Heating It Hotter Made It Cooler — The Boiling Curve and Critical Heat Flux

In 1934, in Tokyo, Shiro Nukiyama was watching water boil around a nichrome wire as he pushed current through it. Each notch up made the bubbles more violent. Then suddenly the wire glowed red-hot, melted, and snapped. He never got to see what happened in the gap between the most vigorous bubbling and the wire burning out. This post walks through exactly the gap he missed — the paradox of the boiling curve, where heating harder can move less heat. We'll see why a critical heat flux exists as a cliff, why a Leidenfrost droplet dances on a hot pan, and back it all with a Python calculation at the end.

The Boiling Curve — Four Regimes in One Graph#

Boiling is evaporation at a solid–liquid interface. The wall heat flux has a simple form.

$q'' = h\,(T_s - T_{sat}) = h\,\Delta T_e$

$T_s$ is the wall temperature, $T_{sat}$ the saturation temperature (100°C for water at 1 atm), and $\Delta T_e$ the difference — the excess temperature. $h$ is the convection coefficient.

Intuition says $q''$ should rise monotonically with $\Delta T_e$. The real boiling curve refuses. Plot $\Delta T_e$ against $q''$ on log–log axes and four regimes appear in sequence.

• Free convection ($\Delta T_e \lesssim 5$°C): almost no bubbles. Buoyancy-driven natural convection carries the heat.
• Nucleate boiling ($5 \lesssim \Delta T_e \lesssim 30$°C): bubbles erupt from nucleation sites. $q''$ climbs steeply.
• Transition boiling ($30 \lesssim \Delta T_e \lesssim 120$°C): a vapor film starts to blanket the wall. $q''$ falls.
• Film boiling ($\Delta T_e \gtrsim 120$°C): a continuous vapor film covers the wall. With radiation added, $q''$ rises again.

Try the simulation below.

Switch the top button to 'Heat-flux control' and push the flux above the critical point. The operating point jumps from the nucleate branch to the film branch. That jump is the burnout that melted Nukiyama's wire.

Nucleate Boiling — Bubbles Pump the Heat#

Nucleate boiling is so effective not because of the bubbles themselves but because of the agitation they stir up. When a bubble breaks free of the wall, cold liquid rushes into the gap. This tiny pumping action churns the liquid near the wall without rest.

Rohsenow captured the flux in this regime with a dimensionless correlation.

$\mu_l$ is liquid viscosity, $h_{fg}$ the latent heat, $\rho_l,\rho_v$ the liquid and vapor densities, $\sigma$ the surface tension, $c_{p,l}$ the liquid specific heat, and $Pr_l$ the liquid Prandtl number. $C_{s,f}$ is an experimental constant for the surface–fluid pair (0.013 for water on platinum).

The key is $q''_s \propto \Delta T_e^3$. A small bump in excess temperature sends the flux up by a cube. That's why nucleate boiling extracts enormous heat across a narrow temperature band. Reactor cores, CPU water cooling, and boilers all aim at this regime.

Critical Heat Flux — The Cliff Zuber Predicted#

The cubic rise doesn't last. Pack the bubbles too densely and they coalesce, starting to blanket the wall. The liquid loses its path to the surface. Just before that, $q''$ hits a maximum. This is the critical heat flux (CHF). For water at 1 atm it exceeds 1 MW/m².

Zuber derived CHF from the condition where vapor columns go unstable (Kelvin–Helmholtz instability).

Symbols are as before, with $\rho_v$ the vapor density. The striking part: this expression carries no wall temperature and no surface material. CHF is set by fluid properties alone.

CHF is a double-edged sword. It's the upper limit of how much heat you can pull, and crossing it is a disaster. In a flux-controlled device (electric heating, nuclear fuel), exceeding CHF flings the operating point from nucleate to film boiling. To carry the same flux in film boiling, the wall temperature must leap by hundreds of degrees. Metal melts first. That jump is exactly what Nukiyama saw.

Film Boiling and Leidenfrost#

At the end of transition boiling, where $q''$ hits its minimum, lies the Leidenfrost point. Beyond it a continuous vapor film fully covers the wall. The liquid never touches it. Heat crosses the vapor only by conduction and radiation. Vapor conducts far worse than liquid, so heat transfer is dismal.

If you've ever flicked water onto a hot pan, you've seen film boiling. On a moderately hot pan, the droplet sizzles and vanishes fast. On a very hot pan, the droplet beads up like a marble and skates across the surface for a long while. It's floating on the vapor film it makes for itself.

