Photonic Lantern: Principles & Applications
- Photonic lanterns are adiabatic devices that convert multimode fields into discrete single-mode channels for efficient light interfacing.
- They employ gradual tapering and precise mode coupling to maintain signal integrity, supporting applications in spectroscopy, imaging, and wavefront sensing.
- Various architectures—fiber-fused, multicore, and laser-inscribed—offer trade-offs in coupling efficiency, bandwidth, and integration versatility.
A photonic lantern is an adiabatic waveguide device that converts a multimode waveguide or fiber into a discrete set of single-mode waveguides, or reciprocally many single-mode channels into one multimode channel, with low loss when the multimode modal content is matched to the number of single-mode outputs. Its practical significance lies in combining multimode collection, which is tolerant of aberrated or seeing-limited input fields, with single-mode delivery, which is compatible with stable spectroscopy, photonic filtering, interferometric processing, and compact integrated optics. In astronomy, photonic lanterns have progressed from enabling multimode-to-single-mode interfacing to on-sky visible and near-infrared operation for spectroscopy, focal-plane wavefront sensing, and subdiffraction-limited measurements; related work extends the same device class to coherent imaging, free-space optical communications, microendoscopy, and optical metrology (Leon-Saval et al., 2015, Birks et al., 2015, Vievard et al., 2024, Sengupta et al., 25 Nov 2025).
1. Fundamental operating principle
The defining feature of a photonic lantern is adiabatic mode conversion. Several isolated single-mode cores are brought together gradually enough that their supermodes evolve continuously into the guided modes of a common multimode region, and the process is reciprocal in the reverse direction. In standard fiber implementations, several single-mode fibers are embedded in a low-index jacket or capillary and tapered until the individual cores cease to guide independently; the collective structure then forms the multimode core. Reviews of the concept emphasize that low loss requires matching the number of available guided degrees of freedom on the multimode side to the number of single-mode channels on the other side, because an arbitrary multimode field cannot be compressed losslessly into fewer modes without violating brightness constraints (Leon-Saval et al., 2015, Birks et al., 2015).
This mode-transformer interpretation is more precise than treating the device as a simple power splitter. A conventional lantern is generally not a true mode sorter: launching a field into the multimode side does not normally map each multimode state to one unique output core. Rather, the multimode input decomposes into a superposition across the single-mode outputs, and in many applications the precise core-to-mode correspondence is unimportant so long as the transition is low loss. That property is central to the historical role of lanterns as low-loss interfaces between multimode and single-mode photonic systems (Birks et al., 2015).
The wavelength dependence of lantern behavior follows from ordinary multimode waveguide scaling. For a step-index multimode core, the approximate number of supported modes scales as
so a fixed-core-count lantern is only perfectly mode matched over a restricted spectral range. The 2015 review reports demonstrated lanterns with bandwidths around $300$ nm and minimum transmission at the edges, illustrating that practical broadband operation is possible but not unlimited (Horton et al., 2014, Leon-Saval et al., 2015).
2. Device architectures and fabrication regimes
Several lantern architectures have been demonstrated, differing mainly in how the single-mode channels are realized and how the adiabatic transition is produced.
| Variant | Realization | Reported characteristics |
|---|---|---|
| Fiber-fused bundle | Separate SMFs inserted into a low-index capillary and tapered | First type #2 lantern: , loss per transition about $0.3$ dB (Birks et al., 2015) |
| Multicore-fiber lantern | Purpose-made multicore fiber tapered inside an F-doped capillary | Example with $120$ single-mode cores and losses below $0.5$ dB (Birks et al., 2015) |
| Ultrafast-laser-inscribed lantern | 3D waveguides written in glass and merged on chip | First MM-SM-MM pair had insertion loss $5.7$ dB; later pairs about $2.1$ dB (Birks et al., 2015) |
| Double-clad mode-selective lantern | Three double-clad fibers with differing first-cladding diameters inside a low-index capillary | Modal isolation above $60$ dB and excess loss lower than $300$0 dB over more than $300$1 nm with fluoride-doped capillary (Becerra-Deana et al., 2024) |
The principal fiber-based geometries remain the fused bundle and the multicore-fiber lantern. The fused-bundle approach is practical for astronomical instrumentation because connectorization and splicing are straightforward, whereas the multicore approach scales more naturally to high port counts and enables operations such as writing identical fiber Bragg gratings into many cores simultaneously (Leon-Saval et al., 2015, Birks et al., 2015). The ultrafast-laser-inscribed route replaces fiber tapering with monolithic 3D photonic writing, offering arbitrary routing and direct integration with slit reformats or beam combiners, albeit historically at higher propagation loss than fiber lanterns (Birks et al., 2015).
