Parameterized Extended Formulations

• Parameterized extended formulations are polyhedral representations that incorporate auxiliary variables and constraints indexed by problem-specific parameters.
• They improve computational tractability by dynamically adjusting formulation strength via closure operations and leveraging combinatorial features.
• Applications include combinatorial optimization and integer programming, offering polynomial-size representations for structured and complex problems.

A parameterized extended formulation is a polyhedral representation in which auxiliary variables, constraints, or the formulation’s size are explicitly tied to problem-specific parameters rather than only the ambient problem dimension, thus enabling a systematic family of formulations indexed by these parameters. This paradigm leverages combinatorial or structural features—such as subgraph types, decomposition width, family closures, protocol complexity, or hitting numbers—to enhance expressiveness or computational properties. The technique has achieved remarkable generality across combinatorial optimization, polytopal convexification, integer programming, and PDE systems.

1. Definition and Formalism

Given a feasible region $P$—often the convex hull of combinatorial objects, solutions to an integer program, or graphs of nonlinear functions—an extended formulation describes $P$ as the linear image (projection) of a higher-dimensional polyhedron $Q$, typically introducing new variables $y$ and constraints $A x + B y \leq b$. In the parameterized setting, $Q = Q(\theta)$ depends on a tunable parameter vector $\theta$ drawn from a set $\Theta$ encoding aspects such as cycle decompositions, hitting sets, flow weights, or state complexities. Each $\theta$ yields a distinct formulation $Q(\theta)$; families $\{Q(\theta) : \theta \in \Theta\}$ admit closure operations, dominance relations, and algorithmic adaptation (Angulo et al., 2024).

This architecture generalizes canonical formulations (MTZ, DL, SCF for ATSP), enables explicit tracking of polyhedral strength via closure/intersection, and provides new handles for both theoretical analysis (extension complexity) and computational practices (dynamic enhancement or compression).

2. Parametric Families: Representative Examples

a. ATSP: MTZ, DL, SCF Formulations

For the Asymmetric Traveling Salesman Problem on a complete digraph $G=(N,A)$, three classical formulations—Miller-Tucker-Zemlin (MTZ), Desrochers-Laporte (DL), and Single-Commodity Flow (SCF)—are analyzed as parametric families, each indexed by vectors $d \in D$ (for MTZ, DL) or $b \in B$ (for SCF):

• MTZ($d$): Polytope $Q_{MTZ}(d)$, with assignment constraints and $u_i-u_j + d_{ij} \leq 1-x_{ij}$ for all $ij \in A_1$. $d_{ij}$ parameterizes violation budget along cycles, generalizing classical MTZ via $d^{MTZ}_{ij}=1/(n-1)$ (Angulo et al., 2024).
• DL($d$): Extended system with lifted inequalities $$u_i-u_j + x_{ij} + (1-d_{ij}-d_{ji}) x_{ji} \leq 1-d_{ij}$$. Closure encompasses richer classes of subtour cuts.
• SCF($b$): Flow constraints parameterized by $b_i > 0$, with total $b_i$ summing to $1$. This recovers cut-family systems and can produce standard cut inequalities for the DFJ polytope via closure.

Analogous families and their closures exist in extended formulations for matroid base polytopes via hitting numbers (Aprile, 2021), bilinear convexifications via graph structure (Gupte et al., 2017), compressed constraint systems via decision diagrams (Kurokawa et al., 2022), and solution polytopes from dynamic programming state-space size (Oliveira et al., 11 Jan 2026).

3. Closure Operations and Polyhedral Strength

The closure of a parameterized family, e.g., $\overline{P}_{MTZ} = \bigcap_{d\in D} P_{MTZ}(d)$, tightens the feasible region and can exactly recover the strongest possible polyhedral relaxations (e.g., DFJ subtour elimination inequalities for ATSP). Comparative analysis quantifies strength via containment and strictness: $$\overline P_{SCF} \subsetneq \overline P_{DL} \subsetneq \overline P_{MTZ}$$ for $n \geq 5$ (Angulo et al., 2024). These inclusions indicate that parametric closures yield strictly increasing relaxation strength, and their explicit intersection characterizes the optimal face of the assignment polytope or matroid base.

