Vacuum, Materials and Gas

The window has a top because the accelerating ions can be lost to the gas they pass through. The mechanism is called charge exchange. A proton at tens of keV that passes a hydrogen molecule can pull an electron off it. A proton with an electron is a hydrogen atom, electrically neutral, and a magnetic field only steers charged particles. So the atom stops curving in the magnet’s field, travels in a straight line, and hits the chamber wall (the vacuum chapter’s opening). The proton is gone from the beam, and nothing the collector measures ever records it. A cyclotron is more exposed to charge exchange than a single pass across the chamber would be, because the beam spirals through many turns, tens of metres on a tabletop machine with of order a hundred turns (How many turns should a proton make before it reaches my target energy? Fewer turns sounds easier, but isn't the whole point of a cyclotron to accelerate over lots of turns?; the path-length calculator). Every metre travelled through gas is another chance to meet a molecule, so the surviving fraction falls exponentially with pressure times path length. The mean free path is the average distance a particle goes before it hits a gas molecule, longer at lower pressure. COLUMBUS’s textbook states the design criterion in one sentence: “the mean free path must be greater than or equal to the total path length” (Prechtl & Wolf (2020), p. 54, translated). Lower dee voltage, meaning a smaller energy gain at each gap crossing and so more turns and a longer path to the same energy, therefore demands lower pressure.

The compromise comes as two numbers. Small cyclotrons that have produced beam commonly pumped into the 10⁻⁶ to 10⁻⁵ torr range before admitting source gas, then ran about ten times higher with the gas flowing. Those are examples, not requirements; the best setting depends on the source, the machine and the beam you are after. The Rutgers 9-inch found its optimum at 5.5 × 10⁻⁵ torr of hydrogen, with less gas starving the source and more cutting the beam (Koeth, p. 6). MIT ran below 10⁻⁶ with no gas and about 2 × 10⁻⁵ with deuterium flowing (Livingston & Blewett (1962), p. 198). COLUMBUS quotes 10⁻⁶ mbar base and 10⁻⁵ with gas (Wolf & Prechtl (2022), p. 2). The Mullins machine gates its filament at 10⁻⁵ torr indicated, an interlock that keeps the filament off above that reading, and ran at 2 to 8 × 10⁻⁵ indicated with hydrogen, 4 to 17 × 10⁻⁵ of actual hydrogen after the gauge correction (an indicated reading is the gauge’s own number; the gauge is calibrated for nitrogen and reads hydrogen low; hosted chamber-vacuum note, §12). Wouters’ 1949 thresholds are 10⁻⁵ before you start and 10⁻⁴ or less to attempt operation (Wouters (1949), p. 6).

If it does not work. Sweep the gas with everything else fixed and watch the beam. Houghton’s single largest gain, an order of magnitude, came from running lower hydrogen pressure together with a smaller filament bias (the voltage held on the filament), at no added RF power (Yuly et al. (2010), p. 17-18). That was a several-variables-at-once change, so it is an illustration rather than a recipe. Beam peaks that broaden and shift as the gas goes up are consistent with the high side of the window (Yuly (2013), p. 5). The same look can come from the source or the RF changing with the gas, so the observation suggests rather than identifies. No beam at any gas setting says little about pressure on its own; the source, the resonance and the collector can each be the cause (How do I find a first beam when all these settings affect one another? Does everything have to be perfect at once before anything happens?). Reading your pressure against a published one needs three checks first: what gas the gauge was calibrated for, where it sits, and whether the published figure was corrected for hydrogen. The Mullins note says which it reports; many sources do not (Can I trust the pressure my vacuum gauge shows? Is that the pressure where the particles are travelling?). The base pressure with gas off is the other half of the diagnosis. If it will not reach the 10⁻⁵ decade (a decade is one factor-of-ten band of pressure), the gas-on pressure will not either, and the question becomes pumping, leaks or outgassing, gas leaving the chamber’s own surfaces (My pump is rated to reach a much better vacuum than I need. Why can't my chamber get there? What are pumping speed and conductance?, How clean do the chamber and the plumbing really need to be? Which of the gloves, the special solvents, the deionized-water rinse and the bake-out do I actually need?).

