WorksPhysics and Astronomy
Energy exchange confined to the perturbations: the frozen dark sector confronted with Planck, DESI, supernovae and cosmic shear
Abstract
We present cosmological constraints on the frozen model, in which dark matter and dark energy exchange energy only at the level of the perturbations. The expansion history is that of flat CDM, while on sub-horizon scales a dimensionless coupling enhances () or suppresses () the growth of the dark-matter density contrast. We use Planck 2018 CMB and lensing data, DESI DR2 baryon acoustic oscillations (BAO), the DES five-year supernovae (Dovekie recalibration), and the KiDS-Legacy and DES Year 3 cosmic-shear amplitudes, corrected for the model's growth. Planck alone gives , a preference driven by the lensing smoothing of the acoustic peaks and the low large-scale temperature power. Since BAO and supernovae do not constrain , the coupling acts only through the Planck posterior, lowering the Planck–DESI tension from to –, and that between Planck plus supernovae and DESI from – to –. Combining Planck, DESI and supernovae we find and relative to CDM, a preference that vanishes when the strongest-coupling region is excluded. The enhanced growth, however, raises the Planck–shear tension from to –. With shear included, and , while favors CDM. and the age of the Universe agree with CDM for every data combination, and the SH0ES discrepancy remains at . Without shear, the favored coupling predicts half the CDM integrated Sachs–Wolfe amplitude at and an enhanced shear signal in the lowest tomographic bins. These signatures and the redshift distribution of cluster counts test the coupling independently of the amplitude of the fluctuations.
Introduction
The standard CDM model describes an impressive range of observations with just six parameters, yet the growing precision of the data has brought to light a few discrepancies between early- and late-time probes. The best known are the difference between the Hubble constant inferred from the CMB and the one measured with the local distance ladder, and the somewhat lower amplitude of the matter fluctuations found by weak-lensing surveys; both have been reviewed extensively [1, 2, 3]. Assessing such discrepancies requires once different estimators answer different questions and react differently to non-Gaussian posteriors and to prior volumes. A detailed comparison of these tools [4], their generalization to several data sets at once [5], and studies of how small-scale inhomogeneities in light propagation or an energy diffusion in unimodular gravity could affect the inferred expansion rate [6, 7] illustrate how much can be learned by quantifying tensions in a uniform way before interpreting them. The second data release of DESI has added a new element: its BAO measurements prefer a lower matter density than the CMB within CDM, a difference that becomes more significant when type Ia supernovae are included and that has been read as a hint of dynamical dark energy [8, 9]. Whether late-time modifications of the expansion can absorb these discrepancies without spoiling the BAO calibration is itself a subtle question [10].
An interaction between dark matter and dark energy is one of the most studied extensions of CDM, and its history is almost as long as that of dynamical dark energy [11, 12, 13]. Interactions were first proposed as a way to address the coincidence problem [14, 15], for instance through exchange rates that drive the ratio of the dark densities to a stationary value [14], and the variety of couplings explored since then, linear and nonlinear in the densities [16, 17], is discussed in the review of Ref. [18]. Closely related ideas include the diffusion of matter into a dark-energy field [19] and the decaying vacuum, whose background and structure formation, including the ISW signal, have been confronted with data [20, 21]. Interest in these models grew when it became clear that an energy flow from dark matter to dark energy can raise the CMB value of and lower the amplitude of the fluctuations at the same time [22, 23, 24], provided that the perturbations are formulated in a stable way [25], also when the dark energy itself is dynamical [26]. These models have since been tested with the full shape of the galaxy power spectrum [27], with forecasts for gravitational-wave standard sirens [28] and with the DESI data [29, 30, 31], which revealed a close degeneracy between interacting and evolving dark energy at the background level [32]. Model-independent reconstructions of the interaction kernel [33] and forecasts for narrow-band photometric surveys [34, 35] show how much room the next generation of data will have to discriminate among these possibilities. Couplings of dark energy to visible matter, on the other hand, are being probed directly in the laboratory [36, 37]. Most of the interacting models mentioned above exchange energy at the level of the background, which changes the expansion history and is therefore constrained mainly by geometric probes. A fluid description must also specify how the exchange enters the perturbations, and some prescriptions lead to large-scale instabilities [38, 39, 40].
A different class of interactions leaves the background untouched and acts only on the inhomogeneities. The best known examples transfer momentum: elastic scattering between dark matter and dark energy [41, 42, 43] and the pure-momentum couplings of the Pourtsidou–Skordis–Copeland classification [44, 45, 46, 47], which have also been described in effective and Lagrangian frameworks [48, 49, 50, 51]. These couplings act on the velocity of the dark matter and slow its infall. In the frozen model, proposed by some of us in Ref. [52], the exchange is instead an energy transfer confined to the perturbations: the background densities evolve as in the noninteracting model, while the local dark-matter density gains or loses energy in proportion to the local density contrast of the dark energy in the frame of the total dark momentum. The dark-matter velocity is unchanged, but the density responds to the convergence of the flow with a modified, time-dependent efficiency. Geometric probes are blind to the coupling, redshift-space distortions are only weakly sensitive to it, and the signal is carried by the density field: the amplitude of the matter fluctuations, the CMB lensing potential, the integrated Sachs–Wolfe (ISW) effect [53, 54] and the abundance of collapsed objects.
The companion analysis [52] allowed for a dark-energy equation of state . Here we fix , so that the expansion history is exactly that of CDM and every effect of the coupling is confined to the growth of structure. Two features make a careful observational analysis of this case worthwhile. First, since the expansion history is that of CDM, any change in the agreement between the CMB, BAO and supernovae must come from how the coupling moves the CMB posterior, which isolates the role of the growth of structure in the tensions among these data sets. Second, the coupling acts precisely on the observables in which the Planck 2018 data show their best-known peculiarities: the preference for more lensing smoothing of the acoustic peaks than predicted by CDM [55, 56] and the low amplitude of the temperature spectrum at the largest scales. A coupling that enhances the late growth of the dark-matter density can absorb both, at the price of a large present amplitude of the fluctuations, which cosmic shear and cluster counts can test.
In this work we confront the frozen model with Planck 2018 temperature, polarization and lensing data, with the DESI DR2 BAO measurements and with the DES five-year supernovae in their Dovekie recalibration, and we add the cosmic-shear measurements of KiDS-Legacy and DES Year 3, compared with the model through an effective amplitude that accounts for its growth history. Before combining the data sets we quantify their agreement in CDM and in the frozen model with a battery of tension estimators based on parameter differences, on the goodness of fit and on the Bayesian evidence. All posteriors and evidences are computed with exact likelihoods, by adaptive importance sampling for the combinations that include the CMB and by nested sampling for the external data alone, with the same priors for every data set. We then derive the joint constraints, compare the model with CDM, and follow the consequences of each data combination for a set of late-time observables: the expansion rate, the Hubble constant, the age of the Universe, the ISW amplitude, and the amplitude of the fluctuations probed by cosmic shear and cluster counts, which we compare with independent measurements. Two appendices study in detail how cosmic shear and cluster abundances respond to the coupling, with the published source distributions of the shear surveys and a halo mass function computed from the linear spectrum of each model.
The paper is organized as follows. Section II presents the model, its sub-horizon limit and its phenomenology, and Sec. III the data. Section IV describes the sampling and the estimators. § 1.4 discusses the constraints from Planck alone and the physical boundary of the model, § 2 the agreement between the data sets, and § 3 the joint analysis and the role of cosmic shear. § 4 compares the predictions for late-time observables with independent measurements, and Sec. IX discusses redshift-space distortions, the relation with other perturbation-only couplings, and the limitations of the analysis. We conclude in § 5.3.
The frozen model
The essential idea of the model is summarized in Figure 1. The two dark components do not exchange energy on average, so the expansion history is that of the noninteracting model; energy flows only between the local perturbations, from the vacuum-energy perturbation into the dark-matter overdensities or in the opposite direction, depending on the sign of the coupling. The construction follows the companion analysis [52], with the two refinements described in the footnotes below.
Figure 1. Schematic view of the frozen model. (a) In the background the dark matter dilutes as and the vacuum energy stays constant, with no exchange between them (); the expansion history is that of CDM and the geometric probes do not see the coupling. (b) In the perturbations, energy is exchanged between the dark-matter overdensities and the vacuum-energy perturbation at the rate of Eq. (3); the velocity field is unchanged, but the dark-matter density responds to the convergence of the flow with the factor of Eq. (11). For the energy flows into the overdensities. (c) The factor of Eq. (12) as a function of the scale factor for , , , and : negative couplings accelerate the growth at late times and approach the physical boundary (shaded region, ). The bottom line summarizes the observational consequences.
Background
We consider a spatially flat universe with photons, three neutrino species (one of them massive), baryons, cold dark matter (DM, subscript c) and a dark-energy fluid (subscript x) with constant equation of state . The two dark components exchange energy only at the level of the perturbations: the background exchange vanishes, , so that
where a prime is a derivative with respect to conformal time and . The expansion history is therefore that of the noninteracting model with the same ; in this work we take , for which it is the expansion history of CDM and all geometric probes are blind to the interaction [52]. Whereas Ref. [52] kept as a free parameter, fixing isolates the effect of the coupling on the growth of structure; the general case is discussed in Sec. IX. We denote by
the ratio of the two dark densities, which grows as for and is today.
Perturbations
We work in the synchronous gauge comoving with the DM, with metric perturbations and in the convention of Ref. [57]. The energy transfer to the DM is
where is the dimensionless coupling, the density contrast of the fluid, and the velocity divergence of the dark sector weighted by ,
The combination is the density contrast of the fluid in the frame of the total dark momentum, so that the exchange is gauge covariant5. The variable remains finite when , and so does , which tends to because the density ratio is finite and positive (the DM density never vanishes); this allows us to set exactly. The DM equations are
and, since the DM is initially at rest in this gauge, at all times: the exchange carries no momentum in the DM frame. The fluid, with rest-frame sound speed and adiabatic sound speed , obeys
We set . The baryons, photons and neutrinos follow the standard equations, and the Einstein equations are unchanged: the total density and momentum perturbations source the metric as usual.
a. Initial conditions. Deep in the radiation era, the adiabatic growing mode has with
and , with in that regime6. For and the fluid is unperturbed and the model reduces exactly to CDM.
Sub-horizon limit and physical boundary
On sub-horizon scales the fluid with does not cluster: its density contrast is suppressed by , and Eq. (7) becomes a constraint for its momentum, . With this gives
Inserting this into Eq. (5) and using that on these scales the Newtonian-gauge velocity divergence of the DM is , we find the continuity equation
while the baryons obey and both components fall in the same gravitational potential. The DM density therefore responds to the convergence of the flow with the factor , where we define the regularity parameter
For the exchange damps the growth of the DM density; for it enhances it, and when .
Since grows with time, the evolution is regular up to today only if , that is, . Following the convention of the companion analyses, we require () and treat the excluded region as a region of vanishing likelihood; we show in § 1.4 how the results depend on .
In this limit the DM mass is not conserved: the exchange converts dark energy into DM where matter converges () or the opposite (). The velocity field is unchanged, so redshift-space distortions, which measure , are much less sensitive to than the density field [52]; quantities that respond to the density, such as , the lensing potential, the integrated Sachs–Wolfe (ISW) effect and the abundance of halos and voids, carry the signal.
Phenomenology
Figure 2 illustrates the effect of at fixed values of the other parameters. Because at early times, the modification is confined to , where it changes by up to 30% for close to the boundary (panel b). The enhanced growth for raises the CMB lensing potential by 2–6% on the scales of the Planck lensing reconstruction (panel c), which also increases the lensing smoothing of the acoustic peaks. At the same time it slows the decay of the gravitational potentials at late times, and the late ISW contribution to the large-angle temperature anisotropy drops by up to 20% at (panel d); for both effects have the opposite sign.
Figure 2. Effect of the coupling at fixed , , , , and . (a) Factor that multiplies the DM velocity divergence in the continuity equation, Eq. (11). (b) Ratio of to its CDM value. (c) Ratio of the CMB lensing potential power spectrum. (d) Relative change of the lensed temperature power spectrum, dominated at low multipoles by the late ISW effect. For the model is at .
Data
We use four groups of observations. The CMB and the two geometric probes enter all the fits, cosmic shear enters the combinations labeled “shear,” and the cluster abundances are used only as a consistency check. Because the frozen model and CDM share the same expansion history, it is instructive to look at each data set together with the best fits of both models; the figures in this section show the data as they enter the likelihoods, so that the reader can see where the coupling acts and where it is invisible.
Planck 2018 (P18)
We use the Planck 2018 low- temperature likelihood (), the low- -mode polarization likelihood, the high- TT, TE and EE likelihood in its foreground-marginalized compressed form, and the CMB lensing reconstruction in its conservative range of lensing multipoles, [55, 58, 59]; as usual, we denote the multipoles of the lensing potential by and those of the CMB spectra by . The overall calibration is sampled with a Gaussian prior . This combination constrains through the three effects of Figure 2: the late ISW contribution at low multipoles, the lensing smoothing of the acoustic peaks, and the amplitude of the reconstructed lensing potential.
