Can local polarization control in optical tweezer arrays enable site-resolved quantum operations?

Why one global polarization setting is not enough

In most optical tweezer arrays, a single laser beam is split into many traps, and position and intensity can be tuned site by site. Polarization is usually set once for the whole beam. That becomes a problem because the trapping light shifts transition frequencies in a way that depends on the angle between the light's polarization and the quantization axis, an effect described by vector and tensor light shifts.

The paper states the practical consequence directly: in arrays with a large spatial extent, the wide angular span of the beam produces site-to-site polarization distortions that global adjustments cannot undo. The authors also note that optical elements in the beam path can introduce parasitic polarization errors. Their earlier experience with wide-angle two-dimensional crossed-AOD arrays is cited as a case where intensity inhomogeneity was hard to push below 2%, partly because polarization errors were converted into intensity variations by downstream polarizing elements.

So the question the paper answers is not whether polarization matters, but whether it can be addressed per site rather than globally. The demonstrated answer is a scheme in which each trap receives its own linear polarization rotation, applied through the local birefringence of a spatial light modulator.

An SLM acts as a site-resolved phase retarder

The array is generated by an acousto-optic deflector driven with multiple radio frequencies, producing a one-dimensional set of 813-nm tweezers that are relayed to the atoms. A spatial light modulator with 1200×1920 pixels sits at an intermediate image plane downstream of the deflector. A quarter waveplate before the modulator turns the incoming linear polarization into an elliptical state; the modulator imprints a phase φ on each focused spot, and a second quarter waveplate converts that phase into a linear polarization rotation of φ/2.

The mask is divided into rectangular superpixels, each carrying a single phase value. In the atom plane the tweezers are spaced by 6.6 µm with a 1/e² waist of 0.8 µm, and the demagnification from the modulator to the atom plane is about 100. That maps to a spacing of 82 pixels and a waist of roughly 10 pixels at the modulator, so a superpixel width of 82 pixels corresponds to about eight waists. In separate tests, the authors reduced the independently programmable superpixel width to 30 pixels, about three waists and containing more than 99% of the tweezer beam power, and still found independent polarization control without appreciable degradation from crosstalk between neighboring pixels. They did not establish a strict minimum width.

This is the mechanism that makes the rest of the results possible: the rotation angle θ between each tweezer's polarization axis and the applied magnetic field of about 450 G can be set per site. The paper also describes an extension, not implemented here, in which a second modulator and a third quarter waveplate would allow independent control of ellipticity and principal-axis orientation at each site.

Calibrating polarization against an atomic transition

To connect applied voltage to actual polarization angle, the authors perform spectroscopy on the narrow ¹S₀↔³P₁ intercombination transition, whose natural linewidth is about 2π×7 kHz, using a single trapped atom. They measure the differential light shift ν as a function of pixel voltage and fit it with a form containing a scalar part and a tensor part proportional to (3cos²θ−1), with θ written as a polynomial in voltage. The fit assumes the atom sees linearly polarized light.

The result is that the modulator delivers a full rotation of θ over the range from 0 to π. This calibration is what turns the device into a usable knob: the angle that governs the tensor light shift can be read out from atomic spectroscopy rather than inferred from optics alone.

The authors also locate the tensor magic angle experimentally, finding that differential trap light shifts are nulled at θ = 49.2(4)°. At that angle the state-dependent differential light shift vanishes, which decouples the cooling resonance from site-to-site intensity non-uniformity. That decoupling is the key link between polarization control and the cooling result that follows.

Homogenizing light shifts across thirteen sites

Without correction and without a clean-up polarizing element, the frequency for optimal tweezer loading varied by more than 1 MHz across sites, because polarization inhomogeneity shifted each site's resonance differently. After applying in-situ corrections to θ for each tweezer, the optimal loading condition collapsed to a single global cooling-laser frequency across all traps.

The authors then probed the spread of ¹S₀↔³P₁ line centers with every site ideally at the magic angle. After one to three rounds of iterative correction, they typically reached a transition-frequency standard deviation of 10–20 kHz. At the roughly 0.5 mK trap depth, the derivative of the light shift with respect to θ at the magic angle is 5.95(8) kHz/mrad, so this spread corresponds to a linear polarization inhomogeneity of 2–3 mrad. That is comparable to the 3 mrad discrete rotation step set by the phase resolution of the modulator controller, which is a useful way to read the number: the residual spread is close to the programming granularity rather than far below it.

The technique is not limited to making every site identical. The authors programmed a staggered pattern in which odd-index tweezers sit at the magic angle for sideband cooling while even-index tweezers are shifted 0.4 MHz away. The two spectral resonances were fully non-overlapping, which is the ingredient that would let a subset of atoms be addressed selectively.

