CO₂ reduction asks for more flux on a smaller area, less infrared in the beam, and hours of output that does not drift. Start from total optical power rather than lamp wattage: a 300 W xenon source delivers 50 W of optical power, and expanding that beam across a large window spends it faster than most experimenters expect.
The gas-phase mass balance and the thermal balance break before the optics do. A water-splitting rig is tuned for hydrogen: produced in large excess, chemically inert on most surfaces, easy to separate on a packed column and easy to detect. Every design choice that follows from that — a generous loop volume, a plain stainless line, a lamp with no infrared management — is wrong here.
Three things change at once when the reaction is switched to CO₂:
Flux density at the sample plane, in mW/cm², sets the pair-generation rate; lamp wattage only sets what you spend to get there. Two sources with the same electrical rating can differ substantially in what reaches a small cell once a filter stack and a window are in the path.
| Source | Lamp configuration | Total optical power | Infrared in the beam | Fit for a small-area CO₂ cell |
|---|---|---|---|---|
| PLS-SXE300E | 300 W xenon, adjustable 150–300 W | 50 W | High; needs a water or IR-cut filter | Good for a cell of a few square centimeters once the beam is condensed |
| PLS-SME400E H1 | 400 W xenon | 100 W total, of which 55 W visible and 6 W UV | High | Headroom for a larger window or a thicker filter stack |
| PCX-50C | Nine 10 W LED channels | Narrowband, set by channel choice | Low | Flux limited but thermally clean; water-cooled −10 to 80 °C |
| PCX-50B | Nine 5 W LED channels | Narrowband, set by channel choice | Low | Screening runs where flux is held constant and spectrum is the variable |
Read the table as a trade, not a ranking. The xenon route buys flux and pays in infrared; the LED route buys spectral control and pays in peak flux. Both can serve a CO₂ cell with a window of a few square centimeters, but they fail differently, and the failure appears in the temperature record before it appears in the rate. Measure the delivered flux with a calibrated meter at the window plane, window in place, and put that number in the method section.
Infrared changes the temperature of the cell rather than the chemistry inside it, and in a sealed CO₂ system a temperature change is also a pressure and water-balance change. Xenon output extends well past the visible, so without a water filter or an IR-cut filter a large share of the delivered power arrives as heat at the window and at the bed.
The symptoms are consistent. Window fogging appears within the first hour; pressure rises in a way that tracks lamp power rather than conversion; and the rate falls when the lamp is dimmed even though the photon flux in the absorption band barely changed. Removing infrared upstream of the cell also protects the seal and the window coating, which age faster under an unfiltered beam.
Where the cell must stay cold, a water-cooled LED head removes the problem at the source: the PCX-50C holds −10 to 80 °C on the cooling circuit and emits little infrared to begin with. Where the xenon spectrum is needed, place the filter before the shutter so the cell never sees an unfiltered beam, not even during alignment.
A CO₂ reduction rate is a small difference between large numbers, so source drift is not a nuisance — it is the signal. Water splitting produces enough hydrogen that a slow change in lamp output disappears into the noise; CO₂ reduction does not have that margin.
The analytical side is rarely the bottleneck. A trace gas analyzer specified at 0.0005 μmol for CH₄ and CO, and 0.05 μmol for H₂, resolves products far below the rate of a typical cell, and hydrogen as a byproduct stays visible even at low selectivity. The limit is the source: output that changes over a long run produces an apparent rate change larger than the effect being studied.
Two habits protect the data. Bring the source to a stable output before the run starts, and log flux at the start, middle and end of every run. A run with only a start value cannot be defended in review.
Verify all three with the reactor mounted, because each one changes when the cell, window and filter stack enter the path. Work in this order.
Yes when the claim is solar comparability, and then only with both spectrum and flux reported. A simulator set to AM1.5G gives a defined reference, but its infrared content and beam uniformity differ from a xenon source with a filter stack, so rates are not transferable between the two without a flux measurement.
No. Above the flux at which pair generation saturates, extra power goes into heating, and heating changes selectivity as well as rate. Raise power in steps, record flux and sample temperature at each step, and keep the step where the rate still responds to flux.
Usually less urgently, because LED emission is narrowband and mostly outside the infrared. The cell still heats from the array and its driver, so measure the sample temperature rather than assuming the cooling circuit controls it.