To understand it. The source and the chamber need different pressures, and the hardware keeps them apart. A chimney, the small gas-fed enclosure around the source, holds the discharge, the ionized gas where the ions are made, at 1 to 10 Pa inside while the chamber sits four orders of magnitude (ten thousand times) lower (the ion-sources chapter, gas-feed section). A tight-fitting insulator between discharge and chamber, leaving less opening for gas to escape, widens the ratio (Forringer (2004), p. 17-29). A source sitting open in the chamber has only about twice the chamber pressure around it, so its window collapses toward one compromise (Rovey (2007), p. 2). The pressure that matters is the one inside the dee, where the beam spends most of its time, and no gauge reads it directly. The vacuum chapter works the correction and shows it is small for an open dee and large for a closed one, whose gas can leave only through the mouth. The chapter’s worked budget at the site’s reference point, 0.6 T and 150 keV with 1 kV on the dee, gives about 90 % proton survival at 10⁻⁵ torr and 73 % at 3 × 10⁻⁵. Five times the dee voltage, a fifth of the turns and of the path, gives the same survival at five times the pressure. Those are calculator outputs under stated assumptions: every crossing at peak voltage and the pressure uniform along the orbit. The molecular ion, H₂⁺, a hydrogen molecule missing one electron, fares worse at the same pressure, because its destruction cross-section (a cross-section is the effective target area for a collision) stays high across the whole amateur energy range (the vacuum chapter; cross-sections section; I'm feeding hydrogen into the chamber. Does that mean the accelerated particles must be protons? Why might I see more than one resonance peak?). The gas-kinetic mean free path the vacuum texts print is about molecules hitting molecules and says nothing about a proton. The beam’s mean free path uses the capture cross-section, the likelihood of the electron pick-up described above; using the gas-kinetic figure in its place is the trap one census machine fell into (Dewan (2007), p. 9-11, with its dated correction).

The Mullins machine’s three-cfm refrigeration-service pump measured 98 microns (a micron is 10⁻³ torr), four times its nameplate figure (hosted chamber-vacuum note, §10). COLUMBUS’s dry diaphragm pump stops near 10⁻² mbar, where its turbopump takes over (Prechtl & Wolf (2020), p. 55-56). One census machine reached 2 × 10⁻³ torr, and its record reports no beam (the vacuum chapter, which also says what that record does not say). Diffusion pumps: Niell, the Rutgers 9-inch and 12-inch on a standard four-inch stack, four inches being the diffusion pump’s inlet diameter and its usual size label, Houghton with a liquid-nitrogen trap, the Mullins machine on a $379 surplus Edwards unit. Turbopumps: COLUMBUS and MIT (census; Koeth, p. 2; Koeth (2013), p. 1; Loucks (2007), p. 41-43; Prechtl & Wolf (2020), p. 55-56; hosted note §10).

The two differ in housekeeping, and each item has a safety side. A diffusion pump moves gas with jets of boiling oil vapour that sweep molecules toward its outlet. It has no moving parts, tolerates abuse, and is what the surplus market offers. It needs its oil level checked and its heater power right (Harvard University Cyclotron Laboratory, p. 1, The University of Washington 60-Inch Cyclotron: Progress and Status…, p. 53). Oil vapour drifts from it toward the chamber unless something stops it, and what stops it depends on the pump. Some have an internal cooled cap or baffle, others rely on an external baffle or trap; the manufacturer’s instructions for the particular pump are the authority on which it needs. And it needs its backing pressure, the pressure the roughing pump maintains at its outlet (the forepressure), held below its critical forepressure, a few tenths of a torr, or its jets collapse and the inlet pressure “rises uncontrollably” (O'Hanlon (2003), p. 232-233). It must cool before the chamber is vented, since air on hot oil degrades it. The Mullins procedure waits twenty minutes, having no gate valve in the path (hosted note §10). A liquid-nitrogen trap, the classic amateur recipe for 10⁻⁶ torr at this scale (Loucks (2007), p. 41-43), is safety equipment in its own right. The trap is a liquid-nitrogen-chilled surface between pump and chamber; oil vapour and water freeze onto it. Left cold during a leak or a vent it can condense oxygen, and it needs pressure relief and cryogen handling, so it belongs in the safety review (the vacuum chapter, pumps section). A turbopump, a high-speed turbine whose spinning blades knock gas molecules toward its outlet, has different operating, maintenance and contamination trade-offs. It is clean, it is usually ready in minutes, and with a dry backing pump it takes oil out of the high-vacuum path. Its spin-down after a stop can take anything from minutes to much longer, depending on the pump and whether it is vented. It compresses hydrogen far less well than nitrogen, so more hydrogen leaks back through it from the outlet side. Under a hydrogen load its ultimate, the lowest pressure it can hold, therefore depends strongly on the backing pressure, and the hydrogen speed is the datasheet figure to design to (the vacuum chapter; citing O’Hanlon and the USPAS notes; Prechtl & Wolf (2020)). Either way an oil-sealed roughing pump’s own oil backstreams, travels up the line against the pumping direction, into the chamber if it is left pumping the chamber alone to too low a pressure. O’Hanlon’s figure for small systems is about 100 to 150 millitorr, and the exact crossover point, the pressure at which the high-vacuum pump takes over, is set by the particular pumps and the valve sequence, not by that number (O'Hanlon (2003), p. 379-381). The roughing pump’s exhaust also carries an oil mist into the room, a health matter the safety page covers under pump exhaust.

