CMB-lite Framework: Efficient CMB Analysis

• CMB-lite framework is a suite of tools and models designed for efficient simulation and analysis of CMB data using compressed likelihoods and modular techniques.
• It employs likelihood compression and foreground marginalization to isolate cosmological signals from systematic noise, ensuring precise parameter estimation.
• Scalable computation via optimized Boltzmann solvers, iterative map-making, and HPC parallelization enables rapid processing for next-generation CMB missions.

The CMB-lite framework refers to a suite of tools, models, and computational techniques for efficiently modeling, simulating, and analyzing cosmic microwave background (CMB) data with an emphasis on reduced complexity, compressed statistical likelihoods, modularity, and scalability. This approach is motivated by the need for rapid, robust, and interpretable cosmological inference in the presence of massive and increasingly complex CMB data sets. The CMB-lite paradigm encompasses specialized Boltzmann solvers optimized for sharp features, modular map-making initiatives, compressed likelihoods via automatic differentiation, and simulation pipelines tailored for next-generation missions. The goal is to decouple essential cosmological information from foregrounds and systematics, enabling lighter computation and streamlined analysis workflows for both current and forthcoming CMB polarization and temperature experiments.

1. Foundational Principles and Motivation

The CMB-lite framework emerges from the convergence of several critical requirements in contemporary CMB research: the exponential scaling of detector arrays ($10^4$–$10^5$ channels), increased complexity of multi-frequency data, and the need to efficiently extract cosmological information despite foreground contamination and instrumental systematics. Key principles include:

• Compression: Reduction of the high-dimensional likelihood function, typically spanning hundreds of bandpowers and dozens of nuisance parameters, into a manageable set of CMB-only bandpowers or compressed statistics (Balkenhol, 2024, Prince et al., 2024).
• Foreground Marginalization: Systematic marginalization over foreground contributions (dust, synchrotron, cross-correlations) at the bandpower extraction stage, producing a "lite" likelihood that dispenses with nuisance parameters in subsequent inference (Prince et al., 2024).
• Modularity: Design and deployment of analysis and simulation frameworks as composable modules (e.g., map-making operators, simulation blocks, Boltzmann solvers) to facilitate adaptation and extension for specialized models or instrument designs (Anand et al., 27 Jan 2025, Kashyap, 2024).
• Scalability: Efficient use of HPC architectures and parallelization (e.g., MPI, OpenMP, GPU offloading) for operations on sparse representations and large-scale linear systems (Anand et al., 27 Jan 2025, Bouhargani et al., 2021).
• Automatic Differentiation and Likelihood Optimization: The use of differentiable programming libraries for rapid likelihood minimization and covariance estimation via Hessian evaluation (Balkenhol, 2024).

2. Likelihood Compression and Statistical Modeling

Central to the CMB-lite approach is likelihood compression, transforming the multi-frequency data and complex foreground models into a set of foreground-marginalized CMB bandpowers, together with their covariance (Balkenhol, 2024).

• Procedure: Compression proceeds by constructing a reconstruction likelihood for the bandpowers, $\mathcal{L}^{\rm recon}$, which jointly fits the CMB signal and foreground model; minimization is utilized to obtain the CMB bandpower estimates, and the Hessian (second derivative) at the minimum provides the covariance matrix.
• Efficiency: The implementation via automatic differentiation (e.g., JAX in the candl library) reduces runtime to minutes on standard hardware, compared to MCMC methods requiring hours to days (Balkenhol, 2024).
• Statistical Form: Resulting compressed likelihoods are typically modeled either as multivariate Gaussians or offset-lognormal distributions to capture non-Gaussian features, especially in bandlimited or small-sky analyses (Prince et al., 2024):

$$p(\mathbf{D}_b) = \frac{1}{(\mathbf{D}_b - \mathbf{D}_0)\sigma \sqrt{2\pi}} \exp \left(-\frac{\left[ \ln(\mathbf{D}_b - \mathbf{D}_0) - \mu \right]^2}{2\sigma^2} \right)$$

• Impact: This lossless compression (with $<0.1\sigma$ biases and $<10\%$ errors versus the full multi-frequency likelihood (Balkenhol, 2024)) enables robust parameter estimation and facilitates joint-probe analyses with other cosmological observables.

3. Modular and Scalable Map-Making

Efficient construction of CMB sky maps from massive time-ordered data forms a core component of CMB-lite frameworks, exemplified by BrahMap (Anand et al., 27 Jan 2025) and MAPPRAISER (Bouhargani et al., 2021).

• Sparse Linear Systems: Map-making is cast as solving $d = P \cdot s + n$, where $P$ (the pointing matrix) is vast and sparse, and $n$ is correlated noise.
• Iterative Solvers and Preconditioning: Use of preconditioned conjugate gradient methods exploiting block-diagonal preconditioners:

$\hat{s} = (P^T N^{-1} P)^{-1} P^T N^{-1} d$

with the preconditioner $M = (P^T \text{diag}(N)^{-1} P)^{-1}$, leading to computationally feasible inversion in the presence of millions of pixels and time samples.

• Scalability: Distributed memory, HPC-optimized C++ routines, MPI/OpenMP parallelization, and Python bindings (e.g., via pybind11) enable strong and weak scaling to hundreds of nodes and GPU platforms.
• Modularity: LinearOperator abstraction from SciPy allows composition and extension, with operations such as filtering and noise modeling added as pluggable modules (Anand et al., 27 Jan 2025).

