Journal of Geometry and Physics 72 (2013) 7–33

Contents lists available at SciVerse ScienceDirect

Journal of Geometry and Physics

journal homepage: www.elsevier.com/locate/jgp

Non-commutative complex differential geometry

Edwin Beggs a , S. Paul Smith b,∗
a
School of Physical Sciences, Swansea University, Swansea SA2 8PP, Wales, United Kingdom
b
Department of Mathematics, Box 354350, University of Washington, Seattle, WA 98195, United States

article info abstract

Article history: This paper defines and examines the basic properties of non-commutative analogues of
Received 20 September 2012 almost complex structures, integrable almost complex structures, holomorphic curvature,
Received in revised form 15 February 2013 cohomology, and holomorphic sheaves. The starting point is a differential structure on a
Available online 15 April 2013
non-commutative algebra defined in terms of a differential graded algebra. This is com-
pared to current ideas on non-commutative algebraic geometry.
Keywords:
© 2013 Elsevier B.V. All rights reserved.
Noncommutative complex geometry
Almost complex structures
Non-commutative holomorphic bundles

1. Introduction

1.1. Philosophy

This paper is about non-commutative complex analytic manifolds and holomorphic sheaf cohomology theory. The
classical theory of complex manifolds begins with a smooth manifold of even dimension endowed with an ‘almost complex
structure’. The Newlander–Nirenberg condition [1] says when this almost complex structure actually comes from a complex
coordinate system.
Every smooth complex algebraic variety is a complex manifold. The Kodaira embedding theorem characterizes the
compact complex manifolds that may be embedded in CPn . Chow’s theorem shows that every compact complex submanifold
of CPn is a smooth complex projective algebraic variety. Serre’s GAGA, the abbreviation of the title Géométrie algébrique et
géométrie analytique of [2], shows that the algebraic and analytic properties of a smooth complex subvariety of CPn are ‘‘the
same’’.
In sum, these results show that the basic definitions in complex differential and complex algebraic geometry are
compatible. This harmony confirms the appropriateness of the basic definitions in the two fields. Hodge theory provides
further compatibilities between complex differential and complex algebraic geometry.
There is nothing like this in non-commutative geometry. There is much less certainty about the ‘correct’ definitions.
This is entirely reasonable: there are more non-commutative algebras than commutative ones, and some non-commutative
algebras can be really horrible. There are many ways in which ideas from commutative geometry might be carried over
into the non-commutative world, and there have been many attempts to do this. But which ideas correspond to ‘geometry’
rather than the general theory of non-commutative algebras?
These ideas, whatever they are, must have a large and diverse collection of examples. They should usually reduce
to the corresponding classical structures, but we should not exclude ideas that only have non-trivial meaning in the
non-commutative world. And most classical geometric ideas should have non-commutative analogues. Certainly, in

∗ Corresponding author.
E-mail addresses: [email protected] (E. Beggs), [email protected] (S. Paul Smith).

0393-0440/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.geomphys.2013.03.018
8 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

mathematical physics, it would make no sense if we were to say that just because the real world should be non-commutative
because of quantum theory, we should no longer be allowed to teach general relativity because the concepts in it cannot
really work. If the universe really is ‘non-commutative’, then there probably are ideas in non-commutative geometry which
reduce to, for example, geodesics and parallel transport in a classical limit.
This, then, is the philosophy of the paper: There ought to be a concept of non-commutative complex analytic manifolds
within which one should be able to carry out many of the classical operations. To be meaningful, there should be a good
number of examples. And, for the future, there should be the possibility of using this as a bridge between non-commutative
complex analytic manifolds and non-commutative complex algebraic varieties.

1.2. Overview of the paper

This paper deals with non-commutative analogues of almost complex structures, integrable almost complex structures,
holomorphic curvature, cohomology, and holomorphic sheaves. We attempt to give minimal conditions for a non-
commutative complex structure that allow the cohomology of holomorphic sheaves to be constructed. We will always refer
to smooth or differentiable manifolds as real manifolds so as to distinguish them from complex manifolds.
An almost complex structure on a non-commutative real manifold having a non-commutative ∗-algebra A of ‘‘C-valued
differentiable functions’’ and a de Rham complex (Ω • A, d, ∗) with complex coefficients (a differential graded algebra) will be
defined as an A-module homomorphism J : Ω 1 A → Ω 1 A whose square is minus the identity, and satisfies a reality condition.
This data, together with an extension of J to a derivation on Ω • A, leads to an A-bimodule decomposition Ω • A = ⊕p,q Ω p,q A
into J-eigenspaces such that (Ω p,q A)∗ = Ω q,p A and

Ω p,q A

Ω nA =
p+q=n

for all n, and to associated operators

∂ : Ω p,q A → Ω p+1,q A and ∂ : Ω p,q A → Ω p,q+1 A .
Motivated by the Newlander–Nirenberg condition we say that the almost complex structure is integrable if dΩ 1,0 A ⊂
Ω 2,0 A ⊕ Ω 1,1 A. If the almost complex structure is integrable we show that
2
d = ∂ + ∂, ∂ 2 = 0, ∂∂ + ∂∂ = 0, ∂ = 0.
Suppose (Ω • A, d, ∗, J ) is an integrable almost complex structure on A. A holomorphic A-module is a pair (E , ∇) where
2
E is a left A-module and ∇ : E → Ω 0,1 A ⊗A E is a ∂ -connection/operator. If ∇ = 0 we call (E , ∇) a holomorphic left
A-module. We define a category Hol(A) of such modules. Associated to each (E , ∇) ∈ Hol(A) is a complex from which
•
we define cohomology groups H (E , ∇). Elements of E play the role of continuous sections of a sheaf on the underlying
non-commutative geometric object and elements of H 0 (E , ∇) = ker ∇ play the role of holomorphic sections.
This category Hol(A) is abelian if Ω 1,0 A is flat as a right A-module. If all Ω p,q A are flat right A-modules every short exact
sequence in Hol(A) yields a long exact sequence of the cohomology groups (Proposition 4.6).

1.3. Relation to non-commutative complex projective algebraic geometry

Let R be a finitely generated connected graded C-algebra of Gelfand–Kirillov dimension n+1. We assume R is left and right
noetherian, Artin–Schelter regular, and a domain. We think of R as a homogeneous coordinate ring (hcr) of an irreducible
non-commutative smooth complex projective variety, Projnc R, of dimension n, that is defined implicitly by declaring that
the category of ‘‘quasi-coherent sheaves’’ on it is
Gr R
Qcoh(Projnc R) := QGr(R) =
Fdim R
where Gr R is the category of Z-graded R-modules and Fdim R is its full subcategory of modules that are the sum of their
finite dimensional submodules and QGr(R) is the quotient category.
Projnc R is a purely algebraic construct. In order to treat Projnc R as a non-commutative differential geometric object on
which one can do calculus one first needs an underlying real structure on Projnc R, presumably defined in terms of a ∗-algebra
A of Gelfand–Kirillov dimension 2n and a de Rham complex (Ω • A, d) which is also a ∗-algebra.

1.3.1
Stafford and Van den Bergh’s survey article [3] shows that many ideas, tools, and results, of projective algebraic geometry
extend to the non-commutative setting in a seamless and satisfying way.
However, there are no non-commutative analogues of Chow’s theorem, the Kodaira embedding theorem, or Serre’s GAGA
principle. In the classical commutative setting those results allow the application of complex-analytic, Hodge-theoretic, and
Kähler geometric, methods to complex projective varieties, and conversely the methods of algebraic geometry apply to
appropriate complex manifolds.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 9

1.3.2
Until this gap, or chasm, is bridged non-commutative geometry will be without a union of algebraic and analytic methods.
If the gap were bridged non-commutative geometry would be greatly enhanced. Such a development might also lead to a
deep connection between non-commutative algebraic geometry and non-commutative geometry as conceived of by Connes,
i.e., based on operator algebras. At present there are few links between the two subjects. The main contacts between the
two schools include the following: the work of Connes and Dubois-Violette on the non-commutative 3-spheres related to
the 4-dimensional Sklyanin algebra; the work of Polishchuk and Schwarz on holomorphic bundles on the non-commutative
2-torus; the work of several people on homogeneous spaces for quantum groups.

1.3.3
Part of the problem is that non-commutative complex projective algebraic geometry deals only with holomorphic aspects
of geometry and the objects appearing in non-commutative complex projective algebraic geometry have no underlying real
structure: although many of these objects behave like smooth complex projective varieties, with few exceptions, there is no
underlying smooth non-commutative real manifold on which a complex structure is then imposed.
Polishchuk and Schwarz’s work on holomorphic structures on non-commutative tori [4] is the great exception to the last
statement and their work illustrates the advantages of being able to impose a complex structure on a non-commutative real
manifold.

1.3.4. Examples of complex structures on some non-commutative projective varieties

We end this paper by showing how the ideas in this paper apply to some particular non-commutative projective algebraic
varieties. The simplest of these examples, in Section 7.2, is CPnθ := Projnc Rθ where θ is a skew-symmetric (n + 1) × (n + 1)
matrix over R, Rθ is the free algebra C⟨z0 , . . . , zn ⟩ modulo relations zµ zν = λµν zν zµ and λµν = eiθµν .
More substantial examples, in Section 7.3, are the quantum group analogues of irreducible generalized flag manifolds
studied by Heckenberger and Kolb, [5,6].
Further specializing, the quantizations CPnq have been examined in detail by various subsets of F. D’Andrea, L. Dabrowski,
M. Khalkhali, G. Landi, A. Moatadelro, and W.D. van Suijlekom [7–11]. The point of view of framed and associated bundles
is discussed in [12].
The reader should be aware that the notation CPnq is used in different ways by different communities. Although there is a
feeling that the same object is being discussed the object called CPnq by one community belongs to a different category than
the object called CPnq by the other community.
Based on these examples we make some speculations in Section 7 about how one might impose integrable almost com-
plex structures on the kinds of non-commutative projective varieties that appear in the ‘‘Artin–Tate–Van den Bergh–Zhang’’
version of non-commutative projective algebraic geometry.

1.4. Relation to other work

Non-commutative differential geometry began with Connes’s 1985 IHES paper of that name [13]. That paper, the birth
of quantum groups, and Woronowicz’s 1987 and 1989 papers, [14,15], in which he developed a ∗-differential calculus for
quantum groups, were the beginning of a quarter-century development of non-commutative calculus.
Non-commutative calculus is an algebraic creation based on a non-commutative algebra that plays the role of, and
is viewed as, either the smooth or holomorphic functions on an imaginary non-commutative real manifold or complex
manifold. Most of this work has focused on developing either an analogue of the classical de Rham complex of smooth
forms or an analogue of the holomorphic de Rham complex.
Less has been done to develop a non-commutative calculus modelled on that for a complex manifold, i.e., one that begins
with the de Rham complex (Ω • A, d) of smooth forms, extends the scalars to C, and then using a non-commutative analogue
of a complex structure obtains a decomposition d = ∂ + ∂ and an associated decomposition ΩC• A = ⊕p,q Ω p,q A into (p, q)-
forms.
A recent paper along these lines by Khalkhali, Landi, and van Suijlekom [9] begins with a short history of non-
commutative complex geometry. That paper and others, for example, [16,17,7,8,18,10,19,20,4,21], examine a range of
interesting examples. Most of those examples fit into our framework.
Here we develop a fairly robust framework for a calculus of (p, q)-forms based on an almost complex structure J. Absent
the J-operator, the definition of the (p, q)-forms in any particular example appears somewhat ad hoc although because the
non-commutative examples that have been examined are closely modelled on classical commutative examples the guidance
provided by the classical case has suggested what the (p, q)-forms should be.
The completely different approach by twisting in [22] is an example of the current theory, as the classical integrability
condition twists into the non-commutative one.
10 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

2. Almost complex structures on ∗-algebras

In this paper we develop the rudiments of a non-commutative complex differential geometry based on the notion of
∗-algebras that applies to some non-commutative algebraic varieties. Several issues must be addressed in doing this. First,
most non-commutative projective varieties are defined globally in terms of their homogeneous coordinate rings. They have
no underlying topological space, and there is no patching together of affine pieces. Second, complex differential geometry is
developed by imposing additional structure on an underlying real manifold but most non-commutative projective varieties
do not seem to have an underlying non-commutative analogue of a real manifold. This forces us to develop a formalism
based on algebraic data that takes the place of the underlying real manifold. That underlying algebraic data consists of an
algebra with a ∗-structure.

2.1. ∗-algebras

Definition 2.1. A ∗-structure on an associative C-algebra A is a map A → A, a → a∗ such that a∗∗ = a, (ab)∗ = b∗ a∗ , and
(λa + µb)∗ = λa∗ + µb∗ for all a, b ∈ A and all λ, µ ∈ C. We then call A a ∗-algebra. We call an element a ∈ A self-adjoint
or Hermitian if a∗ = a. A ∗-homomorphism between ∗-algebras is a C-algebra homomorphism f such that f (a∗ ) = f (a)∗ for
all a.
Throughout this paper A denotes an associative not-necessarily-commutative C-algebra. If the ∗-structure is ignored,
elements of A should be thought of as playing the role of holomorphic functions on a ‘‘non-commutative complex manifold’’
or regular functions on a ‘‘non-commutative quasi-affine variety’’. With the ∗-structure, elements of (A, ∗) should be thought
of as C-valued functions on a ‘‘non-commutative real manifold’’ or ‘‘non-commutative real algebraic variety’’. Often (A, ∗)
plays the role of a dense subalgebra of a C ∗ -algebra.
For example, elements of the Hopf algebra Oq (SL(2, C)) can be thought of as regular functions on the non-commutative
complex algebraic quantum group SLq (2, C) or, when a suitable ∗-Hopf structure is considered, as C-valued polynomial
functions on its compact real form SUq (2).
The matrix algebra Mn (C) is always given the ∗-structure that sends a matrix to its conjugate transpose. An n-dimensional
∗-representation of a ∗-algebra A is a ∗-homomorphism ϕ : A → Mn (C).
The free algebra F = C⟨x1 , . . . , xn , x∗1 , . . . , x∗n ⟩ on 2n variables is a ∗-algebra with (xi )∗ = x∗i , and λ∗ = λ̄ for λ ∈ C. If A
is a finitely generated ∗-algebra there is a surjective ∗-homomorphism F → A for a suitable n.
If B is a commutative R-algebra A := C ⊗R B becomes a ∗-algebra by declaring that (λ ⊗ b)∗ := λ̄ ⊗ b for λ ∈ C and b ∈ B.
We recover B as the subalgebra of self-adjoint elements. A ∗-homomorphism f : A → C restricts to a homomorphism of
R-algebras B → R and, conversely, every R-algebra homomorphism B → R extends to a unique ∗-homomorphism A → C.
Every ∗-homomorphism is obtained by such an extension so extension and restriction are mutually inverse bijections
between ∗-homomorphisms A → C and R-algebra homomorphisms B → R.
These considerations apply when B is the coordinate ring of a real algebraic variety. Suppose X ⊂ Rn is the zero locus
of a set of polynomials with real coefficients. Let B = R[x1 , . . . , xn ]/I where I consists of the polynomials that vanish on X .
Then A := C ⊗ B is a ∗-algebra and X may be recovered as the set of ∗-homomorphisms A → C. We will write C[X ] for A
with this ∗-structure. We may think of C[X ] as the ring of C-valued polynomial functions on X with ∗-structure defined by
f ∗ (x) := f (x).
For example, if S 1 is the unit circle x2 + y2 = 1 then C[S 1 ] is C[x, y]/(x2 + y2 − 1) with x∗ = x and y∗ = y. Thus, if
z = x + i y, then z ∗ = x − i y and zz ∗ = 1 so C[S 1 ] ∼ = C[z , z ∗ ]/(zz ∗ − 1). The unit circle is recovered as the R-valued points
for the subalgebra of self-adjoint elements R[z + z ∗ , i (z − z ∗ )] ∼ = R[x, y]/(x2 + y2 − 1).
When B is an R-algebra that is not commutative the identity map on B is not a C-algebra anti-homomorphism so C ⊗R B
cannot be given a ∗-structure by defining (λ ⊗ b)∗ to be λ̄ ⊗ b. A standard example is provided by the algebra B = R[x, x∗ ]
generated by the annihilation and creation operators x and x∗ with relation xx∗ − x∗ x = 1. Now A = C ⊗R B is the Weyl
algebra. The momentum and position operators p and q are the self-adjoint elements satisfying the equations
1 1
x= (q + ip) and x∗ = (q − ip).
2 2
Then pq − qp = −2i, illustrating the fact that the self-adjoint elements in a non-commutative ∗-algebra need not form a
subalgebra.

