NV-Center Sensing

A negatively charged nitrogen-vacancy center, NV$^-$, combines an electronic spin, optical initialization and fluorescence readout, microwave control, and atomic-scale localization in diamond. Those ingredients make it a versatile transducer, but they do not define one universal sensor. Continuous-wave ODMR, Ramsey magnetometry, echo and dynamical-decoupling spectroscopy, and $T_1$ relaxometry generate different data and estimate different properties of the environment.

The central calculation connects a field-dependent spin population to the detected photons. It explains why longer coherence, brighter fluorescence, and a shallower sensor improve different parts of the measurement, and why none alone determines sensitivity or spatial resolution.

Required background. Magnetometry defines field modes and sensitivity; Ramsey interferometry supplies phase readout. NV centers and solid-state defects explains optical pumping and charge dynamics, and dynamical decoupling develops the control filters.

Minimal Spin Model and Frequency Conventions

Using dimensionless spin-1 operators, a common effective ground-state model is

$$\frac{H}{\hbar} = D S_z^2 + \gamma_e\mathbf B\cdot\mathbf S + E(S_x^2-S_y^2) + \frac{H_{\mathrm{hf}}+H_{\mathrm{other}}}{\hbar}.$$

The NV axis defines $z$, $D/(2\pi)$ is approximately $2.87\,\mathrm{GHz}$ near room temperature, and

$$\frac{|\gamma_e|}{2\pi} \simeq 28.0\,\mathrm{GHz/T}.$$

Here $D$ and $E$ are angular frequencies, whereas $H_{\mathrm{hf}}$ and $H_{\mathrm{other}}$ are energies. We choose $\gamma_e>0$ in the displayed $+\hbar\gamma_e\mathbf B\cdot\mathbf S$ Hamiltonian: it is the magnitude of the electron gyromagnetic ratio. The signed electron magnetic moment has the opposite sign. The simplified strain term uses transverse principal axes; general strain can also require $S_xS_y+S_yS_x$.

For an axial field, ignoring hyperfine interactions, the $m_s=\pm1$ block has eigenfrequencies $D\pm\sqrt{(\gamma_e B_\parallel)^2+E^2}$ relative to $m_s=0$. Thus resolving two lines does not by itself establish linear field response: strain can split the zero-field lines. The following approximation requires $|\gamma_e B_\parallel|\gg|E|$, weak transverse mixing, and $|\gamma_e B_\parallel|\ll D$ so both transitions remain above $m_s=0$. Labeling branches by their adiabatic $m_s=\pm1$ character gives

$$\nu_\pm \simeq \frac{D+\delta_{\mathrm{cm}}}{2\pi} \pm \frac{|\gamma_e|}{2\pi}B_\parallel.$$

The term $\delta_{\mathrm{cm}}$ collects common-mode shifts from temperature, longitudinal strain, and other perturbations at the chosen operating point. The branch average and difference are therefore approximately

$$\begin{aligned} \frac{\nu_++\nu_-}{2} &\simeq \frac{D+\delta_{\mathrm{cm}}}{2\pi}, \\ \nu_+-\nu_- &\simeq 2\frac{|\gamma_e|}{2\pi}B_\parallel. \end{aligned}$$

This common-versus-differential structure is central to nuisance rejection. It is only approximate when transverse fields, strain mixing, hyperfine structure, or level anticrossings matter. Fitting the simple linear formula outside that regime biases the inferred field.

Diamond supplies four crystallographic NV orientation classes. A single resolved class measures one projection $\mathbf n_i\cdot\mathbf B$; it does not directly measure the full vector field.

Fluorescence Is a Likelihood, Not a Spin Value

Off-resonant optical excitation usually prepares population preferentially in $m_s=0$ and produces spin-dependent fluorescence during readout. Let $q_0(\theta)$ be the probability that the final spin is in the bright state. For a readout window in which each conditional count law is adequately approximated as Poisson, a simple single-shot model is a mixture:

$$p(n\mid\theta) = q_0(\theta) e^{-\alpha_0}\frac{\alpha_0^n}{n!} + [1-q_0(\theta)] e^{-\alpha_1}\frac{\alpha_1^n}{n!},$$

where $\alpha_0>\alpha_1$ are the mean detected counts conditioned on the two spin classes. Optical pumping during readout, background fluorescence, charge-state conversion, detector dead time, and time gating can require a richer model.