If it does not work. The chamber will not get below the 10⁻³ decade with the high-vacuum pump supposedly working. Check that the diffusion pump is actually hot and its foreline, the line from its outlet to the roughing pump, below the critical forepressure (O'Hanlon (2003), p. 232-233), or that the turbo is at speed. A forepump (the roughing pump) that has lost its ultimate is the cheapest thing to rule out; blank off its inlet, with only a gauge on it, and read it (O'Hanlon (2003), p. 467-470). The Mullins note’s own pump-only measurement, with no chamber attached, is the number everything else is compared to (§12). It gets to the 10⁻⁵ decade and stops is a conductance (how freely the pipework passes gas), gas-load (gas entering from source, walls and leaks) or leak question rather than a pump-choice one (My pump is rated to reach a much better vacuum than I need. Why can't my chamber get there? What are pumping speed and conductance?). It was fine last month points at oil level, trap temperature, heater voltage or a vent done hot. In the Harvard quarterly, three days went to a cold-trap refrigerator and a fourth to low diffusion-pump oil; support equipment dominated the outages (Harvard University Cyclotron Laboratory, p. 1, Harvard University Cyclotron Laboratory, p. 1).

To understand it. A pump is sized by throughput, the gas it must move per second (pressure times volume flow, torr-litres per second), at the operating pressure. The source gas in torr-litres per second, divided by the pressure you want, is the effective speed you need. The port and baffle between pump and chamber take their share before the pump sees anything (Foss et al. (1952), p. 135-137; My pump is rated to reach a much better vacuum than I need. Why can't my chamber get there? What are pumping speed and conductance?). Livingston and Blewett’s rule of thumb is a litre per second of speed per litre of chamber. The roughing pump is sized to reach backing pressure in fifteen to twenty minutes (Livingston & Blewett (1962), p. 197-198). The old reports’ interlocks are worth copying in spirit. Heaters off and the high-vacuum valve closed when the forepressure rises; a bleed valve that opens when the mechanical pump (the roughing pump) stops and lets air in, so the vacuum left in the line cannot draw oil back up it (Foss et al. (1952), p. 140-142, Foss et al. (1952), p. 140). A scroll pump, an oil-free roughing pump, is clean but passes light gases, which matters when the gas is hydrogen (the vacuum chapter). The vacuum chapter’s pumps section carries the full comparison with its sources; the safety page carries pump exhaust and hydrogen.

No conductance calculation was ever made for the KF25 hose and the narrower KF16 port that connect pump and chamber (small standard vacuum fittings; the number is the nominal bore in millimetres). The note names that as a plausible reason the operating chamber sits an order of magnitude above the pump’s floor (hosted chamber-vacuum note, §10, §12, §16). The measurements alone do not apportion it. The working chamber also has more surface, more components and the source’s gas load, and each of those contributes. COLUMBUS did the arithmetic. A turbopump rated 28 L/s for hydrogen, behind 0.6 m of DN40 pipe (40 mm nominal bore), delivers 17 L/s at the chamber, a 39 % loss; for air the same pipe costs 75 % (Prechtl & Wolf (2020)). A pipe whose conductance equals the pump’s speed halves what the chamber sees, and no pipe can raise it. A short, fat duct straight to the pump is the design answer (Baumgartner (2013), p. 2).

The gas load is the other half, and on a cyclotron most of it is on purpose. When the source runs, the hydrogen you admit is the dominant gas; at MIT the ion-source gas load, not outgassing, set the working pressure (Livingston & Blewett (1962), p. 198). With the gas off, three sources remain. Water leaving the walls (outgassing), which falls off roughly as one over time for the first ten hours of pumping (O'Hanlon (2003), p. 80-81). Air diffusing through the O-rings (the vacuum-chamber chapter). And any real or virtual leak. A virtual leak is air trapped in a blind hole or under an unvented screw, bleeding out through a path too small to find. Vent every trapped volume: the Mullins dee is held by side-vented screws, screws with a slot or hole along them so trapped air has a way out, and Houghton drilled a side hole into each blind screw hole (Bertolini (2004), p. 59-63; Haas (2009), p. 52; hosted note §4). The record’s leak tally is consistent across three construction reports. Gaskets first, then welds between dissimilar metals, then anything carrying water inside the vacuum (Creutz (1950), p. 24; The University of Washington 60-Inch Cyclotron: Progress and Status…, p. 53; the vacuum chapter; seals-and-leaks section).

