⚛️ phenomenology

Stochastic analysis of finite-temperature effects on cosmological parameters by artificial neural networks

This paper employs artificial neural networks and stochastic optimization to analyze finite-temperature quantum gravity effects on cosmological parameters, demonstrating that incorporating new temperature-dependent density terms improves the fit to Planck data and suggests a non-negligible role for thermal quantum corrections in cosmological evolution.

Original authors: Armin Hatefi, Ehsan Hatefi, I. Y. Park

Published 2026-02-18

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Armin Hatefi, Ehsan Hatefi, I. Y. Park

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Fixing the Universe's "Thermostat"

Imagine the Universe as a giant, complex machine. For decades, physicists have been trying to write the perfect instruction manual for how this machine works. This manual is called the Standard Model of Cosmology (specifically the $\Lambda$ CDM model). It's a very good manual; it explains how galaxies form and how the Cosmic Microwave Background (CMB)—the "baby picture" of the Universe—looks.

However, there's a problem. The manual has a few glitches.

The Hubble Tension: When we measure how fast the Universe is expanding today, we get one number. When we look at the baby picture and calculate what the speed should be, we get a slightly different number. They don't match.
The Vacuum Energy Mystery: The manual says empty space should have a huge amount of energy, but observations show it has almost none. It's like a car engine that theoretically should be roaring at 10,000 RPM but is actually idling at 100.

The Authors' Idea:
The authors (Armin, Ehsan, and I. Y. Park) suggest that the manual is missing a crucial ingredient: Temperature.

In the early Universe, things were incredibly hot. The authors propose that when you take the heat of the early Universe into account, quantum physics (the rules of the very small) changes the "thermostat" of the cosmos. They aren't just looking at the cold, empty space we have today; they are looking at how the heat of the past leaves a "thermal fingerprint" on the laws of physics.

The New Ingredients: $\Omega_{\Lambda2}$ and $\Omega_{\Lambda3}$

To fix the manual, they added two new "knobs" or variables to their equations. Let's call them Knob A ( $\Omega_{\Lambda2}$ ) and Knob B ( $\Omega_{\Lambda3}$ ).

The Analogy: Imagine you are baking a cake. The standard recipe calls for flour, sugar, and eggs. But the cake tastes slightly off. The authors say, "Wait, we forgot that the oven was hotter than we thought! That heat changes how the flour reacts."
Knob A ( $\Omega_{\Lambda2}$ ): This represents a correction that gets stronger as the Universe gets hotter (in the past). Interestingly, their math showed this knob has to be negative. Think of it like a "negative sugar" that cancels out some of the sweetness to make the cake taste just right. In physics terms, this negative value is actually a good sign—it matches a specific mathematical rule called "dimensional regularization" used to clean up infinities in quantum physics.
Knob B ( $\Omega_{\Lambda3}$ ): This is another correction related to how the Universe curves and expands.

The Experiment: The "Cosmic Tuner"

To test if these new knobs work, the authors used a super-computer program called CLASS. Think of CLASS as a massive, high-tech simulator that predicts what the "baby picture" of the Universe (the CMB) should look like based on the settings you give it.

The Modification: They hacked the CLASS code to include these new temperature-dependent knobs.
The Comparison: They ran the simulator millions of times, tweaking the knobs, and compared the output to the real data from the Planck satellite (the most accurate map of the early Universe we have).
The Goal: They wanted to find the specific setting of the knobs that makes the simulator's output look exactly like the real Planck data.

The Secret Weapon: Artificial Intelligence (The "Smart Assistant")

Trying to find the perfect setting for 8 different knobs (the standard 6 plus the 2 new ones) is like trying to find a specific grain of sand on a beach by looking at one grain at a time. It would take forever.

So, the authors used Artificial Neural Networks (ANNs).

The Analogy: Imagine you have a very smart assistant who has seen millions of cake recipes. You tell the assistant, "I want a cake that tastes exactly like this specific cake." The assistant doesn't just guess; it learns the relationship between the ingredients and the taste.
How they used it: They trained the AI to predict how "close" a set of settings was to the real Planck data. Once the AI learned the pattern, it could instantly tell them which combination of knobs would get the best result, skipping the need to run millions of slow simulations.

