betaFeedback

WorksPhysics and Astronomy

New Observational Insights into Generalised Axion-Like Dark Energy

Abstract

We consider a quintessence model based on a generalised axion-like scalar field potential, originally introduced as a phenomenological framework to describe the late-time accelerated expansion of the Universe. The model exhibits a tracking regime that alleviates the coincidence problem and naturally leads to a phase of late-time acceleration driven by an effective cosmological constant, without requiring fine-tuning of the scalar field initial conditions. In this work, we confront the background cosmological evolution of this model with current observations by performing a parameter inference analysis using cosmic microwave background distance priors, baryon acoustic oscillation measurements, Type Ia supernovae, the SH0ES distance ladder, cosmic chronometers, and gamma-ray bursts. This analysis allows us to assess the observational viability of the model, its ability to reconcile different cosmological datasets, and its statistical performance relative to ΛΛCDM.

Introduction

Observations over the past two decades have established that the Universe is currently undergoing a phase of accelerated expansion. This behaviour was first inferred from measurements of luminosity distances to high-redshift Type Ia supernovae [1, 2] and has since been independently confirmed by observations of the cosmic microwave background and large-scale structure [3, 4]. Within the standard Λ\LambdaCDM cosmological model, late-time acceleration is attributed to a cosmological constant. While this framework provides an excellent fit to observational data when individual cosmological probes are analysed separately, mild but persistent inconsistencies arise when multiple data sets are combined [5]. Moreover, the physical origin of the cosmological constant remains poorly understood, giving rise to well-known theoretical and conceptual issues. Together, these observational and conceptual issues motivate the exploration of extensions of Λ\LambdaCDM that go beyond a strictly constant dark energy component [6].

A broad class of such extensions assumes that dark energy evolves dynamically over cosmic time. Well-studied examples include scalar field models such as quintessence [7], kk-essence theories with non-canonical kinetic terms [8], and axion-like dark energy scenarios [9, 10, 11, 12]. More general scalar-tensor constructions, including kinetic gravity braiding and subclasses of Horndeski theories, further enlarge the space of viable dark energy dynamics while preserving second-order field equations [13, 14, 15, 16, 17]. The late-time acceleration of the universe can as well be described by a 3-form [18, 19, 20] and can help to alleviate the Hubble tension [21, 22]. Alternative phenomenological approaches describe dark energy as an effective perfect fluid with a parametrised equation of state, enabling systematic studies of departures from Λ\LambdaCDM at both the background and perturbative levels [23, 24, 25]. Closely related scenarios include sign-switching cosmological constant models, in which the effective dark energy density changes sign during cosmic evolution [26, 27, 28, 29, 30, 31, 32, 33].

Alternatively, late-time cosmic acceleration may be interpreted as a consequence of modified gravitational dynamics rather than the presence of a new dark energy component. In this perspective, the accelerated expansion arises from extensions or generalisations of General Relativity, potentially alleviating some of the conceptual difficulties associated with a cosmological constant [6]. Representative examples include f(R)f(R) gravity [34, 35, 36, 37], teleparallel extensions such as f(T)f(T) gravity [38, 39, 40], and theories based on non-metricity, including f(Q)f(Q) gravity [41, 42, 43, 44, 45]. Collectively, these frameworks provide a broad theoretical laboratory for probing physics beyond Λ\LambdaCDM and for confronting the late-time acceleration of the Universe with current and forthcoming observational data.

In this paper, we focus on a class of quintessence models inspired by axion-like dark energy scenarios, motivated by the framework introduced in [46, 47]. Axion-like particles arise naturally in theories involving spontaneously broken global symmetries and are characterised by periodic scalar field potentials [48]. Depending on their characteristic mass and energy scales, these fields can play different roles throughout cosmic history, including driving inflation [49, 50, 51, 52, 53], contributing to early dark energy [54, 55], or acting as ultralight dark matter candidates [56, 57]. In particular, the wave nature of ultralight axion-like dark matter can give rise to distinctive structures on galactic scales, including interference patterns and solitonic cores, as explored in [58, 59, 60]. At sufficiently small mass scales, axion-like fields may instead provide a dynamical explanation for the late-time accelerated expansion of the Universe [9, 10] [61]. Within this context, the model considered here describes cosmic acceleration through the dynamics of a canonical scalar field evolving under a generalised axion-like potential, whose functional form extends the standard periodic axion potential. Constructed as a phenomenological framework, this potential supports an extended tracking regime that reduces the sensitivity to the initial scalar field conditions, followed by a late-time evolution approaching an effective cosmological constant.

In this work, we extend the observational analysis of our previous study [11, 12] by performing a background-level parameter inference of the model using current cosmological data. The constraints are obtained through a Markov Chain Monte Carlo analysis based on the cosmic microwave background (CMB) shift parameters, recent baryon acoustic oscillation (BAO) measurements from DESI Data Release 2 [62], and luminosity distance data from the Pantheon+ type Ia supernovae (SNe) samples [63] combined with the SH0ES determination of the Hubble constant [64] or the cosmic chronometer (CC) based on galaxy cluster star burst history [65] and the luminosity data from the standardised gamma ray burst (GRB) [66]. This approach allows us to assess the consistency of the model with the observed expansion history of the Universe over a wide redshift range. These datasets naturally fall into two categories according to the Hubble parameter H0H_0: the higher H0H_0 branch with H0∼73H_0 \sim73 km/s/Mpc representative in our work by the SH0ES project, and the lower H0H_0 ones with H0∼67H_0 \sim67 km/s/Mpc, as suggested by CC and the combined constraint of CMB and BAO1. The aim of this work is to check if the axion-like dark energy model may alter this picture, by increasing the deceleration parameter during the tracking regime.

This paper is organised as follows. In Section II, we present the theoretical framework of the model and introduce the scalar field dynamics under consideration. Section III describes the observational data sets employed in our analysis and outlines the methodology used to constrain the model parameters. In § 9, we confront the theoretical predictions with current cosmological observations and discuss the resulting parameter constraints. Finally, in Section VI, we summarise our main findings and present our conclusions.

Cosmology of Generalised Axion-like Quintessence

In this section, we briefly review the quintessence model introduced in [46], which employs a generalised axion-like scalar field potential to account for the late-time accelerated expansion of the Universe. The model is formulated within a spatially flat Friedmann–Lemaître–Robertson–Walker (FLRW) framework and includes the standard radiation and pressureless matter components, the latter comprising both cold dark matter and baryonic matter components.

The scalar field dynamics are governed by the Klein–Gordon equation,

ϕ¨+3Hϕ˙+dV(ϕ)dϕ=0,(1)\ddot{\phi} + 3H\dot{\phi} + \frac{\mathrm{d}V(\phi)}{\mathrm{d}\phi} = 0, \tag*{(1)}

where H≡a˙/aH \equiv\dot{a}/a denotes the Hubble expansion rate, an overdot represents differentiation with respect to cosmic time, and V(ϕ)V(\phi) is the scalar field potential.

The defining feature of the model is the axion-inspired potential

V(ϕ)=Λ4[1−cos⁡(ϕf)]−n,n>0,(2)V(\phi) = \Lambda^4 \left[1 - \cos\left(\frac{\phi}{f}\right)\right]^{-n}, \qquad n > 0, \tag*{(2)}

with Λ\Lambda and ff positive constants setting the energy scale and periodicity of the field. Treated as an effective phenomenological ansatz, this potential allows for an extended tracking phase in which the scalar field energy density follows that of the dominant background component, before driving the late-time accelerated expansion.

The background evolution is determined by the Friedmann equation,

H2=8πG3(ρr+ρm+ρϕ),(3)H^2 = \frac{8\pi G}{3}\left(\rho_r + \rho_m + \rho_\phi\right), \tag*{(3)}

where ρr\rho_r denotes the radiation energy density, ρm=ρcdm+ρb\rho_m = \rho_{cdm} + \rho_b is the total matter energy density, with ρcdm\rho_{cdm} and ρb\rho_b corresponding to cold dark matter and baryons, respectively, and ρϕ\rho_\phi is the scalar field energy density, given by

ρϕ=12ϕ˙2+V(ϕ).(4)\rho_\phi= \frac{1}{2}\dot{\phi}^2 + V(\phi). \tag*{(4)}

The corresponding scalar field pressure is given by

pϕ=12ϕ˙2−V(ϕ),(5)p_\phi= \frac{1}{2}\dot{\phi}^2 - V(\phi), \tag*{(5)}

and its equation of state parameter is defined as

wϕ≡pϕρϕ=12ϕ˙2−V(ϕ)12ϕ˙2+V(ϕ).(6)w_\phi\equiv\frac{p_\phi}{\rho_\phi} = \frac{\frac{1}{2}\dot{\phi}^2 - V(\phi)}{\frac{1}{2}\dot{\phi}^2 + V(\phi)}. \tag*{(6)}

The acceleration of the scale factor is given by

a¨a=−4πG3(ρm+2ρr+ρϕ+3pϕ).(7)\frac{\ddot{a}}{a} = -\frac{4\pi G}{3}\left(\rho_m + 2\rho_r + \rho_\phi+ 3p_\phi\right). \tag*{(7)}

It is also useful to introduce the total effective equation of state of the Universe, defined as

wtot≡ptotρtot=13ρr+pϕρr+ρm+ρϕ=13Ωr+wϕΩϕ.(8)w_{\mathrm{tot}} \equiv\frac{p_{\mathrm{tot}}}{\rho_{\mathrm{tot}}} = \frac{\frac{1}{3}\rho_r + p_\phi}{\rho_r + \rho_m + \rho_\phi} = \frac{1}{3}\Omega_r + w_\phi\Omega_\phi. \tag*{(8)}

The acceleration equation can then be rewritten as

a¨a=−H22(1+3wtot),(9)\frac{\ddot{a}}{a}=-\frac{H^2}{2}(1+3w_{\rm tot}), \tag*{(9)}

showing that accelerated expansion occurs when wtot<−1/3w_{\rm tot}<-1/3.

Radiation and matter are minimally coupled and independently conserved,

ρ˙r+4Hρr=0,(10)\dot{\rho}_{r}+4H\rho_{r}=0, \tag*{(10)}
ρ˙m+3Hρm=0.(11)\dot{\rho}_{m}+3H\rho_{m}=0. \tag*{(11)}

Together, these equations fully specify the background cosmological dynamics of the model.

It is useful to introduce the usual density parameters

ΩA≡8πG3H2ρA,A∈{r,m,ϕ},(12)\Omega_A\equiv\frac{8\pi G}{3H^2}\rho_A,\qquad A\in\{r,m,\phi\}, \tag*{(12)}

which satisfy the closure relation Ωr+Ωm+Ωϕ=1\Omega_r+\Omega_m+\Omega_\phi=1 in a spatially flat universe.

Potential properties: minimum, vacuum scale, and effective mass

The potential (2) reflects the periodic structure of the cosine function. Owing to its inverse power-law form, it diverges at ϕ/f=2πk\phi/f=2\pi k, with k∈Zk\in\mathbb{Z}, defining singular boundaries that separate different branches of the potential. Within each branch, the potential possesses a finite minimum at

ϕmin⁡=(2k+1)πf,k∈Z,(13)\phi_{\min}=(2k+1)\pi f,\qquad k\in\mathbb{Z}, \tag*{(13)}

corresponding to V,ϕ(ϕmin⁡)=0V_{,\phi}(\phi_{\min})=0. In this work, we restrict our analysis to the first branch, 0<ϕ/f<2π0<\phi/f<2\pi, whose minimum is located at ϕmin⁡=πf\phi_{\min}=\pi f. This minimum defines the physically relevant late-time attractor for the scalar field trajectories considered here.

The value of the potential at the minimum is

Vmin⁡≡V(ϕmin⁡)=2−nΛ4,(14)V_{\min}\equiv V(\phi_{\min})=2^{-n}\Lambda^4, \tag*{(14)}

which acts as an effective vacuum energy once the scalar field settles near ϕmin⁡\phi_{\min}.

Expanding around the minimum and defining Δϕ≡ϕ−ϕmin⁡\Delta_\phi\equiv\phi-\phi_{\min}, the potential admits the quadratic approximation

V(ϕ)≃Vmin⁡+12meff2Δϕ2+O(Δϕ3),(15)V(\phi)\simeq V_{\min}+\frac{1}{2}m_{\rm eff}^2\Delta_\phi^2+O(\Delta_\phi^3), \tag*{(15)}

which defines the effective mass (curvature) scale

meff2≡V,ϕϕ∣ϕ=ϕmin⁡=n2f2Vmin⁡=n2f22−nΛ4.(16)m_{\rm eff}^2\equiv V_{,\phi\phi}\big|_{\phi=\phi_{\min}}=\frac{n}{2f^2}V_{\min}=\frac{n}{2f^2}2^{-n}\Lambda^4. \tag*{(16)}

This curvature scale determines the late-time dynamics of the scalar field and, in particular, whether the approach to the minimum is overdamped or oscillatory.

The global structure of the potential and the location of its finite minimum are illustrated in Figure 1.

Figure 1. Compactified generalised axion-like potential V(ϕ)/Λ4=[1−cos⁡(ϕ/f)]−nV(\phi)/\Lambda^4 = [1-\cos(\phi/f)]^{-n} as a function of ϕ/f\phi/f over one period, for representative values of nn. The vertical axis uses the compactified mapping 2πarctan⁡(V/Λ4)\frac{2}{\pi}\arctan(V/\Lambda^4), which renders the divergences at ϕ/f=0\phi/f=0 and 2π2\pi finite while preserving the minimum at ϕ/f=π\phi/f=\pi. Dashed curves show the quadratic expansion around the minimum, V/Λ4≃2−n+n22−n(ϕ/f−π)2V/\Lambda^4 \simeq2^{-n} + \frac{n}{2}2^{-n}(\phi/f-\pi)^2.

Tracking behaviour and late-time approach to de Sitter

A central feature of the model is the existence of an early-time tracking regime. In the region of field space relevant during radiation and matter domination, the potential behaves effectively as an inverse power-law. Indeed, for ∣ϕ∣/f≪1|\phi|/f\ll1,

1−cos⁡(ϕf)≃12(ϕf)2,⇒V(ϕ)∝ϕ−2n,(17)1-\cos\left(\frac{\phi}{f}\right)\simeq\frac{1}{2}\left(\frac{\phi}{f}\right)^2, \qquad\Rightarrow\qquad V(\phi)\propto\phi^{-2n}, \tag*{(17)}

so that the dynamics resemble those of inverse power-law quintessence. As a result, ρϕ\rho_\phi can remain subdominant while evolving along an attractor-like trajectory that is only weakly sensitive to initial conditions.