The classical Fisher information in the photon record is

$$F_\theta = \sum_{n=0}^{\infty} \frac{ [\partial_\theta p(n\mid\theta)]^2 }{p(n\mid\theta)}.$$

This expression automatically penalizes poor photon collection and overlapping bright and dark distributions. Replacing fluorescence with an ideal projective spin measurement can overstate the available information by orders of magnitude, especially for room-temperature single-shot readout.

Writing the two conditional Poisson laws as $f_0(n)$ and $f_1(n)$, assume that their calibrated means do not depend on the target parameter. Only $q=q_0(\theta)$ then carries the signal, so

$$F_\theta=(\partial_\theta q)^2 \sum_n\frac{[f_0(n)-f_1(n)]^2}{qf_0(n)+(1-q)f_1(n)}.$$

An ideal measurement of the spin class would instead give $(\partial_\theta q)^2/[q(1-q)]$. The retained fraction

$$\mathcal R(q)=q(1-q)\sum_n \frac{[f_0(n)-f_1(n)]^2}{qf_0(n)+(1-q)f_1(n)} \le1$$

quantifies the readout penalty. The inequality follows from classical data processing: photons are sampled from a spin class without further access to $\theta$. Equal count laws give $\mathcal R=0$; disjoint laws approach one. Thresholding photons into a bright/dark label cannot increase this information. If the target also changes the count laws, their derivatives must be included rather than folded into this expression.

The mixed photon record is generally not Poisson with its average count:

$$\bar\alpha=q\alpha_0+(1-q)\alpha_1,\qquad \operatorname{Var}(n)=\bar\alpha+q(1-q)(\alpha_0-\alpha_1)^2.$$

The second term is the spin-population fluctuation. Treating the mixture as a single Poisson distribution discards this contribution and changes the likelihood.

Reference windows are often used to normalize laser intensity and fluorescence drift. They add data but also consume time and can introduce covariance. The likelihood should include signal and reference counts jointly rather than treating the normalized ratio as noiseless.

For independent cycles of duration $T_c$ with per-cycle information $F_\theta^{(1)}$, the local information rate is

$$\dot F_\theta = \frac{F_\theta^{(1)}}{T_c}.$$

Shared laser, temperature, or magnetic drift requires a joint likelihood before information can be added across cycles.

Continuous-Wave ODMR

In continuous-wave optically detected magnetic resonance (CW ODMR), optical and microwave drives are applied together while fluorescence is recorded as a function of microwave frequency. A simple isolated resonance is

$$R(\nu;B) = R_0 \left[ 1-C_{\mathrm{ODMR}} L(\nu-\nu_0(B)) \right] +R_{\mathrm{bg}},$$

with Lorentzian

$$L(x) = \frac{1}{1+4x^2/(\Delta\nu)^2}.$$

Here $\Delta\nu$ is the full width at half maximum and $C_{\mathrm{ODMR}}$ is the fractional contrast. For Poisson counts collected for time $t$ at one working frequency,

$$F_B = t \frac{[\partial_B R(\nu;B)]^2}{R(\nu;B)}.$$

Defining sensitivity by $\eta_B=\delta B\sqrt{t}$, the local Cramér–Rao coefficient is therefore

$$\eta_B(\nu) = \frac{\sqrt{R(\nu;B)}} {|\partial_B R(\nu;B)|}.$$

For an isolated low-contrast Lorentzian, negligible background, and frequency gyromagnetic ratio $g_e=|\gamma_e|/(2\pi)$, optimizing the slope gives

$$\eta_B^{\mathrm{CW}} \simeq \frac{4}{3\sqrt3} \frac{\Delta\nu} {C_{\mathrm{ODMR}}g_e\sqrt{R_0}}.$$

Here sensitivity has the averaging-time convention, not the one-sided amplitude-spectral-density normalization. For continuous white noise and rectangular averaging, the latter is $\sqrt2\,\eta_B$.

