i = 1, 2, 3, 4,
then QPH(E) is modeled by the blow-up of Int(E) along each of the codimension 2 loci
Reszi Φ = 0, i = 1, 2, 3, 4. (4.1)
There is a natural action of Aut(E) on QPH(E), whose point stabilizers contain Z(Aut(E)),
descending to an action of the quotient group P(Aut(E)). The proof of the following lemma is
obvious.
Lemma 4.1. There is a bijective correspondence between isomorphism classes of quasi-parabolic
Higgs bundles {(E·,Φ)} and P(Aut(E))-orbits in QPH(E).
Since in the present case dimH0(CP1,K2
CP1(D)) = 1, we will define the bulk as
QPHC∗(E) := det−1
(
H0
(
CP1,K2
CP1(D)
)
\{0}
)
,
and the nilpotent locus as
QPH0(E) := det−1(0).
For the embedding ι : QP(E) ↪→ QPH0(E) defined as ι := {0} × id, we will denote
Q(E) := ι(QP(E)),
which is also characterized as the fixed-point locus of the holomorphic involution defined as
τ(F1, F2, F3, F4,Φ) := (F1, F2, F3, F4,−Φ). (4.2)
More generally, QPH(E) is equipped with a C∗-action, defined for any c ∈ C∗ as
c · (F1, F2, F3, F4,Φ) = (F1, F2, F3, F4, cΦ), c ∈ C∗. (4.3)
Since both τ and this C∗-action are P(Aut(E))-equivariant, they descend to an involution and
a C∗-action on the space of P(Aut(E))-orbits of QPH(E). While they are only trivial on Q(E),
we will verify that their descents possess additional fixed loci. The next results characterize the
structure of QPH(E) in terms of the stratification
QPH(E) = QPH0(E) ⊔QPHC∗(E).
12 C. Meneses
Lemma 4.2. A nonzero parabolic Higgs field Φ preserves a line sub-bundle L(Φ) ⊂ E if and
only if it is nilpotent, in which case L(Φ) is unique and determined by ker(Φ).
Proof. Given Φ ̸= 0, the equation Φ · L = sL with s ∈ H0(CP1,KCP1(D)) requires that
s2 = −detΦ ∈ H0(CP1,K2
CP1(D)), from which s ≡ 0, i.e., Φ is nilpotent. Conversely, if Φ is
nilpotent, we can reconstruct L(Φ) in a unique way. Since the zero set of Φ as a holomorphic
section of End(E) ⊗ KCP1(D) is finite, we can associate an effective divisor (Φ) to Φ. The
complement of such zero set is the Zariski open set U of points z ∈ CP1, where ker(Φ)(z) ⊂ E|z
(or ker(Reszi Φ) if z = zi, i = 1, 2, 3, 4) is 1-dimensional. Hence ker(Φ)|U is a line sub-bundle
of E|U . Let σ be a holomorphic section of the line bundle [(Φ)] such that (σ) = (Φ). Then by
construction σ−1Φ is a holomorphic section of EndE ⊗KCP1(D − (Φ)) such that ker(σ−1Φ)(z)
is 1-dimensional ∀ z ∈ CP1. Therefore L(Φ) := ker(σ−1Φ) is a line sub-bundle of E such that
L|U = ker(Φ)|U and invariant under Φ over all CP1. ■
Two important conclusions follow from Lemma 4.2: when it exists, an element of QPHC∗(E)
can never be unstable, and moreover, the elements in QPH0(E)\Q(E) can be classified according
to their unique invariant line sub-bundles, inducing a finer stratification of QPH0(E) that we
will describe in detail.
Lemma 4.3. QPHC∗(E) is nonempty if and only if m1 − m2 ≤ 2. Given a line sub-bundle
L ⊂ E, there is a nonzero nilpotent parabolic Higgs field Φ such that L(Φ) = L if and only if
2(deg(L) + 1) ≥ deg(E). (4.4)
Proof. Any given choice of bundle isomorphism E ∼= O(m1)⊕O(m2) induces an isomorphism
End(E)⊗KCP1(D) ∼=
(
O(2) O(m1 −m2 + 2)
O(m2 −m1 + 2) O(2)
)
,
from which it follows that a parabolic Higgs field on E gets characterized as a holomorphic
section
Φ =
(
u2 −vm1−m2+2
wm2−m1+2 −u2
)
(4.5)
whose pointwise evaluation on D moreover satisfies the set of nilpotency constraints
det(Φ)|zi =
(
−u22 + vm1−m2+2wm2−m1+2
)∣∣
zi
= 0, i = 1, 2, 3, 4.
It follows that the map det is surjective if and only if m1 −m2 ≤ 2, from which the first claim
follows. To prove the second claim, first assume that m1 > m2 and L = E1. A nonzero nilpotent
parabolic Higgs field Φ preserving E1 is characterized in terms of the identity wm2−m1+2 ≡ 0
(which is independent of the choice of isomorphism E ∼= O(m1)⊕O(m2)) since the nilpotency
constraints imply that u2 ≡ 0 necessarily. Since Φ is identified with a nonzero holomorphic
section of O(m1 −m2 + 2), it follows that there is always some Φ such that L(Φ) = E1.
Otherwise, assume that L ̸= E1 if m1 > m2, or equivalently, that deg(L) ≤ m2. Given
a nonzero nilpotent parabolic Higgs field Φ such that L(Φ) = L, we can choose σk ̸= 0 ∈
H0(CP1, [(Φ)]) for some 0 ≤ k ≤ 2−m1 +m2, such that (σk) = (Φ) and express
Φ = σk
(
u2−k −v2−m2+m1−k
w2−m1+m2−k −u2−k
)
, (4.6)
such that w2−m1+m2−k ̸= 0 if m1 > m2, and the following equation holds in H0(CP1,O(4− 2k))
u22−k = v2−m2+m1−kw2−m1+m2−k. (4.7)
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 13
The holomorphic sections u2−k and w2−m1+m2−k could have a (simple) common zero only in the
exceptional case when m1 = m2 = m and k = 0, when a parabolic Higgs field takes the form
Φ =
(
u′1v
′
1 −u′21
v′21 −u′1v
′
1
)
. (4.8)
Since L(Φ) is generated by (u′1, v
′
1), it follows that L(Φ)
∼= O(m−1). Otherwise, we can assume
that u2−k and w2−m1+m2−k don’t have a common zero. Then L(Φ) is generated by
(u2−k, w2−m1+m2−k) ∈ H0
(
CP1,O(2− k)⊕O(2−m1 +m2 − k)
)
and deg(L(Φ)) = m1 − 2 + k ≥ m1 − 2. If m1 − m2 = 2, then k = 0 and deg(L) = m2.
If m1 −m2 = 0 (resp. 1) and k = 0, 1 (resp. 0), then w2−m1+m2−k would have at least one zero,
and it would follow from (4.7) that u2−k and w2−m1+m2−k have a common zero, a contradiction.
Therefore k = 2 (resp. 1), or equivalently, deg(L(Φ)) = m2. In all of the four possible cases, we
conclude that
k = 2(deg(L(Φ)) + 1)− deg(E).
Therefore, k ≥ 0 is equivalent to the lower bound (4.4), and the claim follows.
Conversely, consider first any line sub-bundle L ⊂ E such that deg(L) = m2, generated
by a holomorphic section (um1−m2 , w0) with w0 ̸= 0, and assume that (4.4) is satisfied, i.e.,
m1 −m2 ≤ 2. It readily follows that there exists a holomorphic section
v2m1−2m2 ∈ H0
(
CP1,O(2m1 − 2m2)
)
, (v2m1−2m2) = 2(um1−m2),
solving (4.7), for k = 2−m1+m2. Any Φ constructed according to (4.6), for an arbitrary choice
of σ2−m1+m2 ∈ H0(CP1,O(2 −m1 +m2))\{0}, would be such that L(Φ) = L. The remaining
case m1 = m2 = m, deg(L) = m− 1 follows analogously. ■
The next result, on the stratifications of QPH0(E) for any bundle splitting type, is immediate
from Lemma 4.3. Notice that the C∗-action on QPH(E) preserves its stratification, and that of
QPH0(E) in particular.
Corollary 4.4. Let R(E) (resp. Sj(E)) denote the locus of quasi-parabolic Higgs bundles such
that Φ ̸= 0, for which L(Φ) = E1 (resp. L(Φ) ∈ Lj(E)). QPH0(E) is stratified as
QPH0(E) =
Q(E) ⊔ Sm(E) ⊔ Sm−1(E) if E ∼= O(m)⊕O(m),
Q(E) ⊔ R(E) ⊔ Sm(E) if E ∼= O(m+ 1)⊕O(m),
Q(E) ⊔ R(E) ⊔ Sm−1(E) if E ∼= O(m+ 1)⊕O(m− 1),
Q(E) ⊔ R(E) if m1 −m2 ≥ 3.
Let Pj(E) → Lj(E) be the tautological principal C∗-bundle associated to the canonical
projective embeddings of Lj(E). For any I ⊂ {1, 2, 3, 4}, BlI(CP1) will denote the rational
nodal curve resulting from blowing-up CP1 ⊂ CP2 at zi ∀i ∈ I. In particular, Bl{1,2,3,4}(CP1) is
the D4-configuration. Consider the P(Aut(E))-equivariant maps
LE,j : Sj(E) → Lj(E), (E·,Φ) 7→ L(Φ),
and in the case k = 2(j + 1)− deg(E) > 0, also the P(Aut(E))-invariant maps
DivE,j : Sj(E) → Sk(CP1), (E·,Φ) 7→ (Φ).
14 C. Meneses
Proposition 4.5. The restriction of the C∗-action to each Sj(E) turns it into a principal C∗-
bundle, whose base is modeled by the 3-dimensional subvariety of the product
Lj(E)× S2(j+1)−deg(E)
(
CP1
)
×QP(E),
defined by the incidence constraints Fi = L(Φ)|zi whenever zi ̸∈ (Φ), and projection given by the
map LE,j ×DivE,j ×par. In particular, there is an isomorphism
Sj(E) ∼=
{
Pj(E), 2(j + 1)− deg(E) = 0,
Bl{1,2,3,4}(CP1)× Pj(E), 2(j + 1)− deg(E) = 1.
In general,
par(Sj(E)) =
⋃
|I|=2(j+1)−deg(E)
BI,j(E).
Proof. The map LE,j ×DivE,j ×par is invariant under the C∗-action on QPH(E) by definition.
The first claim follows from the definition of Sj(E) as a stratum of QPH(E), and the expression of
the parabolic Higgs field Φ of any element (E·,Φ) ∈ Sj(E) in one of the two canonical forms (4.6)
and (4.8) under a choice of isomorphism E ∼= O(m1) ⊕ O(m2) as in proof of Lemma 4.3, with
k = 2(j + 1)− deg(E), and in such a way that (Φ) = (σk). ■
Proposition 4.6.
(i) If E is evenly-split, then
par
(
QPHC∗(E)
)
= U′(E) ⊔X(E),
and
par
(
QPH0(E)\Q(E)
)
= X(E) ⊔Y(E).
(ii) If m1 −m2 = 2, then
par
(
QPHC∗(E)
)
= par(Sm−1(E)) = X(E).
(iii) If m1 −m2 ≥ 2, then par|R(E) is surjective.
Proof. (i) Assume that E is evenly-split. We will treat both cases independently using the
parametrization (4.5) of parabolic Higgs fields on a quasi-parabolic bundle E· depending on
a choice of isomorphism E ∼= O(m1) ⊕ O(m2). When m1 = m2 = m, for a partition
{1, 2, 3, 4} = {i} ⊔ {j, k, l}, consider the triple of Lagrange interpolating sections {sj , sk, sl}
spanning H0(CP1,O(2)) relative to the triple {zj , zk, zl}. Since the “residue evaluation” map
n(Fj)⊕ n(Fk)⊕ n(Fl) → sl(E|zi), (ϕj , ϕk, ϕl) 7→ (ϕjsj + ϕksk + ϕlsl)|zi
is an isomorphism, there is a line in n(Fj) ⊕ n(Fk) ⊕ n(Fl) mapping to n(Fi). This implies
that any quasi-parabolic Higgs field Φ is expressed as ϕjsj + ϕksk + ϕlsl, and consequently
par(QPH(E)\Q(E)) = QP(E).