Figure 3 shows these data together with the best fits of CDM and of the frozen model. The low- points are the maximum-likelihood values of the Gibbs-sampled temperature spectrum [58], and the high- points are the foreground-marginalized bandpowers, to which the models are compared after the same binning that enters the likelihood. At the best fit of P18 the coupling lowers the temperature power by 24% at and by 7% at , the multipoles where the observed spectrum lies below the CDM prediction (panel a), while the high- spectra change by less than the size of the error bars (panel b) and the polarization spectra by less than 1 (panel c). The lensing bandpowers, on the other hand, are almost unaffected: the frozen best fit raises the lensing power by 6% in the first bandpower, centered at , which is also the bandpower that lies highest above CDM, and by less than 1% at (panel d).
Figure 3. Planck 2018 data used in this work, shown as differences with respect to the CDM best fit of P18 (horizontal line). The curves are the differences between the best fits of the frozen model to P18 (dashed blue), to P18+DES-Dovekie+DESI DR2 (dotted purple) and to all data (dash-dotted red) and that CDM best fit; they are relative differences in panels (a), (b) and (d) and absolute differences in panel (c). (a) Low- temperature spectrum: maximum-likelihood values of the individual multipoles with their 68% intervals (the upper intervals at and 5 extend beyond the panel). (b) High- TT bandpowers, rebinned by six. (c) TE and EE bandpowers, rebinned by eight; the curves show TE, since the EE differences are below . (d) Lensing-potential bandpowers in the range , with the horizontal bars marking the bin widths. The two fits without shear nearly coincide.
The contrast between the large change of the late ISW signal and the small change of the lensing potential is the key to the phenomenology of the model, and it is easiest to appreciate on the sky. Figure 4 shows Gaussian realizations of the late-ISW temperature map and of the lensing convergence for the best fits of CDM and of the frozen model to P18+DES-Dovekie+DESI DR2, generated with identical random phases so that the maps differ only through the power spectra. The rms of the late-ISW map drops from 25 to 15 , and the late-ISW power at is 0.28–0.40 times the CDM value, whereas the convergence maps, band-limited to the multipoles of the Planck reconstruction, differ by about 2% of their rms. The reconstruction therefore leaves room for a large change of the growth at , where its kernel is small, while the large-scale temperature is sensitive to it.

Figure 4. Simulated full-sky maps (Mollweide projection) for the best fits of CDM and of the frozen model to P18+DES-Dovekie+DESI DR2, drawn with identical random phases. (a, b) Late-ISW contribution to the temperature anisotropy, on a common color scale (). (c) Lensing convergence of CDM, band-limited to and smoothed with a Gaussian beam. (d) Difference between the convergence of the frozen model and that of CDM, multiplied by 50 to make it visible on the color scale of panel (c). These are simulations built from the model spectra, not the Planck maps.
DES-Dovekie
The Dark Energy Survey five-year supernova sample [60], recalibrated with the Dోవekie photometric calibration [61], contains 1820 type Ia supernovae: 1623 photometrically classified supernovae from the DES deep fields at and 197 spectroscopically confirmed supernovae from external low-redshift samples at (Table I). We use the distance moduli with their full statistical and systematic covariance and marginalize analytically over the absolute magnitude; the probability that each DES supernova is not of type Ia is included in its uncertainty. The data measure the shape of the luminosity distance, and therefore .
Figure 5 shows the Hubble diagram and the binned residuals with respect to the CDM best fit of P18+DES-Dovekie+DESI DR2. Because the frozen model has the same expansion history as CDM, the two models would
Figure 5. DES-Dovekie supernovae. (a) Hubble diagram of the DES (blue) and low-redshift (orange) supernovae, with the absolute magnitude fixed to its best-fit value for the CDM best fit of P18+DES-Dovekie+DESI DR2 (black line); points with uncertainties larger than one magnitude, which correspond to a low probability of being of type Ia, are not shown. (b) Inverse-variance weighted residuals in redshift bins (diagonal errors only), and the difference between other best fits and the reference, after marginalization over the absolute magnitude; the full covariance used in the fits, which includes the calibration systematics that correlate the low-redshift points, is not reflected in these error bars. Because the two models share the same background, all curves in panel (b) differ only through the parameters preferred by each data combination.
| Sample | – | median | |
| DES 5-year | 1623 | 0.073–1.144 | 0.519 |
| Foundation | 117 | 0.025–0.087 | 0.043 |
| CfA3K | 31 | 0.025–0.085 | 0.034 |
| CfA4p2 | 22 | 0.025–0.071 | 0.038 |
| CfA3S | 16 | 0.025–0.059 | 0.034 |
| CSP DR3 | 8 | 0.029–0.057 | 0.041 |
| CfA4p3 | 3 | 0.033–0.053 | 0.039 |
| Total | 1820 | 0.025–1.144 | 0.479 |
Table 1. Composition of the DES-Dovekie Hubble diagram: number of supernovae, redshift range and median redshift of each sample (redshifts corrected to the CMB frame and for peculiar velocities). Data from Refs. [60, 61].
give identical Hubble diagrams if fitted to the supernovae alone; the small difference we find stems only from the shift of the cosmological parameters induced by the other data sets in the fit, and it stays below 0.01 mag at all redshifts once the absolute magnitude is marginalized over.
DESI DR2
We use the DESI DR2 measurements of the transverse and line-of-sight BAO distances, and , and of the angle-averaged distance for the bright galaxy sample, for all tracers in , with their covariance [8]. Uncalibrated BAO constrain and . Figure 6 compares the measurements with the best fits of the two models. The curves of CDM and of the frozen model fitted to P18+DES-Dovekie+DESI DR2 are indistinguishable, as they must be for two models with the same background; the best fit of the frozen model to P18 alone, on the other hand, lies at the lower edge of the band allowed by P18 in CDM (gray), closer to the DESI points, which is the geometric counterpart of the reduction of the tension discussed in § 2.
Figure 6. DESI DR2 BAO distances relative to the CDM best fit of P18+DES-Dovekie+DESI DR2: (a) angle-averaged distance and (b) ratio of the transverse and line-of-sight distances, . The gray band is the 68% region of CDM constrained by P18 alone; the lines are the best fits of CDM and of the frozen model to P18+DES-Dovekie+DESI DR2 (they overlap) and of the frozen model to P18 alone. The largest residuals with respect to the CDM best fit of P18+DES-Dovekie+DESI DR2 are for the second LRG bin in and for the first LRG bin in ; neither model can change them, since the coupling does not enter the background.
Cosmic shear
We use the amplitude of the matter fluctuations measured by cosmic shear in the Kilo-Degree Survey (KiDS-Legacy), [62], and in the Dark Energy Survey Year 3, [63, 64], with . These values are derived under CDM growth, while in the frozen model the growth history at is different. We therefore compare them with the effective amplitude
with
where refers to CDM with the same parameters and , and is the redshift kernel of the shear power spectrum at . Because the growth modification is scale independent on the scales probed, the shear power of the frozen model is close to that of a CDM model with amplitude ; Appendix A derives this result, tests it against the published source distributions of both surveys and against a simulated CDM analysis of the DES Year 3 correlation functions, and shows how the tomographic signal of the two models differs. The two surveys are treated as independent split-normal likelihoods. Since at for , is smaller than : the shear kernels peak at –, where part of the late enhancement of the growth has not yet happened.
Table 2. Cluster samples used for comparison: number of clusters in the cosmological sample, redshift range, sky coverage, mass calibration and published amplitude.
Cluster abundances
As a check that does not enter the fits, we compare the predictions with three determinations of the amplitude from cluster counts: SPT clusters with weak-lensing mass calibration, [65]; eROSITA eRASS1, [66]; and the Planck PSZ2 sample, [67]. We use Eq. (13) with kernels that follow the redshift distribution of each sample.
Cluster counts depend on the halo mass function, which has been calibrated only for models in which the DM mass is conserved; Appendix B quantifies how well the effective amplitude describes the counts of the frozen model with a mass function of CDM form, and shows that the redshift evolution of the counts carries additional information.
The three samples are complementary, as Table II summarizes: SPT selects massive clusters through their Sunyaev–Zeldovich signal up to in the southern sky, eRASS1 selects X-ray clusters at over the western Galactic hemisphere, and the Planck sample covers most of the sky outside the Galactic plane with the highest signal-to-noise detections. The masses are calibrated with weak lensing in the first two and with X-ray and weak-lensing data in the third. We list these properties because the selection, the sky coverage and the mass calibration differ among the samples, and these differences underlie the spread of their amplitudes discussed in Sec. VIII.
Theory settings
We assume one massive neutrino with eV and , take the primordial helium fraction from big-bang nucleosynthesis, and use the revised halo-model fit of Ref. [68] for the nonlinear correction to the CMB lensing potential, a prescription calibrated on CDM simulations.
Statistical methodology
Comparing models whose posteriors are strongly non-Gaussian, and combining data sets whose agreement is itself in question, calls for some care in the statistical treatment. In this section we describe the parameters and priors, the sampling strategy, which uses exact likelihoods throughout, and the estimators with which we quantify the agreement between data sets and compare the frozen model with CDM.
Parameters and priors
The frozen model has eight free parameters, , and CDM the same seven without . The priors (Table III) are uniform, except for , and identical for every data set and for both models; this is necessary for the evidence-based estimators of Sec. IV C to be comparable. The prior range of is the domain of validity of the BBN table that sets the primordial helium abundance consistently with , which matters mainly for the recombination history probed by the CMB. Unlike many BAO analyses, which calibrate with a BBN prior on , we impose no prior from the observed light-element abundances, so is constrained by the data alone. Points beyond the physical boundary are assigned zero likelihood; the prior volume of is therefore the full interval [], and the boundary acts as part of the data model, as in a Markov chain that rejects these points.
Table III. Priors. All priors are uniform in the indicated range, except for the Planck calibration, which is Gaussian. They are identical for all data sets and, for the parameters in common, for CDM and the frozen model. Points with have zero likelihood.
Sampling
Each evaluation of the P18 likelihood requires the full solution of the Boltzmann hierarchy, about 6 s. We sample the posteriors that involve P18 by adaptive importance sampling with exact likelihoods, in the form of population Monte Carlo with deterministic-mixture weights [69, 70, 71]. In round we draw points from a mixture of multivariate Student- distributions with five degrees of freedom, and at every draw we evaluate all likelihoods at once: P18, DES-Dovekie, DESI DR2 and the shear and cluster amplitudes. After rounds, every draw carries, for each target , the weight
and the evidence is . Because all likelihoods are computed at every draw, a single set of draws serves all targets: P18, P18+DES-Dovekie, P18+DESI DR2, P18+DES-Dovekie+DESI DR2, P18+shear and P18+DES-Dovekie+DESI DR2+shear.
P18+DESI DR2, P18+DES-Dovekie+DESI DR2, P18+shear and P18+DES-Dovekie+DESI DR2+shear. The weights are exact for any choice of the , which affect only the efficiency, and the effective sample size and the Pareto-smoothing diagnostic [72] monitor their reliability.
The first round of each model draws from the Laplace approximations of the CDM posteriors of the four Planck-based combinations, with covariances inflated by ; for the frozen model the direction is added with the regression of the other parameters on measured in a preliminary Markov chain. Later rounds use one component per target, fitted to its weighted posterior and inflated by 1.15–1.35, plus a broad component with 10% of the weight. In the frozen model the posterior of is strongly skewed and limited by the physical boundary; in the variable , which maps the allowed region onto the real line, it is close to Gaussian, and from the third round on the proposal components are Student- distributions in the space where is replaced by , with the corresponding Jacobian in Eq. (15). After the first two rounds (1878 draws) the effective sample size of P18 was 155; the third round, the first one in , raised it to 648 with 1196 further draws. The last round of each model puts most of its weight on the combinations with cosmic shear. The sampling is summarized in Appendix D.
For DES-Dovekie and DESI DR2 alone the likelihoods depend only on , and on through the boundary, and we use nested sampling [73, 74] with 2000 live points and about effective samples. To make this affordable we compute the expansion history from the densities of photons and neutrinos tabulated with the same Boltzmann solver, and the drag-epoch sound horizon from a bicubic interpolation in ; the log-likelihoods agree with the full computation to (DES-Dovekie) and (DESI DR2).
The best fits (maxima of ) are found with a quadratic surrogate of fitted to all exact evaluations within of the minimum, in coordinates whitened with the posterior covariance (with in place of , so that every step stays inside the boundary) followed by an exact evaluation at the minimum of the surrogate within a trust radius, and iterated until the exact improves by less than 0.03, which takes one to five iterations. For DES-Dovekie and DESI DR2 alone we use a simplex search.
Tension estimators
We quantify the agreement between P18 (denoted ) and an external data set with the same battery of estimators in both models [75, 76, 77, 78, 79, 80, 81, 82, 83]. Each of them answers a slightly different question, and their strengths and limitations for non-Gaussian posteriors and for several data sets at once have been compared in detail in Refs. [4, 5]; using all of them side by side is the most transparent way to report a tension. Their definitions are collected in Appendix C:
the difference in means and the parameter-shift probability in the parameters constrained by : for DES-Dovekie, for DESI DR2, and for cosmic shear;
the update difference in means ;
the goodness-of-fit loss at the best fits, distributed as with degrees of freedom, where counts the parameters constrained beyond the prior;
the evidence ratio , the information ratio , the suspiciousness and the Bayesian model dimensionality , with the probability to exceed obtained from ;
the change of the deviance information criterion, .