The low-contrast optimum lies at $|\nu-\nu_0|=\Delta\nu/(2\sqrt3)$. At finite contrast the denominator $R(\nu;B)$ also shifts the optimum; maximizing slope alone is approximate. The review by Barry and colleagues derives the principal readout and coherence limitations behind these comparisons.

This expression is not universal. Power broadening changes $\Delta\nu$, optical and microwave powers change contrast and photon rate, neighboring hyperfine and orientation lines distort the slope, and technical laser noise violates the Poisson model. Frequency modulation and lock-in detection can reject slow intensity drift, but the measured complex transfer function and demodulation bandwidth still need calibration.

CW ODMR is operationally simple and naturally supports imaging. It is often a good choice when moderate DC bandwidth, robust acquisition, or many camera pixels matter more than maximum per-spin information. A full spectrum is also valuable for identifying branches and nuisance structure before operating a locked sensor.

Pulsed Phase Sensing

A pulsed experiment separates initialization, coherent evolution, and readout. For a selected two-level subspace, the accumulated magnetic phase is

$$\phi_B = \gamma_e \int_0^T y(t)B_\parallel(t)\,dt.$$

With a choice of microwave analysis phase and branch labeling, the final bright-state probability can be written

$$q_0(B) = \frac12 \left[ 1+C_s\sin(\phi_B+\vartheta) \right],$$

where the spin-state contrast $C_s$ describes coherent preparation and evolution. The photon likelihood adds a separate readout penalty through $\alpha_0$ and $\alpha_1$. Conflating those two contrasts hides where information was lost.

For a known waveform $B_\parallel(t)=B\,s(t)$, define $\mathcal A=\int_0^T y(t)s(t)\,dt$. At quadrature, the photon Fisher information is $F_B=C_s^2(\gamma_e\mathcal A)^2\mathcal R(1/2)$. Thus an effective contrast $C_{\mathrm{eff}}=C_s\sqrt{\mathcal R(1/2)}$ gives a compact local benchmark for this calibrated likelihood. For $N_{\mathrm{eff}}$ independent, equivalently coupled NV centers,

$$\eta_B \simeq \frac{\sqrt{T_c}} {C_{\mathrm{eff}}|\gamma_e|\, |\mathcal A|\sqrt{N_{\mathrm{eff}}}}, \qquad \mathcal A = \int_0^T y(t)s(t)\,dt.$$

For actual analysis, evaluating the Fisher information of the measured photon likelihood is preferable to assigning one catch-all contrast.

Ramsey sensing

For Ramsey sensing, $y(t)=1$ during free evolution. It responds to a static or slowly varying field over a time usually limited by $T_2^*$. The local phase slope grows as $T$, while dephasing and phase aliases constrain useful interrogation. The result is a projection along the selected NV axis at the chosen bias point.

Double-quantum coherence

A double-quantum (DQ) protocol prepares coherence between $m_s=+1$ and $m_s=-1$. Its magnetic phase is approximately

$$\phi_{\mathrm{DQ}} = 2\gamma_e \int_0^T y(t)B_\parallel(t)\,dt.$$

The factor of two increases phase slope, while shifts common to both $m_s=\pm1$ levels cancel to first order. DQ sensing can therefore reject temperature and longitudinal-strain shifts that broaden single-quantum ensembles. It is not automatically twice as sensitive: preparation and readout errors, microwave inhomogeneity, altered coherence, and overhead must be included. Transverse mixing and imperfect branch symmetry can leave residual nuisance response.

Echo and dynamical decoupling

Hahn echo and multipulse sequences change the sign of $y(t)$. They suppress slow sensor-frequency noise while retaining target fields whose sign matches the toggling pattern. A sequence with interpulse spacing $\tau$ has response near frequencies of order $1/(2\tau)$, with convention-dependent harmonics and finite-pulse corrections.

