Quantum Complexity Resource

• Quantum complexity resource is an umbrella term describing quantifiable features of quantum systems, including uncomplexity, state complexity, and resource-sensitive circuit complexity.
• It formalizes how deficits from maximal complexity and hard-to-describe state structures serve as resources for quantum computational advantage, especially in MBQC and channel analysis.
• Practical frameworks use concrete resource estimation, oracle optimization, and Gaussian covariance decomposition to translate abstract complexity into measurable implementation criteria.

Searching arXiv for papers on quantum complexity resources and closely related formulations. arxiv_search(query="quantum complexity resource circuit complexity uncomplexity resource theory statistical complexity quantum circuits", max_results=10, sort_by="submittedDate")

Retrieving papers most relevant to resource-theoretic, statistical, and operational notions of quantum complexity. Quantum complexity resource denotes a family of notions that treat complexity-relevant features of quantum systems as quantifiable resources rather than as a single metric. In the literature, the term encompasses at least six technically distinct but partially overlapping usages: resource-theoretic uncomplexity as distance from maximal circuit complexity (Halpern et al., 2021, Brown et al., 2017); state complexity measures such as tree size that can be necessary for quantum computational advantage in measurement-based quantum computation (Cai et al., 2015); resource-sensitive circuit and channel complexity defined relative to chosen gate sets, channels, or generators (Bu et al., 2021, Araiza et al., 2023); nonlocal resources such as shared entanglement or closed timelike curves whose formal complexity power may lack operational certifiability (Fields et al., 24 Jan 2025); nonlocal quantum computation resources studied through reduction-based relative hardness (Bluhm et al., 29 May 2025); and, in Gaussian photonic settings, a quantum complexity resource extracted from covariance matrices by semidefinite programming and linked to hafnian-based $\#P$-hardness (Kocharovsky et al., 17 Jun 2026, Kalra et al., 29 Jun 2026). A plausible implication is that “quantum complexity resource” is best understood as an umbrella expression for resource notions that mediate between formal complexity theory, operational implementability, and physically meaningful nonclassical structure.

1. Resource-theoretic uncomplexity

Brown and Susskind propose a thermodynamic treatment of quantum complexity in which the relevant resource is not complexity itself but the deficit from maximal complexity, called uncomplexity (Brown et al., 2017). For a $K$-qubit system, they use computational or circuit complexity $\mathcal{C}$ in the standard sense: for a unitary $U$, $\mathcal{C}(U)$ is the minimum number of allowed $k$-local gates needed to prepare $U$; for a pure state $|\psi\rangle$, $\mathcal{C}(|\psi\rangle)$ is the minimum number of allowed gates required to prepare it from an unentangled product state. They write a circuit as

$U=g_N g_{N-1}\cdots g_1$

and define $K$0 as the smallest possible $K$1 (Brown et al., 2017).

Within that framework, maximum complexity scales exponentially with system size,

$K$2

more concretely of order $K$3 for state complexity, while generic $K$4-local evolution is conjectured to display an early-time linear regime

$K$5

up to times $K$6, followed by saturation and doubly exponential recurrence times (Brown et al., 2017). Uncomplexity is then defined as

$K$7

Its operational role is explicitly resource-theoretic: it is the available “room” before equilibrium complexity is reached, and Brown and Susskind argue that this quantity can be consumed to perform directed quantum computation (Brown et al., 2017).

That conjectural resource picture is formalized by the resource theory of quantum uncomplexity (Halpern et al., 2021). In that work, no states are free, because tensoring on ancillas is excluded and “any tensored-on state benefits quantum computation” (Halpern et al., 2021). Free operations are fuzzy operations, namely compositions of noisy two-qubit gates. A fuzzy gate $K$8 is sampled around a chosen target $K$9 from a distribution $\mathcal{C}$0 supported within operator-norm distance $\mathcal{C}$1 of $\mathcal{C}$2 and nonzero on an open set containing $\mathcal{C}$3 (Halpern et al., 2021). This noise model prevents exact uncomputation from trivializing the resource theory.

