Phases of the Interstellar Medium

The space between the stars looks empty but is not: it holds gas and dust making up several percent of the baryonic mass of the Milky Way's disk — every gram of raw material from which new stars are built. The surprise is that this material is not one gas but a set of coexisting phases spanning roughly six orders of magnitude in temperature (from $10\ \mathrm{K}$ molecular cores to a $10^7\ \mathrm{K}$ supernova-blown corona) and eight in density, and remarkably they sit side by side at nearly the same pressure. A cold dense cloud floats in warm diffuse gas the way a raindrop floats in fog, and a bubble of million-degree plasma carves a cavity through both. Why should nature partition interstellar gas this way rather than smear it into one smooth fluid? The answer is a competition between heating and cooling. Interstellar heating — photoelectrons off dust grains, cosmic rays, starlight — scales roughly with density, while radiative cooling by fine-structure lines and metal transitions has a steep, non-monotonic temperature dependence. Set the two equal for an equilibrium locus in the density–temperature plane; plot it as equilibrium pressure versus density and the curve folds into an S, with two stable solutions — one cold and dense, one warm and diffuse — over a band of pressures. That fold, found by George Field in 1965, is the engine of the multiphase medium, and everything here follows from it: the five-phase census, the diagnostics that separate the phases, and the supernova-driven cycle that stirs the system.

Key takeaways

The interstellar medium (ISM) is not a single gas but coexisting phases spanning $\sim 10$ – $10^7\ \mathrm{K}$ and $\sim 10^{-3}$ – $10^6\ \mathrm{cm^{-3}}$ , held in rough pressure balance.
Field's (1965) thermal-instability analysis, extended by McKee and Ostriker (1977), gives five phases: molecular gas, cold neutral (CNM), warm neutral (WNM), warm ionized (WIM), and hot ionized medium (HIM).
Distinct phases exist because the equilibrium curve $P_{\rm eq}(n)$ is S-shaped: at one pressure, cold dense and warm diffuse gas are both stable — the two-phase medium.
The diffuse phases share a mean pressure $P/k \approx 3000$ – $4000\ \mathrm{K\,cm^{-3}}$ ; molecular clouds are self-gravitating and over-pressured, the hot phase a transient supernova product.
Each phase has its own diagnostic: 21 cm emission and absorption for atomic gas, CO and dust for molecular, Hα and pulsar dispersion measures for the WIM, soft X-rays and O VI for the HIM.
Key idea: because interstellar heating scales roughly with density while cooling has a strong non-monotonic temperature dependence, the thermal-equilibrium curve is multivalued in pressure — at one pressure the gas can sit cold-and-dense or warm-and-diffuse, both stable, with an unstable branch between. That is why the ISM is intrinsically multiphase.

The core insight

Because interstellar heating scales roughly with density while cooling has a strong non-monotonic temperature dependence, the thermal-equilibrium curve is multivalued in pressure: at a single pressure the gas can sit in a cold dense phase or a warm diffuse phase, both stable, with an unstable branch between. This is why the ISM is intrinsically multiphase rather than a single smooth fluid.

Where this lives today

The multiphase framework is the working language of every effort to understand how galaxies turn gas into stars. Galaxy-formation simulations — FIRE, IllustrisTNG, stratified-disk boxes — live or die on their "sub-grid" ISM: whether they resolve cold clouds, how they inject supernova energy, and how fast thermal instability partitions cooling gas set the star-formation rate and the strength of galactic winds. Observationally the phases are mapped ever more finely: HI4PI and THOR separate CNM absorption from WNM emission in 21 cm; ALMA resolves the molecular phase into filaments and cores where the CO-to-H2 conversion is recalibrated cloud by cloud; and eROSITA has traced the hot phase into the vast "eROSITA bubbles" over the Galactic center — direct images of the supernova-driven fountain. The WIM is a nuisance turned resource: its dispersion of pulsar signals now localizes cosmological fast radio bursts and weighs the invisible baryons between galaxies. The open problems are pointed — what sets the mean ISM pressure, how much mass sits on the "forbidden" branch, how fields and cosmic rays reshape the balance — all attacked with the equilibrium-and-feedback machinery Field set in motion.