If it does not work. One cheap measurement gives the first hint about which of five things to touch next. Rate of rise: valve off the pump and watch the pressure climb. A rise that stays straight suggests a real leak, a constant flow from outside; a rise that flattens suggests outgassing or a virtual leak, a source that runs down (O'Hanlon (2003), p. 468-469, Bertolini (2004), p. 429-433). A steady inflow into a closed volume raises its pressure at a steady rate; a source that is emptying raises it ever more slowly. These are characteristic patterns, not verdicts: several gas sources are usually present at once, and the shape depends on how long you watch and in what state the chamber is. Blank off the pump to learn its real ultimate, then add the plumbing, then the chamber. The Mullins note’s pumps-only, empty-chamber and working-chamber readings are that sequence, taken once and kept (§12; O'Hanlon (2003), p. 467-470). Compare the pump-down curve with the one you logged last time; a curve that departs from last month’s is the first leak detector (Creutz (1950), p. 24). Mind the pump itself: forepressure over the critical value collapses a diffusion pump’s jets (O'Hanlon (2003), p. 232-233); low oil, a warm trap, the wrong heater voltage (Harvard University Cyclotron Laboratory, p. 1, The University of Washington 60-Inch Cyclotron: Progress and Status…, p. 53). Mind the gauge: a gauge on the pump port reads the pump and a gauge on a dead-end stub reads the stub. An uncorrected hydrogen reading is low by about two (Can I trust the pressure my vacuum gauge shows? Is that the pressure where the particles are travelling?). Which of these is first depends on what changed since it last worked; after a vent, outgassing is the ordinary suspect, after a rebuild, a gasket.

To understand it. The effective speed combines the pump’s speed and the pipe’s conductance as reciprocals, 1/S_eff = 1/S + 1/C. The form is that of resistors in parallel, even though the pump and pipe sit in series along the gas path (the vacuum chapter, effective-speed section). Conductance in molecular flow, the regime at these pressures in which molecules fly wall to wall without meeting one another, has closed forms. An opening passes at most 11.6 L/s per square centimetre for room-temperature air, and a long duct a fraction of that set by its diameter cubed over its length (O'Hanlon (2003), p. 48-50; Prechtl & Wolf (2020), p. 82-83; the Carnegie report’s inch-unit formulas; Foss et al. (1952), p. 144-145). Hydrogen passes every aperture 3.8 times faster than air, but in a pump-limited system that gains little, because the pump, not the duct, is the bottleneck (Foss et al. (1952), p. 138-139). Sizing is by throughput against the effective speed at the operating pressure, with margin (Foss et al. (1952), p. 135-137). The old rule of thumb is a litre per second of speed per litre of volume (Livingston & Blewett (1962), p. 197-198). Outgassing tables are for comparison only. A slightly rusty steel surface outgasses hundreds of times worse than clean aluminium or stainless, and the figures change tenfold with the hour they were read (Bertolini (2004), p. 49, O'Hanlon (2003), p. 80-81). The vacuum chapter models the whole system as a network of conductances down to the one place that matters, inside the dee (Foss et al. (1952), p. 132).

Type and range. No single gauge covers the nine decades from atmosphere to high vacuum (Prechtl & Wolf (2020), p. 56-57). Thermocouple and Pirani gauges work by heat conduction, how much heat the gas carries away from a heated wire, and are trustworthy above about 10⁻³ torr; below that too little gas remains to carry a measurable amount. Houghton used them only for the foreline and the pump changeover and read the high-vacuum side with an ion gauge, whose hot filament ionizes some of the gas and whose collected ion current rises with the amount of gas present (Cressman (2006), p. 32, Loucks (2007), p. 41-43). A hot-filament ion gauge is not switched on until the pressure is well down, half a micron at Oak Ridge, and the modern gauge’s own limit governs (Livingston & Boch (1952), p. 45); the general practice behind the rule is that the glowing filament, like a lamp filament exposed to air, is damaged by too much gas. The Mullins machine runs three: a Convectron on the foreline, a process gauge at mid-range that drifted and was recalibrated, and the ion gauge that “carried the decisions”, read as a voltage on a panel meter and converted (§11).