If par(E·,Φ) ∈ BI,m(E) for some |I| ≥ 3 and Φ ̸= 0, then after representing Φ = ϕjsj +
ϕksk + ϕlsl as before for some {j, k, l} ⊂ I, we see that dim(Span{ϕj , ϕk, ϕl}) = 1, and conse-
quently (E·,Φ) ∈ Sm(E). An analogous argument for par(E·,Φ) ∈ B{1,2,3,4},m−1(E) implies that
(E·,Φ) ∈ Sm−1(E). Finally, if par(E·,Φ) ∈ BI,m(E) for some (E·,Φ) ∈ QPHC∗(E) and |I| = 2,
then the same argument implies that par(E·,Φ) ∈ {BI,m(E)∩BIc,m(E)}\B{1,2,3,4},m(E) ⊂ X(E).
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 15
This last possibility occurs when the orbit of (E·,Φ) contains an element with parabolic Higgs
field
Φ =
(
0 −v2
w2 0
)
, (4.9)
with (v2) = zi+zj and (w2) = zk+zl for a partition {i, j}⊔{k, l} = {1, 2, 3, 4}. The P(Aut(E))-
stabilizers of these points are trivial, and their projection exhaust X(E). The surjectivity of
par|QPH(E)\Q(E) and the definition of U′(E) (Proposition 3.1) imply that QPHC∗(E) necessa-
rily projects onto U′(E) ⊔ X(E). The characterization of par(QPH0(E)\Q(E)) follows from
Corollary 4.4 and Proposition 4.5.
When E ∼= O(m+1)⊕O(m), we recall that the divisor (w1) associated to the expression (4.5)
is independent of the choice of Birkhoff–Grothendieck splitting and an invariant of any point
QPH(E)\Q(E). The projection of the Zariski open subset of QPHC∗(E) for which (w1) ∈
CP1\{z1, 0, 1,∞} equals U′(E) (defined in Proposition 3.1), since for any (F1, F2, F3, F4) ∈ U′(E)
and any u2 such that zi ̸∈ (u2) there exists a unique v3 such that ker(Reszi Φ) = Fi, i = 1, 2, 3, 4,
and on the other hand, an element QPH(E)\Q(E) projecting to BI,m(E) for any |I| ≥ 3 necessa-
rily belongs to Sm(E). In turn, the projection of the complementary loci of points in QPHC∗(E)
for which (w1) = zi, i = 1, 2, 3, 4, exhaust the exceptional locus X(E) in Proposition 3.2, via the
P(Aut(E))-orbits of elements (E·,Φ) ∈ QPHC∗(E) of the form
Φ =
(
0 −v3
w1 0
)
, (4.10)
such that (v3) = zj+zk+zl for the partition {i}⊔{j, k, l} = {1, 2, 3, 4}. Moreover, it follows that
the P(Aut(E))-stabilizers of these points are trivial as well. Finally, it also follows from (4.5) and
Proposition 4.5 that any quasi-parabolic structure in the complement of U′(E) can be always
represented as the projection of an element in either R(E) or S0(E), from which the full claim
follows.
(ii) Whenm1−m2 = 2, a point in QPHC∗(E)⊔S0(E), satisfies w0 ̸= 0, and is fully determined
by Φ. Under a choice of isomorphism E ∼= O(m + 1) ⊕ O(m − 1), the P(Aut(E))-orbit of Φ
always contains an element of the form
Φ =
(
0 −v4
w0 0
)
,
from which the claim follows.
(iii) The result follows from (4.5) and the constraints wm2−m1+2 = 0 and u2 = 0 under
a choice of isomorphism E ∼= O(m1)⊕O(m2). ■
5 Combinatorial description of parabolic weight polytopes
We will present a thorough account of the combinatorial structure on [0, 1]4 that encodes wall-
crossing phenomena for the toy model, and refines an explicit embedding of the convex polytope
known as the 4-demicube or demitesseract.5 In order to emphasize the features resulting from
the inclusion of parabolic Higgs fields into the moduli problem, we will construct this structure
from first principles. Consider the functions
βI : [0, 1]4 → R, I ⊂ {1, 2, 3, 4},
5The latter was described by Bauer [1] under the SU(2)-constraints (requiring d to be even in the case of 4
marked points), and later by Biswas [5] for parabolic degree 0, in relation to the moduli problem for parabolic
bundles of rank 2.
16 C. Meneses
which by definition satisfy the relations βI + βIc = 0 and βI′ ≤ βI whenever I ′ ⊂ I, as well as
the bounds
−|Ic| ≤ βI ≤ |I|. (5.1)
Without any loss of generality (see Remark 2.1), we will reformulate the necessary and sufficient
conditions on parabolic weights for the existence of stable parabolic bundles in terms of β-weights
and the inequalities (2.1).
Proposition 5.1 ([1, 5]). There exists a rank 2 semi-stable parabolic bundle E∗ of degree d with
respect to βββ ∈ [0, 1)4 if and only if for any I ⊂ {1, 2, 3, 4} such that d ≡ |I| − 1 (mod 2),
βI(βββ) ≤ |I| − 1. (5.2)
E∗ is necessarily stable if all inequalities are strict.
Proposition 5.1 states necessary and sufficient conditions on βββ ∈ [0, 1)4 to grant the semi-
stability of generic quasi-parabolic structures on evenly-split bundles, since then
|I| = d− 2 deg(L) + 1 = dimP
(
H0
(
CP1, E ⊗ L−1
))
is the minimum number of flags in any E· that can be interpolated by some line sub-bundle
L ⊂ E, for which we have that d ≡ |I| − 1 (mod 2). The inequalities (5.2) only depend on the
parity of d, and each possibility will be treated independently. In the case when d is even, these
inequalities are equivalent to the following interval bounds
−2 ≤ βI ≤ 0, |I| = 1 ⇔ 0 ≤ βI′ ≤ 2, |I ′| = 3, (5.3)
while when d is odd, we obtain the interval bounds
−3 ≤ β∅ ≤ −1 ⇔ 1 ≤ β{1,2,3,4} ≤ 3 (5.4)
−1 ≤ βI ≤ 1, |I| = 2. (5.5)
For both degree parities, the total number of independent interval bounds is equal to four. Given
I ⊂ {1, 2, 3, 4}, let χI : {1, 2, 3, 4} → {0, 1} be its characteristic function. Since
β−1
I (−|Ic|) =
4∑
i=1
χIc(i) ei,
the map I 7→ vI := β−1
I (−|Ic|) determines a bijective correspondence between subsets I ⊂
{1, 2, 3, 4} and vertices in ∂[0, 1]4. A vertex vI ∈ ∂[0, 1]4 will be called even (resp. odd) if |I|
is even (resp. odd). The parity of a vertex is also equal to the parity of its number of nonzero
entries. More generally, we have the following convexity result, whose proof is straightforward.
Lemma 5.2. Any I ⊂ {1, 2, 3, 4} induces a partition of the set of vertices in ∂[0, 1]4 by the level
sets of βI . For every j = 0, . . . , 4, β−1
I (j − |Ic|) contains exactly
(
4
j
)
vertices, namely, those
whose Hamming distance to vI is j. Each level set is a convex hull of its set of vertices. The
partitions induced by I and Ic coincide.
The subset of the power set P ({1, 2, 3, 4}) containing sets of even (resp. odd) cardinality will
be denoted by P0({1, 2, 3, 4}) (resp. P1({1, 2, 3, 4})).
Definition 5.3. An even (resp. odd) partition set is any subset of P0({1, 2, 3, 4}) (resp.
P1({1, 2, 3, 4})) containing exactly one element of the pair {I, I ′} for every partition {1, 2, 3, 4} =
I ⊔ I ′ by even (resp. odd) subsets.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 17
Proposition 5.4. Let B0 (resp. B1) be the convex 4-polytope in [0, 1]4 determined by the interval
bounds (5.3) (resp. (5.4)–(5.5)). B0 is the convex hull of all even vertices in ∂[0, 1]4. Similarly,
B1 is the convex hull of all odd vertices in ∂[0, 1]4. Consequently, both are isomorphic to the
4-demicube. ∂B• is a triangulation with 16 tetrahedral cells, parametrized by partition sets I ⊂
P•({1, 2, 3, 4}) of given parity, as the convex hulls of the sets {vI ∈ B• : I ∈ I}.
Proof. The first two statements are straightforward from the explicit form of the interval
bounds (5.3)–(5.5). Moreover, 3-cells in each boundary ∂B• correspond to convex hulls of sets of
vertices in B• whose Hamming distance is exactly 2. The latter are in bijective correspondence
with partitions sets I ⊂ P•({1, 2, 3, 4}) of the given parity, whose cardinalities are equal to 4,
as I ↔ {vI ∈ B• : I ∈ I}. ■
Definition 5.5. A semi-stability wall is any hyperplanar region in [0, 1]4 of the form
HI,k = β−1
I (k), k ∈ {|I| − 1, |I| − 2, |I| − 3}.
A wall HI,k is even (resp. odd) if k is even (resp. odd). The walls HI,|I|−2 will be called interior ;
otherwise, they will be called boundary.
The definition of semi-stability walls is based on the condition (2.1) and the interval
bounds (5.1). Depending on their parity, semi-stability walls are subsets of either B0 or B1.
Since
HI,|I|−3 = HIc,|Ic|−1, (5.6)
even (resp. odd) boundary semi-stability walls are in bijection with subsets I of odd (resp. even)
cardinality under the correspondence I 7→ HI,|I|−3. In addition, there are four interior semi-
stability walls for each parity, corresponding to the four partitions I⊔I ′ = {1, 2, 3, 4} into subsets
of the same cardinality parity, in the form
HI,|I|−2 = HI′,|I′|−2. (5.7)
For both parities, v1/2 := (1/2, 1/2, 1/2, 1/2) is the intersection of all four interior semi-stability
walls. Tables 1 and 2 list all semi-stability walls for both parities.
Table 1. List of even semi-stability walls.
Boundary
H{i},−2 = H{j,k,l},2, {i} ⊔ {j, k, l} = {1, 2, 3, 4}
H{j,k,l},0 = H{i},0, {i} ⊔ {j, k, l} = {1, 2, 3, 4}
Interior
H∅,−2 = H{1,2,3,4},2
H{i,j},0 = H{k,l},0, {i, j} ⊔ {k, l} = {1, 2, 3, 4}
Table 2. List of odd semi-stability walls.
Boundary
H∅,−3 = H{1,2,3,4},3
H{i,j},−1 = H{k,l},1, {i, j} ⊔ {k, l} = {1, 2, 3, 4}
H{1,2,3,4},1 = H∅,−1
Interior H{i},−1 = H{j,k,l},1, {i} ⊔ {j, k, l} = {1, 2, 3, 4}
18 C. Meneses
Definition 5.6. The even (resp. odd) parabolic weight polytope is the refinement W0 (resp. W1)
of [0, 1]4 resulting after the inclusion of all semi-stability walls into B0 (resp. B1). An even
(resp. odd) open chamber is the interior of a 4-cell in W0 (resp. W1), or equivalently, any
connected component in
(0, 1)4\ ∪ HI,k, k ∈ 2Z (resp. k ∈ Z\2Z).
An open chamber inside B• is called interior, and exterior otherwise. A tetrahedral cell in ∂B•
is of type A if it is also contained in ∂[0, 1]4, and of type B otherwise.
Proposition 5.7. There are exactly 8 tetrahedral cells of type A in ∂B•. Tetrahedral cells of
type B correspond to the 8 boundary semi-stability walls of a given parity.
Proof. A tetrahedral cell belongs to ∂B• ∩ ∂[0, 1]4 if and only if it is the convex hull of a set
of 4 vertices of a given parity with a fixed coordinate component value βi = 0, 1, i = 1, 2, 3, 4,
yielding 8 possibilities for each parity. It follows from Lemma 5.2 and Proposition 5.4 that the
8 remaining tetrahedral cells in ∂B• are exhausted by the boundary semi-stability walls HI,|I|−3
for all I ⊂ {1, 2, 3, 4} of a given parity. ■
Corollary 5.8 (classification of exterior chambers). At most one inequality in either (5.3)
or (5.4)–(5.5) can fail to hold on (0, 1)4, and each boundary semi-stability wall HI,|I|−3 de-
termines a unique exterior chamber whose closure contains it. Every exterior chamber to B0
(resp. B1) is the interior of the convex hull of HI,|I|−3 and the vertex vI for a unique I ⊂
{1, 2, 3, 4} of odd (resp. even) cardinality.
