Compressed Gaussian likelihood for the Planck low- data

arXiv:2606.05166 · astro-ph.CO · Submitted 2026-06-03 · Read on arXiv

Listen

Radio episode about this paper

Transcript

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

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "Compressed Gaussian likelihood for the Planck low- data".

Jocelyn: Here's the gist of the paper. Compressed Gaussian likelihood for the Planck low-l data is presented to provide an analytic Gaussian form of a proxy for non-Gaussian CMB low-l likelihoods, making them compatible with Fisher matrix analyses The Problem Addressed The non-Gaussian nature of CMB low-l TT and EE likelihoods makes them incompatible with Fisher matrix analyses that require an analytic Gaussian χ2 The SRoll2 likelihood, which provides a state-of-the-art constraint on the reionization optical depth τ to date, is non-Gaussian and poses a challenge for analysis methods that assume an analytic Gaussian likelihood Many analyses sidestep this issue by replacing the full SRoll2 likelihood with a simple Gaussian prior on τ or neglecting the low-l data altogether The Proposed Solution The authors show that the χ2 of an offset log-normal likelihood takes a Gaussian form in the log-transformed power spectrum amplitudes and can therefore serve as a proxy for the true Gaussian likelihood of this variable in Fisher matrix analyses, without any explicit change of variables This observation directly justifies the approach used in earlier work

13, 18–21: and makes the compressed likelihood directly compatible with Fisher matrix analyses that previously could not incorporate the non-Gaussian low-l likelihood Construction Methodology The compression method involves several key steps to construct the Gaussian proxy 1.

Vera: Binning scheme: The multipoles are grouped into six bins B, B = 2, 3), (4, 5), (6, 7), (8–11), (12–15), and (16–29) 2.

Jocelyn: Piecewise offset log-normal fit: For each bin i, the binned log-likelihood is fitted with an offset log-normal function Lfit i DEE i = 1/(DEE i − D0,i)σi√2πe−

ln(DEE i −D0,i)−µi: 2/2σ2 3.

Vera: Piecewise scheme: A piecewise fitting scheme is introduced where the central piece containing the peak is fitted freely, and all adjacent pieces are constrained by requiring continuity of both the log-likelihood and its first derivative at each junction Validation and Results The compressed likelihood was validated against the full SRoll2 likelihood via MCMC analysis combined with Planck and ACT DR6 data, finding excellent agreement across all ΛCDM parameters and in extended cosmological models 1.

Paper summary: Vera: So, this paper "Compressed Gaussian likelihood for the Planck low- data" tackles a problem where the standard analysis tools just don't work well because they assume things are normally distributed when they aren't. It shows that you can take those non-Gaussian low-l E-mode polarization likelihoods and turn them into something that looks like a standard Gaussian form, which is what Fisher matrix analyses need.

Jocelyn: Right, so the core thesis here is that the chi square of an offset log-normal likelihood actually becomes a Gaussian form when you look at the log-transformed power spectrum amplitudes. That means we can use this compressed version as a proxy for the real, non-Gaussian likelihood in those Fisher matrix calculations without having to do any complicated variable changes.

Subrahmanyan: That's interesting because it directly addresses how we handle the non-Gaussian nature of CMB low-l data, which has been a headache for years when trying to get reliable constraints on things like the reionization optical depth tau

thirteen eighteen–twenty-one: <ref:2606.05166#pg2>. It basically says that this method justifies using Gaussian likelihoods in Fisher forecasts.

Vera: And they build on that by compressing the SRoll2 likelihood into just a few pieces—piecewise offset log-normal functions—and they validate it against the full SRoll2 likelihood using MCMC data combined with Planck and ACT DR6 data, and they found excellent agreement across all Lambda CDM parameters.

Jocelyn: It’s not just that it agrees well; they specifically found no visible bias in the means or negligible differences in the one sigma uncertainties when comparing their compressed likelihood to the full SRoll2 likelihood posteriors. That's a strong result for model fitting.