Calibration gas. Ion gauges are calibrated for nitrogen. Their sensitivity to hydrogen, the ion current they produce per unit of pressure, is about half, so an uncorrected reading understates hydrogen pressure by roughly two. The factor belongs to the gauge type and the gas actually present, not to all gauges and all mixtures. The instrument manual gives it (O'Hanlon (2003), p. 111; the vacuum calculator states the same). When you compare your pressure with a published one, ask whether theirs was corrected; many papers do not say.

Location. The gauge reads where it sits in the plumbing. A gauge on the pump port reads the pump; one on a dead-end stub reads the stub. Oak Ridge mounted its gauge where the paths from the pumps to the gauge and to the tank were about equal, a statement about that manifold rather than a universal rule (Livingston & Boch (1952), p. 45). The pressure that matters is inside the dee, where the beam spends most of each turn, and no gauge reads it. For an open dee the difference is small, for a closed one it is large (the vacuum chapter, inside-the-dee section).

Interference. An ion gauge works by collecting low-energy ions, and a magnetic field bends them off their path to the collector, so the reading changes. The Rutgers 9-inch’s gauge sat on a chamber port in the fringe field, the part of the magnet’s field that spills out beyond the poles, so the group set the gas pressure with the magnet off (Koeth, p. 3). The Rutgers 12-inch run-up shuts the ion gauge off before the beam run (Koeth (2006), p. 5; What should be grounded and what needs to float? I have a filament heater supply, a filament bias supply, RF drive, a biased collector plate going to an electrometer, a magnet supply, vacuum-gauge supplies, and exposed metal everywhere. Which parts can be tied together, which must be isolated, and what does "floating" actually mean?). A Penning gauge, which has no filament and reads the current of a discharge sustained between cold electrodes by its own small magnet, is more rugged and more sensitive to small changes, and probably less accurate in absolute pressure (Livingston & Boch (1952), p. 46). RF pickup into the gauge controller is the same class of problem as RF into the electrometer (What are ground loops? How do I know if I have one, and how do I avoid them in my cyclotron?).

If it does not work. Two gauges disagree: check range first, then gas, then location. A thermocouple gauge below 10⁻³ torr and an ion gauge above it are both outside their range, and a mid-range process gauge drifting out of calibration is in the Mullins record (§11, §14). The reading changes when the magnet or RF comes on: the gauge may be suffering magnetic or electrical interference, as the 9-inch concluded for its ion gauge in the fringe field (Koeth, p. 3). But the pressure may genuinely be changing too, since RF, a discharge or a heated source can release gas from surfaces. The observation alone does not distinguish the two; reading with the magnet and RF off, and comparing, is the start. The number does not match the paper: apply your gauge’s sensitivity for your gas before concluding anything, and ask whether the paper applied theirs. The beam behaves as if the pressure were higher than the gauge says: it may be, inside the dee. A dee with a solid back and a narrow mouth can run an order of magnitude, ten times, above the gauge (the vacuum chapter).

To understand it. A residual-gas analyzer tells you what the gas is, not just how much. Oxygen beside nitrogen at the atmospheric ratio (air is the two in a fixed proportion) supports an air leak, and a large water peak with a falling rate of rise is outgassing (O'Hanlon (2003), p. 471). Molflow+ computes the pressure at the beam from the geometry when the one-node arithmetic is not enough (the modelling-tools page). The amateur record contains no gauge calibration against a reference, only consistent use of one gauge in one place.

Houghton’s eight-port chamber carries a gas inlet with a needle valve (Cressman (2006), p. 39). The Rutgers 9-inch used a calibrated leak behind a regulator (the pressure-reducing valve on the cylinder) that took the lecture bottle, a small gas cylinder, down to about 10 psi (Koeth, p. 6). COLUMBUS feeds its chimney through a mass-flow controller at 0.10 to 0.20 ml/min from a regulator set to 300 mbar (Prechtl & Wolf (2020), p. 61). The Mullins machine began with a needle valve in 2017 and moved to a used one-sccm mass-flow controller (sccm: standard cubic centimetres per minute, the unit a flow controller reads in), the one significant upgrade in its record (hosted chamber-vacuum note, §8; census). Wouters’ 1949 list is the same menu: a needle valve, a deliberately loose thread, or a heated palladium leak (Wouters (1949)).