Probabilities are converted to an equivalent number of Gaussian standard deviations with the two-tailed convention. For the frozen model the prior used to count is restricted to the physical region, so that the boundary is not counted as a constraint of the external data. For cosmic shear, which enters as a Gaussian constraint on a single amplitude, the shear-alone posterior is known analytically: is the minimum over a common amplitude of the two surveys, and , so , and follow without sampling; the evidence of the shear data alone is not defined independently of the prior on , and we do not quote for this pair.
Model comparison
We compare the frozen model with CDM with the Bayes factor for each data combination and with the difference of the best-fit . Since CDM is the point of the frozen model, includes the Occam penalty of the prior range of ; with the uniform prior adopted here, it changes by for a different prior width that contains the posterior.
Planck alone and the physical boundary
P18 alone constrains the coupling to , with the 95% interval (Table V and Figure 7 and Figure 8). The posterior is strongly asymmetric: it rises steeply toward negative , peaks at , close to the physical boundary, and has a long tail toward positive values. In terms of the regularity parameter, . The best fit sits at , that
Figure 7. Marginalized posteriors (68% and 95% regions) of , and in the frozen model for P18, P18+DESI DR2, P18+DES-Dovekie+DESI DR2 and P18+DES-Dovekie+DESI DR2+shear. Without cosmic shear the coupling is pushed toward the physical boundary and is close to unity; the shear data bring it back to .
Figure 8. (a) Marginalized posteriors of , normalized to their maxima. (b) Samples of the P18+DES-Dovekie+DESI DR2 posterior (purple) and of the combination with cosmic shear (red) in the plane of the regularity parameter and of the amplitude seen by the shear kernel; the band is the 68% interval of the combined KiDS-Legacy and DES Year 3 cosmic-shear measurement, , and the dashed line the physical boundary .
| Parameter | P18 | +Dov. | +DESI | +Dov.+DESI | +shear | +all |
| [Mpc] | ||||||
| (CDM) | ||||||
| (CDM) | ||||||
| (CDM) |
Table V. Marginalized constraints (mean and standard deviation) on the parameters of the frozen model for each data combination. The posterior of is asymmetric for the combinations without shear; its 95% intervals are given in the text. is the amplitude seen by the KiDS-Legacy shear kernel, Eq. (13). in km s Mpc. Columns: P18 alone and P18 combined with DES-Dovekie (Dov.), DESI DR2, both, cosmic shear, or all of them. The last three rows give the corresponding CDM values.
is and , deep in the region where the exchange enhances the late growth of the DM density, and it improves the fit to P18 by with respect to CDM. The amplitude of the fluctuations follows the coupling: the posterior of is and that of is , against in CDM, and the best fit has . The parameters shared with CDM move by less than half a standard deviation: and , against and in CDM.
a. Origin of the preference. Figure 10 decomposes the improvement of the best fit. All four Planck likelihoods contribute with the same sign: the high- spectra by , the low- temperature by , the low- polarization by and the lensing reconstruction by . These are the two effects of Figure 2. The enhanced late growth increases the lensing smoothing of the acoustic peaks, which the high- Planck 2018 spectra prefer to be larger than in CDM, the well-known lensing anomaly [55, 56]; and it suppresses the late ISW contribution at , where the observed temperature spectrum is low. The low- polarization improves slightly because the optical depth moves to lower values, against in CDM, once the amplitude of the peaks is no longer tied to the lensing smoothing. The lensing reconstruction itself is only marginally affected, because the growth enhancement acts at , where its kernel is small; this is why can reach values above unity without being excluded by the CMB lensing data. Since the lensing anomaly is weaker in the Planck PR4 likelihoods [84, 85], we expect the preference for to be weaker with those data.
Figure 10. Difference of the of each likelihood between the best fits of the frozen model and of ΛCDM, for the data combinations that enter each fit.
b. Dependence on the boundary. The posterior does not vanish at the boundary, so the lower limit on is set by the choice of rather than by the data. Raising the cut removes the region of largest growth: for P18 we find , and for , and , and the Bayes factor with respect to CDM drops from to , and (Table VI). The best fits at correspond to to : the DM density responds to the convergence of the flow ten times more efficiently than the baryons. This is a regime in which the linear description of the exchange and the nonlinear corrections calibrated on CDM cannot be taken for granted, and we regard the region as physically questionable even when it is formally allowed.
| Data | 0.1 | 0.2 | 0.5 | |||
| P18 | 1011.1 | 0.72 | 0.63 | 0.41 | ||
| +Dov. | 2643.4 | 0.52 | 0.45 | 0.27 | ||
| +DESI | 1026.9 | 1.64 | 1.48 | 1.15 | 0.01 | |
| +Dov.+DESI | 2661.9 | 1.49 | 1.33 | 1.02 | ||
| +shear | 1017.8 | |||||
| +all | 2666.0 | 0.0 |
Table VI. Model comparison. Best-fit of the data in CDM and in the frozen model, their difference, and the Bayes factor , whose Monte Carlo error is 0.03–0.05. The last three columns give when the physical cut is raised to , 0.2 and 0.5. Data: P18 alone and P18 combined with DES-Dovekie (Dov.), DESI DR2, cosmic shear, or all of them.
Agreement between the data sets
Before combining the data sets we test their agreement with the estimators of Sec. IV C. The results are collected in Table IV and summarized in Figure 12. Because the frozen model has the expansion history of CDM, the posteriors of DES-Dovekie and DESI DR2 alone are the same in the two models, apart from the volume of the prior, which they do not constrain; any change in the level of tension comes from how the coupling moves the Planck posterior in the parameters that the external data measure (Figure 11).
Figure 11. (a) Marginalized posteriors of for DES-Dovekie and DESI DR2 alone and for P18 in CDM and in the frozen model. (b) 68% and 95% regions in the plane measured by uncalibrated BAO, for DESI DR2 and for P18 in CDM (solid) and in the frozen model (dashed).
Figure 12. Tension estimators, in equivalent Gaussian standard deviations, for CDM (squares) and the frozen model (circles): goodness-of-fit loss , difference in means , parameter shift and suspiciousness . The dotted lines mark and .
a. P18 versus DES-Dovekie. The supernovae prefer a higher matter density, , than P18. In CDM the two data sets agree: all estimators give $0.9\sigma$, and . In the frozen model P18 moves to a slightly lower matter density, $\Omega_m =
| P18 vs DES-Dovekie | P18 vs DESI DR2 | P18+DES-Dovekie vs DESI DR2 | P18 vs shear | |||||
| ΛCDM | frozen | ΛCDM | frozen | ΛCDM | frozen | ΛCDM | frozen | |
| [] | 0.91 | 1.11 | 1.85 | 1.01 | 2.40 | 1.69 | 1.93 | 2.74 |
| [] | 0.91 | 1.13 | 1.91 | 1.46 | 2.46 | 2.12 | 1.92 | 2.60 |
| [] | 0.98 | 1.15 | 1.54 | 1.37 | 1.79 | 1.70 | – | – |
| shift [] | 0.89 | 1.13 | 1.88 | 1.37 | 2.50 | 2.03 | 1.94 | 2.51 |
| [] | 0.91 | 1.11 | 1.87 | 1.35 | 2.44 | 2.05 | 1.86 | 2.46 |
| 3.39 | 3.32 | 5.34 | 6.49 | 4.04 | 5.25 | – | – | |
| 0.08 | −0.17 | −1.71 | −0.51 | −3.04 | −1.82 | −1.23 | −3.02 | |
| 0.99 | 1.24 | 1.83 | 1.28 | 1.74 | 1.31 | 1.02 | 1.62 | |
| −1.2 | −0.9 | 1.4 | −0.5 | 4.0 | 2.2 | 1.2 | 4.6 |
Table 4. Agreement between the data sets in ΛCDM and in the frozen model. The first five rows give the equivalent number of Gaussian standard deviations of the goodness-of-fit loss , of the difference in means , of the update difference in means from P18 to the joint posterior , of the parameter shift and of the suspiciousness; the shift parameters are (DES-Dovekie), (DESI DR2) and (shear). The last rows give the log-evidence ratio, the log-suspiciousness, the Bayesian model dimensionality of the tension and the change of the deviance information criterion; , and indicate agreement.
0.3119±0.0075, and the estimators rise to 1.11–1.13; the agreement is still good, with .
b. P18 versus DESI DR2. DESI measures and Mpc. In ΛCDM the moderate discrepancy with P18 ( Mpc) is reproduced by all estimators: , , the shift and the suspiciousness give 1.85, 1.91, 1.88 and 1.87 standard deviations, and the deviance information criterion increases by 1.4 when the two are combined. The frozen model reduces it to 1.01–1.46, with : the coupling lowers and raises of P18 to Mpc along the direction of the DESI measurement. The reduction is largest in , because the best fit of P18+DESI DR2 also improves the lensing likelihood (−1.24) and the DESI (−0.70) at the same time, while the estimators based on parameter differences decrease by about 0.5.
c. P18+DES-Dovekie versus DESI DR2. Adding the supernovae to P18 widens the gap with DESI, as in the companion analyses: in ΛCDM the four estimators give 2.40–2.50 and . The frozen model reduces this to 1.69–2.12 and . In both models the three data sets can be combined, but in ΛCDM the combination is a compromise between discordant measurements of the matter density.
d. P18 versus cosmic shear. The effect of the coupling on the amplitude of the fluctuations goes in the opposite direction. In ΛCDM, P18 predicts against for the combined shear data, a 1.92 difference, and the other estimators agree (1.86–1.94). In the frozen model, P18 alone predicts , and the tension rises to 2.46–2.74, with . The same coupling that relieves the disagreement with DESI therefore increases the one with cosmic shear.
e. Consistency of the estimators. The estimators agree with each other to 0.2 in ΛCDM for every pair. In the frozen model the spread is larger, up to 0.45 for P18 versus DESI DR2, because the posteriors are not Gaussian and uses the best fit at the edge of the allowed region. The Bayesian model dimensionality of the DES-Dovekie and DESI DR2 comparisons is and 1.28 in the frozen model, close to the number of parameters that these data constrain, and the suspiciousness agrees with the parameter-based estimators. The evidence ratio is positive in all cases, –6.49, but it grows with the prior volume of the parameters. that the external data do not constrain, which is why the suspiciousness is the relevant evidence-based estimator. None of the pairs exceeds in either model, so all combinations are allowed.
Joint constraints
Planck, DES-Dovekie and DESI
Since none of the comparisons of § 2 exceeds , we combine the three data sets. For P18+DES-Dovekie+DESI DR2 the coupling is ,
with the 95% interval , which excludes (Table V). The geometric data fix the matter density and the expansion rate, and , which coincide with
CDM ( and km s Mpc), as they must for a model with the same background. Figure 9 shows how the four combinations without shear arrange themselves along the degeneracy direction of the CMB, on which is nearly constant. The supernovae prefer a higher matter density and a lower expansion rate, km s Mpc in the frozen model, and DESI the opposite, km s Mpc; the combination of the three settles on an intermediate point that is the same in both models. None of these values approaches the local distance-ladder measurements, a point to which we return in § 4. The coupling is again pushed toward the boundary, , because DESI selects the region of lower in which P18 prefers more lensing, and the amplitude of the fluctuations becomes large: and , against in CDM.

Figure 9. Marginalized posteriors of (in km s Mpc) and in the frozen model for P18, P18+DES-Dovekie, P18+DESI DR2 and P18+DES-Dovekie+DESI DR2, compared with CDM for P18+DES-Dovekie+DESI DR2 (dashed). All posteriors lie along the CMB degeneracy direction; the supernovae pull toward the upper left and DESI toward the lower right of this line, and the frozen model moves the P18 posterior slightly toward DESI.
The best fit improves the by with respect to CDM, distributed over all the CMB likelihoods (high- , low- TT , low- EE , lensing ) and DESI (), with a small cost for the supernovae (). The Bayes factor is (Table VI), a weak preference on the Jeffreys scale once the Occam penalty of the prior range of is included. This preference rests on the region close to the boundary: with it drops to and with it vanishes, , while the coupling moves to .
Adding cosmic shear
The large amplitude favored by P18+DES-Dovekie+DESI DR2 is not what the shear surveys see. The amplitude relevant for the shear kernels, , is smaller than because the growth enhancement is concentrated at
(Figure 8b), but it is still above the combined shear measurement, . At the best fit of P18+DES-Dovekie+DESI DR2 the two shear likelihoods would add (KiDS-Legacy) and 31.69 (DES Year 3) with respect to CDM.
With cosmic shear included, the coupling is consistent with zero: for P18+shear and for the full combination, with the 95% interval , and . The region near the boundary is excluded, and these constraints no longer depend on . The best fit of the full combination has and the same as CDM (), and the Bayes factor, , now favors CDM: the data constrain the coupling but give no reason for it. The reduction of the tension with DESI found in § 2 therefore cannot be kept once the shear amplitude is taken into account, since both effects come from the same enhancement of the late growth.
Late-time observables
The constraints of the previous sections determine, for each data combination, a prediction for observables that were not used in the fits. Comparing these predictions with independent measurements is a useful complement to the tension estimators: it shows directly which features of the data each model can or cannot reproduce, and it identifies the observables that could decide between them in the future. In the figures of this section the predictions are shown as boxes that span the 68% credible interval, with whiskers at 95% and a line at the median, for the six data combinations and for both models; the measurements are shown with their 68% errors.
