A rugged metallic sphere with an ultra-reflective inner coating. Photons are injected through an input fiber, bounce millions of times inside the cavity, and are extracted on demand through an output fiber — a dynamic reservoir of optical energy.
Formally: a closed-geometry resonant optical cavity operated as a power build-up and flux-integration element. Its figure of merit is not "how much light is stored" but the product pump power × photon lifetime — and the ratio of that lifetime to every loss channel competing for the same photons.
Not a "bottle of light" — a prolonged photon residence time creating high optical energy density.
Laser light enters the sphere through a fiber-coupled port at a controlled angle.
Photons reflect millions of times off the dielectric-coated inner wall. The cavity acts as an optical energy integrator.
A dynamic balance forms between injected power, losses (absorption, scattering) and extraction.
A partially transmissive spot on the coating feeds the output fiber with a stabilized, controllable flux.
The single most common conceptual error is to conflate them. They differ by four orders of magnitude in storage time.
Lambertian coating (Spectralon, ρ ≈ 0.99). Rays are randomized at every hit; mean chord = 2D/3. Storage time for D = 20 cm is only ≈ 44 ns. Perfect for radiometry, useless as a reservoir: no coherence, no build-up beyond the sphere multiplier M = ρ/[1−ρ(1−f)] ≈ 100.
Dielectric mirror coating, ray paths deterministic, fields interfere constructively on resonance. Storage time for the same shell with 5 ppm loss is ≈ 200 µs — a factor 4500 better. This is the actual SPHERION regime.
Solid silica/MgF₂ ball, light confined by total internal reflection near the equator. No coating at all, Q up to 3·10¹¹ measured — but the mode volume is microscopic (µm³) and the stored energy correspondingly tiny.
Δν = ν₀/Q. Storage is therefore always bought with bandwidth, never with reflectivity alone.Why a sphere is not simply "a mirror box that happens to be round".
In a perfect sphere the surface normal is always radial, so specular reflection preserves the plane containing the ray and the centre. A ray injected once never leaves that plane: its trajectory is a polygon inscribed in a great circle, and all its chords are tangent to an inner caustic circle of radius a·sin θ. The 3-D problem collapses to a 2-D billiard.
Every bounce reproduces the same incidence angle θ, so every chord has the same length. This makes the sphere an integrable billiard — no ray chaos, unlike a stadium or a deformed cavity. Consequence: the coating sees a single, well-defined angle, and its reflectivity can be optimized for exactly that angle and polarization.
The dashed violet circle is the caustic: every chord of the orbit is tangent to it, and no ray of this family ever enters it. As q/p → ½ the caustic swells to the wall and the orbit becomes a whispering-gallery mode; as q/p → 0 it collapses to the centre and the orbit becomes diametral.
| Orbit (p/q) | θ | Chord ℓ | Round trip Lrt | Character |
|---|---|---|---|---|
| 2 / 1 — diametral | 0° | 2a | 4a | Longest chord ⇒ longest lifetime, but concentric ⇒ marginally stable |
| 3 / 1 — triangle | 30° | 1.732a | 5.196a | Compact, three well-separated wall spots |
| 4 / 1 — square | 45° | 1.414a | 5.657a | Convenient 90° ports; Brewster-sensitive polarization split |
| 5 / 2 — pentagram | 18° | 1.902a | 9.511a | Star orbit: long path, high spot count |
| 6 / 1 — hexagon | 60° | a | 6a | Textbook grazing-family entry point |
| 36 / 1 — whispering gallery | 85° | 0.174a | 6.27a → 2πa | Grazing incidence, self-guiding, highest R but most bounces per metre |
Treated as a two-mirror resonator, the diametral orbit has cavity length L = 2a and mirror radius R = a, so g = 1 − L/R = −1 and g₁g₂ = 1: exactly on the concentric stability boundary. A ray displaced by δ returns displaced by −δ and never re-focuses. Any real sphericity error or misalignment throws the mode off the diameter. The grazing (whispering-gallery) families, by contrast, are dynamically stable — the curvature continuously refocuses the beam toward the caustic. This is the central design trade of the whole device.
