Real-time Dynamics in 3D for up to 1000 Qubits with Neural Quantum States: Quenches and the Quantum Kibble--Zurek Mechanism
Vighnesh Dattatraya Naik, Zheng-Hang Sun, Markus Heyl
Abstract
Exponential complexity of many-body wave functions limits accurate numerical simulations of real-time dynamics, especially beyond 1D, where rapid entanglement growth poses severe challenges. Neural Quantum States (NQS) have emerged as a powerful approach for real-time dynamics in 2D, but their scalability and accuracy in 3D have remained an open challenge. Here, we establish NQS as a scalable framework for 3D quantum dynamics by introducing a residual-based convolutional architecture tailored to cubic spin lattices. Focusing on the 3D transverse-field Ising model, we demonstrate that NQS reliably capture distinct quench regimes, including collapse-and-revival dynamics and, most challengingly, the dynamics following a sudden quench to the quantum critical point. We perform finite-rate quenches to the critical point on lattices containing up to qubits, an unprecedented system size for numerical simulations of real-time dynamics beyond 1D. This enables the first large-scale numerical demonstration of the 3D quantum Kibble--Zurek mechanism. The QKZM in 3D is particularly intriguing because it lies at the upper critical dimension of the Ising universality class, where the standard power laws are modified by logarithmic factors together with prominent sub-leading logarithmic corrections. By deriving these corrections from renormalization-group flow equations up to two-loop order, we obtain a robust data collapse across all simulated system sizes for the correlation function, the excess energy, and the quantum Fisher information, the latter revealing universal multipartite-entanglement dynamics. In all cases, we find compelling agreement with the expected scaling dimensions. Our findings establish NQS as a scalable and reliable tool for exploring nonequilibrium phenomena in 3D quantum matter and for providing numerical benchmarks for 3D quantum simulators.
AI Impact Assessments
(3 models)Scientific Impact Assessment
1. Core Contribution
This paper makes two intertwined contributions: a methodological advance (extending NQS to 3D real-time quantum dynamics at scale) and a physics result (the first numerical demonstration of the quantum Kibble-Zurek mechanism in 3D at the upper critical dimension).
The authors introduce a 3D residual convolutional neural network (ResNet-CNN) architecture tailored to cubic lattices, extending prior 2D NQS work. The key architectural choices—3D convolutional kernels with circular padding, residual connections with depth-dependent normalization, and a "Pair Complex" layer—are pragmatic rather than conceptually novel, but together they enable simulations on lattices up to 10×10×10 (1000 qubits), which is genuinely unprecedented for real-time dynamics beyond 1D.
The physics contribution is more substantial: demonstrating the quantum Kibble-Zurek mechanism (QKZM) in 3D, where the system lies at the upper critical dimension (d+z=4). Here, standard power-law KZ scaling is modified by logarithmic corrections with prominent sub-leading terms. The authors derive these corrections from two-loop RG flow equations and demonstrate data collapse across multiple observables (correlation functions, excess energy, quantum Fisher information) using consistent nonuniversal fitting parameters.
2. Methodological Rigor
The numerical methodology is carefully executed with several commendable features:
However, there are notable limitations. The paper lacks comparison with any independent method (exact diagonalization on small systems, or other numerical approaches) to validate accuracy. The convergence is shown only between n=3 and n=4 network depths, which demonstrates internal consistency but not absolute accuracy. The timescales accessed remain relatively short (particularly for the critical quench), and the paper does not provide rigorous error bars from the TDVP integration or Monte Carlo sampling. The fitting procedure involves four free parameters (A, C, μ, K) for the correlation collapse, which somewhat weakens the predictive power of the scaling analysis, though the consistency across observables partially mitigates this concern.
3. Potential Impact
Computational physics: This work significantly expands the frontier of what is computationally accessible in 3D quantum dynamics. The demonstration that convolutional NQS can handle 1000-qubit 3D systems positions this approach as a leading method for 3D nonequilibrium quantum matter, complementing tensor networks (limited in 3D) and sparse Pauli dynamics (which the authors note faces convergence issues in 3D).
Quantum simulation benchmarking: The results provide concrete numerical predictions that can serve as benchmarks for 3D quantum simulators (Rydberg atoms, optical lattices). This is practically valuable given the rapid experimental progress in 3D quantum simulation platforms.
Critical phenomena and universality: The verification of logarithmically corrected KZ scaling at the upper critical dimension is a meaningful contribution to the theory of nonequilibrium phase transitions. The inclusion of sub-leading corrections from two-loop RG and their numerical verification is more refined than typical KZ analyses.
Entanglement dynamics: The universal scaling of QFI through the KZ transition provides insight into multipartite entanglement dynamics in 3D, connecting entanglement witnesses to critical scaling—a topic of growing interest.
4. Timeliness & Relevance
The paper addresses a genuine bottleneck: reliable simulation of 3D quantum dynamics has been one of the most challenging open problems in computational quantum physics. Recent work on sparse Pauli dynamics (Begušić & Chan, 2025) highlighted both progress and limitations in 3D, making this a timely contribution. The growing experimental capabilities in 3D quantum simulation (3D Rydberg arrays, 3D optical lattices) create demand for theoretical predictions and benchmarks that this work begins to supply.
The QKZM at the upper critical dimension is also timely, as logarithmic corrections at d_c have been a topic of renewed interest in classical statistical mechanics, and extending this to the quantum nonequilibrium setting fills a conceptual gap.