The same filter acts on target and environmental fields. A narrow coherence dip may indicate a coherent signal, a stochastic spectral feature, a nearby nuclear spin, a pulse harmonic, or control error. Phase cycling, sequence variation, field dependence, and independent calibration are needed to distinguish them. Dynamical Decoupling develops the filter functions in detail.

Correlation and synchronized spectroscopy

One coherent evolution interval need not set the ultimate spectral resolution. Correlation spectroscopy and clock-referenced repeated measurements encode the signal phase across separated windows. Resolution can then be set by the total record length or reference-clock stability, while sensitivity still depends on per-window coherence, readout, dead time, and signal phase memory. Long Fourier records do not evade those resource costs.

Longitudinal Relaxometry

$T_1$ relaxometry asks a different question. Prepare a spin population, wait in the dark for time $\tau$, and read out the survival probability. A simple fit model is

$$P_0(\tau) = P_\infty + [P_0(0)-P_\infty]e^{-\Gamma_1\tau},$$

although the spin-1 level structure can require multiple rates. Additional transverse magnetic noise contributes schematically

$$\Gamma_1^{\mathrm{ext}} \propto \gamma_e^2 S_{B,\perp}(\omega_\pm),$$

with tensor projections and one-sided or two-sided factors fixed by the spectral convention. Sweeping the bias field tunes $\omega_\pm$ and samples different parts of the environmental spectrum.

Relaxometry detects fluctuations that exchange energy with the NV spin. It can probe paramagnetic molecules, Johnson noise, spin waves, and magnetic phase dynamics even when the mean stray field vanishes. It does not directly return a deterministic DC field, and a measured change in $T_1$ is not itself a noise PSD until a coupling geometry and rate model are supplied.

Optimizing the dark time requires the full count likelihood. Waiting much longer than $T_1$ produces equilibrium populations with little rate information; waiting too briefly produces little contrast. Background relaxation should be measured at matched temperature, depth, optical history, and bias field.

Spatial Transfer and Standoff

The atomic scale of an NV wavefunction does not by itself guarantee atomic spatial resolution. A magnetic dipole $\boldsymbol\mu$ at displacement $\mathbf r$ produces

$$\mathbf B_{\mathrm{dip}}(\mathbf r) = \frac{\mu_0}{4\pi r^3} \left[ 3(\boldsymbol\mu\cdot\widehat{\mathbf r}) \widehat{\mathbf r} - \boldsymbol\mu \right].$$

The signal can therefore change sharply with NV depth and tip-to-sample gap. If the distance doubles, a point-dipole field falls by a factor of eight. Uncertainty in depth can dominate the inferred moment or spin density.

For planar sources, a spatial Fourier component with lateral wave number $k$ typically reaches a sensor at height $d$ with a factor proportional to $e^{-kd}$. High-spatial-frequency structure is exponentially suppressed. Deconvolution can amplify those components, but it also amplifies noise and model error. A magnetic image should report the measured projection and height before presenting a source reconstruction.

Scanning single NV

A scanning probe places one shallow NV near a sample and records one field projection or noise statistic point by point. Its strengths are proximity, small sensing volume, and operation over wide temperature and field ranges. Its limitations include photon rate, scan time, tip wear, topographic cross-talk, uncertain NV position and axis, drift, and perturbation from the probe or applied microwaves.

Wide-field and confocal ensembles

An ensemble provides parallel pixels and more photons. The effective number of sensors in a pixel is not simply the implanted density times the nominal volume. Orientation selection, charge-state fraction, optical point-spread function, depth distribution, microwave field, collection efficiency, and analysis weighting all matter.

Increasing density can shorten $T_2^*$ through paramagnetic impurities, dipolar interactions, strain inhomogeneity, and charge instability. Camera binning improves count statistics while reducing spatial resolution. Overlapping processing windows and shared technical fluctuations can correlate pixels; summing disjoint, independently noisy pixels does not itself create cross-pixel covariance. Sensitivity per pixel should always be paired with pixel area, NV-layer thickness, standoff, and measured point-spread response.

