Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks

arXiv:2608.27469 · physics.gen-ph · Submitted 2026-08-14 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks".

Mira: Here's the gist of the paper. Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks aims to directly reconstruct the redshift evolution of a quantity entering the underlying entropy law by combining observational data with physical constraints from cosmological dynamics This approach establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation.

Kai: Theoretical Framework The paper constructs a cosmological framework based on Barrow holographic dark energy with a varying exponent, which serves as the theoretical basis for the reconstruction.

Mira: This involves starting from Barrow entropy where S ∝ A(1+∆/2).

Kai: Incorporating this into the holographic dark-energy construction yields an energy density form of ρDE = CL(∆−2).

Paper summary: Kai: So, wrapping up this talk on "Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks." The paper by Basilakos, Paliathanasis, Saridakis, and Tsilioukas shows that they can reconstruct the function for delta over redshift directly from observations using a physics informed neural network.

Mira: They successfully show that this reconstructed exponent leads to a cosmologically viable late-time history where w DE is near negative one at low redshifts, and both quintessence and phantom behaviors are possible within the results.

Lev: The authors emphasize that they didn't assume any specific linear, CPL-like, exponential, or other parametrizations for how delta changes with time when they performed the reconstruction.

Kai: That’s important because it means the reconstruction isn't just confirming a pre-existing model; it's finding a new functional behavior allowed by the dynamics while still fitting all the data.

Mira: They conclude that this data-driven reconstruction yields a cosmologically viable late-time history, and they show that this approach allows us to infer the entropy deformation itself from observations while preserving that successful background cosmological evolution.

Lev: The main point is using physics informed machine learning to connect cosmological observations with quantities characterizing the more fundamental theoretical structure of the model, which is what they do in "Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks."

Conclusion: Kai: So, we're wrapping up this session on "Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks."

Mira: Basically, they took all those cosmological measurements and used a neural network to figure out how that Bario exponent changes over time.

Kai: They didn't assume a fixed shape for it beforehand; they let the data dictate the functional behavior of that exponent.

Mira: It means we’re not just fitting a curve to the data; we’re letting the physics of dark energy and our observations define what that function actually looks like.

Lev: From an error correction point, that kind of direct reconstruction is tough, but it gives us a real map of the parameter space constraints.

Kai: The paper shows they found this reconstructed exponent follows a smooth path across all the redshift range they looked at.

Mira: And even though the results favor a negative value for that exponent, it’s still compatible with zero or constant negative values, which is an interesting constraint on quantum gravity ideas.

Lev: It's telling us that current data constrains the model pretty tightly without forcing us into one specific theoretical shape for entropy deformation.

Kai: So, the authors are saying they can infer something about how spacetime itself is deforming based on what we measure in the late universe.

Mira: It moves the goalpost a bit; it suggests that even if we don't see direct evidence of this entropy deformation, its effect on dark energy dynamics is still imprinted in our distance measurements.

Lev: This opens up possibilities for how we might eventually design experiments to probe those very fundamental structures through cosmological data alone.

Spyros Basilakos, Andronikos Paliathanasis, Emmanuel N. Saridakis, Stylianos A. Tsilioukas

National Observatory of Athens · Academy of Athens, Research Center for Astronomy and Applied Mathematics · School of Sciences, European University Cyprus · Institute of Systems Science, Durban University of Technology · Centre for Space Research, North-West University · Centro de Investigación, Innovación y Creación (CIIC), Universidad Catˡlica de Temuco · Departamento de Ciencias Matemˡticas y Fisca, Facultad de Ingeniería, Universidad Catˡlica de Temuco · National Institute for Theoretical and Computational Sciences (NITheCS) · Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing, National Observatory of Athens · Departamento de Matemˡticas, Universidad Catˡlica del Norte · CAS Key Laboratory for Research in Galaxies and Cosmology, School of Astronomy and Space Science, University of Science and Technology of China

physics.gen-ph

Submitted: 2026-08-14

Updated: 2026-11-02

Comments: 14 pages, 7 figures; accepted for publication in EPJC

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 78/100

The gist: Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks aims to directly reconstruct the redshift evolution of a quantity entering the underlying entropy…

Key concepts

Barrow Holographic Dark Energy
This model describes dark energy where the energy density depends on a characteristic size $L$ related to entropy via Barrow entropy ($S \propto A^{1+\Delta/2}$). This leads to an energy density $\rho_{DE} \propto L^{\Delta-2}$, linking cosmological evolution directly to the deformation of horizon entropy.
Physics-Informed Neural Network (PINN)
A machine learning framework used here that combines a neural network with the underlying physical laws of cosmology. The network is trained using a total loss function that includes terms enforcing the Barrow dynamics, observational data constraints, and smoothness penalties, allowing it to learn the functional form of $\Delta(z)$ directly.
Generalized Barrow Exponent $\Delta(z)$
Instead of assuming a constant exponent ($\Delta$), this approach treats it as a function of redshift, $\Delta(z)$. This allows the dark energy dynamics to evolve throughout cosmic history. The reconstruction reveals that the posterior mean for this exponent lies predominantly in the negative region over the observed redshift range.
Total Loss Function ($L_{total}$)
The training objective combines four components: LPDE (physics constraint), LDATA (observational data from SNIa, BAO, and CC), LPriors (Gaussian constraints on parameters like $H_0$), and LSP (smoothness penalties). This multi-faceted loss ensures the reconstructed $\Delta(z)$ is physically viable and consistent with current cosmological measurements.

