Brick: Graph Theory, Assembly, and Informatics

• Brick is a nonbipartite matching-covered graph defined by tight cut properties and critical in perfect matching decompositions.
• Algorithmic approaches leverage bricks for generative assembly, using graph-backed and geometry-conditioned models to enhance construction processes.
• The Brick schema integrates building metadata for scalable semantic queries, digital twin development, and advanced HVAC system analytics.

A brick is a foundational and extensively studied structure in graph theory with exacting definitions, crucial roles in decomposition theory, and deep connections to matching theory, combinatorics, chemistry, algorithmic generation, building informatics, astrophysics, and distributed systems. The term also names recent methodologies and systems in machine learning, semantics, and decentralized protocols. The following article provides a comprehensive treatment of the brick across these domains, emphasizing the precise mathematical context, classification results, algorithmic formulations, recent methodological advances, and core applications.

1. Formal Definition and Structural Properties

A brick is a graph-theoretic object arising in the structure theory of matchings. Formally, in the context of finite, loopless, possibly multi-edged graphs, the canonical definition is as follows:

• A matching-covered graph is a connected graph in which every edge is contained in at least one perfect matching, i.e., a set $M \subseteq E(G)$ such that every vertex of $G$ is incident to exactly one edge in $M$.
• A tight cut in a matching-covered graph is an edge-cut $\partial_G(X)$ where every perfect matching meets $\partial_G(X)$ in exactly one edge. A tight cut is nontrivial if neither side has order 1.
• A brick is a nonbipartite matching-covered graph with no nontrivial tight cuts. Equivalently (Edmonds–Lovász–Plummer), a graph $G$ is a brick if and only if $G$ is 3-connected and bicritical: for every pair of distinct vertices $u, v$, the graph $G-u-v$ admits a perfect matching. This places bricks as the 3-connected “atoms” of the tight-cut decomposition of nonbipartite matching-covered graphs (Dai et al., 25 Jun 2026, Lu et al., 2024).

Within this framework, a key dichotomy arises between solid bricks (those with no nontrivial separating cuts/solidity) and nonsolid bricks (admitting at least one such cut). Solidity plays a central role in extremality results and structural characterizations.

2. Extremal Matching Properties and Classification Results

A central question in matching theory concerns the enumeration and structure of perfect matchings in bricks:

• Odd wheels: For $k \geq 3$, the wheel $G$0 consists of a cycle of length $G$1 with an additional hub vertex adjacent to all $G$2 rim vertices. If $G$3 is odd, $G$4 is a nonbipartite solid brick and possesses exactly $G$5 perfect matchings, each corresponding to the unique arrangement of spokes in a matching (Dai et al., 25 Jun 2026, Lu et al., 2024).
• Extremal bricks: Bricks attaining the minimal bound $G$6 on the number of perfect matchings are called extremal. Carvalho–Lucchesi–Murty showed that every solid brick on $G$7 vertices is extremal if and only if it is an odd wheel.
• Counterexamples to lower bounds: Lucchesi and Murty conjectured that for sufficiently large $G$8, every brick on $G$9 vertices has at least $M$0 perfect matchings; this was disproven by Lu and Pan (2024), who constructed infinite families of nonsolid bricks with only $M$1 perfect matchings, demonstrating that nontrivial separating cuts fundamentally reduce matching counts (Lu et al., 2024).

A major recent result by Dai, Lu, and Zhang establishes that every vertex of a brick is incident with a forcing edge (one contained in exactly one perfect matching) if and only if the brick is an odd wheel (up to multiple edges). Forcing edges—a concept rooted in chemical graph theory—provide extremal structure: in an odd wheel, every spoke is forcing, ensuring every vertex meets the condition; no other brick (including all nonwheel bricks) achieves this (Dai et al., 25 Jun 2026).

3. Decomposition Theory and Planar Restrictions

Bricks are fundamental to the decomposition theory of matching-covered graphs, particularly through the tight‐cut and separating‐cut frameworks:

• Wheel-like bricks: These are bricks in which there exists a hub vertex such that every removable class (removable edge or removable doubleton) contains an edge incident to the hub. Lucchesi and Murty conjectured and recent work proved that every planar wheel-like brick is an odd wheel, reinforcing the universality of wheels in the planar, solid regime (Lu et al., 2024).
• Splicing and triangle-insertion: Nonsolid bricks may be constructed by splicing two bricks along corresponding degree-matched vertices or by triangle-insertion (removing a degree-3 vertex and inserting a triangle). These operations preserve the brick property under sufficient conditions and are key to constructing extremal, nonsolid counterexamples to matching bound conjectures (Lu et al., 2024).
• Planarity constraints: It is possible to construct nonwheel, nonplanar wheel-like bricks by splicing odd wheels, but planarity forces bricks into the odd wheel family, as otherwise the resulting graph contains a $M$2 subdivision, violating Kuratowski’s theorem (Lu et al., 2024).

Bricks thus define the maximal, indecomposable nonbipartite structures with respect to perfect matching decompositions and anchor the classification of removable-edge-rich graphs.

