Probing Antiferromagnetic Hysteresis on Programmable Quantum Annealers

arXiv:2511.17779 · quant-ph, cond-mat.other, cond-mat.stat-mech, cond-mat.str-el, physics.comp-ph · Submitted 2025-11-21 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: I'm Kai, and with me are Mira and Lev, guest researcher.

Mira: Today's paper: "Probing Antiferromagnetic Hysteresis on Programmable Quantum Annealers".

Kai: Using programmable analog quantum annealing processors, this work implements a sampling-based magnetic hysteresis protocol to probe the counterintuitive notion of magnetic memory in antiferromagnetic models.

Mira: First, who's behind it and why it matters.

Title and authors: Kai: So we're diving into "Probing Antiferromagnetic Hysteresis on Programmable Quantum Annealers" today. This paper looks at using programmable quantum annealers to investigate magnetic memory effects in antiferromagnetic models, which is pretty counterintuitive.

Mira: Yeah, it’s interesting because magnetic memory in antiferromagnets isn't something you usually see easily, and the paper suggests a sampling-based protocol using a transverse field to enable state transitions while sweeping the longitudinal field.

Lev: From an error correction standpoint, I'm curious how robust these measurements are when you consider running them on actual hardware, since that’s where we have to deal with real noise.

Kai: Exactly, Lev, that’s what we need to figure out—what was actually built and measured in this study. The paper explains that they use D-Wave systems and operate them at about fifteen millikelvin temperatures using superconducting flux qubits connected in a graph structure.

Mira: And the Hamiltonian they implement is a time-dependent transverse field Ising model, specifically H = -A(s)X i sigma x(i) + B(s) squared

g(t)X i h i sigma z(i) + X i not equal to j J ij sigma z(i) sigma z(j): , where s is the programmable annealing parameter.

Lev: That time-dependent term, g(t), that's what makes it complicated for error correction because you have to account for a time-varying driving field in your syndrome extraction cycles.

Kai: Right, and they emphasize that this function g(t) is controlled independently of the annealing schedule parameter s, which is what allows them to sweep the longitudinal magnetic field dynamically.

Mira: The core mechanism hinges on using this transverse field to facilitate state transitions while the longitudinal control field, governed by g(t), handles the actual magnetic field sweep. If you don't have that transverse field in tandem with a strong lattice coupling J, they say no hysteresis can be measured.

Lev: So, it sounds like the requirement for measuring hysteresis is quite stringent regarding both the hardware setup and the model parameters.

Kai: That’s right, and they set each h i to the maximum allowed normalized hardware programmable value in their experiments. They also considered three types of models: one-dimensional with periodic boundary conditions, two-dimensional with open boundary conditions, and a pseudo-three-dimensional model using the full hardware graph.

Mira: The modeling aspect is crucial because they embed these models onto the hardware graph using a Glasgow solver for 1D and 2D cases. They even set all coupler interaction strengths J ij to one, which is the maximum strength specified on the hardware.

Lev: Setting J ij to one means they are pushing those coupling limits, which makes sense if you're trying to see if the intrinsic magnetic memory can survive strong interactions.

Title and authors: Kai: They noted that for 1D embeddings, they tested system sizes ranging from four thousand nine hundred five qubits up to one thousand one hundred thirty-one qubits, and for the 2D model, they achieved lattice sizes up to thirty-two times thirty-two on systems like Advantage system4 point 1 and Advantage system6 point 4.

Mira: They also addressed the non-idealities of quantum annealers by using an iterative statistical-balancing calibration refinement technique, specifically mentioning flux bias offset calibration for quasi-three dee hardware experiments.

Lev: That calibration sounds necessary; if you don't account for those systematic offsets, any dynamical measurement you take will be completely skewed by the hardware specifics.

Kai: For the quasi-three dee models, they performed trial experiments with a zero longitudinal field and iteratively modified the flux bias offsets to reduce the spread in qubit magnetizations.