The physical picture

Start with the water-and-steam analogy, exact in structure if not in physics: at a fixed pressure, liquid and vapour can both be stable, separated by a forbidden intermediate state. The interstellar version is not a phase transition of matter but a bistability of thermal equilibrium. Treat density as the knob and the equilibrium pressure the gas can sustain as the response. Turn density up in a warm diffuse cloud and, over a certain range, that pressure actually falls — the surroundings squeeze it further, it cools and compresses, and it runs away until it lands on the cold, dense branch. The stable ends of the S-curve are the warm neutral medium (WNM, $T\sim 8000\ \mathrm{K}$ ) and the cold neutral medium (CNM, $T\sim 80\ \mathrm{K}$ ); the unstable middle is why gas piles up at the ends. The first figure is that S-curve, drawn with real solar-neighborhood numbers.

Figure

The Field thermal-equilibrium curve: equilibrium pressure P/k versus density for diffuse atomic gas. A horizontal line at the mean ISM pressure P/k ≈ 3800 K cm⁻³ cuts the S-shaped curve three times — a stable warm branch (WNM, n ≈ 0.3 cm⁻³), an unstable middle branch where ∂P/∂n is less than 0, and a stable cold branch (CNM, n ≈ 30 cm⁻³) — so cold and warm neutral gas coexist at one pressure

The hot phase follows a different logic: not a heating–cooling balance but a fossil of energy injection. A supernova shock-heats gas past $10^6\ \mathrm{K}$ , where the cooling function collapses toward its bremsstrahlung floor; with almost nothing to radiate through and $n \sim 10^{-3}\ \mathrm{cm^{-3}}$ , the gas takes tens of millions of years to cool — long enough that fresh supernovae replenish it and hot bubbles overlap into a pervasive corona. The molecular phase sits at the opposite corner: once a cold cloud grows dense and dusty enough to shield ultraviolet light, hydrogen turns molecular, CO forms, the gas cools toward $10\ \mathrm{K}$ , and self-gravity takes over. The second figure lays all five phases on one map, with a single isobar threading through them.

Figure

The five-phase census in the density–temperature plane, with each phase at its canonical n and T. The dashed line is the isobar P/k = nT = 3800 K cm⁻³: the WNM, WIM, and CNM sit on or near it, the hot ionized medium runs slightly under-pressured, and molecular gas floats well above the line because it is self-gravitating and over-pressured

The core equations

The diffuse ISM is nearly an ideal gas, so the quantity coupling the phases is thermal pressure, quoted divided by Boltzmann's constant because $nT$ is what density and temperature measurements return directly.

P/k = n\,T \approx 3000\text{–}4000\ \mathrm{K\,cm^{-3}} \qquad (1)

where: $P$ — thermal pressure; $k$ — Boltzmann constant, $1.381\times10^{-16}\ \mathrm{erg\,K^{-1}}$ ; $n$ — total particle number density; $T$ — kinetic temperature. Reading it: the diffuse phases live at nearly one value of $nT$ , so a hundred-fold density jump offsets a hundred-fold temperature drop — the CNM-versus-WNM contrast. The origin of the phases is the energy equation: heating adds $\rho\Gamma$ per volume, cooling removes $n^2\Lambda(T)$ , and their equality defines thermal equilibrium.

\rho\,\frac{\mathrm{d}u}{\mathrm{d}t} = \rho\,\Gamma - n^2\Lambda(T) \equiv -\rho\,\mathcal{L} \qquad (2)

where: $u$ — specific internal energy per mass; $\Gamma$ — heating rate per mass, nearly density-independent; $\Lambda(T)$ — cooling function, power radiated per volume $= n^2\Lambda$ ; $\mathcal{L}$ — net loss function. Reading it: setting $\mathcal{L}=0$ picks out one curve in the $(n,T)$ plane, and the whole phase structure is its shape. Because heating per particle is nearly constant while cooling per particle scales with $n$ , the condition $\rho\Gamma = n^2\Lambda$ gives a temperature that falls as density rises, $T_{\rm eq}(n)$ ; times $n$ this is the sustainable pressure — the S-curve of the first figure.