4. Algorithmic Generation and Machine Learning with Bricks

Bricks have emerged as a key object in algorithmic generation, particularly in the context of physical and virtual assembly with discrete modules:

• Graph-backed generative assembly: The BrickNet framework initiates large-scale generative modeling of arbitrary brick assembly sequences, using a massive corpus of human-designed LDraw scenes and a graph representation in which each brick is a vertex and its connection (stud, hinge, axle, ball joint, or fixed link) is an annotated edge. Autoregressive models (up to 14B parameters) trained on this representation outperform voxel- or pose-based alternatives, generating build programs that are physically grounded and connectivity-valid for up to ~94 steps before failure (Kulits et al., 24 Apr 2026).
• Geometry-conditioned generation: BrickAnything extends autoregressive brick generation to be directly conditioned on explicit 3D geometry (point clouds), using structure-aware tree tokenization to serialize brick assemblies as local attachment trees. This formulation, coupled with validity-constrained decoding and rollback, yields stable and geometrically faithful assemblies with significantly fewer invalid rollbacks compared to lexicographic or flat tokenizations. Preference-based fine-tuning (DPO) and explicit per-brick static equilibrium testing ensure buildability criteria are met at generation time (Ni et al., 25 May 2026).

Both approaches reinforce the importance of explicit connection semantics, local dependency structures, and physical feasibility in modeling brick-based assemblies for AI-based design and robotic construction.

5. Semantic Metadata, Informatics, and Standardization

The concept of "Brick" extends to ontological schemas for representing building metadata:

• Brick Schema: Brick is a machine-interpretable, ontology-based metadata schema for the built environment. It provides classes, entities, and relationships in an OWL/RDF framework, allowing for scalable, portable analytical queries (e.g., via SPARQL) over diverse building subsystems. Notably, the recent extension for variable refrigerant flow (VRF) systems establishes VRF units, refrigerant circuits, and component roles as explicit Brick schema classes, enabling precise semantic representation and analytic interoperability across domains (Li et al., 2021).
• Service framework and analytics: Brick’s modular design allows rapid integration of novel HVAC subsystems, supports graph-based analytics, and underpins digital twins and automated control/fault detection strategies in facility management.

This data-centric perspective on "brick" aligns with increasing demands for interoperable, machine-actionable building information models and next-generation smart building infrastructure.

6. Applications Across Physics, Distributed Systems, and AI Routing

Bricks also occur as the namesake in diverse scientific and technical applications:

• Astrophysics—The Brick molecular cloud: The term “the Brick” (G0.253+0.016) refers to a massive infrared-dark molecular cloud in the Milky Way’s central molecular zone (CMZ), characterized by extreme density and relevance for star formation. Kinematic measurements of stellar proper motions demonstrate that the Brick is co-spatial with the nuclear stellar disk, forming part of the CMZ, and confirming its direct exposure to the dynamical environment of the galactic center (Martínez-Arranz et al., 2022).
• Blockchains—BRICK payment channels: In distributed systems, BRICK is the first off-chain payment channel protocol that preserves safety under network asynchrony. It uses a rational committee ("wardens") for conflict resolution, combining consistent broadcast for update sequencing and a strict incentive structure for honest behavior, and supports auditability (BRICK+) for permissioned blockchains. This approach achieves sub-second latency, dominant-strategy incentive compatibility, and strong cryptoeconomic guarantees (Avarikioti et al., 2019).
• LLM routing—BRICK for Mixture-of-Models: BRICK also designates a multimodal router for mixture-of-models deployment, assigning queries to models based on a geometric routing rule balancing per-capability matching, difficulty scaling, and dollar cost. A continuous profile knob interpolates between maximal quality and minimal spend, outperforming both single-model baselines and prior routers on cost-quality tradeoffs and latency (Massa et al., 11 Jun 2026).

These deployments highlight the versatility of the brick concept, spanning from physical science to protocol engineering and AI systems.

7. Future Directions, Open Questions, and Significance

Multiple directions remain prominent across the literature:

• Matching theory: Tighter characterizations for the distribution and structure of forcing edges, removable classes, and perfect matchings in arbitrary bricks; the extension of vertex-forcing and removable-edge conditions to “most” rather than “all” vertices; analogous classifications in bipartite matching-covered graphs (braces); and extremality under further connectivity or regularity constraints (Dai et al., 25 Jun 2026, Lu et al., 2024, Lu et al., 2024).
• Algorithmic assembly: Expanding graph-backed and geometry-conditioned frameworks to larger or more diverse part libraries, enhancing validity and stability checkers with differentiable solvers or interactive/real-world data, and integrating human or robotic assembly strategies for physical realization (Kulits et al., 24 Apr 2026, Ni et al., 25 May 2026).
• Semantic informatics: Further extension of the Brick schema to cover additional building systems, richer inter-component relations, and solid integration with analytics, controls, and digital twin ecosystems (Li et al., 2021).
• Systems and routing: Bayesian skill estimation for sparse mixture pools, domain-conditioned query difficulty, and thermodynamic analogues in cost-quality tradeoff optimization for model selection (Massa et al., 11 Jun 2026).
• Observational astrophysics and chemistry: Continued study of “brick”-like structures (e.g., the Brick cloud) in astrophysical and chemical networks, leveraging advances in precision measurement and automated classification (Martínez-Arranz et al., 2022).

Bricks, in their various technical guises, thus remain both central and generative objects in contemporary mathematical, algorithmic, and systems research.