Abstract A formalism is presented that allows cosmological experiments to be tested for consistency, and allows a simple frequentist interpretation of the resulting significance levels. As an example of an application, this formalism is used to place constraints on bulk flows of galaxies using the results of the microwave background anisotropy experiments COBE and SP91, and a few simplifying approximations about the experimental window functions. It is found that if taken at face value, with the quoted errors, the recent detection by Lauer and Postman of a bulk flow of 689 km/s on scales of 150$h^{-1}$Mpc is inconsistent with SP91 at a 95% confidence level within the framework of a Cold Dark Matter (CDM) model. The same consistency test is also used to place constraints that are completely model-independent, in the sense that they hold for any power spectrum whatsoever --- the only assumption being that the random fields are Gaussian. It is shown that the resulting infinite-dimensional optimization problem reduces to a set of coupled non-linear equations that can readily be solved numerically. Applying this technique to the above-mentioned example, we find that the Lauer and Postman result is inconsistent with SP91 even if no assumptions whatsoever are made about the power spectrum.
Introduction
In the early 1990s, a significant tension emerged in cosmology between observations of large-scale structure and the Cosmic Microwave Background (CMB). Measurements of "bulk flows"—the collective motion of galaxies over vast distances—suggested a universe with much higher density fluctuations than those implied by the temperature anisotropies observed in the CMB.
Traditionally, testing the consistency of these disparate datasets required assuming a specific cosmological model, such as Standard Cold Dark Matter (sCDM). However, if a model failed the test, it remained unclear whether the underlying physics was flawed or if the assumed shape of the power spectrum $P(k)$ was merely inaccurate. This paper addresses this ambiguity by introducing a "power-spectrum independent" formalism. This approach tests consistency for any possible Gaussian random field, providing a fundamental check on observational data that transcends specific theoretical models.
The analysis focuses on three key datasets from the era:
- COBE DMR: Satellite-based measurements of large-scale CMB anisotropies.
- SP91: Ground-based CMB observations from the South Pole.
- Lauer & Postman (LP): A 1993 report of a high-velocity bulk flow (689 km/s) of Abell clusters on scales of 150 $h^{-1}$ Mpc.
Methodology
The proposed statistical framework consists of two primary components: the probability product and variational optimization.
The Probability Product ($\eta$)
To quantify consistency, the authors define a goodness-of-fit parameter $\eta$ as the product of individual probabilities ($P_i$) that each experiment would yield a result at least as extreme as the one observed. This frequentist approach is advantageous because its analytical distribution depends solely on the number of experiments ($N$), facilitating the straightforward calculation of significance levels.
Variational Optimization
To achieve model independence, the authors solve an optimization problem to find the "best-case" power spectrum $P(k)$ that maximizes consistency between conflicting datasets. They demonstrate that this infinite-dimensional problem reduces to a finite one where the optimal $P(k)$ is a sum of delta functions at specific scales. This allows for the determination of the absolute upper limit of consistency allowed by Gaussian linear theory, regardless of the specific cosmological model.
Results
The application of this formalism to the LP and CMB datasets reveals a stark discrepancy.
CDM and Universal Inconsistency
Within the standard CDM framework, the Lauer & Postman bulk flow result is found to be inconsistent with the SP91 CMB data at the 95% confidence level. Crucially, this tension persists even when the assumption of a specific power spectrum is removed. Using the model-independent test, the LP result remains statistically inconsistent with SP91.
The analysis highlights the importance of experimental window functions. Since experiments observe the power spectrum through different "windows" of scale, overlapping windows reporting wildly different results cannot be reconciled by simply adjusting the power spectrum.
Discussion and Limitations
The findings suggest that the reported error bars for one of the experiments may be underestimated, the window functions used were inaccurate, or the assumption of Gaussianity is incorrect.
The formalism relies on two key assumptions:
- Gaussianity: The density fields are assumed to be Gaussian random fields. Non-Gaussianity (e.g., from topological defects) would alter these constraints.
- Linear Theory: The window functions are based on linear perturbation theory, which is appropriate for large scales but may be complicated by non-linear effects on smaller scales.
Future work involving more precise, experiment-specific Monte Carlo simulations for window functions could provide a more definitive resolution to these observed tensions.
Summary updated