Weighted Ubiquitous Systems

• Weighted ubiquitous systems are domain-dependent frameworks that assign explicit quantitative weights to data, devices, and links for prioritized fusion and control.
• They are applied in sensor networks, reconfigurable infrastructures, and metric Diophantine approximation to manage context, energy, and anisotropic measurements.
• In network dynamics, these systems reveal invariant synchrony patterns in weighted coupled cell networks, linking matrix balance to robust dynamical behavior.

Searching arXiv for the cited papers and topic to ground the article in current arXiv records. Weighted ubiquitous systems denote several related but technically distinct constructions. In sensor-network-based ubiquitous computing, they are environments that compute, deliver, and act on context by explicitly quantifying the quality, relevance, and reliability of information through weights at different fusion levels (Borges et al., 2013). In reconfigurable ubiquitous computing infrastructure, weighting assigns quantitative importance to devices, components, and links according to device abilities, energy consumption, task priorities, and network capacity so that monitoring, detection, and reconfiguration decisions can be prioritized (Sellami et al., 2018). In metric Diophantine approximation, weighted ubiquitous systems form a framework for local ubiquity for rectangles, resonant sets, and weighted mass transference, yielding ambient measure laws and Hausdorff measure and dimension results for limsup sets in real, $p$-adic, complex, quaternionic, and formal-power-series settings (Robert et al., 2023). In weighted network dynamics, they are naturally modeled as weighted coupled cell networks whose adjacency and Laplacian structure determine synchrony and anti-synchrony subspaces for admissible input-additive systems (Aguiar et al., 2020).

1. Terminology and domain-dependent meaning

In the ubiquitous-computing literature following Weiser’s vision, ubiquitous systems are environments saturated with devices having computing and communication capabilities, gracefully integrated with human users. Wireless Sensor Networks are treated as the enabling substrate, with recurring constraints including energy, bandwidth, heterogeneity, and intermittency, and with desired properties including QoS, fault tolerance, and scalability (Borges et al., 2013). Within that setting, context-awareness means “Any information that characterizes the situation of an entity,” and a system is context-aware if it uses context to provide relevant information and services based on the user’s task. The cited survey categorizes context into computing, physical, time, and user context, and it introduces Quality of Context dimensions including precision, freshness, trustworthiness, completeness, significance, and validity (Borges et al., 2013).

In the load-balancing literature, the same ubiquitous-computing environment is described as a heterogeneous, mobile, resource-constrained collection of smartphones, sensors, actuators, and hosts cooperating to make services available “anytime and anywhere.” Here, “weighted” means assigning quantitative importance to devices, components, and links according to their abilities and state so that monitoring, detection, and reconfiguration decisions can be prioritized (Sellami et al., 2018).

In metric number theory, the phrase has a formal meaning unrelated to middleware or device management. Weighted ubiquitous systems are local ubiquitous systems for rectangles in products of bounded locally compact metric spaces with Ahlfors regular measures, indexed resonant sets, and coordinate-wise approximation radii (Robert et al., 2023). In network dynamics, by contrast, weighted ubiquitous systems are naturally interpreted as weighted coupled cell networks with real-valued edge weights and additive input structure (Aguiar et al., 2020).

A common source of confusion is that the same phrase names materially different objects across the cited literature. This suggests that “weighted ubiquitous systems” is best treated as a domain-dependent term rather than as a single standardized formalism.

2. Context information fusion in sensor-network ubiquitous systems

The survey of context information fusion treats weighting as intrinsic to fusion at the data or measurement, feature, and decision levels (Borges et al., 2013). At the measurement level, weights quantify sensor reliability, noise variance, temporal freshness, or spatial relevance. At the feature level, they scale features by discriminative power, modality confidence, QoC, or learned importance. At the decision level, they express belief, confidence, or trust in hypotheses or outputs from disparate classifiers and sources. The same survey classifies fusion by relationship as complementary, redundant, or cooperative; by abstraction as low, medium, high, or multilevel; and by DFD I/O mapping as DAI-DAO, DAI-FEO, FEI-FEO, FEI-DEO, or DEI-DEO.