Expansion history
Because the two models share the same background, their predictions for the expansion rate differ only through the values of , and preferred by each combination. Figure 13 shows the posterior of , the combination measured by the radial BAO, scaled by for clarity. The bands of the two models are nearly identical: for P18 the frozen model is shifted upward by 0.2–0.4% at , a consequence of its slightly lower matter density, and for P18+DES-Dovekie+DESI DR2 the two models coincide to within 0.2%. The DESI points at and 0.71 lie about above both predictions, while the others agree well, in line with the moderate tension discussed in § 2. Beyond these small parameter shifts, a model that changes only the growth cannot modify the expansion history, and the frozen model leaves untouched the features that motivate a dynamical dark energy in the DESI data.
!Posterior of H(z)r_d/[100(1 + z)] for P18 and P18+DES-Dovekie+DESI DR2 in ΛCDM and in the frozen model, compared with the DESI DR2 measurements of D_H/r_d bbox=0.60,0.08,0.91,0.56
Figure 13. Posterior of (68% and 95% bands) for P18 and P18+DES-Dovekie+DESI DR2 in ΛCDM (top) and in the frozen model (bottom), compared with the DESI DR2 measurements of (points).
Hubble constant
The same reasoning applies to the Hubble constant, shown in Figure 14 together with a selection of recent measurements. Both models predict – for every data combination, and the frozen model moves the P18 value by only with respect to CDM. The distance-ladder measurements of the SH0ES team, [86] and with the additional anchor of the Small Magellanic Cloud [87], therefore remain and – above the prediction of P18+DES-Dovekie+DESI DR2 in both models. The tip-of-the-red-giant-branch and J-region asymptotic giant branch calibrations of the Chicago-Carnegie program with the James Webb Space Telescope, and [88], and the strong-lensing time delays of TDCOSMO, [89], are compatible with both models within . Among the CMB-based determinations, SPT-3G alone, , and its combination with Planck and ACT, [90], agree with our P18 predictions, while the combination of ACT, Planck and CMB lensing with DESI DR1, [91], agrees with our P18+DES-Dovekie+DESI DR2 prediction, illustrating once more that BAO pull the CMB inference of upward by about in CDM-like backgrounds. The small effect of the coupling on is consistent with the way it reduces the tension with DESI: the sound horizon depends only on and , which the coupling leaves untouched, and the shift of the P18 posterior toward DESI, from to , is a small displacement along the CMB degeneracy direction, with a slightly larger and a slightly lower matter density (Figure 9). The frozen model was not designed to address the Hubble tension, and it does not; any model that leaves the expansion history and the sound horizon untouched inherits the CDM value of from the CMB, whatever it does to the growth of structure. Its interest lies in the growth sector, to which we now turn after a brief look at the age of the Universe.
Figure 14. Hubble constant. Top: measurements from the distance ladder [86, 87], from the Chicago-Carnegie program [88], from strong-lensing time delays [89], and from recent CMB analyses in CDM [90, 91], with their 68% errors. Bottom: posteriors of CDM (gray) and of the frozen model (colored by data combination, as in Figure 8) for the six data combinations; boxes enclose the 68% intervals, whiskers the 95% intervals, and the vertical line marks the median.
Age of the Universe
The age of the Universe is another quantity fixed by the background, and its comparison with the oldest objects is a classic consistency test of the cosmological model. Figure 15 and Table VII summarize the results. All models and data combinations give Gyr with an uncertainty of 0.02 Gyr, between 13.76 and 13.81 Gyr; adding DESI lowers the age by 0.031 Gyr through the larger , and the frozen model and CDM differ by less than 0.01 Gyr for the same data. The ages of the Universe inferred from the oldest globular clusters, Gyr [92], Gyr with Gaia distances [93] and Gyr with multiple stellar populations [94] (statistical and systematic errors added in quadrature), agree with the prediction to better than . The ages of individual objects provide lower bounds: the globular cluster M92, Gyr [95], and the white-dwarf cooling sequence of NGC 6397, Gyr [96], are comfortably compatible, while the metal-poor subgiant HD 140283, Gyr [97], lies 0.8–0.9 above the prediction, a mild excess well within its uncertainty. The table also shows that the frozen model leaves the linear structure at essentially untouched, because the exchange becomes effective only at late times; any change in the abundance of the first halos is due to the shift of the parameters shared with CDM.
| Model | Data | [Gyr] | [Gyr] | [Myr] | ||||||||
| CDM | P18 | 67.34 | 0.315 | 0.810 | 0.831 | 13.799 | 13.80 0.02 | 0.950 | 470 | 1.000 | 1.00 | 1.00 |
| CDM | P18+DES-Dovekie | 67.13 | 0.318 | 0.811 | 0.835 | 13.806 | 13.81 0.02 | 0.948 | 470 | 1.001 | 1.02 | 1.03 |
| CDM | P18+DESI DR2 | 68.41 | 0.301 | 0.807 | 0.809 | 13.763 | 13.76 0.02 | 0.963 | 474 | 0.994 | 0.92 | 0.85 |
| CDM | P18+DES-Dovekie+DESI DR2 | 68.28 | 0.303 | 0.808 | 0.811 | 13.768 | 13.77 0.02 | 0.961 | 474 | 0.995 | 0.93 | 0.87 |
| CDM | P18+shear | 67.93 | 0.307 | 0.805 | 0.815 | 13.780 | 13.78 0.02 | 0.957 | 472 | 0.992 | 0.90 | 0.83 |
| CDM | all | 68.37 | 0.301 | 0.805 | 0.807 | 13.765 | 13.77 0.02 | 0.962 | 474 | 0.991 | 0.89 | 0.80 |
| frozen | P18 | 67.79 | 0.309 | 1.097 | 1.113 | 13.784 | 13.79 0.02 | 0.956 | 472 | 0.991 | 0.89 | 0.81 |
| frozen | P18+DES-Dovekie | 67.47 | 0.314 | 1.080 | 1.104 | 13.795 | 13.80 0.02 | 0.952 | 471 | 0.993 | 0.92 | 0.86 |
| frozen | P18+DESI DR2 | 68.53 | 0.299 | 1.115 | 1.114 | 13.759 | 13.76 0.02 | 0.964 | 474 | 0.987 | 0.83 | 0.71 |
| frozen | P18+DES-Dovekie+DESI DR2 | 68.39 | 0.301 | 1.112 | 1.114 | 13.764 | 13.77 0.02 | 0.963 | 474 | 0.987 | 0.84 | 0.73 |
| frozen | P18+shear | 67.70 | 0.310 | 0.787 | 0.800 | 13.788 | 13.79 0.02 | 0.955 | 472 | 0.996 | 0.94 | 0.90 |
| frozen | all | 68.39 | 0.301 | 0.812 | 0.813 | 13.764 | 13.77 0.02 | 0.963 | 474 | 0.991 | 0.88 | 0.79 |
Table VII. Best fits of CDM and of the frozen model: Hubble constant (in km s Mpc), matter density, amplitude of the fluctuations, age of the Universe , the dimensionless product , and linear structure at high redshift: the age at , the rms linear fluctuation of the total matter at in spheres containing , relative to the CDM best fit of P18, and the factors and by which the abundance of collapsed halos of and at changes in the Press–Schechter approximation, with the linear collapse threshold of CDM. The best fits illustrate how the coupling changes these derived quantities; the marginalized constraints on the parameters are given in Table V, the posterior of is shown in Figure 15, and the column gives its posterior mean and standard deviation.
Figure 15. Age of the Universe. Boxes enclose the 68% intervals, whiskers the 95% intervals, and the vertical line marks the median. (a) Ages of the Universe inferred from globular clusters [92, 93, 94] and ages of the cluster M92 [95], of the subgiant HD 140283 [97] and of the white dwarfs of NGC 6397 [96], which give lower bounds (gray; statistical and systematic errors added in quadrature), compared with the posteriors of CDM and of the frozen model for P18+DES-Dovekie+DESI DR2. (b) Posteriors for P18 (open boxes), P18+DES-Dovekie (filled), P18+DES-Dovekie+DESI DR2 (hatched) and the full combination with shear (crosshatched). The gray band is the 68% interval of CDM for P18+DES-Dovekie+DESI DR2.
Integrated Sachs–Wolfe effect
The enhanced growth for slows the decay of the gravitational potentials, and the ISW signal of the large-scale structure is reduced. Figure 16 compares the posterior predictive of the bias-independent ISW amplitude, Eq. (21), at and , with the amplitudes measured from cross-correlating the CMB temperature with tracers of the large-scale structure. These measurements are normalized to the CDM prediction and calibrate the galaxy bias with the autocorrelation of the tracers, so to a good approximation they measure the same ratio of decay rates [98, 99]. For P18+DESI-Dovekie+DESI DR2 the frozen model predicts at and at , and a sign change at for the values of closest to the boundary; with the shear data the prediction is , indistinguishable from CDM, while P18+shear alone, which allows slightly positive couplings, predicts .
Figure 16. ISW amplitude relative to CDM. Top: measurements from the cross-correlation of the CMB temperature with tracers of the large-scale structure [100, 101, 102, 103, 104, 105, 106], with their 68% errors. Bottom: posterior predictive of the bias-independent amplitude, Eq. (21), in the frozen model at and for the six data combinations (boxes: 68%; whiskers: 95%). The vertical gray line is CDM, for which the amplitude is unity by construction. Without shear data the frozen model predicts about half of the CDM signal at ; with shear data the predictions return to unity. The measurements share tracers and sky areas and are therefore correlated with each other.
The measurements are remarkably consistent with CDM: the combination of all tracers in the Planck 2015 analysis gives [100], the unWISE galaxies [101], the DESI Legacy Imaging Survey [102], the RACS radio galaxies [103] and the Quaia quasars [104], while earlier combined analyses found somewhat larger values, [105] and [106]. Taken one at a time, they lie 1.0–1.6 (the recent single-tracer measurements and Planck) and 2.4–2.8 (the earlier combinations) above the prediction of P18+DES-Dovekie+DESI DR2 in the frozen model, and the Quaia measurement lies 1.7 above the prediction at . Since the tracers overlap, these numbers cannot be combined naively, but they consistently point in the same direction as cosmic shear: the strong-coupling region favored by the CMB and BAO alone predicts too little ISW signal. The excess ISW imprint reported for the largest supervoids, [107], cannot be explained by the model either: negative values of reduce the imprint, and the positive values allowed by the shear data raise it by at most about 40%. A tomographic measurement of the bias-independent ratio with a precision of 0.2, within the reach of the combination of CMB maps with DESI and Euclid samples, would test the region of strong coupling independently of the shear amplitude and of the modeling of nonlinear scales.
and in real and harmonic space [109, 110], KiDS-1000, [111], the joint analysis of DES Year 3 and KiDS-1000, [112], DECADE, [113], and UNIONS, [114]. For P18+DES-Dovekie+DESI DR2, CDM predicts , within about of most surveys and 2.0–2.1 above DES Year 3 and KiDS-1000. The frozen model predicts an effective amplitude of for the KiDS kernel and for the DES kernel, which lies 1.8 above KiDS-Legacy, 2.1–2.6 above DECADE, HSC, DES Year 6 and the DES–KiDS combination, and about above DES Year 3 and KiDS-1000. The DES Year 6 result, which was not used in our fits, therefore confirms the conclusion of § 3. When the shear data are included, the frozen model returns to and agrees with all surveys as well as CDM does.
Amplitude of the fluctuations
Figure 17 gathers the measurements of from cosmic shear and from cluster abundances and compares them with the predictions. For the frozen model we show both the true of the model and the effective amplitude seen by a shear survey; the difference between the two, about 0.1 for the combinations without shear, is the growth enhancement at that the shear kernels only partly see. The cosmic-shear surveys are in good mutual agreement, with central values between 0.76 and 0.83: KiDS-Legacy, [62], DES Year 3, [63], and its successor DES Year 6, [108], HSC Year 3,
Figure 17. Amplitude of the matter fluctuations. Top: measurements from cosmic shear [62, 63, 108, 109, 110, 111, 112, 113, 114] and from cluster abundances [65, 66, 67], with their 68% errors. Bottom: posteriors of in CDM (gray), of in the frozen model (open boxes) and of the effective amplitude for the KiDS-Legacy kernel (filled boxes), for the six data combinations (boxes: 68%; whiskers: 95%); the colors of the frozen boxes identify the data combination, as in Figure 8. The open boxes show how large the present amplitude of the frozen model becomes without shear data, and the filled ones what a shear survey analyzed under CDM would measure; only the latter is to be compared with the shear measurements.
Cluster abundances
The three cluster samples, which cover different parts of the sky and different redshift ranges (Sec. III), do not give a coherent verdict (Figure 17). For P18+DES-Dovekie+DESI DR2 in CDM the predicted amplitude is from SPT, from eRASS1 and from PSZ2. The frozen model, which raises the amplitude at the low redshifts of the cluster samples, moves the prediction to (SPT kernel), (eRASS1) and (PSZ2): it removes the tension with eRASS1 () and increases those with SPT () and PSZ2 (). After the shear data are included the predictions return to the CDM values. Given the spread among the cluster samples themselves, which is under active study by the cluster community [65, 66, 67], and the fact that the halo mass function of a model in which the DM mass is not conserved has not yet been calibrated, we do not include the clusters in the fits. We also note that, within CDM, the prediction of P18+DES-Dovekie+DESI DR2 differs from eRASS1 by , a discrepancy comparable to that of the frozen model with PSZ2, so the cluster samples alone cannot currently discriminate between the two
models; a joint analysis of the samples with a common mass calibration would be very valuable in this respect. Their redshift distribution would be a sharper test (Appendix B).
