Non-surjective pullbacks of graph C*-algebras from non-injective pushouts of graphs
Alexandru Chirvasitu, Piotr M. Hajac, Mariusz Tobolski

TL;DR
This paper identifies a broad class of graph C*-algebra homomorphisms whose pullbacks are AF algebras, providing a method to analyze quantum spaces via graph modifications, with applications to noncommutative topology.
Contribution
It introduces a new class of pullback constructions for graph C*-algebras that yield AF algebras and extends the results to more general graph C*-algebras involving sinks.
Findings
Pullback C*-algebras are AF in a substantial class of cases.
The results apply to quantum spaces like quantum spheres and teardrops.
An extension of the theorem to sink-extended graph C*-algebras is provided.
Abstract
We find a substantial class of pairs of -homomorphisms between graph C*-algebras of the form whose pullback C*-algebra is an AF graph C*-algebra. Our result can be interpreted as a recipe for determining the quantum space obtained by shrinking a quantum subspace. There is a variety of examples from noncommutative topology, such as quantum complex projective spaces (including the standard Podle\'s quantum sphere) or quantum teardrops, that instantiate the result. Furthermore, to go beyond AF graph C*-algebras, we consider extensions of graphs over sinks and prove an analogous theorem for the thus obtained graph C*-algebras.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Non-surjective pullbacks of graph C*-algebras
from non-injective pushouts of graphs
Alexandru Chirvasitu
SUNY, Buffalo, USA.
,
Piotr M. Hajac
Instytut Matematyczny, Polska Akademia Nauk, ul. Śniadeckich 8, Warszawa, 00-656 Poland; and Department of Mathematics, University of Colorado Boulder, 2300 Colorado Avenue, Boulder, CO 80309-0395, USA
and
Mariusz Tobolski
Instytut Matematyczny, Polska Akademia Nauk, ul. Śniadeckich 8, Warszawa, 00-656 Poland
Abstract.
We find a substantial class of pairs of -homomorphisms between graph C*-algebras of the form
\textstyle{C^{*}(E)\,\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{C^{*}(G)}$$\textstyle{\ignorespaces\ignorespaces\ignorespaces\ignorespaces C^{*}(F)}
whose pullback C*-algebra is an AF graph C*-algebra. Our result can be interpreted as a recipe for determining the quantum space obtained by shrinking a quantum subspace. There is a variety of examples from noncommutative topology, such as quantum complex projective spaces (including the standard Podleś quantum sphere) or quantum teardrops, that instantiate the result. Furthermore, to go beyond AF graph C*-algebras, we consider extensions of graphs over sinks and prove an analogous theorem for the thus obtained graph C*-algebras.
1. Introduction
The classical two-sphere can be obtained by shrinking the boundary of the disc to a point. In other words, there is a pushout diagram in the category of topological spaces
[TABLE]
Due to the contravariant duality of algebras and spaces, the diagram (1.1) amounts to an isomorphism of C*-algebras of complex-valued continuous functions on the two-sphere and the pushout respectively.
At the same time, the Toeplitz algebra [6] can be viewed as a noncommutative deformation of (see [13, Theorem IV.7]). Therefore, the C*-algebra of the standard Podleś sphere [16, (3a)] provides a noncommutative deformation of the diagram (1.1), namely we have the following pullback diagram in the category of C*-algebras
[TABLE]
The aim of this paper is to generalize the above pullback construction using the concept of a C*-algebra of a directed graph (e.g., see [3]). Graph C*-algebras provide powerful tools in noncommutative topology, and many C*-algebras representing noncommutative deformations of topological spaces are isomorphic with C*-algebras of graphs [5, 11, 12]. These isomorphisms are usually quite complicated and they do not depend on the deformation parameter. Nevertheless, when such an isomorphism is established, it is easier to obtain solutions to many problems, especially concerning K-theory.
Our starting point is that all the C*-algebras in the diagram (1.2) can be viewed as C*-algebras of graphs. We present this pictorially as follows:
[TABLE]
(See the examples in Section 2 for details.)