2.2. Conjugate bimodules

This mixing of differential geometry and ∗-structures must be done with a little care in non-commutative geometry—the
reader is referred to [23] for some of the details, most of which will not be necessary in this paper.
Let A be a ∗-algebra and E an A-bimodule. We define the conjugate bimodule E by declaring that
(1) E = E as an abelian group;
(2) we write e for an element e ∈ E when we consider it as an element of E;
(3) the bimodule operations for E are a.e = e.a∗ and e.a = a∗ .e.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 11

If θ : E → F is any map we define θ : E → F by θ (e) := θ (e).

If θ : E → F is a homomorphism of A-bimodules, so is θ . In this way the bar operation E → E is a functor from the
category A MA of A-bimodules to itself.
We make A an associative algebra by defining the multiplication āb̄ := ba. As an R-algebra, A is isomorphic to Aop via the
map a → a. We make A a C-algebra through the algebra homomorphism C → A, λ → λ∗ . We now define

⋆ : A → A, a → a∗ . (2.1)
Then ⋆ is a homomorphism of C-algebras and an isomorphism in A MA .

2.3. The universal differential calculus

We adopt the standard notation and terminology for non-commutative differential calculus. The reader can find more
details in [24, Chapter 8], [25, Chapter 12], and [13].
Let A be an arbitrary C-algebra. Let Ωuniv
1
A denote the kernel of the multiplication map µ : A ⊗ A → A. Following Cuntz
and Quillen [26], we define the universal differential graded algebra over A to be the tensor algebra
Ωuniv
•
A = TA (Ωuniv
1
A)
endowed with the unique degree one superderivation such that
d(a) := 1 ⊗ a − a ⊗ 1
for a ∈ A.

Definition 2.2. A differential calculus or differential structure on A is a differential graded algebra (Ω • A, d) that is a quotient
of (Ωuniv
•
A, d) by a differential graded ideal whose degree-zero component is zero. The cohomology of (Ω • A, d) is called the
complex-valued de Rham cohomology and denoted by HdR •
(A).

Warning: We will denote the product in Ω • A by wedge ∧ although ξ ∧ η will not usually equal (−1)|ξ | |η| η ∧ ξ when A is
not commutative.
It follows from the definition that Ω 0 A = A; that Ω n+1 A = A · d(Ω n A) for all n ≥ 1; that the multiplication map
A ⊗ d(Ω n A) → Ω n+1 A is surjective for all n ≥ 0; and
d(ξ ∧ η) = dξ ∧ η + (−1)|ξ | ξ ∧ dη.
By hypothesis, Ω • A is generated by A and Ω 1 A, and Ω n A is the C-span of
{a0 da1 ∧ · · · ∧ dan | a0 , . . . , an ∈ A}.
Let (Ω • A, d) and (Ω • B, d) be differential calculi. An algebra homomorphism φ : A → B is differentiable if it extends to a
homomorphism φ : Ω • A → Ω • B of dgas. In particular, d(φ(a)) = φ(da) for all a ∈ A.

2.4. Differential ∗-calculus

Non-commutative ∗-calculus first appears in Woronowicz’s papers [15,14] and [15, Definition 1.4] has been extended to
higher degree forms as follows.

Definition 2.3 ([25, p. 462]). A differential calculus (Ω • A, d) on a ∗-algebra A is compatible with the star operation on A if
the star operation on Ω 0 A = A extends to an involution ξ → ξ ∗ on Ω • A that preserves the grading and has the property
that (dξ )∗ = d(ξ ∗ ) and (ξ ∧ η)∗ = (−1)|η||ξ | η∗ ∧ ξ ∗ for all homogeneous η, ξ ∈ Ω A.
When these conditions hold we call (Ω • A, d, ∗) a differential ∗-calculus on A.

Remark. The definition implies (aξ b)∗ = b∗ ξ ∗ a∗ for all a, b ∈ A and ξ ∈ Ω • A.

We can rephrase the definition of a differential ∗-calculus using the notion of a conjugate bimodule in Section 2.2 and
the fact that the product
ξ ∧ η := (−1)|ξ | |η| η ∧ ξ
makes (Ω • A, d) a differential graded algebra.

Proposition 2.4. Suppose (Ω • A, d, ∗) is a differential ∗-calculus on A.

ai · dbi = 0 in Ω 1 A, then i db∗i · a∗i = 0.
 
1. If
2. The map ⋆ : Ω • A → Ω • A, ⋆ξ := ξ ∗ , is an A-bimodule homomorphism and a homomorphism of differential graded algebras.
Conversely, if ξ → ξ + on Ω • A is an involution that extends the ∗ on A and preserves the grading and has the property that
(dξ )+ = d(ξ + ), then (Ω • A, d, + ) is a differential ∗-calculus on A.
12 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

Proof. We leave this to the reader. 

The map ⋆ : Ω A → Ω • A is a morphism in the category of A-bimodules. The map Ω • A → Ω • A, ξ → ξ ∗ is not.

•

Our ⋆ has nothing to do with the Hodge star operation in Riemannian geometry, which in general changes the degree
of the form. Our ⋆ is a notational device that depends only on the fact that A is a ∗-algebra. In the context of real classical
geometry ⋆ is the identity.

Proposition 2.5. Let F = C⟨x1 , . . . , xn , x∗1 , . . . , x∗n ⟩ be the free ∗-algebra. Then there is a differential ∗-calculus (Ωuniv
•
F , d, ∗)
defined by (da)∗ = d(a∗ ) for a ∈ F and

(a0 da1 · · · dan )∗ := εn (da∗n ) · · · (da∗1 )a∗0 (2.2)

1
n(n−1)
where εn = (−1) 2 ; i.e., εn = 1 if n is 0 or 1 modulo 4, and −1 otherwise.

Proof. If ξ = a0 da1 · · · dan , then

(dξ )∗ = (da0 da1 · · · dan )∗

= εn+1 da∗n · · · da∗1 da∗0
= εn (−1)n da∗n · · · da∗1 da∗0
= εn d(da∗n · · · da∗1 a∗0 )
= d(ξ ∗ ).

Every element in Ωuniv

n
F is a sum of terms of the form a0 da1 · · · dan so (dξ )∗ = d(ξ ∗ ) for all ξ ∈ Ωuniv
n
F.
If a ∈ F and ξ ∈ Ωuniv F it follows at once that (aξ ) = ξ a = aξ so ⋆ : Ω F → Ω F is a homomorphism of left
• ∗ ∗ ∗ ∗ • •
F -modules.
A straightforward calculation shows that
n −1

da0 da1 · · · dan−1 an = (−1)n a0 da1 · · · dan + (−1)n−i+1 da0 · · · d(ai ai+1 ) · · · dan .
i =0

It follows that (da0 da1 · · · dan−1 an ) is equal to

∗

n−1

(−1)n εn da∗n · · · da∗1 a∗0 + (−1)n−i+1 εn da∗n · · · d(ai ai+1 )∗ · · · da∗0
i=0
 
n −1

= (−1) εn (−1)
n n ∗
an da∗n−1 ∗
· · · da0 + (−1)n−i+1 da∗n · · · d(a∗n−i a∗n−i−1 ) · · · da∗0
i =0
n −1

+ (−1)n−i+1 εn da∗n · · · d(a∗i+1 a∗i ) · · · da∗0 .
i=0

The two sums cancel so we obtain

(da0 da1 · · · dan−1 an )∗ = a∗n da∗n−1 · · · da∗0

which shows that ⋆ : Ω • F → Ω • F is a homomorphism of right F -modules. 

We note that εn is the sign of the permutation ( · · · ) and (2.2) therefore agrees with the formula in [14, (3.3)].
1 2 n
n n −1 1

2.5. Almost complex structures

An almost complex structure on a real manifold is an endomorphism J of its tangent bundle such that J 2 = −1. A
complex manifold, when viewed as a real manifold, has a natural almost complex structure [27, Proposition 2.6.2]. Since
we are working with (non-commutative analogues of) differential forms rather than the tangent bundle we must express
the notion of an almost complex structure in terms of differential forms.

Definition 2.6. Let (Ω • A, d, ∗) be a differential ∗-calculus on A. An almost complex structure on (Ω • A, d, ∗) is a degree zero
derivation J : Ω • A → Ω • A such that
(1) J is identically 0 on A and hence an A-bimodule endomorphism of Ω • A;
(2) J 2 = −1 on Ω 1 A; and
(3) J (ξ ∗ ) = (J ξ )∗ for ξ ∈ Ω 1 A (or, equivalently, J ⋆ = ⋆ J on Ω 1 A).
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 13

At this stage, there is no requirement on how J interacts with d. Such a requirement will appear later when we define
what it means for an almost complex structure to be integrable.
Because J 2 = −1 on Ω 1 A, there is an A-bimodule decomposition

Ω 1 A = Ω 1,0 A ⊕ Ω 0,1 A (2.3)

1,0 0 ,1
where Ω A = {ω ∈ Ω A | J ω = i ω} and Ω A = {ω ∈ Ω A | J ω = −i ω}. Condition (3) in the definition implies
1 1

(Ω 0,1 A)∗ = Ω 1,0 A.

The next result is analogous to the trivial fact that a differentiable manifold with an almost complex structure has even
dimension, or, equivalently, the rank of its cotangent bundle is even.

Lemma 2.7. Let (Ω • A, d, ∗) be a differential ∗-calculus on A. Suppose there is a function

rank : {some A-bimodules} → Z

defined on a class of A-bimodules that is closed under finite direct sums and direct summands and contains A and all Ω n A and
has the property that rank M = rank M̄ and rank (M ⊕ N ) = rank M + rank N. If A has an almost complex structure, then
rank Ω 1 A is even.

Proof. The map ⋆ : Ω 1,0 A → Ω 0,1 A is an isomorphism of A-bimodules so Ω 1,0 A and Ω 0,1 A have the same rank. Therefore
rank Ω 1 A = 2 rank Ω 0,1 A. 

Remark 2.8. The hypothesis in Lemma 2.7 is mild. Sometimes there will be a finite set that is a basis for Ω 1 A as both a left
and right A-module and when Ω 1 A is not free it is often a finitely generated projective A-module on the left and on the right,
and becomes free after a suitable localization.
It would be desirable to consider examples having the following additional properties: A is a finitely presented C-algebra,
a domain, and noetherian, or coherent; Ω 1 A is a finitely generated projective A-module of rank 2ron both
  the left and the
right; Ω p,q A, which we define in Lemma 2.10, is a finitely generated projective A-module of rank
r r
p
on both the left
q
and the right; there is a left A-module isomorphism ℓ : Ω A → A and a right A-module isomorphism : Ω 2r A → A.
 2r
r

Proposition 2.9. Let F = C⟨x1 , . . . , xn , x∗1 , . . . , x∗n ⟩ be the free ∗-algebra endowed with the differential ∗-calculus (Ωuniv
•
F , d, ∗)
in Proposition 2.5. Then there is a unique almost complex structure on (Ωuniv F , d, ∗) such that
•

J (dxk ) = −dx∗k and J (dx∗k ) = dxk

for k = 1, . . . , n

Proof. Since Ωuniv

1
F is the free F -bimodule with basis {dxk , dx∗k | 1 ≤ k ≤ n} any action of J on the basis extends in a unique
way to an F -bimodule automorphism of Ωuniv 1
F . Then, because Ωuniv•
F is the tensor algebra over F on Ωuniv
1
F , the action of J
on Ωuniv F extends in a unique way to an action of J as a derivation of Ωuniv F . 
1 •

Warning. When n > 1 the map J : Ω n A → Ω n A does not satisfy the equation J 2 = −1. For example, because J is a derivation
J (da ∧ db) = J (da) ∧ db + da ∧ J (db) for all a, b ∈ A and therefore

J 2 (da ∧ db) = J 2 (da) ∧ db + 2J (da) ∧ J (db) + da ∧ J 2 (db)

= 2 (J (da) ∧ J (db) − da ∧ db) .
More succinctly,

J 2 = 2(J ∧ J − 1) : Ω 2 A → Ω 2 A. (2.4)

2.5.1. (p, q)-forms

Let C[h] denote the commutative polynomial ring viewed as the enveloping algebra of the 1-dimensional Lie algebra.
Thus, if V and W are C[h]-modules, V ⊗C W is given the C[h]-module structure h · (v ⊗ w) = v ⊗ hw + hv ⊗ w .1 In
particular, the tensor algebra T (V ) becomes a C[h]-module and h acts on T (V ) as a derivation.

