Observing the Quantum Mpemba Effect on a Digital Quantum Computer

A fresh quantum twist on a counterintuitive effect

The Mpemba effect is famous because it contradicts a simple expectation: a system that begins “closer” to its final state should not necessarily arrive first. In its classical form, the effect is often described through hot and cold water; in its quantum versions, the comparison is no longer simply about temperature, but about how a quantum state relaxes, restores a symmetry, loses a resource, or gains one. The new work titled “Observing the magic Mpemba effect in localized dynamics on a digital quantum computer” was submitted to arXiv on October 6, 2026, by Han-Ze Li, Xianquan Yan, Yi-Rui Zhang, Jian-Xin Zhong, Shuo Liu, and Ching Hua Lee [1]. It appeared in the arXiv quantum physics new-submission listing for Wednesday, October 7, 2026, placing it within the current 72-hour window for this story [2].

The study narrows the broad idea of quantum Mpemba behavior to a very specific resource: quantum magic, also called nonstabilizerness. In quantum computing, stabilizer states and Clifford operations are powerful but classically simulable; magic is the resource that pushes computations beyond that efficiently simulable sector and is therefore central to universal quantum computation [1]. The paper asks whether a state with less initial magic can overtake one with more initial magic during resource-generating dynamics, reaching a larger asymptotic magic value sooner [1].

That question matters because it reframes the Mpemba effect from the “cooling-side” picture, where a more resourceful state loses its resource faster, to a “heating-side” picture, where a less resourceful state gains the relevant resource more effectively. The authors argue that this reversal can occur in localized many-body dynamics through interaction-induced dephasing, without relying on transport or conventional thermalization [1].

What the researchers claim to have observed

The core reported result is a magic-ordering reversal. For tilted product states in an interacting l-bit model, the authors’ random-phase analysis predicts that states with less initial magic can attain higher saturated magic, and that this endpoint reversal is independent of whether the initial pattern is ferromagnetic, Néel-like, or domain-wall-like [3]. The paper then reports exact simulations showing that, in the studied cases, those initially less magical states also saturate earlier, satisfying the stronger dynamical criterion for a magic Mpemba effect [3].

The team says it observed both the ordering reversal and the earlier saturation signature on IBM superconducting quantum processors [1]. In one hardware implementation, they encoded the effective l-bit evolution directly, putting the simulated time into gate angles so that long dephasing times could be accessed at fixed circuit depth [1]. This matters because long-time many-body dynamics are usually difficult for digital quantum hardware: repeated time steps increase circuit depth, and circuit depth increases exposure to noise.

For six-qubit dynamics, the authors used two identically evolved registers and pairwise Bell measurements to estimate a fourth-order Pauli moment related to the second stabilizer Rényi entropy, the paper’s main magic diagnostic [3]. They explicitly note that this measurement becomes more demanding as system size or magic increases, and they combine readout correction with zero-noise extrapolation to reduce noise-induced upward bias [3].

The reported data compare tilt angles such as θ/π = 0.26, 0.31, and 0.49. In those measurements, the states with less initial magic acquire more late-time magic, matching the random-phase prediction [3]. The same paper also reports a control experiment in the noninteracting Anderson limit, with the interaction terms removed; under that matched condition, magic can still grow, but the magic-ordering reversal is not observed over the measured window [3]. This control is important because it supports the interpretation that the effect comes from interaction-induced many-body dephasing rather than from local precession, localization alone, or measurement noise [3].

Why “magic” changes the meaning of relaxation

In ordinary relaxation problems, a system is often described as moving toward equilibrium. In this work, the language is different. The object being tracked is not temperature, magnetization, or a conventional thermodynamic distance, but the growth and saturation of nonstabilizerness. Magic quantifies how far a state is from the stabilizer framework that can be efficiently simulated classically, and that gives the study a direct link to quantum information rather than only to statistical mechanics [1].

The authors use the second stabilizer Rényi entropy, M2, as their magic measure [3]. For the tilted product states they consider, all three spatial patterns have the same initial magic density, and in the range π/4 < θ < π/2 that initial magic density decreases as θ increases [3]. However, increasing θ also makes the conserved configuration weights more uniform, which leaves more room for dephasing-generated magic at long times [3]. As a result, a state may begin with less magic but end up with more saturated magic [3].