Terminology

Summary

Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks aims to directly reconstruct the redshift evolution of a quantity entering the underlying entropy law by combining observational data with physical constraints from cosmological dynamics This approach establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation. p. 1

Where the paper says it

From the machine-learning perspective, the physicsinformed framework is used not simply to reconstruct a standard cosmological observable such as H(z) or wDE(z), but to infer a function characterizing the underlying horizon entropy. The analysis thus establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation, allowing the data to probe the functional behavior of a quantity associated with the quantum-gravitational structure of the model.

Theoretical Framework

The paper constructs a cosmological framework based on Barrow holographic dark energy with a varying exponent, which serves as the theoretical basis for the reconstruction. This involves starting from Barrow entropy where S ∝ A(1+∆/2). p. 6

Where the paper says it

In both cases the posterior mean is found in the negative-∆ region over the whole redshift interval considered, although the statistical uncertainties are sufficiently large that the standard holographic limit ∆ = 0 remains compatible with the reconstruction within the broader credible region. For the PP combination, the reconstructed mean exhibits a mild evolution: starting from a negative value at

Incorporating this into the holographic dark-energy construction yields an energy density form of ρDE = CL(∆−2). p. 1
Where the paper says it

When the entropy relation (2. 1) is incorporated into the holographic dark-energy construction, the corresponding energy density acquires the form [79] ρDE = CL∆−2 , (2.

The resulting dynamics are governed by a central background equation in redshift space, Equation (2. p. 2
Where the paper says it

The resulting dynamics therefore contains explicitly the derivative of the Barrow exponent [104]. p. 2

Where the paper says it

In this case, the dependence of ρDE on the future event horizon is modified not only through Rh itself but also through the evolution of ∆. The resulting dynamics therefore contains explicitly the derivative of the Barrow exponent [104].

Using ln a = − ln(1+z) and the relation d d ln a = −(1+z) d dz , the evolution equation can be expressed directly in redshift space as (1 + z) ΩDE(1−ΩDE) dΩDE dz = 1 + z ∆ − 2 d∆ dz ln 1 − ΩDE ΩDE P(z) −(2 − ∆)r ΩDE C 1−ΩDE ΩDE P(z) ∆ 2(∆−2) + ∆ + 1, (2.

11), which explicitly contains the derivative of the Barrow exponent [104]. p. 2
Where the paper says it

In this case, the dependence of ρDE on the future event horizon is modified not only through Rh itself but also through the evolution of ∆. The resulting dynamics therefore contains explicitly the derivative of the Barrow exponent [104].

How the training signal is built

The Cosmo-PINN framework is employed to reconstruct the generalized Barrow holographic dark energy scenario directly from cosmological observations. The primary reconstructed quantity is the Barrow exponent ∆(z). p. 3

Where the paper says it

4) such that x ∈ [0, 1], and use it as the input variable of the network. The primary reconstructed quantity is the Barrow entropy exponent ∆(z).

The network represents ∆ through a finite Chebyshev expansion, specifically ∆(x) = A tanh[g(x)], where g(x) = X sum n=0 cn Tn(2x − 1). p. 3
Where the paper says it

The primary reconstructed quantity is the Barrow entropy exponent ∆(z). Following the CosmoPINN architecture [112], and in order to obtain a stable reconstruction while suppressing overfitting, we represent ∆ through a finite Chebyshev expansion, namely ∆(x) = A tanh[g(x)], g(x) = X N n=0 cnTn(2x − 1), (3.

The training is controlled by the total loss function Ltotal = LPDE + LDATA + LPriors + LSP. p. 3
Where the paper says it

Thus, the network reconstructs the relevant cosmological functions while respecting the physical bounds already at the level of its architecture. The training is controlled by the total loss function Ltotal = LPDE + LDATA + LPriors + LSP, (3.

Where the gains come from

The total loss function Ltotal consists of four contributions:

  1. The physics-informed contribution LPDE, defined as λPDE Nc sum i=1 R squared PDE(xi). p. 3

    Where the paper says it

    7) where the four contributions respectively encode the cosmological dynamics, the observational information, soft priors on the cosmological parameters, and smoothness conditions on the reconstructed functions. The physics-informed contribution is defined as LPDE = λPDE Nc X Nc i=1 R2 PDE(xi), (3.

    This term forces the reconstructed ∆(z) and omegaDE(z) to satisfy the underlying cosmological dynamics throughout the redshift interval. p. 3
    Where the paper says it

    11), evaluated at each collocation point xi . Minimization of this contribution therefore forces the reconstructed ∆(z) and ΩDE(z) to satisfy the underlying cosmological dynamics throughout the redshift interval, and not merely at the locations of the observational measurements.

