Signals from the early Universe: a comprehensive search for primordial features in Planck CMB datasets

arXiv:2605.30236 · astro-ph.CO · Submitted 2026-05-28 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: I'm Vera, and with me are Jocelyn and Subrahmanyan, guest researcher.

Jocelyn: Today's paper: "Signals from the early Universe".

Vera: The gist: Several previously reported indications of oscillatory structure persist across different analyses, although none attains global statistical significance,

Jocelyn: First, who's behind it and why it matters.

Paper summary: Vera: So we're looking at this paper titled "Signals from the early Universe: a comprehensive search for primordial features in Planck CMB datasets." Essentially, they are systematically checking for those tiny wiggles or oscillatory patterns in the cosmic microwave background data that might come from inflation.

Jocelyn: It’s about comparing different ways of looking at these potential signals, specifically using two different processing pipelines from Planck data, the PR3 legacy release and the newer PR4 NPIPE processing, to see how much their results change based on which maps you use.

Subrahmanyan: That comparison between the legacy and the new pipeline is important because it tests whether what we see is just an artifact of how we process the data or if it’s something genuinely new coming from physics in the early Universe.

Vera: The main point they’re making here, based on that abstract, is that they are looking for these primordial oscillatory features using several different templates, like sharp features or resonant signals. They want to see if any of these structures show up in the CMB data.

Jocelyn: And what they found is a bit mixed: some indications of oscillatory structure that were reported before still stick across the different analyses they did, but none of them have reached a global statistical significance yet.

Subrahmanyan: That persistence suggests that if these features exist, they are relatively robust against the choice of analysis framework, which is a strong sign for any potential new physics we might be looking for.

Vera: They also mentioned that some anomalies that were reported in earlier studies get significantly reduced when you update the analysis to use the new CamSpec likelihood based on Planck PR4 products.

Jocelyn: That's interesting because it means the results aren't totally stable across all analyses, which points to how sensitive these searches are to the specific data and likelihood framework being used.

Subrahmanyan: From a theoretical standpoint, this reduction when switching likelihoods suggests that those earlier hints might have been statistical noise or artifacts of older methods rather than strong physical predictions.

Vera: They did look for specific frequencies in the range of omega from ten to one hundred and for all the templates they checked, these frequencies improved the fit to the CMB data by up to a delta chi squared of minus ten to minus fifteen compared to a featureless reference model.

Jocelyn: But they immediately followed that up by saying this improvement isn't supported by a Bayesian model comparison, which means it’s not statistically compelling enough on its own.

Subrahmanyan: That’s because when you add those extra parameters, the Occam penalty kicks in; adding complexity usually makes a model less preferred unless the data strongly demands it, and here the data didn't strongly demand those specific frequency improvements.

Vera: They also pointed out that including three or four additional parameters actually reduces how predictable these feature models are because of that penalty.

Jocelyn: So while they found some local improvements in fit quality for certain frequencies, the overall picture remains constrained by the broader statistical framework.

Subrahmanyan: The final point they make is that after properly accounting for the look-elsewhere effect, the significance of those preferred frequencies drops down to a global statistical significance of at most two point six sigma.

Vera: That’s a pretty low level of confidence, so what does this mean for us right now regarding these searches for primordial oscillations?

Jocelyn: It means that while we haven't found definitive proof of these features yet, the persistence across analyses suggests they might be there in some form.

Subrahmanyan: The paper is really pointing toward the need for better data and better observational techniques to confirm what might be subtle signals.

Vera: And they do suggest that future CMB experiments, like the Simons Observatory and LiteBIRD, will play a decisive role in distinguishing genuine primordial oscillations from just statistical fluctuations or modeling systematics.

Jocelyn: That’s important because those next-generation measurements are expected to improve the upper bounds on feature amplitudes by more than one order of magnitude.

Subrahmanyan: If we can get that level of sensitivity, it would allow us to characterize any potential primordial signal with a much higher precision.

Vera: So, looking at this paper "Signals from the early Universe: a comprehensive search for primordial features in Planck CMB datasets," what does it really tell us about where we are in the search for inflationary physics?