A central linear model is weighted averaging:

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$

For inverse-variance weighting,

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$

The same synthesis also gives a confidence- and freshness-aware composite form,

$$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$

where trustworthiness and freshness are QoC parameters. For streaming context, temporal decay is expressed as

$$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$

The survey places these weights within several architecture families. Centralized brokered architectures, including CoBrA, SOCAM, MoCA, and Feel@Home, compute or update weights in brokers or interpreters and embed them in ontology or key-value metadata as QoC properties or confidence scores. Distributed architectures include flat or gossip-based dissemination, in which nodes locally weight neighbor messages by context-aware criteria before probabilistic forwarding, and selection or flooding backbones, in which subscription-driven overlays can carry weighted subscriptions or weighted context. Hierarchical architectures place weighting at gateways, which aggregate and weight inputs from edge nodes and perform higher-level fusion, for example Dempster–Shafer fusion using weights that arrive from local estimators. Activity- and role-based models such as Omnibus and Object-Oriented make weighting integral to Orient, Decide, Perceiver, and Director. Information-based models such as JDL and DFD locate weighting across source selection, object refinement, situation refinement, threat refinement, and process refinement.

The same literature reviews multiple representation layers for weights. Key-value, XML, and object-oriented models represent weights as explicit fields such as confidence or QoC.precision. Logic-based models allow predicates to carry certainty factors. Ontology-based systems in OWL, again including CoBrA, SOCAM, MoCA, and Feel@Home, model weights as datatype properties such as hasConfidence, hasTrust, and hasFreshness; ontological reasoning remains crisp, so hybrid systems attach probabilistic or fuzzy modules for numeric weighting or carry QoC metadata alongside semantic assertions. Probabilistic graphical models and fuzzy systems make weights native as probabilities, covariances, or membership degrees.

The estimation methods surveyed are correspondingly heterogeneous. Weighted least squares is written as

$$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$

Bayesian inference uses

$$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$

with reliability-weighted likelihoods and priors that encode preferences or policies. Kalman filtering uses

$K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$

$x_k = x_k^{-} + K_k (z_k - H x_k^{-}),$

$P_k = (I - K_k H) P_k^{-},$

where $$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$0 and $$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$1 act as implicit weights on measurement trust and adaptation. The survey also identifies particle filters, Dempster–Shafer theory, fuzzy logic and possibility theory, entropy- and information-theoretic weighting, and temporal freshness weighting as recurring techniques.

Application scenarios in the survey include location-aware services, activity recognition from wearables, environmental monitoring and smart spaces, and user preference and personalization. In all four, weighting is tied to QoC, uncertainty, trust, or significance. The survey’s design guidance places lightweight weight computation at the edge, aggregation and semantic policy handling at gateways, and heavier learning or recalibration in the cloud. It also emphasizes trade-offs among accuracy, robustness, latency, energy consumption, bandwidth, scalability, and privacy, and it notes that the survey does not report unified benchmarks across systems (Borges et al., 2013).

3. Energy-aware reconfiguration and load balancing

The reconfigurable infrastructure model is organized around clusters, nodes, a cluster head, a Controller, a DetectionAgent, and a per-node Knowledge Base (Sellami et al., 2018). Nodes know their neighbors. The cluster head hosts a Controller responsible for analysis, detection of high energy consumption, and reconfiguration. A DetectionAgent is dispatched by the Controller to each node when it joins a cluster; it monitors local metrics, compares them to the node’s Knowledge Base of historical “normal” behavior, and reports anomalies. Reconfiguration proceeds by sharing processing across multiple devices in the same cluster and by adapting applications through component redistribution.

The paper explicitly uses device abilities and energy consumption to trigger reconfiguration. The detailed synthesis then gives consistent formulations that make that approach precise. Instantaneous power is represented by

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$2

and cumulative energy by

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$3

A scalar weighted load is

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$4

and a device-placement weight is

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$5

with normalized proportional dispatch

$$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$6

These expressions make explicit how abilities, remaining battery, burn rate, and network constraints can be combined in a single control signal.

The same synthesis formalizes the component model and migration economics. For component $$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$7, the resource demand vector is $$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$8, with priority $$\hat{x} = \sum_{i=1}^{N} w_i x_i,\quad \sum_{i=1}^{N} w_i = 1,\ w_i \ge 0.$$9, state size $$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$0, and latency sensitivity $$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$1. Projected incremental power is

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$2

and projected migration time over link $$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$3 is

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$4

A canonical online objective is weighted load balancing,

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$5

and the trigger-action style decision rule is expressed through

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$6

subject to capacity and policy constraints.

Detection of high-energy consumption is baseline-driven. The DetectionAgent compares current behavior to the Knowledge Base through absolute thresholds, relative deviation from baseline, service-level anomalies, or multi-sensor rules using battery, discharge rate, congestion, and weighted load. To avoid oscillation, the synthesis introduces hysteresis, hold times, and cool-down periods. The operational workflow is monitor, detect, plan, act, and verify. The planning stage ranks components by a priority score, filters feasible targets, computes migration scores, and applies the best positive-gain action.

