Joint Overlap Measurements

• Joint Overlap Measurements are a class of evaluation problems that quantify the co-occurrence or mutual compatibility of multiple observables across diverse fields.
• Techniques like finite-element modeling, biased molecular dynamics, joint POVMs, and collective quantum measurements reveal critical correlation structures often missed by independent marginal analyses.
• Practical applications in engineering, biomolecular spectroscopy, quantum state estimation, LiDAR point cloud registration, and multiway genomic interval analysis underscore both the promise and challenges in accurately measuring overlaps.

Searching arXiv for the cited papers and nearby context to ground the article in published work. Joint overlap measurements denote a family of measurement and inference problems in which the quantity of interest is not an isolated marginal event but the co-occurrence, mutual compatibility, or shared spatial extent of multiple structures, observables, or states. In the literature summarized here, the term ranges from the effective gap width of mechanical overlap joints in cylindrical magnetic shields, to the reconstruction of joint conformational distributions from separately acquired spectroscopic marginals, to informationally complete joint measurements of conjugate quantum observables, optimal estimation of quantum-state overlap, per-pillar overlap estimation in bird’s-eye-view point clouds, and multiway overlap of genomic interval sets (Crawford, 2015, Hays et al., 2020, Carmeli et al., 2011, Fanizza et al., 2019, Li et al., 2023, Otlu et al., 2018).

1. Scope of the term across disciplines

The literature does not use a single universal definition of “joint overlap measurement.” Instead, the phrase refers to several technically distinct but structurally related tasks in which overlap is defined on a coupled object rather than on isolated measurements.

Domain	Measured object	Operational notion of overlap
Magnetic shielding	Mechanical overlap joint	Effective gap width in the overlap region
Biomolecular ensembles	Multiple observables $O_i$	Joint distribution $p(O_1,\ldots,O_N)$ and co-occurrence of conformational sub-populations
Finite quantum systems	Two conjugate observables	Joint POVM enabling state reconstruction
Quantum overlap estimation	Two unknown pure states	Squared overlap $c = |\langle\psi|\phi\rangle|^2$
Point-cloud registration	Two LiDAR scans	Per-pillar overlapping region and global score $\tau$
Genomic interval analysis	Multiple interval sets	Regions covered jointly by intervals from many sets

In magnetic shielding, the overlap is geometrical and electromagnetic: one shell edge extends radially outward under the other, and the relevant measured quantity is the leakage field produced by a small radial gap in the overlap region (Crawford, 2015). In biomolecular ensemble inference, overlap is probabilistic: separately acquired marginals $p(O_i)$ do not determine the true joint distribution $p(O_1,\ldots,O_N)$, and assuming independence creates spurious overlap between conformational sub-populations (Hays et al., 2020).

In finite-dimensional quantum measurement theory, the relevant object is a joint observable $C$ whose marginals reproduce noisy conjugate observables, with informational completeness determined by whether the POVM spans the full operator space (Carmeli et al., 2011). In quantum overlap estimation, the overlap is the scalar parameter $c = |\langle\psi|\phi\rangle|^2$, and the central question is whether pairwise swap tests are optimal or whether collective measurements on all copies yield lower mean square error (Fanizza et al., 2019). In LiDAR registration, overlap is defined per BEV pillar and then aggregated into a symmetric similarity score $\tau$ for loop closure (Li et al., 2023). In genomics, overlap is coordinate intersection across multiple interval sets, implemented through segment tree and indexed segment tree forest data structures (Otlu et al., 2018).

This suggests a common abstract pattern: the overlap is often a latent multivariate quantity, while the experimentally or computationally accessible data are local, pairwise, or marginal summaries.

2. Mechanical overlap joints in magnetic shielding

In "Modeling of Mechanical Overlap Joints in Magnetic Shields" (Crawford, 2015), joint overlap measurements are embedded in a tightly coupled modeling-measurement framework for high-permeability cylindrical shields. The shields are cylindrical shells split into two half-cylinders and reassembled around a cylindrical object, with the halves joined by a mechanical overlap joint rather than a butt joint. The modeled shield is a thin cylinder with 1 mm wall thickness, the overlap length in the example mesh is 25 mm, and the key geometric parameter is the gap width $g$, varied from 0.05 mm to 0.5 mm.

The computations use FEMM as a 2D magnetostatic finite-element code. The cylinder is assumed infinitely long in the axial direction, end effects are excluded, the ambient field is 437 mG, and the shield material is treated with linear relative permeability $p(O_1,\ldots,O_N)$0, with $p(O_1,\ldots,O_N)$1 used for Cryoperm10. The governing equation is written in terms of the vector potential $p(O_1,\ldots,O_N)$2,

$p(O_1,\ldots,O_N)$3

with the overlap represented numerically as an air layer of thickness $p(O_1,\ldots,O_N)$4 between overlapping metal surfaces. The comparison metric is the transverse field magnitude along the radial line from the joint into the shield interior.