The Results: A Better Fit

The results were promising:

The Negative Knob: They found that Knob A ( $\Omega_{\Lambda2}$ ) must be negative to fit the data. This wasn't a mistake; it was a confirmation of their quantum theory.
Better Accuracy: When they included these two new temperature-based knobs, the simulation matched the real Planck data much better than the old standard model. The "distance" between their prediction and reality shrank significantly.
The Hubble Tension: While they didn't completely solve the Hubble Tension (the disagreement on expansion speed), they showed that these thermal quantum effects are a piece of the puzzle that we can't ignore. They act like a fine-tuning mechanism that smooths out the rough edges of the standard model.

The Conclusion: Why This Matters

This paper is like a mechanic realizing that a car engine runs better when you account for the temperature of the air intake, not just the fuel mixture.

It bridges the gap: It connects the hot, chaotic early Universe with the cold, calm Universe we see today.
It uses new tools: It shows that combining deep physics with modern AI (machine learning) is a powerful way to solve problems that are too complex for human intuition alone.
It opens the door: While they didn't solve every mystery, they proved that "finite-temperature quantum gravity" (the heat of the early universe affecting gravity) is real and important. It suggests that to fully understand the Universe, we need to stop treating it as a cold, static place and start accounting for the heat of its birth.

In short: The authors took the standard model of the Universe, added a "heat correction" derived from quantum physics, used AI to tune the settings, and found that the Universe fits the data much better when we remember that it was once very, very hot.

1. Problem Statement

The paper addresses the Cosmological Constant (CC) problem, specifically the discrepancy between the observed vacuum energy and the massive values predicted by Quantum Field Theory (QFT). While conventional cosmology treats temperature effects via classical thermodynamics, the authors propose that finite-temperature Quantum Gravity (finite-T QG) effects introduce perturbative corrections to the cosmological constant.

The core hypothesis is that the total cosmological constant ( $\Lambda_{tot}$ ) is not a static value but a time-dependent quantity driven by the scale factor $a(t)$ due to loop-generated corrections in a finite-temperature environment. This leads to a modified expansion history:
$\Lambda_{tot} = \Lambda_1 + \Lambda_2 a^{-4} + \Lambda_3 a^{-2} + \dots$
This introduces two new density parameters, $\Omega_{\Lambda2}$ and $\Omega_{\Lambda3}$ , which scale similarly to radiation ( $a^{-4}$ ) and curvature ( $a^{-2}$ ), respectively. The central challenge is to determine if these quantum-induced parameters can be distinguished from standard radiation ( $\Omega_r$ ) and curvature ( $\Omega_K$ ) and whether their inclusion improves the fit to observational data (specifically the 2018 Planck CMB data) and potentially alleviates the Hubble tension.

2. Methodology

The authors employ a hybrid approach combining theoretical modification of Boltzmann codes with advanced machine learning (ML) techniques.

A. Theoretical Framework & Code Modification

Modified CLASS: The authors modified the Cosmic Linear Anisotropy Solving System (CLASS) Boltzmann code. They replaced the standard cosmological constant term with the temperature-dependent effective quantity derived from the 1PI (one-particle irreducible) effective action.
Modified Hubble Parameter: The Hubble parameter $H(t)$ was updated to include the new terms:
$H \propto \sqrt{1.523 \times 10^{-5}T + \left(\frac{2.725}{T}\right)^3 \Omega_{\Lambda1}h^2 + \Omega_M h^2 + \frac{T}{2.725}\Omega_{\Lambda2}h^2 + \frac{2.725}{T}\Omega_{\Lambda3}h^2}$
Non-Degeneracy Proof: The authors explicitly demonstrated (via Appendix A and numerical analysis) that $\Omega_{\Lambda2}$ and $\Omega_{\Lambda3}$ are not degenerate with standard radiation ( $\Omega_r$ ) or curvature ( $\Omega_K$ ). While they share scaling behaviors, their physical origins differ: $\Omega_r$ arises from on-shell particles (kinetic theory), whereas $\Omega_{\Lambda2/3}$ arise from virtual particle loop corrections (vacuum energy shifts).

