CO as a Molecular Gas Tracer and the X-Factor

The mass budget of a star-forming galaxy is dominated by its coldest, densest phase — the molecular clouds out of which stars condense. Yet the overwhelming constituent of that gas, molecular hydrogen $\mathrm{H_2}$ , is almost perfectly silent at the $10\text{--}20\,\mathrm{K}$ temperatures that prevail there. Being a homonuclear diatomic molecule, $\mathrm{H_2}$ carries no permanent electric dipole moment and so has no allowed dipole rotational spectrum at all; its lowest radiative transition is a weak electric-quadrupole line whose upper level sits some $510\,\mathrm{K}$ above ground. Direct $\mathrm{H_2}$ emission therefore selects only warm, shocked, or ultraviolet-illuminated gas and misses the entire cold reservoir. This is the central observational problem that carbon monoxide solves. CO is the workhorse substitute: chemically robust, forming wherever $\mathrm{H_2}$ is shielded, and — because its C–O charge distribution is lopsided — carrying a small permanent dipole moment ( $\mu \approx 0.11\,\mathrm{D}$ ) and a ladder of low-energy rotational lines. The ground $J{=}1\to0$ transition at $115.27\,\mathrm{GHz}$ ( $\lambda = 2.6\,\mathrm{mm}$ ) has an excitation energy of only $E_{10}/k_B \approx 5.5\,\mathrm{K}$ and is collisionally excited throughout the medium. The proportionality between its velocity-integrated brightness and the underlying $\mathrm{H_2}$ column — the X-factor — makes CO the standard molecular-gas ruler of Galactic and extragalactic astronomy. The subtlety, and the reason the X-factor is empirical rather than derived, is that the $\mathrm{CO}(1{\to}0)$ line is optically thick in ordinary clouds, so a naive column estimate from the saturated line fails badly. Instead the calibration rests on the fact that giant molecular clouds are gravitationally bound, approximately virialized structures: the integrated CO luminosity of an unresolved ensemble counts the clouds, and virial scaling ties that count to total mass. This is why $X_{\rm CO}$ is remarkably stable across normal disks — and precisely why it drifts wherever the assumptions break.

Key takeaways

Molecular hydrogen has no permanent dipole moment and its first emitting level lies $\sim 510\,\mathrm{K}$ above ground, so it is invisible in the cold ( $10\text{--}20\,\mathrm{K}$ ) gas that holds most of the molecular mass.
CO, second most abundant molecule (abundance $\sim 10^{-4}$ relative to $\mathrm{H_2}$ ), has low-lying rotational lines — $J{=}1{\to}0$ at $115.27\,\mathrm{GHz}$ , $E_{10}/k_B\approx 5.5\,\mathrm{K}$ — that are easily excited and observed.
The velocity-integrated line intensity $W_{\rm CO}$ maps to an $\mathrm{H_2}$ column through the empirical X-factor, $N_{\mathrm{H_2}}=X_{\rm CO}\,W_{\rm CO}$ , with a Milky Way value $X_{\rm CO}\approx 2\times10^{20}\,\mathrm{cm^{-2}\,(K\,km\,s^{-1})^{-1}}$ .
The justification is not that the line is thin — it is highly optically thick — but that clouds are virialized clumps whose CO luminosity tracks their gravitating mass.
$X_{\rm CO}$ is not universal: it rises in low-metallicity, dust-poor gas ('CO-dark' $\mathrm{H_2}$ ) and falls in warm, turbulent starburst and ULIRG nuclei — the dominant systematic in molecular-mass estimates.
Key idea: $X_{\rm CO}$ works not because CO emission is optically thin — it is not — but because molecular clouds are ensembles of virialized clumps whose total CO luminosity scales with their gravitating mass; the conversion counts clouds and inherits the near-universality of the virial relation.

