AI summary by TwinMind
How good is the evidence for Dark Energy?
Sabine Hossenfelder
45m 15s
Background and Context
- Interview with Sue Berserker, theoretical physics professor at University of Oxford
- Discussion of his 2016 paper challenging supernova data evidence for dark energy
- 2011 Nobel Prize awarded for dark energy discovery based on supernova data
- Original supernova data made public only in 2014 by Joint Likelihood Analysis collaboration
Statistical Analysis Problems with Original Supernova Data
- Original analysis used unprincipled statistical methods
- Error bars adjusted until good fit achieved with assumed standard cosmological model
- Appropriate for parameter estimation but not model selection
- Berserker's team used maximum likelihood estimator (industry standard)
- Work done by master student Yep Nissen and statistical expert Albert Ponty
- Found evidence for acceleration was marginal - less than 3 standard deviations
- Far short of 5 sigma required for fundamental discovery
- First detailed independent analysis of the data since 2014 release
Radio Source Dipole Analysis
- Studied anisotropy in radio sources from Very Large Array, New Mexico
- Sources at high redshift (z > 1) used to test isotropy assumption
- Cross-correlated with 2MASS survey to remove nearby contamination
- Found velocity 4x larger than CMB dipole but in same direction
- Confirmed earlier result by radio astronomer Single
- Catalog contained ~800,000 galaxies after cleaning
- Statistical significance less than 3 sigma
- Square Kilometre Array will provide much larger catalog for definitive test
Implications of Radio Source Results
- CMB dipole may not be entirely kinematic in origin
- Challenges assumption of convergence to cosmological rest frame at 100 megaparsecs
- Affects standard analysis assuming special relativistic boost to isotropic frame
- Questions fundamental assumptions of Friedmann-Robertson-Walker cosmology
Supernova Dipole Discovery
- Tomography of local Hubble flow using Union 2 catalog
- Found dipole in supernova distribution extending beyond 100 megaparsecs
- Flow extends to Shapley supercluster at ~260 megaparsecs
- Confirmed by Nearby Supernova Factory (led by Saul Perlmutter)
- Challenges assumption of homogeneity at large scales
Reanalysis Without Isotropy Assumption
- Undid corrections for peculiar velocities in original analysis
- Found covariance matrix had unphysical negative off-diagonal terms
- Analyzed data in heliocentric frame without isotropy assumption
- Discovered acceleration is almost entirely dipolar
- Universe accelerating in one direction, decelerating in opposite
- Dipole direction within 23 degrees of CMB dipole
- Monopole (isotropic) acceleration consistent with zero at 1.4 sigma
Cosmological Constant and String Theory Implications
- Isotropic acceleration required for cosmological constant interpretation
- Evidence for isotropic acceleration is non-existent
- Impacts string theory swampland conjecture discussions
- String theory community assumed positive cosmological constant from supernova data
- Berserker suggests string theorists could "uplift to zero" instead
Criticism of Concordance Cosmology
- Cultural difference: particle physics seeks anomalies, cosmology defends standard model
- Most confirmations assume standard model being tested
- Baryon Acoustic Oscillations (BAO) example:
- Need several million spectroscopic redshifts for 5 sigma detection
- First claim used only ~40,000 luminous red galaxies
- Uses template fitting rather than blind peak detection
- Only 10% chance of seeing peak in any given sample
- Same data consistent with non-accelerating models (Blanchard et al.)
CMB and Dark Energy Connection
- CMB insensitive to dark energy (decoupled when universe 1000x smaller)
- Dark energy inference relies on cosmic sum rule (matter + curvature + Λ = 1)
- Sum rule only valid for simplified three-component model
- Concordance may be manufactured through confirmation bias
- Rupert Croft analysis: only 2/28 lambda measurements outside 1-sigma of WMAP (expected ~9)
Future Tests and Data Quality Issues
- Dark Energy Spectroscopic Instrument will measure millions of redshifts
- Late Integrated Sachs-Wolfe effect requires 10 million redshifts for 5 sigma detection
- Current claims based on <3 sigma detections
- Pantheon catalog shows discrepancies with JLA catalog
- Redshift differences exceed quoted uncertainties
- Systematic errors in data presentation (identical columns for different frames)
- Limited availability of detailed data for independent analysis
Hubble Tension Discussion
- Local Hubble constant measurements show 7% variation across sky
- Systematic uncertainties not properly accounted for
- Hubble parameter varies depending on supernova catalog used
- Tension may be premature claim given systematic uncertainties
- Specialized analysis creates "black box" problem for community assessment