The central quantitative object is the complexity entropy

$\mathcal{C}$4

where $\mathcal{C}$5 is the set of measurements implementable by at most $\mathcal{C}$6 two-local gates followed by a zero-complexity measurement (Halpern et al., 2021). The associated complexity negentropy,

$\mathcal{C}$7

acts as a resource measure in one-shot tasks.

Two operational tasks are defined. In uncomplexity extraction, one seeks to extract as many clean qubits $\mathcal{C}$8 as possible from $\mathcal{C}$9 using at most $U$0 fuzzy gates. If $U$1, then for $U$2 some protocol extracts

$U$3

qubits $U$4-close to $U$5, while every protocol obeys

$U$6

(Halpern et al., 2021). In uncomplexity expenditure, one borrows $U$7 clean qubits and uses them to imitate a target state to a bounded-complexity referee; if $U$8, then $U$9 can be imitated with

$\mathcal{C}(U)$0

clean qubits (Halpern et al., 2021). This gives uncomplexity an explicitly operational meaning.

This line of work treats uncomplexity as analogous to free energy. The analogy is exact at the level of the proposed auxiliary thermodynamic system: $\mathcal{C}(U)$1 so the useful resource is controlled by the gap from maximal complexity rather than by complexity itself (Brown et al., 2017). A common misconception is that high complexity is the computational resource; these papers instead argue that maximal complexity is operationally exhausted, while low complexity or clean ancillary structure is what enables useful computation (Brown et al., 2017, Halpern et al., 2021).

2. State complexity as a computational resource

A different usage of quantum complexity resource centers on state complexity. In “State complexity and quantum computation” (Cai et al., 2015), the resource is tree size (TS), a measure of the shortest rooted-tree decomposition of a pure state using additions and tensor products of single-qubit states. For an $\mathcal{C}(U)$2-qubit state

$\mathcal{C}(U)$3

tree size is defined as the minimum number of leaves over all rooted trees built from $\mathcal{C}(U)$4 and $\mathcal{C}(U)$5 nodes that represent $\mathcal{C}(U)$6 (Cai et al., 2015). The $\mathcal{C}(U)$7-approximate version is

$\mathcal{C}(U)$8

and mixed-state tree size is defined by minimizing the maximal pure-state tree size over decompositions of the mixed state (Cai et al., 2015).

The importance of tree size derives from two distinctive properties emphasized in the paper: it is “in principle computable,” and nontrivial lower bounds can be obtained using the connection to multilinear formula size and Raz’s lower bound theorem (Cai et al., 2015). The associated amplitude function is

$\mathcal{C}(U)$9

and multilinear formula lower bounds transfer to tree-size lower bounds because $k$0 (Cai et al., 2015).

The strongest resource-theoretic conclusion in that paper concerns measurement-based quantum computation (MBQC). The authors prove that if the resource state of an MBQC protocol has polynomial tree size, then the computation can be simulated efficiently classically; accordingly, superpolynomial tree size is necessary for MBQC quantum advantage (Cai et al., 2015). They further prove that the universal 2D cluster state has superpolynomial tree size,

$k$1

and also

$k$2

(Cai et al., 2015).

The paper is explicit that the situation is subtler in the circuit model. The stronger conjecture $k$3 remains open; only a weaker simulability result is proved for circuit families whose polynomial-size trees satisfy an additional separating-tree condition implying polynomial Schmidt rank across every bipartition (Cai et al., 2015). Large tree size is therefore necessary for MBQC speedup, conjecturally important for the circuit model, but not sufficient for quantum speedup, because stabilizer/subgroup states can have superpolynomial tree size while remaining classically simulable (Cai et al., 2015).

This suggests a more precise interpretation: tree size functions as a resource witness for hardness of classical representation, but not as a complete resource theory of quantum advantage. A common misconception addressed directly in the paper is that entanglement alone or state complexity alone should fully characterize quantum speedup; the results show that large tree size is informative but incomplete (Cai et al., 2015).

