Select by optical path first, because light, not volume, sets the rate. Slurries run 10–50 mm paths at 0.2–2 g/L catalyst; flat-plate reactors use 1–10 mm gaps; fixed beds hold 1–5 mm particles; microreactors use 100–1000 µm channels at 1000–10000 m²/m³ of illuminated area per unit volume.
In a photocatalytic reactor, the illuminated volume is not the working volume. Photon absorption follows Beer–Lambert, and for a typical anatase slurry at 1 g/L more than 90% of 365 nm light is absorbed within the first 2–3 mm. A 50 mm annular reactor therefore spends most of its volume in the dark, and adding catalyst or volume cannot recover that. The useful figure of merit is illuminated area per unit liquid volume, which for a parallel gap of thickness L millimetres is roughly A/V ≈ 1000/L m²/m³: 20 mm gives 50, 2 mm gives 500, 0.5 mm gives 2000.
That single ratio explains the whole family tree below. Thin paths give high quantum efficiency and low throughput per unit area; thick paths give throughput but waste photons. Every commercial design is a position on that trade-off.
| Type | Optical path | Catalyst form | A/V (m²/m³) | Typical throughput | Main limitation |
|---|---|---|---|---|---|
| Slurry, immersion or annular | 10–50 mm | Powder, 0.2–2 g/L | 20–100 | 0.1–5 L batch | Light gradient, catalyst recovery |
| Fixed bed, packed or monolith | 1–5 mm voids | 1–5 mm granules or washcoat | 100–400 | 1–100 mL/min gas | Channelling, pressure drop |
| Flat plate, thin film | 1–10 mm | Immobilised film, 1–10 µm | 100–1000 | 10–500 mL/min | Sealing, film adhesion |
| Microfluidic or capillary | 100–1000 µm | Wall coating or segmented slug | 1000–10000 | 10–500 µL/min per channel | Fouling, low absolute output |
| Spinning disc | 20–100 µm film | Immobilised on disc | 300–2000 | 0.5–20 L/h | Mechanical wear, scale-up by area |
Slurries win on mass transfer because catalyst and reactant share one phase, and on optics because suspended powder scatters light deeper than a flat film. Loading has a sharp optimum: below 0.2 g/L you are photon-starved, above about 2 g/L extra catalyst only raises turbidity, shortening penetration depth and lifting the rate per gram while lowering it per reactor. Keep the powder suspended at 300–800 rpm in a 500 mL vessel, without creating a vortex that admits air.
The cost appears at the end: sub-micron titania neither settles nor filters. Centrifugation at 8000 g for 10–15 min, or 0.1–0.2 µm membrane filtration, sets the experiment cycle time rather than the photoreaction.
A packed bed removes the separation step and makes continuous operation trivial, at the price of two hydraulic problems. Channelling appears when the tube-to-particle diameter ratio falls below 10, because the wall void fraction exceeds the bulk void fraction and flow short-circuits along the wall; keep D/d at 10–20. Pressure drop rises as 1/d² for a given superficial velocity, so a 1 mm granule bed can cost 5–10 kPa/m where a 3 mm bed costs 1–2 kPa/m. Monoliths avoid both by offering straight, uniform channels, but their washcoat loading is only 20–80 g/m² of geometric area, which caps the catalyst inventory per reactor volume.
Light access is the hidden constraint: a packed bed is opaque after 1–2 mm, so unless the tube is thin-walled quartz and the lamp surrounds it, only the first particle layer works.
Flat plates put a thin, uniform liquid layer behind a quartz window and illuminate it normal to the surface. With a 2 mm gap and 100 mW/cm² incident at 365 nm, the whole volume receives 20–90 mW/cm² depending on catalyst loading, which is why flat plates are the standard geometry for apparent quantum yield. Scale-up is by numbering up — adding plates in parallel — because enlarging one changes both path length and irradiance uniformity.
Microreactors take the logic to its limit. A 500 µm × 10 mm × 50 mm channel holds 250 µL and, at 100 µL/min, gives a 2.5 min residence time with a path shorter than the oxygen diffusion distance. That suits screening and fast photoredox chemistry, and fails anything needing grams per day, since output scales by adding channels rather than by raising flow.
Usually the flat plate or the microreactor, because both keep the optical path near the photon penetration depth. A 1 mm gap with an immobilised film can exceed 10% AQY where a 50 mm slurry of the same catalyst reaches 1–2%, and the difference is geometry, not chemistry. Report the path length with any AQY value or the number cannot be compared.
Only after normalising to absorbed photons and to catalyst mass. A slurry rate per gram and a fixed-bed rate per gram differ by an order of magnitude for the same material, because the fixed bed illuminates a small fraction of its catalyst. Convert both to mol per photon absorbed per second, and state the illuminated fraction explicitly.
Three common causes, in order of frequency: fines migration blocking the inlet void space, condensable products coating the particles, and bypassing as the packing settles. Check the pressure drop trend before suspecting the catalyst; a rise of more than 20% over a run points to physical blockage rather than deactivation.
When the limiting step is oxygen transfer rather than photon supply. The thin film on a disc spinning at 100–1000 rpm is continuously renewed and re-oxygenated, giving mass transfer coefficients well above a stirred tank, and the catalyst stays immobilised. If your reaction is already photon-limited, the disc adds mechanical complexity for no gain.