Discussion
The results of the previous sections can be summarized in a simple way: the data that measure the geometry do not see the coupling, the CMB welcomes a moderate enhancement of the late growth, and the probes of the present amplitude of the fluctuations do not. In this section we place the model in the broader context of perturbation-only couplings, discuss which observables could break the remaining degeneracies, and list the limitations of our analysis.
Redshift-space distortions and other perturbation-only couplings
Redshift-space distortions measure the divergence of the velocity field, which the exchange does not modify directly (Sec. II C). At fixed primordial amplitude and , the value of built from the velocity divergence changes by +2.0% for and by -0.6% for , three times less than (+6.9% and -1.8%), and well below the precision of individual RSD measurements. The region of strong coupling favored by P18+DES-Dovekie+DESI DR2 is therefore only mildly constrained by current RSD data, which were not included here. The relevant combination is , with obtained from the velocity divergence, and not , built from the logarithmic growth rate of the density: in CDM the two coincide, but in the frozen model they differ precisely because the exchange changes the density at fixed velocity, so a comparison of with RSD data would overestimate their sensitivity to . On the scales used by RSD analyses the growth of the model is scale independent to good accuracy (Appendix A), so no additional dependence on arises. This is the main difference with the momentum-transfer models [41, 43, 44, 115], which slow the infall of the DM, suppress the growth of the velocity and the density alike, and are typically constrained by RSD and favored by the low shear amplitude. In the frozen model the two observables decouple: the coupling acts on the density at fixed velocity, and its sign is fixed by the data in the opposite way, , by the CMB. The same decoupling also distinguishes the model from interactions that exchange energy in the background, whose signatures in the expansion history are the main target of the analyses of Refs. [22, 23, 27, 29, 30, 32]: here the geometric data can only fix the parameters shared with CDM, and the growth observables carry all the information on the coupling.
Breaking the degeneracies
Three observables emerge from our analysis as the most promising tests of the region of strong coupling. The first is the ISW effect discussed in § 4: the bias-independent ratio of the ISW–galaxy and lensing–galaxy cross-spectra measures the decay rate of the potentials directly, and the frozen model predicts a reduction by a factor of about two at for the couplings favored without shear. The second is the tomographic structure of cosmic shear (Appendix A): the model enhances the signal of the lowest source bins relative to the highest ones in a way that no rescaling of the CDM amplitude can reproduce. The third is the redshift distribution of cluster counts (Appendix B), which the model steepens at low redshift. All three are within the reach of the combination of the next CMB experiments with the Euclid, Rubin and DESI samples, and two of them do not depend on the absolute calibration of the amplitude of the fluctuations, which makes them complementary to the comparison that currently limits the model. A further handle is the cross-correlation of CMB lensing with galaxy samples at , which isolates the redshift range where the growth departs from CDM and is free of the calibration uncertainties of cosmic shear; and the abundance and density profiles of cosmic voids, which respond to the growth of underdense regions, offer a complementary view of the same physics.
Limitations
Five aspects of the analysis deserve further work. (i) The cosmic-shear data enter as compressed amplitudes, corrected for the growth history of the model through an effective kernel. Appendix A shows that a CDM analysis of the DES Year 3 correlation functions of the frozen model recovers this effective amplitude to better than 0.5%, but a full analysis would require the nonlinear power spectrum of the model and the response of baryonic feedback and intrinsic alignments to the modified growth. (ii) The nonlinear correction to the CMB lensing potential and to the shear spectra uses a fit calibrated on CDM, which may not describe the model near the boundary, where and the nonlinear evolution with a varying DM mass is unknown. (iii) We used the Planck 2018 likelihoods. The preference for is driven by the lensing smoothing of the high- spectra and by the low- temperature deficit, and the former is weaker in the PR4 analyses [84, 85]; with those likelihoods we expect a weaker preference and a weaker tension with shear. (iv) We fixed . For the background changes, the geometric probes constrain directly, and the boundary moves to . (v) The cluster comparison relies on a halo mass function calibrated for a conserved DM mass; a dedicated study of collapse with energy exchange is needed before cluster counts and void abundances can test the model on their own.
Conclusions
We have confronted the frozen model, in which dark matter and dark energy exchange energy only at the level of the perturbations, with Planck 2018, DESI DR2, the DES five-year supernovae in their Dovekie recalibration and the cosmic-shear amplitudes of KiDS-Legacy and DES Year 3. On sub-horizon scales the exchange changes the efficiency with which the dark-matter density responds to the convergence of the flow by the factor , with , while it leaves the expansion history and the velocity field untouched. The evolution is regular only for , a physical boundary that plays a central role in the results. All posteriors and evidences were computed with exact likelihoods, and the agreement between the data sets was assessed with the same battery of estimators in CDM and in the frozen model. Our main findings are the following.
Planck 2018 alone prefers a negative coupling, , with a posterior that rises toward the boundary and a best fit at that improves the by . The preference is driven by the lensing smoothing of the high- spectra and by the low- temperature deficit, which the enhanced late growth reproduces at the price of ; the CMB lensing reconstruction is barely affected.
Since the background is that of CDM, the coupling changes the agreement between data sets only through the Planck posterior. It reduces the tension between P18 and DESI DR2 from – to –, and that between P18+DES-Dovekie and DESI DR2 from – to –; the agreement with DES-Dovekie remains good (–).
The joint fit of P18, DES-Dovekie and DESI DR2 gives , with outside the 95% interval, and . This weak preference rests on the region next to the boundary: it disappears when the cut is raised to ().
The same enhancement of the growth raises the amplitude seen by cosmic shear, and the tension between P18 and the shear data grows from in CDM to –. With the shear data included the coupling is consistent with zero, , the best fit coincides with CDM and favors CDM. The relief of the BAO tension and the shear amplitude cannot be reconciled within the model.
The late-time observables that depend only on the background are those of CDM: for every data combination the expansion rate, the Hubble constant and the age of the Universe agree between the two models within a fraction of their errors, so
the model neither eases nor worsens the discrepancy with the local distance ladder, and the age remains compatible with the oldest stars and globular clusters.
The region favored without shear predicts a reduced ISW signal, at for the bias-independent ratio of the ISW–galaxy and lensing–galaxy cross-spectra, and a sign change at very low redshift. With the published source distributions of DES Year 3 and KiDS-Legacy, the model also enhances the shear signal of the lowest tomographic bins relative to the highest ones (Appendix A), and a simulated CDM analysis of the DES Year 3 correlation functions confirms that the effective amplitude we use is what such an analysis measures. Cluster counts give contradictory indications at the level of the amplitude, but their redshift distribution is steeper in the frozen model than in any CDM model with the same effective amplitude.
The natural next steps are the analysis with the Planck PR4 likelihoods, in which the lensing anomaly that drives the preference for is weaker, together with the ACT DR6 and SPT-3G spectra [90, 91]; the extension to , where the background becomes informative; a full cosmic-shear likelihood with the nonlinear evolution of the model and its tomographic signature; a halo mass function for a dark matter whose mass is not conserved, which would turn the redshift distribution of cluster counts into a sharp test; and the covariant formulation of the exchange, which will tell whether the region close to the boundary, where the data pull the model, is physically acceptable. We hope that the tools assembled here, from the uniform treatment of tensions to the validation of compressed growth probes, will also be useful for other couplings that leave the expansion history untouched. The coming years will bring CMB lensing maps of much higher fidelity, the full DESI spectroscopic samples and the first cosmological analyses of Euclid and of the Rubin Observatory, which together will measure the growth of structure in the very redshift range where the frozen model departs from CDM; they offer an excellent opportunity to settle the question.
Acknowledgments
M.G.R. is partially supported by UFABC and by WFSA. L.F.G. thanks Fundação Araucária for financial support. S.J.L. and M.L. are supported by grant PIP 11220250100565CO CONICET and grant 20020170100129BA UBACYT. L.F.G. thanks João Victor Silva Rebouças for useful discussions about the DES-Dovekie data. We acknowledge the use of the CHE cluster, managed by the COSMO/CBPF/MCTI, with financial support from FINEP and FAPERJ, and operating at the RSDC–Datacenter para Ciência Alfredo Marques de Oliveira/CBPF.
Data Availability
Some of the data that support the findings of this article are openly available [116, 117, 118, 119, 120]. The Planck 2018 temperature, polarization and lensing likelihoods [55, 58, 59] are distributed by the Planck Legacy Archive at https://pla.esac.esa.int [116]. The distance moduli of the DES five-year supernova sample [60] in the Dovekie recalibration [61], together with their covariance, are available at https://github.com/des-science/DES-SNSYR [117]. The DESI DR2 BAO measurements and their covariance [8] are available through the DESI public data releases at https://data.desi.lbl.gov/doc/releases/ [118]. The source redshift distributions of KiDS-Legacy [62] are available at https://kids.strw.leidenuniv.nl/DR5/legacy_wl.php [119], and those of DES Year 3, together with its cosmic-shear correlation functions and their covariance [63, 64], at https://des.ncsa.illinois.edu/releases/y3a2 [120]. The cosmic-shear and cluster-count amplitudes, Hubble-constant measurements, stellar ages and ISW amplitudes used in Secs. III and § 4 are quoted from the cited publications. The posterior samples, best-fit parameters, tension estimates and numerical codes produced in this work cannot be made publicly available at present because they are being used in ongoing work by the authors. The data are available upon reasonable request from the authors.
Cosmic shear in CDM and in the frozen model
This appendix explains how the shear data enter our analysis, compares the shear signal of the two models with the published source distributions of DES Year 3 and KiDS-Legacy, and quantifies how well the effective amplitude of Eq. (13) summarizes the frozen model for a survey analyzed under CDM.
Tomographic shear spectra
In the Limber approximation, the cross-spectrum of the shear (or convergence) between tomographic bins and is
where is the comoving distance, the nonlinear matter power spectrum, and the lensing efficiency of bin , with normalized source distribution , is
The real-space correlation functions measured by DES are the Hankel transforms
with for and for . In CDM the amplitude of these spectra scales approximately as – on the scales used by current surveys, which is why is the best-measured combination.
The frozen model changes Eq. (16) only through , since the distances and the lensing efficiencies depend on the background, which is that of CDM. On linear scales the change is a rescaling of the amplitude at each redshift,
because the growth modification is scale independent on the relevant scales: the ratio of the density growth rate of the frozen model to that of CDM is constant to better than 0.2% for at , even close to the boundary (). On larger scales, , the dark-energy perturbations with introduce a scale dependence of up to 3% at for , which is irrelevant for the shear and cluster amplitudes. On nonlinear scales we apply the same halo-model correction as for the CMB lensing potential to the linear spectrum of each model. The ratio in Eq. (19) is shown in Figure 18b for the best fits: it is close to unity at and grows steeply at low redshift, reaching 1.36 and 1.39 today at the best fits of P18 and P18+DES-Dovekie+DESI DR2, respectively.
Figure 18. (a) Published source distributions of the four DES Year 3 bins (solid blue) and of the six KiDS-Legacy bins (dashed orange), each normalized to its maximum, together with the redshift kernels of the shear power at of the two surveys (thick lines). (b) Ratio of in the frozen model and in CDM with the same parameters, at the best fits of three data combinations. (c) Auto-spectra of the four DES Year 3 bins of the frozen best fit to P18+DES-Dovekie+DESI DR2, divided by those of a CDM model with the same parameters and the amplitude rescaled to : the low-redshift bins are enhanced and the high-redshift bins suppressed relative to CDM, a tomographic signature of the late growth of the frozen model.
Effective amplitude
Since the shear signal of a survey is dominated by the redshifts where its kernel peaks, a model whose amplitude evolves differently from CDM is seen by a CDM analysis as an effective . We define it through the redshift kernel of the convergence power at , near the middle of the scales that carry most of the signal,
which leads to Eq. (13). In the fits we tabulate with the Boltzmann solver on a grid of 30 values of and five values of , which is sufficient because at fixed background the late-time growth depends on alone [the combination of Eq. (12)], and interpolate it with bicubic splines; the effective amplitudes agree with a direct computation to 0.2% over the posterior, an order of magnitude below the errors of the shear measurements. For the kernels
we approximated the source distributions by Gaussian tomographic bins at the published mean redshifts. We tested this approximation with the published source distributions, shown in Figure 18a. Their kernels have mean redshifts 0.30 (DES Year 3) and 0.35 (KiDS-Legacy), and at the best fit of P18+DES-Dovekie+DESI DR2 they give and 0.896, against 0.909 and 0.895 with the Gaussian bins, for a model with . The two kernels thus agree to 0.5%, at the most extreme point of the posterior.
