---
title: "The Boltzmann equation and freeze-out"
description: "Derive the equation for particles per unit entropy, solve it near freeze-out and find the cross-section that leaves the observed dark matter density."
canonical_url: "https://sandboxphysics.com/stories/dark-matter-relics/the-boltzmann-equation-and-freeze-out/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/dark-matter-relics/the-boltzmann-equation-and-freeze-out/"
last_modified: "2026-11-14"
---

[How dark matter was left over](https://sandboxphysics.com/stories/dark-matter-relics/)

# The Boltzmann equation and freeze-out

Sandbox Physics · 2026-11-14

The number of dark particles per unit entropy obeys one equation: it falls at a rate set by annihilation minus inverse production, and nothing else changes it. Solved for a 100 GeV particle, it leaves $Y_\infty=4.37\times10^{-12}$ when the low-temperature cross-section is $2.136\times10^{-26}\,\mathrm{cm^3\,s^{-1}}$, which is today's measured density $\Omega h^2=0.120$. A two-line approximation gets within ten per cent of that, and it shows why the leftover is inversely proportional to the cross-section.

This chapter follows [Equilibrium, expansion and the reaction race](https://sandboxphysics.com/stories/dark-matter-relics/equilibrium-expansion-and-the-reaction-race/) and uses its units and symbols. You need separable differential equations. The aims are to derive the equation for $Y$, solve it in its two regimes, convert the result into a density today and see why a computer needs an implicit method.

## From particle number to particles per entropy

Count dark particles in a box that grows with the universe. Annihilation removes pairs at the rate $\langle\sigma v\rangle n^2$ per unit volume. Inverse production, bath particles colliding to make a pair, adds them at $\langle\sigma v\rangle n_{\mathrm{eq}}^2$, because detailed balance requires the two to cancel in equilibrium. Dilution by the growing volume removes $3Hn$. So

$$
\frac{\mathrm dn}{\mathrm dt}+3Hn=-\langle\sigma v\rangle\left(n^2-n_{\mathrm{eq}}^2\right).
$$

The Particle Data Group review prints the dilution term with a minus sign, $\mathrm dn/\mathrm dt-3Hn$. [Particle Data Group, 2025, eq. 27.2](https://pdg.lbl.gov/2025/reviews/rpp2025-rev-dark-matter.pdf) That sign would make the density grow without reactions. The plus sign used here is the one in D'Eramo, Fernandez and Profumo's eq. (15) and in Steigman and colleagues' eq. (1), and it follows from $na^3$ being constant when nothing reacts. [D'Eramo, Fernandez and Profumo, 2017, eq. 15](https://arxiv.org/abs/1703.04793v1)

Now divide by $s$. Because $sa^3$ is constant, $\mathrm d(n/s)/\mathrm dt=(\dot n+3Hn)/s$. With $T\propto1/a$ the clock $x=m/T$ obeys $\mathrm dx/\mathrm dt=Hx$, so

$$
\frac{\mathrm dY}{\mathrm d\ln x}=-\frac{s\langle\sigma v\rangle}{H}\left(Y^2-Y_{\mathrm{eq}}^2\right).
$$

In the radiation era $s\propto T^3$ and $H\propto T^2$, so $s/H$ is a constant times $1/x$. Write $\langle\sigma v\rangle=\langle\sigma v\rangle_0F(x)$, where $F\to1$ for slow particles. Then

$$
\frac{\mathrm dY}{\mathrm d\ln x}=-\frac{\lambda F(x)}{x}\left(Y^2-Y_{\mathrm{eq}}^2\right),\qquad \lambda=\frac{s\langle\sigma v\rangle_0}{H}\bigg|_{T=m}=\frac{2\pi\sqrt{10}}{15}\sqrt{g_*}\,mM_{\mathrm{Pl}}\langle\sigma v\rangle_0.
$$

For $m=100\,\mathrm{GeV}$ and $\langle\sigma v\rangle_0=1.830\times10^{-9}\,\mathrm{GeV^{-2}}$, $\lambda=5.48\times10^{12}$. This single number sets the whole problem. Note that $\lambda Y_{\mathrm{eq}}F/x$ is exactly the $\Gamma/H$ of the previous chapter.

