Metasurfaces and Flat Optics

A metasurface welds two ideas from elsewhere in physics into optics that is essentially two-dimensional. The first is electromagnetic: at any interface the wavevector component parallel to the surface is conserved, and Snell's law is nothing more than that bookkeeping. The second is geometric: a beam carries a phase that can be advanced not only by travelling farther but by cyclically transforming its polarization — the Pancharatnam–Berry phase, a cousin of Berry's quantum adiabatic phase. Put an engineered, position-dependent phase jump on an interface and the momentum bookkeeping sends light off anew; wire that phase to a rotated birefringent scatterer and it becomes broadband. A conventional lens shapes a wavefront the slow way, letting rays accumulate different optical path lengths $\int n\,\mathrm{d}\ell$ through a curved slab of glass. A metasurface throws that away: it tiles a flat wafer with subwavelength meta-atoms — each a dielectric nanopillar or plasmonic antenna tens to hundreds of nanometres across — whose scattered field carries a phase and amplitude fixed by geometry. Varying the design cell to cell writes an almost arbitrary $t(x,y)=A(x,y)\,e^{i\Phi(x,y)}$ onto a layer far thinner than a wavelength. The founding result, the generalized Snell's law of Yu and Capasso (2011), says a phase gradient $\mathrm{d}\Phi/\mathrm{d}x$ adds an in-plane wavevector, bending light anomalously. Choose $\Phi(x,y)$ to be the hyperbolic profile of an ideal lens and you get a flat metalens; a ramp gives a single-order beam deflector; a spiral gives an optical vortex. The recurring tension — the strong chromatic focal shift from fixing $\Phi$ at one wavelength — is the same dispersion trade-off that haunts guided-wave photonics, now pushed onto a plane.

Key takeaways

A metasurface is a planar array of subwavelength scatterers (meta-atoms) whose geometry, orientation, or resonance imprints a chosen phase, amplitude, and polarization onto light within a thickness far below one wavelength.
The governing law is the generalized Snell's law: a position-dependent phase $\Phi(x,y)$ contributes a tangential wavevector $\nabla_{\parallel}\Phi$ , so $n_t\sin\theta_t-n_i\sin\theta_i=(\lambda_0/2\pi)\,\mathrm{d}\Phi/\mathrm{d}x$ and light refracts anomalously.
Full $0$ to $2\pi$ coverage comes from truncated-waveguide meta-atoms (tune the diameter) or the geometric Pancharatnam–Berry phase $\Phi=2\sigma\theta$ , set purely by rotating each half-wave element by $\theta$ .
A flat metalens synthesizes the hyperbolic phase $\Phi(r)=-(2\pi/\lambda_d)(\sqrt{r^2+f^2}-f)$ , replacing a curved surface with a patterned wafer that focuses without spherical aberration; subwavelength periodicity keeps it single-order and efficient.
Because $\Phi$ is fixed at one wavelength while the focusing phase scales as $1/\lambda$ , an ordinary metalens has strong chromatic focal shift $f(\lambda)\approx f_d\lambda_d/\lambda$ — the problem achromatic designs attack.
Key idea: a metasurface is a spatially programmed phase mask — engineer the local scattering phase $\Phi(x,y)$ of subwavelength meta-atoms and the generalized Snell's law $n_t\sin\theta_t-n_i\sin\theta_i=(\lambda_0/2\pi)\nabla_{\parallel}\Phi$ turns that pattern into arbitrary wavefront control in a flat, ultrathin device.