The measurements were performed on a Fermilab ILC cryomodule magnetic shield assembly at room temperature. The overlap is clamped by screws spaced every 125 mm along the joint, and the measurements are taken midway between screws, explicitly to probe the worst gap region. The measured quantity is the transverse internal field $p(O_1,\ldots,O_N)$5 as a function of radial distance from the inner joint region. The gap in hardware is not directly measured; instead, the field measurements are used to back out an effective average gap.

The central result is explicit: “The average value of 0.1 mm is found to agree with measurements.” In the vertical 437 mG ambient field configuration, which is identified as the worst case scenario for leakage field at a mechanical joint, a model gap size of 0.1 mm with $p(O_1,\ldots,O_N)$6 reproduces the measured radial field profile. The paper states that varying the permeability of the shield material in the model does not have a large effect on the distance that field penetrates the joint, so the leakage profile is more sensitive to gap width than to the exact value of $p(O_1,\ldots,O_N)$7 in this regime. In the horizontal 437 mG ambient field configuration, leakage through the joint is very low, the internal field profile approximates that of a continuous cylindrical profile far from the joint, and the data are instead used to adjust $p(O_1,\ldots,O_N)$8, with about 24,000 giving the best fit.

The magnetic-circuit interpretation is that the overlap behaves like a continuous high-$p(O_1,\ldots,O_N)$9 path interrupted by a localized low-$c = |\langle\psi|\phi\rangle|^2$0 section. The gap reluctance is written as

$c = |\langle\psi|\phi\rangle|^2$1

so larger $c = |\langle\psi|\phi\rangle|^2$2 implies higher reluctance and stronger leakage into the interior. The engineering conclusion is equally explicit: it is reasonable to use an average gap width of 0.1 mm when modeling overlap joints in cylindrical shields. This 0.1 mm is an effective gap width parameter in the FEM model, not a direct mechanical measurement.

3. Reconstructing overlap in conformational ensembles

"Inference of joint conformational distributions from separately-acquired experimental measurements" (Hays et al., 2020) treats joint overlap measurements as a missing-data problem in biomolecular spectroscopy. DEER, smFRET, and related methods typically provide one pairwise distance distribution per experiment, so one obtains separately acquired marginal distributions

$c = |\langle\psi|\phi\rangle|^2$3

rather than the full joint distribution $c = |\langle\psi|\phi\rangle|^2$4. When these marginals are assumed independent,

$c = |\langle\psi|\phi\rangle|^2$5

the result is an artificial joint distribution that allows combinations of observables that never occur simultaneously in any real conformation.

The paper’s central object is the joint distribution over observables and its relation to the underlying conformational ensemble $c = |\langle\psi|\phi\rangle|^2$6. The induced observable-space distribution is written as

$c = |\langle\psi|\phi\rangle|^2$7

What is missing from separate experiments is the correlation structure: which distances co-occur in the same conformation, and which combinations are energetically inaccessible. The paper states that this is exactly the information that joint or overlap measurements would provide if multiple distances could be measured simultaneously on the same molecule.

The proposed method, Ensemble Estimation from Separate Measurements (EESM), reconstructs the lost joint information from separate marginals using biased MD and Jarzynski’s equality,

$c = |\langle\psi|\phi\rangle|^2$8

The algorithm begins with a reference ensemble, draws target observable values from the experimental marginals, applies a linear biasing potential to drive conformations toward those values, computes the non-equilibrium work, and uses the work distribution to estimate the conditional equilibrium probabilities and, by aggregation, the joint probabilities. Physically incompatible combinations of observables correspond to large $c = |\langle\psi|\phi\rangle|^2$9 and therefore small probability, even if they are allowed by the product of marginals.

The method is demonstrated first on a toy alternating-access transporter. In that system, the true joint distribution $\tau$0 is known, while the independence approximation $\tau$1 allows many combinations that are never realized. The authors run 500 aggregate iterations of EESM and quantify convergence using Jensen–Shannon divergence. EESM reconstructs the correct correlation structure between $\tau$2 and $\tau$3, thereby recovering what a true 2D distance-overlap measurement would have shown.

The second application concerns syntaxin-1a, using three published DEER measurements. Here only separate marginals $\tau$4 are available. EESM uses an aggregate of 1.3 $\tau$5s of MD simulations and converges to a final joint distribution over ten iterations. The inferred joint $\tau$6 differs substantially from the naive product $\tau$7, and one conformation that is highly probable under the product-of-marginals model is strongly down-weighted. Structural analysis identifies that state as biochemically implausible.