Where this lives today

Every quantitative statement about molecular gas in galaxies passes through an adopted $X_{\rm CO}$ . ALMA and NOEMA now resolve CO across the cosmic star-formation peak at $z\sim1\text{--}3$ , where the gas fraction is far higher than today and the choice of $\alpha_{\rm CO}$ decides whether a galaxy is judged gas-rich or merely efficient. Nearby, IRAM and ALMA legacy surveys (PHANGS, EMPIRE) map CO cloud-by-cloud across whole disks to test whether $\alpha_{\rm CO}$ really tracks metallicity and dynamical state as the virial argument predicts. The 'CO-dark' gas problem — $\mathrm{H_2}$ with little or no CO — is attacked with the $[\mathrm{C\,II}]\,158\,\mu\mathrm{m}$ line, dust continuum from Herschel and Planck, and the diffuse $\gamma$ -ray method Fermi-LAT sharpened for the Milky Way. The open frontier is the low-metallicity, high-redshift regime: as surveys push to primordial dwarfs and the first galaxies, the Galactic $X_{\rm CO}$ becomes untrustworthy by more than an order of magnitude, and calibrating the conversion in metal-poor gas is a rate-limiting uncertainty in the cosmic history of molecular gas.

The physical picture

Two facts, held together, make the whole subject. First, the tracer we want ( $\mathrm{H_2}$ ) cannot radiate in the cold, while the tracer we use (CO) can — a statement about molecular energy ladders. Second, the line we detect is saturated, so its brightness cannot report a column directly; what saves the measurement is that clouds are gravitationally bound, so their line widths report their masses. The $\mathrm{H_2}$ molecule is symmetric, with no dipole to shake and only faint quadrupole lines far up the ladder; CO is asymmetric, with a gently rung rotational bell whose lowest note ( $2.6\,\mathrm{mm}$ ) is struck by collisions with the very $\mathrm{H_2}$ we are trying to weigh. The first figure fixes the excitation contrast; the second shows how a saturated line still weighs a cloud.

Figure

Rotational energy ladders of CO and H₂ set against the cloud temperature band kT ≈ 10 to 20 K. CO's J = 1 to 0 line sits at only E₁₀/k_B = 5.5 K, easily populated by collisions, whereas H₂ is homonuclear so its dipole ladder is forbidden and the lowest allowed quadrupole line J = 2 to 0 lies at E₂₀/k_B = 510 K, some 30 times above the gas — hence CO radiates and H₂ does not

Now the harder half. For a single cloud the $\mathrm{CO}(1{\to}0)$ line is optically thick, so piling more gas along the sightline does not raise the peak brightness — it is already pinned near $T_{\rm ex}\approx T_{\rm kin}$ . Instead the extra bound gas moves at more velocities and broadens the line, so the integrated intensity $W_{\rm CO}=\int T_B\,\mathrm{d}v$ grows with the velocity spread, which for a virialized cloud grows with mass. The second figure makes this concrete: two clouds of the same brightness temperature but different mass share a peak and differ in area.

Figure

Two optically thick CO lines with the same peak brightness T_B ≈ T_ex ≈ 10 K but different masses. The line does not grow taller with column — it grows wider, so the integrated area W_CO = ∫T_B dv tracks the velocity dispersion σ_v, which for a bound cloud tracks √M. Anchor values: X_CO ≈ 2×10²⁰ cm⁻² per K km s⁻¹ and α_CO ≈ 4.3 M_⊙ pc⁻² per K km s⁻¹

The X-factor breaks exactly where these pictures break. Strip out dust and metals — a dwarf galaxy — and far-ultraviolet photons photodissociate CO deep into the cloud while $\mathrm{H_2}$ self-shields and survives, so gas is under-counted and the effective $X_{\rm CO}$ rises. Heat and stir the gas — a starburst nucleus — and the discrete clouds merge into a warm, high-dispersion sea that radiates more CO per unit mass, so $X_{\rm CO}$ falls. Everything downstream is the interplay of the ladder and the virial relation.

The core equations

The X-factor is a single empirical definition; the physics lives in the relations that explain why it has a nearly universal value. We start from the definition, then supply the excitation, radiative-transfer, and virial ingredients in turn. The definition converts an observable — the velocity-integrated CO brightness — into the quantity we want, the hydrogen column. It is a calibration, not a derived constant.

