21-cm Hydrogen Line

A single hydrogen atom, left alone in its ground state with electron and proton spins aligned, will wait on average eleven million years before it flips one spin and emits a photon. That photon has a wavelength of $21$ centimetres and a frequency known to twelve digits. It is the most useful "forbidden" transition in physics: the hyperfine spin-flip of neutral hydrogen, and it is how astronomers weigh and map the neutral hydrogen of the entire universe. This article takes the hyperfine splitting computed in Hyperfine Structure — the singlet–triplet gap $\Delta E\approx5.9\times10^{-6}$ eV — and asks the astrophysical questions the atomic-physics chapter did not. Why is the transition so slow (it is magnetic-dipole, $M1$ , not the fast electric-dipole kind)? If a single atom emits once per eleven million years, how do we detect the line at all (a galactic sightline holds $\sim10^{21}$ atoms per square centimetre — immensity beats forbiddenness)? Why does the line trace neutral-hydrogen mass rather than temperature (the levels are populated $3{:}1$ by their statistical weights, since $\Delta E/k\approx0.07$ K is colder than anything in space)? And how does a razor-thin line become a Doppler map of a spinning galaxy? The frequency/wavelength triangle from the hyperfine gap, the $M1$ lifetime to order of magnitude, the $3{:}1$ weighting, and the column-density detectability argument are machine-verified (18 checks).

Key takeaways

The line is the $F=1\to F=0$ hyperfine spin-flip of the hydrogen $1s$ ground state, energy $\Delta E$ from Hyperfine Structure.
Precise values: $\nu=1420.405751768$ MHz, $\lambda=21.106$ cm — twelve significant figures, a metrological triumph.
It is a magnetic-dipole ( $M1$ ) transition, not electric-dipole: the rate $A_{10}\approx2.85\times10^{-15}$ s $^{-1}$ (order-of-magnitude reproduced) is $\sim23$ orders below a typical optical line.
Single-atom lifetime $\tau=1/A_{10}\approx1.1\times10^7$ yr — eleven million years — the "most forbidden useful line."
We see it anyway because there is so much hydrogen: a sightline with $\sim10^{21}$ atoms/cm $^2$ emits $\sim10^{6}$ – $10^{7}$ photons/s/cm $^2$ despite the per-atom rate.
The upper and lower states are populated $3{:}1$ by statistical weight ( $\Delta E/k\approx0.068$ K $\ll$ any astrophysical temperature), so the line strength tracks the amount of neutral hydrogen, not its temperature.
Key idea: an eleven-million-year atomic transition becomes astronomy's workhorse because column densities of $10^{21}$ turn a hopeless per-atom rate into a bright, mass-tracing, Doppler-mappable line.

The core insight

The $21$ cm line is the hyperfine spin-flip $F=1\to F=0$ of ground-state hydrogen. Its per-atom rate is minuscule (magnetic-dipole, $\tau\approx11$ Myr), but the sheer number of hydrogen atoms along any galactic sightline makes it bright. Because $\Delta E/k\approx0.07$ K is far below any astrophysical temperature, the levels are populated $3{:}1$ by statistical weight, so the line's strength measures the column density of neutral hydrogen — which is why it maps the mass and motion of galaxies.

Where this lives today

The $21$ cm line is one of radio astronomy's foundational tools. Large surveys map neutral hydrogen across the Milky Way and out to distant galaxies, tracing spiral structure, gas dynamics, and the flat rotation curves that remain a pillar of the dark-matter case. Because the line is a clean mass tracer, HI surveys measure the gas reservoirs that feed star formation across cosmic time. The frontier is cosmological. Before the first stars, the universe was mostly neutral hydrogen, and its $21$ cm signal — redshifted today to metre wavelengths — encodes the epoch of reionization and the "cosmic dawn." Experiments like EDGES and the upcoming Square Kilometre Array hunt for this faint global and fluctuating signal, aiming to map the universe's neutral-hydrogen history the way optical surveys map galaxies. The same atomic transition that Ewen and Purcell caught on a windowsill is now a probe of the first billion years. The full transition-rate formalism behind the $M1$ estimate — and the selection rules that make it $M1$ rather than $E1$ — belong to the time-dependent perturbation theory of a later chapter.

It is the hyperfine spin-flip: electron and proton spins go from parallel (triplet $F=1$ ) to antiparallel (singlet $F=0$ ) within the $1s$ ground state, emitting a $\Delta E\approx5.9\,\mu$ eV photon.

