The Vacuum Chamber and Its Seals

A cyclotron's chamber serves two masters at once. Outside it is the magnet gap, where every millimetre of lid steel or clearance costs field; inside it is the vacuum the beam needs, held by joints that must survive repeated opening. That makes the vessel a budget exercise (gap, deflection, ports, clearances all drawn from the same few millimetres) and makes its reliability a matter of joint engineering: gland dimensions, elastomer choice, weld geometry, and the feedthroughs that carry motion and power through the wall. Both halves can be budgeted before metal is cut and verified by test, and the published record shows what happens when they are not: a chamber that could not hold vacuum rebuilt from scratch, a glued assembly that one spark turned into a total loss, lids that bowed until the RF detuned. The design guide's chamber domain holds 47 sourced rules and its seals domain 43 more; the lid-deflection calculator is this page's worked tool.

Elsewhere: gas loads, pumping speed and the pressure target are the vacuum budget. That page owns the gas; this one owns the vessel. Gap-versus-field physics is magnet design; dee and RF electrical design is dee coupling; what the beam does inside is beam dynamics.

The chamber serves two masters, and the magnet always wins

For a gap-dominated, unsaturated magnet the field follows B ≈ μ₀NI/g: at fixed ampere-turns, every millimetre taken out of the pole gap comes back as field, and for a nonrelativistic machine energy at fixed radius scales as B². The builders of a 2 MeV amateur machine put it plainly: "This tight spacing made the design of the vacuum chamber more difficult, but it was essential" (dg-130). The chamber is what the gap squeezes. The professional literature frames the same trade from both sides: a small gap cuts ampere-turns and lets orbits run close to the pole edge, but leaves no room for the source, probes and pumping, and is sensitive to errors; a large gap buys space for everything at the price of field and power (dg-163, dg-076).

So the chamber design starts as a stack-up. Zickler's aperture budget for accelerator magnets is the form to copy: required aperture = good-field region + vacuum chamber wall (0.3–2 mm in his table) + installation and alignment margin (0–5 mm), with a further closed-orbit allowance inside the good-field region (dg-082). Sum the terms in one consistent convention: full-gap or half-gap, but not a mixture, which double-counts the allowances. For a cyclotron the stack is: internal height the dee and beam need, plus two lid thicknesses, plus the clearances on both sides of the dee, plus the chamber-to-pole clearance and whatever fasteners or flanges protrude. The total is the pole gap the magnet must energize.

The oldest advice on the layout still stands. Wouters' 1949 recommendations for small cyclotrons: make the chamber top and bottom "thin, circular steel plates … to decrease the magnetic gap as much as possible," and make the side wall non-magnetic — "preferably brass" — so the field is not bypassed around the gap (dg-129). With steel lids the plates become part of the magnetic circuit; a Houghton College thesis modelled the aggressive version, ferromagnetic stainless lids reaching 2.2 cm past the poles that act as wide pole faces, cutting the effective gap from 3.9 to 2.54 cm and raising the modelled field from 1.27 to 1.77 T (dg-141). That is a model of one geometry, not a measurement, and it presumes a stainless grade that is actually ferromagnetic, which the common 300-series chamber alloys are not; but it shows the leverage.

The neatest packaging solution in the record is COLUMBUS's: a 150 mm diameter, 12 mm deep recess milled into the chamber lid, into which the upper pole is lowered. The recess gives the chamber a fixed seat in the magnet, houses the Hall probe, and brought the chamber height down to about 72 mm inside a minimum pole spacing of about 75 mm (dg-1381): the lid doing double duty as vacuum wall and pole-gap liner, with roughly three millimetres of the stack left over for everything else.

Thin lids bow, and bowing is an RF problem before it is a vacuum problem

Atmosphere is a heavy tool. A 300 mm evacuated lid carries about 7.2 kN (some 730 kg-force), and even the 8-inch-class chambers of the site's reference point carry a few hundred kilograms per lid. A flat plate answers that load by deflecting, and the deflection is set almost entirely by thickness: plate stiffness goes as D ∝ Et³, so halving the lid thickness multiplies the sag roughly eightfold, and switching to a stronger aluminium alloy changes it almost not at all: 7075-T6 buys yield margin over 6061-T6, not stiffness, because their elastic moduli are nearly identical (dg-490). The lid-deflection calculator runs the circular Roark cases; the worked example from the record, a 10 cm radius aluminium lid 3.5 mm thick under full atmospheric load, deflects about 0.6 mm with a truly clamped edge, and about 2.3 mm if the edge is merely rested and bolted. The edge condition is worth a factor of four; check which one the real flange provides. Deflection is only the serviceability check: stress and external-pressure buckling for the actual geometry and alloy are their own calculations (dg-1407). And the formula has a measured check in the census: the Mullins Cyclotron's build log records its 0.200-inch steel lids designed for 0.030 in of centre deflection under vacuum and measured at 0.033 in: plate theory within ten percent, the record's one predicted-versus-measured pair.

