---
title: Orbit Dissimilarity Function (D-Function) Overview
url: https://www.emergentmind.com/topics/orbit-dissimilarity-function-d-function
type: topic
---

# Orbit Dissimilarity Function (D-Function) Overview

Searching arXiv for recent and foundational papers on orbit dissimilarity functions across meteor science, operator algebras, and geometric orbit distance.
{"query":"all:\"orbit dissimilarity\" OR title:\"orbit dissimilarity\"","max_results":10,"sort_by":"submittedDate","sort_order":"descending"}
Searching arXiv for exact cited works to anchor the article.
{"query":"1910.10729","max_results":5,"sort_by":"relevance","sort_order":"descending"}
Orbit dissimilarity function, or **D-function**, denotes a scalar comparison construction whose precise meaning depends on the surrounding theory. In meteor science, a D-function assigns one value to two meteoroid orbits or to two meteors represented through observed or geocentric quantities, and small values are interpreted as orbital similarity; together with a threshold, it becomes a **D-criterion** used to form meteor groups [2507.19075]. In operator algebras, \(D_c\) compares normal elements through Cuntz comparison and is designed to control the distance between closures of unitary orbits [1403.0927]. In diffeomorphism-invariant learning, DID can be read as an orbit-aware dissimilarity for functions under smooth reparameterizations of the domain, although it is not defined as a quotient metric on orbit space [2202.05614]. In celestial mechanics, the Keplerian distance and the MOID are geometric orbit-to-orbit distances measuring actual Euclidean separation between confocal Keplerian trajectories [2305.13900][1905.10096].

## 1. General definition and structural role

In the literature considered here, a D-function is always a scalar object attached to a pair of mathematical objects regarded through an orbit relation, a similarity relation, or an actual geometric orbit. What changes from field to field is the meaning of “orbit,” the data used in the comparison, and the logical role of the resulting scalar.

| Context | Objects compared | Role of the D-function |
|---|---|---|
| Meteor science | Two meteoroid orbits or observed/geocentric parameters | Pairwise similarity used with a threshold and clustering algorithm |
| Simple \(C^*\)-algebras | Normal elements or associated homomorphisms | Controls distance between closures of unitary orbits |
| Diffeomorphism-invariant learning | Functions \(f,g:X\to Y\) under \(f\mapsto f\circ Q\) | Regularized certificate of similarity under diffeomorphic reparameterization |
| Celestial mechanics | Two confocal elliptic or circular Keplerian orbits | Actual geometric orbit distance or computable bounds on it |

A recurring pattern is that the scalar is not self-sufficient. In meteor science, a D-function must be paired with a threshold \(D_S\) and a grouping algorithm; in operator algebras, \(D_c\) must be supplemented by \(K_1\)-information in the nontrivial case; in DID, regularization and a mask break symmetry; and in MOID work, the distance captures closest spatial approach but not timing or long-term dynamical accessibility [2507.19075][1403.0927][2202.05614][1905.10096].

## 2. Meteor-shower D-functions and D-criteria

In meteor science, the review literature distinguishes sharply between a **D-function** and a **D-criterion**. The D-function is the scalar comparison map; the D-criterion is the function plus a threshold \(D_S\). Operationally, if
\[
D(\text{object}_1,\text{object}_2)<D_S,
\]
the two meteors are treated as orbitally similar and may be assigned to the same meteor group. The review also distinguishes “meteor group,” meaning a set of meteors with similar orbits, from “meteor shower,” meaning a group additionally shown to share a common origin [2507.19075].

