Constraint-Preserving Mixers in QAOA

• Constraint-preserving mixers are mixing operators in QAOA that maintain evolution strictly within the feasible subspace defined by hard constraints.
• They encompass designs such as XY-mixers, Grover-style, and logical mixers, offering efficient exploration while reducing search space via exact compilation and shallow circuits.
• These mixers enable techniques like warm-starting and iterative biasing to enhance optimization performance on problems including graph coloring, TSP, and portfolio selection.

Constraint-preserving mixers are mixing operators in QAOA and related alternating-operator constructions that are designed so that quantum evolution remains inside a feasible subspace defined by hard constraints, rather than relying solely on penalty terms in the cost Hamiltonian. In the standard formulation, a level-$p$ circuit has the form

$$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$

and the central design question is how to choose $U_M(\beta)=e^{-i\beta H_M}$ and $|\psi_0\rangle$ so that hard constraints are preserved exactly while the feasible subspace remains sufficiently connected for optimization. Across the literature, this program appears as symmetry-preserving $XY$-mixers for fixed Hamming weight, Grover-style mixers based on projectors onto feasible superpositions, stabilizer-based logical mixers for arbitrary specified subspaces, graph-controlled permutation mixers for scheduling, and warm-started variants that bias the search without leaving the constrained sector (Wang et al., 2019, Bärtschi et al., 2020, Fuchs et al., 2023, Bucher et al., 2 Apr 2026).

1. Feasible-subspace dynamics in constrained QAOA

The original $X$-mixer,

$H_X = \sum_i X_i,$

induces transitions between all computational-basis states and therefore does not preserve hard constraints such as one-hot or fixed-cardinality conditions. In graph coloring with one-hot encoding, for example, each vertex $v$ is represented by qubits $\{x_{v,c}\}_{c=1}^{\kappa}$ with

$\sum_{c=1}^{\kappa} x_{v,c} = 1,$

so the feasible subspace is exponentially smaller than the full Hilbert space: for $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$0 vertices and $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$1 colors, the full dimension is $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$2, the feasible dimension is $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$3, and the ratio is

$$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$4

This makes amplitude leakage into infeasible states especially costly (Wang et al., 2019).

In the constraint-preserving formulation, the target is a mixer family that is feasibility preserving and explores the feasible subspace. One explicit definition requires that $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$5 for every feasible $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$6, and that for all feasible $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$7 there exist $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$8 and $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$9 such that $U_M(\beta)=e^{-i\beta H_M}$0. This is the precise sense in which a mixer is both constraint-preserving and ergodic on the feasible sector (Palackal et al., 2023).

Penalty methods remain a baseline, but they enlarge the search space and complicate the energy landscape. In the multi-constrained setting, one standard QUBO route adds every constraint as a quadratic penalty, converts inequalities with slack variables, and then applies the $U_M(\beta)=e^{-i\beta H_M}$1-mixer over the full $U_M(\beta)=e^{-i\beta H_M}$2-dimensional space. Later work explicitly contrasts this with workflows that encode one-hot constraints directly into the mixer and initial state, thereby reducing the effective search space to $U_M(\beta)=e^{-i\beta H_M}$3 for one-hot groups of sizes $U_M(\beta)=e^{-i\beta H_M}$4 (Bucher et al., 3 Jun 2025).

2. Conserved quantities and the central role of $U_M(\beta)=e^{-i\beta H_M}$5-mixers

The canonical constraint-preserving mixer for one-hot and fixed-cardinality constraints is the $U_M(\beta)=e^{-i\beta H_M}$6-mixer. For a block of $U_M(\beta)=e^{-i\beta H_M}$7 qubits, the full $U_M(\beta)=e^{-i\beta H_M}$8-mixer is

$U_M(\beta)=e^{-i\beta H_M}$9

and the ring version is

$|\psi_0\rangle$0

with indices modulo $|\psi_0\rangle$1. The conserved quantity is the Hamming-weight operator

$|\psi_0\rangle$2

and the key relation is

$|\psi_0\rangle$3

Thus, if the initial state lies in the weight-1 sector, evolution under the $|\psi_0\rangle$4 mixer stays in that sector exactly (Bucher et al., 3 Jun 2025).

