🔭 astrophysics

Nonparametric Variational Inference Reconstruction of the Cosmic Expansion History from SNe Ia -- the \texttt{charm2} code

The paper introduces `charm2`, a new nonparametric reconstruction code based on Information Field Theory and geometric variational inference, which finds that while some recent supernova datasets suggest potential deviations from the standard $\Lambda$ CDM model, current statistical evidence remains inconclusive in favoring evolving dark energy over the standard paradigm.

Original authors: Iason Saganas, Matteo Guardiani, Natalia Porqueres, Torsten Enßlin

Published 2026-04-28

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Iason Saganas, Matteo Guardiani, Natalia Porqueres, Torsten Enßlin

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 Cosmic Detective: Searching for the "Ghost" in the Universe

Imagine you are a detective trying to figure out how a car has been driving down a long, winding road. You weren't there to watch the whole trip, but you have a few snapshots: a photo of the speedometer at mile 10, another at mile 50, and a third at mile 100.

Based on these snapshots, you could assume the driver was using cruise control (this is the standard scientific model called $\Lambda$ CDM). It’s the simplest explanation: the car moves at a steady, predictable pace. But what if the driver was actually tapping the gas pedal and the brakes in a strange, rhythmic pattern? If you only assume cruise control, you’ll miss the "dance" the driver was doing.

This paper introduces a new tool called charm2 to help astronomers catch that "dance."

The Problem: The "Cruise Control" Bias

For decades, astronomers have used a model called $\Lambda$ CDM to explain the universe. It assumes that "Dark Energy"—the mysterious force pushing the universe apart—is constant, like a car set to a perfect, unchanging speed.

The problem is that scientists worry they might be biased. If we assume the universe is on cruise control, our math will force the data to look like cruise control, even if the universe is actually speeding up and slowing down in a complex way. This is called "model bias."

The Solution: The "charm2" Detective

The authors created charm2, a piece of software that acts like a "nonparametric" detective.

Instead of walking into the crime scene saying, "I bet the driver used cruise control," charm2 walks in and says, "I don't care how the driver drove; I'm just going to look at the snapshots and tell you exactly what the speed was at every single moment."

It doesn't assume a specific shape or pattern. It uses a sophisticated mathematical framework (called Information Field Theory) to let the data "speak for itself." It’s like looking at the tire tracks in the mud to reconstruct the car's path, rather than just guessing based on a map.

The Investigation: What did they find?

The team tested charm2 on several different "snapshots" (sets of Supernova data, which act as our cosmic speedometers). Here is the breakdown of their findings:

The Usual Suspects (Union2.1 & Pantheon+): When they looked at these data sets, the "detective" found that the universe mostly looks like it is on cruise control. The data and the standard model agreed.
The Strange Case (DESY5): When they looked at a newer, different set of data (DESY5), the detective saw something weird! The data suggested the universe might be "dancing"—speeding up or slowing down in a way that doesn't fit the cruise control model. This hints at Evolving Dark Energy (the idea that Dark Energy changes over time).
The "Wait a Minute" Moment (DESY5-Dovekie): When they applied the tool to a corrected, updated version of that same data, the "dance" mostly disappeared. This taught the scientists that the "weirdness" they saw might have just been a "glitch in the camera" (systematic errors in the data) rather than a real cosmic mystery.

The Catch: The "Noise" Problem

The authors end with a very important warning. They found that even though their tool is great, the current data is a bit "blurry" (noisy).

Imagine trying to watch a dance through a very foggy window. You might think you see a dancer spinning, but it might just be a smudge on the glass. The authors calculated that until our "telescope cameras" get about 7 times clearer, we can't be 100% sure if the universe is actually dancing or if we're just looking at cosmic fog.

Summary in a Nutshell

charm2 is a new, unbiased way to map the history of the universe's expansion. While it found some "weird" behavior in recent data, it also showed that much of that weirdness might just be errors in our measurements. It’s a more honest way of looking at the stars, ensuring we don't mistake a "glitch" for a "discovery."

This paper introduces charm2, a next-generation nonparametric Bayesian inference code designed to reconstruct the cosmic expansion history (specifically the evolution of the universe's energy density) using Type Ia Supernova (SNIa) data.