Proof. The first claim follows from the convexity of the regions βI < |I|−3. Similarly, it follows
from the identity (5.6) that there is a bijective correspondence between exterior chambers and
regions β−1
I ((−|Ic|, |I| − 3)) for each I ⊂ {1, 2, 3, 4}, in such a way that their chamber parity is
given by the parity of |I| − 3. ■
On the other hand, interior chambers on each parabolic weight polytope are effectively clas-
sified in terms of the convex geometry of interior semi-stability walls. This is formalized in the
next two results.
Proposition 5.9. Every 2-cell in the boundary of a given tetrahedral cell in ∂B• is the inter-
section of the latter and an interior semi-stability wall of the given parity. Consequently, the
restriction W•|B• is equal to the pyramid of B• over the apex v1/2.
Proof. By Lemma 5.2, the interior semi-stability wall corresponding to a partition I ⊔ I ′ =
{1, 2, 3, 4} as in (5.7) is the convex hull of 6 vertices, obtained by removing {vI , vI′} (of pairwise
Hamming distance 4) from the set of 8 vertices of given parity. In this set of 6 vertices, there
exist exactly 8 triples of vertices whose pairwise Hamming distance is equal to 2, so that each
triple spans a 2-cell in ∂B•. It follows from Proposition 5.4 that every tetrahedral cell in ∂B• is
uniquely expressed as the convex hull of one of these triples and either vI or vI′ , in such a way
that its intersection with each of the 4 interior semi-stability walls in B• retrieves the 2-cells in
its boundary. The second claim is immediate from the convexity of B•, since v1/2 is the common
intersection of all interior semi-stability walls of given parity. ■
Corollary 5.10 (classification of interior chambers). Every interior chamber is the intersection
of a choice of side regions {βI < |I| − 2: I ∈ I} to all interior semi-stability walls for a unique
partition set I ⊂ P•({1, 2, 3, 4}). Equivalently, it is the interior of the convex hull of the apex v1/2
and the tetrahedral cell in ∂B• spanned by {vI′ : I ′ ∈ I}.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 19
Proof. It follows from Proposition 5.9 that interior chambers in each polytope B• can be
parametrized by a choice of side of all interior semi-stability walls. Any of these choices is
defined and parametrized by a unique partition set under the correspondence
I 7→
⋂
I∈I
RI , RI := β−1
I
(
(−|Ic|, |I| − 2)
)
,
given that for any partition {1, 2, 3, 4} = I ⊔ I ′, the open regions RI and RI′ satisfy
vI ∈ RI , vI′ ∈ RI′ , RI ∪RI′ = [0, 1]4, RI ∩RI′ = HI,|I|−2 = HI′,|I′|−2,
from which the first claim follows. Similary, it follows from Proposition 5.4 that the closure of
any interior chamber contains a unique tetrahedral cell in ∂B•, which is equal to the intersection
of ∂B• and the former. Any additional vertex in W• would necessarily arise from the intersection
of interior semi-stability walls of a given parity. Since their common intersection equals v1/2,
the second claim follows. ■
We will stick to the following notational convention. Interior chambers will be denoted as CI ,
where I is the partition set parametrizing each of them under the previous correspondence.
On the other hand, the exterior chamber to the boundary semi-stability wall HI,|I|−3 will be
denoted as CI .
The intersection of ∂B• and the closure of an open chamber of the same parity is a tetrahedral
cell and characterizes an interior chamber uniquely. Since tetrahedral cells of type B are exterior
semi-stability walls, they characterize uniquely both an interior and an exterior chamber as the
common boundary of their closures. The type of an interior chamber in W• (or equivalently,
its associated partition set) is defined as the type of its tetrahedral cell. Reflection along an
exterior semi-stability wall HI,|I|−3 exchanges the vertex vI and the apex v1/2, and determines
a bijection between its exterior chamber and its interior chamber of type B.
We will say that two interior chambers CI and CI′ are neighboring if a reflection along a semi-
stability wall HI,|I|−2 = HI′,|I′|−2 transforms one into the other. The proof of the next corollary
is immediate from Corollaries 5.8 and 5.10.
Corollary 5.11 (combinatorial wall-crossing). Every interior semi-stability wall of given parity
HI,|I|−2 = HI′,|I′|−2 is the union of 4 tetrahedral cells in W•, each of which is the common
boundary of one of the four possible pairs of neighboring interior chambers {CI , CI′} of opposite
type whose partitions sets satisfy
I\{I} = I ′\{I ′}.
Every even (resp. odd) exterior semi-stability wall HI,|I|−3 is the common boundary between the
exterior chamber CI and the even (resp. odd) interior chamber CI(I) of type B, where
I ′ ∈ I(I) ⇔ vI′ ∈ HI,|I|−3.
Tables 3 and 4 provide explicit lists of partition sets for each parity, together with the tetra-
hedral cells in ∂B· for the corresponding open chambers they parametrize, according to Propo-
sition 5.4. Notice that the collections of partition sets of a given type are closed under the
operation of taking complements in P•({1, 2, 3, 4}).
The group of symmetries of [0, 1]4 is called hexadecachoric group, denoted as B4. It is a group
of order 384 generated by permutation of coordinates and reflections along the hyperplanes
βi = 1/2, i = 1, 2, 3, 4. Its restriction to reflections along pairs of hyperplanes leads to a distin-
guished index 2 subgroup, the Coxeter group D4.
20 C. Meneses
Table 3. Even partition sets, i = 1, 2, 3, 4, {j, k, l} = {1, 2, 3, 4}\{i}.
Partition set I Tetrahedral cell in ∂B0
Type A
{∅, {k, l}, {j, l}, {j, k}} B0 ∩ {βi = 1}
{{1, 2, 3, 4}, {i, j}, {i, k}, {i, l}} B0 ∩ {βi = 0}
Type B
{∅, {i, j}, {i, k}, {i, l}} H{i},−2
{{1, 2, 3, 4}, {k, l}, {j, l}, {j, k}} H{j,k,l},0
Table 4. Odd partition sets, i = 1, 2, 3, 4, {j, k, l} = {1, 2, 3, 4}\{i}.
Partition set I Tetrahedral cell in ∂B1
Type A
{{i}, {i, k, l}, {i, j, l}, {i, j, k}} B1 ∩ {βi = 0}
{{j, k, l}, {j}, {k}, {l}} B1 ∩ {βi = 1}
Type B
{{1}, {2}, {3}, {4}} H∅,−3
{{i}, {j}, {i, j, l}, {i, j, k}} H{i,j},−1
{{2, 3, 4}, {1, 3, 4}, {1, 2, 4}, {1, 2, 3}} H{1,2,3,4},1
Corollary 5.12. D4 is the group of symmetries of both W0 and W1. Both polytopes are iso-
morphic under the action of the group B4/D4
∼= Z2. Moreover, D4 acts transitively on the
sets of exterior chambers, and on the sets of interior chambers of type A and B respectively.
The stabilizer of any open chamber is the permutation group of the vertices of its characteristic
tetrahedral cell.
Definition 5.13. The wall-crossing graph of W• is the 1-skeleton of its dual polytope. It assigns
a vertex to every open chamber in W•, and an edge to any pair of open chambers whose closures
intersect at a tetrahedral cell.
Figure 2. The wall-crossing graph for both W0 and W1. Square vertices correspond to the 8 exterior
open chambers. White round vertices correspond to the 8 interior open chambers of type A, while black
round vertices to the 8 interior open chambers of type B.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 21
The wall-crossing graph of W• inherits a coloring from the three possible types of open
chambers. The qualitative type of wall-crossing in W• is then measured by the possible vertex
types that are connected by a given edge in the wall-crossing graph. Up to the action of the
symmetry groupD4, there are only two qualitative types of wall-crossing: (i) between an exterior
chamber and its neighboring chamber of type B, whose closures intersect in a boundary wall,
and (ii) between neighboring chambers of type A and B, whose closures intersect in an interior
wall (Figure 2).
6 Conditional stability and basic building blocks
The construction of geometric models for all possible Harder–Narasimhan strata as spaces of
stable orbits under P(Aut(E))-actions, as well as the characterization of their wall-crossing under
variation of parabolic weights, will be achieved by linking the classification of conditionally
stable strata (Definition 6.3) to the combinatorics of weight polytopes. This problem depends
on a choice of degree parity for a rank 2 vector bundle. We will consider the two possibilities
simultaneously.
In order to classify the geometry and topology of all orbit spaces of conditionally stable
strata, which in general contain non-Hausdorff loci even after restricting the type of invariant
sub-bundles, their “taming” is achieved by dissecting them into elementary Hausdorff irreducible
components. The details are summarized in Definition 6.6 and Propositions 6.5, 6.7 and 6.9.
We begin with the following corollary.
Corollary 6.1. If m1−m2 ≤ 2, then for any βββ := (β1, β2, β3, β4) ∈ (0, 1)4, the set QPHs
βββ(E) of
stable elements in QPH(E) is nonempty, Zariski open, and contains QPHC∗(E). If m1−m2 ≥ 4,
then QPHs
βββ(E) = ∅ with respect to any βββ ∈ [0, 1)4.
Proof. The first claim follows from Lemmas 4.2 and 4.3, since QPHC∗(E) ̸= ∅ if and only if
m1 −m2 ≤ 2. If m1 −m2 ≥ 4, Lemma 4.3 implies that QPH(E) = QPH0(E) = Q(E) ⊔ R(E),
and every element in QPH(E) would be destabilized by E1. ■
Remark 6.2. The absence of weight constraints for the existence of moduli spaces of semi-stable
parabolic Higgs bundles in genus 0, following from Corollary 6.1, is already implicit, in terms
of the general non-abelian Hodge correspondence, in Simpson’s solution of the Deligne–Simpson
problem [31]. This is a sharp contrast with moduli spaces of semi-stable parabolic bundles [1, 5],
for which Φ ≡ 0.
Corollary 6.1 states that for even degree d = 2m, the only Birkhoff–Grothendieck splitting
types that can support stable parabolic Higgs bundles are
O(m)⊕O(m) and O(m+ 1)⊕O(m− 1),
while for odd degree d = 2m+ 1, these are
O(m+ 1)⊕O(m) and O(m+ 2)⊕O(m− 1).
It follows from general principles that stability of quasi-parabolic Higgs bundles is an invariant
along a given open chamber, and moreover, that the subset of points in QPH(E) that are
stable with respect to every choice of open chamber in W• is Zariski open. Nevertheless, we will
verify that the intersection of the former Zariski open subset with QPH0(E) is always empty with
respect to any choice of degree parity, i.e., it either coincides with QPHC∗(E) when m1−m2 ≤ 2,
or is empty when m1 − m2 ≥ 3. For any choice of open chamber C ⊂ W•, we will denote the
stable locus associated to any βββ ∈ C by QPHs
C(E).
22 C. Meneses
Definition 6.3. A point in QPH(E) will be called conditionally stable if it is stable with respect
to some choice of open chamber C ⊂ W• of corresponding degree parity. The different substrata
of conditionally stable quasi-parabolic Higgs bundles in QPH0(E) will be accordingly denoted
by Qcs(E), Rcs(E), and Scsj (E).
The subset of conditionally stable parabolic Higgs bundles in QPH(E) is equal to the com-
plement of the Zariski closed subset of points that are either unstable or strictly semi-stable for
any choice of weights βββ ∈ [0, 1)4. By Corollary 6.1, the latter is a subset of QPH0(E). The next
lemma is the key to characterize conditionally stable loci systematically. The characterization
is performed in Proposition 6.5.
Lemma 6.4. (E·,Φ) ∈ QPH0(E)\Q(E) is not stable with respect to any βββ ∈ [0, 1)4 if and only if
−
∣∣IcL(Φ),E·
∣∣ ≥ deg(E)− 2 deg(L(Φ)). (6.1)
A point (E·, 0) ∈ Q(E) is not stable with respect to any βββ ∈ [0, 1)4 if and only if there is a pair
of line sub-bundles L,L′ ⊂ E, L ∼= O(m1) and L′ ∼= O(m2), satisfying
{1, 2, 3, 4} = IL,E· ∪ IL′,E· . (6.2)
Proof. Since −|Ic| ≤ βI , the inequality (6.1) for some (E·,Φ) ∈ QPH0(E)\Q(E) implies that
the inequality (2.1) would never be satisfied in the strict sense for any given choice of βββ ∈ [0, 1)4.
