Pair Distribution Function¶
A normalized radial function comparing particle-pair occurrence at each separation with the occurrence expected from the system's mean number density.
Core Idea¶
For particles whose positions can be treated statistically as homogeneous and isotropic, the normalized pair distribution function \(g(r)\) asks how densely other particles occur at distance \(r\) from a reference particle, relative to the system's mean number density. Under the stated radial convention, if \(\rho(r)\) is this conditional pair density and \(\rho_0\) the mean density, \(g(r)=\rho(r)/\rho_0\). Thus \(g(r)>1\) indicates an excess of partners at that separation relative to the same-density baseline; \(g(r)<1\) indicates depletion. This is a dimensionless pair-structure descriptor, not a probability density whose integral over \(r\) equals one.[1]
The function's structural value is to retain separation dependence while averaging over which particle played the reference role. In a homogeneous three-dimensional isotropic convention, the expected number of partners in a thin shell is \(4\pi r^2\rho_0g(r)\,dr\); counting a named coordination shell therefore requires declared integration bounds. Peaks can express common neighbor separations, but no fixed number, height or shape of peaks belongs to the definition. A hard-core system may have \(g(r)=0\) at forbidden separations; a decorrelating simple liquid tends toward \(g(r)=1\) at large \(r\). Those are conditional models, not universal axioms for soft particles, crystals or critical states.[1][2]
Structural Signature¶
Sig role-phrases: reference particle in a spatial ensemble → separation-conditioned partner density → same-system mean-density baseline → declared radial normalization → bounded reading of excess, depletion and peaks.
- Reference particle and spatial carrier. Each counted pair has a focal particle and another particle at a displacement. Averaging over admissible focal particles makes the function an ensemble descriptor rather than a story about one particular atom. Remove the particle-conditioned origin and a generic distance histogram is not yet this normalized pair function.[1]
- Separation-resolved partner density. The question is how many partners occur at each radial separation, not merely which one is closest. Retaining every pair distance is what lets neighboring shells and longer-range organization remain visible.[1]
- Same-system density baseline. Dividing conditional density \(\rho(r)\) by mean number density \(\rho_0\) gives the unitless relative result. A different normalization changes the numerical object; the reduced scattering function \(G(r)\), for example, requires a conversion before its height can be read as \(g(r)\).[1]
- Radial convention. A scalar \(g(r)\) uses distance rather than full displacement direction and needs a declared spatial dimension, averaging window and normalization. For anisotropic or multicomponent structure, vector-valued or species-partial pair functions may be needed instead of pretending one scalar retains everything.[1]
- Conditional structural interpretation. Excesses, depletions and peak sequences indicate relative pair occurrence under that convention. A zero excluded core, large-\(r\) approach to one, sharp crystal peaks or an inferred coordination number are results under additional conditions, not mandatory roles of every \(g(r)\).[2][1][3]
Neither neutron scattering nor particle imaging is constitutive. Both can supply data from which a pair function is obtained, but they bring different correction and sampling problems.[1][3]
What It Is Not¶
It is not the nearest-neighbor distance distribution. That live identity asks, for each typical focal point, only where the first other point appears; \(g(r)\) counts the relative frequency of partners at all separations. It is also not a unit-integral probability density over \(r\): shell volume grows with \(r\), and \(g(r)\) is normalized to a spatial number-density baseline rather than to total mass one.[1]
Nor is every published “PDF” numerically this \(g(r)\). Total-scattering studies use reduced \(G(r)\) and other weighted forms. In Peterson and colleagues' monatomic convention, \(G(r)=4\pi r\rho_0[g(r)-1]\); common underlying local-order information does not erase the additive and scale differences. Multicomponent scattering can additionally weight unlike atomic pairs differently. A raw Fourier transform of \(g(r)\), with no subtraction of its baseline or specification of scattering convention, is not the ordinary structure-factor relation.[1][2]
Scope of Application¶
The scalar form is most literal for statistically homogeneous and isotropic particle ensembles: atomic liquids, simple fluids, colloidal suspensions and suitably qualified materials data. A model or measurement must identify the particle species, spatial density, dimensionality, pair-separation calculation and averaging window. Liquids may show local shell peaks and weak distant correlation; an ordered sample may retain prominent long-range features. The same definition can organize these cases, but the simple-liquid limiting picture must not be projected onto all of them.[1][2][3]
