Peak Model of Structure Formation

• The Peak Model of Structure Formation is an analytic framework identifying local density maxima in the primordial field as the seeds for cosmic halo development.
• It incorporates tensorial features like the density Hessian and tidal shear to capture anisotropic collapse, linking simple peak counts with complex cosmic web dynamics.
• By merging excursion set theory with peak nesting and nonlocal bias, the model refines predictions for halo mass functions and clustering properties.

The peak model of structure formation is an analytic framework in which nonlinear cosmic structures are traced back to special points of the initial, smoothed density field, principally local maxima of a Gaussian random field, and are then evolved by collapse dynamics, tidal anisotropy, and multi-scale selection rules. In its modern form, the model is no longer limited to the statement that haloes form at peaks of the density field: it incorporates the density Hessian, the initial shear tensor, excursion-set first-crossing conditions, peak nesting, nonlocal bias, and, in more recent refinements, non-spherical protohalo boundaries and field-level deformation hierarchies (Paranjape et al., 2012, Rossi, 2014, Musso et al., 2023, Salvador-Solé et al., 2024).

1. Statistical setting and the basic peak hypothesis

The standard cosmological setting of the peak model is a homogeneous and isotropic FLRW background perturbed by a nearly Gaussian, nearly scale-invariant primordial fluctuation field. In this description, structure is seeded by the linear density contrast

$$\delta(\mathbf{x}) \equiv \frac{\rho(\mathbf{x})-\bar\rho}{\bar\rho},$$

or, equivalently, by the primordial curvature perturbation whose statistics are set during inflation and conserved on super-horizon scales for adiabatic modes (Malik, 2010). The smoothed density field is

$$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$

and its spectral moments are

$$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$

The dimensionless peak height is then

$\nu \equiv \frac{\delta}{\sigma_0}.$

In the BBKS formulation, protohaloes are statistically associated with local maxima of $\delta_R$. The peak conditions are $\nabla\delta_R=0$ together with a negative-definite Hessian. This is intrinsically a single-scale description: one counts peaks on a chosen smoothing scale, derives their abundance and clustering, and then identifies that scale with a halo mass (Paranjape et al., 2012, Musso et al., 2023). The core hypothesis is therefore Lagrangian and geometrical: haloes form preferentially around special points of the initial field rather than at generic spatial positions.

This hypothesis is supported, but not exhausted, by simulation tests. A direct numerical study found that as many as ~70% of well-resolved dark matter haloes form preferentially near peaks whose characteristic masses are similar to that of the halo, with more massive haloes showing a stronger tendency to reside near peaks initially (Ludlow et al., 2010). That result fixes the empirical status of the peak ansatz: it is statistically strong, but not exact in the naive one-peak/one-halo sense.

2. Peaks, Hessians, and tidal anisotropy

The simplest version of the peak model is density-based, but its physically relevant degrees of freedom are tensorial. Two tensors are central. The density Hessian is

$$H_{ij}(\mathbf{x}) \equiv \frac{\partial^2\delta(\mathbf{x})}{\partial x_i\partial x_j},$$

and the initial shear, or tidal tensor, is

$$T_{ij}(\mathbf{x}) \equiv \frac{\partial^2\phi(\mathbf{x})}{\partial x_i\partial x_j}.$$

Because $\nabla^2\phi \propto \delta$, the trace of $T_{ij}$ is proportional to the local overdensity, while its traceless part governs anisotropic collapse. The ordered eigenvalues $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$0 of the shear tensor determine collapse or expansion along principal axes; in Lagrangian dynamics they encode the sequence sheet $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$1 filament $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$2 node, or conversely void-like expansion (Rossi, 2014).

Rossi’s peak/dip picture generalizes the older Doroshkevich formalism by conditioning the shear statistics on the density Hessian at special points, i.e. on peaks and dips rather than random positions. The central object is a conditional probability density $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$3, where $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$4 measures the correlation between density curvature and tidal shear. In this formulation, halo and void formation are associated with local extrema of the density field, but their geometry is controlled by the correlated shear field at those extrema (Rossi, 2014).

This substantially changes the interpretation of the peak model. The relevant initial conditions are no longer just peak height and curvature. The conditional mean and scatter of the shear eigenvalues at peaks/dips differ from those at random locations, so the model acquires explicit predictions for ellipticity, prolateness, alignments, and anisotropic collapse histories. A plausible implication is that the peak model becomes, in practice, a peak-plus-tidal-environment model rather than a purely scalar peak-counting theory.