Why the millimetre matters: the lid is one plate of the dee-to-lid capacitance. The Cyclotron Kids' machine documented the failure chain: its thin lids bowed inward under vacuum, "which affected the capacitance of the dee and reduced the maximum voltage that the dee could withstand before flashing over" (dg-283). A lid that moves detunes the resonant RF system and eats the flashover clearance at the same time. Their fix, tack-welded internal support posts, has prior art at every scale: steel rods carrying the atmospheric load so the plates can stay thin (dg-468), with the caveat that ferromagnetic posts in or near the useful field distort or shunt it, so posts go outside the beam's region or are made non-magnetic, and the plate arithmetic gets done anyway.

Ferromagnetic lids add a second load the vacuum arithmetic never sees. The University of Washington's model study measured the magnetic pull on its mild-steel tank cover plates at more than 35 tons, against 24 tons of atmospheric force (dg-1338): half again the vacuum load. The scaling survives to the tabletop: magnetic surface pressure is B²/2μ₀, which at 0.59 T is about 138 kPa, or 1.4 atmospheres. A steel lid on a small machine sees magnetic surface pressure of the same order as the atmospheric load (the net force depends on the actual field map and geometry and must be integrated for the real lid); it appears and disappears with the coil current, and assembly and disassembly fight it. A non-magnetic lid opts out of the term entirely; a magnetic one gets checked in both states.

The extreme thin-lid construction in the record is glass: a historical account, not a qualified design, and not one to copy without the engineering it never had. Fred Niell's high-school machine used annealed glass plates as chamber faces, held on by nothing but external air pressure against a greased steel ring, and Knox College sealed its top plate the same way (dg-1446). Atmosphere supplies about 10 N/cm² of clamping for free; the same pressure is an implosion load, nearly 1.8 kN across the face of a 15 cm glass lid. Stress and deflection for the actual glass, positive retention against venting transients, and a shield between glass and bystanders come before the visual access pays for itself.

Material choice is magnetic first, nuclear second, weldable third

The magnetic test comes before any other property, because it is the one a vacuum catalogue will not flag. Everything living in the field must be magnetically transparent — "aluminum, copper, or brass" in the MIT thesis's phrasing (dg-491) — or part of the magnetic circuit on purpose, like Wouters' steel lids (dg-129). The trap is the hardware in between: screws, feedthrough bodies, gauge fittings and off-the-shelf KF clamps are frequently steel of unstated grade. A hand magnet at the bench, before installation, is the first screen: cheap enough to apply to every part; a part whose permeability actually matters gets bought to a specified grade instead. Choosing a deliberately magnetic lid is a legitimate design (it buys gap, per above), but it couples the chamber to the field map, which is why Creutz's synchrocyclotron chamber was built to split and withdraw "without disturbing the magnet pole tips," so chamber service never invalidated the shimming (dg-645).

The nuclear consideration is the one that matters for the machine the builder hopes this one becomes. The University of Washington chose its tank material with the end of the machine's life in mind: "Aluminum was chosen over stainless steel because of its short half-life property" (dg-1345): their reasoning being that aluminium's dominant activation products die away quickly where stainless steel's cobalt traces produce long-lived ⁶⁰Co. The comparison is between materials, not an absolution: fast neutrons make 15-hour ²⁴Na in aluminium, and alloying elements add their own products. The lower a machine's energy sits below its materials' neutron-producing reactions, the smaller the stake, with the scoping that "below threshold" is a statement about specific reactions in specific nuclides, not a blanket clearance. A machine aimed at neutron-producing energies should pick its beam-facing metal the way UW did. The thresholds themselves are on the experiments-by-energy pages.