A needle valve is a fine-taper metering valve. Choose one with a vernier handle so that a setting can be written down and returned to, and log turns-open as the process setting (Parker Hannifin, p. 1-2). Give it a separate shutoff valve in series, because the manufacturer states the metering series is “not recommended for positive shut-off” (Parker Hannifin, p. 2). Its flow depends on the upstream pressure, so the regulator setting is part of the recipe. A mass-flow controller holds the flow you ask for, within its specified inlet-pressure and control range, regardless of upstream drift. With the flow known, COLUMBUS used its measured chamber pressure as a check on its pumping-speed arithmetic (Prechtl & Wolf (2020), p. 53). A leak valve sits between the two in cost and finesse. None of them controls the chamber pressure directly. The pressure changes whenever the pumping changes: as the walls load or unload with hydrogen, as the pump warms, as a filament ages. The Mullins log records that after running its controller high, “it took a while for the low vacuum to return”, hydrogen loading of the walls, walls that had taken up hydrogen giving it back for a while (§14). The hydrogen upstream of whatever valve you choose is a gas-handling problem the safety page covers. The lines are flushed after any opening to air (hosted note §13; dg-1488 gives Houghton’s purge procedure).

If it does not work. Two phases need different things from the valve. Finding the window is a sweep. Admit gas until the chamber rises by about 10⁻⁴ torr while watching the arc current (the current of the source’s discharge), which is the 1949 commissioning step (Wouters (1949)). Then move it in small steps with everything else fixed and watch the beam (What pressure should I aim for when the cyclotron is running? Don't I need more gas to make ions, but less gas to let them spiral around?). The Mullins log’s version is peak shape, the shape of the beam-current peak, against controller setting. The best peak shape came at low flow, and a wider, uglier peak with more current at ten times the flow (the hosted performance note’s knob table). Holding the point is where the device matters: a vernier needle valve returns to its mark if the regulator has not moved; a controller returns to its flow regardless. Either way the number to log is the chamber pressure that resulted, since that is what the beam saw. A lag between turning the valve and seeing the change is a long gas line, whose volume must fill or empty before the chamber sees the new flow. Ohio State’s inherited lines were long enough that beam-versus-feed measurements were not valid until they were shortened (Hausman (1954), p. 6). A setting that gives a different pressure after a vent is most often the walls reloaded with water, and it recovers with pumping time (How clean do the chamber and the plumbing really need to be? Which of the gloves, the special solvents, the deionized-water rinse and the bake-out do I actually need?). It is not the only possibility: the regulator pressure, the valve, the pump’s condition and the gauge can all have moved while the chamber was open.

Leaving the setting alone is not the same as leaving the machine alone. A fixed flow is a fixed gas load into a system whose pumping, walls and source drift, so the pressure and the peak both move during a session. The chamber pressure is read and logged on every run; the Mullins log’s pressure columns exist for that reason (§12). Unattended operation with hydrogen flowing is a safety question, not a convenience, and the hydrogen handling, isolation valve and ventilation live on the safety page.

To understand it. The steady-state balance is one line: chamber pressure equals throughput over effective speed, and COLUMBUS’s worked point lands within a few percent of its measured pressure (Prechtl & Wolf (2020), p. 51-53). One sccm is about 0.0127 torr-litres per second (the vacuum chapter, gas-load section). Published feeds run from a tenth of a sccm for a hooded arc to a few sccm for a Penning source (two ion-source designs). A discharge with good confinement, one that keeps its ionizing electrons in the gas for longer, ionizes up to half the gas fed to it, a poorly confined one a tenth (the ion-sources chapter). Running at the lowest flow that still gives usable source current is COLUMBUS’s stated practice, with a table of flow against pressure and mean free path to show why (Prechtl & Wolf (2020), p. 58). Starving a Penning source, on the other hand, flips its arc mode and brings out the molecular ion (Forringer (2004), p. 79-80). The change is in species, which kind of ion the source emits, rather than pressure (I'm feeding hydrogen into the chamber. Does that mean the accelerated particles must be protons? Why might I see more than one resonance peak?).

Houghton’s 2013 chamber is a milled aluminium ring with ten KF flanges in vacuum epoxy, reaching 2 × 10⁻⁶ torr (Yuly (2013)). Its 2006 chamber was soldered brass (Cressman (2006), p. 38), its dee stands on machinable-ceramic strips (Haas (2009)), and its first insulation was glass slides glued with vacuum epoxy (Cressman (2006), p. 41). El Cerrito’s 1947 chamber was a brass ring with a soldered bottom plate (Cressman (2006), p. 15), and Niell’s lids were annealed glass held on by air pressure against a greased ring (Cressman (2006), p. 16).