The reported evaluation uses GTNetS, wireless nodes, and services with limited capabilities including Print, View, Send e-mail, Update DB, and Scanner (Sellami et al., 2018). The experiment artificially overloads services to test detection and correction. Table 3 indicates that all overloaded services were detected in the tested cases, and Figure 1 shows that most overloads were corrected after reconfiguration. At the same time, the paper does not provide numerical energy savings, response time, or overhead figures, and it does not quantify a baseline such as static placement or unweighted selection. The stated limitations include algorithmic detail, migration overhead, network contention, controller availability, and the absence of discussion of security and trust.

4. Weighted ubiquitous systems in metric Diophantine approximation

In metric number theory, weighted ubiquitous systems are defined on a product of bounded locally compact metric spaces with Ahlfors regular measures, a countably infinite index set, resonant sets, and rectangular neighbourhoods (Robert et al., 2023). The limsup set generated by weighted rectangles is

$$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$7

The framework requires a $$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$8-scaling property for the coordinate-wise resonant sets and a local ubiquitous system for rectangles with respect to $$w_i = \frac{\sigma_i^{-2}}{\sum_{j=1}^{N} \sigma_j^{-2}}.$$9. In this setting, the paper recalls the ambient product space, Ahlfors regularity, resonant sets, layers $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$0, and rectangular neighbourhoods $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$1.

Two general results organize the theory. Theorem 5, attributed to Kleinbock–Wang, gives an ambient measure law in the divergence case for local ubiquitous systems for rectangles under monotonicity, c-regularity, $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$2-scaling, and Ahlfors regularity assumptions. Theorem 6, attributed to Wang–Wu, is the weighted mass transference principle: it passes from the measure-theoretic divergence law to Hausdorff dimension lower bounds for weighted rectangular targets. The paper explicitly states that these tools are used to prove ambient measure laws and Hausdorff measure and dimension results for limsup sets defined by weighted rectangular neighbourhoods of resonant sets across real, $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$3-adic, complex, quaternionic, and formal-power-series settings, and for sequences with prescribed discrepancy (Robert et al., 2023).

The real case is the weighted Diophantine approximation of systems of linear forms. For an $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$4-tuple $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$5,

$$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$6

Theorem 1 gives the weighted Khintchine–Groshev theorem:

$$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$7

and

$$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$8

Theorem 2 gives a weighted Jarník–Besicovitch-type dimension formula for power-law approximating functions.

The paper then develops analogues in several fields. Theorems 7 and 8 treat the $$w_i = \frac{t_i^\alpha\, f_i^\beta\, \sigma_i^{-2}}{\sum_j t_j^\alpha\, f_j^\beta\, \sigma_j^{-2}},$$9-adic setting with Haar measure on $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$0. Theorems 9 and 10 treat the complex case. Theorems 11 and 12 treat quaternions. Theorems 13 and 14 treat formal power series. Theorem 15 gives a $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$1–$$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$2 dichotomy for uniformly distributed sequences under discrepancy assumptions. Theorems 16, 17, and 19 establish divergence and Hausdorff-dimension results under discrepancy restrictions.

The paper also contrasts classical and weighted ubiquity. Classical ubiquitous systems for balls rely on a single-radius function and isotropic coverings. Weighted systems for rectangles use vector-valued radii $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$3, rectangular neighbourhoods, and anisotropic approximation rates. This change modifies the ubiquity function, covering scales, summation criteria, and dimension outcomes. The worked example $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$4 illustrates the mechanism: Proposition 1 constructs a ubiquitous system of rectangles, Corollary 1 gives the measure dichotomy, and Corollary 2 gives the dimension formula for $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$5 when $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$6 (Robert et al., 2023).

5. Weighted coupled cell networks, synchrony, and anti-synchrony

Aguiar and Dias study weighted networks through a dynamical-systems formalism in which a weighted coupled cell network on $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$7 cells is encoded by a real adjacency matrix $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$8, where $$w_i(t) = \frac{e^{-\lambda \Delta t_i}}{\sum_j e^{-\lambda \Delta t_j}}.$$9 is the weight of the directed edge from cell $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$0 to cell $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$1 (Aguiar et al., 2020). The weighted in-degree of cell $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$2 is

$$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$3

with diagonal degree matrix $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$4 and Laplacian

$$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$5

For weighted networks it is natural for admissible coupled cell systems to have additive input structure, and a canonical input-additive admissible system is

$$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$6

Special cases include adjacency-coupled linear input and difference coupling leading to Laplacian coupling.

