---
title: Joyce Hypercomplex Manifolds
url: https://www.emergentmind.com/topics/joyce-hypercomplex-manifolds
type: topic
---

# Joyce Hypercomplex Manifolds

Searching arXiv for recent and foundational papers on Joyce hypercomplex manifolds.
Joyce hypercomplex manifolds are compact homogeneous spaces \(M=G/L\) endowed with \(G\)-invariant hypercomplex structures arising from Joyce’s root-theoretic construction, together with the metric and holonomy structures naturally attached to them. In the homogeneous setting treated in recent work, they are quotients \(M=G/L\) of a compact connected Lie group \(G\) by a closed subgroup \(L\), equipped with integrable complex structures \(I,J\) satisfying \(I^2=J^2=-\mathrm{id}\) and \(IJ=-JI=:K\), so that one obtains the quaternionic \(2\)-sphere \(\mathsf H=\{aI+bJ+cK:a^2+b^2+c^2=1\}\) [2604.22404]. Their significance lies in the fact that they furnish an exact class of compact homogeneous hypercomplex manifolds, admit a complete classification of invariant hypercomplex and HKT metrics, and connect representation theory, Obata and Bismut holonomy, strong HKT geometry, and twistor constructions in Donaldson–Thomas theory [2604.22404] [1004.5238] [2407.18229].

## 1. Homogeneous definition and Joyce’s root-theoretic construction

In the homogeneous setting, the basic objects are compact homogeneous spaces
\[
M=G/L
\]
with \(G\) a compact connected Lie group, \(L\subset G\) a closed subgroup, and a \(G\)-invariant hypercomplex structure [2604.22404]. A hypercomplex structure on a \(4n\)-dimensional manifold is given by integrable complex structures \(I,J\) with
\[
I^2=J^2=-\mathrm{id},\quad IJ=-JI=:K,
\]
which generate the quaternionic family \(\mathsf H\) of complex structures [2604.22404]. A hyperhermitian metric \(g\) is Hermitian with respect to every \(P\in\mathsf H\), and the associated \((2,0)\)-form with respect to \(I\) is
\[
\Omega=\frac12(\omega_J+i\omega_K),\qquad g=2\operatorname{Re}\big(\Omega(\cdot,J\cdot)\big)
\]
[2604.22404].

The homogeneous hypercomplex structures considered are exactly those produced by Joyce’s construction and classified by Dimitrov–Tsanov [2604.22404]. One fixes a maximal torus in \(G\), a maximal set of strongly orthogonal roots \(\{\alpha_1,\dots,\alpha_d\}\), and obtains a decomposition of the Lie algebra
\[
\mathfrak g=\mathfrak b_m\oplus\bigoplus_{j=1}^m\mathfrak d_j\oplus\bigoplus_{j=1}^m\mathfrak f_j,
\]
where each \(\mathfrak d_j\cong\mathfrak{su}(2)\), each \(\mathfrak f_j\) is built from root spaces in the corresponding layer, and \(\mathfrak b_m\) centralizes all \(\mathfrak d_j\) [2604.22404]. Under the condition
\[
\dim\mathfrak b_m-\dim\mathfrak l=4,
\]
the tangent module \(\mathfrak m=\mathfrak l^\perp\subset\mathfrak g\) decomposes as
\[
\mathfrak m=\bigoplus_{j=1}^m\mathfrak m_j,\qquad \mathfrak m_j=\mathbb RX_1^j\oplus\mathfrak d_j\oplus\mathfrak f_j,
\]
and the hypercomplex structure is defined layerwise [2604.22404].

On the \(4\)-plane spanned by \(X_1^j,X_2^j,X_3^j,X_4^j\), with \(\{X_2^j,X_3^j,X_4^j\}\) satisfying the \(\mathfrak{su}(2)\) relations
\[
[X_2^j,X_3^j]=2X_4^j,\quad [X_3^j,X_4^j]=2X_2^j,\quad [X_4^j,X_2^j]=2X_3^j,
\]
one sets
\[
IX_1^j=X_2^j,\quad IX_3^j=X_4^j,\quad JX_1^j=X_3^j,\quad JX_2^j=-X_4^j,
\]
and on \(\mathfrak f_j\),
\[
I|_{\mathfrak f_j}=\mathrm{ad}(X_2^j),\qquad J|_{\mathfrak f_j}=\mathrm{ad}(X_3^j)
\]
[2604.22404]. This algebraic construction yields the invariant hypercomplex structure, and when \(L\) is connected it integrates to a \(G\)-invariant hypercomplex structure on \(G/L\) [2604.22404].