N_{\mathrm{H_2}} = X_{\rm CO}\, W_{\rm CO},\qquad W_{\rm CO} \equiv \int T_B(v)\,\mathrm{d}v \qquad (1)

where: $N_{\mathrm{H_2}}$ — $\mathrm{H_2}$ column density along the sightline ( $\mathrm{cm^{-2}}$ ); $X_{\rm CO}$ — CO-to- $\mathrm{H_2}$ conversion factor ( $\mathrm{cm^{-2}\,(K\,km\,s^{-1})^{-1}}$ ); $W_{\rm CO}$ — velocity-integrated $\mathrm{CO}(1{\to}0)$ brightness temperature ( $\mathrm{K\,km\,s^{-1}}$ ); $T_B(v)$ — brightness temperature at velocity $v$ . Read it as: the area under the line profile, not its peak, sets the column. The excitation contrast between the two molecules is set entirely by the rigid-rotor spectrum. A molecule's rotational lines are cheap to excite only if its moment of inertia is large.

E_J = \frac{\hbar^2}{2I}\,J(J+1) = h B\, J(J+1),\qquad \nu_{J\to J-1} = 2BJ \qquad (2)

where: $J$ — rotational quantum number; $I=\mu_{\rm r} r_0^2$ — moment of inertia ( $\mu_{\rm r}$ reduced mass, $r_0$ bond length); $B$ — rotational constant, $B\approx 57.6\,\mathrm{GHz}$ for CO; $\nu_{J\to J-1}$ — line frequency, giving $115.27\,\mathrm{GHz}$ for $J{=}1{\to}0$ . Read it as: heavy, long-bonded CO has a large $I$ , small $B$ , and closely spaced low-energy rungs; light $\mathrm{H_2}$ has the opposite. Once excited, the line's brightness is fixed by radiative transfer through a slab at excitation temperature $T_{\rm ex}$ . This equation is why a thick line saturates.

T_B = \big[J_\nu(T_{\rm ex}) - J_\nu(T_{\rm bg})\big]\,(1 - e^{-\tau_\nu}),\quad J_\nu(T)=\frac{h\nu/k_B}{e^{h\nu/k_B T}-1} \qquad (3)

where: $T_{\rm ex}$ — excitation temperature of the transition; $T_{\rm bg}=2.725\,\mathrm{K}$ — CMB background; $\tau_\nu$ — line optical depth, typically $\gg1$ for $\mathrm{CO}(1{\to}0)$ ; $J_\nu(T)$ — Rayleigh–Jeans-equivalent radiation temperature. Read it as: when $\tau_\nu\gg1$ the factor $(1-e^{-\tau_\nu})\to1$ and $T_B$ saturates near $T_{\rm ex}$ , independent of column. (The bare critical density $n_{\rm crit}=A_{10}/\gamma_{10}\sim2\times10^{3}\,\mathrm{cm^{-3}}$ is lowered to a few $\times10^{2}\,\mathrm{cm^{-3}}$ by radiative trapping in the thick line, so CO is excited throughout the cloud.) The mass all of this is trying to measure enters through the virial theorem for a self-gravitating cloud — the load-bearing physical assumption.

M_{\rm vir} = \frac{c_1\, R\, \sigma_v^2}{G}, \qquad \sigma_v^2 \propto \frac{M_{\rm vir}}{R} \qquad (4)

where: $M_{\rm vir}$ — gravitating mass; $R$ — cloud radius; $\sigma_v$ — one-dimensional velocity dispersion (line width); $c_1$ — order-unity geometry factor ( $c_1=5$ for a uniform sphere). Read it as: for a bound cloud the observable line width is not free — it is locked to mass and size by gravity. Combining the thick-line brightness with the virial relation yields the pivotal scaling: the conversion factor depends only weakly on cloud properties, as the square root of density over brightness temperature.

X_{\rm CO} \simeq \frac{\text{const}\times\sqrt{n_{\mathrm{H_2}}}}{T_B} \qquad (5)

where: $n_{\mathrm{H_2}}$ — mean $\mathrm{H_2}$ volume density; $T_B\approx T_{\rm ex}$ — brightness temperature of the saturated line. Read it as: both $n_{\mathrm{H_2}}$ and $T_B$ vary only mildly among normal disk clouds, so $X_{\rm CO}$ is nearly constant — the entire empirical success of the method, in one line. For practical work the column-based $X_{\rm CO}$ is repackaged as a mass-to-light coefficient $\alpha_{\rm CO}$ that multiplies a CO luminosity straight into a mass; the second line bridges the two conventions.