## Two regimes and a match

While $\Gamma/H\gg1$, any excess $\Delta=Y-Y_{\mathrm{eq}}$ is destroyed almost at once. Setting $\mathrm dY/\mathrm d\ln x\simeq\mathrm dY_{\mathrm{eq}}/\mathrm d\ln x$ gives a small lag, $\Delta\simeq x^2/(2\lambda F)$ for $x\gg1$, so $Y$ tracks equilibrium. At $x=20$ this is $4.2\times10^{-11}$, 6 per cent of $Y_{\mathrm{eq}}$, as the full solution confirms.

Once $Y_{\mathrm{eq}}$ has collapsed, drop it. The equation separates:

$$
\frac{1}{Y_\infty}=\frac{1}{Y(x_f)}+\lambda\int_{x_f}^{\infty}\frac{F(x)}{x^2}\,\mathrm dx\;\simeq\;\frac{\lambda}{x_f}.
$$

The last step uses $F\simeq1$ and $Y(x_f)\gg Y_\infty$. The matching point $x_f$ is fixed by asking where the lag $\Delta$ becomes comparable to $Y_{\mathrm{eq}}$. D'Eramo and colleagues set $\Delta=cY_{\mathrm{eq}}$ and obtain

$$
\mathrm e^{x_f}x_f^{1/2}\simeq\frac{c(c+2)}{c+1}\,\frac{3\sqrt5\,g}{2\pi^{5/2}\sqrt{g_*}}\,mM_{\mathrm{Pl}}\langle\sigma v\rangle_0,
$$

their eq. (C12) with a constant cross-section, where $c$ is a number of order one. [D'Eramo, Fernandez and Profumo, 2017, appendix C](https://arxiv.org/abs/1703.04793v1) It follows from putting $Y_{\mathrm{eq}}\propto x^{3/2}\mathrm e^{-x}$ into the condition that the lag reaches $cY_{\mathrm{eq}}$, and its constant is the same combination as $\lambda$ times the prefactor of $Y_{\mathrm{eq}}$. Taking $c=\sqrt2-1$ and iterating $x_f=\ln(\ldots)-\tfrac12\ln x_f$ gives $x_f=21.7$. This matching point sits earlier than the crossing $\Gamma/H=1$ at $x=25.2$; both are markers on a gradual process.

## Worked example: the observed density

With $x_f=21.7$ and $\lambda=5.48\times10^{12}$,

$$
Y_\infty\simeq\frac{x_f}{\lambda}=3.97\times10^{-12}.
$$

To turn this into a density today, multiply by the mass and today's entropy density $s_0=2891\,\mathrm{cm^{-3}}$, and divide by the critical density $\rho_c/h^2=1.054\times10^{-5}\,\mathrm{GeV\,cm^{-3}}$:

$$
\Omega h^2=\frac{m\,Y_\infty\,s_0}{\rho_c/h^2}=2.74\times10^{8}\,\frac{m}{\mathrm{GeV}}\,Y_\infty=0.109.
$$

The numerical solution with a constant cross-section gives $Y_\infty=4.11\times10^{-12}$, so the matching estimate is 3.5 per cent low. With the Lab's $F(x)$, which is 0.86 to 0.9 near freeze-out, annihilation is slightly weaker and the full answer is $4.37\times10^{-12}$, or $\Omega h^2=0.120$. Planck measures $0.120\pm0.001$. [Planck Collaboration, 2018, abstract](https://arxiv.org/abs/1807.06209v4)