1 Equivalently, if ∆ : C[h] → C[h] ⊗ C[h] is the C-algebra homomorphism defined by ∆(h) := 1 ⊗ h + h ⊗ 1, then V ⊗ W is naturally a C[h] ⊗ C[h]-
module and is made into a C[h]-module via ∆. Alternatively, V and W can be viewed as representations of the trivial Lie algebra Ch and V ⊗r ⊗ W ⊗n is
then made into a representation of Ch in the standard way.
14 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

Lemma 2.10. Let V be a C[h]-module annihilated by h2 + 1. Let n ≥ 1. Suppose Ω n is a C[h]-module quotient of V ⊗n . Then
there is a C[h]-module decomposition

Ω p,q

Ωn =
p+q=n

where

Ω p,q = {ξ ∈ Ω ⊗n | h · ξ = (p − q)iξ }.

n=0 Ω is a quotient algebra of T (V ) by a C[h]-stable graded ideal, then the multiplication in Ω is

∞
Furthermore, if Ω • = n •
p,q p ,q
′ ′ p+p′ ,q+q′ 0,•
such that Ω ⊗ Ω →Ω . In particular, Ω and Ω are subalgebras of Ω .
•,0 •

Proof. By hypothesis, there is a C[h]-module direct sum decomposition V = V 1,0 ⊕ V 0,1 where h acts as multiplication by
i on V 1,0 and as multiplication by −i on V 0,1 .
If W and W ′ are C[h]-modules annihilated by h − λ and h − λ′ respectively, then W ⊗ W ′ is annihilated by h − (λ + λ′ ).
Applying this observation inductively to (V 1,0 ⊕ V 0,1 )⊗n it follows that

V p,q

V ⊗n =
p+q=n

p,q
where V = {ξ ∈ V ⊗n | h · ξ = (p − q) iξ }. This proves the lemma for Ω n = V ⊗n . Since V ⊗n is a semisimple C[h]-module,
so is every quotient of it, and the set of h-eigenvalues for a quotient of V ⊗n is a subset of the set of h-eigenvalues for V ⊗n .
This completes the proof of the first part of the lemma. The second part involving Ω • is equally easy. 

2.5.2
Let J be an almost complex structure on (Ω • A, d, ∗).
By hypothesis the multiplication Ω 1 A ⊗A Ω n A → Ω n+1 A is surjective for all n ≥ 1. Therefore C ⊕ Ω ≥1 A is a quotient of
the tensor algebra TC (Ω 1 A) where the tensor is taken over C.
We may apply Lemma 2.10 with V = Ω 1 A and h acting as J does because J 2 ξ = −ξ for all ξ ∈ Ω 1 A. Because TC (Ω 1 A) is
generated by Ω 1 A as a C-algebra there is a unique extension of h to a derivation on TC (Ω 1 A). Since J is a derivation of Ω • A
the action of J on Ω n A is induced by the action of h on (Ω 1 A)⊗n .
For all p, q ≥ 0 we define
Ω p,q A := {ξ ∈ Ω p+q A | J ξ = (p − q)i ξ }.
Elements in Ω p,q A are called (p, q)-forms. Because J is a derivation on Ω • A it follows from Lemma 2.10 that

Ω p,q A

Ω nA =
p+q=n

for all n. Because J vanishes on A, it is a homomorphism of A-bimodules and therefore each Ω p,q A is an A-bimodule. By the
last part of Lemma 2.10,

Ω p,q A ∧ Ω p ,q A ⊂ Ω p+p ,q+q A

′ ′ ′ ′
(2.5)
0,•
and Ω A and Ω •,0
A are subalgebras.

2.5.3. The ∂ and ∂ operators

For all pairs of non-negative integers (p, q), let
π p,q : Ω p+q A → Ω p,q A
be the projections associated to the direct sum decomposition Ω n A = ⊕p+q=n Ω p,q A.

Definition 2.11. Let

∂ : Ω p,q A → Ω p+1,q A
be the composition π p+1,q d : Ω p,q A → Ω p+1,q A. Let

∂ : Ω p,q A → Ω p,q+1 A
be the composition π p,q+1 d : Ω p,q A → Ω p,q+1 A.

Because Ω 1 A is generated as a left A-module by {da | a ∈ A}, Ω 1,0 A is generated as a left A-module by {∂ a | a ∈ A};
likewise, Ω 0,1 A is generated as a left A-module by {∂ a | a ∈ A}.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 15

Proposition 2.12. Let J be an almost complex structure on (Ω • A, d, ∗). Then for all ξ ∈ Ω • A and for all p, q:

J (ξ ∗ ) = (J ξ )∗ (2.6)
p,q q ,p
(Ω A) = Ω
∗
A. (2.7)

Proof. Certainly (2.6) holds when ξ ∈ A ⊕ Ω 1 A. We now argue by induction on degree. Suppose (2.6) holds when ξ ∈ Ω n A.
Every n + 1 form is a sum of elements ξ ∧ η where ξ ∈ Ω n A and η ∈ Ω 1 A. Because J is a derivation on Ω • A,
(J (ξ ∧ η))∗ = (J ξ ∧ η + ξ ∧ J η)∗
= (−1)n η∗ ∧ (J ξ )∗ + (J η)∗ ∧ ξ ∗
 

= (−1)n η∗ ∧ J (ξ ∗ ) + J (η∗ ) ∧ ξ ∗
 

= (−1)n J (η∗ ∧ ξ ∗ )
= J ((ξ ∧ η)∗ ).
Hence (2.6) holds when ξ ∈ Ω n+1 A, and therefore holds for all ξ ∈ Ω • A. Finally, a simple calculation shows that (2.7)
follows from (2.6). 

3. Integrable complex structures

3.1. The Newlander–Nirenberg integrability condition

By definition, an almost complex structure J on a smooth manifold X is a vector bundle isomorphism J : TX → TX

such that J 2 = −1. Such a J leads to a decomposition T 1,0 X ⊕ T 0,1 X of the complexified tangent bundle TX ⊗ C into the
±i-eigenspaces for J. A complex manifold has a canonical almost complex structure and one says that (X , J ) is integrable if
there is a complex structure on X such that J is the canonical almost complex structure.
The Newlander–Nirenberg theorem says J is integrable if and only if
[T 0,1 X , T 0,1 X ] ⊂ T 0,1 X .
In terms of differential forms, the Newlander–Nirenberg theorem says J is integrable if and only if
1 ,0
d ΩX ⊂ ΩX2,0 ⊕ ΩX1,1
(see, e.g., [27, Proposition 2.6.15] or [28, p. 54]). Furthermore, J is integrable if and only if dω = ∂ω + ∂ω for all ω ∈ Ω •
2
[27, Proposition 2.6.15] if and only if ∂ = 0 on A [27, Corollary 2.6.18].

Definition 3.1. An almost complex structure J on (Ω • A, d, ∗) is integrable if any of the equivalent conditions in Lemma 3.2
hold (cf., [27, Proposition 2.6.15 and Definition 2.6.16]).

Lemma 3.2. Let J be an almost complex structure on (Ω • A, d, ∗). The following conditions are equivalent:
2
(1) ∂ = 0 as an operator A → Ω 2 A;
(2) ∂ 2 = 0 as an operator A → Ω 2 A;
(3) d = ∂ + ∂ as operators Ω 1 A → Ω 2 A;
(4) dΩ 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A;
(5) dΩ 0,1 A ⊂ Ω 1,1 A ⊕ Ω 0,2 A.
Proof. If a ∈ A, then d2 a = 0 so
π 2,0 d(∂ a + ∂ a) = π 1,1 d(∂ a + ∂ a) = π 0,2 d(∂ a + ∂ a) = 0.
In other words,
2
∂ 2 a + π 2,0 d∂ a = ∂∂ a + ∂∂ a = π 0,2 d∂ a + ∂ a = 0. (3.8)
2
(1) ⇒ (4) Suppose (1) holds. Since ∂ a = 0, (3.8) implies π 0,2 d∂ a = 0, i.e., d(∂ a) ∈ Ω 2,0 A ⊕ Ω 1,1 A. If b ∈ A, then
d(b · ∂ a) = db ∧ ∂ a + b · d(∂ a) ∈ Ω 1 A ∧ Ω 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A.
Since Ω 1,0 A is generated as a left A-module by {b.∂ a | a, b ∈ A} the last calculation shows that dΩ 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A.
2
(4) ⇒ (1) If dΩ 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A, then π 0,2 d(∂ a) = 0 for all a ∈ A so ∂ a = 0 by (3.8).
(2) ⇔ (5) This is proved by the same  kind of∗argument as
∗was used to
∗prove the equivalence of (1) and (4).
(4) ⇔ (5) Suppose (4) holds. Then dΩ 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A . Condition (5) now follows by applying (2.6) and

(2.7). The implication (5) ⇒ (4) is proved in a similar way.
16 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

(3) ⇒ (4) If d = ∂ + ∂ on Ω 1 A, then

dΩ 1,0 A ⊂ ∂ Ω 1 A + ∂ Ω 1,0 A ⊂ Ω 2,0 A + Ω 1,1 A.
(4) ⇒ (3) Let ω ∈ Ω 1,0 A and η ∈ Ω 0,1 A. Then
dω ∈ dΩ 1,0 A ⊂ Ω 2,0 A + Ω 1,1 A
so dω = ∂ω + ∂ω. Since (4) holds so does (5), and (5) implies that dη ∈ Ω 1,1 A + Ω 0,2 A so dη = ∂η + ∂η. Thus
d(ω + η) = (∂ + ∂)(ω + η). 

Lemma 3.3. Let J be almost complex structure on (Ω • A, d, ∗). The following conditions are equivalent:
(1) J is integrable;
(2) (1 − J ∧ J )dJ = Jd : Ω 1 A → Ω 2 A;
(3) J 2 dJ = −2Jd : Ω 1 A → Ω 2 A;
(4) J 2 d = 2JdJ : Ω 1 A → Ω 2 A;
(5) JdJd = 0 : A → Ω 2 A.
Proof. (1) ⇒ (3) and (4). Because Ω 1,0 A = (J + i)Ω 1 A and because Ω 2,0 A, Ω 1,1 A, and Ω 0,2 A are the 2i-, 0-, and −2i-
eigenspaces for J acting on Ω 2 A, Lemma 3.2(4) implies that (J − 2i)Jd(J + i) = 0 on Ω 1 A. Lemma 3.2(5) implies that
(J + 2i)Jd(J − i) = 0 on Ω 1 A. But
(J − 2i)Jd(J + i) = J 2 dJ + iJ 2 d − 2iJdJ + 2Jd = 0
and
(J + 2i)Jd(J − i) = J 2 dJ − iJ 2 d + 2iJdJ + 2Jd = 0.
It is now clear that (3) and (4) hold.
(2) ⇔ (3) We already showed (2.4) that as endomorphisms of Ω 2 A, J 2 = 2(J ∧ J − 1). Hence

1
(1 − J ∧ J )dJ − Jd = − (J 2 dJ + 2Jd). (3.9)
2
The equivalence of (2) and (3) now follows.
(3) ⇔ (4) Since J : Ω 1 A → Ω 1 A is an invertible map, (3) holds if and only if (4) holds (multiply on the right by J).
(4) ⇒ (1) Assume (4) holds. Then (3) holds too.
A 2-form ξ belongs to Ω 2,0 A ⊕ Ω 1,1 A if and only if (J + 2i)J ξ = 0. Since Ω 1,0 A = (J + i)Ω 1 A it follows that Lemma 3.2(4)
holds if and only if (J + 2i)Jd(J + i) vanishes identically on Ω 1 A. But
(J + 2i)Jd(J − i) = J 2 dJ + 2Jd − i J 2 d − 2JdJ
   

and this is zero by (3) and (4).

(4) ⇒ (5) Since J 2 d = 2JdJ, 0 = J 2 dd = 2JdJd.
(5) ⇒ (4) Let a, b ∈ A. Then
J 2 d(adb) = 2(J ∧ J − 1)(da ∧ db) = 2(Jda ∧ Jdb − da ∧ db).
On the other hand 2JdJ (adb) = 2J (da ∧ Jdb + adJdb) and J (adJdb) = aJdJdb = 0, so
2JdJ (adb) = 2J (da ∧ Jdb) = 2(Jda ∧ Jdb + da ∧ J 2 db).
But da ∧ J 2 db = −da ∧ db, so J 2 d = 2JdJ : Ω 1 A → Ω 2 A. 

Proposition 3.4. The operators

(1 − J ∧ J )dJ − Jd, J 2 dJ + 2Jd, J 2 d − 2JdJ
are left A-module homomorphisms Ω 1 A → Ω 2 A. Hence J is integrable if any one of these homomorphisms vanishes on a set of
left A-module generators for Ω 1 A. (Similarly changing right for left everywhere.)
Proof. Let us show that the first of these three operators is a left A-module homomorphism. Let aξ ∈ Ω 1 A. Then
((1 − J ∧ J )dJ − Jd) (aξ ) = (1 − J ∧ J )d(aJ ξ ) − J (da ∧ ξ + adξ )
= (1 − J ∧ J )(da ∧ J ξ + adJ ξ ) − J (da ∧ ξ + adξ )
= da ∧ J ξ + Jda ∧ ξ + a(1 − J ∧ J )dJ ξ − J (da ∧ ξ + adξ )
= a(1 − J ∧ J )dJ ξ − aJdξ .
Hence (1 − J ∧ J )dJ − Jd is a left A-module homomorphism. It then follows from (3.9) that the second operator is also
a left A-module homomorphism. The third operator is obtained from the second one by composing on the right with the
A-module homomorphism J so it too is a left A-module homomorphism. (The right module proof is similar.) 
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 17

Proposition 3.5. Suppose (Ω • A, d, ∗) is a differential ∗-calculus with an integrable almost complex structure J. Then for all p
and q,

d Ω p,q A ⊂ Ω p+1,q A ⊕ Ω p,q+1 A.

 

Proof. We argue by induction on p + q. The case p + q = 1 is true by the definition of integrability. Suppose n ≥ 2 and that
the result is true for all p′ + q′ = n − 1.
Let p + q = n. Then

Ω n A = Ω 1 A ∧ Ω n−1 A = Ω 1,0 A ⊕ Ω 0,1 A ∧ Ω n−1 A.

 

It therefore follows from (2.5) that

Ω p,q A = Ω 1,0 A ∧ Ω p−1,q A + Ω 0,1 A ∧ Ω p,q−1 A .

However, dΩ 1,0 A ⊂ Ω 2,0 A ⊕ Ω 1,1 A and dΩ 0,1 A ⊂ Ω 1,1 A ⊕ Ω 0,2 A so, in conjunction with the induction hypothesis applied
to dΩ p−1,q A and dΩ p,q−1 A, we obtain

dΩ p,q = dΩ 1,0 A ∧ Ω p−1,q A + Ω 1,0 A ∧ dΩ p−1,q A + dΩ 0,1 A ∧ Ω p,q−1 A + Ω 0,1 A ∧ dΩ p,q−1 A

⊂ Ω p+1,q A + Ω p,q+1 A,
as claimed. 

