Interacting Dark Matter (IDM)

• Interacting dark matter (IDM) is a class of models where dark matter undergoes nongravitational interactions that modify detection kinematics, cosmic perturbations, and halo formation.
• IDM theories encompass elastic and inelastic scattering with baryons, photons, neutrinos, as well as self-interactions mediated by light or continuum states, influencing both early-universe dynamics and laboratory signals.
• Astrophysical simulations of IDM reveal measurable impacts on halo core sizes, subhalo concentrations, and star formation rates, with important consequences for 21-cm cosmology and direct detection experiments.

Interacting dark matter (IDM) denotes departures from collisionless cold dark matter in which the dark-matter sector undergoes nongravitational interactions that affect direct-detection kinematics, linear perturbation evolution, halo formation, or the background cosmology. In the literature summarized here, the same acronym is used for several distinct constructions: elastic dark-matter scattering with baryons, photons, or neutrinos; interactions with dark radiation or dark energy; self-interactions mediated by light or continuum states; and, in direct-detection studies, inelastic dark matter with a mass-split excited state (Fischer et al., 16 Apr 2025, Moliné et al., 2019, Harko et al., 2022, Scopel et al., 2014). This suggests that IDM is best understood as a family of non-collisionless dark-matter hypotheses rather than a single model.

1. Taxonomy and defining interaction structures

A common starting point for IDM is an elastic differential cross section in the center-of-mass frame,

$\frac{d\sigma}{d\Omega}(v,\theta),$

with explicit dependence on the relative velocity $v$ and the scattering angle $\theta$. In dark-matter–baryon applications, the velocity dependence is frequently parameterized as a power law, $\sigma(v)=\sigma_0 (v/v_0)^n$, and the momentum-transfer cross section is defined by

$$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$

Fischer et al. treat both a forward-peaked frequent-scattering limit at fixed $\sigma_T$ and isotropic hard-sphere scattering, which makes the formalism applicable to both drag-dominated and rare large-angle regimes (Fischer et al., 16 Apr 2025).

In early-universe IDM with photons or neutrinos, the interaction does not primarily appear as isolated particle collisions inside halos; instead it enters the Boltzmann hierarchy as a drag term that suppresses the growth of small-scale perturbations before kinetic decoupling. In that setting the linear matter spectrum is written as

$$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$

with a collisional-damping cutoff and, in some models, dark acoustic oscillations (Moliné et al., 2019, Lopez-Honorez et al., 2018).

A related but distinct branch is self-interacting dark matter, often described by a Yukawa potential

$V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$

or, in continuum-mediated realizations, by a non-integer power-law potential $V(r)\sim -1/r^{2\Delta-1}$. The relevant transport quantity is again $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$, but here it controls halo heat conduction and orbit isotropization rather than DM–Standard Model coupling. The benchmark SIDM models summarized by Tulin, Yu, and Zurek favor mediator masses in the $v$0–$v$1 MeV range and $v$2–$v$3 at dwarf-galaxy velocities, while continuum mediation replaces the usual Yukawa velocity scalings by non-integer exponents such as $v$4 in the Born regime and $v$5 in the classical regime (Kaplinghat et al., 2013, Chaffey et al., 2021).

In direct-detection phenomenology, “IDM” often means inelastic dark matter: a ground state $v$6 or $v$7 upscatters to a heavier state $v$8 or $v$9 with mass splitting $\theta$0. The sign of $\theta$1 distinguishes endothermic from exothermic scattering, and the minimal incoming speed becomes recoil dependent. That terminology is standard in the halo-independent analysis of Scopel and Yoon and in terrestrial upscattering/downscattering studies using xenon detectors (Scopel et al., 2014, Alam et al., 14 Oct 2025).

2. DM–baryon scattering in $\theta$2-body and hydrodynamics simulations

A central technical development in recent IDM work is the explicit embedding of DM–baryon scattering into astrophysical simulation codes. Fischer et al. implement such interactions in OpenGadget3, using both smoothed-particle hydrodynamics with 230 neighbors and a Wendland $\theta$3 kernel, and meshless finite mass with 32 neighbors and a cubic spline kernel (Fischer et al., 16 Apr 2025).