3. Resource-sensitive complexity of circuits and channels

A third usage of quantum complexity resource is explicitly relative to a chosen resource set. Two complementary strands illustrate this.

The first strand studies the statistical complexity of quantum circuit classes in the sense of learning theory. “On the statistical complexity of quantum circuits” (Bu et al., 2021) defines the empirical Rademacher complexity of a hypothesis class induced by quantum circuits,

$k$4

where

$k$5

The key resource quantity is a $k$6 group norm of the Pauli transfer matrix $k$7,

$k$8

which acts as a measure of magic for channels and layered circuits (Bu et al., 2021). Clifford unitaries give signed permutation matrices in the Pauli basis, and for suitable $k$9 the norm is faithful relative to Clifford structure, invariant under Clifford pre- and post-processing, multiplicative under tensor products, and convex (Bu et al., 2021). The paper derives upper bounds on statistical complexity in terms of this resource. For example, for layered circuits of depth $U$0 with bounded average layer resource $U$1,

$U$2

(Bu et al., 2021). This treats magic as a resource constraining model capacity.

The second strand asks how the addition of non-free channels changes the statistical richness of circuit classes. “Effects of quantum resources on the statistical complexity of quantum circuits” (Bu et al., 2021) uses resource theory and the free robustness of a channel,

$U$3

to upper-bound the increase in Rademacher and Gaussian complexity when a resource channel $U$4 is adjoined to a free set $U$5 (Bu et al., 2021). The main one-resource bound is

$U$6

and with up to $U$7 uses of $U$8,

$U$9

(Bu et al., 2021). Here free robustness becomes a direct upper bound on learnability-relevant expressivity.

A broader channel-theoretic generalization appears in “Resource-Dependent Complexity of Quantum Channels” (Araiza et al., 2023). That paper defines a resource-sensitive channel complexity from a set $|\psi\rangle$0 via the noncommutative Lipschitz seminorm

$|\psi\rangle$1

and then

$|\psi\rangle$2

A complete, correlation-assisted version is also defined: $|\psi\rangle$3 This measure is designed for both unitary and open-system dynamics, satisfies faithfulness, subadditivity, convexity, and, for the $|\psi\rangle$4 version, exact tensor additivity (Araiza et al., 2023). It provides lower bounds on mixed-unitary gate complexity, Carnot–Carathéodory geometric complexity, Hamiltonian simulation cost, and open-system simulation cost. For instance, if

$|\psi\rangle$5

then with resources $|\psi\rangle$6,

$|\psi\rangle$7

for $|\psi\rangle$8, while short-time lower bounds scale linearly in $|\psi\rangle$9 under a commutator ratio parameter $\mathcal{C}(|\psi\rangle)$0 (Araiza et al., 2023).

These works collectively imply that “quantum complexity resource” can mean a chosen gate set, channel class, generator family, or resource monotone that determines what counts as elementary and how far a target process lies from identity. This suggests a common structural pattern: complexity is not absolute but relative to an admissible resource set.

4. Concrete resource estimation and oracle-aware programming

A more implementation-facing usage of quantum complexity resource concerns resource estimation rather than resource monotones. “Quipper: Concrete Resource Estimation in Quantum Algorithms” (Smith et al., 2014) argues that asymptotic complexity is insufficient for assessing practical implementability because actual counts of qubits, ancillas, and logical gates can differ by orders of magnitude from high-level descriptions. Quipper is introduced as “a formal framework to write, and reason about, quantum algorithms,” embedded in Haskell and based on a generalized circuit model with two runtimes: circuit generation time and circuit execution time (Smith et al., 2014). It explicitly tracks initializations and terminations of qubits “for the purpose of ancilla management,” supports hierarchical boxing, distinguishes parameters from inputs, and can automatically generate quantum oracles from classical code via Template Haskell (Smith et al., 2014).

