Text from the interactive explanation
Photonic nanojets as emergent free-space power flux funnels
Mirza Karamehmedović, Cristian Placinta, Tobias Abilock Mikkelsen, Jesper Glückstad
Original paperPhotonic nanojets · a live wave experiment
Shine light on a glass bead a few wavelengths across and a slender, brilliant jet appears just behind it — narrower than the light’s own wavelength, in open space, with nothing around it. What holds it together?
Drag the bead in the picture to move it (or focus the picture and use the arrow keys).
λ = one wavelength of the light (532 nanometres for a green laser). Refractive index = how much a material slows light down, and so how strongly it bends it.
Live simulation A simplified 2D wave model computed in your browser (a long glass rod seen end-on, plane-wave light, field polarized along the rod). Illustrative — not the authors’ calculations. Amber tracers follow the flow of light’s energy.
Where does the jet go?
Try water (1.33), then sapphire (1.77). Watch the jet and the distance readout.
Does a square block make a jet?
Rays pass straight through flat faces, so ray optics predicts no focus at all. Switch to a square block — does the jet vanish?
Too small for rays
The textbook way to explain a focus is ray optics: straight lines of light that bend where they enter and leave glass (Snell’s law). Seen that way, the bead is a tiny magnifying glass and the rays cross behind it.
But rays have no width. They meet at points, so they cannot say how wide the jet’s neck is, or why it stays narrow instead of shrinking to nothing. And nanojets are mesoscopic — a few to a few tens of wavelengths long — a scale at which, the authors write, “geometric-optics constructions are invalid”.
Full computer simulations of light’s waves do reproduce nanojets faithfully. Yet they “offer limited physical insight: the jet appears in the computed field, but its origin, internal structure, and sensitivity to material and geometric parameters remain unclear.”
How light is usually held: walls and modes
Light that is squeezed narrow does not stay narrow. It spreads out — physicists call it diffraction — and the narrower the beam, the faster it fans out. Try it: make the beam thinner and watch its dashed edges open up.
To keep light narrow over a long distance, we normally build something around it. A graded-index optical fibre is glass whose refractive index is highest at the centre and falls toward the edges. It keeps bending light back toward the middle, the way a bowl keeps a marble at the bottom.
Light held like that can only settle into a few allowed shapes, called modes: one hump, two humps, three… Each one keeps its shape as it travels.
Illustration: textbook beam formulas (Gaussian and Hermite–Gaussian beams), computed live. Dashed amber lines mark the beam’s edges.
A nanojet keeps its narrow neck in open space — no glass around it, no wall, no fibre. So where is the bowl?
Live simulation The same 2D simulation as at the top, zoomed on the jet. Physical units assume 532-nm light.
Read the crests
A light wave is a pattern of crests and troughs sweeping forward. Where a point sits in that cycle is its phase, and a line joining neighbouring crests is a crest line (a wavefront). In open space, crests are one wavelength apart and, for a wide beam, straight.
The authors did something simple. Instead of staring at the bright jet, they looked at the crests around its neck. In every simulated jet they examined — square glass rods in 2D, a glass sphere in 3D — they found the same pattern.
Your prediction
As the probe moves through the neck (left to right), how will the bend of the crests change?
Drag the probe (or the white line) through the jet. Before the neck, the crests bow forward at their edges — they are converging, aimed at the axis. At the neck they are flat. After it they bow the other way and spread. The bend changes steadily along the jet and passes through zero right at the neck. The crests are also stretched: farther apart than in open space.
With silica glass the neck sits just behind the glass, so much of the converging half lies inside the bead. A lower-index droplet pushes the neck out into open space, where the whole hourglass can be read.