The graph-algebraic decomposition (1.3) manifests a certain general phenomenon that can be explained in terms of non-injective pushouts of graphs. The goal of this paper is to explore this phenomenon to arrive at a general setting. To this end, we search for a new concept of morphisms of graphs, so as to ensure that, in the thus defined category of graphs, the assignment of graph algebras to graphs becomes a contravariant functor translating pushouts of graphs into pullbacks of graphs algebras. While this task seems to be completed in [10] (cf. [14, Corollary 3.4]) for injective pushouts of row-finite graphs (each vertex emits only finitely many edges), herein we handle a non-injective case without row-finiteness assumption.
To accommodate this naturally occuring non-injectivity, we replace the standard idea of mapping vertices to vertices and edges to edges by the more flexible idea of mapping finite paths to finite paths. We arrive at a general result for a class of unital AF graph C*-algebras including the standard Podleś sphere, complex quantum projective spaces [18, Definition on p. 109], and quantum teardrops [4]. Finally, we go beyond AF graph C*-algebras by extending their acyclic graphs over sinks.
2. Graph-algebraic preliminaries
A directed graph is a quadruple , where is the set of vertices, is the set of edges (arrows), and are the source map and the range (target) map respectively. Throughout the paper, we consider only directed graphs with countable sets of vertices and edges, and we will often simply refer to them as graphs.
Definition 2.1** *(Graph C-algebra).
The graph C-algebra of a directed graph is the universal C*-algebra generated by mutually orthogonal projections \big{\{}P_{v}\;|\;v\in E^{0}\big{\}} and partial isometries \big{\{}S_{e}\;|\;e\in E^{1}\big{\}} satisfying the following conditions:*
[TABLE]
A vertex in is called a sink if and only if . A vertex is called regular iff it is not a sink and it emits finitely many edges. A graph is called row finite iff all its vertices are either regular or sinks. By a finite path in we mean a sequence of edges satisfying for all . The length of a path is the number of edges in the sequence. We consider vertices as paths of length zero, and denote the set of finite paths by . The notation , along with the source and the range map, naturally extend to any . As we consider only finite paths throughout this paper, we will simply refer to them as paths.
A path is called a loop if and only if and is not a vertex. We say that a loop is short iff it is an edge.
Definition 2.2**.**
We call a path pointed iff its final edge is not a loop.
We say that a path is a prolongation of a path if and only if for a path such that . We write when is a prolongation of . Observe that gives a partial order on .
Lemma 2.3**.**
Let and be finite paths in an arbitrary graph . Then
[TABLE]
Proof.
Assume that , i.e. that with . Then, as is an element of a linear basis of for any [1, Corollary 1.5.12], we obtain
[TABLE]
Much in the same way, we see that when .
Conversely, assume that for some finite paths
[TABLE]
If , then
[TABLE]
and (GA1) imply that for . This means that . Otherwise, when , we get that . ∎
Next, to make the condition (GA3) easier to check, we prove the following lemma:
Lemma 2.4**.**
Let be an arbitrary graph and a path in with its origin at . Then
[TABLE]
Proof.
Write , where is an edge with its origin at and is an initial subpath of ending at . Then
[TABLE]
where the middle inequality is due to (GA3). Now, the claim follows by the induction on the length of . ∎
To get ready for examples in the last section, we present graph-algebraic presentations of some well-known C*-algebras.
Example 2.5**.**
The algebra of complex numbers is isomorphic with the graph C-algebra of the graph with one vertex and no edges.*
[TABLE]
Example 2.6**.**
The C-algebra of all continuous complex-valued functions on the circle is the universal unital C*-algebra generated by a single unitary . It is isomorphic with the graph C*-algebra of the graph given below through the isomorphism given by .*
[TABLE]
Example 2.7**.**
The Toeplitz algebra [6] is the universal unital C-algebra generated by a single isometry . It is isomorphic with the graph C*-algebra of the graph given below through the isomorphism given by .*
[TABLE]
Example 2.8**.**
The Cuntz algebra [7] is the universal unital C-algebra generated by isometries , , subject to the relation . It is isomorphic with the graph C*-algebra of the graph given below through the isomorphism given by .*
[TABLE]
Example 2.9**.**
Let . The C*-algebra [16, (3a)] of the standard Podleś quantum sphere coincides with the C*-algebra of the Vaksman–Soibelman quantum complex projective line [18, p. 109], which has a graph-algebraic presentation as the graph C*-algebra of the graph given below (see [11, Section 2.3]):*
[TABLE]
Here the arrow decorated by denotes countably infinitely many arrows.**
We end this section by recalling some standard results that we will use throughout the paper. Let be a directed graph. A subset is called hereditary iff, for any such that there is a path starting at and ending at , we have . If is hereditary, then the ideal generated by the projections associated with the elements of is of the form (cf. the equation (1) in [3]):
[TABLE]
Here denotes the closed linear span.
