Skip to content

A Method For Robust Spin Relaxometry in the Presence of Imperfect State Preparation

FieldValue
JournalAdvanced Quantum Technologies
Published2026-05-01
AuthorsElla P. Walsh, Sepehr Ahmadi, Alexander J. Healey, David A. Simpson, Liam T. Hall
DOI10.1002/qute.70291

The two-state model recovered relaxation rates across the full analyzed widefield image, while a stretched exponential restricted useful imaging to about 50 µm. It returned a bare-diamond rate of 0.197 ± 0.052 kHz and a particle-region rate of 0.353 ± 0.093 kHz. The stretched exponential returned 0.626 ± 0.177 kHz and 1.242 ± 0.444 kHz for the same regions. These results show that incomplete spin preparation can dominate the fitted decay unless the model includes the preparation error.

Two-state model separates preparation from relaxation

Section titled “Two-state model separates preparation from relaxation”

Spin relaxometry measures decay from a polarized spin population during a dark interval, τ. A green laser prepares the NV spin state, and red photoluminescence provides the readout. Repeated preparation, dark time, and readout cycles build the decay curve. The quantity of interest is Γ1 = 1/T1, while Γp describes how quickly optical pumping reaches the prepared state.

The authors represent the spin as two coupled populations and separate the polarization and relaxation propagators. The model assumes weak relaxation during pumping, Γp ≫ Γ1, and no pumping during the dark interval. Polarization inefficiency η ranges from 0 to 1, with η = 0 denoting perfect polarization. Finite pump rate and pulse duration produce η above zero.

For repeated cycles, the analytic solution uses the effective propagator raised to N. Populations settle after a few repetitions, so the authors use the steady-state limit for experiments with thousands of repeats. The normalized photoluminescence expression includes a constant offset c for baseline noise. Fitting Γ1 and η separates intrinsic relaxation from the preparation-dependent curve shape.

The simulation set Γ1 to 0.5 kHz, corresponding to T1 = 2 ms. It collected 1000 repetitions at dark times from 1 µs to 5 ms while reducing Γp relative to the fixed preparation time. A single exponential continued to fit the normalized curves, but its fitted Γ1 increased as preparation became less effective. At η ≈ 0.4, the fitted rate exceeded the true rate by about 50%.

The authors performed the fits with the lmfit Python package. They show that the artifact cannot be removed by normalization with π pulses. Imperfect preparation creates a distinct decay contribution that obscures Γ1. This matters when estimating a target’s contribution from the difference between a bare-diamond measurement and a measurement taken after the target is present.

The test sample was a 2 mm × 2 mm commercially available [100]-oriented CVD diamond. It was implanted at 4 keV with a fluence of 10^13 cm^-2 and annealed at 900 °C. Assuming a 1% nitrogen-to-NV conversion, the authors used an NV density of 1 ppm.

A custom inverted widefield microscope supplied 14 kW/μm² over a 100 µm diameter laser spot. A magnetic field of about 30 G was aligned with one of the four NV axes to isolate one NV family. Initialization and readout used a 10 µs pulse, and the camera exposure was 30 ms. Dark times were spaced exponentially from 10 µs to 9 ms. The 12 hour sequence provided sufficient signal-to-noise at single-pixel resolution.

A 300 ns π pulse supported common-mode normalization for laser fluctuations. The authors cropped a 362 × 362 pixel region without binning to include strongly and weakly illuminated areas. For the target measurement, they deposited 2 µL of AMS-40-10H paramagnetic microparticles in MilliQ at 100 µg/mL, then let the liquid evaporate.

Calibration maps polarization inefficiency

Section titled “Calibration maps polarization inefficiency”

The two-state analysis needs a separate polarization estimate because η and Γ1 are correlated. The authors fitted the bare-diamond reference while holding Γ1 at the empirical ensemble-relaxation estimate of Jarmola et al. The NV-density coefficient was 0.82 s^-1 ppm^-1. At 1 ppm, this gave A1 = 0.82 s^-1 and an optimal relaxation rate of about 180 Hz.

