Radiative Diffusion and the Rosseland Mean Opacity

Drop a photon into the core of the Sun and it does not fly out — it staggers. Before it can travel even a millimetre it is absorbed or scattered, re-emitted in a random direction, and the walk begins again; a packet of energy needs of order ten thousand to a hundred thousand years to grope its way from centre to surface. In this enormously optically thick interior the radiation field is almost perfectly isotropic and almost exactly a Planck spectrum at the local temperature. Almost — but not quite. The whisper of anisotropy from the outward temperature gradient is the entire net energy flow of the star, and describing it is the business of the radiative diffusion approximation: energy leaks down the gradient exactly as heat conducts down a bar, with the photon mean free path as the diffusion length. Two ideas make this quantitative. First, expanding the transfer equation to first order in the small parameter $\ell/L$ collapses it into a Fick-type law, $F = -\tfrac{c}{3\kappa\rho}\,\mathrm{d}u/\mathrm{d}r$ , whose diffusion coefficient carries the same factor of $1/3$ that governs radiation pressure. Second, real opacity depends violently on frequency, so a single grey coefficient must be defined by averaging correctly — and because diffusion is set by where photons most easily escape, the right average is the flux-weighted harmonic mean of the monochromatic opacity: the Rosseland mean. Fold it into the definition of luminosity and out drops the radiative temperature-gradient equation, one of the four pillars of stellar structure and the quantity the Schwarzschild criterion tests to decide where a star convects.

Key takeaways

Deep in a star a photon's mean free path is minuscule compared with the temperature scale height, so radiation transport becomes a slow diffusion of energy on an almost isotropic, near-blackbody bath.
Expanding the transfer equation to first order in the anisotropy gives Fick's law for radiation, $F_{\mathrm{rad}} = -\dfrac{c}{3\kappa\rho}\dfrac{\mathrm{d}u}{\mathrm{d}r} = -\dfrac{4acT^3}{3\kappa\rho}\dfrac{\mathrm{d}T}{\mathrm{d}r}$ , with $D=c\ell/3$ .
Frequency dependence folds into one grey number, the Rosseland mean $\bar\kappa$ : a harmonic average of $1/\kappa_\nu$ weighted by $\partial B_\nu/\partial T$ , which hands the flux to the most transparent windows.
Combining diffusion with $L_r=4\pi r^2 F$ yields the stellar-structure equation $\dfrac{\mathrm{d}T}{\mathrm{d}r} = -\dfrac{3\bar\kappa\rho L_r}{16\pi ac\,r^2T^3}$ ; comparing its gradient to the adiabatic one is the Schwarzschild test for convection.
The approximation fails near the photosphere ( $\tau\lesssim$ a few), where the field is no longer isotropic and full transfer is required.
Key idea: in an optically thick medium radiation transport reduces to diffusion of energy down the temperature gradient, $\mathbf{F}_{\mathrm{rad}} = -\dfrac{c}{3\bar\kappa\rho}\nabla(aT^4)$ , where the single grey coefficient $\bar\kappa$ is the flux-weighted harmonic mean of the monochromatic opacity — the Rosseland mean.

Where this lives today

The Rosseland mean is a live, precision-critical input to every modern stellar model. Codes such as MESA integrate equation (3) using opacity tables (OPAL, OP, and the low-temperature AESOPUS and Ferguson tables) that tabulate $\bar\kappa$ over density, temperature, and composition; a computed stellar age, radius, or convection-zone depth is only as accurate as those tables. Two contemporary stories show why. First, the "iron opacity bump" near $2\times10^5$ K — a spike in $\bar\kappa$ from millions of iron lines — drives the pulsations of Cepheids and $\beta$ Cephei stars by the kappa-mechanism, and getting it right was a triumph of the OPAL recalculations in the 1990s. Second, the solar abundance problem: when three-dimensional spectroscopy lowered the Sun's metallicity in the 2000s, standard models lost the opacity they needed and could no longer match the helioseismic sound-speed profile — and laboratory iron-opacity measurements at the Sandia Z machine in 2015 found the true opacity higher than the tables predicted, reopening the question. In every one of these debates the quantity under the microscope is a Rosseland mean.