What a CDM analysis would infer
A more demanding test is to ask what a ΛCDM analysis of the shear data would infer if the frozen model were true. We computed the DES Year 3 correlation functions of the frozen best fit of P18+DES-Dovekie+DESI DR2 for all ten bin pairs, kept the 220 data points with for and for , and fitted them with ΛCDM spectra that share all the cosmological parameters of the frozen model except the amplitude, using the published covariance of the survey. The fit gives , in excellent agreement with and very different from the true of the model; for the best fit of P18 alone the analogous numbers are , 0.917 and 1.113. The effective amplitude is therefore what a ΛCDM analysis measures, and our comparison of the frozen model with the published values is consistent.
The fit is not perfect, and the residual is instructive. Relative to the rescaled ΛCDM model, the frozen model enhances the auto-spectrum of the lowest DES bin by 14% and suppresses that of the highest one by 3% at (Figure 18c); for KiDS-Legacy the lowest bin is enhanced by 18% and the highest suppressed by 4%. This differential growth across the bins is a genuine signature of the model. With the DES Year 3 covariance it costs only for 220 data points, and it would in practice be partly absorbed by the photometric-redshift and intrinsic-alignment parameters of a full analysis. Since grows in proportion to the inverse covariance, a survey with errors several times smaller, such as Euclid or the Rubin Observatory LSST, could detect it for the couplings favored by P18+DESI DR2 if the nuisance parameters are well controlled.
Figure 19. DES Year 3 measurements of for the four auto-correlations of the source bins, with the predictions of the ΛCDM best fit of P18+DES-Dovekie+DESI DR2 (solid), the frozen best fit of the same combination (dashed) and the frozen best fit of all data (dash-dotted, overlapping with ΛCDM). The predictions do not include intrinsic alignments, baryonic effects or survey nuisance parameters; the shaded region marks small scales that are most affected by them. The steep rise of the growth enhancement at .
Figure 19 shows the DES Year 3 measurements of for the four auto-correlations together with the predictions of the best fits. The predictions do not include intrinsic alignments, baryonic feedback or the nuisance parameters of the survey, so the figure is meant only to illustrate the size of the effect: the frozen best fit of P18+DES-Dovekie+DESI DR2 lies above the data in all bins, while the best fit of the full combination is indistinguishable from ΛCDM. The excess of the frozen best fit is largest in the lowest-redshift bin, about 50% in at , and decreases to about 20% in the highest one, in line with the tomographic pattern of Figure 18c and with the steep rise of the growth enhancement at .
ISW amplitude
The same tabulation gives the ISW amplitude. Since the Poisson equation holds with the total matter density, the Weyl potential scales as and its decay rate is , with the growth rate of the density. The ratio of the ISW–galaxy and lensing–galaxy cross-spectra in a narrow redshift bin, which is independent of galaxy bias and of the amplitude of the fluctuations [98], is proportional to this decay rate, and its value relative to CDM,
agrees with the full Limber computation of the two cross-spectra for galaxy samples at to –.
Cluster abundances in CDM and in the frozen model
The abundance of massive halos is exponentially sensitive to the amplitude of the linear fluctuations, which makes cluster counts a natural probe of a coupling that changes the late growth. This appendix summarizes the standard formalism, describes how we apply it to the frozen model, and quantifies the accuracy of the effective amplitude used in Sec. III.
Formalism
The number of clusters per unit redshift and solid angle above a mass threshold is
where, in a real survey, the sharp threshold is replaced by the selection function in the observable and by the scaling relation between the observable and the mass. The halo mass function is written as
with the variance of the linear density field smoothed with a top-hat filter of radius ,
where , and the multiplicity function of Ref. [121],
whose parameters depend on the overdensity that defines the halo mass and on redshift. We use masses , defined with respect to 500 times the critical density, which corresponds to an overdensity with respect to the mean, and interpolate the published parameters in this overdensity.
Application to the frozen model
The linear power spectrum of the frozen model differs from that of CDM only by the scale-independent factor of Eq. (19), so is that of CDM multiplied by . The multiplicity function, on the other hand, has been calibrated with -body simulations of CDM-like cosmologies [121], in which the mass of the dark matter is conserved. In the frozen model the exchange converts vacuum energy into dark matter where matter converges, which modifies the spherical collapse and may change the collapse threshold and the universality of . We use the CDM form of , which is the assumption implicit in any comparison with the published cluster constraints, and we regard the results as indicative.
Figure 20a compares the mass function at for the best fits of CDM and of the frozen model. For P18+DES-Dovekie+DESI DR2 the frozen model predicts 2.0 times more halos above at than CDM, and 2.8 times more above . The excess grows toward low redshift, where the growth is enhanced, and toward high masses, where the mass function is steepest (Figure 20b). For the best fit of the full combination the two models are indistinguishable.
Figure 20. Cluster abundances in CDM and in the frozen model. (a) Mass function at for the best fits of CDM and of the frozen model to P18+DES-Dovekie+DESI DR2, and of the frozen model to all data. (b) Ratio of the comoving number densities of halos above three mass thresholds in the frozen model and in CDM, for the best fits of P18+DES-Dovekie+DESI DR2. (c) Ratio of the counts per unit redshift of the frozen best fit and of a CDM model with the same parameters and the amplitude rescaled to the effective amplitude of each cluster sample, for the mass thresholds (SPT and PSZ2) and (eRASS1). The thresholds are representative of the mass range of each sample and are not meant to reproduce its selection function; the kernel-weighted ratios quoted in the text are close to unity, while the redshift dependence of the ratio shows that the frozen model changes the shape of the redshift distribution of the counts, and not only their amplitude.
Accuracy of the effective amplitude
To test the compression used in the main text, we compare the counts of the frozen best fit of P18+DES-Dovekie+DESI DR2 with those of a CDM model that has the same parameters and an amplitude rescaled to the effective amplitude of each sample. With representative mass thresholds and the redshift kernels of the three samples, the kernel-weighted counts of the frozen model are 1.02 (SPT), 0.92 (eRASS1) and 0.90 (PSZ2) times those of the rescaled CDM model. Since the counts of these samples scale as with , 6 and 8, respectively, these differences correspond to shifts of 0.2%, 1.4% and 1.4% in the effective amplitude, comparable to the error of the most precise sample and smaller than the others. The compression is therefore adequate for the consistency check of § 4.
The redshift distribution of the counts is a different matter. Figure 20c shows that, relative to the rescaled CDM model, the frozen model predicts more clusters at and fewer at higher redshift, by factors of two or more at the extremes of the samples. A full analysis of the counts in bins of redshift, which all the cluster surveys perform, would therefore be considerably more sensitive to the coupling than the compressed amplitude we use, once a mass function appropriate for a dark matter whose mass is not conserved becomes available.
Tension estimators
Let , and be the posteriors of P18, of the external data set and of their combination, with means and covariances , and the prior covariance.
a. Effective number of parameters. counts the parameters constrained beyond the prior [75].
b. Difference in means. In the parameters constrained by both data sets,
distributed as with degrees of freedom.
c. Update difference in means.
restricted to the directions in which the update improves the constraint [76].
d. Goodness-of-fit loss. at the respective best fits, with degrees of freedom [75].
e. Parameter shift. With drawn from the product of the two posteriors, the shift probability is
estimated with a kernel density estimate evaluated by fast Fourier transform [77].
f. Evidence ratio, information and suspiciousness.
and the tension probability is [79, 80]. depends on the prior volume of the parameters that does not constrain; does not.
g. Deviance information criterion. with , and ; indicates disagreement [81, 82].
Sampling and validation
Table Table VIII summarizes the importance samples. Each model was sampled with a single adaptive importance-sampling run (Sec. IV B), whose successive rounds refine the proposal and accumulate draws: CDM in four rounds, with 3298 exact likelihood evaluations in total, and the frozen model in five rounds, with 5966; every target has at least 650 effective samples and a Pareto diagnostic , well below the threshold 0.7 above which the weights become unreliable [72]. The raw and Pareto-smoothed evidences agree to better than 0.002. In the frozen model 6.2% of the draws fall beyond the physical boundary; they are rejected before the perturbations are computed and cost a negligible fraction of the time. No evaluation failed inside the prior and the boundary.
| Data | |||
| CDM, | |||
| P18 | 1129 | ||
| +Dov. | 881 | ||
| +DESI | 974 | ||
| +Dov.+DESI | 1087 | ||
| +shear | 1024 | ||
| +all | 894 | ||
| Frozen model, | |||
| P18 | 1369 | ||
| +Dov. | 992 | ||
| +DESI | 963 | ||
| +Dov.+DESI | 1104 | ||
| +shear | 650 | ||
| +all | 763 |
Table VIII. TABLE VIII. Importance sampling. is the total number of exact likelihood evaluations, accumulated over all the rounds of the sampling and shared by all data combinations, the effective sample size after Pareto smoothing, the Pareto diagnostic and the log-evidence with its Monte Carlo error. The evidences of DES-Dovekie and DESI DR2 alone, from nested sampling, are and in CDM and and in the frozen model.
a. Validation. With the fluid is unperturbed and the code reduces to CDM exactly. Our CDM results reproduce the companion analysis of the same data, obtained with a different version of the Boltzmann solver: the evidences of P18, P18+DES-Dovekie and P18+DES-Dovekie+DESI DR2 agree to 0.08, those of DES-Dovekie and DESI DR2 alone to 0.01, and the tension between P18+DES-Dovekie and DESI DR2 is 2.3–2.5 in both. Evaluations of the same points on two computers with different processor architectures agree to in .
b. Best fits. The surrogate refinement of Sec. IV B improves the best exact draw by 0.2–1.3 in and converges in one to five iterations. In the frozen model the best fits of the combinations without shear lie at –0.12, inside the allowed region; the coordinate keeps every trial point inside it.
Full posterior of the joint analysis
For completeness, Figure 21 shows the marginalized posteriors of all the sampled cosmological parameters and of the main derived ones for the two largest data combinations, in both models. The parameters shared with CDM are almost unaffected by the coupling: the only appreciable correlations of are with and , which it controls directly, and, more weakly, with the amplitude and the optical depth, which adjust the height of the acoustic peaks once the lensing smoothing is provided by the late growth. Adding cosmic shear removes the region of large without moving any of the other parameters, and the two models then give practically identical posteriors.
Figure 21. Marginalized posteriors (68% and 95% regions) of the cosmological parameters for P18+DES-Dovekie+DESI DR2 and P18+DES-Dovekie+DESI DR2+shear in the frozen model (filled) and in CDM (dashed and dotted lines). The last three parameters are derived: the matter density, the amplitude of the fluctuations and the age of the Universe.
References
- [1]E. Di Valentino et al., In the realm of the Hubble tension—a review of solutions, Class. Quant. Grav. 38, 153001 (2021), arXiv:2103.01183.DOI
- [2]E. Abdalla et al., Cosmology intertwined: A review of the particle physics, astrophysics, and cosmology associated with the cosmological tensions and anomalies, J. High Energy Astrophys. 34, 49 (2022), arXiv:2203.06142.DOI
- [3]E. Di Valentino, J. Levi Said, A. Riess, A. Pollo, V. Poulin, A. Gómez-Valent, et al., The CosmoVerse White Paper: Addressing observational tensions in cosmology with systematics and fundamental physics, Phys. Dark Univ. 49, 101965 (2025), arXiv:2504.01669.DOI
- [4]M. Leizerovich, S. J. Landau, and C. G. Scóccola, Tensions in cosmology: A discussion of statistical tools to determine inconsistencies, Phys. Lett. B 855, 138844 (2024), arXiv:2312.08542.DOI
- [5]M. Leizerovich, S. J. Landau, and C. G. Scóccola, Generalized tension metrics for multiple cosmological datasets, arXiv:2512.06086 (2025), arXiv:2512.06086.arxiv.org/abs/2512.06086
- [6]L. Kraiselburd, C. Pigozzo, S. J. Landau, and J. Alcaniz, Hubble tension and small-scale inhomogeneities on light propagation, Astron. Astrophys. 711, A199 (2026), arXiv:2504.06831.DOI
- [7]S. J. Landau, M. Benetti, A. Perez, and D. Sudarsky, Cosmological constraints on unimodular gravity models with diffusion, Phys. Rev. D 108, 043524 (2023), arXiv:2211.07424.arxiv.org/abs/2211.07424
- [8]DESI Collaboration, M. Abdul-Karim, et al., DESI DR2 results. II. Measurements of baryon acoustic oscillations and cosmological constraints, Phys. Rev. D 112, 083515 (2025), arXiv:2503.14738.