They then treated η as a free parameter at each pixel and built a polarization map for the particle measurement. The best-polarized region had an average η of about 0.36, while the weakly illuminated edge reached about 0.77. The map gave the subsequent fits a measured preparation condition instead of assigning all curve curvature to Γ1.

The calibration assumes that nitrogen or vacancy clusters, the target particles, and laser-power changes do not alter polarization. It omits surface noise, which the authors consider likely in this shallow ensemble. The empirical rate therefore gives a lower bound on ensemble Γ1 and η. A supporting analysis with Γ1 = 333 Hz and T1 = 3 ms showed similar model behavior.

Fit methodBare diamond Γ_intrinsicParticle region Γ_particle
Two-state model0.197 ± 0.052 kHz0.353 ± 0.093 kHz
Stretched exponential0.626 ± 0.177 kHz1.242 ± 0.444 kHz

The table gives means with standard deviations over the analyzed pixels. The stretched exponential produced a faster and broader background distribution, with strong covariance between rate and stretch parameter. Its particle rates also varied with laser illumination and deviated strongly in low-polarization regions. The two-state fit returned rates that were much more consistent across the full field of view.

For each of 19 particles, the authors averaged a 10 × 10 pixel region centered on the particle and used the regional standard deviation as the error bar. The two-state rates remained consistent across the polarization range, while the stretched exponential restricted practical use to a roughly 50 µm region. The lower background rate also supports the stated sensing goal, because a slower intrinsic Γ1 leaves more range for target-induced changes.

The model does not remove every systematic error. Its absolute rates depend on the η map, and the surface-noise omission can bias the reference estimate. The authors also limit the two-state approximation to the weak-relaxation condition during pumping, Γp ≫ Γ1. A stretched exponential can fit ensemble curves, but its effective rate lacks a defined relation to the mean of a distributed set of rates. It therefore also changes with polarization.

The current procedure requires a reference relaxometry measurement to characterize η. The authors propose using microwave-field inhomogeneity, ODMR or Rabi contrast, or photoluminescence information to constrain η more efficiently. They also suggest fitting η and Γ1 together. These options could reduce reliance on the fixed reference rate, but the paper does not validate them experimentally.

The result changes how optical relaxometry data should be collected and fitted. Near-surface NVs can suffer charge-state changes at the high excitation powers needed for full polarization. Lowering laser power avoids some of that risk, but it increases preparation error. The two-state model makes the resulting decay usable without presenting the preparation error as intrinsic relaxation.

For widefield instruments, the reference measurement should span the target measurement’s illumination profile. Each pixel can then use its measured preparation efficiency. The approach supports ensemble or nanopillar measurements where the laser profile varies across sensing sites. It does not shorten the reported 12 hour acquisition, but it retains more of the field and reduces rate-estimation scatter.

Engineers should distinguish polarization error, intrinsic diamond relaxation, and target-induced relaxation in the fit. The method is most useful when laser power is limited or spatially nonuniform. It also cautions against reading a low-polarization decay as evidence of a fast intrinsic T1.

The paper requires an NV-doped [100]-oriented CVD diamond sample for a widefield ensemble relaxometry sensor, with nitrogen implantation and annealing used to create the centers. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal and polycrystalline diamond plates and wafers up to 125 mm. The engineering support page can be used to confirm the base material, [100] orientation, implantation, annealing, and surface requirements for the finished device. The reported setup does not use boron-doped diamond or diamond metallization.

View Original Abstract

ABSTRACT Spin relaxometry using optically addressable spin systems has developed as a valuable tool in medical and condensed matter systems, offering the advantage of operating without the need for external DC or RF fields. Among optically addressable spin systems, nitrogen‐vacancy (NV) centers have been applied to paramagnetic sensing using both single crystal diamond and nanodiamond materials. However, these methods often suffer from artifacts and systematic uncertainties, particularly due to imperfect spin state preparation, leading to artificially fast relaxation times. Current analysis techniques fail to adequately account for these issues, limiting the precision of parameter estimation. In this work, we introduce a minimal fitting procedure that enables more robust parameter estimation in the presence of imperfect spin polarization. Our model improves upon existing approaches by offering more accurate fits and providing a framework for efficiently parallelizing single‐spin dynamics studies.