Graph theory has long promised that a modest number of well-placed sensors can uniquely pinpoint any fault in a network. The theory of locating-total dominating sets makes this precise: in a "twin-free" graph (one where no two vertices share exactly the same neighbors), a set $S$ of at most $\frac{3}{4}n$ vertices can be chosen so that every vertex is watched, the watchers watch each other, and the pattern of which watchers see a vertex acts as a fingerprint that identifies it uniquely Locating-Total Dominating Sets in Twin-Free Graphs: a Conjecture. But this entire edifice rests on an assumption that a decade of wireless networking research has empirically demolished: that "seeing" is a clean, binary, purely local relationship. In real physical systems, signals attenuate continuously with distance and, crucially, accumulate — many weak, individually harmless contributions can collectively drown the very difference that was supposed to distinguish one vertex from another Effects of Interference on Wireless Mesh Networks: Pathologies and a Preliminary Solution.
This proposal asks what happens to locating-total domination when the binary model is replaced by a physical, additive one, and it answers the question on a benchmark where it matters biologically: the microvascular bed after stroke. There, in vivo mapping of more than 40,000 capillaries shows a 34.9% stall rate and a 30.6% decrease in total perfused capillary length, distributed very unevenly across flow-topology junction types In Vivo Network-Level Cerebrovascular Mapping Reveals the Impact of Flow Topology on Capillary Stalls After Stroke — and the stalling culprit is plausibly neutrophil extracellular traps (NETs), web-like DNA strands released by immune cells that lodge in microvessels and are cleared by the enzyme DNase I Stroke and myocardial infarction induce neutrophil extracellular trap release disrupting lymphoid organ structure and immunoglobulin secretion. Monitoring where stalls form, and degrading them without poisoning the surrounding tissue, is exactly a locating-and-scheduling problem under cumulative "interference." No existing work connects these threads: the interference literature stops at pairwise link scheduling, and the vascular literature stops at empirical topology.
Core Claim
The core claim is two-sided. On the theory side, we define an analog locating-total dominating set: sensors do not return binary neighbor sets but continuous readings that attenuate as $d^{-\alpha}$ and sum additively at every point, so a vertex $v$ is identified not by the set $N(v)\cap S$ but by a real-valued signature vector $\Sigma_S(v) = (P_u\, d(u,v)^{-\alpha})_{u\in S}$, and $v$ counts as distinguishable from $u$ only when $\|\Sigma_S(u)-\Sigma_S(v)\|$ exceeds a threshold $\beta$ set by measurement noise. We claim that under this model the classical bounds do not simply transfer: additive interference creates new shadow twins — vertices whose binary neighbor sets differ but whose analog signatures are indistinguishable at realistic noise levels — while power-law attenuation also resolves some binary twins, so the worst-case sensor fraction moves away from $\frac{3}{4}n$ in a direction that depends on $\beta$, $\alpha$, and the noise floor, and can be bounded. On the applied side, we claim that on a reconstructed cortical microvascular graph, this analog model measurably changes both how many measurement points are needed to uniquely localize NET-induced stalls and how a DNase-based degradation therapy should be scheduled: each enzyme release is a transmitter whose local benefit attenuates with distance while its toxic spillover (freed elastase and histones) accumulates across the whole bed, exactly the good-signal-versus-cumulative-interference structure of the wireless problem Protocol Design Beyond Graph-Based Models.
Research Program
The theoretical program has three parts. First, we formalize SINR-locating-total domination, replacing the binary distinguishability axiom $N(u)\cap S \ne N(v)\cap S$ of Locating-dominating sets in twin-free graphs with the analog condition above, and we pose the natural analogue of the two-thirds conjecture: does there exist a constant $c(\beta,\alpha)$, uniformly bounded over twin-free graphs, with $\gamma_t^{L,\mathrm{SINR}}(G)\le c\cdot n$? Our working hypothesis, derived from the structure of 2-coronas and the extremal families identified in Locating-Total Dominating Sets in Twin-Free Graphs: a Conjecture, is that $c$ degrades smoothly from the binary value as noise grows, with a phase transition when $\beta$ exceeds the typical gap between signature vectors of adjacent vertices. Second, we characterize shadow twins: pairs distinguishable in the binary model but not the analog one, and we show they are controlled by local geometry (near-symmetric neighborhoods), which suggests a strengthening of the Tutte–Berge matching decomposition used for cubic graphs Location-domination and matching in cubic graphs. Third, we build placement algorithms. The grid-shifting and buffer-zone machinery of Maximum Weighted Independent Set of Links under Physical Interference Model and Throughput Optimizing Localized Link Scheduling for Multihop Wireless Networks under Physical Interference Model was designed to buy spatial separation against aggregate interference; we adapt it to buy signature separation — guaranteeing that any two vertices in different cells receive signatures that differ by more than $\beta$ — and we aim for a constant-factor approximation to the minimum SINR-locating set, mirroring the $(K+1)^2\omega$ ratio proved for link selection in Maximum Weighted Independent Set of Links under Physical Interference Model.