The physical picture

Picture a single photon deep in the solar interior. Its mean free path is $\ell = 1/(\kappa\rho)$ , a fraction of a millimetre; to cross a region of size $L$ it must take roughly $(L/\ell)^2$ random steps, each with a tiny outward bias. The field it belongs to is therefore isotropic to about one part in $\ell/L$ and Planckian to the same accuracy, and the net transport of energy is not this vast isotropic bath but the faint outward residue left when the bath is slightly hotter inside than out. Diffusion runs down the gradient of energy density, and since $u = aT^4$ , the excess leaks from hot interior to cooler exterior at a rate set by how far a photon travels between interactions — the logic of kinetic-theory heat conduction, where conductivity scales as (speed) × (mean free path). Here the speed is $c$ and $D = c\ell/3$ .

Figure

A photon random-walks down the temperature gradient. Each free flight is only ℓ = 1 over κρ ≈ 0.1 cm, so crossing the star takes of order L over ℓ squared steps; the net outward drift (green) is the whole radiative flux F = minus 4acT³ over 3κρ times dT/dr, with D = cℓ/3

Now let the opacity depend on frequency, as it truly does. Bound–free edges, the free–free continuum, and thousands of spectral lines make $\kappa_\nu$ swing over orders of magnitude, while between them lie transparent windows. Radiation does not care about the opaque frequencies — it pours out through the windows, the way water in a leaky reservoir finds the lowest hole in the dam. Any sensible grey opacity must therefore be dominated by the transparent frequencies and weighted by those that carry the diffusive flux. That weighting is $\partial B_\nu/\partial T$ , because diffusion is driven by the temperature gradient and it is the temperature derivative of the Planck spectrum that sets how much flux each frequency conducts. The result is a harmonic (reciprocal) average — the Rosseland mean — shown in the second figure.

Figure

The Rosseland mean bows to transparent windows. Monochromatic opacity κ_ν (blue) spikes at bound-free edges and lines but plunges in a window; the weight ∂B_ν/∂T (amber) marks the flux-carrying band. A two-band model, a 10 percent window at κ = 1 and 90 percent opaque at κ = 1000, gives harmonic κ̄ ≈ 9.9 cm² per g versus an arithmetic mean of 900

The factor $1/3$ recurring in the diffusion coefficient $D=c\ell/3$ , in the flux law, and in the radiation pressure $P_{\mathrm{rad}}=\tfrac13 aT^4$ is no coincidence: it is the angular average $\langle\cos^2\theta\rangle = 1/3$ over the sphere. Diffusion and the pressure it exerts are two moments of one nearly isotropic field — the flux the first (odd) moment of the intensity, the pressure the second (even) — and the diffusion approximation is just the statement that these two moments suffice.

The core equations

The starting point is the grey diffusion law: deep inside the star the small residual flux is proportional to the gradient of the radiation energy density, with the mean free path setting the constant — Fick's law with a photon diffusion coefficient. Writing $u_{\mathrm{rad}}=aT^4$ turns that into a temperature gradient.

F_{\mathrm{rad}}(r) = -\frac{c}{3\kappa\rho}\frac{\mathrm{d}u_{\mathrm{rad}}}{\mathrm{d}r} = -\frac{4acT^3}{3\kappa\rho}\frac{\mathrm{d}T}{\mathrm{d}r} \qquad (1)

where: $F_{\mathrm{rad}}$ — radiative energy flux (erg s⁻¹ cm⁻²); $u_{\mathrm{rad}}=aT^4$ — radiation energy density; $a=4\sigma/c=7.566\times10^{-15}$ erg cm⁻³ K⁻⁴ — radiation constant; $\kappa$ — opacity (cm² g⁻¹); $\rho$ — mass density; $c$ — speed of light. Reading: the flux points down the temperature gradient, growing as $T^3$ but shrinking as opacity rises — an opaque layer throttles the outward energy flow. Because $\kappa_\nu$ varies with frequency, the grey $\kappa$ in equation (1) must be an average that reproduces the true total flux: the Rosseland mean, a harmonic mean of the monochromatic opacity weighted by the temperature derivative of the Planck function.