M_{\rm mol} = \alpha_{\rm CO}\, L_{\rm CO},\qquad \alpha_{\rm CO} = \left(\frac{X_{\rm CO}}{6.3\times10^{19}}\right)\times 1.36 \approx 4.3\ M_\odot\,\mathrm{pc^{-2}\,(K\,km\,s^{-1})^{-1}} \qquad (6)

where: $M_{\rm mol}$ — total molecular mass, including the $\times1.36$ helium correction; $L_{\rm CO}$ — CO luminosity in $\mathrm{K\,km\,s^{-1}\,pc^2}$ ; $6.3\times10^{19}$ — helium-free constant $M_\odot\,\mathrm{pc^{-2}}$ per $\mathrm{cm^{-2}}$ of $\mathrm{H_2}$ . Read it as: point a telescope, integrate the line, multiply by $\approx4.3$ , and read off a mass in solar units — with $X_{\rm CO}=2\times10^{20}$ giving that $4.3$ .

Derivation AWhy molecular hydrogen is invisible in cold clouds while CO is not

The goal is to quantify, from the rigid-rotor spectrum alone, why cold molecular gas radiates copiously in CO and negligibly in $\mathrm{H_2}$ — and thereby to justify using an indirect tracer at all. We model each molecule as a rigid rotor, locate its lowest allowed emitting transition, and compare that energy to the thermal energy $k_B T$ of gas at $10\text{--}20\,\mathrm{K}$ : the whole argument is a competition between an excitation energy and a temperature, decided by a Boltzmann factor.

Step 1. Write the rotational energies as $E_J = hB\,J(J+1)$ with $B=\hbar/(4\pi I)$ and $I=\mu_{\rm r}r_0^2$ . The energy of the photon emitted in $J{\to}J{-}1$ is $E_{J,J-1}=2hBJ$ , so the ground line $J{=}1{\to}0$ costs $2hB$ . Why: rotational spacing is set by one number, $B$ , which is large for a light, short-bonded molecule and small for a heavy, long-bonded one.

Step 2. For CO, $B\approx57.6\,\mathrm{GHz}$ , so $E_{10}/k_B = 2hB/k_B \approx 5.5\,\mathrm{K}$ . At $T\sim15\,\mathrm{K}$ the Boltzmann factor $e^{-5.5/15}\approx0.7$ : the upper level is well populated and the line is readily emitted. Why: CO is heavy and long-bonded, so its first rung sits inside the thermal band the cloud already supplies.

Step 3. For $\mathrm{H_2}$ , $B\approx1.76\times10^{3}\,\mathrm{GHz}$ — thirty times larger — and the molecule is homonuclear. A homonuclear molecule has no permanent dipole, so every dipole rotational line, including $J{=}1{\to}0$ , is strictly forbidden; the lowest allowed line is the electric-quadrupole $J{=}2{\to}0$ at $E_{20}/k_B\approx510\,\mathrm{K}$ . Why: symmetry, not just energy, silences $\mathrm{H_2}$ — even if the $J{=}1$ level were populated it could not radiate a dipole photon.

Step 4. Compare with the thermal bath: for $\mathrm{H_2}$ 's $J{=}2$ level, $e^{-510/15}\approx1.7\times10^{-15}$ , utterly negligible, while CO's $J{=}1$ level sits at $\approx0.7$ . The two populations differ by fourteen orders of magnitude. Why: this single Boltzmann ratio is the entire case for CO as a tracer — cold gas simply cannot put $\mathrm{H_2}$ into an emitting state.

Step 5. Conclude that cold molecular gas emits in CO and is dark in $\mathrm{H_2}$ , so any census of the cold reservoir must be built on CO (or another proxy) rather than on the dominant species itself. Verified: $E_{10}/k_B=5.53\,\mathrm{K}$ and $E_{20}^{\mathrm{H_2}}/k_B\approx510\,\mathrm{K}$ recomputed from $E=hB\,J(J+1)$ ; consistent with Draine and Carroll & Ostlie.

Go deeper · Derivation B · 1 minThe virial calibration of the X-factor

Derivation BThe virial calibration of the X-factor

The goal is to show that for an ensemble of self-gravitating, optically thick clouds the conversion factor $X_{\rm CO}$ is nearly constant, depending only on the square root of the mean density over the brightness temperature — quantities that barely vary among normal clouds. We form the ratio of $\mathrm{H_2}$ column to integrated CO intensity for one cloud, fix the peak with thick-line saturation and the width with the virial theorem, then watch the radius cancel.

