Does spherical mean the data must lie on a sphere?

No. It refers to growing balls around locations in the ambient space; in a plane these are disks.

Is this the same as the nearest-neighbor distribution?

No. That distribution starts from a point in the process, whereas the empty-space function starts from an ordinary spatial location.

Why exclude D = 0 for some random sets?

Covered locations already touch the set, so conditioning on being outside separates positive contact distances from the coverage probability.

AI Glossary / Mathematics and optimization / Advanced

Spherical contact distribution function

Definition

The spherical contact distribution function describes how far a location in space is from the nearest part of a random spatial pattern. For a stationary point process, it is the probability that a ball of radius r around a fixed location contains at least one point. It is also called the empty-space distribution function, usually written F(r). For random sets with area or volume, a conditional version excludes locations already inside the set.

What it means

The starting location is an ordinary location in space, not a point selected from the pattern. This distinction separates contact distance from point-to-point nearest-neighbor distance. If facilities cluster in one area, facilities may be close to one another while many ordinary locations remain far from any facility. The contact distribution summarizes the sizes of gaps from the perspective of space. It is a cumulative probability curve, not one average distance or a complete description of spatial structure.

How it works

Let D be the shortest distance from a fixed location to the random set. For a point process with no point at that fixed location almost surely, F(r) = P(D ≤ r) = 1 − P(no points in the radius-r ball). Stationarity makes the theoretical distribution independent of which fixed location is chosen. For a random set that can cover the location, the conditional spherical contact function is H(r) = P(D ≤ r | D > 0). Estimation from a finite observation window uses distances from many locations, often through a distance map. Edge correction matters because unseen objects beyond the window may be closer than the observed ones.

Three practical examples

A spatial-service study can summarize how much geographic space lies within specified straight-line distances of observed facilities, while distinguishing that quantity from population-weighted access.
An ecological analysis can compare empty-space distances in a mapped point pattern with a stated spatial model.
A materials-image study can examine distances from the surrounding material to a segmented phase, using the conditional random-set version when pixels inside that phase should be excluded.

Think of it this way

Think of placing a pin at an ordinary map location and expanding a circle until it first touches a marked site. Repeating the exercise describes contact distances. Starting at a marked site instead asks a different question about its neighbors. The analogy uses circles in two dimensions; the mathematical idea extends to balls in higher dimensions and requires a clear rule for selecting locations and measuring distance.

Worked example

For a homogeneous planar Poisson point process with intensity λ, the number of points in a radius-r disk is Poisson with mean λπr². The probability of no points is exp(−λπr²), so F(r) = 1 − exp(−λπr²). Suppose λ = 0.01 points per square meter. At r = 5 meters, λπr² = π/4, giving F(5) approximately 0.544. At r = 10 meters, the mean count is π, giving F(10) approximately 0.957. These are model probabilities, not measurements of an actual site map. For a separate conditional-versus-unconditional illustration, take five equally weighted sampled locations with distances [0, 0, 1, 2, 3] from an area-covering set. The uncorrected empirical fraction with distance at most 2 is 4/5. Excluding the two covered locations leaves distances [1, 2, 3], so the corresponding conditional fraction is 2/3. This small table illustrates conditioning only; it is not an edge-corrected estimate or evidence for the Poisson model.

Illustrative case study

In a fictional project, Willow Spatial Lab compares two maps of observation sites. It calculates an empty-space curve over the same study boundary and distance units, then checks sensitivity to boundary correction and missing sites near the edges. Analysts also create a separate population-weighted accessibility measure because a uniform location in land area is not the same as a randomly selected resident. The case illustrates matching a statistic to a question, without claiming that either map provides better real-world access.

Common misconceptions and mistakes

Despite the name empty-space function, F(r) is the probability of contact by radius r, while 1 − F(r) is the probability that the ball remains empty. It is not the probability density of an exact distance. For clustered processes, the contact and nearest-neighbor distributions can differ substantially; their equality under a homogeneous Poisson model is not universal. A curve close to a model's curve does not prove the entire spatial process follows that model. Nonstationarity, observation boundaries, measurement resolution and incomplete detection can all affect interpretation.

Business use

Use the statistic when distance from space to a pattern is the intended viewpoint. Document the observation window, metric, units, sampling or pixel resolution and whether conditioning excludes covered locations. Choose an estimator and edge correction appropriate to the data. Compare against a justified reference model, including its intensity assumptions, rather than using the Poisson formula automatically. If the operational question involves road travel, terrain barriers or population locations, a Euclidean stationary contact curve may be only a partial description and should be labeled accordingly.

Check your understanding

Under the Poisson example, F(5) is approximately 0.544. What does 1 − F(5) describe?

Answer: The probability that the five-meter disk around the fixed location contains no points, approximately 0.456.

Related terms

empty-space function
nearest-neighbor distribution
point process
stationary random set
distance transform
edge correction

Learn next

plot the Poisson contact curve for several intensities using consistent units. Then compare distances from a regular grid of query locations with distances between the observed sites themselves. Keep the two histograms separate and explain which question each answers.

Sources

Sources: Adrian Baddeley, Rolf Turner and Ege Rubak, spatstat documentation, Estimate the Empty Space Function, https://search.r-project.org/CRAN/refmans/spatstat.explore/html/Fest.html ; Spherical Contact Distribution Function, https://search.r-project.org/CRAN/refmans/spatstat.explore/html/Hest.html . The Poisson calculations, five-location table and fictional mapping study are original examples.

Previous: Spell checker Next: Sphericity