In numerical simulations evaluating gate-based quantum algorithms for multi-state financial decisions, the architectural structure of quantum mixer operators directly dictates solution quality, with highly connected designs excelling in idealized environments but deteriorating rapidly in the presence of noise. Researchers investigating the quantum approximate optimization algorithm applied to ternary portfolio optimization found that while advanced constraint-preserving mixers outperform the conventional Standard Mixer in noiseless simulations, environmental decoherence fundamentally reverses this hierarchy. The team conducted detailed evaluations across multiple algorithmic variations to determine how mixer connectivity and initial state preparation govern practical performance when balancing risk and return across investment assets.
Portfolio optimization represents a foundational challenge in quantitative financial engineering, originally formalized under the Markowitz mean-variance framework to balance expected investment return against risk quantified by portfolio variance. Although conventional quantum formulations typically employ binary representations where an asset is either held or excluded, practical market strategies frequently permit short selling, requiring a ternary representation that models long positions, no positions, and short positions. To capture these mechanics, the research team implemented a discrete combinatorial formulation governed by ternary variables that explicitly incorporate short positions alongside a fixed net investment budget constraint. Under this modeling framework, identifying the asset configuration that minimizes total risk while penalizing portfolio underperformance requires navigating a discrete search space subject to strict combinatorial constraints.
To solve this constrained problem using gate-based quantum circuits, the researchers mapped the ternary asset states onto qubits using a two-qubit encoding scheme for each asset. In this encoding, individual selection variables are represented by two binary quantities indicating whether an asset is held long or sold short, yielding two-qubit basis states representing long, short, and no-position holdings. Because this mapping introduces a degeneracy in which multiple distinct computational basis states correspond to having no position in an asset, the choice of quantum mixer and initial state becomes critical. The researchers evaluated the conventional Standard Mixer, four distinct XY-based mixer configurations—comprising the XY Ring Mixer, the XY Parity Ring Mixer, the XY Full Mixer, and the Quantum Alternating Mixer-Phase Ansatz—and the Quantum Walk Optimization Algorithm, which executes continuous-time quantum walks over non-degenerate feasible state graphs.
A critical technical distinction emphasized in the investigation centers on how constraints are enforced throughout algorithmic execution. When employing the unconstrained Standard Mixer, which applies transverse single-qubit rotations across the entire register, the net investment constraint must be treated as a soft constraint enforced through an additive quadratic penalty term in the cost function. Conversely, XY-based mixers inherently preserve the total Hamming weight of the register, enabling the circuit to preserve the net investment constraint as a strict hard constraint throughout the variational evolution. Unlike prior studies that initialized XY mixers from simplified, binomially biased feasible states, the authors constructed an unbiased, uniform superposition across all constraint-satisfying computational basis encodings using tailored Dicke state preparation circuits, preventing initialization bias from persisting through subsequent circuit layers.
Numerical benchmarks conducted in idealized, noiseless conditions demonstrated that mixer connectivity strongly influences algorithmic convergence across varying circuit depths. Evaluating twenty random portfolio instances generated from historical German stock index market data with five assets and a net investment budget constraint, the researchers tracked the average approximation ratio and optimal solution probability as circuit depth scaled. In noiseless state-vector simulations, all constraint-preserving XY mixers achieved superior performance compared to the penalized Standard Mixer even at depth one. The XY Full Mixer and the integrated phase ansatz reached optimal solution probabilities near sixty percent and approximation ratios exceeding ninety percent at shallow depths, driven by all-to-all pairwise qubit couplings that facilitate rapid state exploration across the feasible subspace.
The Quantum Walk Optimization Algorithm exhibited an entirely different depth-dependent behavior compared to the qubit-encoded XY mixers. Because this quantum walk operates on a complete graph linking all non-degenerate canonical feasible portfolios without encoding redundancies, its mixing operations are uniformly and symmetrically distributed across states. At shallow circuit depths, this global symmetry dispersed probability amplitudes broadly, causing the walk-based algorithm to perform slightly below the highly connected XY mixers. However, as circuit depth increased, cumulative constructive interference across successive layers enabled the quantum walk to concentrate amplitude effectively, eventually achieving the smallest mean approximation error among all tested approaches for five-asset instances at depth seven.
When problem instances were expanded to eight assets with a net investment constraint of four, the underlying scaling properties of the different algorithmic formulations became more pronounced. Although the XY Full Mixer and the integrated phase ansatz maintained superior solution quality, the overall performance across all methods experienced degradation due to the combinatorial expansion of the search space. In this larger setting, the quantum walk algorithm trailed the most expressive XY mixers in optimal solution probability despite operating within a strictly non-degenerate subspace of 266 canonical portfolios. The findings indicate that simply reducing the dimension of the feasible subspace is insufficient to guarantee superior optimization at shallow circuit depths, as structured pairwise exchange interactions provide essential expressibility to steer probability amplitudes toward the global optimum.