A foundational classification result states that any invariant hypercomplex structure on a compact homogeneous space \(M=G/L\) admitting a hyper-Hermitian naturally reductive invariant metric is obtained via Joyce’s construction [1004.5238]. More precisely, if \(M=G/L\) is compact homogeneous, \(Q\) is a \(G\)-invariant hypercomplex structure, and \(g\) is a \(G\)-invariant hyper-Hermitian naturally reductive metric, then \(Q\) is exactly the one obtained by Joyce’s construction; if \(G\) is semisimple, every simple factor of \(\mathfrak g\) is of type \(A_n\) [1004.5238]. This identifies Joyce hypercomplex manifolds, under the natural reductiveness hypothesis, with the full compact homogeneous hypercomplex class in that setting.

## 2. Classification of invariant HKT metrics

For a hyperhermitian structure \((I,J,g)\), the relevant metric condition is the HKT condition. Let \(\nabla^{\Bis}_P\) be the Bismut connection of \((P,g)\). The structure is HKT when the three Bismut connections of \((I,g)\), \((J,g)\), and \((K,g)\) coincide [2604.22404]. A fundamental characterization used in the homogeneous theory is
\[
(I,J,g)\text{ is HKT}\iff \partial\Omega=0,
\]
where \(\partial\) is the Dolbeault operator with respect to \(I\) [2604.22404]. In the Lie algebra formulation this becomes an algebraic condition on \(\mathfrak m^{1,0}\) [2604.22404].

The decisive metric classification theorem for Joyce hypercomplex manifolds gives a complete description of all invariant HKT metrics [2604.22404]. There is a distinguished \(\mathrm{Ad}(L)\)-invariant hyperhermitian scalar product \(h\) on
\[
\mathfrak m=\bigoplus_{j=1}^m\mathfrak m_j
\]
such that each layer \(\mathfrak m_j\) is orthogonal, with
\[
h|_{\mathfrak d_j\oplus\mathfrak f_j}=-(\cdot,\cdot)|_{\mathfrak d_j\oplus\mathfrak f_j},\quad h(X_1^j,X_1^j)=\frac{4}{\|\alpha_j\|^2}
\]
[2604.22404]. The HKT condition forces strong restrictions: on each layer \(\mathfrak m_j\), the restriction of any invariant HKT metric must be a scalar multiple of \(h\), and different layers are orthogonal [2604.22404].

Accordingly, every invariant HKT metric has the form
\[
g=\sum_{j=1}^m g_j\,h|_{\mathfrak m_j\times\mathfrak m_j},\qquad g_j>0,
\]
and, conversely, if \(L\) is connected, every such choice defines a \(G\)-invariant HKT metric [2604.22404]. Thus the invariant HKT cone is an \(m\)-dimensional positive cone parametrized by the Joyce layers. This result is representation-theoretic: the root decomposition and the strongly orthogonal roots completely control the HKT geometry [2604.22404].

The connection with naturally reductive geometry is also structural. On a homogeneous Hermitian manifold, if the invariant metric is naturally reductive,
\[
g([X,Y]_{\mathfrak m},Z)+g(Y,[X,Z]_{\mathfrak m})=0,
\]
then the canonical homogeneous connection has torsion
\[
T(X,Y)=-[X,Y]_{\mathfrak m},
\]
its torsion \(3\)-form is
\[
c(X,Y,Z)=-g([X,Y]_{\mathfrak m},Z),
\]
and this canonical connection coincides with the Bismut connection [2604.22404]. However, on Joyce hypercomplex manifolds the canonical connection is not necessarily HKT for an arbitrary hyperhermitian metric; the classification above isolates exactly those invariant metrics for which the HKT condition holds [2604.22404].