Proposition 3.6. Suppose (Ω • A, d, ∗) is a differential ∗-calculus with an integrable almost complex structure J. Then d = ∂ + ∂¯
and

∂ 2 = 0, ∂ ∂¯ + ∂∂
¯ = 0, ∂¯ 2 = 0. (3.10)

Proof. By Proposition 3.5, d = ∂ + ∂¯ . We have

0 = d2 Ω p,q A
2
⊂ ∂ 2 Ω p,q A + ∂∂ + ∂∂ Ω p,q A + ∂ Ω p,q A
 

⊂ Ω p+2,q A ⊕ Ω p+1,q+1 A ⊕ Ω p,q+2 A.

The equalities in (3.10) follow from the direct sum decomposition of Ω n A. 

Proposition 3.7. Suppose (Ω • A, d, ∗) is a differential ∗-calculus with an integrable almost complex structure J. Both ∂¯ and ∂
are superderivations.

Proof. Let π r ,s : Ω r +s A → Ω r ,s A denote the obvious projection (and its restrictions) for the direct sum decomposition of
Ω r +s A onto the direct sum of its subspaces of (p, q)-forms.
Let ξ ∈ Ω p,q A and η ∈ Ω p ,q A. Then
′ ′

∂(ξ ∧ η) = π p+p +1,q+q d(ξ ∧ η)

′ ′

= π p+p +1,q+q dξ ∧ η + (−1)|ξ | ξ ∧ dη

′ ′  
 ′ 
= π p+1,q dξ ∧ η + (−1)|ξ | ξ ∧ π p +1,q dη
  ′

= ∂ξ ∧ η + (−1)|ξ | ξ ∧ ∂η
as required. The proof for ∂ is essentially the same. 

Proposition 3.8. For all ξ ∈ Ω ∗ A,

∂(ξ )∗ = ∂(ξ ∗ ) and ∂(ξ )∗ = ∂(ξ ∗ ).

Proof. Since it suffices to prove the proposition when ξ ∈ Ω p,q A we make that assumption. Since d(ξ ∗ ) = (dξ )∗ ,

(∂ξ )∗ + (∂ξ )∗ = ∂(ξ ∗ ) + ∂(ξ ∗ ).

But ξ ∗ ∈Ω q,p A by Proposition 2.12, so = ∂(ξ∗ ) ∈ Ω q+1,p A and ∂(ξ ∗ ) ∈ Ω q,p+1 A. The result now follows from the fact that
∗ ∗
(∂ξ )∗ ∈ Ω p+1,q A = Ω q,p+1 A and (∂ξ )∗ ∈ Ω p,q+1 A = Ω q+1,p A. 
18 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

3.2. Holomorphic forms and holomorphic elements of A

Let J be an integrable almost complex structure on a differential ∗-calculus (Ω • A, d, ∗).

An element f ∈ A is holomorphic if ∂ f = 0. We define

Ahol := {f ∈ A | ∂ f = 0} (3.11)
and
Ωhol A := {ω ∈ Ω p,0 A | ∂ω = 0}.
p
(3.12)
p
Elements in Ahol are called holomorphic (functions) and elements in Ωhol A
are called holomorphic p-forms.
Elements in Ahol play the role of ‘‘holomorphic functions on a non-commutative complex variety’’ and elements in A play
the role of ‘‘C-valued differentiable functions on an underlying real variety’’.
As some reassurance, we check the analogue of the fact that the only R-valued holomorphic functions on a connected
complex manifold are the constants. The analogue of an R-valued function is a self-adjoint element of A. Suppose then that
f = f ∗ and ∂ f = 0. Then df ∈ Ω 1,0 A so (df )∗ ∈ Ω 0,1 A. However, (df )∗ = d(f ∗ ) = df ∈ Ω 1,0 A so df ∈ Ω 0,1 A ∩ Ω 1,0 A. Hence
df = 0. The analogue of connectedness for a differential calculus is that d : A → Ω 1 A vanishes only on C; we therefore
deduce that f ∈ C, i.e., f is constant.
p
Proposition 3.9. Ahol is a C-subalgebra of A and every Ωhol A is a bimodule over Ahol .
p
Proof. Since ∂ is a C-linear derivation, Ahol is a C-subalgebra of A. If f ∈ Ahol and ω ∈ Ωhol A, then ∂(f ω) = ∂ f · ω + f ∂ω = 0
p p
so f ω ∈ Ωhol A. Similarly, ωf ∈ Ωhol A. 
The holomorphic de Rham complex for A is the complex
∂ ∂ ∂
0 −→ Ahol −→ Ωhol
1
A −→ Ωhol
2
A −→ · · · .

3.2.1
Omitting some crucial definitions, consider a non-commutative projective algebraic variety X := Projnc R. For F ∈
p
QcohX , one defines H q (X , F ) := Extq (OX , F ). There is no general definition of objects ΩX in QcohX though in some cases
p
where R is a Koszul algebra with properties like a polynomial ring there are reasonable candidates for such ΩX defined in
terms of the Koszul resolution of the trivial R-module. It would be very interesting to examine whether there are situations
p,q p
in which H (A) is isomorphic to H q (X , ΩX ) or the qth cohomology group of the holomorphic de Rham complex for an
∂
appropriate A endowed with a suitable integrable almost complex structure on (Ω • A, d, ∗).

4. Holomorphic modules

4.1. The Koszul–Malgrange theorem

Throughout this section J is an integrable almost complex structure on a differential ∗-calculus (Ω • A, d, ∗). In this section
we will write Ω p,q A for the A-bimodules Ω p,q A.
Holomorphic A-modules are defined in Definition 4.3. The justification for our definition is the Koszul–Malgrange
theorem as we now explain.
Let E be a complex vector bundle on a complex manifold M. A holomorphic structure on E is a complex manifold structure
on the total space of E such that the transition functions are holomorphic. Although there is no natural exterior derivative
on a general complex vector bundle a holomorphic bundle E does have a naturally defined C-linear ∂ -operator
∂ : Ω p,q A ⊗A E → Ω p,q+1 A ⊗A E
2
satisfying ∂ = 0 and the Leibniz rule ∂(f · α) = ∂(f ) ∧ α + f ∂(α) (see, e.g., [29, p. 70] and [27, Lemma 2.6.23]).
Let E be a complex vector bundle with a connection ∇ : E → Ω 1 A ⊗A E. We can write ∇ = ∇ 1,0 + ∇ 0,1 where
∇ : E → Ω p,q A ⊗A E. The Koszul–Malgrange theorem, which is analogous to the Newlander–Nirenberg criterion, says
p,q

that if the composition

∇ 0,1 / Ω 0,1 A ⊗A E ∇ 0,2 / Ω 0,2 A ⊗A E
E

is zero, in which case ∇ 0,1 is said to be integrable, then there is a unique holomorphic structure on E with respect to which
the natural ∂ -operator is ∇ 0,1 .
A given complex vector bundle E might have many holomorphic structures each of which determines a natural ∂ -operator
which, in turn, determines the holomorphic structure on E. Moreover, every C-linear operator ∇ : E → Ω 1,0 A ⊗A E
satisfying the conditions in Definition 4.3 is the ∂ -operator for a unique holomorphic structure on E (see, e.g., [27, Theorem
2.6.26]).
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 19

4.2. ∂ -operators on modules

Definition 4.1. Let (Ω • A, d, ∗) be a differential ∗-calculus with an integrable almost complex structure J. A ∂ -operator on
a left A module E is a C-linear map
∇ : E → Ω 0,1 A ⊗A E
such that
∇(a · e) = ∂ a ⊗ e + a · ∇ e
for all a ∈ A and e ∈ E. For q ≥ 1 we define
∇ : Ω 0,q A ⊗A E → Ω 0,q+1 A ⊗A E
by

∇(ξ ⊗ e) := ∂ξ
¯ ⊗ e + (−1)q ξ ∧ ∇(e). (4.13)
The holomorphic curvature of ∇ is defined to be
2
∇ : E → Ω 0 ,2 A ⊗ E .

It is easy to check that

(1) ∇ : Ω 0,q A ⊗A E → Ω 0,q+1 A ⊗A E is a homomorphism of left Ahol -modules and
2
(2) ∇ is a left A-module map.

Lemma 4.2. Let (E , ∇) be a module with a ∂ -operator. Then

2
∇ : Ω 0,∗ A ⊗A E −→ Ω 0,∗+2 A ⊗A E
is a homomorphism of left Ω 0,∗ A-modules. In fact, for ξ ∈ Ω 0,∗ A and e ∈ E,
2 2
∇ (ξ ⊗ e) = ξ ∧ ∇ (e).
2 2
Proof. An easy computation shows that ∇ (ξ ⊗ e) = ξ ∧ ∇ e. Hence for every η ∈ Ω 0,∗ A,
2 2 2
∇ (η ∧ ξ ⊗ e) = η ∧ ξ ∧ ∇ e = η ∧ ∇ (ξ ⊗ e).
2
In other words, ∇ commutes with ‘‘left multiplication by η’’ so is a homomorphism of left Ω 0,∗ A-modules. 

4.3. Holomorphic modules

Definition 4.3. Let (Ω • A, d, ∗) be a differential ∗-calculus with an integrable almost complex structure J. A holomorphic
structure on a left A-module E is a ∂ -operator ∇ on E whose holomorphic curvature vanishes. We then call (E , ∇) a
holomorphic A-module and an element e ∈ E such that ∇ e = 0 is said to be holomorphic.

Definition 4.4. The holomorphic modules (E , ∇) are the objects in a category Hol(A). A morphism φ : (E1 , ∇ 1 ) → (E2 , ∇ 2 )
in Hol(A) is an A-module homomorphism φ : E1 → E2 such that ∇ 2 φ = (id ⊗ φ)∇ 1 .

Proposition 4.5. If Ω 0,1 A is a flat right A-module, then Hol(A) is an abelian category.
Proof. Let f : (E1 , ∇ 1 ) → (E2 , ∇ 2 ) be a morphism in Hol(A). Let K be the kernel and C the cokernel of f in the category of
right A-modules. The hypothesis that Ω 0,1 A is flat implies the second row in the diagram

0 /K / E1 f
/ E2 /C /0

∇1 ∇2
 
0 / Ω 0,1 A ⊗A K / Ω 0,1 A ⊗A E1 / Ω 0,1 A ⊗A E2 / Ω 0,1 A ⊗A C /0
1⊗f

is exact. There are unique maps ∇ K : K → Ω 0,1 A ⊗A K and ∇ C : C → Ω 0,1 A ⊗A C making the diagram commute. Since ∇ K
is, in effect, the restriction of ∇ 1 it is a ∂ -operator. Likewise, ∇ C is a ∂ -operator because it is induced by the ∂ -operator ∇ 2 .
It is easy to check that (K , ∇ K ) and (C , ∇ C ) are a kernel and cokernel for f in Hol(A). 
20 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

If (E , ∇) is a holomorphic module Lemma 4.2 implies that

∇ ∇ ∇
0 −→ E −→ Ω 0,1 A ⊗A E −→ Ω 0,2 A ⊗A E −→ · · ·

is a complex. The cohomology groups of the complex Ω 0,• A ⊗A E will be denoted by

H • (E , ∇).

If φ : (E1 , ∇ 1 ) → (E2 , ∇ 2 ) is a homomorphism between holomorphic modules, then id ⊗ φ is a map of cochain complexes
so induces a map in cohomology

φ • : H • (E1 , ∇ 1 ) → H • (E2 , ∇ 2 ).

Proposition 4.6. Suppose every Ω n A is flat as a right A-module. Then every exact sequence

φ ψ
0 / (E , ∇ E ) / (F , ∇ F ) / (G, ∇ G ) /0

of holomorphic modules gives rise to a long exact sequence

φ∗ ψ∗
0 / H 0 (E , ∇ E ) / H 0 (F , ∇ F ) / H 0 (G, ∇ G ) δ /

φ∗ ψ∗
H 1 (E , ∇ E ) / H 1 (F , ∇ F ) / ···

of left Ahol -modules.

Proof. A direct summand of a flat module is flat so each Ω 0,n A is flat as a right A-module. It follows that

0 / Ω 0,• A ⊗A E / Ω 0,• A ⊗A F / Ω 0,• A ⊗A G /0

is an exact sequence of cochain complexes. The existence of the long exact sequence follows in the usual way (see, e.g.,
[30, Theorem 1.3.1]). 

A Question. Let R be one of the non-commutative homogeneous coordinate rings that appears in non-commutative
projective algebraic geometry over C. Let X = Projnc R. Let M be a finitely generated graded left R-module that corresponds
to an OX -module M . Is there a relation between Ext• (OX , M ) and H • (E , ∇) for some holomorphic module (E , ∇) over some
∗-calculus (Ω • A, d, ∗) having an integrable structure J?

4.4. A connection whose (0, 2)-curvature component vanishes induces a holomorphic structure

A connection on a left A-module E is a C-linear map ∇ : E → Ω 1 A ⊗A E such that

∇(a · e) = da ⊗ e + a · ∇ e
for all a ∈ A and e ∈ E. Because Ω 1 A = Ω 1,0 A ⊕ Ω 0,1 A we write ∇ = ∇ 1,0 + ∇ 0,1 where ∇ p,q : E → Ω p,q A ⊗A E. We define
higher covariant exterior derivatives

∇ : Ω n A ⊗A E → Ω n+1 A ⊗A E by ∇(ξ ⊗ e) := dξ ⊗ e + (−1)n ξ ∧ ∇ e

for ξ ∈ Ω n A and e ∈ E. The curvature of the connection ∇ is

R := ∇ ◦ ∇ : E → Ω 2 A ⊗A E .