To evaluate performance under conditions representative of noisy intermediate-scale quantum hardware, the authors subjected the QAOA variants to simulated depolarizing noise applied across all single-qubit and two-qubit logic gates. As the depolarizing noise parameter increased from zero to one percent, a pronounced crossover occurred: beyond specific noise thresholds, the elementary Standard Mixer systematically surpassed the advanced XY-based mixers in both average approximation ratio and optimal solution probability. For deeper circuits, this transition took place at increasingly lower noise intensities, demonstrating that the structural overhead required to maintain hard constraints can severely undermine algorithmic performance in realistic computing environments.
Analysis of the circuit resource requirements revealed the precise structural mechanism behind this noise-induced performance collapse. While the Standard Mixer requires only single-qubit rotations and constant initialization depth without two-qubit entanglement, preparing the uniform Dicke superposition for XY mixers demands entangling circuits whose gate counts scale substantially with system size. Circuit complexity simulations demonstrated that for a ten-qubit system, preparing the constrained initial state requires hundreds of controlled-NOT gates before the actual variational layers begin. Under depolarizing noise, this extensive gate overhead causes rapid state corruption, driving the system out of the constraint-satisfying subspace and negating the primary theoretical advantage of constraint-preserving mixer designs.
The authors contextualized their findings by emphasizing that the numerical demonstrations were conducted on modest system sizes of five and eight assets, where all feasible portfolios can be completely enumerated classical methods. Consequently, the observed advantages of expressive mixers should be interpreted as structural comparative insights rather than a demonstration of practical quantum advantage or unconstrained scalability. Furthermore, the simplified discrete portfolio model allocates equal position weights rather than continuous capital allocations, and actual quantum hardware features complex, device-dependent noise structures beyond uniform depolarizing channels. The researchers concluded that selecting optimal quantum optimization architectures requires balancing mathematical expressibility against hardware decoherence, noting that while shallow, low-overhead mixers are preferable for noisy near-term processors, highly connected mixers and quantum walk formulations remain promising candidates for future fault-tolerant implementations.
Beyond structural connectivity, the performance of each variational algorithm was shaped by the classical optimization routines paired with quantum simulators. To optimize the continuous variational parameters, the researchers used the gradient-based SLSQP algorithm during noiseless state-vector calculations, whereas finite-shot sampling evaluations with the Qiskit simulator relied on the Nelder-Mead method across 3000 shots for five-asset models and 10,000 shots for eight-asset models. For the quantum walk optimization algorithm, classical parameter tuning utilized the BFGS method for state vectors and Nelder-Mead for sampled distributions. In the presence of simulated depolarizing noise, parameter optimization was handled by the COBYLA routine across 8192 measurement shots, evaluated using five specific DAX stocks: LIN.DE, BAYN.DE, VNA.DE, MTX.DE, and MUV2.DE.
The compilation requirements and gate decompositions also introduced distinct differences in operational complexity among the evaluated mixer designs. Although all gate-based QAOA variants share an asymptotic scaling with circuit depth and the square of asset count due to dense quadratic cost interactions, their internal circuit schedules differ substantially. The XY Parity Ring Mixer partitions adjacent qubit couplings into alternating even and odd site subsets, executing disjoint two-qubit operations in parallel to reduce circuit depth relative to the sequential XY Ring Mixer. Meanwhile, the Quantum Alternating Mixer-Phase Ansatz consolidates full pairwise exchange operations and phase rotations into unified two-qubit blocks, reducing controlled-NOT gate counts to approximately three-quarters of those required by separate implementations, despite requiring additional single-qubit operations.
In contrast, gate-based implementations of the quantum walk algorithm bypass the redundant no-position encoding entirely, but face significant hardware scaling barriers. Physical realization of the complete-graph quantum walk requires dedicated indexing operations, arithmetic registers, and ancillary qubits, driving resource scaling to a cubic dependence on asset count when decomposed with standard ripple-carry circuits. To address the heavy classical optimization burden of deep quantum walks, the authors highlighted recent non-variational QWOA frameworks that govern phase and walk angles via predetermined parameter schedules with only a few global hyperparameters. This design strategy could potentially eliminate variational parameter instabilities while preserving the strict feasibility advantages of canonical state-space evolution in future fault-tolerant systems.
Subject of Research: Performance comparison of QAOA mixers and QWOA for ternary portfolio optimization under noiseless and depolarizing noise conditions
Article Title: Performance comparison of QAOA mixers for ternary portfolio optimization
Article References: Yamamura, S., Watanabe, S., Kunimi, M., Saito, K., & Nikuni, T. (2026). Performance comparison of QAOA mixers for ternary portfolio optimization. Quantum Information Processing, 25(12), Article 377. https://doi.org/10.1007/s11128-026-05322-x
Image Credits: AI Generated
DOI: 10.1007/s11128-026-05322-x
Keywords: Quantum approximate optimization algorithm, QAOA, Portfolio optimization, Ternary optimization, Quantum mixers, XY mixers, QAMPA, Quantum walk optimization algorithm, QWOA, Depolarizing noise, Dicke state preparation, NISQ devices, Financial engineering, DAX 30, Short selling, Constraint satisfaction
News Source: Katie Riggs. (November 19, 2026). Simulations Show Quantum Mixer Design Shapes Performance in Ternary Portfolio Optimization. Scienmag.



