Pipeline 2 · Galaxy Chirality Catalog · DESI Legacy Survey DR8

Galaxy Chirality Explorer

Browse the largest galaxy chirality catalog ever produced: 8,474,531 galaxies classified by spiral chirality using a ViT-Small vision transformer with test-time equivariant averaging. Dipole significance: 0.43σ (null — no parity violation detected). CW/(CW+CCW) = 0.4974, all sky regions within 0.5% of 50/50. Search by sky position, filter by classification and confidence, and explore the all-sky distribution.

8,474,531
Total Galaxies
DESI LS DR8
1,592,107
Clockwise (CW)
18.8%
1,609,053
Counter-CW (CCW)
19.0%
5,273,371
Not Spiral
62.2%
0.43σ
Dipole Significance
null — no parity violation
0.4974
CW / (CW+CCW)
equivariant (essentially 50/50)

Key Science Findings

Comprehensive analysis of 8,474,531 galaxies — the largest galaxy chirality catalog ever produced. 100% coordinate match with GZ DESI Zenodo catalog. 949,584 high-confidence spirals (confidence > 0.6).

Global Dipole: Null

Simple dipole: 0.43σ. Power-spectrum dipole: 2.75σ (marginal). No significant large-scale parity violation detected.

Hemisphere Asymmetry: 3.05σ

p = 0.002. North CW = 49.81%, South CW = 49.64%, difference 0.17%. Does not survive look-elsewhere correction.

l=5 Multipole: 2.46σ

Marginal signal at the fifth multipole. Does not survive look-elsewhere correction across all tested multipoles.

Shamir 3% Claim: REFUTED

Our maximum regional asymmetry is 0.47% — 7× smaller than Shamir's claimed 3%. The claimed signal is not reproducible with proper bias control.

CMB Dipole Alignment: Null

85.4° separation between chirality dipole and CMB dipole — essentially perpendicular. No alignment detected.

Scale Dependence: Null

No chirality signal at any angular scale tested. The universe shows no preferred handedness at any resolution.

Bottom line: No signal survives look-elsewhere correction. The universe appears to have no preferred handedness at the scales probed by this catalog.

Visual Analysis

Figures from the chirality catalog analysis of 8,474,531 galaxies. Click any figure to view full resolution.

Note: Figures with the "fig_" prefix that show a broken image icon are currently being generated and will appear once processing completes.
Mollweide asymmetry map
Fig 1. Mollweide Asymmetry Map
Equivariant CW/(CW+CCW) fraction across the full sky in Mollweide projection. No large-scale pattern is visible. Global dipole significance: 0.43σ.
Finding: Global Dipole Null
Raw vs equivariant Mollweide comparison
Fig 2. Raw vs Equivariant Comparison
Before/after equivariant averaging. Raw classifier shows systematic CW bias from CNN mirror asymmetry; equivariant averaging eliminates it entirely. This is why prior claims of 3% parity violation are artifacts.
Finding: Shamir 3% Claim Refuted
Galaxy density map
Fig 3. Spiral Galaxy Density Map
Spatial distribution of classified spiral galaxies across the DESI Legacy Survey DR8 footprint. Higher density regions correspond to deeper survey coverage. The uneven footprint is accounted for in all statistical tests.
Context: Survey geometry
RA/DEC CW fraction heatmap
Fig 4. CW Fraction Heatmap (RA/DEC)
Binned CW/(CW+CCW) fraction as a function of right ascension and declination. All bins cluster tightly around 0.50, confirming no preferred handedness at any sky position.
Finding: Scale Dependence Null
Angular power spectrum multipole analysis
Fig 5. Angular Power Spectrum (Multipoles)
Power in the CW asymmetry field decomposed by angular multipole l. The dipole (l=1) is at 0.43σ; l=5 shows a marginal 2.46σ excess that does not survive look-elsewhere correction. No significant power at any scale.
Finding: l=5 Multipole 2.46σ (does not survive LEE)
Raw vs equivariant CW fraction by sky region
Fig 6. CW Fraction by Sky Region
Raw versus equivariant CW/(CW+CCW) fraction across RA quadrants and DEC bands. Raw classifications show ~1% systematic bias that vanishes after equivariant averaging. All equivariant regions fall within 0.62% of 50/50.
Finding: All regions balanced after equivariant correction
North/South hemisphere CW comparison
Fig 7. Hemisphere Asymmetry (N/S)
North CW = 49.81%, South CW = 49.64% — a 0.17 percentage point difference at 3.05σ (p = 0.002). This is the strongest individual signal but does not survive look-elsewhere correction across all tested multipoles and scales.
Finding: Hemisphere Asymmetry 3.05σ (does not survive LEE)
Classification confidence distribution
Fig 8. Confidence Distribution
Distribution of classifier confidence across all 8.47M galaxies, broken down by class. The model is highly confident on NOT_SPIRAL classifications. Among spirals, 949,584 exceed the 0.6 confidence threshold.
Context: Classifier performance
CW/CCW/NOT_SPIRAL class breakdown
Fig 9. Classification Breakdown
CW (18.8%), CCW (19.0%), NOT_SPIRAL (62.2%). The near-equal CW/CCW split is the key result: the equivariant classifier finds no global handedness preference among the 3.2M spiral galaxies.
Finding: CW/(CW+CCW) = 0.4974
Equivariant averaging demonstration
Fig 10. Equivariant Averaging Demonstration
Illustration of the test-time equivariant averaging procedure. Each galaxy image is classified in its original orientation and its mirror reflection; the final prediction averages both, guaranteeing exact CW/CCW symmetry at the classifier level.
Context: Methodology — bias elimination

Example Galaxy Galleries

Representative galaxies from each classification category.