P_{\rm eq}(n) = n\,k\,T_{\rm eq}(n)\quad\text{— multivalued: stable, unstable, stable} \qquad (3)

where: $P_{\rm eq}$ — pressure along the equilibrium locus; $T_{\rm eq}(n)$ — equilibrium temperature at density $n$ . Reading it: over a band of pressures a horizontal line meets the curve three times, so three equilibria share one pressure. Which of the three the gas can occupy is set by a stability test: compress a parcel at fixed pressure, and if the pressure it can hold drops, the surroundings overwhelm it.

\left(\frac{\partial P}{\partial n}\right)_{\rm eq} < 0 \;\Longleftrightarrow\; \text{thermally unstable} \qquad (4)

where: $(\partial P/\partial n)_{\rm eq}$ — slope of the equilibrium curve. Reading it: the positive-slope branches (WNM, CNM) are stable and populated; the negative-slope middle is Field-unstable and empties out. The hot phase survives because it cannot radiate fast enough. Dividing thermal energy density by radiated power gives the cooling time that controls how long a phase persists once made.

t_{\rm cool} = \frac{(3/2)\,n\,k\,T}{n^2\,\Lambda(T)} = \frac{3\,k\,T}{2\,n\,\Lambda(T)} \qquad (5)

where: $t_{\rm cool}$ — time to radiate the thermal content; $\Lambda(T)$ — cooling function. Reading it: the inverse dependence on $n$ is decisive — the hot phase's tiny density plus a collapsed $\Lambda$ above $10^6\ \mathrm{K}$ give $t_{\rm cool} \sim 10^7$ – $10^8\ \mathrm{yr}$ . Finally, the ionized phases are read through electron columns: the dispersion measure to a pulsar is the workhorse WIM diagnostic, and the emission measure sets its Hα brightness.

DM = \int_0^L n_e\,\mathrm{d}\ell,\qquad EM = \int_0^L n_e^2\,\mathrm{d}\ell \qquad (6)

where: $DM$ — dispersion measure (electron column); $EM$ — emission measure, fixing recombination-line and free–free brightness; $n_e$ — free-electron density; $L$ — path length. Reading it: $DM$ is linear in $n_e$ and $EM$ quadratic, so comparing them separates mean density from clumpiness (the filling factor).

Derivation AThe two-phase medium from a heating and cooling balance

The goal is to show, without special assumptions about the exact cooling curve, that a heating rate proportional to density combined with a peaked cooling function forces the equilibrium pressure to be multivalued in density, so two stable phases of neutral gas must coexist at one pressure. We build the equilibrium curve step by step and read its stability off its slope, reproducing the Field–Goldsmith–Habing construction from the energy equation alone.

Step 1. Set the net loss to zero in equation (2): $\rho\Gamma = n^2\Lambda(T)$ . With $\rho \propto n$ , divide by density to get $\Gamma \propto n\,\Lambda(T)$ — heating per particle balancing cooling per particle. Why: dividing by $n$ removes the trivial density scaling and exposes the one genuine $n$ – $T$ relation equilibrium demands.

Step 2. Use the measured shape of $\Lambda(T)$ . Below $\sim 10^4\ \mathrm{K}$ cooling is dominated by fine-structure lines — singly-ionized carbon at $158\ \mu\mathrm{m}$ , neutral oxygen at $63\ \mu\mathrm{m}$ — rising steeply with temperature, and near $10^4\ \mathrm{K}$ collisionally excited Lyman-α spikes. With $\Gamma$ nearly fixed, this peaked $\Lambda$ forces $T_{\rm eq}(n)$ from $\sim 8000\ \mathrm{K}$ down to $\sim 80\ \mathrm{K}$ as $n$ rises. Why: the steep, non-monotonic cooling — not gravity — maps one density to wildly different equilibrium temperatures at each end.

