🔭 astrophysics

The choice of Planck CMB likelihood in cosmological analyses

This paper demonstrates that cosmological parameter constraints, particularly for extended models, are robust across different Planck map and likelihood combinations when augmented with ground-based data, while introducing a new marginalized dataset and likelihood to facilitate accurate multi-dataset analyses and revealing that the improved constraining power of the PR4 release stems primarily from polarization and high-multipole temperature data.

Original authors: Hidde Jense, Marc Viña, Erminia Calabrese, J. Colin Hill

Published 2026-02-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hidde Jense, Marc Viña, Erminia Calabrese, J. Colin Hill

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

Imagine the universe as a giant, ancient room that has been echoing with a faint, cosmic hum since its birth. This hum is the Cosmic Microwave Background (CMB). For decades, scientists have tried to listen to this hum to figure out the room's blueprint: How big is it? What is it made of? How old is it?

For a long time, the Planck satellite was the only microphone powerful enough to hear this hum clearly. But recently, new microphones on the ground (like the ACT and SPT telescopes) have been added to the mix, listening to the higher-pitched, finer details of the sound.

This paper is essentially a "quality control" report. The authors are asking: "Does it matter which version of the Planck microphone recording we use when we combine it with the new ground-based microphones?"

Here is the breakdown of their investigation using simple analogies:

1. The Two Versions of the Recording

The Planck team released two different "editions" of their data over the years:

The "Legacy" Edition (PR3): This is the classic, well-loved recording. It was processed with a specific set of rules (called the Plik likelihood).
The "NPIPE" Edition (PR4): This is the newer, remastered version. The scientists cleaned up some static and noise in the recording process, allowing them to use a slightly larger portion of the sky. It was processed with a different set of rules (called CamSpec).

The Problem: When scientists used these two versions alone, they got slightly different answers about the universe's blueprint. It was like listening to the same song on two different speakers and hearing the bass slightly differently.

2. The "Noise" in the Room

The universe isn't just a clean hum; there's "noise" in the room. Dust from our own galaxy and other cosmic clutter can muddy the signal.

The Old Way: The "Legacy" version tried to mathematically subtract this noise while analyzing the data.
The New Way: The "NPIPE" version tried to clean the noise before the analysis even started (like cleaning the microphone before recording).

The authors created a new tool called CamSpec-NPIPE-lite. Think of this as a "stripped-down" version of the new recording. They removed all the messy noise and "nuisance" factors beforehand, leaving only the pure cosmic signal. This makes it much easier to combine with the new ground-based microphones without the math getting confused.

3. The Big Test: Mixing the Microphones

The main goal of the paper was to see what happens when you mix the Planck satellite (listening to the deep, low-frequency rumble of the universe) with the ground telescopes (listening to the high-frequency details).

Scenario A: Mix the "Legacy" Planck data with ground data.
Scenario B: Mix the "NPIPE" Planck data with ground data.

The Result:
When they looked at the full Planck data alone, the two versions disagreed slightly (about 1 "sigma," which is a statistical way of saying "a little bit different").
However, when they combined Planck with the ground telescopes, the disagreement vanished.

The Analogy:
Imagine you are trying to guess the weight of a giant elephant.

Planck alone is like looking at the elephant from far away. You can see it's big, but you aren't sure if it's 5,000 lbs or 6,000 lbs.
Ground telescopes are like standing right next to the elephant and measuring its legs.
The Paper's Finding: When you combine the "far view" (Planck) with the "close view" (Ground), it doesn't matter if you used the "Legacy" or "NPIPE" version of the far view. The close-up measurements are so precise that they override the tiny differences in the far view. The final answer is the same regardless of which Planck version you started with.

4. Why This Matters

The authors also looked at "Extended Models." These are theories that say the universe might be weirder than we thought (e.g., maybe there are extra types of invisible particles, or the universe is expanding in a weird way).

They found that when you add the ground telescope data, the results for these weird theories become completely insensitive to which Planck version you use. The ground data is so strong that it locks the answer in place, making the choice of Planck software irrelevant.

The Bottom Line

The paper is a huge relief for cosmologists. It confirms that:

The new "remastered" Planck data (NPIPE) is valid and robust.
We can safely mix the old and new Planck data with ground-based telescopes without worrying about getting conflicting results.
The "noise" (foregrounds) has been handled correctly in their new tools.

In short: The universe's blueprint is stable. Whether you use the classic Planck recording or the new remastered one, once you add the high-definition ground data, everyone agrees on the shape of the cosmos.

1. Problem Statement

Cosmological analyses relying on the Cosmic Microwave Background (CMB) have historically depended on the Planck satellite mission. However, the community currently utilizes two distinct sets of Planck sky maps and likelihood pipelines:

PR3 (Legacy): Uses "Legacy" maps processed with the Plik likelihood (the standard reference for PR3).
PR4 (NPIPE): Uses "NPIPE" maps (processed with improved systematics handling) typically analyzed with the CamSpec likelihood (the standard reference for PR4).

While both datasets yield similar cosmological parameters when used in isolation, they exhibit small but non-negligible shifts (up to $\sim 1\sigma$ ) in parameters like the baryon density ( $\Omega_b h^2$ ). Furthermore, modern cosmology increasingly combines Planck data with ground-based experiments (ACT, SPT) that probe smaller angular scales ( $\ell > 1500$ ). A critical open question is whether the choice between the Plik-Legacy and CamSpec-NPIPE pipelines affects the robustness of these combined constraints, particularly for extended cosmological models (e.g., varying $N_{eff}$ , $A_L$ , or dark energy equation of state $w_0, w_a$ ). Additionally, there is a need for a "lite" likelihood version of CamSpec-NPIPE that marginalizes over foregrounds to allow for correct statistical combination with other datasets.

