When a photocatalyst or persulfate activator treats water that holds several pollutants, it produces one pool of short-lived oxidants. These include hydroxyl radicals, superoxide, singlet oxygen, sulfate radicals and surface holes, and every pollutant in the water competes for that same pool. Studies almost never describe this competition. They test one pollutant at a time and report a single removal percentage. They also infer which oxidant did the work with "scavenger" experiments, in which a chemical that is supposed to destroy one oxidant is added and the drop in degradation is measured. The scavengers are known to be cross-reactive, so the resulting attributions are often disputed. We propose to borrow a single object from network information theory, the capacity region, and apply it to mixtures. In information theory, a capacity region is the set of rate combinations a single transmitter can deliver to several receivers at once. In our setting, the "transmitter" is the catalyst, the "receivers" are the pollutants, and the region is the set of removal combinations reachable under a fixed photon or oxidant budget.
The idea works in two directions, which is why it is worth doing. Going forward, an inner bound built from measured operating points and an outer bound built from species attributions together summarize what a catalyst can do on a mixture. Going backward, the outer bound gives a falsification test for reactive-species identification. If any measured removal combination lies outside the outer bound computed from the scavenger-derived species contributions, then those contributions are wrong. This turns the ambiguity of scavenger experiments into something an experiment can check.
Core Claim
The analogy we use is deliberately limited. We do not claim that radicals carry messages, and we use no mutual information. What we borrow from broadcast theory is its geometry and its computational habits. A multi-receiver system is described by a region rather than a number. The region is bracketed from inside by achievable schemes and from outside by conservation-type limits. Its boundary is traced by maximizing weighted sums. Mixing two strategies over time ("time-sharing") makes any achievable region convex.
The chemical version has three testable parts. The first concerns the outer bound. Let $y_j$ be the yield of oxidant species $j$ per unit budget $B$ (photons absorbed or oxidant dosed), and let $n_{ij}$ be the moles of pollutant $i$ destroyed through species $j$, which scavenger experiments estimate. Conservation then requires
$$\sum_i s_{ij}\, n_{ij} \le y_j B - \ell_j, \qquad n_i=\sum_j n_{ij},$$
where $s_{ij}$ is the number of species-$j$ equivalents consumed per pollutant molecule and $\ell_j$ is the loss to background reactants such as chloride. Projecting this set onto the removal vector $(n_1,n_2,\dots)$ gives the outer bound.
The second part is the inner bound. It is the convex hull of removal vectors actually measured under static operating conditions, enlarged by staged operation, meaning switching pH, oxidant dose or a selective quencher partway through the run. We predict that static operation alone gives a non-convex set and that staging fills the hull.
The third part is the falsification test. A measured point outside the outer bound refutes the attribution $\{n_{ij}\}$ that was used to build it.
We also test one structural hypothesis borrowed from Superposition Coding is Almost Always Optimal for the Poisson Broadcast Channel. That paper proves that when one receiver is "effectively less noisy" than another, a simple layered scheme traces the whole boundary. Our chemical analogue is about reactivity profiles. If pollutant A reacts with every species that B reacts with, plus more, we predict that the boundary is traced by a single control knob, such as the oxidant-to-pollutant ratio. Where the profiles cross, we predict that multiple knobs and staging are needed.
Research Program
The work has three linked pieces: an attribution layer, a computation layer and a measurement layer.
The attribution layer estimates $n_{ij}$ for each pollutant alone. It combines scavenger panels with spin-trapping electron paramagnetic resonance (EPR), the spectroscopy that confirms which radicals are present by trapping them as stable adducts. It follows the protocols in the source studies: tert-butanol and isopropanol for hydroxyl radicals, p-benzoquinone for superoxide, methanol for sulfate plus hydroxyl, and potassium iodide or ammonium oxalate for holes (Keyi Zhou et al., Linfeng Xiao et al., Yan Zhang et al.). Each attribution carries an uncertainty interval, obtained by repeating the panels at two scavenger concentrations. This brackets the known non-selectivity problem without pretending to resolve it.
The computation layer traces both bounds by maximizing $\sum_i w_i n_i$ over a grid of weights $w$, the supporting-hyperplane method. For the outer bound this is a small linear program. For a kinetic inner-bound model, built from competition kinetics with measured rate constants, it is a non-convex problem. Here we reuse the structural lesson of Blahut-Arimoto Algorithms for Inner and Outer Bounds on Capacity Regions of Broadcast Channels. That paper splits each supporting hyperplane into nested max-min subproblems and reports large speedups over generic solvers. Its algorithms target discrete channels and cannot be applied directly to continuous kinetics, so we take the decomposition idea rather than the algorithm.
The measurement layer runs batch experiments on two-pollutant mixtures. It sweeps static conditions and staged protocols at a matched total budget, and repeats the sweep under a matrix stressor.
We chose two catalyst systems from the file because their reactivity profiles differ. The first is Bi@(BiO)₂CO₃/TiO₂ with fluoroquinolone antibiotics, where superoxide is reported as primary (The Bi-Modified (BiO)2CO3/TiO2 Heterojunction Enhances the Photocatalytic Degradation of Antibiotics). Its pollutants are chemically similar, so we expect nested profiles and a single-knob boundary. The second is KOH-modified activated coke activating persulfate (Yan Zhang et al.). It produces sulfate radicals, hydroxyl radicals and singlet oxygen together. We pair phenol with a dye such as Rhodamine B, chosen so the two pollutants' reactivity profiles cross and the staging hypothesis has room to show.