That is the key inversion. A naive expectation would rank the states by their starting magic and assume the more magical state remains ahead. The paper’s random-phase argument predicts the opposite in the relevant tilt range: the initial ordering and saturated ordering are reversed [3]. The authors then add a stronger time-domain test: the state that starts with less magic and farther from its own final value must reach that final value sooner [3]. In exact N = 14 simulations across the interacting l-bit model and several localized spin-chain settings, the paper reports that θ = 0.50π saturates earlier than θ = 0.35π for every localized model and initial-state family studied [3].

Localized dynamics without ordinary thermalization

One of the most interesting aspects of the work is what it says the effect does not require. Many discussions of relaxation emphasize transport, scrambling, and thermalization. Here, the claimed mechanism is interaction-induced dephasing in a localized setting. In the effective l-bit description, the conserved quantities remain fixed while relative phases evolve coherently, creating a route for magic to change without ordinary transport [3].

The authors contrast the localized regime with an ergodic regime. In their description, the localized regime displays pattern-robust reversal across tilted ferromagnetic, Néel, and domain-wall initial states, while the ergodic regime is more selective about which initial patterns exhibit the reversal [1]. The paper further reports that microscopic spin models show the reversal universally across the studied initial patterns in the localized regime but only selectively in the ergodic regime [1].

This distinction matters because it suggests that the magic Mpemba effect is not merely a byproduct of fast chaotic scrambling. Instead, the localized system’s restricted dynamics can create a structured dephasing process that lets an initially “less resourceful” state exploit a larger asymptotic capacity for magic. The authors also report that comparable saturation-time inversions appear in several qualitatively different localized settings: an effective interacting l-bit model, random-disorder localization, quasiperiodic localization, and Stark localization [3].

The digital-quantum-computer angle

The work is not only a theory paper. Its current significance comes from the reported observation on IBM superconducting processors. The arXiv abstract names IBM superconducting quantum processors as the platform for observing the two central features: states with less initial magic attaining higher saturated magic, and those states saturating earlier [1].

The hardware details show the practical challenge. The l-bit implementation used ibm boston for dynamical measurements and ibm kingston for polarization-based reconstruction of saturated magic, according to the supplemental details in the PDF [3]. Both processors are described as 156-qubit heavy-hex devices, though the dynamical Bell-protocol evidence comes from six-qubit measurements on ibm boston [3]. The authors state that the 156-qubit estimate serves as a model-assisted reference for tilt dependence, not as the time-resolved dynamical evidence [3].

That distinction is worth keeping clear. The headline result is not a claim that a full 156-qubit many-body system was dynamically tracked through the magic Mpemba effect. Rather, the strongest time-resolved hardware evidence described in the paper comes from carefully designed six-qubit protocols, while larger-device measurements help reconstruct and compare saturated-magic trends [3]. This is a measured but meaningful step: it uses present-day digital quantum hardware to probe a subtle resource-relaxation phenomenon that would be hard to access by a simple classical intuition.

What is new, and what remains open

As of the current fresh sources, the work is a newly submitted arXiv preprint, not a final journal publication [1]. The arXiv listing shows it among new quantum physics submissions on October 7, 2026 [2], and an arXiv mirror records the same paper with submission in October 2026 and last update on October 7, 2026 [4]. The current state, therefore, is best described as a recent preprint report of a hardware observation, accompanied by theory, simulations, control measurements, and supplemental hardware-method details.

The study’s main conceptual contribution is to separate a magic Mpemba effect from conventional thermalization and transport. Its main experimental contribution is to implement a digital-quantum-computer protocol that detects a magic-ordering reversal and compares it with a noninteracting control. Its main caveat is the usual one for near-term quantum hardware: the signals require error mitigation, limited qubit counts for the most direct dynamics, and careful interpretation of nonlinear magic estimators [3].

If the result holds up under peer review and independent reproduction, it could add a new layer to how researchers think about relaxation in quantum computers. The lesson would not be simply that “farther can be faster,” but that the relevant notion of distance depends on the resource being measured. For quantum computation, that resource can be magic. In this story, the state that starts with less of it can end up generating more of it, faster, because interactions dephase many-body phases in a way that reverses the expected ordering.

Sources from the last 72 hours

1. [1][2610.08769] Observing the magic Mpemba effect in localized dynamics on a digital quantum computerOct 6, 2026, 7:54 PM
2. [2]Quantum Physics: New submissions for Wednesday, 7 October 2026Oct 7, 2026, 2:00 AM
3. [3]Observing the magic Mpemba effect in localized dynamics on a digital quantum computerOct 6, 2026, 7:54 PM
4. [4]Observing the magic Mpemba effect in localized dynamics on a digital quantum computerOct 6, 2026, 2:00 AM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.