The paper’s best-known quantitative example is Triangle Finding, where the command $K$40 produces

• total gates: $\mathcal{C}(|\psi\rangle)$1
• qubits: $\mathcal{C}(|\psi\rangle)$2

in under two minutes on a laptop computer (Smith et al., 2014). The point is not a new resource theory but a concrete logical-level accounting discipline. The paper stresses that these are logical, not physical, resources, and that the numbers can themselves pose “a fundamental barrier to quantum computing unless significant optimizations in the transformations from algorithm to gates can be found” (Smith et al., 2014).

A related development appears in “Resource-Aware Hybrid Quantum Programming with General Recursion and Quantum Control” (Chardonnet et al., 23 Oct 2025). That paper introduces Hyrql, a hybrid quantum language “driven towards resource-analysis” and deliberately not tied to a fixed initial set of quantum gates (Chardonnet et al., 23 Oct 2025). Its key idea is to compile well-typed programs to simply-typed term rewrite systems (STTRSs), thereby importing rewrite-based termination and complexity methods. The main complexity-preservation proposition states that if the generated STTRS terminates on input $\mathcal{C}(|\psi\rangle)$3 in time $\mathcal{C}(|\psi\rangle)$4, then the original Hyrql program reduces in

$\mathcal{C}(|\psi\rangle)$5

and

$\mathcal{C}(|\psi\rangle)$6

steps (Chardonnet et al., 23 Oct 2025). This is a program-level resource analysis framework supporting hybrid control, quantum control, higher-order functions, and general recursion.

Oracle construction is treated directly in “Modeling and Resource Optimization for Quantum Oracles” (Li et al., 20 May 2026). That paper introduces the Hierarchical Recursive Synthesis-Evaluation (HRSE) model, representing a multi-function oracle by a rooted tree $\mathcal{C}(|\psi\rangle)$7 with node attributes

$\mathcal{C}(|\psi\rangle)$8

where $\mathcal{C}(|\psi\rangle)$9 is node size in auxiliary qubits, $U=g_N g_{N-1}\cdots g_1$0 recursion depth, $U=g_N g_{N-1}\cdots g_1$1 out-degree, $U=g_N g_{N-1}\cdots g_1$2 gate complexity, and $U=g_N g_{N-1}\cdots g_1$3 covered leaf count (Li et al., 20 May 2026). For non-leaf nodes,

$U=g_N g_{N-1}\cdots g_1$4

and globally,

$U=g_N g_{N-1}\cdots g_1$5

(Li et al., 20 May 2026). The associated Adaptive Space-depth Trade-off (ASDT) algorithm is claimed to achieve the optimal gate count for a given number of qubits, and experimentally reduces the average quantum circuit depth by $U=g_N g_{N-1}\cdots g_1$6 compared with the W-cycle approach, with the number of variables being $U=g_N g_{N-1}\cdots g_1$7, $U=g_N g_{N-1}\cdots g_1$8, and $U=g_N g_{N-1}\cdots g_1$9, respectively (Li et al., 20 May 2026).

These papers do not define a single scalar “complexity resource,” but they do establish a common methodology: formalize the algorithmic structure, make hidden oracle and ancilla costs explicit, and analyze trade-offs among gates, depth, width, and qubits. A plausible implication is that concrete resource estimation is the implementation-level counterpart of more abstract resource-theoretic approaches.

5. Nonlocal and operationally undecidable resources

Another major meaning of quantum complexity resource concerns nonlocal or otherwise nonclassical resources whose formal computational power may exceed what can be operationally certified.

In “Whether a quantum computation employs nonlocal resources is operationally undecidable” (Fields et al., 24 Jan 2025), the central claim is that once quantum computation is allowed to use genuinely nonlocal spatial or temporal resources, the usual operational interpretation of computational complexity breaks down. The paper focuses on two paradigmatic models: multiple interactive proofs with entangled provers, $K$00, and computers using closed timelike curves (CTCs) (Fields et al., 24 Jan 2025). For $K$01, the formal background includes

$K$02

but the paper argues that a classical verifier cannot operationally certify that purportedly multiple entangled provers are genuinely independent (Fields et al., 24 Jan 2025).

