Clausal Deletion Backdoors for QBF: a Parameterized Complexity Approach

Abstract: Determining the validity of a quantified Boolean formula (QBF) is a PSPACE-complete problem with rich expressive power. Despite interest in efficient solvers, there is, compared to problems in NP, a lack of positive theoretical results, and in the parameterized complexity setting one often has to restrict the quantifier prefix (e.g., bounding alternations) to obtain fixed parameter tractability (FPT). We propose a new parameter: the number of variables in clauses that has to be removed before reaching a tractable class (a clause covering (CC) backdoor). We are then interested in solving QBF in FPT time given a CC-backdoor of size $k$. We consider the three classical, tractable cases of QBF as base classes: Horn, 2-CNF, and linear equations. We establish W[1]-hardness for Horn but prove FPT for the others, and prove that in a precise, algebraic sense, we are only missing one important case for a full dichotomy. Our algorithms are non-trivial and depend on propagation, and Gaussian elimination, respectively, and are comparably unexplored for QBF.

• The paper introduces Clause Covering Backdoors (CC-backdoors) for QBF to address challenges posed by quantifier interleavings.
• The authors develop FPT algorithms using branching for 2-CNF formulas and Gaussian elimination for affine formulas, achieving optimal performance under SETH.
• The work establishes a clear dichotomy by proving W[1]-hardness for Horn classes and providing actionable insights for practical QBF solving.

Parameterized Complexity and Clause Covering Backdoors for QBF

Introduction

Quantified Boolean Formula (QBF) validity is a canonical PSPACE-complete decision problem with substantial expressive power for both theoretical complexity and practical applications, notably in AI, planning, and verification. While SAT solvers and parameterized algorithms achieve impressive results for NP-complete instances, PSPACE-complete problems such as QBF remain resistant to similar tractability results. Classical parameters such as treewidth have limited applicability, and existing backdoor methods for SAT do not easily transfer to QBF due to the quantifier structure.

This paper introduces a novel parameterization: the Clause Covering Backdoor (CC-backdoor). This parameter measures the minimum number of variables required to cover all "out-of-class" clauses, i.e., clauses which prevent the formula from being entirely within a tractable base class. The paper systematically investigates whether QBF can be solved in FPT time given a CC-backdoor of size $k$, considering canonical base classes: Horn, 2-CNF, and affine (systems of linear equations over $\mathrm{GF}(2)$). The authors provide hardness and tractability results, and their framework leverages both branching and algebraic methods, notably Gaussian elimination for affine formulas.

Clausal Deletion Backdoors: Definition and Motivation

The motivation for CC-backdoors arises from the inadequacy of existing variable-based backdoor definitions for QBF, due to the quantifier interleaving and non-trivial dependencies. Unlike traditional backdoor sets (where branching on variables can violate the quantifier prefix), CC-backdoors avoid such restrictions and instead focus on variables covering problematic clauses. The detection problem (finding a CC-backdoor) is trivial for fixed base classes, but the evaluation problem (solving QBF given a CC-backdoor) is highly non-trivial.

The formal definition: Given class $C$ (e.g., 2-CNF), a set $B$ is a CC-backdoor for formula $\Phi$ if after removing all clauses containing variables in $B$, the remaining formula is in $C$. The interest lies in whether QBF can be solved in FPT time as a function of $k = |B|$.

Parameterized Complexity Results

FPT Algorithms for 2-CNF and AFF

For QBF with CC-backdoors into 2-CNF and AFF classes:

• 2-CNF: The paper presents an FPT algorithm with running time $O^*(2^k)$, where $k$ is the size of the CC-backdoor. The algorithm employs a branching strategy with look-ahead and unit propagation. Crucially, it leverages the structure of the quantifier prefix to ensure that branchings reduce the parameter and that propagation reveals satisfying assignments efficiently.
• AFF (Affine): For affine formulas (conjunctions of linear equations), the classical branching techniques do not suffice. Instead, the authors construct a kernel via Gaussian elimination coupled with strategic pivoting and elimination of variables. The kernel size is at most $\mathrm{GF}(2)$0 variables, permitting evaluation in $\mathrm{GF}(2)$1 time. This approach is algebraic and overlooked in prior QBF literature.