3. Excursion sets, cloud-in-cloud, and the emergence of excursion-set peaks

BBKS peak theory is single-scale, whereas halo formation is intrinsically multi-scale. Excursion set theory addresses this by following $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$5 or $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$6, with $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$7, as a random walk and identifying collapse with first crossing of a barrier $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$8. Its classical advantage is a treatment of the cloud-in-cloud problem, but it averages over random positions, not over the special subset of positions occupied by peaks (Paranjape et al., 2012).

Paranjape and Sheth unified these viewpoints in the excursion set peaks (ESP) formalism. In Gaussian smoothing, the Laplacian of the smoothed field is proportional to the derivative of the walk with respect to smoothing scale, so the BBKS curvature variable and the excursion-set up-crossing variable become directly linked. ESP therefore imposes the one-step up-crossing condition not on all walks, but on the restricted ensemble of peak trajectories (Paranjape et al., 2012).

This change has two major consequences. First, it provides a cloud-in-cloud treatment for peaks. Second, it changes the predicted mass function and bias in a distinctly nonlocal way. At large $$\delta_R(\mathbf{x})=\int d^3x'\,W_R(|\mathbf{x}-\mathbf{x}'|)\,\delta(\mathbf{x}'),$$9, $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$0, so ESP boosts the high-mass tail relative to correlated-step excursion sets at random positions. The same framework predicts that even the linear halo bias factor is $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$1-dependent as a consequence of the nonlocality associated with the peak constraint (Paranjape et al., 2012).

A stress test of this logic was provided in a warm-dark-matter-like spectrum with a small-scale cut-off. In that regime, standard excursion set theory with correlated steps completely fails to reproduce the mass function, while inclusion of the peaks constraint leads to the correct number of haloes but significantly underpredicts the masses of low-mass objects, with the predicted low-mass behavior

$$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$2

That asymptotic form was shown to be very robust within a single-collapse-barrier framework (Hahn et al., 2013). The lesson is that the peaks constraint is necessary, but the mapping from first crossing to final halo mass requires a more detailed treatment of collapse time and mass reassignment.

4. Nesting, first-principles bias, and tests of the excursion assumption

A central unresolved issue in peak-based models is peak nesting: smaller peaks can lie inside larger ones, so not every peak corresponds to an independent halo. The CUSP formalism addresses this by constructing a ConflUent System of Peak trajectories, deriving non-nested peak abundances, and fixing the mapping between Gaussian peaks and collapsed haloes from first principles, including ellipsoidal collapse of triaxial peaks in the Gaussian-smoothed density field (Salvador-Solé et al., 2024).

Within this framework, the Lagrangian local peak bias parameters adopt very simple analytic expressions similar to those found in the PS and ES models, but without free parameters, and the predicted Eulerian linear halo bias recovers the results of simulations. The paper further argues that the only small departure observed at intermediate and low masses can be due to the spurious halo splitting and grouping caused by the Spherical Overdensity halo-finding algorithm used in simulations (Salvador-Solé et al., 2024). In encyclopedic terms, this marks a shift from phenomenological peak barriers toward a self-consistent peak theory of both abundance and bias.

Recent particle-by-particle tests of excursion-set logic sharpen the status of its assumptions. Using $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$3-body simulations as a “perfect” collapse model, one study found that the core excursion set assumption—small-scale perturbations do not impact larger-scale collapse—is approximately fulfilled but exhibits small quantitative violations dependent on the smoothing filter. For a sharp $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$4-space cut-off $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$5 of mass elements revert collapse as the smoothing scale decreases, while only $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$6 do for a Gaussian and $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$7 for a top-hat (Wisłocka et al., 10 Mar 2025). The same work found that particles first accreted into haloes at the same smoothing scale may end up in haloes of significantly different masses, with a scatter of $$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$8-$$\sigma_j^2(R)\equiv \int \frac{d^3k}{(2\pi)^3}\,P(k)\,k^{2j}\,W^2(kR).$$9 dex, and that the proportionality constant in the deterministic mapping $\nu \equiv \frac{\delta}{\sigma_0}.$0 should be considered a degree of freedom (Wisłocka et al., 10 Mar 2025).

These results do not refute the peak model. They indicate, rather, that physically realistic implementations require explicit treatment of filter dependence, nesting, and stochastic mass mapping.