Vector Reconstruction from Four Orientations

Choose directed unit vectors $\mathbf n_i$ along the four tetrahedral NV axes. In the ideal geometry,

$$\sum_{i=1}^{4}\mathbf n_i=0, \qquad \sum_{i=1}^{4} \mathbf n_i\mathbf n_i^{\mathsf T} = \frac43 I.$$

If spectroscopy identifies the signed projections $b_i=\mathbf n_i\cdot\mathbf B$, then

$$\mathbf B = \frac34 \sum_{i=1}^{4}b_i\mathbf n_i.$$

Real reconstruction is a weighted inverse problem. Resonance overlap, orientation-dependent contrast, unequal linewidths, strain, hyperfine structure, bias-field uncertainty, and camera-pixel covariance should enter a design matrix and covariance model. If only transition splittings are known, sign ambiguities may remain unless a bias field or branch assignment resolves them.

The four-orientation formula estimates the local vector field at the NV layer. Inferring current density, magnetization, or spin texture from that field is a second inverse problem with boundary conditions and regularization. Maxwell constraints can help, but they do not make an underdetermined source unique.

Nanoscale Nuclear Magnetic Resonance

Near-surface NV centers can detect nuclear-spin fields from volumes far smaller than conventional inductive NMR. At that scale, statistical polarization can exceed the mean thermal polarization. The observed quantity is then often a field variance or correlation rather than a phase-coherent mean field:

$$\langle B(t)\rangle \simeq0, \qquad \langle B(t)B(0)\rangle \ne0.$$

For independent, approximately unpolarized nuclei of spin density $\rho$ in a half-space above a shallow NV, dipolar coupling gives the scaling

$$B_{\mathrm{rms}}^2 \propto \rho \int d^3r\,r^{-6} \propto \frac{\rho}{d^3},$$

up to angular factors and sample geometry. Thus $B_{\mathrm{rms}}\propto\sqrt\rho\,d^{-3/2}$ in the ideal half-space model. The depth inferred from a proton layer is model-dependent; surface roughness, adsorbates, finite thickness, diffusion, and pulse errors can shift it.

An XY8 or related sequence can reveal a response near a nuclear Larmor frequency. Establishing chemical identity or molecular structure requires more than matching one dip frequency. Field scaling, isotope controls, harmonic checks, correlation times, spectral resolution, surface chemistry, and a quantitative coupling model are needed. Diffusion can broaden nanoscale liquid signals because molecules move through a strongly inhomogeneous near-field kernel.

Detection of an ensemble’s statistical polarization should not be described as single-nucleus detection. Conversely, a coherently coupled individual nuclear spin in the diamond host is a different regime from external ensemble NMR.

Magnetic and Nonmagnetic Nuisance Responses

NV centers respond to several environmental variables. That versatility is a feature when the variable is the target and a cross-sensitivity when it is not.

Temperature, strain, and electric fields

Temperature changes $D$; strain and electric fields shift and mix levels. Branch differencing or DQ coherence rejects common-mode shifts to first order, while branch averaging can be used for thermometry. Neither operation removes all effects of transverse fields, line-shape changes, or spatially varying contrast. The response matrix should be calibrated over the actual operating range.

Charge and optical history

Optical power affects NV$^-$ population, NV$^0$ background, spin polarization, readout contrast, local heating, and sample photochemistry. A dark interval does not necessarily erase the preparation history. Laser-power references and charge-state diagnostics are especially important near surfaces and in nanodiamonds.

Microwave control

Microwave amplitude and detuning errors alter pulse area and filter functions. Conductors or magnetic samples can distort the near field, while microwave dissipation can heat the sample. A sequence that performs well on an isolated calibration NV may not have the same transfer function across an ensemble image.

Hyperfine and orientation structure

The host nitrogen and nearby $^{13}\mathrm C$ spins produce resolved or unresolved hyperfine features. They can serve as resources, spectral labels, or coherent ancillas, but ignoring them can bias resonance centers and create spurious filter responses. The chosen nuclear-spin state and polarization protocol belong in the measurement model.