Photon lifetime is τ = ℓ / (c·𝓛bounce) — it depends on the chord, not the round trip. Grazing orbits pack more bounces per metre of path, so unless the coating's reflectivity improves faster than cos θ falls, going grazing shortens the lifetime. Dielectric stacks do gain reflectivity at high incidence (especially s-polarization) but simultaneously blue-shift their stop-band by λ·cos θ — the coating must be designed for the chosen orbit, not for normal incidence.
The vector Helmholtz equation in a spherical shell gives TE/TM modes indexed by radial n, polar ℓ and azimuthal m, with eigenfrequencies from the zeros of the spherical Bessel functions. Perfect spherical symmetry makes every mode (2ℓ+1)-fold degenerate in m. Real sphericity errors lift that degeneracy by a fractional splitting of order the fractional deformation — a 10 nm figure error on a 100 mm radius (10⁻⁷) already smears each multiplet across many linewidths at Q = 10¹¹.
The number of electromagnetic modes inside bandwidth Δν is N ≈ 8πVν²Δν/c³. For a 20 cm sphere at 1550 nm this is ≈ 1.5·10⁵ modes per kHz. Even a Hz-linewidth laser therefore addresses a dense quasi-continuum, not one clean resonance. That is precisely why the honest description is energy integrator: the device is macroscopically multimode by construction, and single-mode operation would require deliberately spoiling the degeneracy (deformation, aperture, or mode-selective coupling).
Four subsystems, one sealed sphere — plus the two that are always forgotten: the vacuum and the metrology.
R > 99.99%, ideally R > 99.999%TIS ≈ (4πσ/λ)², where σ is the RMS micro-roughness. At 1550 nm, 1 ppm of scatter demands σ ≲ 0.12 nm — sub-atomic-step polishing over a full sphere's inner surface. This, not the coating recipe, is the true manufacturing bottleneck: superpolishing a concave closed sphere is far harder than polishing a flat or a mirror blank, and is the single most likely reason a real build lands at 10⁻⁴ rather than 10⁻⁶ loss.The complete steady-state formalism — and the ceilings nature refuses to move.
1 − e−t/τ and free ring-down e−t/τ. Tick labels are live-linked to the calculator below.Even at R = 99.999%, photons survive microseconds to milliseconds — not minutes. A full millisecond at 1550 nm requires Q ≈ 1.2·10¹², an order of magnitude beyond the best resonator ever measured at any wavelength in the optical band.
Every lost photon becomes heat in the coating and shell — concentrated on a handful of sub-millimetre spots, not spread over the sphere. Local flux, not total watts, sets the damage margin.
Dielectric stacks are near-perfect over roughly Δλ/λ ≈ 10%, and ppm-class only over a fraction of that. The reservoir is inherently monochromatic, and its stop-band shifts as λ·cos θ with orbit angle and as dn/dT with temperature.
F = 2Pcirc/c gives 6.7 mN per megawatt circulating — enough to deform a thin shell at the picometre scale, to create an optical spring, and above threshold to drive parametric instabilities that steal power into mechanical modes.
Circulating power ≈ ℱ/π × input. A finesse of 10⁶ turns 10 W of pump into ~3 MW circulating — comparable to a LIGO arm cavity, in an object the size of a grapefruit.
Micron-scale deformation shifts cavity modes; at Q = 10¹¹ the linewidth is ~2 kHz, i.e. a length stability requirement of δL/L ≈ 10⁻¹¹. The shell's rigidity is not cosmetic — it is optical.
Passive optics conserve étendue: you cannot make the intracavity brightness exceed the pump laser's. Build-up increases intensity by recirculation, never by concentration — a distinction that quietly kills most "light concentrator" proposals.
A passive cavity fed by a source of brightness temperature T cannot exceed blackbody occupancy at T in any mode. The second law, not engineering, forbids the "optical battery" reading of this device.
Extraction is ultimately shot-noise limited: relative intensity noise ≥ √(2hν/P). Below the cavity pole the reservoir filters classical pump noise; above it, it cannot beat the vacuum.