Subrahmanyan: The implication for cosmology is that we can now incorporate these low-l E-mode constraints into Fisher matrix analyses, which was previously impossible because of that non-Gaussian issue, meaning we get much tighter uncertainty estimates on parameters like tau and the logarithm of ten to the power ten to the power A <ref:2606.05166#pg1>.

Vera: They show this method works even when you extend the model a bit—like in a non-flat Lambda CDM model or a dark energy model with a time-varying equation of state, w zero and wa. This suggests this compression isn't just an artifact of one specific cosmology but is robust across different parameter spaces.

Jocelyn: It’s also useful because they show that the Fisher matrix uncertainty estimates from their compressed Gaussian likelihood actually agree well with the results from the full MCMC analysis, which is a key validation point for how this proxy works.

Subrahmanyan: The construction methodology involves binning multipoles into six bins, and then fitting each bin with a piecewise offset log-normal function, where they make sure the pieces connect smoothly in terms of both the likelihood and its first derivative. That's how they handle those asymmetric tails they mentioned earlier.

Vera: So, to summarize this paper "Compressed Gaussian likelihood for the Planck low- data", it proposes turning a complex, non-Gaussian low-l E-mode likelihood into a tractable Gaussian form using piecewise offset log-normal functions and showing that this can accurately represent the true likelihood for Fisher matrix calculations.

Jocelyn: And they validate this by comparing their compressed model against the full SRoll2 likelihood using MCMC data from Planck and ACT DR6, finding excellent agreement across all Lambda CDM parameters including tau and ln(ten to the power ten to the power A) <ref:2606.05166#pg1,against the full SRoll2 likelihood>.

Subrahmanyan: The conclusion of this work is that fitting these binned DEE l posteriors with those offset log-normal functions yields a chi square that takes a Gaussian form in the log-transformed power spectrum amplitudes, which can serve as a proxy for the true Gaussian likelihood in Fisher matrix analyses.

Vera: So, to wrap up, they release this compressed likelihood planckgaussianlowl as a Python package so that anyone doing low-l analysis can plug this into their Gaussian-likelihood based analyses easily. This makes the information from Planck low-l data much more accessible for parameter estimation tools.

Conclusion: Vera: So, we're wrapping up on this paper about the "Compressed Gaussian likelihood for the Planck low- data."

Jocelyn: Yeah, it’s about taking these messy, non-Gaussian measurements from Planck’s low-l E-mode polarization and turning them into something that fits nicely into standard analysis tools.

Subrahmanyan: The authors are showing how you can use a specific fitting technique, involving binning and log-normal functions, to get a Gaussian approximation of the likelihood.

Vera: Exactly. They’re making it compatible with Fisher matrix analyses where we usually need that smooth Gaussian shape to work properly. It’s about making the data easier to handle for parameter estimation tools like those.

Jocelyn: The implication is that we can now use these low-l constraints in more standard ways, without having to completely bypass the non-Gaussian nature of the original data.

Subrahmanyan: What it means for cosmology is that we get better, more reliable error bars on things like the reionization optical depth tau and parameters related to dark energy.

Vera: It’s a practical tool. They release this compressed likelihood as a Python package so anyone can easily plug it into their existing Gaussian-based analyses.

Jocelyn: So, it’s moving us toward making these low-l constraints more accessible for routine use in CMB studies.

Subrahmanyan: And the work they did validating this against the full SRoll2 likelihood across different cosmological models shows that this approximation holds up pretty well.