The converse statement follows after taking the infimum of (2.1) over the values of βIL(Φ),E·
.
Similarly, the condition (6.2) implies that for any βββ ∈ [0, 1)4, (E·, 0) ∈ Q(E) is not stable
with respect to at least one sub-bundle of the pair {L,L′}. On the other hand, if (6.1) holds
for a point (E·, 0) ∈ Q(E) and some L ∼= O(m1), then it follows from the properties of line
bundle interpolation that there always exists L′ ∼= O(m2) such that the pair L,L′ satisfies (6.2).
Therefore, to prove the remaining implication, it suffices to assume that (E·, 0) ∈ Q(E) is not
stable for any βββ ∈ [0, 1)4 while (6.1) doesn’t hold for any L ∼= O(m1).
We will treat the possibilities m1 = m2 and m1 > m2 separately. If we assume that m1 > m2,
instability of (E·, 0) for all βββ ∈ [0, 1)4, and |IE1,E· | < 4−m1 +m2, then the instability of (E·, 0)
for βIE1,E· < m2 −m1 implies the existence of a destabilizing L′ ∼= O(m2), and we obtain
IL′,E· = IcE1,E· .
In particular, (6.2) holds. Similarly, if we assume that m1 = m2 = m and L ∼= O(m) with
|IL,E· | < 4, then the instability of (E·, 0) for all βIL,E· < 0 implies the existence of L′ ∼= O(m)
such that
IL′,E· = IcL,E· .
The case when (F1, F2, F3, F4) ∈ B{1,2,3,4},m(E) holds trivially, since this is the only instance for
which IL,E· ∩ IL′,E· ̸= ∅, and then (6.2) holds if we let L′ = L. ■
Proposition 6.5. Let6
RI(E) := R(E) ∩ par−1(BI,m1(E)), SI,j(E) := Sj(E) ∩ par−1(BI,j(E)).
(i) The different types of conditionally stable substrata are classified as follows:
Qcs(E) =
{
ι(U(E)) if m1 −m2 ≤ 2,
∅ if m1 −m2 = 3,
6The notation for SI,j(E) is redundant when 2(deg(L(Φ))+1)−deg(E) = 4, that is, when E ∼= O(m)⊕O(m)
or E ∼= O(m+ 1)⊕O(m− 1) and j = m− 1, since then S{1,2,3,4},m−1(E) = Sm−1(E).
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 23
Rcs(E) =
⊔
|I|≤3−m1+m2
RI(E),
Scsj (E) =
{⊔
|I|=2,3 SI,m(E) if m1 = m2 = m, j = m,
Sj(E) otherwise.
(ii) An element in Qcs(E) is stable with respect to any choice of interior open chamber if and
only if it lies in ι(U′(E)).
(iii) P(Aut(E)) acts freely on Rcs(E). If m1 = m2 = m then P(Aut(E)) acts freely on Scsm(E)
and P(Aut(E)) × C∗ acts freely and transitively on Scsm−1(E). If m1 − m2 = 1, 2 then
P(Aut(E)) acts freely and transitively in the factor Pm2(E) of Sm2(E).
Proof. The first and second claims follow from Lemma 6.4. The third claim follows from
Proposition 4.5, the definition of the subspaces BI,j(E) ⊂ QP(E) in (3.1) together with prop-
erty (3.2), the definition of the strata BI,m1(E) when m1 > m2, and the specializations of the
canonical form (4.6) (including (4.8)) in each case. Notice that
⊔
|I|≤2RI(E) =
⊔
|I|=1,2RI(E)
when m1 −m2 = 1, since R∅(E) = ∅. ■
Definition 6.6. The (possibly empty) basic building blocks of a subset I ⊂ {1, 2, 3, 4}, denoted
by KI (resp. LI whenever j = m2), are the P(Aut(E))-orbit spaces of RI(E) (resp. SI,j(E)).
In turn, the individual orbits NI are unambiguously defined as
NI := ι(BI,j(E) ∩U(E))/P(Aut(E)).
When E is evenly-split we will denote the “bulk” of orbits of quasi-parabolic bundles in QP(E)
that are stable with respect to any interior open chamber as
NB• := ι(U′(E))/P(Aut(E)) ∼= CP1\{z1, 0, 1,∞}.
Propositions 4.5 and 6.5 imply that the orbit space of Scsm−1(E) consists of a single basic
building block, which is a P(Aut(E))-principal bundle with base a point when E ∼= O(m+ 1)⊕
O(m− 1) and C∗ when E ∼= O(m)⊕O(m). We will denote these as L{1,2,3,4},0 and L{1,2,3,4},1
respectively. Even though they are associated to different bundle splitting types, they glue
into a punctured line (Proposition 7.4). For notational simplicity, we will denote the resulting
punctured line by L{1,2,3,4}.
Proposition 6.7.
(i) There is a bijective correspondence between basic building blocks NI (resp. KI , resp. LI)
and subsets I ⊂ {1, 2, 3, 4}.
(ii) When nonempty, the action of P(Aut(E)) on SI,m2(E) and RI(E) is free and proper.
Apart from L{1,2,3,4},0 = pt and L{1,2,3,4},1 ∼= C∗, the nonempty basic building blocks are
classified as follows:
KI
∼=
{
C, |I|+m1 −m2 = 2,
CP1, |I|+m1 −m2 = 3,
LI
∼=
{
C, |I| −m1 +m2 = 2,
CP1, |I| −m1 +m2 = 3.
Moreover, for any |I| = 3−m1 +m2, there is an isomorphism
KI ⊔
{ ⊔
I′⊂I, |I\I′|=1
KI′
}
∼= BlI
(
CP1
)
24 C. Meneses
and for any |I| = 3 +m1 −m2, there is an isomorphism
LI ⊔
{ ⊔
I′⊂I, |I\I′|=1
LI′
}
∼= BlI
(
CP1
)
.
Proof. The first claim is already implicit in Proposition 3.1 for the basic building blocks of
the form NI , and the remaining possibilities are implicit in the second claim. In terms of this
correspondence, the resulting orbits are individual points. The second claim is a consequence
of the P(Aut(E))-equivariance of par combined with Propositions 4.5 and 6.5. When I is such
that |I| = 3−m1 +m2 (resp. |I| = 3 +m1 −m2), the divisor (Φ) associated to any element in
RI(E) (resp. SI,m(E)) takes the form
(Φ) = z +
∑
j∈Ic
zj ,
in such a way that if z = zi for any i ∈ I, then Fi = L(Φ)|zi . In this case, the correspondence
(E·,Φ) 7→ z
determines the isomorphism KI
∼= CP1 (resp. LI
∼= CP1). Moreover, when I ′ ⊂ I is such that
|I ′| = 2 − m1 + m2 (resp. |I ′| = 2 + m1 − m2), letting I\I ′ = {i} we have that z = zi and
Fi ̸= L(Φ)|zi . Then
(E·,Φ) 7→ Fi
gives a biholomorphism KI′
∼= C (resp. LI′
∼= C), and moreover, the biholomorphism
KI ⊔ KI′
∼= Bl{i}
(
CP1
) (
resp. LI ⊔ LI′
∼= Bl{i}
(
CP1
))
,
from which the claim follows. ■
There are three non-Hausdorff orbit spaces of conditionally stable strata in QPH0(E)\Q(E),
resulting from fixing a type of kernel lines, as a consequence of the pathological orbit space
topology of U(E)\U′(E) and Y(E). These are parametrized by the incidence properties of
certain projective plane configurations. When m1 = m2, the topology of the union of basic
building blocks LI is determined by the incidence properties of the complete quadrilateral
(6243) (Figure 3), with all basic building blocks of the form L{j,k,l} (resp. all pairs of the
form {L{i,j},L{k,l}} for any partition {i, j} ⊔ {k, l} = {1, 2, 3, 4}) identified under the quotient
topology in the full orbit space. When m1 − m2 = 1, the same is true for all basic building
blocks KI and the incidence properties of the complete quadrangle (4362) (Figure 4), with all
basic building blocks of the form K{k,l} being identified under the quotient topology in the
full orbit space. When m1 − m2 = 2, the basic building blocks KI are superimposed as an
identification of four blow-ups of a line at a point (Figure 5). A full list of conditionally stable
loci grouped by QPH0(E)-stratum type, and their induced orbit space type, is compiled in
Table 5.
For each degree parity, there is a bijective correspondence between nontrivial basic building
blocks and subsets I ⊂ {1, 2, 3, 4}. We conclude with a combinatorial rule to organize their
families according to their biholomorphism type.
When the parities of |I| and d differ, basic building blocks are biholomorphic to CP1 and
in bijective correspondence with the sets of exterior semi-stability walls. Concretely, the collec-
tions are{
{KI : |I| = 1} and {LI : |I| = 3} if d is even,
K∅, {KI : |I| = 2} and L{1,2,3,4} if d is odd.
(6.3)
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 25
Table 5. Master kit of orbit spaces for all possible conditionally stable loci, following from Proposi-
tions 6.5 and 6.7.
m1 −m2 Conditionally stable locus Orbit space
0
Qcs(E) = ι(U(E)) CP1 w. 3 double points
Scsm(E) =
⊔
|I|=2,3 SI,m(E) (6243)-family (Figure 3)
Scsm−1(E) = Sm−1(E) C∗
1
Qcs(E) = ι(U(E)) CP1 w. 4 double points
Rcs(E) =
⊔
|I|=1,2 RI(E) (4362)-family (Figure 4)
Scsm(E) = Sm(E) D4-config. (Figure 1)
2
Qcs(E) = ι
(
B∅(E)\B{1,2,3,4},0(E)
)
point
Rcs(E) =
⊔
|I|=0,1 RI(E) (1− 4)-family (Figure 5)
Scsm−1(E) = Sm−1(E) point
3
Qcs(E) = ∅ –
Rcs(E) = R∅(E) CP1
a b
Figure 3. (a) The complete quadrilateral or (6243)-configuration. (b) Maximal Hausdorff unions of
basic building blocks in the orbit space of Scs0 (E) when E ∼= O(m) ⊕ O(m), parametrized by the sets
{{j, k}, {j, l}, {k, l}, {j, k, l}}, j ̸= k, j ̸= l, k ̸= l.
a b
Figure 4. (a) The complete quadrangle or (4362)-configuration. (b) Maximal Hausdorff union of basic
building blocks in the orbit space of Rcs(E) when E ∼= O(m + 1) ⊕ O(m), parametrized by the sets
{{k}, {l}, {k, l}}, k ̸= l.
26 C. Meneses
∅
Figure 5. The (non-Hausdorff) orbit space of Rcs(E) when E ∼= O(m + 1) ⊕ O(m − 1), formed by
union of the punctured line K∅ and the quadruple line ∪4
i=1K{i}, with maximal Hausdorff subspaces
K{i} ⊔ K∅ ∼= Bl{i}(CP1), i = 1, 2, 3, 4.
In turn, a nontrivial basic building block is biholomorphic to C if the parities of |I| and d
coincide. It is convenient to group these in pairs corresponding to partitions I ⊔ I ′ = {1, 2, 3, 4}
of even and odd type, as those pairs are in bijective correspondence with interior semi-stability
walls. This will be the basis to the classification of interior wall-crossing. The resulting pairs
are {{
L{i,j},L{k,l}
}
and
{
K∅,L{1,2,3,4}
}
if d is even,{
K{i},L{j,k,l}
}
if d is odd.
(6.4)
We will also group N{1,2,3,4} := ι(B{1,2,3,4},m−1(E))/P(Aut(E)) when m1 = m2 = m, and
N∅ := ι(U(E))/P(Aut(E)) when m1 − m2 = 2. With the exception of the previous pair of
orbits occurring over different bundles, whose topology will be considered subsequently, all other
pairs of the form {NI ,NIc} are orbits in the same ambient space having intersecting closures.
When E evenly-split, these are the orbits determining double points in the orbit space of Qcs(E),
locally compactifying the corresponding puncture sphere model of NB• .
∅
∅
a
b
Figure 6. Nilpotent cone assembly kits for even degree interior chambers CI of type A. Left (resp.
right) columns correspond to E ∼= O(m) ⊕ O(m) (resp. E ∼= O(m + 1) ⊕ O(m − 1)). (a) I =
{∅, {k, l}, {j, l}, {j, k}}. (b) I = {{1, 2, 3, 4}, {i, j}, {i, k}, {i, l}}.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 27
Definition 6.8. The nilpotent cone assembly kit of an open chamber C ⊂ W• of given parity
is the collection of stable components in all conditionally stable orbit spaces associated to all
strata in the two admissible nilpotent loci.