Step 1. The $\mathrm{H_2}$ column through a cloud of mean density $n_{\mathrm{H_2}}$ and radius $R$ is $N_{\mathrm{H_2}}\sim n_{\mathrm{H_2}}R$ , and its mass is $M\sim n_{\mathrm{H_2}}\,m\,R^3$ with $m$ the mean mass per molecule. Why: these are just geometry — a column is density times path, a mass is density times volume.

Step 2. Impose the virial relation $\sigma_v\sim(GM/R)^{1/2}$ . Substituting $M\propto n_{\mathrm{H_2}}R^3$ gives $\sigma_v\propto n_{\mathrm{H_2}}^{1/2}R$ . Why: gravity ties the observable line width to the cloud's density and size — the line width is not an independent quantity.

Step 3. For the optically thick line the peak saturates at $T_B\approx T_{\rm ex}$ , so the integrated intensity is $W_{\rm CO}=\int T_B\,\mathrm{d}v\approx T_B\,\sigma_v$ . Why: since the height is pinned, the area is set by the width — this is the step that replaces a radiative column measurement with a dynamical one.

Step 4. Form the ratio $X_{\rm CO}=N_{\mathrm{H_2}}/W_{\rm CO}\propto n_{\mathrm{H_2}}R/(T_B\,n_{\mathrm{H_2}}^{1/2}R)=n_{\mathrm{H_2}}^{1/2}/T_B$ . The radius cancels exactly; only $\sqrt{n_{\mathrm{H_2}}}$ and $T_B$ survive. Why: because $R$ drops out, clouds of wildly different sizes share one conversion factor — the origin of its near-universality.

Step 5. Evaluate the sensitivity. For normal disk clouds $n_{\mathrm{H_2}}\sim10^{2\text{--}3}\,\mathrm{cm^{-3}}$ and $T_B\sim10\,\mathrm{K}$ , and neither varies much, so $X_{\rm CO}$ is stable to tens of percent. Warm gas (large $T_B$ ) lowers it; hidden $\mathrm{H_2}$ from CO photodissociation raises the effective value. Why: the same formula that explains the constancy also predicts, quantitatively, the two directions in which it fails. *Verified: the scaling $X_{\rm CO}\propto\sqrt{n_{\mathrm{H_2}}}/T_B$ reproduces Solomon et al. (1987) and the Bolatto, Wolfire & Leroy (2013) virial derivation.*

Limiting cases

The Galactic value is a default, not a law of nature. It applies to normal, solar-metallicity, self-gravitating disk clouds; each row below is a way of leaving that regime.

Case	Result	Meaning
Milky Way disk (fiducial)	$X_{\rm CO}\approx2\times10^{20}\,\mathrm{cm^{-2}\,(K\,km\,s^{-1})^{-1}}$	Equivalently $\alpha_{\rm CO}\approx4.3$ with helium; the standard default.
Low metallicity / dwarfs	$X_{\rm CO}\uparrow$ (few to $>10$ )	Weak dust shielding lets far-UV photodissociate CO while $\mathrm{H_2}$ self-shields — large 'CO-dark' envelopes; roughly $X_{\rm CO}\propto Z^{-\beta}$ , $\beta\sim0.5\text{--}1$ .
Starburst / ULIRG nuclei	$X_{\rm CO}\downarrow\ \sim0.4\text{--}1\times10^{20}$	Warm ( $T\gtrsim30\,\mathrm{K}$ ), high- $\sigma_v$ , non-self-gravitating gas is over-luminous per unit mass; the Galactic value overestimates by a factor of a few.
Optically thick single cloud	$T_B\to T_{\rm ex}(1-e^{-\tau})\approx T_{\rm ex}$ ( $\tau\gg1$ )	Peak saturates; the column lives in the width, not the peak — why the calibration is virial.

Worked examples

1 · The excitation gap, from scratch (confidence builder)

Confirm the numbers behind the whole method: the excitation temperature of $\mathrm{CO}(1{\to}0)$ and of $\mathrm{H_2}$ 's lowest allowed line, and the resulting level populations at a cloud temperature $T=15\,\mathrm{K}$ .

Step 1. For CO, use the measured line frequency directly: $E_{10}/k_B=h\nu_{10}/k_B=(6.626\times10^{-34}\times115.27\times10^{9})/(1.381\times10^{-23})=\mathbf{5.53\,K}$ . This is comparable to the gas temperature, so the level is easy to reach.