For one-hot graph coloring, this conservation law is often written in terms of the total $|\psi_0\rangle$5 operator on a vertex register,

$|\psi_0\rangle$6

with feasible states satisfying $|\psi_0\rangle$7. Because

$|\psi_0\rangle$8

the mixer acts as a continuous-time quantum walk on the color graph while never leaving the Hamming-weight-1 subspace. On that subspace, each $|\psi_0\rangle$9 term acts essentially as a swap between $XY$0 and $XY$1 (Wang et al., 2019).

The natural unbiased initial state for such dynamics is the generalized W-state,

$XY$2

and for multiple one-hot blocks one uses $XY$3. This state lies exactly in the feasible subspace and is an eigenstate of the $XY$4-mixer, making it the analogue of $XY$5 for $XY$6-mixer QAOA (Wang et al., 2019).

The same mechanism extends beyond one-hot to general cardinality constraints. For binary selection with $XY$7, the embedded constraint operator is proportional to total $XY$8-magnetization, and each $XY$9 commutes with it. This makes $X$0-mixers prototypical constraint-preserving mixers for cardinality-constrained optimization (Kordonowy et al., 23 May 2025).

3. Exact compilation, shallow circuits, and hardware-aware architectures

A major result of the $X$1-mixer literature is that constraint preservation need not imply prohibitive circuit depth. For the ring mixer on $X$2 qubits, Jordan–Wigner mapping yields a quadratic fermionic Hamiltonian diagonalizable by a fermionic Fourier transform. This gives an exact implementation of $X$3 by: applying the fermionic fast Fourier transform, performing single-qubit $X$4-rotations in momentum space, and applying the inverse transform. Under all-to-all connectivity, the FFFT can be implemented in depth $X$5; under nearest-neighbor connectivity, Givens rotation networks give depth $X$6. The implementation is exact and introduces no Trotter error (Wang et al., 2019).

For the complete-graph $X$7-mixer on a one-hot block, the same work gives an exact decomposition into $X$8 commuting partitions when $X$9. Each partition consists of disjoint pairwise $H_X = \sum_i X_i,$0 interactions induced by bit-flip patterns on the binary labels of the colors, so each layer is constant depth and the total depth is $H_X = \sum_i X_i,$1 on all-to-all hardware. This avoids the $H_X = \sum_i X_i,$2 or worse depth that a naïve Trotterization would incur for high precision (Wang et al., 2019).

Later architecture work combines these ideas with other constraint encodings. In a multi-constrained QAOA workflow, one-hot constraints are enforced by ring $H_X = \sum_i X_i,$3-mixers and $H_X = \sum_i X_i,$4-state initialization, while inequality constraints are implemented through QPE-based Indicator Functions. For a group of size $H_X = \sum_i X_i,$5, the initialization depth is

$H_X = \sum_i X_i,$6

and the ring mixer cost per QAOA layer is summarized as

$H_X = \sum_i X_i,$7

Because groups are independent, mixer depth scales with the largest one-hot block size rather than the total number of variables (Bucher et al., 3 Jun 2025).

A separate practical caveat arises under Trotterized Adiabatic Evolution. For fully connected $H_X = \sum_i X_i,$8-mixers on a single large constraint block, the number of non-commuting term pairs scales as $H_X = \sum_i X_i,$9, and Trotter errors can dominate. For multiple disjoint local constraints, by contrast, blockwise $v$0-mixers commute across blocks and the dominant Trotter error depends on the size of each block rather than total problem size. This identifies constraint locality as the main hardware-level criterion for when $v$1-mixers remain advantageous (Awasthi et al., 4 May 2026).

4. Beyond $v$2: Grover, logical, permutation, and algebraic mixer constructions

Grover-type mixers provide a different route. If one can efficiently prepare

$v$3

then GM-QAOA uses the projector mixer

$v$4

This mixer is automatically constraint-preserving because it acts only through the feasible superposition, and it can be implemented exactly as

$v$5

The trade-off is that complexity shifts from Hamiltonian design to state preparation and a global multi-controlled phase-shift gate (Bärtschi et al., 2020).