In scattering, a corrected structure factor can be transformed into a real-space pair representation under specified weighting, number-density and Fourier conventions. In microscopy, three-dimensional particle positions can be used to count pairs directly. The observations are not interchangeable as raw values: scattering has finite-\(Q\) truncation and species-weighting issues, while microscopy has localization and field-of-view limitations. Neither route removes the need to state which \(g(r)\) convention is reported.[1][2][3]
Clarity¶
The essential question is not “How likely is a pair?” without qualification. It is “Compared with how many partners would be expected at this separation from the same mean density?” In three-dimensional isotropic sampling, a shell at larger \(r\) contains more volume; a rising raw pair count can therefore coexist with a flat \(g(r)=1\). This separates geometric opportunities for pairs from actual relative structure.[1]
It also separates similar-looking plots that answer different questions. A \(G(r)\) curve can carry the same pair information yet have different units and baseline; a nearest-neighbor distribution conditions on only the first arrival; a structure factor is reciprocal-space information. Before comparing peak heights or asymptotes, identify the plotted function and its density/scattering convention.[1][2]
Manages Complexity¶
A particle configuration contains many coordinates and \(N(N-1)/2\) unordered pair relationships for \(N\) particles. Radial averaging compresses those relationships into a one-variable function, making excess, exclusion zones and coordination shells comparable without retaining the identity of every particle. For a homogeneous isotropic three-dimensional system, \(4\pi\rho_0\int_{r_1}^{r_2}r^2g(r)\,dr\) then estimates partners within declared shell bounds, rather than relying on visual peak height alone.[1]
That compression has a price. Distinct directional organizations or species-pair mixtures can share a similar scalar \(g(r)\). Total-scattering intensity may also be represented by several real-space functions whose differences matter in absolutely normalized analysis. The useful simplification is therefore separation-resolved relative density with its convention attached, not a claim that a single curve reconstructs a unique microscopic configuration.[1]
Abstract Reasoning¶
Given particle positions or a properly reduced scattering result, first choose the population and averaging window. In the radial case, count partner density in a shell around reference particles, divide by the shell's volume to obtain a conditional density, and divide again by \(\rho_0\) to obtain \(g(r)\). A peak means partners are overrepresented at that distance relative to the stated baseline; it does not itself prove a unique lattice, bond or interaction potential. For a hard-core simple-liquid model, a forbidden-distance gap and damped shell peaks have more specific interpretations, but those interpretations depend on the model.[1][2]
If converting from scattering, state the relevant \(S(Q)\) and finite-\(Q\) conventions. The spatial Fourier relation involves the deviation \(g(r)-1\) multiplied by number density under the homogeneous convention; it is not “\(S\) is the Fourier transform of \(g\)” without qualification. If using a coordination integral, state dimension and shell limits. Inferring energy or pressure from a pair curve requires additional equilibrium and interaction-potential assumptions, so the curve alone is not a universal thermodynamic equation of state.[1][2]
Knowledge Transfer¶
Within particle physics and materials science, the same pair-conditioning and baseline-ratio operation can be applied to atomic liquids and confocal-resolved colloids. The first may infer a real-space function from reciprocal-space scattering; the second can count resolved three-dimensional separations. The transfer is literal only after each method produces the same normalized \(g(r)\), rather than a visually similar but differently scaled real-space plot.[1][3]
The broader thought of comparing pair co-occurrence with a baseline resembles portable correlation or normalization reasoning. But the checked live Correlation describes variable co-variation; it does not automatically supply a strict parent for a conditioned spatial number-density ratio. Extending a \(g\)-like comparison to nonparticle records would require defining units, spatial or other separation, reference density and independence baseline anew. Such an extension is a future-prime question, not proof that this named physical function is prime.