It can be shown that

∇ ◦ ∇ = id ∧ R : Ω n A ⊗A E → Ω n+2 A ⊗A E
so, if the curvature vanishes, there is a complex (Ω • A ⊗A E , ∇) and an associated cohomology theory [31].
Now we show how, in parallel with the Koszul–Malgrange theorem, a connection induces a holomorphic structure if the
(0, 2) part of its curvature vanishes. The vanishing of the holomorphic curvature is a weaker condition than the vanishing
of the curvature for a standard covariant derivative, as shall be explained.
As before, π p,q : Ω p+q A → Ω p,q A are the orthogonal projections.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 21

Proposition 4.7. Let (E , ∇) be a left A-module with a connection. Then the map
∇ := (π 0,1 ⊗ idE )∇ : E → Ω 0,1 A ⊗A E
is a ∂ -operator. Its holomorphic curvature in terms of the curvature ∇ 2 is given by the formula
2
∇ = (π 0,2 ⊗ idE )∇ 2 : E → Ω 0,2 A ⊗A E . (4.14)
Hence if the (0, 2) component of the curvature ∇ vanishes, then (E , ∇) is a holomorphic left A-module.
2

Proof. First we check that ∇ is a ∂ -operator, i.e., that the left Leibniz rule holds. Let a ∈ A and e ∈ E. Because π 0,1 is a left
A-module homomorphism
(π 0,1 ⊗ idE )∇(a.e) = (π 0,1 ⊗ idE ) (da ⊗ e + a · ∇(e))
= π 0,1 (da) ⊗ e + a · (π 0,1 ⊗ idE )∇(e)
= ∂¯ a ⊗ e + a · ∇(e).
Thus ∇ is a ∂ -operator.
If ω ∈ Ω p+q A we will write ωp,q for π p,q (ω), i.e., for the component of ω ∈ Ω p,q A.
Let e ∈ E and suppose that ∇ e = ω ⊗ f (or, more correctly, a sum of such terms). Then ∇ 2 (e) = ∇(ω ⊗ f ) =
dω ⊗ f − ω ∧ ∇ f so
(π 0,2 ⊗ idE )∇ 2 (e) = π 0,2 (dω) ⊗ f − (π 0,2 ⊗ idE )(ω ∧ ∇ f )
= π 0,2 (dω1,0 + dω0,1 ) ⊗ f − ω0,1 ∧ (π 0,1 ⊗ idE )(∇ f )
= ∂ω0,1 ⊗ f − ω0,1 ∧ ∇ f
= ∇(ω0,1 ⊗ f )
∇ (π 0,1 ⊗ idE )(ω ⊗ f )
 
=
= ∇ ∇(e)
so proving the formula in (4.14). The second line of the calculation used the fact that if ω, η ∈ Ω 1 A, then π 0,2 (ω ∧ η) =
ω0,1 ∧ η0,1 ; see (2.5).
The last sentence of the proposition follows at once. 

5. Dolbeault cohomology

Let (Ω • A, d, ∗) be a differential ∗-calculus with an integrable almost complex structure J.

5.1

Definition 5.1. For each integer p we call

∂¯ ∂¯
0 −→ Ω p,0 A −→ Ω p,1 A −→ Ω p,2 A −→ · · · (5.15)
the pth Dolbeault complex and call its cohomology groups
p,q
H
∂
(A)
the (p, q)-Dolbeault cohomology of A.
The q = 0 case of the classical theorem that the Dolbeault cohomology of a complex manifold X computes the cohomol-
p,0
ogy of the sheaf ΩX , i.e., H p,q (X ) ∼
p p p p
= H q (X , ΩX ), becomes a tautology here: Ωhol A is equal to H (A) because we defined Ωhol
∂
to be {ω ∈ Ω p,0 | ∂ω = 0} in (3.12).

5.2

More generally, if (E , ∇) is a holomorphic A-module we define Dolbeault complexes

∇ ∇
0 −→ Ω p,0 A ⊗A E −→ Ω p,1 A ⊗A E −→ Ω p,2 A ⊗A E −→ · · · (5.16)
by
∇(ξ ⊗ e) := ∂ξ ⊗ e + (−1)|ξ | ξ ∧ ∇ e
and Dolbeault cohomology groups
H p,q (E , ∇).
22 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

Let M be a complex manifold and ΩM its sheaf of holomorphic

p1-forms, i.e., the holomorphic sections of the cotangent
p
bundle. Then the sheaf of holomorphic p-forms on M is ΩM := ΩM . If A is the C-valued differentiable functions on M,
then
p,q p
H
∂
(A) ∼
= H q ( M , ΩM )
the qth sheaf cohomology group of the sheaf ΩM ; see e.g., [27, Section 2.6] and [28, Corollary 4.38 and Definition 2.37]. The
p,q
dimensions hp,q of H (A) are the Hodge numbers.
∂
Similarly, H p,q (E , ∇) ∼
p
= H q (M , E ⊗ ΩM ).

6. The Hodge to de Rham spectral sequence

The Hodge to de Rham spectral sequence for a complex manifold can be used to state some of the results of Hodge theory
without assuming the underlying analysis and integrals that underpin Hodge theory (see [28]).
Let (Ω • A, d, ∗) be a differential ∗-calculus with integrable almost complex structure J.
The Hodge–Frölicher, or Hodge to de Rham, spectral sequence is the spectral sequence associated to the double complex
Ω •,• A with its total differential d = ∂ +∂ . The complex-valued de Rham complex (Ω • , d)A has a decreasing filtration F p Ω • A
where
Ω r ,n−r A .

F p Ω n A :=
p≤r

Since the quotient F p Ω A/F p+1 Ω A is the pth Dolbeault complex

∂¯ ∂¯ ∂¯
Ω p,0 A −→ Ω p,1 A −→ Ω p,2 A −→ · · · (6.17)
the terms on the first page of the spectral sequence associated to this filtration are the Dolbeault cohomology groups,
p,q
E1 = H∂p,q (A),
and the differential on E1 is given by restricting ∂ : Ω p,q A → Ω p+1,q A to the kernel of ∂¯ . This spectral sequence converges
•,•

to the cohomology of the total complex (i.e. HdR•

(A)) in the sense that
p+q
p,q ∼ F p HdR (A)
E∞ = p+q
,
F p+1 HdR (A)
•
where F p HdR (A) is the image of the map H • (F p Ω A, d) → HdR
•
(A) induced by the inclusion F p Ω A → Ω A.

Proposition 6.1. Suppose the integrable almost complex structure on A satisfies the following condition:
for all p, q ∈ N, the map ∧ : Ω 0,q A ⊗A Ω p,0 A → Ω p,q A
is an isomorphism with inverse Θ p,q .
Then each Ω p,0 A is a holomorphic left A-module with respect to the ∂ -operator
∇ := Θ p,1 ∂¯ : Ω p,0 A → Ω 0,1 A ⊗A Ω p,0 A.
Furthermore, the terms on the first page of the Frölicher, or Hodge to de Rham, spectral sequence are
p,q
E1 = H∂p,q (A) ∼
= H q (Ω p,0 A, ∇)
and this spectral sequence converges to the de Rham cohomology of A.2
2
Proof. We need to check that the holomorphic curvature vanishes. By (4.13), the holomorphic curvature ∇ is the
composition

∂⊗
Ω p,0 A
∇ / Ω 0,1 A ⊗A Ω p,0 A ¯ id−id∧∇
/ Ω 0,2 A ⊗A Ω p,0 A .
2
As ∧ : Ω 0,2 A ⊗A Ω p,0 A → Ω p,2 A is an isomorphism, to show ∇ = 0 it suffices to show that the

∂⊗
Ω p,0 A
∇ / Ω 0,1 A ⊗A Ω p,0 A ¯ id−id∧∇
/ Ω 0,2 A ⊗A Ω p,0 A ∧ / Ω p,2 A (6.18)

vanishes.

2 Classically, if the conditions for Hodge theory are satisfied, this spectral sequence has all derivatives zero, and thus converges at the first page.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 23

The associativity of the multiplication ∧ implies that the diagram

Ω 0,1 A ⊗A Ω 0,1 A ⊗A Ω p,0 A

∧⊗id
/ Ω 0,2 A ⊗A Ω p,0 A

id⊗∧ ∧
 
Ω 0,1 A ⊗A Ω p,1 A / Ω p,2 A
∧

commutes. By hypothesis, the vertical maps are isomorphisms with inverses id ⊗ Θ p,1 and Θ p,2 so the rectangle in the
diagram

Ω 0,1 A ⊗A ΩOp,0 A
id⊗∂
/ Ω 0,1 A ⊗A Ω p,1 A ∧ / Ω p,2 A (6.19)
OOO
OOO
OOO
OOO
OOO id⊗Θ p,1 Θ p,2
OOO
id⊗∇
OOO
OOO
'  
Ω A ⊗A Ω 0,1 A ⊗A Ω p,0 A
0 ,1 / Ω 0,2 A ⊗A Ω p,0 A
∧⊗id

∧

Ω p,2 A
commutes; the triangle commutes by the definition of ∇ ; by the commutativity of (6.19),
∧ ◦ (∧ ⊗ id) ◦ (id ⊗ ∇) = ∧ ◦ Θ p,2 ◦ ∧ ◦ (id ⊗ ∂) = ∧ ◦ (id ⊗ ∂).
It follows that the composition in (6.18) is equal to
(∂ ∧ id − id ∧ ∂)Θ p,1 ∂.
This is zero because ∂¯ 2 = 0.
p,q
By the discussion prior to the lemma the Hodge to de Rham spectral sequence converges and E1 is the Dolbeault
p,q
cohomology group H (A). It therefore suffices to show that the pth Dolbeault complex
∂

∂¯ ∂¯ ∂¯
Ω p,0 A −→ Ω p,1 A −→ Ω p,2 A −→ · · · (6.20)
is isomorphic to

∇ ∇ ∇
Ω p,0 A −→ Ω 0,1 A ⊗ Ω p,0 A −→ Ω 0,2 A ⊗ Ω p,0 A −→ · · · . (6.21)
0 ,q p,0 p,q+1
Let ξ ⊗ η ∈ Ω A ⊗A Ω A. The image of ξ ⊗ η in Ω A after going clockwise around the diagram

Ω pO ,q A
∂ / Ω p,q+1 A
O
∧ ∧

Ω 0,q A ⊗A Ω p,0 A / Ω 0,q+1 A ⊗A Ω p,0 A

∇

is ∂ξ ∧ η + (−1)q ξ ∧ ∂η. Going counter-clockwise, ξ ⊗ η is first sent to

∂ξ ⊗ η + (−1)q ξ ∧ ∇η = ∂ξ ⊗ η + (−1)q ξ ∧ Θ p,1 ∂η

which is then sent to ∂ξ ∧ η + (−1)q ξ ∧ ∂η. The rectangle therefore commutes and it follows that (6.20) is isomorphic to
(6.21). 

Example 6.2. The complex-valued forms on a complex manifold are given locally in terms of the coordinates zi by wedges
of dzi and dz̄i . To get the map Θ , locally, just permute all the dz̄i to the left, introducing the appropriate power of −1, and
replace the ∧ separating the dz̄i from the dzi by ⊗. For example,
dz1 ∧ dz̄2 −→ −dz̄2 ⊗ dz1 ,
dz̄1 ∧ dz1 ∧ dz̄2 −→ −dz̄1 ∧ dz̄2 ⊗ dz1 ,
dz3 ∧ dz1 ∧ dz̄2 ∧ dz̄4 −→ dz̄2 ∧ dz̄4 ⊗ dz3 ∧ dz1 .
24 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

7. Non-commutative complex differential geometry and non-commutative projective algebraic geometry

We begin this section with some speculations in Section 7.1 about the existence of integrable almost complex structures
on non-commutative complex projective varieties. A foundation for these speculations is provided by the examples in
Sections 7.2 and 7.3.

7.1. Speculating about the big picture

Let R be a C-algebra having all the properties attributed to the ring R at the beginning of Section 1.3. For brevity, we write
p
X = Projnc (R). Following [32–34], we write OX for the image of R in QcohX and define H p (X , −) := ExtQcohX (OX , −).
This section speculates about non-commutative spaces X , Xreal , X, Xreal , and Y , and associated structures, that play roles
analogous to the spaces in the right-hand rectangle below:

2n+2
6 Xreal − {0} X − {0} − {0} Cn+1 − {0}
6R
σ R+ σ R+
 
Y C× S 2n+1 C×

U (1) U (1)
   
Xreal X CPn CPn

The left-hand columns represent the underlying real varieties of the complex varieties in the right-hand columns.
The examples in Sections 7.2 and 7.3 suggest there should be finitely generated non-commutative noetherian C-algebras
that correspond to the non-commutative spaces in the left-hand diagram just above and fit together as in the following
picture:

Z-graded
∗-algebra /o /o /o o/ / R o incl
Rhol R o o/ o/ o/ o/ N-graded
 GK = n + 1
GK = 2n + 2  
 incl 
  
 + 
R R0 [L]

 
 incl  C×
  
strongly graded 
∗-algebra o
/ o
/ o
/ / o ∼
R0 [L±1 ]

RO
GK = 2n + 1 

incl 

∗-algebra /o /o /o /o / R Hol(R0 ) _ _ _ QGr(R)
0
GK = 2n

The objects in this diagram should have the following properties:

(1) R is a finitely generated noetherian ∗-algebra of Gelfand–Kirillov dimension 2n + 2 and is a domain. It plays the role
of C-valued polynomial functions on a non-commutative real variety Xreal of real dimension 2n + 2.
(2) The C-algebra R should have an integrable almost complex structure that induces an integrable almost complex
structure R0 .
(3) R should have a Z-grading that extends to Ω • R and (Rn )∗ should equal R−n .
(4) We will write R for Rhol .
(5) It is clear that R is an N-graded C-algebra. We want R0 = C and R should be generated by a finite set of elements zi ∈ R1 .
We also want GK(R) = n + 1 but it might be over optimistic to hope that R is noetherian. Let m = R≥1 = (z1 , z2 , . . .).
(6) As a C-algebra R should be generated by R1 and its conjugate R∗1 ⊂ R−1 . Let M for the two-sided ∗-ideal of R generated
by R1 . We want R/M ∼ = C; if this is the case, then m = R ∩ M. Since C is the ∗-algebra of C-valued functions on the
point Spec(R) the homomorphism R → C with kernel M is the algebraic analogue of an inclusion Spec(R) → Xreal ;
we think of the image of this inclusion as the vertex of the cone over X and write 0 for that vertex.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 25

(7) The ring R plays two roles: that of the coordinate ring of the non-commutative complex affine variety X, and that of
the homogeneous coordinate ring of a non-commutative complex projective variety X := (X − {0})/C× . The objects
in QGr(R) are ‘‘quasi-coherent sheaves on X ’’ and objects in Mod(R) are ‘‘quasi-coherent sheaves on X’’; from this
perspective, X is the cone over X . The ‘‘quasi-coherent sheaves’’ on the punctured cone X − {0} are the objects in the
quotient category Mod(R)/Mod0 (R) where Mod0 (R) is the full subcategory of Mod(R) consisting of the R-modules that
are the sum of their finite dimensional submodules and those submodules are annihilated by mn for n ≫ 0.
(8) Let Mod0 (R) be the full subcategory of Mod(R) consisting of the modules M such that each element of M is annihilated
by some power of M. Let us write Mod(Xreal − {0}) for the quotient category Mod(R)/Mod0 (R).
(9) There should be a central element c ∈ R0 that is an R-linear combination of the elements zi zi∗ and also an R-linear
combination of the elements zi∗ zi . The requirement that the coefficient of zi zi∗ belongs to R ensures that c, and hence
c − 1, is self-adjoint which implies that R/(c − 1) is a ∗-algebra. We think of c as playing the role of the ‘‘norm-squared’’
function.
(10) R = R/(c − 1) is the algebra of C-valued polynomial functions on Y , the norm = 1 part of Xreal − {0}, and Xreal − {0}
behaves like Y × R+ .
(11) Because c − 1 is homogeneous, R inherits a grading from R. Its degree-zero component R0 is the U (1)-invariant
subalgebra, of R and plays the role of C-valued polynomial functions on Xreal , the non-commutative real variety
underlying the non-commutative complex projective variety X whose category of ‘‘quasi-coherent sheaves’’ is QGr(R).
(12) The solid arrows in the previous diagram represent C-algebra homomorphisms and those labelled incl are injective.
The dashed lines—should correspond to adjoint pairs of functors between certain abelian categories attached to the
rings.
(13) Hol(R0 ) is the category of holomorphic R0 -modules defined in Section 4.3 and should be related to QGr(R) in such a way
that the cohomology groups H • (E , ∇) for (E , ∇) ∈ Hol(R0 ) coincide with appropriate cohomology groups H q (X , −) in
QGr(R).
(14) L is an invertible R0 -bimodule, hence a rank one projective R0 -module on both the left and the right,
∞
 ∞

R0 [L±1 ] = L⊗ m and R0 [L] = L⊗ m .
m=−∞ m=0

7.1.1. Complex line bundles on Xreal

A closed subvariety X of CPn has a homogeneous coordinate ring R whose homogeneous components are (holomorphic)
sections of complex line bundles on Xreal . In the situation we are considering, invertible bimodules over R0 play the role of
complex line bundles on Xreal .
Each Rm is an R0 -bimodule.
From now on we will assume that R, and hence R0 , is a domain. This is a mild hypothesis that roughly corresponds to Y
and Xreal being irreducible varieties.
Because R0 has finite Gelfand–Kirillov dimension, it has a division ring of fractions, D say. The rank of a finitely generated
projective right R-module P is dimD (P ⊗R0 D). In particular, if P is isomorphic to a non-zero right ideal of R0 , then rank P = 1.
If conditions (9) and (10) above hold, then 1 ∈ R1 R−1 whence R0 = R1 R−1 . Likewise, 1 ∈ R−1 R1 and R0 = R−1 R1 .