During matter domination, the tracking solution is well approximated by [46]

wϕ≃−11+n,(matter domination),(18)w_\phi\simeq-\frac{1}{1+n},\qquad\text{(matter domination)}, \tag*{(18)}

so that the scalar field energy density redshifts more slowly than matter and its density fraction grows with time. For the representative case n=1n=1 considered below, this yields wϕ≃−1/2w_\phi\simeq-1/2, independently of the value of ff. This reflects the fact that the tracking dynamics are governed by the effective inverse power-law behaviour of the potential at small field values, and are therefore insensitive to the curvature scale near the minimum. As a result, both the oscillatory and non-oscillatory solutions exhibit the same tracking behaviour prior to the onset of scalar field domination. This behaviour is illustrated in Figure 2a, where varying ϕini/f\phi_{\mathrm{ini}}/f shows that the tracker is reached from a range of initial scalar field displacements. This gradual growth naturally triggers a transition to a late-time phase in which the scalar field becomes dynamically important and drives accelerated expansion as it settles near the minimum.

Oscillatory versus non-oscillatory late-time dynamics

Once the field reaches the vicinity of the minimum, the displacement Δϕ≡ϕ−ϕmin⁡\Delta_\phi\equiv\phi-\phi_{\min} obeys, to leading order,

Δ¨ϕ+3HΔ˙ϕ+meff2Δϕ≃0,(19)\ddot{\Delta}_{\phi}+3H\dot{\Delta}_{\phi}+m_{\mathrm{eff}}^2\Delta_{\phi}\simeq0, \tag*{(19)}

corresponding to a damped harmonic oscillator with Hubble friction. Assuming that HH varies slowly over one oscillation period, the characteristic exponents are

λ±=−3H2±(3H2)2−meff2.(20)\lambda_{\pm}=-\frac{3H}{2}\pm\sqrt{\left(\frac{3H}{2}\right)^2-m_{\mathrm{eff}}^2}. \tag*{(20)}

The motion becomes underdamped and oscillatory when

meff>32H,(21)m_{\mathrm{eff}}>\frac{3}{2}H, \tag*{(21)}

whereas for meff<32Hm_{\mathrm{eff}}<\frac{3}{2}H the evolution is overdamped and the field approaches the minimum monotonically. Since HH decreases with time, a transition from overdamped to underdamped behaviour may occur dynamically as the Universe expands. A practical late-time diagnostic is therefore the comparison with the present expansion rate,

meff≳H0⇒oscillations can occur by today,meff≲H0⇒overdamped approach.(22)\begin{aligned} m_{\mathrm{eff}} &\gtrsim H_0 \quad\Rightarrow\quad\text{oscillations can occur by today},\\ m_{\mathrm{eff}} &\lesssim H_0 \quad\Rightarrow\quad\text{overdamped approach}. \tag*{(22)} \end{aligned}

Using the expression

meff2=n2f2Vmin⁡,(23)m_{\mathrm{eff}}^2=\frac{n}{2f^2}V_{\min}, \tag*{(23)}

and assuming that the field reaches the minimum during the onset of scalar field domination, so that

3H2≃8πGVmin⁡,(24)3H^2\simeq8\pi G V_{\min}, \tag*{(24)}

one obtains the approximate relation

meff2≃3n16πGf2H2.(25)m_{\mathrm{eff}}^2\simeq\frac{3n}{16\pi G f^2}H^2. \tag*{(25)}

The condition meff=3H/2m_{\mathrm{eff}}=3H/2 then yields a characteristic critical scale

fcrit=n12πG=2n3MPl,(26)f_{\mathrm{crit}}=\sqrt{\frac{n}{12\pi G}}=\sqrt{\frac{2n}{3}}M_{\mathrm{Pl}}, \tag*{(26)}

where MPl=(8πG)−1/2M_{\mathrm{Pl}}=(8\pi G)^{-1/2}. Therefore, oscillatory behaviour is expected when f≲fcritf\lesssim f_{\mathrm{crit}}, whereas for f≳fcritf\gtrsim f_{\mathrm{crit}} the evolution remains overdamped.

Regardless of whether the late-time evolution is overdamped or oscillatory, both branches converge toward the same asymptotic de Sitter solution characterised by

H2→8πG3Vmin⁡,wϕ→−1.(27)H^2\to\frac{8\pi G}{3}V_{\min},\qquad w_\phi\to-1. \tag*{(27)}

The distinction between the two regimes therefore concerns only the dynamical path by which the scalar field approaches the minimum, while the final background state is identical. The two late-time regimes are illustrated in Figure 2 for the representative case n=1n=1. In Figure 2b and Figure 2c we compare an oscillatory trajectory, with f=0.1MPlf=0.1M_{\mathrm{Pl}}, and a non-oscillatory trajectory, with f=1MPlf=1M_{\mathrm{Pl}}, fixing in both cases the same initial conditions, ϕini/f=10−4\phi_{\mathrm{ini}}/f=10^{-4} and ϕ˙ini=0\dot{\phi}_{\mathrm{ini}}=0. In the oscillatory case, the field reaches the vicinity of the minimum earlier and the subsequent oscillations are efficiently damped by the Hubble expansion. As a result, the scalar field rapidly approaches the Sitter attractor and the background evolution remains very close to that of Λ\LambdaCDM, with only mild late-time departures in the density parameters. By contrast, in the non-oscillatory case, the approach to the minimum is overdamped and more gradual, leading to more visible deviations from Λ\LambdaCDM in the recent expansion history.

(a) Evolution of the scalar field equation of state wϕw_{\phi} and of the total effective equation of state wtotw_{\mathrm{tot}} for several initial displacements ϕini/f\phi_{\mathrm{ini}}/f, with ϕ˙ini=0\dot{\phi}_{\mathrm{ini}}=0. The values of nn and ff are kept fixed separately for the non-oscillatory and oscillatory branches, with n=1n=1, f=1 MPlf=1\,M_{\mathrm{Pl}} and f=0.1 MPlf=0.1\,M_{\mathrm{Pl}}, respectively. The different initial conditions mainly affect the time at which the field joins the tracker, while the trajectories subsequently converge toward the common matter-era tracking behaviour and the same late-time de Sitter regime. The inset highlights the damped oscillations of wϕw_{\phi} around −1-1 in the oscillatory branch.

(b) Evolution of the density parameters Ωr\Omega_{r}, Ωm\Omega_{m} and Ωϕ\Omega_{\phi} for the generalised axion-like quintessence model with n=1n=1. We compare a non-oscillatory trajectory with f=1 MPlf=1\,M_{\mathrm{Pl}} and an oscillatory trajectory with f=0.1 MPlf=0.1\,M_{\mathrm{Pl}}, together with the corresponding Λ\LambdaCDM reference. In both scalar field trajectories the initial conditions are fixed to ϕini/f=10−4\phi_{\mathrm{ini}}/f=10^{-4} and ϕ˙ini=0\dot{\phi}_{\mathrm{ini}}=0.

(c) Residuals of the density parameters with respect to Λ\LambdaCDM, ΔΩi≡Ωiaxion−ΩiΛCDM\Delta\Omega_i \equiv\Omega_i^{\mathrm{axion}}-\Omega_i^{\Lambda\mathrm{CDM}}, for i={r,m,ϕ}i=\{r,m,\phi\}. The same parameter choices and initial conditions as in panel (b) are used, showing that the non-oscillatory branch produces larger late-time deviations from Λ\LambdaCDM than the oscillatory branch.

Figure 2. Representative background evolution of the generalised axion-like quintessence model for n=1n=1. Panel (a) tests the dependence on the initial scalar field displacement by varying ϕini/f\phi_{\mathrm{ini}}/f while keeping the values of nn and ff fixed within each branch. The different trajectories exhibit the expected tracking behaviour, approaching wϕ≃−1/2w_{\phi}\simeq-1/2 during matter domination for n=1n=1, and later evolving toward the same asymptotic de Sitter solution with wϕ→−1w_{\phi}\to-1. Panels (b) and (c) compare two late-time regimes: a non-oscillatory case with f=1 MPlf=1\,M_{\mathrm{Pl}} and an oscillatory case with f=0.1 MPlf=0.1\,M_{\mathrm{Pl}}, both evolved from ϕini/f=10−4\phi_{\mathrm{ini}}/f=10^{-4} and ϕ˙ini=0\dot{\phi}_{\mathrm{ini}}=0. Panel (b) shows the radiation, matter and scalar field density parameters, while panel (c) shows their residuals with respect to the Λ\LambdaCDM reference. The comparison shows that the value of ff controls the late-time approach to the minimum of the potential, distinguishing oscillatory from overdamped evolution, whereas the tracker reduces the sensitivity to the initial value of the scalar field.

Figure 2a complements this comparison by varying the initial displacement ϕini/f\phi_{\mathrm{ini}}/f, while keeping the values of nn and ff fixed separately in the oscillatory and non-oscillatory cases. This makes the tracking behaviour explicit: different initial conditions may delay or advance the moment at which the field joins the tracker, but once the tracking regime is reached the trajectories evolve toward the same matter-era behaviour and subsequently toward the same late-time de Sitter limit. Thus, the late-time evolution is controlled primarily by the structure of the potential and by the value of ff, rather than by a fine tuning of the initial scalar field displacement.

Observational Data

In this section we describe the cosmological datasets used to constrain the generalised axion-like quintessence model. Since the analysis is restricted to the homogeneous background evolution, we employ probes that primarily measure cosmological distances or the expansion rate over a broad range of redshifts. The datasets considered here are CMB distance priors, baryon acoustic oscillation measurements from DESI DR2, Type Ia supernovae from Pantheon+, local measurements of the Hubble constant from SH0ES, cosmic chronometer measurements of H(z)H(z), and gamma-ray burst distance data. These observations provide complementary information on different epochs of the cosmic expansion history. The CMB distance priors constrain the geometry of the

Universe close to photon decoupling. BAO measurements use the sound horizon at the baryon drag epoch as a standard ruler and therefore provide absolute distance information at low and intermediate redshifts. Type Ia supernovae trace the relative luminosity-distance relation at late-time, while the SH0ES calibration fixes the corresponding absolute scale. Cosmic chronometers give direct estimates of the expansion rate, and gamma-ray bursts extend the Hubble diagram to redshifts beyond those typically reached by supernovae.

For the purposes of the present analysis, we organise the data into three classes. The first class is a high-redshift set formed by CMB distance priors and BAO measurements. The second class is a high-H0H_0 late-time distance-ladder set, consisting of the Pantheon+ supernova compilation calibrated with SH0ES Cepheid measurements, which we denote by PPS. The third class is a lower-H0H_0 late-time set formed by cosmic chronometers and gamma-ray bursts, together with a binned Pantheon+ supernova sample without the SH0ES calibration, which we denote by Pan.

We analyse the following dataset combinations:

  1. BAO;

  2. CMB + BAO;

  3. PPS;

  4. CC + GRB + Pan;

  5. CMB + BAO + PPS;

  6. CMB + BAO + CC + GRB + Pan.

We do not consider the CMB distance priors as an independent dataset in the parameter inference. The compressed CMB likelihood contains only three effective observables, which is fewer than the number of free parameters varied in the model.

CMB distance priors

The Cosmic Microwave Background contains precise information about the geometry and physical conditions of the Universe at the time of photon decoupling. For models in which the main deviations from Λ\LambdaCDM occur at late-time, much of the relevant background information can be encoded in a small set of compressed variables known as distance priors or shift parameters.

We use the CMB distance priors derived in [67]. The data vector is given by

XCMB=(R,ℓa,ωb0),(28)X_{\mathrm{CMB}} = (R,\ell_a,\omega_{b0}), \tag*{(28)}

where RR is the shift parameter, ℓa\ell_a is the acoustic angular scale, and

ωb0=Ωb0h2(29)\omega_{b0} = \Omega_{b0}h^2 \tag*{(29)}

is the present physical baryon density. Here hh is defined through

H0=100h km s−1 Mpc−1.(30)H_0 = 100h\ {\rm km\ s^{-1}\ Mpc^{-1}}. \tag*{(30)}

The shift parameter is sensitive to the distance to the surface of last scattering weighted by the present matter density, whereas the acoustic scale measures the angular size of the sound horizon at photon decoupling. The baryon density controls the baryonic contribution to the pre-recombination plasma and enters the calculation of both the sound horizon and the decoupling redshift. In the likelihood analysis we use the covariance matrix associated with these three quantities, thereby retaining the correlations among them.

DESI DR2 BAO measurements

Baryon acoustic oscillations originate from acoustic waves in the tightly coupled photon-baryon fluid before recombination. After photons decouple, the same physical scale remains imprinted in the late-time distribution of matter. This scale is set by the comoving sound horizon at the baryon drag epoch, rdr_d, and can be used as a standard ruler for cosmological distance measurements.

We employ the full BAO dataset from the second data release of DESI [62, 68]. This compilation contains 13 measurements covering redshifts up to z=2.33z = 2.33. Depending on the tracer and redshift bin, the observational constraints are reported in terms of the dimensionless ratios

DM(z)rd,DH(z)rd,DV(z)rd.(31)\frac{D_M(z)}{r_d},\qquad\frac{D_H(z)}{r_d},\qquad\frac{D_V(z)}{r_d}. \tag*{(31)}

Here DM(z)D_M(z) is the transverse comoving distance, DH(z)D_H(z) is the Hubble distance, and DV(z)D_V(z) is the volume-averaged distance. These quantities probe, respectively, the angular BAO scale, the radial BAO scale, and the isotropic BAO scale. The DESI DR2 measurements therefore provide a precise reconstruction of the expansion history over a wide redshift interval.

Type Ia supernovae

Type Ia supernovae are standardisable candles and provide direct information on the luminosity-distance relation. Their absolute luminosity is not known a priori, but empirical corrections based on light-curve properties allow them to trace relative distances with high precision.

We use the Pantheon+ compilation [63], which contains 1701 supernova light curves in the approximate redshift range

0.001<z<2.26.(32)0.001 < z < 2.26. \tag*{(32)}

In this work the Pantheon+ sample is used in two different ways. First, we consider the Pantheon+ and SH0ES calibrated compilation, which combines the supernova data with Cepheid-based distance-ladder information. We denote this dataset by PPS. Since the SH0ES calibration fixes the absolute magnitude of Type Ia supernovae, this dataset provides an absolute late-time distance probe and is associated with a relatively high value of the Hubble constant.

Second, we also consider a Pantheon+ supernova sample without the SH0ES calibration, denoted by Pan. In the absence of an external calibration, the supernovae constrain the shape of the luminosity-distance relation but not its absolute normalisation. For this case, we use the binned version of the Pantheon+ sample, which consists of 40 data points and provides a compressed description of the supernova Hubble diagram. This binned, uncalibrated supernova information is combined with other probes, such as cosmic chronometers and gamma-ray bursts, when constructing the lower-H0H_0 late-time dataset.

Cosmic chronometers

Cosmic chronometers provide direct measurements of the Hubble expansion rate H(z)H(z). The method is based on the differential ageing of passively evolving galaxies. By estimating the variation of cosmic time with redshift, one obtains

H(z)=−11+zdzdt.(33)H(z)=-\frac{1}{1+z}\frac{dz}{dt}. \tag*{(33)}

This approach gives expansion rate measurements that do not rely on the integrated distance-redshift relation. We adopt the cosmic chronometer compilation of [65]. The sample contains

NCC=33(34)N_{\mathrm{CC}}=33 \tag*{(34)}

measurements covering approximately

0.07≲z≲2.0.(35)0.07\lesssim z\lesssim2.0. \tag*{(35)}

These data are used as part of the CC + GRB + Pan dataset combination.