Jocelyn: It tells us that the search is still very much exploratory, but it’s also telling us that we need to be more rigorous about how we compare different data processing methods.

Subrahmanyan: The implication is that while the simple featureless power-law model remains the minimal description consistent with current data, a rich landscape of inflationary scenarios predicts these departures from slow-roll dynamics.

Vera: So, we're still waiting for those next big experiments to give us the definitive answer on whether those subtle wiggles in the CMB are real physics or just noise.

Conclusion: Vera: So we've been looking at this paper, "Signals from the early Universe: a comprehensive search for primordial features in Planck CMB datasets," and what they found is that while they didn't find any huge, obvious signals, there are still some hints of those tiny wiggles in the data.

Jocelyn: It’s about them systematically comparing two different ways of looking at the Planck data—the older PR3 version versus the newer PR4 version—to see if those potential primordial oscillations show up depending on how you process it.

Subrahmanyan: And what they discovered is that some earlier hints of structure still stick across those different analyses, even though none of them have reached a high level of statistical certainty yet.

Vera: That’s right, and the authors are really digging into these four different templates they tested, like sharp features or resonant signals, to see if any specific physical pattern is hiding in there.

Jocelyn: They also showed how updating the analysis to use the newer likelihoods actually helps reduce some of those earlier anomalies they found in older studies.

Subrahmanyan: It suggests that those previous hints might have been statistical noise or artifacts from the older methods, which is a pretty interesting thing for theoretical work because it means we need to be careful about what we interpret.

Vera: And the most important part for us right now is how this whole search points toward future experiments, like Simons Observatory and LiteBIRD, to get those features down to a much finer level of detail.

Jocelyn: They are expecting those next-gen measurements to improve the upper limits on the size of any potential feature by more than ten times what we currently have.

Subrahmanyan: If we can get that sensitivity, it means we could actually characterize any primordial signal with a lot more precision, which is exactly what theory needs to see to make a real connection.

Vera: So even though this specific search didn't find the definitive answer yet, it really underscores the need for those next-generation polarization measurements we talked about earlier.

Dipartimento di Fisica e Scienze della Terra, Universita degli Studi di Ferrara · INFN, Sezione di Ferrara

astro-ph.CO

Submitted: 2026-05-28

Updated: 2026-05-28

Comments: 21 pages + appendices, 22 figures, 9 tables

Journal ref: JCAP 10 (2026) 024

DOI: 10.1088/1475-7516/2026/10/024

Code: https://github.com/simonsobs/so_noise_models

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

Importance score: 79/100

The gist: The gist: Several previously reported indications of oscillatory structure persist across different analyses, although none attains global statistical significance, and some anomalies reported in

Key concepts

Phenomenological Templates: These are mathematical models used to describe potential deviations from the standard nearly scale-invariant power spectrum of primordial fluctuations. The study tested four specific templates: sharp features, resonant features, potential bumps/dips, and standard clock signals to see which best fit the data.
Likelihood Frameworks (PR3 vs. PR4): The analysis compared results from two different ways of processing Planck data: the legacy PR3 release and the newer PR4 (NPIPE) processing. This comparison showed how changing the underlying CMB maps and likelihood methods affects the detection of oscillatory signals.
Bayesian Model Comparison: This method is used to rigorously assess whether adding extra parameters (like those describing a feature) actually improves the model fit enough to justify those extra parameters. The study found that while localized fits improved, this improvement was not strong enough to overcome the 'Occam penalty' associated with adding complexity.
Look-Elsewhere Effect: This effect occurs when searching for signals across many possible frequencies or locations. It reduces the statistical significance of any single detected anomaly because you are essentially increasing the number of tests performed, making random fluctuations more likely to appear significant.