Locally acting Grover mixers reduce that overhead when the initial state factorizes over disjoint subsystems. If

$v$6

the mixer can be replaced by

$v$7

On a 7-qubit exact-cover instance, the mixer CNOT count drops from 218 for global GM-QAOA to 28 for the local construction. On a 9-qubit TSP instance, the corresponding counts are 572 and 54, while convergence behavior remains comparable (Choi et al., 10 Jun 2026).

LX-mixers generalize the well-known $v$8 and $v$9 mixers and relax the Grover mixer by using the stabilizer formalism. For a pair of feasible basis states $\{x_{v,c}\}_{c=1}^{\kappa}$0, the primitive mixer

$\{x_{v,c}\}_{c=1}^{\kappa}$1

can be written as a logical $\{x_{v,c}\}_{c=1}^{\kappa}$2 times a stabilizer projector,

$\{x_{v,c}\}_{c=1}^{\kappa}$3

This gives a systematic recipe: partition the feasible subspace into stabilizer codespaces, apply logical rotational $\{x_{v,c}\}_{c=1}^{\kappa}$4 gates within them, and choose a connected union graph over feasible states. Numerical examples show dramatic CX reductions relative to previous constructions (Fuchs et al., 2023).

A separate line of work gives a fully general algebraic formulation. Classical constraints are embedded as observables $\{x_{v,c}\}_{c=1}^{\kappa}$5, and one seeks Hamiltonians or unitary primitives that commute with all $\{x_{v,c}\}_{c=1}^{\kappa}$6. In the general case, finding such operators is NP-Complete, whereas the locality-bounded variant is in P for constant locality $\{x_{v,c}\}_{c=1}^{\kappa}$7 because all $\{x_{v,c}\}_{c=1}^{\kappa}$8-local candidates can be enumerated and checked in worst-case polynomial time (Leipold et al., 2024). Graph-controlled permutation mixers for scheduling instantiate the same principle in a different language: constraints are encoded by a conflict graph, and graph-controlled permutation unitaries are proven to be both feasibility preserving and capable of exploring the feasible subspace (Palackal et al., 2023).

5. Warm-starting, trainability, and iterative biasing

Warm-starting introduces bias toward promising regions of the feasible subspace, but it creates a structural problem if the initial state ceases to be the ground state of the mixer. For one-hot constraints, a biased W-state

$\{x_{v,c}\}_{c=1}^{\kappa}$9

is generally not an eigenstate of the standard $\sum_{c=1}^{\kappa} x_{v,c} = 1,$0-mixer unless all $\sum_{c=1}^{\kappa} x_{v,c} = 1,$1. A recent construction resolves this by defining

$\sum_{c=1}^{\kappa} x_{v,c} = 1,$2

and proving that $\sum_{c=1}^{\kappa} x_{v,c} = 1,$3 is the unique ground state of $\sum_{c=1}^{\kappa} x_{v,c} = 1,$4 in the Hamming-weight-1 sector. The construction extends to arbitrary connected topologies $\sum_{c=1}^{\kappa} x_{v,c} = 1,$5, preserves Hamming weight, and admits a shallow decomposition in terms of standard $\sum_{c=1}^{\kappa} x_{v,c} = 1,$6 gates and single-qubit $\sum_{c=1}^{\kappa} x_{v,c} = 1,$7-rotations (Bucher et al., 2 Apr 2026).

This enables Iterative Warm-Starting (IWS), where the QAOA angles are optimized once for an initial uniform distribution and then a classical update rule repeatedly modifies the probabilities $\sum_{c=1}^{\kappa} x_{v,c} = 1,$8 of each one-hot block based on previous samples. The mixer Hamiltonian is updated accordingly, so the current biased W-state remains the mixer ground state at every iteration. On Max-$\sum_{c=1}^{\kappa} x_{v,c} = 1,$9-Cut and TSP instances, IWS-QAOA increases the probability of sampling optimal solutions by orders of magnitude compared to standard XY-QAOA, and hardware experiments on ibm_boston with 144-qubit one-hot instances successfully identify optimal solutions after greedy post-processing repairs infeasible measurements caused by noise (Bucher et al., 2 Apr 2026).