Step 2. For $\mathrm{H_2}$ , the lowest allowed transition is the quadrupole $J{=}2{\to}0$ ; the tabulated value is $E_{20}/k_B\approx\mathbf{510\,K}$ , roughly $34$ times the gas temperature. (You can recover it from $E_{20}=6hB_{\mathrm{H_2}}$ with $B_{\mathrm{H_2}}\approx1.76\times10^{3}\,\mathrm{GHz}$ .)

Step 3. Compare populations with Boltzmann factors at $T=15\,\mathrm{K}$ : for CO, $e^{-5.53/15}=\mathbf{0.69}$ ; for $\mathrm{H_2}$ , $e^{-510/15}=e^{-34}=\mathbf{1.7\times10^{-15}}$ . The emitting level is well populated in CO and empty in $\mathrm{H_2}$ by fourteen orders of magnitude. CO radiates, H₂ does not — the factor is $\sim4\times10^{14}$ .

2 · Weighing a galaxy — and the starburst trap (fully worked)

A survey measures a CO luminosity $L_{\rm CO}=1.0\times10^{9}\,\mathrm{K\,km\,s^{-1}\,pc^2}$ from a galaxy. Find its molecular mass with the Galactic factor. Then suppose the emission comes from a warm ULIRG nucleus where $\alpha_{\rm CO}\approx0.8$ ; how large is the error from using the Galactic value?

Step 1. With the fiducial factor, $M_{\rm mol}=\alpha_{\rm CO}L_{\rm CO}=4.3\times(1.0\times10^{9})=\mathbf{4.3\times10^{9}\,M_\odot}$ — a few billion solar masses of molecular gas, typical of a gas-rich spiral.

Step 2. For the ULIRG nucleus the true factor is smaller because warm, turbulent, non-self-gravitating gas is over-luminous per unit mass: $M_{\rm mol}=0.8\times(1.0\times10^{9})=\mathbf{8.0\times10^{8}\,M_\odot}$ .

Step 3. The ratio is $4.3/0.8=\mathbf{5.4}$ . Using the Galactic value in a starburst overestimates the molecular mass — and lengthens the inferred gas-depletion time $t_{\rm dep}=M_{\rm mol}/\mathrm{SFR}$ — by roughly a factor of five. The choice of $\alpha_{\rm CO}$ is the dominant systematic, not the photometry.

3 · The virial mass of a resolved cloud

A nearby GMC is resolved with radius $R=20\,\mathrm{pc}$ and 1-D velocity dispersion $\sigma_v=3.0\,\mathrm{km\,s^{-1}}$ . Estimate its virial mass and mean density.

Step 1. With $M_{\rm vir}=5\,\sigma_v^2 R/G$ ( $c_1=5$ ) and $G=4.302\times10^{-3}\,\mathrm{pc\,M_\odot^{-1}\,(km\,s^{-1})^2}$ : $M_{\rm vir}=5\times9\times20/(4.302\times10^{-3})=\mathbf{2.1\times10^{5}\,M_\odot}$ — a typical GMC. Spreading that mass over the volume with mean mass $2.72\,m_{\rm H}$ per molecule gives $n_{\mathrm{H_2}}\approx\mathbf{90\,cm^{-3}}$ , close to the effective critical density — so the line is thermalized throughout. Comparing $M_{\rm vir}$ with $\alpha_{\rm CO}L_{\rm CO}$ for the same cloud is how the factor is locally recalibrated. $M_{\rm vir}\approx2.1\times10^{5}\,M_\odot$ , $n_{\mathrm{H_2}}\approx90\,\mathrm{cm^{-3}}$ .

Physics of the portrait

The X-factor is less a discovery than a hard-won consensus — a number pinned by three independent methods before anyone trusted it, from the first detection of interstellar CO to the realization that the conversion is fundamentally a statement about cloud dynamics.