Mira: They extracted two key observables: M z = one/N sum X i sigma z(i) for average longitudinal magnetization, and M s = one/N sum (-one) i sigma z(i) for the antiferromagnetic order parameter, which they noted is only well-defined for bipartite graphs.

Lev: Extracting that staggered order parameter M s from raw expectation values is a big step because it lets us probe that magnetic structure without needing direct neutron scattering data.

Kai: They also reconstructed the magnetic structure factor S(q) from individual spins using the formula S(q) = sum i,j e iq times(r i-r j) sigma z i sigma z j, which is how they see the spatial organization of the magnetism.

Mira: The results showed that in 1D antiferromagnetic experiments, a "pinchpoint" appears at s = zero point six on three of the QPUs, and when /J is sufficiently small, full polarization is achieved.

Lev: That pinchpoint observation suggests there's a specific dynamical regime where the system's response to the annealing schedule hits a critical point before saturation occurs.

Kai: They also found that for odd rings with an odd number of spins N, the domain wall number D is constrained to be odd, and applying a non-zero longitudinal field h not equal to zero breaks that global Z two symmetry by splitting two local configurations in energy.

Mira: That splitting is very telling because it directly relates to the emergence of magnetic domains mediated by quantum fluctuations, which gives rise to the magnetic memory effect in antiferromagnets, as mentioned at the start of this paper.

Lev: If that splitting happens due to quantum fluctuations, it means we're looking at a highly sensitive region where classical approximations definitely won't work without including those effects.

Kai: Moving into 2D experiments with open boundary conditions, they observed that reversal appears to happen through a competition between the nucleation of favorably oriented antiferromagnetic droplets and the propagation and roughening of pre-existing interfaces.

Mira: The magnetic structure factors showed clear transitions, specifically that near saturation they concentrate around q=zero reflecting large polarized regions, but toward the demagnetized regime, intensity migrates to antiferromagnetic wavevectors at the corners of the Brillouin zone.

Title and authors: Lev: That migration of intensity in the structure factor across different momentum vectors tells us a lot about how long-range correlations decay or evolve as you move away from saturation.

Kai: In their three dee hardware-defined antiferromagnets with high coordination number, they found that it gets harder to fully magnetize systems, and hysteresis effects are substantially suppressed because of a large interfacial cost scaling as lambda proportional to cJ.

Mira: That scaling with cJ suggests that in denser structures, the energy penalty for creating domain walls or interfaces becomes too high for the dynamic sweep to easily induce reversal.

Lev: So, while they can measure it, they also found that certain physical constraints imposed by hardware architecture actively suppress the very memory effects they are trying to probe.

Kai: Finally, they noted that the loop morphology—how the hysteresis behaves—is set by several tunable parameters: geometry and coordination, the sweep protocol itself, and the ratio /J. The non-monotonicity near h about one at s about zero point six in odd rings is understood as a result of synchronized depinning at reproducible nucleation fields set by hardware pinning.

Mira: That final point ties together the dynamical aspect with the hardware specifics, suggesting that the non-monotonic behavior isn't just a feature of the physics but is directly linked to how that specific D-Wave topology pins those configurations.

Lev: It seems like this work provides a very detailed map linking fundamental magnetic dynamics in antiferromagnets to specific constraints imposed by current quantum annealer architectures.

Kai: Exactly, and overall, the main implication is that the full programmability of these devices gives us a new way to examine antiferromagnetic hysteresis that we really struggle to study systematically in traditional magnetic laboratories.

Mira: The robustness they found across different quantum processors despite intrinsic noise differences and timescale variations is what makes this paper significant for condensed matter theory right now.

Lev: If we take this result seriously, it means that the complex, non-equilibrium dynamics of frustrated systems can be mapped onto these programmable architectures in a controlled way.

Kai: So, to wrap up on "Probing Antiferromagnetic Hysteresis on Programmable Quantum Annealers," this work establishes robust dynamical hysteresis in programmable antiferromagnets across various geometries.

Mira: It confirms that the full programmability of these devices provides a capability to examine antiferromagnetic hysteresis, which is hard to systematically study in magnetic laboratories.

