---
title: Non-Hitting Index in Finite Geometry & Probability
url: https://www.emergentmind.com/topics/non-hitting-index
type: topic
---

# Non-Hitting Index in Finite Geometry & Probability

The term **non-hitting index** has two closely related uses. In finite geometry and finite-field polynomial theory, it is an explicit numerical invariant counting lines that do not meet a prescribed point set or graph. In stochastic analysis, the same phrase is not always used formally, but several papers identify canonical critical numbers that play the same role: they separate target sets that are hit with positive probability from polar sets, or quantify the long-time decay of survival probabilities. Across these settings, the central theme is the same: a non-hitting index measures how strongly an object avoids intersections, either combinatorially, through incidence counts, or probabilistically, through capacity, Hausdorff measure, and hitting-time asymptotics [2003.06678] [1810.05386].

## 1. Finite-geometric definition and basic formalism

In the finite-geometric setting introduced by Li and Pott, the non-hitting index is defined for point sets in the classical projective plane \(PG(2,q)\) and for polynomials over \(\mathbb{F}_q\). If \(D \subset PG(2,q)\) with \(|D|=q+1\), its projective intersection distribution is the sequence
\[
u_i(D)=\#\{\text{lines } \ell \subset PG(2,q): |\ell\cap D|=i\}, \qquad 0\le i\le q+1.
\]
The non-hitting index is
\[
u_0(D),
\]
the number of lines in \(PG(2,q)\) that contain no point of \(D\). The largest \(i\) with \(u_i(D)\neq 0\) is called the degree of \(D\) [2003.06678] [2510.04675].

The affine polynomial version is defined as follows. For \(f\in\mathbb{F}_q[x]\), let
\[
v_i(f)=\#\{(a,b)\in\mathbb{F}_q^2 : f(x)-ax-b=0 \text{ has precisely } i \text{ solutions in } \mathbb{F}_q\}, \qquad 0\le i\le q.
\]
Then the affine non-hitting index of \(f\) is
\[
v_0(f),
\]
which counts the non-vertical affine lines \(y=ax+b\) that do not intersect the graph \(\Gamma_f=\{(x,f(x)):x\in\mathbb{F}_q\}\) [2510.04675].

The affine and projective viewpoints are canonically linked by the \((q+1)\)-set
\[
S_f=\{(x,f(x),1):x\in\mathbb{F}_q\}\cup\{(0,1,0)\}\subset PG(2,q).
\]
For this construction,
\[
v_0(f)=u_0(S_f),
\]
so the non-hitting index is the same invariant seen in affine and projective coordinates [2510.04675]. The point \((0,1,0)\) is an internal nucleus of \(S_f\), and conversely every \((q+1)\)-set with an internal nucleus is projectively equivalent to some \(S_f\) [2003.06678] [2510.04675].

## 2. Algebraic and geometric meaning

Geometrically, a large non-hitting index means that a \((q+1)\)-set is sparse with respect to lines: many lines of the ambient plane miss it entirely. Algebraically, for a polynomial \(f\), the same quantity measures how often the equations
\[
f(x)=ax+b
\]
have no solution in \(\mathbb{F}_q\). This is the most direct interpretation of \(v_0(f)\) [2510.04675].

The polynomial formulation has an equivalent value-set description. For each \(c\in\mathbb{F}_q\), define
\[
V_{f,c}=\{f(x)+cx:x\in\mathbb{F}_q\}.
\]
Then
\[
v_0(f)=q^2-\sum_{c\in\mathbb{F}_q}|V_{f,c}|.
\]
Thus the non-hitting index records the total deficiency of the \(q\) value sets \(\{f(x)+cx:x\in\mathbb{F}_q\}\): the larger \(v_0(f)\) is, the smaller these value sets are on average [2003.06678].

This invariant is constrained by standard incidence identities. For a \((q+1)\)-set \(S\subset PG(2,q)\),
\[
\sum_{i=0}^{q+1}u_i(S)=q^2+q+1,\qquad
\sum_{i=1}^{q+1} i\,u_i(S)=(q+1)^2,
\]
and
\[
\sum_{i=2}^{q+1} i(i-1)\,u_i(S)=q(q+1).
\]
From these identities one obtains
\[
u_0(S)=\frac{q(q-1)}{2}-\sum_{i=3}^{q+1}\binom{i-1}{2}u_i(S),
\]
so higher-order secants necessarily reduce the non-hitting index [2003.06678]. In the later work on intersection distributions, the same counting philosophy is extended to reconstruct \(u_0\) or \(v_0\) once the higher intersection numbers are known [2510.04675].