  2. The observational contribution LDATA, which is split into contributions from Type Ia supernovae (SNIa), Baryon acoustic oscillations (BAO), and cosmic chronometers (CC). p. 0

    Where the paper says it

    In this work we reconstruct ∆(z) directly from observations, without assuming any particular functional form, using the Cosmo-PINN physics-informed neural-network framework. The generalized Barrow holographic evolution equation is incorporated into the training, while PantheonPlus supernovae, DESI DR2 baryon acoustic oscillations and cosmic chronometers constrain the reconstruction.

    The weights λCC, λBAO and λSN are dynamically adjusted using the GradNorm method to balance the gradient contributions of the different observational probes. p. 3
    Where the paper says it

    The observational contribution takes the form LDATA = λCCLCC + λBAOLBAO + λSNLSN. 9) During training, the weights λCC, λBAO and λSN are dynamically adjusted using the GradNorm method, in order to balance the gradient contributions of the differen

  3. Soft priors LPriors, which include Gaussian anchors for cosmological parameters like H0 and C. p. 4

    Where the paper says it

    12) are treated as trainable cosmological parameters. Since their values are coupled to the reconstruction of ∆(z) and ΩDE(z), we introduce soft Gaussian anchors through LPriors = X θ λθ 2 θ PINN − ¯θ σθ 2 .

  4. Smoothness penalties LSP, which contain weak smoothness penalties on the first and second derivatives of ∆(z) and omegaDE(z). p. 4

    Where the paper says it

    17) These terms act as soft anchors rather than fixed constraints, allowing the corresponding parameters to be adjusted by the combined observational and physicsinformed loss. Finally, LSP contains weak smoothness penalties on the first and second derivatives of ∆(z) and ΩDE(z).

Reconstruction of the varying Barrow exponent

The analysis reveals that the posterior mean of ∆(z) lies in the negative region over the whole redshift interval considered for both observational combinations. The inclusion of the Cepheid calibration produces an expected shift in the distance-scale parameters, favoring a larger value of H0 and a smaller value of rdrag. p. 5

Where the paper says it

The same is true for the holographic parameter C, whose posterior distributions are broad and largely compatible between PP and PPS. On the other hand, the inclusion of the Cepheid calibration produces the expected shift in the distance-scale parameters: the PPS reconstruction favors a larger value of H0 and, correspondingly, a smaller value of rdrag than the PP case.

The reconstruction displays a smooth redshift evolution, increasing towards less negative values at higher redshift before turning slightly downwards near the upper edge of the reconstruction range. p. 7
Where the paper says it

4, the PPS+BAO+CC combination shifts the posterior mean of ∆(z) further towards negative values, especially at low and intermediate redshifts. The reconstructed function again displays a smooth redshift evolution, increasing towards less negative values at higher redshift before turning slightly downwards near the upper edge of the reconstruction range.

Cosmological implications

The reconstructed effective dark-energy equation-of-state parameter wDE(z) remains close to the cosmological constant value wDE = −1 at low redshifts for both combinations. Both quintessence-like behavior, wDE > −1, and phantom behavior, wDE < −1, are allowed within the posterior region. p. 8

Where the paper says it

For the PP combination, wDE(z) exhibits only a mild evolution throughout most of the reconstructed range, while the uncertainty band progressively broadens towards higher redshift. Consequently, both quintessence-like behavior, wDE > −1, and phantom behavior, wDE < −1, are allowed within the posterior region, and a crossing of the phantom divide is not excluded.

p. 8
Where the paper says it

For the PP combination, wDE(z) exhibits only a mild evolution throughout most of the reconstructed range, while the uncertainty band progressively broadens towards higher redshift. Consequently, both quintessence-like behavior, wDE > −1, and phantom behavior, wDE < −1, are allowed within the posterior region, and a crossing of the phantom divide is not excluded.

The deceleration parameter q(z) exhibits the expected transition from a past decelerating phase to the current accelerated epoch, crossing zero at a redshift below z = 1. p. 8
Where the paper says it

The two observational combinations lead to very similar transition histories, although the PPS reconstruction presents slightly larger uncertainties at the upper end of the redshift interval. Thus, despite the freedom allowed in the redshift dependence of ∆(z), the reconstructed model naturally recovers the standard sequence of a past decelerating epoch followed by the current accelerated phase.

Conclusions

The central result is not a detection of an evolving entropy deformation but the direct determination of the functional behavior of ∆(z) allowed simultaneously by observations and by the generalized Barrow holographic dynamics. The reconstructed exponent leads to a consistent and phenomenologically viable late-time cosmological evolution. p. 0

Where the paper says it

The reconstructed cosmology yields a viable late-time evolution, with wDE close to −1 and the expected transition to accelerated expansion. Our results demonstrate that cosmological observations can directly probe the functional behavior of a quantity entering the underlying entropy law itself.

This allows the entropy deformation itself to be inferred from observations preserves the successful background cosmological evolution while considerably relaxing assumptions about its underlying functional form. p. 1
Where the paper says it

From the machine-learning perspective, the physicsinformed framework is used not simply to reconstruct a standard cosmological observable such as H(z) or wDE(z), but to infer a function characterizing the underlying horizon entropy. The analysis thus establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation, allowing the data to probe the functional behavior of a quantity associated with the quantum-gravitational structure of the model.