Weldability is third because it can be designed around entirely: several chambers in the record are soldered, epoxied or clamped rather than welded. Where welds are used, the accelerator-shop preference is 304L over 304, the low-carbon grade resisting weld sensitization; the USPAS notes call it "most commonly used in vacuum," a preference with reasons rather than an exclusion, and put the real leak-tightness in joint design and cleanliness (dg-489). TIG-welded aluminium has been adequate vacuum practice since at least UW's 1951 tank (dg-1345). What welding does to thin flat chambers is its own section below.

Ports are the budget nobody allocates enough of

Count the services before drawing the wall. The Houghton College record carries a complete port budget for a minimal gas-fed machine with internal diagnostics: two glass viewports, one power feedthrough for the dee, one multi-conductor feedthrough shared by the filament and the dummy-dee ground, a Faraday-collector port, a gas inlet with needle valve, an ion-gauge port, and a pumping port: eight, each with a named catalogue part (dg-1483). COLUMBUS's chamber carries ten radial ports (dg-1407); the later Houghton chamber, ten KF-16s (dg-472). Port count is the classic regret on small chambers: the service that was not foreseen (a second probe, an RF pickup, a second gauge) has nowhere to enter, so spares cost least at design time.

Where the ports go matters as much as how many. The Houghton eight-port design put every penetration on the cylinder wall, leaving both removable lids free of vacuum services, which is what made the lids simple to re-machine or replace (dg-1481). Ports on a lid travel with the lid at every opening; ports on the wall stay put. And the wall itself need not be welded to accept them: the 2013 Houghton chamber's ten KF-16 flanges were secured with vacuum epoxy into a milled aluminium ring, and the chamber reached 2 × 10⁻⁶ torr (dg-472): a fabrication route open to a shop with a mill and no TIG set, provided the joints are leak-checked as workmanship rather than trusted as design. The pattern has an independent twin in the census: the Mullins Cyclotron's chamber is a ring cut from 10-inch schedule-80S stainless pipe with ten KF16 flanges set in the same Hysol 1C vacuum epoxy, operating in the 10⁻⁶-torr range since 2018 per its build log. Two builds of the same pattern, arrived at separately; both work.

Serviceability is a layout decision, not a virtue. Argonne mounted its entire dee assembly on a motor-driven rail carriage so the dee system rolls out of the chamber for maintenance (dg-492); the tabletop translation is a chamber that slides out of the gap, a dee that comes out through a lid, a pump cart that disconnects. The counter-example is equally documented: the 2006 Houghton chamber's glued glass insulation could not be repaired after a dee-to-wall spark — "impossible to fix a single component … without replacing the entire piece" — and the machine needed a complete rebuild (dg-355). Screws where possible; where adhesive or solder is structural, the bonded assembly itself becomes the replaceable unit. Small reliefs earn their keep the same way: Houghton milled a 0.3 cm depression into a lid specifically to stop a repeat filament-to-lid discharge (dg-404). A pocket is cheaper than a spark hunt.

Two staging strategies bracket the sizing question, both documented. Rutgers built its 22.9 cm prototype's stainless chamber — ports, flanges and all — to the dimensions of the 30.5 cm machine it planned next, and the final cyclotron reused the chamber with only a new ion source (dg-1451). The Houghton program ran the opposite way: an as-built chamber smaller than the magnet's full pole diameter, with the larger chamber deferred to "longer range plans" (dg-1498). Oversizing preserves the highest-labour component across the upgrade; undersizing reaches first beam sooner. Which rework hurts more is a per-project fact. And for a school build, the hardest items may not need buying at all: COLUMBUS's magnet was donated by Jülich and its ten-port chamber fabricated free of charge by a vacuum-component firm (dg-1518).

The chamber is also an electrode

The dee runs at kilovolts of RF a few millimetres from a grounded lid, which makes the chamber wall the other half of a vacuum-insulated capacitor. The historical calibration points: MIT's 1.25-inch dee-to-lid clearance capped its dee voltage at about 70 kV by breakdown (dg-247), and Oak Ridge's ORIC design took 1.5 inches of dee-to-liner clearance for 100 kV peak, at the minimum, "since the magnetic gap is so precious" (dg-321). Neither number is a scaling law; there is no universal safe kV per centimetre for vacuum gaps, because breakdown rides on edges, finish, contamination and conditioning state. A kilovolt-class dee has large margin against these breakdown figures specifically (margin against one failure mode, not a pass on the electrical design), and a 10–15 kV upgrade re-opens the analysis for every element in the path, where the weakest element is often not a gap at all: COLUMBUS's dee amplitude is bounded at 3 kV by the voltage rating of its vacuum feedthrough (dg-1388). Rate the whole path for peak RF, in vacuum, with margin: feedthrough, stem supports, matching network.