Plastics and rubber are gas sources, and the compounds differ by more than a factor of ten. Viton loses 0.07 % of its weight under vacuum (the weight lost is gas given off) where nitrile loses one to three percent, and silicone passes helium through its bulk (permeation) twenty times faster than Viton (Parker Hannifin (1998), p. 6, Parker Hannifin (1998), p. 7). Viton is therefore the small-machine default and silicone a poor one despite its temperature range. Elastomer seals add permeation and outgassing loads that can limit the attainable pressure. The accelerator notes put ordinary elastomer-sealed flange systems in the 10⁻⁶ decade in practice, better under the right conditions (Bertolini (2004), p. 443-446), and the 10⁻⁶ to 10⁻⁵ decades are where the documented first machines ran (What pressure should I aim for when the cyclotron is running? Don't I need more gas to make ions, but less gas to let them spiral around?). Thick plastic as an insulator needs thick-sample data: Teflon’s dielectric strength, the voltage per unit thickness it withstands before it breaks down, falls from 240 kV/mm in a thin film to 70 kV/mm at 5 mm (Wolf (ed.), p. 523). Adhesives have a documented success and a documented failure, and the difference is the application. The Mullins flanges are epoxied into holes bored a few thousandths over, so the bond line is thin and the epoxy is never asked to bridge a void. The note names that, and the full cure before handling, as the reason no joint has leaked in eight years, and, separately, records that the chamber cannot be baked in place, because there is no way to heat it between the poles without overheating the coils (§13). That is one adhesive in one geometry under one load at one temperature; it says an epoxied port can work, not that a structural epoxy joint is generally safe or appropriate. Houghton’s glued glass insulation could not be repaired after one dee-to-wall spark and forced a complete rebuild (Haas (2009), p. 47). Solder has carried vacuum joints since the first machines (Cressman (2006), p. 15, Cressman (2006), p. 38). It fails with heat: a silver-soldered joint “fails fast” in a discharge that keeps the electrode incandescent (Hull (1997), p. 5-8, a fusor grid), so nothing soldered goes near the filament. Workshop metals are judged on the same four properties. Aluminium, brass and copper are the usual choices for anything in the field and stainless for the wall; magnetic behaviour, corrosion, RF current and temperature decide the rest. The hot-zone limits for metals and insulators are tabulated in the ion-source handbook (Wolf (ed.), Wolf (ed.), p. 356). A rusty steel surface outgasses hundreds of times worse than a clean one (Bertolini (2004), p. 49).

If it does not work. A substitute is accepted on a comparison, and the cheapest comparison screens without giving a verdict. Pump the chamber with the part in and with it out, same pumping time, and read the gauge. The Mullins workbook recorded a “sticky bolt” test of exactly this kind after 25 minutes of pumping, an early, measured version of the cleanliness discipline in its procedures (§13). A part that passes this can still fail on voltage, heat or RF, which the other checks cover. A pressure that falls slowly for hours with the part in points to outgassing, and a rate-of-rise test helps separate it from a leak (How clean do the chamber and the plumbing really need to be? Which of the gloves, the special solvents, the deionized-water rinse and the bake-out do I actually need?, My pump is rated to reach a much better vacuum than I need. Why can't my chamber get there? What are pumping speed and conductance?). A spark or a dark track across an insulator is the voltage job failing, and the cure is geometry and cleanliness before material. Particulates on the surface dominated breakdown in the one careful study (Werner (2004), p. 78). A thin air gap between a conductor and a solid insulator sparks first, because the solid multiplies the electric field in the gap (Miley & Murali (2014)). A joint that cannot be undone is a repairability problem. Screws where possible; where glue or solder is structural, the bonded assembly becomes the replaceable unit (Haas (2009), p. 47; the vacuum-chamber chapter).

To understand it. Outgassing tables are comparative: aluminium, polished copper, stainless and rusty steel at one hour and four hours, with the instruction to use them that way (Bertolini (2004), p. 49). The rates fall roughly as one over time for the first ten hours (O'Hanlon (2003), p. 80-81). Elastomer selection is three axes, permeability, weight loss and compression set, how much the rubber stays flattened after it has been squeezed (Apple Rubber Products, p. 84), and the gland arithmetic is in the chamber chapter. Activation, a material becoming radioactive after irradiation, is the material decision for the machine this one may become. The University of Washington chose aluminium over stainless for its tank because aluminium’s activation products die away quickly where stainless’s cobalt traces make long-lived cobalt-60 (The University of Washington 60-Inch Cyclotron: Progress and Status…, p. 46). That compares two materials rather than clearing one, and that comparison concerns neutron activation, which matters where the machine makes neutrons (the experiments-by-energy pages). For welded chambers the accelerator-shop preference is 304L over 304, two stainless-steel grades (Bertolini (2004), p. 355-360), a preference with reasons; several documented chambers are soldered, epoxied or clamped instead.