Data Availability

Data are available on reasonable request. The work demonstrates how physics-informed machine learning can connect cosmological observations with quantities characterizing the more fundamental theoretical structure of a model. p. 1

Where the paper says it

From the machine-learning perspective, the physicsinformed framework is used not simply to reconstruct a standard cosmological observable such as H(z) or wDE(z), but to infer a function characterizing the underlying horizon entropy. The analysis thus establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation, allowing the data to probe the functional behavior of a quantity associated with the quantum-gravitational structure of the model.

Acknowledgments

This work was partially supported by FONDECYT Grant 1240514. The authors would like to acknowledge the contribution of the LISA CosWG, and of COST Actions CA21136 “Addressing observational tensions in cosmology with systematics and fundamental physics (CosmoVerse)”, CA21106 “COSMIC WISPers in the Dark Universe: Theory, astrophysics and experiments (CosmicWISPers)”, and CA23130 “Bridging high and low energies in search of quantum gravity (BridgeQG)”. p. 1

Where the paper says it

The implementation of Barrow entropy in the holographic framework gives rise to Barrow holographic dark energy [79], characterized by ρDE ∝ L ∆−2 and exhibiting a rich cosmological behavior [79–103]. There is, however, no fundamental reason for the Barrow exponent to remain constant throughout cosmic history, since energy-scale dependence is common in quantum field theory and quantum-gravity settings.

The paper is 2608. p. 0
Where the paper says it

It is given by S ∝ A1+∆/2 , with A the horizon area and where the exponent ∆ quantifies the horizon deformation and thus the departure from standard Bekenstein-Hawking entropy, which is recovered for ∆ = 0. Let us mention that the possibility that quantumgravitational effects can drive the late-time acceleration has also been explored through different routes, such arXiv:2608.

27469.

--- Page 1 ---

Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks Spyros Basilakos,, Andronikos Paliathanasis Emmanuel N. Saridakis and Stylianos A. p. 0

Where the paper says it

Reconstructing the generalized Barrow holographic dark energy with physics-informed neural networks Spyros Basilakos , , 1, 2, 3, ∗ Andronikos Paliathanasis , , 4, 5, 6, 7, 8, † Emmanuel N. Saridakis , , 9, 10, 11, ‡ and Stylianos A.

Tsilioukas National Observatory of Athens, Lofos Nymfon, 11852 Athens, Greece Academy of Athens, Research Center for Astronomy and Applied Mathematics, Soranou Efesiou 4, 11527, Athens, Greece School of Sciences European University Cyprus Diogenes Street, Engomi, 1516 Nicosia Cyprus Institute of Systems Science Durban University of Technology Durban 4000 South Africa Centre for Space Research North-West University Potchefstroom 2520 South Africa Centro de Investigaci´on Innovaci´on y Creaci´on (CIIC) Universidad Cat´olica de Temuco Temuco Chile Departamento de Ciencias Matem´aticas y F´ısicas Facultad de Ingenier´ıa Universidad Cat´olica de Temuco Temuco Chile National Institute for Theoretical and Computational Sciences NITheCS South Africa Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing National Observatory of Athens 15236 Penteli Greece Departamento de Matem´aticas Universidad Cat´olica del Norte Avda. p. 0
Where the paper says it

Tsilioukas , 9, § 1National Observatory of Athens, Lofos Nymfon, 11852 Athens, Greece 2Academy of Athens, Research Center for Astronomy and Applied Mathematics, Soranou Efesiou 4, 11527, Athens, Greece 3School of Sciences, European University Cyprus, Diogenes Street, Engomi, 1516 Nicosia, Cyprus 4 Institute of Systems Science, Durban University of Technology, Durban 4000, South Africa 5Centre for Space Research, North-West University, Potchefstroom 2520, South Africa. 6Centro de Investigaci´on, Innovaci´on y Creaci´on (CIIC), Universidad Cat´olica de Temuco, Temuco, Chile 7Departamento de Ciencias Matem´aticas y F´ısicas, Facultad de Ingenier´ıa, Universidad Cat´olica de Temuco, Temuco, Chile 8National Institute for Theoretical and Computational Sciences (NITheCS), South Africa 9 Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing, National Observatory of Athens, 15236 Penteli, Greece 10Departamento de Matem´aticas, Universidad Cat´olica del Norte, Avda.

Angamos 0610 Casilla 1280 Antofagasta Chile CAS Key Laboratory for Research in Galaxies and Cosmology School of Astronomy and Space Science University of Science and Technology of China Hefei 230026 Barrow holographic dark energy connects cosmic acceleration with possible quantum-gravitational deformations of horizon entropy, encoded in the Barrow exponent ∆. p. 0
Where the paper says it

6Centro de Investigaci´on, Innovaci´on y Creaci´on (CIIC), Universidad Cat´olica de Temuco, Temuco, Chile 7Departamento de Ciencias Matem´aticas y F´ısicas, Facultad de Ingenier´ıa, Universidad Cat´olica de Temuco, Temuco, Chile 8National Institute for Theoretical and Computational Sciences (NITheCS), South Africa 9 Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing, National Observatory of Athens, 15236 Penteli, Greece 10Departamento de Matem´aticas, Universidad Cat´olica del Norte, Avda. Angamos 0610, Casilla 1280, Antofagasta, Chile 11CAS Key Laboratory for Research in Galaxies and Cosmology, School of Astronomy and Space Science, University of Science and Technology of China, Hefei 230026, China Barrow holographic dark energy connects cosmic acceleration with possible quantum-gravitational deformations of horizon entropy, encoded in the Barrow exponent ∆.

