Kac–Rice Formulae: Theory and Applications

• Kac–Rice formulae are a fundamental analytic tool for evaluating level sets and zeros of random fields using integrals with Jacobians and density functions.
• They leverage classical and coarea formulae to compute expected measures and factorial moments, enhancing our understanding of crossings and critical points.
• The methodology extends to applications in algebraic geometry, signal processing, and random matrix theory, offering precise analytic and statistical insights.

The Kac–Rice formulae constitute a fundamental analytic framework for computing the expected number, measure, or other functionals of solution sets—typically level sets or zeros—of random fields, stochastic processes, or more general random mappings. Originating from the works of Kac (for real roots of random polynomials) and Rice (for level crossings of stochastic processes), the theory now encompasses a broad spectrum of random settings, ranging from real Gaussian processes and fields, through vector-valued and piecewise-deterministic processes, to random sections of bundles over manifolds and applications in algebraic geometry, statistical inference, and random matrix theory.

1. The Classical Kac–Rice Formula and Generalizations

The prototypical Kac–Rice formula expresses the expectation of the number (or $(d{-}j)$-dimensional Hausdorff measure) of points in the set

$$C_u(f) = \{x \in D \subset \mathbb{R}^d : f(x) = u\}$$

where $f$ is typically a $C^1$ (or smoother) random field $f: D \to \mathbb{R}^j$. Under mild nondegeneracy and regularity, the expectation of $C_u(f)$ in a Borel set $B \subset D$ is

$\E[ \mathcal{H}_{d-j}(C_u(f) \cap B) ] = \int_B \E[ | \det Df(x) | \,|\, f(x) = u ]\; p_{f(x)}(u)\, dx,$

where $Df(x)$ is the Jacobian matrix, and $p_{f(x)}$ is the density of $f(x)$ at $u$ (Berzin et al., 2022, Armentano et al., 2023).

Higher factorial moments admit analogous expressions in terms of joint distributions at $k$ points (Berzin et al., 2022): $\E\left[ \prod_{m=1}^{k} \mathcal{H}_{d-j}(C_u(f) \cap B_m) \right] = \int_{B_1 \times \dots \times B_k} \E\left[ \prod_{m=1}^k |\det Df(x_m)| \mid f(x_1) = \dots = f(x_k) = u \right] \prod_{m=1}^k p_{f(x_m)}(u)\, dx_1 \dots dx_k.$

The classical one-dimensional level-crossing formula (for real-valued $C^1$ process $X$) is recovered as

$\E\left[\#\{t \in [a, b] : X(t) = u\}\right] = \int_a^b \E[|X'(t)| \mid X(t) = u]\, p_{X(t)}(u)\, dt.$

2. Foundational Principles: Area and Coarea Formulae

At the heart of the Kac–Rice theory is the coarea formula for $C^1$ maps $G: D\subset\mathbb{R}^d \to \mathbb{R}^j$ and nonnegative, measurable $h$ (Berzin et al., 2022): $$\int_B h(x, G(x)) |\det [DG(x) DG(x)^T]|^{1/2} dx = \int_{\mathbb{R}^j} \left[ \int_{G^{-1}(y) \cap B} h(x, y) d\sigma_{d-j}(x) \right] dy,$$ where $\sigma_{d-j}$ is the $(d-j)$-Hausdorff measure along each fiber. The Kac–Rice formula emerges as a probabilistic conditioning of this geometric change-of-variables, using test functions and properties of the random field.

3. Multivariate and Nonstationary Kac–Rice Formulae for Piecewise-Deterministic and Jump Processes

For stationary multivariate piecewise deterministic Markov processes (PDMP) such as those moving along ODE flows between random jumps, the mean intensity of continuous crossings of a smooth hypersurface $S \subset \mathbb{R}^d$ is given by (Borovkov et al., 2010): $$v_c(B) = \int_B (n(x),\, p(x))\, p(x)\, dH_{d-1}(x)$$ where $(n(x),\,p(x))$ is the scalar product of the surface normal and the vector field at $x$, $p(x)$ is the stationary density of $X_0$, and $dH_{d-1}$ is $(d{-}1)$-Hausdorff measure on $S$. This result underlies applications in queueing theory and seismology.

For nonstationary or non-diffusion processes, the Kac–Rice framework extends using time-dependent densities $p_{X(t)}(x)$ and vector fields $r(x)$ (Azaïs et al., 2017): $\E[\#\ \text{crossings of } S] = \int_S |r(x) \cdot \nu(x)| \left[\int_0^T p_{X(t)}(x)dt \right] d\sigma_{d-1}(x)$ where $\nu(x)$ is the normal and $r(x)$ the local velocity field.

Corrections for processes with jumps require accounting for both continuous and discontinuous crossings, often via compensators of the jump measure and the independence structure between continuous and jump parts (Dalmao et al., 2012).

