DQC1: Minimal-Resource Quantum Computation

• DQC1 is a quantum model using one pure qubit with a register of maximally mixed qubits to enable normalized trace estimation.
• The model leverages quantum discord over entanglement, demonstrating computational advantages despite minimal pure resources.
• DQC1 establishes quantum-classical separations by solving classically intractable problems and posing challenges for efficient classical simulation.

Deterministic Quantum Computation with One Clean Qubit (DQC1) is a highly influential subuniversal model of quantum computation notable for its ability to efficiently solve certain problems believed to be classically intractable, despite operating with only minimal quantum resources: a single pure ("clean") qubit and a register of maximally mixed qubits. DQC1 enables efficient normalized trace estimation and forms a natural framework for analyzing the roles of quantum correlations, the limits of quantum state synthesis, and computational complexity separations between quantum and classical models.

1. Formal Model: Circuit Structure and Measurement Protocol

In DQC1, the input state is

$$\rho_\mathrm{in} = |0\rangle\langle0| \otimes \frac{I^{\otimes n}}{2^n}$$

where the first qubit is pure and the remaining $n$ qubits are maximally mixed. The protocol applies a polynomial-size unitary $U$ to all qubits, followed by measurement of the clean qubit in the computational basis. The probability of measurement outcome $a \in \{0,1\}$ is

$$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$

For the ubiquitous "trace-estimation" subcircuit—preparing the clean qubit in $|+\rangle$, applying a controlled-$U$, and measuring the $X$ or $Y$ operator—the reduced state immediately before measurement is

$$\rho_\mathrm{DQC1} = \frac{1}{2^{n+1}} \begin{pmatrix} I & U^\dagger \ U & I \end{pmatrix}$$

and measurement yields the normalized trace: $n$0 This defines a framework for estimating spectral invariants and certain correlation functions using only a single measured qubit and a maximally mixed bath (Morimae et al., 2014, Moulik et al., 2024, Yu et al., 2013, Passante et al., 2012).

2. Complexity-Theoretic Position and DQC1-Completeness

DQC1 constitutes a promise-problem complexity class: $n$1 if, for polynomial-size quantum circuits $n$2, the accept probability $n$3 is separated by $n$4 between yes- and no-instances (Morimae et al., 2014, Moulik et al., 2024, Ji et al., 2 Apr 2026). Known strict containments are

$n$5

and DQC1 is not believed to be universal for quantum computation (not BQP-complete), but is widely conjectured to be strictly more powerful than BPP.

Canonical DQC1-complete problems include:

• Normalized trace estimation $n$6 (Morimae et al., 2014).
• Estimation of spectral sums $n$7 for functions $n$8 with high approximate degree on log-local Hamiltonians (Ji et al., 2 Apr 2026).
• Evaluation of out-of-time-ordered correlators (OTOCs) and general $n$9-point correlation functions at infinite temperature (Moulik et al., 2024).
• Certain link invariants (e.g., fixed-point evaluations of the Jones polynomial).

Classical simulation of DQC1 output distributions is intractable under standard complexity assumptions; efficient weak simulation with multiplicative error $U$0 for $U$1 measured bits in DQC1$U$2 circuits collapses PH to $U$3 (Morimae et al., 2013), and even for one measured output collapses PH to AM (Fujii et al., 2014).

3. Quantum Resources: Entanglement, Discord, and Coherence

DQC1 demonstrates that exponential quantum computational advantages can be achieved with little or vanishing entanglement, but with robust nonclassical correlations or quantum discord. In the standard DQC1 output state,

• Multiplicative negativity (as a measure of bipartite entanglement) is maximized at $U$4 for any partition, and typically exponentially small in $U$5 for Haar-random unitaries (Kay, 2015).
• The geometric quantum discord (GQD), quantifying nonclassical correlations, is nonzero and admits a closed-form expression for all $U$6-qubit DQC1 outputs (Passante et al., 2012): $U$7 GQD decays exponentially with $U$8 for typical unitaries, while standard quantum discord remains $U$9.
• Quantum dissonance, the component of quantum correlations in the closest separable state—the "genuine quantum resource" in DQC1—can be demonstrated to power computational speedup in the absence of entanglement (Ali, 2013).

Experimental work confirms that DQC1 acts as a resource converter: initial coherence in the clean qubit is consumed to generate discord with the register, with bounds of the form $a \in \{0,1\}$0 (relative entropy of coherence difference), all verified in superconducting implementations (Wang et al., 2018).