1. The Problem: Model Bias in Cosmology

Standard cosmological analyses typically rely on the $\Lambda$ CDM paradigm, which assumes dark energy is a constant ( $\Lambda$ ). While $\Lambda$ CDM is highly successful, recent observations (e.g., from DESI) suggest the possibility of Evolving Dark Energy (EDE).

The core problem is model bias: if researchers assume a specific parametric form (like $\Lambda$ CDM or a specific $w_0w_a$ model) and that form is incorrect, the resulting cosmological parameters will be biased. To avoid this, "agnostic" or "nonparametric" methods are needed to let the data "speak for itself" without imposing a predefined functional shape on the dark energy evolution.

2. Methodology: Information Field Theory & geoVI

The authors utilize Information Field Theory (IFT), a Bayesian framework for optimal nonparametric reconstruction. The technical workflow is as follows:

Signal Definition: Instead of modeling parameters like $\Omega_m$ or $w$ , the code reconstructs the signal $s(x) = \ln(\rho(x)/\rho_0)$ on a double-logarithmic scale, where $x = \ln(1+z)$ . In this coordinate system, the $\Lambda$ CDM model appears as a piecewise linear function, making deviations easier to identify.
Generative Model: The signal is modeled as a combination of a linear trend (representing the dominant cosmic constituent) and a Correlated Field Model (CFM). The CFM uses a stationary Gaussian process to capture oscillations or deviations around that linear trend.
Geometric Variational Inference (geoVI): This is the primary technical advancement over the predecessor (charm1). Standard variational inference often fails when the posterior distribution is "curved" (non-Gaussian). charm2 employs geoVI, which uses a coordinate transformation to "flatten" the curved posterior, allowing for a much more accurate Gaussian approximation and more reliable uncertainty estimates.
Response Operator: The code maps the reconstructed energy density to the observable distance modulus ( $\mu$ ) via the Friedmann equation and the luminosity distance integral.

3. Key Contributions

charm2 Code: A successor to charm1 built on the NIFTy Python package, offering improved handling of nonlinearities and more accurate posterior uncertainties.
Improved Uncertainty Quantification: By using geoVI, the code avoids the significant underestimation of uncertainties at high redshifts seen in previous linearized methods.
Robustness Testing: The authors demonstrate that the reconstruction is robust against different initializations of the "fluctuation parameter" ( $b$ ), which controls the expected amplitude of deviations from the mean trend.

4. Results and Findings

The authors applied charm2 to four datasets: Union2.1, Pantheon+, DESY5, and the recalibrated DESY5-Dovekie.

Consistency with $\Lambda$ CDM: For the Union2.1 and Pantheon+ datasets, the nonparametric reconstructions are consistent with the flat $\Lambda$ CDM signal. The Pantheon+ results notably show a deviation from Planck-based $\Lambda$ CDM at low redshifts, reflecting the ongoing "Hubble Tension."
The DESY5 "Anomaly": The original DESY5 data showed a feature at $z \sim [0.57, 1.17]$ that appeared compatible with Evolving Dark Energy. However, when applying the code to the updated DESY5-Dovekie (which corrected photometric calibrations), this feature was significantly reduced. This suggests the "signal" in the original DESY5 may have been a result of systematic errors rather than new physics.
Model Selection (ELBO): Using the Evidence Lower Bound (ELBO) as a proxy for Bayesian evidence, the authors found no conclusive evidence to prefer a nonparametric or EDE model over flat $\Lambda$ CDM. The simplicity of $\Lambda$ CDM (2 parameters) outweighs the better fit provided by more complex models.

5. Significance and Caveats

The paper provides a critical warning regarding signal-to-noise ratios. Through synthetic testing, the authors proved that at current SNIa noise levels, the ELBO (and Bayesian evidence) is "overconfident"—it tends to favor the simpler $\Lambda$ CDM model even when the underlying truth is actually Evolving Dark Energy.

Conclusion: The study concludes that while charm2 is a powerful tool for detecting deviations from standard cosmology, current supernova data is not yet precise enough to statistically validate nonparametric features over $\Lambda$ CDM. To truly distinguish between these models, the noise covariance in future surveys would need to be reduced by a factor of approximately 7.