4. Manifold and Geometric Generalizations of Kac–Rice

The Kac–Rice paradigm generalizes to random fields over manifolds, including vector bundles, stratified targets, or intersections with submanifolds. On a compact Riemannian manifold $(M,g)$, the expected $(n-1)$-volume of the nodal set $Z_f = f^{-1}(0)$ of a nondegenerate $C^2$ function $f: M \to \mathbb{R}$ is expressible via divergence-theorem-based closed formulas (Jubin, 2019): $$\mathrm{vol}_{n-1}(Z_f) = -\int_M (\mathrm{div}\ X)\, d\mathrm{vol}_M, \quad X = F(V_f)$$ for suitable $F$. By specific choices of regularizers $F$, this yields various closed-form integrals involving $|\nabla f|$, $\mathrm{Hess}\ f$, and $\Delta f$, without recourse to limiting $\delta$-functions.

Recent measure-theoretic developments (e.g., Stecconi (Stecconi, 2021)) provide a generalized density formula for the number of preimages of a submanifold $W \subset N$ under a random $C^1$ map $X: M \to N$. Under transversality and absolute continuity, the expected count is given by integrating a density explicitly involving the Jacobian of $X$, the angle between tangent spaces, and the one-point density of $X(p)$.

5. Specialized Settings and High-Dimensional Extensions

Critical points of random fields: The Kac–Rice framework computes not just level-set measures but also the expected number or density of critical points, classified by Hessian index, for smooth (often Gaussian or chi-field) random fields on manifolds (Marinucci et al., 2024): $\E\left\{\#\left\{p \in A : f(p) \in B,\, \nabla f(p) = 0 \right\}\right\} = \int_A \E\left[ |\det \nabla^2 f(p) | \mathbf{1}_{ \{ f(p) \in B \} } \mid \nabla f(p) = 0 \right] p_{\nabla f(p)}(0) d\mathrm{vol}_g(p)$

Random matrix theory: The joint density of eigenvalues and right eigenvectors of random non-Hermitian matrices can be interpreted through a Kac–Rice counting formula for the roots of the characteristic polynomial and its derivatives. For a matrix $X$, the density on eigenvalue-eigenvector pairs $(z,\mathbf{v})$ is (Fyodorov, 26 Jun 2025): $$\Pi_N(z,\mathbf{v}) = \frac{1}{\pi} \delta^{(2N)}((X-zI)\mathbf{v}) |P_N'(z)|^2$$ where $P_N(z)$ is the characteristic polynomial.

Singularities and intersections in algebraic geometry: For parametrized systems $f(t;\theta)=0$, Kac–Rice provides a means to express the expected number of solutions through integrals involving Jacobians and parametric densities, enabling practical high-dimensional calculations via Monte Carlo methods (Feliu et al., 2020).

6. Applications: Nodal Geometry, Statistical Inference, and Topological Quantities

The Kac–Rice approach underpins the computation of length, area, Euler characteristic, and other geometric or topological functionals of level sets—including expected nodal volumes of eigenfunctions, random wave dislocation statistics, and the mean Euler characteristic of superlevel sets (Lachièze-Rey, 2016, Azaïs et al., 2011). Beyond expectation formulas, second or higher moment Kac–Rice formulae yield variance and central-limit theorems for such geometrical quantities, allowing rigorous probabilistic statements about fluctuations and asymptotic behavior (Berzin et al., 2022, Azaïs et al., 2011).

In statistical applications, the Kac–Rice machinery provides non-asymptotic, exact pivotal distributions for test statistics in adaptive regression and high-dimensional inference—yielding powerful, finite-sample-valid p-values for global null hypotheses in regularized models such as the lasso, group lasso, or principal components (Taylor et al., 2013).

7. Technical Foundations, Limitations, and Contemporary Directions

The flexibility of Kac–Rice formulas relies fundamentally on regularity hypotheses: $C^1$ (or $C^2$) smoothness of random fields, nondegeneracy (full-rank Jacobians at zeros), and the existence and continuity of marginal and joint densities. Failure of these can lead to singularities (e.g., non-Morse critical points, degenerate crossings) requiring amended or limiting arguments (Armentano et al., 2023, Jubin, 2019). Recent advances address singularities in structured random fields and extend the range of validity to broad classes of non-Gaussian and dependent fields (Armentano et al., 2023, Stecconi, 2021).

Active research domains include the cohomological or index-theoretic analysis of random field zeros, universality of limiting distributions, manifold and stratified-target generalizations, large-deviation phenomena in random matrices, and the interplay with modern probabilistic numerical algebraic geometry (Berzin et al., 2022, Fyodorov, 26 Jun 2025). The ongoing development of generalized area/coarea theorems and probabilistic measure-theoretic tools further expands the scope of the Kac–Rice apparatus for high-dimensional, geometric, and applied probabilistic analysis.