Examples¶
Simulated liquid argon in total scattering. Peterson and coauthors use molecular-dynamics simulations to display several real-space functions for liquid Ar, including normalized \(g(r)\), and distinguish it from reduced \(G(r)\). Mapped back: carrier = Ar atoms in the simulated liquid, each eligible as reference; partner density = separation-resolved Ar–Ar information represented in simulated total-scattering functions; baseline = model mean Ar number density \(\rho_0\); radial convention = the article's monatomic conversion to \(g(r)\); interpretation = local peaks indicate relatively favored pair separations without assigning each peak a unique microscopic cause. This is a modeled positive case, not an experimental Ar measurement, and does not equate \(G(r)\) to dimensionless \(g(r)\).[1]
Confocal-imaged charged colloids. Kodger, Guerra and Sprakel locate particles in three dimensions and calculate \(g(r)\) from those positions. Their screened sample has a fluid-like arrangement, whereas deionization yields a colloidal Wigner crystal with distinct sharp peaks. Mapped back: carrier = imaged colloidal particles; partner density = pair separations accumulated around focal colloids; baseline = the sample's mean particle number density at its stated volume fraction; radial convention = a three-dimensional radial \(g(r)\) from coordinates; interpretation = changed peak structure documents a different pair arrangement. The paper's interaction interpretation uses its controlled-charge context; the peak pattern alone would not identify the cause.[3]
Normalization boundary. A reduced total-scattering \(G(r)\) for the same sample can encode the same local-order information, but under Peterson and coauthors' convention its ordinate is \(4\pi r\rho_0[g(r)-1]\). Reading that ordinate as a relative probability without conversion is a negative, not a third instance of normalized \(g(r)\).[1]
Structural Tensions¶
Compact radial comparison versus retained directional or species detail. Radial averaging makes samples and distances easy to compare, but it can erase anisotropy or unlike-species pair distinctions. Favor one curve and chemically or directionally different arrangements may look alike; favor full vector or partial functions and interpretation, estimation and visualization become more demanding. Diagnostic: Could the conclusion change if pair direction or species identity were retained?[1]
Scattering reach versus normalization and sampling transparency. Total scattering reaches atomic separations unavailable to direct imaging but carries correction, finite-\(Q\) and scattering-weight conventions. Position-resolved microscopy shows particles directly, yet its finite field of view and localization still limit pair estimation. Favor scattering without auditing the transform and truncation ripples can be mistaken for real short-range features; favor imaging without a declared window and missing neighbors near its boundary can skew counts. Diagnostic: What transformation, baseline density, weighting and spatial window produced this reported curve?[1][2][3]
Structural–Framed Character¶
Evaluative weight. \(g(r)\) is not a judgment that an arrangement is desirable; an excess at a distance is a relative statistical statement. Deciding whether a peak supports crystallization, packing or a particular interaction is a further, model-dependent inference.[1][3]
Human-practice dependence. Scientists choose the averaging window, radial convention and measurement route; nevertheless a specified particle configuration and density constrain the function's value. Institutional origin. Scattering laboratories and microscopy groups have developed different PDF vocabularies, but no laboratory or instrument constitutes the normalized pair relation.[1]
Vocabulary travel. “Pair distribution” may be used for \(g(r)\), reduced \(G(r)\) or other conventions in adjacent communities, so the words travel farther than the precise normalization. Import versus recognition. Recognize this entry by its reference-conditioned partner density divided by a same-system baseline; calling an arbitrary co-occurrence chart a “pair distribution” merely imports the label. Its character: strongly structural within statistical particle descriptions, with a measurement vocabulary that must be normalized explicitly and without established prime-level cross-domain reach.[1]
Structural Core vs. Domain Accent¶
Portable skeleton. A relative pair-occurrence pattern—condition on a focal item, compare neighboring occurrence with a declared baseline—might recur outside particles. That is an explicit future-prime question, not an accepted prime or typed parent here. The checked live Correlation handles co-variation of variables; resemblance to it does not prove that every spatial pair-density ratio instantiates its exact signature.