Filter Galaxies

Results

Showing 1–20 of 8,474,531 galaxies
Page 1 of 423,727
Image dr8_id Class P(CW) P(CCW) P(NS) Confidence RA DEC
Page 1 of 423,727

All-Sky Chirality Map

Mollweide projection of spiral galaxy classifications across 8.47M galaxies. The distribution is essentially uniform. Global dipole is null (0.43σ). A marginal hemisphere asymmetry (3.05σ, north CW=49.81% vs south CW=49.64%) does not survive look-elsewhere correction. CMB dipole alignment is null (85.4° separation).

Mollweide all-sky asymmetry map of 3.2M spiral galaxies
Equivariant CW Asymmetry — Mollweide Projection
All-sky map of CW/(CW+CCW) fraction from 3,201,160 equivariant spiral classifications. Blue indicates slight CW excess, red indicates slight CCW excess. The distribution is essentially uniform with no significant large-scale pattern.
Raw vs equivariant Mollweide comparison
Raw vs Equivariant Classification — Before/After Bias Removal
Side-by-side comparison showing raw classifier output (left) versus equivariant test-time averaging (right). The raw classifier shows systematic CW bias from CNN mirror asymmetry; equivariant averaging eliminates this, producing balanced 49.74% CW. This is the key methodological advance that distinguishes this catalog from prior work.
Key figure — demonstrates why prior claims of parity violation (e.g., Shamir 3%) are artifacts of classifier bias.

Classification Pipeline

Details of the inference pipeline that produced this catalog.

Model: ViT-Small + 3-class head
Training data: GZ1 + CE-ResNet + synthetic (26.6K images)
Bias tests: 8/8 PASS
Equivariant CW/(CW+CCW): 0.4974
Test accuracy: 93.7%
Source images: DESI Legacy Survey DR8/DR9
Status: COMPLETE — Catalog C production run
Dipole: 0.43σ (null — no significant parity violation)
Runtime: 4.1 hours on H200 (32× DataLoader speedup)
High-confidence spirals: 949,584 (confidence > 0.6)
Cutout size: 150 × 150 px
Coordinate match: 100% vs GZ DESI Zenodo catalog

Deeper Analysis

Full multipole decomposition, sky region balance, and confidence stratification of the equivariant Catalog C (3,201,160 spirals).

Multipole Analysis

TestSignificanceSurvives LEE?
Global dipole (simple)0.43σN/A (null)
Global dipole (power spectrum)2.75σNo
Hemisphere asymmetry (N/S)3.05σNo
l=5 multipole2.46σNo
CMB dipole alignmentNull (85.4°)N/A (null)
Scale dependence (all scales)NullN/A (null)
Shamir 3% claimREFUTEDMax 0.47% (7× smaller)

Sky Region Balance (Equivariant)

RegionCW/(CW+CCW)Deviation from 0.5
RA 0°–90°0.4950−0.50%
RA 90°–180°0.4952−0.48%
RA 180°–270°0.4966−0.34%
RA 270°–360°0.4986−0.14%
DEC South (< −30°)0.4938−0.62%
DEC Equatorial (−30° to +30°)0.4968−0.32%
DEC North (> +30°)0.4960−0.40%

All regions within 0.62% of 50/50. Global equivariant CW/(CW+CCW) = 0.4974.

Hemisphere Asymmetry Detail

The strongest individual signal is a north-south hemisphere asymmetry at 3.05σ (p = 0.002): northern hemisphere CW fraction = 49.81%, southern hemisphere CW fraction = 49.64%, a difference of 0.17 percentage points. This is a marginal result — it does not survive look-elsewhere correction across the multiple tests performed (dipole, hemisphere, multipoles l=1–10, CMB alignment, scale dependence). The physical interpretation is ambiguous: it could reflect a subtle residual systematic in survey coverage, or a genuine but weak statistical fluctuation. It does not constitute evidence for large-scale parity violation.

Publications & Data Access

HuggingFace Model: bamfai/galaxy-chirality-v2
HuggingFace Dataset: bamfai/galaxy-chirality-catalog
Convex Database: 8,474,531 rows live
Source catalog: GZ DESI DR8 (Zenodo, 100% coordinate match)
Catalog paper: “No Evidence for Large-Scale Parity Violation in Galaxy Morphology: A Survey-Scale Chirality Catalog of 8.47 Million Galaxies” — Draft complete (1,096 lines, 11 figures, 4 tables). Read PDF · LaTeX source · Paper 4 on Papers page