4. Parameterization in Algorithmic Construction

Parameterization directly influences the size and tractability of extended formulations:

• Decision Diagrams: Extended formulations for integer constraint systems leverage compressed zero-suppressed decision diagrams (NZDD/ZDD). The LP size depends on the diagram vertex/edge count, not the number of original constraints, providing exponential compression in highly structured cases (Kurokawa et al., 2022).
• Dynamic Programming Table Size: For vertex-subset polytopes, solution-preserving DP algorithms over decompositions of width $k$ and table size $\alpha(k,n)$ yield extended formulations of size $O(\alpha(k,n) n)$, with ETH-based lower bounds proving this optimal (Oliveira et al., 11 Jan 2026).
• Matroid Hitting Number: In matroid base polytopes, parameterization by hitting number $\tau(M)$ of flacets yields an extended formulation of size $O(\tau(M) n r)$ (Aprile, 2021). This matches classic bounds for graphic matroids when using star-bases.

Formulation Size Table

Problem Class	Parameter	Ext. Form Size
ATSP (MTZ/DL/SCF)	$d$, $b$	$O(n^2)$, $O(n|A|)$
Matroid base polytope	$\tau(M)$	$O(\tau(M)n r)$
Bilinear convexification	graph structure	$O(n^2)$ or linear
Constraint system (NZDD/ZDD)	$|V'|$, $|E'|$	$O(n + |V'|)$
DP solution polytopes	$\alpha(k,n),k$	$O(\alpha(k,n) n)$

5. Applications and Computational Implications

Parameterized extended formulations offer multiple practical advantages:

• Polynomial-size formulations for structured instances (cycles, cacti, cliques, bounded treewidth) where classical convex hulls are exponentially large (Gupte et al., 2017, Kurokawa et al., 2022, Oliveira et al., 11 Jan 2026).
• Reduced computational complexity in branch-and-cut methods, enabling on-the-fly enhancement by adjusting parameters (e.g., adding $d$-vectors in MTZ closure if fractional solutions violate specific subtours) (Angulo et al., 2024).
• Compression and efficient linear programming in high-dimensional learning and combinatorial optimization tasks, notably for SVMs and integer programs with redundancy (Kurokawa et al., 2022).
• Transfer of lower bounds: Intractability results on extension complexity translate to algorithmic lower bounds for DP/state-space approaches, and vice versa (Oliveira et al., 11 Jan 2026).

6. Connections to Related Frameworks

The parameterized viewpoint unifies multiple advanced methodologies:

• Extended Formulations via Communication Protocols: The communication cost in randomized protocols gives size bounds on extended formulations, with matroid polytopes as a premier application (Aprile, 2021).
• Zuckerberg-style Geometric Convexification: Geometric certificates for facet families in bilinear graphs provide tight convex hull characterizations, showing that parameterized structural restrictions suffice for polynomial-size formulations in many cases (Gupte et al., 2017).
• Closure and Intersection: Parameterized families can be closed under intersection to capture combinatorially complete inequality systems (e.g., all subtour cuts for ATSP).
• Dynamic Enhancement: Algorithms can dynamically select or expand parameter sets, optimizing polyhedral strength and cutting plane effectiveness in practical solvers.

7. Open Directions and Research Frontiers

Several open problems persist in parameterized extended formulations:

• For complex graph classes (wheels, Halin graphs, bounded treewidth/cactus), identification of minimal facet-bases or structural parameters for compact formulations remains ongoing (Gupte et al., 2017).
• Extension of dynamic or on-demand parameter selection strategies in cutting-plane and branch-and-cut frameworks could yield improved computational performance (Angulo et al., 2024).
• Further exploration of decision diagram parameterizations for integer and mixed-integer programs with elaborate combinatorial redundancy or symmetries.
• Investigation of the transfer principle between extension complexity and DP table size, aiming for new unconditional lower bounds or sharper thresholds in width-based decomposition algorithms (Oliveira et al., 11 Jan 2026).

In summary, parameterized extended formulations harness structural, combinatorial, and algorithmic properties to yield efficient and expressive representations tailored to problem instances, with fundamental implications for complexity, computation, and polyhedral analysis across a spectrum of mathematical programming domains.