The classical orbital-element criteria are \(D_{SH}\) of Southworth and Hawkins, \(D_D\) of Drummond, and \(D_H\) of Jopek. The review presents \(D_{SH}\) as historically foundational and still dominant in meteor-shower identification, \(D_D\) as a dimensionless normalized alternative, and \(D_H\) as a hybrid intended to combine a physically motivated structure with partial normalization. Beyond these, the review discusses geocentric or observational formulations such as Valsecchi’s \(D_N\) and \(D_R\), Rudawska’s \(D_X\), vectorial criteria such as Neslušan’s momentum-vector distance and Jopek et al.’s \(D_V\), and reduced-variable or dynamical comparators such as Jenniskens’ \(D_B\) and \(D_T\) [2507.19075].

The meteor-astronomy use case is not limited to a choice of formula. The same D-function can behave very differently depending on the threshold and on the clustering procedure. The review identifies several operational modes: all-pairs searches to discover new groups, comparisons against one or more reference orbits to assign meteors to known showers, single-neighbour or single-linkage chaining, iterative mean-orbit procedures, two-stage procedures using multiple D-functions, and density-based methods such as DBSCAN. A central conclusion is that D-functions are pairwise similarity kernels embedded inside broader workflows rather than standalone identification methods [2507.19075].

Several recurring criticisms are also emphasized. Many classical criteria incorporate \(\Omega\) or solar longitude, so time is effectively built into the similarity score. This can split a long-duration shower into shorter subgroups. Standard orbital-element functions also operate in a five-dimensional element space even though Earth intersection makes the meteor-shower similarity problem effectively four-variable. The review further notes that \(D_{SH}\) and \(D_D\) fail the triangle inequality, so they are not true distances in the mathematical sense. Threshold choice is treated as a balance between false positives and false negatives, and the review strongly favors dataset-specific calibration against the sporadic background rather than blind reuse of literature values [2507.19075].

## 3. Static-reference bias in shower activity profiles

A major limitation of meteor D-criteria appears when they are used not merely to isolate dynamically similar meteors, but to recover a shower’s activity profile as a function of solar longitude. The central physical point is that meteoroids in a real shower can disperse in longitude of the ascending node, and therefore in solar longitude, while preserving a common Sun-centered ecliptic radiant and geocentric speed. A static reference orbit plus a fixed D-threshold therefore penalizes true members observed far from the shower peak, making the shower “appear briefer than it actually is” [1910.10729].

The paper makes this argument explicitly with the Drummond criterion
\[
D^2=
\left(\frac{q_m-q_s}{q_m+q_s}\right)^2+
\left(\frac{e_m-e_s}{e_m+e_s}\right)^2+
\left(\frac{I}{180^\circ}\right)^2+
\left(\frac{e_m+e_s}{2}\cdot \frac{\theta}{180^\circ}\right)^2,
\]
where \(q\) is perihelion distance, \(e\) is eccentricity, and the subscripts \(m\) and \(s\) denote the meteor and the shower reference orbit. The angular terms are
\[
I=\arccos\!\left[\cos i_m \cos i_s+\sin i_m \sin i_s \cos(\Omega_m-\Omega_s)\right]
\]
and
\[
\theta=\arccos [\sin \eta_m \sin \eta_s+\cos \eta_m \cos \eta_s \cos(\gamma_s-\gamma_m)],
\]
with
\[
\gamma=\Omega+\arctan (\cos i \tan \omega),\qquad
\eta=\arcsin (\sin i \sin \omega).
\]
The paper states that the shortening effect is not peculiar to Drummond \(D\); it “will occur for any method that incorporates time, solar longitude, or longitude of the ascending node into a single measure of shower member likelihood,” including most other variants [1910.10729].

The demonstration uses simulated Perseid and Southern Taurid showers with double-exponential activity profiles in solar longitude. Radiants are scattered with a circularly symmetric angular offset having standard deviation \(3^\circ\), and geocentric speeds are drawn from a normal distribution with standard deviation equal to \(10\%\) of the shower speed. Two selection methods are then compared. Method 1 is a direct cut in observational space: within \(5^\circ\) of the Sun-centered ecliptic radiant and within \(20\%\) of the geocentric speed. Method 2 uses Drummond \(D\) with \(D<0.18\) for Perseids and \(D<0.06\) for Southern Taurids, following Galligan’s thresholds for approximately 70% stream retrieval in AMOR radar data. In the present simulation those thresholds recover only 52% of Perseids and 25% of Taurids, excluding noise, whereas the direct radiant/velocity cut recovers 86% of meteors in both showers, regardless of whether they are shower or noise meteors. This is used to argue that D-thresholds are not universal and depend on instrument precision [1910.10729].