## 3. HKT-Einstein geometry

The HKT-Einstein condition used in the homogeneous theory is defined via the Chern connection rather than the Levi-Civita or Bismut Ricci tensor [2604.22404]. For a Hermitian structure \((I,g)\), let \(\nabla^{\Ch}\) be its Chern connection and \(\Ric_{\omega_I}\) the Chern-Ricci form. An HKT metric is HKT-Einstein if there exists a smooth function \(\lambda\) such that
\[
\frac{\Ric_{\omega_I}-J\Ric_{\omega_I}}{2}=\lambda\,\omega_I.
\]
On compact manifolds, \(\lambda\) must be a constant \(\ge 0\), and \(\lambda=0\) is equivalent to the structure being balanced and the canonical bundle \(K_{M,I}\) being holomorphically trivial [2604.22404].

For a Joyce hypercomplex manifold, the Chern-Ricci form can be computed explicitly from the root system. Writing
\[
H_{\hat\delta}:=\frac12\sum_{\alpha\in\hat R^+}H_\alpha,
\]
one has
\[
\Ric_{\omega_I}(E_\gamma,\overline{E}_\gamma)=2i\,\gamma(H_{\hat\delta}),
\]
and \(\Ric_{\omega_I}\) vanishes on \(\mathfrak t^{1,0}\) [2604.22404]. Evaluating the HKT-Einstein condition on the Joyce layers shows that it is equivalent to the existence of \(\lambda\) with
\[
\alpha_j(H_{\hat\delta})=\lambda\,g_j\qquad \forall j.
\]
Up to overall scaling, this forces
\[
g_j=\alpha_j(H_{\hat\delta}),\qquad j=1,\dots,m
\]
[2604.22404].

These coefficients are positive and admit the root-theoretic expression
\[
\alpha_j(H_{\hat\delta})=\frac{\|\alpha_j\|^2}{4}\,(2+\dim_{\mathbb C}\mathfrak f_j)
\]
[2604.22404]. Consequently, every Joyce hypercomplex manifold admits a unique invariant HKT-Einstein metric up to scale, at least in the simply connected case [2604.22404]. This metric is the canonical representative in the invariant HKT cone.

The relation to other Einstein notions is limited. The HKT-Einstein condition is imposed on the \(J\)-antiinvariant part of the Chern-Ricci form, not on the Levi-Civita Ricci tensor [2604.22404]. Hyperkähler metrics form a special case where the Chern-Ricci form vanishes and \(\lambda=0\), but in general HKT-Einstein metrics need not be Levi-Civita Einstein [2604.22404].

## 4. Strong HKT metrics, Bismut parallelism, and rigidity

A HKT structure is called strong HKT if one, hence all, of the Hermitian structures \((P,g)\) is SKT:
\[
dT_P=0\quad\Leftrightarrow\quad dd^c_P\omega_P=0
\]
[2604.22404]. On Joyce hypercomplex manifolds, strong HKT geometry is highly restrictive.

For invariant HKT metrics
\[
g=\sum_{j=1}^m g_j\,h|_{\mathfrak m_j\times\mathfrak m_j},
\]
the condition \(\nabla^{\Bis}T^{\Bis}=0\) holds if and only if
\[
g_j=g_l\quad \text{whenever }\alpha_j,\alpha_l\text{ lie in the same irreducible component of }R
\]
[2604.22404]. Equivalently, the restriction of \(g\) to the root part \(\mathfrak n\) must come from an \(\mathrm{Ad}(G)\)-invariant inner product on \(\mathfrak g\) [2604.22404]. Under the same hypothesis, the Bismut curvature is also parallel:
\[
\nabla^{\Bis}R^{\Bis}=0.
\]
Such manifolds are Bismut–Ambrose–Singer manifolds [2604.22404].

Strong HKT implies these parallelism properties automatically. If \((M=G/L,I,J)\) is a Joyce hypercomplex manifold with an invariant strong HKT metric \(g\), then
\[
\nabla^{\Bis}T^{\Bis}=\nabla^{\Bis}R^{\Bis}=0
\]
[2604.22404]. Moreover, if
\[
M=\mathbb T^\ell\times(K_1/L_1)\times\cdots\times(K_t/L_t)
\]
with each \(K_j\) compact, connected, simple and \(L_j\subset K_j\) closed and connected, then each \(L_j\) must be trivial, so \(M\) itself is a compact Lie group and \((I,J,g)\) is left-invariant [2604.22404].