5. Strengths & Limitations
Strengths:
Limitations:
Overall Assessment: This is a strong paper that makes a meaningful contribution at the intersection of machine learning methods for quantum physics and nonequilibrium critical phenomena. The combination of methodological scaling to 3D and the first numerical verification of logarithmically corrected QKZM at the upper critical dimension represents genuine progress. The main concern is the absence of independent validation, which means the accuracy claims rest entirely on internal convergence checks.
Generated Apr 8, 2026
Comparison History (43)
Paper 1 likely has higher impact: it introduces a fundamentally new fault-tolerant “quantum uploading” paradigm with provable exponential separations under experimentally motivated noise constraints, plus a new lower-bound technique (Heisenberg learning tree). This combines strong novelty, methodological rigor (theorems/lower bounds), and broad relevance across quantum computing, metrology, and experimental quantum science, with clear real-world applicability (e.g., imaging) and timeliness as fault-tolerant QC emerges. Paper 2 is strong and timely, but its impact is more domain-specific and relies on empirical validation of NQS accuracy.
Paper 1 presents a profound breakthrough by reducing the estimated physical qubits needed to break RSA-2048 by an order of magnitude (to under 100,000). This drastically accelerates the timeline for utility-scale quantum computing and poses an immediate, tangible threat to current global cybersecurity protocols. While Paper 2 offers significant advancements in 3D quantum simulations and many-body physics, Paper 1 has vastly broader implications, fundamentally altering post-quantum cryptography timelines and real-world technological roadmaps.
Paper 1 dramatically reduces the physical qubit requirement for breaking RSA-2048 to 100,000, bringing cryptographically relevant quantum computers much closer to reality. This has massive, urgent implications for global cybersecurity, post-quantum cryptography, and quantum hardware design, offering significantly broader real-world impact than Paper 2's advance in computational physics.
Paper 2 has higher likely impact: it demonstrates scalable real-time 3D quantum dynamics with Neural Quantum States up to 1000 qubits and provides the first large-scale numerical evidence of the 3D quantum Kibble–Zurek mechanism, combining algorithmic innovation with new physics results and RG-based validation. The breadth spans ML-for-physics, nonequilibrium many-body theory, critical phenomena, and benchmarking quantum simulators, making it timely and broadly relevant. Paper 1 is methodologically strong and useful for quantum error/noise simulation, but its novelty and cross-field reach appear narrower and more incremental.
Paper 2 achieves an unprecedented scale (simulating 1000 qubits in 3D real-time dynamics) by leveraging Neural Quantum States, solving a major bottleneck in computational many-body physics. Its demonstration of the 3D quantum Kibble-Zurek mechanism and ability to benchmark near-term quantum simulators give it broader immediate relevance and higher potential impact across the highly active intersections of machine learning, quantum computing, and condensed matter physics compared to the theoretical, albeit rigorous, algorithmic guarantees provided in Paper 1.
Paper 1 presents a major breakthrough in quantum many-body simulation, achieving real-time dynamics for up to 1000 qubits in 3D using Neural Quantum States. This unprecedented scale allows for the first numerical demonstration of the 3D quantum Kibble-Zurek mechanism, offering profound implications for quantum simulation benchmarks and AI-driven physics. While Paper 2 provides valuable theoretical guarantees for tensor networks, Paper 1 pushes practical computational boundaries and unlocks new physical insights in previously intractable 3D regimes.
Paper 1 demonstrates a breakthrough in simulating real-time quantum dynamics in 3D for up to 1000 qubits using neural quantum states, providing the first large-scale numerical demonstration of the 3D quantum Kibble-Zurek mechanism at the upper critical dimension. This has broad impact across quantum simulation, condensed matter physics, and machine learning for physics, with direct relevance to validating quantum simulators. While Paper 2 makes important theoretical contributions to quantum channel tomography with a novel phase transition result, Paper 1's combination of methodological innovation, unprecedented scale, physical insight into universal dynamics with logarithmic corrections, and practical relevance to experimental quantum simulation gives it broader and higher potential impact.
Paper 2 demonstrates significantly higher scientific impact through multiple factors: (1) It achieves an unprecedented scale of 1000 qubits for real-time 3D quantum dynamics simulations, a major breakthrough; (2) It provides the first large-scale numerical demonstration of the 3D quantum Kibble-Zurek mechanism at the upper critical dimension, including novel logarithmic corrections derived from two-loop RG equations; (3) It establishes NQS as a scalable framework for 3D quantum dynamics, opening broad applications across condensed matter and quantum simulation; (4) The methodological innovation (residual convolutional architecture) and theoretical depth (universal scaling with sub-leading corrections) are substantial. Paper 1, while useful for quantum fluid simulation, addresses a more incremental engineering optimization with modest system sizes.
Paper 2 has higher impact potential due to a concrete methodological advance (scalable 3D neural-quantum-state architecture) demonstrated on unprecedented system sizes (up to 1000 qubits) and used to obtain the first large-scale numerical verification of the 3D quantum Kibble–Zurek mechanism with two-loop RG corrections and multiple observables. This is timely for quantum simulation/benchmarking, broadly relevant across condensed matter, quantum information, and ML-for-physics, and shows strong rigor via scaling collapses and theory–numerics agreement. Paper 1 is promising but more design/forward-looking and less validated in the abstract.
Paper 2 likely has higher impact due to broad, timely relevance and cross-field applicability: it advances scalable simulation of real-time 3D many-body dynamics with neural quantum states up to 1000 qubits, enabling the first large-scale numerical demonstration of the 3D quantum Kibble–Zurek mechanism with RG-derived logarithmic corrections and multiple observables (including quantum Fisher information). This provides a widely usable computational tool and benchmarks for quantum simulators. Paper 1 is novel for programmable multiphoton sources but is more specialized and appears primarily theoretical/engineering-focused within cavity QED.