The planned CMB-S4 experiment targets the tensor-to-scalar ratio rr at the σ(r)≈10−3\sigma(r)\approx 10^{-3} level through degree-scale B-mode polarization of the cosmic microwave background, but at these noise levels the measurement is gated by methodology rather than statistics: the CMB-S4 pipeline-comparison study of Bianchini et al. (ApJ 993, 105, 2025) found that three independent foreground-cleaning pipelines agree for simple foregrounds yet develop ≳1σ\gtrsim 1\sigma biases on rr for realistic spatially varying Galactic dust, and that marginalizing over foreground residuals removes the bias at a pipeline-dependent cost of up to 160% in σ(r)\sigma(r). That study explicitly leaves open how to quantify omitted real-data complications (time-ordered-data filtering, anisotropic noise, calibration/beam/bandpass systematics), how to debias the delensing-template noise spectrum without perfect foreground knowledge, how to build spatially adaptive component separation and principled bias-robust foreground-residual models, and how to combine mutually consistent pipelines into a defensible detection claim.

Background

The cosmic microwave background (CMB) is relic radiation from the early Universe. Its polarization decomposes into even-parity E modes and odd-parity B modes; a stochastic background of primordial gravitational waves (tensor spacetime perturbations) generated during inflation would imprint degree-scale B modes whose amplitude relative to scalar density perturbations is the tensor-to-scalar ratio rr. Measuring rr discriminates between inflation models and probes energy scales far beyond particle colliders. The tightest current limit, r0.05<0.036r_{0.05}<0.036 (95% credibility) from BICEP/Keck observations through 2018 combined with Planck, still sits well above the r∼0.001r\sim 0.001–0.003 range that popular models such as Starobinsky R2R^2 and Higgs inflation can reach; the CMB-S4 experiment — a next-generation ground-based array of small-aperture telescopes (SATs) surveying a low-foreground 'deep patch' around the South Pole, plus large-aperture telescopes (LATs) — is designed to either bound r≤0.001r\le 0.001 (95%) if r=0r=0 or detect r>0.003r>0.003 at ≥5σ\ge 5\sigma.

Three contaminants stand between the data and rr. First, gravitational lensing by large-scale structure converts E modes into B modes with power exceeding the primordial target; 'delensing' subtracts a lensing B-mode template reconstructed internally from LAT maps, and the likelihood must be debiased for the template's reconstruction-noise power spectrum NℓLTN_\ell^{\rm LT}. Second, polarized Galactic foregrounds — thermal dust at high frequencies and synchrotron at low frequencies — have spectral energy distributions (SEDs) that vary across the sky and can decorrelate between observing frequencies. Third, instrumental noise and systematics (calibration, beam, bandpass mismatches) enter below the target signal. Component separation removes foregrounds either parametrically — fitting foreground SED parameters via a 'mixing matrix' that maps foreground amplitudes and spectral indices to the observed frequency maps, by maximum likelihood — or blindly, e.g. the internal linear combination (ILC), a variance-minimizing linear combination of frequency maps with unit response to the CMB, and its variants: needlet ILC (NILC, ILC in a localized wavelet frame), constrained ILC (cILC, additionally nulling modeled components), and SMICA (spectral-matching independent component analysis).

The source study (Bianchini et al. 2025) compared three foreground-cleaning pipelines on map-based simulations of the CMB-S4 South Pole Deep Patch: (A) parametric maximum likelihood on auto- and cross-spectra between frequency maps; (B) harmonic-space ILC; (C) map-based parametric maximum likelihood — all with iterative internal delensing. For simple and intermediate-complexity foreground suites the pipelines agreed, with σ(r)=3\sigma(r)=3–5×10−45\times 10^{-4}; but for complex suites featuring spatially varying dust spectral index and frequency decorrelation, biases at the 11–2σ2\sigma level appeared. Extending each pipeline to marginalize over foreground residuals (freely varying decorrelation parameters; residual foreground power templates) restored unbiased rr at the cost of 15–20% larger σ(r)\sigma(r) for Pipelines A and B and up to 160% for Pipeline C, i.e. detection significances of only 1.8–4.2σ\sigma for an input r=0.003r=0.003. The simulations omitted time-ordered-data (TOD) filtering effects, anisotropic correlated noise, and instrument systematics; the lensing-template noise was debiased assuming perfect knowledge of the foregrounds. The authors conclude that mutual validation of independent pipelines is essential at these noise levels, and defer the remediation of all these gaps to future work.