U = 2 mJ stored in a 4.2 litre sphere, i.e. ≈ 0.5 J·m⁻³. A lithium-ion cell holds ≈ 2·10⁹ J·m⁻³ — ten orders of magnitude more. SPHERION is not an energy store in any engineering sense; it is a power device. What it genuinely delivers is megawatt-class circulating intensity, kilohertz-class spectral selectivity and a 10⁻¹¹-level loss transducer — all from a watt-class fibre laser.At the parts-per-million level, every physical effect you normally ignore becomes the design driver.
Representative per-bounce budget for a state-of-the-art ion-beam-sputtered SiO₂/Ta₂O₅ stack at 1550 nm.
Crystalline AlGaAs/GaAs coatings shift this budget: absorption drops below 0.2 ppm and Brownian noise falls ~10×, but scatter rises with the bonded-transfer process and the usable aperture shrinks. There is no coating that wins on every axis simultaneously.
Because the steady-state share of injected power ending in any channel equals that channel's share of the round-trip loss, this single pie answers two different questions at once: where do photons die, and where does the pump power go. With this budget, a 10 W pump delivers 3.6 W out of the coupler, dissipates 0.9 W as heat in a few sub-millimetre spots, and scatters 4.5 W into the shell interior as stray light — which must itself be baffled, or it re-couples and corrupts the ring-down.
Mechanical dissipation in the multilayer drives displacement noise ∝ √(kBT·φcoating). It is the limiting noise of gravitational-wave detectors in their most sensitive band, and it sets the frequency stability floor of any ultrastable cavity. Crystalline coatings and cryogenic operation are the only known mitigations.
Temperature fluctuations in the mode volume modulate both n(T) and the geometry, converting thermal noise directly into frequency noise. These scale inversely with mode volume — one of the few places where a big sphere genuinely beats a microresonator.
Absorbed power heats the illuminated spot on a thermal timescale of milliseconds, bulging the coating and detuning the mode. The result is a delayed, power-dependent feedback path that can turn the servo loop unstable even when everything is nominally in lock.
Because absorbed power depends on detuning and detuning depends on absorbed power, the resonance curve becomes asymmetric and finally bistable. Sweeping the laser up and down in frequency traces different curves — the classic triangular lineshape of a high-power cavity. Locking must approach the resonance from the thermally stable side.
At MW-class intensities the coating's n₂ produces an intensity-dependent phase shift, enabling four-wave mixing, Kerr comb formation and modulational instability. In a mode-dense sphere this is a feature (broadband generation) and a hazard (uncontrolled power transfer between the 10⁵ nearby modes).
Radiation pressure couples optical modes to acoustic modes of the shell. When an acoustic mode is resonant with an optical mode spacing and the gain exceeds mechanical damping, the shell rings up and steals circulating power. Observed in advanced LIGO at ~10 kHz; damped there by active feedback and lossy dampers.
CW damage for good dielectric stacks sits around 1–10 MW·cm⁻² and is defect-, not field-, limited: one absorbing nodule sets the threshold for the whole mirror. A 1 mm spot carrying 3 MW is already at 10⁸ W·cm⁻² — hence the mandatory beam-expansion / spot-distribution strategy.
In vacuum the intracavity medium contributes nothing, but any residual gas or any solid element in the beam (windows, gain media, coupling optics) brings Brillouin and Raman thresholds that scale as 1/(g·L·ℱ). High finesse lowers every nonlinear threshold by exactly the build-up factor.
Loss the textbooks omit: the overlap integral between the injected fibre mode and the cavity eigenmode. 90% coupling means 10% of the pump never enters — comparable to everything else in the budget combined, and by far the easiest to get wrong.
Every number below is computed live from the formalism in §4 — no lookup tables, no fitting.
The ultimate physics-limited design — and the version you could build first.