Where this lives today

The most visible payoff is the metalens itself. Near-infrared metalenses, fabricated on standard semiconductor lines, already ship inside consumer devices for face-recognition and depth-sensing cameras, where the flat, wafer-scale, lightweight form beats a stack of moulded plastic elements. Research metalenses focus everything from visible light for endoscopes to X-rays, and dispersion-engineered achromatic designs now hold a common focus across the whole visible band — the hard problem Example 2 exposes. Reconfigurable metasurfaces on liquid crystals, phase-change materials, or MEMS actuators steer beams electrically, the technology behind solid-state LiDAR with no moving parts. Beyond imaging, the same phase-mask idea drives structured light and quantum optics: metasurface holograms encode full images into a film, vortex generators stamp an $e^{i\ell\varphi}$ spiral phase for light carrying orbital angular momentum, and single metasurfaces have generated and sorted entangled photon states in one flat element. Each is the generalized Snell's law of 2011 pushed to a spatially varying $\Phi(x,y)$ — the promise that if you can specify a wavefront, you can etch it.

The physical picture

Think of a metasurface as a pixelated wavefront stamp. Each pixel — one meta-atom — delays or advances the local phase by a chosen amount and, if desired, rotates the polarization. Because the pixels are packed more finely than a wavelength, the propagating field cannot resolve them individually; it sees a smooth phase surface, not a row of teeth. That is why a single clean beam emerges instead of a fan of grating orders — the property that separates a metasurface from an ordinary diffractive plate. Anomalous refraction is then just momentum bookkeeping. A flat interface carrying a phase ramp $\Phi=\xi x$ hands each photon a fixed transverse kick $\hbar\xi$ ; since Snell's law conserves the tangential wavevector, adding $\xi$ shifts the outgoing angle, and a lens is the case where $\xi$ grows with radius fast enough to steer every ray onto one focus. The first figure makes the ramp concrete with a real deflection angle.

Figure

Anomalous refraction from a phase ramp: a metasurface whose imposed phase advances by 2π over a period Λ_Φ = 2λ₀ hands the beam a transverse momentum kick and steers normally-incident light to θ_t = 30° in a single dominant order, whereas a coarse grating of the same period would smear the power across many orders

The Pancharatnam–Berry route reaches the same phase map geometrically: rotate identical half-wave nano-elements, and circularly polarized light passing each flips helicity and picks up a phase $\Phi=2\sigma\theta$ , twice the orientation angle. Orientation alone writes the phase — size and resonance do not enter — so the same law holds at every frequency, intrinsically broadband (only the half-wave efficiency is dispersive). The second figure shows the payoff: bending the ramp into a hyperbola turns the wafer into a lens.

Figure

A flat metalens: meta-atoms of graded design synthesize the hyperbolic phase Φ(r) = −(2π/λ)(√(r²+f²) − f) so that every ray leaves the flat aperture and arrives at the focus in phase — for D = 200 μm and f = 100 μm this gives NA ≈ 0.71 and a diffraction-limited spot w₀ ≈ 0.38 μm at λ = 532 nm

The core equations

The founding statement of the field is the generalized law of refraction. Besides the ordinary optical path, a ray crossing a metasurface picks up the local phase jump $\Phi(x)$ ; requiring the total phase of neighbouring rays to be stationary is Fermat's principle with an extra term, so the phase gradient acts as an added tangential wavevector.

n_t\sin\theta_t-n_i\sin\theta_i=\frac{\lambda_0}{2\pi}\,\frac{\mathrm{d}\Phi}{\mathrm{d}x} \qquad (1)

where: $n_i,n_t$ — refractive indices of the incidence and transmission media; $\theta_i,\theta_t$ — incidence and refraction angles from the normal; $\lambda_0$ — free-space wavelength; $\mathrm{d}\Phi/\mathrm{d}x$ — gradient of the imposed interfacial phase. Reading: with zero gradient this is ordinary Snell's law; a nonzero gradient is a transverse momentum source that steers the beam. The same matching, applied to the ray that stays in the incidence medium, gives an anomalous law of reflection. Its striking consequence is that the reflection angle need not equal the incidence angle — a beam can be reflected off-axis even at normal incidence.