Terminology

Non-Gaussianity — 6× in this paper · explained in 2 episodes
Simons Observatory — 4× in this paper
Anisotropy — 3× in this paper · explained in 4 episodes
Bispectrum — 2× in this paper · explained in 3 episodes
Primordial non-Gaussianity — 2× in this paper · explained in 2 episodes
Robustness — 2× in this paper · explained in 5 episodes
Weak Lensing — 2× in this paper · explained in 2 episodes
Abundance — 1× in this paper · explained in 1 episode
Accuracy — 1× in this paper · explained in 1 episode
Adam — 1× in this paper · explained in 2 episodes
Bayesian Inference — 1× in this paper · explained in 5 episodes
Cold Dark Matter — 1× in this paper
Core — 1× in this paper · explained in 2 episodes
Decoupling — 1× in this paper · explained in 2 episodes
Homogeneity — 1× in this paper · explained in 2 episodes
Matter Power Spectrum — 1× in this paper · explained in 2 episodes
Quantum fluctuations — 1× in this paper · explained in 2 episodes
Wilkinson Microwave Anisotropy Probe — 1× in this paper

Summary

The gist: Several previously reported indications of oscillatory structure persist across different analyses, although none attains global statistical significance, and some anomalies reported in earlier studies are substantially reduced when updated to the new versions of the CamSpec likelihood using Planck PR4 products<ref:2605.30236#pg19>.

Search Methodology

The investigation systematically compared phenomenological templates for primordial oscillatory features using measurements from cosmic microwave background (CMB) anisotropies, specifically comparing the full PR3 legacy release with the PR4 (NPIPE) processing to assess how results depend on the choice of CMB maps and likelihood framework<ref:2605.30236#pg6>. To maximize sensitivity to rapidly varying oscillatory signals, unbinned likelihoods were employed<ref:2605.30236#pg19>. The analysis focused only on Planck datasets and likelihoods because ground-based CMB measurements are not expected to lead to a significant improvement in the feature constraints obtained from the unbinned Planck likelihoods<ref:2605.30236#pg5>.

Feature Templates Tested

The study investigated four different feature templates for the primordial power spectrum, defined by how deviations from a nearly scale-invariant power-law PPS are parametrized<ref:2605.30236#pg19>. These templates include:

Sharp feature signal (LIN): Linear oscillations arising from a temporary departure from slow-roll<ref:2605.30236#pg5>
Resonant feature signal (LOG): Logarithmic oscillations arising when the slow-roll potential is subject to a periodic modulation<ref:2605.30236#pg19>
Bump/dip in the potential (BUMP): Linear oscillations generated by a step, dip, or bump in the inflationary potential<ref:2605.30236#pg19>
Primordial standard clock signal (PSC): Classical or quantum oscillations of massive fields acting as standard clocks<ref:2605.30236#pg19>.

Constraint Analysis and Likelihood Comparison

The analysis utilized three different high-l CMB likelihoods: Plik-PR3 (unbinned version of the Planck PR3 legacy likelihood), CamSpec-PR3 (CamSpec likelihood applied to the Planck PR3 maps), and CamSpec-PR4 (CamSpec high-l likelihood based on the Planck PR4 NPIPE maps)<ref:2605.30236#pg19>. The analysis employed nested sampling via PolyChord with Cobaya for parameter inference<ref:2605.30236#pg19>. Global amplitude constraints showed that for the LOG template, upper bounds on the feature amplitude were broadly consistent across the three likelihood combinations<ref:2605.30236#pg19>.

Preferred Frequencies and Statistical Significance

A dedicated study analyzed the quantity ∆χ2 ≡ χ2X − χ2ΛCDM to identify preferred frequencies that yield the largest improvement in fit relative to the featureless PPS model<ref:2605.30236#pg19>. The analysis found that for all templates and likelihood combinations, preferred frequencies remained in the range ω ∼ 10–100, yielding improvements of ∆χ2 ∼ −10 to −15 relative to the featureless reference model<ref:2605.30236#pg5>. However, this improvement is not supported by a Bayesian model comparison, as the Bayes factor consistently does not favour the extended models<ref:2605.30236#pg19>. After accounting for the look-elsewhere effect, the significance of preferred frequencies was reduced to a global statistical significance of at most 2.6σ<ref:2605.30236#pg5>.