This leads to a sharp classification statement: if a simply connected Joyce hypercomplex manifold carries an invariant strong HKT metric, then it is isomorphic to a product
\[
M=\mathrm{SU}(2n_1+1)\times\cdots\times\mathrm{SU}(2n_r+1)
\]
equipped with a bi-invariant hyperhermitian metric [2604.22404]. In particular, no nontrivial fibrations \(G/L\) with \(L\neq\{e\}\) admit strong invariant HKT metrics [2604.22404].

A related group-manifold phenomenon appears at the level of explicit examples. When \(L=\{e\}\), so that \(M=G\) is a compact Lie group with a Joyce hypercomplex structure, the reference metric \(h\) is bi-invariant and naturally reductive for simple factors of types \(B_\ell,C_\ell,D_{2\ell},E_7,E_8,F_4,G_2\); in these cases it is strong HKT and Bismut-parallel [2604.22404]. For \(G=\mathrm{SU}(2n+1)\), the invariant HKT-Einstein metric coincides with the one previously constructed by modifying the Killing form [2604.22404].

## 5. Obata holonomy and twisted Calabi–Yau phenomena

The Obata connection is the unique torsion-free connection preserving the hypercomplex structure [2509.07722]. On a Lie group with a left-invariant hypercomplex structure, it is given by
\[
\nabla_X Y = \frac{1}{2}\Big([X,Y] + I[IX,Y] - J[X,JY] + K[IX,JY]\Big)
\]
for left-invariant vector fields [2509.07722]. Its holonomy lies in \(\mathrm{GL}(n,\mathbb H)\), and the restricted holonomy lies in \(\mathrm{SL}(n,\mathbb H)\) precisely under special geometric conditions [2509.07722].

For Joyce hypercomplex group manifolds, a recent holonomy analysis shows that maximal Obata holonomy is not generic [2509.07722]. If in the Joyce decomposition
\[
\mathfrak g=\mathfrak b\oplus\bigoplus_{i=1}^m\mathfrak d_i\oplus\bigoplus_{i=1}^m\mathfrak f_i
\]
there exists an index \(i\) with \(\mathfrak f_i=0\), then the quaternionic subspace \(\mathfrak h_i\cong\mathbb H\) is parallel under the Obata connection, and the holonomy is strictly smaller than \(\mathrm{GL}(n,\mathbb H)\) [2509.07722]. This yields a general reduction theorem: for all compact Lie groups in Joyce’s list except \(\mathrm{SU}(2n+1)\) and the flat Hopf surface case, every left-invariant Joyce hypercomplex structure has
\[
\mathrm{Hol}(\nabla)\subsetneq \mathrm{GL}(n,\mathbb H)
\]
[2509.07722].

The case \(\mathrm{SU}(2n+1)\) is exceptional. For \(\mathrm{SU}(3)\), Soldatenkov proved
\[
\mathrm{Hol}(\nabla)=\mathrm{GL}(2,\mathbb H),
\]
and there is now a higher-dimensional full-holonomy example on \(\mathrm{SU}(5)\) with
\[
\mathrm{Hol}(\nabla)=\mathrm{GL}(6,\mathbb H)
\]
for a specific Joyce hypercomplex structure [2509.07722]. At the same time, for every \(n>1\) there exist infinitely many Joyce hypercomplex structures on \(\mathrm{SU}(2n+1)\) with reduced Obata holonomy, produced by choosing the Joyce basis so that a large \(\mathfrak{su}(2k+1)\subset\mathfrak{su}(2n+1)\) is preserved by the hypercomplex structure and by the Obata connection [2509.07722].

The restricted holonomy condition \(\mathrm{Hol}^0(\nabla)\subset \mathrm{SL}(n,\mathbb H)\) has a different geometric meaning. On Joyce hypercomplex manifolds in the list
\[
\mathbb{T}^k\times \mathrm{SO}(2k+1),\ 
\mathbb{T}^{2k}\times \mathrm{SO}(4k),\ 
\mathbb{T}^k\times \mathrm{Sp}(k),\ 
\mathbb{T}^7\times \mathrm{E}_7,\ 
\mathbb{T}^8\times \mathrm{E}_8,\ 
\mathbb{T}^4\times \mathrm{F}_4,\ 
\mathbb{T}^2\times \mathrm{G}_2,
\]
the strong HKT metric built from the Killing form has closed Lee form, and one obtains
\[
\mathrm{Hol}^0(\nabla)\subset \mathrm{SL}(n,\mathbb H)
\]
[2509.07722]. Moreover, the pair \((g,\Psi)\), with \(\Psi=\Omega^n\), solves the twisted Calabi–Yau system on the complex manifold \((M,I)\) [2509.07722]. This produces compact twisted Calabi–Yau manifolds from Joyce hypercomplex geometry.