Both states are $1s$ with the same spatial wavefunction and parity; only spin flips, so the fast electric-dipole channel is unavailable. The $M1$ rate scales as $\omega^3|\mu|^2$ and, at $1420$ MHz, is $\sim10^{23}$ times below an optical line.

Detectability is rate times column density. With $N_H\sim10^{21}$ atoms/cm $^2$ , the sightline emits $\sim10^{6}$ – $10^{7}$ photons/s/cm $^2$ — immensity beats forbiddenness.

The tiny energy scale means $e^{-\Delta E/kT}\approx1$ for any real $T$ , pinning the population ratio at $3{:}1$ (statistical weights). The line strength then depends only on the number of atoms — the column density.

Because the line is intrinsically sharp, galactic rotation Doppler-shifts it into a velocity-resolved profile — the basis of HI rotation curves and spiral-structure maps.

The physical picture

Everything quantitative about the line comes from one number: the hyperfine gap $\Delta E$ that Hyperfine Structure computed from the Fermi contact interaction. That article established the two levels — triplet $F=1$ (electron and proton spins parallel) sitting $\Delta E/4$ above centre, singlet $F=0$ (antiparallel) sitting $3\Delta E/4$ below — separated by $\Delta E\approx5.9\times10^{-6}$ eV. This article does not re-derive that splitting; it uses it. Three facts turn that gap into astronomy. First, it is astonishingly precise. Because the transition is between two sharp, long-lived levels of the simplest atom, its frequency is one of the best-known numbers in all of science: $1420.405751768$ MHz. Second, it is astonishingly slow. A photon of $21$ cm carries essentially no orbital angular momentum change and no parity change — it is a magnetic-dipole transition, and magnetic-dipole rates are tiny compared to the electric-dipole transitions that make ordinary spectral lines. The single-atom lifetime is eleven million years. Third — and this is the resolution of the apparent paradox — the universe is made mostly of hydrogen, and there is a staggering amount of it. A line of sight across a galaxy passes through something like $10^{21}$ hydrogen atoms per square centimetre of beam. Even if each atom emits only once per eleven million years, that many atoms produce a steady, detectable glow. Forbiddenness loses to immensity.

Figure

Why we see an 11-million-year line: a hopeless per-atom rate times an enormous column density gives a bright line

Figure

The spin-flip and its Doppler map: one sharp line becomes a velocity profile across a spinning galaxy

The final gift is the population ratio. The two levels are separated by only $\Delta E/k\approx0.068$ K — colder than the coldest interstellar cloud. So in any real environment the Boltzmann factor $e^{-\Delta E/kT}$ is essentially $1$ , and the two levels are populated purely by their statistical weights $2F+1$ : three for the triplet, one for the singlet, a $3{:}1$ ratio. Because that ratio is fixed regardless of temperature, the strength of the line depends only on how many hydrogen atoms are present — which is why $21$ cm astronomy measures hydrogen mass and maps its distribution, not its temperature.

The frequency, to twelve digits

The transition energy is the hyperfine gap of Hyperfine Structure. Converting an energy to a frequency and wavelength is the Planck–de Broglie bookkeeping, but the precision is what matters here:

\nu=\frac{\Delta E}{h},\qquad \lambda=\frac{c}{\nu} \qquad (1)

where: $\Delta E\approx5.9\times10^{-6}$ eV is the singlet–triplet gap (owned and computed by Hyperfine Structure), $h$ is Planck's constant, and $c$ the speed of light. The simple contact formula of the previous article gives $\nu\approx1421$ MHz and $\lambda\approx21$ cm — right to a few parts in $10^3$ . The measured line is

\nu=1420.405751768\ \text{MHz},\qquad \lambda=21.106\ \text{cm} \qquad (2)

where: the twelve significant figures come from decades of maser and atomic-beam metrology. The gap between the simple $1421$ MHz estimate and this value is reduced-mass, bound-state QED, and finite-proton-size corrections — the same corrections the hyperfine article flagged. What the elementary theory gets exactly right is the physics; the last digits are a precision-QED triumph, and they are what make the line a Doppler ruler: a velocity of $1$ km/s shifts a $1420$ MHz line by about $5$ kHz, easily resolved. The corresponding temperature scale $\Delta E/k=0.068$ K is the number that decouples the line from thermometry: no astrophysical environment is that cold, so the level populations never depart from their $3{:}1$ statistical-weight ratio.

