Boiling Crisis: When Bubbles Merge and the Metal Melts

The Nukiyama curve and the cliff hidden inside it

Plot the heat flux leaving a submerged hot surface against its wall superheat — how far it sits above the liquid's saturation temperature — and you get the boiling curve. Shiro Nukiyama measured it at Tohoku University in 1934, and how he did it is the whole story: he heated thin nichrome and platinum wires electrically in saturated water at 1 atm, controlling the heat flux, not the temperature.

At a few kelvin of superheat the wire merely stirs the water. Past about 5 K, vapour nucleates in microscopic pits in the metal, bubbles release, and the flux climbs steeply. Then, near 30 K of superheat and roughly 1.1 MW/m², the curve stops. Nukiyama turned the current up one notch and the wire's temperature leapt by hundreds of kelvin. His nichrome wire melted. A platinum wire survived and settled into a stable, glowing, vapour-blanketed state — film boiling — at over 1000 K of superheat.

A whole branch was missing between the two: falling flux with rising temperature, which a flux-controlled experiment can never visit, because a surface driven at constant power cannot sit on a negatively sloped branch. Drew and Mueller reached it in 1937 by heating with condensing organic vapours instead. Nukiyama's leap is the boiling crisis, also called burnout or, in reactor language, departure from nucleate boiling (DNB).

How a bubble field turns into a dry patch

Nucleate boiling starts in cavities: machining pits a few micrometres across trap residual gas, and once the liquid above them is superheated enough that vapour pressure beats surface tension's inward squeeze, they grow. A bubble does not grow by conduction through its own vapour — it grows by evaporating liquid at its own interface, and a large share of that comes from a microlayer of liquid, typically 1–10 µm thick, wedged between its base and the metal (the remainder from the superheated liquid layer around the dome). Fritz's 1935 force balance puts departure where buoyancy beats the surface-tension grip on the contact line: for water at 1 atm, 1–3 mm bubbles leaving at tens of hertz.

Here is the counterintuitive part. A 2 mm steam bubble holds only ~2.5 µg of vapour, so it carries only ~6 mJ of latent heat. Even at ~10⁵ sites per square metre firing at 30 Hz, that is a few tens of kW/m² — a small fraction of the megawatt the wall sheds. Most of the flux is transient conduction: a departing bubble drags cold bulk liquid onto the freshly stripped wall, which quenches like a hot plate hit with cold water until the next bubble nucleates (Mikic and Rohsenow, 1969). Boiling cools so well because it is a bubble-driven pump, not because vapour carries heat.

Raise the flux and the pump jams. Sites multiply and bubbles merge into columns and then flattened mushroom bubbles, leaving a macrolayer a few hundred micrometres thick pierced by vapour stems (Haramura and Katto, 1983). Locally, the microlayer under a coalesced bubble thins to nothing and a dry spot appears. Normally the contact line recedes, the bubble leaves, and liquid floods back. But a dry spot has almost no cooling, so it heats fast, and once it passes the Leidenfrost point arriving liquid can no longer wet it — it is repelled on its own vapour cushion. The patch stops rewetting and spreads irreversibly across the heater in 1–10 ms. Thin foils fail fastest; a thick copper block can smear the hot spot with its own heat capacity and hang on slightly longer.

Zuber's number: the hydrodynamic limit, and how big it is

The classic prediction comes from Novak Zuber's 1959 UCLA dissertation. He argued that vapour leaves in an organised array of jets, spaced by the wavelength the Rayleigh–Taylor instability selects for heavy liquid sitting on light vapour, and that the crisis occurs when those jets go Helmholtz-unstable and choke the returning liquid. The result has no fitted constant:

q″_CHF = (π/24) · h_fg · √ρ_v · [σ · g · (ρ_l − ρ_v)]^1/4

For saturated water at 1 atm — h_fg = 2257 kJ/kg, ρ_v = 0.60 kg/m³, ρ_l = 958 kg/m³, σ = 59 mN/m — the bracket is 553 in SI units, its fourth root 4.85, √ρ_v is 0.77, and π/24 is 0.131. The product is 1.11 MW/m², within tens of percent of every careful flat-plate measurement made. Lienhard and Dhir refined the coefficient to 0.149 in 1973. The same grouping had appeared in 1948, from dimensional analysis alone, in work by S. S. Kutateladze, who fitted the constant at ~0.16; the ratio now bears his name and lands between 0.13 and 0.16 for water, cryogens, refrigerants and liquid metals alike.