Future Forecasts

Forecasts for forthcoming CMB experiments like the Simons Observatory (SO) and LiteBIRD highlight the decisive role of next-generation polarisation measurements in distinguishing genuine primordial oscillations from statistical fluctuations and modelling systematics<ref:2605.30236#pg19>. The upper bounds or uncertainties on the feature amplitudes expected from the combination of Simons Observatory and LiteBIRD improve by more than one order of magnitude<ref:2605.30236#pg19>. In a scenario where a primordial feature is present in the data, next-generation experiments will be able to characterise the signal with high precision<ref:2605.30236#pg5>.

Conclusions

The search for primordial oscillatory features provides a unique window into early Universe dynamics, and while specific frequency bands yield an improved local fit to the data, a rigorous Bayesian evidence framework demonstrates that these improvements are insufficient to overcome the Occam penalty associated with the enlarged parameter space<ref:2605.30236#pg19>. The statistical significance of most localized anomalies is drastically reduced when accounting for the look-elsewhere effect, reaching a global significance of at most 1.7σ within the PR4 dataset<ref:2605.30236#pg5>. Exploiting the physical distinction that E-mode polarisation transfer functions preserve sharp oscillatory patterns with high contrast, future polarisation measurements will be essential for conclusively resolving current tentative hints<ref:2605.30236#pg19>.

How it works

The analysis relies on comparing different CMB likelihoods to test the robustness of results against alternative high-l likelihood constructions applied to the same PR3 maps<ref:2605.30236#pg19>. The methodology involves deriving a local Bayesian test statistic qS that accounts for the vast search volume of extra parameters while avoiding the arbitrary prior dependence associated with amplitude parameters<ref:2605.30236#pg5>. This framework maps local peaks to a calibrated global significance by moving through sequential quantities: the local fit quality (qL), the local Bayesian mass (qB), and the final look-elsewhere-corrected frequentist statistic (qS)<ref:2605.30236#pg5>.

Analytic Considerations

The analysis of oscillatory features in temperature and E-mode polarisation anisotropies involves several analytic steps<ref:2605.30236#pg19>. For linear oscillations, the angular power spectrum is approximated by an integral involving the Bessel function, leading to a formula that shows how the amplitude of linear oscillations is further suppressed by a factor of √l<ref:2605.30236#pg19>. Conversely, in the logarithmic case, the oscillation period widens proportionally to k, maintaining constant efficiency in the line-of-sight integration and preventing features from being washed out at high multipoles<ref:2605.30236#pg19>. The analysis also considers sub-Hubble regime acoustic oscillations where photon perturbations are governed by driven harmonic oscillations of the photon-baryon fluid<ref:2605.30236#pg19>.

Signal-to-Noise Ratio Diagnostic

The detectability of oscillatory features is quantified using the cumulative signal-to-noise ratio as a function of the maximum multipole lmax, defined by equation (D.25)<ref:2605.30236#pg5>. This diagnostic allows researchers to identify multipole ranges that contribute most significantly to the total signal-to-noise ratio and compare the relative constraining power of the T T and EE spectra for different feature templates<ref:2605.30236#pg5>. The results show that while detectability decreases as oscillation frequency is increased, this degradation is gradual rather than abrupt<ref:2605.30236#pg5>.

Standard Cosmological Parameters

The constraints on the six standard cosmological parameters remain stable across the different parametrisations adopted for the PPS, showing consistency between Plik-PR3, CamSpec-PR3, and CamSpec-PR4 datasets<ref:2605.30236#pg5>. Table 5 reports mean values and marginalised constraints at 68% CL on these parameters for the different datasets<ref:2605.30236#pg5>. The constraints on the scalar spectral index, ns, are consistently found around 0.965 to 0.967 across all likelihoods<ref:2605.30236#pg5>.

Forecast Results

In the absence of a primordial signal, future CMB observations will significantly tighten current constraints on the feature amplitudes for all templates, improving 95% CL upper limits by more than an order of magnitude<ref:2605.30236#pg5>. In the scenario where a primordial feature is present in the data, next-generation experiments will be able to characterise the signal with high precision, with the significance of the LOG template reaching 5.4σ<ref:2605.30236#pg5>. The projected sensitivity to feature amplitudes aligns well with general expectations derived from information matrix techniques<ref:2605.30236#pg5>.

