Chapter 23Research Directions
This chapter outlines research directions that can shape the next generation of GQL systems and standards, with an emphasis on topics where formal foundations and engineering practice still diverge. We organize the discussion around core themes—expressiveness vs. complexity, query processing, updates and transactions, graph-to-graph compositionality, interoperability, and schema management—that repeatedly emerge in recent surveys and theoretical analyses of property-graph querying. Rather than prescribing a single roadmap, the chapter highlights open problems and promising alternatives that can be studied independently yet ultimately fit together into a coherent, implementable, and optimizable language design.
23.1 Expressiveness and Complexity
Understanding the trade-offs between the expressiveness and complexity of GQL queries is crucial (Angles et al. 2017; Gheerbrant et al. 2025b). Similar to the rich landscape of fragments and extensions in relational query languages, GQL requires a theoretical foundation to evaluate its expressiveness and computational complexity (Rogova 2024; Figueira et al. 2025). Researchers should focus on developing specific fragments of GQL, such as conjunctive regular path queries (CRPQs), and their extensions with counting, aggregation, and recursion (Barceló 2013; Gheerbrant et al. 2025a). They should also study the expressiveness of GQL’s complex label expressions and path modes, including how they can be leveraged to solve real-world problems efficiently (Deutsch et al. 2022; Francis et al. 2023a). Analyzing the computational complexity of various GQL operations, particularly in pattern matching and aggregation, is also essential.
Recent theoretical work on GQL and related graph querying formalisms has begun to provide abstract models for these languages, clarifying their expressive power and limitations as primarily graph-to-relation query systems (Francis et al. 2023a; Gheerbrant et al. 2025b). A natural research direction is to refine such models so that they also capture genuinely graph-to-graph constructs, compositional query algebras, and the interaction between path modes, label expressions, and schema constraints (Angles et al. 2017; Kondylakis et al. 2025). This would enable a systematic comparison of GQL fragments with existing formalisms, such as CRPQs and their extensions, and support formal frameworks for query rewriting and optimization.
23.2 Query Processing and Optimization
Efficient query processing and optimization are central to the practical use of GQL. There is a need for advanced algorithms and data structures that can handle complex graph queries, both in native graph engines and in relational implementations of SQL/PGQ (Angles et al. 2017; Deutsch et al. 2022). Research in this area should focus on developing theories and algorithms for query equivalence and containment, which are essential for optimizing GQL queries (Gheerbrant et al. 2025b). Designing and implementing efficient algorithms for GQL query processing, particularly for native graph database systems and relational implementations, is crucial. Additionally, exploring innovative indexing and storage techniques that can enhance the performance of GQL queries, especially in large-scale graph datasets, is important (Kondylakis et al. 2025).
Schema-aware containment provides a concrete basis for semantic optimization, because inclusion and equivalence need only hold over graphs conforming to the schema. Recent results prove -completeness for unions of conjunctive regular path queries (UCRPQs) under or schemas and for simple unions of conjunctive two-way regular path queries (UC2RPQs) under ; they also give a upper bound for UC2RPQs under schemas without participation constraints (Gutiérrez-Basulto et al. 2024).
Beyond traditional batch workloads, scalable processing of continuous and streaming graph queries remains an open challenge. Recent work on window-based continuous queries and stream processing for property graphs indicates that even relatively simple patterns can be difficult to evaluate efficiently under high update rates (Rost et al. 2024). Future research should therefore investigate cost models, indexing schemes, and execution architectures tailored to continuous GQL queries, including hybrid engines that jointly optimize over static and streaming graph data (Kondylakis et al. 2025).
23.3 Design Decisions and Alternatives
The current design of GQL includes specific decisions that might benefit from further exploration and validation. Academics should investigate alternative approaches and their implications, informed by emerging formal models of GQL and SQL/PGQ (Francis et al. 2023a; Gheerbrant et al. 2025b). This includes examining the implications of allowing concatenation of patterns containing different kinds of variables, which could enable more complex and flexible queries. Another area of interest is developing automaton models that incorporate group variables, enabling classical evaluation techniques and automata-theoretic constructions, such as product and determinization, in the spirit of existing work on graph pattern matching and regular path queries (Deutsch et al. 2022; Gheerbrant et al. 2025a). It is also important to conduct empirical studies to validate current design decisions in GQL and propose alternatives that could enhance its usability and functionality.
23.4 Updates and Transaction Processing
The update and transaction processing capabilities of GQL are critical for maintaining data integrity and supporting dynamic graph modifications. Research should investigate the design of robust update languages that can efficiently handle complex graph updates, similar to SQL’s transaction processing facilities, but adapted to property graph workloads. Exploring mechanisms for ensuring consistency and concurrency control in graph updates and preventing issues such as inconsistent changes and non-determinism is also necessary. Understanding how isolation levels, transactional semantics, and failure handling should be adapted or extended for graph workloads is another promising direction, complementing the current focus of the standards and surveys on query and schema aspects (Gheerbrant et al. 2025b; Kondylakis et al. 2025).