Assume additionally that there are no vertices that emit infinitely many arrows into and finitely many (but not zero) arrows outside of . Assume also that is saturated, i.e. that there does not exist a regular vertex such that . Then, the quotient algebra is again a graph C*-algebra (cf. the discussion below the equation (1) in [3]):
[TABLE]
and and are the restrictions-corestrictions of and respectively.
3. Non-surjective pullbacks of graph C*-algebras
In this section we prove a non-surjective pullback theorem generalizing the diagram (1.2). First, we need some preliminaries on graphs and their morphisms.
Let and be directed graphs. A morphism of graphs is a pair of mappings and satisfying
[TABLE]
If there is an injective morphism of graphs , we say that is a subgraph of and write .
Definition 3.1**.**
An injective graph morphism is called an admissible inclusion iff the following conditions are satisfied:
- (A1)
* is hereditary and saturated,* 2. (A2)
, 3. (A3)
no vertex in emits infinitely many edges into while emitting finitely many (but not zero) edges into .
Next, let us state the following elementary fact (cf. (2.7) and the discussion preceding it).
Proposition 3.2**.**
Let be an admissible inclusion. Then, we have an isomorphism of graph C-algebras*
[TABLE]
To phrase our main result, it is convenient to view graphs as small categories whose objects are vertices and morphisms are finite paths. Then functors between such categories are what we want as morphisms between graphs. Using the thus understood functors as morphisms, we generalize the Cuntz–Krieger graph category [8, p. 172] (cf. [1, Definition 1.6.2]) by allowing egdes to be mapped to finite paths intead of only edges.
Lemma 3.3**.**
Let be a functor between graphs such that:
- (1)
* is compatible with the prolongation relation as follows*
[TABLE] 2. (2)
for any vertex that emits at least one and at most finitely many edges, restricts-corestricts to a bijection
[TABLE]
Then induces a -homomorphism given by
[TABLE]
Proof.
Since graph C*-algebras are universal, it suffices to show that all defining relations are preserved. For starters, since the condition (1) implies the injectivity of , we infer that the set of mutually orthogonal projections is sent to the set of mutually orthogonal projections:
[TABLE]
Next, to show that (GA1) is preserved, it suffices to prove the implication
[TABLE]
which follows from combining Lemma 2.3 with the condition (1). Finally, showing that (GA2) and (GA3) are preserved is also straightforward: the former follows directly from the condition (2) and the latter from Lemma 2.4. ∎
We are now ready to prove our first main result:
Theorem 3.4**.**
Let , , be admissible inclusions of graphs such that
- (1)
* has no loops, has no short loops at vertices in , and , ;* 2. (2)
there is a functor such that: it satisfies the condition (1) in Lemma 3.3, it is on objects, and its image is the set of all pointed paths.
Then the induced -homomorphisms exist and render the diagram
[TABLE]
a pullback diagram of C-algebras. (If is finite, then this is a pullback diagram of unital C*-algebras.) Here and are the canonical surjections (3.2), is a -homomorphism of Lemma 3.3, and is its restriction-corestriction.*
Proof.
We begin by proving that and are well-defined injective -homomorphisms. To see that is well defined, by Lemma 3.3 and the assumption (2), it suffices to check the condition (2) of Lemma 3.3. To this end, take any regular vertex and any edge . Then, as the image of is the set of pointed paths, is a pointed path from to .