Origins · 1 minthe CO-to-H2 conversion factor and how this was discovered

the CO-to-H2 conversion factor

the empirical ruler X_CO = N(H₂)/W_CO ≈ 2×10²⁰ · molecular-gas astronomy of galaxies

That a saturated line could weigh unseen gas is one of the stranger successes of observational astrophysics. When Wilson, Jefferts and Penzias detected the $2.6\,\mathrm{mm}$ CO line toward Orion in 1970, CO was merely a new interstellar molecule; within a decade it was the only practical map of the Galaxy's molecular content, because $\mathrm{H_2}$ stays dark. The conceptual leap was to treat the thick line not as a failed column measurement but as a cloud counter. Philip Solomon and collaborators, mapping Galactic clouds through the 1970s and culminating in the 1987 survey, showed that giant molecular clouds obey tight size–line-width and luminosity–mass relations — Larson's laws — that make them an approximately one-parameter family; that regularity is what turns $\int T_B\,\mathrm{d}v$ into a mass, with the virial theorem supplying the rest. Independent calibrations followed a different logic: because cosmic-ray protons strike every nucleon regardless of chemical form, the diffuse $\gamma$ -ray emissivity measures total gas, and comparing $\gamma$ -ray maps with $\mathrm{H\,I}$ and CO isolates the molecular column with no radiative-transfer assumption; dust continuum gives a third handle. The outcome — consolidated by Bolatto, Wolfire and Leroy in 2013 — is that three unrelated methods converge on $X_{\rm CO}\approx2\times10^{20}$ for the Milky Way, and all three agree it climbs in metal-poor dwarfs and falls in starbursts. The X-factor is thus not a constant of nature but a calibrated regularity, trustworthy exactly to the extent that clouds stay discrete, cold, and virialized.

History

1970

— Wilson, Jefferts and Penzias detect the 2.6 mm CO line toward Orion

1975

— Scoville, Solomon and collaborators map the Galactic molecular ring in CO

1983

— diffuse gamma-ray emission from cosmic rays gives a tracer-independent gas column

1987

— Solomon, Rivolo, Barrett and Yahil calibrate the virial X-factor from GMC surveys

2001

— Dame, Hartmann and Thaddeus complete the CfA CO survey of the Milky Way

2013

— Bolatto, Wolfire and Leroy consolidate X_CO and its variations in an ARA and A review

Going further

Texts. Carroll & Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 12 (the interstellar medium and star formation) · Draine, Physics of the Interstellar and Intergalactic Medium, Ch. 5, 19 · Binney & Tremaine, Galactic Dynamics, 2nd ed. (virial estimates) · Mo, van den Bosch & White, Galaxy Formation and Evolution, Ch. 9 (cold gas and star formation) · Bolatto, Wolfire & Leroy, ARA&A 51, 207 (2013), 'The CO-to-H₂ Conversion Factor' · Solomon, Rivolo, Barrett & Yahil, ApJ 319, 730 (1987).

Provenance. The excitation energies are recomputable from $E_J=hB\,J(J+1)$ : $B_{\rm CO}\approx57.6\,\mathrm{GHz}$ gives $E_{10}/k_B=5.53\,\mathrm{K}$ , and $B_{\mathrm{H_2}}\approx1.76\times10^{3}\,\mathrm{GHz}$ gives $E_{20}/k_B\approx510\,\mathrm{K}$ (Draine; Carroll & Ostlie). The fiducial $X_{\rm CO}=2\times10^{20}\,\mathrm{cm^{-2}\,(K\,km\,s^{-1})^{-1}}$ and $\alpha_{\rm CO}\approx4.3\,M_\odot\,\mathrm{pc^{-2}\,(K\,km\,s^{-1})^{-1}}$ are from Bolatto, Wolfire & Leroy (2013); a direct cgs evaluation of eq. (6) gives $\alpha_{\rm CO}=4.36$ , confirming the pairing (note: the He-free bridge constant is $6.3\times10^{19}$ , corrected here from a value that appeared in the source draft). The virial scaling $X_{\rm CO}\propto\sqrt{n}/T_B$ follows Solomon et al. (1987) and the B&T virial treatment. A verifier should re-check the $E_J$ numbers, the $X_{\rm CO}\leftrightarrow\alpha_{\rm CO}$ conversion including the $1.36$ helium factor, the virial mass $M_{\rm vir}=5\sigma_v^2 R/G=2.1\times10^{5}\,M_\odot$ , and the ULIRG factor-of- $5.4$ example.