Integrated astronomical lanterns fabricated by femtosecond laser direct-write provide a useful intermediate case. Devices designed for operation around $300$2 and based on a circular array of $300$3 waveguides showed coupling and transition losses of less than $300$4 for $300$5 injection and total throughput, including substrate absorption, as high as $300$6–$300$7 (1311.0578). The same study identified an optimum multimode pitch near $300$8 and emphasized the role of a cosine taper in approaching simulated normalized throughput of about $300$9 for a single transition (1311.0578).
Mode-selective lanterns introduce a distinct design objective: one single-mode input should map predominantly to one guided mode of the few-mode or multimode end. The double-clad-fiber implementation reported in 2024 achieved symmetry breaking by varying first-cladding diameters while preserving telecom-compatible single-mode cores, allowing shorter and more robust mode-selective lanterns and enabling real-time characterization during tapering. The best devices used fluoride-doped silica capillaries and maintained modal isolation above 0 dB with excess loss below 1 dB across more than 2 nm; lower-cost synthetic fused silica capillaries still delivered modal isolation above 3 dB with excess loss lower than 4 dB over the same bandwidth (Becerra-Deana et al., 2024).
3. Coupling behavior, throughput, and system design
Lantern acceptance differs qualitatively from that of a conventional step-index multimode fiber. Rather than a hard numerical-aperture cutoff with approximately flat transmission up to the edge, a photonic lantern exhibits a smooth angular response that is better captured empirically by a Generalised Squared Logistic Function. Laboratory measurements on 5-core and 6-core devices showed that this smooth angular dependence matters directly for instrument design, including the optimum field of view per microlens and the thermal-background budget of systems such as PRAXIS. In that case, the selected field of view per microlens was about 7 for 8 and about 9 for 0 (Horton et al., 2014).
A useful first-order description is an effective lantern numerical aperture derived from multimode mode-count scaling. In practice, however, the acceptance must be measured because lanterns occupy an intermediate regime between single-mode and highly multimoded behavior. That intermediate regime is precisely why they are attractive for adaptive-optics-fed instruments: the telescope point-spread function can be collected with higher tolerance than for a lone single-mode fiber, while the outputs remain single-mode and therefore compatible with diffraction-limited downstream optics (Horton et al., 2014, Vievard et al., 2024).
The visible Subaru/SCExAO lantern provides a concrete example of these trade-offs. The device has one multimode input and 1 single-mode outputs; its hardware transmission, measured by retro-injecting a 2 nm laser, averaged
3
with a range from 4 to 5. Coupling efficiency was normalized as
6
where 7 is measured flux after the lantern, 8 is flux before it, 9 accounts for Fresnel reflections at $0.3$0 per surface, and $0.3$1 is the average lantern throughput (Vievard et al., 2024).
For that visible device, the field of view evolves hyperbolically with focal ratio because the fiber acceptance angle is inversely proportional to focal ratio. Lower $0.3$2-ratio injection increases acceptance cone and field of view but risks modal overfilling and spectral loss; higher $0.3$3-ratio improves spectral reconstruction but narrows the field of view. The best compromise was found near $0.3$4, with average injection efficiency $0.3$5 over an $0.3$6 mas field of view, on-axis efficiency $0.3$7, and maximum efficiency $0.3$8. At lower focal ratios, reconstructed supercontinuum spectra exhibited spectral drops and low-frequency ripples attributed to modal overfilling (Vievard et al., 2024).
These measurements clarify why lanterns are especially useful when direct single-mode coupling is marginal. The 2014 astrophotonics study notes that direct coupling of seeing-limited telescope light into a single-mode fiber is roughly $0.3$9 times the Strehl ratio and is often far below $120$0 in large-telescope seeing-limited operation. A lantern relaxes that requirement by accepting a larger effective input while preserving single-mode delivery downstream (Horton et al., 2014).
4. Spectroscopic deployment in astronomy
Photonic lanterns entered astronomical instrumentation first as interfaces to single-mode photonic functions such as OH suppression, arrayed-waveguide spectroscopy, and stable diffraction-limited feeds (Leon-Saval et al., 2015, Birks et al., 2015). The most developed recent implementations are on the Subaru Coronagraphic Extreme Adaptive Optics system, where a visible-light lantern was integrated as a new injection mode for the FIRST spectro-interferometer. In this configuration, the $120$1 single-mode outputs are spliced into a linear V-groove and fed to a visible mid-resolution spectrograph around $120$2–$120$3; a Wollaston prism splits the two polarizations, so the detector records $120$4 traces corresponding to $120$5 outputs times two polarizations (Vievard et al., 2024).