The invariant geometric objects are generalized polydiagonal subspaces characterized by conditions of the forms $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$7, $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$8, and $$J(\theta) = \sum_{i=1}^{N} w_i\, \|y_i - f_{\theta}(x_i)\|^2,\qquad \theta = (X^\top W X)^{-1} X^\top W y.$$9. Synchrony corresponds to equalities inside clusters, anti-synchrony to sign flips, and zero clusters to pinned coordinates. For a partition $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$0, the synchrony criterion is stated in terms of cluster row-sums

$$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$1

The synchrony subspace is $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$2-invariant if and only if, for every cluster $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$3, for all $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$4 and every cluster $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$5,

$$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$6

This is the weighted equitable partition condition. The synchrony subspace is $$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$7-invariant if and only if the same off-block regularity holds and the degrees are constant within each cluster:

$$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$8

This is the exo-balanced condition.

For anti-synchrony, the paper introduces signed cluster row-sums

$$p(\theta\mid D) = \frac{p(D\mid\theta)\, p(\theta)}{p(D)},$$9

The anti-synchrony subspace is $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$0-invariant if and only if, for every cluster $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$1, for all $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$2 and every cluster $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$3,

$K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$4

together with zero-cluster conditions

$K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$5

for each $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$6 in a zero cluster and every nonzero cluster. The corresponding $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$7-invariance criterion requires the signed equitable conditions off-diagonal together with equal degrees within each nonzero cluster.

The significance of these matrix invariance criteria is dynamical. If a generalized polydiagonal is $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$8-invariant, it is flow-invariant for all even-odd-input-additive systems. If it is $K_k = P_k^{-} H^\top (H P_k^{-} H^\top + R)^{-1},$9-invariant, it is flow-invariant for exo-, odd-, and linear-input-additive systems (Aguiar et al., 2020). The paper therefore links weighted network structure directly to robust dynamical patterns, quotient networks, model reduction, and symbolic reduced equations. It also distinguishes adjacency-based coupling from Laplacian or diffusive coupling: adjacency invariance requires stricter weighted balance, whereas Laplacian coupling is more permissive because $x_k = x_k^{-} + K_k (z_k - H x_k^{-}),$0 has zero row sums and full synchrony is always $x_k = x_k^{-} + K_k (z_k - H x_k^{-}),$1-invariant.

6. Comparative perspective, limitations, and open directions

Across these literatures, weighting serves a common technical role: it converts heterogeneity into explicit coefficients that affect inference, control, approximation, or dynamics. In sensor-network ubiquitous systems, weights attach to data, features, and decisions through QoC, uncertainty, trustworthiness, and significance (Borges et al., 2013). In reconfigurable infrastructure, weights prioritize devices and migration targets by abilities, battery state, burn rate, and network constraints (Sellami et al., 2018). In metric number theory, weights enter through coordinate-wise radii, anisotropic approximation scales, and weighted rectangular neighbourhoods of resonant sets (Robert et al., 2023). In weighted network dynamics, they appear as edge strengths whose block row sums and degree structure determine synchrony and anti-synchrony subspaces (Aguiar et al., 2020).

The cited works also expose distinct limitations. The sensor-fusion survey states that dynamic context-aware weighting is underdeveloped, that few systems close the loop through JDL Level 4 to adapt weights online based on performance and changing QoC, that learning-based fusion and probabilistic context modeling remain fragmented across middleware, and that privacy-preserving weighting and cross-layer optimization need stronger mechanisms; it also notes clock synchronization challenges for filters in WSNs and the absence of unified benchmarks (Borges et al., 2013). The reconfiguration paper reports qualitative success in detection and correction but does not specify exact thresholds or optimization criteria in the original paper and does not provide numerical energy or latency results or comparisons against baselines; it further leaves migration overhead, controller availability, and security and trust unresolved (Sellami et al., 2018).

The number-theoretic framework is powerful but specialized: it depends on verifying local ubiquity for rectangles, $x_k = x_k^{-} + K_k (z_k - H x_k^{-}),$2-scaling, regularity assumptions, and appropriate discrepancy or counting estimates in each ambient field (Robert et al., 2023). The dynamical-systems framework is structurally exact but class-dependent: different admissible input-additive classes require different invariance criteria, and the distinction between adjacency and Laplacian coupling has direct consequences for which synchronized or anti-synchronized manifolds persist (Aguiar et al., 2020).

A plausible implication is that the phrase “weighted ubiquitous systems” is less a single theory than a recurring methodological pattern. In all four settings, weights make heterogeneous structure computable: they determine how information is fused, how components are migrated, how anisotropic limsup sets are measured, or how invariant manifolds emerge from a weighted graph.