William H. Miller III Department of Physics & Astronomy Johns Hopkins University

astro-ph.CO

Submitted: 2026-06-03

Updated: 2026-11-10

Comments: The version published in PRD

Code: https://github.com/nanoomlee/planck-gaussian-lowl

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 86/100

The gist: Compressed Gaussian likelihood for the Planck low-l data is presented to provide an analytic Gaussian form of a proxy for non-Gaussian CMB low-l likelihoods…

Key concepts

Non-Gaussian Likelihood
CMB data, specifically the EE power spectrum at low multipoles (low-$\ell$), often exhibits non-Gaussian behavior. This means the data distribution does not follow a simple bell curve. Standard analysis methods that assume a Gaussian shape cannot accurately use this type of likelihood directly, leading to potential inaccuracies in parameter estimation.
Compressed Gaussian Likelihood
This is an analytic Gaussian approximation created for the non-Gaussian low-$\ell$ data. The authors show that by binning the data and fitting each bin with an offset log-normal function, the resulting $\chi^2$ statistic takes a Gaussian form in transformed variables. This allows it to serve as a reliable substitute (proxy) for the true likelihood in analyses requiring a Gaussian input.
Fisher Matrix Analysis
Fisher matrix analysis is a statistical tool used to estimate parameter uncertainties when analyzing data. It requires an analytic, Gaussian $\chi^2$ function to calculate these uncertainties reliably. The paper demonstrates that their compressed likelihood provides a valid Gaussian form, allowing Fisher matrix estimates for parameters like optical depth ($\tau$) and $A_s$ to agree well with full MCMC results.
Offset Log-Normal Fit
This is the mathematical technique used to model the binned low-$\ell$ data. For each group of multipoles (a bin), the likelihood is fitted using an offset log-normal function. The authors use a piecewise scheme, fitting a central piece freely and constraining adjacent pieces for continuity, to accurately represent the complex shape of the true posterior distribution.

Terminology

Summary

Compressed Gaussian likelihood for the Planck low-l data is presented to provide an analytic Gaussian form of a proxy for non-Gaussian CMB low-l likelihoods, making them compatible with Fisher matrix analyses

The Problem Addressed

The non-Gaussian nature of CMB low-l TT and EE likelihoods makes them incompatible with Fisher matrix analyses that require an analytic Gaussian χ2 The SRoll2 likelihood, which provides a state-of-the-art constraint on the reionization optical depth τ to date, is non-Gaussian and poses a challenge for analysis methods that assume an analytic Gaussian likelihood Many analyses sidestep this issue by replacing the full SRoll2 likelihood with a simple Gaussian prior on τ or neglecting the low-l data altogether

The Proposed Solution

The authors show that the χ2 of an offset log-normal likelihood takes a Gaussian form in the log-transformed power spectrum amplitudes and can therefore serve as a proxy for the true Gaussian likelihood of this variable in Fisher matrix analyses, without any explicit change of variables This observation directly justifies the approach used in earlier work [13, 18–21] and makes the compressed likelihood directly compatible with Fisher matrix analyses that previously could not incorporate the non-Gaussian low-l likelihood

Construction Methodology

The compression method involves several key steps to construct the Gaussian proxy

  1. Binning scheme: The multipoles are grouped into six bins B, B = 2, 3), (4, 5), (6, 7), (8–11), (12–15), and (16–29)

  2. Piecewise offset log-normal fit: For each bin i, the binned log-likelihood is fitted with an offset log-normal function Lfit i DEE i = 1/(DEE i − D0,i)σi√2πe−[ln(DEE i −D0,i)−µi]2/2σ2

  3. Piecewise scheme: A piecewise fitting scheme is introduced where the central piece containing the peak is fitted freely, and all adjacent pieces are constrained by requiring continuity of both the log-likelihood and its first derivative at each junction

Validation and Results

The compressed likelihood was validated against the full SRoll2 likelihood via MCMC analysis combined with Planck and ACT DR6 data, finding excellent agreement across all ΛCDM parameters and in extended cosmological models

  1. Full Likelihood Agreement: The compressed likelihood reproduces the posterior distributions of all six ΛCDM parameters in excellent agreement with the SRoll2 likelihood, with no visible bias in the means and negligible differences in the widths