\frac{1}{\bar\kappa} \equiv \frac{\displaystyle\int_0^\infty \frac{1}{\kappa_\nu}\frac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu}{\displaystyle\int_0^\infty \frac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu} \qquad (2)

where: $\bar\kappa$ — Rosseland mean opacity, the single grey value; $\kappa_\nu$ — monochromatic opacity at $\nu$ ; $\partial B_\nu/\partial T$ — temperature derivative of the Planck function, the weighting. Reading: it is $1/\kappa_\nu$ (the mean free path) that is averaged, so transparent windows dominate — a single clear band can set $\bar\kappa$ even when the gas is opaque almost everywhere else. Folding the grey diffusion law into the luminosity crossing a sphere gives the radiative temperature gradient — one of the four coupled equations of stellar structure.

\frac{\mathrm{d}T}{\mathrm{d}r} = -\frac{3\bar\kappa\rho\,L_r}{16\pi a c\,r^2 T^3} \qquad (3)

where: $L_r=4\pi r^2 F_{\mathrm{rad}}$ — luminosity crossing radius $r$ ; $r$ — radial coordinate; $T$ — local temperature. Reading: a denser, more opaque, or cooler layer needs a steeper gradient to carry a given luminosity — and when that demand exceeds what the gas can supply adiabatically, the layer convects. The microphysics behind the diffusion coefficient is the photon mean free path: the transport coefficient is the product of step speed $c$ and step length $\ell$ , reduced by the angular factor $1/3$ .

\ell = \frac{1}{\kappa\rho}, \qquad D = \frac{c\,\ell}{3} = \frac{c}{3\kappa\rho} \qquad (4)

where: $\ell$ — photon mean free path; $D$ — radiative diffusion coefficient. Reading: with $\kappa\sim1$ cm² g⁻¹ and $\rho\sim1$ g cm⁻³, $\ell\sim1$ cm — the photon is trapped, and the huge ratio $L/\ell$ makes the escape time astronomically long. Finally, the diffusion law in momentum form shows the flux delivers momentum to the gas at a rate $\bar\kappa\rho F/c$ per unit volume — the physics that caps a star's luminosity.

P_{\mathrm{rad}} = \frac{1}{3}u_{\mathrm{rad}} = \frac{1}{3}aT^4, \qquad \frac{\mathrm{d}P_{\mathrm{rad}}}{\mathrm{d}r} = -\frac{\bar\kappa\rho}{c}\,F_{\mathrm{rad}} \qquad (5)

where: $P_{\mathrm{rad}}$ — radiation pressure. Reading: diffusion and radiation-pressure support are one equation read two ways; requiring the radiative force not exceed gravity gives $L_{\mathrm{Edd}}=4\pi GMc/\bar\kappa$ .

Derivation ARadiative diffusion from the transfer equation

The goal is to show that in the optically thick limit the full equation of radiative transfer collapses into a simple diffusion law, and to see precisely where the diffusion coefficient $c/3\kappa\rho$ comes from. We work at grey (frequency-integrated) level in plane-parallel geometry, treat $\ell/L$ as a small parameter, expand the intensity to first order in the anisotropy, and take the flux moment — from which the recurring factor of $1/3$ emerges as an angular integral tying this result to the radiation-pressure moment.

Step 1. Write the plane-parallel transfer equation $\mu\,\mathrm{d}I/\mathrm{d}\tau = I - S$ , where $\mu=\cos\theta$ , $I$ is the specific intensity, $S$ the source function, and $\mathrm{d}\tau=-\kappa\rho\,\mathrm{d}r$ . Why: this is the exact bookkeeping of gains and losses along a ray, and every approximation below is a controlled truncation of it.

Step 2. Impose the LTE closure. Deep inside, Kirchhoff's law makes the source function the local Planck function, $S=B(T)=acT^4/4\pi$ , which is isotropic. Why: at large optical depth a photon is absorbed and re-emitted so many times that it thermalizes, so emission is set by $T$ alone, not by the incoming radiation's history.

Step 3. Take the zeroth order. If $\ell/L\to0$ the gradient term is negligible and $I^{(0)}=S=B(T)$ , isotropic. Why: an isotropic intensity carries no net flux — the leading-order bath just sits there — so all transport lives in the next order.

Step 4. Get the first-order correction by inserting $I^{(0)}=B$ into the gradient term: $I \approx B - \mu\,\mathrm{d}B/\mathrm{d}\tau$ . Why: the anisotropy is linear in $\mu=\cos\theta$ and proportional to the tiny gradient $\mathrm{d}B/\mathrm{d}\tau\sim\ell/L$ — the forward-backward imbalance a temperature gradient imprints.