The 2024 Subaru study reported the first on-sky demonstration of a visible photonic lantern coupled with an extreme adaptive optics system. On Ikiiki ($120$6 Leo), observed for $120$7 minutes at $120$8 Hz with injection set to $120$9, the reconstructed stellar spectrum showed H$0.5$0 absorption near $0.5$1 nm together with the atmospheric O$0.5$2-B band near $0.5$3 nm and O$0.5$4-A band near $0.5$5 nm. On Po'a (Algol), observed to compare the lantern with a single-mode fiber at $0.5$6, the lantern data contained, on average, $0.5$7 times more flux, with a peak gain of $0.5$8 around $0.5$9 nm; the paper summarizes the result as a $5.7$0 throughput improvement over a sole single-mode fiber (Vievard et al., 2024).
A second 2024 Subaru report extended the visible demonstration to Ikiiki and Aua ($5.7$1 Ori) using the FIRST $5.7$2 spectrograph over roughly $5.7$3–$5.7$4 nm. Under median seeing conditions of about $5.7$5 arcsec measured in H band and large tip/tilt residuals over $5.7$6 mas, the average on-sky light-coupling efficiency was estimated as $5.7$7, with a maximum of $5.7$8 at $5.7$9 nm. The reconstructedAua spectrum contained O$2.1$0 A and B bands, H$2.1$1, TiO absorption bands, VO bands from about $2.1$2–$2.1$3 nm, and Ba II near $2.1$4 nm. The same study notes that the demonstrated performance would not have been achievable with a single visible SMF under similar residual tip/tilt, partly because the SMF field of view was about $2.1$5 mas at $2.1$6 nm (Vievard et al., 2024).
The practical reason for the gain over a lone SMF is the difference in angular acceptance. In the Subaru comparison, the lantern field of view was $2.1$7 mas, whereas the SMF field of view was $2.1$8 mas, so residual tip-tilt and low-order aberrations that displace the point-spread function outside an SMF can still be accommodated by the lantern (Vievard et al., 2024). This establishes the device as a viable feed for high-throughput, high-angular-resolution spectroscopy in regimes where visible Strehl remains limited.
5. Focal-plane wavefront sensing and second-stage control
The same modal sensitivity that makes lantern outputs useful for spectroscopy also makes them useful for wavefront sensing. When a focal-plane point-spread function is injected into a lantern, the distribution of power across its output ports depends on the incoming phase structure. Calibrated against known aberration modes, that intensity vector becomes a focal-plane wavefront-sensing signal capable of measuring non-common-path aberrations and other residuals that a pupil-plane sensor does not see (Lin et al., 2023, Sengupta et al., 25 Nov 2025).
Real-time laboratory demonstrations on the Subaru/SCExAO testbed established the basic performance of the photonic lantern wavefront sensor. Using a $2.1$9-port lantern at $60$0 nm, CACAO control at $60$1 kHz, and a leaky integrator with regularization $60$2, the loop corrected about $60$3 of injected wavefront error in the first five non-piston Zernike modes, remained stable out to roughly $60$4 nm low-order wavefront error, and corrected petaling modes. For injected time-varying low-order error of about $60$5 nm RMS with a $60$6 s decorrelation timescale, the system achieved about $60$7 rejection at $60$8 s timescales; the same study reported that the $60$9-port sensor could sense $300$00 of $300$01 low-wind-effect degrees of freedom in the segmented SCExAO pupil, including all four segment piston modes (Lin et al., 2023).
On-sky second-stage control was then demonstrated at Lick Observatory on the Shane $300$02 m telescope. In that architecture, ShaneAO’s Shack-Hartmann loop performed first-stage correction, while a non-dispersed $300$03-port lantern at the science focal plane sensed residual low-order aberrations and non-common-path aberrations. Using a leaky integrator with gain $300$04, leak $300$05, and three stacked $300$06 s frames per update, the lantern-measured RMS wavefront error decreased from about $300$07 nm open loop to $300$08 nm closed loop, and the closed-loop PSF peak count increased by $300$09 relative to open loop (Sengupta et al., 25 Nov 2025).
