Compatibility of Planck and BICEP2 results in light of inflation

Jérôme Martin, Christophe Ringeval, Roberto Trotta, Vincent VenninPhysical 2014 · DOI · arXiv

Abstract: We investigate the implications for inflation of the detection of $B$-modes polarization in the cosmic microwave background by BICEP2. We show that the hypothesis of the primordial origin of the measurement is favored only by the first four band powers, while the others would prefer unreasonably large values of the tensor-to-scalar ratio. Using only those four band powers, we carry out a complete analysis in the cosmological and inflationary slow-roll parameter space using the BICEP2 polarization measurements alone and extract the Bayesian evidences and complexities for all the Encyclopædia Inflationaris models. This allows us to determine the most probable and simplest BICEP2 inflationary scenarios. Although this list contains the simplest monomial potentials, it also includes many other scenarios, suggesting that focusing model building efforts on large field models only is unjustified at this stage. We demonstrate that the sets of inflationary models preferred by Planck alone and BICEP2 alone are almost disjoint, indicating a clear tension between the two data sets. We address this tension with a Bayesian measure of compatibility between BICEP2 and Planck. We find that for models favored by Planck the two data sets tend to be incompatible, whereas there is moderate evidence of compatibility for the BICEP2 preferred models. As a result, it would be premature to draw any conclusion on the best Planck models, such as Starobinsky and/or K"ahler moduli inflation. For the subset of scenarios not exhibiting data sets incompatibility, we update the evidences and complexities using both data sets together.

The BICEP2 Challenge

In early 2014, the BICEP2 experiment announced a groundbreaking detection of "B-mode" polarization in the Cosmic Microwave Background (CMB). This signal was initially interpreted as a "smoking gun" for primordial gravitational waves, suggesting a high tensor-to-scalar ratio ($r \approx 0.2$). However, this result stood in direct contrast to earlier data from the Planck satellite, which had set an upper limit of $r < 0.11$ based on temperature fluctuations.

This paper provides a rigorous Bayesian autopsy of this "tension," comparing the datasets across the Encyclopædia Inflationaris—a library of 193 distinct inflationary scenarios—to determine if any single model can reconcile the two experiments.

Selective Evidence and the B-Mode Spectrum

A critical finding of this study is that the hypothesis of a primordial origin for the BICEP2 signal is not uniformly supported across all its data points. The authors argue that only the first four "bandpowers" of the BICEP2 spectrum are physically plausible. Higher multipoles would require a tensor-to-scalar ratio so extreme that it would contradict almost all other cosmological observations.

The BICEP2 B-mode angular power spectrum. The authors' analysis focuses on the first four bandpowers, as higher multipoles suggest physically implausible values for the tensor-to-scalar ratio.

Quantifying the Tension

To measure the compatibility between Planck and BICEP2, the researchers employed a sophisticated Bayesian framework, calculating the "evidence" (predictive power) and "complexity" (parameter efficiency) for each model. They utilized the $R$-factor, a measure where $R > 1$ indicates compatibility and $R < 1$ indicates tension.

The results reveal a deep divide: BICEP2 strongly favors "large-field" models (such as simple monomial potentials $V \propto \phi^p$), while Planck prefers "plateau" models like Starobinsky inflation. The sets of preferred models for each experiment are almost entirely disjoint.

1D posterior distributions for cosmological and inflationary parameters. The red dashed lines (Planck) and black solid lines (BICEP2) show significant displacement, particularly in the Hubble flow functions (ε) that dictate the shape of primordial spectra.

As shown in the parameter distributions, the tension is most visible in the first Hubble flow function, $\epsilon_1$ (related to $r$), and $\epsilon_2$ (related to the spectral index $n_s$). For models that Planck favors, the BICEP2 data is almost certainly incompatible. Conversely, for models BICEP2 prefers, Planck is only "moderately" compatible.

Key Results and the "Dust" Factor

If the BICEP2 signal were truly primordial, it would fix the energy scale of inflation at approximately $2 \times 10^{16}$ GeV—the Grand Unified Theory (GUT) scale. Under BICEP2 alone, the most probable model is LFI3 ($V(\phi) \propto \phi^3$). However, when attempting to combine the datasets, the Bayesian measure of compatibility for general slow-roll models remains inconclusive ($\ln(R_{SR}) \approx -0.01$), meaning individual models often fail to bridge the gap.

Ultimately, the authors acknowledge the "elephant in the room": galactic foreground dust. If the BICEP2 signal was caused by polarized dust rather than the Big Bang, the tension vanishes. While subsequent joint analyses confirmed dust as the primary source, this paper remains a landmark study in how to statistically navigate conflicting datasets in precision cosmology.

Summary updated

Abstract

We investigate the implications for inflation of the detection of $B$-modes polarization in the cosmic microwave background by BICEP2. We show that the hypothesis of the primordial origin of the measurement is favored only by the first four band powers, while the others would prefer unreasonably large values of the tensor-to-scalar ratio. Using only those four band powers, we carry out a complete analysis in the cosmological and inflationary slow-roll parameter space using the BICEP2 polarization measurements alone and extract the Bayesian evidences and complexities for all the Encyclop\ae{}dia Inflationaris models. This allows us to determine the most probable and simplest BICEP2 inflationary scenarios. Although this list contains the simplest monomial potentials, it also includes many other scenarios, suggesting that focusing model building efforts on large field models only is unjustified at this stage. We demonstrate that the sets of inflationary models preferred by Planck alone and BICEP2 alone are almost disjoint, indicating a clear tension between the two data sets. We address this tension with a Bayesian measure of compatibility between BICEP2 and Planck. We find that for models favored by Planck the two data sets tend to be incompatible, whereas there is moderate evidence of compatibility for the BICEP2 preferred models. As a result, it would be premature to draw any conclusion on the best Planck models, such as Starobinsky and/or K\"ahler moduli inflation. For the subset of scenarios not exhibiting data sets incompatibility, we update the evidences and complexities using both data sets together.
BibTeX
@inproceedings{martin2014compatibility,
  author = {Jérôme Martin and Christophe Ringeval and Roberto Trotta and Vincent Vennin},
  title = {{Compatibility of Planck and BICEP2 results in light of inflation}},
  booktitle = {Physical},
  year = {2014},
  doi = {10.1103/physrevd.90.063501},
}
Figure