Gate Optimization via Efficient Two-Qubit Benchmarking for NV Centers in Diamond
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | Open MIND |
| Published | 2026-03-09 |
| Authors | Alessandro Marcomini, Philipp J. Vetter, Tommaso Calarco, Felix Motzoi, Fedor Jelezko |
| DOI | 10.48550/arxiv.2603.08581 |
A simulated closed-loop CNOT reached about 99.9% fidelity while requiring four measurements per figure-of-merit evaluation instead of 144 for standard process tomography. The result applies to an NV center coupled to a 13C nuclear spin, with feedback limited to four pulse parameters. All performance values in the paper come from numerical simulation rather than a hardware experiment.
Reduced benchmark for feedback calibration
Section titled “Reduced benchmark for feedback calibration”Full quantum process tomography reconstructs the complete two-qubit process, but its measurement cost becomes a major burden during closed-loop optimization. The conclusion compares the four-measurement protocol with 144 measurements for standard quantum process tomography. The methods section allows up to 256 measurements for a general completely positive trace-preserving map. The abstract describes this lower demand as a two-order-of-magnitude reduction.
Each evaluation uses two probe-state types instead of reconstructing a full process map. The figure of merit F_J is not identical to gate fidelity F_sm. For near-unitary evolution, maximizing F_J corresponds to maximizing F_sm, but identical behavior is not guaranteed away from the optimum.
NV center and nuclear spin system
Section titled “NV center and nuclear spin system”The modeled register contains one NV center and one nearby 13C spin from the G family. A 600 G magnetic bias field lies along the NV symmetry axis. The nitrogen spin is treated as fully polarized through polarization transfer from the electron spin, so its hyperfine interaction is neglected.
The NV zero-field splitting is D = 2π · 2.87 GHz, and the nuclear Larmor frequency is ω_n = 2π · 642 kHz. Nominal hyperfine parameters are A_zz = 2π · 2.281 MHz and A_zx = A_zy = 2π · 0.240 MHz. After rotating the coordinate frame, the model uses A_zx = 2π · 0.339 MHz and A_zy = 0.
The effective electron qubit uses |m_s = 0⟩ and |m_s = −1⟩, while the nuclear qubit uses |m_I = ±1/2⟩. The target controlled-NOT gate flips the electron spin conditionally on the nuclear spin and includes a relative phase.
Control amplitudes are limited to 2π · 30 MHz for microwave and 2π · 40 kHz for radiofrequency. The gate itself uses only microwave control, while radiofrequency pulses appear in state preparation and readout. The full Hamiltonian captures off-resonant driving among the electron and nuclear transitions. Effects of microwave control on the nuclear spin and radiofrequency control on the electron spin are neglected because they are far detuned. The two-qubit unit could serve as a register element, a sensing memory, or a light-matter interface, according to the authors.
Two probe states for the reduced metric
Section titled “Two probe states for the reduced metric”The practical metric uses a mixed probe ρ_1 with distinct diagonal populations and no coherences. It also uses a pure probe ρ_2 that is nonorthogonal to every computational basis state. The general F_J construction includes a third state for leakage, but this study omits it because evolution stays inside the qubit subspace.
Unitary control pulses cannot directly prepare the mixed state, so the protocol averages four unitary preparations. The phase tuples are (0, 0), (0, π), (π, 0), and (π, π), each combining one nuclear pulse and one electron pulse. Their average recreates the diagonal state required by the fidelity calculation.
The protocol can replace complementary phases with random samples from a uniform distribution. The authors report that 10^5 to 10^6 random realizations approximate the target ensemble average with sufficient precision for this system.
State preparation and optical readout
Section titled “State preparation and optical readout”The simulated closed-loop trials include state preparation and optical readout. A green laser initializes the electron spin, transfers nuclear polarization, and reads the |m_s = 0⟩ population.
Each probe-state preparation uses a 49.5 µs nuclear π/2 rotation followed by a 10 ns electron π/2 rotation. The prepared ρ_2 state reaches 99.93% canonical fidelity with the ideal target. The sequences keep the electron in superposition for about 1 µs, below typical dephasing times in isotopically diluted samples. The simulations omit dephasing because of this short exposure.
One readout uses a 1.74 µs electron π pulse at Ω_mw ≈ 2π · 300 kHz. That pulse has 99.3% reference fidelity to the target gate but 99.96% conditional-flip contrast. Three measured populations reconstruct the diagonal entries needed by F_J. A final basis rotation maps the target action on ρ_2 back to the initial basis for readout. These routines supply four measurements per F_J evaluation without reconstructing state coherences.
