On the Complexity of the Circuit Width Problem

Abstract: Montanaro's polynomial representation expresses amplitudes of quantum circuits over the gates $H$, $Z$, $CZ$, and $CCZ$ as normalized gaps of degree-three polynomials over $\mathbb{F}_2$. The normalization is governed by the circuit width $w(f)$, the minimum number of qubits in any circuit realizing a polynomial $f$. Thus, efficient width minimization would give an approximate-counting route toward a combinatorial characterization of $BQP$. We study the computational complexity of this parameter. For degree-three polynomials with no constant term, deciding whether $w(f)\le k$ is $NP$-complete, resolving Montanaro's open question. We also prove $NP$-hardness of approximation within any factor $49/48-ε$, and show via a twin-copy construction that the exact and approximation hardness results also hold for degree-two polynomials. Under the Exponential Time Hypothesis, the exact problem admits no $2^{o(n)}$-time algorithm when $k=Θ(n)$. Complementing these hardness results, we give a nondeterministic polynomial-time search algorithm using $2\log_2\binom{n}{k}=O(k\log(en/k))$ witness bits, and a constructive fixed-parameter algorithm parameterized by $k$ with running time $k^{6k+o(k)}n+O(m)$.

• The paper establishes that deciding whether a polynomial’s circuit width is at most k is NP-complete via a structured reduction from 3-SAT.
• It shows that approximating circuit width within a factor of (49/48 - ε) is NP-hard, even when the input is restricted to quadratic polynomials.
• The study provides nondeterministic and fixed-parameter tractable algorithms that exploit the circuit’s structural properties to guide practical quantum circuit synthesis.

Complexity of the Circuit Width Problem: Formal Summary and Analysis

Introduction and Context

The paper "On the Complexity of the Circuit Width Problem" (2606.18201) provides a rigorous complexity-theoretic analysis of the circuit width parameter for quantum circuits, as formalized in the polynomial representation framework developed by Montanaro and others. In this setting, every quantum circuit over a restricted gate set corresponds to a low-degree polynomial over $\mathbb{F}_2$, and the circuit width $w(f)$ of a polynomial $f$ is the minimal number of qubits required by any quantum circuit whose "associated polynomial" is $f$. This parameter is closely related to the achievable precision in quantum amplitude estimation via these algebraic encodings.

The central motivation is to determine how "tractable" it is to compute $w(f)$, both exactly and approximately, for arbitrary polynomials $f$. This has implications both for attempts at a combinatorial characterization of BQP based on approximate counting, and for practical resource optimization in quantum circuit compilation. The authors resolve multiple open questions in this direction and delineate both upper and lower bounds for the circuit width problem.

Main Results

Hardness of the Circuit Width Problem

The key result is a tight complexity classification of the decision version of the width problem, denoted $\mathrm{CW}(f, k)$:

• NP-completeness: The decision problem of whether $w(f) \leq k$, given an explicit degree-3 polynomial $f$ over $\mathbb{F}_2$, is NP-complete. This is established via a polynomial-time reduction from 3-SAT; specifically, the authors encode a 3-SAT formula $w(f)$0 as a polynomial $w(f)$1 such that $w(f)$2 if and only if $w(f)$3 is satisfiable.
• Hardness of approximation: It remains NP-hard to approximately compute the circuit width: specifically, for any $w(f)$4, it is NP-hard to distinguish whether $w(f)$5 or $w(f)$6. Notably, this rules out any quantum or classical algorithm that could, even with the power of BQP, approximate $w(f)$7 within an additive $w(f)$8 error unless $w(f)$9.
• Hardness for degree-2 polynomials: The above hardness extends to quadratic polynomials, not just cubic ones. The reduction leverages a duplicate-or-clique expansion (the "twin" construction) to simulate cubic constraints in the quadratic regime, ensuring the same inapproximability factor of $f$0.

Algorithmic Results

Despite these strong hardness results, the paper also presents tight positive results:

• Nondeterministic search algorithm: There is a nondeterministic polynomial-time algorithm for the search version of the problem—explicitly, an NP algorithm that, on witness sets of $f$1 bits (specifying the boundary segments for each wire), can reconstruct a width-$f$2 circuit realizing a given polynomial, if one exists.
• Fixed-parameter tractability (FPT): Parameterized by $f$3, there exists an explicit fixed-parameter tractable algorithm for $f$4 with runtime $f$5. This leverages the fact that the "connection graph" of a realizable polynomial has pathwidth at most $f$6, permitting dynamic programming across nice tree decompositions. This is faster than the trivial brute-force algorithm ($f$7) for small $f$8 and saturates the best known lower bound (under ETH) when $f$9.