Why the transition is so slow

An ordinary bright spectral line — the hydrogen Lyman- $\alpha$ , a sodium D line — is an electric-dipole ( $E1$ ) transition: the electron's charge distribution oscillates, radiating efficiently, with rates around $10^{8}$ s $^{-1}$ (nanosecond lifetimes). The $21$ cm transition cannot do this. Both hyperfine levels are $1s$ states with the same spatial wavefunction and the same parity; only the spin orientation changes. An oscillating charge distribution is not available. What flips is a magnetic moment — a magnetic-dipole ( $M1$ ) transition, radiating far more weakly.

Derivation AThe Magnetic-Dipole Rate and the Eleven-Million-Year Lifetime

The goal is an order-of-magnitude estimate of the spontaneous-emission rate $A_{10}$ for the $21$ cm line and the lifetime $\tau=1/A_{10}$ it implies, showing why the transition is so slow: it is magnetic-dipole and low-frequency, a double suppression. The strategy is the standard radiated-power scaling for a dipole, applied with a magnetic moment $\sim\mu_B$ instead of an electric one and the tiny hyperfine frequency. This is an estimate by design — the exact coefficient is a QED calculation — but it lands within a factor of a few of the measured value, which is the honest goal.

Step 1. A magnetic-dipole transition of angular frequency $\omega$ and moment $|m|$ has a spontaneous rate scaling as

A_{M1}\sim\frac{\mu_0\,\omega^3\,|m|^2}{3\pi\hbar c^3},

the magnetic analogue of the electric-dipole formula, with the near-field magnetic coupling $\mu_0$ replacing the electric $1/\varepsilon_0$ .

Step 2. Put in the transition's numbers: $\omega=2\pi\nu$ with $\nu=1420$ MHz, and $|m|\sim\mu_B=9.27\times10^{-24}$ J/T (a Bohr magneton — the electron spin does the flipping). Evaluating gives $A_{10}\sim3\times10^{-15}$ s $^{-1}$ , within a factor of a few of the accepted $2.85\times10^{-15}$ s $^{-1}$ .

Step 3. Read off the two suppressions against a typical optical $E1$ line ( $A_{E1}\sim10^{8}$ s $^{-1}$ ). The $\omega^3$ factor: the $21$ cm frequency is $\sim10^{6}$ times below an optical frequency, costing $\sim10^{18}$ in rate. The $M1$ -vs- $E1$ nature costs a further $\sim(v/c)^2\sim\alpha^2$ . Together $A_{10}/A_{E1}\sim10^{-23}$ — the line is roughly twenty-three orders of magnitude slower than an ordinary transition.

Step 4. The single-atom lifetime is the reciprocal: $\tau=1/A_{10}=1/(2.85\times10^{-15}\ \text{s}^{-1})=3.5\times10^{14}$ s $=1.1\times10^{7}$ yr — eleven million years. This is the lifetime grammar $\tau=\hbar/\Gamma$ of Conservation of Normalization applied to a genuinely metastable state. Why the estimate is honest, not fitted: the coefficient in Step 1 is order-unity-ambiguous, so we claim only the order of magnitude and the two scalings ( $\omega^3$ and $M1$ -suppression). The exact $2.85\times10^{-15}$ s $^{-1}$ is a full QED matrix-element calculation; the point here is to see why eleven million years, not to reproduce the last digit. *Verified: the $M1$ estimate $A_{est}=2.87\times10^{-15}$ s $^{-1}$ (within a few percent of $2.85\times10^{-15}$ , the closeness partly luck given the order-unity coefficient); $\tau=1.11\times10^{7}$ yr; the $\sim10^{-23}$ suppression vs an optical $E1$ line.*

Why we see it anyway: the column-density argument

An eleven-million-year lifetime sounds like a death sentence for detectability, but the arithmetic runs the other way once you count atoms. The rate of photons emitted per unit column is the number of upper-state atoms in the column times the per-atom rate:

\text{photon rate per cm}^2 \sim N_{\text{upper}}\times A_{10},\qquad N_{\text{upper}}\approx\tfrac{3}{4}N_H \qquad (3)

where: $N_H$ is the total hydrogen column density (atoms per cm $^2$ along the sightline), and $\tfrac34$ is the triplet fraction from the $3{:}1$ weighting. For a typical galactic sightline, $N_H\sim10^{21}$ atoms/cm $^2$ (one atom per cm $^3$ over a kiloparsec). Then the emission rate is $\sim\tfrac34\times10^{21}\times2.85\times10^{-15}\approx10^{6}$ – $10^{7}$ photons/s/cm $^2$ — bright and steady. The forbiddenness of the single-atom transition is completely overwhelmed by the number of atoms.