Whatever the clearances, a freshly opened chamber must be conditioned. Livingston and Blewett: "no amount of smoothing or polishing will eliminate the necessity of some high-voltage conditioning under vacuum": assemble clean (dust controlled, all grease and fingerprints off, never steel wool or coarse abrasives), and the 1962 practice chapter asks the oscillator to drive through the sparking without manual resets while the surfaces settle (dg-253). A modern build meets the same requirement with current limiting, monitoring and supervision rather than brute persistence. Conditioning time after each opening is part of the machine's duty cycle; budget for it. Creepage along insulators is managed the 1949 way: Wouters slid a Pyrex sleeve over the dee support rod from the dee edge to at least two inches past the vacuum seal (dg-352), though a modern build sizes stem insulation from peak voltage and flashover behaviour rather than copying the geometry.

Two RF-geometry traps live specifically in the chamber's shape; both matter more as dee voltage and machine size grow, and cost little to check. The assembled pole-chamber structure is itself a cavity: the 184-inch found a re-entrant resonator mode between its pole pieces with the tank walls as return circuit, suppressed by strapping the poles together (dg-666). Sweep or model the assembled machine, and bond the poles only if a mode actually lands in band, since straps perturb the intended RF structure too. And long open paths along the magnetic field sustain oscillating-electron discharges: at 20 MHz an electron gains ionizing energy over centimetres, and the 184-inch's trouble came from its ~20 cm paths while the short dee-region gaps stayed quiet (dg-681). Keep open RF-exposed volumes small or shielded; lower frequency makes it worse.

Chambers from the record

The census's chambers, read together, are a design course in themselves. Each entry below is cited to its public record on the builds page.

• El Cerrito (1947) — the rebuild that defined the pattern. The first chamber was curved copper sheets clamped and gasketed around the magnet poles; it made 1.5 µA but "could not maintain vacuum." Its replacement — a rigid 16.5 cm brass ring, bottom plate soldered on permanently, top plate screwed down onto a rubber gasket — carried the beam to 7 µA (dg-1439). One permanent seal, one demountable seal, a rigid ring between them: the architecture most small chambers have used since.
• Niell (1994–95) and Knox (2001) — atmosphere as the clamp. Glass lids and a top plate held on by external air pressure alone (dg-1446). Knox's improvised penetrations — nylon plugs, brass screws and sealant for feedthroughs, drilled rubber stoppers for gas lines (dg-1461) — mark the other boundary: the machine never ran to publication, and a leak-checked commercial feedthrough retires that whole liability class for tens of dollars.
• Houghton College (2006 → 2013) — three documented generations. The 2006 build: an eight-port ring of soldered brass, lids of 6.4 mm aluminium with milled Viton grooves, every port on the wall so the lids stay clean (dg-1481, dg-1482). Its glued insulation cost a full rebuild after one spark (dg-355). The 2013 chamber: a milled aluminium ring, ten KF-16 flanges set in vacuum epoxy, 6.5 mm lids, 2 × 10⁻⁶ torr (dg-472), and a dee assembly standing on three KF-16 feedthroughs with vented screws throughout (dg-358).
• Cyclotron Kids (2008–13) — the weld-distortion lesson. Welding the bottom plate onto their thin flat chamber pulled the whole frame out of shape; the cure was grinding the weld off and sealing with a flat Viton gasket instead (dg-482). Their bowing-lid RF episode is above; their gap-first doctrine (dg-130) is the reason the chamber was hard in the first place.
• COLUMBUS (2013–) — packaging depth. A 200 mm chamber rolled from 2 mm stainless tube, ISO200 lid on four claw clamps, ten radial ports (dg-1407), and the recessed lid that seats the upper pole and holds the Hall probe (dg-1381).
• Rutgers (1995–2001) — the chamber that outlived its magnet. The 9-inch prototype's stainless chamber was built for the 12-inch machine from the start and served both (dg-1451).
• Mullins (2017–) — the pattern twin, with the measured numbers. A ring cut from 10-inch schedule-80S stainless pipe, 0.200-inch lids, ten epoxied KF16 ports, a Viton lid gland at a documented 42% squeeze, and the record's one predicted-versus-measured lid-deflection pair (0.030 in designed, 0.033 in measured), running in the 10⁻⁶-torr range since 2018 per its build log.
• The printed chamber (2019) — from the COLUMBUS group's vacuum partner: a laser-printed 316L base body with CFD-shaped heating/cooling channels integrated into the wall, finished by welding on standard commercial flanges (dg-1543, dg-1542). Printed-to-wrought welds needed no rework (dg-1537), and in the high-vacuum tests the measured leakage was "essentially attributable to the Viton flange gaskets": the printed walls were not the limit (dg-1544).