A related trainability analysis studies the dynamical Lie algebras generated by different $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$00-mixer topologies. A ring $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$01-mixer with arbitrary $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$02 gates yields a DLA of size $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$03, whereas all-to-all $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$04-mixers or the addition of $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$05 gates produce exponentially large DLAs. This motivates a warm-start strategy in which optimization begins in a polynomial-sized DLA and then lifts into the exponentially expressive one. Numerical results on Portfolio Optimization, Sparsest $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$06-Subgraph, and Graph Partitioning show improved convergence and higher-quality optima for both shared-angle and multi-angle QAOA (Kordonowy et al., 23 May 2025).

Warm-starting can also be combined with constraint-preserving $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$07-evolution without redefining the mixer. For 5-city TSP instances, a MaxCut-based warm-start is projected into the one-hot subspace to create a biased product state compatible with an $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$08-mixer. The combined warm-started $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$09 method reaches a mean of $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$10 true optimal tours at $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$11, substantially outperforming both a pure $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$12-mixer with W-state initialization and a warm-start-only variant (Carmo et al., 28 Apr 2025).

6. Empirical regimes, trade-offs, and open problems

The strongest empirical case for constraint-preserving mixers comes from problems where the feasible subspace is naturally one-hot or block-cardinality constrained. On graph coloring, level-1 QAOA with an $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$13-mixer and optimized penalties gives approximation ratio $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$14 on 3-coloring a triangle, whereas an $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$15 ring mixer with W-state initial state reaches values near $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$16. For 2-coloring a triangle, the optimized $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$17-mixer reaches $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$18, while the $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$19-mixer reaches $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$20 at level 1. On the Prism graph, the simultaneous ring $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$21-mixer and W-state initial state give average approximation ratio $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$22 at $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$23, and by $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$24 the probability of sampling an optimal coloring exceeds 0.6 (Wang et al., 2019).

For multi-constrained industrial-style problems, the best results come from hybrid architectures. On Prosumer Problem instances, IF+XY achieves RAAR $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$25 across all instances at $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$26, retains the highest $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$27 among the tested methods, and has the smallest TTS$$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$28. Compared to QUBO on the smaller instances where QUBO is simulable, the average speedup factor is reported as

$$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$29

This suggests that search-space restriction can outweigh per-layer circuit complexity when evaluated through time-to-solution rather than raw depth (Bucher et al., 3 Jun 2025).

A common misconception is that $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$30-mixers are uniformly preferable to penalty methods. The Trotterized adiabatic study rejects that blanket statement. For a single global equality constraint spanning all variables, Trotter errors significantly impair $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$31-mixer performance, and standard $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$32-mixers can be more robust. For multiple disjoint local blocks, however, $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$33-mixers outperform $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$34-mixers by several orders of magnitude even under Trotterized evolution. This suggests that the relevant discriminator is not merely “constraint-preserving versus penalties,” but the locality structure of the constraint decomposition (Awasthi et al., 4 May 2026).

Practical applications also reveal an objective-design trade-off. In a direct-indexing-style portfolio selection experiment with Dicke-state initialization, an $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$35-mixer, and a Trotterized parameter schedule, the QAOA approach achieves a Sharpe Ratio of $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$36, compared to $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$37 for Simulated Annealing and $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$38 for HRP. At the same time, the reported turnover is $$|\psi(\boldsymbol\gamma,\boldsymbol\beta)\rangle = \prod_{\ell=1}^p \Big[ U_M(\beta_\ell)\, e^{-i\gamma_\ell H_{PS}} \Big] \,|\psi_0\rangle,$$39, which indicates that constraint-preserving optimization can improve the discrete selection stage while still requiring explicit transaction-cost or turnover regularization if implementation frictions matter (Mancilla et al., 16 Feb 2026).

Open problems remain stable across the literature. Scaling beyond the small-instance regime is unresolved for several constructions; the effect of noise on elaborate mixers and on nontrivial state preparation remains incompletely characterized; overlapping constraint families such as TSP’s two-way one-hot structure still require specialized higher-order mixers or a mixer-plus-penalty split; and general-purpose synthesis of low-depth, high-connectivity mixers for arbitrary feasible subspaces remains computationally hard in the worst case (Wang et al., 2019, Leipold et al., 2024, Awasthi et al., 4 May 2026). A plausible implication is that future progress will continue to combine three ingredients rather than relying on any single primitive: conserved-quantity mixer design, problem-structured compilation, and warm-start or oracle mechanisms that reduce effective search without destroying alignment inside the feasible sector.