Variational Iterative Rotation Algorithm: Combinatorial Optimization with Classical Kicked Tops

arXiv:2604.01512 · cond-mat.dis-nn · Submitted 2026-04-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Variational Iterative Rotation Algorithm".

Mira: Here's the gist of the paper. Variational Iterative Rotation Algorithm (VIRAL) is a classical optimization algorithm based on kicked tops that outperforms the Quantum Approximate Optimization Algorithm (QAOA) on canonical spin-glass benchmarks by leveraging an underlying iterative rotation structure and reducing the many-body problem to an effective Landau-Lifshitz dynamics.

Kai: Core Methodology The VIRAL algorithm is fundamentally a dynamical system composed of classical 3-component unit vectors (kicked tops) governed by a deterministic Hamiltonian.

Mira: The protocol initializes all N spins identically in the xz-plane at an angle α, and then evolves through p discrete layers where each layer consists of a site-dependent rotation around the z-axis followed by a uniform rotation around the x-axis.

Kai: The variational parameters are the transverse and longitudinal rotation angles at each of the p layers.

Paper summary: Kai: So to recap, this paper introduces VIRAL as a classical formulation of QAOA using kicked tops. The main claim is that this classical approach outperforms QAOA on canonical spin-glass benchmarks across all circuit depths.

Mira: They argue that this is because the underlying iterative rotation structure allows the system to navigate the optimization landscape more effectively than its quantum counterpart, and it reduces the many-body problem to an effective Landau-Lifshitz dynamics.

Lev: So they are claiming a classical implementation can be more efficient in this context, which is a big deal when you consider the overhead of error correction in real hardware.

Kai: They show that their classical protocol converges to the ground state value linearly in one over p, and for large depths, it follows a Floquet protocol where spins are driven toward polar Ising configurations.

Mira: The paper also points out that the effectiveness of QAOA-like protocols is derived from that iterative rotation structure itself, and it suggests that quantum fluctuations actually hamper the optimization process rather than helping it.

Lev: That's a strong pushback against the idea that quantum fluctuations are inherently beneficial for these kinds of control schemes.

Kai: So what does this mean for the world? It suggests that we might not need to rely on complex quantum algorithms just because they are quantum, if a well-designed classical dynamical system can capture the necessary iterative structure.

Conclusion: Kai: So, looking at the Variational Iterative Rotation Algorithm paper, it’s Flaviano Morone, Andrew D. Kent, and Dries Sels that put this classical kicked-top idea forward. The implication is that we have a way to systematically optimize many-body problems using purely classical dynamics if you can structure the rotations correctly.

Mira: It suggests that the advantage isn't necessarily in quantum mechanics itself, but in how we set up the iterative process—specifically, how you control the rotation angles at each layer.

Lev: From my side of error correction, it means that if we can find a classical dynamical system that captures this kind of structure reliably without needing full quantum resources, then the overhead for running algorithms on actual hardware drops significantly.

Kai: So, when we talk about QAOA versus VIRAL, the paper is basically saying the iterative rotation structure is what matters most here and that classical dynamics are better controlled with minimal fluctuations.

Mira: It really shifts the question from "which algorithm is better" to "how do we build a classical system that mirrors this ideal iterative control?"

Lev: And for people who only listen to our show, it means they don't have to assume quantum optimization is always the best tool available.

Kai: Exactly. We’re looking at where the practical advantage actually lands when you translate these ideas from theory into something you can actually build and measure.

Flaviano Morone, Andrew D. Kent, Dries Sels

Center for Quantum Phenomena, Department of Physics, New York University · Department of Physics, Boston University · Center for Computational Quantum Physics, Flatiron Institute

cond-mat.dis-nn

Submitted: 2026-04-02

Updated: 2026-10-28

Comments: 31 pages, 7 figures

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 87/100

The gist: Variational Iterative Rotation Algorithm (VIRAL) is a classical optimization algorithm based on kicked tops that outperforms the Quantum Approximate Optimization Algorithm (QAOA) on canonical…