**Table 3.1** The full solution for the default Lab settings.

| $x$ | $Y$ | $Y/Y_{\mathrm{eq}}$ | $Y/Y_\infty$ |
| --- | --- | --- | --- |
| $10$ | $5.69\times10^{-6}$ | $1.00$ | $1.3\times10^{6}$ |
| $20$ | $7.14\times10^{-10}$ | $1.06$ | $163$ |
| $25.2$ | $3.42\times10^{-11}$ | $6.76$ | $7.83$ |
| $50$ | $7.81\times10^{-12}$ | $4\times10^{10}$ | $1.78$ |
| $100$ | $5.55\times10^{-12}$ | $6\times10^{31}$ | $1.27$ |
| $208$ | $4.81\times10^{-12}$ | $2\times10^{78}$ | $1.10$ |
| $1000$ | $4.37\times10^{-12}$ |  | $1.00$ |

Freeze-out is a process spread over a factor of ten in temperature. At $\Gamma/H=1$ there are still almost eight times the final number, and the last ten per cent disappears only by $x\simeq200$.

## Why the leftover goes as one over the cross-section

$Y_\infty\simeq x_f/\lambda$ and $\lambda\propto\langle\sigma v\rangle_0$, while $x_f$ grows only logarithmically. A larger cross-section therefore leaves proportionally less. The saved scan gives $\Omega h^2=2.23$, $0.248$ and $0.0273$ at $10^{-27}$, $10^{-26}$ and $10^{-25}\,\mathrm{cm^3\,s^{-1}}$: each decade costs a factor of nine rather than ten, because $x_f$ rises by about 2.3 per decade.

The required cross-section barely depends on the mass. In this model it is $1.92\times10^{-26}$ for 10 GeV, $2.14\times10^{-26}$ for 100 GeV and $2.36\times10^{-26}\,\mathrm{cm^3\,s^{-1}}$ for 1 TeV. The careful calculation of Steigman, Dasgupta and Beacom, with the full temperature dependence of $g_*$, gives $2.2\times10^{-26}$ within about 5 per cent above 10 GeV for $\Omega h^2=0.11$; their fit, rescaled to 0.120, gives $1.98\times10^{-26}$. [Steigman, Dasgupta and Beacom, 2012, section II.D](https://arxiv.org/abs/1204.3622v3) The Lab is 8 per cent higher, mostly because it holds $g_*$ fixed.

## Why a computer needs an implicit step

At $x=1$, $\Gamma/H=3.3\times10^{9}$: the equation tries to restore equilibrium a billion times per e-fold of cooling. An explicit step of any practical size overshoots and blows up. The saved model therefore uses the second-order backward differentiation formula in $\ln x$. Each step asks for the new $Y$ on both sides of the equation, which turns it into a quadratic,

$$
aY^2+Y-c=0,\qquad Y=\frac{2c}{1+\sqrt{1+4ac}},
$$

with $a$ and $c$ built from the step size, $\lambda F/x$, the two previous values and $Y_{\mathrm{eq}}^2$. This root is never negative and never loses precision by subtraction. The method guide [Recompute a relic abundance with the saved scripts](https://sandboxphysics.com/stories/dark-matter-relics/recompute-a-relic-abundance/) shows the explicit step failing and checks the implicit one against an independent time-domain solution.

## Lab tasks

Open the [dark matter relic Lab](https://sandboxphysics.com/labs/dark-matter-freeze-out-freeze-in/).

1. On the default run, read $Y_\infty$ and the settling point. Then compare them with the last two rows of Table 3.1.
2. Move the cross-section slider up by exactly one in the logarithm. Predict $\Omega h^2$ from the factor-of-nine rule, then read it.
3. Switch the start to "From nothing". Explain why the final density does not change.

## Exercises

1. Starting from $\mathrm dn/\mathrm dt+3Hn=\ldots$ and $s\propto T^3$, $H\propto T^2$, derive the equation for $\mathrm dY/\mathrm d\ln x$ and the expression for $\lambda$.
2. At $x=25.2$ the full solution has $Y=3.42\times10^{-11}$. Use $\int_{25.2}^\infty F/x^2\,\mathrm dx=0.0374$ to estimate $Y_\infty$, and explain the remaining difference from $4.37\times10^{-12}$.
3. A particle of 1 TeV needs $\langle\sigma v\rangle_0=2.36\times10^{-26}\,\mathrm{cm^3\,s^{-1}}$ in this model. Show from $Y_\infty\simeq x_f/\lambda$ why the value rises only slightly with mass.
4. Show that $\mathrm dn/\mathrm dt-3Hn=0$ would make $na^{-3}$ constant, and explain why that cannot describe dilution.