Go deeper · Derivation B · 1 minThe Spin Temperature and the Three-to-One Population

Derivation BThe Spin Temperature and the Three-to-One Population

The goal is to show that the two hyperfine levels are populated in the fixed ratio $3{:}1$ set by their statistical weights, independent of temperature — the reason the $21$ cm line measures the amount of neutral hydrogen rather than its temperature. The strategy is one Boltzmann factor with the tiny hyperfine energy scale, and a comparison of $\Delta E/k$ to any real astrophysical temperature.

Step 1. In thermal equilibrium at temperature $T$ (the "spin temperature"), the population ratio of the two hyperfine levels is

\frac{n_1}{n_0}=\frac{g_1}{g_0}\,e^{-\Delta E/kT},

with statistical weights $g_1=2F+1=3$ (triplet) and $g_0=1$ (singlet).

Step 2. Evaluate the energy scale: $\Delta E/k=(5.87\times10^{-6}\ \text{eV})/(8.62\times10^{-5}\ \text{eV/K})=0.068$ K. This is far colder than any astrophysical environment — the coldest molecular clouds are $\sim10$ K, the cosmic microwave background is $2.7$ K, warm neutral gas is thousands of K.

Step 3. For any $T\gg0.068$ K the Boltzmann factor $e^{-\Delta E/kT}\to1$ , so

\frac{n_1}{n_0}\to\frac{g_1}{g_0}=3.

At $T=100$ K the ratio is $2.998$ ; at $T=10$ K it is $2.98$ — never departing from $3$ by more than a percent under realistic conditions.

Step 4. Because the ratio is pinned at $3{:}1$ regardless of temperature, the number of atoms available to emit is a fixed fraction of the total, so the line strength is proportional to the total hydrogen column density $N_H$ . The line measures how much neutral hydrogen lies along the sightline — its mass — not how hot it is. Why this is the whole reason 21 cm maps galaxies: a temperature-independent line strength is a clean mass tracer. Combined with the Doppler shift (below), it gives both the amount and the velocity of neutral hydrogen at every point — which is how rotation curves, and the dark-matter problem, were charted. *Verified: $\Delta E/k=0.0682$ K; the population ratio $=2.98$ at $10$ K, $3.00$ at $100$ K and above; the $3{:}1$ weighting from $2F+1$ .*

Finally, the line's narrowness makes it a velocity ruler. Because it is intrinsically sharp, any observed shift is a Doppler shift, and the differential rotation of a spiral galaxy smears the line across a range of frequencies. Decoding that smear gives a velocity-resolved map of neutral hydrogen — the technique that traced spiral-arm structure and, when the outer rotation curves refused to fall off, supplied one of the earliest strong cases for dark matter. The dynamics themselves belong to the astrophysics tracks; here the point is only that a razor-thin, mass-tracing line is exactly what such maps require.

Limiting cases

case	result	meaning
single isolated atom	emits once per $\tau\approx11$ Myr	the per-atom rate is hopeless in a lab
galactic sightline, $N_H\sim10^{21}$ /cm $^2$	$\sim10^{6}$ – $10^{7}$ photons/s/cm $^2$	immensity beats forbiddenness — bright line
any astrophysical $T\gg0.068$ K	$n_1{:}n_0=3{:}1$ fixed	line strength $\propto$ hydrogen mass, not $T$
narrow intrinsic line + galactic rotation	Doppler-broadened velocity map	HI rotation curves, spiral structure, dark matter
ionized hydrogen (HII)	no $21$ cm line	the electron is gone; $21$ cm traces neutral H only

Worked examples

Example 1 — the frequency/wavelength/temperature triangle. From the hyperfine gap of Hyperfine Structure, compute the line's frequency, wavelength, and the temperature scale that decouples it from thermometry.

Step 1. Take the measured gap $\Delta E=h\nu$ with the standard line $\nu=1420.405751768$ MHz. Cross-check against the contact estimate: the previous article's $\Delta E\approx5.88\times10^{-6}$ eV gives $\nu=\Delta E/h=(5.88\times10^{-6})/(4.136\times10^{-15})=\mathbf{1.42\times10^{9}\ Hz}$ (verified) — the estimate lands on the measured line to a few parts in $10^3$ .