Choosing a Protocol

Target	Natural starting protocol	Main response	Frequent limiting factors
Static or slowly varying field	CW ODMR or Ramsey	Resonance shift or free phase	Linewidth, $T_2^*$, drift, phase range
Known AC waveform	Echo or dynamical decoupling	Filtered coherent phase	$T_2$, pulse errors, timing, harmonics
Unknown narrowband coherent tone	Correlation or synchronized spectroscopy	Phase across repeated windows	Clock stability, dead time, phase memory
Magnetic noise near the NV transition	$T_1$ relaxometry	Population-relaxation rate	Background $T_1$, rate model, geometry
Nanoscale nuclear-spin fluctuations	Multipulse correlation spectroscopy	Coherence loss or correlation	Depth, diffusion, surface noise, aliases
Wide-field vector map	Multi-orientation ODMR	Several resonance projections	Line overlap, strain, pixel covariance

The table gives starting points, not exclusive assignments. Hybrid protocols can combine a coarse ODMR spectrum, a locked Ramsey channel, and intermittent calibration. The final choice should maximize information for the declared task per unit wall time, sample dose, and acceptable disturbance.

Worked CW Benchmark

Suppose an isolated Lorentzian has

$$\Delta\nu=1.0\,\mathrm{MHz}, \qquad C_{\mathrm{ODMR}}=0.030, \qquad R_0=5.0\times10^5\,\mathrm{s^{-1}}.$$

Using $g_e=28.0\,\mathrm{GHz/T}$ gives

$$\begin{aligned} \eta_B^{\mathrm{CW}} &\simeq \frac{4}{3\sqrt3} \frac{10^6} {0.030(28.0\times10^9) \sqrt{5.0\times10^5}} \\ &\simeq 1.3\times10^{-6} \ \mathrm{T}/\sqrt{\mathrm{Hz}}. \end{aligned}$$

This is a photon-shot-noise benchmark near the optimum slope. A measured spectrum can be worse because of laser noise, frequency drift, background, overlapping lines, and estimator inefficiency. A pulsed protocol may improve the result by separating optical broadening from coherent evolution, but only after preparation and readout overhead are counted.

Separating the sample response from the measurement

A feature that follows the nuclear gyromagnetic ratio under a bias-field sweep is stronger evidence for a nuclear signal than a dip at one frequency. Changing pulse spacings and phases tests harmonics; isotope substitution tests the species assignment. Neither establishes single-molecule detection without a coupling and occupancy model. Chemical resolution additionally requires distinguishable chemical shifts or couplings rather than only a resolved Larmor frequency.

Optical and microwave heating, photoactivation, bias fields, and tip forces can change the sample. Repeating a measurement at different powers or standoff distances helps separate those changes from the intended response. For reconstructed source maps, the measured field projection and its uncertainty remain the direct observations; inversion adds assumptions about height, boundary conditions, and source geometry.

Exercises

1. Common and differential resonance shifts

Suppose

$$\nu_\pm = 2.870000\,\mathrm{GHz} +50\,\mathrm{kHz} \pm g_e B_\parallel,$$

with $g_e=28.0\,\mathrm{GHz/T}$, and the measured resonances are $\nu_+=2.871450\,\mathrm{GHz}$ and $\nu_-=2.868650\,\mathrm{GHz}$. Infer the common-mode shift and $B_\parallel$.

Solution

The average is

$$\frac{\nu_++\nu_-}{2} = 2.870050\,\mathrm{GHz},$$

so the common-mode shift from $2.870000\,\mathrm{GHz}$ is $50\,\mathrm{kHz}$. The splitting is

$$\nu_+-\nu_- = 2.800\,\mathrm{MHz}.$$

Therefore

$$B_\parallel = \frac{2.800\times10^6} {2(28.0\times10^9)} = 5.00\times10^{-5}\ \mathrm T = 50.0\ \mu\mathrm T.$$

2. CW photon-shot-noise floor

Use the Lorentzian benchmark to evaluate a resonance with $\Delta\nu=2.0\,\mathrm{MHz}$, $C_{\mathrm{ODMR}}=0.020$, and $R_0=2.0\times10^5\,\mathrm{s^{-1}}$. Take $g_e=28.0\,\mathrm{GHz/T}$.