The instability also sets a length. The most-dangerous Taylor wavelength λ_d = 2π√(3σ/gΔρ) is about 27 mm for water, and the capillary length √(σ/gΔρ) is 2.5 mm — which is why water bubbles depart at a couple of millimetres. So heater size matters: a wire smaller than about three capillary lengths cannot host the full jet array, and its CHF exceeds the flat-plate value, sometimes twofold. The formula also predicts CHF ∝ g^1/4, a peak near one third of the critical pressure (for water, ~7–8 MPa at ~3.5 MW/m²), and much lower values for electronics coolants: FC-72, with h_fg = 88 kJ/kg and σ = 8.5 mN/m, gives ~0.14 MW/m², matching the ~15 W/cm² measured in immersion cooling.

Why burnout melts the metal instead of stabilising

Once the blanket forms, heat must cross vapour. Steam at 100 °C conducts at 0.025 W/m·K against liquid water's 0.68 — and about 100× in the heat-transfer coefficient once film thickness and the loss of the bubble pump are included. Nucleate boiling at 10–50 kW/m²·K becomes film boiling at 0.1–0.3 kW/m²·K.

What decides whether this is interesting or fatal is what is held constant. On a temperature-controlled surface — condensing steam on the far side, or a quenching billet — the wall slides down the transition branch and nothing dramatic happens. On a heat-flux-controlled surface the power source does not care: a wire keeps dissipating I²R, a fuel rod keeps fissioning. The wall must find a superheat at which film boiling still carries the imposed flux, and at ~1 MW/m² that is enormous. Bromley's 1950 correlation gives only a few hundred W/m²·K, so conduction alone would need thousands of kelvin. What can save a surface is radiation: σT⁴ reaches 290 kW/m² at 1500 K and 900 kW/m² at 2000 K, so the new equilibrium sits near 1500–2000 K. That is exactly why Nukiyama's platinum wire lived and his nichrome one did not — platinum melts at 2041 K, nichrome near 1670 K.

And it is fast. A 1 mm nichrome wire has a thermal capacity per unit surface area of ρ·c·(d/4) ≈ 950 J/m²·K, so ~1.1 MW/m² with almost nothing leaving raises it at ~10³ K/s — a thousand kelvin in under a second, four times faster for a quarter-millimetre wire. It glows, sags and parts.

Watching it happen: infrared thermometry and the dry-spot picture

For four decades the hydrodynamic theory stood largely because it got the number right. Seeing the mechanism meant looking through the heater. The enabling technique is infrared thermometry through an IR-transparent substrate: coat sapphire or calcium fluoride with a few hundred nanometres of indium tin oxide or titanium, pass current through it to make the heater, and film the back with an IR camera. The film is optically thin and thermally negligible, so the camera reads true wall temperature at tens of micrometres and kilohertz frame rates.

Theofanous, Tu, Dinh and Dinh (2002) did this with thin titanium films and reported that the crisis was not a jet instability but the failure of individual dry spots to rewet — a contact-line event. Groups at MIT under Matteo Bucci added optical phase detection, where total internal reflection maps exactly which patches are dry. The time-averaged dry-area fraction grows smoothly with flux, rewetting wait times lengthen, and CHF arrives when one patch outruns rewetting. Nothing in the data requires the jets.

Zuber's formula survives anyway, because the properties setting the dry-spot force balance — surface tension pulling the contact line back, evaporation momentum pushing it out, buoyancy clearing vapour — are the same σ, ρ and h_fg in his group. Kandlikar's 2001 force-balance model makes this explicit and predicts what Zuber's cannot: CHF rises as the receding contact angle falls. Gravity has been tested directly too — NASA flew the Boiling eXperiment Facility to the ISS in 2010 for the Microgravity Science Glovebox. CHF does fall roughly as g^1/4, but at very low g the scaling breaks: bubbles stop detaching at all, and Marangoni flow takes over vapour removal.