The applied program maps this onto NET pathology. We treat the microvascular bed as a directed flow graph (divergent/convergent junction labels, as in the OMNIMap study), treat measurement points as analog sensors whose perfusion signals attenuate along vessels, and treat DNase delivery sites as transmitters whose cleavage signal is local while their released protease load is additive and global. The scheduling objective is to recover perfused capillary length while holding cumulative released-toxic load below the cytotoxicity class exemplified by the finding that NETs at $\ge 500$ ng/mL directly kill lung epithelial cells Necrostatin-1 Alleviates Diffuse Pulmonary Haemorrhage by Preventing the Release of NETs via Inhibiting NE/GSDMD Activation in Murine Lupus. This reframing is directly motivated by the tension in the literature: systemic DNase I rescues organ architecture Stroke and myocardial infarction induce neutrophil extracellular trap release disrupting lymphoid organ structure and immunoglobulin secretion, yet the same dissolution mobilizes the cytotoxic cargo the traps were carrying.
Proposed Experiments
The computational benchmark is the OMNIMap cortical vascular volume: a $1000\times1000\times360\ \mu\text{m}^3$ graph with more than 40,000 capillaries and per-junction flow-topology labels In Vivo Network-Level Cerebrovascular Mapping Reveals the Impact of Flow Topology on Capillary Stalls After Stroke, cross-checked for reconstruction fidelity against the lumen-staining CLARITY pipeline, which reports a 10-fold SNR gain and true-positive detection above 90% at 240 µm depth Whole-Brain Vasculature Reconstruction at the Single Capillary Level. Experiment one asks how many measurement points uniquely localize a simulated stall placed at a random vertex: we compare the binary locating-total dominating placement against the analog placement across a swept noise threshold $\beta$, with the headline metric being the fraction of stall positions uniquely identifiable per sensor spent, and the secondary metric being whether identification errors concentrate at the C-to-D junctions that stall most in vivo. Experiment two runs the degradation scheduler on the same graph with stalls seeded at the empirically observed 34.9% rate, biased toward C-to-D junctions: policies compared are a uniform systemic bolus, a greedy binary conflict-graph scheduler, and our SINR-scheduled staggered release with spatial buffer zones, each given an identical total enzyme budget, scored on recovered perfused length versus peak local toxic load. Experiment three is a calibration bridge to the biology: an in vitro A549 viability assay exposing epithelial cells to stepwise NET concentrations around the 500 ng/mL threshold reported in Necrostatin-1 Alleviates Diffuse Pulmonary Haemorrhage by Preventing the Release of NETs via Inhibiting NE/GSDMD Activation in Murine Lupus, used to fix the toxicity ceiling in the simulator rather than assumed. A fourth, external check reruns the analog placement machinery on the classic wireless setting — two-pair and three-pair capture scenarios on the mica2 testbed configuration, where SINR-aware scheduling cut transmission of 20,000 messages from ~600 seconds to 267 seconds Protocol Design Beyond Graph-Based Models — to confirm our signature-separation algorithm degrades gracefully to known results in the regime where ground truth exists.
Baselines
The "do nothing" reference point is uniform systemic DNase infusion, which is the de facto standard in the NET literature: it is the intervention shown to restore mucosal IgA and lymphoid architecture in the stroke/MI models of Stroke and myocardial infarction induce neutrophil extracellular trap release disrupting lymphoid organ structure and immunoglobulin secretion, and it requires no sensing at all. In the simulator it corresponds to activating every actuator simultaneously at the same rate; it is the strongest competitor on simplicity and clinical plausibility, and our method must beat it on the joint metric of perfused-length recovered per unit of peak toxic load, not on perfused length alone, since a bolus can clear aggressively while dumping cytotoxic cargo systemically. The strongest existing method for the sensing half is a classical greedy minimum locating-total dominating set computed on the binary vessel-adjacency graph, the direct algorithmic descendant of the bounds in Locating-dominating sets in twin-free graphs and Locating-Total Dominating Sets in Twin-Free Graphs: a Conjecture; it is the strongest competitor because it is the incumbent theory with proven worst-case guarantees of at most $\frac{3}{4}n$ sensors on twin-free graphs, and if the analog model offers no identification advantage on real vascular geometry, the binary method should win on cost. We run it on the identical graph, with the same candidate sensor positions, the same number of sensors, and the same compute budget (identical greedy evaluation counts), and we count our approach as beating it only if it maintains or improves the unique-localization rate under injected measurement noise while using no more sensors. For the therapy half, the strongest competitor is a greedy binary conflict-graph scheduler that activates degradation sites in mutually exclusive neighborhoods, in the spirit of the buffer-zone scheduler of Throughput Optimizing Localized Link Scheduling for Multihop Wireless Networks under Physical Interference Model; it is strongest because buffer zones are the established cure for cumulative-interference instability, and the same study reports its localized variant achieving roughly 50% of the centralized optimal traffic rate, so any scheduler must at minimum not fall below that kind of constant-factor behavior. The key ablation isolating our ingredient runs the full pipeline — analog signatures, additive interference, staggered release — but with the additive interference term switched off, reverting each sensor and actuator to a purely local binary radius; any performance gap between this ablation and the full model is attributable solely to the physical coupling, and if that gap is negligible on the vascular graph, the honest conclusion is that the analog machinery is unnecessary at this geometry.