If such effects are scale dependent, however, there is no fundamental reason for ∆ to remain constant throughout cosmic history. p. 0
Where the paper says it

Angamos 0610, Casilla 1280, Antofagasta, Chile 11CAS Key Laboratory for Research in Galaxies and Cosmology, School of Astronomy and Space Science, University of Science and Technology of China, Hefei 230026, China Barrow holographic dark energy connects cosmic acceleration with possible quantum-gravitational deformations of horizon entropy, encoded in the Barrow exponent ∆. If such effects are scale dependent, however, there is no fundamental reason for ∆ to remain constant throughout cosmic history.

In this work we reconstruct ∆(z) directly from observations, without assuming any particular functional form, using the Cosmo-PINN physics-informed neural-network framework. p. 0
Where the paper says it

If such effects are scale dependent, however, there is no fundamental reason for ∆ to remain constant throughout cosmic history. In this work we reconstruct ∆(z) directly from observations, without assuming any particular functional form, using the Cosmo-PINN physics-informed neural-network framework.

The generalized Barrow holographic evolution equation is incorporated into the training, while PantheonPlus supernovae DESI DR2 baryon acoustic oscillations and cosmic chronometers constrain the reconstruction. p. 0
Where the paper says it

In this work we reconstruct ∆(z) directly from observations, without assuming any particular functional form, using the Cosmo-PINN physics-informed neural-network framework. The generalized Barrow holographic evolution equation is incorporated into the training, while PantheonPlus supernovae, DESI DR2 baryon acoustic oscillations and cosmic chronometers constrain the reconstruction.

We find a mild and smooth redshift evolution, with the posterior mean favoring negative ∆ and this tendency becoming stronger when the Cepheid calibration is included. p. 0
Where the paper says it

The generalized Barrow holographic evolution equation is incorporated into the training, while PantheonPlus supernovae, DESI DR2 baryon acoustic oscillations and cosmic chronometers constrain the reconstruction. We find a mild and smooth redshift evolution, with the posterior mean favoring negative ∆ and this tendency becoming stronger when the Cepheid calibration is included.

Nevertheless, ∆ = 0 and constant negative values remain compatible with current uncertainties. p. 0
Where the paper says it

We find a mild and smooth redshift evolution, with the posterior mean favoring negative ∆ and this tendency becoming stronger when the Cepheid calibration is included. Nevertheless, ∆ = 0 and constant negative values remain compatible with current uncertainties.

The reconstructed cosmology yields a viable late-time evolution, with wDE close to −1 and the expected transition to accelerated expansion. p. 0
Where the paper says it

Nevertheless, ∆ = 0 and constant negative values remain compatible with current uncertainties. The reconstructed cosmology yields a viable late-time evolution, with wDE close to −1 and the expected transition to accelerated expansion.

--- Page 2 ---

In its standard formulation, requiring that the energy contained in a region of characteristic size L does not exceed the mass of a black hole of the same size leads to a dark-energy density controlled by an infrared cutoff [8, 9]. The resulting sector is intrinsically dynamical and directly connects cosmological evolution with horizon thermodynamics. p. 0

Where the paper says it

In its standard formulation, requiring that the energy contained in a region of characteristic size L does not exceed the mass of a black hole of the same size leads to a dark-energy density controlled by an infrared cutoff [8, 9]. The resulting sector is intrinsically dynamical and directly connects cosmological evolution with horizon thermodynamics.

One particularly interesting case is Barrow entropy [73], which incorporates possible quantum-gravitational deformations. p. 0
Where the paper says it

Such phenomenological success led to the investigation of its extensions, either by changing the infrared cutoff [41–51], or by considering modified entropy-area relations [52–72]. One particularly interesting case is Barrow entropy [73], which incorporates possible quantum-gravitational deformations.

The implementation of Barrow entropy in the holographic framework gives rise to Barrow holographic dark energy [79], characterized by ρDE ∝ L(∆−2). p. 1
Where the paper says it

2 as the topological dark energy at the spacetime-foam level [74, 75] or through Wald-Gauss-Bonnet entropy [76– 78]. The implementation of Barrow entropy in the holographic framework gives rise to Barrow holographic dark energy [79], characterized by ρDE ∝ L ∆−2 and exhibiting a rich cosmological behavior [79–103].