Chloride is the stressor. It is known to remove hydroxyl radicals and to push systems toward superoxide and singlet oxygen (Zhiwei Shang et al.). We predict that it shrinks the region asymmetrically, toward whichever pollutant relies less on hydroxyl radicals. This puts a number on the selectivity-versus-potency tension that the file identifies between non-radical and radical pathways (Juexiu Li et al.).
Baselines
The naive reference point is current practice. Each pollutant's single-compound removal is taken as its performance in the mixture, which implicitly predicts the corner point (r₁, r₂). For the Bi@(BiO)₂CO₃/TiO₂ system those single-compound figures are 93.2% for ciprofloxacin, 97.5% for lomefloxacin and 100% for norfloxacin after 60 minutes of simulated sunlight (Yue Gao et al.). For AC-K/persulfate the figure is 98.75% phenol removal (Yan Zhang et al.). The region framework beats this baseline if mixture measurements fall measurably short of that corner and our outer bound predicts the shortfall.
The strongest competitor is a static multi-objective Pareto front fitted directly to measured mixture data. This means running the same operating-condition sweep and keeping the non-dominated points, with no species attribution and no staging. It is what a careful process engineer would do. It gets the identical experimental budget: the same number of batch runs and the same condition grid. The region framework beats it in two ways. Staged protocols must produce points that lie strictly beyond the static front, outside its run-to-run error. And the attribution-based outer bound must predict where held-out operating points land more accurately than interpolating the front.
A pseudo-first-order competition-kinetics model is a second competitor. It fits one apparent rate constant per pollutant and splits the shared oxidant by rate-constant-weighted competition, and it gets the same fitting data.
Two ablations isolate the proposal's key ingredients. The first lumps all species into one oxidant, which tests whether species-resolved identification adds anything to the bound. The second removes staging, which tests whether time-sharing matters.
There is no reported result found on Lacuna for removal rate regions, Pareto fronts or competition-kinetics predictions on any two-pollutant mixture with these catalysts. All mixture numbers will therefore be new.
Why not just…?
Why not just fit a multi-objective Pareto front and skip the information-theory language? This is a real threat. If static operation already yields a convex front, and attribution adds no predictive power, then our proposal reduces to Pareto mapping and we will say so. Two things are not available from a Pareto front. The first is a certificate: an outer bound that no operating condition can exceed, which is what lets a measured violation falsify a species attribution. The second is the time-sharing prediction that staged protocols reach points a static sweep cannot. The baseline comparison and the staging ablation above decide both.
Why not just use pseudo-first-order kinetics with scavenger rate drops? This approach collapses coupled pathways into one apparent constant. The source studies show that pathways feed each other, for example superoxide acting as a precursor to other radicals (Keyi Zhou et al.), and that scavengers give partial, overlapping inhibitions (Linfeng Xiao et al.). We nonetheless run the kinetic model as a baseline. If it predicts held-out mixture points as well as the species-resolved bound does, then the extra structure is not earning its keep.
Why not just model radicals as a molecular communication channel? Existing molecular-communication work studies inert molecules diffusing to Poisson or ligand-receptor receivers (Capacity of Diffusion-based Molecular Communication with Ligand Receptors, Identification for Molecular Communication Based on Diffusion Channel with Poisson Reception Process). Nobody encodes data in hydroxyl radicals, and a capacity in bits would answer no question a water chemist asks. We keep the region geometry and drop the bits.
Why not just settle attribution with EPR? In these studies EPR confirms which radicals are present but does not apportion pollutant destruction among them (Yan Zhang et al., Zijun Dong et al.). We use EPR as a qualitative constraint on the attribution, and we use the outer-bound test as the quantitative check that EPR cannot supply.
Decisive Checks
Three results would decide the proposal. The first check is staging. In the crossed-profile system (AC-K/persulfate with phenol and dye), staged protocols must reach points strictly outside the static achievable set at matched budget. If they do not, the time-sharing claim fails.
The second check is the outer bound. It is computed from scavenger attributions at both scavenger concentrations. It must contain every measured point in the nested-profile antibiotic system. It must also be tight there, with the gap to the inner hull no larger than run-to-run variability. As a deliberate control, we use an attribution known to be biased, namely one taken from a single high scavenger dose. That attribution should produce violations, which shows the test can actually detect a wrong mechanism.
The third check is chloride. Adding chloride should shrink the region preferentially along the axis of the pollutant that depends more on hydroxyl radicals, matching the direction predicted from the attribution.
Success on the second check alone would already give reactive-species identification a falsifiable consistency test that the field currently lacks.
References
- Superposition Coding Is Almost Always Optimal for the Poisson Broadcast Channel
- Highly efficient photocatalytic performance of BiI/Bi2WO6 for degradation of tetracycline hydrochloride in an aqueous phase
- Facile Fabrication of F-Doped SnO2 Nanomaterials for Improved Photocatalytic Activity
- Activation of Persulfates Using Alkali-Modified Activated Coke to Promote Phenol Removal
- Blahut-Arimoto Algorithms for Inner and Outer Bounds on Capacity Regions of Broadcast Channels
- The Bi-Modified (BiO)2CO3/TiO2 Heterojunction Enhances the Photocatalytic Degradation of Antibiotics
- Treatment of cationic red X-GRL in high-salt printing and dyeing wastewater by an electrocatalytic ozonation system: treatment efficiency and degradation mechanism
- Peroxymonosulfate Activation by Facile Fabrication of α-MnO2 for Rhodamine B Degradation: Reaction Kinetics and Mechanism
- Capacity of diffusion-based molecular communication with ligand receptors
- Identification for Molecular Communication Based on Diffusion Channel with Poisson Reception Process
- Fe(II)-activated persulfate oxidation to degrade iopamidol in water: parameters optimization and degradation paths