Gland design is arithmetic, not folklore

An O-ring seals because the gland squeezes it to a computed interference; every number in the joint is in a manufacturer's chart, and the charts disagree with folklore in useful ways. The accelerator-vacuum recommendation is face-type grooves with "as heavy a squeeze as possible" (dg-459). For the standard 1/8-inch (0.139 in) cross-section, Parker's face-seal chart gives gland depth 0.101–0.107 in (20–30% squeeze), with the vacuum groove width narrower than the liquid-service column (dg-429). Squeeze does real work: in Parker's own butyl face-seal test, going from 15% to 30% to 50% squeeze cut the helium leak rate steeply (dg-434), and Apple Rubber sanctions up to 40% for vacuum if the groove is widened to take the displaced volume (dg-478). The census carries a working point beyond even that: the Mullins Cyclotron's lid gland is documented at 42% squeeze and has sealed in service since 2018: a single in-service datum on the heavy-squeeze side, and any gland cut past the chart still has to survive the fill arithmetic that follows. The absolute geometric limit is fill: ring volume at worst-case tolerance must stay under gland volume, or thermal expansion finishes the crushing; compression set, extrusion and assembly damage cap real designs short of that bound (dg-484, dg-480).

The details that decide whether the chart numbers arrive intact on the metal:

• Locate the groove by the diameter the pressure pushes toward. A vacuum chamber is external pressure (atmosphere shoves the ring inward), so the groove's inside diameter is the controlled dimension, sized to the ring's mean ID (dg-431), with the ring itself at 1–5% stretch, 2% ideal (dg-483).
• Finish 16 RMS for vacuum, with a circular lay. The sealing face spec is 16 RMS for vacuum and gases against 32 for liquids (dg-430, dg-485), and tool marks perpendicular to the seal line are built-in leak paths: face the seat on a lathe so the lay runs concentric, and never sand a groove crosswise (dg-436).
• Non-circular grooves have their own rules. A round ring in a rectangular groove needs inside corner radii of at least three cross-section diameters, and ring length matched to groove centreline length (dg-487). Dovetail grooves hold the ring in a lid that opens; their corner radius is the critical dimension (dg-481).
• The squeeze must be bolted in. O'Hanlon's figures: 15–20% compression for Viton, minimum initial contact pressure around 13 kg/cm², which for a 3.2 mm ring at 75 Shore is about 2.7 kg per centimetre of ring, some 215 kg of clamping for a 10-inch-diameter seal before atmosphere helps (dg-442). And bolt groups interact elastically: tightening later bolts relaxes earlier ones, in some cases almost completely; the cross pattern in multiple passes with a recheck of the first bolts is the standard counter (dg-479). A lid seal that leaks in a different place after each reassembly is describing its bolt pattern.

Two older gland patterns from the cyclotron record are still worth knowing. Oak Ridge sealed its large flanges with a continuous square-section gasket in a groove cut to swallow the entire gasket under pressure, so the metal faces land metal-to-metal: a hard stop that makes reassembly repeatable (dg-473). And the 1962 practice chapter adds the RF detail elastomer charts never carry: where RF current must cross a gasketed joint, lay a thin copper-foil strip half over the gasket to carry it (dg-255): the cure for mysterious Q loss and local heating at a sealed joint in the RF path.

The elastomer is a gas source and a gas path

The compound is chosen on the seal guide's three vacuum axes (gas permeability, weight loss under vacuum, compression-set resistance; dg-478), plus the temperature and chemistry of the service in question (dg-1346) — and the spread between compounds is larger than most builders expect. Helium permeability at room temperature spans a factor of ~37: butyl and neoprene at 6.5 (×10⁻⁸ std cc·cm/cm²·s·bar), nitrile 8.0, fluorocarbon (Viton) 12.7, EPDM 19.7, silicone 238 (dg-438). Vacuum weight loss splits the field by an order of magnitude: fluorocarbon 0.07%, neoprene 0.13%, butyl 0.18%, against nitrile's 1.06–3.45% (dg-437). Viton's position on both tables plus its temperature and compression-set behaviour is why it is the small- machine default. Historical note for readers of the old reports: UW's 1951 "Hycar" gaskets are the nitrile family, not a Viton ancestor (dg-1346); the 1962 practice preferred neoprene because most rubbers of the day had "unacceptable vapor pressures" (dg-255).