Lev: I just want to say that if we can build simulators based on these observations, it could dramatically speed up our understanding of how quantum fluctuations mediate memory in these systems.

Kai: Agreed, and we'll see what the next steps are for testing these findings on larger or different hardware setups.

The paper's summary: Kai: So, to recap, this paper is essentially showing how you can use these programmable quantum annealers to actually measure magnetic hysteresis in antiferromagnets, which is usually a tricky thing to observe in standard labs.

Mira: Exactly; they're using a sampling protocol involving a transverse field and sweeping the longitudinal field to probe that memory effect. The core idea is using the annealer's flexibility as an experimental tool for condensed matter physics problems that are notoriously difficult classically.

Lev: From my angle, what really strikes me is the practical demonstration of extracting meaningful order parameters like M s from noisy hardware outputs, which is a huge hurdle for anyone trying to run these kinds of experiments on actual machines.

Kai: Right, Lev, and it's not just about measuring something; it's about seeing how the specific physical constraints of the hardware—like those fifteen millikelvin temperatures and the graph structure—actually influence whether we see that hysteresis loop or not.

Mira: I think the most important part is their detailed analysis of domain wall dynamics in different geometries, showing how things like droplet nucleation compete with interface propagation in 2D systems. That's a deep dive into non-equilibrium physics right there.

Lev: And that dynamic competition is precisely what makes this work so interesting for error correction research; understanding the noise landscape under these specific driving conditions gives us better tools for building fault-tolerant systems.

Kai: Speaking of those hardware specifics, they found that the loop shape itself depends on geometry, coordination number, and the sweep rate in a non-trivial way. That suggests we have a set of tunable knobs to explore magnetic memory effects systematically.

Mira: That leads me to think about the broader implications for simulating frustrated magnetism; if we can reliably map these dynamical behaviors onto hardware constraints, it opens up new avenues for computational simulations that handle strong interactions better than current methods.

Lev: If the AI system I mentioned earlier could actually ingest all this data—the structure factor evolution and the magnetization curves—it could serve as a powerful simulator for predicting how different hardware topologies would affect magnetic memory before we even run the experiment.

Kai: That's a big leap, connecting the experimental observation to predictive modeling. It moves us from just reporting what happened to actually designing experiments that test specific physical hypotheses about quantum materials.

Mira: And for me, it suggests that antiferromagnetic order, which is often studied using classical mean-field theories because of its frustration, might have subtle dynamical features that only become apparent when you introduce the complexity of quantum annealing dynamics.

Lev: I agree; the paper's success in achieving these results despite intrinsic noise differences across various QPUs really validates the approach for future work on real-world implementations.

Kai: It sounds like this isn't just an academic curiosity anymore; it’s a new experimental paradigm for probing magnetic memory that leverages the unique capabilities of programmable quantum hardware.

The paper's improvements: Tom: So, to wrap up on the paper's improvements, they aren't just stopping at measurement; they’re proposing ways to make these measurements much smarter and more robust by incorporating hardware realities into the analysis.

Kai: Right, that’s what I mean; they are suggesting iterative calibration techniques like flux bias offset balancing not just as a fix for noise, but as a way to systematically improve the fidelity of those quasi-three dee experiments.

Mira: I think that's a smart move because it acknowledges that the hardware itself has imperfections, and by learning to compensate for them through statistical balancing, they can get cleaner data on the underlying physics.

Lev: From my point of view, this iterative refinement is crucial for anyone planning to run these protocols on actual D-Wave hardware; it shows a path toward extracting reliable physical signals even when you're dealing with non-ideal systems.

Kai: And they also suggest performing trial experiments with zero longitudinal fields to systematically reduce the spread in qubit magnetizations before running the main protocol, which really cleans up their input data.

Mira: That makes sense because if the initial state preparation is too varied, you end up diluting your ability to observe subtle physical transitions like those domain wall behaviors they’re studying.

Lev: It's a practical suggestion that moves beyond just "it works on this chip" and into "here's how you make it work reliably across different chips."

