A Bayesian estimator for peculiar velocity correction in cosmological inference from supernovae data

This paper presents a novel Bayesian estimator that simultaneously corrects for peculiar velocity effects and fits cosmological models to supernova data by treating the magnitude-redshift relation as a non-linear errors-in-variables problem, thereby relaxing traditional assumptions of linearity and Gaussianity without requiring independent velocity measurements.

The Gist

Imagine you are trying to map the entire universe. You have a giant, cosmic ruler, and you are measuring how fast galaxies are moving away from us. This is how cosmologists figure out the age, size, and expansion rate of the universe. The most famous "ruler" they use is a specific type of exploding star called a Type Ia Supernova.

However, there's a problem. These stars aren't just floating in a calm, empty void. They are riding on "waves" of gravity created by massive clusters of galaxies. Just like a surfer on a wave, a galaxy has its own local motion (called peculiar velocity) on top of the general expansion of the universe.

The Problem: The "Wobbly" Ruler

Think of the universe's expansion like a giant balloon being blown up. Every point on the balloon moves away from every other point. But if you draw a dot on the balloon and then shake the balloon vigorously, that dot jiggles around.

When astronomers look at a supernova, they measure two things:

1. How bright it is (which tells us the distance).
2. How much its light is stretched (redshift, which tells us the speed).

The "jiggle" (peculiar velocity) messes up the speed measurement. If a galaxy is moving toward us because of a local gravitational tug, it looks like it's moving away slower than it actually is. If it's moving away, it looks faster.

The Old Way of Fixing It:
For a long time, scientists used two "band-aids" to fix this:

1. For the big waves (Coherent motion): They tried to map the entire ocean of galaxies to predict the waves. But to do this, they had to guess the shape of the balloon (the cosmology) first. It's like trying to predict the wind by assuming you already know the weather forecast. This creates a circular logic trap.
2. For the small jiggles (Random motion): They just added a "fudge factor" to their error bars. They assumed the jiggles were perfectly random and followed a nice, neat bell curve (Gaussian distribution). They also assumed the relationship between distance and speed was a straight line.

The Flaw:
The universe isn't always a straight line, and the jiggles aren't always perfectly neat. As our telescopes get better and our measurements become incredibly precise (the era of "Precision Cosmology"), these old band-aids start to fail. The tiny errors in the "jiggles" can now hide big secrets about Dark Energy or the Hubble Constant.

The New Solution: The "Smart Navigator"

This paper introduces a new, smarter way to handle these errors. Instead of using band-aids, they built a Bayesian Estimator.

Here is the analogy:
Imagine you are trying to find a hidden treasure on a map.

• The Old Method: You look at the map, see a landmark, and say, "The treasure is exactly 5 miles North." If the landmark is slightly blurry (due to the jiggles), you just say, "Well, it might be 5.1 or 4.9 miles." You treat the landmark's position as fixed and just widen your search circle.
• The New Method (This Paper): You say, "I don't know exactly where that landmark is. It could be anywhere within a small range." So, you treat the location of the landmark itself as a mystery variable. You run a simulation where you move the landmark around in its possible range, checking which position makes the most sense with the rest of the map.

What makes this special?

1. No Guessing the Weather: It doesn't need to assume a specific model of the universe to correct the motion. It figures out the motion while it figures out the universe's shape.
2. No Straight Lines: It doesn't force the relationship between distance and speed to be a straight line. It handles the curves and twists of the real universe.
3. No Perfect Circles: It doesn't assume the "jiggles" are perfectly random. It can handle weird, messy distributions.

The Results: Why Should We Care?

The authors tested this new method in two ways:

1. Fake Data: They created a computer simulation of the universe with known answers.
   • When the data was "noisy" (like our current telescopes), the old methods and the new method gave similar results.
   • BUT, when they simulated "super-precise" data (like what we will get from future telescopes like the LSST), the old methods started to drift away from the truth. The new method stayed perfectly on target.
2. Real Data: They applied it to the famous Pantheon sample of supernovae.
   • The results were very similar to the old methods (which is good; it means the old methods were "good enough" for now).
   • However, the new method found a tiny, subtle shift that the old methods missed. This suggests there might be a tiny bit of "leftover" motion or weirdness in the data that we haven't accounted for yet.

The Bottom Line

This paper is like upgrading from a compass to a GPS.

• The compass (old methods) works great when you are walking in a straight line on a calm day.
• The GPS (this new Bayesian method) knows that the ground might be shifting, the wind might be blowing, and the path might curve. It recalculates your position in real-time, considering all the uncertainties simultaneously.

As we move into an era where we can measure the universe with sub-percent precision, we need this GPS. If we don't correct for the "jiggles" of galaxies properly, we might think Dark Energy is changing or that the universe is expanding at the wrong rate, simply because we didn't account for the local traffic jams of gravity. This new tool ensures that when we look at the future of the universe, we aren't just looking at a reflection of our own measurement errors.

Technical Summary

1. Problem Statement

The estimation of cosmological parameters from Type Ia Supernovae (SNIa) data is complicated by the peculiar motion of host galaxies. This motion introduces a bias in the observed redshift ($z_{obs}$), which deviates from the true cosmological redshift ($z_{cos}$).

• Current Limitations:
  • Coherent Component: Large-scale bulk flows are typically corrected using velocity field reconstruction based on galaxy density distributions. However, this method requires assuming an underlying cosmology, potentially introducing circular bias into the final inference.
  • Random Component: Small-scale random motions are usually treated statistically by inflating magnitude uncertainties. This relies on two restrictive assumptions:
    1. A locally linear approximation of the magnitude-redshift relation (which fails at very low redshifts).
    2. A Gaussian distribution for peculiar velocities (which may have extended tails in the non-linear regime).
  • General Statistical Issue: Standard cosmological fitting often treats independent variables (redshift) as perfectly measured, ignoring errors-in-variables (EIV) in non-linear models.