Answers

1. $\dot Y=(\dot n+3Hn)/s=-(\langle\sigma v\rangle s)(Y^2-Y_{\mathrm{eq}}^2)$. Divide by $\dot x=Hx$: $\mathrm dY/\mathrm d\ln x=-(s\langle\sigma v\rangle/H)(Y^2-Y_{\mathrm{eq}}^2)$. With $s=(2\pi^2/45)g_*T^3$ and $H=\sqrt{\pi^2g_*/90}\,T^2/M_{\mathrm{Pl}}$, $s/H=(2\pi\sqrt{10}/15)\sqrt{g_*}\,M_{\mathrm{Pl}}T=(2\pi\sqrt{10}/15)\sqrt{g_*}\,mM_{\mathrm{Pl}}/x$.
2. $1/Y_\infty=1/(3.42\times10^{-11})+5.48\times10^{12}\times0.0374=2.92\times10^{10}+2.05\times10^{11}$, so $Y_\infty=4.27\times10^{-12}$, 2.4 per cent below the full value. Just after $x=25.2$ inverse production, the $Y_{\mathrm{eq}}^2$ term, still adds a little.
3. $\lambda\propto m$, so at fixed cross-section $Y_\infty\propto x_f/m$ and $\Omega h^2\propto mY_\infty\propto x_f$. The mass cancels except through $x_f$, which grows by $\ln10\simeq2.3$ per decade of mass. Keeping $\Omega h^2$ fixed then needs a cross-section larger by about $x_f(1\,\mathrm{TeV})/x_f(100\,\mathrm{GeV})\simeq1.1$.
4. $\mathrm d(na^{-3})/\mathrm dt=a^{-3}(\dot n-3Hn)=0$, so $n\propto a^3$ would grow as the volume grows. A fixed number of particles in a growing volume must have $n\propto a^{-3}$, which is $\dot n+3Hn=0$.

## Sources

1. [Particle Data Group, 2025, eq. 27.2](https://pdg.lbl.gov/2025/reviews/rpp2025-rev-dark-matter.pdf)
2. [D'Eramo, Fernandez and Profumo, 2017, eq. 15](https://arxiv.org/abs/1703.04793v1)
3. [Planck Collaboration, 2018, abstract](https://arxiv.org/abs/1807.06209v4)
4. [Steigman, Dasgupta and Beacom, 2012, section II.D](https://arxiv.org/abs/1204.3622v3)

## Keep reading

How dark matter was left over · Reading path

### From thermal equilibrium to relentless dark matter

1. [Article **Why did any dark matter survive the hot early universe?**](https://sandboxphysics.com/stories/dark-matter-relics/why-some-dark-matter-survived-the-cooling/)
2. [Article **Equilibrium, expansion and the reaction race**](https://sandboxphysics.com/stories/dark-matter-relics/equilibrium-expansion-and-the-reaction-race/)
3. [You are here **The Boltzmann equation and freeze-out**](https://sandboxphysics.com/stories/dark-matter-relics/the-boltzmann-equation-and-freeze-out/)
4. [Article **Freeze-in, hot relics and faster expansion**](https://sandboxphysics.com/stories/dark-matter-relics/freeze-in-hot-relics-and-faster-expansion/)
5. [Article **Recompute a relic abundance with the saved scripts**](https://sandboxphysics.com/stories/dark-matter-relics/recompute-a-relic-abundance/)
6. [Article **How to read the relentless dark matter paper of D'Eramo and colleagues**](https://sandboxphysics.com/stories/dark-matter-relics/reading-relentless-dark-matter/)
