A four-spin proof of concept

On September 4, 2026, Carlos H. S. Vieira, Xinfang Nie, Dawei Lu and Fernando Parisio submitted the first version of a preprint combining a mathematical framework for quantum-state texture with controlled experiments on four nuclear-spin qubits. Their practical result is a compact diagnostic that used local measurements to identify whether a tested circuit layer contained a CNOT entangling operation. In one benchmark, it also distinguished which qubit acted as control and which as target. This was a laboratory proof of concept in liquid-state nuclear magnetic resonance, not a demonstration on a large programmable processor.

Quantum-state texture is a basis-dependent way of describing how far a quantum state sits from a designated flat, or textureless, reference state. The paper works with a quantity called the grand sum, which adds all entries of a state’s density matrix in the chosen basis. Because the diagonal entries already sum to one, changes in the grand sum contain information about coherence between basis states. For a single spin-1/2 qubit in this experiment, the researchers obtained it from one component of transverse magnetization: the grand sum equals one plus the measured x-magnetization.

A smaller description with explicit assumptions

That shortcut matters because full quantum-state tomography attempts to reconstruct every parameter of a state, while quantum-process tomography seeks the much larger description of an operation. Both become increasingly demanding as systems grow. The authors prove that, for any finite-dimensional quantum channel, the texture response for arbitrary inputs is encoded by how the channel’s dual map acts on one fixed textureless reference state. They estimate that this reduces a generic characterization from order D^4 parameters to order D^2, where D is the state-space dimension. It is a mathematical reduction, not a measured scaling result.

The reference cannot be omitted from the claim. Texture is defined relative to a fixed basis and its equal-superposition state. Changing that basis can change the measured texture. The theory also separates “free” operations, which leave the reference state fixed, from other channels. A free operation preserves texture only when the reference is also fixed by the dual map. The paper proves that free, unital channels satisfy that condition. Unital means the channel leaves the maximally mixed state unchanged, a property that excludes many dissipative processes found in real devices.

What the NMR processor measured

The experiments used carbon-13-labeled trans-crotonic acid dissolved in deuterated acetone. Four coupled carbon-13 nuclear spins served as the four-qubit register inside a magnetic field of about 7.050 tesla. Radio-frequency pulses controlled the spins, and the apparatus measured an ensemble containing many identical molecules rather than reading one individually addressable molecule. The experiment used effective pseudopure initial states. Those are established tools for testing quantum-control ideas with NMR, but they are physically and operationally different from scalable superconducting, trapped-ion or neutral-atom processors.

Before applying the gate diagnostic, the team tested the theory with several controlled channel implementations. These included phase damping that was unital but not free, phase damping that was both free and unital, and amplitude damping that was free but non-unital. The measured grand-sum behavior followed the predicted categories within the reported uncertainties. For one reconstruction, the researchers used a tomographically complete set of inputs to determine the dual action on the reference state. Avoiding reconstruction of the entire channel does not mean every supporting quantity can always be learned from a single measurement.

Two probes establish a local imbalance

The gate protocol rests on a balance rule. For a unital operation, averaging the grand sum across an orthonormal basis produces the value one. For a qubit, that basis contains two orthogonal states. The researchers prepared the same two product-state probes across the four-spin register, using an angle of pi divided by four, then measured each spin locally after the circuit layer. Local unitary operations preserved the balanced average. An entangling operation can break the balance on the participating qubits because those two global product preparations do not form a complete basis for the larger multi-qubit state space.

In the known-basis benchmark, a CNOT acted on qubits A and B while C and D were spectators. The ideal texture averages for control and target were 1.5 and 1.0. Reversing the CNOT reversed that pattern, and the measured bars followed the expected orientation. This establishes that the chosen probes and local readout distinguished two deliberately implemented CNOT directions in this four-spin system. It does not show that the method can identify an arbitrary unknown gate, reconstruct its complete action or measure its fidelity.

The hidden-basis test found the interacting pair

The second test withheld the gate’s rotated local basis from the diagnostic. After the CNOT, the measured averages were about 1.390 and 1.373 for participating qubits A and B. Spectator values remained near the balanced reference at 1.002 and 1.014. The local data therefore located the interacting pair without using the basis transformation that defined the implemented CNOT. In this test, both active qubits moved away from one, so the result located connectivity. It did not recover the hidden basis or provide a complete description of the entangling operation.

There are logical boundaries as well as hardware boundaries. The authors’ CNOT analysis contains exceptional basis choices for which the relevant local average can remain balanced even after entanglement. They describe those exceptions as a zero-measure set, supporting detection in almost all theoretical cases rather than every case. The argument concerns unital circuit layers and controlled product-state probes. Non-unital noise, state-preparation error, measurement bias or drift could also shift a local average on another platform. A practical diagnostic would need calibrated thresholds and tests separating an intended interaction from those confounding effects.

Local does not mean experimentally effortless

The NMR measurements themselves involved substantial processing. Each carbon resonance was split into an eight-line pattern by coupling to the other three spins. The researchers fitted the spectral lines, combined their absorptive amplitudes and normalized them against a pseudopure reference to extract x-magnetization. Reported error bars came from Monte Carlo propagation of fluctuations estimated from fit residuals. The paper identifies radio-frequency inhomogeneity, pulse deviations and magnetic-field inhomogeneity as experimental imperfections. Local measurement here means that the reported observable belongs to one spin, not that the entire experimental procedure was trivial.

The progress is a narrower, more useful claim than universal circuit verification. The work supplies a theory for how one basis-dependent resource changes under quantum channels and demonstrates that carefully selected local observables can reveal a known class of entangling layer in a four-spin register. If comparable signatures remain reliable on larger hardware, engineers could gain a lower-overhead way to screen circuit connectivity before deploying expensive full characterization. That scale-up is still unknown. The preprint does not measure performance as qubit count grows, test multiple simultaneous entangling gates, establish robustness under realistic device noise or compare total experimental cost against process tomography on another platform. The next decisive test is a larger, individually addressable processor with independently varied noise, gate types and connectivity.