Key concepts

Kicked Tops
These are classical 3-component unit vectors that form the core dynamical system of VIRAL. They evolve under a deterministic Hamiltonian, meaning their motion is entirely predictable based on their current state and the applied forces. This classical framework allows for the systematic exploration of complex optimization landscapes.
Landau-Lifshitz Dynamics
The problem is reduced to an effective Landau-Lifshitz dynamics for a single spin in a stochastic magnetic field. This means the complex many-body spin system behaves like a simpler, single-spin model under certain conditions. The underlying structure reveals that the effective field has a nearly rank-one covariance matrix, simplifying the stochastic dynamics significantly.
Iterative Rotation Structure
The effectiveness of VIRAL stems from its iterative rotation structure, which involves site-dependent rotations around the z-axis followed by uniform rotations around the x-axis in discrete layers. This structure drives the system toward polarized Ising configurations, and this classical implementation is found to be more controlled than quantum counterparts.
QAOA Performance Limitation
The paper argues that QAOA's success is due to its iterative rotation structure, but quantum fluctuations hinder optimization rather than aiding it. In contrast, VIRAL's classical nature ensures highly controlled dynamics with minimal noise covariance, allowing for superior convergence toward the ground state energy.

Terminology

Summary

Variational Iterative Rotation Algorithm (VIRAL) is a classical optimization algorithm based on kicked tops that outperforms the Quantum Approximate Optimization Algorithm (QAOA) on canonical spin-glass benchmarks by leveraging an underlying iterative rotation structure and reducing the many-body problem to an effective Landau-Lifshitz dynamics. p. 0

Where the paper says it

Our results demonstrate that the effectiveness of QAOA-like protocols derives primarily from their underlying iterative rotation structure, and that a classical implementation of it outperforms its quantum counterpart. We further elucidate its efficiency by reducing the many-body classical evolution to an effective Landau-Lifshitz dynamics for a single spin in a stochastic magnetic field.

Core Methodology

The VIRAL algorithm is fundamentally a dynamical system composed of classical 3-component unit vectors (kicked tops) governed by a deterministic Hamiltonian. The protocol initializes all N spins identically in the xz-plane at an angle α, and then evolves through p discrete layers where each layer consists of a site-dependent rotation around the z-axis followed by a uniform rotation around the x-axis. p. 1

Where the paper says it

The system then evolves through p discrete layers. Each layer n comprises two successive rotations: first, a site-dependent rotation Rz(γn) around the z-axis, where each spin i precesses by an angle γnh z i determined by its local field h z i = P j̸=i Jij s z j ; second, a uniform rotation Rx(βn) around the x-axis, kicking all spins by a common angle βn.

The variational parameters are the transverse and longitudinal rotation angles at each of the p layers. p. 0
Where the paper says it

, Boston, Massachusetts 02215, USA 3Center for Computational Quantum Physics, Flatiron Institute, New York, NY, USA We investigate a classical formulation of the Quantum Approximate Optimization Algorithm (QAOA), realized as a Hamiltonian dynamical system of classical kicked tops, which we call the Variational Iterative Rotation Algorithm (VIRAL). The variational parameters are the transverse and longitudinal rotation angles at each of the p layers of the circuit.

Performance Comparison

The authors find that VIRAL outperforms QAOA on the canonical Sherrington-Kirkpatrick spin-glass benchmark at all circuit depths, with the energy density converging to the ground state value linearly in 1/p. For large circuit depths, the optimized dynamics follow a Floquet protocol in which a pitchfork bifurcation destabilizes the equatorial fixed point and drives the spins toward polar Ising configurations. p. 0

Where the paper says it

We find that VIRAL outperforms QAOA on the canonical Sherrington-Kirkpatrick spin-glass benchmark at all circuit depths, with the energy density converging to the ground state value linearly in 1/p. For large circuit depths, the optimized dynamics follows a Floquet protocol in which a pitchfork bifurcation destabilizes the equatorial fixed point and drives the spins toward polar Ising configurations.

Numerical simulations confirm that VIRAL reaches an energy density of approximately E/N ≈ −0. p. 1
Where the paper says it

Numerical simulations shown in Fig. 2A confirm that our algorithm reaches an energy density of approximately E/N ≈ −0.

761, scaling linearly in 1/p and notably outperforming QAOA which extrapolates to a higher energy value of roughly E/N ≈ −0. p. 1
Where the paper says it

761 (very close to the exact Parisi ground state value of −0. ), scaling linearly in 1/p and notably outperforming QAOA which extrapolates to a higher energy value of roughly E/N ≈ −0.

