Kernel dependence of the Gaussian Process reconstruction of late Universe expansion history
Joseph P Johnson e1,addr1, H. K. Jassal e2,addr1,addr2 · 2025
Read Kernel dependence of the Gaussian Process reconstruction of late Universe expansion history by Joseph P Johnson e1,addr1, H. K. Jassal e2,addr1,addr2 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-04273.txt.
Introduction
Joseph Johnson and Jassal examine, in a 2025 paper, how the choice of a smoothing recipe steers reconstructions of the universe's recent expansion. Gaussian process methods rebuild cosmic history from data without assuming a model. The kernel, the recipe's core, decides how the curve may bend.
The authors rebuilt the expansion rate and its slowdown using both fixed-shape and more flexible kernels. Flexible, lower-order kernels handled noisy telescope data better, per a standard scoring test. Derivative quantities, which sharpen the physics, also shifted with kernel choice. The caution matters: a supposedly model-free method still smuggles in assumptions through its kernel, and honest cosmology should report that dependence.
Catalog record
| CREATOR | Joseph P Johnson e1,addr1, H. K. Jassal e2,addr1,addr2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-04273
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Kernel dependence of the Gaussian Process reconstruction of late Universe expansion history Joseph P Johnson e1,addr1, H. K. Jassal e2,addr1,addr2 Abstract In this work, we discuss model-independent reconstruction of the expansion history of the late Universe. We use Gaussian Process Regression (GPR) to reconstruct the evolution of various cosmological parameters such as Hubble parameter H ( z ) H(z) and deceleration parameter q ( z ) q(z) using observational data to train the GPR model. We look at the GP reconstruction of these parameters using stationary and non-stationary kernel functions. We examine the effect of the choice of kernel functions on the reconstructions. We find that using non-stationary kernels such as lower-order polynomial kernels is a better choice for the reconstruction if the training data set is noisy (such as H ( z ) H(z) data) as shown by the log marginal likelihood analysis. We also look at the reconstructions of the derivatives of H ( z ) H(z) and study the kernel dependence on the reconstruction other cosmological parameters such as the q ( z ) q(z) and the redshift of transition to the accelerated expansion. We see that reconstructed evolution of q ( z ) q(z) also indicate that lower-order polynomial kernels are a better choice for the reconstruction compared to the stationary kernels. ∎ 1 1 1 e-mail: [email protected] 2 2 2 e-mail: [email protected] Journal: Eur. Phys. J. C 1 Introduction Cosmological parameters continue to be determined to better and better precision. The availability of a large amount of good quality and diverse datasets is a key factor in getting to this precision. The datasets include observations of the type Ia supernovae, Baryon acoustic oscillations (BAO), Cosmic Microwave Background (CMB) Radiation, direct measurements of the Hubble parameter, and more recently, dark energy survey and dark energy spectroscopic instrument data [ 1 , 2 , 3 ] . These different datasets probe different cosmological quantities and their combinations give very tight constraints on the cosmological parameters. One of the most significant breakthroughs in cosmology happened when observations of Type Ia supernovae revealed that the expansion of the late Universe is accelerating [ 4 ] . This observation required the introduction of dark energy to the standard cosmological model, as ordinary matter or dark matter could not explain the accelerated expansion. The nature of dark energy is not known, although the cosmological constant is a simple and an elegant explanation. The \Lambda CDM model, a cosmological constant and cold dark matter universe model has been highly successful in explaining various cosmological observations [ 4 , 5 , 6 ] . However, the known theoretical problems such as fine-tuning problem and coincidence problem indicated that one might have to look at the alternatives to the \Lambda CDM model [ 7 , 8 , 9 ] . Typically, the alternatives include those based on the assumption that dark energy is a fluid, and those where the model is described by a scalar field and various other descriptions [ 10 ] . There have been many dynamic dark energy models proposed including parametric models such as CPL model, scalar field models like quintessence and k-essence, all within the framework of GR [ 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ] . Modified gravity theories such as f(R) gravity models also have been considered to be viable alternatives [ 20 , 21 , 22 , 23 , 24 , 25 ] . However, none of these models have been able to solve the issues completely [ 26 , 27 ] . With these issues, it is important to obtain the expansion history of the Universe from the observations in a model-independent manner [ 28 , 29 ] . There have been methods of model-independent reconstruction of the Hubble parameter. One of the most popular methods is cosmography, where the Hubble parameter is expanded as a Taylor series expansion using its derivatives [ 30 , 31 , 32 , 33 , 34 ] . Although it gives the evolution of the Hubble parameter without assuming any underlying model, at higher redshifts, it falls short of the desired accuracy. Hence we need to look for a better reconstruction method that is model-independent and accurate in the redshift range of observations [ 35 , 36 , 37 , 38 , 39 , 40 ] . Moreover, as the observations became more accurate and precise, discrepancies between the cosmological parameters estimated from the late and early Universe observations using \Lambda CDM model were revealed, most prominent of them being the H_{0} tension and the \sigma_{8} tension [ 26 , 41 , 27 , 42 ] . Therefore there is still room for improvement in determining cosmological parameters; this is especially true in constraining dark energy equation of state parameter. While there are theoretical models that attempt to solve this problem, attempts are also underway to understand if calibration and systematics are the main reasons for this tension [ 43 , 44 ] . Since there is no reason to prefer one phenomenological model over the other, model-independent methods of cosmological parameter fitting are therefore needed to reconstruct the expansion history of the Universe. A theoretical reconstruction can be done by determining the form of a scalar field potential or a fluid parameterisation by assuming an evolution history of the scale factor [ 45 , 46 , 47 , 48 ] . The expansion history of the Universe can be reconstructed with methods that do not assume an underlying cosmological model. Such methods include the Principal Component Analysis, Singular Value Decomposition, Gaussian Processes, Artificial Neural Networks and other machine learning tools [ 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 ] . While being efficient tools in cosmology; all the methods have their own individual merits and limitations. For example, the principal component analysis is very sensitive to the errors in the data and is also sensitive to outliers which makes the analysis unreliable. Moreover, the nonlinear relations between cosmological quantities can lead to spurious results [ 28 , 58 ] . Similarly, single value decomposition also assumed linearity [ 59 ] . Gaussian Processes are also a powerful tool which has been used for cosmological analysis [ 60 , 32 , 61 , 62 , 63 , 64 , 65 , 66 , 67 , 68 ] . By definition, any subset of a Gaussian Process follows a multivariate Gaussian distribution. This property can be used to predict the mean and variance of a new test point using a closed-form equation [ 69 ] . This makes it a powerful regression tool to reconstruct the expansion history of the Universe using the available data points. Compared to other regression/interpolation tools, the Gaussian Process Regression (GPR) has the advantage of being model-independent and being able to predict the Gaussian error at a test point in addition to the mean value, making it an ideal tool to study the derived cosmological parameters as well [ 60 , 32 , 65 , 70 ] . Even though it is a powerful tool, care must be