--- Page 3 ---

We consider a spatially flat Friedmann-RobertsonWalker geometry, ds squared = -dt squared + a 2(t)δijdxi dxj. We choose the future event horizon as the infrared cutoff, namely L = Rh ≡ a(t) Z ∞ t dt' a(t') = a Z ∞ da' H(a')a'squared. The Barrow holographic darkenergy density becomes ρDE = CR(∆−2)h. p. 2

Where the paper says it

4) where H ≡ a/a ˙ . Hence, the Barrow holographic darkenergy density becomes ρDE = CR∆−2 h .

The Friedmann equations are then 3M 2p H squared = ρm + ρDE. p. 2
Where the paper says it

Hence, the Barrow holographic darkenergy density becomes ρDE = CR∆−2 h . 5) The Friedmann equations are then 3M2 p H2 = ρm + ρDE, (2.

--- Page 4 ---

We now generalize the construction by allowing the Barrow exponent to evolve throughout the cosmological history, namely ∆ = ∆(z). The resulting dynamics therefore contains explicitly the derivative of the Barrow exponent [104]. Using ln a = − ln(1+z) and the relation d/d ln a = -(1+z) d/dz, the evolution equation can be expressed directly in redshift space as (1 + z) omegaDE(1−omegaDE) domegaDE/dz = 1 + z ∆ − 2 d∆/dz ln 1 − omegaDE / ΩDE P(z) - (2 − ∆)r / ΩDE C / (1−omegaDE P(z)) ∆ 2(∆−2) + ∆ + 1. p. 2

Where the paper says it

The resulting dynamics therefore contains explicitly the derivative of the Barrow exponent [104]. Using ln a = − ln(1+z) and the relation d d ln a = −(1+z) d dz , the evolution equation can be expressed directly in redshift space as (1 + z) ΩDE(1−ΩDE) dΩDE dz = 1 + z ∆ − 2 d∆ dz ln 1 − ΩDE ΩDE P(z) −(2 − ∆)r ΩDE C 1−ΩDE ΩDE P(z) ∆ 2(∆−2) + ∆ + 1, (2.

--- Page 5 ---

Thus, both the value of the Barrow exponent and its redshift variation affect the effective dark-energy equation of state. In the constant-exponent limit, d∆/dz = 0, the standard Barrow holographic dark-energy equations are recovered. p. 2

Where the paper says it

16) Thus, both the value of the Barrow exponent and its redshift variation affect the effective dark-energy equation of state. In the constant-exponent limit, d∆/dz = 0, the standard Barrow holographic dark-energy equations are recovered, while for ∆ = 0 one further obtains ordinary holographic dark energy [8, 9, 79].

--- Page 6 ---

Equations (2.10), (2.11), and (2.16) form the background system that will be implemented in the physics-informed reconstruction. p. 2

Where the paper says it

11), and (2. 16) form the background system that will be implemented in the physicsinformed reconstruction.

Contrary to previous analyses, where a particular functional dependence for ∆(z) was imposed beforehand, in the present work no phenomenological ansatz for its redshift evolution is assumed. p. 2
Where the paper says it

16) form the background system that will be implemented in the physicsinformed reconstruction. Contrary to previous analyses, where a particular functional dependence for ∆(z) was imposed beforehand, in the present work no phenomenological ansatz for its redshift evolution is assumed.

--- Page 7 ---

We now turn to the central result of the analysis, namely the data-driven reconstruction of ∆(z). In both cases the posterior mean is found in the negative-∆ region over the whole redshift interval considered. p. 6

Where the paper says it

4 we present the reconstructed ∆(z) together with its 68% and 95% credible intervals for the PP+BAO+CC and PPS+BAO+CC combinations. In both cases the posterior mean is found in the negative-∆ region over the whole redshift interval considered, although the statistical uncertainties are sufficiently large that the standard holographic limit ∆ = 0 remains compatible with the reconstruction within the broader credible region.

The inclusion of the Cepheid calibration has a visible effect on the reconstructed exponent. p. 7
Where the paper says it

In particular, both the standard holographic value ∆ = 0 and constant negative values of ∆ remain compatible with the reconstruction over substantial parts of the redshift interval. The inclusion of the Cepheid calibration has a visible effect on the reconstructed exponent.

The reconstructed function again displays a smooth redshift evolution. p. 7
Where the paper says it

4, the PPS+BAO+CC combination shifts the posterior mean of ∆(z) further towards negative values, especially at low and intermediate redshifts. The reconstructed function again displays a smooth redshift evolution, increasing towards less negative values at higher redshift before turning slightly downwards near the upper edge of the reconstruction range.

--- Page 8 ---

It is worth emphasizing that these results have been obtained without assuming any of the linear, CPL-like, exponential or other parametrizations previously considered for a varying Barrow exponent. The suppression of the higher-order Chebyshev coefficients discussed in the previous subsection further indicates that the reconstructed behavior is mainly controlled by the lowest-order modes and is therefore smooth rather than being generated by rapid oscillations of the functional basis. p. 8

Where the paper says it

Instead, the shape of ∆(z) emerges from the combined observational information and the generalized Barrow holographic evolution equation. The suppression of the higher-order Chebyshev coefficients discussed in the previous subsection further indicates that the reconstructed behavior is mainly controlled by the lowest-order modes and is therefore smooth rather than being generated by rapid oscillations of the functional basis.