The degree of \(S_f\) interacts strongly with the non-hitting index. Since \(\deg(S_f)\) is the largest possible number of \(\mathbb{F}_q\)-solutions of \(f(x)-ax-b=0\), it measures the maximum line-intersection multiplicity, while \(v_0(f)\) measures the number of lines with zero intersection. The papers emphasize that these are complementary descriptors of the same incidence distribution [2510.04675].

## 3. Extremal bounds, spectrum, and classification in finite planes

For a \((q+1)\)-set \(S\subset PG(2,q)\), the non-hitting index satisfies
\[
0\le u_0(S)\le \frac{q(q-1)}{2}.
\]
The lower extreme \(u_0(S)=0\) holds if and only if \(S\) is a line, while the upper extreme \(u_0(S)=\frac{q(q-1)}{2}\) holds if and only if \(S\) is a \((q+1)\)-arc [2003.06678]. These are the fundamental extremal cases: a line is maximally hit, and an arc is maximally avoided.

A finer lower bound depends on the degree \(n\) of \(S\). If \(S\) has degree \(n\), then
\[
u_0(S)\ge n(q+2-n)-(q+1).
\]
In particular, if \(n=q\), then \(u_0(S)=q-1\) if and only if \(q\) points of \(S\) lie on a line and the remaining point is off that line. For \(3\le n\le q-1\), one has \(u_0(S)\ge 2q-4\) [2003.06678]. This shows that small non-hitting index corresponds to highly collinear configurations.

For polynomials, the same extremal picture becomes a classification statement. One always has
\[
v_0(f)\ge q-1,
\]
with equality if and only if \(f\) is linear. At the other end,
\[
v_0(f)\le \frac{q(q-1)}{2}.
\]
Equality holds if and only if \(S_f\) is a \((q+1)\)-arc; for even \(q\), this means that for exactly one \(c\in\mathbb{F}_q\), the polynomial \(f(x)-cx\) is an o-polynomial, while for odd \(q\), \(f\) is projectively equivalent to \(x^2\) [2003.06678].

The later paper develops the **non-hitting spectrum**
\[
\mathrm{Spec}(q)=\{u_0(S):S\subset PG(2,q),\ |S|=q+1\},
\]
equivalently the set of possible values \(v_0(f)\) for polynomially representable \((q+1)\)-sets [2510.04675]. Li–Pott had previously identified the smallest values
\[
0,\ q-1,\ 2q-4,\ 2q-3,
\]
and the later analysis shows that, for large \(q\), the next values include
\[
3q-9,\ 3q-8,\ 3q-7,\ 3q-6,
\]
while also demonstrating that these larger values no longer determine a set up to projective equivalence [2510.04675]. This corrects a possible misconception: the non-hitting index is a strong invariant, but it is not a complete classifier once one moves beyond the extremal part of the spectrum.

Concrete families illustrate the range of behavior. For a linear polynomial \(f(x)=ax+b\), one has \(v_q(f)=1\), \(v_1(f)=q(q-1)\), and \(v_0(f)=q-1\), so the non-hitting index is minimal among polynomial examples. For cubic polynomials \(f=x^3+ax^2\), the non-hitting index is computed through irreducible cubic counts:
\[
v_0(f)=\frac{q^2-1}{3}\quad (p\neq 3),\qquad
v_0(f)=\frac{q^2}{3}\quad (p=3,\ a\neq 0),\qquad
v_0(f)=\frac{q(q-1)}{3}\quad (p=3,\ a=0).
\]
For monomials \(x^d\), the degree \(\deg(S_f)\) is controlled by \(\gcd(d,q-1)\), \(\gcd(d-1,q-1)\), and \(\lfloor \tfrac12+\sqrt{q-1}\rfloor\), which in turn constrains the non-hitting index through the global intersection identities [2510.04675].