Their scheme treats each DM particle as interacting pairwise with nearby baryonic particles through a virtual baryon constructed at the baryon position,

$\theta$4

with mass

$\theta$5

and velocity

$\theta$6

The stochastic interaction probability for a DM–virtual-baryon pair depends on $\theta$7, the relative speed, the reduced-mass factor, the timestep $\theta$8, and the kernel-overlap integral $\theta$9. If a scattering occurs, the relative velocity is rotated in the center-of-mass frame according to the sampled angular distribution, after which the virtual particle is destroyed and its momentum and energy changes are deposited back into the real baryon. The bulk-velocity update is

$\sigma(v)=\sigma_0 (v/v_0)^n$0

The implementation is designed to preserve energy and momentum in each collision, while also retaining arbitrary angular and velocity dependence and the physical mass ratio between the dark and baryonic scattering partners. Each DM particle searches $\sigma(v)=\sigma_0 (v/v_0)^n$1 neighbors in the baryon species; to control the number of pairs, the smoothing lengths are rescaled so that on average $\sigma(v)=\sigma_0 (v/v_0)^n$2 overlaps occur. Time stepping is synchronized and chosen so that the largest pairwise probability remains $\sigma(v)=\sigma_0 (v/v_0)^n$3 and the fractional velocity kick is small (Fischer et al., 16 Apr 2025).

The validation strategy uses periodic boxes without gravity. The code reproduces the analytic thermalization law for two Maxwellian components with percent-level accuracy in both SPH and MFM, for both forward-peaked and isotropic scattering. It also matches analytic momentum-transfer evolution in drag tests, and it preserves the same analytic thermalization behavior at extreme mass ratio, including runs with $\sigma(v)=\sigma_0 (v/v_0)^n$4 and isotropic scattering (Fischer et al., 16 Apr 2025).

An idealized halo-collapse application follows a $\sigma(v)=\sigma_0 (v/v_0)^n$5 spherical overdensity from $\sigma(v)=\sigma_0 (v/v_0)^n$6 to $\sigma(v)=\sigma_0 (v/v_0)^n$7 with $\sigma(v)=\sigma_0 (v/v_0)^n$8 particles. For forward-peaked $\sigma(v)=\sigma_0 (v/v_0)^n$9, the $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$0 halo develops a central dark-matter core whose size and depth increase with $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$1. The baryonic response is non-monotonic: at $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$2 the baryon density peaks because of net cooling, whereas larger $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$3 reheats the gas and lowers the baryon density. The local interaction rate peaks early and falls to $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$4 by $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$5 (Fischer et al., 16 Apr 2025).

The same work emphasizes nontrivial numerical limits. Artificial small-scale turbulence from discrete kicks must be damped by viscosity, but excessive viscosity can erase real turbulence. Energy conservation in leapfrog KDK integration is easier to restore in SPH than in MFM; increasing $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$6 yields little improvement, whereas halving $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$7 significantly reduces energy errors. The formal continuum limit is $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$8, $$\sigma_T(v)=2\pi\int_{-1}^{1}(1-\cos\theta)\,\frac{d\sigma}{d\Omega}(v,\theta)\,d\cos\theta.$$9, $\sigma_T$0, and inner-profile measurements remain resolution sensitive at the $\sigma_T$1kpc scale for $\sigma_T$2 particles (Fischer et al., 16 Apr 2025).

3. Collisional damping, linear perturbations, and suppressed small-scale power

When dark matter scatters with photons, neutrinos, or baryons before or around recombination, the primary signature is a modification of linear perturbation growth. For DM–photon scattering, the Euler equation acquires a momentum-exchange term proportional to $\sigma_T$3, and the linear power spectrum can be parameterized by a transfer function cutoff. A representative estimate gives a comoving damping scale $\sigma_T$4 and a cutoff halo mass $\sigma_T$5–$\sigma_T$6 for

$\sigma_T$7

showing how modest early-time drag can reorganize the low-mass halo population without altering large scales (Moliné et al., 2019).