  2. Fisher Matrix Agreement: The Fisher matrix uncertainty estimates on τ and ln(1010As) from our compressed likelihood agree well with the full MCMC posteriors, validating that the Gaussian χ2 formulation correctly captures the information content of the low-l EE likelihood for Fisher matrix analyses

  3. Model Extension: The compressed likelihood yields posteriors consistent with the SRoll2 likelihood across all parameters, with mean shifts ≲ 0. p. 61σ for all parameters in both a non-flat ΛCDM model and a dark energy model with a time-varying equation of state (w0, wa)

Implementation and Utility

The resulting compressed Gaussian likelihood planck-gaussian-lowl is released as a lightweight Python package allowing for straightforward incorporation of the Planck CMB low-l data into any Gaussian-likelihood-based analysis This package also incorporates the compressed low-l TT likelihood of Ref. [17] in a consistent Gaussian χ2 form It provides a simple and accurate tool for cosmological analyses that require a tractable analytic form of the low-l EE likelihood

Conclusion

The key observation is that fitting the binned DEE l posteriors with offset log-normal functions yields a χ2 that takes a Gaussian form in the log-transformed power spectrum amplitudes and can therefore serve as a proxy for the true Gaussian likelihood of this variable in Fisher matrix analyses To accurately capture the asymmetric tails of the binned posteriors, we introduce a piecewise fitting scheme where each piece is fitted by minimizing the likelihood-weighted sum of squared log-likelihood residuals We validate the compressed likelihood against the SRoll2 likelihood in three ways, confirming its fidelity across various cosmological models Our compressed likelihood, planck-gaussian-lowl, is publicly available at github.com/nanoomlee/planckgaussian-lowl as a lightweight Python package The central piece is exactly Gaussian in this variable by construction, peaked at zero, and is extended beyond the fitting range for illustration purposes

Acknowledgments

We thank Laura Herold and Benjamin Wandelt for useful discussions NL was supported by the Horizon Fellowship from Johns Hopkins University This work was carried out at the Advanced Research Computing at Hopkins (ARCH) core facility (rockfish.jhu.edu), which is supported by the National Science Foundation (NSF) grant number OAC1920103 The resulting marginalized posterior distributions are shown in Fig. p. 7 3

Appendix A

Figure 4 shows the normalized likelihood Li for each bin as a function of the transformed variable ln DEE i − Dcentral 0,i − µ˜central i, where Dcentral 0,i and µ˜central i are the parameters of the central piece By construction, the central piece (blue, dot-dashed) is exactly Gaussian in this variable by construction, peaked at zero, and is extended beyond the fitting range for illustration purposes The true SRoll2 posterior (black) closely follows this Gaussian shape within the fitting range (shaded region), confirming that the central piece accurately captures the curvature of the likelihood near the peak and validating its use in Fisher matrix analyses [Eq. (16)] The blue vertical line shows the PACT best-fit fiducial (τ = 0. p. 7 p. 706), which falls within the fitting range for all bins

Table I

The fitted parameters of the piecewise offset log-normal likelihood for the 6-bin scheme adopted in this work are summarized in Table I The fitted parameters for all bins are summarised in Table I The fitted pieces are shown in Fig. 2

Table II

The comparison of 1σ uncertainties on τ and ln(1010As) from MCMC and the Fisher matrix, using the same likelihood combination as in Fig. 3 is presented in Table II The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties

Table I (Continued)

The fitted parameters for all bins are summarised in Table I The fitted pieces are shown in Fig. 2

Figure 3

The marginalized posterior distributions of the six ΛCDM parameters from MCMC analyses using the full SRoll2 likelihood (black solid) and our compressed Gaussian likelihood (orange dashed), combined with Planck low-l TT, Planck high-l, and ACT DR6 data is shown in Fig. 3 The posteriors from our compressed likelihood (orange dashed) are in excellent agreement with those from the original SRoll2 likelihood (black solid) across all parameters In particular, τ and ln(1010As), which are most sensitive to the low-l EE likelihood, show no statistically significant bias in the means and negligible differences in the widths