\sin\theta_r-\sin\theta_i=\frac{\lambda_0}{2\pi n_i}\,\frac{\mathrm{d}\Phi}{\mathrm{d}x} \qquad (2)

where: $\theta_r$ — reflection angle, generally not equal to $\theta_i$ ; $\theta_i$ — incidence angle; $n_i$ — index of the incidence medium. Reading: the imposed gradient breaks the mirror symmetry of ordinary reflection — the basis of flat reflect-arrays. To build a lens we do not pick a gradient by hand; we demand equal optical path from every aperture point to a single focus. The phase the surface must impose is the hyperbolic profile below, the exact aberration-free prescription at the design wavelength.

\Phi(r)=-\frac{2\pi}{\lambda_d}\left(\sqrt{r^2+f^2}-f\right) \qquad (3)

where: $r$ — radial coordinate from the lens centre; $f$ — design focal length; $\lambda_d$ — design wavelength; $\Phi(r)$ — required phase at radius $r$ , wrapped modulo $2\pi$ in fabrication. Reading: the surface removes exactly the extra propagation phase off-axis rays accumulate on their way to $F$ . Spanning a full $0$ to $2\pi$ without any resonance is the job of the Pancharatnam–Berry phase. Rotating a half-wave meta-atom by an in-plane angle $\theta$ converts the incident circular polarization to the opposite handedness and stamps a purely geometric phase on it, as the Jones-matrix form on the right shows.

\Phi_{\mathrm{PB}}=2\sigma\theta,\qquad t=\cos\!\Big(\tfrac{\delta}{2}\Big)\hat{\sigma}+i\sin\!\Big(\tfrac{\delta}{2}\Big)e^{i2\sigma\theta}\hat{\sigma}^{*} \qquad (4)

where: $\sigma=\pm1$ — incident circular helicity; $\theta$ — in-plane rotation angle of the meta-atom; $\delta$ — retardance ( $\delta=\pi$ for a half-wave element); $\hat{\sigma},\hat{\sigma}^{*}$ — co- and cross-handed circular basis vectors. Reading: the cross-handed term carries a phase set only by orientation, the same at every frequency to first order — the origin of broadband metasurfaces. Everything above assumes the array acts as a single-order element. That holds only if the meta-atom lattice is subwavelength — the period must be small enough that no diffraction order but the designed one propagates over the working angular range.

\Lambda<\frac{\lambda_0}{n_t+n_i\sin\theta_{i,\max}} \qquad (5)

where: $\Lambda$ — meta-atom lattice period; $\theta_{i,\max}$ — largest incidence angle handled. Reading: violate this and the device leaks power into spurious orders like a coarse grating; satisfy it and one designed beam dominates. Once built, a metalens's resolving power follows from the marginal-ray geometry. The numerical aperture is the sine of the steepest focused ray, and the focal spot is the diffraction-limited Airy scale.

\mathrm{NA}=\sin\theta_{\max}=\frac{D/2}{\sqrt{(D/2)^2+f^2}},\qquad w_0\approx\frac{0.51\,\lambda}{\mathrm{NA}} \qquad (6)

where: $D$ — aperture diameter; $\theta_{\max}$ — marginal-ray angle to the focus; $w_0$ — focal-spot FWHM (Airy scale). Reading: shorter focal length at fixed aperture raises NA and shrinks the spot to the diffraction floor. Finally, there is a price for writing $\Phi$ at one wavelength. The etched profile is frozen, but the focusing phase a metalens needs scales as $1/\lambda$ , so the focus walks with colour — the defining defect of a simple metalens.

f(\lambda)\approx f_d\,\frac{\lambda_d}{\lambda} \qquad (7)

where: $f_d$ — focal length at the design wavelength $\lambda_d$ ; $f(\lambda)$ — focal length at operating wavelength $\lambda$ . Reading: shorter wavelengths focus farther and longer ones nearer, so an achromatic design must add tailored group delay, not just the right phase.

Derivation AGeneralized Snell's law from tangential-wavevector matching

The goal is to show that the generalized law of refraction is ordinary Fermat's principle with one added term, so a metasurface's phase gradient sits on exactly the same footing as an added tangential wavevector. We compare two rays crossing the interface an infinitesimal distance $\mathrm{d}x$ apart, require their total phase to be stationary, and read off the refraction angle — everything reducing to conservation of the wavevector component along the surface.

