Prove three things before the first experiment: the assembled system holds a leak rate at or below 0.1 μmol/h measured as oxygen; a degassing cycle leaves a background the detector can resolve; and a catalyst-free blank under the same light shows no trend above that background. Each gate is a number, not an impression.
Treat the first day as commissioning and expect to discard the data. The purpose is to establish three baselines — how fast gas crosses the system boundary, how much background the line contributes, and what the light and the reactor produce with no catalyst present — because every later result is read against them. Run the gates in order: a blank measured before the leak test proves a signal exists but not where it came from.
The order matters for attribution. A leak admits air, and air carries oxygen and nitrogen that shift the baseline and suppress reductive chemistry. Incomplete degassing leaves dissolved gas that desorbs slowly across the run, which looks like a rising product curve. A contaminated reactor adds a constant offset that survives every later change of catalyst. Three faults, three distinct symptoms, and only a sequenced start-up separates them.
Write the criteria down before you open a valve, because a criterion chosen after the measurement is not a criterion.
| Gate | What you measure | Pass criterion | What failure points to |
|---|---|---|---|
| Leak rate | Pressure rise in an isolated, evacuated section, or a gas signal over a sealed hold | At or below 0.1 μmol/h measured as O₂ on a 65 mL line volume | A joint, valve seat, septum or feedthrough |
| Degassing endpoint | Background H₂, CH₄ and CO after evacuation and purge, read on the analyzer | Below the analyzer's detection limit — 0.05 μmol for H₂, 0.0005 μmol for CH₄ and CO | Dissolved gas still desorbing, or a purge that never reached a dead volume |
| Blank run | Product signal over the full run length with no catalyst, same light and same stirring | Flat within the noise band for the whole run, with no upward slope | The reactor, seals, grease or light-induced desorption from the walls |
The leak criterion is the one people relax first. At 0.1 μmol/h an eight-hour run admits under 1 μmol of oxygen equivalent, below what most experiments resolve — that is why the number sits there rather than at "no visible bubble". The degassing and blank criteria are detector-relative, so record them as absolute values with units, never as "background acceptable".
Pressurize to the top of your working range, isolate the section, and watch the gauge for longer than feels necessary — thermal equilibration after a pressure change can take tens of minutes and mimics a slow leak in the first few. A drop that continues at a constant rate after the system has settled is a leak; a drop that flattens is gas dissolving into liquid, or the vessel cooling.
Segment the system instead of hunting the whole rig. Close valves to isolate the reactor, the sampling loop, the analyzer feed and any cold trap, then hold each section against the same criterion. The segment that fails contains the fault, and you have narrowed it without opening a joint that was sealing. Repeat at operating temperature if the run is heated: a joint that holds cold can weep hot.
Most first-run leaks sit in the same five places, in roughly this order of frequency.
Chase them with a pressure hold on each segment and, where the geometry allows, leak-detection fluid on the outside of the joint; bubbles confirm the location immediately. Keep the fluid off vacuum-side joints that will later see the reaction, since the residue outgasses for weeks.
The degassing endpoint is a measurement, not a cycle count. Evacuate to the limit of your pump, purge with the carrier or reactant gas you will actually use, and repeat until the analyzer's reading for oxygen and nitrogen stops falling between cycles. Two consecutive cycles that agree within the noise band is the practical stopping rule; a fixed "three cycles" rule ignores line volume, liquid volume and how long the system sat open.
Then run the blank: same light, same stirring, same duration, no catalyst. A blank that is flat and slightly above zero is usable, and subtracting it is legitimate. A blank that rises steadily is not, because a slope is indistinguishable from a slow reaction, and subtraction cannot fix a slope that changes with temperature and light history. If the blank drifts, fix the blank before you screen catalysts.
Long enough for the pressure to stabilize after the change, then long enough to see whether the remaining drift is linear. In practice that is usually thirty minutes to an hour for a small glass rig, and longer with a large liquid volume, because dissolved gas keeps equilibrating. Record the gauge at fixed intervals rather than only at the start and the end.
No. A tight system can still produce a signal: grease, septum bleed, impurities in the water and desorption from reactor walls all generate products without a catalyst. The leak test proves the boundary is closed; only the blank proves the interior is clean. Run both, and run the blank at the light intensity you intend to use.
Usually because something in the line is still giving up gas — a seal warming up, a solvent film on the walls, or liquid that was never fully degassed. It can also be a real leak that opens only at temperature. Repeat the pressure hold hot and compare the background at the start and end of a catalyst-free run; that pair of readings separates the two causes.