Figure 4

The normalized likelihood Li(D EE i) for each of the six bins (black solid) as a function of the transformed variable ln D EE i − Dcentral 0,i − µ˜central i, where Dcentral 0,i and µ˜central i are the parameters of the central piece is shown in Fig. 4 The central piece (blue dot-dashed) is exactly Gaussian in this variable by construction, peaked at zero, and is extended beyond the fitting range for illustration purposes The shaded region indicates the fitting range of the central piece, within which the true SRoll2 posterior closely follows the Gaussian shape, confirming that the central piece accurately captures the curvature of the likelihood near the peak and validating its use in Fisher matrix analyses [Eq. p. 7 (16)] The blue vertical line shows the PACT best-fit fiducial (τ = 0.06), which falls within the fitting range for all bins

Figure 2

The normalized likelihood Li D EE i for each of the six bins (black solid) compared to the piecewise offset log-normal fit (orange dashed) and the single offset log-normal fitted over the ±1σ region for bins 1–3 and the ±3σ region for bins 4–6 is shown in Fig. 2 The piecewise fit accurately captures both the peak and the tails of the true binned posteriors, while the single log-normal significantly under(over)estimates the high-D EE tail for bins 1 and 3 (bin 2)

Figure 1

The Planck low-l EE power spectrum from the SRoll2 likelihood (black points) compared to the ACT DR6 best-fit ΛCDM spectrum with τ = 0.06 (blue curve) is shown in Fig. p. 3 1 The black points mark the maximum likelihood values of the individual D EE l posteriors, with error bars indicating the 1σ credible intervals The reionization bump is clearly visible at l ≲ 6

Figure 5

The TT likelihood, the Planck high-l temperature and polarization likelihood, and the ACT DR6 likelihood (ACTlite), replacing only the low-l EE component with either the SRoll2 likelihood or our compressed Gaussian likelihood is shown in Fig. 5 The resulting marginalized posterior distributions are shown in Fig. p. 5 3 The posteriors from our compressed likelihood (orange dashed) are in excellent agreement with those from the original SRoll2 likelihood (black solid) across all parameters In particular, τ and ln(1010As), which are most sensitive to the low-l EE likelihood, show no statistically significant bias in the means and negligible differences in the widths

Table II (Continued)

The 1σ uncertainties on τ and ln(1010As) from both Fisher matrix and MCMC analyses are presented in Table II The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties

The fitted parameters for all bins are summarised in Table I The fitted pieces are shown in Fig. 2 The fitted parameters for all bins are summarised in Table I

The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties

The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties The Fisher matrix uncertainties from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties

The Fisher matrix uncertainty estimates from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties The Fisher matrix uncertainty estimates from our compressed Gaussian likelihood agree well with the MCMC results, while those computed from the SRoll2 likelihood assuming Gaussianity in DEE l overestimate the uncertainties

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. Support Fisher matrix analyses with non-Gaussian likelihoods by using a proxy Gaussian chi squared, specifically showing that the χ2 of an offset log-normal likelihood takes a Gaussian form in the log-transformed power spectrum amplitudes. This allows for direct compatibility with methods like the Fisher bias formalism without explicit change of variables.

  2. Implement compressed low-l likelihoods in MCMC analyses by using a piecewise offset log-normal fitting scheme to more accurately capture the asymmetric tails of the binned posteriors, which a single log-normal fails to reproduce. This improved fidelity ensures that parameter constraints are more accurate than those obtained from simpler Gaussian approximations.

  3. Create a lightweight Python package named planck-gaussian-lowl that incorporates compressed low-l TT likelihoods in a consistent Gaussian chi squared form. This tool allows users to straightforwardly combine both compressed lowl TT and EE likelihoods, simplifying the incorporation of Planck CMB data into any Gaussian-likelihood based analysis.

Sources

Related papers