Prerequisites. Molecular Clouds and Giant Molecular Clouds · Phases of the Interstellar Medium · Virial Theorem for Molecular Clouds · Interstellar Extinction and Reddening. Leads to. Thermal Dust Emission and Infrared Radiation · The Jeans Instability and Jeans Mass · Turbulence in the Interstellar Medium · The 21 cm Line of Neutral Hydrogen.

Ask the tutor: Ask me to walk through why a saturated CO line still weighs a cloud, or to work out how much your molecular mass changes if you swap the Galactic $\alpha_{\rm CO}$ for a starburst or low-metallicity value.

If CO is only a trace species, why is it more useful than H₂ itself?

Abundance is not the whole story — emissivity is. CO is only about one part in $10^{4}$ of the gas by number, but it can radiate at cloud temperatures whereas the dominant $\mathrm{H_2}$ cannot: being homonuclear, $\mathrm{H_2}$ has no permanent dipole and its lowest allowed line sits $\sim510\,\mathrm{K}$ up, empty by fourteen orders of magnitude at $15\,\mathrm{K}$ . CO's lowest line costs only $5.5\,\mathrm{K}$ and is switched on by collisions with $\mathrm{H_2}$ itself. So we watch the rare molecule that talks to infer the common one that stays silent.

How can a saturated, optically thick line carry any information about mass?

For a thick line the peak saturates near the excitation temperature and stops responding to column — extra gas along the sightline does not brighten the peak, it adds gas at more velocities and widens the line. The integrated intensity $W_{\rm CO}=\int T_B\,\mathrm{d}v\approx T_B\,\sigma_v$ therefore grows with the width, and for a bound cloud the width is fixed by the mass through the virial theorem. So the area under the profile, not its height, encodes the mass — which is why the calibration is dynamical, not radiative.

Why does the X-factor go up at low metallicity when H₂ still forms?

It is a shielding mismatch. Both CO and $\mathrm{H_2}$ are destroyed by far-ultraviolet photons, but $\mathrm{H_2}$ self-shields very efficiently — a thin surface layer protects the interior — while CO relies mainly on dust to block the ultraviolet. In a metal-poor, dust-poor cloud that dust shield is thin, so CO is photodissociated well into the cloud while $\mathrm{H_2}$ survives as a large 'CO-dark' envelope. Less CO per unit $\mathrm{H_2}$ means you must adopt a larger $X_{\rm CO}$ to recover the true mass, sometimes by more than an order of magnitude at a tenth solar metallicity.

Why is the starburst value of α_CO lower than the Galactic one?

Because starburst and ULIRG nuclei violate the virial picture behind the Galactic value. The gas is warm ( $T\gtrsim30\,\mathrm{K}$ ), highly turbulent, and often not in discrete self-gravitating clouds but smeared into a hot, high-dispersion medium filling the whole nuclear potential. Such gas radiates more CO per unit mass, so the mass-to-light factor $\alpha_{\rm CO}$ drops — typically to $\sim0.8$ rather than $4.3$ . Applying the Galactic value there overestimates the molecular mass by a factor of a few, which is why starburst studies adopt a separate, lower calibration.

Why is the diffuse γ-ray method considered such a clean calibration?

Because it bypasses chemistry and radiative transfer entirely. Cosmic-ray protons strike every interstellar nucleon — atomic, molecular, or 'CO-dark' — producing $\gamma$ -rays in proportion to the total gas column, with no dependence on CO's excitation, optical depth, or abundance. Mapping the $\gamma$ -ray emissivity, subtracting the atomic column from $\mathrm{H\,I}$ 21 cm, and comparing the molecular remainder with the CO map isolates $X_{\rm CO}$ with none of the assumptions the CO line requires. That independence is why the $\gamma$ -ray, dust-continuum, and virial values all agreeing on $\approx2\times10^{20}$ is so persuasive.

Does the critical density mean CO only lights up in the densest cores?

Not for the $J{=}1{\to}0$ line. The bare critical density — spontaneous decay rate over collisional de-excitation rate — is a couple of thousand per cubic centimetre, above a cloud's mean density. But the thick line traps its own photons: an emitted photon is usually re-absorbed nearby, so a level de-excites collisionally after many internal scatterings, lowering the effective critical density to a few hundred per cubic centimetre. That is comparable to the mean GMC density, so $\mathrm{CO}(1{\to}0)$ is bright throughout the bulk of a cloud, not merely in its densest clumps.