Towards Design of System Hierarchy (research survey)

The paper addresses design/building frameworks for some kinds of tree-like and hierarchical structures of systems. The following approaches are examined: (1) expert-based procedures, (2) hierarchical clustering; (3) spanning problems (e.g., minimum spanning tree, minimum Steiner tree, maximum leaf spanning tree problem; (4) design of organizational ‘optimal’ hierarchies; (5) design of multi-layer (e.g., three-layer) k-connected network; (6) modification of hierarchies or networks: (i) modification of tree via condensing of neighbor nodes, (ii) hotlink assignment, (iii) transformation of tree into Steiner tree, (iv) restructuring as modification of an initial structural solution into a solution that is the most close to a goal solution while taking into account a cost of the modification. Combinatorial optimization problems are considered as basic ones (e.g., classification, knapsack problem, multiple choice problem, assignment problem). Some numerical examples illustrate the suggested problems and solving frameworks.

💡 Research Summary

The paper presents a comprehensive survey of methods for designing and constructing tree‑like and hierarchical system structures. It begins by emphasizing the growing importance of hierarchical architectures in complex engineered systems, where efficient design directly impacts cost, performance, and maintainability. Six major families of approaches are examined.

1. Expert‑Based Procedures – Traditional, knowledge‑driven techniques where domain specialists manually define components and their hierarchical relationships. While fast for small‑scale problems, they suffer from subjectivity and scalability issues.

2. Hierarchical Clustering – Data‑driven bottom‑up or top‑down clustering algorithms (Ward’s method, average linkage, complete linkage) are used to infer a hierarchy from similarity or distance matrices. The paper discusses how the choice of linkage, the number of clusters, and distance metrics shape the resulting tree, and it provides experimental comparisons that illustrate trade‑offs between intra‑cluster cohesion and overall tree depth.

3. Spanning‑Tree Problems – Classical combinatorial models are adapted to hierarchical design. Minimum Spanning Tree (MST) minimizes total link cost; Minimum Steiner Tree (SMT) allows insertion of auxiliary Steiner nodes to further reduce overall distance, albeit at NP‑hard complexity, for which 2‑approximation and meta‑heuristic solutions are surveyed. The Maximum‑Leaf Spanning Tree problem seeks to maximize the number of leaf nodes, thereby enhancing extensibility. The authors apply each model to realistic network‑design case studies, comparing cost, total length, and leaf‑node count.

4. Optimal Organizational Hierarchies – A multi‑objective formulation incorporates hierarchy depth, span of control, communication overhead, and security constraints. Role and resource allocation are modeled as a combination of 0‑1 knapsack and multiple‑choice problems, leading to integer linear programs whose Pareto‑optimal solutions are explored. Experiments on corporate structures and IT service management demonstrate measurable reductions in management overhead and communication latency relative to conventional flat or ad‑hoc hierarchies.

5. Multi‑Layer k‑Connected Networks – The paper addresses the design of resilient three‑layer networks (core, distribution, access) that guarantee k‑connectivity between any pair of nodes. By integrating minimum‑cost flow, graph covering, and k‑connectivity verification algorithms, the authors produce designs that maintain service continuity under failures while minimizing construction cost.

6. Modification and Restructuring of Existing Hierarchies – Four transformation techniques are investigated: (i) condensing adjacent nodes to reduce depth and search cost; (ii) hot‑link assignment, adding direct shortcuts to improve accessibility; (iii) converting a tree into a Steiner tree by inserting auxiliary nodes; and (iv) a general restructuring model that minimizes the “edit distance” between an initial solution and a target solution subject to a bounded modification cost. The restructuring problem is cast as a constrained 0‑1 integer program or tackled with evolutionary algorithms and simulated annealing. Numerical examples illustrate how each technique can be employed during system upgrades, refactoring, or adaptation to new requirements.

Across all sections, the authors emphasize that these seemingly disparate problems share a common combinatorial‑optimization core: classification, knapsack, multiple‑choice, and assignment problems. By recognizing this commonality, designers can reuse algorithmic frameworks, exploit existing approximation guarantees, and integrate multiple objectives more systematically.

The concluding discussion synthesizes the six families into a unified design framework, recommending a selection matrix that matches system scale, complexity, and performance criteria to the most appropriate method(s). It also outlines future research directions, including real‑time hierarchical reconfiguration in dynamic environments, efficient multi‑objective approximation schemes, and the development of human‑AI collaborative tools that blend expert intuition with algorithmic rigor.

Overall, the paper provides both a theoretical taxonomy of hierarchical design problems and practical solution pathways, making it a valuable reference for researchers and practitioners seeking to construct cost‑effective, scalable, and adaptable hierarchical systems.