- [9]K. Lodha et al., Extended Dark Energy analysis using DESI DR2 BAO measurements, Phys. Rev. D 112, 083511 (2025), arXiv:2503.14743.arxiv.org/abs/2503.14743
- [10]D. Pedrotti, L. A. Escamilla, V. Marra, L. Perivolaropoulos, and S. Vagnozzi, BAO mis-calibration cannot rescue late-time solutions to the Hubble tension, Phys. Rev. D 113, 043507 (2026), arXiv:2510.01974.DOI
- [11]C. Wetterich, The cosmon model for an asymptotically vanishing time-dependent cosmological “constant”, Astron. Astrophys. 301, 321 (1995), arXiv:hep-th/9408025.arxiv.org/abs/hep-th/9408025
- [12]L. Amendola, Coupled quintessence, Phys. Rev. D 62, 043511 (2000), arXiv:astro-ph/9908023.DOI
- [13]T. Damour, G. W. Gibbons, and C. Gundlach, Dark matter, time-varying G, and a dilaton field, Phys. Rev. Lett. 64, 123 (1990).pubmed.ncbi.nlm.nih.gov/10041656
- [14]L. P. Chimento, A. S. Jakubi, D. Pavón, and W. Zimdahl, Interacting quintessence solution to the coincidence problem, Phys. Rev. D 67, 083513 (2003), arXiv:astro-ph/0303145.DOI
- [15]L. P. Chimento and D. Pavón, Dual interacting cosmologies and late accelerated expansion, Phys. Rev. D 73, 063511 (2006), arXiv:gr-qc/0505096.DOI
- [16]L. P. Chimento, Linear and nonlinear interactions in the dark sector, Phys. Rev. D 81, 043525 (2010), arXiv:0911.5687.DOI
- [17]L. P. Chimento and M. G. Richarte, Interacting dark matter and modified holographic Ricci dark energy induce a relaxed Chaplygin gas, Phys. Rev. D 84, 123507 (2011), arXiv:1107.4816.DOI
- [18]H. E. S. Velten, R. F. vom Marttens, and W. Zimdahl, Aspects of the cosmological “coincidence problem”, Eur. Phys. J. C 74, 3160 (2014), arXiv:1410.2509.DOI
- [19]S. Calogero and H. Velten, Cosmology with matter diffusion, J. Cosmol. Astropart. Phys. 11, 025, arXiv:1308.3393.DOI
- [20]H. Velten, A. Montiel, and S. Carneiro, GRB Hubble diagram and constraints on a Λ(t)CDM model, Mon. Not. R. Astron. Soc. 431, 3301 (2013), arXiv:1209.6590.
- [21]H. Velten, H. A. Borges, S. Carneiro, R. F. Fazolo, and S. Gomes, Large-scale structure and integrated Sachs–Wolfe effect in decaying vacuum cosmology, Mon. Not. R. Astron. Soc. 452, 2220 (2015), arXiv:1504.02416.
- [22]E. Di Valentino, A. Melchiorri, and O. Mena, Can interacting dark energy solve the H0 tension?, Phys. Rev. D 96, 043503 (2017), arXiv:1704.08342.DOI
- [23]E. Di Valentino, A. Melchiorri, O. Mena, and S. Vagnozzi, Interacting dark energy in the early 2020s: A promising solution to the H0 and cosmic shear tensions, Phys. Dark Univ. 30, 100666 (2020), arXiv:1908.04281.DOI
- [24]E. Di Valentino, A. Melchiorri, O. Mena, and S. Vagnozzi, Nonminimal dark sector physics and cosmological tensions, Phys. Rev. D 101, 063502 (2020), arXiv:1910.09853.DOI
- [25]W. Yang, S. Pan, E. Di Valentino, R. C. Nunes, S. Vagnozzi, and D. F. Mota, Tale of stable interacting dark energy, observational signatures, and the H0 tension, J. Cosmol. Astropart. Phys. 2018 (09), 019, arXiv:1805.08252.DOI
- [26]S. Pan, W. Yang, E. Di Valentino, E. N. Saridakis, and S. Chakraborty, Interacting scenarios with dynamical dark energy: Observational constraints and alleviation of the H0 tension, Phys. Rev. D 100, 103520 (2019), arXiv:1907.07540.DOI
- [27]R. C. Nunes, S. Vagnozzi, S. Kumar, E. Di Valentino, and O. Mena, New tests of dark sector interactions from the full-shape galaxy power spectrum, Phys. Rev. D 105, 123506 (2022), arXiv:2203.08093.DOI
- [28]W. Yang, S. Vagnozzi, E. Di Valentino, R. C. Nunes, S. Pan, and D. F. Mota, Listening to the sound of dark sector interactions with gravitational wave standard sirens, J. Cosmol. Astropart. Phys. 07, 037, arXiv:1905.08286.DOI
- [29]W. Giarè, M. A. Sabogal, R. C. Nunes, and E. Di Valentino, Interacting dark energy after DESI baryon acoustic oscillation measurements, Phys. Rev. Lett. 133, 251003 (2024), arXiv:2404.15232.DOI
- [30]E. Silva, M. A. Sabogal, M. Scherer, R. C. Nunes, E. Di Valentino, and S. Kumar, New constraints on interacting dark energy from DESI DR2 BAO observations, Phys. Rev. D 111, 123511 (2025), arXiv:2503.23225.DOI
- [31]M. A. Sabogal, S. Vagnozzi, and E. Di Valentino, Cosmological anatomy of interacting dark energy, arXiv:2609.18288 (2026), arXiv:2609.18288.arxiv.org/abs/2609.18288
- [32]V. Petri, V. Marra, and R. von Marttens, Dark degeneracy in DESI DR2 data: Interacting or evolving dark energy?, Phys. Rev. D 113, 023504 (2026), arXiv:2508.17955.DOI
- [33]R. von Marttens, J. E. Gonzalez, J. Alcaniz, V. Marra, and L. Casarini, Model-independent reconstruction of dark sector interactions, Phys. Rev. D 104, 043515 (2021), arXiv:2011.10846.DOI
- [34]V. Salzano, C. Pigozzo, M. Benetti, H. A. Borges, R. von Marttens, S. Carneiro, J. S. Alcaniz, J. C. Fabris, S. Tsujikawa, et al., J-PAS: forecasts on interacting vacuum energy models, J. Cosmol. Astropart. Phys. 2021 (09), 033, arXiv:2102.06417.DOI
- [35]D. Figueruelo, M. Aparicio Resco, F. A. Teppa Pannia, J. Beltrán Jiménez, D. Bettoni, A. L. Maroto, et al., J-PAS: Forecasts for dark matter–dark energy elastic couplings, J. Cosmol. Astropart. Phys. 07, 022, arXiv:2103.01571.DOI
- [36]S. Vagnozzi, L. Visinelli, P. Brax, A.-C. Davis, and J. Sakstein, Direct detection of dark energy: The XENON1T excess and future prospects, Phys. Rev. D 104, 063023 (2021), arXiv:2103.15834.DOI
- [37]G.-W. Yuan, A.-C. Davis, M. Giannotti, S. Vagnozzi, L. Visinelli, and J. K. Vogel, Direct detection of solar chameleons with electron recoil data from XENONnT, Phys. Rev. D 113, 123024 (2026), arXiv:2511.01655.DOI
- [38]J. Valiviita, E. Majerotto, and R. Maartens, Large-scale instability in interacting dark energy and dark matter fluids, J. Cosmol. Astropart. Phys. 2008 (07), 020, arXiv:0804.0232.
- [39]Y.-H. Li, L.-J.-F. Zhang, and X. Zhang, Parametrized post-Friedmann framework for interacting dark energy, Phys. Rev. D 90, 063005 (2014), arXiv:1404.5220.DOI
- [40]M. G. Richarte and L. Xu, Interacting parametrized post-Friedmann method, Gen. Relativ. Gravit. 48, 39 (2016), arXiv:1407.4348.
- [41]F. Simpson, Scattering of dark matter and dark energy, Phys. Rev. D 82, 083505 (2010), arXiv:1007.1034.DOI
- [42]M. Baldi and F. Simpson, Simulating momentum exchange in the dark sector, Mon. Not. R. Astron. Soc. 449, 2239 (2015), arXiv:1412.1080.DOI
- [43]M. Baldi and F. Simpson, Structure formation simulations with momentum exchange: alleviating tensions between high-redshift and low-redshift cosmological probes, Mon. Not. R. Astron. Soc. 465, 653 (2017), arXiv:1605.05623.DOI
- [44]A. Pourtsidou, C. Skordis, and E. J. Copeland, Models of dark matter coupled to dark energy, Phys. Rev. D 88, 083505 (2013), arXiv:1307.0458.DOI
- [45]C. Skordis, A. Pourtsidou, and E. J. Copeland, Parametrized post-Friedmannian framework for interacting dark energy theories, Phys. Rev. D 91, 083537 (2015), arXiv:1502.07297.DOI
- [46]A. Pourtsidou and T. Tram, Reconciling CMB and structure growth measurements with dark energy interactions, Phys. Rev. D 94, 043518 (2016), arXiv:1604.04222.DOI
- [47]M. S. Linton, A. Pourtsidou, R. Crittenden, and R. Maartens, Variable sound speed in interacting dark energy models, J. Cosmol. Astropart. Phys. 2018 (04), 043, arXiv:1711.05196.DOI
- [48]R. Kase and S. Tsujikawa, Weak cosmic growth in coupled dark energy with a Lagrangian formulation, Phys. Lett. B 804, 135400 (2020), arXiv:1911.02179.DOI
- [49]L. Amendola and S. Tsujikawa, Scaling solutions and weak gravity in dark energy with energy and momentum couplings, J. Cosmol. Astropart. Phys. 2020 (06), 020, arXiv:2003.02686.DOI
- [50]J. Beltrán Jiménez, D. Bettoni, D. Figueruelo, F. A. Teppa Pannia, and S. Tsujikawa, Velocity-dependent interacting dark energy and dark matter with a La grangian description of perfect fluids, J. Cosmol. As-tropart. Phys. 2021 (03), 085, arXiv:2012.12204.
- [51]K. Aoki, J. Beltrán Jiménez, M. C. Pookkillath, and S. Tsujikawa, Effective field theory of coupled dark energy and dark matter, Phys. Rev. D 10.1103/5wsb-1v6k (2025), arXiv:2504.17293.DOI
- [52]M. G. Richarte, L. F. Guimarães, S. J. Landau, and J. C. Fabris, The frozen vanilla model: Exploring dark sector interactions with delta effective theories, Phys. Rev. D 112, 043510 (2025), arXiv:2504.21110.DOI
- [53]R. K. Sachs and A. M. Wolfe, Perturbations of a cosmological model and angular variations of the microwave background, Astrophys. J. 147, 73 (1967).
- [54]R. G. Crittenden and N. Turok, Looking for a Cosmological Constant with the Rees-Sciama Effect, Phys. Rev. Lett. 76, 575 (1996), arXiv:astro-ph/9510072.
- [55]Planck Collaboration, N. Aghanim, et al., Planck 2018 results. VI. Cosmological parameters, Astron. Astrophys. 641, A6 (2020), arXiv:1807.06209.
- [56]E. Calabrese, A. Slosar, A. Melchiorri, G. F. Smoot, and O. Zahn, Cosmic microwave background lensing data as a test for the dark universe, Phys. Rev. D 77, 123531 (2008), arXiv:0803.2309.DOI
- [57]C.-P. Ma and E. Bertschinger, Cosmological perturbation theory in the synchronous and conformal Newtonian gauges, Astrophys. J. 455, 7 (1995), arXiv:astro-ph/9506072.DOI
- [58]Planck Collaboration, N. Aghanim, et al., Planck 2018 results. V. CMB power spectra and likelihoods, Astron. Astrophys. 641, A5 (2020), arXiv:1907.12875.
- [59]Planck Collaboration, N. Aghanim, et al., Planck 2018 results. VIII. Gravitational lensing, Astron. Astrophys. 641, A8 (2020), arXiv:1807.06210.
- [60]DES Collaboration, T. M. C. Abbott, et al., The Dark Energy Survey: Cosmology Results with 1500 New High-redshift Type Ia Supernovae Using the Full 5 yr Data Set, Astrophys. J. Lett. 973, L14 (2024), arXiv:2401.02929.
- [61]B. Popovic et al., The Dark Energy Survey Supernova Program: A Reanalysis Of Cosmology Results And Evidence For Evolving Dark Energy With An Updated Type Ia Supernova Calibration, Mon. Not. R. Astron. Soc. 10.1093/mnras/stag632 (2026), arXiv:2511.07517.DOI
- [62]A. H. Wright et al., KiDS-Legacy: Cosmological constraints from cosmic shear with the complete kilodegree survey, Astron. Astrophys. 703, A158 (2025), arXiv:2503.19441.DOI
- [63]A. Amon et al. (DES), Dark Energy Survey Year 3 results: Cosmology from cosmic shear and robustness to data calibration, Phys. Rev. D 105, 023514 (2022), arXiv:2105.13543.DOI
- [64]L. F. Secco et al. (DES), Dark Energy Survey Year 3 results: Cosmology from cosmic shear and robustness to modeling uncertainty, Phys. Rev. D 105, 023515 (2022), arXiv:2105.13544.DOI
- [65]S. Bocquet et al. (SPT, DES), SPT clusters with DES and HST weak lensing. II. Cosmological constraints from the abundance of massive halos, Phys. Rev. D 110, 083510 (2024), arXiv:2401.02075.
- [66]V. Ghirardini et al., The SRG/eROSITA all-sky survey: Cosmology constraints from cluster abundances in the western Galactic hemisphere, Astron. Astrophys. 689, A298 (2024), arXiv:2402.08458.DOI
- [67]G. Aymerich, M. Douspis, G. W. Pratt, et al., Cosmological constraints from the Planck cluster catalogue with new multi-wavelength mass calibration from Chandra and CFHT, Astron. Astrophys. 690, A238 (2024), arXiv:2402.04006.