757.

Physical Insights from Dynamics

The effectiveness of the QAOA-like protocols is derived primarily from their underlying iterative rotation structure, and that a classical implementation of it outperforms its quantum counterpart. The problem reduces to an effective Landau-Lifshitz dynamics for a single spin in a stochastic magnetic field where the covariance matrix of the effective field reveals a nearly rank-one structure in which a single mode dominates the stochastic dynamics. p. 0

Where the paper says it

We further elucidate its efficiency by reducing the many-body classical evolution to an effective Landau-Lifshitz dynamics for a single spin in a stochastic magnetic field. In this picture, the covariance matrix of the effective field reveals a nearly rank-one structure in which a single mode dominates the stochastic dynamics.

This contrasts with quantum fluctuations which make the noise covariance of the effective quantum model of higher rank, hampering the control of the system. p. 0
Where the paper says it

In this picture, the covariance matrix of the effective field reveals a nearly rank-one structure in which a single mode dominates the stochastic dynamics. In contrast, quantum fluctuations make the noise covariance of the effective quantum model of higher rank, hampering the control of the system.

Experimental Realization

Nanometer-scale magnetic tunnel junctions (MTJ) free layers are proposed as a natural physical platform for implementing VIRAL. At these length scales, the magnetization of the MTJ free layer is well approximated as a single macrospin whose dynamics is accurately described by the classical Landau-Lifshitz–Gilbert equation. p. 6

Where the paper says it

PHYSICAL REALIZATION OF THE ITERATIVE ROTATION ALGORITHM We propose nanometer-scale magnetic tunnel junction (MTJ) free layers as a natural physical platform for the Variational Iterative Rotation Algorithm. At these length scales, the magnetization of the MTJ free layer is well approximated as a single macrospin whose dynamics is accurately described by the classical Landau-Lifshitz–Gilbert equation.

p. 6
Where the paper says it

PHYSICAL REALIZATION OF THE ITERATIVE ROTATION ALGORITHM We propose nanometer-scale magnetic tunnel junction (MTJ) free layers as a natural physical platform for the Variational Iterative Rotation Algorithm. At these length scales, the magnetization of the MTJ free layer is well approximated as a single macrospin whose dynamics is accurately described by the classical Landau-Lifshitz–Gilbert equation.

Deterministic spin rotations required by the algorithm can be implemented using spin-transfer torque, spin-orbit torque, and applied magnetic fields. p. 6
Where the paper says it

Nanoscale MTJ free layers are therefore well suited for implementing deterministic classical spin dynamics, with magnetic anisotropy engineering, macrospin behavior, and spin-torquedriven control well established experimentally. Deterministic spin rotations required by the algorithm can be implemented using spin-transfer torque, spin-orbit torque, and applied magnetic fields, which generate rotations about spin axes set by the direction of the injected electron spin polarization.

Conclusion on Quantum Advantage

The results indicate that the performance of QAOA is actually due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations. p. 6

Where the paper says it

Our results indicate that the performance of QAOA is actually due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations.

p. 6
Where the paper says it

Our results indicate that the performance of QAOA is actually due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations.

p. 6
Where the paper says it

Our results indicate that the performance of QAOA is actually due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations.

The paper concludes by questioning where quantum optimization is going to present a clear advantage. p. 7
Where the paper says it

For more complex optimization problems than the SK model, such as p-spin models, one can further consider adding controlled noise to the dynamics to improve the phase space exploration. With all of this in mind, one really wonders, where is quantum optimization going to present a clear advantage?

p. 7
Where the paper says it

For more complex optimization problems than the SK model, such as p-spin models, one can further consider adding controlled noise to the dynamics to improve the phase space exploration. With all of this in mind, one really wonders, where is quantum optimization going to present a clear advantage?