The core impossibility theorem states:

Theorem 2. An observer $K$03 embedded in an environment $K$04 cannot determine, either by monitoring classical communication between $K$05 and $K$06, or by performing local measurements within $K$07, whether or not $K$08 and $K$09 are employing a LOCC protocol with classical and quantum channels traversing $K$10 (Fields et al., 24 Jan 2025).

The proof is built around CHSH-style tests. Even Bell-violation data only certify nonclassical correlations, not the separability architecture

$K$11

required to instantiate a genuine multi-prover system (Fields et al., 24 Jan 2025). The paper concludes that the verifier “cannot operationally distinguish between a MIP* machine and a monolithic quantum computer” (Fields et al., 24 Jan 2025). Parallel arguments are given for CTC-based resources via an inability to determine whether a channel is, in the internal metric of the system, effectively a closed timelike curve (Fields et al., 24 Jan 2025).

The main methodological warning is that complexity measures are operationally meaningful only if the relevant resources are user-measurable. In ordinary Turing-style computation, time and space are local and countable. With nonlocal spatial or temporal resources, an external observer may be unable to determine whether those resources are present at all; in that case, statements about their complexity usage may “cease to be reliable indicators of practical computational capability” (Fields et al., 24 Jan 2025).

A different but related treatment appears in “A complexity theory for non-local quantum computation” (Bluhm et al., 29 May 2025). There, the problem is not operational undecidability but the obstruction to exact entanglement-cost lower bounds. The paper argues that characterizing entanglement cost directly for tasks such as $K$12-route would imply major breakthroughs in complexity theory, because entanglement upper bounds are already tied to hard measures such as memory complexity and span-program size (Bluhm et al., 29 May 2025). The proposed solution is a reduction-based complexity theory for non-local quantum computation (NLQC).

Its main equivalence theorem states that three central tasks—$K$13-route, $K$14-measure, and CDQS—are equivalent under $K$15-overhead reductions: $K$16 (Bluhm et al., 29 May 2025). This transfers known sub-exponential upper bounds,

$K$17

to $K$18-measure for all Boolean functions and extends efficient protocols to functions in $K$19 (Bluhm et al., 29 May 2025). The paper also studies coherent controlled tasks such as $K$20-SWAP, $K$21-PHASE, and $K$22-PAULI and situates them within a reduction hierarchy (Bluhm et al., 29 May 2025).

Together, these papers show two distinct senses in which nonlocality functions as a complexity resource: formally, it can enable stronger task classes; operationally, its presence may be difficult or impossible to certify. A common misconception is that a theorem about an abstract model immediately licenses a claim about physical computational power. The operational undecidability paper rejects that inference, while the NLQC paper reframes resource comparison in relative-hardness terms rather than exact entanglement formulas (Fields et al., 24 Jan 2025, Bluhm et al., 29 May 2025).

6. Gaussian boson sampling and covariance-based quantum complexity resource

In recent Gaussian photonic work, “quantum complexity resource” acquires a very specific meaning: the irreducible pure Gaussian core of a mixed covariance matrix.

“The quantum-advantage resource in multimode OPA light: Identification, optimization, extraction” (Kocharovsky et al., 17 Jun 2026) defines the resource via the Oh semidefinite decomposition. For an $K$23-mode Gaussian covariance matrix $K$24, one solves

$K$25

The decomposition

$K$26

defines $K$27 as the quantum complexity resource and $K$28 as the maximal positive-semidefinite classically simulable part (Kocharovsky et al., 17 Jun 2026). A central theorem proves that every optimizer $K$29 is pure, equivalently

$K$30

so the extracted resource is not merely low-noise but a pure Gaussian covariance (Kocharovsky et al., 17 Jun 2026).