Step 3. Form $P_{\rm eq}(n) = n\,k\,T_{\rm eq}(n)$ . Multiplying a decreasing $T_{\rm eq}(n)$ by an increasing $n$ gives a curve that first rises (WNM, the $n$ factor winning), then falls (the intermediate branch, the plunge in $T_{\rm eq}$ winning), then rises again (CNM) — the S-shape of the first figure. Why: the fold is a pure consequence of a product of one rising and one sharply falling factor; no fine tuning is needed.

Step 4. Read stability from the slope, equation (4). A horizontal line at the mean pressure crosses the fold three times: the outer crossings have $(\partial P/\partial n)_{\rm eq} > 0$ and are stable, the middle has negative slope and is unstable, so gas there condenses onto a branch — cold and warm neutral gas coexisting at one pressure. Why: a positive-slope parcel resists compression and pushes back; a negative-slope parcel has its supportable pressure lowered by compression and runs away.

Step 5. Bound the coexistence range by the fold's turning points $P_{\min}$ and $P_{\max}$ . For solar-neighborhood rates this band is $P/k \sim 10^3$ – $10^4\ \mathrm{K\,cm^{-3}}$ , bracketing the observed mean $\approx 3800\ \mathrm{K\,cm^{-3}}$ . Why: the mean pressure landing inside the bistable band turns a mathematical possibility into the observed CNM/WNM coexistence. *Verified: reproduces the Field–Goldsmith–Habing (1969) two-phase result; the S-shaped $P_{\rm eq}(n)$ and the sign change of $(\partial P/\partial n)_{\rm eq}$ delimit the coexistence range, which contains the observed mean pressure.*

Go deeper · Derivation B · 1 minField's isobaric instability criterion

Derivation BField's isobaric instability criterion

The goal is to turn the qualitative slope argument into the sharp condition Field derived in 1965, expressing thermal instability through how the net loss function responds to a temperature perturbation at constant pressure. This matters because the criterion in terms of $\mathcal{L}$ is what appears in the literature and is not obviously the same statement as "negative slope of $P_{\rm eq}(n)$ ", yet the two are identical.

Step 1. Perturb an equilibrium parcel isobarically: hold $P = nkT$ fixed and let the temperature drop by $\delta T < 0$ , so the density rises, $\delta n/n = -\delta T/T$ , and the parcel condenses as it cools. Why: interstellar perturbations relax pressure at the sound-crossing rate, much faster than they cool, so the relevant perturbation is at constant pressure, not constant volume.

Step 2. The perturbation grows if the net loss function increases as the gas cools at constant pressure — if radiating harder makes it cool faster still:

\left(\frac{\partial \mathcal{L}}{\partial T}\right)_{P} < 0 \qquad (\text{Field 1965})

Why: a runaway needs positive feedback; the colder parcel must lose energy even more efficiently, or it just relaxes back.

Step 3. Along the locus $\mathcal{L}=0$ , differentiating shows $(\partial \mathcal{L}/\partial T)_P < 0$ is algebraically equivalent to $(\partial P/\partial n)_{\rm eq} < 0$ : the branch where cooling runs away is exactly the negative-slope branch. Why: both are one fact — on the unstable branch an imposed compression is amplified, not resisted, whether phrased through $\mathcal{L}(T)$ or $P_{\rm eq}(n)$ . *Verified: the isobaric criterion $(\partial \mathcal{L}/\partial T)_P < 0$ is the standard Field (1965) condition and maps onto the negative-slope segment of the previous derivation's curve.*

Go deeper · Derivation C · 1 minCooling time and the persistence of the hot phase

Derivation CCooling time and the persistence of the hot phase

The goal is to estimate how long supernova-shocked gas stays hot, and thereby explain why a $10^6$ – $10^7\ \mathrm{K}$ phase can fill much of the disk volume rather than radiating away almost instantly. The argument rests on the ratio of thermal energy to radiated power and on the collapse of the cooling function above $10^6\ \mathrm{K}$ , so we assemble it from the energy content and the emissivity, then insert hot-phase numbers.