Step 5. Take the flux moment, $F=2\pi\int_{-1}^{1} I\,\mu\,\mathrm{d}\mu$ . The isotropic part integrates to zero and $\int_{-1}^{1}\mu^2\,\mathrm{d}\mu=2/3$ , so $F=-\tfrac{4\pi}{3}\,\mathrm{d}B/\mathrm{d}\tau$ . Why: this is where the $1/3$ is born — the average of $\cos^2\theta$ over the sphere, the same moment that gives $P_{\mathrm{rad}}=\tfrac13 u$ .

Step 6. Restore radial variables with $B=cu/4\pi$ and $\mathrm{d}\tau=-\kappa\rho\,\mathrm{d}r$ to obtain $F=-\dfrac{c}{3\kappa\rho}\dfrac{\mathrm{d}u_{\mathrm{rad}}}{\mathrm{d}r}$ . Why: this is Fick's law for radiation with $D=c/3\kappa\rho$ — a diffusion current driven by the gradient of radiation energy density, exactly analogous to heat conduction. *Verified: the moments $\int_{-1}^1\mu\,\mathrm{d}\mu=0$ and $\int_{-1}^1\mu^2\,\mathrm{d}\mu=2/3$ give the $1/3$ , and $B=acT^4/4\pi$ recovers Carroll and Ostlie Eq. (9.35).*

Go deeper · Derivation B · 1 minThe Rosseland harmonic mean as the correct grey opacity

Derivation BThe Rosseland harmonic mean as the correct grey opacity

The aim is to derive the flux-weighted harmonic average of equation (2) rather than merely assert it, so it becomes clear why the average is reciprocal and why the weight is $\partial B_\nu/\partial T$ . We start from the monochromatic diffusion law, convert its gradient into a temperature gradient via LTE, integrate over frequency, and demand that a single grey opacity reproduce the true total flux — with the normalization integral tying the answer back to the Stefan-Boltzmann law.

Step 1. Apply the diffusion law frequency by frequency: $F_\nu=-\dfrac{c}{3\kappa_\nu\rho}\dfrac{\mathrm{d}U_\nu}{\mathrm{d}r}$ , with $U_\nu=(4\pi/c)B_\nu$ in LTE. Why: each frequency diffuses independently with its own mean free path $1/(\kappa_\nu\rho)$ , so transparent frequencies conduct far more flux than opaque ones.

Step 2. Convert the gradient to a temperature gradient. Only $T$ varies with $r$ , so the chain rule gives $\mathrm{d}U_\nu/\mathrm{d}r=(4\pi/c)(\partial B_\nu/\partial T)(\mathrm{d}T/\mathrm{d}r)$ . Why: this is the origin of the weight — a frequency contributes to the flux in proportion to how sharply its Planck intensity responds to temperature, $\partial B_\nu/\partial T$ , not $B_\nu$ itself.

Step 3. Integrate over frequency to get the total flux $F=\int_0^\infty F_\nu\,\mathrm{d}\nu=-\dfrac{4\pi}{3\rho}\dfrac{\mathrm{d}T}{\mathrm{d}r}\displaystyle\int_0^\infty \dfrac{1}{\kappa_\nu}\dfrac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu$ . Why: the total flux sums the reciprocal opacity $1/\kappa_\nu$ weighted by $\partial B_\nu/\partial T$ — a harmonic mean already staring back at us.

Step 4. Demand that the grey flux $F=-\dfrac{4acT^3}{3\bar\kappa\rho}\dfrac{\mathrm{d}T}{\mathrm{d}r}$ equal the exact total; matching gives $\dfrac{1}{\bar\kappa}\displaystyle\int_0^\infty\dfrac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu=\displaystyle\int_0^\infty\dfrac{1}{\kappa_\nu}\dfrac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu$ . Why: this is a definition, not an approximation — $\bar\kappa$ is built to make the single-opacity diffusion law reproduce the frequency-resolved flux exactly.