Conclusion

The search for primordial oscillatory features provides a unique, yet challenging, window into the dynamics of the early Universe<ref:2605.30236#pg19>. The work concludes that while specific frequency bands yield an improved local fit to the data, a rigorous Bayesian evidence framework demonstrates that these improvements are insufficient to overcome the Occam penalty associated with the enlarged parameter space<ref:2605.30236#pg5>.

Improvements for AI systems

textbfStronger Likelihood Framework Robustness Testing: The improved AI system can systematically assess how results depend on the choice of CMB maps and likelihood framework by comparing full PR3 legacy release with PR4 (NPIPE) processing, ensuring that inferred features are not artifacts of specific data pre-processing choices, as stated in the abstract.
textbfEnhanced Sensitivity to Rapidly Varying Signals: The system can be optimized to maximise our sensitivity to rapidly varying oscillatory signals by employing unbinned likelihoods, which allows for a more precise assessment of high-frequency features compared to standard binning procedures that might smooth out the signal.
textbfContextual Model Comparison: The AI can rigorously perform Bayesian model comparison by assessing if improvements are supported, noting that the improvement is not supported by a Bayesian model comparison, as the Bayes factor consistently does not favour the extended models, thereby providing a more cautious interpretation of statistical significance.
textbfQuantifying Occam Penalty: The system can explicitly incorporate the penalty for complexity, recognizing that the inclusion of three or four additional parameters can reduces the overall predictability of the feature models and leads to an Occam penalty, allowing for a more conservative assessment when comparing feature models to the standard model.
textbfLook-Elsewhere Effect Correction: The improved system can utilize a calibrated frequentist test statistic (qS) derived from mapping local peaks to a calibrated global significance using the trial factor N, which addresses the look-elsewhere effect, leading to more reliable statistical conclusions regarding localized anomalies.
textbfModel-Specific Signal Characterization: The AI can differentiate between feature types by analyzing the distinct spectral signatures, such as how linear oscillations imprint a pattern of constant frequency on the CMB angular power spectra versus how logarithmic features stretch out towards higher multipoles, based on their respective integral solutions (D.13 and D.17).
textbf-Mode Specific Constraint Mapping: The system can map the sensitivity across different cosmological templates by analyzing the cumulative signal-to-noise ratio as a function of the maximum multipole lmax for each template, providing a diagnostic on which frequency regimes contribute most significantly to detectability (Figure 21 and 22).

Abstract

We investigate the presence of primordial oscillatory features in measurements of CMB anisotropies through a systematic comparison of phenomenological templates. Building upon previous searches for primordial features using Planck data, we compare the full PR3 legacy release with the PR4 (NPIPE) processing to assess how the results depend on the choice of CMB maps and likelihood framework. To maximise our sensitivity to rapidly varying oscillatory signals, we employ unbinned likelihoods. We find that several previously reported indications of oscillatory structure persist across different analyses, although none attains global statistical significance. Furthermore, some anomalies reported in earlier studies are substantially reduced when updated to the new versions of the CamSpec likelihood using Planck PR4 products. For all templates considered, we identify a small number of frequencies in the range ω about 10-100 that improve the fit to the CMB data by up to Δχ squared-10 to-15 relative to the featureless reference model. However, this improvement is not supported by a Bayesian model comparison. The inclusion of three or four additional parameters can reduces the overall predictability of the feature models and leads to an Occam penalty. Finally, after properly accounting for the look-elsewhere effect, the significance of the preferred frequencies is reduced, corresponding to a global statistical significance of at most 2.6σ. We present forecasts for forthcoming CMB experiments, highlighting the decisive role of next-generation polarisation measurements in distinguishing genuine primordial oscillations from statistical fluctuations and modelling systematics. The upper bounds or uncertainties on the feature amplitudes, expected from the combination of SO and LiteBIRD, improve by more than one order of magnitude.