Proposition 7.1. Suppose that R is a domain and that conditions (6), (9), and (10) above hold. Then each Rm is an invertible
R0 -bimodule and a finitely generated rank one projective R0 -module on both the left and the right.

Proof. For each integer m, the multiplication in R gives an R0 -bimodule homomorphism

µm : Rm ⊗R0 R−m → R0 .

Claim. µm is surjective.

Proof. This is a triviality when m = 0. The image of µm is a two-sided ideal of R so it suffices to show that 1 is in the image.
The image of µ1 contains zj zj∗ for all j and hence the element c in (9); but c = 1 in R0 so µ1 is surjective. Similarly, µ−1 is
surjective.
Suppose m ≥ 1. Because R is generated as a C-algebra by the image of {zj∗ , zj | 0 ≤ j ≤ n}, R≥0 is generated as an algebra
by R0 and R1 . Hence, if m ≥ 1, Rm = (R1 )m by which we mean that Rm is spanned by the products of m elements belonging
to R1 . Likewise, R−m = (R−1 )m . The image of µm is therefore (R1 )m (R−1 )m = R1 (Rm−1 R1−m )R1 . The truth of the claim for
m ≥ 1 now follows by induction.
A similar argument shows µm is surjective when m ≤ −1. 

By the dual basis lemma, Rm is a projective right R0 -module if the identity map is in the image of the map

Ψ : Rm ⊗R0 HomR0 (Rm , (R0 )R0 ) −→ EndR0 (Rm ), Ψ (s ⊗ φ)(s′ ) := sφ(s′ ).

26 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

Given a ∈ R−m define φa ∈ HomR0 (Rm , (R0 )R0 ) by φa (s′ ) = as′ . By the claim, there are elements si ∈ Rm and ai ∈ R−m such

that si ai = 1. Hence
 

Ψ si ⊗ φai = idR0
i

so Rm is a projective right R0 -module. A similar argument shows that Rm is a projective left R0 -module.
Because R is a domain, left multiplication by a fixed non-zero element in R−m is an injective homomorphism Rm → R0
of right R0 -modules. Hence rank Rm = 1 on the right. As similar argument shows that Rm has rank 1 as a left R0 -module.
The invertibility of Rm as an R0 -bimodule is part (3) of the next corollary. 

Corollary 7.2. Suppose that R is a domain and that conditions (6), (9), and (10) above hold. Then

(1) R is a strongly graded C-algebra.

(2) There is an equivalence of categories Mod(R0 ) ≡ Gr (R).
(3) The multiplication map Rk ⊗R0 Rm → Rk+m is an isomorphism of R0 -bimodules for all k, m ∈ Z.
(4) Condition (14) holds.

Proof. (1) By definition, a Z-graded ring A is strongly graded if Ai A−i = A0 for all i. The proof of Proposition 7.1 shows that
1 ∈ R−m Rm for all m ∈ Z so R is strongly graded.
(2), (3), and (4), are standard facts about strongly graded rings. 

Given Proposition 7.1, the philosophy of non-commutative geometry says that elements of Rm are smooth sections of
a complex line bundle on the real algebraic variety Xreal . After we give Xreal , equivalently R0 , an integrable almost complex
structure there should be a ∂ -operator ∇ on each Rm such that Rm consists of the elements of Rm on which ∇ vanishes.

7.1.2. The (punctured) affine cone over X

As in algebraic geometry, there is a non-commutative complex affine variety X that is a cone over X defined implicitly
by saying that the category of quasi-coherent sheaves on X is ModR. We then think of R as a ring of ‘‘holomorphic functions’’
on X.
The Artin–Schelter regularity conditions require R to have good homological properties that one should probably think
of as saying X is smooth, so not quite what most people would think of when they say ‘‘cone’’. If necessary, one could
weaken the hypothesis on R by allowing it to be a quotient of an Artin–Schelter regular ring by a central regular sequence
such that the vertex of the cone is the only singular point, i.e., the ‘‘deviation’’ of R from being Artin–Schelter regular is
‘‘concentrated’’ at the module R/R≥1 . Formally, one would ask that the quotient category (ModR)/S, where S is the smallest
localizing subcategory containing R/R≥1 , has finite global dimension, i.e., Ext2n+1 vanishes on (ModR)/S. In such a case it is
more appropriate to think of R as a ring of ‘‘holomorphic functions’’ on X − {0}.

7.1.3. The action of C× on X

The homogeneous components of R are the eigenspaces for an action of C× as automorphisms of R. If we think of this as
being inherited from an action of C× on X, then Gr R is, in effect, the category of C× -equivariant (i.e., constant along orbits)
quasi-coherent modules on X. The vertex 0 ∈ X is, by definition, the unique point fixed by C× ; formally, {0} = Spec(R/R≥1 ).
By quotienting out Fdim R we are removing those modules supported at the vertex. Thus X behaves as the orbit space
(X − {0})/C× .

7.1.4. Viewing X as a non-commutative real algebraic variety with an almost complex structure
Viewed through the lens of the ring R, X is a non-commutative complex variety of complex dimension n + 1. If we wish
to treat X as a real algebraic variety of real dimension 2(n + 1) we need a ∗-algebra R of GK-dimension 2n + 2 that will play
the role of C-valued differentiable functions on Xreal − {0}.
An appropriate almost complex structure on Xreal would be an almost complex structure (Ω • R, d, ∗, J ) such that

(1) Rhol = R,
(2) there is a Z-grading on R that is compatible with that on R, and
(3) (Rn )∗ = R−n .

Because we want to retain the underlying real structure on X we use the factorization C× = R+ × U (1) to pass down
the left-hand sides of the diagrams at the beginning of Section 7.1. Sections 7.1.5 and 7.1.6 address this.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 27

7.1.5. Xreal − {0} as a trivial R+ -bundle over Y

Let S 2n+1 be the sphere of radius 1 in R2n+2 centred at the origin. Then R2n+2 − {0} is diffeomorphic to R+ × S 2n+1 . A
non-commutative analogue of the inclusion S 2n+1 → R2n+2 is a surjective homomorphism
R
R → R :=
(c − 1)
where c − 1 is a self-adjoint central element in R0 . Since R is to play the role of C-valued polynomials on a non-commutative
real algebraic variety Y of dimension 2n + 1 we want the GK-dimension of R to be 2n + 1. By Krause and Lenagan
[35, Proposition 3.15] GK(R) ≤ 2n + 1.
The homomorphism R → R induces a closed immersion σ : Y → Xreal − {0} and there should be a non-commutative
map π : Xreal − {0} → Y such that π σ = idY . The existence of such π and σ can be phrased in terms of the associated
module categories by using the ideas and language in [36,37]. A closed immersion is a triple of functors (σ ∗ , σ∗ , σ ! ) such
that σ∗ : Mod (R/(c − 1)) → Mod(Xreal − {0}) is fully faithful, the essential image of σ∗ is closed under subquotients, and
σ∗ has a left adjoint σ ∗ and a right adjoint σ∗ .
For example, the functor β∗ : Mod (R/(c − 1)) → Mod(R) that sends an R/(c − 1) module M to M viewed as an
R-module determines a closed immersion β : Y → Xreal . Similarly, The homomorphism R → R/M ∼ = C corresponds to a
closed immersion α : {0} → Xreal .
The localization functor j∗ : Mod(R) → Mod(Xreal − {0}) and its right adjoint j∗ implicitly define an open immersion
j : Xreal − {0} → X.

Lemma 7.3. Let β∗ be the forgetful functor associated to the homomorphism R → R/(c − 1), and let β ∗ and β ! the usual left
and right adjoints to β∗ . The functors σ ∗ := β ∗ j∗ , σ∗ := j∗ β∗ , and σ ! := β ! j∗ , define a closed immersion σ : Y → Xreal − {0}
such that the diagram
j
/ Xreal o α
6 Xreal − {0} t9
{0}
tt
tt
tt β
σ
t
tt
Y

commutes. Furthermore, σ ∗ σ∗ ∼
= id and σ ! σ∗ ∼
= id.
Proof. Because β∗ is left adjoint to β ! and j∗ is left adjoint to j∗ , j∗ β∗ is left adjoint to β ! j∗ .
To key step in showing β ∗ j∗ is left adjoint to σ∗ := j∗ β∗ is to show that HomR (M , N ) = Ext1R (M , N ) = 0 for all R/(c − 1)-
modules N and all R/M-modules M (see [37, Proposition 7.1] for example). Since every R/M-module is a direct sum of
copies of R/M it suffices to show that HomR (R/M, N ) = Ext1R (R/M, N ) = 0 for all R/(c − 1)-modules N.
Since c belongs to M, R/M is not annihilated by c − 1, whence HomR (R/M, N ) = 0.
Now consider an exact sequence 0 → N → E → R/M → 0 of left R-modules such that (c − 1)N = 0. Since c is central
multiplication by c − 1 gives an R-module homomorphism E → (c − 1)E. Hence (c − 1)E ∼ = R/M. Since (c − 1)N = 0,
E = N ⊕ (c − 1)E and this leads to a splitting of the sequence 0 → N → E → R/M → 0.
We now show σ ∗ is left adjoint to σ∗ . Let N be an R-module. It is a standard fact [38] about localizing subcategories that
there is an exact sequence

0 → lim HomR (R/Mn , N ) → N → j∗ j∗ N → lim Ext1R (R/Mn , N ) → 0.

−
→ −
→
n n

From this sequence and what we proved above we see that the natural map N → j∗ j∗ N is an isomorphism for all R/(c − 1)-
modules N. Hence j∗ j∗ β∗ ∼
= β∗ . It is also a standard fact about localizations that j∗ j∗ ∼
= id. Let N be an R/(c − 1)-module.
Then

Hom(β ∗ j∗ M , N ) ∼
= Hom(j∗ M , j∗ j∗ β∗ N ) ∼
= Hom(j∗ j∗ M , j∗ β∗ N ) ∼
= Hom(M , j∗ β∗ N ).
Thus σ ∗ is left adjoint to σ∗ . It is easy to check that σ ∗ σ∗ ∼
= id and σ ! σ∗ ∼
= id. 

For this to be compatible with the principal C× -bundle on the right-hand sides of the diagrams at the beginning of
Section 7.1 it should be the case that Rm = L⊗m .
As mentioned before the previous result we would like a non-commutative map π : Xreal −{0} → Y such that π σ = idY .
Since σ ∗ σ∗ ∼
= id one might define π∗ = σ ∗ . However, we do not know whether this is appropriate since it does not appear
that such a π∗ has a left adjoint.

7.1.6. The action of U(1) on Y

There is an action of U (1) as ∗-algebra automorphisms of R and R/(c − 1) given by ξ · r = ξ m r if r ∈ L⊗m . If x ∈ Rm and
ξ ∈ U (1), then ξ · x∗ = (ξ · x)∗ = (ξ m x)∗ = ξ̄ m x∗ = ξ −m x∗ so x∗ ∈ R−m .
28 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

p
7.1.7. Analogues of the sheaves ΩX
Even for those non-commutative projective varieties X that are well understood and have excellent homological
p
properties one knows no objects in Qcoh(X ) that deserve being denoted ΩX . There is no effective general theory of exterior
p
powers in non-commutative ring theory so it is unwise to seek objects ΩX that are exterior powers of ΩX1 .
An alternative is to ask that R, which is connected graded, be a twisted Calabi–Yau algebra of dimension n, i.e., the
projective dimension of R as an R-bimodule is n and the shifted Hochschild cohomology HH • (R, R ⊗ R)[n] is an invertible
R-bimodule. Over a connected graded algebra the only invertible graded bimodules are of the form 1 Rσ for some graded
algebra automorphism σ of R. Because R is connected graded it has a unique minimal projective resolution as an R-bimodule.
p
The wished for ΩX might be defined in terms of this resolution. Beilinson’s resolution of the diagonal is the motivation for
this approach.
A test as to whether these ideas are promising is to ask if, after proceeding as above, for some X ’s that we understand,
p,q p q p
H (R) is isomorphic to H q (X , ΩX ) where the latter is defined as ExtQcohX (OX , ΩX ).
∂

7.1.8. Testing these ideas on non-commutative analogues of PnC

Let R be an Artin–Schelter regular algebra [39] such that the minimal resolution of the trivial module C = R/R≥1 is
0 → R(−n) → R(−n + 1)n → · · · → R(−1)n → R → C → 0.
For such an R it is generally accepted that X is a good non-commutative analogue of PnC . The image in Qcoh(X ) of the
resolution is an exact sequence
0 → O (−n) → O (−n + 1)n → · · · → O (−1)n → OX → 0.
When X is PnC ,
    
n n
p
ker O (−p) p
→ O (−p + 1) p−1
= Ωhol (7.22)

p
and Bott’s Vanishing Theorem (see, e.g., [40, Theorem 7.2.3]) says H q (PC
n −1
, Ωhol (k)) vanishes except for
(1) p = q and k = 0;
(2) q = 0 and k > p;
(3) q = n − 1 and k < −n + 1 + p.
The object Ωhol in (7.22) is not the same as the object called Ωhol in (3.12); the latter was defined to be {ω ∈ Ω p,0 R | ∂ω =
p p

0}. However, if {ω ∈ Ω p,0 R | ∂ω = 0} is a graded R-module it would have an image in the quotient category Qcoh(X ) and
we could ask if that image is isomorphic to the kernel in (7.22). If so, Bott’s Vanishing Theorem would hold for X : because R
is Gorenstein, which is part of the definition of Artin–Schelter regularity, the cohomology for OX (k) is the same as that for
OPn−1 (k).
p
For Ωhol , as defined in (3.12), to be a graded R-module the C× -action on R must extend to a suitable C× -action on Ω • R.