Suppose that factorizes through a third vertex . Then we can write , where is a pointed path from to and is a pointed path from to . Indeed, deleting any intial subpath from a pointed path always yields a pointed path, and making all loops based at part of makes a pointed path. Furthermore, as is surjective on the set of pointed paths, we can write . Combining it with the injectivity of , which follows from the condition (1) in Lemma 3.3, we get a contradiction (the edge is not a path factorizing through the vertex ). Hence is a pointed path from to that does not factorize through any third vertex.
If there is a loop in based at , then there are infinitely many non-factorizing pointed paths from to , and (because is a functor) none of them can be the image of a path that factorizes through a third vertex. Consequently, as is bijective when corestricted to the set of pointed paths, and there are no loops in , there must be infinitely many edges in from to , which contradicts the assumption that is a regular vertex in . Hence, there is no loop in based at , so is an edge.
Next, if , then because is an injective functor that is on the set of veritices. Indeed, suppose that , where ’s are edges. Then is of length at least , as cannot be a vertex. Hence , i.e. is an edge, so any edge emitted from in comes from an edge emitted from in . Combining this with the injectivity of and the above established fact that is an edge, we conclude that the condition (2) in Lemma 3.3 is satisfied.
Thus we obtain a well-defined -homomorphism that is injective by [17, Corollary 1.3] because has no loops. Furthermore, by the admissibility condition Definition 3.1(A2), it is clear that restricted to the subgraph corestricts to yielding a restriction-corestriction of . The -homomorphism is injective because is injective.
It is straightforward to check that the maps , , and make the diagram (3.5) commutative. Therefore, as and are surjective and and are injective, due to [15, 3.1 Proposition], to show that (3.5) is a pullback diagram, it suffices to prove that
[TABLE]
To obtain the above inclusion, we use the characterization of ideals associated to hereditary subsets (2.6):
[TABLE]
By the assumption (1), all paths in terminating in are pointed, so they are in the image of . Therefore, as by the assumption (1), we conclude that the inclusion (3.6) holds at the algebraic level. Finally, as any -homomorphism between C*-algebras is a continuous map whose image is closed, we infer the desired inclusion at the C*-level. ∎
4. Extending graphs over sinks
To generalize the diagram (1.2) even further (e.g. to allow loops in in the pullback theorem of the previous section), we first need to determine suitable conditions under which the graph-algebra construction preserves pushouts of graphs over sinks.
The general setup assumptions (GS) are as follows:
- •
and are graphs;
- •
is a set regarded as a graph with no edges;
- •
and are injective maps defining the pushout
[TABLE]
- •
, where is the canonical quotient map.
Next, let and be the induced -homomorphisms (see Lemma 3.3). Define
[TABLE]
Here we divide the amalgamated free product by the ideal generated by the product of non-identified projections.
Lemma 4.1**.**
Assume that at least one of the maps and takes its values in the sinks of the respective graph. Then the natural assignment of elements defines an isomorphism of C-algebras:*
[TABLE]
Proof.
Since or takes values in the sinks of or , respectively, all edge relations in involving vertices in the image of are of one of two types: either they refer to edges only in , or to edges only in . Hence, there are -homomorphisms
[TABLE]
given the natural assignment of elements. Furthermore, as , they induce a surjective -homomorphism
[TABLE]
Finally, as the kernel of coincides with the kernel of the defining surjection
[TABLE]
the claim follows. ∎
Now, consider three graphs , and with injective maps , , and . Assume also that and consist of sinks of the two respective graphs and . Now, consider a C*-algebra homomorphism
[TABLE]
annihilating the vertex projections of . Then and the zero map induce a -homomorphism on the amalgamated product that annihilates the kernel of . Hence, by Lemma 4.1, extends to
[TABLE]
Lemma 4.2**.**
Let be the map defined in (4.3). Then
[TABLE]
Proof.
The inclusion is clear by the construction of . For the other inclusion, note that, as annihilates , it factors as
[TABLE]
where the last map is induced by . The inclusion follows from this factorization. Finally, as
[TABLE]
we infer that
[TABLE]
which ends the proof. ∎
Assume now that, under the general setup assumptions (GS) and the assumptions preceding (4.6), we have a pullback diagram
[TABLE]
of -homomorphism of C*-algebras. Here is an injective -homomorphism sending all vertex projections to vertex projections and all partial isometries associated with edges to partial isometries associated with paths, intertwining with , and sending the projections labeled by to projections labeled by . We also assume that all partial isometries in associated with paths ending in are in the image of . Furthermore, we assume that is a -homomorphism annihilating the vertex projections labeled by , and and are arbitrary C*-algebras fitting into the pullback diagram (4.11) for some -homomorphisms and .