Step 5. Fix the normalization with $\displaystyle\int_0^\infty\dfrac{\partial B_\nu}{\partial T}\,\mathrm{d}\nu=\dfrac{\mathrm{d}}{\mathrm{d}T}\left(\dfrac{acT^4}{4\pi}\right)=\dfrac{acT^3}{\pi}$ , reducing the condition to equation (2). Why: the denominator is the temperature derivative of the Stefan-Boltzmann integral, anchoring the whole construction to the total blackbody energy. *Verified: $\int(\partial B_\nu/\partial T)\,\mathrm{d}\nu=acT^3/\pi$ follows from $\int B_\nu\,\mathrm{d}\nu=acT^4/4\pi$ , and equation (2) reproduces Carroll and Ostlie Eq. (9.20).*

Go deeper · Derivation C · 1 minThe radiative temperature gradient and the Schwarzschild criterion

Derivation CThe radiative temperature gradient and the Schwarzschild criterion

Here we assemble the frequency-averaged diffusion law into the differential equation that stellar-structure codes actually integrate, then read off the condition that decides whether a layer transports heat by radiation or convection. The route is short — substitute the grey flux into the local luminosity and solve for the gradient — but the payoff is one of the four fundamental equations of stellar structure plus the trigger for convection zones.

Step 1. Write the grey diffusion flux with the Rosseland mean, $F_{\mathrm{rad}}=-\dfrac{4acT^3}{3\bar\kappa\rho}\dfrac{\mathrm{d}T}{\mathrm{d}r}$ . Why: this is equation (1) with $\bar\kappa$ inserted, and by construction it is the exact frequency-integrated flux in the diffusion limit.

Step 2. Define the local luminosity, $L_r=4\pi r^2 F_{\mathrm{rad}}$ . Why: $L_r$ is the total power threading the sphere of radius $r$ , the quantity the energy-generation equation supplies from below.

Step 3. Substitute and solve: $L_r=-4\pi r^2\dfrac{4acT^3}{3\bar\kappa\rho}\dfrac{\mathrm{d}T}{\mathrm{d}r}$ rearranges to $\dfrac{\mathrm{d}T}{\mathrm{d}r}=-\dfrac{3\bar\kappa\rho L_r}{16\pi ac\,r^2 T^3}$ . Why: this is equation (3), the radiative structure equation — the temperature profile a star needs to carry its luminosity purely by radiation.

Step 4. Recast as the logarithmic gradient $\nabla_{\mathrm{rad}}\equiv\left(\dfrac{\mathrm{d}\ln T}{\mathrm{d}\ln P}\right)_{\mathrm{rad}}=\dfrac{3\bar\kappa L_r P}{16\pi ac\,G M_r T^4}$ and compare with the adiabatic value $\nabla_{\mathrm{ad}}$ . Why: the Schwarzschild criterion holds a layer stable only if $\nabla_{\mathrm{rad}}<\nabla_{\mathrm{ad}}$ ; when opacity spikes or luminosity concentrates, $\nabla_{\mathrm{rad}}$ exceeds $\nabla_{\mathrm{ad}}$ and the layer convects. *Verified: equation (3) matches Carroll and Ostlie Eq. (10.68); the dimensional check gives K cm⁻¹ with $[a]=$ erg cm⁻³ K⁻⁴ and $[L_r]=$ erg s⁻¹, and $\nabla_{\mathrm{rad}}$ is dimensionless.*

Limiting cases

The form and validity of the diffusion picture depend on the optical depth and the frequency structure of the opacity.

Case	Result	Meaning
Optically thick interior ( $\tau\gg1$ )	$F_{\mathrm{rad}} = -\dfrac{c}{3\kappa\rho}\dfrac{\mathrm{d}(aT^4)}{\mathrm{d}r}$	Diffusion holds; the field is nearly isotropic with $S_\nu=B_\nu$ . This is the regime of stellar interiors.
Near the surface ( $\tau\lesssim$ few)	$I_\nu(\mu)$ strongly anisotropic	Diffusion breaks down; radiation streams outward and a two-stream or Eddington closure, or full transfer, is required.
Pure electron (Thomson) scattering	$\kappa_{\mathrm{es}} = 0.20\,(1+X)$ cm² g⁻¹	Frequency-independent, so $\bar\kappa=\kappa_{\mathrm{es}}$ exactly; the floor opacity of hot ionized interiors ( $X$ = hydrogen mass fraction).
Free–free (Kramers-like) opacity	$\bar\kappa_{\mathrm{ff}} \propto \rho\,T^{-7/2}$	Rosseland averaging over the free–free cross section yields the classic Kramers density and temperature scaling.
Two frequency bands (opaque + window)	$\dfrac{1}{\bar\kappa} \approx \dfrac{w_{\mathrm{win}}}{\kappa_{\mathrm{win}}}$	The harmonic mean is set by the transparent band even when it carries little weight $w_{\mathrm{win}}$ — the qualitative heart of the Rosseland mean.

