Lagrangian Bias as a Gaussian Random Field

1. Problem Statement

In modern cosmology, halo bias (the relationship between the clustering of dark matter halos and the underlying matter field) is traditionally treated as a set of perturbative coefficients generated by the gravitational collapse process. Standard frameworks (Peak-Background Split, Effective Field Theory, Separate Universe) view bias as a property of tracers conditional on their formation history.

The paper identifies a conceptual gap: these approaches treat bias as an emergent property of collapse, whereas the fundamental statistical structure of the initial conditions (a Gaussian Random Field, or GRF) may inherently possess a scale-independent bias field before any collapse occurs. The author seeks to demonstrate that bias is not generated by collapse but is a pre-existing field that collapse merely selects from.

2. Methodology

The author employs a rigorous statistical analysis of Gaussian Random Fields (GRF) in Lagrangian space, utilizing both analytical derivations and numerical simulations (Quijote simulations).

• Analytical Derivation:

  • The density field $\delta(\mathbf{q})$ is decomposed into independent Fourier shells $\delta_k(\mathbf{q})$.
  • Using the properties of multivariate Gaussians, the author calculates the conditional mean of a specific Fourier shell $\delta_k$ given the local density $\delta(\mathbf{q})$ at a point $\mathbf{q}$.
  • The derivation shows that the conditional mean shift is proportional to the variance of that shell, $\sigma_k^2$, leading to a scale-independent ratio.
  • The author defines a bias field $b(\mathbf{q})$ as the ratio of the local density to the total variance: $b(\mathbf{q}) = \delta(\mathbf{q})/\sigma^2$.
  • The framework is extended to linear operators (e.g., $\nabla^2\delta$, tidal tensors $s_{ij}$), generating "secondary bias fields."

• Numerical Verification:

  • The theory is tested using initial conditions from Quijote simulations at $z=127$.
  • The author measures the cross-correlation between local density shells and the full field to verify the scale-independence of the bias.
  • Simulations track particles from initial Lagrangian patches to Eulerian positions ($z=0$) to compare predicted bias against measured halo bias.

3. Key Contributions

A. Bias as a Fundamental Field

The paper establishes that scale-independent Lagrangian bias is a fundamental property of every point in a Gaussian density field, not just a property of collapsed objects.

• Result: For any point $\mathbf{q}$, the mean response of large-scale modes to the local density is scale-independent:
  $$\langle \delta_k(\mathbf{q}) | \delta(\mathbf{q}) \rangle = \frac{\sigma_k^2}{\sigma^2} \delta(\mathbf{q})$$
  This implies a bias field $b(\mathbf{q}) = \delta(\mathbf{q})/\sigma^2$ exists everywhere in the initial conditions.

B. Geometric Selection vs. Collapse Generation

The paper reframes halo bias:

• Old View: Collapse generates bias coefficients.
• New View: Collapse is a geometric selection of patches within a pre-existing bias field.
• Halos are simply regions where the collapse criterion is met; their large-scale clustering is the average of the bias field values of the constituent particles in that Lagrangian patch.

C. Secondary Bias Fields and "Active" Operators

The framework introduces secondary bias fields generated by linear operators acting on the density (e.g., $b_{\nabla^2\delta} \propto \nabla^2\delta$).

• Redundant vs. Active:
  • If the collapse criterion depends only on density ($\delta_s$), secondary fields are "redundant" (their clustering is fully determined by the primary bias).
  • If the collapse criterion depends on multiple operators (e.g., density and curvature $\nabla^2\delta_s$), these operators become "active."
  • Active operators distort the covariance structure of the selected patches relative to the full field, generating independent clustering signals (scale-dependent bias).

D. Unified Explanation of Assembly Bias

The paper provides a unified origin for three distinct phenomena:

1. Gaussian width of bias: The distribution of bias at fixed mass is Gaussian due to the underlying GRF structure.
2. Higher-derivative bias: Arises from active operators (like $\nabla^2\delta$) in the collapse criterion.
3. Assembly Bias: Occurs when the collapse selection correlates with secondary operators (e.g., selecting high-curvature regions at fixed density), inducing a split in the primary bias distribution.

4. Key Results

• Scale Independence: Numerical measurements confirm that the conditional mean of Fourier shells, normalized by their variance, is identical across all $k$-shells, collapsing to the theoretical prediction $\delta(\mathbf{q})/\sigma^2$.
• Halo Bias Prediction: The framework predicts the measured halo bias $b(M)$ at $z=0$ without free parameters. The prediction is simply the mean Lagrangian bias of the particles in the halo's initial patch, evolved by the linear growth factor $D(a)$:
  $$b_{Eulerian} = \frac{\langle b_{Lagrangian} \rangle}{D(a)} + 1$$
• Assembly Bias Demonstration: By splitting halos of fixed mass based on their Lagrangian curvature ($\nabla^2\delta_s$), the author observes a significant split in their large-scale bias. This confirms that $\nabla^2\delta_s$ is an "active" operator in the collapse criterion, modifying the covariance structure and generating assembly bias.
• Non-Halo Patches: The results hold for arbitrary geometric selections (not just halos), confirming the universality of the bias field framework.

5. Significance

• Theoretical Unification: The paper unifies disparate concepts (Peak-Background Split, EFT bias expansions, and Assembly Bias) under a single probabilistic framework rooted in the statistics of Gaussian fields.
• Parameter-Free Prediction: It shifts the paradigm from fitting bias coefficients to calculating them ab initio from initial conditions. The bias is determined entirely by the linear power spectrum and the geometric selection criteria.
• New Language for Non-Linear Evolution: The bias field is a Lagrangian quantity conserved along fluid trajectories. This offers a new way to track how information from initial conditions is redistributed and mixed during non-linear gravitational evolution.
• Future Applications: The framework provides a natural path for modeling cosmologies with massive neutrinos (scale-dependent bias) and primordial non-Gaussianity (where the collapse volume rotates in the multi-dimensional operator space). It suggests a future where forward models of halo clustering can be built directly from initial conditions without perturbative bias expansions.

In summary, Banerjee demonstrates that bias is not a consequence of collapse, but a pre-existing statistical property of the initial density field, which collapse simply samples. This insight fundamentally alters how cosmologists should model the connection between the early universe and the large-scale structure observed today.