4. Cosmological Boltzmann Solvers and Feature Sensitivity

Cosmological Boltzmann codes adapted for CMB-lite frameworks focus on both speed and accurate resolution of sharp features, crucial for inflation model testing.

• Optimized Integration: CosmoLib implements an $\ell$-by-$\ell$ brute-force integration algorithm for $C_\ell$:

$$C_\ell = \int |\Delta_\ell^k|^2 \mathcal{P}(k) d\ln k$$

with $$\Delta_\ell^k = \int_0^{\tau_0} S(k,\tau) j_\ell[k(\tau_0-\tau)] d\tau$$ and efficient updates of $j_\ell(x)$ via recurrence relations and Chebyshev polynomial fits (Huang, 2012).

• Gauge Choices and Perturbations: Implementation in Newtonian gauge, with explicit evolution of dark energy perturbations and baryon physics; arbitrary $w(a)$ and sound speed models supported.
• Validation: Enhanced integrators provide sufficient accuracy for models predicting rapid oscillations, such as axion monodromy, showing sensitivity limits of current CMB spectra (Huang, 2012).
• Modularity: Modern solvers like CMBAns provide object-oriented (class-based) C interfaces for component swapping, truncated hierarchy control, and adaptive integration (Das et al., 2019).

5. Simulation Pipelines for Space Missions

Mission-specific simulation frameworks such as LiteBIRD Simulation Framework (LBS) (Tomasi et al., 7 Jul 2025) illustrate the essential features required for comprehensive E2E CMB experiment modeling.

• Component Modules: LBS provides modules for scanning strategy (quaternion-based pointing), generation of input sky maps (including CMB, dipoles, and diffuse foregrounds), noise models (white + $1/f$), and polarization systematics.
• Data Provenance: Provenance tracking via parameter files, environment snapshots, and random seed propagation ensures reproducibility in simulations.
• Integration and Partitioning: Support for MPI partitioning, modular pipeline assembly, and output of time-ordered data matrices for subsequent map-making.
• Scientific Validation: E2E pipelines based on LBS have been used to characterize instrumental sensitivity, destriping, and systematic control for LiteBIRD science goals (Tomasi et al., 7 Jul 2025).

6. Applications to Cosmological Inference and Model Selection

Compressed CMB-lite likelihoods facilitate cosmological parameter estimation and model selection with minimum computational overhead (Petretti et al., 2024).

• Foreground Marginalization: Lite likelihoods (e.g., BK-lite, SPT-3G lite) have no foreground nuisance parameters, allowing rapid and interpretable estimation of primordial parameters such as $r$ (tensor-to-scalar ratio), with $<$10% maximal effect from foreground modeling assumptions (Prince et al., 2024).
• Bandpower Distributions: Small-sky analyses necessitate the use of offset-lognormal statistics for bandpower distributions to handle likelihood non-Gaussianity (Prince et al., 2024).
• Model Selection: Synergy with missions such as LiteBIRD and CMB-S4 allows discrimination between various large-scale primordial features (e.g., step, kink, amplification) using cosmic-variance-limited $E$-mode data, with model selection bolstered by compressed likelihood pipelines (Petretti et al., 2024).

7. Summary Table: CMB-lite Frameworks and Their Domains

Component	Primary Function	Key Features / Methods
Likelihood Compression (Balkenhol, 2024, Prince et al., 2024)	Statistical marginalization, inference	Automatic differentiation, offset-lognormal bandpowers
Map-making (Anand et al., 27 Jan 2025, Bouhargani et al., 2021)	Sky map production from TOD	Sparse operators, iterative solvers, scalability
Boltzmann Solver (Huang, 2012, Das et al., 2019)	$C_\ell$ computation, feature sensitivity	$\ell$-by-$\ell$ brute-force integration, modular libraries
Simulation Pipeline (Tomasi et al., 7 Jul 2025)	Instrumental and sky modeling	Modular blocks, provenance tracking, realistic data generation

All frameworks emphasize modularity, scalability, and the ability to compress or isolate essential cosmological information for rapid, robust analysis.

• "A Cosmology Forecast Toolkit – CosmoLib" (Huang, 2012)
• "CMBAnalysis: A Modern Framework for High-Precision Cosmic Microwave Background Analysis" (Kashyap, 2024)
• "Compressed 'CMB-lite' Likelihoods Using Automatic Differentiation" (Balkenhol, 2024)
• "A foreground-marginalized 'BK-lite' likelihood for the tensor-to-scalar ratio" (Prince et al., 2024)
• "\texttt{BrahMap}: A scalable and modular map-making framework for the CMB experiments" (Anand et al., 27 Jan 2025)
• "A Simulation Framework for the LiteBIRD Instruments" (Tomasi et al., 7 Jul 2025)
• "Cosmic Microwave Background Anisotropy numerical solution (CMBAns) I: An introduction to $C_l$ calculation" (Das et al., 2019)
• "CMB Telescopes and Optical Systems" (Hanany et al., 2012)
• "Investigating the Origin of CMB Large-Scale Features Using LiteBIRD and CMB-S4" (Petretti et al., 2024)

The CMB-lite framework thus designates a collection of interconnected software and statistical practices for reduced-complexity, high-fidelity CMB analysis, simulation, and mission support, with continued evolution toward greater interoperability and computational efficiency as experimental demands scale.