Step 1. Write the thermal energy density of an ideal monatomic gas, $E_{\rm th} = \tfrac{3}{2}\,n\,k\,T$ . Why: only the thermal energy — not magnetic or bulk-kinetic — feeds the radiative channels that set the cooling time.

Step 2. Write the radiated power per volume, $\mathcal{P} = n^2\Lambda(T)$ , quadratic in density because collisional excitation needs two particles to meet. Why: the $n^2$ scaling is what makes cooling collapse in tenuous gas — the single most important fact about the hot phase.

Step 3. Divide: $t_{\rm cool} = E_{\rm th}/\mathcal{P} = 3kT/(2n\Lambda)$ , falling linearly with density. Why: the inverse dependence on $n$ means a phase a thousand times more tenuous than the CNM cools a thousand times more slowly at fixed $\Lambda$ .

Step 4. Insert hot-phase values: above $10^6\ \mathrm{K}$ , $\Lambda \sim 10^{-22}\ \mathrm{erg\,cm^3\,s^{-1}}$ near the bremsstrahlung floor and $n \sim 10^{-3}\ \mathrm{cm^{-3}}$ , giving $t_{\rm cool} \sim 10^7$ – $10^8\ \mathrm{yr}$ — comparable to or longer than the interval between nearby supernovae. Why: since $t_{\rm cool}$ exceeds the supernova recurrence time, fresh explosions top the gas up faster than it cools, so hot bubbles overlap into a pervasive corona — the third phase of McKee and Ostriker. *Verified: order-of-magnitude estimate consistent with McKee and Ostriker (1977); the long cooling time follows directly from the low density and the reduced $\Lambda(T)$ above $10^6\ \mathrm{K}$ , and is computed explicitly in the worked examples.*

Limiting cases

Each phase occupies a characteristic corner of the $(n,T)$ plane, and its dominant physics changes accordingly:

Case	Result	Meaning
Molecular gas	$T\sim 10\text{–}50\ \mathrm{K},\ n\gtrsim 10^2\ \mathrm{cm^{-3}}$	Self-gravitating and over-pressured; cooling by CO and dust; UV-shielded; the site of star formation.
Cold + warm neutral (CNM/WNM)	$T\sim 80\ \mathrm{K}$ and $\sim 8000\ \mathrm{K}$	Two stable atomic branches coexisting near $P/k\sim 3800\ \mathrm{K\,cm^{-3}}$ ; the classic Field two-phase medium.
Warm ionized (WIM)	$T\sim 8000\ \mathrm{K},\ n\sim 0.1\ \mathrm{cm^{-3}}$	Photoionized diffuse gas (the Reynolds layer), traced by Hα and pulsar dispersion measures.
Hot ionized (HIM)	$T\sim 10^6\text{–}10^7\ \mathrm{K},\ n\sim 10^{-3}\ \mathrm{cm^{-3}}$	Shock-heated by supernovae; long cooling time; fills a large volume fraction; seen in soft X-rays and O VI.

Common trapDo not treat the phases as being in strict, static pressure equilibrium

"Pressure equilibrium" is a useful zeroth-order idea, not a law. Three caveats matter. First, molecular clouds are held by their own gravity and are internally over-pressured — they do not float at the mean ISM pressure at all. Second, the hot phase is a transient, supernova-driven structure whose pressure tracks the local star-formation history, not a thermal-equilibrium branch. Third, and most easily forgotten, the total pressure balancing the diffuse phases includes magnetic, cosmic-ray, and turbulent terms each comparable to the thermal one, so even the CNM and WNM are only in rough thermal-pressure balance, with large parcel-to-parcel deviations. Use it as a guide to the mean, never as a constraint you can impose exactly.