Gamma-ray bursts

Gamma-ray bursts can be used as high-redshift distance indicators. Although they are less standardised than Type Ia supernovae, they are valuable because they extend the Hubble diagram to significantly higher redshifts. We use the Mayflower gamma-ray burst sample of [66]. This compilation contains

NGRB=79(36)N_{\mathrm{GRB}}=79 \tag*{(36)}

objects in the redshift interval

1.44<z<8.1.(37)1.44<z<8.1. \tag*{(37)}

The GRB data therefore complement the supernova and cosmic chronometer measurements by adding distance information at redshifts where late-time probes are otherwise sparse.

Statistical methodology

We constrain the parameters of the generalised axion-like quintessence model within a Bayesian framework. Given a model MM, a set of parameters θM\theta_M, and a dataset D\mathcal{D}, Bayes’ theorem gives

P(θM∣D,M)=P(D∣θM,M)P(θM∣M)P(D∣M).(38)P(\theta_M\mid\mathcal{D},M)=\frac{P(\mathcal{D}\mid\theta_M,M)P(\theta_M\mid M)}{P(\mathcal{D}\mid M)}. \tag*{(38)}

Here P(θM∣D,M)P(\theta_M\mid\mathcal{D},M) is the posterior probability distribution, P(D∣θM,M)P(\mathcal{D}\mid\theta_M,M) is the likelihood, and P(θM∣M)P(\theta_M\mid M) is the prior probability distribution. The Bayesian evidence 2^2,

P(D∣M)=∫P(D∣θM,M)P(θM∣M) dθM,(39)P(\mathcal{D}\mid M)=\int P(\mathcal{D}\mid\theta_M,M)P(\theta_M\mid M)\,d\theta_M, \tag*{(39)}

acts as the normalisation of the posterior.

The posterior distributions are sampled using the Markov Chain Monte Carlo sampler implemented in Cobaya [69–72]. The resulting chains are analysed and visualised with GetDist [73]. For each cosmological model and dataset combination, we run between two and four independent MCMC chains in order to assess convergence and the numerical stability of the posterior constraints. Convergence is monitored using the Gelman–Rubin criterion [71], which we evaluate both within each individual chain and across independent chains 3^3, requiring

R−1<0.1(40)R-1<0.1 \tag*{(40)}

for intra-chain convergence and

R−1<0.01(41)R - 1 < 0.01 \tag*{(41)}

for inter-chain convergence. After these convergence criteria are satisfied, we discard the earliest 30% of samples in each chain as burn-in. The remaining samples from all chains are then combined and used to compute the marginalised constraints and derived quantities.

Likelihood functions

For all datasets considered in this work, we assume Gaussian likelihoods. The likelihood of a dataset can therefore be written in terms of its chi-squared statistic as

P(D∣θM,M)∝exp⁡(−χ22).(42)P(D|\theta_M, M) \propto\exp\left(-\frac{\chi^2}{2}\right). \tag*{(42)}

When several statistically independent datasets are combined, the total likelihood is the product of the individual likelihoods. Equivalently, the total chi-squared is obtained by summing the relevant contributions. For the dataset combinations considered in this work, the total chi-squared is constructed by summing the statistically independent contributions included in each case. In particular, the late-time sector is taken either from the Pantheon+ & SH0ES calibrated sample or from the combination CC + GRB + Pan, but the two are not used simultaneously. Thus, schematically,

χobs2=χCMB2+χBAO2+χlate2,(43)\chi^2_{\mathrm{obs}} = \chi^2_{\mathrm{CMB}} + \chi^2_{\mathrm{BAO}} + \chi^2_{\mathrm{late}}, \tag*{(43)}

where the relevant terms are included only when present in the dataset combination, and

χlate2={χPPS2,for PPS combinations,χCC2+χGRB2+χPan2,for CC + GRB + Pan combinations.(44)\chi^2_{\mathrm{late}} = \begin{cases} \chi^2_{\mathrm{PPS}}, & \text{for PPS combinations,}\\ \chi^2_{\mathrm{CC}}+\chi^2_{\mathrm{GRB}}+\chi^2_{\mathrm{Pan}}, & \text{for CC + GRB + Pan combinations.} \end{cases} \tag*{(44)}

CMB distance-prior likelihood

The theoretical CMB distance-prior vector is

XCMBth=(R,ℓa,ωb0).(45)X^{\mathrm{th}}_{\mathrm{CMB}} = (R,\ell_a,\omega_{b0}). \tag*{(45)}

The shift parameter is defined as

R(z∗)=Ωm0H02c(1+z∗)DA(z∗),(46)R(z_*) = \frac{\sqrt{\Omega_{m0}H_0^2}}{c}(1+z_*)D_A(z_*), \tag*{(46)}

where z∗z_* is the redshift of photon decoupling and DA(z)D_A(z) is the angular diameter distance.

The acoustic angular scale is

ℓa(z∗)=π(1+z∗)DA(z∗)rs(z∗),(47)\ell_a(z_*) = \pi(1+z_*)\frac{D_A(z_*)}{r_s(z_*)}, \tag*{(47)}

where rs(z∗)r_s(z_*) is the comoving sound horizon at photon decoupling.

For the spatially flat cosmologies studied here, the angular diameter distance is

DA(z)=c1+z∫0zdz′H(z′).(48)D_A(z) = \frac{c}{1+z}\int_0^z \frac{dz'}{H(z')}. \tag*{(48)}

The comoving sound horizon is computed as

rs(z)=∫z∞c dz′H(z′)3[1+Rb(1+z′)−1],(49)r_s(z) = \int_z^\infty\frac{c\,dz'}{H(z')\sqrt{3[1+R_b(1+z')^{-1}]}}, \tag*{(49)}

where [74]

Rb=31500ωb0(TCMB2.7 K)−4.(50)R_b = 31500\omega_{b0}\left(\frac{T_{\mathrm{CMB}}}{2.7\,\mathrm{K}}\right)^{-4}. \tag*{(50)}

We fix the CMB temperature to [75]

TCMB=2.7255 K.(51)T_{\mathrm{CMB}} = 2.7255\,\mathrm{K}. \tag*{(51)}

The photon decoupling redshift is evaluated using the fitting formula [74, 76]

z∗=1048(1+0.00124ωb0−0.738)(1+g1ωm0g2),(52)z_* = 1048\left(1+0.00124\omega_{b0}^{-0.738}\right)\left(1+g_1\omega_{m0}^{g_2}\right), \tag*{(52)}

with

g1=0.0783ωb0−0.238(1+39.5ωb00.763)−1,(53)g_1 = 0.0783\omega_{b0}^{-0.238}\left(1+39.5\omega_{b0}^{0.763}\right)^{-1}, \tag*{(53)}

and

g2=0.560(1+21.1ωb01.81)−1.(54)g_2 = 0.560\left(1+21.1\omega_{b0}^{1.81}\right)^{-1}. \tag*{(54)}

The physical matter density is

ωm0=Ωm0h2.(55)\omega_{m0} = \Omega_{m0}h^2. \tag*{(55)}

The residual vector is defined as

ΔXCMB=XCMBobs−XCMBth,(56)\Delta X_{\mathrm{CMB}} = X^{\mathrm{obs}}_{\mathrm{CMB}} - X^{\mathrm{th}}_{\mathrm{CMB}}, \tag*{(56)}

and the corresponding chi-squared is

χCMB2=ΔXCMBTCCMB−1ΔXCMB,(57)\chi^2_{\mathrm{CMB}} = \Delta X_{\mathrm{CMB}}^{\mathrm{T}}C_{\mathrm{CMB}}^{-1}\Delta X_{\mathrm{CMB}}, \tag*{(57)}

where CCMBC_{\mathrm{CMB}} is the covariance matrix of the CMB distance-prior.

DESI DR2 BAO likelihood

The DESI DR2 BAO likelihood is constructed from measurements of

DM(z)rd,DH(z)rd,DV(z)rd.(58)\frac{D_M(z)}{r_d}, \qquad\frac{D_H(z)}{r_d}, \qquad\frac{D_V(z)}{r_d}. \tag*{(58)}

The Hubble distance is

DH(z)=cH(z),(59)D_H(z) = \frac{c}{H(z)}, \tag*{(59)}

and the transverse comoving distance is

DM(z)=(1+z)DA(z).(60)D_M(z) = (1+z)D_A(z). \tag*{(60)}

The volume-averaged distance is defined by

DV(z)=[zDM2(z)DH(z)]1/3.(61)D_V(z) = \left[zD_M^2(z)D_H(z)\right]^{1/3}. \tag*{(61)}

The sound horizon at the drag epoch is calculated using the fitting expression

rd=147.05 Mpc(ωb00.02236)−0.13(ωm00.1432)−0.23.(62)r_d = 147.05\,\mathrm{Mpc}\left(\frac{\omega_{\mathrm{b0}}}{0.02236}\right)^{-0.13}\left(\frac{\omega_{\mathrm{m0}}}{0.1432}\right)^{-0.23}. \tag*{(62)}

The original fitting relation also includes an additional correction factor depending on the effective number of relativistic species, NeffN_{\mathrm{eff}}. Throughout our analysis we fix this parameter to its standard value, Neff=3.04N_{\mathrm{eff}}=3.04, so that the corresponding factor is equal to unity and is not written explicitly in (62).

The BAO residual vector is

ΔXBAO=XBAOobs−XBAOth.(63)\Delta X_{\mathrm{BAO}} = X_{\mathrm{BAO}}^{\mathrm{obs}} - X_{\mathrm{BAO}}^{\mathrm{th}}. \tag*{(63)}

The BAO chi-squared is then

χBAO2=ΔXBAOTCBAO−1ΔXBAO,(64)\chi_{\mathrm{BAO}}^2 = \Delta X_{\mathrm{BAO}}^{\mathrm{T}} C_{\mathrm{BAO}}^{-1}\Delta X_{\mathrm{BAO}}, \tag*{(64)}

where CBAOC_{\mathrm{BAO}} denotes the DESI DR2 covariance matrix.

Pantheon+ and SH0ES likelihood

We use Type Ia supernovae from the Pantheon+ compilation calibrated with the SH0ES distance ladder. The data are effectively given in terms of the observed apparent magnitude, heliocentric redshift, and CMB-frame redshift, (mi,zhel,i,zCMB,i)(m_i,z_{\mathrm{hel},i},z_{\mathrm{CMB},i}). The observational distance modulus is defined as

μSN,iobs=mi−M,(65)\mu_{\mathrm{SN},i}^{\mathrm{obs}} = m_i-M, \tag*{(65)}

where MM is the standardised absolute magnitude of Type Ia supernovae.

The theoretical distance modulus is obtained from the Hubble-free luminosity distance as

μSNth(zhel,zCMB)=5log⁡10[DLfree(zhel,zCMB)]+μ0,(66)\mu_{\mathrm{SN}}^{\mathrm{th}}(z_{\mathrm{hel}},z_{\mathrm{CMB}}) = 5\log_{10}\left[D_L^{\mathrm{free}}(z_{\mathrm{hel}},z_{\mathrm{CMB}})\right]+\mu_0, \tag*{(66)}

where

μ0≡5log⁡10[cH0−1 Mpc−1]+25(67)\mu_0 \equiv5\log_{10}\left[cH_0^{-1}\,\mathrm{Mpc}^{-1}\right]+25 \tag*{(67)}

collects the unit conversion from Mpc to the reference distance of 10 pc. The dimensionless Hubble-free luminosity distance is

DLfree(zhel,zCMB)=(1+zhel)∫0zCMBH0H(z′) dz′.(68)D_L^{\mathrm{free}}(z_{\mathrm{hel}},z_{\mathrm{CMB}})=(1+z_{\mathrm{hel}})\int_0^{z_{\mathrm{CMB}}}\frac{H_0}{H(z')}\,\mathrm{d}z'. \tag*{(68)}

The Pantheon+ & SH0ES compilation includes a set of Cepheid-host supernovae that calibrate the distance scale through the SH0ES distance ladder. For Hubble-flow supernovae, the standard cosmological prediction above is used. For Cepheid-host supernovae, the distance entering the likelihood is instead replaced by the measured Cepheid-based host distance modulus. This calibration is incorporated through the full Pantheon+ & SH0ES covariance matrix, which accounts for statistical and systematic uncertainties as well as correlations among supernovae.

Defining the residual vector

ΔμSN=μSNobs−μSNth,(69)\Delta\mu_{\mathrm{SN}}=\mu_{\mathrm{SN}}^{\mathrm{obs}}-\mu_{\mathrm{SN}}^{\mathrm{th}}, \tag*{(69)}

the Pantheon+ & SH0ES chi-squared is written as

χPPS2=ΔμSNTCPPS−1ΔμSN,(70)\chi_{\mathrm{PPS}}^2=\Delta\mu_{\mathrm{SN}}^{\mathrm{T}}C_{\mathrm{PPS}}^{-1}\Delta\mu_{\mathrm{SN}}, \tag*{(70)}

where CPPSC_{\mathrm{PPS}} is the covariance matrix of the Pantheon+ & SH0ES dataset.

Uncalibrated Pantheon+ likelihood

For the Pan dataset, we use the binned Pantheon+ sample without the SH0ES Cepheid calibration. This compressed sample consists of 40 measurements of the supernova distance modulus and provides a binned representation of the Pantheon+ Hubble diagram. The observed distance modulus of each binned supernova point is defined, as previously, as

μSN,iobs=mi−M,(71)\mu_{\mathrm{SN},i}^{\mathrm{obs}}=m_i-M, \tag*{(71)}

where mim_i is the observed apparent magnitude of the iith binned data point and MM is the standardised absolute magnitude of Type Ia supernovae. For a given cosmological model, the theoretical prediction is computed from the dimensionless Hubble-free luminosity distance defined above,

μSNth(zhel,zCMB)=5log⁡10[DLfree(zhel,zCMB)]+μ0,(72)\mu_{\mathrm{SN}}^{\mathrm{th}}(z_{\mathrm{hel}},z_{\mathrm{CMB}})=5\log_{10}\left[D_L^{\mathrm{free}}(z_{\mathrm{hel}},z_{\mathrm{CMB}})\right]+\mu_0, \tag*{(72)}

with the same definition of μ0\mu_0 as in the Pantheon+ & SH0ES case.

Defining the residual vector

ΔμSN=μSNobs−μSNth,(73)\Delta\mu_{\mathrm{SN}}=\mu_{\mathrm{SN}}^{\mathrm{obs}}-\mu_{\mathrm{SN}}^{\mathrm{th}}, \tag*{(73)}

the corresponding chi-squared statistic is

χPan2=ΔμSNTCPan−1ΔμSN,(74)\chi_{\mathrm{Pan}}^2=\Delta\mu_{\mathrm{SN}}^{\mathrm{T}}C_{\mathrm{Pan}}^{-1}\Delta\mu_{\mathrm{SN}}, \tag*{(74)}

where CPanC_{\mathrm{Pan}} is the covariance matrix of the binned Pantheon+ sample.