Lopez-Honorez et al. express the linear suppression in both warm dark matter and IDM through

$\sigma_T$8

and use the half-mode mass $\sigma_T$9 to connect transfer-function suppression to halo abundances. For IDM with photon scattering, they employ a halo-mass-function correction calibrated to simulations rather than the simpler warm-dark-matter fit, which makes the faint-end response shallower than a matched free-streaming cutoff (Lopez-Honorez et al., 2018).

A different realization appears in the partial-DM baryon-scattering model studied by He et al. Only a fraction

$$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$0

of the total dark matter is interacting; their baseline choice is $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$1 with $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$2 and velocity-independent $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$3. Solving the modified Boltzmann equations yields a plateau-like power suppression, with

$$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$4

dropping to $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$5, $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$6, and $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$7 at $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$8, $$P_{\rm IDM}(k,z)=P_{\rm CDM}(k,z)\,[T_{\rm IDM}(k)]^2,$$9, and $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$0, respectively, for the best-fit model with $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$1 (He et al., 2023).

For DM–neutrino interactions, the same logic carries into late-time full-shape analyses through the Effective Field Theory of Large-Scale Structure. The DESI ELG forecast of Li et al. uses the dimensionless coupling

$V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$2

and shows that optimistic priors on EFTofLSS nuisance parameters can produce a $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$3 upper limit $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$4 for a CDM fiducial, or a detection

$V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$5

for an interacting fiducial with $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$6 and $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$7. Under intermediate or pessimistic priors, however, the bounds weaken by roughly an order of magnitude because the interaction is degenerate with counterterms and stochastic contributions (Mosbech et al., 2024).

4. Halo substructure, galaxy formation, and nonlinear consequences

The nonlinear realization of IDM depends on the channel, but a recurring outcome is a depletion or restructuring of low-mass halos. In high-resolution cosmological simulations with DM–photon scattering, halo concentrations below $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$8 are systematically lower than in CDM by $V(r)=\pm \alpha_\chi \frac{e^{-m_\phi r}}{r},$9–$V(r)\sim -1/r^{2\Delta-1}$0, while subhalo concentrations at $V(r)\sim -1/r^{2\Delta-1}$1 are lower by $V(r)\sim -1/r^{2\Delta-1}$2–$V(r)\sim -1/r^{2\Delta-1}$3. The cumulative subhalo mass function in Local Group zoom runs is fit by

$V(r)\sim -1/r^{2\Delta-1}$4

with $V(r)\sim -1/r^{2\Delta-1}$5, $V(r)\sim -1/r^{2\Delta-1}$6 in CDM and $V(r)\sim -1/r^{2\Delta-1}$7, $V(r)\sim -1/r^{2\Delta-1}$8 in IDM over $V(r)\sim -1/r^{2\Delta-1}$9–$$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$0. The radial profile of subhalo number density retains the same qualitative shape, but with an overall normalization smaller by $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$1–$$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$2 (Moliné et al., 2019).

These changes propagate directly into indirect-detection forecasts. Because annihilation luminosity depends on $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$3, and because the subhalo boost factor is sensitive to both concentration and abundance, the same simulations suggest

$$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$4

A plausible implication is that interacting scenarios reduce, but do not eliminate, the contribution of unresolved halo substructure to Galactic and extragalactic annihilation signals (Moliné et al., 2019).

Radiation-hydrodynamic simulations during the Epoch of Reionization extend this picture to galaxies. In THESAN-HR, the strong-DAO interacting-dark-matter benchmark has $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$5, $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$6, and

$$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$7

Below this scale the halo mass function is suppressed by up to a factor of $$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$8–$$\sigma_T=\int d\Omega\,(1-\cos\theta)\,d\sigma/d\Omega$$9, and the UV luminosity function at $v$00 is reduced by $v$01–$v$02 dex relative to CDM. Yet the nonlinear response is not purely suppressive: the specific star-formation rate in halos below $v$03 is enhanced by roughly $v$04–$v$05, the gas fraction at $v$06 rises from $v$07 to $v$08, gas-depletion times shorten by $v$09, galaxy sizes shrink by $v$10, and metallicity gradients steepen at the outskirts (Shen et al., 2023).