Proposition 6.9 (open chamber stable orbit loci).
(i) The point NI is stable with respect to an interior chamber CI if and only if I ∈ I.
(ii) A I-basic building block is stable with respect to an interior chamber CI if and only if I ∈ I,
and stable with respect to an exterior chamber CI′ if and only if I ∈ I(I ′) and I ̸= I ′.
Proof. Immediate from the definition of stability and Corollaries 5.8, 5.10 and 5.11. ■
Figures 6–11, built from Definition 6.8 and Propositions 6.7 and 6.9, compile the geometry
and combinatorics of all nilpotent cone assembly kits in both parities, with central spheres
represented as horizontal lines. In particular, when d is odd, there is a single Harder–Narasimhan
stratum for all open chambers (corresponding to E ∼= O(m+1)⊕O(m)), except for the exterior
chamber C∅.
∅
∅
a
b
Figure 7. Nilpotent cone assembly kits for even degree interior chambers CI of type B. Left (resp.
right) columns correspond to E ∼= O(m) ⊕ O(m) (resp. E ∼= O(m + 1) ⊕ O(m − 1)). (a) I =
{∅, {i, j}, {i, k}, {i, l}}. (b) I = {{1, 2, 3, 4}, {k, l}, {j, l}, {j, k}}.
7 Proof of main results and further properties
7.1 Proof of Theorem 1.1
We will combine all previous results to construct all Harder–Narasimhan strata MC(E) of
a given open chamber C as P(Aut(E))-orbit spaces. The missing information to do so is con-
tained in Proposition 7.1, where we relate the geometry of bulk orbit spaces and basic building
blocks.
28 C. Meneses
∅
a
b
Figure 8. Nilpotent cone assembly kits for even degree exterior chambers CI . Left (resp. right) columns
correspond to E ∼= O(m)⊕O(m) (resp. E ∼= O(m+ 1)⊕O(m− 1)). (a) I = {i}. (b) I = {j, k, l}.
a b
Figure 9. Nilpotent cone assembly kits for odd degree interior chambers of type A, all in E ∼= O(m+
1)⊕O(m). (a) I = {{i}, {i, k, l}, {i, j, l}, {i, j, k}}. (b) I = {{j, k, l}, {j}, {k}, {l}}.
Proposition 7.1 (orbit space models in QPH(E)).
(i) The group P(Aut(E)) acts freely on QPH(E)\ι(U(E)c).
(ii) For E evenly-split, P(Aut(E)) acts properly on QPHC∗(E), turning it into a principal
P(Aut(E))-bundle over (ΣD\{w1}) × C∗ when E ∼= O(m) ⊕ O(m) and ΣD × C∗ when
E ∼= O(m+ 1)⊕O(m).
(iii) If m1 −m2 = 2, the map par× det determines the isomorphism
QPHC∗(E) ⊔ S0(E) ∼= Pm−1(E)×H0
(
CP1,K2
CP1(D)
)
,
inducing the biholomorphism{
QPHC∗(E) ⊔ S0(E)
}
/P(Aut(E)) ∼= H0
(
CP1,K2
CP1(D)
)
.
Proof. (i) Propositions 3.1 and 3.2 establish the claim over Q(E). The triviality of P(Aut(E))-
stabilizers for points in QPH(E)\Q(E) follows from Proposition 4.6(i). Namely, the claim is
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 29
a b
c
Figure 10. Nilpotent cone assembly kits for odd degree interior chambers of type B, all in
E ∼= O(m + 1) ⊕ O(m). (a) I = {{1}, {2}, {3}, {4}}. (b) I = {{i}, {j}, {i, j, l}, {i, j, k}}. (c)
I = {{2, 3, 4}, {1, 3, 4}, {1, 2, 4}, {1, 2, 3}}.
∅
a
b c
Figure 11. Nilpotent cone assembly kits for odd degree exterior chambers CI . Left (resp. right) columns
correspond to E ∼= O(m + 1) ⊕ O(m) (resp. E ∼= O(m + 2) ⊕ O(m − 1)). 1) I = ∅. 2) I = {i, j}. 3)
I = {1, 2, 3, 4}.
trivial over par−1(U′(E)), and on QPHC∗(E) ∩ par−1(X(E)) it follows from the explicit form
of the residual C∗-automorphism action on the canonical forms (4.9)–(4.10). Corollary 4.4,
Proposition 4.5, and the canonical form (4.5) for elements in R(E) when m1 − m2 = 1 imply
the claim over QPH0(E)\Q(E).
30 C. Meneses
(ii) Both claims follow in analogy to (i), after restriction to QPHC∗(E). For any q ∈
H0(CP1,K2
CP1(D))\{0}, the orbits in det−1(q) ∩ par−1(X(E)) are fixed by τ , and the limit
properties of X(E) with respect to U(E) described in Proposition 3.2 imply that the orbit space
of det−1(q) is biholomorphic to the punctured elliptic curve ΣD\{w1} when m1 = m2 = m, and
to ΣD when m1 −m2 = 1.
(iii) In analogy to (ii), the divisor of any parabolic Higgs field Φ is necessarily trivial for
both strata in QPHC∗(E) ⊔ S0(E) when m1 −m2 = 2, and every point is uniquely determined
by Φ, which admits a unique decomposition of the form
Φ = Φ0 +Φ1, det(Φ0) = 0, Φ1 ∈ H0
(
CP1,End(E)⊗KCP1
)
.
Moreover, the map
Φ1 7→ det(Φ0 +Φ1)
is a linear isomorphism to H0(CP1,K2
CP1(D)) for any fixed Φ0, from which the first claim follows.
The second claim follows from Proposition 6.5. ■
Proof of Theorem 1.1. The case m1 − m2 = 3 follows from Propositions 6.5, 6.7 and 6.9,
since QPHC∗(E) = ∅ (Lemma 4.3), and is trivial for all open chambers other than C∅, in which
case the Harder–Narasimhan stratum is a projective line (Figures 9–11).
The case m1 −m2 = 2 can be concluded from Propositions 6.9 and 7.1(iii) (see Figures 6–8;
Figure 12 illustrates the topology of the Harder–Narasimhan stratum when its nilpotent cone
component is a nodal rational curve).
When E is evenly-split, both claims would follow as a consequence of Proposition 7.1(i)
and (ii) once the orbit limits of QPHC∗(E) as det(Φ) → 0 are established. Proposition 6.9
ensures that for every possible choice of open chamber, this limit would be a rational curve,
possibly noncompact and disconnected, with at most nodal points. The possibilities for both
degree parities are illustrated in detail in the left column of Figures 6–11. By construction, the
resulting orbit spaces are smooth surfaces. ■
∅
∅
∅
Figure 12. Maximal Harder–Narasimhan strata in the splitting E ∼= O(m+1)⊕O(m− 1) correspond-
ing to the exterior chambers C{i}, consisting of the image of a Hitchin section and two nilpotent cone
irreducible components.
7.2 Proof of Theorems 1.2 and 1.3
In order to understand the geometry of gluing of strata, we will present a series of results on orbit
space limits associated to jumping families of even (resp. odd) degree bundles, i.e., 1-parameter
families F = {Et : t ∈ C} for which
E0 ∼= O(m+ 1)⊕O(m− 1)
(
resp. E0 ∼= O(m+ 2)⊕O(m− 1)
)
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 31
and Et is evenly-split if t ∈ C∗. A holomorphic family of cocycles {gt01 : t ∈ C} with respect to
the standard affine cover {U0 = CP1\{∞}, U1 = CP1\{0}} of CP1, with coordinates z ∈ U0,
ζ ∈ U1 related on the intersection U01 as ζ = 1/z, can be given as follows
gt01(z) =
(
zm+1 0
t zm−1
)
, d even,(
zm+2 0
t zm−1
)
, d odd.
That the corresponding vector bundles Et in each degree parity are evenly-split when t ∈ C∗, is
elucidated in the matrix factorizations of the form gt01 = gt0g01(g
t
1)
−1,
(
zm+1 0
t zm−1
)
=
(
1 −z
0 −t
)(
zm 0
0 zm
)(
z−1 −1
−t 0
)−1
, d even,
(
zm+2 0
t zm−1
)
=
(
1 −z2
0 −t
)(
zm+1 0
0 zm
)(
z−1 −1
−t 0
)−1
, d odd,
(7.1)
inducing the cocycle equivalences gt01
∼= g01 with the standard evenly-split cocycle. The impli-
cations of the existence of jumping families on the limiting behavior of the relevant orbit spaces
of quasi-parabolic structures are gathered in the next lemma.
Lemma 7.2. Let {Et : t ∈ C} be a jumping family of vector bundles of arbitrary degree. Then
lim
t→0
Y
(
Et
)
⊂ Y
(
E0
)
.
If d is even, then
lim
t→0
B{1,2,3,4},m−1
(
Et
)
= X
(
E0
)
.
Proof. It follows from the definition of Y(E) in Proposition 3.2, when E is not evenly-split,
that
Y(E) =
4⊔
i=1
{{E1|zi}} ×
{∏
j ̸=i
P(E|zi)
}
, d even,
QP(E), d odd.
The first claim follows from the cocycle factorizations (7.1), after acting with the local change
of trivializations gt0 and gt0 on the elements of Y(E), E evenly-split. If d is even, the second
statement is verified in the same way after normalizing elements in B{1,2,3,4},1(E
t), t ∈ C∗, in
the form (z1, 0, 1,∞), under the identification C (Et) ∼= Conf4(L0(E
t)) and the isomorphism
L0(E
t) ∼= CP1. ■
We will introduce additional notation for special components of orbit spaces in QPHC∗(E).
Recall that for m1 −m2 ≤ 2, the connected components of the sets X(E) are parametrized by
I ⊔ I ′ = {1, 2, 3, 4} of the corresponding degree parity. Concretely, these are
{i, j} ⊔ {k, l}, m1 = m2 = m,
{i} ⊔ {j, k, l}, m1 −m2 = 1,
∅ ⊔ {1, 2, 3, 4}, m1 −m2 = 2.
32 C. Meneses
For every partition I⊔I ′ = {1, 2, 3, 4}, we will denote its corresponding orbit space in QPHC∗(E)
by H{I,I′}. In all cases, the isomorphism H{I,I′} ∼= C∗ defined by the determinant map Φ 7→
det(Φ) readily follows from Proposition 7.1. Moreover, in the case when m1 −m2 = 2, we have
that H{∅,{1,2,3,4}} = QPHC∗(E)/P(Aut(E)). The next corollary is a consequence of Lemma 7.2
(see Figure 12).
Corollary 7.3. Every orbit in H{∅,{1,2,3,4}} is the limit of a holomorphic family of orbits in
QPHC∗(E)/P(Aut(E)), with E ∼= O(m)⊕O(m).
Proposition 7.4 (limits of P(Aut(E))-orbit families).
(i) If d is even, there is a jumping family such that
lim
t→0
N t
{1,2,3,4} = N 0
∅ ,
while
lim
t→0
⊔
L t
{j,k,l} =
⊔
K 0
{i},
and moreover, there is a family of orbits {(Et
· ,Φ
t)} ∈ L{1,2,3,4},1 such that
lim
t→0
{(
Et
· ,
1
t
Φt
)}
= L 0
{1,2,3,4},0.
In particular, L{1,2,3,4},1 ∼= C∗ is compactified by the orbits L{1,2,3,4},0 and N{1,2,3,4}.
(ii) If d is odd, then there is a jumping family for which
lim
t→0
L t
{1,2,3,4} = K 0
∅ .
Proof. Let E = O(m)⊕O(m) and consider any nonzero parabolic Higgs field of the form
Φ =
(
u1v1 −u21
v21 −u1v1
)
, (u1) ̸= (v1)
whose kernel line is an arbitrary element of L1(E), and consider the family of quasi-parabolic
Higgs bundles {(Et,Φt) : t ∈ C∗} that is induced in terms of the local form
Φt|U0 := tgt0(Φ|U0)
(
gt0
)−1
.