23.5 Graph-to-Graph Queries
One of the significant challenges in graph databases is designing a language that supports graph-to-graph queries, enabling composability and advanced query capabilities. Researchers should focus on designing languages that allow queries to produce graphs as output, supporting concepts like views and subqueries at the graph level rather than solely at the table level (Angles et al. 2017; Francis et al. 2023b; Bonifati 2025). Ensuring that the language supports the composability of queries, and allowing the output of one query to be used as the input for another, is also crucial. Investigating the algebraic properties of such graph-to-graph operators and their relationship to existing GQL constructs, path expressions, and prospective graph builders and graph views is an important aspect of this line of work (Deutsch et al. 2022; Gheerbrant et al. 2025b). Connections to work on graph-to-graph transformations between property graphs and RDF further motivate this direction (Bruyat et al. 2024).
23.6 Interoperability and Data Integration
Interoperability between property graphs and other data models, particularly RDF and relational data, is a central challenge for the broader adoption of graph technologies. One research direction is the design of meta-models that can faithfully capture both RDF and property graphs, enabling uniform reasoning and transformation across models (Angles et al. 2021; Angles et al. 2023). Complementary to this, semantics-preserving transformation frameworks are needed to translate RDF graphs into property graphs and back, while respecting schema information, constraints, and inference rules (Bruyat et al. 2024). Such frameworks should account for the different typing, labeling, and identity mechanisms in the underlying models, and be closely aligned with the evolving property graph standards (Kondylakis et al. 2025).
Three direct mappings have been formalized: a simple mapping for schema-less RDF data, a schema-independent mapping to a generic property graph schema, and a schema-dependent mapping of complete RDF schema–instance pairs. The latter two preserve both semantics and information; under their formalization, this provides a basis for lossless RDF-to-property-graph translation and shows that the property-graph model subsumes the information capacity of RDF (Angles et al. 2020).
The S3PG approach extends this line by using SHACL and PG-Schema to coordinate schema and data transformations. It preserves RDF instance data and supported SHACL integrity constraints and is monotonic under source changes; it achieved query-result accuracy on the evaluated workloads, while open problems include RDF-star, ontological definitions and inference, automated query translation, and optimization of generated property graphs (Rabbani et al. 2024).
A closely related topic is maintaining graph views across heterogeneous systems. As applications increasingly combine RDF knowledge graphs, property graphs, and relational data, there is a need for incremental, low-overhead mechanisms that maintain the consistency of derived or materialized graph views as the underlying sources evolve. Understanding how GQL graph-to-graph queries and transformation rules can serve as a unifying abstraction for such integration tasks, and how they can interoperate with schema languages such as PG-Schema and PG-Keys, is a promising research direction (Angles et al. 2021; Angles et al. 2023; Kondylakis et al. 2025).
23.7 Schema and Metadata Management
Effective schema and metadata management are essential for maintaining data integrity and facilitating advanced query optimization. Research in this area should include developing a more comprehensive schema language for GQL, with richer constraints and validations beyond the current key and unique constraints, as advocated by PG-Schema and PG-Keys (Angles et al. 2021; Angles et al. 2023). Investigating techniques for semantic query optimization based on schema information and enhancing the efficiency and accuracy of query processing are important (Gheerbrant et al. 2025b). Exploring methods for incremental validation of schemas and constraints and ensuring that changes to the graph data are validated efficiently is also essential.
Validation complexity is an important boundary for extensions to GQL’s schema facilities. For PG-Schema graph types without integrity constraints, combined complexity is -complete and data complexity is in ; combined complexity drops to when alternation between type combinations and unions is suitably restricted (Ciszewski et al. 2026).
A broader comparison of PG-Schema, ShEx, and SHACL shows that graph schemas should be studied not only for validation but also as inputs to optimization, data exchange, interoperability, and schema discovery. The comparison identifies central design questions relevant to GQL: whether constraints should be expressed in a controlled language or backed by an injected query language, and how expressive power should be divided between types and constraints (Hidders and Murlak 2026).
A complementary line of work is also important, focusing on the automatic discovery of schemas for property graphs. Existing approaches combine label-based and property-based heuristics, as well as probabilistic clustering, to infer hierarchical schemas that summarize large, heterogeneous graphs (Lbath et al. 2021; Bonifati et al. 2022). More recent work applies formal concept analysis to schema discovery and merging, yielding methods that can cope with multiple and evolving graphs while remaining interpretable for human designers. A key research problem is to design hybrid symbolic–statistical methods that remain robust under schema drift and graph evolution, and that can propose candidate GQL schemas or constraints that are both expressive and understandable to human designers.
Closely related is the largely unexplored topic of normalization for property graphs. While normalization theory is well-developed for relational schemas, only initial proposals exist for defining normal forms in the property graph setting. Future work should investigate how key constraints, cardinalities, and higher-arity relationships interact with redundancy, and how normalization procedures can be formulated to respect the graph structure and be expressed using GQL DDL and graph-to-graph transformations (Angles et al. 2021; Angles et al. 2023; Kondylakis et al. 2025).