The strongest quantitative example concerns the Southern Taurids. The simulated true profile uses \(B=0.026\), but fitting the \(D\)-selected sample yields \(B\approx 0.08\), so the recovered profile is between 3 and 4 times steeper than the true one. At the same time, the same Drummond-\(D\) filter recovers 25% of simulated Taurids but only 2% of simulated noise, showing why D-based selection remains attractive when the aim is to derive representative stream orbits rather than durations or rise and decay rates. The paper therefore recommends two ways to avoid temporal compression: detect showers initially using only Sun-centered ecliptic radiant and velocity, as in the wavelet coefficient method of Brown et al. (2008), or replace a single static reference orbit with a time-dependent set of reference orbits or radiants/velocities, as in the “look-up table” approach of Jenniskens et al. (2018) [1910.10729].

## 4. Comparison-theoretic orbit dissimilarity in simple \(C^*\)-algebras

In the \(C^*\)-algebraic setting, the relevant orbit is the **unitary orbit** of a normal element. For a unital \(C^*\)-algebra \(A\), a normal element \(x\in A\), and unitary group \(U(A)\), the orbit closure is
\[
U(x)=\overline{\{u^*xu:u\in U(A)\}}.
\]
For two normal elements \(x,y\in A\), the geometric quantity under study is \(\mathrm{dist}(U(x),U(y))\). The paper introduces \(D_c\) as a comparison-theoretic dissimilarity for the associated homomorphisms \(\varphi_x,\varphi_y:C(\Omega)\to A\), where \(\Omega=\mathrm{sp}(x)\cup \mathrm{sp}(y)\) and \(\varphi_x(f)=f(x)\), \(\varphi_y(f)=f(y)\) [1403.0927].

The core definition is
\[
D_c(\kappa_1,\kappa_2)=
\sup\Bigl\{
\inf\{d>0:\kappa_1(f_O)\lesssim \kappa_2(f_{O_d})\}
:\ O\subset \Omega,\ {\rm open}
\Bigr\},
\]
where \(f_O\) is a positive function with support exactly \(O\), \(O_d=\{\xi\in\Omega:\mathrm{dist}(\xi,O)<d\}\), and \(\lesssim\) denotes Cuntz subequivalence of positive elements. For normal elements,
\[
D_c(x,y):=D_c(\varphi_x,\varphi_y).
\]
This construction does not compare \(x\) and \(y\) directly in norm. It compares how spectral pieces of \(x\) and \(y\) sit in the algebra through the Cuntz semigroup. In the stable-rank-one simple setting, a priori asymmetry disappears, and \((H_{c,1}(C(\Omega),A),D_c)\) is a metric space [1403.0927].

The main orbit-distance statements are upper and lower bounds under increasingly restrictive hypotheses. If \(A\) is a unital simple separable \(C^*\)-algebra with real rank zero, stable rank one, and weakly unperforated \(K_0(A)\), and if \([\lambda-x]=0\) and \([\mu-y]=0\) in \(K_1(A)\) outside the spectra, then
\[
\mathrm{dist}(U(x),U(y))\le D_c(x,y).
\]
For unital separable AF-algebras, no \(K_1\)-assumption is needed:
\[
\mathrm{dist}(U(x),U(y))\le D_c(x,y).
\]
A refined essential version \(D_e\) is introduced by splitting off arbitrarily small common finite-dimensional pieces supported in \(X\cap Y\), and under matching \(K_1\)-data one has
\[
\mathrm{dist}(U(x),U(y))\le D_e(x,y),
\]
while in favorable “hub” configurations the stronger \(\mathrm{dist}(U(x),U(y))\le D_c(x,y)\) is recovered [1403.0927].