- [68]R. Takahashi, M. Sato, T. Nishimichi, A. Taruya, and M. Oguri, Revising the Halofit Model for the Nonlinear Matter Power Spectrum, Astrophys. J. 761, 152 (2012), arXiv:1208.2701.DOI
- [69]E. Veach and L. J. Guibas, Optimally combining sampling techniques for Monte Carlo rendering, in Proceedings of SIGGRAPH ’95 (1995) p. 419.DOI
- [70]A. Owen and Y. Zhou, Safe and effective importance sampling, J. Am. Stat. Assoc. 95, 135 (2000).DOI
- [71]J.-M. Cornuet, J.-M. Marin, A. Mira, and C. P. Robert, Adaptive Multiple Importance Sampling, Scand. J. Stat. 39, 798 (2012), arXiv:0907.1254.arxiv.org/abs/0907.1254
- [72]A. Vehtari, D. Simpson, A. Gelman, Y. Yao, and J. Gabry, Pareto Smoothed Importance Sampling, J. Mach. Learn. Res. 25, 1 (2024), arXiv:1507.02646.
- [73]J. Skilling, Nested sampling for general Bayesian computation, Bayesian Anal. 1, 833 (2006).DOI
- [74]J. U. Lange, Nautilus: boosting Bayesian importance nested sampling with deep learning, Mon. Not. R. Astron. Soc. 525, 3181 (2023), arXiv:2306.16923.arxiv.org/abs/2306.16923
- [75]M. Raveri and W. Hu, Concordance and Discordance in Cosmology, Phys. Rev. D 99, 043506 (2019), arXiv:1806.04649.DOI
- [76]M. Raveri, G. Zacharegkas, and W. Hu, Quantifying concordance of correlated cosmological data sets, Phys. Rev. D 101, 103527 (2020), arXiv:1912.04880.DOI
- [77]M. Raveri and C. Doux, Non-Gaussian estimates of tensions in cosmological parameters, Phys. Rev. D 104, 043504 (2021), arXiv:2105.03324.DOI
- [78]P. Marshall, N. Rajguru, and A. Slosar, Bayesian evidence as a tool for comparing datasets, Phys. Rev. D 73, 067302 (2006), arXiv:astro-ph/0412535.DOI
- [79]W. Handley and P. Lemos, Quantifying tensions in cosmological parameters: Interpreting the DES evidence ratio, Phys. Rev. D 100, 043504 (2019), arXiv:1902.04029.DOI
- [80]W. Handley and P. Lemos, Quantifying dimensionality: Bayesian cosmological model complexity, Phys. Rev. D 100, 023512 (2019), arXiv:1903.06682.DOI
- [81]D. J. Spiegelhalter, N. G. Best, B. P. Carlin, and A. van der Linde, Bayesian measures of model complexity and fit, J. R. Stat. Soc. B 64, 583 (2002).DOI
- [82]S. Joudaki et al., CFHTLenS revisited: assessing concordance with Planck including astrophysical systematics, Mon. Not. R. Astron. Soc. 465, 2033 (2017), arXiv:1610.04606.
- [83]P. Lemos et al., Assessing tension metrics with dark energy survey and Planck data, Mon. Not. R. Astron. Soc. 505, 6179 (2021), arXiv:2012.09554.DOI
- [84]E. Rosenberg, S. Gratton, and G. Efstathiou, CMB power spectra and cosmological parameters from Planck PR4 with CamSpec, Mon. Not. R. Astron. Soc. 517, 4620 (2022), arXiv:2205.10869.DOI
- [85]M. Tristram et al., Cosmological parameters derived from the final Planck data release (PR4), Astron. Astrophys. 682, A37 (2024), arXiv:2309.10034.DOI
- [86]A. G. Riess, W. Yuan, L. M. Macri, D. Scolnic, D. Brout, S. Casertano, et al., A comprehensive measurement of the local value of the Hubble constant with 1 km s−1 Mpc−1 uncertainty from the Hubble Space Telescope and the SH0ES team, Astrophys. J. Lett. 934, L7 (2022), arXiv:2112.04510.
- [87]L. Breuval, A. G. Riess, S. Casertano, W. Yuan, L. M. Macri, M. Romaniello, et al., The Small Magellanic Cloud Cepheids observed with the Hubble Space Telescope provide a new anchor for the SH0ES distance ladder, Astrophys. J. 973, 30 (2024), arXiv:2404.08038.DOI
- [88]W. L. Freedman, B. F. Madore, I. S. Jang, T. J. Hoyt, A. J. Lee, and K. A. Owens, Status report on the Chicago-Carnegie Hubble Program (CCHP): Measurement of the Hubble constant using the Hubble and James Webb Space Telescopes, Astrophys. J. 985, 203 (2025), arXiv:2408.06153.arxiv.org/abs/2408.06153
- [89]TDCOSMO Collaboration, S. Birrer, E. J. Buckley-Geer, M. Cappellari, F. Courbin, F. Dux, et al., TDCOSMO 2025: Cosmological constraints from strong lensing time delays, Astron. Astrophys. 704, A63 (2025), arXiv:2506.03023.
- [90]E. Camphuis, W. Quan, L. Bakenhol, A. R. Khalife, F. Ge, F. Guidi, et al., SPT-3G D1: CMB temperature and polarization power spectra and cosmology from 2019 and 2020 observations of the SPT-3G main field, Phys. Rev. D 113, 083504 (2026), arXiv:2506.20707.
- [91]T. Louis, A. La Posta, Z. Atkins, H. T. Jense, I. Abril-Cabezas, G. E. Addison, et al., The Atacama Cosmology Telescope: DR6 power spectra, likelihoods and ΛCDM parameters, J. Cosmol. Astropart. Phys. 11, 062, arXiv:2503.14452.
- [92]D. Valcin, R. Jimenez, L. Verde, J. L. Bernal, and B. D. Wandelt, The age of the Universe with globular clusters: Reducing systematic uncertainties, J. Cosmol. Astropart. Phys. 2021 (08), 017, arXiv:2102.04486.DOI
- [93]D. Valcin, R. Jimenez, U. Seljak, and L. Verde, The age of the Universe with globular clusters. Part III. Gaia distances and hierarchical modeling, J. Cosmol. Astropart. Phys. 10, 030, arXiv:2503.19481.
- [94]D. Valcin, R. Jimenez, C. Lardo, U. Seljak, and L. Verde, The age of the Universe with globular clusters IV: Multiple stellar populations, arXiv:2603.04872 (2026), arXiv:2603.04872.arxiv.org/abs/2603.04872
- [95]J. Ying, B. Chaboyer, E. M. Boudreaux, C. Slaughter, M. Boylan-Kolchin, and D. Weisz, The absolute age of M92, Astron. J. 166, 18 (2023), arXiv:2306.02180.DOI
- [96]B. M. S. Hansen, J. Anderson, J. Brewer, A. Dotter, G. G. Fahlman, J. Hurley, et al., The white dwarf cooling sequence of NGC 6397, Astrophys. J. 671, 380 (2007), arXiv:astro-ph/0701738.DOI
- [97]H. E. Bond, F. P. Nelan, D. A. VandenBerg, G. H. Schaefer, and D. Harmer, HD 140283: A star in the solar neighborhood that formed shortly after the Big Bang, Astrophys. J. Lett. 765, L12 (2013), arXiv:1302.3180.
- [98]P. Zhang, Isolating the decay rate of cosmological gravitational potential, Astrophys. J. Lett. 647, L55 (2006), arXiv:astro-ph/0512422.DOI
- [99]F. Dong, P. Zhang, Z. Sun, and C. Park, The first direct measurement of gravitational potential decay rate at cosmological scales and improved dark energy constraint, Astrophys. J. (2022), arXiv:2206.04917.DOI
- [100]Planck Collaboration, P. A. R. Ade, N. Aghanim, M. Arnaud, M. Ashdown, J. Aumont, et al., Planck 2015 results. XXI. The integrated Sachs–Wolfe effect, Astron. Astrophys. 594, A21 (2016), arXiv:1502.01595.
- [101]A. Krolewski and S. Ferraro, The integrated Sachs Wolfe effect: unWISE and Planck constraints on dynamical dark energy, J. Cosmol. Astropart. Phys. 04, 033, arXiv:2110.13959.DOI
- [102]Q. Hang, S. Alam, J. A. Peacock, and Y.-C. Cai, Galaxy clustering in the DESI Legacy Survey and its imprint on the CMB, Mon. Not. R. Astron. Soc. 501, 1481 (2021), arXiv:2010.00466.DOI
- [103]B. Bahr-Kalus, D. Parkinson, J. Asorey, S. Camera, C. Hale, and F. Qin, A measurement of the integrated Sachs–Wolfe effect with the Rapid ASKAP Continuum Survey, Mon. Not. R. Astron. Soc. 517, 3785 (2022), arXiv:2204.13436.
- [104]M. Ghodsi Yengejeh, J. R. Bermejo-Climent, A. Kovács, I. Szapudi, and I. Csabai, Probing dark energy evolution with Quaia quasars through the integrated Sachs–Wolfe effect, arXiv:2607.22461 (2026), arXiv:2607.22461.arxiv.org/abs/2607.22461
- [105]T. Giannantonio, R. Crittenden, R. Nichol, and A. J. Ross, The significance of the integrated Sachs–Wolfe effect revisited, Mon. Not. R. Astron. Soc. 426, 2581 (2012), arXiv:1209.2125.DOI
- [106]B. Stölzner, A. Cuoco, J. Lesgourgues, and M. Bilicki, Updated tomographic analysis of the integrated Sachs–Wolfe effect and implications for dark energy, Phys. Rev. D 97, 063506 (2018), arXiv:1710.03238.DOI
- [107]A. Kovács, C. Sánchez, J. García-Bellido, J. Elvin-Poole, N. Hamaus, V. Miranda, et al., More out of less: an excess integrated Sachs–Wolfe signal from supervoids mapped out by the Dark Energy Survey, Mon. Not. R. Astron. Soc. 484, 5267 (2019), arXiv:1811.07812.DOI
- [108]DES Collaboration, T. M. C. Abbott, M. Aguena, A. Alarcon, O. Alves, A. Amon, et al., Dark Energy Survey Year 6 results: Cosmological constraints from cosmic shear, arXiv:2602.10065 (2026), arXiv:2602.10065.
- [109]X. Li, T. Zhang, S. Sugiyama, R. Dalal, R. Terasawa, M. M. Rau, et al., Hyper Supreme-Cam Year 3 results: Cosmology from cosmic shear two-point correlation functions, Phys. Rev. D 108, 123518 (2023), arXiv:2304.00702.arxiv.org/abs/2304.00702
- [110]R. Dalal, X. Li, A. Nicola, J. Zuntz, M. A. Strauss, S. Sugiyama, et al., Hyper Supreme-Cam Year 3 results: Cosmology from cosmic shear power spectra, Phys. Rev. D 108, 123519 (2023), arXiv:2304.00701.DOI
- [111]M. Asgari, C.-A. Lin, B. Joachimi, B. Giblin, C. Heymans, H. Hildebrandt, et al., KiDS-1000 cosmology: Cosmic shear constraints and comparison between two point statistics, Astron. Astrophys. 645, A104 (2021), arXiv:2007.15633.DOI
- [112]Dark Energy Survey and Kilo-Degree Survey Collaboration, T. M. C. Abbott, M. Aguena, A. Alarcon, O. Alves, A. Amon, et al., DES Y3 + KiDS-1000: Consistent cosmology combining cosmic shear surveys, Open J. Astrophys. 6 (2023), arXiv:2305.17173.DOI
- [113]D. Anbajagane, C. Chang, A. Drlica-Wagner, C. Y. Tan, M. Adamow, R. A. Gruendl, et al., The DECADE cosmic shear project IV: cosmological constraints from 107 million galaxies across 5,400 deg² of the sky, Open J. Astrophys. 8 (2025), arXiv:2502.17677.DOI
- [114]L. W. K. Goh, S. Guerrini, C. Daley, F. Hervas-Peters, M. Kilbinger, A. Wittje, et al., UNIONS-3500 weak lensing: III. 2D cosmological constraints in configuration space, Mon. Not. R. Astron. Soc. (2026), arXiv:2605.13547.DOI
- [115]M. S. Linton, R. Crittenden, and A. Pourtsidou, Momentum transfer models of interacting dark energy, J. Cosmol. Astropart. Phys. 2022 (08), 075, arXiv:2107.03235.DOI
- [116]Planck Collaboration, Planck 2018 likelihood data, COM_Likelihood_Data-baseline, version R3.00, Planck Legacy Archive, European Space Agency (2019).
- [117]DES Collaboration, DES-SN5YR: Dark Energy Survey Supernova Program five-year data release, including the DES-Dovekie recalibration, GitHub repository desscience/DES-SN5YR (2024).
- [118]DESI Collaboration, DESI DR2 baryon acoustic oscillation measurements and cosmology data products, DESI public data releases (2025).
- [119]KiDS Collaboration, KiDS-Legacy weak-lensing catalogue and cosmic-shear data products (KiDS DR5), KiLo-Degree Survey data release (2025).
- [120]DES Collaboration, DES Y3A2 data release: cosmic-shear data products and source redshift distributions, DES Data Management, National Center for Supercomputing Applications (2021).
- [121]J. Tinker, A. V. Kravtsov, A. Klypin, K. Abazajian, M. Warren, G. Yepes, et al., Toward a halo mass function for precision cosmology: The limits of universality, Astrophys. J. 688, 709 (2008), arXiv:0803.2706.DOI

