The resource photon number can then be written through the Bloch–Messiah parameters of $K$31: $K$32 To connect more directly to hardness, the paper defines the dimension of multimode-state quantum complexity

$K$33

The cap at $K$34 reflects the claim that repeated photons in the same mode do not exponentially increase hafnian hardness and that only the irreducible pure core should count toward quantum advantage (Kocharovsky et al., 17 Jun 2026). The paper explicitly states that if $K$35, the state lies beyond the reach of the best known classical simulation algorithm of Oh et al., and proposes this as an experimental quantum-advantage benchmark (Kocharovsky et al., 17 Jun 2026).

The relation to computation comes from Gaussian boson sampling: photon-counting probabilities are given by hafnians of matrices built from the covariance, and hafnian computation is $K$36-hard (Kocharovsky et al., 17 Jun 2026). The resource is therefore the part of the state whose photon-number statistics remain irreducibly hafnian-hard after stripping away classical Gaussian camouflage.

A refined structural analysis appears in “Quantum complexity resource in Gaussian boson sampling: Core structure of the semidefinite program” (Kalra et al., 29 Jun 2026). That paper studies the same primal SDP,

$K$37

and proves that the optimizer is unique and pure (Kalra et al., 29 Jun 2026). It solves the dual inner problem in closed form through the oracle map

$K$38

which obeys the algebraic Riccati identity

$K$39

(Kalra et al., 29 Jun 2026). The paper further shows that the problem compresses exactly onto the active symplectic sector generated by the dual support and that passive-diagonalizable states admit closed-form solutions (Kalra et al., 29 Jun 2026).

These Gaussian papers use the phrase “quantum complexity resource” in a literal, singular sense: a specific matrix-valued object extracted by SDP. This is more concrete than the umbrella usage elsewhere. A plausible implication is that the phrase may be stabilizing into a technical term in Gaussian boson sampling, where it denotes the pure covariance component carrying hafnian hardness.

7. Synthesis and conceptual boundaries

The literature does not support a single universal definition of quantum complexity resource. Instead, it supports a structured plurality.

First, some papers use the term for a deficit resource: uncomplexity, the gap to maximal complexity, which can be extracted, spent, and constrained by monotones (Brown et al., 2017, Halpern et al., 2021). Second, others use it for hard-to-describe state structure, such as superpolynomial tree size, which can be necessary for MBQC advantage (Cai et al., 2015). Third, several works define complexity relative to chosen resources—free channels, magic-bearing operations, generator sets, or gate bases—and analyze how these resources control expressivity, learnability, or implementation difficulty (Bu et al., 2021, Bu et al., 2021, Araiza et al., 2023). Fourth, some papers emphasize that purported resources such as nonlocality may be formally powerful yet operationally uncertifiable, limiting the physical significance of complexity claims (Fields et al., 24 Jan 2025). Fifth, concrete systems papers identify implementation resources—ancillas, circuit depth, oracle structure, multimode covariance components—and optimize them explicitly (Smith et al., 2014, Li et al., 20 May 2026, Kocharovsky et al., 17 Jun 2026).

The main conceptual boundary is between formal resource theories and practical resource accounting. Formal approaches ask what monotones, free operations, and task conversion rates characterize a resource. Practical approaches ask what circuits, qubits, depth, ancillas, or covariance components are actually needed. The two are related but not interchangeable.

Another boundary separates computational hardness resources from operational usefulness resources. High complexity, hafnian hardness, and large tree size point toward classical intractability; uncomplexity points toward remaining computational usefulness. These are not opposites, but they emphasize different operational questions.

Finally, there is a recurring caution across the literature: asymptotic or abstract complexity claims can be misleading without an operational anchor. Quipper demonstrates this at the level of logical gate counts (Smith et al., 2014). The nonlocal-resource paper demonstrates it at the level of certifying architecture (Fields et al., 24 Jan 2025). The Gaussian boson sampling papers demonstrate it by subtracting classically simulable covariance before calling the remainder a resource (Kocharovsky et al., 17 Jun 2026, Kalra et al., 29 Jun 2026). This suggests that the most robust uses of “quantum complexity resource” are those that explicitly specify: the allowed operations, the reference notion of simplicity, the operational task, and the measurable quantity being optimized or bounded.