Everything pushed to known physical limits: crystalline-coating mirrors, cryogenic operation, superpolished substrate, vacuum-sealed interior.
| Parameter | Limit value |
|---|---|
| Reflectivity | R ≈ 99.9999% (crystalline GaAs/AlGaAs coatings, <1 ppm loss) |
| Finesse / Q | Q > 10¹¹ (best supercavity regime) |
| Photon lifetime | ~ milliseconds → light travels 100s of km inside |
| Circulating power | MW–GW class from watt-level pump (coating damage-limited) |
| Environment | Cryogenic (~10 K), ultra-high vacuum, vibration isolation |
| Shell | Titanium sphere, actively cooled, interferometric shape control |
Hard ceiling set by coating absorption (~ppm), scattering from atomic-scale roughness, and thermal noise. Beyond this, no known material improves the reservoir.
Achievable with commercial off-the-shelf optics in months, not decades — a pragmatic proof of concept.
| Parameter | Practical value |
|---|---|
| Reflectivity | R ≈ 99.99% (commercial IBS dielectric coating @ 1550 nm) |
| Photon lifetime | ~ microseconds (thousands of bounces) |
| Shell | Stainless-steel sphere, Ø 10–30 cm, standard vacuum ports |
| Coupling | Standard SMF-28 fibers + fiber collimators + AR windows |
| Pump | 1550 nm telecom laser (1–10 W), fully fiber-connectorized |
| Cooling | Passive or simple water jacket |
Enough to demonstrate energy integration, output-flux stabilization, and loss-based sensing — the fastest credible path to a working reservoir.
| Figure of merit | Tier 1 — near-term | Tier 2 — credible lab | Tier Ω — physics limit |
|---|---|---|---|
| Per-bounce loss 𝓛 | 100 ppm | 20 ppm | 5 ppm |
| Finesse ℱ | 1.6·10⁴ | 7.9·10⁴ | 3.1·10⁵ |
| Photon lifetime τp | ~ 4.7 µs | ~ 24 µs | ~ 94 µs |
| Optical path | 1.4 km | 7 km | 28 km |
| Circulating power @ 10 W | 50 kW | 250 kW | 1.0 MW |
| Stored energy @ 10 W | 47 µJ | 240 µJ | 0.94 mJ |
| Environment | Bench, passive cooling | 10⁻⁶ mbar, thermal enclosure | UHV, ~10 K, seismic isolation |
| Dominant limit | Substrate roughness | Mode matching & port leakage | Coating absorption & thermal noise |
Computed for a 20 cm sphere on the square orbit (p = 4) at 1550 nm. Reproduce any row with the calculator in §6.
A high-finesse cavity that is not locked is simply an expensive mirror. The servo chain is not an accessory — it is half the physics.
Phase-modulate the pump at a frequency well outside the cavity linewidth; the beat between the promptly reflected sidebands and the leaked intracavity carrier yields an antisymmetric error signal that is (a) linear through resonance, (b) immune to intensity noise to first order, and (c) has a capture range set by the modulation frequency rather than the linewidth. Without PDH, a 2 kHz linewidth at 193 THz is a 10⁻¹¹ fractional target — unlockable by side-of-fringe methods.
Two topologies, opposite consequences. Laser → cavity: the sphere becomes a frequency reference and the laser inherits its stability — this is how ultrastable lasers reach 10⁻¹⁶. Cavity → laser: the sphere becomes a power reservoir tracking a free-running source. SPHERION as an energy integrator wants the second; SPHERION as a sensor wants the first. The mechanical design differs completely.
Shutter the pump and time the decay. Because τ is a pure time, ring-down is insensitive to pump power, coupling efficiency and detector gain — the three things hardest to calibrate. This is why cavity ring-down spectroscopy (O'Keefe & Deacon, 1988) still holds absorption-sensitivity records, and why it is the correct acceptance test for the coating.
The output coupling can be made dynamic: an electro-optic or acousto-optic element at the output port converts the reservoir from a passive filter into a controlled release valve. Fast dumping (≪ τ) gives a shaped pulse carrying most of U; slow bleeding (≫ τ) gives a low-noise CW output whose fluctuations above 1/2πτ are suppressed by the cavity pole.