Solution

Substitution gives

$$\eta_B^{\mathrm{CW}} = \frac{4}{3\sqrt3} \frac{2.0\times10^6} {0.020(28.0\times10^9) \sqrt{2.0\times10^5}}.$$

Numerically,

$$\eta_B^{\mathrm{CW}} \simeq 6.1\times10^{-6} \ \mathrm{T}/\sqrt{\mathrm{Hz}}.$$

This is the idealized photon floor at the optimum point, not the complete measured noise spectrum.

3. Photon-count Fisher information

For the mixture likelihood, show that if $\alpha_0=\alpha_1$, the photon record contains no information about a parameter that enters only through $q_0(\theta)$.

Solution

If $\alpha_0=\alpha_1=\alpha$, then both conditional count distributions are the same. The likelihood becomes

$$p(n\mid\theta) = [q_0(\theta)+1-q_0(\theta)] e^{-\alpha}\frac{\alpha^n}{n!} = e^{-\alpha}\frac{\alpha^n}{n!}.$$

It is independent of $\theta$, so $\partial_\theta p(n\mid\theta)=0$ for every $n$ and

$$F_\theta=0.$$

Coherent phase accumulation is not useful unless readout converts it into a distinguishable record.

4. Single- versus double-quantum gain

An SQ protocol has phase slope $\gamma_eT$, contrast $C$, and cycle time $T_c$. A DQ protocol has slope $2\gamma_eT$, contrast $0.80C$, and cycle time $1.10T_c$. Assume equal coherence and photon statistics. Find the ratio $\eta_B^{\mathrm{DQ}}/\eta_B^{\mathrm{SQ}}$.

Solution

Using $\eta_B\propto\sqrt{T_c}/(C|\partial_B\phi|)$,

$$\frac{\eta_B^{\mathrm{DQ}}} {\eta_B^{\mathrm{SQ}}} = \frac{\sqrt{1.10}} {0.80\times2} \simeq 0.66.$$

The DQ protocol improves this matched local sensitivity by about a factor of $1/0.66\simeq1.53$, not a full factor of two, because contrast and overhead changed.

5. Standoff scaling

A point magnetic dipole produces $80\,\mathrm{nT}$ at an NV located $20\,\mathrm{nm}$ away along a fixed geometry. What field is expected at $30\,\mathrm{nm}$?

Solution

For fixed orientation, the dipole field scales as $r^{-3}$. Hence

$$B(30\,\mathrm{nm}) = 80\,\mathrm{nT} \left(\frac{20}{30}\right)^3 \simeq 23.7\,\mathrm{nT}.$$

A $10\,\mathrm{nm}$ standoff change reduces the signal by more than a factor of three. This is why depth and gap uncertainty must accompany a source claim.

6. Tetrahedral vector reconstruction

Use the four ideal axes

$$\begin{aligned} \mathbf n_1&=(1,1,1)/\sqrt3, &\mathbf n_2&=(1,-1,-1)/\sqrt3, \\ \mathbf n_3&=(-1,1,-1)/\sqrt3, &\mathbf n_4&=(-1,-1,1)/\sqrt3. \end{aligned}$$

Show that $\sum_i\mathbf n_i\mathbf n_i^{\mathsf T}=4I/3$ and derive the reconstruction formula for $b_i=\mathbf n_i\cdot\mathbf B$.

Solution

Each diagonal entry of the sum is four times $1/3$, giving $4/3$. For every off-diagonal entry, two axes contribute $+1/3$ and two contribute $-1/3$, so the sum vanishes. Therefore

$$\sum_i\mathbf n_i\mathbf n_i^{\mathsf T} = \frac43I.$$

Now

$$\sum_i b_i\mathbf n_i = \sum_i \mathbf n_i\mathbf n_i^{\mathsf T}\mathbf B = \frac43\mathbf B,$$

which gives

$$\mathbf B = \frac34\sum_i b_i\mathbf n_i.$$

Unequal projection uncertainties require a weighted least-squares version.

