Turn the cross-section down far enough and the same Boltzmann equation stops describing freeze-out. If the particles start in equilibrium they keep the hot plateau, Y≃5.3×10−3. If they start with none, they are made from the bath with a yield that rises in proportion to the cross-section, YFI=(3π2/32)a2λ for the Lab's contact interaction. A faster early expansion changes the freeze-out branch: with a kination phase that ends at 20 MeV, annihilation keeps thinning the particles for a factor of a few hundred in temperature after they leave equilibrium.
This chapter follows The Boltzmann equation and freeze-out. It uses λ, F(x) and Yeq=ax2K2(x) with a=45g/(4π4g∗s)=2.68×10−3. You need integrals of Bessel functions, or a table of them. The aims are to compute the hot-relic and freeze-in yields, map all three regimes on one curve and derive the relentless law.
The hot-relic plateau
If Γ/H never exceeds one but the particles start in equilibrium, nothing changes Y and it stays at its value while relativistic:
Yhot=2a=4π4g∗s90g=5.36×10−3.
The Lab's hot-relic preset uses ⟨σv⟩0=10−38cm3s−1, whose peak Γ/H is 1.5×10−3; that much annihilation removes 0.5 per cent, so it reports 5.33×10−3. For m=100GeV this is Ωh2=1.46×108, a billion times too much. Only a particle lighter than about 82 eV in this model ends with Ωh2=0.120, and a relic that light is hot dark matter, which is excluded. Hall, Jedamzik, March-Russell and West, 2010, section 5.1 The same reasoning for Standard Model neutrinos, which decouple near 1 MeV, gives the familiar result Ωh2=m/93eV. Particle Data Group, 2025, eq. 27.4
Freeze-in from an empty start
Start with Y=0 and a tiny λ. As long as Y≪Yeq, drop Y2:
dxdY=x2λF(x)Yeq2,YFI=λa2∫0∞x2K2(x)2F(x)dx.
The integral can be done exactly for the contact interaction. Its thermal average obeys
where s is the pair energy squared. Put T=m/x and swap the order of integration. The x integral is ∫0∞x3K1(bx)dx=3π/(2b4) with b=s/m. What remains is
where the last integral becomes the beta function B(21,23)=π/2 after substituting s=4m2/w.
Worked example: freezing in the observed amount
The observed density needs Y=0.120/(2.74×108×100)=4.37×10−12. With 3π2a2/32=6.64×10−6, that requires λ=6.59×10−7, or
⟨σv⟩0=(2π10/15)g∗mMPlλ=2.57×10−45cm3s−1.
That is eight billion billion times below the freeze-out value. The full solution gives Y∞=4.373×10−12, 0.1 per cent below the closed form, the small difference coming from the start at x=10−3 and the dropped Y2. Dropping Y2 is safe: by x=3, when 99 per cent of the yield is made, Y is still below 3×10−9 of Yeq, and later the produced particles are far too dilute to find each other.
Production is concentrated near x≃1. In the saved run Y reaches 47 per cent of its final value by x=0.5, 75 per cent by x=1 and 99 per cent by x=3. Hall and colleagues call this IR domination: the yield is set by temperatures near the mass and does not care what happened earlier. Hall, Jedamzik, March-Russell and West, 2010, section 1 It works here because F∝x2 at small x. If the cross-section were constant, dY/dx∝x−2 at small x and the yield would grow as 1/xstart: the saved scripts find a factor of 10.1 between starts at x=10−3 and 10−2.
One curve, three regimes
Table 4.1 Density today for m=100GeV and standard expansion, from the saved scan.
⟨σv⟩0 in cm3s−1
Start in equilibrium
Start empty
10−45
1.47×108
0.0467
10−40
1.47×108
4.67×103
10−37
1.40×108
4.66×106
10−36
1.01×108
3.99×107
10−35
3.05×107
3.05×107
10−30
1.48×103
1.48×103
10−26
0.248
0.248
Reading down, the empty start rises tenfold per decade (freeze-in), meets the plateau near 10−35, and from there both starts fall together (freeze-out). The meeting point is where the peak of Γ/H reaches about one. Above it, particles made from nothing reach equilibrium before freeze-out and the starting state is forgotten. This is the slice through the phase diagram drawn in Figure 4b of Hall and colleagues. Hall, Jedamzik, March-Russell and West, 2010, section 5.1
with xr=m/Tr. Kination, a scalar field dominated by its kinetic energy, has n=2. Light-element abundances require Tr≳(15.4)1/nMeV, their eq. (14), so the Lab keeps Tr≥20MeV, which satisfies the bound for every n it offers.
The larger H makes freeze-out earlier: with the standard cross-section and Tr=20MeV the crossing moves from x=25.2 to 22.4, 19.5 and 12.9 for n=1, 2 and 4.
The relentless law
For xf≪x≪xr, E≃(xr/x)n/2 and Yeq is negligible, so
dlnxdY1=xrn/2λFxn/2−1.
For n<2 the right side shrinks with x and 1/Y converges soon after freeze-out, as in the standard case. For n=2 it is constant, so 1/Y grows by (λ/xr)ln(x/xf) all the way to x≃xr. For n>2 it grows as a power. This is the relentless annihilation of eqs. (C16) and (C18). D'Eramo, Fernandez and Profumo, 2017, appendix C
Check it on the kination run, xr=5000. The time-domain solution gives 1/Y=2.57×109 at x=200 and 3.56×109 at x=500. The law predicts an increase of (λ/xr)∫Fdlnx=1.10×109×0.90=9.9×108, matching to 0.1 per cent. The final densities with the standard cross-section are Ωh2=0.871, 4.02 and 13.7 for n=1, 2 and 4, against 0.120 with standard expansion. The settling point moves from x=208 to 1140, 8080 and 26,900. These densities are 2 to 3 times below the paper's Figure 1, because the paper keeps the full g∗(T) and the fall from 86.25 to 10.75 by Tr raises the extra fluid's share of H2. The guide to the paper works through the difference.
Choose Freeze-in from nothing. On the abundance chart, find the x where the violet curve reaches its plateau and compare with the 99 per cent point above.
On the scan chart, follow the dashed empty-start curve up to where it joins the solid curve. Read off the cross-section and compare with Table 4.1.
Choose Faster expansion. Watch the dashed curve for the particles present on the rate chart: it stays near one for a long time instead of dropping away. Explain this with the relentless law.
Exercises
Check the closed form numerically: with a=2.678×10−3 and λ=6.597×10−7, compute YFI.
In freeze-in, YFI∝λ∝m⟨σv⟩0, while the observed yield is proportional to 1/m. What cross-section does a 1 TeV particle need?
For a constant cross-section, show that ∫xstartYeq2x−2dx≃4a2/xstart when xstart≪1.
For n=1, integrate the relentless equation from xf to infinity, ignoring the eventual return to E≃1, and show that 1/Y stays finite.
AnswersShow the answer
3π2/32=0.9253 and a2=7.172×10−6, so YFI=0.9253×7.172×10−6×6.597×10−7=4.378×10−12.
Fixing Ωh2 means Y∝1/m, so m⟨σv⟩0∝1/m and ⟨σv⟩0∝m−2. Ten times the mass needs a hundred times less: 2.57×10−47cm3s−1.
For x≪1, Yeq≃2a, so the integrand is 4a2/x2 and ∫xstart∼14a2x−2dx≃4a2/xstart. The yield depends on where one starts, which no measurement fixes.
d(1/Y)/dx=λx−3/2xr−1/2, so Δ(1/Y)=2λxr−1/2xf−1/2, which is finite. The extra annihilation after freeze-out adds only a fixed correction.