Problem Statement

Develop and validate the foreground-cleaning and rr-inference methodology that can convert CMB-S4 South-Pole-Deep-Patch-class (ultra-low-noise) B-mode observations into a defensible constraint on, or detection claim for, the tensor-to-scalar ratio rr. All five requested components serve this single objective and must hold jointly at CMB-S4 noise levels:

  1. Quantify the impact on foreground residuals, lensing-template construction, and final rr inference of the real-data complications omitted in the source simulations: time-ordered-data (TOD) filtering effects, anisotropic and correlated inhomogeneous noise, calibration errors, beam systematics, bandpass mismatches, and realistic map-level filtering.

  2. Construct data-driven (non-oracle) estimates and debiasing of the lensing-template noise spectrum NℓLTN_\ell^{\rm LT} and of foreground-residual power that do not rely on perfect foreground knowledge and remain robust to unknown foreground higher-order statistics and to mismatch between assumed and true foreground models.

  3. Develop operational, data-driven spatially adaptive component separation that scales to realistic survey footprints while keeping uncertainties controlled — for example, via spatially varying mixing matrices (multi-patch or clustering schemes) or localized weighting (needlet/NILC-type variants).

  4. Establish a principled tradeoff between foreground-residual bias control and statistical power: minimally informative yet bias-robust residual models that reduce, and where possible eliminate, the pipeline-dependent sensitivity loss observed in the source study (σ(r)\sigma(r) inflation of up to 160% for one pipeline, versus 15–20% for the other two), preserving the ability to claim high-significance detections.

  5. Establish a validated multi-pipeline mutual-consistency framework for single-realization data — which and how many independent pipelines to include, and how to compare or combine their posteriors — that yields defensible detection claims.

The completion boundary, consistent with the source's validation methodology, is end-to-end demonstration on progressively realistic mock data yielding unbiased recovery of rr across foreground-complexity suites, exercising all five components above jointly and delivering each as an explicit, independently checkable artifact: (i) a quantified residual and systematics error budget for the omitted real-data complications; (ii) operationally defined, data-driven (non-oracle) debiasing of the lensing-template noise spectrum NℓLTN_\ell^{\rm LT} and of the foreground-residual power; (iii) an explicitly specified spatially adaptive component-separation method, demonstrated on at least the source's most complex foreground suite with spatially varying dust spectral index and frequency decorrelation; (iv) a stated residual-model family and its resulting bias-versus-σ(r)\sigma(r) tradeoff, with the recovered bias quoted in units of σ(r)\sigma(r) against the tolerance adopted for 'unbiased' recovery; and (v) a stated multi-pipeline selection-and-combination rule — which and how many independent pipelines are used, and how their posteriors are compared or combined. A demonstration that covers only a subset of these components, or that leaves its validation mock suites, pipeline set, or residual-model family unspecified, does not resolve the task. Deployment on actual CMB-S4 data is the eventual application but cannot precede the observations; the reviewable target is the validated framework and its analysis protocols.