The lower-bound side is equally important. In AF-algebras there exists a constant \(C>0\) such that
\[
C\cdot D_e(x,y)\le \mathrm{dist}(U(x),U(y))\le D_c(x,y).
\]
The introduction states that this constant is universal, that Davidson computed it at least \(1/3\), and that it cannot be improved to \(1\) even in matrices. The same paper also warns against a common misconception: \(D_c\) is not a complete invariant of orbit distance in the presence of nontrivial \(K_1\)-information. Theorem 7.2 gives examples with
\[
D_c(u_1,u_2)=0
\qquad\text{and}\qquad
\mathrm{dist}(U(u_1),U(u_2))=2,
\]
so \(D_c\) alone measures only the Cuntz/spectral-comparison side of dissimilarity and must be supplemented by \(K_1\)-obstruction terms [1403.0927].

## 5. Diffeomorphism-invariant dissimilarity and orbit actions

DID is introduced as a pairwise dissimilarity for data represented as functions
\[
f,g:X\to Y,
\]
with invariance sought under right-composition by a diffeomorphism \(Q:X\to X\). A natural orbit interpretation is
\[
\mathcal O(f)=\{f\circ Q:\ Q \text{ a diffeomorphism of the domain}\},
\]
although the paper itself does not define DID by minimizing over this orbit. Instead, it uses the change-of-variables formula and an RKHS optimization over weighting functions \(q\) to eliminate \(Q\) implicitly [2202.05614].

The definition is
\[
D_\lambda(f,g):=
\max_{\|h\|\le 1}\min_{q\in \mathcal H}
\Delta_{f,g}(h,q)+\lambda\|q\|^2,
\]
where
\[
\Delta_{f,g}(h,q)=
\max_{\|v\|_{\mathcal F}\le 1}
\left|
\int_X v(g(x))q(x)\,dx-
\int_X v(f(x))\mu(x)h(x)\,dx
\right|^2.
\]
With the operators
\[
F_\mu=\int_X \Phi(f(x))\otimes \psi(x)\mu(x)\,dx,
\qquad
G=\int_X \Phi(g(x))\otimes \psi(x)\,dx,
\]
the paper shows
\[
\Delta_{f,g}(h,q)=\|F_\mu h-Gq\|_{\mathcal F}^2
\]
and derives the closed form
\[
D_\lambda(f,g)=
\lambda\left\|(GG^*+\lambda I)^{-1/2}F_\mu\right\|_{op}^2.
\]
The inner minimization is \(2\lambda\)-strongly convex, so it has a unique global minimizer [2202.05614].

The invariance is quantitative and approximate rather than exact for fixed \(\lambda>0\). If \(X\subset \mathbb R^d\) is open and bounded with Lipschitz boundary, \(\mu\in C^\infty(\mathbb R^d)\) has compact support \(\Omega\subset X\), \(k_X\) is a Sobolev kernel of smoothness \(m>d/2\), and \(Q\) is a \(C^{m+1}\) diffeomorphism with \(Q^{-1}(\Omega)\subset X\), then
\[
D_\lambda(f,f\circ Q)\le \lambda\, C_\mu^2 C_Q^2
\qquad \forall f\ \text{measurable},
\]
so
\[
D_\lambda(f,f\circ Q)\to 0 \quad \text{as } \lambda\to 0.
\]
This makes DID approximately orbit-invariant for smooth domain warps in a precise asymptotic sense [2202.05614].