7. Statistical-polarization scaling

In the ideal half-space model, $B_{\mathrm{rms}}\propto\sqrt\rho\,d^{-3/2}$. By what factor does the RMS signal change if the spin density increases by four while the NV depth doubles?

Solution

The density change contributes $\sqrt4=2$, while the depth change contributes

$$2^{-3/2} = \frac{1}{2\sqrt2}.$$

The total factor is

$$2\times2^{-3/2} = \frac1{\sqrt2} \simeq0.707.$$

The higher spin density does not compensate for the increased depth.

8. Strain splitting near zero field

Neglect hyperfine interactions and take an axial field. In the $\{|+1\rangle,|-1\rangle\}$ subspace, show that transverse strain produces an avoided crossing. Determine the field derivative of the line separation and explain why assigning the splitting to $2\gamma_e|B_\parallel|$ fails near zero field.

Solution

The angular-frequency matrix is

$$\frac{H_{\pm1}}{\hbar}= \begin{pmatrix}D+\gamma_e B_\parallel&E\\E&D-\gamma_e B_\parallel\end{pmatrix}.$$

Its eigenvalues are $D\pm R$, where $R=\sqrt{(\gamma_e B_\parallel)^2+E^2}$. The angular-frequency separation is $2R$, with derivative

$$\frac{d(2R)}{dB_\parallel} =\frac{2\gamma_e^2 B_\parallel}{R}.$$

For $E\ne0$ the separation is $2|E|$ at zero field but its first-order field response vanishes. For $|\gamma_e B_\parallel|\gg|E|$, it approaches $2\gamma_e|B_\parallel|$. Splittings alone are even in the field and do not determine its sign; a known bias and branch tracking supply that additional information. Divide angular frequencies by $2\pi$ for hertz.

References

• V. M. Acosta et al., “Temperature dependence of the nitrogen-vacancy magnetic resonance in diamond,” Physical Review Letters 104, 070801 (2010).
• J. F. Barry et al., “Sensitivity optimization for NV-diamond magnetometry,” Reviews of Modern Physics 92, 015004 (2020).
• E. Bauch et al., “Ultralong dephasing times in solid-state spin ensembles via quantum control,” Physical Review X 8, 031025 (2018).
• J. M. Boss, K. S. Cujia, J. Zopes, and C. L. Degen, “Quantum sensing with arbitrary frequency resolution,” Science 356, 837–840 (2017).
• C. L. Degen, F. Reinhard, and P. Cappellaro, “Quantum sensing,” Reviews of Modern Physics 89, 035002 (2017).
• M. W. Doherty et al., “The nitrogen-vacancy colour centre in diamond,” Physics Reports 528, 1–45 (2013).
• H. J. Mamin et al., “Multipulse double-quantum magnetometry with near-surface nitrogen-vacancy centers,” Physical Review Letters 113, 030803 (2014).
• H. J. Mamin et al., “Nanoscale nuclear magnetic resonance with a nitrogen-vacancy spin sensor,” Science 339, 557–560 (2013).
• L. Rondin, J.-P. Tetienne, T. Hingant, J.-F. Roch, P. Maletinsky, and V. Jacques, “Magnetometry with nitrogen-vacancy defects in diamond,” Reports on Progress in Physics 77, 056503 (2014).
• J. M. Schloss, J. F. Barry, M. J. Turner, and R. L. Walsworth, “Simultaneous broadband vector magnetometry using solid-state spins,” Physical Review Applied 10, 034044 (2018).
• S. Schmitt et al., “Submillihertz magnetic spectroscopy performed with a nanoscale quantum sensor,” Science 356, 832–837 (2017).
• J.-P. Tetienne et al., “Spin relaxometry of single nitrogen-vacancy defects in diamond nanocrystals for magnetic noise sensing,” Physical Review B 87, 235436 (2013).