Near the neck, the phase “consistently exhibits a distinct hourglass-like structure”. Average mismatch between the formula and their full-wave simulations:
2D square glass rod (8 µm, silica, 532-nm light, field along the rod): 1.5 % — keff ≈ 0.887 k0, Ω = 100 µm−3
2D, other polarization , jet steered by 20°: 4.6 % — keff ≈ 0.975 k0 (mean over both 2D cases: 3.1 %)
3D glass sphere (radius 2.5 µm, n = 1.49, focused beam): 0.039 % in a narrow region near the waist — keff ≈ 0.931 k0, Ω ≈ 8.56 µm−3
Fig. 6: computed phase near the 3D jet’s waist with the keffz trend removed (left) and the funnel formula (right). The saddle is the hourglass: the bend flips sign along z. Axes in µm.
Energy flows across the crests
Light’s energy travels perpendicular to its crests — straight across them, like a swimmer cutting through waves. (In the paper’s terms: the time-averaged power flow points along the negative phase gradient.)
So hourglass crests make an hourglass of energy flow. Before the neck, energy pours in toward the axis; at the neck it runs straight along the jet; after it, it spreads out again. Nothing leaks sideways at the neck. The authors call the jet a power flux funnel.
The paper’s equations Computed from the 3D, axially symmetric version of the model (eqs. 5–12), for 532-nm light.
A bowl made of phase
Now the leap. The authors put the hourglass crest pattern into the wave equation that every light field must obey (the Helmholtz equation). What comes out, for the jet’s brightness, is the equation of a marble in a bowl — a harmonic oscillator, the same mathematics as a pendulum or the simplest quantum system.
Two numbers set the scene. The bowl’s steepness is Ω·keff: the faster the crests change their bend along the jet, the steeper the bowl and the tighter the light is squeezed. The budget is k0² − keff²: it grows the more the crests are stretched. A brightness shape — a mode — is held only if its level sits below the budget line. Otherwise it escapes, and there is no jet.
The allowed shapes are the same kind you met in the fibre. In a flat 2D slice they are exactly its humps (Hermite–Gaussian modes); in a round 3D jet seen end-on they become a single spot, a spot with one ring, two rings… (Laguerre–Gaussian modes). But here there is no glass. The bowl is drawn by the light’s own crests. In the paper’s words, this “quantization” is “not imposed by an external boundary condition”.
Challenge
Squeeze harder: drag Ω up. What happens to the jet? Then stretch the crests (lower keff). Can you hold it again?
The narrowest neck allowed
Because a mode is held only below the budget line, squeezing has a price: a tighter neck needs more stretched crests. Push that to the limit and a floor appears. The neck’s intensity half-width — the distance from the axis to where the brightness halves — can never be smaller than 0.187 λ0 / √(1 − (keff/k0)²).
Even with the most stretched crests that is at least 0.187 wavelengths — about 100 nm for green light. It holds whatever the bead’s shape, material or position: it needs only the phase funnel. It explains, the authors write, “the empirical observation that PNJ waists do not collapse arbitrarily below the diffraction scale” — and it matches the observation that jets get narrower at shorter wavelengths.
The paper’s equations Bound from eq. 12. Your point follows the sliders above; drag across the chart to change keff. ✕ marks the authors’ 3D sphere fit.
Try it elsewhere
Go back to the live simulation, change the bead, and read the crests again: the hourglass comes back with new numbers. In the simulation, narrower necks come with more stretched crests (smaller keff). Every time it settles after a change, the jet is logged here.
Live simulation 2D, illustrative. Jets whose neck touches the glass are not logged (their funnel can’t be read). The 3D bound above is not drawn here: the 2D version of the model has a different factor.
What the full simulations show
Everything in this section was computed by the authors with full-wave simulations: COMSOL Multiphysics for the 2D jet of Fig. 1, the Ceviche library with time reversal for Fig. 2, and generalized Lorenz–Mie theory (scattnlay) for the 3D glass sphere. In the paper’s axes, light travels toward −z.
Model versus computed jet
The ground mode of the free-space oscillator — a single bright spot — fitted to the computed brightness near the neck. Drag the divider to compare.
