Construction Classics

Hot-filament arc sources: the species, the filament, the bench procedure

Most small cyclotrons share one component: a hot tungsten or tantalum filament feeding electrons into a low-voltage arc in hydrogen. These five documents cover it from three directions. Two Physical Review papers from 1925 settle which hydrogen ions an electron-impact discharge makes and how pressure and fields shift the mix. Two Oak Ridge reports from the calutron programme measure why filaments die and show two circuit changes that double their life. And a university cyclotron’s final report records how its source was positioned, tuned and measured in 1954.

Why these documents can be hosted

Smyth and Hogness & Lunn were published in the United States in 1925, more than ninety-five years ago, so their copyright terms have expired. Both scans are from the Internet Archive’s serial scans of Physical Review and carry no later publisher’s stamp or watermark.

ORNL-1525 and ORNL-1758 are Oak Ridge National Laboratory reports under AEC contract W-7405-eng-26, operated by Carbide and Carbon Chemicals Company, marked Unclassified with their internal distribution lists bound in: the same class and the same contract as the hosted ORNL-1347 and ORNL-1348, and AEC contractor reports are treated as US Government works. One line was checked specifically. ORNL-1758’s reference 11 says the curves of its Figures 10 and 11 were used with permission of C. E. Normand; Normand appears on the report’s own ORNL distribution list and his source is cited as AEC document Y-563, so this is a courtesy credit between colleagues inside the same AEC laboratory, not a licence from a third-party publisher.

AECU-2907 is a contractor final report delivered to the Atomic Energy Commission by the Ohio State University Research Foundation under contract AT(11-1)-158 and issued by the AEC in its unclassified AECU series, the same class as the hosted AECU-1951. It is a project report, not a thesis, and carries no copyright notice. The cover of the copy OSTI scanned bears a library stamp reading DO NOT PHOTOSTAT; that is a handling instruction on a physical file copy, and the AEC’s successor distributes the same scan publicly. All five were read page by page before this determination; made October 2026.

Which ion the arc makes (Smyth; Hogness & Lunn)

Both papers put a known-energy electron beam through hydrogen, pulled the ions out through a slit, bent them through a semicircle in a magnetic field and counted them by mass. Smyth at Princeton concluded that the only primary product is H₂⁺, with H⁺ arising by secondary dissociation of H₂⁺ and H₃⁺ appearing in large numbers as a secondary product at any voltage above 16 V; the Design Guide labels that a 1925 interpretation under his conditions and its note carries the current appearance energies (dg-2084). His Fig. 4 is the picture to keep: at 0.0007 mm Hg the spectrum was essentially H₂⁺, and at 0.053 mm that peak was almost suppressed with H₃⁺ and H⁺ strong (dg-2085). He fits the pressure dependence and concludes the molecular ion is destroyed about fifteen times more readily than a kinetic-theory collision rate would allow (dg-2086), and that the apparent appearance potential of the secondary species, anywhere from 16 to 21 V, reflects the tube conditions rather than a separate primary process (dg-2088).

Hogness and Lunn at Berkeley tabulated the mix. Their Table I, at 48 V electron energy, runs from about 97% H₂⁺ at 2 × 10⁻⁴ mm to 39% H₂⁺ and about 58% H₃⁺ at 0.0056 mm, with the two curves crossing near 0.0045 mm (dg-2090); the abstract’s round numbers and the table’s differ slightly, and the rule brackets both (dg-2089). Raising the field that draws ions out of the ionization chamber raised the atomic fraction (dg-2091), and lowering the electron energy toward 18 V raised the H₃⁺ share, which they suggest was a path-length effect of their retarding field rather than the impact energy itself (dg-2093). Their summary on p. 9 is the recipe a source designer wants: H⁺ with a large potential difference and some helium, H₂⁺ at low pressure, H₃⁺ at higher pressure and small potential difference; with helium and a little hydrogen their last table line reads 68.6% H⁺ and no H₃⁺ (dg-2092).

Why filaments die, and two ways to make them last (ORNL-1525, ORNL-1758)

The calutron was a mass separator, not a cyclotron, but its ion source was a hot-filament arc run harder than any cyclotron’s, and Yonts’s two reports are the only public-domain study of filament wear at that intensity. The cathode was a four-inch U of 170- or 145-mil tantalum heated by up to 500 A, and its life depended on the element being processed: 200 hours for mercury, 15 to 20 for iron and titanium, 30 to 50 for most (dg-2066). He names three failure mechanisms: growths on the filament, alloying or chemical reaction with the charge, and sputtering by ions arriving from the arc with energies up to the arc voltage (dg-2067). The uranium case is the vivid one, a low-melting alloy growth that migrated to one side of the filament under the force on the heating current in the magnetic field, cured by heating with alternating current (dg-2068).

The first discovery was accidental. Calutron sources ran with a filament-arc regulator that held arc current constant by trimming filament current (dg-2070); when it was bypassed and the filament run manually at nearly constant current, iron-separation filament life almost doubled (dg-2071). The second was designed: run the arc like a gas-filled diode, on the vertical part of the emission curve with the arc drop near the ionization potential, where ion energies stay below the sputtering threshold (dg-2080). The circuit change was four 150 W light bulbs in parallel as a ballast in series with the arc supply (dg-2081). Full diode operation cut average arc power from 300 W to 46 W and halved the ion output; the arc no longer filled the chamber (dg-2072). Raising the arc voltage back to 80–100 V by slightly lowering the filament temperature, the modified diode mode, restored output to near normal with substantially longer life (dg-2073); ORNL-1525’s conclusion is a factor of 2 to 3 with no significant output loss (dg-2082), and ORNL-1758’s weighed filaments gave 49.4 mg/h against 95.5 mg/h in normal running (dg-2077).