Pruning redundant reasoning tokens

The authors introduce the Variational Iterative Rotation Algorithm (VIRAL), a purely classical algorithm built on a dynamical system of three-component unit vectors (kicked tops) that systematically outperforms QAOA on the paradigmatic Sherrington-Kirkpatrick spin-glass problem. Since the entire algorithm is classical, it can be implemented on a large number of spins and optimized at much larger depths p then typically reported for QAOA. p. 0

Where the paper says it

In this work, we introduce the Variational Iterative Rotation Algorithm (VIRAL), a purely classical algorithm built on a dynamical system of three-component unit vectors (kicked tops) that systematically outperforms QAOA on the paradigmatic Sherrington-Kirkpatrick spin-glass problem. Since the entire algorithm is classical, it can be implemented on a large number of spins and optimized at much larger depths p then typically reported for QAOA.

How the training signal is built

The total energy is the sum of contributions from each layer n, where each contribution is the product of three factors: the interaction strength γn, the correlation function C(n), and the response function R(n). The correlation and response operate in opposition throughout the optimization protocol. p. 4

Where the paper says it

The total energy is the sum of contributions from each layer n, where each contribution is the product of three factors: the interaction strength γn, the correlation function C(n), and the response function R(n). The correlation and response operate in opposition throughout the optimization protocol.

In the early layers, the system exhibits a large response, when the spins are near the equator and experience large transverse kicks βn. p. 4
Where the paper says it

The correlation and response operate in opposition throughout the optimization protocol. In the early layers, the system exhibits a large response, when the spins are near the equator and experience large transverse kicks βn.

On the contrary the correlation is negligible in the early layers, because these same large kicks effectively erase the memory of the spin. p. 4
Where the paper says it

A small perturbation of the angle θn applied at this stage drastically alters the spin’s subsequent trajectory, yielding a high response (or susceptibility). On the contrary the correlation is negligible in the early layers, because these same large kicks effectively erase the memory of the spin.

Where the gains come from

The efficiency of our classical protocol is embodied in the spectral properties of the self-consistent noise covariance matrix Σ(t, t′), shown in Fig. 4D. In the classical case, the spectrum is dominated by a single eigenvalue, reflecting a nearly rank-one structure of Σ(t, t′). p. 5

Where the paper says it

Within the dynamical mean field theory, the efficiency of our classical protocol is embodied in the spectral properties of the self-consistent noise covariance matrix Σ(t, t′ ), shown in Fig. In the classical case, the spectrum is dominated by a single eigenvalue, reflecting a nearly rank-one structure of Σ(t, t′ ).

This implies that Σ(t, t′) ≈ s z(t)s z(t').

The optimization is driven by a bifurcation in the phase space of the kicked top, with a soft mode frequency that vanishes like √γ − γc. By carefully choosing the rotation angles, the spins are transported from the equator to the polarized Ising states, with the energy density converging toward the optimal Parisi value linearly in 1/p. p. 7

Where the paper says it

8 By reducing the many-body problem to an effective Landau-Lifshitz dynamics, we show that optimization is driven by a bifurcation in the phase space of the kicked top, with a soft mode frequency that vanishes like √ γ − γc. By carefully choosing the rotation angles, the spins are transported from the equator to the polarized Ising states, with the energy density converging toward the optimal Parisi value linearly in 1/p.

The calculation of the average energy E N = X p n=1 γ n ⟨⟨s z(p)s z(n-1)⟩⟩ ∂s z(p)/∂θk EE is precisely Eq. (7) in the main text. This expression is derived by integrating out the response fields r and finding that E N = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf). p. 0

Where the paper says it

We propose nanometer-scale magnetic tunnel junctions as a natural physical platform for implementing VIRAL, where spin rotations can be realized using magnetic fields and spin torques. INTRODUCTION Hard combinatorial optimization problems can often be reformulated as finding the ground state of an Ising spinglass system, a perspective that has brought together statistical physics, computer science, and electrical engineering, and motivated both classical approaches, such as Ising machines, and quantum algorithms.

The final energy expression is found to be E N = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf), where the average is over the self-consistent measure in Eq. (58). If one succeeds to find the ground state, the energy should go to the Parisi value lim t->∞ ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf) = −0. p. 1

Where the paper says it

2A confirm that our algorithm reaches an energy density of approximately E/N ≈ −0. 761 (very close to the exact Parisi ground state value of −0.

p. 1
Where the paper says it

2A confirm that our algorithm reaches an energy density of approximately E/N ≈ −0. 761 (very close to the exact Parisi ground state value of −0.