Kai: They are also outlining how the loop morphology, that shape of the hysteresis, is determined by geometry and the sweep protocol ratio in a way we can actually tune. That gives us more control over what physical phenomena we see.

Mira: And linking that tuning back to the hardware pinning mechanism for odd rings is important because it suggests that some of those non-monotonic features might be intrinsically tied to how the quantum annealer's physical layout dictates those specific energy barriers.

Lev: If we can tune the protocol to exploit these pinning effects, it could lead to better control over domain wall dynamics than just running a fixed sequence.

Kai: Ultimately, these improvements suggest that future work should focus on developing software tools that can learn these calibration curves and optimization parameters automatically across different hardware architectures.

Mira: That would be incredibly valuable for condensed matter theory because it would allow us to test more complex theoretical models against the actual observed dynamical constraints of real quantum annealers.

Lev: I'm eager to see how the error correction community can use these calibration insights to develop better diagnostic tools for noisy quantum hardware in general.

Kai: It seems like the direction is toward making this method less about a single experiment and more about building a robust framework for studying magnetic systems on these platforms.

Conclusion: Kai: So, to wrap up, this paper on "Probing Antiferromagnetic Hysteresis on Programmable Quantum Annealers" shows that we have a new way to examine magnetic memory effects in antiferromagnets using programmable hardware.

Mira: It confirms that the full programmability of these devices offers a capability to study antiferromagnetic hysteresis in ways that are quite challenging to do systematically in traditional magnetic laboratories.

Lev: I just want to say that this work provides a really strong foundation for how we should think about running these protocols on real-world hardware, especially concerning noise mitigation.

Kai: Exactly; the robust results they found across different quantum processors despite intrinsic noise variations is what makes this paper so significant for experimental physics right now.

Mira: It confirms that the dynamical characteristics identified are stable even when you account for timescale variations in a real quantum annealer setting.

Lev: If we can build simulators based on these observations, it could dramatically speed up our understanding of how quantum fluctuations mediate memory in these systems.

Kai: And as we look ahead, I think the next big step is seeing how this methodology applies to even larger or different hardware setups to test its general applicability.

Mira: I'm particularly interested in how this approach might inform our simulations of frustrated magnetism, giving us a new lens through which to view these complex systems.

Lev: For me, I think the real impact is moving toward better error correction strategies because we now have a clearer picture of the physical constraints that limit or enable these dynamical measurements.

Kai: It’s exciting to see how hardware and theory are merging this closely for problems like antiferromagnetic dynamics in programmable annealers.

Los Alamos National Laboratory

quant-ph, cond-mat.other, cond-mat.stat-mech, cond-mat.str-el, physics.comp-ph

Submitted: 2025-11-21

Updated: 2026-09-30

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 78/100

The gist: Using programmable analog quantum annealing processors, this work implements a sampling-based magnetic hysteresis protocol to probe the counterintuitive notion of magnetic memory in antiferromagnetic

Key concepts

Programmable Analog Quantum Annealers: These are specialized quantum processors, like D-Wave's superconducting flux qubits, that allow researchers to program the system's behavior dynamically. They operate at extremely low temperatures (15 mK) and are used here to simulate complex magnetic systems.
Time-Dependent Transverse Field Ising Model: This is the mathematical model used to describe the quantum system. It involves a transverse field that drives state transitions and a longitudinal magnetic field sweep controlled by a time-dependent parameter 's'. This model is essential for simulating how magnetism changes over time.
Magnetic Hysteresis: Hysteresis refers to the memory effect in magnetism, where the system's state depends not just on the current external field but also on its previous history. The protocol specifically probes this counterintuitive memory effect within antiferromagnetic materials.
Antiferromagnetic Order Parameter (Ms): This is a measure used to quantify how ordered an antiferromagnet is. It involves calculating the average magnetization component along the negative direction, which is only well-defined for systems with bipartite graphs. This helps characterize the magnetic structure.