## 4. Capacity and Hausdorff thresholds in stochastic PDEs

In stochastic PDEs, the phrase **non-hitting index** is not always part of the paper’s formal terminology, but several works identify exact critical quantities with the same operational meaning: they determine whether sets are polar or non-polar. This is explicit in the fractional stochastic heat system of spatial dimension \(1\),
\[
\frac{\partial u_i}{\partial t}(t,x)
=
D^\alpha u_i(t,x)
+
\sum_{j=1}^d \sigma_{ij}(u(t,x))\,\dot W^j(t,x)
+
b_i(u(t,x)),
\]
with \(1<\alpha\le 2\), globally Lipschitz coefficients, and a uniformly elliptic diffusion matrix \(\sigma\) [1810.05386].

The analytic input is a sharp Gaussian-type bound for the two-point density of \((u(s,y),u(t,x))\):
\[
p_{s,y;t,x}(z_1,z_2)
\le
C\,\Delta_\alpha((t,x);(s,y))^{-d}
\exp\!\Big(
-\frac{c\,|z_1-z_2|^2}{\Delta_\alpha((t,x);(s,y))^2}
\Big),
\]
where the adapted parabolic distance is
\[
\Delta_\alpha((t,x);(s,y))
=
|t-s|^{\frac{\alpha-1}{\alpha}}
+
|x-y|^{\alpha-1}.
\]
This estimate yields lower bounds on hitting probabilities in terms of Newtonian capacity and upper bounds in terms of Hausdorff measure [1810.05386].

For the three natural observation sets, the critical quantities are
\[
\gamma_{space\text{-}time}=\frac{2(\alpha+1)}{\alpha-1},\qquad
\gamma_{time\;section}=\frac{2}{\alpha-1},\qquad
\gamma_{space\;section}=\frac{2\alpha}{\alpha-1}.
\]
Equivalently, the capacity indices are
\[
d-\frac{2(\alpha+1)}{\alpha-1},\qquad
d-\frac{2}{\alpha-1},\qquad
d-\frac{2\alpha}{\alpha-1}.
\]
The paper states that, although it does not use the term “non-hitting index,” these numbers naturally define such an index: if \(d\) is larger than the relevant threshold, points are polar; if \(d\) is smaller, points are non-polar [1810.05386]. For \(\alpha=2\), these become
\[
6,\qquad 2,\qquad 4,
\]
recovering the classical heat-equation critical dimensions and removing the extra \(+\eta\) present in earlier multiplicative-noise results [1810.05386].

A related picture appears for stochastic heat and wave equations driven by an additive fractional Brownian sheet with temporal index \(1/2\) and spatial index \(H\le 1/2\). For the Gaussian field \(u\), the anisotropic Hölder exponents are
\[
(H/2,H)\quad\text{for the heat equation},\qquad
(H,H)\quad\text{for the wave equation},
\]
so the effective parameter
\[
Q=\sum_{j=1}^N \frac1{\tau_j}
\]
becomes
\[
Q=\frac{3}{H}\quad\text{for the stochastic heat equation},\qquad
Q=\frac{2}{H}\quad\text{for the stochastic wave equation}.
\]
The hitting-probability bounds are
\[
C^{-1}\operatorname{Cap}_{d-Q}(A)
\le
\mathbb{P}\{u(I\times J)\cap A\neq\emptyset\}
\le
C\,\mathscr{H}_{d-Q}(A).
\]
For singleton targets, points are non-polar when \(d<Q\) and polar when \(d>Q\). The paper explicitly calls \(Q\) the critical dimension of hitting probabilities, and this is precisely the role of a non-hitting index in the SPDE context [1608.00085].

## 5. Hitting-time formulations and survival exponents

A different probabilistic usage arises from first hitting times. For a one-dimensional strictly \(\alpha\)-stable Lévy process with \(\alpha\in(1,2)\), killed upon hitting the origin,
\[
\tau_0=\inf\{t>0:X_t=0\},
\]
the survival probability admits the spectral representation
\[
\mathbb{P}_x(\tau_0>t)
=
\frac{\alpha\sin(\pi/\alpha)}{\pi\cos\theta}
\int_0^\infty e^{-s^\alpha t}\,\frac1s\,F_-(sx)\,ds,
\qquad x\neq 0,\ t>0,
\]
where \(F_-\) is a generalized eigenfunction and \(\theta\) encodes skewness [1910.12821]. The reconstruction of the paper’s consequences identifies a natural temporal non-hitting index
\[
\beta=1-\frac1\alpha,
\]
through the large-time asymptotic
\[
\mathbb{P}_x(\tau_0>t)\asymp t^{-\beta}.
\]
This interpretation is presented as a natural definition suggested by the spectral formula rather than as the paper’s own terminology [1910.12821].