## 6. Cohomological and twistor-theoretic extensions of the Joyce framework

The term “Joyce” also appears in a second, distinct but related framework: Joyce structures on complex manifolds arising from Donaldson–Thomas theory. In that setting, a Joyce structure on a complex manifold \(M\) consists of a period structure with skew form and a compatible pre-Joyce structure, and it induces a complex hyperkähler structure on the total space \(X=TM\) [2407.18229] [2006.13059]. Here the relevant objects are not compact homogeneous spaces \(G/L\), but tangent bundles of moduli spaces such as spaces of stability conditions or moduli of quadratic differentials [2407.18229] [2203.17148].

The local form of such a Joyce structure is governed by a holomorphic potential \(W(z,\theta)\) satisfying a heavenly-type equation [2006.13059]. Under the affine symplectic fibration condition, one obtains a complex hyperkähler structure on \(X=TM\) with
\[
\Omega_- = T^*\omega,
\]
and the geometry is characterized locally by a nonlinear PDE for \(W\) [2006.13059]. Strong Joyce structures include further conditions: oddness under fibre inversion, homogeneity under an Euler vector field, and lattice periodicity [2006.13059]. This setting provides a twistor-space interpretation of Joyce structures and their associated hyperkähler metrics [2407.18229].

Concrete nontrivial examples arise from spaces of quadratic differentials. For the moduli space \(M(C,Q)\) of genus \(g>1\) curves with a quadratic differential with simple zeroes, one obtains a meromorphic Joyce structure and hence a complex hyperkähler structure on \(TM(C,Q)\) [2203.17148]. The period structure is built from the anti-invariant homology of the spectral curve \(\Sigma\), and the non-linear connection comes from isomonodromic deformation theory via spectral correspondence and flat connections [2203.17148]. This suggests a geometric realization of Joyce structures expected from Donaldson–Thomas theory.

A further development constructs Joyce structures from moduli spaces of meromorphic quadratic differentials on the sphere, especially in cases with odd pole orders, again producing complex hyper-Kähler metrics with homothetic symmetry [2509.05275]. Another explicit family is obtained from deformed polynomial oscillators of odd degree \(2n+1\), where the base \(M\) is the unfolding of the \(A_{2n}\)-singularity and the total space \(X\) carries a complex hyper-Kähler metric compatible with an affine symplectic fibration [2402.14352]. These are Joyce-type hyper-Kähler geometries rather than homogeneous Joyce manifolds in the Lie-theoretic sense.

A broader cohomological extension concerns hypercomplex manifolds in the sense of Joyce, where quaternionic Dolbeault and Bott–Chern cohomologies admit \(J\)-invariant and \(J\)-anti-invariant decompositions [2303.04890]. On compact hypercomplex manifolds satisfying the \(\partial\partial_J\)-Lemma, the structure is \(C^\infty\)-pure-and-full [2303.04890]. In real dimension \(8\), HKT existence on compact \(SL(2,\mathbb H)\)-manifolds can be characterized in terms of dimensions of \(J\)-invariant and Bott–Chern cohomology groups [2303.04890]. This cohomological viewpoint is not the homogeneous metric classification of Joyce hypercomplex manifolds, but it extends the Joyce framework into HKT detection, deformation theory, and almost abelian solvmanifolds [2303.04890].

These later developments suggest a bifurcation of terminology. In one usage, Joyce hypercomplex manifolds are the compact homogeneous manifolds \(G/L\) obtained from strongly orthogonal roots and Joyce decompositions [2604.22404] [1004.5238]. In another, Joyce structures are geometric structures on moduli spaces whose tangent bundles carry complex hyperkähler metrics of “Joyce type” [2407.18229] [2006.13059]. A plausible implication is that the shared terminology reflects a common structural core: quaternionic or hyperkähler geometry controlled by period data, symmetry, and nonlinear flatness conditions.

Source: https://www.emergentmind.com/topics/joyce-hypercomplex-manifolds