The paper’s limitations are equally explicit. EESM relies on MD and force fields, Jarzynski-based free-energy estimates require many trajectories for convergence, the joint space grows combinatorially with the number of observables, and label placement strongly influences informativeness. Nevertheless, the method restores the true pattern of overlap by suppressing spurious combinations that arise from assuming independent marginals.

4. Quantum joint measurements and state-overlap estimation

Two quantum-information papers address joint overlap measurements at different levels of abstraction (Carmeli et al., 2011, Fanizza et al., 2019). The first studies joint measurements of two conjugate observables on a finite quantum system; the second studies direct estimation of the overlap between two unknown pure states.

In "Informationally complete joint measurements on finite quantum systems" (Carmeli et al., 2011), the setting is a $\tau$8-dimensional Hilbert space with two mutually unbiased bases $\tau$9 and $p(O_i)$0, giving sharp observables

$p(O_i)$1

Because sharp conjugate observables are not jointly measurable, the paper introduces noisy marginals

$p(O_i)$2

and shows that their joint observables can be taken to be Weyl-covariant phase space observables

$p(O_i)$3

If minimal noise is required, the joint observable is unique. The parity effect is central: if $p(O_i)$4 is odd, the minimally noisy joint observable is informationally complete; if $p(O_i)$5 is even, it is not informationally complete, and one has to allow more noise in order to obtain informational completeness. The paper further shows that a joint observable can be implemented as a sequential measurement, for instance by first measuring $p(O_i)$6 via the Lüders instrument and then measuring $p(O_i)$7 sharply. In this setting, “joint overlap measurements” are not named as such, but the core idea is to obtain one joint POVM from which all quantum states, and therefore overlap-type quantities such as fidelities, can be reconstructed.

"Beyond the swap test: optimal estimation of quantum state overlap" (Fanizza et al., 2019) addresses the scalar overlap parameter

$p(O_i)$8

for two unknown pure states given $p(O_i)$9 copies of $p(O_1,\ldots,O_N)$0 and $p(O_1,\ldots,O_N)$1 copies of $p(O_1,\ldots,O_N)$2. The paper analyzes both the global Bayesian regime and the local pointwise regime. Under a Haar prior, the induced prior on $p(O_1,\ldots,O_N)$3 is

$p(O_1,\ldots,O_N)$4

The standard primitive is the swap test, a joint measurement on one copy of each type whose symmetric outcome probability is $p(O_1,\ldots,O_N)$5. For $p(O_1,\ldots,O_N)$6 repeated swap tests with $p(O_1,\ldots,O_N)$7, the local MSE is

$p(O_1,\ldots,O_N)$8

The paper shows that a more precise estimate is obtained by allowing general collective measurements on all copies. The optimal measurement is the projective measurement $p(O_1,\ldots,O_N)$9 arising from the Schur–Weyl decomposition, so weak Schur sampling suffices. In the asymptotic local regime, the optimal estimator achieves

$C$0

This is especially important for small overlaps, because the swap test is extremely inefficient for small values of the overlap, which become exponentially more likely as the dimension increases. The paper also studies estimate-and-project and estimate-and-estimate strategies, both of which are suboptimal but outperform the swap test. It further shows that the optimal measurement is less invasive than the swap test and derives, for depolarizing noise on qubit states,

$C$1

Taken together, these two works formalize two different senses of quantum joint overlap measurement: a joint POVM whose statistics are informationally complete for state reconstruction, and a collective measurement optimized specifically for the overlap parameter $C$2.

5. Overlap estimation in geometric and genomic data

"A Unified BEV Model for Joint Learning of 3D Local Features and Overlap Estimation" (Li et al., 2023) defines overlap for partially overlapping LiDAR point clouds $C$3 and $C$4 after discretization into a BEV grid of pillars. A cell of $C$5’s BEV and a cell of $C$6’s BEV are considered overlapping if their 3D spatial regions intersect given the ground-truth relative pose. Training labels $C$7 mark the true overlapping region, and the network predicts overlap score maps $C$8. The global similarity score used for loop closure is

$C$9

where $c = |\langle\psi|\phi\rangle|^2$0 and $c = |\langle\psi|\phi\rangle|^2$1 are the numbers of occupied BEV cells.

The architecture uses a 2D sparse UNet-like BEV backbone, with local feature description, 2D keypoint detection, height regression, and overlap estimation trained jointly. Overlap estimation is performed on the deepest feature maps using bilateral cross-attention,

$c = |\langle\psi|\phi\rangle|^2$2

with the symmetric expression for $c = |\langle\psi|\phi\rangle|^2$3, followed by a classification head that predicts $c = |\langle\psi|\phi\rangle|^2$4. The overlap losses are binary cross-entropy $c = |\langle\psi|\phi\rangle|^2$5 and an additional circle-loss term $c = |\langle\psi|\phi\rangle|^2$6 on the deepest feature map. Overlap classification is evaluated with IOU, Precision, and Recall. On KITTI, the reported mean IOU is 78.6%, mean Precision is 90.0%, and mean Recall is 84.7%. For loop closure, the reported average Recall@1 is 96.7%. The overlap estimate is not only descriptive; it is used to mask BEV cells so that corresponding keypoints are searched only within the predicted overlapping area. The registration recall improvements reported for overlap masking are from 33.0% to 67.9% at 50 m on KITTI and from 47.4% to 76.7% at 80 m on Apollo.