4. Algorithmic and Applied Aspects

DQC1 efficiently solves a specific class of problems:

• Estimation of $a \in \{0,1\}$1 for black-box $a \in \{0,1\}$2-qubit unitaries or log-local Hamiltonian functions $a \in \{0,1\}$3 up to inverse-polynomial additive error (Ji et al., 2 Apr 2026).
• Algorithms for certain knot invariants (Morimae et al., 2014), spectral density estimation, and specific classically hard correlation functions (Moulik et al., 2024).

In DQC1-parameterized quantum machine learning, the normalized trace of data-parameterized unitaries $a \in \{0,1\}$4 implements a wide range of function classes. The class of learnable functions can be characterized as partial Fourier series, with the set of Fourier modes scaling as $a \in \{0,1\}$5 for $a \in \{0,1\}$6 mixed qubits and $a \in \{0,1\}$7 layers, matching the expressivity of universal variational quantum circuits up to a constant overhead in $a \in \{0,1\}$8 or $a \in \{0,1\}$9 (Kim et al., 2024). Training via analytic gradients reduces to further DQC1-type trace estimation, with gradients also efficiently measurable within the same protocol.

Quantum kernels for supervised learning can be estimated via DQC1 circuits by preparing two data-encoded unitaries $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$0, computing $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$1, and extracting

$$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$2

from the $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$3-basis measurement on the clean qubit. Coherence consumption on the clean qubit directly reflects the kernel value, and positive geometric discord persists in the presence of noise, accounting for robustness of this protocol on NISQ platforms (Karimi et al., 2022).

5. Quantum Correlations, Resource Limits, and No-Go Theorems

Despite DQC1’s computational power, there are sharp limitations on state synthesis:

• It is impossible, via unitary evolution and partial trace (discarding or measuring qubits), to prepare more than the initial number of clean qubits in a pure state on any subsystem; specifically, from $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$4 clean qubits plus $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$5 maximally mixed qubits, no pure state on $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$6 output qubits can be synthesized (Stier, 2024).
• The probability of obtaining an $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$7-qubit pure state (via measurement and postselection) is bounded by $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$8, exponentially small unless $$P(a) = \operatorname{Tr}\big[(|a\rangle\langle a| \otimes I^{\otimes n})\, U\, \rho_\mathrm{in}\, U^\dagger\big].$$9.
• No similar protocol can generate low-temperature Gibbs states on more than $|+\rangle$0 qubits, with the success probability for preparing such states also decaying exponentially with $|+\rangle$1.

These no-go results place intrinsic, information-theoretic lower bounds on the runtime and resource requirements of any repeated-interaction or state-preparation strategy based purely on DQC1 primitives. Preparation of high-purity states on $|+\rangle$2 qubits from a maximally mixed register requires at least $|+\rangle$3 fresh clean-qubit operations (Stier, 2024).

6. Quantum Advantage, Simulatability, and Open Directions

Although separable from spectrum under generic bipartitions, DQC1 circuits cannot be efficiently simulated classically unless unexpected complexity collapses occur. For instance:

• Efficient simulation of DQC1 output probabilities with weak multiplicative error collapses PH to AM (Fujii et al., 2014).
• Simulation for DQC1$|+\rangle$4 ($|+\rangle$5) with multiplicative approximation $|+\rangle$6 collapses PH to $|+\rangle$7 (Morimae et al., 2013).
• For trace estimation over sparse log-local Hamiltonians and functions of high approximate degree, the classical query complexity for normalized trace estimation is exponential in the approximate degree, while the quantum complexity is polynomial, yielding explicit quantum-classical separations (Ji et al., 2 Apr 2026).

Furthermore, in specific algorithmic contexts, the presence of quantum discord or coherence, not entanglement, is sufficient for quantum computational advantage; entanglement may be vanishingly small or absent (Kay, 2015, Boyer et al., 2016, Ali, 2013). Not all quantum correlations confer computational benefit: in the Deutsch–Jozsa problem, intermediate states in DQC1 may exhibit transient discord, but do not outpace classical probabilistic sampling (Santos et al., 2013).

These features identify DQC1 as a precise setting for distinguishing the role of different quantum resources, for benchmarking quantum advantage with minimal hardware requirements, and for concretely delimiting the operational power of highly-mixed-state quantum computing.