References

  1. F. Bianchini, D. Beck, W. L. K. Wu, Z. Ahmed, S. Belkner, J. Carron, B. S. Hensley, C. L. Pryke, C. Umiltà (CMB-S4 Collaboration), 'CMB-S4: Foreground-Cleaning Pipeline Comparison for Measuring Primordial Gravitational Waves', Astrophys. J. 993, 105 (2025), DOI: 10.3847/1538-4357/ae03be, arXiv:2502.04300, https://arxiv.org/abs/2502.04300 (source paper)
  2. K. Wolz, S. Azzoni, C. Hervías-Caimapo, J. Errard, N. Krachmalnicoff, D. Alonso, et al., 'The Simons Observatory: Pipeline comparison and validation for large-scale B-modes', Astron. Astrophys. 686, A16 (2024), DOI: 10.1051/0004-6361/202346105, arXiv:2302.04276, https://arxiv.org/abs/2302.04276
  3. K. Wolz, S. Azzoni, C. Hervías-Caimapo, J. Errard, N. Krachmalnicoff, D. Alonso, et al., 'The Simons Observatory: component separation pipelines for B-modes' (updated version), arXiv:2407.06891, https://arxiv.org/abs/2407.06891
  4. Y. Liu, S. Azzoni, S. E. Clark, B. Hensley, L. Vacher, D. Alonso, et al., 'The Simons Observatory: Assessing the Impact of Dust Complexity on the Recovery of Primordial B-modes', J. Cosmol. Astropart. Phys. 2025, 024 (2025), DOI: 10.1088/1475-7516/2025-11-024, arXiv:2508.00073, https://arxiv.org/abs/2508.00073
  5. A. Steier, S. Ghosh, J. Delabrouille, 'Unbiased Primordial Gravitational Wave Inference from the CMB with SMICA', J. Cosmol. Astropart. Phys. 2026, 080 (2026), DOI: 10.1088/1475-7516/2026-02-080, arXiv:2510.26767, https://arxiv.org/abs/2510.26767
  6. A. Carones, 'Debiasing cosmological parameters from large-scale foreground contamination in Cosmic Microwave Background data', J. Cosmol. Astropart. Phys. 2026, 028 (2026), DOI: 10.1088/1475-7516/2026-05-028, arXiv:2510.20785, https://arxiv.org/abs/2510.20785
  7. BICEP/Keck Collaboration (P. A. R. Ade, Z. Ahmed, M. Amiri, et al.), 'BICEP/Keck XX: Component-separated maps of polarized CMB and thermal dust emission using Planck and BICEP/Keck Observations through the 2018 Observing Season', Astrophys. J. 999, 89 (2026), DOI: 10.3847/1538-4357/ae3e53, arXiv:2509.21648, https://arxiv.org/abs/2509.21648
  8. F. Rahman, P. Chingangbam, K. Huffenberger, T. Ghosh, D. Pogosyan, T. Souradeep, C. Park, 'Understanding the non-Gaussian nature of Galactic foreground emissions towards small scales', arXiv:2606.22596, https://arxiv.org/abs/2606.22596
  9. W.-Z. Chen, Y. Liu, Y.-M. Wang, H. Li, 'From South to North: Leveraging Ground-Based LATs for Full-Sky CMB Delensing and Constraints on rr', J. Cosmol. Astropart. Phys. 2026, 057 (2026), DOI: 10.1088/1475-7516/2026-02-057, arXiv:2507.19897, https://arxiv.org/abs/2507.19897
  10. A. Rizzieri, C. Leloup, J. Errard, D. Poletti, 'Cleaning Galactic foregrounds with spatially varying spectral dependence from CMB observations with fgbuster', arXiv:2510.08534, https://arxiv.org/abs/2510.08534
  11. W. Kabalan, A. Rizzieri, W. Sohn, A. Basyrov, A. Boucaud, B. Beringue, P. Chanial, E. Tsang King Sang, J. Errard, 'A GPU-Accelerated JAX Framework for Robust Parametric Component Separation and Clustering Optimization for CMB Polarization Satellites', arXiv:2604.08463, https://arxiv.org/abs/2604.08463
  12. Simons Observatory Collaboration (I. Abril-Cabezas, S. Adachi, P. Ade, et al.), 'The Simons Observatory: forecasted constraints on primordial gravitational waves with the expanded array of Small Aperture Telescopes', arXiv:2512.15833, https://arxiv.org/abs/2512.15833
  13. A. Carones, M. Migliaccio, G. Puglisi, C. Baccigalupi, D. Marinucci, N. Vittorio, D. Poletti, 'Multi-Clustering Needlet-ILC for CMB B-modes component separation', Mon. Not. R. Astron. Soc. 525, 3117 (2023), DOI: 10.1093/mnras/stad2423, arXiv:2212.04456, https://arxiv.org/abs/2212.04456
  14. BICEP/Keck Collaboration (P. A. R. Ade, Z. Ahmed, M. Amiri, et al.), 'BICEP/Keck XIII: Improved Constraints on Primordial Gravitational Waves using Planck, WMAP, and BICEP/Keck Observations through the 2018 Observing Season', Phys. Rev. Lett. 127, 151301 (2021), DOI: 10.1103/PhysRevLett.127.151301, arXiv:2110.00483, https://arxiv.org/abs/2110.00483
  15. CMB-S4 Collaboration (K. Abazajian, G. E. Addison, P. Adshead, et al.), 'CMB-S4: Forecasting Constraints on Primordial Gravitational Waves', Astrophys. J. 926, 54 (2022), DOI: 10.3847/1538-4357/ac1596, arXiv:2008.12619, https://arxiv.org/abs/2008.12619
  16. LiteBIRD Collaboration (K. Aizawa, H. Akamatsu, R. Akizawa, et al.), 'LiteBIRD Mission Overview after Mission Reformation', arXiv:2607.25285, https://arxiv.org/abs/2607.25285