Note that the assumptions made on allow us to define its extension
[TABLE]
Indeed, we can use the isomorphism (4.2) and observe that the conditions on allow us to extend it by to
[TABLE]
This brings us to the second main result of the paper:
Theorem 4.3**.**
Under the general setup assumptions (GS) and the additional assumptions preceding (4.6), the pullback diagram (4.11) of -homomorphisms of C-algebras induces the following pullback diagram of -homomorphisms of C*-algebras:*
[TABLE]
Here and are defined by (4.7), and is defined by (4.13).
Proof.
The commutativity of the diagram (4.14) is immediate by construction. To prove that it is a pullback diagram, first we establish the injectivity of . It follows from: the injectivity of , the assumption that does not annihilate vertex projections, the fact that loops without exit in remain loops without exit both in and , and the general Cuntz–Krieger uniqueness theorem [17, Theorem 1.2].
Next, using the injectivity of and appealing to [15, 3.1 Proposition], we note that to conclude the proof of the theorem, it suffices to check the following two conditions:
[TABLE]
The first condition is immediate from our assumption that (4.11) is a pullback. Indeed, the analogous equation holds for the unprimed maps and , and the images of these maps coincide with those of and respectively because the primed maps are obtained from the unprimed maps by extending them by the zero map on .
To show the second condition, we apply Lemma 4.2 and (4.10) to obtain:
[TABLE]
Here both and are defined as in (4.3). Now, since (4.11) is a pullback diagram, we have . Furthermore, it follows from the construction of and that
[TABLE]
Hence , so, by (4.18), we are left having to argue that
[TABLE]
To this end, note first that
[TABLE]
because and . Furthermore, observe that
[TABLE]
and
[TABLE]
Next, since all partial isometries in associated with paths ending in are in the image of by assumption, we conclude that
[TABLE]
Indeed, it boils down to showing that
[TABLE]
Since any graph C*-algebra is the closed linear span of elements of the form with (see [1, Corollary 1.5.12]), we are looking at such that or can be non-zero. This means that or . However, as is a sink in , we infer that or . Hence and , or and . Furthermore, any -subpath of any path ending in has to end in . Consequently, , so . As for any , we conclude that all elements that can multiply nontrivially with are in the image of , which proves (4.25).
Now, taking advantage of the assumption that for any , we infer that
[TABLE]
Hence, by (4.17),
[TABLE]
Consequently, combining (4.23), (4.24) and (4.27), we arrive at
[TABLE]
which, together with (4.21) and (4.22) proves (4.20). ∎
5. Examples and applications
This section is devoted to the study of special cases of Theorem 3.4 and Theorem 4.3 leading to interesting examples in noncommutative topology.
5.1. The standard Podleś quantum sphere
Observe that the assumptions of Theorem 3.4 are true for the standard Podleś quantum sphere. Here , , and (see the diagram (1.3)).
Our next example generalizes a simple gluing construction in topology. Recall that the real projective plane may be represented as a closed hemisphere with the antipodal points on the equator identified. If we further identify all those antipodal points, we obtain the sphere . Here we present a -deformed analog of this procedure.
The C*-algebra of the quantum real projective plane [9, Section 4] admits a graph-algebraic presentation (see [11, Section 3.2]) as the C*-algebra of the graph given below:
[TABLE]
Due to Theorem 3.4 and , we have the following pullback diagram:
[TABLE]
Observe that the diagram (5.2) reflects the aforementioned procedure of shrinking the copy of inside to a point.
5.2. The quantum teardrop
The classical teardrop may be represented as the wedge of two spheres, namely we have the following pushout diagram:
[TABLE]
To obtain a noncommutative counterpart of the diagram (5.3), we need to introduce a different kind of a noncommutative sphere. The C*-algebra of the equatorial Podleś quantum sphere [16, (3b)] admits a graph-algebraic presentation (see [11, Section 3.1]) as the C*-algebra of the graph given below.
