Adler-Taylor Metric Factor

• The Adler–Taylor metric factor is defined via the covariance of a Gaussian field’s gradient, inducing a Riemannian metric on a manifold.
• In three-dimensional non-isotropic settings, it is computed as the Jacobian volume-distortion from the eigenvalues of the induced metric and relates to scalar curvature.
• Under isotropy, the metric factor simplifies to the first spectral moment, serving as the normalization in Gaussian kinematic formulas for excursion-set geometry.

The Adler–Taylor metric factor is a term used in several closely related parts of Gaussian random field geometry. For a centered, nondegenerate Gaussian field $f:M\to\mathbb R$ of unit variance on a Riemannian manifold $(M,g)$, the associated Adler–Taylor metric $g^f$ is defined by

$g^f_p(v,v):=\mathbb E[d_pf(v)^2],$

so that the field induces a second Riemannian metric through the local covariance of its gradient. In three-dimensional non-isotropic settings, the relevant local metric factor is the Jacobian of volume, $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$, where $a_1,a_2,a_3$ are the eigenvalues of $g^f$ with respect to the background metric $g$. In isotropic spherical settings, the same expression is often compressed to a single scalar, the first spectral moment $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$, which enters Kac–Rice formulas for expected boundary length. In the Gaussian kinematic formula, the phrase “metric factor” is also used for the universal coefficient

$$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$

which multiplies $(M,g)$0 in the Adler–Taylor expansion (Pistolato et al., 2024, Caponera et al., 14 Jul 2025, Fu, 2024).

1. Definition of the Adler–Taylor metric

Let $(M,g)$1 be a smooth Riemannian manifold of dimension $(M,g)$2, and let $(M,g)$3 be a centered, nondegenerate Gaussian field with unit variance. The Adler–Taylor metric, also called the induced metric of the field, is the Riemannian metric $(M,g)$4 defined pointwise by

$(M,g)$5

Nondegeneracy of $(M,g)$6 is exactly the condition that $(M,g)$7 be positive-definite for every $(M,g)$8, so $(M,g)$9 is a bona fide Riemannian metric (Pistolato et al., 2024).

In the isotropic sphere-cross-time setting of a zero-mean, unit-variance Gaussian field $g^f$0 on $g^f$1, one instead starts from the multipole expansion

$g^f$2

with angular power spectrum at lag zero given by

$g^f$3

Under isotropy, the variance of the gradient is spatially constant and equals

$g^f$4

Much of the excursion-set literature works with its square root,

$g^f$5

and refers to $g^f$6 as the metric factor or first spectral moment (Caponera et al., 14 Jul 2025).

A separate but standard usage appears in the Gaussian kinematic formula, where the metric factor is the universal constant

$g^f$7

The three usages are not identical, but they all arise from the same geometric principle: local covariance structure is converted into the coefficients governing expected geometric functionals (Fu, 2024).

2. Local metric factor and scalar curvature in dimension three

In the three-dimensional setting of $g^f$8, let $g^f$9 denote the eigenvalues of $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$0 with respect to $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$1. Equivalently, in any $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$2-orthonormal basis of $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$3, the matrix of $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$4 is $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$5. Two derived quantities enter the excursion-set formulas.

The first is the local volume-distortion factor,

$g^f_p(v,v):=\mathbb E[d_pf(v)^2],$6

This is the Jacobian factor relating the volume form of $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$7 to the volume form of the ambient metric $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$8. In the terminology of the paper, it is the local “metric-factor” of the induced metric (Pistolato et al., 2024).

The second is the scalar curvature of $g^f_p(v,v):=\mathbb E[d_pf(v)^2],$9, denoted $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$0. It may be computed through sectional curvatures via

$\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$1

or, in local coordinates $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$2, by contracting the Riemann tensor: $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$3 The paper emphasizes that once one has the Christoffel symbols of $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$4, one can assemble the Riemann tensor $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$5 and then its scalar curvature (Pistolato et al., 2024).