Since
lim
t→0
ker
(
Φt
)
= E0
1 ,
it follows from Lemma 7.2 that the corresponding P(Aut(E0))-orbit equals L 0
{1,2,3,4},0. The
remaining claims follow analogously from Lemma 7.2 and Proposition 6.7, given that
par
( ⊔
|I|=3+m1−m2
SI,m(E)
)
= Y(E)
when E is evenly-split, and otherwise
par
( ⊔
|I|=3−m1+m2
RI(E)
)
= Y(E). ■
It follows from Propositions 7.1 and 7.4 that for each choice of degree parity, the non-Hausdorff
phenomena on conditionally stable loci concentrates on the basic building block data (6.3)
and (6.4). This is stated in the following corollary.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 33
Definition 7.5. The central sphere SI for an interior open chamber CI is the associated moduli
space of stable parabolic bundles NCI . The central sphere SI for an exterior open chamber CI
is its corresponding stable basic building block (6.3).
Corollary 7.6. The union of all exterior central spheres SI of a given degree parity determines
a line of 8-tuple points. The pair (6.4) of a given partition I ⊔ I ′ = {1, 2, 3, 4} determines
a punctured line of double points compactified by {NI ,NI′}. The union of all conditionally
stable loci of given degree parity conforms a D4-configuration whose nodes are 10-tuple points,
its central punctured sphere is formed by 9-tuple points, and its 4 complementary components
are formed by double points.
Proof of Theorem 1.2. It follows from Proposition 7.1 together with Corollary 7.3 that
when d is even, H{∅,{1,2,3,4}} glues into the bulk QPHC∗(E) for m1 = m2 = m as a branch locus
component under τ over the ramification point w1, in a compatible way with respect to the map
det. This determines an isomorphism for both degree parities between det−1(H0(CP1,K2
CP1(D))
\{0}) and ΣD×C∗ which is independent of the choice of open chamber. Similarly, it follows from
Proposition 7.4 that for all open chambers, the different components in the nilpotent cone as-
sembly kit (whose explicit classification is compiled in Figures 6–11) glue into a D4-configuration
(Figure 1), on which det extends to 0 as a consequence of Proposition 7.1. The resulting explicit
geometry of the map det implies the isomorphism MC ∼= Mtoy for all degrees and all choices of
open chambers. ■
Proof of Theorem 1.3. Since wall-crossing is concentrated along the locus det−1(0) in
each MC , all that remains to be done after having established Theorems 1.1 and 1.2 is to
understand the resulting transformations in det−1(0) when either an interior or exterior semi-
stability wall is crossed. The claim for both possibilities follows if we combine Corollaries 5.11
and 5.12 with Corollary 7.6, as the latter ensures that all nilpotent cone components that can be
pairwise exchanged according to the combinatorial rules dictated by the former share the same
topology in their respective geometric model. ■
Remark 7.7. In [6, Section 6.7], Blaavand considers parabolic Higgs bundles of parabolic
degree zero on vector bundles of degree −1, subject to the constraints α11 = α21 = α31 = 0 and
α12 + α22 + α32 + α41 + α42 = 1. Those constraints confine the parabolic weights βββ to lie in the
open chamber C{1,2,3,4}. Consequently, the moduli spaces that he considers correspond to the
geometric model MC{1,2,3,4} .
Remark 7.8. The isomorphisms for the different open chamber geometric models are also
implicit in Corollary 5.12. In fact, all the constructed geometric models are obviously isomorphic
as complex surfaces, independently of the choice of degree parity. One standard strategy to
construct isomorphisms between moduli spaces associated to parabolic structures is known in
the literature under the name of elementary transformations [19, 20]. We have chosen not to
invoque that notion in this work, as our method of construction of geometric models is not
compatible with it: the groups and associated orbits are fundamentally different for each choice
of degree parity.
7.3 The Hitchin fibration, C∗-action, and Hitchin sections
Each geometric model MC can be endowed with a product of transversal projections
π1 × π2 : MC → CP1 ×H0
(
CP1,K2
CP1(D)
)
,
where π1 maps onto the central sphere of MC , while π2 is the standard Hitchin elliptic fibration
π2([(E∗,Φ)]) = det(Φ).
34 C. Meneses
By Proposition 7.1, the nonzero fibers π−1
2 (q), q ̸= 0 are isomorphic to the elliptic curve ΣD
of the pair (CP1, D), with an involution induced from (4.2) that we will keep denoting by τ .
It follows from Proposition 7.1 that the resulting branch loci are parametrized by the connected
components of the union of the sets X(E) of a given degree parity, and consequently, correspond
to the four components H{I,I′} parametrized by partitions I⊔I ′ = {1, 2, 3, 4} of the given parity.
The zero fiber π−1
2 (0) is called the nilpotent cone of the geometric model MC . In general the
Zariski open sets MC\π−1
2 (0) are independent of the choice of open chamber C.
The definition of π1 requires careful attention to details and will be presented in several
stages. In essence, the restriction π1 × π2|MC\π−1
2 (0) is defined as the induced branched cover
MC\π−1
2 (0)
2:1−−→ CP1 ×H0
(
CP1,K2
CP1(D)
)
\{0}, (7.2)
under identification of the first factor in the image with the central sphere, and the extension
to the nilpotent cone is then defined as its collapse to the former. We will make this precise in
terms of the global properties of the C∗-action induced by (4.3) on any geometric model MC ,
described in Propositions 7.9 and 7.10.
Yet another consequence of Propositions 4.6 and 6.5 – in analogy to the proof of Proposi-
tion 6.7 – is that for any basic building block in a pair (6.4), there is exactly one P(Aut(E))-orbit
contained in par−1(X(E)). In the special case when m1 −m2 = 2 and I = {1, 2, 3, 4}, this orbit
coincides with L{1,2,3,4},0. Depending on the basic building block they belong to in a given pair,
we will denote them as
K +
I
(
resp. L +
I
)
.
For any partition I ⊔ I ′ = {1, 2, 3, 4}, Corollary 7.6(ii) implies that the pair {K +
I ,L +
I′ } (or
{L +
I ,L +
I′ } if |I| = |I ′| = 2) conforms a double point.
Proposition 7.9. Any central sphere is pointwise-fixed by the C∗-action. For any complemen-
tary nilpotent cone component in a pair (6.4), K +
I (resp. L +
I ) is fixed by the C∗-action, while
KI\K +
I (resp. LI\L +
I ) is a C∗-principal homogeneous space. Moreover, in the latter case, any
orbit {(E·,Φ)} ∈ KI\K +
I (resp. LI\L +
I ) satisfies
lim
c→∞
c · {(E·,Φ)} = K +
I
(
resp. L +
I
)
,
while for any partition I ⊔ I ′ = {1, 2, 3, 4} of given parity and any orbit {(E·,Φ)} ∈ H{I,I′},
lim
c→0
c · {(E·,Φ)} =
{
K +
I ,L +
I′
} (
or
{
L +
I ,L +
I′
}
if |I| = |I ′| = 2
)
.
Proof. The first statement is clear when the central sphere is a moduli space of stable parabolic
bundles. Otherwise, a conditionally stable P(Aut(E))-orbit is fixed under the induced C∗-action
if and only if for any choice of c ∈ C∗, the equation
(F1, F2, F3, F4, cΦ) = (g · F1, g · F2, g · F3, g · F4,Ad(g)(Φ)) (7.3)
can be solved for some g ∈ P(Aut(E)). We will independently consider the two families of basic
building blocks according to their stratum type in QPH0(E). Under a choice of bundle isomor-
phism E ∼= O(m1)⊕O(m2), every orbit in a basic building block LI contains a representative
(F1, F2, F3, F4,Φ) whose parabolic Higgs field takes the form
Φ =
(
0 0
w2−m1+m2 0
)
, m1 −m2 = 0, 1, 2.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 35
In all cases, the set of these canonical forms is a principal homogeneous space for a subgroup in
P(Aut(E)) isomorphic to C∗. It follows from Proposition 6.7 and the definition of Y(E) for every
admissible splitting type, that when an orbit belongs to the central sphere SI of an exterior
chamber CI , the equation (7.3) can be solved on its subset of canonical form elements, and the C∗-
action leaves the orbit invariant. In the case of a nilpotent cone’s complementary component, the
same holds for the special orbit L +
I by the definition of X(E), while the action of the subgroup
stabilizing canonical forms is trivial on quasi-parabolic structures associated to LI\L +
I . Hence it
coincides with the standard C∗-action, and determines a principal homogeneous space structure
on it. The final claim, on the orbit limits under the C∗-action, follows once again from the
limiting properties of X(E) and Y(E) with respect to the corresponding P(Aut(E)) action on
the loci U(E).
On the other hand, in the case of conditionally stable basic building blocks KI , the parabolic
Higgs field of all elements in every given orbit is always in the normal form
Φ =
(
0 −v2+m1−m2
0 0
)
, m1 −m2 = 1, 2, 3.
Such parabolic Higgs field’s normal forms are R(Aut(E))-invariant by definition, and the residual
group of bundle automorphisms acting on the underlying space {E·} of quasi-parabolic bundles
associated to the orbit is isomorphic to C∗. The same argument as before also verifies the claim
in this case. ■
Proposition 7.10. For every interior open chamber CI of given parity, the elliptic fibration π2
possesses four natural holomorphic sections, extending the connected components of the branch
locus of (7.2) to π−1
2 (0), parametrized by the subsets I ∈ I,
σI : H0
(
CP1,K2
CP1(D)
)
→ MCI
such that
σI
(
H0
(
CP1,K2
CP1(D)
)
\{0}
)
= H{I,I′}, σI(0) = K +
I′
(
or L +
I
)
.
For every exterior chamber CI′′, an analogous construction holds in terms of its neighboring
interior chamber CI(I′′) of type B.
Proof. Recall that (4.9) and (4.10) establish correspondences between nonzero elements
q ∈ H0
(
CP1,K2
CP1(D)
)
\{0}
and orbits of parabolic Higgs bundles with parabolic Higgs field
Φ =
(
0 −v2+m1−m1
w2−m1+m2 0
)
in terms of factorizations q = v2+m1−m1w2−m1+m2 , parametrized by partitions I⊔I ′ = {1, 2, 3, 4}
induced by fixing the divisors (v2+m1−m1) and (w2−m1+m2). For each degree parity, there are
exactly four partitions, and for any such partition, we can assume without loss of generality that
I ∈ I. By Corollary 7.6 and Proposition 7.9, the induced isomorphism H0(CP1,K2
CP1(D))\{0}
∼= H{I,I′} can be extended uniquely to a section σI as σI(0) := K +
I (or σI(0) := L +
I if that were
the case for I). Since exterior wall-crossing leaves all basic building blocks in the pairs (6.4)
invariant, in the case of an exterior chamber CI′′ , the corresponding holomorphic sections can
be defined to be the same as the corresponding sections of MCI(I′′) . ■
36 C. Meneses
We will refer to the previous holomorphic sections as the Hitchin sections of the geometric
model MC . When d is even, three of the Hitchin sections are defined in the Zariski open Harder–
Narasimhan stratum of MC , while the fourth one belongs to the complementary stratum. This
fourth section corresponds to the holomorphic section constructed by Gothen–Oliveira [11].
In turn, when d is odd, all Hitchin sections always belong to the Zariski open Harder–Narasimhan
stratum of MC .
Corollary 7.11. Under the conventions of Proposition 7.10, for every open chamber C ⊂ W•
there is a biholomorphism
MC
∖⊔
σI
(
H0
(
CP1,K2
CP1(D)
)) ∼= T ∗CP1 (7.4)
mapping the C∗-action into the standard cotangent bundle C∗-action. The limit c → 0 collapses
the Zariski open set MC\
⊔
σI′(H
0(CP1,K2
CP1(D))) to the central sphere, and every component
H{I,I′} to the point K +
I (or L +
I ).
Remark 7.12. For each interior chamber CI , the points {NI : I ∈ I} ⊂ SI correspond to
isomorphism classes of stable parabolic bundles supporting nilpotent Higgs fields. Altogether,
they are called the wobbly locus in the literature. By Corollary 7.6, the induced loci on all
central spheres recovers the pair (CP1, D) for any open chamber.
Summarizing, we conclude that the projection π1 for an interior chamber CI can be defined
as an extension of the forgetful map π1([(E∗,Φ)]) = [E∗] over T
∗SI ⊂ MCI , under the implicit
isomorphism SI ∼= CP1. In virtue of Corollary 7.6, π1 is also defined for each exterior chamber CI
as a projection onto the corresponding exterior central sphere SI
∼= CP1. In both cases, the
restriction π1|π−1
2 (0) is the map collapsing the nilpotent cone to its central sphere SC along the
wobbly locus.