[TABLE]
Theorem 3.4 applies and we obtain the pullback diagram
[TABLE]
which can be regarded as a noncommutative deformation of the diagram (5.3).
5.3. The quantum complex projective spaces
The CW-complex decomposition of complex projectives spaces may be described in terms of pushout diagrams
[TABLE]
Let us recall the graph-algebraic presentation of -deformations of the spaces in the diagram (5.6).
- •
The C*-algebra of the quantum complex projective space [18] is the graph C*-algebra of a graph that, for , is given below (see [11, Section 4.3]):
[TABLE]
- •
The C*-algebra of the Hong–Szymański quantum even-dimensional ball [12] is the graph C*-algebra of a graph that, for , is given below (see [12, Section 3.1]):
[TABLE]
- •
The C*-algebra of the Vaskman–Soibelman quantum odd-dimensional sphere [18, Definition on p. 106] is the graph C*-algebra of a graph that, for , is given below (see [11, Section 4.1]):
[TABLE]
Applying Theorem 3.4, we obtain the pullback diagram
[TABLE]
Note that the diagram (5.10) was obtained in [2, Proposition 4.1] using equivariant pullback structures.
5.4. The quantum teardrops
Let . Consider the following graph :
[TABLE]
Observe that . Moreover, one can show (see [5, Section 3]) that, in general, the graph C*-algebra is isomorphic with the C*-algebra [4, Section 3]. We will also need the following -sink extension of the graph (2.5):
[TABLE]
Here the notation means that there are many edges from to . Now, due to Theorem 3.4, we obtain the pullback diagram
[TABLE]
Let us now consider the graph defined as a pushout of (see (5.11)) and an another graph over the sinks of . The only restriction on the graph is that there exists an inclusion . The graph is represented pictorially as follows:
[TABLE]
Next, we consider an analogous construction for the graph (see (5.12)) using the same graph , and we denote the resulting graph by . The graph is represented pictorially as follows:
[TABLE]
Theorem 4.3 applies and, for any , we obtain the following pullback diagram:
[TABLE]
Acknowledgement
The work on this project was partially supported by NCN grant 2015/19/B/ST1/03098 (Piotr M. Hajac, Mariusz Tobolski) and by NSF grant DMS-1801011 (Alexandru Chirvasitu). It is a pleasure to thank Sarah Reznikoff for a helpful discussion. P.M.H. is also grateful to SUNY Buffalo for its hospitality and financial support.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] Abrams, Gene; Ara, Pere; Siles Molina, Mercedes. Leavitt path algebras . Lecture Notes in Mathematics, 2191. Springer, London, 2017.
- 2[2] Arici, Francesca; D’Andrea, Francesco; Hajac, Piotr M.; Tobolski, Mariusz. An equivariant pullback structure of trimmable graph C*-algebras. To appear in J. Noncommut. Geom.
- 3[3] Bates, Teresa; Hong, Jeong Hee; Raeburn, Iain; Szymański, Wojciech. The ideal structure of the C*-algebras of infinite graphs. Illinois J. Math., 46 (4), pp. 1159–1176, 2002.
- 4[4] Brzeziński, Tomasz; Fairfax, Simon A. Quantum teardrops. Comm. Math. Phys. , 316 (1), pp. 151–170, 2012.
- 5[5] Brzeziński, Tomasz; Szymański, Wojciech. The C*-algebras of quantum lens and weighted projective spaces. J. Noncommut. Geom. , 12 (1), pp. 195–215, 2018.
- 6[6] Coburn, Lewis A. The C*-algebra generated by an isometry. Bull. Amer. Math. Soc. , 73, pp. 722–726, 1967.
- 7[7] Cuntz, Joachim. Simple C ∗ superscript 𝐶 C^{*} -algebras generated by isometries. Comm. Math. Phys. , 57 (2), pp. 173–185, 1977.
- 8[8] Goodearl, K. R. Leavitt path algebras and direct limits. Rings, modules and representations, 165–187, Contemp. Math. , 480, Amer. Math. Soc. , Providence, RI, 2009.