Try the simulation below.

Set the surface near 130°C and the droplet bubbles furiously and evaporates fast. Push past 200°C and it lifts onto a vapor film and skitters. Hotter, yet it lives longer — that non-monotonicity is the Leidenfrost paradox.

Film-boiling flux on a horizontal cylinder follows this form.

$D$ is the cylinder diameter, $\nu_v, k_v$ the vapor's kinematic viscosity and conductivity, and $C$ a shape constant (0.62 for a horizontal cylinder, 0.67 for a sphere). $h'_{fg} = h_{fg} + 0.8\,c_{p,v}(T_s - T_{sat})$ corrects the latent heat for vapor superheat.

Python — Drawing the Boiling Curve Yourself#

Using water properties at 1 atm, let's compute Rohsenow nucleate boiling and Zuber CHF. Where the two meet, and where they part, jumps right out.

import numpy as np
 
# Water, saturated properties at 1 atm
g    = 9.81       # m/s^2
rho_l = 957.9     # kg/m^3
rho_v = 0.5956
h_fg  = 2257e3    # J/kg
sigma = 0.0589    # N/m
mu_l  = 279e-6    # N.s/m^2
cp_l  = 4217.0    # J/kg.K
Pr_l  = 1.76
Csf, n = 0.013, 1.0   # water-platinum
 
def rohsenow_nucleate_flux(dTe):
    """Rohsenow nucleate boiling heat flux q'' (W/m^2)."""
    bracket = np.sqrt(g * (rho_l - rho_v) / sigma)
    ja = cp_l * dTe / (Csf * h_fg * Pr_l**n)
    return mu_l * h_fg * bracket * ja**3
 
def zuber_critical_heat_flux():
    """Zuber critical heat flux (W/m^2)."""
    return 0.149 * h_fg * rho_v * (
        sigma * g * (rho_l - rho_v) / rho_v**2) ** 0.25
 
chf = zuber_critical_heat_flux()
print(f"CHF (Zuber)      = {chf/1e6:.3f} MW/m^2")
 
# Scan the nucleate branch and find where it reaches CHF
for dTe in [5, 10, 20, 30, 40]:
    q = rohsenow_nucleate_flux(dTe)
    flag = "  <-- exceeds CHF (Rohsenow invalid)" if q > chf else ""
    print(f"dTe={dTe:3d} C : q''={q/1e6:6.3f} MW/m^2{flag}")

The run prints:

CHF (Zuber)      = 1.259 MW/m^2
dTe=  5 C : q''= 0.017 MW/m^2
dTe= 10 C : q''= 0.137 MW/m^2
dTe= 20 C : q''= 1.094 MW/m^2
dTe= 30 C : q''= 3.692 MW/m^2  <-- exceeds CHF (Rohsenow invalid)

Rohsenow already overshoots CHF near $\Delta T_e \approx 30$°C. Not because the formula is wrong, but because it has left its valid range. The real curve bends down at CHF. The nucleate correlation only holds up to that bend. When you use both formulas together, you must always enforce CHF as the ceiling.

The Language of Dimensionless Numbers — Jakob and Bond#

Two dimensionless numbers govern boiling. The first is the Jakob number.

$Ja = \frac{c_{p,l}\,\Delta T_e}{h_{fg}}$

It's the ratio of sensible heat the liquid absorbs to latent heat. A small $Ja$ means latent heat dominates sensible heat. Water's latent heat is so large that $Ja$ stays small in boiling. The bracketed term in the Rohsenow formula is exactly $Ja$.

The second is the Bond number.

$Bo = \frac{g(\rho_l - \rho_v)\,L^2}{\sigma}$

It's the ratio of buoyancy to surface tension. A bubble's characteristic size $L$ comes from balancing these two. The bubble diameter scales as $D_b \propto \sqrt{\sigma / [g(\rho_l-\rho_v)]}$. That's why surface tension and density difference sit inside the fourth root of the CHF formula. Bubble size and departure frequency ultimately set the ceiling on heat flux.

Three Lines to Remember#

• The boiling curve is not monotonic. Raising the excess temperature drops the heat flux through the transition regime, and a critical heat flux (CHF) cliff exists.
• Cross CHF and the operating point jumps to film boiling, sending the wall temperature up by hundreds of degrees. That's why Nukiyama's wire melted.
• The Leidenfrost effect is the everyday version of the same non-monotonicity. On a hotter pan, a droplet floats on a vapor film and lasts longer.