This local viewpoint is important because it separates the geometry induced by the Gaussian field from the background geometry in which one wishes to measure Lipschitz–Killing curvatures. The resulting formulas do not require $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$6, and that distinction is the core of the non-isotropic extension.

3. Expected Lipschitz–Killing curvatures under an arbitrary background metric

The classical Adler–Taylor expectation formula gives $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$7 in terms of the Lipschitz–Killing curvatures $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$8 of $\sqrt{\det g^f_p}=\sqrt{a_1a_2a_3}$9. The three-dimensional non-isotropic extension instead expresses $a_1,a_2,a_3$0 directly in terms of quantities computed under the background metric $a_1,a_2,a_3$1, together with the eigenvalues $a_1,a_2,a_3$2 of $a_1,a_2,a_3$3 relative to $a_1,a_2,a_3$4 (Pistolato et al., 2024).

Under the standing assumptions that $a_1,a_2,a_3$5 is three-dimensional and that $a_1,a_2,a_3$6 is nondegenerate and unit-variance, with $a_1,a_2,a_3$7, the formulas are

$a_1,a_2,a_3$8

$a_1,a_2,a_3$9

$g^f$0

and

$g^f$1

The auxiliary functions are

$g^f$2

$g^f$3

where $g^f$4 are i.i.d. $g^f$5. When two of the $g^f$6 coincide, these one-dimensional integrals collapse to elementary functions (Pistolato et al., 2024).

A key methodological point is that the formulas depend on $g^f$7 and $g^f$8, but not on the full Levi–Civita connection of $g^f$9. The paper states explicitly that one never needs the Christoffel symbols of $g$0 once the $g$1 are known. This sharply distinguishes the non-isotropic three-dimensional theory from approaches that reconstruct the induced metric in full.

4. Closed-form induced metric for spin fields on $g$2

The principal example is the real part $g$3 of a left-invariant spin-$g$4 Gaussian field $g$5 on the group manifold $g$6. In Euler angles $g$7, the induced Adler–Taylor metric is itself left-invariant, and its $g$8 Gram matrix is

$g$9

The constants are determined by spectral moments: $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$0 where $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$1 is the circular covariance of $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$2 (Pistolato et al., 2024).

With respect to the standard $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$3-metric $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$4, the Adler–Taylor eigenvalues are constant: $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$5 The determinant and volume-distortion factor are therefore

$\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$6

The scalar curvature is also constant: $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$7 In this example, the functions $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$8 and $\lambda_1=\sqrt{\operatorname{Var}(\nabla Z)}$9 reduce to closed-form elementary combinations of $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$0, and substitution into the general three-dimensional formulas yields fully explicit expressions for $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$1, $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$2, depending only on $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$3, $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$4, and $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$5 (Pistolato et al., 2024).

The same paper places this computation in the context of spherical spin random fields used to model the Cosmic Microwave Background polarization. A plausible implication is that explicit control of $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$6 and $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$7 makes the induced geometry accessible for geometric statistical analyses of non-isotropic spin data.

5. First spectral moment and frequency-domain estimation

For isotropic Gaussian fields on $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$8, the Adler–Taylor metric factor is identified with the variance of the gradient, or more commonly with its square root $$\mathrm{MF}_{k,m}=(2\pi)^{-(k+m)/2}k!\,\omega_{k+m},$$9. Its direct geometric role is in the expected boundary length of excursion sets: $(M,g)$00 where $(M,g)$01 is the standard normal density (Caponera et al., 14 Jul 2025).

The estimation strategy in the long-memory setting begins from Wiener–Itô expansions for the excursion area $(M,g)$02 and boundary length $(M,g)$03. Defining

$(M,g)$04

the low-frequency leading-order relation is

$(M,g)$05

This yields a narrow-band least-squares estimator in the frequency domain. For Fourier frequencies $(M,g)$06, $(M,g)$07, the NBLS estimator is

$(M,g)$08

Its validity is formulated under explicit assumptions. The monopole $(M,g)$09 must satisfy

$(M,g)$10

higher Hermite chaoses must have strictly shorter memory, the bandwidth must obey Condition B,

$(M,g)$11

and the underlying NBLS theory requires Gaussianity together with suitable “linear-process” or cumulant conditions in Robinson 1994, formulated in the paper as Conjecture 1 (Caponera et al., 14 Jul 2025).