An installed ring is a gas load before it is a leak. Unbaked Viton outgasses about 10⁻³ Pa·m/s initially; a 4-hour 150 °C bake plus 12 hours of pumping brings it down 2500-fold, re-exposure to air reloads the elastomer with water, and solvent washing does not substitute (dg-443). Below the outgassing sits the permeation floor: atmosphere diffusing through the ring itself, estimable from Parker's formula (dg-435). That floor is why all-elastomer flange systems ordinarily run in the 10⁻⁶-torr regime whatever their rating says (well-designed ones go lower, but permeation and outgassing usually hold operation near it), and why a joint that must ever bake hotter or hold UHV gets a metal-sealed CF flange from the start (dg-454). The printed-chamber tests above measured exactly this hierarchy: the Viton gaskets, not the walls, set the leakage (dg-1544). Whether gaskets dominate your gas load is a budget question; the vacuum page owns that arithmetic.

Permeation also fools the leak detector. Helium walks through a typical Viton ring in about 20 minutes (faster hot), so during helium leak checking the background creeps up and will not fall until the gaskets degas; O'Hanlon's advice is literally to take a coffee break rather than chase the phantom, and never to leak check during bakeout (dg-448). On silicone the effect is an order of magnitude worse, which is half the reason silicone rings are a poor vacuum choice despite their temperature range (dg-438).

The grease question has a chart answer. Grease is not needed for a static elastomer-to-metal seal, and it traps gas pockets that release as pressure bursts (dg-444); in Parker's squeeze test the benefit of grease shrank as squeeze rose and was undetectable at 50% (dg-434). Grease is the compensation for a light squeeze or a damaged seat: a repair-queue flag, not a practice. Where a thin film is justified (dynamic seals, a scratched flange awaiting repair), the one firm compatibility rule is never to lubricate a ring with its own chemistry: like dissolves like, silicone grease on a silicone ring being the classic mistake (dg-486). And gloves throughout — fingerprints are contamination (dg-444).

Two O-rings and a pumped interspace

The heaviest tool in the sealing kit is the double gasket. Two concentric O-rings with a pump-out connection between them "makes it possible to test the seal for vacuum-tightness quickly and with certainty" (dg-433): valve a gauge or sniffer onto the interspace and the joint interrogates itself, no spraying required. Held at rough vacuum continuously, the interspace becomes a guard vacuum: the pressure differential across the inner ring drops from an atmosphere to millitorr, attacking the permeation term directly (dg-459). The pattern is old cyclotron doctrine: Carnegie Tech's component reports specify welding-neck flanges with double O-ring grooves and a pump-out port as the standard demountable joint (dg-848), and UW used "double gaskets with a pump-out space … wherever possible" across its whole 4800-litre envelope, with a leak record to match: its leaks appeared at two stuffing-box seals and one flat gasket, none at a double-gasket joint (dg-1346). One system's tally is evidence, not proof; the design logic stands on its own. On a tabletop machine the complexity pays best on the largest lid and any joint opened often: one extra groove and a hose barb, against the worst leak hunts the machine will ever generate. The inner ring's own outgassing remains either way: the floor drops, it does not disappear (dg-459).

Welds fail as geometry: virtual leaks and distortion

A virtual leak is trapped atmosphere bleeding into the chamber through a path too small to find and too large to ignore: the pump-down that stalls for hours with no leak the sniffer can see. The classic culprits are a catalogue of innocent-looking geometry: unvented screws in blind tapped holes, two welds in series enclosing a void, unvented double O-rings (dg-458). (The guard-vacuum joint above escapes the third culprit precisely because its interspace is pumped.) The rule that retires the whole class: vent every trapped volume. Every internal socket-head screw gets a drilled vent, a slotted thread or a vented washer; a slot in the groove floor vents an O-ring's dead space; and pole-tip liners get drilled through "to facilitate vacuum pumping," as Carnegie Tech's chamber did (dg-647).