Worked examples

1 · Densities of the diffuse phases at one pressure (fully worked)

The diffuse ISM sits at $P/k \approx 3800\ \mathrm{K\,cm^{-3}}$ . Find the densities of the WNM ( $T = 8000\ \mathrm{K}$ ), CNM ( $T = 80\ \mathrm{K}$ ), and WIM ( $T = 8000\ \mathrm{K}$ ), and decide whether a molecular cloud with $n = 10^3\ \mathrm{cm^{-3}}$ , $T = 20\ \mathrm{K}$ is over- or under-pressured.

Step 1. Invert equation (1), $n = (P/k)/T$ . WNM: $n = 3800/8000 = \mathbf{0.48\ cm^{-3}}$ . CNM: $n = 3800/80 = \mathbf{48\ cm^{-3}}$ . Both match canonical values, and the hundred-fold density contrast is exactly the hundred-fold temperature contrast, as pressure balance demands.

Step 2. The WIM shares the warm temperature but only partly fills its volume: the mean pressure would give $n \approx 0.48\ \mathrm{cm^{-3}}$ , yet the observed WIM density is lower, $n_e \sim 0.1\ \mathrm{cm^{-3}}$ — the bright Hα gas is clumped into a small filling fraction.

Step 3. For the molecular cloud, $P/k = nT = 10^3 \times 20 = 2\times10^4\ \mathrm{K\,cm^{-3}}$ , about five times the mean pressure — over-pressured, so bound by self-gravity, not confined by the ambient medium. WNM 0.48 cm⁻³, CNM 48 cm⁻³ (both on the isobar); the molecular cloud is ~5× over-pressured and self-gravitating.

2 · Cooling time of the hot ionized medium (fully worked)

Estimate the cooling time of supernova-heated gas at $T = 10^6\ \mathrm{K}$ , $n = 10^{-3}\ \mathrm{cm^{-3}}$ , with $\Lambda \approx 10^{-22}\ \mathrm{erg\,cm^3\,s^{-1}}$ . Compare to the local supernova interval, a few $\times 10^6\ \mathrm{yr}$ .

Step 1. Thermal energy density: $E_{\rm th} = \tfrac{3}{2}nkT = 1.5 \times 10^{-3} \times 1.381\times10^{-16} \times 10^6 = 2.07\times10^{-13}\ \mathrm{erg\,cm^{-3}}$ .

Step 2. Radiated power density: $\mathcal{P} = n^2\Lambda = (10^{-3})^2 \times 10^{-22} = 10^{-28}\ \mathrm{erg\,cm^{-3}\,s^{-1}}$ .

Step 3. Divide: $t_{\rm cool} = 2.07\times10^{-13}/10^{-28} = 2.1\times10^{15}\ \mathrm{s}$ , and with $1\ \mathrm{yr} = 3.16\times10^7\ \mathrm{s}$ , $t_{\rm cool} \approx \mathbf{6.6\times10^{7}\ yr}$ .

Step 4. This is roughly ten times the supernova recurrence interval, so new explosions refresh a hot bubble before it cools, and bubbles overlap into a pervasive corona filling much of the disk. (The exact value swings over $\sim 10^7$ – $10^8\ \mathrm{yr}$ with $\Lambda$ , but $t_{\rm cool}$ exceeding the supernova interval is robust.) t_cool ≈ 7×10⁷ yr, longer than the supernova recurrence time, so the hot phase persists and fills volume.

3 · Reading the warm ionized medium from a pulsar dispersion measure

A pulsar at distance $L = 1\ \mathrm{kpc}$ has $DM = 30\ \mathrm{pc\,cm^{-3}}$ . Estimate the mean free-electron density and reconcile it with the WIM's local $n_e \sim 0.1\ \mathrm{cm^{-3}}$ .

Step 1. By equation (6), $\langle n_e\rangle = DM/L = 30/1000 = \mathbf{0.03\ cm^{-3}}$ .