In the “Case a” regime

$(M,g)$12

Proposition 4.6 gives

$(M,g)$13

Thus $(M,g)$14 is consistent, with rate determined by the memory gap $(M,g)$15. The same paper reports Monte Carlo evidence of negligible bias and shrinking variance as $(M,g)$16 grows, and gives a step-by-step implementation recipe based on centering $(M,g)$17 and $(M,g)$18, choosing $(M,g)$19, computing the discrete Fourier transforms, and recovering $(M,g)$20 from $(M,g)$21 (Caponera et al., 14 Jul 2025).

6. Universal metric factor in the Gaussian kinematic formula

In Fu’s treatment of the Gaussian kinematic formula, the metric factor is identified with a universal constant rather than with a field-specific gradient variance. If $(M,g)$22 is a compact positive-reach subset, $(M,g)$23 is the canonical Gaussian vector field of $(M,g)$24 independent standard spherical harmonics, and $(M,g)$25 is a compact positive-reach set, then for each $(M,g)$26,

$(M,g)$27

Here $(M,g)$28 is the $(M,g)$29-th Lipschitz–Killing valuation on $(M,g)$30, normalized so that $(M,g)$31 is volume and $(M,g)$32 is Euler characteristic, while $(M,g)$33 is the $(M,g)$34-th Gaussian intrinsic volume, defined through the tube formula

$(M,g)$35

The metric factor is therefore

$(M,g)$36

It depends only on $(M,g)$37 and is universal in the sense of not depending on the particular field $(M,g)$38, set $(M,g)$39, or excursion set $(M,g)$40 (Fu, 2024).

The paper explains the origin of each part of the constant. The factor $(M,g)$41 comes from the normalization of the Gaussian density and the Poincaré-limit approximation of large-dimensional spherical density; $(M,g)$42 comes from the $(M,g)$43-th derivative in the tube formula and from the combinatorics of the spherical kinematic operator; and $(M,g)$44 is the Euclidean volume of the unit $(M,g)$45-ball (Fu, 2024).

In the proof, the constant emerges from the Poincaré-limit of spherical kinematic formulas. The crucial asymptotic identification is

$(M,g)$46

combined with the convergence of projected spherical densities to the Gaussian density and the limiting identification of spherical curvature valuations with Gaussian intrinsic volumes (Fu, 2024).

7. Conceptual relations and common points of confusion

The phrase “Adler–Taylor metric factor” is not used in a single uniform way across the literature represented here. One work uses it for the local Jacobian $(M,g)$47 associated with the induced metric $(M,g)$48; another uses it for the isotropic scalar $(M,g)$49; and another uses “metric factor” for the universal coefficient $(M,g)$50 in the Gaussian kinematic formula (Pistolato et al., 2024, Caponera et al., 14 Jul 2025, Fu, 2024).

The principal misconception is to treat these quantities as interchangeable. They are related, but they arise at different levels of the theory. In the three-dimensional non-isotropic setting, the geometry of excursion sets under a background metric $(M,g)$51 depends on the full eigenvalue triple $(M,g)$52 and on $(M,g)$53, not on a single scalar. In isotropic spherical settings, that local structure collapses to the single first spectral moment $(M,g)$54, which is sufficient for the boundary-length expectation. In the Gaussian kinematic formula, the “metric factor” is instead a universal normalization constant converting curvature valuations and Gaussian intrinsic volumes into the expected curvature of random intersections.

This suggests that the shared terminology reflects a common function rather than a single object: each “metric factor” encodes the scalar normalization by which the covariance structure of a Gaussian field enters excursion-set geometry. The degree of compression depends on the symmetry class. Under isotropy it can reduce to $(M,g)$55; in non-isotropic three-dimensional problems it is mediated by the eigenvalues of $(M,g)$56 and the scalar curvature of the induced metric; and in the abstract Gaussian kinematic formula it appears as the universal constant multiplying $(M,g)$57.