Terminology

Hysteresis — 155× in this paper · explained in 3 episodes
Calibration — 41× in this paper · explained in 7 episodes
Embeddings — 11× in this paper · explained in 2 episodes
Domain Walls — 7× in this paper · explained in 2 episodes
Antiferromagnetic Ordering — 6× in this paper · explained in 1 episode
Frustration — 4× in this paper · explained in 2 episodes
Physical Constraints — 3× in this paper · explained in 7 episodes
Core — 2× in this paper · explained in 2 episodes
Effective Hamiltonian — 2× in this paper · explained in 2 episodes
Fidelity — 2× in this paper · explained in 2 episodes
Phase Transitions — 2× in this paper · explained in 2 episodes
Static Structure Factor — 2× in this paper
Symmetry Breaking — 2× in this paper · explained in 3 episodes
Accuracy — 1× in this paper · explained in 1 episode
Adiabatic Evolution — 1× in this paper · explained in 2 episodes
Benchmarking — 1× in this paper · explained in 3 episodes
Diagnostic Tools — 1× in this paper · explained in 2 episodes
Entanglement — 1× in this paper · explained in 3 episodes
Iterative Refinement — 1× in this paper · explained in 3 episodes
Metastability — 1× in this paper · explained in 2 episodes
Predictive Modeling — 1× in this paper · explained in 2 episodes
Robustness — 1× in this paper · explained in 5 episodes
Scaling Laws — 1× in this paper · explained in 4 episodes
Trace — 1× in this paper · explained in 2 episodes
Uncertainty Quantification — 1× in this paper · explained in 8 episodes

Summary

Using programmable analog quantum annealing processors, this work implements a sampling-based magnetic hysteresis protocol to probe the counterintuitive notion of magnetic memory in antiferromagnetic models.

How it works

The protocol utilizes programmable analog quantum annealing processors, specifically D-Wave Systems' superconducting flux qubits, which are connected in a graph comprising repeating subgraph units and operate at approximately 15 millikelvin (15 mK).
The Hamiltonian implemented is a time-dependent transverse field Ising model:

H = −A(s)Xiσˆ(i)x + B(s)2 [g(t)Xihiσˆz(i) + Xi̸=jJijσˆz(i)σˆz(j)], where s ∈ [0, 1] is a programmable time-dependent annealing parameter.

The key component responsible for the hysteresis is a transverse field that enables state transitions, while the longitudinal magnetic field sweep is done via a longitudinal control field controlled by g(t).
The protocol requires both the transverse field on the D-Wave QPUs and a sufficiently strong lattice coupling J, without which no hysteresis can be measured.

Models and Embeddings

Three varieties of models are considered: one-dimensional with periodic boundary conditions, two-dimensional with open boundary conditions, and a specific type of “pseudo-3-dimensional” model involving the full hardware graph.
For all models, the coupler J values are set to 1, which is the maximum coupler strength specified on the hardware.
The one and two-dimensional models are embedded onto the hardware graph using a Glasgow solver, a graph isomorphism finder from minorminer package.
Maximum system sizes for 1D embeddings range from 4905 qubits to 1131 qubits, while for the 2D model, maximum sizes realized are up to a 32 × 32 lattice on Advantage system4.1 and Advantage system6.4.

Calibration and Observables

Non-idealities in quantum annealers can be mitigated by leveraging the symmetries of the model via an iterative statistical-balancing calibration refinement technique, such as flux bias offset (FBO) calibration for quasi-3D hardware-defined experiments.
For quasi-3D models, trial experiments with zero longitudinal field are performed to reduce the spread in qubit magnetizations by iteratively modifying FBOs.
The average longitudinal magnetization is extracted as Mz = 1/N Σ Xi ⟨σˆz(i)⟩, and the antiferromagnetic order parameter is defined as Ms = 1/N Σ (-1)i ⟨σˆz(i)⟩, which is well-defined only for bipartite graphs.
The magnetic structure factor S(q) can be reconstructed from individual spins using S(q) = Σ Xi,j e(iq·(ri−rⱼ)σˆi zσˆj z).