Colour: |Ex| normalized, 0.45 (black) to 1 (white). Vertical: z from −2.75 µm (top) to −4.0 µm (bottom). Mean relative error 3.52 %. Fit: keff = 11.03 µm−1 ≈ 0.934 k0, Ω = 6.70 µm−3.
Colour: |E| from 0.15 to 0.26 V/m. Vertical: z from −5.7 µm (top) to −6.5 µm (bottom). Mean relative error 3.53 %. Fit: keff = 9.84 ± 0.03 µm−1 ≈ 0.83 k0, Ω = 160 ± 6 µm−3.
The numbers
All values computed by the authors with full-wave simulations (paper, Sections 2 and 4), for 532-nm light (k0 = 11.81 µm−1). The phase fits test the funnel shape; the brightness fits aim for the best match to the jet’s brightness, so their parameters differ.
Back to the bead
What holds a nanojet together? Its own crests. Stretched along the beam and bent into an hourglass, they act as a bowl in empty space — and the light settles into it.
The authors call this a free-space oscillator: a jet held by the shape of its own phase rather than by any wall, fibre or cavity.
What it brings closer, in their framing: one geometry-independent principle in place of case-by-case explanations for particular beads; an explicit trade-off between how tight a jet can be and how stretched its crests are; a floor for the neck; and “design principles for nanojet-assisted imaging, lithography, and subwavelength field localization”. Nanojets are already used in label-free microscopy, for moving nanoparticles, and in super-resolution measurement.
Scope, as the paper states it: the model is local. It describes the field near the jet’s axis and near its neck, for the dominant polarization of the light. Its two numbers, Ω and keff, depend on the bead’s geometry and material — but the structure, an hourglass that acts as a harmonic oscillator, is robust.
Source & credits
Mirza Karamehmedović, Cristian Placinta, Tobias Abilock Mikkelsen, Jesper Glückstad. Photonic nanojets as emergent free-space power flux funnels. arXiv:2603.10831 [physics.optics], 11 March 2026 (preprint).
https://arxiv.org/abs/2603.10831 · DOI 10.48550/arXiv.2603.10831
Figures 1, 2, 4, 5, 6, 9, 10 from Karamehmedović, Placinta, Mikkelsen & Glückstad (2026), arXiv:2603.10831, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); cropped, rescaled and relabelled.
Live simulation Simplified 2D wave simulation computed in your browser: the scalar wave equation (exact for light polarized along a long rod), plane-wave illumination, 16 grid cells per wavelength, steady state read out over one optical period. Illustrative — not the authors’ calculations.
The paper’s equations Trap and limit tools computed from the paper’s equations (8)–(12) (3D axially symmetric reduction). The sphere preset uses the authors’ fitted values (Section 4); its neck and minimum widths are our arithmetic from those values. Fibre beams: textbook Gaussian and Hermite–Gaussian beam formulas (illustration).
Photonic nanojets were first observed experimentally in 2006 (Heifetz et al., Appl. Phys. Lett. 89, 221118, as cited in the paper).
Topics in this iPaper
Physics
Photonic nanojets (PNJs) - highly localized, high-intensity beams forming in the
Title page header and Section 1, page 1
Mathematics
yields the radial form of a 2D quantum harmonic oscillator equation
Section 3, page 10
photonic nanojets
Photonic nanojets (PNJs) - highly localized, high-intensity beams forming in the
Section 1, page 1
electromagnetic scattering
Consider the time-harmonic scattering of a Gaussian beam
Section 2 (3D case), page 4
helmholtz equation
based on a local amplitude–phase reduction of the Helmholtz equa-
Section 1, page 2
harmonic oscillator
yields the radial form of a 2D quantum harmonic oscillator equation
Section 3, page 10
lorenz mie theory
which realizes generalized Lorenz–Mie theory
Section 2 (3D case), page 5