Technical Contributions

Hardness Reductions

The reduction from 3-SAT is highly structured. The authors construct $f$0 in such a way that:

• Each clause and variable is represented by distinct sets of symbols (segments), arranged on qubit wires subject to overlap constraints realized via quadratic (CZ) and cubic (CCZ) monomials.
• Satisfaction of a clause is enforced geometrically: only if the segments corresponding to the literals in a clause may overlap in certain ways, the minimal qubit realization corresponds to a satisfying assignment. Any violation requires the use of additional qubits.
• These gadgets ensure that width minimization is combinatorially equivalent to Boolean assignment satisfaction.

For the approximation hardness, the work bootstraps the classical inapproximability of Max-3-SAT (via Håstad's Theorem), resulting in an explicit constant-factor gap in the hardness of approximating circuit width.

The extension to degree-2 (quadratic) polynomials is achieved by expanding each variable into multiple "twins" and enforcing clique constraints. The geometric essence of the cubic overlap is simulated via enforcing pairwise overlaps across many copies, preserving the necessary complexity-theoretic properties.

Algorithmic Framework

The nondeterministic algorithm and the FPT result both build upon a clean structural characterization of circuit realizations:

• Partition into wires: Any feasible realization corresponds to a partition of the symbols into $f$1 ordered lists, one per wire, encoding the temporal sequence of segments.
• Connection graph and acyclicity: Realizability implies that the connection graph—incorporating both adjacency (due to CZ or Hadamard) and temporal intersection constraints (due to monomials)—must be acyclic. A compatible topological ordering then yields valid placements of Hadamard and diagonal gates.
• Dynamic programming on decomposition: Exploiting the bounded pathwidth structure (guaranteed by the wire partitioning), the FPT algorithm propagates partial solutions up a nice tree decomposition, tracking wire assignments and temporal orderings on each bag boundary.

Theoretical and Practical Implications

Combinatorial characterization of BQP is fundamentally limited in the "algebraic counting via width" paradigm: the NP-hardness and inapproximability of circuit width—even approximating to within a tiny constant—preclude any unqualified reduction of BQP to approximate classical counting equipped with an oracle for circuit width.

Quantum resource minimization is inherently intractable: Even with quantum computational resources, determining how to optimally "minimize precision" in state preparation or amplitude estimation—when formulated as width minimization for arbitrary polynomials—is computationally hard. Thus, heuristic or domain-constrained methods are essential in practical quantum circuit compilation.

Circuit width mirrors classical structural graph parameters (e.g., treewidth/pathwidth): The complexity of width aligns more naturally with these combinatorial structures than with any direct quantum computational power metric. Intractability persists even when only a small fraction of qubits could be saved, or when the input is restricted to limited-degree polynomials deriving from practical quantum gate sets.

Robustness of parameterized algorithms: The FPT result guarantees that for applications where the target width $f$2 is small (e.g., very low-qubit syntheses), feasible algorithms exist. The parameterized complexity boundary is tight up to known lower bounds under ETH.

Strong Numerical and Qualitative Claims

• Approximation hardness factor is explicit ($f$3): No polynomial-time (or quantum polynomial-time) method can efficiently approximate $f$4 better than this factor unless $f$5.
• Exponential lower bound under ETH: Solving $f$6 on unrestricted instances in time $f$7 violates the Exponential Time Hypothesis.
• Concrete FPT runtime: The presented FPT algorithm achieves $f$8 runtime; this is provably superior to all previously known bounds for relevant $f$9.
• Quadratic polynomials retain full hardness: Restricting to degree-2 polynomials does not alleviate inapproximability in any meaningful way, due to the twin construction.

Future Directions

• Hardness amplification: Can the approximation lower bound be raised, potentially linking to PCP-based reductions for larger constant factors, without abandoning the structure-preserving geometric encoding?
• Complexity for natural subclasses: Which (if any) significant subclasses of polynomials admit tractable width computation? Are there structural dichotomies (as in treewidth for graphs) that separate easy and hard cases?
• Optimal FPT bounds: Is a single-exponential FPT algorithm (i.e., $w(f)$0) achievable, or do strong parameterized lower bounds apply?

Conclusion

This work establishes definitive computational complexity boundaries for the quantum circuit width problem, excluding broad avenues for combinatorial, approximate-counting-based characterizations of quantum computation and sharply delimiting the tractability in classical and parameterized settings. The implications pervade both the theoretical understanding of quantum computational resources and the practical domain of circuit synthesis and optimization. The tools developed also illuminate connections between quantum circuit topology and classical graph-theoretic structure, suggesting future research directions at this intersection.