Quantinuum and HQS reproduced a difficult diphosphane NMR feature on trapped-ion hardware after compressing a 22-spin model into 21 effective spins, using 70 Trotter steps and circuits containing more than 1,400 two-qubit gates
A trapped-ion quantum computer has reproduced the subtle double peak that earlier hardware experiments failed to resolve in a simulated molecular NMR spectrum. Quantinuum and HQS Quantum Simulations obtained the feature at roughly 0.07 ppm resolution on the Quantinuum System Model H2-1 while running a digital simulation of 1,2-di-tert-butyl-diphosphane. Quantinuum describes the work as an end-to-end NMR simulation on the System Model H2 platform: the reconstructed spectrum reproduced key features of a classical reference calculation, including the previously unresolved double-peak structure.
The result is best understood as a hardware benchmark rather than a claim of chemical superiority over classical methods. The experiment tested whether a quantum processor could preserve the spin correlations responsible for a difficult observable during deep digital Hamiltonian simulation. The reported run used 70 Trotter steps and represented approximately 29 milliseconds of simulated physical evolution, with spectral features resolved at about 0.07 ppm.
The benchmark is presented in a Quantinuum technical account, while the available independent industry description reports agreement between the reconstructed spectrum and classical reference calculations in the 3.9-4.3 ppm region. No official external validation or regulatory determination has been reported for this experiment.
The target is a liquid-state proton NMR spectrum for a 22-spin heteronuclear molecule containing two 31P nuclei and twenty 1H nuclei. Rather than represent every phosphorus degree of freedom independently, the team projected the two-phosphorus subsystem into its singlet-triplet (|S⟩, |T0⟩) subspace. That hardware-efficient mapping produced a 21-spin effective Hamiltonian while retaining the coupled-spin structure needed for the spectral calculation.
This is not a shortcut that turns the molecule into a smaller chemical system. It is a change in representation: the simulator uses a reduced computational description of the phosphorus pair so that the relevant dynamics can be implemented with fewer interactions. The practical consequence was substantial. The reported two-qubit gate count per Trotter step fell from 69 to 20 native ZZPhase gates.
The full architecture included 21 qubits for the effective spin system and 21 ancillary qubits. The H2-1 run therefore used 42 qubits across the reported workflow, while the deepest circuits exceeded 1,400 two-qubit gates. These figures make the demonstration significant as a control-and-compilation test, but they do not by themselves establish fault-tolerant quantum computation or a logical-qubit advantage.
The difficult test lies in the 3.9-4.3 ppm region of the diphosphane spectrum. The double-peak structure is weak enough that resolving it requires tracking coupled-spin correlations through as many as fourth order. Earlier experiments using 22 qubits on superconducting and trapped-ion platforms struggled to recover that feature, making it a more demanding benchmark than simply reproducing a broad spectral envelope.
The H2-1 result is a hardware demonstration of digital Hamiltonian simulation rather than an assertion that the processor has surpassed classical chemistry software. The measured spectrum reproduced the double peak with an approximate resolution of 0.07 ppm, the central experimental outcome reported for the benchmark. In this context, resolution refers to the ability to distinguish nearby spectral structure, not to a measurement of molecular structure beyond the modeled Hamiltonian.
Two properties of the trapped-ion architecture were important to the reported result. The processor provides all-to-all connected ionic channels, so the mapped interactions do not require the same routing overhead imposed by restricted-connectivity devices. Quantinuum also attributes the preservation of spectral features to high-fidelity gate operation and error-suppression methods used during the computation.
Those details matter because a digital simulation accumulates error during state preparation, Trotterized evolution and readout. Reducing the number of native two-qubit operations lowers one source of exposure, but it does not make the computation error-free. The reported techniques concern hardware control and error mitigation, not a fault-tolerant logical simulation with encoded qubits.
The comparison with classical computation is equally important. Classical solvers such as Spinach remain more computationally efficient for routine high-field liquid-state spectroscopy. The quantum experiment therefore does not establish general quantum advantage, a faster production workflow or a replacement for established NMR simulation tools. It shows that a carefully mapped Hamiltonian can be executed deeply enough on present hardware to recover a chemically meaningful spectral feature.
The strongest case for this work is methodological. It connects molecular-spin modeling, circuit compilation, native trapped-ion gates and spectral readout in one end-to-end experiment. That integration is more informative than a raw qubit count because it tests whether the hardware can preserve the correlations that determine an observable rather than merely run an isolated gate sequence. The same emphasis on reproducible instrumentation and benchmark definitions is familiar from large research ecosystems such as CERN and MIT, although this experiment remains a distinct quantum-chemistry demonstration.
The result also clarifies the boundary between a benchmark and a useful application. The molecule is a demanding test for current digital simulation, but the reported comparison still favors classical methods in the routine regime. The authors identify zero- to ultralow-field NMR and solid-state materials characterization as areas where deep digital Hamiltonian simulation could become relevant. The work is presented through a preprint manuscript, so its technical claims should be read as preliminary rather than as independently established consensus or a peer-reviewed finding in Nature.
That boundary is familiar across quantum engineering: the earlier network analysis likewise separates a demonstrated software capability from the much larger engineering problem of dependable system operation. Here the unresolved issue is not whether the spectrum can be reconstructed once, but how reliably similar circuits can be mapped, calibrated and repeated as models grow.
An effective Hamiltonian is the operator that governs the simulated spin dynamics, while a Trotter step approximates its continuous evolution with a sequence of implementable gates. The singlet-triplet projection reduces the number of active degrees of freedom without turning physical qubits into logical qubits or removing hardware noise. This demonstration therefore signals genuine progress in full-stack quantum simulation, but not quantum advantage: its value lies in showing that trapped-ion control can recover a difficult NMR signature while classical solvers remain the practical baseline.