Both algorithms are shown to be optimal under the Strong Exponential-Time Hypothesis (SETH): if the CC-backdoor covers all variables, they revert to the full QBF evaluation with $\mathrm{GF}(2)$2 time, which cannot be improved under SETH. The results hold independently of quantifier prefix structure.

Hardness for Horn and Hitting Set-Bounded Classes

• Horn: For QBF instances with CC-backdoors into Horn formulas, the evaluation problem is $\mathrm{GF}(2)$3-hard, even with a single quantifier alternation (i.e., $\mathrm{GF}(2)$4). This is shown by reduction from the multipartite independent set problem. The proof establishes that even a single clause outside Horn can reintroduce PSPACE-completeness.
• Hitting Set-Bounded Classes: The hardness extends to implicative hitting set-bounded fragments of Horn (and their duals), including formulas permitting implications and negative clauses. If implication is excluded, the problem becomes FPT.

Dichotomy and Classification

The paper analyzes the full landscape via an algebraic approach, considering constraint languages and polymorphism preservation. The main dichotomy is:

• FPT: When the base class admits either majority, minority, or threshold polymorphisms (covering 2-CNF and affine).
• W[1]-hard: When only conjunction ($\mathrm{GF}(2)$5) or disjunction ($\mathrm{GF}(2)$6) is preserved, but not majority.
• Para-PSPACE-hard: Otherwise.

A complete dichotomy is achieved except for $\mathrm{GF}(2)$7-ary positive Horn ($\mathrm{GF}(2)$8-$\mathrm{GF}(2)$9) for $C$0, where complexity remains open.

Algorithmic Techniques

The paper introduces two algorithmic paradigms tailored to parameterized QBF:

• Branching with Propagation: Used for 2-CNF, exploiting resolution closure and propagation under quantifier alternations, ensuring parameter reduction at each step.
• Gaussian Elimination and Kernelization: Used for affine formulas, leveraging algebraic structure for kernelization, and careful management of variable elimination/pivoting under quantifiers.

These techniques handle the interdependence between quantifiers and clause structure, overcoming the pitfalls of prior methods.

Implications and Future Directions

Practical Impact

• QBF Solvers: The results suggest that clause covering backdoors can enable efficient QBF solving even for formulas with unrestricted quantifier prefixes, provided the problematic clauses are covered by a small set of variables. This opens avenues for integrating propagation or algebraic methods into QBF solvers, potentially improving PSPACE-complete AI/verification tasks.
• Kernelization Lower Bounds: The affine kernel size cannot be reduced below $C$1 without collapsing the polynomial hierarchy, providing sharp lower bounds on preprocessing.

Theoretical Impact

• Complexity Landscape: The dichotomy offers clarity on which base classes admit FPT evaluation via CC-backdoors and which are inherently resistant, up to strong parameterized complexity assumptions.
• Constraint Satisfaction and Beyond: The algebraic framework generalizes to arbitrary constraint languages, linking polymorphism theory and parameterized complexity for QCSP and related decision problems.

Open Problems

• The complexity status for $C$2-$C$3 backdoors for $C$4 is unresolved. Developing FPT algorithms or proving hardness for these remains a challenging direction, with implications for a complete classification.
• Extensions to non-Boolean QCSP, as well as more expressive forms (e.g., quantified CSP for finite domains), are identified as further points for investigation.
• Understanding the precise parameterized complexity (e.g., XP membership, W[2]-hardness) for Horn formulas with more elaborate quantifier prefixes.

Conclusion

This work establishes the parameterized complexity landscape for QBF via Clause Covering Backdoors. It delivers FPT results for 2-CNF and affine classes using novel algorithmic frameworks, and proves strong hardness for Horn-based classes, constructing a near-complete dichotomy for tractable CC-backdoor evaluation. The theoretical classification is buttressed by sharp algorithmic insights and optimal lower bounds under SETH, and sets the stage for practical QBF solver improvements and further research into parameterized structural decompositions for PSPACE-complete problems (2605.12073).