Cooling and transport inside polarization zones

With local polarization tuning, the authors performed sideband cooling on the ¹S₀↔³P₁ transition at the magic-angle condition. Three independent beams addressed the tweezers: two drove radial motional transitions and a third, delivered through the microscope objective, coupled to the axial mode. At a saturation parameter of about 60 and a Lamb-Dicke parameter of about 0.2, the carrier and red-sideband transitions were power broadened to roughly 8Γ and 2Γ respectively.

Cooling was applied simultaneously with fluorescence imaging, followed by additional cooling cycles, to mimic conditions where heating occurs mainly during imaging. Radial motional occupation was then measured by sideband asymmetry on the ultranarrow ¹S₀↔³P₀ clock transition. After 40 cycles, or 9.6 ms, of sideband cooling, atoms were recooled to a site-averaged radial occupation of 0.137(19). The authors report this as faster and lower in final occupation than Sisyphus cooling of strontium in tweezers with comparable trapping frequencies, while describing it as a proof-of-concept on a 13-site one-dimensional array.

The paper also sketches a zone architecture for a two-dimensional strontium processor: atoms are cooled in a cooling zone tuned near the magic angle, then transported by the same tweezers to a computing zone, while polarization rotation elsewhere spectrally detunes the ¹S₀↔³P₁ transition so those atoms are isolated from the cooling light. As a building block, the authors demonstrate transport between zones with orthogonal polarizations. With a smooth polarization transition zone they observe high-survival transport, and in a separate measurement they observe good coherence preservation for the clock qubit during transport, with a slight contrast reduction largely attributed to atom heating that a smoother transition zone could further improve.

Measuring and stabilizing polarization noise

Polarization noise matters because it converts into frequency noise through the same tensor light shift the technique exploits. The authors characterize it two ways: optical polarimetry and site-resolved ¹S₀↔³P₁ spectroscopy.

For the spectroscopic route, several effects broaden the carrier resonance, including probe-laser power broadening, the finite 120 µs pulse time, laser linewidth, and the natural linewidth Γ. The trap-depth-independent contributions are separated from the polarization contribution by extrapolating the full width at half maximum to zero probe power, then measuring that extrapolated width at three trap depths: 1×, 0.75×, and 0.5× the nominal 0.5 mK depth. Modeling the excess broadening as frequency shifts drawn from a Gaussian distribution whose width scales linearly with trap depth, the fit gives a standard deviation of 18(3) kHz at nominal depth. That corresponds to a polarization-angle standard deviation of 3.0(4) mrad, consistent with the integrated noise spectrum from polarimetry.

The authors then implement closed-loop stabilization referenced to the atomic transition frequency, which suppresses slow polarization-angle drifts to the few-mrad level on long timescales. The abstract describes this as reaching mrad-level stability. The distinction worth keeping is between the intrinsic noise floor measured spectroscopically and the residual drift after feedback: the feedback addresses slow drifts, and the reported stabilization is at the milliradian scale rather than below it.

What the demonstration does and does not establish

The evidence supports a specific claim: independent, site-resolved linear polarization rotation is experimentally feasible in a one-dimensional array, and it is sufficient to homogenize tensor light shifts, enable magic-angle sideband cooling, allow transport across polarization zones, and hold polarization noise near the milliradian level with atomic feedback. The array in the experiment is 13 sites, the transport and coherence results are described as feasibility demonstrations, and the two-dimensional zone architecture is presented as a proposal rather than a realized processor.

The numbers also define the practical ceiling of the current implementation. Residual polarization inhomogeneity of 2–3 mrad sits at the 3 mrad rotation step imposed by the modulator controller's phase resolution, so finer phase control, not the optical concept, is the near-term lever. The 30-pixel superpixel test suggests the scheme can be packed more densely than the 82-pixel spacing used for most measurements, but the authors explicitly did not establish a minimum superpixel width, and they defer scalability to larger arrays to their outlook.

The most decision-relevant comparison in the paper is therefore not against a competing polarization technique but against the global-polarization status quo. Global tuning has sufficed for small arrays, where one setting can reach a magic condition everywhere. The case for local control rests on the regime where it fails: arrays large enough that the beam's angular span produces site-to-site polarization distortion, and architectures that need different polarization in different regions at the same time. For a group whose array is small and uniform, the demonstrated benefit is a margin against parasitic errors; for a group building multi-zone or wide-angle arrays, the paper's staggered-pattern result, with two fully non-overlapping resonances separated by 0.4 MHz, is the result that matters most.

Sources used in this answer