Step 2. This is a few times below the WIM's in-situ $\sim 0.1\ \mathrm{cm^{-3}}$ — not an error but the filling factor: the ionized gas occupies only $f \sim \langle n_e\rangle/n_e \sim 0.3$ of the path volume, so the smooth average dilutes the denser clumps. Mean n_e ≈ 0.03 cm⁻³ implies a WIM filling factor of order 0.2–0.3.

Physics of the portrait

The multiphase ISM rests on one insight applied relentlessly: the balance of heating and cooling, not gravity alone, sets the structure of interstellar gas — supplied by George Field and grown into the modern supernova-regulated picture.

Origins · 1 minGeorge B. Field and how this was discovered

George B. Field

b. 1929 · thermal instability and the multiphase interstellar medium

George Brooks Field is an American astrophysicist whose 1965 paper "Thermal Instability" (Astrophysical Journal 142, 531) supplied the theoretical spine of the subject. From the linearized energy equation he derived when small temperature perturbations in a radiating gas grow rather than damp, producing the isobaric, isochoric, and isentropic criteria that carry his name. The isobaric criterion showed diffuse interstellar gas is unstable to condensation over a range of conditions — why a smooth medium cannot persist. Four years later, with Goldsmith and Habing (1969), he turned that instability into a prediction: a two-phase neutral medium in which cold dense and warm diffuse gas coexist in pressure equilibrium, the CNM and WNM confirmed decades later in 21 cm absorption. Field had also done foundational work on the 21 cm line's excitation — the Wouthuysen–Field effect now underpins studies of cosmic reionization — so one physicist tied together the theory of the phases and the tool used to observe them. When McKee and Ostriker added the hot phase in 1977 they extended his framework, and the modern $P_{\rm eq}(n)$ curves of Wolfire and collaborators are recognizably his 1965 argument with better atomic data.

History

1930

— Trumpler proves interstellar dust exists from the reddening of star clusters

1965

— Field derives the thermal-instability criterion for diffuse interstellar gas

1969

— Field, Goldsmith and Habing formulate the two-phase neutral medium

1977

— McKee and Ostriker add the supernova-driven hot phase, giving three phases

2003

— Wolfire and colleagues compute modern CNM–WNM equilibrium curves with updated rates

Going further

Texts. Draine, Physics of the Interstellar and Intergalactic Medium (2011), Chs. 30–34 · Carroll and Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 12 · Binney and Tremaine, Galactic Dynamics, 2nd ed., §8 · Tielens, The Physics and Chemistry of the Interstellar Medium (2005).

Papers. Field, G. B. (1965), ApJ 142, 531 — thermal instability · Field, Goldsmith and Habing (1969), ApJ 155, L149 — the two-phase model · McKee and Ostriker (1977), ApJ 218, 148 — the three-phase model · Wolfire et al. (2003), ApJ 587, 278 — modern neutral-phase equilibria.

Provenance. Phase temperatures, densities, and the mean pressure $P/k \approx 3000$ – $4000\ \mathrm{K\,cm^{-3}}$ are standard values from Draine (2011) and Carroll and Ostlie Ch. 12; a verifier can recompute $n = (P/k)/T$ for the WNM ( $0.48\ \mathrm{cm^{-3}}$ ) and CNM ( $48\ \mathrm{cm^{-3}}$ ), the HIM cooling time ( $\sim 7\times10^7\ \mathrm{yr}$ for $\Lambda \approx 10^{-22}\ \mathrm{erg\,cm^3\,s^{-1}}$ ), and $\langle n_e\rangle$ from a pulsar $DM$ . The two-phase construction and the criterion $(\partial\mathcal{L}/\partial T)_P < 0$ follow from Field (1965) and Field, Goldsmith and Habing (1969); the cooling-time argument from McKee and Ostriker (1977). The cooling function is quoted at order-of-magnitude precision, as is standard.

