Fiber delivery is governed by étendue, not by lamp power: a 1 mm core, NA 0.22 fiber accepts about 4% of a 1 × 1 mm LED die but 30–50% of a xenon arc through a liquid light guide. Budget 0.1–0.3 dB/m attenuation at 365–450 nm, 12.7° half-angle divergence, and ±10% uniformity.
Every optical system conserves throughput, G = A · π · NA², where A is the emitting area and NA the collection cone. A fiber can only swallow the photons that fit inside its own G, so the maximum fraction of a source you can couple in is Gfiber / Gsource. Nothing downstream — bigger lenses, shorter focal lengths, silvered reflectors — changes that ratio. This is why LED-to-fiber losses of 90% are normal and are not a fault of the coupling optics.
| Source | Emitting area | Guide | Practical coupling | Delivered at 300 W / 100 W input |
|---|---|---|---|---|
| Xenon arc, ellipsoidal reflector | 0.5 × 1.0 mm | 3–5 mm liquid light guide, NA 0.5 | 30–50% | 90–150 W |
| Xenon arc, direct focus | 0.5 × 1.0 mm | 1 mm silica, NA 0.22 | 5–10% | 15–30 W |
| LED, 1 × 1 mm die | 1.0 mm² | 1 mm silica, NA 0.22 | 3–5% | 3–5 W |
| LED, 0.5 × 0.5 mm die | 0.25 mm² | 1 mm silica, NA 0.22 | 12–16% | 12–16 W |
| Diode laser, 450 nm | < 0.01 mm² | 0.2 mm silica, NA 0.12 | 70–85% | — |
If you need high irradiance at the end of a fiber, start with a small, bright emitter: halving the die dimension quadruples the coupled fraction. If you need total watts rather than watts per square centimetre, use a liquid light guide and accept its 340 nm cut-off.
Coupling is only the first term. A realistic run through a xenon source, one mated connector, 3 m of UV-grade silica and an output collimator looks like this:
| Stage | Loss | Power remaining |
|---|---|---|
| Arc to fiber (40% coupling) | −4.0 dB | 400 mW |
| SMA connector, one mate | −0.5 dB | 356 mW |
| 3 m fiber at 0.3 dB/m (365 nm) | −0.9 dB | 289 mW |
| Uncoated exit face, Fresnel | −0.16 dB | 278 mW |
| Collimating lens (92% transmission) | −0.36 dB | 256 mW |
Attenuation is strongly wavelength dependent. Visible-grade silica runs 0.05–0.2 dB/m from 400 to 700 nm, UV-grade high-OH silica 0.3–0.8 dB/m at 365 nm, and 1–3 dB/m at 254 nm where solarisation also darkens the core. Liquid light guides are 0.2–0.5 dB/m in the visible but opaque below 340 nm. Every mated SMA pair costs 0.3–1.0 dB, so each additional connector in a multi-reactor manifold is expensive.
Bending is the quiet killer: a 1 mm core needs a radius above roughly 100 mm, and a 30 mm loop adds 0.5–1.5 dB while pushing light into higher-order modes that change the spot size at the reactor.
A multimode fiber output is not a flat disc. It diverges at the fiber NA: at a working distance z, the spot diameter is roughly d + 2z·tanθ, which for NA 0.22 (θ = 12.7°) means a 1 mm core grows to 14.5 mm at 30 mm. Irradiance therefore falls faster than 1/z² once z exceeds a few core diameters: 1 W out of a 1 mm fiber gives about 600 mW/cm² at 30 mm, but 60 mW/cm² at 95 mm.
For anything larger than a 10 mm target, use a bundle. A 7-fiber bundle randomised round-to-round holds ±10% across a 12 mm spot; a 19-fiber bundle mapped round-to-line holds ±5% across a 50 mm line, which is what a flat-plate window needs. Measure on a 9-point grid and state the index you used — minimum/maximum, or (max−min)/(max+min) — because the two differ by a factor of two on the same data. Target ±10% for rate comparisons and ±5% for a published quantum yield.
Fiber is the wrong tool above roughly 500 mW/cm² over 50 cm²: that is 25 W at the window, or a 70 W source after 40% coupling and 1.5 dB of run loss, and the bundle face will heat. A collimated xenon source or a close-packed LED array is cheaper and more uniform at that scale.
Fiber earns its cost in four situations: the lamp must stay outside a glovebox; one lamp feeds several reactors through a splitter; the reactor sits in a strong magnetic field; or the sample is a photoelectrochemical cell where electrode and beam must be mechanically decoupled. A 1 × 4 multimode splitter yields 20–22% per port after excess loss, not the 25% a naive division predicts.
Yes, but only over a small area. One watt out of a 1 mm core expands to a 14.5 mm spot at 30 mm working distance, which is 1.65 cm² and therefore about 600 mW/cm². To hold 100 mW/cm² over 1 cm² you need roughly 0.17 W at the window; over 10 cm² you need a bundle and about 2 W of coupled power.
Not by shaping it. Silica transmits from about 200 nm to 2.4 µm with a smooth attenuation curve, so the spectrum at the exit is the input spectrum attenuated, not filtered. Two exceptions matter: liquid light guides cut off below 340 nm, and some LLG formulations fluoresce faintly under 365 nm, adding a broadband background that a monochromator will see.
After every connector mating and at least quarterly. Put a power meter at the exit and log the reading against the date and the lamp hours. A drop beyond 10% at constant lamp current points to a damaged or contaminated fiber face rather than to lamp ageing, and the two are easily confused when only the reactor rate is monitored.
Because the datasheet quotes wall-plug radiant power, not radiance into a small cone. A 100 W LED with a 1 mm² die spreads that power over a hemisphere, and a 1 mm, NA 0.22 fiber can only accept about 4% of it. Use a smaller die, a larger core, or a higher-NA bundle; adding a bigger lens only improves the fill of the fiber's own acceptance cone, which is already the limit.