Common trapThe Rosseland mean is not a straight average of the opacity

The single most common error is to average $\kappa_\nu$ directly — a simple or Planck-weighted mean. That is wrong twice over. First, diffusion adds fluxes, and each frequency's flux scales as $1/\kappa_\nu$ , so it is the reciprocal opacity that must be averaged; the Rosseland mean is harmonic. Second, the weight is $\partial B_\nu/\partial T$ , not $B_\nu$ , because it is the temperature response of the spectrum that drives the diffusive current. A straight average, dominated by opaque lines, badly overestimates the resistance to radiation flow; the correct harmonic mean is dominated by transparent windows. And whatever the average, the diffusion approximation is valid only at large optical depth — never apply it at the photosphere or in a thin nebula, where the field is strongly anisotropic.

Worked examples

1 · A two-band opacity and the tyranny of the window (fully worked)

Over the band where $\partial B_\nu/\partial T$ is appreciable, the gas is opaque ( $\kappa_{\mathrm{op}}=1000$ cm² g⁻¹) across a fraction $w_{\mathrm{op}}=0.90$ of the weight but has a clear window ( $\kappa_{\mathrm{win}}=1$ cm² g⁻¹) over the remaining $w_{\mathrm{win}}=0.10$ . Find the Rosseland mean and compare it with the naive arithmetic average.

Step 1. Discretized into two bands, the Rosseland definition is a weighted average of the reciprocal opacity: $\dfrac{1}{\bar\kappa}=\dfrac{w_{\mathrm{op}}}{\kappa_{\mathrm{op}}}+\dfrac{w_{\mathrm{win}}}{\kappa_{\mathrm{win}}}=\dfrac{0.90}{1000}+\dfrac{0.10}{1}=0.0009+0.1000=0.1009$ cm² g⁻¹ inverse.

Step 2. Invert: $\bar\kappa=1/0.1009=\mathbf{9.9\ cm^2\,g^{-1}}$ . The opaque band, despite 90 percent of the weight, contributes under one percent of $1/\bar\kappa$ — essentially invisible to the flux.

Step 3. Contrast with the arithmetic mean: $\langle\kappa\rangle=0.90\times1000+0.10\times1=\mathbf{900\ cm^2\,g^{-1}}$ — nearly a hundred times larger, and wrong. Only the harmonic value $9.9$ predicts the correct flux. A window over just 10 percent of the weight drags the grey opacity from 900 down to about 10 cm² g⁻¹ — why line-blanketed tables still let so much flux through.

2 · The Eddington luminosity of the Sun

Find the luminosity at which radiation pressure alone would lift off the outer layers of a solar-mass star, using electron-scattering opacity for solar composition ( $X=0.70$ ).

Step 1. Electron-scattering opacity: $\kappa_{\mathrm{es}}=0.20(1+X)=\mathbf{0.34\ cm^2\,g^{-1}}$ , frequency-independent, so $\bar\kappa=\kappa_{\mathrm{es}}$ exactly.

Step 2. Balance radiative force against gravity (equation 5): setting the outward force per gram $\bar\kappa L/(4\pi r^2 c)$ equal to $GM/r^2$ gives $L_{\mathrm{Edd}}=4\pi GMc/\bar\kappa$ , independent of radius.

Step 3. Insert cgs values ( $G=6.67\times10^{-8}$ , $M_\odot=1.99\times10^{33}$ g, $c=3.00\times10^{10}$ cm s⁻¹): $L_{\mathrm{Edd}}=\dfrac{4\pi(6.67\times10^{-8})(1.99\times10^{33})(3.00\times10^{10})}{0.34}\approx\mathbf{1.5\times10^{38}\ erg\,s^{-1}}$ , about $3.8\times10^4\,L_\odot$ . The Sun radiates roughly ten thousand times below its Eddington limit and is comfortably stable, while the most luminous massive stars live close to this ceiling and drive powerful radiation-pressure winds.

3 · How long is a photon trapped in the Sun?

Estimate the time for energy to random-walk from the solar centre to the surface, taking $\ell\approx0.1$ cm and $R_\odot=7.0\times10^{10}$ cm.