Weld discipline is the same idea applied with a torch. Argonne's chamber used single continuous seams welded on the atmosphere side only, with internal braces stagger-welded "to keep virtual leaks at a minimum" (dg-461). The invariant behind their practice is the question to ask of every joint: is there a pocket sealed on both sides, or a crevice open to vacuum? Vent it, or weld it through; atmosphere-side-only seams are one way of satisfying it. Distortion is the other weld failure, and thin flat chambers are its natural victims: the Cyclotron Kids' bottom plate contracted enough in welding to bend the whole frame, and the working fix was to grind the weld off and go demountable (dg-482). Material choice (304L, dg-489) matters less than these geometric habits: when Carnegie Tech leak-tested its finished chamber, the only two leaks found were both at gasket seals, attributed to non-uniform gasket stock, not at a weld (dg-649). That tally is also the standing prior for any leak hunt: check the gasketed joints first.

Everything that moves through the wall is its own failure class

Static seals fail at assembly time; motion seals fail on schedule. The NRL cyclotron's operations reports single them out: the chevron-stack seals on its source drive mechanisms "were unreliable and displayed a short life expectancy" (dg-1148). A moving shaft into vacuum deserves a real specification: compound, gland dimensions and squeeze (dynamic service tops out near 30% where static seals take 40%, dg-485), shaft finish, lubrication, and the motion profile it must survive — not a static gland pressed into dynamic duty.

The patterns that work are all in the record. The classic is the Wilson seal, an elastomer disc or O-ring gripping a polished sliding shaft: it is how Berkeley put an auxiliary beam probe into the 184-inch tank with its radial depth adjustable under vacuum, the probe itself a copper finger inside a grounded RF shield (dg-689); the same construction carries a Faraday cup or target manipulator on a tabletop machine. For rotation, the ferrofluidic seal (a magnetically retained low-vapour-pressure liquid in thousandths-of-an-inch gaps, commercial since 1971) holds a full atmosphere with no sliding contact (dg-1093); near a cyclotron, its behaviour in the stray field is part of the purchase spec. Electrical penetrations are the easiest call in the chamber: Knox's nylon-plug-and-sealant improvisations (dg-1461) against Houghton's catalogue KF-16 feedthroughs sturdy enough to carry the dee assembly itself (dg-358): a leak-checked commercial part, rated for the voltage with creepage to spare, costs tens of dollars and replaces that whole liability class with a data sheet, its ratings still checked like any other component's. Whatever enters the wall, the chamber gets reconditioned after: every opening restarts the spark-conditioning clock (dg-253).

Go deeper

Before cutting metal, in the order the budget runs:

1. Write the gap stack-up in one convention: internal height + lids + clearances = pole gap (dg-082).
2. Size the lids with the lid-deflection calculator, both edge conditions; add the magnetic term if the lid is ferromagnetic (dg-1338).
3. Hand-magnet every part bound for the gap (dg-491); choose wall and lid materials for the machine this one hopes to become (dg-1345).
4. List every service, then add spares. Wall ports leave the lids free (dg-1481, dg-1483).
5. Cut glands from the manufacturer's vacuum chart: squeeze, fill, finish, stretch, groove located by ID (dg-429–431, dg-484).
6. Default Viton; bake what can be baked; no grease on static seals (dg-443, dg-444).
7. Consider a double-gasket groove with pump-out on the main lid (dg-433).
8. Vent every trapped volume: screws, liners, dead grooves (dg-458).
9. Torque the lid in cross-pattern passes; recheck the first bolts (dg-479).
10. Leak-hunt gaskets first (dg-649), and know the helium-through-Viton trap (dg-448).

• Lid-deflection calculator — the Roark circular-plate cases with both edge conditions; share a sized lid by URL.
• Design guide: chamber domain (47 rules) and seals domain (43 rules) — the sourced corpus behind this page.
• The vacuum budget — what the sealed vessel must then achieve: gas loads, effective speed, the pressure the beam needs.
• TID-454 and AECU-1951 — the two most chamber-rich hosted classics: Carnegie Tech's component design reports and UW's construction report with the double-gasket doctrine.
• Resources — the general vacuum-practice communities, O'Hanlon, and surplus sources.