Domain-bound mechanism. Spatial particle coordinates, radial shells, reference-particle conditioning, mean number density and physical pair statistics make this particular \(g(r)\) identifiable. Scattering relates it to \(S(Q)\) only with normalization and transform conventions; microscopy computes it only after detecting positions. These are not decorative examples of a generic chart but conditions that determine what the curve means.[1][2][3]
Why not prime. Atomic liquids and colloids are distinct scales and acquisition methods but still inhabit particle structure. The evidence does not show the exact \(g(r)=\rho(r)/\rho_0\) construction operating as a native abstraction in several independent nonphysical domains. The broader compare-to-baseline skeleton is separable from the named radial function, so the latter remains domain-specific.
Instantiates / Related Primes¶
Correlation is a useful portable neighbor, not an asserted necessary genus: its live identity is systematic variable co-variation, whereas the present function is a conditioned spatial pair-density ratio. Nearest neighbour distribution retains only minimum distance and is neither equivalent to nor a strict parent of all-pair \(g(r)\). Scattering is an optional acquisition process, not a required parent; confocal imaging supplies a different positive route. Probability Density Function and Spatial Distribution are also neighboring descriptions, not tested exact identities. No canonical DAG edge is asserted.
Neighborhood in Abstraction Space¶
Pair Distribution Function sits in a sparse region of the domain-specific corpus (72nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Statistical Mechanics & Particle Phenomena (15 abstractions)
Nearest neighbors
- Laser Diffraction Analysis — 0.85
- Phi Coefficient — 0.83
- Quantum Concentration — 0.83
- Geometrical Frustration — 0.83
- Boltzmann–Matano Analysis — 0.83
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
Nearest-neighbor distribution: a distribution of the closest partner's distance from a typical point, not relative density of all pairs at every separation. Reduced \(G(r)\): a related real-space total-scattering representation with different scale, baseline and units; it may be converted under stated conventions. Structure factor \(S(Q)\): reciprocal-space correlation information, linked to a real-space deviation from baseline rather than numerically identical to \(g(r)\). Probability density over distance: integrates to one under its measure, unlike the normalized density ratio \(g(r)\). Pair potential: an interaction model that may help explain a curve but is not contained uniquely in its plotted peaks.[1][2]
References¶
[1] Peter F. Peterson, Daniel Olds, Marshall T. McDonnell and Katharine Page, “Illustrated formalisms for total scattering data: a guide for new practitioners,” Journal of Applied Crystallography 54 (2021), 317–332, §2 simulation setup, §4, Figures 3–4, Table 4 and Appendix A equations (32)–(38). The article distinguishes normalized \(g(r)\), reduced \(G(r)\) and scattering-weighted or partial functions. Its published corrigendum corrects some other Figure 3–4 ordinates and multicomponent limits; no affected plotted ordinate is used here. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28
[2] Roger Pynn, “Lecture 1: Introduction & Neutron Scattering ‘Theory’,” NIST NCNR summer-school hosted teaching slides (2009), PDF pp. 15–16, especially “Static pair correlation” and “S(Q) and g® for Simple Liquids.” These pages support the conditional simple-liquid core, shell-peak and transform-caution claims, not a universal hard-core rule. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k
[3] Thomas E. Kodger, Rodrigo E. Guerra and Joris Sprakel, “Precise colloids with tunable interactions for confocal microscopy,” Scientific Reports 5 (2015), 14635, Results text around Figure 5 and Figure 5 caption. Original research computes radial \(g(r)\) from three-dimensional confocal particle positions and contrasts screened-fluid and deionized crystal cases. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i