In view of the restrictions (7.2) and (7.4) and Corollary 7.11, every geometric model MC is
the simultaneous extension of each of the two complex surfaces
T ∗CP1 and ΣD × C∗
towards the Hitchin section loci and the nilpotent cone, respectively, in such a way that the
intersection of any elliptic fiber of π2 and the Hitchin section loci corresponds to the set of
Weierstrass points of the former. In the case when C is an interior chamber and NC ̸= ∅, π1
corresponds to the dominant abelianization morphism constructed in [2] for the only possible
spectral curve ΣD, whose very stable locus equals NB• . The C∗-action on MC corresponds to
the standard C∗-action on both surfaces. The characterization of the limits of this action follows
from Propositions 7.9 and 7.10 (Figure 13). There is always a bijective correspondence between
C∗-fixed points not in the central sphere and points in the wobbly locus.
7.4 Weight degeneration to semi-stability walls
We will finish this section with a brief discussion of an interpretation of the notion of parabolic
S-equivalence for any choice of parabolic weights along the intersection of any given semi-
stability wall with the interior of [0, 1]4. Strictly speaking, it is necessary to exclude the points in
∂[0, 1]4, as they lead to a degeneration of the moduli problem by definition, to a strictly smaller
number of marked points in CP1. Recently, Godinho–Mandini [10] have studied the geometry
of the special degeneration βββ = 0 of the Zariski open set of holomorphically trivial parabolic
Higgs bundles (in the case when d = 0), in terms of a correspondence with the so-called null
hyperpolygons.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 37
Figure 13. The topology of the union of the nilpotent cone and the four Hitchin sections, indicating in
red the four 0-dimensional and one 1-dimensional components of fixed points of the global C∗-action in
the geometric model, to which MC collapses in the limit c → 0.
Proposition 7.13.
(i) For any βββ ∈ H̊I,|I|−3, all points in the central spheres SI and SI(I) = NCI(I)
are strictly
semi-stable.
(ii) For any βββ ∈ H̊I,|I|−2 = H̊I′,|I′|−2 and a corresponding pair {CI , CI′} of neighboring interior
chambers, all points in the associated pair (6.4), as well as the pair {NI ,NI′}, are strictly
semi-stable. Additionally, the orbits of ι(BI,j(E) ∩ BI′,j(E)) ⊂ ι(X(E)) are strictly semi-
stable.
Proof. The result is a straightforward consequence from Lemma 6.4, Proposition 6.9, and the
definition of strict semi-stability for parabolic Higgs bundles. ■
Corollary 7.14 (parabolic S-equivalence on geometric models). For any βββ in the interior
of a semi-stability wall, the pointwise identification of strictly semi-stable components as in
Proposition 7.13 determines a surface biholomorphic to Mtoy.
A Residue models
We will describe another approach to explicitly construct moduli spaces of stable parabolic
Higgs bundles based on residue evaluations. We will refer to the resulting orbit spaces M ev
C (E)
as residue models for Harder–Narasimhan strata.
The desingularization by blow-up of the cone of nilpotent endomorphisms n(V ) of a 2-
dimensional vector space V is constructed as the smooth complex hypersurface
ñ(V ) :=
{
(ϕ, F ) ∈ n(V )× P(V ) : F ⊂ ker(ϕ)
}
.
The restriction to ñ(V ) of the projection onto the second factor is again a projection which we
will denote by pr, and endows ñ(V ) with the structure of a holomorphic line bundle of degree −2,
identifying its exceptional divisor E ∼= P(V ) with the zero section of the latter. The fiber n(F )
over a given line F ∈ P(V ) is identified with the Lie algebra of nilpotent endomorphisms of V
whose kernel contains F . ñ(V ) is the fiber product of n(V )\{0} (viewed as a principal C∗-bundle)
under the standard C∗-representation ρ on C,
ñ(V ) =
(
n(V )\{0}
)
×ρ C.
Our goal is to parametrize the spaces QPH(E) and their stratifications in terms of the geometry
of a natural evaluation map associated to D. Consider the maps
evzi : QPH(E) → ñ(E|zi), (E·,Φ) 7→ (Reszi Φ, Fi), i = 1, 2, 3, 4.
38 C. Meneses
We define the total evaluation to be the map Ev =
∏4
i=1 evzi . Letting Pr be the projection
map associated to the product of projections
∏4
i=1 przi , we obtain the following commutative
diagram
QPH(E)
Ev //
par
##
4∏
i=1
ñ(E|zi)
Pr
{{
QP(E)
(A.1)
which is equivariant with respect to the induced P(Aut(E))-actions. In other words, the total
evaluation map is derived from the residue map on parabolic Higgs fields
ResE : Int(E) →
4∏
i=1
n(E|zi), Φ 7→ (Resz1 Φ,Resz2 Φ,Resz3 Φ,Resz4 Φ),
by taking simultaneous blow-ups on its domain and codomain along the loci defined by the
identities (4.1).
Remark A.1. The surface ñ(V ) can be also described in terms of explicit coordinate charts.7
These coordinates determine another way to parametrize QPH(E) from local trivializations
on E, from which a projective model for Harder–Narasimhan strata can be constructed repeating
the ideas of this section (cf. [18]). We do not pursue that possibility here, since a coordinate-
invariant approach is self-sufficient for our general purposes.
For every z ∈ CP1, there is a natural action of the group Aut(E)|z on the space n(E|z)
by conjugation. Since the action of central elements is trivial, there is an induced action of
P(Aut(E|z)) on ñ(E|z) and P(E|z), making the projection prz equivariant. When m1 = m2 the
induced action of P(Aut(E)|z) on P(E|z) is transitive with stabilizer
P(Aut(E|z))F = P(P(F ))
for any F ∈ P(E|z). Moreover, since the action of P(Aut(E)|z) on n(E|z)\{0} is also transitive,
and such that for any ϕ ∈ n(E|z)\{0}, its stabilizer is equal to the subgroup
P(Aut(E|z))ϕ = P(R(F )),
this action partitions the surface ñ(E|z) into a total of two orbits, corresponding to the excep-
tional divisor Ez and its complement, which we will denote by Oz, that is
ñ(E|z) = Oz ⊔ Ez, Oz
∼= n(E|z)\{0}.
7A choice of isomorphism V ∼= C2 determines an identification n(V ) ∼= Z2\C2 under the explicit parametrization
(Z0, Z1) 7→
(
Z0Z1 −Z2
0
Z2
1 −Z0Z1
)
,
which is invariant under the Z2-action given by (Z0, Z1) 7→ (−Z0,−Z1), and associates the kernel line [Z0 :
Z1] ∈ P(C2) whenever (Z0, Z1) ̸= (0, 0). Thus, the introduction of the standard affine charts U0, U1 on P(C2)
induces two affine charts W0,W1
∼= C2 for the blown-up surface Z̃2\C2, defined on their domain intersections in
the complement of the exceptional divisor by the maps
(Z0/Z1, Z
2
1 ) 7→Z2
1
(
Z0/Z1 −(Z0/Z1)
2
1 −Z0/Z1
)
= Z2
0
(
Z1/Z0 −1
(Z1/Z0)
2 −Z1/Z0
)
7→
(
Z1/Z0, Z
2
0
)
.
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 39
When m1 > m2, the prz-inverse image of each of the two orbits Vz(E) ⊔ {E1|z} = P(E|z) is in
turn partitioned into the two P(Aut(E)|z)-orbits which result from the intersection with Ez and
its complement Oz. This way, the action of P(Aut(E)|z) on ñ(E|z) partitions it into a total of
four orbits (Figure 14), that we will denote by Oz,1, Oz,2, Oz,3 and Oz,4, where
Oz,1 := pr−1
z (Vz(E))\Ez =
(
Ez ∪ n(E1|z)
)c
is Zariski open and a principal homogeneous space for P(Aut(E)|z). The remaining orbits
Oz,2 := n(E1|z)\Oz,4, Oz,3 := Ez\Oz,4, Oz,4 := Ez ∩ n(E1|z)
are respectively isomorphic to C∗, C, and a single point, since the stabilizer of any point
(ϕ,E1|z) ∈ Oz,2 is equal to P(R(E1|z)).
Figure 14. Orbit stratification of ñ(E|z) when m1 > m2 ∀z ∈ CP1.
Given that Ev(Q(E)) =
∏4
i=1 Ezi , the inclusion ι : QP(E) → QPH(E) satisfies Pr◦Ev◦ι = id.
Since the sum of an arbitrary End(E)-valued holomorphic differential and any parabolic Higgs
field is again a parabolic Higgs field, there is an additional affine action of the (possibly trivial)
vector space H0(CP1,End(E) ⊗ KCP1) on QPH(E). The orbit of any (E·,Φ) ∈ QPH(E) is
the level set Ev−1(Ev(E·,Φ)), and Ev is precisely the invariant map for this action, endowing
QPH(E) with the structure of an (possibly trivial) affine bundle over its image.
Proposition A.2.
(i) If E is evenly-split, then Ev is a P(Aut(E))-equivariant holomorphic embedding, whose
image is the smooth 5-dimensional variety of residue constraints.
(ii) When m1−m2 = 2, QPHC∗(E)⊔Sm−1(E) and Q(E)⊔R(E) are H0(CP1,End(E)⊗KCP1)-
invariant, Ev(QPHC∗(E) ⊔ Sm−1(E)) ⊂
∏4
i=1 Ozi,1 and Ev(Q(E) ⊔ R(E)) =
∏4
i=1 Oc
zi,1
.
(iii) When m1 −m2 = 3, Ev(R∅(E)) =
∏4
i=1 Ozi,2.
Proof. (i) Ev is injective if and only if E is evenly-split. The definition of Ev implies that, in
general, dEv has constant rank at any point (E·,Φ) ∈ QPH(E). Consequently, Ev is a holomor-
phic embedding if E is evenly-split.
(ii) Since nonzero elements in H0(CP1,End(E) ⊗KCP1) preserve E1, and these exist if and
only ifm1−m2 ≥ 2, while QPH(E) = Q(E)⊔R(E) whenm1−m2 ≥ 3, H0(CP1,End(E)⊗KCP1)-
invariance is trivial unless m1−m2 = 2, and then it follows from (4.5), since QPHC∗(E)⊔S0(E)
is characterized by w0 ̸= 0, which is unchanged under addition of End(E)-valued holomorphic
differentials, and is independent of the isomorphism E ∼= O(m + 1) ⊕ O(m − 1). The second
claim is verified from (4.5) in analogy to the proof of Proposition 4.6(ii) and (iii).
(iii) When m1 −m2 = 3 by definition we have that Ev(R∅(E)) =
∏4
i=1 Ozi,2. ■
40 C. Meneses
Corollary A.3. Assume that QPHs
C(E) ̸= ∅.
(i) If E evenly-split, there is an isomorphism MC(E) ∼= M ev
C (E), where M ev
C (E) is the
P(Aut(E))-orbit space of
Ev(QPHs
C(E)) ⊂
4∏
i=1
ñ(E|zi).
(ii) When m1 −m2 = 2, M ev
C (E) := Ev(QPHs
C(E))/R(Aut(E)) ∼= MC(E).
(iii) When m1 −m2 = 3, R(Aut(E)) acts freely and transitively on Ev(QPHs
C(E)) and
MC(E) ∼= P
(
H0
(
CP1,End(E)⊗KCP1
))
.
Proof. (i) The claim follows from the equivariance of the commutative diagram (A.1), Propo-
sition A.2(i), and Theorem 1.1.
(ii)–(iii) When m1 − m2 ≥ 2, the affine H0(CP1,End(E) ⊗ KCP1)-action on QPH(E) is
nontrivial and commutes with the action of the subgroup R(Aut(E)). R(Aut(E)) acts trivially
on H0(CP1,End(E) ⊗ KCP1), and the induced action of P(Aut(E))/R(Aut(E)) on it equals
the square of its standard C∗-action. When m1 − m2 = 2, it follows from the classification of
nilpotent cone assembly kits and Proposition A.2(ii) that the action of P(Aut(E)) on the fibers
of Ev|QPHs
C(E) is transitive, and the result follows. When m1 − m2 = 3 there is a single case
when QPHs
C(E) ̸= ∅, namely C = C∅, for which QPHs
C(E) = R∅(E). Every element in R∅(E)
can be uniquely expressed as the sum of an element in ι(B∅,m+2(E)) and a nonzero element in
H0(CP1,End(E) ⊗ KCP1). The result follows from Proposition A.2(iii), since R(Aut(E)) acts
freely and transitively on
∏4
i=1 Ozi,2. ■
Acknowledgements
I would like to thank Hartmut Weiß, whose encouragement and support where crucial in promp-
ting the appearance of the present manuscript, Steven Rayan for providing many insightful
remarks, and the anonymous referee for the careful revision of the manuscript and the construc-
tive criticism provided. This work was supported by the DFG SPP 2026 priority programme
“Geometry at infinity”.