Step 1. Set the total phase of two neighbouring rays equal. Ray one crosses at $x$ with $n_i k_0\,\mathrm{d}x\sin\theta_i+\Phi(x)$ ; ray two crosses at $x+\mathrm{d}x$ with $n_t k_0\,\mathrm{d}x\sin\theta_t+\Phi(x+\mathrm{d}x)$ , where $k_0=2\pi/\lambda_0$ . Why: Fermat's principle picks the ray whose phase is stationary against small transverse shifts, so equating neighbouring paths is the whole content of refraction.

Step 2. Taylor-expand $\Phi(x+\mathrm{d}x)=\Phi(x)+(\mathrm{d}\Phi/\mathrm{d}x)\,\mathrm{d}x$ , cancel $\Phi(x)$ , divide by $\mathrm{d}x$ , then by $k_0$ : $n_t\sin\theta_t-n_i\sin\theta_i=(\lambda_0/2\pi)\,\mathrm{d}\Phi/\mathrm{d}x$ . Why: the derivative of the imposed phase is exactly the extra tangential wavevector the surface injects — equation (1), a controllable transverse-momentum source that returns plain Snell's law at zero gradient.

Step 3. Sanity-check against a grating. A period $\Lambda_\Phi$ over which $\Phi$ advances by $2\pi$ gives $\sin\theta_t=\lambda_0/\Lambda_\Phi$ at normal incidence — the form of the first-order grating equation $\sin\theta=m\lambda/\Lambda$ , but routed into one order. Why: matching a known result in the right limit checks the derivation and its dimensions.

Go deeper · Derivation B · 1 minWhy the hyperbolic profile focuses to a point

Derivation BWhy the hyperbolic profile focuses to a point

The goal is to derive the metalens phase from first principles rather than borrow it: requiring equal optical path from every aperture point to a common focus forces exactly the hyperbolic profile of equation (3), aberration-free by construction at the design wavelength. We set up the path length from radius $r$ to the on-axis focus, demand the imposed phase plus the free-space propagation phase be constant across the aperture, solve for $\Phi(r)$ , and expand it to recover the thin-lens quadratic as a check.

Step 1. The distance from a surface point at radius $r$ to the axial focus $F$ , a distance $f$ behind the lens, is $L(r)=\sqrt{r^2+f^2}$ by Pythagoras. Why: a lens works only if all partial waves reach $F$ in step, so the path each travels is the first thing to pin down.

Step 2. Demand a flat total phase across the aperture, $\Phi(r)+k_0 L(r)=\Phi(0)+k_0 f$ (on-axis ray as reference), and solve: $\Phi(r)=-k_0\big(\sqrt{r^2+f^2}-f\big)=-(2\pi/\lambda_d)\big(\sqrt{r^2+f^2}-f\big)$ . Why: constant total phase is exactly the condition for a perfectly converging wave, so the surface must subtract the off-axis rays' extra propagation phase — no residual aberration at $\lambda_d$ .

Step 3. Expand for $r\ll f$ : $\sqrt{r^2+f^2}\approx f+r^2/2f$ , giving $\Phi(r)\approx-\pi r^2/\lambda_d f$ . Why: this paraxial quadratic is the standard thin-lens phase, so the exact hyperbola reduces to a textbook converging lens near the axis — the limit that identifies $f$ as a focal length.