The THESAN-HR results also show that alternative-dark-matter models can “catch up” in star formation and reionization. In the strong-DAO run, the star-formation rate density at $v$11 is about half the CDM value, but by $v$12 the difference narrows to $v$13. Reionization is delayed only modestly, with $v$14 at $v$15 rather than $v$16. At the same time, the faint-end luminosity function is sensitive to reionization morphology, and uniform-UVB approximations can mimic or obscure the signatures of suppressed small-scale structure. That systematic uncertainty is itself a central part of the IDM phenomenology (Shen et al., 2023).

5. 21-cm cosmology as a probe of IDM

The 21-cm signal is sensitive to IDM because scattering changes both the thermal state of the gas and the timing of star formation. In the global-signal analysis of Lopez-Honorez et al., the brightness temperature is

$v$17

and the IDM effect enters mainly through delayed Ly$v$18 coupling, delayed X-ray heating, and delayed UV ionization. Requiring either $v$19 or an absorption minimum above the EDGES central value $v$20 yields

$v$21

for $v$22 and

$v$23

for $v$24, improving the previous Planck + LSS limit of $v$25 by roughly an order of magnitude under the adopted astrophysical assumptions (Lopez-Honorez et al., 2018).

For Coulomb-like DM–baryon interactions, Mittal et al. emphasize that cooling alone is insufficient: one must also include the suppression of small-scale structure, which delays star formation and weakens the Ly$v$26, X-ray, and ionizing backgrounds. Their cross section is

$v$27

with both heat exchange and drag contributing to the gas-temperature evolution. A joint Bayesian fit of a 21-cm signal model and a flexible foreground model to the SARAS3 antenna temperature finds that the signal parameters remain weakly constrained after foreground marginalization. The principal functional bound is

$v$28

at $v$29, and the evidence ratio between IDM and CDM is only $v$30, which is statistically inconclusive (Mittal et al., 1 May 2026).

Interferometric power-spectrum forecasts are substantially stronger. Rahimieh et al. use a Fisher analysis for HERA over $v$31–$v$32 MHz, marginalizing over star-formation efficiency, escape fraction, and X-ray luminosity. For the optimistic HERA configuration, the projected $v$33 upper bounds at $v$34 are

$v$35

and

$v$36

The same forecasts improve global-signal sensitivities by at least a factor of five for $v$37 and by more than an order of magnitude for $v$38, and they improve on existing CMB and Milky Way satellite bounds. The inferred IDM cross section is essentially uncorrelated with the Population-II star-formation efficiency and escape fraction, whereas the Coulomb-like case is positively correlated with X-ray luminosity (Rahimieh et al., 28 Aug 2025).

6. Inelastic dark matter in direct detection and terrestrial scattering

In the direct-detection literature, IDM usually refers to inelastic dark matter. Scopel and Yoon formulate the basic kinematics through an excited state $v$39, with $v$40 for endothermic and $v$41 for exothermic scattering. The minimum incoming speed required to produce recoil energy $v$42 is

$v$43

and the differential recoil rate factors into nuclear response functions and the halo function

$v$44

By rebinning multiple experiments into common $v$45 intervals, one can compare measurements and limits without specifying a halo model. Their scan over $v$46 and $v$47 identifies five benchmark regions where at least one of the DAMA, CDMS-Si, or CRESST excesses can evade the strongest null limits through non-overlapping $v$48 ranges. No single benchmark, however, accommodates all three excesses, and the kinematic “rescue” regions disappear if one assumes the standard Maxwellian halo (Scopel et al., 2014).

A more recent extension replaces detector-only upscattering by a two-step terrestrial process. In “Dark Matter Boosted by Terrestrial Collisions,” the dark sector contains $v$49 and $v$50 with splitting $v$51, no tree-level elastic coupling, and only off-diagonal scattering $v$52 or $v$53. Halo $v$54 can upscatter in the Earth, primarily on heavy nuclei such as Pb, and the resulting excited $v$55 can then downscatter in an underground detector, releasing the stored excitation energy (Alam et al., 14 Oct 2025).