References
[1] Bauer S., Parabolic bundles, elliptic surfaces and SU(2)-representation spaces of genus zero Fuchsian groups,
Math. Ann. 290 (1991), 509–526.
[2] Beauville A., Narasimhan M.S., Ramanan S., Spectral curves and the generalised theta divisor, J. Reine
Angew. Math. 398 (1989), 169–179.
[3] Bérczi G., Doran B., Hawes T., Kirwan F., Geometric invariant theory for graded unipotent groups and
applications, J. Topol. 11 (2018), 826–855, arXiv:1601.00340.
[4] Bérczi G., Jackson J., Kirwan F., Variation of non-reductive geometric invariant theory, in Surveys in
Differential Geometry 2017. Celebrating the 50th Anniversary of the Journal of Differential Geometry, Surv.
Differ. Geom., Vol. 22, Int. Press, Somerville, MA, 2018, 49–69, arXiv:1712.02576.
[5] Biswas I., A criterion for the existence of a parabolic stable bundle of rank two over the projective line,
Internat. J. Math. 9 (1998), 523–533.
[6] Blaavand J.L., The Dirac–Higgs bundle, Ph.D. Thesis, University of Oxford, 2015.
[7] Fredrickson L., Mazzeo R., Swoboda J., Weiss H., Asymptotic geometry of the moduli space of parabolic
SL(2,C)-Higgs bundles, Proc. London Math. Soc., to appear, arXiv:2001.03682.
https://doi.org/10.1007/BF01459257
https://doi.org/10.1515/crll.1989.398.169
https://doi.org/10.1515/crll.1989.398.169
https://doi.org/10.1112/topo.12075
https://arxiv.org/abs/1601.00340
https://doi.org/10.4310/SDG.2017.v22.n1.a2
https://arxiv.org/abs/1712.02576
https://doi.org/10.1142/S0129167X98000233
https://doi.org/10.1112/jlms.12581
https://arxiv.org/abs/2001.03682
Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles 41
[8] Gaiotto D., Moore G.W., Neitzke A., Wall-crossing, Hitchin systems, and the WKB approximation, Adv.
Math. 234 (2013), 239–403, arXiv:0907.3987.
[9] Godinho L., Mandini A., Hyperpolygon spaces and moduli spaces of parabolic Higgs bundles, Adv. Math.
244 (2013), 465–532, arXiv:1101.3241.
[10] Godinho L., Mandini A., Quasi-parabolic Higgs bundles and null hyperpolygon spaces, Trans. Amer. Math.
Soc. 374 (2021), 7411–7447, arXiv:1907.01937.
[11] Gothen P.B., Oliveira A.G., Topological mirror symmetry for parabolic Higgs bundles, J. Geom. Phys. 137
(2019), 7–34, arXiv:1707.08536.
[12] Hamilton E., Stratifications and quasi-projective coarse moduli spaces for the stack of Higgs bundles,
arXiv:1911.13194.
[13] Hausel T., Compactification of moduli of Higgs bundles, J. Reine Angew. Math. 503 (1998), 169–192,
arXiv:math.AG/9804083.
[14] Heller L., Heller S., Abelianization of Fuchsian systems on a 4-punctured sphere and applications, J. Sym-
plectic Geom. 14 (2016), 1059–1088, arXiv:1404.7707.
[15] Heu V., Loray F., Flat rank two vector bundles on genus two curves, Mem. Amer. Math. Soc. 259 (2019),
v+103 pages, arXiv:1401.2449.
[16] Hitchin N.J., The self-duality equations on a Riemann surface, Proc. London Math. Soc. 55 (1987), 59–126.
[17] Kim S., Wilkin G., Analytic convergence of harmonic metrics for parabolic Higgs bundles, J. Geom. Phys.
127 (2018), 55–67, arXiv:1705.08065.
[18] Komyo A., Saito M.H., Explicit description of jumping phenomena on moduli spaces of parabolic connections
and Hilbert schemes of points on surfaces, Kyoto J. Math. 59 (2019), 515–552, arXiv:1611.00971.
[19] Loray F., Saito M.H., Lagrangian fibrations in duality on moduli spaces of rank 2 logarithmic connections
over the projective line, Int. Math. Res. Not. 2015 (2015), 995–1043, arXiv:1302.4113.
[20] Loray F., Saito M.H., Simpson C.T., Foliations on the moduli space of rank two connections on the projective
line minus four points, in Geometric and Differential Galois Theories, Sémin. Congr., Vol. 27, Soc. Math.
France, Paris, 2013, 117–170, arXiv:1012.3612.
[21] Mehta V.B., Seshadri C.S., Moduli of vector bundles on curves with parabolic structures, Math. Ann. 248
(1980), 205–239.
[22] Meneses C., Remarks on groups of bundle automorphisms over the Riemann sphere, Geom. Dedicata 196
(2018), 63–90, arXiv:1607.03865.
[23] Meneses C., Geometric models for moduli of rank 2 parabolic Higgs bundles in genus 0 and applications, in
preparation.
[24] Meneses C., Takhtajan L.A., Logarithmic connections, WZNW action, and moduli of parabolic bundles on
the sphere, Comm. Math. Phys. 387 (2021), 649–680, arXiv:1407.6752.
[25] Miranda R., The basic theory of elliptic surfaces, Dottorato di Ricerca in Matematica, ETS Editrice, Pisa,
1989.
[26] Mukai S., An introduction to invariants and moduli, Cambridge Studies in Advanced Mathematics, Vol. 81,
Cambridge University Press, Cambridge, 2003.
[27] Nakajima H., Hyper-Kähler structures on moduli spaces of parabolic Higgs bundles on Riemann surfaces,
in Moduli of Vector Bundles (Sanda, 1994; Kyoto, 1994), Lecture Notes in Pure and Appl. Math., Vol. 179,
Dekker, New York, 1996, 199–208.
[28] Rayan S., The quiver at the bottom of the twisted nilpotent cone on P1, Eur. J. Math. 3 (2017), 1–21,
arXiv:1609.08226.
[29] Rayan S., Schaposnik L.P., Moduli spaces of generalized hyperpolygons, Q. J. Math. 72 (2021), 137–161,
arXiv:2001.06911.
[30] Simpson C.T., Harmonic bundles on noncompact curves, J. Amer. Math. Soc. 3 (1990), 713–770.
[31] Simpson C.T., Products of matrices, in Differential Geometry, Global Analysis, and Topology (Halifax, NS,
1990), CMS Conf. Proc., Vol. 12, Amer. Math. Soc., Providence, RI, 1991, 157–185.
[32] Thaddeus M., Variation of moduli of parabolic Higgs bundles, J. Reine Angew. Math. 547 (2002), 1–14,
arXiv:math.AG/0003222.
https://doi.org/10.1016/j.aim.2012.09.027
https://doi.org/10.1016/j.aim.2012.09.027
https://arxiv.org/abs/0907.3987
https://doi.org/10.1016/j.aim.2013.04.026
https://arxiv.org/abs/1101.3241
https://doi.org/10.1090/tran/8450
https://doi.org/10.1090/tran/8450
https://arxiv.org/abs/1907.01937
https://doi.org/10.1016/j.geomphys.2018.08.020
https://arxiv.org/abs/1707.08536
https://arxiv.org/abs/1911.13194
https://doi.org/10.1515/crll.1998.096
https://arxiv.org/abs/math.AG/9804083
https://doi.org/10.4310/JSG.2016.v14.n4.a4
https://doi.org/10.4310/JSG.2016.v14.n4.a4
https://arxiv.org/abs/1404.7707
https://doi.org/10.1090/memo/1247
https://arxiv.org/abs/1401.2449
https://doi.org/10.1112/plms/s3-55.1.59
https://doi.org/10.1016/j.geomphys.2018.01.023
https://arxiv.org/abs/1705.08065
https://doi.org/10.1215/21562261-2019-0016
https://arxiv.org/abs/1611.00971
https://doi.org/10.1093/imrn/rnt232
https://arxiv.org/abs/1302.4113
https://arxiv.org/abs/1012.3612
https://doi.org/10.1007/BF01420526
https://doi.org/10.1007/s10711-017-0309-y
https://arxiv.org/abs/1607.03865
https://doi.org/10.1007/s00220-021-04183-y
https://arxiv.org/abs/1407.6752
https://doi.org/10.1017/CBO9781316257074
https://doi.org/10.1007/s40879-016-0120-6
https://arxiv.org/abs/1609.08226
https://doi.org/10.1093/qmath/haaa036
https://arxiv.org/abs/2001.06911
https://doi.org/10.2307/1990935
https://doi.org/10.1515/crll.2002.051
https://arxiv.org/abs/math.AG/0003222
1 Introduction
1.1 Structure of the paper
1.2 Relation to other work
2 Conventions and index of notation
3 Geometry of P(Aut(E))-actions and stratifications in QP(E)
3.1 Line sub-bundles and interpolation of quasi-parabolic structures
3.2 Combinatorial stratification of orbit spaces in QP(E)
4 Structural results and stratification of QPH(E)
5 Combinatorial description of parabolic weight polytopes
6 Conditional stability and basic building blocks
7 Proof of main results and further properties
7.1 Proof of Theorem 1.1
7.2 Proof of Theorems 1.2 and 1.3
7.3 The Hitchin fibration, C^*-action, and Hitchin sections
7.4 Weight degeneration to semi-stability walls
A Residue models
References
|
| id | nasplib_isofts_kiev_ua-123456789-211725 |
| institution | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| issn | 1815-0659 |
| language | English |
| last_indexed | 2026-03-16T23:40:37Z |
| publishDate | 2022 |
| publisher | Інститут математики НАН України |
| record_format | dspace |
| spelling | Meneses, Claudio 2026-01-09T12:50:58Z 2022 Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study. Claudio Meneses. SIGMA 18 (2022), 062, 41 pages 1815-0659 2020 Mathematics Subject Classification: 14H60; 14D22; 32G13; 22E25 arXiv:2012.13389 https://nasplib.isofts.kiev.ua/handle/123456789/211725 https://doi.org/10.3842/SIGMA.2022.062 We construct explicit geometric models for moduli spaces of semi-stable strongly parabolic Higgs bundles over the Riemann sphere, in the case of rank two, four marked points, arbitrary degree, and arbitrary weights. The mechanism of construction relies on elementary geometric and combinatorial techniques, based on a detailed study of orbit stability of (in general non-reductive) bundle automorphism groups on certain carefully crafted spaces. The aforementioned techniques are not exclusive to the case we examine, and this work elucidates a general approach to construct arbitrary moduli spaces of semi-stable parabolic Higgs bundles in genus 0, which is encoded into the combinatorics of weight polytopes. We also present a comprehensive analysis of the geometric models' behavior under variation of parabolic weights and wall-crossing, which is concentrated on their nilpotent cones. I would like to thank Hartmut Weiß, whose encouragement and support were crucial in prompting the appearance of the present manuscript, Steven Rayan for providing many insightful remarks, and the anonymous referee for the careful revision of the manuscript and the constructive criticism provided. This work was supported by the DFG SPP 2026 priority programme “Geometry at infinity”. en Інститут математики НАН України Symmetry, Integrability and Geometry: Methods and Applications Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study Article published earlier |
| spellingShingle | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study Meneses, Claudio |
| title | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study |
| title_full | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study |
| title_fullStr | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study |
| title_full_unstemmed | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study |
| title_short | Geometric Models and Variation of Weights on Moduli of Parabolic Higgs Bundles over the Riemann Sphere: a Case Study |
| title_sort | geometric models and variation of weights on moduli of parabolic higgs bundles over the riemann sphere: a case study |
| url | https://nasplib.isofts.kiev.ua/handle/123456789/211725 |
| work_keys_str_mv | AT menesesclaudio geometricmodelsandvariationofweightsonmoduliofparabolichiggsbundlesovertheriemannsphereacasestudy |