The injected fibre mode must overlap the target cavity eigenmode. In a mode-dense sphere this is not automatic: bad matching does not just lose power, it excites the wrong orbit family, dumping energy into short-lifetime modes and corrupting the ring-down measurement with a multi-exponential tail. Aperturing at the caustic radius is the practical fix.
The cavity is a single-pole low-pass filter of corner frequency fc = 1/(2πτp) for amplitude noise — about 800 Hz at τ = 200 µs. Above it, pump RIN is suppressed by 20 dB/decade. Below it, the reservoir passes noise through unchanged and adds its own thermal drift. The device is a superb high-frequency stabilizer and a poor low-frequency one.
Where a spherical reservoir sits among every other way humanity has found to delay a photon.
| Platform | Q (optical) | Storage time | Path travelled | Character |
|---|---|---|---|---|
| SMF-28 fibre delay line | — | ~22 µs (1/e at 0.2 dB/km) | 4.3 km | No build-up; loss-limited; broadband |
| Photonic-crystal nanocavity | 10⁶–10⁷ | ~10 ns | ~3 m | Wavelength-scale mode volume; chip-integrable |
| Silica WGM microsphere | ~8·10⁹ | ~7 µs | ~2 km | No coating; surface-tension-perfect sphere |
| Crystalline MgF₂ WGM resonator | ~3·10¹¹ | ~250 µs | 75 km | Highest optical Q measured; µm³ mode volume |
| Supermirror Fabry–Pérot (ℱ ≈ 2·10⁶) | ~2·10¹¹ | ~200 µs | 60 km | Open geometry; the direct SPHERION analogue |
| LIGO arm cavity (4 km, ℱ ≈ 450) | ~10¹⁵ | ~1 ms | ~300 km | ~750 kW circulating; largest stored optical path built |
| SPHERION Tier Ω (20 cm) | ~2·10¹¹ | ~100–200 µs | 30–60 km | Closed geometry, massively multimode, µJ–mJ stored |
| Superconducting microwave FP (Haroche) | ~4·10¹⁰ | ~130 ms | ~39 000 km | Microwave, not optical — the true record holder |
| Rare-earth spin memory (Eu:YSO) | — | up to hours | — | Not photon storage: coherence is transferred to spins |
What a photonic reservoir is actually good for — ordered from "already demonstrated elsewhere" to "genuinely speculative".
The cavity is a one-pole low-pass filter on amplitude noise above 1/2πτp. At τ = 200 µs that is 20 dB/decade of RIN suppression above ~800 Hz, without a single electronic component in the optical path.
Loss transduction: an added absorption δα changes τ measurably when δα·Lrt ≈ 𝓛/100. At 5 ppm round-trip loss this means detecting 10⁻¹⁰ cm⁻¹ — trace gases at parts-per-trillion, or a sub-monolayer of adsorbate on the coating.
Lock a laser to the sphere instead of the reverse and the shell's dimensional stability becomes spectral purity. A rigid, sealed, athermal sphere is a mechanically far better reference body than the usual ULE spacer — this may be the device's strongest genuine claim.
TRL 5Dump the stored energy in a time short compared to τp: 1 mJ released in 100 ns is a 10 kW peak pulse from a 10 W CW fibre laser, with pulse energy set purely by the dump timing.
TRL 4Any process with a threshold ∝ 1/Pcirc — second-harmonic generation, OPO, Raman conversion, Kerr comb formation — sees its pump requirement divided by the build-up factor ℱ/π. A 3·10⁵ build-up turns milliwatt-threshold physics into microwatt-threshold physics.
Insert a gain medium in the high-intensity field. The catch is that any solid element must itself have ppm-class loss, or it instantly dominates the budget — in practice this means a thin, Brewster-cut, superpolished crystal, or a dilute gas.
TRL 36.7 mN per circulating megawatt acting on a closed elastic shell is a clean, symmetric optomechanical system: optical springs, radiation-pressure cooling of breathing modes, and parametric instability all become accessible in a table-top sealed unit.
TRL 3Long optical paths in a small, magnetically shieldable volume suit searches for vacuum birefringence, axion-like-particle-induced polarization rotation, and short-range force tests — all of which scale with accumulated path, i.e. with c·τp.