76321.

The paper proposes nanometer-scale magnetic tunnel junctions (MTJ) free layers as a natural physical platform for the Variational Iterative Rotation Algorithm. At these length scales, the magnetization of the MTJ free layer is well approximated as a single macrospin whose dynamics is accurately described by the classical Landau-Lifshitz–Gilbert equation.

The discrete version of the dynamics does not require to solve the self-consistent fixed point condition in Eq. (61). Instead, it can be computed iteratively by sampling angles and applying rotations layer by layer. p. 4

Where the paper says it

Once the dynamics is reduced to a discrete sequence of rotations driven by the stochastic angles θn, the goal of the algorithm is to tune the control parameters γn and βn to minimize the expected energy. In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p.

The calculation of the average energy E N = X p k=1 γ k ⟨⟨s z(p) s z(k-1)⟩⟩ ∂s z(p)/∂θk EE is precisely Eq. p. 1
Where the paper says it

This sequence is repeated p times. The variational parameters α, γn, and βn are optimized to minimize the average energy function of the final state ⟨H[sp]⟩J .

(7) in the main text.

The final expression for the average energy is E N = X p k=1 γ k ⟨⟨s z(p) s z(k-1)⟩⟩ ∂s z(p)/∂θk EE, which is precisely Eq. (7) in the main text. This expression is derived by integrating out r and finding that E N = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf). p. 4

Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

p. 4
Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

The paper demonstrates that the effectiveness of QAOA is due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations.

The effective single-spin LLG dynamics is given by ∂s α/∂t = ε αβγg β(t)s γ. The discrete map of the effective single-spin LLG dynamics evolves through p layers, where a layer n consists of two successive operations: first the local z-rotation driven by the continuous precession in the interval t ∈ (n − 1, n), and second the global x-rotation driven by the kick at the end of the interval t = n. p. 1

Where the paper says it

The system then evolves through p discrete layers. Each layer n comprises two successive rotations: first, a site-dependent rotation Rz(γn) around the z-axis, where each spin i precesses by an angle γnh z i determined by its local field h z i = P j̸=i Jij s z j ; second, a uniform rotation Rx(βn) around the x-axis, kicking all spins by a common angle βn.

The calculation of θ n is a Gaussian random variable (it is the integral of the Gaussian process g z(t)) with zero mean and covariance ⟨⟨θmθn⟩⟩ = γ mγ nΣmn. This covariance matrix Σ that dictates the angles θ m and θ n during layers m and n is defined by the spin states from the previous layers m − 1 and n − 1. p. 5

Where the paper says it

4C, showing that the energy is built up across all layers, with the largest contributions coming from the intermediate layers where the product of correlation and response is the largest. Within the dynamical mean field theory, the efficiency of our classical protocol is embodied in the spectral properties of the self-consistent noise covariance matrix Σ(t, t′ ), shown in Fig.

The calculation of g z(p)s z(p) is reduced to X p k=1 X p m=1 C-1 nm um⟨θns z(p)⟩. This expression is derived by integrating out r and finding that E = −i Z DrDsDg z e−S[g z] + iΦ(s,r,g) g z(tf)s z(tf) γ(tf). p. 4

Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

The final energy expression is E N = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf), where the average is over the self-consistent measure in Eq. (58). If one succeeds to find the ground state, the energy should go to the Parisi value lim t->∞ ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf) = −0.76321.

The calculation of s z(p) = f(θ1, …, θp) and then computing ⟨⟨θns z(p)⟩⟩ = Z dθ P(θ) θn f(θ) is done by using the identity ∂P(θ)/∂θk = −P(θ) X p l=1 C-1 k l → X p k=1 Cnk ∂kP(θ). This leads to the expression for the average energy E N = X p k=1 γ k ⟨⟨s z(p) s z(k-1)⟩⟩ ∂s z(p)/∂θk EE. p. 1

Where the paper says it

This sequence is repeated p times. The variational parameters α, γn, and βn are optimized to minimize the average energy function of the final state ⟨H[sp]⟩J .

The calculation of the average energy E N = X p k=1 γ k ⟨⟨s z(p) s z(k-1)⟩⟩ ∂s z(p)/∂θk EE is precisely Eq. (7) in the main text. This expression is derived by integrating out r and finding that E N = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf).