2. Methodology

The authors employed a rigorous statistical framework to compare the two pipelines and develop a new combined product:

Framework Implementation: All likelihoods were implemented within the Simons Observatory MFLike framework to ensure consistent handling of input spectra, covariances, and marginalization procedures.
Derivation of CamSpec-NPIPE-lite:
- The authors developed a foreground-marginalized version of the CamSpec-NPIPE likelihood.
- They utilized Gibbs sampling to extract CMB-only bandpowers ( $D^{CMB}_\ell$ ) from the multi-frequency data, marginalizing over nuisance parameters (foreground amplitudes and spectral indices) independently of the cosmological model.
- This resulted in a compressed Gaussian likelihood (CamSpec-NPIPE-lite) that retains the CMB constraining power while removing the need to model foregrounds explicitly during cosmological inference.
Comparative Analysis:
- Full vs. Lite: Validated the new lite likelihood against the full multi-frequency CamSpec-NPIPE and the official Plik-Legacy.
- Signal-to-Noise (SNR) Localization: Calculated per- $\ell$ SNR to identify where the additional constraining power of NPIPE maps resides compared to Legacy maps.
- Restricted Multipole Ranges: Tested the pipelines using truncated multipole ranges ( $\ell < 1000/600/600$ in TT/TE/EE), which is the standard practice when combining Planck with ground-based data (ACT/SPT).
- Combined Analyses: Performed joint analyses combining Planck (both Plik and CamSpec variants) with ACT DR6 data to form the "P-ACT" combination.
- Extended Models: Tested standard $\Lambda$ CDM and extensions ( $\Lambda$ CDM+ $A_L$ , $\Lambda$ CDM+ $N_{eff}$ , $\Lambda$ CDM+ $dn_s/d\ln k$ ) and dynamical dark energy ( $w_0 w_a$ ).

3. Key Contributions

Release of CamSpec-NPIPE-lite: The authors derived and released a nuisance-marginalized dataset and likelihood for the Planck NPIPE data. This product allows for the correct combination of Planck NPIPE data with other CMB datasets (like ACT) without double-counting foreground uncertainties.
Quantification of Constraining Power: They demonstrated that the additional constraining power of the PR4 (NPIPE) maps over PR3 (Legacy) comes primarily from:
- Polarization at all scales.
- Temperature at multipoles $\ell > 1500$ .
- This is attributed to the increased sky fraction retained in NPIPE maps ( $\sim 10\%$ more data) and improved foreground cleaning.
Robustness of Combined Constraints: They established that when Planck large-scale data is combined with small-scale ground-based data (ACT DR6), the choice of Planck likelihood (Plik vs. CamSpec) becomes completely irrelevant for cosmological parameter inference.
Foreground Marginalization Impact: They showed that failing to properly marginalize over foreground nuisance parameters when truncating scales can impact parameter inference and model selection, a pitfall their new "lite" likelihood avoids.

4. Key Results

Parameter Consistency:
- For the full dataset, Plik-Legacy and CamSpec-NPIPE yield consistent results but with shifts of up to $\sim 1\sigma$ in $\Omega_b h^2$ and smaller shifts in other parameters.
- The CamSpec-NPIPE-lite likelihood recovers the full likelihood results with excellent agreement, showing only a minor ( $\sim 5-10\%$ ) broadening of uncertainties due to foreground marginalization.
Restricted Multipole Ranges:
- When restricting Planck data to $\ell < 1000$ (TT) / $600$ (TE/EE), the differences between Plik and CamSpec shrink significantly. The shift in $\Omega_b h^2$ reduces from $\sim 1\sigma$ (full data) to $\sim 0.4\sigma$ (cut data).
Combined Analyses (P-ACT):
- Combining Planck (either Plik or CamSpec) with ACT DR6 reduces the parameter differences to negligible levels (e.g., $\Delta \chi^2 \approx 0.4-0.7$ ).
- The largest remaining difference is $0.15\sigma$ in $\Omega_b h^2$ and $0.7\sigma$ in $\Omega_c h^2$ , which is statistically insignificant.
- Extended Models: Constraints on $A_L$ and $N_{eff}$ are identical regardless of the Planck likelihood choice when combined with ACT.
Dark Energy ( $w_0 w_a$ ):
- In the P-ACT-LBS combination (Planck + ACT + BAO + Supernovae), the statistical preference for dynamical dark energy ( $w_0 w_a$ ) over $\Lambda$ CDM is $\sim 2.4\sigma - 2.5\sigma$ .
- This preference is insensitive to whether Plik-Legacy or CamSpec-NPIPE is used, nor is it significantly affected by the specific treatment of foreground marginalization (changing the significance by only $\sim 0.1\sigma$ ).

5. Significance

This work provides a crucial validation for the current era of "hybrid" CMB cosmology. As the field moves toward combining space-based (Planck) and ground-based (ACT, SPT) data to constrain extended models and dark energy, the choice of the specific Planck release (PR3 vs. PR4) or likelihood pipeline (Plik vs. CamSpec) is shown to be robust.

The release of CamSpec-NPIPE-lite is a significant technical contribution, enabling the community to utilize the improved NPIPE maps and foreground cleaning of PR4 in combined analyses without the computational burden or statistical risks of modeling foregrounds manually. The results confirm that the leading cosmological constraints derived from the combination of Planck and ground-based experiments are stable and reliable, regardless of the specific Planck product chosen.