A further extension was spectrally dispersed wavefront sensing on Subaru/SCExAO. In the laboratory, dispersing the outputs of a $300$10-port near-infrared lantern increased the number of clearly corrected control modes from $300$11 in earlier undispersed work to $300$12, with stable correction for residual wavefront errors up to about $300$13–$300$14 nm RMS. On sky, the spectrally dispersed lantern was calibrated on Humu (Altair), operated at $300$15 kHz with leak $300$16, gain $300$17, and $300$18-frame latency, and delivered a $300$19 Strehl improvement in H band, corresponding to reduction of RMS wavefront error from about $300$20 nm to $300$21 nm (Lin et al., 1 May 2025).
Recent sensitivity studies have made the control trade-off more explicit. Work on the muirSEAL testbed reports that lantern performance depends not only on dynamic range but on photon-noise sensitivity, port allocation, and input focal ratio. In joint WFS/imager configurations, using more ports for wavefront sensing enables greater aberration sensitivity but leaves less spatial information for image reconstruction; experimental throughput and linear reconstruction tests on muirSEAL also showed strong alignment sensitivity and substantial crosstalk in the linear regime (Sengupta et al., 27 Aug 2025, Sengupta et al., 25 Jun 2026).
6. Imaging, interferometry, and broader photonic-lantern applications
Photonic lanterns do not only transport light and sense aberrations; they also retain spatial information. In coherent-imaging formalisms, the focal-plane field is projected onto lantern principal modes, and each output behaves like an effective aperture in the pupil. For a lantern transfer matrix $300$22, the modal coefficients $300$23 and output fields $300$24 can be written as
$300$25
so the measured intensities encode subdiffraction spatial structure. Simulations of a $300$26-port lantern show that pairwise interference of outputs yields observables analogous to aperture-masking visibilities but with weighted, overlapping effective apertures and strong sensitivity to small-scale asymmetries (Kim et al., 2024, Kim et al., 22 Oct 2025).
This mode-based perspective has now been verified on sky. The FIRST-PL module on SCExAO/Subaru recorded simultaneous focal-plane images and visible lantern spectra to self-calibrate time-varying aberrations, then used spectral-differential response maps to reconstruct the H$300$27-emitting disk of $300$28 CMi. The experiment achieved an H$300$29 photocenter precision of $300$30as in about $300$31 minutes of useful observation, recovered blue- and red-shifted photocenters along the disk major axis, and detected a minor-axis photocenter shift of about $300$32 mas associated with near-side/far-side asymmetry in the disk (Kim et al., 22 Oct 2025). Related theoretical work argues that extremely large telescopes should extend lantern spectroastrometry and interferometric imaging to higher contrast and smaller angular scales, with science cases including exomoons, quasar broad-line regions, and inner circumstellar disks (Kim et al., 2023).
Outside astronomy, the same reciprocal mapping between multimode fields and single-mode channels has been exploited in several distinct ways. In compressive microendoscopy, a tapered $300$33-core multicore-fiber lantern generated distinct multimode illumination patterns by exciting one core at a time; those patterns were highly stable under fiber bending and enabled single-pixel imaging reconstructed with the SARA-COIL algorithm (Choudhury et al., 2019). In free-space optical communications, a $300$34 lantern has been used as the front end of an all-fiber coherent beam-combining receiver, raising combining efficiency from $300$35 to $300$36 without turbulence and from $300$37 to $300$38 with turbulence (Zhang et al., 2021). In optical metrology, simulation of a $300$39-port non-mode-selective lantern driven from the single-mode side generated a custom nulling wavefront with $300$40 nm RMS error from the target, suggesting a reconfigurable non-interferometric null test for spherical, aspheric, and freeform surfaces (Romer et al., 21 Jul 2025).
A current systems-level extension is the Hybrid-Mode Selective Photonic Lantern proposed for deep-contrast exoplanet spectroscopy. In that concept, the lantern is positioned at the focal plane and directs object light into a central SMF feeding a mid-resolution spectrograph, while adjacent SMFs route surrounding speckle light into a low-resolution spectrograph for rapid wavefront sensing. The design is intended to combine spectroscopy and wavefront control in one compact photonic component and is being prepared for tests at UTSA’s high-contrast imaging laboratory and on SCExAO at Subaru (Morsy et al., 15 Apr 2026). Collectively, these developments indicate that the photonic lantern has evolved from a multimode-to-single-mode interface into a broader class of multifunctional photonic devices for sensing, imaging, and coherent optical control.