The paper is equally explicit about limitations. Because of the mask \(\mu\), symmetry between \(f\) and \(g\) is broken: \(f\) becomes the reference and one searches in \(g\) for matching statistics. Identity of indiscernibles is not established; for \(\lambda>0\), even \(D_\lambda(f,f)\) need not be zero because regularization introduces bias. No triangle inequality is proved, and no converse theorem states that small DID implies the existence of a diffeomorphism \(Q\) with \(g=f\circ Q\). DID is therefore best read as a regularized RKHS-based certificate of diffeomorphic equivalence rather than a metric on orbit space [2202.05614].

## 6. Geometric orbit distance, MOID, and explicit lower bounds

In celestial mechanics, the relevant “orbit dissimilarity” is often not a heuristic element-space similarity score but the actual Euclidean distance between two confocal Keplerian trajectories. For two elliptic orbits \({\cal E}_1,{\cal E}_2\) with common focus, the Keplerian distance is
\[
d(V)=\sqrt{\langle {\cal X}_1-{\cal X}_2,{\cal X}_1-{\cal X}_2\rangle},
\]
where \(V=(v_1,v_2)\) specifies one point on each orbit. The absolute minimum
\[
d_{\min}=\min_V d(V)
\]
is the MOID. The computation is organized through the critical points of \(d^2\), not \(d\), because “we squared the distance \(d\) to include crossing points among the critical ones.” The paper revisits two algebraic routes to all critical points of \(d^2\): one using ordinary polynomials and one using trigonometric polynomials. In both formulations the generic problem reduces to a univariate polynomial of degree \(16\), which the paper describes as minimal in the general elliptic case [2305.13900].

The same work emphasizes that MOID is a genuine geometric dissimilarity measure between two confocal Keplerian trajectories. It measures actual minimum spatial separation rather than a weighted difference of orbital elements. At the same time, it is not presented as a metric on orbital-element space, and it does not encode timing, phase, dynamical accessibility, or long-term evolution. Two orbits can have small MOID yet be dynamically very different, and conversely element-wise similar orbits can have a nontrivial geometric separation. This places MOID in a different category from classical meteor \(D\)-criteria [2305.13900].

A related paper gives a computable positive lower bound for MOID between two noncoplanar bounded Keplerian orbits with a common focus. If \(\rho(\mathcal E,\mathcal E')\) denotes MOID and \(\sigma(\mathcal E,\mathcal E')\) is the nodal-distance-based upper bound constructed from the four distinguished nodal pairings, then the new lower bound is
\[
\tau(\mathcal E,\mathcal E')=C(e,e',I)\,\sigma(\mathcal E,\mathcal E'),
\]
with
\[
\tau(\mathcal E,\mathcal E') \le \rho(\mathcal E,\mathcal E') \le \sigma(\mathcal E,\mathcal E').
\]
The coefficient is
\[
C(e, e', I) = \sqrt{ \frac{(1-e)(1-e')\sin^2\!I} {(1-e)(1-e')\sin^2\!I +2\left(1+|\cos I|\sqrt{(1-e^2)(1-e'^2)}-ee'\right)} }.
\]
For noncoplanar pairs, this lower bound is positive and vanishes if and only if the orbits intersect, so it is a rigorous geometric filter rather than a heuristic similarity score [1905.10096].

The computational significance is substantial. On the first \(20{,}000\) asteroids of the MPC numbered catalog, yielding \(199{,}990{,}000\) distinct pairs, the lower-bound test skipped \(63.77\%\), \(78.62\%\), and \(87.65\%\) of pairs for MOID thresholds \(0.01\), \(0.005\), and \(0.0026\) AU, with corresponding total speedups of \(2.41\), \(3.78\), and \(5.92\). This suggests a useful distinction within orbit dissimilarity research: some D-functions are thresholded similarity surrogates, whereas MOID and its explicit lower bounds are geometrically certified orbit-distance quantities that can be used as safe rejectors in large catalogs [1905.10096].

Source: https://www.emergentmind.com/topics/orbit-dissimilarity-function-d-function