Limiting cases

The generalized law and the meta-atom picture reduce to familiar regimes:

Case	Result	Meaning
No phase gradient	$\nabla_{\parallel}\Phi=0\Rightarrow n_t\sin\theta_t=n_i\sin\theta_i$	Ordinary Snell's law returns; a uniform metasurface is a flat window.
Period much larger than $\lambda$	$\Lambda\gg\lambda\Rightarrow$ many propagating orders	The device becomes a conventional grating; single-order operation needs $\Lambda<\lambda/(n_i+n_t)$ roughly.
Paraxial metalens	$\Phi(r)\approx-\pi r^2/\lambda f$	The hyperbola for $r\ll f$ gives the thin-lens phase; an ideal paraxial lens of focal length $f$ .
Geometric-phase element	$\Phi_{\mathrm{PB}}=2\sigma\theta,\ \sigma=\pm1$	Converted light acquires a phase fixed by orientation alone, independent of resonant details.
Design wavelength	$f(\lambda_d)=f_d$ , aberration-free	At $\lambda_d$ the frozen profile is exactly right; away from it the focus shifts as $f_d\lambda_d/\lambda$ .

Common trapA metasurface is just a very thin diffraction grating

It is not, and the difference is the whole point. A blazed grating has a period comparable to or larger than the wavelength, so it supports several propagating orders and leaks power into all of them; its efficiency into any one order is fundamentally limited. A metasurface uses a subwavelength lattice that cuts off every order but the designed one — which is precisely why it can route almost all the power into a single beam. The subtler half of the misconception: because $\Phi(x,y)$ is fixed at one design wavelength while the focusing phase a lens needs scales as $1/\lambda$ , a metasurface is not achromatic — the same pattern deflects different colours to different angles and focuses them at different distances. Treating the etched profile as a fixed blaze that works at all wavelengths is exactly what makes a metalens's chromatic aberration a surprise.

Worked examples

1 · Designing a visible metalens (the concrete build)

You want a metalens for green light, $\lambda_d=532$ nm, with aperture $D=200\ \mu$ m and focal length $f=100\ \mu$ m. Find its numerical aperture, its diffraction-limited focal spot, and the meta-atom period demanded at the very edge of the lens.

Step 1. Numerical aperture from the marginal-ray geometry (equation 6): $\mathrm{NA}=\dfrac{D/2}{\sqrt{(D/2)^2+f^2}}=\dfrac{100}{\sqrt{100^2+100^2}}=\dfrac{1}{\sqrt2}=\mathbf{0.71}$ . Aperture and focal length are equal, so the marginal ray leaves the edge at $45^\circ$ .

Step 2. Focal spot: $w_0\approx\dfrac{0.51\,\lambda}{\mathrm{NA}}=\dfrac{0.51\times532\ \mathrm{nm}}{0.7071}=\mathbf{384\ nm}\approx0.38\ \mu$ m. A high-NA metalens focuses green light to a sub-wavelength, diffraction-limited spot.

Step 3. Edge period. At the rim the hyperbolic phase gradient is $|\mathrm{d}\Phi/\mathrm{d}r|=(2\pi/\lambda)\,\mathrm{NA}$ , so the phase wraps by $2\pi$ over $\Lambda_\Phi=\lambda/\mathrm{NA}=532/0.7071=\mathbf{752\ nm}$ . Sampling that ramp needs a lattice several times finer — a $250$ nm unit cell packs only about three pillars into the outermost zone, the hardest region to fabricate. NA = 0.71, w₀ ≈ 0.38 μm, outer zones wrap every ≈ 0.75 μm.

2 · Chromatic focal shift of that same lens

Illuminate the metalens of Example 1 (designed at $\lambda_d=532$ nm, $f_d=100\ \mu$ m) with a red HeNe laser at $\lambda=633$ nm instead. Where does it focus, and is the shift serious?

Step 1. The etched phase is frozen while the required focusing phase scales as $1/\lambda$ , so the focus moves as $f(\lambda)\approx f_d\,\lambda_d/\lambda$ (equation 7): $f(633)=100\ \mu\mathrm{m}\times\dfrac{532}{633}=\mathbf{84.0\ \mu m}$ . The red focus sits $16\ \mu$ m closer to the lens than the green one.

Step 2. Compare it to the depth of focus, $\mathrm{DOF}\approx\lambda/\mathrm{NA}^2=532\ \mathrm{nm}/0.5\approx1.06\ \mu$ m: the shift is about $15\times$ the DOF, so at red the green focal plane is wildly out of focus.