Wouters’ 1949 recommendation is a wash of all tank parts before final assembly, with a modern degreaser in place of his carbon tetrachloride (Wouters (1949), p. 9). No grease on static (non-sliding) O-ring seals, since grease is not needed there and traps gas that releases in bursts (O'Hanlon (2003), p. 340). The Mullins bake of 2017, on the bench, was to find the chamber’s floor, the lowest pressure it would reach; the chamber cannot be heated between the poles without cooking the coils, and taking it out means venting it (§13). The 2024 recovery was from 2.5 × 10⁻⁵ to 6.6 × 10⁻⁶ torr after “hours of vacuum pumping” (§14). The Rutgers 9-inch routinely reached 10⁻⁵ in half an hour and below 5 × 10⁻⁶ in under two (census). Time under vacuum can be sufficient for a first beam. It is not the same as a bake, which drives off more strongly held gas faster, and it does not accomplish everything a bake would.

Cleanliness is not optional in two places, and neither is about pressure. The first is the high-voltage surfaces. In the one careful breakdown study, particles on the electrode, rather than the metal, set the breakdown voltage, the voltage at which a spark jumps the gap. Forty of fifty-two contaminated sites broke down against one of sixteen clean ones (Werner (2004), p. 78, Werner (2004), p. 84). Livingston and Blewett’s assembly rule for the dee region is no dust, grease or fingerprints, and never steel wool or coarse abrasives. After that the chamber still needs conditioning under vacuum, running voltage on the electrodes until their surfaces settle, a need that “no amount of smoothing or polishing” removes (Livingston & Blewett (1962), p. 189). The second is the hot parts. An evaporating filament and a sputtering electrode (one whose atoms are knocked loose by the ions striking it) put their own films on everything nearby; the Mullins chamber floor carries concentric heat-tint rings centred on the filament, cleaned off once or twice over the years (hosted rf-dee-source note, §6.2).

If it does not work. The symptom that gets blamed on cleanliness has other causes, and one measurement gives a hint about which. A pressure that keeps falling for hours after a vent is mostly water leaving the walls, and the cure is time or heat (O'Hanlon (2003), p. 80-81). A pressure that stops at a floor and stays there can be a leak, a pump limit, or outgassing that has reached a steady balance against the pumping. A floor by itself does not say which. Valve off the pump and watch: a rise that stays straight suggests a leak, a rise that flattens suggests outgassing or trapped gas, and mixed cases are common (O'Hanlon (2003), p. 468-469, Bertolini (2004), p. 429-433). A chamber that was fine and now is not, with no vent in between, is rarely cleanliness. The Mullins log has no leak-hunting episode in seven years, and every “bad vacuum” entry traces to time open to air or to a fresh filament outgassing (§14). A solvent that made things worse is a compatibility problem. Solvent washing does not help an O-ring, whose water comes back on every air exposure (O'Hanlon (2003), p. 340). The generic cleaning sequence in the accelerator notes is explicitly a sequence of baths each matched to the material (Bertolini (2004), p. 497-513). Check the elastomer, the plastic and the adhesive against the solvent before it touches them. Conditioning is not cleaning. Running RF on the dee does drive gas off the electrodes and settles their surfaces. Fusor builders describe the same pressure rise and fall when new hardware is first run hot (Hull (1997), p. 5-6). But operating an apparatus capable of discharges, X-rays and electrode damage is not a general substitute for cleaning it first; the controlled version, with its limits and safety controls, is the conditioning entry (People talk about RF "conditioning". What does that mean, and how do I tell conditioning from a fault that means I should stop?).

To understand it. The accelerator-shop sequence, for the machine this one may become, runs: mechanical cleaning, acetone for tape and ink, detergent, rinses, a deionized rinse to a stated resistivity (purer water conducts less), filtered dry, lint-free wrap, and bake-out last (Bertolini (2004), p. 497-513). Outgassing rates are quoted per unit area and per hour of pumping, and they differ tenfold between the first and tenth hour (O'Hanlon (2003), p. 80-81). The tables are therefore comparative (Bertolini (2004), p. 49). A Viton ring baked four hours at 150 °C and pumped twelve drops its outgassing 2,500-fold, until the next air exposure reloads it (O'Hanlon (2003), p. 340). Adsorbed gas, gas molecules stuck to a surface, rebuilds a monolayer, a layer one molecule thick, in seconds at 10⁻⁶ torr (Bertolini (2004), p. 42-43), so “clean” is a state the pump maintains. Monolayer physics, electropolishing and ultra-high-vacuum practice belong to the vacuum chapter, beyond a first beam; a residue-free rinse is needed earlier wherever a surface must hold voltage.