In genomics, "Joint Overlap Analysis of Multiple Genomic Interval Sets" (Otlu et al., 2018) defines overlap on genomic intervals. For half-open intervals $c = |\langle\psi|\phi\rangle|^2$7 and $c = |\langle\psi|\phi\rangle|^2$8, overlap is given by

$c = |\langle\psi|\phi\rangle|^2$9

or equivalently $\tau$0 and $\tau$1. The tool Joint Overlap Analysis (JOA) takes $\tau$2 interval sets and finds overlapping intervals with no constraints on the given intervals. It introduces segment tree (ST) and indexed segment tree forest (ISTF) based solutions for the intersection of multiple genomic interval sets in parallel, implemented in Java using the fork/join framework.

The segment tree backend stores intervals from multiple sets in a single structure, while the indexed segment tree forest is a novel composite data structure that leverages indexing and natural binning of a segment tree. The query index is defined by

$\tau$3

allowing direct access to local trees and forward/backward search through linked nodes. JOA is intended for regions overlapped by all sets, by at least $\tau$4 sets, or by user-defined subsets. The paper reports comparisons between JOA ST and JOA ISTF and states that they are comparable with each other in terms of execution time and memory usage. As a concrete application, JOA was applied to 141 ENCODE DNase hypersensitive site datasets to compute all jointly overlapping intervals.

These two papers illustrate two data-centric meanings of joint overlap measurement: explicit prediction of overlapping support in a learned representation, and exact coordinate-based multiway intersection over large heterogeneous interval collections.

6. Shared methodological themes, misconceptions, and limitations

A recurring theme is that overlap is often not directly measured. In magnetic shielding, the 0.1 mm value is not a direct mechanical measurement but an effective gap width parameter inferred from agreement between FEMM and measured field profiles (Crawford, 2015). In conformational spectroscopy, the joint distribution $\tau$5 is not observed; it is reconstructed from separate marginals by combining biased MD with non-equilibrium work estimates (Hays et al., 2020). In finite quantum systems, joint measurability requires deliberately noisy marginals, and informational completeness depends on dimension parity (Carmeli et al., 2011). In quantum overlap estimation, the overlap $\tau$6 is inferred from collective measurement statistics rather than accessed directly by repeated pairwise swap tests (Fanizza et al., 2019). In BEV registration, overlap labels depend on the ground-truth relative pose used during training (Li et al., 2023). In genomic analysis, joint overlap is determined by coordinate intersection and the chosen multiway query condition (Otlu et al., 2018).

A common misconception is that pairwise or marginal information is sufficient. The biomolecular results directly show that multiplying marginals invents spurious overlap between conformational states (Hays et al., 2020). The quantum overlap-estimation results show that repeated swap tests discard useful collective information and are especially inefficient for small overlaps (Fanizza et al., 2019). The LiDAR study shows that feature matching without explicit overlap masking degrades severely in low-overlap conditions (Li et al., 2023). These examples support the broader view that joint overlap is not, in general, recoverable from independent local measurements without either an explicit joint sensor model or an additional structural prior.

The limitations are domain-specific. The magnetic-shield model is 2D, infinitely long, magnetostatic, and excludes end effects; it assumes linear material properties and focuses on low-field conditions (Crawford, 2015). EESM depends on the accuracy of MD force fields, the convergence of Jarzynski-based estimates, and the tractability of a rapidly expanding joint space (Hays et al., 2020). In the finite-quantum-system setting, the minimally noisy joint observable is not informationally complete in even dimension (Carmeli et al., 2011). In direct quantum overlap estimation, the optimal collective measurement is implemented through the Schur transform, which is polynomial but still demanding for large copy numbers (Fanizza et al., 2019). The BEV model assumes outdoor urban LiDAR, a BEV grid, and reasonably flat ground, and still relies on RANSAC post-processing (Li et al., 2023). JOA can require substantial RAM for very large analyses, and ISTF performance depends on the choice of presetValue (Otlu et al., 2018).

Taken together, these studies suggest that joint overlap measurements are best understood not as a single protocol but as a class of coupled measurement-model procedures for recovering co-occurrence structure that is hidden by marginalization, low overlap, or local sensing. The central technical issue is always the same: whether the representation of overlap retains the dependencies that matter for the underlying physical, biological, quantum, geometric, or genomic system.