Why not just…?
Why not just use binary conflict graphs or standard protocol models? This is the most serious objection because the binary model is simpler and the whole domination literature already lives there. But the wireless evidence is unusually direct: when Mahajan et al. tested binary conflict graphs on real mesh networks, they produced flow starvation affecting up to 50% of active flows and severe capacity misestimation Effects of Interference on Wireless Mesh Networks: Pathologies and a Preliminary Solution, and on physical hardware the SINR-aware MAC halved the transmission time of 20,000 messages relative to the graph-governed standard Protocol Design Beyond Graph-Based Models. The analogue is not rhetorical: many low-level DNase releases, each locally below a toxicity threshold, sum to a global load that harms tissue exactly as summed weak interferers corrupt reception. Our ablation (binary radius, interference off) is precisely the experiment that settles whether the vascular geometry is bad enough for this to matter, and we pre-commit to reporting it either way.
Why not just infuse DNase systemically and skip the graph theory? Systemic infusion demonstrably works in the mouse models Stroke and myocardial infarction induce neutrophil extracellular trap release disrupting lymphoid organ structure and immunoglobulin secretion, and no reported result found on Lacuna for any topologically-targeted DNase delivery strategy. The counter-evidence is also in the file: dissolution of the DNA scaffold releases its anchored elastase and histone cargo, and that cargo kills epithelial cells at $\ge 500$ ng/mL Necrostatin-1 Alleviates Diffuse Pulmonary Haemorrhage by Preventing the Release of NETs via Inhibiting NE/GSDMD Activation in Murine Lupus; meanwhile stalls are strongly non-uniform across junction types In Vivo Network-Level Cerebrovascular Mapping Reveals the Impact of Flow Topology on Capillary Stalls After Stroke, so a uniform dose over-treats resilient regions for no benefit. Experiment two is designed to answer this directly: if the bolus matches our scheduler on the joint recovery-per-toxic-load metric, the targeting hypothesis fails and we will say so.
Why not just use ordinary dominating sets or independent sets, without the locating and total constraints? A plain dominating set guarantees only that a stall is adjacent to some sensor, not which vertex failed; the distinction between "coverage" and "unique identification" is the entire content of the locating condition $N(u)\cap S\ne N(v)\cap S$ Locating-dominating sets in twin-free graphs. The total condition is not decoration either: sensors in inflamed tissue can themselves be degraded, and a total dominating set is one in which every sensor is watched by another sensor, a property unique to the total variant Locating-Total Dominating Sets in Twin-Free Graphs: a Conjecture. The honest cost is real — worst-case bounds rise from $\frac{1}{2}n$ in restricted classes Location-domination in line graphs toward $\frac{3}{4}n$ in general — so we will report, for the actual cortical graph, how much extra sensor budget the locating and total constraints cost over a plain dominating set; if the marginal cost is small in practice, the objection dissolves, and if it is large, that itself quantifies the price of fault-localizing monitoring in tissue.
Decisive Checks
Three outcomes would decide the proposal. If, on the reconstructed vascular graph, the analog signature model raises the unique-localization rate at fixed sensor budget under realistic noise while the binary-radius ablation does not, the physical model earns its place. If the SINR-scheduled degradation recovers a materially larger fraction of the 30.6% perfused-length loss per unit of peak toxic load than both the bolus and the binary scheduler, the scheduling theory earns its place. And if the theory yields a proven bound $\gamma_t^{L,\mathrm{SINR}}(G)\le c(\beta,\alpha)\,n$ with an explicit characterization of when shadow twins force $c$ above the binary constant, the result stands on its own as the first domination bound proved in a cumulative-interference model — a setting the combinatorial community has not entered and the wireless community has entered only for pairwise transmission, never for identification.
References
- Locating-Total Dominating Sets in Twin-Free Graphs: a Conjecture
- Effects of Interference on Wireless Mesh Networks: Pathologies and a Preliminary Solution
- In Vivo Network-Level Cerebrovascular Mapping Reveals the Impact of Flow Topology on Capillary Stalls After Stroke
- Stroke and myocardial infarction induce neutrophil extracellular trap release disrupting lymphoid organ structure and immunoglobulin secretion
- Protocol Design Beyond Graph-Based Models
- Locating-dominating sets in twin-free graphs
- Location-domination and matching in cubic graphs
- Maximum Weighted Independent Set of Links under Physical Interference Model
- Throughput Optimizing Localized Link Scheduling for Multihop Wireless Networks under Physical Interference Model
- Necrostatin-1 Alleviates Diffuse Pulmonary Haemorrhage by Preventing the Release of NETs via Inhibiting NE/GSDMD Activation in Murine Lupus
- Whole-Brain Vasculature Reconstruction at the Single Capillary Level
- Location-domination in line graphs