Step 3. The lesson: a simple metalens is strongly chromatic. An achromatic design fixes this with meta-atoms that supply both the correct phase and a tailored group delay $\mathrm{d}\Phi/\mathrm{d}\omega$ across the aperture — the spatial analogue of fibre dispersion compensation. f(633 nm) = 84.0 μm, a 16 % focal shift and ≈ 15× the depth of focus.

Physics of the portrait

Flat optics has one unmistakable modern architect: the group that turned the generalized Snell's law from a boundary-condition exercise into metalenses you can hold up to the light.

Origins · 1 minFederico Capasso and how this was discovered

Federico Capasso

born 1949 · flat optics and the generalized laws of refraction

Federico Capasso is an Italian-American applied physicist who co-invented the quantum cascade laser at Bell Labs in 1994 — itself a triumph of engineering light by design. At Harvard from 2003 he turned that band-structure instinct onto the wavefront. In 2011, with Nanfang Yu and coworkers, he published the generalized laws of reflection and refraction, showing that an interface of subwavelength phased antennas hands light an extra tangential wavevector and bends it anomalously — the paper that launched the modern metasurface field. His group then built the pieces that made flat optics real: high-efficiency dielectric metasurfaces, polarization optics realizing arbitrary Jones matrices, and in 2016 (with Mohammadreza Khorasaninejad) a titanium-dioxide metalens focusing visible light to a diffraction-limited spot at usable efficiency. His throughline is a refusal to accept the shapes nature hands you: if a lens needs a hyperbolic phase, do not grind glass — write the phase, pixel by subwavelength pixel, on a flat wafer.

History

1956

— Pancharatnam identifies the geometric phase acquired by polarized light cycled through a sequence of states

1984

— Berry derives the geometric phase of a cyclically and adiabatically varied quantum system

2002

— Bomzon, Kleiner and Hasman realize Pancharatnam–Berry phase optical elements from subwavelength gratings

2011

— Yu, Capasso and coworkers announce the generalized laws of reflection and refraction

2016

— Khorasaninejad and Capasso demonstrate a diffraction-limited visible-wavelength metalens

Going further

Texts. Hecht, Optics, 5th ed., Ch. 4 (refraction, wavefronts) and Ch. 10 (diffraction) · Saleh and Teich, Fundamentals of Photonics, 3rd ed., Ch. 2 (ray optics), Ch. 4 (Fourier optics), Ch. 5 (electromagnetic optics) · Yu and Capasso, "Flat optics with designer metasurfaces," Nature Materials 13, 139 (2014) · Yu et al., "Light propagation with phase discontinuities: generalized laws of reflection and refraction," Science 334, 333 (2011) · Khorasaninejad and Capasso, "Metalenses: versatile multifunctional photonic components," Science 358, eaam8100 (2017) · Loudon, The Quantum Theory of Light, 3rd ed. (scattering and polarization formalism).

Provenance. A verifier should reconfirm: (1) the generalized laws of refraction and reflection by tangential-wavevector conservation, following Yu et al., Science 334, 333 (2011); (2) the hyperbolic metalens phase from the equal-optical-path condition, paraxial limit $-\pi r^2/\lambda_d f$ matching the thin-lens phase in Saleh and Teich; (3) the Jones-calculus Pancharatnam–Berry phase $2\sigma\theta$ for a half-wave element; (4) the numerical checks by substitution — deflector $\Lambda_\Phi=2\lambda_0\Rightarrow\theta_t=30^\circ$ ; metalens $D=200\ \mu$ m, $f=100\ \mu$ m give $\mathrm{NA}=1/\sqrt2\approx0.71$ and $w_0\approx0.38\ \mu$ m at $532$ nm; chromatic shift $f(633)=100\times532/633=84.0\ \mu$ m; (5) the grating cross-check $\sin\theta=m\lambda/\Lambda$ from Hecht. No fabricated material constants are used; the amorphous-silicon and titanium-dioxide references are illustrative of common visible-band meta-atom materials.