Designing against it: reactors, divertors and better surfaces

In a pressurised water reactor the coolant is at 15.5 MPa and ~320 °C, moving past 9.5 mm fuel rods at thousands of kilograms per square metre per second. Forced convection and subcooling push flow-boiling CHF to roughly 4–7 MW/m², while peak rod surface flux is around 1–1.5 MW/m². The margin is the DNB ratio — predicted CHF over actual local flux — required to exceed about 1.3 with 95/95 confidence in anticipated transients. That number is not physics; it is the uncertainty band on the correlations used, from Tong's W-3 of 1967 through WRB-2 to the Groeneveld look-up tables built from tens of thousands of tube measurements. Boiling water reactors face a different failure — dryout of the liquid film in high-quality annular flow — and are licensed on a minimum critical power ratio near 1.2–1.3 instead.

Fusion pushes harder. The ITER divertor takes 10 MW/m² steady and 20 MW/m² in slow transients on tungsten monoblocks bonded to CuCrZr tubes carrying water at ~4 MPa and roughly 10 m/s; twisted-tape swirl inserts centrifuge liquid onto the wall to hold CHF near 25–30 MW/m². At the other extreme, two-phase immersion cooling of datacentre processors is capped near 150–200 kW/m² by the poor properties of dielectric fluids, which is why surface engineering became a field. Making the wall strongly hydrophilic and adding wicking structure — microporous coatings, nanowire forests, sintered particles, the nanoparticle deposits that form when boiling a nanofluid, as MIT's group under Jacopo Buongiorno showed after 2006 — pulls liquid into an incipient dry patch by capillarity and routinely raises CHF by 50–150%, the best hierarchical surfaces reaching ~2–2.5 MW/m² in water. The physics runs backwards in reactors, where CRUD deposits on cladding alter wettability and so the licensed margin.

Look-alikes, confusions and open questions

The Leidenfrost effect is the same vapour film at the opposite end of the same curve. Johann Gottlob Leidenfrost described the skittering droplet in 1756. The Leidenfrost point is the minimum of film boiling — for water at 1 atm roughly 20 kW/m² at 120–200 K of superheat, which Zuber's companion formula q″_min = 0.09 ρ_vh_fg[σgΔρ/(ρ_l+ρ_v)²]^1/4 puts at 19 kW/m². Both are vapour blankets; the difference is how you arrive. From a hot, temperature-controlled surface the film forms gently and a droplet hovers. From below on a flux-controlled surface, the same film forms while a megawatt per square metre is still being pushed through it. The curve is hysteretic too: coming down from film boiling, a surface stays blanketed until it cools to the Leidenfrost point, which is why quenching a forging or reflooding a reactor core proceeds as a visible rewetting front rather than all at once.

Other confusions worth separating: cavitation makes vapour by dropping pressure, not by adding heat; dryout is film depletion in annular flow, not a local contact-line failure; and the Ledinegg flow excursion (1938) is a system instability in which one channel among many loses flow and hits CHF far below the local hydrodynamic limit — the coolant fails first, not the boiling.

Open questions remain. No single mechanistic model predicts pool boiling, subcooled flow boiling and high-quality dryout from one starting point; industry runs on look-up tables with fitted uncertainty bands. Predicting CHF on an engineered surface from geometry and wettability alone is out of reach, and enhanced surfaces degrade as they foul or their wicking clogs.

The pool-boiling curve for saturated water at 1 atm on a horizontal surface. Wall superheat is ΔT = T_wall − 100 °C.

Regime	Wall superheat	Heat flux	Heat-transfer coefficient
Natural convection (no bubbles)	0–5 K	< ~10 kW/m²	~0.5–1 kW/m²·K
Isolated-bubble nucleate boiling	~5–15 K	~10–200 kW/m²	~5–15 kW/m²·K
Coalesced / jet nucleate boiling (up to CHF)	~15–30 K	up to ~1.1 MW/m²	~20–50 kW/m²·K
Transition boiling (unstable branch)	~30–120 K	falls 1.1 MW/m² → ~20 kW/m²	falling, negative slope
Film boiling above the Leidenfrost minimum	> ~120–200 K	~20 kW/m² at the minimum, rising with superheat as radiation (∝T⁴) takes over	~0.1–0.3 kW/m²·K