The effective dark-energy equation-of-state parameter wDE(z) remains close to the cosmological-constant value at low redshift for both combinations. Both quintessence-like behavior, wDE > −1, and phantom behavior, wDE < −1, are allowed within the posterior region. The deceleration parameter q(z) exhibits the expected transition from a past decelerating phase to the current accelerated epoch. p. 8

Where the paper says it

The two observational combinations lead to very similar transition histories, although the PPS reconstruction presents slightly larger uncertainties at the upper end of the redshift interval. Thus, despite the freedom allowed in the redshift dependence of ∆(z), the reconstructed model naturally recovers the standard sequence of a past decelerating epoch followed by the current accelerated phase.

The density parameters reproduce the standard evolution from matter domination to dark-energy domination. p. 9
Where the paper says it

5-7 therefore shows that the data-driven reconstruction of the Barrow exponent produces a cosmologically viable late-time history. In particular, it yields a dark-energy sector close to wDE = −1 at low redshift, a transition from deceleration to acceleration at recent times, and the expected succession from matter domination to dark-energy domination.

Final Remarks

The data-driven reconstruction of the Barrow exponent produces a cosmologically viable late-time history. The analysis illustrates how physics-informed machine learning can connect cosmological observations with quantities characterizing the more fundamental theoretical structure of a model. p. 1

Where the paper says it

From the machine-learning perspective, the physicsinformed framework is used not simply to reconstruct a standard cosmological observable such as H(z) or wDE(z), but to infer a function characterizing the underlying horizon entropy. The analysis thus establishes a direct link between late-time cosmological observations and the effective evolution of the entropy deformation, allowing the data to probe the functional behavior of a quantity associated with the quantum-gravitational structure of the model.

The crossing of omegaDE(z) and omegam(z) marks the transition from matter to dark-energy domination. p. 9
Where the paper says it

5-7 therefore shows that the data-driven reconstruction of the Barrow exponent produces a cosmologically viable late-time history. In particular, it yields a dark-energy sector close to wDE = −1 at low redshift, a transition from deceleration to acceleration at recent times, and the expected succession from matter domination to dark-energy domination.

The fact that ∆ = 0 remains allowed shows that present data constrain, but do not require, a departure from the standard Bekenstein-Hawking holographic framework. p. 8
Where the paper says it

Hence, the reconstruction reveals a mild preference of the posterior mean for negative values of the generalized Barrow exponent, with this tendency becoming stronger when the Cepheid calibration is included. At the same time, the present late-time data do not require a departure from the standard holographic limit, nor do they establish a statistically significant redshift variation of ∆.

--- Page 9 ---

FIG. 1: Evolution of the different loss components during the two training phases for the PP+BAO+CC (left panels) and PPS+BAO+CC (right panels) datasets. p. 5

Where the paper says it

6 0 5000 10000 15000 Epoch 10 1 10 2 total PP PPS 0 5000 10000 15000 Epoch 10 1 10 0 P D E 0 5000 10000 15000 Epoch 10 0 10 1 C C PP PPS 0 5000 10000 15000 Epoch 10 0 10 1 B A O 0 5000 10000 15000 Epoch 5 × 10 1 6 × 10 1 7 × 10 1 8 × 10 1 S N FIG. 1: Evolution of the different loss components during the two training phases for the PP+BAO+CC (left panels) and PPS+BAO+CC (right panels) datasets.

Upper panels: total loss Ltotal and physics-informed loss LPDE. p. 5
Where the paper says it

1: Evolution of the different loss components during the two training phases for the PP+BAO+CC (left panels) and PPS+BAO+CC (right panels) datasets. Upper panels: total loss Ltotal and physics-informed loss LPDE.

Lower panels: individual observational loss contributions LCC, LBAO, and LSN. p. 5
Where the paper says it

Upper panels: total loss Ltotal and physics-informed loss LPDE. Lower panels: individual observational loss contributions LCC, LBAO, and LSN.

The behavior is very similar for the PP+BAO+CC and PPS+BAO+CC combinations. p. 5
Where the paper says it

In the upper panels we show the total loss Ltotal and the physics-informed contribution LPDE, while the lower panels display separately the contributions associated with the CC, BAO and SNIa datasets. The behavior is very similar for the PP+BAO+CC and PPS+BAO+CC combinations.

--- Page 10 ---

FIG. 2: One- and two-dimensional marginalized posterior distributions for the cosmological parameters H0,omegam0, rdrag and C, obtained from the HMC/NUTS analysis for the PP+BAO+CC and PPS+BAO+CC dataset combinations. p. 6

Where the paper says it

0 rdrag 1 2 C PP PPS FIG. 2: One- and two-dimensional marginalized posterior distributions for the cosmological parameters H0, Ωm0, rdrag, and C, obtained from the HMC/NUTS analysis for the PP+BAO+CC and PPS+BAO+CC dataset combinations.

The matter density parameter is very similar in the two reconstructions with strongly overlapping posterior regions. p. 5
Where the paper says it

2 we present the one- and two-dimensional marginalized posterior distributions for Ωm0, H0, rdrag and C, for the two observational combinations. The matter density parameter is very similar in the two reconstructions, with strongly overlapping posterior regions.