Step 1. A random walk of step $\ell$ covers net distance $R$ only after $N\approx(R/\ell)^2=(7.0\times10^{11})^2\approx\mathbf{4.9\times10^{23}}$ steps.

Step 2. Each step takes $\ell/c$ , so $t\approx N\ell/c=R^2/(\ell c)\approx\dfrac{(7.0\times10^{10})^2}{(0.1)(3.0\times10^{10})}\approx1.6\times10^{12}$ s $\approx\mathbf{5\times10^{4}\ yr}$ .

Step 3. The estimate is crude — $\ell$ varies by orders of magnitude from centre to photosphere, and careful models give ten thousand to a few hundred thousand years — but the lesson is exact: the light reaching us today carries energy released in the core tens of thousands of years ago, and diffusion, not free flight, sets that clock.

Physics of the portrait

The name on this equation belongs to a Norwegian theorist who, in a few pages in 1924, solved how to compress a wildly frequency-dependent opacity into the one number a stellar model can use.

Origins · 1 minSvein Rosseland and how this was discovered

Svein Rosseland

1894–1985 · radiative transfer and the internal constitution of the stars

Svein Rosseland trained in the great school of early astrophysics — he worked with Niels Bohr in Copenhagen with Eddington's ideas fresh in the field — and turned to the deep interiors of stars just as Eddington was establishing that radiation, not convection, carries most of their energy. The obstacle was opacity: bound–free edges, free–free absorption, and spectral lines make the true absorption coefficient a jagged function of frequency, and it was not obvious what single "mean" belonged in the diffusion law. In his 1924 note on the absorption of radiation within a star, Rosseland gave the answer: because each frequency conducts flux in proportion to its mean free path $1/\kappa_\nu$ , the correct grey opacity is the harmonic mean of $\kappa_\nu$ , weighted by $\partial B_\nu/\partial T$ . The construction hands transport to the transparent windows between lines, exactly where photons escape, and made the radiative temperature-gradient equation computable for the first time. He went on to write an influential textbook on stellar atmospheres and to found the Institute of Theoretical Astrophysics in Oslo, but it is the harmonic mean that carries his name into every stellar-structure code and opacity table since — OPAL, OP, and their successors all deliver, in the end, a Rosseland mean.

History

1906

— Schwarzschild shows the solar atmosphere is in radiative equilibrium and introduces optical depth

1916

— Eddington applies radiative diffusion to the deep interiors of stars

1924

— Rosseland defines the harmonic, flux-weighted mean opacity that bears his name

1926

— Eddington's Internal Constitution of the Stars codifies radiative energy transport

1992

— the OPAL project releases modern Rosseland-mean opacity tables and resolves long-standing pulsation puzzles

Going further

Texts. Carroll and Ostlie, An Introduction to Modern Astrophysics, 2nd ed., §9.3 and §10.4 (radiative diffusion, the Rosseland mean) · Rybicki and Lightman, Radiative Processes in Astrophysics, Ch. 1 · Kippenhahn, Weigert and Weiss, Stellar Structure and Evolution, 2nd ed., Ch. 5 · Clayton, Principles of Stellar Evolution and Nucleosynthesis, Ch. 3 · Mihalas, Stellar Atmospheres, 2nd ed., Ch. 2.

Provenance. The diffusion law and Rosseland mean follow Carroll and Ostlie §9.3–§10.4 and Rybicki and Lightman Ch. 1. A verifier should recompute the angular moments ( $\int_{-1}^1\mu\,\mathrm{d}\mu=0$ , $\int_{-1}^1\mu^2\,\mathrm{d}\mu=2/3$ ), confirm $\int(\partial B_\nu/\partial T)\,\mathrm{d}\nu=acT^3/\pi$ , and check equation (3) against Carroll and Ostlie Eq. (10.68). Anchor numbers: $a=7.566\times10^{-15}$ erg cm⁻³ K⁻⁴; $\kappa_{\mathrm{es}}=0.20(1+X)=0.34$ cm² g⁻¹ at $X=0.70$ ; $L_{\mathrm{Edd}}\approx1.5\times10^{38}(M/M_\odot)$ erg s⁻¹ (the canonical $1.26\times10^{38}$ for pure ionized hydrogen at $\kappa=0.40$ ); two-band $\bar\kappa\approx9.9$ cm² g⁻¹; solar photon escape time $\sim5\times10^4$ yr.