Since the uncalibrated binned Pantheon+ sample does not fix the absolute magnitude scale, the likelihood depends on the nuisance parameter M‾≡M+μ0\overline{M}\equiv M+\mu_0. Equivalently, the residual can be written as

μSN,iobs−μSN,ith=mi−5log⁡10[DLfree(zhel,i,zCMB,i)]−M‾.(75)\mu_{\mathrm{SN},i}^{\mathrm{obs}}-\mu_{\mathrm{SN},i}^{\mathrm{th}}=m_i-5\log_{10}\left[D_L^{\mathrm{free}}(z_{\mathrm{hel},i},z_{\mathrm{CMB},i})\right]-\overline{M}. \tag*{(75)}

We marginalise analytically over M~\widetilde{M}. Following [77], the resulting marginalised statistic is

χ~Pan2=APan+ln⁡(EPan2π)−BPan2EPan,(76)\widetilde{\chi}^{2}_{\mathrm{Pan}}=A_{\mathrm{Pan}}+\ln\left(\frac{E_{\mathrm{Pan}}}{2\pi}\right)-\frac{B_{\mathrm{Pan}}^{2}}{E_{\mathrm{Pan}}}, \tag*{(76)}

where

APan=Δμ~SNTCPan−1Δμ~SN,BPan=Δμ~SNTCPan−11,EPan=1TCPan−11,Δμ~SN=ΔμSN(M~=0).(77)\begin{aligned} A_{\mathrm{Pan}}&=\Delta\widetilde{\mu}_{\mathrm{SN}}^{\mathsf{T}}C_{\mathrm{Pan}}^{-1}\Delta\widetilde{\mu}_{\mathrm{SN}},\\ B_{\mathrm{Pan}}&=\Delta\widetilde{\mu}_{\mathrm{SN}}^{\mathsf{T}}C_{\mathrm{Pan}}^{-1}\mathbf{1},\\ E_{\mathrm{Pan}}&=\mathbf{1}^{\mathsf{T}}C_{\mathrm{Pan}}^{-1}\mathbf{1},\\ \Delta\widetilde{\mu}_{\mathrm{SN}}&=\Delta\mu_{\mathrm{SN}}(\widetilde{M}=0). \tag*{(77)} \end{aligned}

In the present analysis, the uncalibrated binned Pantheon+ sample is always combined with cosmic chronometer measurements, which provide an independent calibration of the expansion scale. This dataset combination is also always used together with the GRB Hubble-diagram sample.

Cosmic chronometer likelihood

The cosmic chronometer data provide direct measurements of H(z)H(z). For each redshift ziz_i in the compilation, the theoretical prediction is obtained by evaluating the background expansion rate of the model at that redshift. Assuming uncorrelated measurements, the chi-squared is

χCC2=∑i=1NCC[Hobs(zi)−Hth(zi)]2σH,i2,(78)\chi^{2}_{\mathrm{CC}}=\sum_{i=1}^{N_{\mathrm{CC}}}\frac{\left[H^{\mathrm{obs}}(z_i)-H^{\mathrm{th}}(z_i)\right]^{2}}{\sigma_{H,i}^{2}}, \tag*{(78)}

where σH,i\sigma_{H,i} is the uncertainty of the iith measurement.

Gamma-ray burst likelihood

The GRB dataset is provided in terms of the observed distance modulus μGRBobs\mu_{\mathrm{GRB}}^{\mathrm{obs}} at a given redshift, together with the corresponding uncertainty. For a given cosmological model, we compute the theoretical prediction μGRBth(z)\mu_{\mathrm{GRB}}^{\mathrm{th}}(z) from the luminosity distance in a way analogous to the one used for Type Ia supernovae, except that there is no need to distinguish between heliocentric and CMB-frame redshifts. Thus, we use

μGRBth(z)=5log⁡10[DLfree(z)]+μ0,(79)\mu_{\mathrm{GRB}}^{\mathrm{th}}(z)=5\log_{10}\left[D_{L}^{\mathrm{free}}(z)\right]+\mu_{0}, \tag*{(79)}

where DLfree(z)D_{L}^{\mathrm{free}}(z) denotes the dimensionless Hubble-free luminosity distance.

Defining the residual vector

ΔμGRB=μGRBobs−μGRBth,(80)\Delta\mu_{\mathrm{GRB}}=\mu_{\mathrm{GRB}}^{\mathrm{obs}}-\mu_{\mathrm{GRB}}^{\mathrm{th}}, \tag*{(80)}

the corresponding chi-squared statistic is

χGRB2=ΔμGRBTCGRB−1ΔμGRB,(81)\chi^{2}_{\mathrm{GRB}}=\Delta\mu_{\mathrm{GRB}}^{\mathsf{T}}C_{\mathrm{GRB}}^{-1}\Delta\mu_{\mathrm{GRB}}, \tag*{(81)}

where CGRBC_{\mathrm{GRB}} is the covariance matrix of the GRB sample. In this work, we assume this matrix to be diagonal, CGRB=diag⁡(σμ,i2)C_{\mathrm{GRB}}=\operatorname{diag}(\sigma_{\mu,i}^{2}), with σμ,i\sigma_{\mu,i} the reported uncertainty of the iith data point.

Since μGRB\mu_{\mathrm{GRB}} depends on the nuisance parameter μ0\mu_{0}, we marginalise analytically over μ0\mu_{0}. Following [77], the resulting marginalised statistic is

χ~GRB2=AGRB+ln⁡(EGRB2π)−BGRB2EGRB,(82)\widetilde{\chi}^{2}_{\mathrm{GRB}}=A_{\mathrm{GRB}}+\ln\left(\frac{E_{\mathrm{GRB}}}{2\pi}\right)-\frac{B_{\mathrm{GRB}}^{2}}{E_{\mathrm{GRB}}}, \tag*{(82)}

where

AGRB=Δμ~GRBTCGRB−1Δμ~GRB,BGRB=Δμ~GRBTCGRB−11,EGRB=1TCGRB−11,Δμ~GRB=ΔμGRB(μ0=0).(83)\begin{aligned} A_{\mathrm{GRB}}&=\Delta\widetilde{\mu}_{\mathrm{GRB}}^{\mathsf{T}}C_{\mathrm{GRB}}^{-1}\Delta\widetilde{\mu}_{\mathrm{GRB}},\\ B_{\mathrm{GRB}}&=\Delta\widetilde{\mu}_{\mathrm{GRB}}^{\mathsf{T}}C_{\mathrm{GRB}}^{-1}\mathbf{1},\\ E_{\mathrm{GRB}}&=\mathbf{1}^{\mathsf{T}}C_{\mathrm{GRB}}^{-1}\mathbf{1},\\ \Delta\widetilde{\mu}_{\mathrm{GRB}}&=\Delta\mu_{\mathrm{GRB}}(\mu_{0}=0). \tag*{(83)} \end{aligned}

Parameter space and prior specification

The parameter space explored in this work consists of the standard late-time cosmological parameters together with a set of quantities describing the scalar potential of the generalised axion-like dark energy model. The cosmological parameters are the Hubble constant H0H_{0} and the present-day matter and baryon density parameters Ωm0\Omega_{\mathrm{m}0} and Ωb0\Omega_{\mathrm{b}0}.4 The additional parameters determine the shape of the scalar potential and the initial conditions of the scalar field evolution.

For the background cosmological sector we adopt broad flat priors on the parameters H0H_{0}, Ωm0\Omega_{\mathrm{m}0}, and Ωb0\Omega_{\mathrm{b}0}. In particular we impose the cut 40<H0<10040 < H_{0} < 100 on the Hubble constant, while the matter and baryon density parameters are sampled within the ranges 0.1<Ωm0<0.90.1 < \Omega_{\mathrm{m}0} < 0.9 and 0.01<Ωb0<0.090.01 < \Omega_{\mathrm{b}0} < 0.09.

We also include a Big Bang Nucleosynthesis (BBN) constraint on the physical baryon density ωb0=Ωb0h2\omega_{\mathrm{b}0}=\Omega_{\mathrm{b}0}h^{2} using the measurement

ωb0=0.0222±0.0005\omega_{\mathrm{b}0}=0.0222\pm0.0005

from [78]. This constraint is implemented as a Gaussian likelihood term.

Since the background expansion history is obtained by solving the scalar field equation of motion, we additionally enforce consistency between the sampled value of the Hubble constant and the value obtained from the numerical evolution. To this end we introduce a Gaussian constraint on

EH0≡1−H0,EOM2H02,\mathcal{E}_{H_0} \equiv1-\frac{H_{0,\mathrm{EOM}}^2}{H_0^2},

where H0,EOMH_{0,\mathrm{EOM}} denotes the Hubble constant derived from the equations of motion. The Gaussian width is chosen to be 0.001/20.001/\sqrt{2}, effectively enforcing the condition H0,EOM≃H0H_{0,\mathrm{EOM}}\simeq H_0.

Both Gaussian constraints are included in the likelihood used for parameter inference, ensuring that the sampled cosmological models satisfy the BBN measurement and the present-day Friedmann constraint. However, their direct contributions are excluded from the observational likelihood employed in the statistical model comparison and dataset concordance estimators introduced in the following subsection. In this way, the BBN measurement is treated as external information, while the Friedmann constraint restricts the allowed parameter space of the generalised axion-like model.

The dark energy sector follows the scalar field model described in Section II. The parameters controlling the scalar potential and the initial field value are sampled using flat priors within the ranges 10−4<n<110^{-4}<n<1, −2<log⁡10(Vmin⁡/ρφ0)<0-2<\log_{10}(V_{\min}/\rho_{\varphi0})<0, −2<log⁡10(f/MPl)<1-2<\log_{10}(f/M_{\mathrm{Pl}})<1, and −4.5<log⁡10(ϕi/f)<0.5-4.5<\log_{10}(\phi_i/f)<0.5, where the subscript 0 denotes present-day quantities and ii refers to the initial value at the starting scale factor aia_i. The field evolution is initialised at an early epoch, ai/a0=10−6a_i/a_0=10^{-6}, with vanishing initial field velocity,

ϕ˙(ai)=0,\dot{\phi}(a_i)=0,

which is consistent with the freezing behaviour of the scalar field during the radiation-dominated era.

The parameters and their prior ranges adopted in the analysis are summarised in Table I. The first set corresponds to the cosmological parameters common to both Λ\LambdaCDM and the scalar field model, while the remaining parameters describe the scalar potential and the initial conditions of the field.

ParameterMin.Max.
H0H_040100
Ωm0\Omega_{\mathrm{m}0}0.10.9
Ωb0\Omega_{\mathrm{b}0}0.010.09
nn10−410^{-4}1
log⁡10(Vmin⁡/ρφ0)\log_{10}(V_{\min}/\rho_{\varphi0})−2-20
log⁡10(f/MPl)\log_{10}(f/M_{\mathrm{Pl}})−2-21
log⁡10(ϕi/f)\log_{10}(\phi_i/f)−4.5-4.50.5

Table I. Prior ranges adopted for the cosmological and scalar field parameters of the generalised axion-like model. Flat priors are used in all cases.

For each sampled parameter set we solve the scalar field equation of motion from the initial scale factor aia_i to the present-time. The resulting background evolution is then used to compute the theoretical predictions entering the likelihood functions described in the previous subsection.

Model comparison and dataset concordance

To evaluate the observational performance of the generalised axion-like quintessence model relative to Λ\LambdaCDM, we employ a collection of statistical estimators that account for both the quality of the fit and the complexity of the cosmological model. These diagnostics are particularly useful when comparing models with different numbers of free parameters, since additional degrees of freedom may improve the maximum likelihood without necessarily providing a statistically preferred description of the data.

Throughout the following analysis, we distinguish between the observational likelihood associated with a given dataset combination and the auxiliary Gaussian constraints described in Section IV B. The latter are retained when constructing the posterior distribution and determining the best-fitting parameters, but their direct contributions are excluded from the likelihood entering the statistical estimators.

Denoting the observational dataset by D\mathcal{D} and the auxiliary constraints by A\mathcal{A}, the total likelihood can be written as

P(D,A∣θM,M)=P(D∣θM,M)P(A∣θM,M).(84)P(\mathcal{D},\mathcal{A}\mid\theta_M,M)=P(\mathcal{D}\mid\theta_M,M)P(\mathcal{A}\mid\theta_M,M). \tag*{(84)}

For the generalised axion-like model, the auxiliary factor contains both the BBN and Friedmann consistency likelihoods, whereas for Λ\LambdaCDM only the BBN constraint is required.

Accordingly, the observational chi-squared employed in our statistical diagnostics is obtained by removing the auxiliary contributions from the total chi-squared,

χobs2=χtot2−χBBN2−χEH02,(85)\chi^2_{\mathrm{obs}}=\chi^2_{\mathrm{tot}}-\chi^2_{\mathrm{BBN}}-\chi^2_{\mathcal{E}_{H_0}}, \tag*{(85)}

where the last term is absent for Λ\LambdaCDM. The posterior distribution nevertheless retains the information provided by the auxiliary constraints. All posterior-based estimators are therefore evaluated conditionally on A\mathcal{A}.

We consider estimators constructed from the maximum likelihood, namely the corrected Akaike information criterion (AICc) and the Bayesian information criterion (BIC), together with posterior-based quantities such as the Bayesian evidence, the deviance information criterion (DIC), and the widely applicable information criterion (WAIC). These complementary approaches allow us to examine the preference between the generalised axion-like model and Λ\LambdaCDM from different statistical perspectives.

The maximum-likelihood estimators are evaluated using the observational likelihood at the best-fitting parameter values obtained with the minimize sampler implemented in Cobaya [79–81]. The auxiliary Gaussian constraints are retained during minimisation, while their direct contributions are excluded from the likelihood entering the information criteria. The corrected Akaike information criterion is defined as [82, 83]

AICc=−2ln⁡P(D∣θM,bf,M)+2k+2k(k+1)N−k−1.(86)\mathrm{AIC}_{\mathrm{c}}=-2\ln P(\mathcal{D}\mid\theta_{M,\mathrm{bf}},M)+2k+\frac{2k(k+1)}{N-k-1}. \tag*{(86)}

where θM,bf\theta_{M,\mathrm{bf}} denotes the best-fitting parameter vector obtained under the auxiliary constraints, kk is the number of independent free parameters of the cosmological model5, and NN is the total number of observational data points in the corresponding dataset combination. The final term corrects for the finite size of the dataset and becomes negligible when N≫k2N\gg k^{2}, recovering the standard Akaike information criterion,

AIC=−2ln⁡P(D∣θM,bf,M)+2k.(87)\mathrm{AIC}=-2\ln P(\mathcal{D}\mid\theta_{M,\mathrm{bf}},M)+2k. \tag*{(87)}

In this way, AICc\mathrm{AIC}_{\mathrm{c}} penalises additional parameters while retaining the information provided by the best-fitting solution. The Bayesian information criterion is given by [84]

BIC=−2ln⁡P(D∣θM,bf,M)+kln⁡N.(88)\mathrm{BIC}=-2\ln P(\mathcal{D}\mid\theta_{M,\mathrm{bf}},M)+k\ln N. \tag*{(88)}

Although both criteria depend on the maximum likelihood, their complexity penalties differ. In particular, the BIC penalty increases logarithmically with the number of data points and generally becomes more restrictive than that of AICc\mathrm{AIC}_{\mathrm{c}} for sufficiently large datasets.