5. Numerical validation, peakless haloes, and protohalo geometry

Direct numerical tests reveal both the reach and the limits of the peak ansatz. In $\nu \equiv \frac{\delta}{\sigma_0}.$1CDM simulations, virtually all haloes with $\nu \equiv \frac{\delta}{\sigma_0}.$2 particles have at least one peak particle from some smoothing scale, and ≈71% of haloes in the $\nu \equiv \frac{\delta}{\sigma_0}.$3 box and ≈79% in the $\nu \equiv \frac{\delta}{\sigma_0}.$4 box have a main peak with $\nu \equiv \frac{\delta}{\sigma_0}.$5 very close to $\nu \equiv \frac{\delta}{\sigma_0}.$6 (Ludlow et al., 2010). At the high-mass end, ~91% of haloes with $\nu \equiv \frac{\delta}{\sigma_0}.$7 form near peaks with $\nu \equiv \frac{\delta}{\sigma_0}.$8 (Ludlow et al., 2010). The peak correspondence therefore strengthens with mass.

The same simulations also isolate a genuine minority population of “peakless haloes,” defined by $\nu \equiv \frac{\delta}{\sigma_0}.$9. These objects are not literally peak-free; rather, no peak of comparable mass is found in their vicinity. They differ systematically from the bulk protohalo population in two ways: their initial shapes are significantly flatter and more elongated than the predominantly triaxial majority, and they are, on average, more strongly compressed by tidal forces associated with their surrounding large scale structure (Ludlow et al., 2010). Peakless haloes also emerge from highly clustered regions of the initial density field, linking the limitations of the simplest peak model to assembly bias and environmental dependence (Ludlow et al., 2010).

The geometry of protohaloes is another major correction to the classical spherical picture. Analytical work has long represented protohaloes as spherical Lagrangian patches centered on peaks of a spherically smoothed field, but this does not describe their actual boundaries. A variational reformulation replaces the spherical patch by a region of fixed volume and arbitrary shape that minimizes the enclosed energy. The resulting boundaries are equipotential surfaces of a slightly modified gravitational potential, and these equipotential surfaces provide an excellent description of protohalo shapes, orientations and associated torques (Musso et al., 2023). In this refinement, peaks still identify centres, but protohalo boundaries are non-spherical and explicitly sensitive to the tidal field.

6. Field-level generalizations and current directions

Recent work has extended the peak model from a point-based theory of special locations to a field-level theory of deformation and transport. One such reformulation starts from the same maximum-entropy Gaussian baseline as BBKS, but interprets peaks as special cases within a hierarchy of deformation tensors $\delta_R$0 carried by a Lagrangian–Eulerian transport map,

$\delta_R$1

Here the Jacobian $\delta_R$2 factorizes in terms of deformation eigenvalues, shell crossing occurs at $\delta_R$3, and anisotropic collapse becomes a generic field-level property rather than a feature attached only to named peaks (Kitaura, 13 Apr 2026).

This framework makes two points that are directly relevant to the mature peak model. First, the nonlocal tidal level $\delta_R$4 becomes important already at moderate overdensity, with an activation threshold around

$\delta_R$5

whereas a more local curvature level $\delta_R$6 activates only later, with

$\delta_R$7

Second, the framework reinterprets anisotropy as an order parameter in a Landau–Ginzburg free energy, so that halos and filaments seeded by peaks appear as self-organized phases of the density-plus-anisotropy field (Kitaura, 13 Apr 2026).

The broader trajectory of the subject is therefore clear. The peak model began as a theory of the abundance and clustering of density maxima on a fixed smoothing scale. It was then merged with excursion sets to address cloud-in-cloud and mass assignment, generalized to peak/dip-conditioned shear statistics to describe the cosmic web, refined by first-principles nesting and bias calculations, and supplemented by non-spherical protohalo boundaries and transport-based deformation hierarchies (Paranjape et al., 2012, Rossi, 2014, Salvador-Solé et al., 2024, Kitaura, 13 Apr 2026). Current open issues follow directly from these developments: the treatment of collapse time for general ellipsoidal perturbations, stochastic mapping from peak scale to final halo mass, extension beyond Gaussian initial conditions, and incorporation of baryonic physics and fully nonlinear environmental effects. The modern peak model is therefore best understood not as a single formula, but as a family of analytically connected constructions organized around one central claim: collapsed cosmic structure forms preferentially out of special regions of the initial field, and the correct theory is obtained by specifying ever more faithfully which regions are special, how they are nested, and how anisotropic gravity evolves them.