A Structural Equivalence of Symmetric TSP to a Constrained Group Steiner Tree Problem

We present a brief structural equivalence between the symmetric TSP and a constrained Group Steiner Tree Problem (cGSTP) defined on a simplicial incidence graph. Given the complete weighted graph on the city set V, we form the bipartite incidence graph between triangles and edges. Selecting an admissible, disk-like set of triangles induces a unique boundary cycle. With global connectivity and local regularity constraints, maximizing net weight in the cGSTP is exactly equivalent to minimizing the TSP tour length.

💡 Research Summary

The paper establishes an exact structural equivalence between the symmetric Traveling Salesman Problem (TSP) and a constrained variant of the Group Steiner Tree Problem (cGSTP) defined on a simplicial incidence graph. Starting from a complete weighted graph G=(V,E) with symmetric edge lengths L_e, the authors construct a bipartite incidence graph B=(U∪W,A) where U consists of all possible triangles (3‑vertex subsets of V) and W consists of the original edges. An arc (t,e)∈A indicates that edge e belongs to triangle t.

In the cGSTP formulation each edge node e carries a cost c(e)=2·L_e, while each incidence arc (t,e) yields a profit p(t,e)=L_e. Binary variables x_t, y_e, and z_{t,e} indicate whether a triangle, an edge, or an incidence is selected. The objective is to maximize the net profit

 W(B′)=∑{(t,e)∈A} z{t,e}·L_e – ∑_{e∈W} y_e·2L_e.

Five families of constraints enforce a “disk‑like” selection of triangles:

• (C1) Terminal degree – every active triangle must use all three of its edges (z_{t,e}≤y_e, z_{t,e}≤x_t, ∑{e⊂t}z{t,e}=3x_t).
• (C2) Manifold regularity – each primal edge may belong to at most two selected triangles (y_e ≤ ∑{t⊃e}z{t,e} ≤ 2y_e).
• (C3) Global Euler counts – the numbers of selected triangles and edges satisfy |K|=n−2 and |∂K|=2n−3, which is exactly the Euler characteristic of a triangulated disk with n vertices.
• (C4) Global connectivity – the induced subgraph B′ must be a tree, guaranteeing that the selected triangles form a connected complex.
• (C5) Local connectivity – for every city v, the subgraph H_v induced by selected triangles and edges incident to v must satisfy χ(H_v)=|V(H_v)|−|E(H_v)|=1, i.e., the vertex link is a simple path (no bow‑ties or cycles).

Lemma 1 shows that under (C1)–(C5) the net profit equals the negative total length of the boundary edges ∂K, because internal edges are incident to two triangles (profit 2L_e, cost 2L_e → net 0) while boundary edges are incident to exactly one triangle (profit L_e, cost 2L_e → net –L_e). Lemma 2 proves that any feasible solution defines a connected simplicial surface whose boundary is a single simple Hamiltonian cycle. The proof uses (C3) to guarantee the surface is a topological disk (Euler characteristic 1) and (C5) to ensure every vertex appears on the boundary.

Theorem 3 combines the lemmas to establish the equivalence: the optimal TSP tour length equals the negative of the optimal cGSTP objective value (OPT_TSP = –OPT_cGSTP). The “completeness” direction constructs, from any Hamiltonian cycle C, a set K of n−2 triangles that fill the interior of C, sets the corresponding variables to 1, and shows that this yields a feasible cGSTP solution with objective –L(C).

In the discussion the authors highlight several implications. First, the construction is input‑independent: when the full set of triangles is available the reduction is exact; when only a sparse geometric complex (e.g., Delaunay triangulation) is used, the model becomes a controlled heuristic that remains exact if the optimal tour lies within the chosen complex. Second, the global tree constraint together with local Euler regularity naturally enforce the cancellation of internal edge costs, making the tour objective emerge automatically from local profit‑cost balances. This suggests new avenues for integer programming or meta‑heuristic formulations of TSP that exploit the richer combinatorial structure of simplicial complexes. Finally, the paper leaves algorithmic development and computational experiments as future work, inviting researchers to explore whether the cGSTP perspective can lead to practical improvements for solving large‑scale symmetric TSP instances.