---
title: f-Normality in Mathematics
url: https://www.emergentmind.com/topics/f-normality
type: topic
---

# f-Normality in Mathematics

$f$-normality is a multifaceted concept spanning several mathematical disciplines, each with a precise technical meaning adapted to its context—probability theory, Boolean and $p$-ary function analysis, ergodic theory, holomorphic dynamics, computational complexity, and finite field theory. Across these domains, $f$-normality captures a property of extremal behavior or structural regularity: convexity of likelihood ratios, maximal dimensions of constant flats, equidistribution under general representations, or algebraic independence of roots. Modern research reveals that $f$-normality is a unifying theme with deep characterizations, tight bounds, and powerful algorithms at the interface of analysis, algebra, and computation.

## 1. Convex Likelihood Ratios and the Characterization of Gaussian Densities

In probability theory, $f$-normality, as introduced by Jacobovic and Kella, identifies those positive densities on $\mathbb{R}^d$ for which shifted likelihood ratios $L_y(x) = f(x+y)/f(x)$ are convex functions of $x$ for all $y$. The main theorem provides a striking analytic characterization:

- For a positive Borel density $f:\mathbb{R}^d \to \mathbb{R}$, the following are equivalent:
  1. For every $y$, $x \mapsto f(x+y)/f(x)$ is convex.
  2. For every $y$, $x \mapsto f(x+y)/f(x)$ is log-convex or log-concave.
  3. $f$ is a (multivariate) Gaussian density.

The proof hinges on convex analysis: defining $g(x) = 1/f(x)$ and exploiting convexity properties, one deduces that $g$ is an exponential-quadratic, leading to $f$ being precisely Gaussian ($A$ positive-definite).

The result has both theoretical and applied implications:
- It supplies a new convex-analytic characterization of normality, distinguishing Gaussian distributions by the convexity of all their location-shift likelihood ratios.
- It resolves a longstanding question in statistical testing: only for Gaussian densities does the convexity condition force inadmissibility of Moran's two-stage test for location parameters.

A concrete one-dimensional counterexample is the Laplace density $f(x) = \frac{1}{2}e^{-|x|}$, for which $L_y(x)$ fails convexity for $y \neq 0$; thus, it is not $f$-normal in this sense [2110.14173].

## 2. $f$-Normality in Boolean and $p$-ary Function Theory

In finite field settings, $f$-normality typically quantifies the largest dimensional affine flat on which a function is structurally simplified—being either constant or affine. In the Boolean setting ($f:\mathbb{F}_2^n \to \mathbb{F}_2$):

- A function is $k$-normal if it is constant on some affine $k$-flat; the normality $\mathcal{N}(f)$ is the maximal $k$ achievable.
- A central result relates algebraic sparsity (ANF sparsity, $T(f)$) to normality. Specifically, if $T(f)\leq n^{3-\epsilon}$, then $\mathcal{N}(f)\geq c n^{\epsilon/2}$, and an explicit polynomial-time algorithm finds such a flat.
- For $p$-ary functions $f:\mathbb{F}_p^n \to \mathbb{F}_p$ (notably for bent functions), theory provides sharp upper bounds: for regular bent functions in even $n$, $k\le n/2$; for weakly regular but non-regular, $k\le n/2-1$; and for $n$ odd, $k\le (n-1)/2$. Algorithms efficiently test and certify (non-)normality of these functions [1410.1318, 1706.06427].

This notion of $f$-normality is essential for coding theory, cryptography, and complexity, with explicit constructions matching the theoretical tradeoff bounds for a wide spectrum of Boolean and $p$-ary functions.

## 3. $f$-Normality in Ergodic Theory: Separator Enumerators and Finite-State Dimension

A distinct, representation-driven notion of $f$-normality emerges in symbolic dynamics and effective randomness, as formalized via separator enumerators (SEs):

- An SE $f: \Sigma^* \to [0,1)$ assigns to each finite word a real number, generating a dense countable separator in $[0,1)$. For fixed $x$ and $f$, the $f$-normality of $x$—denoted $\dim^{f}_{\mathrm{FS}}(x)=1$—means that $x$ is maximally incompressible relative to $f$ under any finite-state transducer, capturing the generalized finite-state dimension.
- Mayordomo posed whether $f$-normality could be characterized by equidistribution of derived numeric sequences (e.g., $(k^n a_n^f(x))$). The answer is negative in full generality: there exist computable $f_0, f_1$ and $x$ with identical approximation sequences but different $f$-normalities. Thus, no equidistribution property of a single numeric sequence suffices uniformly across all SEs.
- However, restricting to finite-state coherent SEs—those arising from bijective synchronous Mealy automata—restores the equidistribution characterization: $x$ is $f$-normal if and only if $(k^n a_n^f(x))$ is uniformly distributed modulo $k^m$ for all $m$ [2602.01199].