This mechanism materially extends the accessible inelastic frontier. For XENON1T with a $v$56–$v$57 keV recoil window and $v$58 ton-yr exposure, including Earth upscattering pushes sensitivity from the conventional $v$59 to $v$60 for $v$61. In the XENON100 high-energy $v$62–$v$63 keV analysis, the smaller exposure weakens the low-$v$64 reach but allows sensitivity up to $v$65, exceeding previous direct limits by $v$66. The quoted constrained region corresponds to $v$67–$v$68 (Alam et al., 14 Oct 2025).

A recurrent misconception is that inelastic models are invisible once detector upscattering becomes kinematically forbidden. The terrestrial-collision channel shows that this is not generally correct: excitation outside the detector can reactivate the signal through downscattering inside it (Alam et al., 14 Oct 2025).

7. Interacting dark sectors: dark energy, dark radiation, and cosmological tensions

Some IDM constructions move the interaction entirely into the dark sector or into the cosmological background equations. In the open-system interacting dark energy–dark matter model of Harko et al., each species satisfies a particle-balance equation $v$69, and the effective creation pressure is

$v$70

The background energy transfer is encoded by $v$71 through

$v$72

$v$73

together with a scaling ansatz $v$74. With Planck+R20 they obtain $v$75, $v$76, and $v$77, while Planck+BAO+Pantheon gives $v$78 and $v$79 (Harko et al., 2022).

A phenomenological but perturbatively explicit interacting DM–DE analysis using KiDS and Planck arrives at a related conclusion for the $v$80 discrepancy. Yang et al. consider four couplings of the form

$v$81

with different allowed regions in $v$82. In $v$83CDM the residual tension is $v$84; in the four IDMDE models it drops to $v$85, $v$86, $v$87, and $v$88, and the joint KiDS+Planck fit yields $v$89 to $v$90, indicating moderate preference for interaction (An et al., 2017).

Dark-matter–dark-radiation models supply another route to altered early-universe dynamics. Buen-Abad et al. study an interacting subcomponent with fraction $v$91 coupled to dark radiation until $v$92–$v$93, with momentum-exchange rate $v$94. Among their benchmark decoupling histories, the exponential or “step” case yields the best tension relief. Without ACT DR6, fitting Planck+BAO+Pantheon+SH0ES gives

$v$95

reducing the tension to below $v$96. Once ACT DR6 is included, the best-fit value falls to $v$97, and the maximal $v$98 in the $v$99 region is $\theta$00 without SH0ES or $\theta$01 with SH0ES, so the improvement becomes only mild (Buen-Abad et al., 20 Nov 2025).

More structured dark-sector realizations combine dark radiation, dark recombination, and inelasticity. The “SIDR+$\theta$02” framework introduces an inelastic fermion doublet coupled to self-interacting dark radiation under a $\theta$03 gauge symmetry. The DM–DR rate scales as $\theta$04, and the model yields best-fit cosmological parameters

$\theta$05

with the CMB-inferred Hubble constant shifting from $\theta$06 to $\theta$07–$\theta$08 and $\theta$09 moving to $\theta$10–$\theta$11 (Cho et al., 2024).

The NuADaM model provides an atomic-dark-matter realization of the same general strategy. A subcomponent of dark atoms interacts strongly with an interacting dark-radiation sector and decouples during the CMB epoch following dark recombination. In global fits to Planck 2018 TT,TE,EE+lensing+BAO+Pantheon with the SH0ES prior, the model returns

$\theta$12

and improves the fit relative to both $\theta$13CDM and conventional atomic dark matter in the data combinations reported (Buen-Abad et al., 2024).

Across these cosmological applications, the main controversy is not whether interactions can shift $\theta$14 or $\theta$15 in principle, but whether the required parameter regions remain viable once higher-$\theta$16 CMB data, full-shape large-scale structure, Ly$\theta$17, or other small-scale probes are included. The existing results are therefore best read as model-dependent demonstrations of degeneracy lifting and tension redistribution, rather than as a settled empirical confirmation of any single interacting-dark-sector scenario (Buen-Abad et al., 20 Nov 2025, Buen-Abad et al., 2024).