The calculation of g z(p) = X p n=1 a nθ n + b, where b is an independent Gaussian variable and a is given by an = X p m=1 C-1 nm um. This expression is derived by integrating out r and finding that E = −iηX i Z tf 0 dt γ(t) vi(t) Σ(t, tf) s z i(tf). p. 4

Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

p. 4
Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

The final energy expression is E = −i Z DrDsDg z e−S[g z] + iΦ(s,r,g) g z(tf)s z(tf) γ(tf). This expression is derived by integrating out r and finding that E = ⟨⟨g z(tf)s z(tf)⟩⟩ γ(tf). p. 4

Where the paper says it

In the thermodynamic limit, the disorder-averaged energy can be evaluated exactly at the final layer p. The resulting energy density, derived in Supplementary Sec.

The paper demonstrates that the effectiveness of QAOA is due to the iterative rotation structure of the protocol and quantum fluctuations hamper the optimization process rather than aiding it. In contrast, the classical nature of our protocol ensures that the system evolves under a highly controlled dynamics with minimal fluctuations. The paper concludes by questioning where quantum optimization is going to present a clear advantage.

The calculation of g z(p)s z(p) is reduced to X p k=1 X p m=1 C-1 nm um⟨θns z(p)⟩. This expression is derived by integrating out r and finding that E = −iηX i Z tf 0 dt γ(t) vi(t) Σ(t, tf) s z i(tf).

The calculation of the average energy E N = X p k=1 γ k ⟨⟨s z(p) s z(k-1)⟩⟩ ∂s z(p)/∂θk EE is precisely Eq. (7) in the main text.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Classical optimization of QAOA-like protocols can be used to find low-energy states of Ising spinglass systems, potentially offering a faster or more resource-efficient alternative to quantum algorithms on classical hardware.

  2. The system can be implemented using nanometer-scale magnetic tunnel junctions where spin rotations are realized via magnetic fields and spin torques, allowing for the simulation of complex combinatorial optimization problems in physical devices.

  3. The algorithm's success is rooted in a pitchfork bifurcation of the Floquet map, which destabilizes the equatorial fixed point and drives spins toward polar Ising configurations, enabling precise steering from an initial configuration to the ground state.

  4. Optimization can be performed by learning control parameters on small system instances and deploying them to solve much larger instances due to the universality of the optimized control parameters exhibited in Figure 2B.

  5. The many-body problem is reduced to an effective Landau-Lifshitz dynamics for a single spin in a stochastic magnetic field, which simplifies the computational cost by reducing it to solving a single spin problem with propagator P(sf, tf s0, 0).

Abstract

We investigate a classical formulation of the Quantum Approximate Optimization Algorithm (QAOA), realized as a Hamiltonian dynamical system of classical kicked tops, which we call the Variational Iterative Rotation Algorithm (VIRAL). The protocol is fully variational, with the initial tilt of the spins and the 2p rotation angles of a depth- p circuit optimized by backpropagation through the deterministic dynamics. On the canonical Sherrington-Kirkpatrick (SK) spin glass and on Max-Cut for random 3-regular graphs, VIRAL systematically outperforms QAOA, with the SK energy density converging to the Parisi ground state value and the cut fraction to the 2RSB upper bound, both linearly in 1/p. The optimization is driven by a pitchfork bifurcation of the Floquet map, which destabilizes the equatorial fixed point and adiabatically transports the spins to the polar Ising configurations encoding the solution. These results indicate that the effectiveness of QAOA-like protocols is rooted in their iterative rotation structure. To explain the classical advantage, we exactly reduce the many-body evolution to an effective Landau-Lifshitz dynamics of a single spin in a self-consistent stochastic magnetic field. The temporal covariance of the effective field is nearly rank-one, so that a single noise mode governs the stochastic dynamics and the system remains highly controllable. Quantum fluctuations, by contrast, increase the rank of the noise covariance, hampering control and slowing convergence to the solution. Finally, we propose nanometer-scale magnetic tunnel junctions, where spin rotations are naturally realized by magnetic fields and spin torques, as a physical platform for implementing VIRAL.

Sources

Related papers