We complement these maximum-likelihood criteria with estimators that depend on the posterior distribution. For any function of the model parameters, we define the posterior average conditional on the auxiliary constraints as

⟨⋯ ⟩M∣D,A≡∫(⋯ )P(θM∣D,A,M) dθM.(89)\langle\cdots\rangle_{M\mid\mathcal{D},\mathcal{A}}\equiv\int(\cdots)P(\theta_{M}\mid\mathcal{D},\mathcal{A},M)\,\mathrm{d}\theta_{M}. \tag*{(89)}

The posterior distribution appearing in this expression is obtained from the complete likelihood, including the auxiliary constraints, whereas the log-likelihood functions entering the statistical estimators contain only the observational contributions. Using this notation, we define the posterior expectation value of the log-likelihood as

F(D∣A,M)≡⟨ln⁡P(D∣θM,M)⟩M∣D,A.(90)F(\mathcal{D}\mid\mathcal{A},M)\equiv\langle\ln P(\mathcal{D}\mid\theta_{M},M)\rangle_{M\mid\mathcal{D},\mathcal{A}}. \tag*{(90)}

Another useful posterior quantity is the Bayesian model dimensionality (BMD) [85], defined as

BMD(D∣A,M)=2⟨[ln⁡P(D∣θM,M)]2⟩M∣D,A−2⟨ln⁡P(D∣θM,M)⟩M∣D,A2.(91)\mathrm{BMD}(\mathcal{D}\mid\mathcal{A},M)=2\left\langle\left[\ln P(\mathcal{D}\mid\theta_{M},M)\right]^{2}\right\rangle_{M\mid\mathcal{D},\mathcal{A}}-2\left\langle\ln P(\mathcal{D}\mid\theta_{M},M)\right\rangle^{2}_{M\mid\mathcal{D},\mathcal{A}}. \tag*{(91)}

This estimator characterises the effective dimensionality of the parameter space constrained by the observations, conditional on the auxiliary information. Unlike the nominal number of free parameters, the BMD accounts for the information contained in the posterior and can therefore be smaller when some parameter combinations remain poorly determined by the data.

We next consider the Bayesian evidence, which measures the statistical support for a cosmological model after integrating the likelihood over its prior parameter space. Since the auxiliary Gaussian constraints are included in the parameter inference, we work with the Bayesian evidence conditional on these constraints.

Let P(θM∣M)P(\theta_{M}\mid M) denote the parameter prior, and let A\mathcal{A} represent the auxiliary information introduced in Section IV B. We define the corresponding normalised effective prior as

P(θM∣A,M)≡P(A∣θM,M)P(θM∣M)VA,M,(92)P(\theta_{M}\mid\mathcal{A},M)\equiv\frac{P(\mathcal{A}\mid\theta_{M},M)P(\theta_{M}\mid M)}{V_{\mathcal{A},M}}, \tag*{(92)}

where

VA,M=∫ΠMP(A∣θM,M)P(θM∣M) dθM.(93)V_{\mathcal{A},M}=\int_{\Pi_{M}}P(\mathcal{A}\mid\theta_{M},M)P(\theta_{M}\mid M)\,\mathrm{d}\theta_{M}. \tag*{(93)}

The conditional Bayesian evidence is then defined as

P(D∣A,M)≡∫ΠMP(D∣θM,M)×P(θM∣A,M) dθM.(94)P(\mathcal{D}\mid\mathcal{A},M)\equiv\int_{\Pi_{M}}P(\mathcal{D}\mid\theta_{M},M)\times P(\theta_{M}\mid\mathcal{A},M)\,\mathrm{d}\theta_{M}. \tag*{(94)}

Equivalently, it can be expressed as the ratio

P(D∣A,M)=P(D,A∣M)P(A∣M),(95)P(\mathcal{D}\mid\mathcal{A},M)=\frac{P(\mathcal{D},\mathcal{A}\mid M)}{P(\mathcal{A}\mid M)}, \tag*{(95)}

where P(D,A∣M)P(\mathcal{D},\mathcal{A}\mid M) denotes the joint evidence obtained by integrating the product of the observational and auxiliary likelihoods over the parameter prior, while P(A∣M)=VA,MP(\mathcal{A}\mid M)=V_{\mathcal{A},M} is the evidence associated with the auxiliary constraints alone.

For consistency with the information criteria, we introduce the rescaled logarithmic conditional evidence

ln⁡B(D∣A,M)≡2ln⁡P(D∣A,M),(96)\ln\mathrm{B}(\mathcal{D}\mid\mathcal{A},M)\equiv2\ln P(\mathcal{D}\mid\mathcal{A},M), \tag*{(96)}

so that

−ln⁡B(D∣A,M)=−2ln⁡P(D∣A,M).(97)-\ln\mathrm{B}(\mathcal{D}\mid\mathcal{A},M)=-2\ln P(\mathcal{D}\mid\mathcal{A},M). \tag*{(97)}

Using Bayes’ theorem, the inverse conditional evidence can be expressed as a posterior average,

P(D∣A,M)−1=⟨[P(D∣θM,M)]−1⟩M∣D,A.(98)P(\mathcal{D}\mid\mathcal{A},M)^{-1}=\left\langle\left[P(\mathcal{D}\mid\theta_{M},M)\right]^{-1}\right\rangle_{M\mid\mathcal{D},\mathcal{A}}. \tag*{(98)}

Consequently,

−ln⁡B(D∣A,M)=2ln⁡⟨[P(D∣θM,M)]−1⟩M∣D,A.(99)-\ln\mathrm{B}(\mathcal{D}\mid\mathcal{A},M)=2\ln\left\langle\left[P(\mathcal{D}\mid\theta_{M},M)\right]^{-1}\right\rangle_{M\mid\mathcal{D},\mathcal{A}}. \tag*{(99)}

We also employ the deviance information criterion, which combines the likelihood evaluated at the maximum-a-posteriori point with the posterior-averaged log-likelihood. Following [86], we define

DIC⁡(D∣A,M)=2ln⁡P(D∣θM,map,M)−4F(D∣A,M).(100)\operatorname{DIC}(\mathcal{D}|\mathcal{A},\mathcal{M}) = 2\ln P(\mathcal{D}|\theta_{\mathcal{M},\mathrm{map}},\mathcal{M}) - 4F(\mathcal{D}|\mathcal{A},\mathcal{M}). \tag*{(100)}

where θM,map\theta_{\mathcal{M},\mathrm{map}} denotes the parameter vector that maximises the posterior probability, including the auxiliary Gaussian constraints described in Section IV B. This point is determined using the minimize sampler implemented in Cobaya [79–81]. However, the likelihood entering the DIC is evaluated using only the observational contributions, consistently with the conditional framework adopted throughout this analysis. The DIC incorporates the effective complexity of the posterior distribution, rather than penalising the model solely according to its nominal number of parameters.

Finally, following the posterior-based convention adopted here, we define the widely applicable information criterion as [87]

WAIC⁡(D∣A,M)=−2F(D∣A,M)+BMD⁡(D∣A,M).(101)\operatorname{WAIC}(\mathcal{D}|\mathcal{A},\mathcal{M}) = -2F(\mathcal{D}|\mathcal{A},\mathcal{M}) + \operatorname{BMD}(\mathcal{D}|\mathcal{A},\mathcal{M}). \tag*{(101)}

This estimator includes both the average log-likelihood and the posterior fluctuations of the log-likelihood, thereby incorporating information about the distribution of statistically allowed parameter values.

To compare the generalised axion-like quintessence model with the reference Λ\LambdaCDM cosmology, we introduce

ΔIC⁡≡IC⁡axion−IC⁡ΛCDM,(102)\Delta\operatorname{IC} \equiv\operatorname{IC}_{\mathrm{axion}} - \operatorname{IC}_{\Lambda\mathrm{CDM}}, \tag*{(102)}

where

IC⁡∈{AIC⁡C,BIC⁡,−ln⁡B,DIC⁡,WAIC⁡}.(103)\operatorname{IC} \in\{\operatorname{AIC}_{\mathrm{C}}, \operatorname{BIC}, -\ln B, \operatorname{DIC}, \operatorname{WAIC}\}. \tag*{(103)}

According to this convention, ΔIC⁡>0\Delta\operatorname{IC} > 0 indicates a preference for Λ\LambdaCDM, whereas ΔIC⁡<0\Delta\operatorname{IC} < 0 favours the generalised axion-like quintessence model. We interpret the differences according to the Jeffreys’-scale convention summarised in Table II.

ΔIC⁡\Delta\operatorname{IC}Interpretation
>10> 10Strongly disfavoured / tensioned
5∼105 \sim10Moderately disfavoured / tensioned
2∼52 \sim5Weakly disfavoured / tensioned
−2∼2-2 \sim2Inconclusive
−5∼−2-5 \sim-2Weakly favoured / aligned
−10∼−5-10 \sim-5Moderately favoured / aligned
<−10< -10Strongly favoured / aligned

Table II. Interpretation of information criterion differences and dataset concordance diagnostics following a Jeffreys’-scale convention. Positive values indicate that the generalised axion-like quintessence model is disfavoured relative to Λ\LambdaCDM, or that the datasets exhibit more tension. Negative values correspond to model preference or improved dataset agreement, whereas values close to zero are inconclusive.

Beyond the relative statistical preference between cosmological models, we investigate whether the generalised axion-like quintessence scenario changes the consistency between different observational datasets. This question is particularly relevant in the context of the Hubble tension, since the distance-ladder measurements and the combination of CMB and BAO observations favour different values of the present expansion rate.

We therefore examine the concordance between the CMB+BAO dataset and the two alternative late-time combinations considered in this work, namely PPS and CC+GRB+Pan. The latter two are analysed separately and are never included simultaneously in the same dataset combination. To characterise their compatibility, we consider the Bayesian evidence ratio, the goodness-of-fit statistic, and suspiciousness.

For two observational datasets D1\mathcal{D}_1 and D2\mathcal{D}_2, we define the Bayesian concordance ratio conditionally on the auxiliary constraints. The corresponding rescaled logarithmic statistic is

−ln⁡R(D1,D2∣A,M)=−ln⁡B(D1D2∣A,M)+ln⁡B(D1∣A,M)+ln⁡B(D2∣A,M).(104)-\ln\mathcal{R}(\mathcal{D}_1,\mathcal{D}_2|\mathcal{A},\mathcal{M}) = -\ln B(\mathcal{D}_1\mathcal{D}_2|\mathcal{A},\mathcal{M}) + \ln B(\mathcal{D}_1|\mathcal{A},\mathcal{M}) + \ln B(\mathcal{D}_2|\mathcal{A},\mathcal{M}). \tag*{(104)}

Here, the same auxiliary information A\mathcal{A} is imposed in the separate and joint dataset analyses. The resulting statistic measures the concordance between the observational datasets after conditioning on the BBN constraint and, for the generalised axion-like model, on the present-day Friedmann consistency condition.

To assess whether the generalised axion-like quintessence scenario modifies the dataset agreement relative to Λ\LambdaCDM, we define

Δ(−ln⁡R)=[−ln⁡R]axion−[−ln⁡R]ΛCDM,(105)\Delta(-\ln\mathcal{R}) = [-\ln\mathcal{R}]_{\mathrm{axion}} - [-\ln\mathcal{R}]_{\Lambda\mathrm{CDM}}, \tag*{(105)}

where both terms are evaluated using their corresponding conditional Bayesian evidences. A positive value indicates that the generalised axion-like model increases the discrepancy between the datasets compared with Λ\LambdaCDM, whereas a negative value indicates an improvement in their concordance. The magnitude of this difference is interpreted using Table II.

We further consider the goodness-of-fit concordance statistic [88, 89], which quantifies the loss in best-fit quality when two datasets are described by a common set of cosmological parameters. It is defined as

GoF⁡(D1,D2∣M)=−ln⁡P(D1D2∣θM,map12,M)+ln⁡P(D1∣θM,map1,M)+ln⁡P(D2∣θM,map2,M),(106)\begin{aligned} \operatorname{GoF}(\mathcal{D}_1,\mathcal{D}_2|\mathcal{M}) ={}& -\ln P(\mathcal{D}_1\mathcal{D}_2|\theta_{\mathcal{M},\mathrm{map}}^{12},\mathcal{M}) \\ &+ \ln P(\mathcal{D}_1|\theta_{\mathcal{M},\mathrm{map}}^{1},\mathcal{M}) \\ &+ \ln P(\mathcal{D}_2|\theta_{\mathcal{M},\mathrm{map}}^{2},\mathcal{M}), \tag*{(106)} \end{aligned}

where θM,map12\theta^{12}_{M,\mathrm{map}} denotes the maximum-a-posteriori parameter vector obtained from the combined dataset, while θM,map1\theta^{1}_{M,\mathrm{map}} and θM,map2\theta^{2}_{M,\mathrm{map}} correspond to the separate analyses6. This statistic measures the extent to which combining the datasets forces a deterioration of the best-fitting solutions.

Another concordance diagnostic is suspiciousness, which is constructed from the posterior-averaged log-likelihoods rather than their values at a single point in parameter space [89]. It is given by

S(D1,D2∣A,M)≡−F(D1D2∣A,M)+F(D1∣A,M)+F(D2∣A,M).(107)S(\mathcal{D}_1,\mathcal{D}_2|\mathcal{A},M) \equiv-F(\mathcal{D}_1\mathcal{D}_2|\mathcal{A},M) +F(\mathcal{D}_1|\mathcal{A},M)+F(\mathcal{D}_2|\mathcal{A},M). \tag*{(107)}

Since suspiciousness incorporates the likelihood information averaged over the posterior, it provides a complementary assessment of dataset compatibility to that obtained from the goodness-of-fit statistic.

To quantify the significance associated with these concordance estimators, we introduce the effective Bayesian model dimensionality of the dataset comparison, conditional on the auxiliary constraints,

BMDeff=−BMD(D1D2∣A,M)+BMD(D1∣A,M)+BMD(D2∣A,M).(108)\mathrm{BMD}_{\mathrm{eff}}=-\mathrm{BMD}(\mathcal{D}_1\mathcal{D}_2|\mathcal{A},M) +\mathrm{BMD}(\mathcal{D}_1|\mathcal{A},M) +\mathrm{BMD}(\mathcal{D}_2|\mathcal{A},M). \tag*{(108)}

Under the Gaussian approximation, the goodness-of-fit and suspiciousness can be related to an effective chi-squared statistic through

QC=BMDeff+2C,(109)Q_C=\mathrm{BMD}_{\mathrm{eff}}+2C, \tag*{(109)}

where CC denotes either GoF or S. The quantity QCQ_C approximately follows a chi-squared distribution with BMDeff\mathrm{BMD}_{\mathrm{eff}} effective degrees of freedom [90, 91].