B. Numerical Analysis & Machine Learning

Instead of standard Bayesian inference (which requires priors that are difficult to define for these new parameters), the authors adopted a frequentist framework augmented by Machine Learning:

Brute-Force Scanning: Generated tens of millions of data points across an 8-dimensional parameter space ( $\Omega_{\Lambda2}, \Omega_{\Lambda3}, h, \omega_b, \omega_{cdm}, A_s, n_s, \tau_{reio}$ ) to map the "distance" (Euclidean metric) between the model's power spectrum and the Planck 2018 data.
Artificial Neural Networks (ANN):
- Trained ANNs to act as surrogate models for the distance function $y = f(x)$ , where $x$ are the cosmological parameters.
- Architecture: 4 hidden layers with 16 neurons each, using ReLU activation.
- Purpose: To efficiently navigate the high-dimensional space and predict the minimum distance without running the computationally expensive CLASS code for every iteration.
Feature Ablation: Systematically removed input features (parameters) to quantify their importance. The increase in Mean Squared Error (MSE) upon removal indicated the parameter's contribution to model accuracy.
Stochastic Optimization: Used Simulated Annealing (SA) guided by the ANN gradients (BFGS algorithm) to locate the global minimum of the distance function, accounting for numerical noise and sampling variability.

3. Key Contributions

Implementation of Finite-T QG in Cosmology: Successfully integrated finite-temperature quantum corrections into the standard $\Lambda$ CDM framework via a modified CLASS code.
Introduction of New Parameters: Defined and constrained $\Omega_{\Lambda2}$ and $\Omega_{\Lambda3}$ as distinct, non-redundant cosmological parameters arising from quantum gravity.
Novel ML-Driven Inference: Developed a workflow using ANNs for parameter estimation and feature importance analysis in a context where likelihood functions are unavailable or difficult to construct.
Theoretical Consistency: Provided a physical justification for the sign of $\Omega_{\Lambda2}$ , linking it to dimensional regularization in renormalization procedures.

4. Key Results

Negative $\Omega_{\Lambda2}$ : The analysis revealed that the optimal value for $\Omega_{\Lambda2}$ is negative ( $\approx -2.06 \times 10^{-8}$ ). This aligns with theoretical expectations from dimensional regularization, where specific loop contributions yield negative shifts.
Improved Model Fit:
- The 8-parameter model (including $\Omega_{\Lambda2}$ and $\Omega_{\Lambda3}$ ) achieved a significantly lower distance to the Planck 2018 data (minimum distance $\approx 26.83$ ) compared to the 7-parameter model (minimum distance $\approx 28.12$ ).
- The 8-parameter model converged faster and showed higher predictive accuracy despite using fewer data points and a coarser grid.
Parameter Importance: Feature ablation studies showed that removing $\Omega_{\Lambda2}$ caused a larger increase in prediction error (MSE) than removing some standard parameters, indicating its critical role in defining the fine structure of the CMB power spectrum.
Hubble Tension: While the inclusion of these parameters improved the overall fit to the CMB power spectrum, the study did not fully resolve the Hubble tension. However, it suggests that higher-order finite-temperature effects play a non-negligible role in cosmological evolution, particularly during the radiation-dominated epoch.

5. Significance and Future Directions

Refining Cosmological Models: The work demonstrates that quantum gravitational corrections, often ignored in standard cosmology, can have measurable impacts on the CMB power spectrum and should be considered in precision cosmology.
Methodological Shift: The paper validates the use of ANN-based surrogate models and stochastic optimization as effective alternatives to traditional Bayesian inference when dealing with complex, non-analytic objective functions and poorly constrained new physics parameters.
Future Work: The authors note that while their results are promising, a full Bayesian analysis (using tools like Cobaya or MontePython) is required to rigorously quantify statistical uncertainties. Future investigations should also focus on polarization data constraints and higher-order thermal effects.

In summary, this paper provides a rigorous numerical and theoretical framework for testing finite-temperature quantum gravity effects, showing that these effects improve the fit to observational data and introduce physically meaningful, non-degenerate parameters to the standard cosmological model.