7.2. An example: θ -deformed Cn+1 and CPn

There is a large physics literature on the subject of θ -deformed spaces, or Moyal products. This has been used in
quantization, gauge theory, string theory, cosmology and integrable systems. We will quote the relations for the algebra
and differential calculus from [41].

7.2.1. The non-commutative (2n + 2)-plane R2n

θ
+2

Let θ be a real skew-symmetric (n + 1) × (n + 1) matrix, and set λµν = eiθµν .

The algebra of complex-valued polynomial functions on Rθ2n+2 is the ∗-algebra C R2n generated by {z µ , z̄ µ | 0 ≤
 +2 
θ
µ ≤ n} with relations
z µ z ν = λµν z ν z µ , z̄ µ z̄ ν = λµν z̄ ν z̄ µ , z̄ µ z ν = λνµ z ν z̄ µ (7.23)
and (z µ )∗ = z̄ µ . The elements
(z 0 )i0 · · · (z n )in (z̄ 0 )j0 · · · (z̄ n )jn , (i0 , . . . , in , j0 , . . . , jn ) ∈ N2n+2
 +2 
are a C-vector space basis for C R2n θ .
There is an action of C× as algebra automorphisms of C Rθ2n+2 defined by ξ · z µ = ξ z µ and ξ · z̄ µ = ξ̄ z̄ µ . Its subgroup
 

U (1) acts as ∗-algebra automorphisms of C Rθ . The C× -action induces a Z-grading on C Rθ2n+2 with deg z µ = 1 and
 2n+2   
deg z̄ µ = −1.
2n+2

Since C Rθ is a quadratic algebra having a PBW basis it is a Koszul algebra by Priddy’s Theorem [42, Theorem 5.3].
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 29

+2
7.2.2. A differential ∗-calculus on R2n
θ
A ∗-differential calculus (Ω • , d, ∗) on C R2n+2
is defined in [41]: Ω • C R2n +2
has generators z µ , z̄ µ ∈ Ω 0 and
   
θ θ
dz µ , dz̄ µ ∈ Ω 1 with relations (7.23) and

z µ dz ν = λµν dz ν z µ ,
z̄ µ dz̄ ν = λµν dz̄ ν z̄ µ ,
z̄ µ dz ν = λνµ dz ν z̄ µ ,
z µ dz̄ ν = λνµ dz̄ ν z µ , (7.24)

and

dz µ ∧ dz ν + λµν dz ν ∧ dz µ = 0,
dz̄ µ ∧ dz̄ ν + λµν dz̄ ν ∧ dz̄ µ = 0,
dz̄ µ ∧ dz ν + λνµ dz ν ∧ dz̄ µ = 0, (7.25)

and ∗-structure defined by

(z µ )∗ = z̄ µ and (dz µ )∗ = dz̄ µ

and extended to Ω • C R2n by (ωη)∗ := (−1)mn η∗ ω∗ for ω ∈ Ω m C R2n and η ∈ Ω n C Rθ2n+2 .
 +2   +2   
θ θ
The C× -action extends to Ω • C R2n +2
by ξ · dz µ = ξ dz µ and ξ · dz̄ µ = ξ̄ dz̄ µ .
 
θ

+2
7.2.3. An integrable almost complex structure on R2n
θ
There is a unique degree-preserving map J : Ω • C R2n+2
→ Ω • C Rθ2n+2 that is
   
θ

(1) an A-bimodule endomorphism, and

(2) a derivation, and satisfies
(3) J (dz µ ) := i dz µ and J (dz̄ µ ) := −i dz̄ µ .
In other words, J is an almost complex structure on R2n +2
. We therefore have spaces Ω p,q C Rθ2n+2 and derivations ∂ and ∂
 
θ
defined as in Sections 2.5.2 and 2.5.3.
The C× -action extends with ξ · ω = ξ p ξ̄ q ω for ω ∈ Ω p,q C Rθ2n+2 .
 

Lemma 7.4. J is integrable.

Proof. Every element in Ω 1,0 C R2n +2

can be written in the form fµ .dz µ for some fµ ∈ C R2n +2
   
. Applying d to this gives
µ 2 ,0
 2n+θ2  1,1
θ
dfµ ∧ dz , which is in Ω C Rθ ⊕ Ω C Rθ
 2n+2 
. Hence d satisfies condition (4) in Lemma 3.2. 

The subalgebra C R2n +2

is generated by the z µ , 0 ≤ µ ≤ n. We denote it by O (Cθn+1 ) and think of it as the coordinate
 
θ hol
ring of a non-commutative affine variety Cnθ +1 . The algebra O (Cθn+1 ) has all the properties attributed to the ring R at the
beginning of Section 1.3. It is well-known that Projnc O (Cθn+1 ) behaves very much like CPn .

7.2.4. The non-commutative sphere Sθ2n+1

For each µ, the element z̄ µ z µ = z µ z̄ µ is central. We define the C-valued polynomial functions on Sθ2n+1 to be
+2
C R2n
 
θ
2n+1
.
 
C Sθ := 
z µ z̄ µ

−1
µ

z µ z̄ µ − 1 is self-adjoint C Sθ2n+1 inherits a ∗-algebra structure from C Rθ2n+2 . Since z µ z̄ µ − 1 is invariant

     
Since µ µ
under the C× -action the unit circle U (1) acts as ∗-algebra automorphisms of C Sθ2n+1 .
 
 µ µ µ µ
 µ µ
µ z dz̄ = 0, so we define Ω C Sθ
 •
 2n+1 
Differentiating the relation µ z z̄ = 1 gives µ dz z̄ + to be the quotient
of Ω • C Rθ 2n+2
by the relations µ z µ z̄ µ = 1 and µ dz µ z̄ µ + µ z µ dz̄ µ = 0.
    

7.2.5. CPnθ
We define C CPnθ to be the U (1)-invariant subalgebra of C Sθ2n+1 . Since U (1) acts as ∗-algebra automorphisms, C CPnθ
     
is a ∗-algebra.
30 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

We define Ω • C CPθ to be the U (1)-invariant subalgebra of Ω • C Sθ2n+1 , with the additional relations dz µ z̄ µ = 0
 n
   
µ
and µ z µ dz̄ µ = 0.

µ µ
= 0 in Ω 1,0 C CPnθ and µ z µ dz̄ µ = 0 in Ω 0,1 C CPnθ correspond to the following
     
These relations µ dz z̄
projection matrices for finitely generated projective modules:
Pµν = δµν − z µ z̄ ν , Qµν , = δµν − z̄ µ z ν .
The complex structure is given by J from Section 7.2.3, i.e. J (dz µ ) := i dz µ and J (dz̄ µ ) := −i dz̄ µ . A brief check shows that
this is consistent with the additional relations given here.

7.2.6. Complex line bundles on CPnθ

The U (1)-action on C Sθ2n+1 corresponds to a Z-grading on C Sθ2n+1 and C CPnθ is its degree-zero component. The
     

homogeneous components of C Sθ2n+1 are therefore C CPnθ -bimodules.

   
   
They are, in fact, invertible C CPnθ -bimodules, hence finitely generated rank one projective C CPnθ -modules on both
the right and left so should be thought of as polynomial sections of complex line bundles on CPnθ . The holomorphic sections
should be thought of as the global sections of invertible OCPn -modules.
θ

7.3. A complex structure on quantum irreducible flag varieties

In [43,5,6], Kolb and Heckenberger construct an integrable almost complex structure on the quantum group analogues
of irreducible flag varieties. Stefan Kolb kindly provided us with some notes about that work and has allowed us to use them
in writing this section. We thank him. Of course, any inaccuracies below are due to the present authors.

7.3.1. Generalized flag varieties

Let g be a finite dimensional, complex, simple Lie algebra, p ⊂ g a parabolic subalgebra, p = l ⊕ n its Levi decomposition,
and pop := l ⊕ n− the opposite parabolic.
Let G be the connected, simply connected, affine algebraic group with Lie algebra g, P ⊂ G the connected subgroup with
Lie algebra p, and L the Levi factor of P. The quotients G/L and G/P are affine and projective algebraic varieties, respectively.
We make the additional assumption that g/p is irreducible as a p-module under the adjoint action. In this situation
G/P is called an irreducible flag manifold.3 Irreducible flag manifolds coincide with compact Hermitian symmetric
spaces [45, Section X.6.3].
Suppose that µ is a dominant integral weight, V (µ) is the finite dimensional simple U (g)-module with highest weight
µ, and vµ is a highest weight vector in V (µ).
For each parabolic subgroup P ⊂ G there is a weight λ such that G/P is isomorphic to the G-orbit in P (V (λ)) of vλ , the
image of vλ .4 The vector space
∞

S [G/P ] := V (nλ)∗
n=0

endowed with the Cartan multiplication and Z-grading given by deg V (λ)∗ = +1 is a homogeneous coordinate ring of G/P.
It coincides with the subalgebra of O (G) generated by the matrix coefficients {cfλ,vλ | f ∈ V (λ)∗ } where
µ
cf ,v (u) := f (uv), f ∈ V (µ)∗ , v ∈ V (µ), u ∈ U (g).

There are invertible OG/P -modules Lnλ such that H 0 (G/P , Lnλ ) ∼
= V (nλ)∗ .

7.3.2. Quantum group analogues

Let q ∈ R − {0}.
Let Uq (g) be the quantized enveloping algebra of g defined at [25, 6.1.2]. The quantum group Gq is defined implicitly by
µ
declaring that its coordinate ring, O (Gq ), is the linear span of the matrix coefficients cf ,v ∈ Uq (g)∗ ,
µ
cf ,v (u) := f (uv), f ∈ V (µ)∗ , v ∈ V (µ), u ∈ Uq (g),

as µ ranges over the dominant integral weights and V (µ) is the simple Uq (g)-module with highest weight µ.

3 For a classification of such pairs (g, p) see [44, Section 3.1].

4 The weight λ is not unique but there is a ‘‘smallest’’ such λ, the one that is a sum of fundamental weights with coefficients 0 or 1. From now on it is
that λ which we will use.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 31

Following Soibelman [46], and subsequent authors, let S [Gq /P ] be the subalgebra of O (Gq ) generated by {cfλ,vλ | f ∈
V (λ)∗ }. There is an isomorphism
∞

S [Gq /P ] ∼
= V (nλ)∗
n =0

of vector spaces that becomes an algebra isomorphism when the right-hand side is given a quantized analogue of the Cartan
product. We view S [Gq /P ] as a homogeneous coordinate ring of a non-commutative analogue of G/P.5 In this setting, S [Gq /P ]
is the algebra denoted R in Section 7.1, and the non-commutative analogue of G · vλ plays the role of X .
So far we are only thinking of Gq /P as a non-commutative complex projective variety, Projnc (S [Gq /P ]) where S [Gq /P ] is
thought of as holomorphic polynomial functions on the ‘‘punctured cone’’ over Gq /P, i.e., the non-commutative space X −{0}
in Section 7.1. To give Gq /P, or, more precisely, X − {0}, an underlying real structure we need a ∗-algebra that plays the role
played by R in Section 7.1.4.
At a minimum, we need an algebra to play the role of anti-holomorphic polynomial functions on X − {0}. The G-orbit
in P(V (λ)∗ ) of a lowest weight vector f−λ ∈ V (λ)∗ ∼ = V (−w0 λ) is isomorphic to G/P op so, following the classical case
(e.g., [43, Sections 2–3]), we define the ring of anti-holomorphic polynomial functions on Gq /P to be the subalgebra
−w0 λ
S [Gq /P op ] ⊂ O (Gq ) generated by {cv,f−λ | v ∈ V (−w0 λ)∗ ∼
= V (λ)}. It is isomorphic to
∞

V (nλ)
n =0

endowed with the quantized Cartan multiplication. We consider S [Gq /P op ] as a Z-graded algebra concentrated in degree
≤ 0 with deg V (λ) = −1.

7.3.3. The underlying real structure on Gq /P

First, we need an underlying real structure on Gq by viewing it as the complexification of a non-commutative real
algebraic variety. Recall, G is the complexification of its compact connected real form, U ⊂ G and, if K = U ∩ P, the
natural inclusion U /K → G/P is a diffeomorphism.
The Hopf ∗-algebra structure on Uq (g) induces a Hopf ∗-algebra structure on O (Gq ). The explicit form of ∗ appears at
[47, (3.5)]. We write C[Uq ] for O (Gq ) viewed as a ∗-algebra; as a ∗-algebra C[Uq ] behaves as C-valued polynomial functions
on the quantum group analogue Uq of the compact real form U. The involution ∗ is an anti-isomorphism between S [Gq /P ]
and S [Gq /P op ] (see [48, Definition 2.2]).
Let R be the Z-graded vector space S [Gq /P ]⊗ S [Gq /P op ] with the following algebra structure: let v1 , . . . , vN and f1 , . . . , fN
be dual bases for V (λ) and V (λ)∗ consisting of weight vectors; declare that the subalgebra of R generated by f1 , . . . , fN (resp.,
v1 , . . . , vN ) is isomorphic to S [Gq /P ] (resp., S [Gq /P op ]), and declare that (1 ⊗ vi )(fi ⊗ 1) is the linear combination of the
elements fk ⊗ vℓ given by Heckenberger and Kolb [5, (28)].
The degree-zero component, R0 , can also be described in the following alternative ways:
q
(1) R0 = Åλ , the unital ∗-subalgebra of C[Uq ] generated by the products {f · g ∗ | f , g ∈ V (λ)∗ } ([48, Definition 2.3(b)]);
(2) R0 = C[Uq /L ∩ Uq ] = C[Gq /L], the subalgebra {φ ∈ C[Uq ] | u.φ = ε(u)φ ∀u ∈ Uq (l)} of C[Uq ] ([47, (4.4)]).
Equality of the algebras in (1) and (2) is proved at [47, Theorem 4.10] and [48, Theorem 2.5] by passing to C ∗ -algebra com-
pletions, and at [43, Proposition 2] without using C ∗ -algebras.
Following Heckenberger and Kolb [5, Section 3.1.2] there is a central element and an associated quotient algebra,
N
 R
c := vi ⊗ fi ∈ R0 and R := .
i=1
(c − 1)

Because deg(c ) = 0, R inherits the Z-grading on R. Because c is self-adjoint it also inherits the ∗-algebra structure on R. The
c =1
algebra R is denoted Sq [G/P ]C by Heckenberger and Kolb in [5, Section 3].