For non-backtracking random walks on \(ER(N,p)\) networks, the paper studies the first hitting time \(d\), defined as the path length before termination by retracing or trapping. The tail distribution satisfies
\[
P(d>\ell)\simeq \exp\left[-\frac{\ell(\ell-1)}{2\alpha^2}-\beta\ell\right],
\]
with
\[
\alpha=\sqrt{\frac{N(c+1)}{c}},\qquad
\beta=-\ln(1-e^{-c}),\qquad c=(N-1)p.
\]
The tail therefore factors into a discrete Rayleigh component and an exponential component. The same reconstruction proposes several possible non-hitting indices for this setting, including the normalized mean first hitting time \(\mathbb{E}[d]/\sqrt{N}\), the Rayleigh scale parameter \(\alpha\), and composite quantities involving \(\beta\) [1609.08375]. Here again, the underlying idea is that non-hitting is quantified not by a line-incidence count but by how long a dynamics remains self-avoiding before a first hit occurs.

These examples show that, in stochastic processes, a non-hitting index need not be a single canonical number. Depending on the model, it may appear as a persistence exponent, a capacity threshold, or a scale parameter in a hitting-time distribution. This suggests that the concept is structural rather than purely terminological.

## 6. Conceptual unification and scope

Across the cited literatures, the non-hitting index has a precise core meaning: it quantifies avoidance relative to a family of potential intersections. In finite geometry, the family consists of projective or affine lines, and the index is exactly the number of lines that miss the set. In stochastic PDEs, the relevant family is a class of target sets \(A\subset\mathbb{R}^d\), and the index appears as the critical exponent in \(\operatorname{Cap}_{d-\gamma}(A)\) or \(\mathscr{H}^{d-\gamma}(A)\). In hitting-time problems, the same role is played by survival exponents or effective scale parameters controlling the decay of \(\mathbb{P}(\tau>t)\) [2003.06678] [1810.05386] [1910.12821].

The finite-geometric theory is the most explicit and terminologically settled. There, the non-hitting index is a coarse but robust invariant, tightly linked to projective equivalence, internal nuclei, and the degree of \(S_f\). It controls extremal configurations, constrains the full intersection distribution, and has direct applications to Kakeya sets in affine planes through the identity
\[
|K|=q^2-u_0(D_K),
\]
where \(D_K\) is the dual \((q+2)\)-set associated with a Kakeya set \(K\) [2003.06678] [2510.04675].

The stochastic literature suggests a broader interpretation. For the stochastic fractional heat system, smaller \(\alpha\) increases the indices
\[
\frac{2(\alpha+1)}{\alpha-1},\qquad \frac{2}{\alpha-1},\qquad \frac{2\alpha}{\alpha-1},
\]
so, in a fixed state-space dimension \(d\), points and other thin sets become less likely to be polar [1810.05386]. For rough fractional noises, smaller \(H\) increases
\[
Q=\frac{3}{H}\quad\text{or}\quad Q=\frac{2}{H},
\]
again making hitting of points possible in higher dimensions [1608.00085]. In both cases, rougher or more dispersive dynamics correspond to a larger critical hitting dimension and therefore to weaker non-hitting behavior for small targets.

A plausible synthesis is that a non-hitting index always identifies a transition between regimes of guaranteed avoidance and possible intersection. In finite planes, this transition is exact and combinatorial. In stochastic systems, it is potential-theoretic or asymptotic. The common structure is the existence of a threshold parameter—\(u_0\), \(v_0\), \(d-\gamma\), \(Q\), or \(1-\frac1\alpha\)—that summarizes how an object interacts with the ambient family of probes, whether those probes are lines, compact target sets, or the origin itself under time evolution [2510.04675] [1810.05386] [1910.12821].

Source: https://www.emergentmind.com/topics/non-hitting-index