Sources

• L. F. Wouters, General Recommendations for Design of Small Cyclotrons, UCRL-476, 1949 — chamber architecture p. 5, stem sleeve p. 7. Hosted.
• M. S. Livingston & J. P. Blewett, Particle Accelerators, McGraw-Hill, 1962 — clearances and conditioning pp. 175, 189; gaskets and double seals pp. 199–201. Library entry.
• Parker Hannifin, O-Ring Vacuum Sealing (5705B, 1998) and the O-Ring Handbook face-seal chart — gland charts, squeeze test, permeation formula, dovetails. Library entries lib-033, lib-011.
• J. F. O'Hanlon, A User's Guide to Vacuum Technology, 3rd ed., Wiley, 2003 — elastomer compression, outgassing and bake figures, helium permeation, pp. 337–340, 469–470. Library entry.
• L. R. Bertolini, Accelerator Vacuum and Mechanical Engineering, USPAS course notes, 2004 — flange ratings, virtual-leak catalogue, O-ring recommendations, materials. Library entry.
• Apple Rubber Products, Seal Design Guide — stretch, fill, finish, vacuum compounds, dynamic seals. Library entry.
• Fastenal, Technical Reference Guide, Rev. 9, 2005 — bolt elastic interaction, p. 26. Library entry.
• The University of Washington 60 Inch Cyclotron, AECU-1951, 1951 — magnetic force on covers p. 30, material choice p. 46, double-gasket practice and leak record p. 51. Hosted.
• R. S. Livingston et al., The Oak Ridge 86-Inch Cyclotron, ORNL-1196, 1952 — square-gasket flange seals p. 41. Hosted.
• The Argonne 60-Inch Cyclotron, ANL-5907, 1959 — weld discipline and service carriage pp. 6–7. Hosted.
• Cyclotron Component Design Technical Reports, TID-454, 1952 — welding-neck double-O-ring flanges p. 141. Hosted.
• E. Creutz et al., Design and Construction of Synchro-Cyclotron, NYO-780, 1952 — split chamber p. 15, liner venting p. 19, leak tally p. 24. Hosted.
• K. R. MacKenzie et al., UCRL-64, 1948 — pole-gap cavity mode p. 22 (hosted); MDDC-1045, 1946 — along-field discharge paths p. 12 (hosted); the 184-inch commissioning papers, 1946–47 — Wilson-seal probe (hosted).
• The Oak Ridge Relativistic Isochronous Cyclotron, ORNL-2648, 1958 — dee-to-liner clearance p. 100. Hosted.
• Cyclotron Branch, NRL, Report of Cyclotron Operation, July–December 1969, NRL-MR-2103 — drive-mechanism seal failures p. 26. Hosted.
• Rainwater et al., The Columbia University Nevis Synchrocyclotron Major Modification, NEVIS-189, 1971 — ferrofluidic seal p. 10. Library entry.
• T. Zickler, "Basic design and engineering of normal-conducting, iron-dominated electromagnets," 2010 — aperture budget p. 14; D. Zaremba and W. Beeckman's cyclotron-magnet lectures — the gap trade. Library entries lib-024, lib-032, lib-078.
• Houghton College theses and papers: P. Cressman 2006 (survey and eight-port chamber, pp. 15–48), D. Haas 2009 (glued-chamber rebuild, pp. 47–50), J. Morrow 2015 (steel-lid model, pp. 54–55), M. Yuly, Cyclotrons 2013 WE1PB01 (2013 chamber). Library entries lib-256, lib-003, lib-028, lib-041.
• P. Baumgartner & P. Heuer, the Cyclotron Kids' talks and Cyclotrons 2013 WE1PB05 — gap doctrine, lid bowing, weld distortion, support posts. Library entries lib-035, lib-036.
• COLUMBUS: Wolf, Frank & Held, Cyclotrons 2013 WE1PB03 (donations); Prechtl & Wolf, Das Lehr-Zyklotron COLUMBUS, Springer 2020, pp. 37–57 (chamber, recess, feedthrough limit; German, commercial). Library entries lib-258, lib-255.
• Wolf, Beck, Franz & Neumaier, "3D Printing for High Vacuum Applications," Cyclotrons 2019 THC04. Library entry.
• L. Dewan, MIT S.B. thesis, 2007 — lid-deflection example p. 12, materials p. 9. Library entry.
• ²⁴Na half-life 14.956 h — NNDC NuDat 3 decay data, nndc.bnl.gov/nudat3, retrieved 2026-08-28.
• Census entries cited by anchor are sourced on the builds page, each to its own public record.