Buy a thermopile head for 50–2000 mW/cm2 broadband work and a UV-enhanced photodiode below 10 mW/cm2. Require a calibration certificate at ±5 % (k=2), a stated spectral range, and an active area in mm. Recalibrate every 12 months, and check the new meter against a reference within ±5 % on arrival.
A power meter does not measure photons. It measures a photocurrent or a temperature rise across an active area of a few millimetres, then multiplies by a stored responsivity curve. The reading in mW/cm2 is therefore valid only when three conditions hold: the beam is wider than the aperture, the wavelength setting matches the light, and the stored curve belongs to that detector head. Break any one of them and the instrument still displays four digits, which is why two meters placed in the same beam routinely disagree by 20–40 % while both are “calibrated”.
The first purchase decision is the detector, not the display unit. Almost every meter body accepts only one detector family, so this choice locks in the rest of the specification.
| Detector | Useful range | Spectral span | Wavelength setting | Best for | Watch out for |
|---|---|---|---|---|---|
| Thermopile (disc or cone) | 1–10,000 mW/cm2 | 190 nm–20 µm, flat | Not required | Xenon and Hg-Xe sources, photothermal work, any full-spectrum flux | Reads IR leakage that a photocatalyst never uses; slow, 0.5–3 s response |
| Si photodiode | 0.01–200 mW/cm2 | 350–1100 nm | Mandatory | LED panels, low-flux screening, long stability runs | Responsivity falls to about 0.15 A/W at 365 nm against 0.55 A/W near 900 nm |
| UV-enhanced Si or GaAsP | 0.01–100 mW/cm2 | 200–400 nm emphasis | Mandatory | 365 and 254 nm work, quantum-yield photon counting | Degrades with UV dose; recheck responsivity yearly |
| Spectroradiometer | 0.001–100 mW/cm2 | 200–1100 nm in 1 nm bins | None, it resolves the spectrum | Broadband AQY, filter validation, spectral mismatch audits | 5–10× the cost; stray light and cosine error at the fibre tip |
| Chemical actinometry | Equivalent 1–50 mW/cm2 | Defined by the actinometer band | None | Independent check on any radiometer | Agrees with a calibrated head only to ±10–15 %; hours per measurement |
A lab that runs both filtered UV and full-spectrum photothermal experiments needs two detectors, and should expect them to disagree. The disagreement is information: the gap between a thermopile reading and a photodiode reading on a 365 nm bandpass beam is roughly the IR leakage of that filter, usually 1–5 % of the transmitted power with a good OD3 block, and much more with a cheap one.
Certificates quote an uncertainty at one wavelength, often 550 nm, on a reference source. Your lamp is not that source. The correction factor is the ratio of two integrals, the measured spectrum weighted by the detector responsivity, divided by the same spectrum weighted by the responsivity used at calibration. Two practical consequences follow. First, a photodiode operated at the wrong wavelength setting on a narrow UV band can be off by 20–40 %, in either direction. Second, a thermopile needs no wavelength correction but does integrate everything the optics pass, so a “365 nm” source that leaks 3 % in the near IR reads 3 % high on a thermopile and correctly on a photodiode.
For work that will be published, state four things with the irradiance: the detector type, the band over which the value applies, the active area, and the calibration date. “120 mW/cm2, thermopile, 320–500 nm, calibrated 2026-03” is checkable. “120 mW/cm2” is not.
Reviewers rarely challenge a reading; they challenge a reading with no budget behind it. In a typical photocatalysis setup the random error is small and the systematic terms dominate.
| Source | Typical contribution | How to cut it |
|---|---|---|
| Calibration certificate | ±3–5 % (k=2) | Buy ±2–3 % traceable heads; keep the certificate number in the log |
| Wavelength setting error | 1–3 % on a UV band edge | Set the actual peak wavelength, not a rounded value |
| Position and geometry | 1–3 % per mm near focus | Fixed jig at the reaction plane, same window and liquid depth as the run |
| Linearity | ±0.5 % of full scale | Work at 20–80 % of the selected range |
| Thermal drift | 0.05 %/°C on photodiodes | 15 min warm-up; log ambient temperature |
| Beam non-uniformity | 10–30 % centre to edge | Nine-point grid, report mean and deviation, not the hot spot |
Combined in quadrature, a careful measurement lands at 5–7 % (k=1). That is the number to put next to the irradiance. Claiming better than 3 % without a mismatch correction and a position jig is usually wishful.
Quantum yields need a photon count, and the conversion is where units get lost. Incident photon flux in mol/s equals optical power in watts times wavelength in metres, divided by 0.1196 J·m/mol. One milliwatt at 365 nm is therefore 3.05 × 10-9 mol photons/s, and the same milliwatt at 450 nm is 3.76 × 10-9. Over 19.6 cm2 at 150 mW/cm2 the 365 nm flux is 2.94 W, or 8.97 × 10-6 mol/s, which is 32 mmol/h. Any apparent quantum yield quoted without that chain of arithmetic cannot be reproduced, however good the meter was.
Yes if your work spans more than two decades of flux, which is normal once a lab runs both a 300 W xenon source and LED screening panels. The thermopile covers 50–2000 mW/cm2 without saturation; the photodiode resolves 0.01 mW/cm2 where the thermopile is at its noise floor.
Every 12 months for weekly use, every 6 months for daily use or any UV-heavy duty. Recalibrate immediately after a drop, an over-range exposure, or a reading that disagrees with a reference by more than 5 %. Photodiode responsivity in the UV falls measurably within two to three years of constant use.
No. A lux meter applies the photopic weighting curve, which peaks at 555 nm and collapses in the UV, so the same 365 nm beam can read 10–30× lower than its true irradiance. An uncalibrated solar cell has an unknown spectral response and no traceability at all.
Almost always spectral mismatch, IR leakage, or a beam narrower than one of the apertures. Run both on the same beam at the same plane with the wavelength set explicitly; if the gap persists, put a spectroradiometer or ferrioxalate actinometry between them before trusting either number.