The same is true for the holographic parameter C, whose posterior distributions are broad and largely compatible between PP and PPS. p. 5
Where the paper says it

The matter density parameter is very similar in the two reconstructions, with strongly overlapping posterior regions. The same is true for the holographic parameter C, whose posterior distributions are broad and largely compatible between PP and PPS.

On the other hand, the inclusion of the Cepheid calibration produces the expected shift in the distance-scale parameters. p. 5
Where the paper says it

The same is true for the holographic parameter C, whose posterior distributions are broad and largely compatible between PP and PPS. On the other hand, the inclusion of the Cepheid calibration produces the expected shift in the distance-scale parameters: the PPS reconstruction favors a larger value of H0 and, correspondingly, a smaller value of rdrag than the PP case.

--- Page 11 ---

FIG. 3: One- and two-dimensional marginalized posterior distributions for the Chebyshev coefficients cn entering (3. p. 7

Where the paper says it

2 c5 PP PPS FIG. 3: One- and two-dimensional marginalized posterior distributions for the Chebyshev coefficients cn entering the reconstruction of the Barrow exponent ∆(z), for the PP+BAO+CC and PPS+BAO+CC dataset combinations.

5) The coefficients of the lowest-order modes possess the broadest posterior distributions and therefore carry most of the freedom required to describe the reconstructed redshift dependence. p. 6
Where the paper says it

3, where we show the posterior distributions of the Chebyshev coefficients cn entering (3. The coefficients of the lowest-order modes possess the broadest posterior distributions and therefore carry most of the freedom required to describe the reconstructed redshift dependence.

--- Page 12 ---

FIG. 4: Reconstructed generalized Barrow exponent ∆(z) for the PP+BAO+CC and PPS+BAO+CC dataset combinations. p. 8

Where the paper says it

4 (z) PPS FIG. 4: Reconstructed generalized Barrow exponent ∆(z) for the PP+BAO+CC (left panel) and PPS+BAO+CC (right panel) dataset combinations.

The solid curves denote the posterior mean reconstruction, while the shaded regions correspond to the 68% and 95% credible intervals obtained from the HMC/NUTS posterior sampling. p. 8
Where the paper says it

4: Reconstructed generalized Barrow exponent ∆(z) for the PP+BAO+CC (left panel) and PPS+BAO+CC (right panel) dataset combinations. The solid curves denote the posterior mean reconstruction, while the shaded regions correspond to the 68% and 95% credible intervals obtained from the HMC/NUTS posterior sampling.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Reconstruct cosmological functions directly from observational data by using physics-informed neural networks to infer a quantity entering an underlying entropy law, such as the Barrow exponent delta z. This improved system can probe the functional behavior of a quantity associated with the quantum-gravitational structure of a model without imposing any specific functional form beforehand.

  2. Implement a training procedure that incorporates governing differential equations directly into the learning process, as described by incorporating the generalized Barrow holographic evolution equation directly into the learning process. This allows the reconstructed solution to satisfy both observational data and physical laws simultaneously, ensuring dynamical consistency throughout the training domain.

  3. Perform posterior inference using Hamiltonian Monte Carlo and No-U-Turn Sampler methods to quantify uncertainties for reconstructed cosmological parameters and functions, such as H(z), wDE(z), and q(z). This enables the system to determine 68% and 95% credible intervals for the inferred redshift evolution of delta z.

  4. Utilize a network architecture that represents the quantity being reconstructed through a finite Chebyshev expansion, ∆(x) = A tanh[g(x)], g(x) = X N n=0 cn Tn(2x − 1). This structure is designed to ensure stability and suppress spurious high-frequency oscillations, as evidenced by the suppression of the higher-order Chebyshev coefficients.

  5. Dynamically adjust the relative weights between different loss components using the GradNorm method during training, balancing the contributions of the physics and observational terms to the optimization. This prevents one data set from dominating optimization early on and ensures a compromise where both observational and physics-informed requirements are fulfilled.

Abstract

Barrow holographic dark energy connects cosmic acceleration with possible quantum-gravitational deformations of horizon entropy, encoded in the Barrow exponent Δ. If such effects are scale dependent, however, there is no fundamental reason for Δ to remain constant throughout cosmic history. In this work we reconstruct Δ(z) directly from observations, without assuming any particular functional form, using the Cosmo-PINN physics-informed neural-network framework. The generalized Barrow holographic evolution equation is incorporated into the training, while PantheonPlus supernovae, DESI DR2 baryon acoustic oscillations and cosmic chronometers constrain the reconstruction. We find a mild and smooth redshift evolution, with the posterior mean favoring negative Δ and this tendency becoming stronger when the Cepheid calibration is included. Nevertheless, Δ=0 and constant negative values remain compatible with current uncertainties. The reconstructed cosmology yields a viable late-time evolution, with w DE close to-1 and the expected transition to accelerated expansion. Our results demonstrate that cosmological observations can directly probe the functional behavior of a quantity entering the underlying entropy law itself.

Sources

Related papers