2. Methodology

The authors propose a Bayesian estimator that simultaneously fits cosmological models and corrects for peculiar velocities without assuming linearity or Gaussianity.

A. General Framework: Non-linear Errors-in-Variables (EIV)

The authors develop a general method for fitting arbitrary non-linear models where both dependent ($y$) and independent ($x$) variables have errors.

• Formulation: Instead of propagating $x$-errors into $y$-errors, they treat the true values of the independent variable ($x^*$) as latent parameters.
• Posterior Distribution: The posterior probability for model parameters $\theta$ is derived by marginalizing over the unknown true values $x^*$:
  $$P(\theta | D) \propto \int \prod_i P_Y(y_i - f(\theta, x^*_i)) P_X(x_i - x^*_i) P(\theta) P_{\chi^2}(\chi^2) \, d^N x^*$$
  • $P_Y$: Likelihood of observed magnitude given the model.
  • $P_X$: Prior distribution of the true redshift given the observed redshift (modeling the peculiar velocity).
  • Regularization: A $\chi^2$ regularization term is added to prevent overfitting, ensuring the mean $\chi^2$ aligns with the degrees of freedom.
• Implementation: Due to the high dimensionality (one $x^*$ parameter per data point), the integration is performed using Markov Chain Monte Carlo (MCMC) (Metropolis-Hastings algorithm).

B. Application to Cosmology

The method is specialized for the magnitude-redshift relation ($m$ vs. $z$):

• Variables: $x = z$ (redshift), $y = m$ (apparent magnitude).
• Model: The luminosity distance relation for $\Lambda$CDM and $w$CDM models.
• Three Estimators Compared:
  1. E1 (Baseline): Ignores peculiar velocities; assumes $z_{obs} = z_{cos}$.
  2. E2 (Standard): Uses the linear approximation and Gaussian assumption. Propagates redshift error to magnitude uncertainty (Eq. 21).
  3. E3 (Proposed): The exact Bayesian EIV estimator. Treats each observed redshift as a parameter drawn from a uniform prior within a range defined by the peculiar velocity uncertainty. It handles non-linearity and non-Gaussianity explicitly.

3. Key Contributions

1. General EIV Method: Development of a robust Bayesian framework for fitting non-linear models with errors in both variables, applicable beyond cosmology.
2. Self-Consistent Coherent Correction: Unlike velocity field reconstruction, the proposed method corrects for coherent bulk flows self-consistently during the model fitting process by treating redshifts as free parameters. It does not require an external cosmological assumption for the correction.
3. Relaxation of Assumptions: The method removes the need for the locally linear approximation and Gaussian velocity distribution assumptions, which are critical for low-redshift data and future high-precision surveys.
4. Unified Treatment: It simultaneously accounts for both random and coherent peculiar velocity components within a single statistical framework.

4. Results

The authors tested the estimators on synthetic data (mimicking Pantheon and future LSST/ZTF precision) and the real Pantheon SNIa sample.

Synthetic Data Analysis

• Current Precision ($\sigma_m = 0.2$): The impact of peculiar velocities is small due to large magnitude errors. E1, E2, and E3 yield statistically consistent results, though E2 and E3 show slightly larger uncertainties.
• Future Precision ($\sigma_m = 0.05$):
  • Bias in E1: Neglecting peculiar velocities (E1) leads to significant bias in cosmological parameters ($\Omega_M$, $H_0$, $w$).
  • Limitations of E2: While E2 reduces bias compared to E1 by down-weighting low-redshift data (due to propagated errors), it does not fully recover the true parameters.
  • Success of E3: The proposed estimator (E3) recovers the true input cosmological parameters within $1\sigma$, even in the presence of coherent bulk flows (simulated as a constant shift of 300 km/s).
• Coherent vs. Random:
  • For purely random (Gaussian) motions, E2 and E3 yield similar results.
  • For coherent motions, E2 fails to correct the bias (it only down-weights data), whereas E3 correctly shifts the redshift parameters toward their true values.

Pantheon Data Application

• Applied to the Pantheon sample (1048 SNe).
• All three estimators (E1, E2, E3) produced statistically consistent results for $\Lambda$CDM and $w$CDM parameters.
• Significance: A small shift was observed between E2 and E3 in the $w$ parameter. The authors suggest this could indicate residual coherent motion or non-Gaussianity in the Pantheon data, which E2 misses but E3 captures. This highlights the potential for E3 to detect subtle systematics in current data and future high-precision datasets.

5. Significance and Future Outlook

• Precision Cosmology: As upcoming surveys (LSST, ZTF, DESI) reduce magnitude uncertainties, the systematic bias from peculiar velocities will become the dominant source of error. The standard linear/Gaussian approximations (E2) may become insufficient, especially for very low-redshift supernovae where the magnitude-redshift relation is non-linear.
• Model Independence: The method eliminates the circularity of using a fiducial cosmology to correct velocities before inferring that same cosmology.
• Broader Applicability: The general EIV Bayesian framework developed here is applicable to various other problems in astronomy and cosmology where independent variables have significant measurement errors.
• Future Work: The authors plan to apply this to combined datasets (Pantheon+, DESY5, Union3) and explore direct detection of peculiar velocity fields from SNIa data to constrain structure growth and gravity models.

In conclusion, this paper presents a rigorous, non-approximate Bayesian framework that significantly improves the robustness of cosmological parameter inference from supernovae, particularly as data precision increases and the limitations of current linear approximations become apparent.