Open-loop pulse design
Section titled “Open-loop pulse design”The open-loop optimizer uses the dressed Chopped RAndom Basis method in QuOCS. It uses 10 basis functions per super-iteration and 15 super-iterations. The optimizer adjusts the microwave amplitude envelope, phase, carrier frequency, and pulse duration with Nelder-Mead. The amplitude envelope remains within the stated microwave limit.
The nominal optimization reaches F_sm = 99.97%. The best duration is approximately T_hf, the characteristic period of the hyperfine interaction. Directly applying that pulse to 20 systems with uncertain parameters usually lowers performance. Each sample system draws A_zx and A_zz from Gaussian distributions with standard deviations of 5% of nominal values. The simulations include the full Hamiltonian rather than a model with selective addressing of all four transitions. The authors state that no NV-center two-qubit gate had previously been optimized in closed-loop mode.
Closed-loop calibration
Section titled “Closed-loop calibration”Closed-loop calibration starts from the open-loop pulse and considers four scaling parameters: Ω_mw,0, ω_mw, φ_mw, and T. Nelder-Mead maximizes F_J while full state preparation and readout errors remain in each trial. The optimizer does not adjust the full pulse shape.
Calibrating amplitude, duration, and carrier frequency improves average infidelity by over an order of magnitude. Across the sample systems, F_sm settles near 99.9% after about 1000 iterations. Optimizing duration and carrier frequency is essential because omitting either produces the highest dispersion. Adding phase as a free parameter slightly worsens results because the larger search reaches local minima. The authors therefore keep the simplex small and the search near the open-loop solution.
One example used 1666 F_J evaluations, with no visible improvement after about 400 evaluations. Closed-loop optimization uses only measurement data accessible through the reduced protocol, not the simulated density matrix.
| Performance measure | Simulated value | Applied condition | Comparison |
|---|---|---|---|
| Open-loop gate fidelity | 99.97% | Nominal NV-13C model | Best open-loop result |
| Closed-loop gate fidelity | about 99.9% | 20 systems, about 1000 iterations | Average infidelity fell by over one order of magnitude |
| Probe preparation fidelity | 99.93% | State preparation and readout for ρ_2 | Canonical fidelity to the ideal state |
| Conditional-flip contrast | 99.96% | 1.74 µs electron π pulse | 99.3% reference fidelity to the target |
| Measurements per F_J evaluation | 4 | Practical closed-loop protocol | 144 for standard process tomography |
Limits and device implications
Section titled “Limits and device implications”The protocol is designed for experimental feedback, but the paper reports no hardware test. Two limitations dominate the simulated fidelity. First, state preparation and measurement errors cause F_J to differ from the error-free figure of merit. Second, the four-parameter ansatz limits how far the open-loop pulse can be corrected.
Readout noise is omitted, and the simulations assume evolution remains inside the computational subspace. The proposed remedy is repetitive benchmarking, which applies the gate multiple times in each measurement sequence. The authors also suggest full pulse-shape optimization if more evaluations are feasible.
Extending the protocol to three qubits or Group-IV diamond defects is proposed future work, not demonstrated performance. The simulation data and figure-generation code are deposited on Zenodo.
For device engineering, the protocol makes feedback calibration practical when the microwave carrier and pulse duration drift from the model. It prioritizes the two controls that most affect fidelity in the tested system.
Materials and support
Section titled “Materials and support”The paper models an NV-center register in diamond with a 13C spin, a 600 G axial field, microwave and radiofrequency control, and optical readout. The diamond materials catalog from 6CCVD includes MPCVD-grown single-crystal plates and wafers up to 125 mm, with thicknesses from 0.1 to 500 micrometers. The engineering support page can address custom fabrication questions, while defect fabrication and isotope preparation remain unspecified in the paper.
View Original Abstract
High-fidelity gate implementation requires sophisticated control pulses that steer the quantum system to undergo the desired transformation. Quantum Optimal Control allows to derive these control pulses in an open-loop fashion based on numerical simulations. However, their precision can be limited by incomplete knowledge of the system. Closed-loop optimization overcomes this limitation by incorporating feedback from measurements, provided a suitable and efficient measure of the gate performance can be defined. In this article, we present an efficient method to evaluate the performance of a two-qubit gate by preparation and measurement of only two quantum states, enabling experimental closed-loop optimization with a metric previously believed to be limited to open-loop control. We tailor the approach to nitrogen-vacancy centers in diamond and, through numerical simulations, demonstrate how the method can optimize a two-qubit gate while reducing the number of required measurements by two orders of magnitude compared to standard process tomography under realistic experimental settings.
High Concentration Quantum Grade Single Crystal Diamond
View Specs →Products
View Specs →