Builds on. The Equation of Radiative Transfer · Optical Depth · Opacity and Absorption Coefficients · Blackbody Radiation and the Planck Law · Local Thermodynamic Equilibrium · Specific Intensity and Radiative Flux. Leads to. Stefan-Boltzmann and Wien Displacement Laws · Emission and the Source Function · Kirchhoff's Laws and Thermal Emission.

Ask the tutor: ask me to derive the harmonic mean from the two-band model, or to estimate the opacity, mean free path, and photon escape time for any star you name.

Why is the Rosseland mean a harmonic average and not a simple one?

Because diffusion adds fluxes, and each frequency's flux is proportional to its mean free path $1/\kappa_\nu$ . Summing over frequencies sums $1/\kappa_\nu$ values, and the average of reciprocals is a harmonic mean. Physically it is a parallel-resistor problem: radiation flows most easily through the least resistive (most transparent) channels, so those channels dominate the total conductance. A straight arithmetic average of $\kappa_\nu$ would instead be dominated by the opaque lines and drastically overstate how hard it is for radiation to escape.

Why is the weighting function the temperature derivative of the Planck function?

Diffusion is driven by the temperature gradient, and in LTE the only thing that varies with position at fixed frequency is the temperature. Differentiating the monochromatic energy density with respect to radius brings out $\partial B_\nu/\partial T$ by the chain rule: each frequency contributes to the flux in proportion to how strongly its Planck intensity responds to a small temperature change. Frequencies where $\partial B_\nu/\partial T$ is large are the ones carrying the diffusive current, so they set the mean. Weighting by $B_\nu$ itself would answer a different question — the total energy present, not the flux transported.

At what optical depth does the diffusion approximation break down?

Around $\tau\lesssim$ a few, meaning near the photosphere and everywhere above it. Deep down, a photon is absorbed and re-emitted so many times over any temperature scale height that the field is isotropic to one part in $\ell/L$ , and the first-order expansion $I\approx B-\mu\,\mathrm{d}B/\mathrm{d}\tau$ is excellent. As the optical depth falls toward unity, the mean free path grows comparable to the scale height, photons begin to stream freely outward into empty space, and the intensity develops a strong one-sided anisotropy that the two-term expansion cannot capture. There you must solve the full transfer equation or use a two-stream / Eddington closure.

How is this the same 1 over 3 that appears in radiation pressure?

The intensity's angular moments form a hierarchy: the zeroth moment is the energy density, the first (odd) moment is the flux, and the second (even) moment is the pressure. The diffusion approximation keeps just the first two nontrivial moments, and both the flux law and the pressure involve the same angular integral $\int\mu^2\,\mathrm{d}\mu$ , whose sphere average is $\langle\cos^2\theta\rangle=1/3$ . That is why the diffusion coefficient carries a $1/3$ and the radiation pressure is $P_{\mathrm{rad}}=\tfrac13 u_{\mathrm{rad}}$ — they are two moments of one nearly isotropic field, not two independent facts.

Does electron scattering really give an exact Rosseland mean?

Yes, in the non-relativistic Thomson limit. Thomson scattering has a cross section independent of frequency, so $\kappa_\nu=\kappa_{\mathrm{es}}$ for every $\nu$ ; the harmonic mean of a constant is that constant and the weighting integrates out, giving $\bar\kappa=\kappa_{\mathrm{es}}=0.20(1+X)$ cm² g⁻¹ exactly. This is the opacity floor of hot, fully ionized interiors: however transparent the gas is to true absorption, free electrons still scatter, and in the densest cores scattering dominates. Relativistic corrections lower it slightly at very high temperature, but the frequency independence — and hence the exact mean — holds well for stellar interiors.

Why do stellar models need the Rosseland mean tabulated so precisely?

Because the temperature gradient, and through it the whole structure of a star, scales linearly with $\bar\kappa$ : a ten-percent opacity error shifts the radiative gradient by ten percent, moving convection-zone boundaries and changing predicted radii, effective temperatures, and ages. Opacity also governs pulsation driving (the kappa-mechanism) and the Eddington limit that caps stellar masses. That is why enormous effort goes into the OPAL, OP, and molecular tables, why the solar abundance revisions caused a helioseismology crisis, and why laboratories now measure iron opacity directly to check the calculations. The single number the model reads off the table is always a Rosseland mean.