Prerequisites. Interstellar Gas Heating and Cooling · Thermal Instability and the Two-Phase Medium. Leads to. The 21 cm Line of Neutral Hydrogen · Molecular Clouds · Turbulence in the Interstellar Medium · CO as a Molecular Gas Tracer and the X-Factor.

Ask the tutor: Ask me to walk through why the same mean pressure lands cold and warm neutral gas on two different stable branches — or to estimate the cooling time and filling factor of any phase from its density and temperature.

How many phases are there really — two, three, or five?

All three counts are correct at different resolutions. Field's original result is a two-phase neutral medium: CNM and WNM, the two stable branches of one equilibrium curve. McKee and Ostriker added the supernova-blown hot ionized medium for a three-phase model that also accounts for volume filling. The modern working census is five, further distinguishing molecular gas (colder, denser, self-gravitating) and the warm ionized medium (photoionized rather than neutral). These are not five independent theories but the two-phase equilibrium plus the transient hot phase plus two chemically distinct additions.

Are the phases actually in pressure equilibrium?

Only roughly, and only for the diffuse gas. The CNM, WNM, and WIM cluster around $P/k \approx 3000$ – $4000\ \mathrm{K\,cm^{-3}}$ — coexistence requires a shared pressure. But molecular clouds are self-gravitating and over-pressured, sometimes tenfold, and the hot phase is a transient feedback product whose pressure tracks recent supernovae. Even among the diffuse phases the balance is approximate, because magnetic, cosmic-ray, and turbulent pressures each rival the thermal term, so the total pressure is what is really shared. Treat it as a statement about the mean, not a constraint on every parcel.

Why is molecular hydrogen invisible, and why use CO instead?

Molecular hydrogen, H2, is symmetric with no permanent dipole moment, so it has no allowed rotational dipole transitions, and at the $10$ – $20\ \mathrm{K}$ of molecular clouds there is not enough energy to excite its lowest quadrupole line — the dominant mass component radiates essentially nothing. Carbon monoxide solves this: it is next most abundant, it does have a dipole moment, and its lowest rotational line at $2.6\ \mathrm{mm}$ is easily excited at cloud temperatures. Astronomers observe CO and convert to H2 column via the calibrated CO-to-H2 X-factor, cross-checked against dust and extinction maps.

What is the galactic fountain?

The galactic fountain is the large-scale circulation the hot phase drives. Supernovae heat gas past $10^6\ \mathrm{K}$ ; this buoyant, over-pressured plasma rises into the halo, cools, and condenses into clouds that rain back onto the disk. The cycle matters twice: it redistributes freshly synthesized metals far from where they were made, flattening abundance gradients, and it reshuffles the mass budget among the phases as gas leaves hot and returns cool. The eROSITA bubbles above the Galactic center are a resolved image of the hot outflow that powers it.

Does thermal instability create the phases, or just permit them?

It permits and organizes them but does not supply the energy. Thermal instability is a statement about equilibria: intermediate-temperature gas is unstable, so gas sorts itself onto the cold and warm branches. What drives parcels between branches is dynamics — spiral-arm compression, supernova shells, turbulent flows — pushing gas across the unstable region, where instability finishes quickly. It is a division of labor: turbulence and feedback stir and compress, thermal instability selects the end states, and gravity gathers the densest gas into clouds. Simulations show real mass caught in transit across the "forbidden" branch, so it is not entirely empty.

How is the warm ionized medium kept ionized so far from hot stars?

This puzzled astronomers when Reynolds mapped the diffuse Hα layer in the 1970s–80s: the WIM is spread through the disk and halo, far from the O and B stars that make ionizing photons, yet stays ionized at $\sim 8000\ \mathrm{K}$ . The answer is that ionizing photons leak out of classical H II regions through low-density channels carved by earlier supernovae and winds; because recombination is slow at the WIM's low density, an escaped photon travels far and keeps gas ionized. The Galaxy's massive stars, with a modest escape fraction, supply enough — though extra heating is needed to explain the WIM's line ratios, and its source is still debated.