Step 2. Wavelength: $\lambda=c/\nu=(3.00\times10^{8}\ \text{m/s})/(1.420\times10^{9}\ \text{Hz})=\mathbf{0.2110\ m}=21.106$ cm (verified). This is the " $21$ cm" the line is named for.

Step 3. Temperature scale: $\Delta E/k=(5.874\times10^{-6}\ \text{eV})/(8.617\times10^{-5}\ \text{eV/K})=\mathbf{0.0682\ K}$ (verified). Because this is far below any astrophysical temperature, the Boltzmann factor is $\approx1$ and the levels sit at the $3{:}1$ statistical-weight ratio — the line measures hydrogen amount, not temperature.

Example 2 — detectability: why an 11-Myr line is bright. Estimate the $21$ cm photon emission rate from a galactic sightline of $1$ hydrogen atom per cm $^3$ over $1$ kpc.

Step 1. Column density: $N_H=n_H\times L=(1\ \text{cm}^{-3})\times(1\ \text{kpc})=(1)\times(3.09\times10^{21}\ \text{cm})=\mathbf{3.1\times10^{21}}$ atoms/cm $^2$ (verified).

Step 2. Upper-state atoms: the triplet holds $\tfrac34$ of the population, so $N_{\text{upper}}=0.75\times3.1\times10^{21}=2.3\times10^{21}$ atoms/cm $^2$ .

Step 3. Photon rate: multiply by the per-atom rate $A_{10}=2.85\times10^{-15}$ s $^{-1}$ : rate $=2.3\times10^{21}\times2.85\times10^{-15}=\mathbf{6.6\times10^{6}}$ photons/s/cm $^2$ (verified). A per-atom transition that fires just once in ~11 million years produces millions of photons per second per square centimetre of column — because there are so many atoms. This is why Ewen and Purcell needed only a horn antenna on a Harvard windowsill.

The observers

Origins · 1 minRadio Astronomy and how this was discovered

Radio Astronomy

1944–1951 · Leiden, Harvard, Sydney

Radio astronomy earned its place as a precision science through this one spectral line. In German-occupied Leiden in 1944, Hendrik van de Hulst worked out that the hyperfine flip of interstellar hydrogen should be observable — a prediction made with no working radio telescope on the continent. At Harvard, Edward Mills Purcell — the rare physicist equally at home with a soldering iron and a Hamiltonian, co-discoverer of NMR in bulk matter in 1946 (the 1952 Nobel, shared with Felix Bloch's independent work) — set his graduate student Harold Ewen to build a horn antenna and a "frequency-switching" receiver that could pull a faint, steady line out of the receiver's own noise. In March 1951 they detected the $21$ cm line of interstellar hydrogen, and Purcell held publication so that the Dutch and Australian groups confirming it could appear alongside — three continents announcing one line. The detection opened astronomy's window onto the neutral universe: for the first time the sky could be mapped in a single quantum transition whose physics was known exactly.

History

1944

— van de Hulst predicts the 21 cm line in German-occupied Leiden, arguing it should be observable

1949

— Shklovsky independently analyses the line and argues it is detectable

1951

— Ewen and Purcell detect the 21 cm line at Harvard; Dutch and Australian groups confirm within weeks

1954

— the first 21 cm surveys begin mapping the spiral structure of the Milky Way

1970

s — 21 cm rotation curves of galaxies stay flat to large radii, strengthening the case for dark matter

Going further

Griffiths and Schroeter, Introduction to Quantum Mechanics, 3rd ed., §7.5 — the hyperfine origin of the line.
Rybicki and Lightman, Radiative Processes in Astrophysics — the $M1$ rate and radiative transfer for the line.
Draine, Physics of the Interstellar and Intergalactic Medium — the $21$ cm line as an ISM diagnostic.
Pritchard and Loeb, Rep. Prog. Phys. 75, 086901 (2012) — $21$ cm cosmology and the cosmic dawn.