Results and Phenomena

In 1D antiferromagnetic experiments, a “pinchpoint” occurs at s = 0.6 on three of the QPUs, and when Γ/J is sufficiently small, full polarization is achieved.
For odd rings (odd N), the domain wall number D is constrained to be odd, and a longitudinal field h ≠ 0 breaks the global Z2 symmetry, splitting two local configurations in energy.
In 2D antiferromagnetic experiments on finite samples with open boundary conditions, reversal appears to proceed through a competition between nucleation of favorably oriented antiferromagnetic droplets and propagation and roughening of pre-existing interfaces.
The magnetic structure factors show clear transitions: near saturation, they concentrate around q=0 (reflecting large polarized regions), while toward the demagnetized regime, intensity migrates to antiferromagnetic wavevectors at the corners of the Brillouin zone.
In 3D hardware-defined antiferromagnets with high coordination number, it becomes harder to fully magnetize systems, and hysteresis effects are substantially suppressed due to a large interfacial cost scaling as λ ∝ cJ.
The loop morphology is set by tunable parameters: geometry and coordination, the sweep protocol, and the ratio Γ/J. The non-monotonicity near h ≈ 1 at s ≈ 0.6 in odd rings is understood as a result of synchronized depinning at reproducible nucleation fields set by hardware pinning.

Conclusion

The study demonstrates robust dynamical hysteresis in programmable antiferromagnets across various geometries, establishing that the full programmability of these devices provides a capability to examine antiferromagnetic hysteresis, which is hard to systematically study in magnetic laboratories. The hysteretic characteristics identified are robust across different quantum processors despite intrinsic noise differences and timescale variations.

Improvements for AI systems

As a fastidious researcher, I have analyzed the core contributions of this paper: probing antiferromagnetic hysteresis using programmable quantum annealers via a sampling-based protocol involving transverse fields and longitudinal field sweeps. The key findings relate to extracting magnetic order parameters, domain wall dynamics (especially in 1D rings), and the role of hardware structure in suppressing or enabling these phenomena.

Here are the specific improvements to AI systems that can be derived from this research:

)

Improvement Focus: Enhanced understanding and simulation of complex, non-equilibrium quantum dynamics, particularly in frustrated magnetic systems where classical methods fail.
Improved AI System Capability: Development of a Quantum Magnet Dynamics Simulator (QMD Simulator). This system would be trained on the observed hysteresis loops and structure factor evolution data from various D-Wave QPUs (as detailed in Figures 5–27).
Specific Functions:

) 1. Simulate Non-Monotonic Magnetization: The AI can predict the non-monotonic magnetization features (like the pinchpoint at specific ratios of transverse field/coupling, e.g., s=0.6) as a function of the longitudinal field sweep rate and magnitude, which are characteristic of quantum annealing dynamics in 1D antiferromagnets.

) 2. Predict Domain Wall Statistics: The system can predict the domain wall density (Fig. 4 and Fig. 23). It can distinguish between single-wall regimes (odd rings) where domain walls are driven by the transverse field, and many-wall regimes (high coordination/large graphs) where they are suppressed due to high surface tension scaling as λ∝cJ.

) 3. Infer Latent Order Parameters: The AI can take raw simulation outputs (magnetization curves and structure factor snapshots, Fig. 6–19) and use them to infer the staggered antiferromagnetic order parameter (Ms, Eq. 3), even in systems where direct measurement is inaccessible, providing a computational bridge for neutron scattering-like observables.

) 4. Optimize Annealing Protocols: The system can learn the optimal annealing schedule parameters (s = Γ/J) required to maximize the hysteresis loop area, effectively optimizing the protocol for detecting magnetic memory effects in specific hardware architectures.

) 5. Hardware-Aware Error Correction: By incorporating calibration refinement techniques (like Flux Bias Offset balancing, Fig. 12) into its training loop, the AI can learn to compensate for systematic hardware noise and non-idealities unique to specific QPUs (e.g., identifying the noisier Advantage system7.1 vs. Advantage system4.1), leading to more robust and accurate results across different analog devices.