This dichotomy suggests a deep sensitivity of $f$-normality to the structure of the representation and the complexity class of encoders.

## 4. $f$-Normality in Complex Analysis: Normal Families and Holomorphic Correspondences

In complex analysis and dynamics, $f$-normality manifests as the classical notion of normality for families of meromorphic or holomorphic functions:

- A family $\mathcal{F}$ is normal if every sequence has a subsequence which converges locally uniformly (either with respect to the spherical or the chordal metric).
- For an individual meromorphic function $f$ on the unit disc $\mathbb{D}$, normality (in the sense of Lehto–Virtanen) is equivalent to a uniform bound on $(1-|z|^2)f^\#(z)$, where $f^\#$ is the spherical derivative.
- A function-sharing criterion stipulates that if three boundary-distinct meromorphic functions $\psi_1,\psi_2,\psi_3$ all share the same solution set with $f$ in $\mathbb{D}$, then $f$ is normal in this sense. A two-share plus one-avoidance version also suffices.

For holomorphic correspondences on $\mathbb{P}^1$, the normality set $N(F)$ generalizes the classical Fatou set, comprising points where all local inverse branches of all iterates of $F$ form a normal family. The support of canonical invariant (Dinh–Sibony) measures constructed from $F$ is always disjoint from $N(F)$ [2509.16740, 1210.6011].

## 5. Normality Criteria for Polynomials over Finite Fields

In the algebraic context of finite fields, normality (also called $f$-normality) for irreducible polynomials $f \in \mathbb{F}_q[X]$ of degree $n$ requires that the roots of $f$ form a normal basis of the extension $\mathbb{F}_{q^n}/\mathbb{F}_q$:

- This property can be checked via the non-vanishing of a circulant determinant $\Delta_n(r_0, ..., r_{n-1})$, where $r_i = r^{q^i}$.
- Symmetrization and symmetric reduction techniques enable translation of the root-based criterion into explicit conditions on coefficients via particular polynomials $\theta_{n,m}(a_1, ..., a_n)$. For families $f(X) = X^n + X^{n-1} + a$, explicit forms for $n = 6,7$ have been computed.
- More generally, for an arbitrary finite Galois extension and Galois group $G$, vanishing of group-determinant-derived symmetric polynomials on the coefficients provides a sufficient criterion for normality.

This approach enables algorithmic certification of $f$-normality for large-degree or structurally special polynomials and links group representations with field theory [2309.05470].

## 6. Connections, Limitations, and Open Problems

Across all domains, $f$-normality acts as a structural separator, uniquely identifying normal or extremal objects (such as the Gaussian, maximal flats, or normal bases) via analytic, algebraic, or computational properties. Notably, $f$-normality conditions are often tight, with explicit counterexamples revealing the limits of each characterization:

- In convex likelihood ratios, only Gaussian densities satisfy the $f$-normality property globally; the Laplace distribution provides a concrete failure.
- In separator enumerator theory, equidistribution fails to universally capture $f$-normality; only for finite-state coherent cases does the criterion recover.
- For $p$-ary functions (bent functions), algebraic and combinatorial properties tightly constrain possible normality, and infinite families of non-maximally normal bent functions have been constructed.

Open problems persist regarding the prevalence of $f$-normality in broad classes (e.g., regular non-$n/2$-normal bent functions), algorithmic optimality for normality testing, and generalizations to non-cyclic group actions or more complex representation models.

## 7. Table of $f$-Normality Across Domains

| Domain                    | Definition/Characterization                                                     | Canonical Objects          |
|---------------------------|--------------------------------------------------------------------------------|----------------------------|
| Probability (densities)   | $L_y(x) = f(x+y)/f(x)$ convex $\forall y$                                      | Gaussians                  |
| Boolean/$p$-ary functions | Maximal dimension $k$-flat where $f$ is constant/affine                        | Sparse/flat-support functions |
| Ergodic/FS-dim/Enumerators  | $\dim^{f}_{\mathrm{FS}}(x)=1$, finite-state compressibility, Mealy machine relabeling   | (Generalized) normal numbers|
| Complex analysis dynamics | Normality set: orbits/forms normal families                                    | Fatou set, Schlicht class  |
| Polynomial algebra        | Nonvanishing of group/circulant determinant on roots or coefficients            | Normal bases of field extensions|

This multidisciplinary architecture of $f$-normality continues to spur algorithmic, combinatorial, and analytic advances with ramifications in statistics, information theory, coding, and algebraic geometry.

Source: https://www.emergentmind.com/topics/f-normality