Provenance note: Verifier should recompute: (1) the spin dot-product eigenvalues $+\tfrac14\hbar^2$ (triplet) and $-\tfrac34\hbar^2$ (singlet) from $\mathbf{S}_e\cdot\mathbf{S}_p=\tfrac12[F(F+1)-3/2]\hbar^2$ ; (2) the contact splitting $\Delta E=\tfrac43 g_p (m_e/m_p)\alpha^4 m_e c^2\approx 5.9\times10^{-6}\,\mathrm{eV}$ using $g_p=5.586$ , $\alpha=1/137.036$ , $m_ec^2=0.511\,\mathrm{MeV}$ (Griffiths §7.5); (3) frequency $\nu=\Delta E/h\approx 1.42\,\mathrm{GHz}$ and $\lambda=c/\nu\approx 21.1\,\mathrm{cm}$ . Standard values: $\nu=1420.405751768\,\mathrm{MHz}$ , $A_{10}=2.85\times10^{-15}\,\mathrm{s}^{-1}$ , $\tau\approx 1.1\times10^{7}\,\mathrm{yr}$ (Bethe & Salpeter; NIST). Cross-check derivation against Griffiths §7.5, Cohen-Tannoudji vol. II ch. XII. — The items this article owns (the frequency/wavelength/temperature triangle from the hyperfine gap, the $M1$ rate and lifetime to order of magnitude, the $3{:}1$ spin-temperature population, and the column-density detectability argument) are executed in its verify script (18 checks); the hyperfine gap $\Delta E$ itself is consumed by name from Hyperfine Structure and never re-derived here.

Builds on: Hyperfine Structure · Fine Structure of Hydrogen · Degenerate Perturbation Theory · Good Quantum Numbers and Symmetry-Adapted Bases Leads to: Zeeman Effect · Lamb Shift

Where does the 5.9 μeV energy actually come from — is it derived here?

No — it is imported. The singlet–triplet gap $\Delta E\approx5.9\times10^{-6}$ eV is derived in Hyperfine Structure from the Fermi contact interaction ( $\propto|\psi(0)|^2$ ). This article consumes that number and asks the astrophysical questions: the precise frequency, the lifetime, why we see it, and what it measures. Re-deriving the splitting here would duplicate the previous article; the chapter's rule is that each result has one owner.

Why is it magnetic-dipole and not electric-dipole?

Because nothing about the electron's position changes. Both hyperfine levels are $1s$ states with identical spatial wavefunctions and the same (even) parity; only the relative spin orientation flips. An electric-dipole transition needs an oscillating charge distribution and a parity change — neither is available. What does change is a magnetic moment, so the transition proceeds by the much weaker magnetic-dipole channel, roughly $\alpha^2$ times slower even before the low-frequency $\omega^3$ suppression.

Is the 2.85 × 10⁻¹⁵ s⁻¹ rate something I can derive, or must I look it up?

You can get the order of magnitude yourself — the $M1$ scaling $A\sim\mu_0\omega^3\mu_B^2/(3\pi\hbar c^3)$ lands within a factor of a few (the article's estimate gives $\approx2.9\times10^{-15}$ s $^{-1}$ , close partly by luck given the order-unity coefficient). The exact value is a full quantum-electrodynamic matrix-element calculation. The article deliberately claims only the order of magnitude and the two physical scalings, not the last digit — that is the honest level for a scaling argument.

If the line traces mass, how do radio astronomers get velocities and temperatures too?

Velocities come from the Doppler shift: the line is intrinsically narrow, so its observed frequency shift maps directly to line-of-sight velocity, and galactic rotation spreads it into a velocity profile. The gas temperature (more precisely the "spin temperature") enters through subtler effects — the line can appear in emission or absorption against a background source, and the ratio encodes the excitation temperature — but the leading measurement, the integrated line strength, is a clean column-density (mass) tracer because the $3{:}1$ population is temperature-independent.

Does ionized hydrogen produce a 21 cm line?

No. The $21$ cm line is a hyperfine transition of the neutral hydrogen atom — it needs the electron bound to the proton so their spins can couple. In ionized regions (HII, around hot stars) the electron is free and there is no atom, hence no line. This is a feature: $21$ cm surveys map specifically the neutral hydrogen reservoir, complementary to the optical and radio tracers of ionized gas.

Why did van de Hulst's 1944 prediction matter if the line is such a natural consequence of hyperfine structure?

The hyperfine splitting was known, but whether the line could be observed was not obvious — an eleven-million-year lifetime looks discouraging. Van de Hulst's contribution, made in occupied Leiden during a wartime seminar, was the detectability argument: the enormous column densities of interstellar hydrogen would make the line bright despite the per-atom rate. That insight — the same column-density argument in this article — is what motivated Ewen and Purcell to build the receiver. Prediction and detection were seven years apart.