The measurement craft in ORNL-1758 is as useful as its result. Filament temperature was read with an optical pyrometer through a quartz window and corrected from brightness to true temperature, 1950 to 2150 °C across the three modes (dg-2074). Erosion was tracked on line from the fall of the filament-transformer primary current, the time to fall through a fixed range being inversely proportional to the average erosion rate (dg-2075). Yonts also found the ion current at the receiver peaking and then falling as arc current rose, from defocusing by ion density in the exit region, and warns that the optimum is specific to the electrodes and the extraction voltages (dg-2076). His empirical fit, weight loss per ampere-hour proportional to arc power to the 3/5, is on p. 42 (dg-2078).

Two cautions. The charges were ferrous chloride, titanium and silicon tetrachlorides and uranium, and Yonts lists cyclotron sources running hydrogen or deuterium, and thoriated or coated filaments, as untested next steps (dg-2082); the mechanisms transfer, the hours do not. And ORNL-1758 is a typescript whose scan loses decimal points: its Child’s-law sheath estimate prints 67 mm where the inputs give 0.67 mm, and the rule brackets both the lost decimals and an arithmetic slip in the source (dg-2069). Read the page images for any number that matters.

A university cyclotron’s source, as run (AECU-2907)

Hausman’s Part II final report covers leak testing of the aluminium tank, the ion source, the self-excited tuned-grid tuned-plate oscillator, dee-voltage measurement and beam-current measurement on the rebuilt Ohio State cyclotron, in twenty pages. The source section is the one the Design Guide mined. The source was a low-voltage gas discharge in a small Elkonite cavity at the tank centre, fed by a heated U-shaped tungsten filament and inserted between the dee manifolds through a vacuum lock, so that filaments could be changed without losing tank vacuum (dg-2060). The position of the filament relative to the vertical opening in the discharge cone was found to be critical, so the filament assembly was made to slide inside the housing and be adjusted from outside the vacuum while the beam was watched (dg-2058).

The procedure for finding that position is written down, which is rarer than the finding. Put the target close to the source, 8 inches, so that focusing of the resonant beam does not confound the reading; hold the magnetic field and dee voltage fixed; and at each filament position re-maximise the beam by adjusting filament current and arc voltage before comparing (dg-2059). The operating point that resulted was 160 A of filament heater current and 60–100 V of arc (dg-2062). The report is candid about what it could not measure: the gas lines carried over from the earlier system were long enough that the lag in gas-input changes prevented any valid beam-versus-gas-feed measurement (dg-2063), and the beam current read on the insulated target carried a large secondary-emission component, so it was treated as a relative tuning indication (dg-2064). Absolute beam power came from the temperature rise of the target cooling water on a strip-chart recorder, calibrated in place with a 500 W immersion heater in the water line (dg-2065). The source assembly drawing on p. 14, with its hand-lettered insulation notes, is transcribed in dg-2061.

The OSTI scan is a photostat of a carbon copy and its text layer is poor; the rules were verified against the page images, and a reader should expect to do the same.

Why these five belong together

A hydrogen arc source is a chemistry experiment wrapped round a consumable. The 1925 papers say what the chemistry produces and which knobs move it; the Oak Ridge reports say what the consumable costs and how to pay less; the Ohio State report shows the component installed, adjusted and measured on a machine of the size this site is written for. None of the five is about a modern PIG source, and the species physics in particular has been remeasured many times since 1925; what is still current is the method of each, and the fact that the first questions a builder asks about a filament source were answered, in public, before the cyclotron existed.

Design rules extracted from these documents

36 Design Guide rules come from these five documents, almost all in the ion-source domain, with the Ohio State report also feeding beam measurement. Representative examples:

• dg-2092 — which hydrogen ion predominates: H⁺ with a large potential difference and some helium, H₂⁺ at low pressure, H₃⁺ at higher pressure
• dg-2085 — at 0.0007 mm Hg the spectrum was essentially H₂⁺; at 0.053 mm the H₂⁺ peak was almost suppressed
• dg-2067 — three ways a filament fails: growths, alloying with the processed element, and sputtering by ions at up to the arc voltage
• dg-2071 — bypassing the filament-arc regulator nearly doubled filament life
• dg-2082 — modified diode operation: life ×2–3 with no significant loss of output
• dg-2075 — track filament erosion on line from the decay of the filament heating current
• dg-2059 — the filament-position procedure: target close, field and dee voltage fixed, re-maximise at every position
• dg-2064 — an insulated target reads secondary emission along with beam; treat it as a tuning indicator

Every rule, by document:

• Smyth 1925 (5): dg-2084, dg-2085, dg-2086, dg-2087, dg-2088
• Hogness & Lunn 1925 (5): dg-2089, dg-2090, dg-2091, dg-2092, dg-2093
• ORNL-1525 (5): dg-2079, dg-2080, dg-2081, dg-2082, dg-2083
• ORNL-1758 (13): dg-2066, dg-2067, dg-2068, dg-2069, dg-2070, dg-2071, dg-2072, dg-2073, dg-2074, dg-2075, dg-2076, dg-2077, dg-2078
• AECU-2907 (8): dg-2058, dg-2059, dg-2060, dg-2061, dg-2062, dg-2063, dg-2064, dg-2065