Prerequisites. Numerical Aperture and Acceptance Angle · Waveguide Modes and the Wave Equation · Total Internal Reflection and Optical Guiding. Leads to. Integrated Photonic Waveguides and Circuits · Chromatic and Modal Dispersion in Fibres · Dispersion Management and Compensation.

Ask the tutor: Ask me to design a metalens for a wavelength and NA you choose, and I will walk through the phase profile, meta-atom library, and the chromatic budget step by step.

How is a metasurface different from an ordinary diffraction grating?

The lattice period separates them. A grating has features comparable to or larger than the wavelength, so several diffraction orders propagate and share the power. A metasurface uses a subwavelength lattice, making every order but the designed one evanescent, so a single beam carries almost all the power. Just as important, a metasurface sets phase, amplitude, and polarization independently at each pixel, whereas a grating repeats one profile — which is why the same platform builds lenses, holograms, and polarization optics, not just deflectors.

Why does a flat metalens have such strong chromatic aberration?

Because the phase it imposes is etched into geometry and cannot change with colour, while the phase a lens needs scales as $1/\lambda$ . The profile is exactly right at one design wavelength; at any other it supplies the wrong phase, and the focus moves as $f(\lambda)\approx f_d\lambda_d/\lambda$ , as Worked Example 2 shows. Achromatic metalenses fix this by adding a controlled wavelength dependence — meta-atoms chosen to deliver the right phase and the right group delay together — so the focal length stays put across a band.

What exactly is a meta-atom, and how does it set the phase?

A meta-atom is one subwavelength scatterer, the pixel of the surface, and it fixes the local phase in one of two dominant ways. The first is the truncated-waveguide picture: a high-index dielectric pillar acts as a short multimode guide, so its diameter tunes the effective index and hence the propagation phase $\Phi=k_0 n_{\mathrm{eff}}(d)h$ , covering $0$ to $2\pi$ at fixed height. The second is geometric: identical half-wave elements are rotated, and the Pancharatnam–Berry phase $2\sigma\theta$ follows from orientation alone. Resonant plasmonic antennas use the Lorentzian phase swing near a resonance but are lossier.

Why must the meta-atoms be smaller than a wavelength?

Subwavelength spacing is what lets the array act as a single optical element rather than a grating. If the period exceeds roughly $\lambda/(n_i+n_t)$ , extra diffraction orders become propagating waves and steal power from the designed beam. Below that threshold those orders are evanescent — they decay within a wavelength of the surface and carry no energy away — so only the intended wavefront survives into the far field. Physically, the propagating field cannot resolve the individual pixels; it responds to the smooth effective phase they collectively impose, which is what turns a nanostructured film into a high-efficiency lens.

Does a metalens really beat a conventional lens?

It wins on some axes and not others. A metalens is flat, micron-thin, lightweight, and mass-produced on a semiconductor wafer alongside detectors — decisive for compact cameras, endoscopes, and depth sensors — and is free of spherical aberration at its design wavelength. The costs are real: efficiency below a good refractive lens, sensitivity to nanometre fabrication error, and strong chromatic aberration in a simple design. For narrowband jobs a metalens is already competitive; for broadband white-light imaging it needs an achromatic design, and conventional glass still often wins.

How does the Pancharatnam–Berry phase relate to Berry's phase in quantum mechanics?

They are the same idea in different clothing. Berry's phase is the extra phase a quantum state gains when its parameters are carried adiabatically around a closed loop — set by the geometry of the path, not the dynamics. Pancharatnam found the polarization version first, in 1956: cycle a beam's polarization around a loop on the Poincaré sphere and it gains a phase equal to half the enclosed solid angle. Rotating a half-wave meta-atom traces exactly such a loop for the converted circular component, so the phase is $2\sigma\theta$ , orientation-only. Being geometric rather than dynamical, it carries no intrinsic frequency dependence — the root of broadband performance.