Microsecond-scale energy buffering for pulsed-power and burst-mode systems. Honest limit: buffering, not storage — the reservoir empties in τp, and its capacity is Pin·τp, not "however much you keep pumping in".
Each stage answers exactly one question and is falsifiable on its own.
Non-sequential ray tracing on the closed-orbit families, then a full vector mode solve including port apertures and measured figure error. Kill criterion: if port leakage alone exceeds 10 ppm, the closed-sphere topology loses to a conventional two-mirror cavity and the concept should be abandoned in favour of a linear resonator.
Two hemispheres, superpolished to σ < 0.2 nm RMS, interferometrically figured, then optically contacted or diffusion-bonded. Measure scatter directly with a TIS bench before any coating is applied. Kill criterion: σ > 0.5 nm ⇒ scatter alone caps finesse near 10⁴.
The genuinely unsolved manufacturing problem: ion-beam sputtering is a line-of-sight process and a closed concave shell is the worst possible target geometry. Options are coat-then-bond (risking the seam), rotating-planetary deposition through a large port, or atomic-layer deposition at the cost of higher absorption.
First light. A single-exponential decay proves clean single-orbit excitation; a multi-exponential tail proves mode mixing. This one measurement validates or refutes the entire geometrical model of §2.
Map the thermal-bistability lineshape, find the damage-free operating envelope, and demonstrate the RIN-suppression transfer function predicted in §8. First point where the device does something a mirror cannot.
These three have incompatible optimizations (loss sensitivity vs. dimensional stability vs. fast extraction). Committing to one is a design decision, not a marketing one.
The honest list of what nobody currently knows about this specific geometry.
Every ppm-class mirror ever made is a small, open, convex-side-accessible optic. There is no published demonstration of a ppm coating on a closed concave shell, and the deposition physics (angle-dependent stoichiometry, stress-driven delamination on a curved seam) is genuinely unresolved.
Mode degeneracy could make the device robust — energy that leaves one mode lands in another with a similar lifetime, so total stored energy is preserved even as the spatial pattern scrambles. Or it could be fatal: any weak coupling channel then drains the entire manifold. Which regime applies depends on the ratio of inter-mode scattering rate to loss rate, and that ratio has not been measured.
The diametral orbit sits exactly on g₁g₂ = 1. Perturbation theory near a marginal stability boundary is delicate; whether a real sphere's imperfections push the mode into stability or into rapid walk-off is a numerical question with no closed-form answer.
Longer chords give longer lifetime; grazing incidence gives higher reflectivity but shorter chords and a shifted stop-band. The optimum is a joint optimization over coating design and orbit angle that has, to our knowledge, never been performed for a closed sphere.
The uncomfortable question. A linear supermirror cavity achieves the same Q with a fraction of the coated area and full access for polishing. The sphere's only structural advantages are hermetic sealing, mechanical rigidity, and long path in a compact convex envelope. Whether those are worth the manufacturing penalty is the concept's central open bet.
Any manufacturable sphere is two bonded hemispheres. The equatorial seam is a scattering line crossing every whispering-gallery orbit exactly twice per round trip. Placing the orbit plane to avoid the seam is geometrically impossible for the grazing families — a constraint that may by itself select the diametral or polygonal orbits.
Notation used throughout, and the real published results the numbers are calibrated against.
| a, D | Sphere radius, diameter |
| θ | Angle of incidence (constant of motion) |
| ℓ | Chord length between bounces, 2a·cos θ |
| p, q | Bounces per closed orbit; turns before closure |
| 𝓛 | Fractional power loss (per bounce or per round trip) |
| T, A, S | Transmission, absorption, scatter, in ppm |
| ℱ | Finesse, 2π/𝓛rt |
| Q | Quality factor, ω₀τp |
| τp | Photon (energy) lifetime |
| νFSR, Δν | Free spectral range; cavity linewidth |
| U, Nγ | Stored energy; stored photon number |
| g | Resonator stability parameter, 1 − L/R |