The corresponding tension can be expressed as an equivalent one-dimensional Gaussian significance,

σC=CDF1−1[CDFBMDeff(QC)],(110)\sigma_C=\mathrm{CDF}^{-1}_{1}\left[\mathrm{CDF}_{\mathrm{BMD}_{\mathrm{eff}}}\left(\sqrt{Q_C}\right)\right], \tag*{(110)}

where CDFd(r)\mathrm{CDF}_{d}(r) is the radial cumulative distribution function of a dd-dimensional standard Gaussian,

CDFd(r)=12d/2Γ(d/2)∫0r2xd/2−1e−x/2 dx,(111)\mathrm{CDF}_{d}(r)=\frac{1}{2^{d/2}\Gamma(d/2)}\int_{0}^{r^2}x^{d/2-1}e^{-x/2}\,dx, \tag*{(111)}

with

Γ(x)=∫0∞tx−1e−t dt(112)\Gamma(x)=\int_{0}^{\infty}t^{x-1}e^{-t}\,dt \tag*{(112)}

denoting the Gamma function.

The information criteria and concordance estimators provide different but complementary tests of the generalised axion-like quintessence model. While the former establish whether the additional scalar field parameters lead to a statistically justified improvement over Λ\LambdaCDM, the latter determine whether the model changes the level of consistency between cosmological observations obtained at different redshifts.

In practice, we evaluate these estimators using the independent MCMC chains generated for each cosmological model and dataset combination. Denoting the number of available chains by NcN_c, which varies between two and four in our analysis, we assess the stability of the numerical results by comparing the values obtained from the different chains. This procedure is applied to AICC_C, BIC, Bayesian evidence, DIC, WAIC, Bayes ratio, goodness-of-fit, and suspiciousness.

For a generic estimator O\mathcal{O}, with Oi\mathcal{O}_i representing the value obtained from the iith chain, we define the inter-chain mean as

O‾=1Nc∑i=1NcOi,(113)\overline{\mathcal{O}}=\frac{1}{N_c}\sum_{i=1}^{N_c}\mathcal{O}_i, \tag*{(113)}

and the corresponding standard deviation as

sO=1Nc−1∑i=1Nc(Oi−O‾)2.(114)s_{\mathcal{O}}=\sqrt{\frac{1}{N_c-1}\sum_{i=1}^{N_c}\left(\mathcal{O}_i-\overline{\mathcal{O}}\right)^2}. \tag*{(114)}

All chains contribute with equal weight to these estimates. Unless explicitly stated otherwise, the statistical estimators are reported in the form O‾±sO\overline{\mathcal{O}}\pm s_{\mathcal{O}}.

Observational constraints of our model

In this section, we present the result of MCMC according to the methodology presented in section IV. As mentioned previously, the dataset we consider are grouped into 2 categories: CMB-excluding datasets of BAO, PPS, CC+GRB+Pan, and CMB-including datasets of CMB+BAO, CMB+BAO+PPS, CMB+BAO+CC+GRB+Pan, with CMB representing CMB distance prior [67] rather than the full CMB power spectrum fit.

The 68% confidence intervals of cosmological parameters and the tension probes for Λ\LambdaCDM model and the axion-like dark energy model considered in Section II given various combination of datasets are shown in Table III and Table IV.

ParameterBAOCMB+BAOPPSCC+GRB+PanCMB+BAO+PPSCMB+BAO+CC+GRB+Pan
Λ\LambdaCDM
H0H_068.62±0.5168.62 \pm0.5168.40±0.3068.40 \pm0.3073.6±1.073.6 \pm1.068.9±1.968.9 \pm1.968.71±0.2768.71 \pm0.2768.42±0.2868.42 \pm0.28
Ωm0\Omega_{\mathrm{m}0}0.2975±0.00890.2975 \pm0.00890.3011±0.00380.3011 \pm0.00380.332±0.0180.332 \pm0.0180.300±0.0200.300 \pm0.0200.2974±0.00340.2974 \pm0.00340.3010±0.00350.3010 \pm0.0035
Ωb0\Omega_{\mathrm{b}0}0.04713±0.000730.04713 \pm0.000730.04813±0.000340.04813 \pm0.000340.0411±0.00150.0411 \pm0.00150.0469−0.0030+0.00240.0469^{+0.0024}_{-0.0030}0.04789±0.000320.04789 \pm0.000320.04813±0.000320.04813 \pm0.00032
AICC\mathrm{AIC}_{\mathrm{C}}18.9532±0.002318.9532 \pm0.002324.2468±0.005624.2468 \pm0.00561458.0370±0.00651458.0370 \pm0.0065103.63971±0.00052103.63971 \pm0.000521506.099±0.0281506.099 \pm0.028119.9878±0.0074119.9878 \pm0.0074
BIC\mathrm{BIC}17.9814±0.002317.9814 \pm0.002324.5646±0.005624.5646 \pm0.00561474.3397±0.00651474.3397 \pm0.0065112.54919±0.00052112.54919 \pm0.000521522.430±0.0281522.430 \pm0.028129.2133±0.0074129.2133 \pm0.0074
DIC\mathrm{DIC}14.36±0.2014.36 \pm0.2022.20±0.1522.20 \pm0.151455.963±0.0591455.963 \pm0.059101.58±0.22101.58 \pm0.221505.92±0.191505.92 \pm0.19119.64±0.23119.64 \pm0.23
WAIC\mathrm{WAIC}14.59±0.5314.59 \pm0.5322.21±0.4422.21 \pm0.441455.94±0.101455.94 \pm0.10101.73±0.36101.73 \pm0.361505.95±0.361505.95 \pm0.36119.55±0.22119.55 \pm0.22
−ln⁡B-\ln B16.46±0.6416.46 \pm0.6422.37±0.6822.37 \pm0.681456.29±0.511456.29 \pm0.51102.36±0.89102.36 \pm0.891506.49±0.441506.49 \pm0.44119.34±0.12119.34 \pm0.12
Generalised axion-like
H0H_066.9−1.1+2.266.9^{+2.2}_{-1.1}68.23±0.3968.23 \pm0.3973.6±1.073.6 \pm1.068.9±1.868.9 \pm1.868.70±0.2868.70 \pm0.2868.37±0.3168.37 \pm0.31
Ωm0\Omega_{\mathrm{m}0}0.294−0.0098+0.0120.294^{+0.012}_{-0.0098}0.3025±0.00430.3025 \pm0.00430.285−0.032+0.0680.285^{+0.068}_{-0.032}0.286−0.020+0.0340.286^{+0.034}_{-0.020}0.2973±0.00350.2973 \pm0.00350.3011−0.0040+0.00360.3011^{+0.0036}_{-0.0040}
Ωb0\Omega_{\mathrm{b}0}0.0497−0.0033+0.00160.0497^{+0.0016}_{-0.0033}0.04840−0.00057+0.000410.04840^{+0.00041}_{-0.00057}0.0411±0.00150.0411 \pm0.00150.0469±0.00260.0469 \pm0.00260.04792±0.000330.04792 \pm0.000330.04820−0.00040+0.000330.04820^{+0.00033}_{-0.00040}
nn<0.519< 0.519<0.606< 0.606–<0.612< 0.612––
log⁡10(Vmin⁡/ρφ0)\log_{10}(V_{\min}/\rho_{\varphi0})>−0.163> -0.163>−0.0239> -0.0239>−0.189> -0.189>−0.0267> -0.0267>−0.000573> -0.000573>−0.000863> -0.000863
log⁡10(f/MPl)\log_{10}(f/M_{\mathrm{Pl}})>−0.722> -0.722–−0.34−0.70+0.99-0.34^{+0.99}_{-0.70}−0.66−1.1+0.52-0.66^{+0.52}_{-1.1}<−1.45< -1.45<−1.21< -1.21
log⁡10(φi/f)\log_{10}(\varphi_i/f)–>−1.89> -1.89––>−2.73> -2.73>−2.49> -2.49
AICC\mathrm{AIC}_{\mathrm{C}}34.643±0.09734.643 \pm0.09737.5909±0.008437.5909 \pm0.00841463.7860±0.00861463.7860 \pm0.0086109.867±0.044109.867 \pm0.0441512.1870±0.00191512.1870 \pm0.0019126.4002±0.0036126.4002 \pm0.0036
BIC\mathrm{BIC}24.033±0.09724.033 \pm0.09732.8931±0.008432.8931 \pm0.00841496.3702±0.00861496.3702 \pm0.0086127.431±0.044127.431 \pm0.0441544.8279±0.00191544.8279 \pm0.0019144.6223±0.0036144.6223 \pm0.0036
DIC\mathrm{DIC}15.30±0.4315.30 \pm0.4323.391±0.09323.391 \pm0.0931456.34±0.161456.34 \pm0.16102.35±0.13102.35 \pm0.131507.588±0.0741507.588 \pm0.074121.05±0.31121.05 \pm0.31
WAIC\mathrm{WAIC}14.24±0.4614.24 \pm0.4623.17±0.1623.17 \pm0.161456.030±0.0191456.030 \pm0.019102.2719±0.0083102.2719 \pm0.00831507.382±0.0691507.382 \pm0.069120.85±0.26120.85 \pm0.26
−ln⁡B-\ln B15.09±0.7315.09 \pm0.7324.16±0.3724.16 \pm0.371456.9±1.41456.9 \pm1.4104.4±2.5104.4 \pm2.51508.263±0.0801508.263 \pm0.080121.27±0.62121.27 \pm0.62
Differences relative to Λ\LambdaCDM
ΔAICC\Delta\mathrm{AIC}_{\mathrm{C}}15.690±0.09715.690 \pm0.09713.344±0.01013.344 \pm0.0105.749±0.0115.749 \pm0.0116.227±0.0446.227 \pm0.0446.088±0.0286.088 \pm0.0286.4124±0.00826.4124 \pm0.0082
ΔBIC\Delta\mathrm{BIC}6.052±0.0976.052 \pm0.0978.329±0.0108.329 \pm0.01022.031±0.01122.031 \pm0.01114.882±0.04414.882 \pm0.04422.398±0.02822.398 \pm0.02815.4090±0.008215.4090 \pm0.0082
ΔDIC\Delta\mathrm{DIC}0.94±0.470.94 \pm0.471.19±0.181.19 \pm0.180.38±0.170.38 \pm0.170.77±0.260.77 \pm0.261.67±0.201.67 \pm0.201.41±0.391.41 \pm0.39
ΔWAIC\Delta\mathrm{WAIC}−0.35±0.70-0.35 \pm0.700.96±0.470.96 \pm0.470.09±0.100.09 \pm0.100.54±0.360.54 \pm0.361.43±0.371.43 \pm0.371.30±0.341.30 \pm0.34
Δ(−ln⁡B)\Delta(-\ln B)−1.37±0.97-1.37 \pm0.971.79±0.771.79 \pm0.770.6±1.50.6 \pm1.52.04±2.72.04 \pm2.71.77±0.451.77 \pm0.451.93±0.631.93 \pm0.63

Table III. Summary of the observational constraints obtained for Λ\LambdaCDM and the generalised axion-like dark energy model, using the different dataset combinations considered in this work: BAO, CMB+BAO, PPS, CC+GRB+Pan, CMB+BAO+PPS, and CMB+BAO+CC+GRB+Pan. Reported parameter constraints correspond to posterior mean values and 68% credible intervals. Parameters that are unconstrained are omitted. The table also reports the model comparison statistics AICC\mathrm{AIC}_{\mathrm{C}}, BIC\mathrm{BIC}, DIC\mathrm{DIC}, WAIC\mathrm{WAIC}, and −ln⁡B-\ln B, together with their differences relative to Λ\LambdaCDM. Positive values of ΔAICC\Delta\mathrm{AIC}_{\mathrm{C}}, ΔBIC\Delta\mathrm{BIC}, ΔDIC\Delta\mathrm{DIC}, ΔWAIC\Delta\mathrm{WAIC}, and Δ(−ln⁡B)\Delta(-\ln B) indicate that the quintessence model is disfavoured relative to Λ\LambdaCDM, according to the convention adopted in Section IV C.

Dataset comparisonModelGoFΔ\DeltaGoFSΔ\DeltaS−ln⁡R-\ln RΔ(−ln⁡R)\Delta(-\ln R)
CMB+BAO vs. PPSΛ\LambdaCDM5.472±0.038 σ5.472 \pm0.038\,\sigma–5.289±0.037 σ5.289 \pm0.037\,\sigma–27.83±0.7627.83 \pm0.76–
CMB+BAO vs. PPSScalar field5.5377±0.0055 σ5.5377 \pm0.0055\,\sigma0.066±0.038 σ0.066 \pm0.038\,\sigma5.3458±0.0028 σ5.3458 \pm0.0028\,\sigma0.057±0.037 σ0.057 \pm0.037\,\sigma27.21±0.9327.21 \pm0.93−0.6±1.2-0.6 \pm1.2
CMB+BAO vs. CC+GRB+PanΛ\LambdaCDM0.920±0.032 σ0.920 \pm0.032\,\sigma–0.12±0.11 σ0.12 \pm0.11\,\sigma–−5.4±1.7-5.4 \pm1.7–
CMB+BAO vs. CC+GRB+PanScalar field0.9813±0.0084 σ0.9813 \pm0.0084\,\sigma0.061±0.033 σ0.061 \pm0.033\,\sigma0.075±0.041 σ0.075 \pm0.041\,\sigma−0.05±0.12 σ-0.05 \pm0.12\,\sigma−7.3±3.5-7.3 \pm3.5−1.9±3.9-1.9 \pm3.9

Table IV. Comparison of the goodness-of-fit (GoF), suspiciousness statistic S, and Bayes ratio R for the two dataset concordance analyses considered in this work. The first comparison considers CMB+BAO versus PPS, while the second considers CMB+BAO versus CC+GRB+Pan. Results are reported for Λ\LambdaCDM and the generalised axion-like dark energy model. The quantities Δ\DeltaGoF, Δ\DeltaS, and Δ(−ln⁡R)\Delta(-\ln R) denote the differences between quintessence and Λ\LambdaCDM. Positive differences indicate increased discordance relative to Λ\LambdaCDM, whereas negative differences indicate improved concordance.