7.3.4. The de Rham complex for C[Gq /L]

By Heckenberger and Kolb [43, Theorem 2, p. 483], there are exactly two non-isomorphic finite dimensional irreducible
left covariant first order differential calculi, C[Gq /L] → Γ+ and C[Gq /L] → Γ− , in the sense of Woronowicz [25, Section
12.1.2]. In [6, Section 7.1], Heckenberger and Kolb relabel these (Γ∂ , ∂) and (Γ∂ , ∂). To conform with our notation we label
them (Ω 1,0 C[Gq /L], ∂) and (Ω 0,1 C[Gq /L], ∂). Heckenberger and Kolb now define

(Ω 1 C[Gq /L], d) := (Ω 1,0 C[Gq /L] ⊕ Ω 0,1 C[Gq /L], ∂ ⊕ ∂).

5 We are writing S [G /P ] for the algebra denoted S [G/P ] by Heckenberger and Kolb [5, Section 3.1.1] and, later, will write S [G /P op ] for the algebra they
q q q
denote by Sq [G/P op ].
32 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33

Thus, Ω 1 C[Gq /L] and the derivation d : C[Gq /L] → Ω 1 C[Gq /L] provide a non-commutative analogue of the module of
Kähler differentials for the affine variety G/L.
Associated to each first order calculus (Ω 1 C[Gq /L], d), (Ω 1,0 C[Gq /L], ∂), and (Ω 0,1 C[Gq /L], ∂), is a unique universal
higher order differential calculus (see, e.g., [25, 12.2.3]) that we denote by (Ω • C[Gq /L], d), (Ω •,0 C[Gq /L], ∂), and
(Ω 0,• C[Gq /L], ∂), respectively.6
For example, (Ω • C[Gq /L], d) is the differential graded algebra that is the quotient of the tensor algebra TC[Gq /L] (Ω 1 ) by
the ideal generated by
  
 
dai ⊗ dbi  a db = 0 (7.26)

i
 i i i

with differential defined by d(a0 da1 ∧ · · · ∧ dan ) = da0 ∧ da1 ∧ · · · ∧ dan .7 It follows from the definition of Ω • C[Gq /L] and
d : Ω • C[Gq /L] → Ω •+1 C[Gq /L] that d2 = 0.
Heckenberger and Kolb call (Ω • C[Gq /L], d) the de Rham complex for Cq [G/L]. By construction, Ω 0 C[Gq /L] = Cq [G/L]
and d : Cq [G/L] → Ω 1 C[Gq /L] is, as above, ∂ ⊕ ∂ .
The main results in [5, Proposition 3.6, 3.7, and 3.11], show that the homogeneous components of (Ω • C[Gq /L], d),
•C[Gq /L],0
(Ω , ∂), and (Ω 0,• C[Gq /L], ∂), have the same dimensions as their classical counterparts. In particular, Ω 2 dim(g/p)
C[Gq /L] is a free left and right C[Gq /L]-module of rank one, a generator being interpreted as a volume form on Gq /P.

7.3.5. (p, q)-forms on Gq /P

In [6], it is shown that (Ω • C[Gq /P ], d) may be constructed in terms of the Bernstein–Gelfand–Gelfand resolution and
that construction [6, above Theorem 7.14] gives a decomposition of (Ω • C[Gq /P ], d) as a direct sum of covariant Cq [G/L]-
bimodules

Ω p,q C[Gq /P ]

Ω n C[Gq /P ] = (7.27)
p+q=n

where Ω 1,0 C[Gq /P ] and Ω 0,1 C[Gq /P ] are as in Section 7.3.4 and d : Ω n C[Gq /P ] → Ω n+1 C[Gq /P ] satisfies d = ∂ + ∂ with
∂(Ω p,q C[Gq /P ]) ⊂ Ω p+1,q C[Gq /P ] and ∂(Ω p,q C[Gq /P ]) ⊂ Ω p,q+1 C[Gq /P ]. The differentials ∂ and ∂ coincide with those
defined in [5, Proposition 3.8, Remark 3.10]. Proceeding as in [6, Lemmata 7.6, 7.17], one shows that the multiplication maps

∧ : Ω 1,0 C[Gq /P ] : Ω p,q C[Gq /P ] → Ω p+1,q C[Gq /P ]

∧ : Ω 0,1 C[Gq /P ] : Ω p,q C[Gq /P ] → Ω p,q+1 C[Gq /P ]
are surjective, and hence

Ω p,q C[Gq /P ] ∧ Ω p ,q C[Gq /P ] = Ω p+p ,q+q C[Gq /P ] .

′ ′ ′ ′
(7.28)

7.3.6. Definition of J
Let J : Ω 1 C[Gq /P ] → Ω 1 C[Gq /P ] be the unique C-linear map such that J 2 = −1 and Ω 1,0 is the +1-eigenspace and Ω 0,1
is the −1-eigenspace. It is clear that J is a Cq [G/L]-bimodule automorphism of both Ω 1,0 C[Gq /P ] and Ω 0,1 C[Gq /P ] hence a
Cq [G/L]-bimodule automorphism of Ω 1 C[Gq /P ].
Using the direct sum decomposition (7.27) we can extend J to a bimodule isomorphism J : Ω • C[Gq /P ] → Ω • C[Gq /P ]
by

J (ω) = (p − q)iω for all ω ∈ Ω p,q C[Gq /P ].

By (7.28), J is a derivation: if ω ∈ Ω p,q C[Gq /P ] and τ ∈ Ω p ,q C[Gq /P ], then

′ ′

J (ω ∧ τ ) = (p + p′ − (q + q′ ))i ω ∧ τ
= (p − q)i ω ∧ τ + (p′ − q′ )i ω ∧ τ
= J (ω) ∧ τ + ω ∧ J (τ ).

6 These are denoted (Γ , d), (Γ , ∂), and (Γ , ∂), in the sentence before [6, Theorem 7.1].
d ∂ ∂
7 Here, as throughout this paper, ∧ is used to denote the multiplication and is not assumed to be skew-commutative.
 E. Beggs, S. Paul Smith / Journal of Geometry and Physics 72 (2013) 7–33 33

Acknowledgements

The authors would like to thank the Isaac Newton Institute, where this work began during the program Non-commutative
Geometry, 24 July–22 December 2006. The authors are also grateful to the Milliman Fund of the University of Washington’s
Mathematics Department for funding the first author’s visit there in September 2007. Edwin Beggs would also like to
thank the Royal Society for a grant to allow him to give a lecture on this material ‘Non-commutative sheaves and
complex structures’ at the International Conference on K -theory, C ∗-algebras and topology of Manifolds, Chern Institute
of Mathematics, Tianjin, 1–5 June, 2009. The second author is grateful for support from the NSF (grant DMS-0602347). The
authors would like to thank Stefan Kolb for providing notes relating to the papers [43,5,6], which were used in writing
Section 7.3.

References

[1] A. Newlander, L. Nirenberg, Complex analytic coordinates in almost complex manifolds, Ann. of Math. 65 (1957) 391–404.
[2] J.P. Serre, Géométrie algébrique et géométrie analytique, Ann. Inst. Fourier (Grenoble) 6 (1956) 1–42.
[3] J.T. Stafford, M. Van den Bergh, Noncommutative curves and noncommutative surfaces, Bull. Amer. Math. Soc. 38 (2001) 171–216.
[4] A. Polishchuk, A. Schwarz, Categories of holomorphic vector bundles on non-commutative two-tori, Comm. Math. Phys. 236 (1) (2003) 135–159.
[5] I. Heckenberger, S. Kolb, De Rham complex for quantized irreducible flag manifolds, J. Algebra 305 (2) (2006) 704–741.
[6] I. Heckenberger, S. Kolb, De Rham complex via the Bernstein–Gelfand–Gelfand resolution for quantized irreducible flag manifolds, J. Geom. Phys. 57
(2007) 2316–2344.
[7] F. D’Andrea, L. Dabrowski, G. Landi, The non-commutative geometry of the quantum projective plane, Rev. Math. Phys. 20 (2008) 979–1006. arXiv:
0712.3401v2.
[8] F. D’Andrea, G. Landi, Anti-selfdual connections on the quantum projective plane: monopoles, Comm. Math. Phys. 297 (2010) 841–893. arXiv:0903.
3551v1.
[9] M. Khalkhali, G. Landi, W.D. van Suijlekom, Holomorphic structures on the quantum projective line, Int. Math. Res. Not. IMRN 4 (2011) 851–884. arXiv:
0907.0154.
[10] M. Khalkhali, A. Moatadelro, The homogeneous coordinate ring of the quantum projective plane, J. Geom. Phys. 61 (2011) 276–289.
[11] M. Khalkhali, A. Moatadelro, Noncommutative complex geometry of the quantum projective space, J. Geom. Phys. 61 (2011) 2436–2452. arXiv:1105.
0456.
[12] R. Ó Buachalla, Quantum bundles and noncommutative complex structures, arXiv:1108.2374.
[13] A. Connes, Non-Commutative Differential Geometry, Publ. I.H.E.S., 1985, pp. 41–144. Tome 62.
[14] S.L. Woronowicz, Twisted SU(2) group, an example of a noncommutative differential calculus, in: Publ. Res. Inst. Math. Sci., vol. 23, Kyoto Univ., 1987
pp. 117–181.
[15] S.L. Woronowicz, Differential calculus on compact matrix pseudogroups (quantum groups), Comm. Math. Phys. 122 (1) (1989) 125–170.
[16] T. Brzeziński, H. Dabrowski, J. Rembieliński, On the quantum differential calculus and the quantum holomorphicity, J. Math. Phys. 33 (1992) 19–24.
[17] C.-S. Chu, P.-M. Ho, B. Zumino, Some complex quantum manifolds and their geometry, in: Quantum Fields and Quantum Space Time (Cargse, 1996),
in: NATO Adv. Sci. Inst. Ser. B Phys., vol. 364, Plenum, New York, 1997, pp. 281–322.
[18] M. Dieng, A. Schwarz, Differential and complex geometry of two-dimensional noncommutative tori, Lett. Math. Phys. 61 (2002) 263–270. arXiv:math.
QA/0203160v3.
[19] S. Klimyk, A note on noncommutative holomorphic and harmonic functions on the unit disk, in: Analysis, Geometry and Topology of Elliptic Operators,
World Sci. Publ, Hackensack, NJ, 2006, pp. 383–400. arXiv:0409040v1.
[20] S. Majid, Noncommutative Riemannian and spin geometry of the standard q-sphere, Comm. Math. Phys. 256 (2005) 255–285.
[21] S. Sinel’shchikov, L. Vaksman, Quantum groups and non-commutative complex analysis, arXiv:0410396v1.
[22] S.J. Brain, S. Majid, Quantisation of twistor theory by cocycle twist, Comm. Math. Phys. 284 (2008) 713–774.
[23] E.J. Beggs, S. Majid, Bar categories and star operations, Algebr. Represent. Theory 12 (2009) 103–152.
[24] J.M. Gracia-Bondía, J.C. Várailly, H. Figueroa, Elements of Noncommutative Geometry, Birkhäuser, 2000.
[25] A. Klimek, K. Schmüdgen, Quantum Groups and Their Representations, in: Texts and Monog. in Physics, Springer-Verlag, Berlin, 1997.
[26] J. Cuntz, D. Quillen, Algebra extensions and nonsingularity, J. Amer. Math. Soc. 8 (1995) 251–289.
[27] D. Huybrechts, Complex Geometry: An Introduction, in: Universitext, Springer-Verlag, 2004.
[28] C. Voisin, Hodge theory and complex algebraic geometry I, Cambridge Stud. Adv. Math. (76) (2002).
[29] P. Griffiths, J. Harris, Principles of Algebraic Geometry, Wiley, New York, 1978.
[30] C. Weibel, An Introduction to Homological Algebra, in: Camb. Studies in Adv. Math., No. 38, Camb. Univ. Press, 1994.
[31] E.J. Beggs, T. Brzeziński, The Serre spectral sequence of a noncommutative fibration for de Rham cohomology, Acta Math. 195 (2005) 155–196.
[32] M. Artin, J. Tate, M. Van den Bergh, Some algebras related to automorphisms of elliptic curves, in: The Grothendieck Festschrift, vol. 1, Birkhauser,
Boston, 1990, pp. 33–85.
[33] M. Artin, J. Tate, M. Van den Bergh, Modules over regular algebras of dimension 3, Invent. Math. 106 (1991) 335–388.
[34] M. Artin, J.J. Zhang, Non-commutative projective schemes, Adv. Math. 109 (1994) 228–287.
[35] G. Krause, T.H. Lenagan, Growth of Algebras and Gelfand–Kirillov Dimension, Revised Edition, in: Grad. Studies in Math., vol. 22, Amer. Math. Soc.,
Providence, RI, 2000.
[36] M. Van den Bergh, Blowing up on non-commutative smooth surfaces, Mem. Amer. Math. Soc. 154 (734) (2001).
[37] S.P. Smith, Subspaces of non-commutative spaces, Trans. Amer. Math. Soc. 354 (2002) 2131–2171.
[38] P. Gabriel, Des catégories abéliennes, Bull. Soc. Math. France 90 (1962) 323–448.
[39] M. Artin, W. Schelter, Graded algebras of global dimension, 3, Adv. Math. 66 (1987) 171–216.
[40] J. Carlson, S. Müller-Stach, C. Peters, Period Mappings and Period Domains, in: Camb. Studies in Adv. Math., vol. 85, Camb. Univ. Press, 2003.
[41] P.J. Martínez, J.L. Peña, F. Panaite, F. Van Oystaeyen, On iterated twisted tensor products of algebras, Int. J. Math. 19 (9) (2008) 1053–1101.
[42] S. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970) 39–60.
[43] I. Heckenberger, S. Kolb, The locally finite part of the dual coalgebra of quantized irreducible flag manifolds, Proc. Lond. Math. Soc. 89 (2) (2004)
457–484.
[44] R.J. Baston, M.G. Eastwood, The Penrose Transform, Oxford University Press, Oxford, 1989.
[45] S. Helgason, Differential Geometry, Lie Groups and Symmetric Spaces, in: Pure and Appl. Math., vol. 80, Academic Press, New York, 1978.
[46] Y. Soibelman, On quantum flag manifolds, Funct. Anal. Appl. 26 (1992) 225–227.
[47] J.V. Stokman, M. Dijkhuizen, Quantized flag manifolds and irreducible ∗-representations, Comm. Math. Phys. 203 (1999) 297–324.
[48] J.V. Stokman, The quantum orbit method for generalized flag manifolds, Math. Res. Lett. 10 (2003) 469–481.