The BAO-only constraints already illustrate the additional freedom of the scalar field model. Whereas Λ\LambdaCDM gives H0=68.62±0.51 km s−1 Mpc−1H_0=68.62\pm0.51\,\mathrm{km\,s^{-1}\,Mpc^{-1}}, the axion-like model yields H0=66.9−2.1+2.2 km s−1 Mpc−1H_0=66.9^{+2.2}_{-2.1}\,\mathrm{km\,s^{-1}\,Mpc^{-1}}. The broader and asymmetric scalar field posterior is narrowed considerably when the CMB distance prior is included. For consistency check, we verify that the background-level fit of CMB based on the distance prior approach is consistent with full perturbation analysis of [11] (c.f. CMB + BAO and CMB + BAO + PPS result in Tables III and IV). As shown in Figure 3, for PPS dataset the axion-like dark energy model we consider tends to have a lower Ωm0\Omega_{\mathrm{m}0} to compensate for the temporary reduction of the deceleration parameter by the increase of dark energy equation of state to wφ∼−(1+n)−1w_{\varphi} \sim-(1+n)^{-1} in the tracking regime. This effect on Ωm0\Omega_{\mathrm{m}0} is absent in the case of CMB + BAO dataset, which is well explained by the anchoring of BAO distance measure and the CMB distance prior.

Figure 3. Marginalised one-dimensional posterior distributions and two-dimensional 68% and 95% credible regions for H0H_0, Ωm0\Omega_{\mathrm{m}0}, and Ωb0\Omega_{\mathrm{b}0}. We compare the Λ\LambdaCDM and scalar field models using CMB+BAO constraints (blue) against BAO-only constraints (top), PPS (bottom left), and CC+GRB+Pan (bottom right).

We then turn our attention to the newly introduced CC + GRB + Pan. As CC provides tighter constraint on cosmic evolution at mid-to-low redshift than SNe Ia alone via direct Hubble parameter measurements, the Ωm0\Omega_{\mathrm{m}0} contour is clearly narrower in Figure 3. This is further confirmed by Figure 4 where the parameter space leading to a lower Ωm0\Omega_{\mathrm{m}0} is clearly disfavoured. Effectively, CC + GRB + Pan prefers Λ\LambdaCDM-esque branch. The combined dataset result of CMB + BAO + CC + GRB + Pan shows similar non-preference away from Λ\LambdaCDM, with strong consistency between CMB + BAO and CC + GRB + Pan. As such, the added parameters serves no purpose and only incurs strong debt on IC probes, as shown in Table III.

Figure 4. Marginalised one-dimensional posterior distributions and two-dimensional posterior scatter plots for the generalised axion-like model parameters log⁡10(Vmin⁡/ρϕ0)\log_{10}(V_{\min}/\rho_{\phi0}), log⁡10(f/MPl)\log_{10}(f/M_{\mathrm{Pl}}), and log⁡10(ϕi/f)\log_{10}(\phi_i/f). The panels correspond to BAO-only constraints (top), PPS (bottom left), and CC+GRB+Pan. (bottom right). The scatter points are coloured according to H0H_0 in the BAO-only panel and Ωm0\Omega_{\mathrm{m}0} in the two lower panels, as indicated by their respective colour bars.

We would like to emphasise that while eventually the scalar field behaves like a cosmological constant at late-time, the tracking still happens in earlier matter-dominated era. The scalar dark energy model we consider therefore serves as a natural extension to Λ\LambdaCDM. With additional degrees of freedom, the tracking may serves as a window to more complicated phenomenon [92].

The scalar field parameter bounds provide a complementary view of this preference. Table III shows that the four additional parameters are not all independently determined: several posteriors admit only one-sided bounds or remain unconstrained. In particular, the CMB-containing combinations require log⁡10(Vmin⁡/ρφ0)\log_{10}(V_{\min}/\rho_{\varphi0}) close to its upper prior boundary, which corresponds to a potential minimum close to the reference present-day dark energy density. The correlations between the potential parameters and the initial field displacement in Figure 4 illustrate how different scalar field configurations can reproduce similar distance observables. Bounds adjacent to prior limits should therefore be interpreted with their assumed prior ranges in mind.

Model comparison and dataset concordance

We now assess whether the extra freedom of the scalar field is supported by the data after accounting for model complexity. Throughout this comparison, ΔX≡Xscalar−XΛCDM\Delta X \equiv X_{\mathrm{scalar}} - X_{\Lambda\mathrm{CDM}}, so that a positive difference in the information criteria indicates a preference for the simpler Λ\LambdaCDM description. As shown in Table III, Δ\DeltaAICC_{\mathrm{C}} is positive for all six combinations, ranging from 5.749±0.0115.749 \pm0.011 for PPS to 15.690±0.09715.690 \pm0.097 for BAO. The BIC differences are also positive throughout, with particularly large penalties in the PPS and CMB+BAO+PPS fits, Δ\DeltaBIC =22.031±0.011= 22.031 \pm0.011 and 22.398±0.02822.398 \pm0.028, respectively. Thus, the improvement in the best-fit likelihood is not sufficient to overcome the complexity penalty imposed by these criteria.

The posterior-based criteria give a less pronounced, but broadly consistent, picture. DIC differences lie between 0.38±0.170.38 \pm0.17 (PPS) and 1.67±0.201.67 \pm0.20 (CMB+BAO+PPS), while WAIC differences are generally small; in particular, BAO gives Δ\DeltaWAIC =−0.35±0.70= -0.35 \pm0.70, which is consistent with no preference. The values of Δ(−ln⁡B)\Delta(-\ln\mathrm{B}) also depend on the dataset and have comparatively large numerical uncertainties in several cases. For example, BAO gives −1.37±0.97-1.37 \pm0.97, whereas CMB+BAO gives 1.79±0.771.79 \pm0.77. These evidence estimates should be interpreted cautiously because the harmonic-mean evaluation is sensitive to low-likelihood posterior tails. Taken together, the different criteria provide no robust statistical preference for the additional scalar field degrees of freedom, even where they broaden the range of allowed cosmological histories.

To investigate dataset consistency, we compare CMB+BAO separately with PPS and CC+GRB+Pan, using the joint fits to evaluate the goodness-of-fit (GoF), suspiciousness (S), and Bayesian ratio (R) in Table IV. Both GoF and S are expressed in equivalent Gaussian standard deviations. For CMB+BAO versus PPS, the discrepancy remains strong in both models: the GoF significance changes from 5.472±0.038 σ5.472 \pm0.038\,\sigma in Λ\LambdaCDM to 5.5377±0.0055 σ5.5377 \pm0.0055\,\sigma in the scalar field model, while the corresponding S significance changes from 5.289±0.037 σ5.289 \pm0.037\,\sigma to 5.3458±0.0028 σ5.3458 \pm0.0028\,\sigma. The differences, Δ\DeltaGoF =0.066±0.038 σ= 0.066 \pm0.038\,\sigma and Δ\DeltaS =0.057±0.037 σ= 0.057 \pm0.037\,\sigma, are small and do not point to any alleviation of the Hubble tension. The Bayesian ratio difference, Δ(−ln⁡R)=−0.6±1.2\Delta(-\ln\mathrm{R}) = -0.6 \pm1.2, is likewise compatible with zero. The ability of PPS to accommodate a smaller Ωm0\Omega_{m0} therefore does not translate into a more concordant joint fit with CMB+BAO.

By contrast, CMB+BAO and CC+GRB+Pan are mutually consistent. The GoF significance is 0.920±0.032 σ0.920 \pm0.032\,\sigma in Λ\LambdaCDM and 0.9813±0.0084 σ0.9813 \pm0.0084\,\sigma in the scalar field model, while the corresponding suspiciousness significances are 0.12±0.11 σ0.12 \pm0.11\,\sigma and 0.075±0.041 σ0.075 \pm0.041\,\sigma. Their differences, Δ\DeltaGoF =0.061±0.033 σ= 0.061 \pm0.033\,\sigma and Δ\DeltaS =−0.05±0.12 σ= -0.05 \pm0.12\,\sigma, are too small to establish a meaningful scalar field change in concordance. The Bayesian ratio difference of −1.9±3.9-1.9 \pm3.9 is also inconclusive. These results are consistent with the restricted scalar field parameter space inferred from the separate CC+GRB+Pan fit: no substantial change to the background dynamics is required for agreement with CMB+BAO.

These contrasting levels of dataset concordance are also illustrated by the posterior distributions shown in Figure 3, where the tension between PPS and CMB+BAO is evident, while CC+GRB+Pan exhibits substantial overlap with CMB+BAO.

Reconstructed scalar field and density evolution

Finally, we examine the background histories reconstructed from posterior samples in Figure 5 and Figure 6. The upper panels show the scalar field equation of state wϕ(a)w_{\phi}(a), while the lower panels show the fractional densities Ωr(a)\Omega_{r}(a), Ωm(a)\Omega_{m}(a), and Ωϕ(a)\Omega_{\phi}(a). These reconstructions allow us to examine how the freedom in the scalar field dynamics changes when the late-time datasets are combined with CMB+BAO.

Cosmological evolution of the generalised axion-like scalar field reconstructed from posterior samples

Figure 5. Cosmological evolution of the generalised axion-like scalar field reconstructed from posterior samples obtained using PPS (left) and CMB+BAO+PPS (right). The upper panels show the evolution of the scalar field equation of state wϕ(a)w_{\phi}(a), with the blue curves representing posterior realisations and the darker solid curve corresponding to the best sampled fit. The dashed horizontal line indicates the cosmological constant value wΛ=−1w_{\Lambda}=-1. The lower panels show the evolution of the fractional energy densities of radiation Ωr\Omega_{\mathrm{r}} (blue), matter Ωm\Omega_{\mathrm{m}} (gold), and the scalar field Ωϕ\Omega_{\phi} (purple).

Cosmological evolution of the generalised axion-like scalar field

Figure 6. Cosmological evolution of the generalised axion-like scalar field reconstructed from posterior samples obtained using CC+GRB+Pan (left) and CMB+BAO+CC+GRB+Pan (right). The upper panels show the evolution of the scalar field equation of state wϕ(a)w_{\phi}(a), with the blue curves representing posterior realisations and the darker solid curve corresponding to the best sampled fit. The dashed horizontal line indicates the cosmological constant value wΛ=−1w_{\Lambda}=-1. The lower panels show the evolution of the fractional energy densities of radiation Ωr\Omega_{\mathrm{r}} (blue), matter Ωm\Omega_{\mathrm{m}} (gold), and the scalar field Ωϕ\Omega_{\phi} (purple).

We first consider PPS, shown in Figure 5. The PPS-only constraints permit a broad variety of scalar field trajectories, including substantial departures from wϕ=−1w_{\phi} = -1. Such behaviour is compatible with a temporary tracking phase, whose approximate equation of state is wϕ≃−(1+n)−1w_{\phi} \simeq-(1+n)^{-1}, followed by evolution towards the potential minimum. The corresponding freedom in the scalar field dynamics also allows a wider range of histories for Ωϕ(a)\Omega_{\phi}(a). However, once CMB+BAO is included, the allowed evolution becomes much more tightly restricted. The posterior realisations cluster around histories close

to the Λ\LambdaCDM limit, with only limited departures from wϕ=−1w_\phi=-1 remaining allowed. This illustrates that the additional freedom permitted by PPS alone is strongly constrained by the CMB+BAO information, consistent with the persistence of the dataset tension discussed above.

A similar, although less pronounced, behaviour is found for CC+GRB+Pan, as shown in Figure 6. This dataset combination also permits temporary departures from wϕ=−1w_{\phi}=-1, but the range of allowed scalar field histories is more restricted than in the PPS-only case. This is consistent with the preference for a Λ\LambdaCDM-like branch already inferred from the corresponding parameter constraints. Adding CMB+BAO further reduces the allowed deviations, leading to a much narrower set of background histories. In particular, the radiation-to-matter transition and the onset of late-time scalar field domination become very similar across the reconstructed posterior realisations. The resulting evolution remains close to Λ\LambdaCDM, in agreement with the strong concordance between CMB+BAO and CC+GRB+Pan.

These results illustrate that late-time observations alone can accommodate non-trivial scalar field dynamics, with the largest freedom found for PPS. Nevertheless,

the inclusion of CMB+BAO strongly restricts such deviations, driving the allowed background evolution towards Λ\LambdaCDM-like histories. The same qualitative behaviour is found when comparing BAO alone with CMB+BAO, although we omit the corresponding reconstructions since they do not provide additional information beyond the comparisons shown here. The best sampled-fit trajectories should be regarded as representative realisations rather than a complete characterisation of the posterior uncertainty. Overall, while temporary deviations from wϕ=−1w_{\phi}=-1 remain possible, the statistical comparison does not establish a preference for the additional scalar field degrees of freedom.

Conclusions

Scalar fields have long played a central role in theoretical attempts to explain the late-time accelerated expansion of the Universe, offering a dynamical alternative to a rigid cosmological constant. Through an evolving equation of state, scalar field dark energy models can accommodate a wide range of expansion histories while addressing some of the conceptual challenges associated with Λ\LambdaCDM. In this work, we have focused on a class of axion-like scalar field models characterised by a generalised potential that supports a prolonged tracking regime followed by late-time acceleration. By confronting this scenario with background cosmological observations from the cosmic microwave background (reproducing full Planck power spectrum result in [11]), baryon acoustic oscillations, Type Ia supernovae, SH0ES determination of the local expansion rate, cosmic chronometers, and gamma-ray bursts, we have assessed its phenomenological viability and shown that axion-inspired quintessence models provide a flexible and testable framework for describing the dark energy component driving the present cosmic acceleration.

Our results show that the additional freedom introduced by the scalar field allows departures from the Λ\LambdaCDM expansion history, particularly for the PPS (and to a lesser degree CC+GRB+Pan) dataset, which admits a wider range of scalar field trajectories and lower values of Ωm0\Omega_{\mathrm{m}0}. However, these deviations become strongly restricted when CMB+BAO information is included. Moreover, the additional degrees of freedom do not alleviate the discrepancy between CMB+BAO and PPS, which remains at approximately 5.5σ5.5\sigma in both models, nor disrupt the concordance between CMB+BAO and CC+GRB+Pan. The model comparison statistics also provide no significant preference for the generalised axion-like scenario over Λ\LambdaCDM, with the information criteria generally favouring the simpler model. Therefore, although the scalar field can exhibit tracking regime during matter-dominated era, current background observations do not require departures from a cosmological constant. On the other hand, it does provide rich physics in earlier epoch that may serve as a platform for future extension, without any undesirable side effects.

Acknowledgements

M. B.-L. is supported by the Basque Foundation of Science Ikerbasque. HWC is supported by the NSFC Grant No. 12250410250, No. 12505074 and No. 12347133. C.G.B. acknowledges financial support from the French National Research Agency (ANR) under project No. ANR-25-CE57-1445-01 and from the Spanish Ministry of Science, Innovation and Universities under the FPI fellowship PRE2021-100340. T.J.B. is supported by the Basque Foundation of Science Ikerbasque. Our work is supported by the Spanish Grant PID2023-149016NB-I00 (MINECO/AEI/FEDER, UE). This work is also supported by the Basque government Grant No. IT1628-22 (Spain). The authors acknowledge the contribution of the COST Action CA21136 “Addressing observational tensions in cosmology with systematics and fundamental physics (CosmoVerse)”.

References

  1. [1]A. G. Riess et al. (Supernova Search Team), Observational evidence from supernovae for an accelerating universe and a cosmological constant, Astron. J. 116, 1009 (1998), arXiv:astro-ph/9805201.DOI
  2. [2]S. Perlmutter et al. (Supernova Cosmology Project), Measurements of Ω and Λ from 42 High Redshift Supernovae, Astrophys. J. 517, 565 (1999), arXiv:astro-ph/9812133.DOI
  3. [3]Planck Collaboration, Planck 2018 results. VI. Cosmological parameters, Astron. Astrophys. 641, A6 (2020).
  4. [4]S. e. a. Alam, Completed sdss-iv extended baryon oscillation spectroscopic survey, Physical Review D 103, 083533 (2021).
  5. [5]E. Di Valentino et al. (CosmoVerse Network), The CosmoVerse White Paper: Addressing observational tensions in cosmology with systematics and fundamental physics, Phys. Dark Univ. 49, 101965 (2025), arXiv:2504.01669 [astro-ph.CO].
  6. [6]Y. Akrami et al. (CANTATA), Modified Gravity and Cosmology. An Update by the CANTATA Network, edited by E. N. Saridakis, R. Lazkoz, V. Salzano, P. Vargas Moniz, S. Capozziello, J. Beltrán Jiménez, M. De Laurentis, and G. J. Olmo (Springer, 2021) arXiv:2105.12582 [gr-qc].DOI
  7. [7]B. Ratra and P. J. E. Peebles, Cosmological Consequences of a Rolling Homogeneous Scalar Field, Phys. Rev. D 37, 3406 (1988).pubmed.ncbi.nlm.nih.gov/9958635
  8. [8]C. Armendariz-Picon, V. F. Mukhanov, and P. J. Steinhardt, Essentials of k essence, Phys. Rev. D 63, 103510 (2001), arXiv:astro-ph/0006373.
  9. [9]M. Kamionkowski, J. Pradler, and D. G. E. Walker, Dark energy from the string axiverse, Phys. Rev. Lett. 113, 251302 (2014), arXiv:1409.0549 [hep-ph].DOI
  10. [10]R. Emami, D. Crin, J. Pradler, A. Raccanelli, and M. Kamionkowski, Cosmological tests of an axiverse-inspired quintessence field, Phys. Rev. D 93, 123005 (2016), arXiv:1603.04851 [astro-ph.CO].DOI
  11. [11]H.-W. Chiang, C. G. Boiza, and M. Bouhmadi-López, Observational constraints on generalised axion-like potentials for the late Universe, JCAP 08 (2025), 064, arXiv:2503.04898 [astro-ph.CO].arxiv.org/abs/2503.04898
  12. [12]M. Bouhmadi-López, H.-W. Chiang, and C. G. Boiza, Generalised axion-like dark energy: From theory to observations, Phys. Dark Univ. 49, 101968 (2025).
  13. [13]C. Deffayet, O. Pujolas, I. Sawicki, and A. Vikman, Imperfect Dark Energy from Kinetic Gravity Braiding, JCAP 10 (2010), 026, arXiv:1008.0048 [hep-th].DOI
  14. [14]O. Pujolas, I. Sawicki, and A. Vikman, The Imperfect Fluid behind Kinetic Gravity Braiding, JHEP 11 (2011), 156, arXiv:1103.5360 [hep-th].
  15. [15]T. Borislavov Vasilev, M. Bouhmadi-López, and P. Martín-Moruno, Phantom attractors in kinetic gravity braiding theories: a dynamical system approach, JCAP 06 (2023), 026, arXiv:2212.02547 [gr-qc].
  16. [16]T. Borislavov Vasilev, M. Bouhmadi-López, and P. Martín-Moruno, Dark energy with a shift-symmetric scalar field: Obstacles, loophole hunting and dead ends, Phys. Dark Univ. 46, 101679 (2024), arXiv:2406.12576 [gr-qc].
  17. [17]T. Kobayashi, Horndeski theory and beyond: a review, Rept. Prog. Phys. 82, 086901 (2019), arXiv:1901.07183 [gr-qc].DOI
  18. [18]T. S. Koivisto and N. J. Nunes, Inflation and dark energy from three-forms, Phys. Rev. D 80, 103509 (2009), arXiv:0908.0920 [astro-ph.CO].DOI
  19. [19]J. Morais, M. Bouhmadi-López, K. Sravan Kumar, J. Marto, and Y. Tavakoli, Interacting 3-form dark energy models: distinguishing interactions and avoiding the Little Sibling of the Big Rip, Phys. Dark Univ. 15, 7 (2017), arXiv:1608.01679 [gr-qc].DOI
  20. [20]M. Bouhmadi-López, J. Marto, J. Morais, and C. M. Silva, Cosmic infinity: A dynamical system approach, JCAP 03 (2017), 042, arXiv:1611.03100 [gr-qc].
  21. [21]M. Bouhmadi-López, H.-W. Chiang, C. G. Boiza, and P. Chen, Observational constraints on 3-forms dark energy, JCAP 06 (2026), 084, arXiv:2512.09991 [astro-ph.CO].
  22. [22]M. Bouhmadi-López, H.-W. Chiang, C. G. Boiza, J. O. del Río, T. J. Broadhurst, and P. Chen, Three-form dark energy: Constraits and multi-probe comparison with ΛCDM, Phys. Dark Univ. 54, 102509 (2026), arXiv:2606.27436 [astro-ph.CO].
  23. [23]A. Y. Kamenshchik, U. Moschella, and V. Pasquier, An Alternative to quintessence, Phys. Lett. B 511, 265 (2001), arXiv:gr-qc/0103004.DOI
  24. [24]M. Bouhmadi-López and J. A. Jimenez Madrid, Escaping the big rip?, JCAP 05 (2005), 005, arXiv:astro-ph/0404540.
  25. [25]M. Bouhmadi-López, P. F. González-Díaz, and P. Martín-Moruno, Worse than a big rip?, Phys. Lett. B 659, 1 (2008), arXiv:gr-qc/0612135.
  26. [26]O. Akarsu, S. Kumar, E. Özülker, and J. A. Vazquez, Relaxing cosmological tensions with a sign switching cos- mological constant, Phys. Rev. D 104, 123512 (2021), arXiv:2108.09239 [astro-ph.CO].
  27. [27]O. Akarsu, S. Kumar, E. Özülker, J. A. Vazquez, and A. Yadav, Relaxing cosmological tensions with a sign switching cosmological constant: Improved results with Planck, BAO, and Pantheon data, Phys. Rev. D 108, 025513 (2023), arXiv:2211.05742 [astro-ph.CO].DOI
  28. [28]O. Akarsu, E. Di Valentino, S. Kumar, R. C. Nunes, J. A. Vazquez, and A. Yadav, ΛsCDM model: A promising scenario for alleviation of cosmological tensions, To appear (2023), arXiv:2307.10899 [astro-ph.CO].arxiv.org/abs/2307.10899
  29. [29]M. Bouhmadi-López and B. Ibarra-Uriondo, Cosmographic analysis of sign-switching dark energy, Phys. Rev. D 112, 063559 (2025), arXiv:2506.12139 [gr-qc].
  30. [30]M. Bouhmadi-López and B. Ibarra-Uriondo, Cosmological perturbations for smooth sign-switching dark energy models, Phys. Dark Univ. 50, 102129 (2025), arXiv:2506.18992 [gr-qc].
  31. [31]P. Ghafari, M. Najafi, M. Ghodsi Yengejeh, E. Özülker, E. Di Valentino, and J. T. Firouzjaee, A Multi-Probe ISW Study of Dark Energy Models with Negative Energy Density: Galaxy Correlations, Lensing Bispectrum, and Planck ISW-Lensing Likelihood, To appear (2025), arXiv:2512.07060 [astro-ph.CO].arxiv.org/abs/2512.07060
  32. [32]Ö. Akarsu, E. Di Valentino, J. Vyskočil, E. Yılmaz, A. E. Yükselci, and A. Zhuk, Nonlinear matter power spectrum from relativistic N-body simulations: ΛsCDM versus ΛCDM, Phys. Rev. D 113, 083508 (2026), arXiv:2510.18741 [astro-ph.CO].DOI
  33. [33]M. Bouhmadi-López, H.-W. Chiang, and B. Ibarra-Uriondo, Alleviating the Hubble Tension with Smooth Sign-Switching Dark Energy: Full CMB Constraints with DESI and PantheonPlus (2026), arXiv:2607.05044 [astro-ph.CO].
  34. [34]S. Capozziello and M. De Laurentis, Extended Theories of Gravity, Phys. Rept. 509, 167 (2011), arXiv:1108.6266 [gr-qc].DOI
  35. [35]S. Nojiri and S. D. Odintsov, Unified cosmic history in modified gravity: from F(R) theory to Lorentz non-invariant models, Phys. Rept. 505, 59 (2011), arXiv:1011.0544 [gr-qc].DOI
  36. [36]S. Nojiri, S. D. Odintsov, and V. K. Oikonomou, Modified Gravity Theories on a Nutshell: Inflation, Bounce and Late-time Evolution, Phys. Rept. 692, 1 (2017), arXiv:1705.11098 [gr-qc].DOI
  37. [37]J. Morais, M. Bouhmadi-López, and S. Capozziello, Can f(R) gravity contribute to (dark) radiation?, JCAP 09 (2015), 041, arXiv:1507.02623 [gr-qc].DOI
  38. [38]G. R. Bengochea and R. Ferraro, Ferrara, Dark torsion as the cosmic speed-up, Phys. Rev. D 79, 124019 (2009), arXiv:0812.1205 [astro-ph].DOI
  39. [39]R. Ferraro and F. Fiorini, Modified teleparallel gravity: Inflation without inflation, Phys. Rev. D 75, 084031 (2007), arXiv:gr-qc/0610067.DOI
  40. [40]Y.-F. Cai, S. Capozziello, M. De Laurentis, and E. N. Saridakis, f(T) teleparallel gravity and cosmology, Rept. Prog. Phys. 79, 106901 (2016), arXiv:1511.07586 [gr-qc].DOI
  41. [41]J. Beltrán Jiménez, L. Heisenberg, and T. S. Koivisto, Teleparallel Palatini theories, JCAP 08 (2018), 039, arXiv:1803.10185 [gr-qc].DOI
  42. [42]J. Beltrán Jiménez, L. Heisenberg, T. S. Koivisto, and S. Pekar, Cosmology in f(Q) geometry, Phys. Rev. D 101, 103507 (2020), arXiv:1906.10027 [gr-qc].DOI
  43. [43]I. Ayuso, R. Lazkoz, and V. Salzano, Observational constraints on cosmological solutions of f(Q) theories, Phys. Rev. D 103, 063505 (2021), arXiv:2012.00046 [astro-ph.CO].DOI
  44. [44]C. G. Boiza, M. Petronikколou, M. Bouhmadi-López, and E. N. Saridakis, Addressing H 0 and S 8 tensions within f(Q) cosmology, JCAP 12 (2025), 011, arXiv:2505.18264 [astro-ph.CO].DOI
  45. [45]I. Ayuso, M. Bouhmadi-López, C.-Y. Chen, X. Y. Chew, K. Dialektopoulos, and Y. C. Ong, Insights in f(Q) cosmology: the relevance of the connection, JCAP 11 (2025), 068, arXiv:2506.03506 [gr-qc].DOI
  46. [46]C. G. Boiza and M. Bouhmadi-López, Speeding up the Universe with a generalised axion-like potential, Eur. Phys. J. C 85, 777 (2025), arXiv:2409.18184 [astro-ph.CO].arxiv.org/abs/2409.18184
  47. [47]C. G. Boiza and M. Bouhmadi-López, Cosmological perturbations in a generalised axion-like dark energy model, Phys. Dark Univ. 48, 101845 (2025), arXiv:2410.22467 [astro-ph.CO].arxiv.org/abs/2410.22467
  48. [48]D. J. E. Marsh, Axion Cosmology, Phys. Rept. 643, 1 (2016), arXiv:1510.07633 [astro-ph.CO].DOI
  49. [49]K. Freese, J. A. Frieman, and A. V. Olinto, Natural inflation with pseudo - Nambu-Goldstone bosons, Phys. Rev. Lett. 65, 3233 (1990).pubmed.ncbi.nlm.nih.gov/10042817
  50. [50]F. C. Adams, J. R. Bond, K. Freese, J. A. Frieman, and A. V. Olinto, Natural inflation: Particle physics models, power law spectra for large scale structure, and constraints from COBE, Phys. Rev. D 47, 426 (1993), arXiv:hep-ph/9207245.DOI
  51. [51]S. Dimopoulos, S. Kachru, J. McGreevy, and J. G. Wacker, N-flation, JCAP 08 (2008), 003, arXiv:hep-th/0507205.DOI
  52. [52]R. Easther and L. McAllister, Random matrices and the spectrum of N-flation, JCAP 05 (2006), 018, arXiv:hep-th/0512102.DOI
  53. [53]T. C. Bachlechner, M. Dias, J. Frazer, and L. McAllister, Chaotic inflation with kinetic alignment of axion fields, Phys. Rev. D 91, 023520 (2015), arXiv:1404.7496 [hep-th].
  54. [54]V. Poulin, T. L. Smith, T. Karwal, and M. Kamionkowski, Early Dark Energy Can Resolve The Hubble Tension, Phys. Rev. Lett. 122, 221301 (2019), arXiv:1810.04083 [astro-ph.CO].
  55. [55]M. Kamionkowski and A. G. Riess, The Hubble Tension and Early Dark Energy, Ann. Rev. Nucl. Part. Sci. 73, 153 (2023), arXiv:2211.04492 [astro-ph.CO].arxiv.org/abs/2211.04492
  56. [56]W. Hu, R. Barkana, and A. Gruzinov, Cold and fuzzy dark matter, Phys. Rev. Lett. 85, 1158 (2000), arXiv:astro-ph/0003365.DOI
  57. [57]L. Hui, J. P. Ostriker, S. Tremaine, and E. Witten, Ultralight scalars as cosmological dark matter, Phys. Rev. D 95, 083541 (2017), arXiv:1610.08297 [astro-ph.CO].DOI
  58. [58]H.-Y. Schive, T. Chiueh, and T. Broadhurst, Cosmic Structure as the Quantum Interference of a Coherent Dark Wave, Nature Phys. 10, 496 (2014), arXiv:1406.6586 [astro-ph.GA].DOI
  59. [59]H.-Y. Schive, M.-H. Liao, T.-P. Woo, S.-K. Wong, T. Chiueh, T. Broadhurst, and W. Y. P. Hwang, Understanding the Core-Halo Relation of Quantum Wave Dark Matter from 3D Simulations, Phys. Rev. Lett. 113, 261302 (2014), arXiv:1407.7762 [astro-ph.GA].DOI
  60. [60]P. Mocz et al., Galaxy formation with BECDM – II. Cosmic filaments and first galaxies, Mon. Not. Roy. Astron.

Paper details

vlnc:W_yzu3ana7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents