Benefits and yields of simultaneous free chlorine and hydrogen generation in side-emitting optical fiber photocatalytic reactors

Abstract

Decentralized water treatment requires resilient, consumable-free technologies that operate independently of complex supply chains. This study presents a side-emitting polymer optical fiber photocatalytic reactor immobilized with modified SrTiO₃ for the simultaneous generation of gaseous hydrogen fuel and disinfecting oxidants (free chlorine) in water. A three-factor response surface methodology was employed to quantify the interactive effects of the factors (salinity, catalyst loading, and light intensity) on the system responses. Statistical analysis identified two distinct limiting regimes: the aqueous free chlorine response is chemically limited by the salinity factor, allowing for precise control of disinfection capacity, whereas the hydrogen response is optically limited by the catalyst loading factor due to a “shielding effect” where excess particles block the light from the fiber core. Leveraging the independent relationship between these factors and responses, we established a tunable strategy with three distinct modes: a “Rapid Response” mode for emergency use that achieves critical disinfection levels (0.5 mg L⁻¹) in 10–20 minutes; a “Maintenance” mode that uses a continuous flow rate of 13 mL min⁻¹ to maintain a steady-state chlorine level within WHO safety standards (0.5–4.0 mg L⁻¹) to prevent microbial regrowth; and a “Fuel-Priority” mode that maximizes the hydrogen response (∼80 µmol h⁻¹). This work demonstrates a robust, dual-purpose infrastructure that transforms passive treatment units into active utility nodes which can provide both clean water and energy for off-grid applications.

---

1 Introduction

Access to safe drinking water remains one of the most persistent global challenges, particularly in decentralized, rural, and resource-limited areas where conventional water treatment infrastructure is usually unavailable or unreliable.^1–3 Chlorination is the most widely implemented disinfection strategy because aqueous free chlorine provides broad-spectrum pathogen inactivation, low cost, ease of dosing, and lasting efficacy to protect water during storage and distribution.^4–7 However, chlorine is introduced as pre-packaged tablets or liquid concentrates that must be replenished periodically.^8–10 This dependence on consumables, rather than on-site generation, creates supply and handling issues that become more acute when transportation or storage conditions are inconsistent. These limitations have motivated a shift toward in situ oxidant generation technologies that can produce disinfecting species directly from the available water matrix.

Electrochlorination is currently the dominant method for onsite chlorine production, converting chloride in saline or brackish water into free chlorine via anodic oxidation.^11–13 Although effective, electrochlorination systems inherently depend on electrodes, applied electrical current, electronic control modules, and materials resistant to passivation.¹⁴ These requirements create vulnerability to electrode instability, increase maintenance needs, and raise safety concerns associated with exposed electrical components and gas evolution.¹⁴ For off-grid communities, the necessity of electrical infrastructure presents a barrier to widespread utilization. As water treatment transitions toward smaller and more distributed architectures, there is a growing need for low-power, low-complexity, photon-driven systems capable of generating disinfecting oxidants without relying on electrochemical devices.

Instead, photocatalysis offers a promising alternative by using absorbed photons to directly generate redox equivalents capable of oxidizing chloride and reducing protons. Photocatalysts such as SrTiO₃ have been widely explored for overall water splitting, hydrogen evolution, and oxidative reactions due to their chemical stability, suitable band structure, and high quantum efficiency under UV excitation.^15–18 Literature shows that SrTiO₃ can remain stable in a range of aqueous environments, from deionized to saline and buffered systems, particularly near neutral to mildly alkaline pH.^19,20 In chloride-containing media, photogenerated holes can form aqueous and gaseous molecular chlorine (Cl₂), hypochlorous acid (HOCl), or hypochlorite (OCl⁻), depending on pH and reaction environment.²¹ Simultaneously, photogenerated electrons can reduce protons to molecular hydrogen gas (H₂),²¹ which preferentially evolves into the headspace, thereby achieving natural phase separation of the oxidant and fuel. This generated H₂ can serve as a renewable energy source to offset power consumption in water treatment processes, such as driving circulation pumps. This dual oxidative–reductive pathway opens the possibility of simultaneous disinfectant production and energy recovery within a single photocatalytic device.

Several challenges have limited photocatalytic chlorine generation for realistic water treatment. Most previous studies rely on slurry-based or film-based photocatalysts suspended in batch reactors, where photon utilization is inefficient due to light scattering, turbidity sensitivity, catalyst sedimentation/aggregation, and short light penetration depths.^22–24 Furthermore, no systematic evaluation exists on how salinity, catalyst amount, and light intensity jointly control chlorine formation, hydrogen production, and energy consumption in a photocatalytic system. These gaps create uncertainties about whether photocatalytic oxidant generation can be engineered to consistently achieve disinfection-relevant chlorine levels under variable operating conditions.

Polymer optical fibers (POFs) provide a unique platform to overcome these limitations of classical photocatalytic devices. When coated with photocatalysts, POFs act as both light-delivery channels and catalytic substrates, distributing photons over extended reactor volumes and reducing the optical losses that limit traditional slurry systems.^22,25–28 This architecture allows high catalyst loading over large surface areas, long optical path lengths, and tunable photon flux independent of water turbidity.²² POF-based photocatalytic systems have recently been explored for oxidant generation and pollutant degradation, demonstrating superior photon utilization efficiencies and compact form factors compared to conventional reactors.^22,29 However, the ability of the POF platform to drive chloride oxidation and hydrogen evolution simultaneously in saline water has not yet been investigated completely. Furthermore, the fundamental dependencies between operating factors and system outputs (e.g. H₂ evolution rate, aqueous free chlorine generation, and energy consumption) remain unresolved.

In this work, we develop a POF photocatalytic reactor that produces aqueous free chlorine and gaseous hydrogen using LED illumination as the energy input. We focus on three operational variables that govern photocatalytic performance under real-world conditions: (i) salinity, representing the availability of chloride and ionic strength effects; (ii) catalyst amount, corresponding to the accessible photocatalytic surface area across the POF bundle; and (iii) light intensity, which determines photon-limited or catalyst-limited reaction regimes. A three-factor experimental design is constructed to quantify how these variables individually and interactively influence aqueous free chlorine concentration, hydrogen evolution rate, and energy consumption. This approach enables the identification of factor sensitivities and nonlinear interactions that are not observable in single-factor studies.

By mapping the multidimensional relationships between these experimental factors and their corresponding responses, we established a design framework consisting of three distinct operational modes: “Rapid Response,” “Maintenance,” and “Fuel-Priority.” These modes were further validated through experiments to confirm their performance. The results provide insight into how the POF system can be tuned to maintain disinfection-relevant chlorine responses (0.5–4.0 mg L⁻¹) while managing energy efficiency and hydrogen yield. Ultimately, this work demonstrates the feasibility of using POF-based reactors for tunable, light-driven oxidant and fuel production, offering a practical strategy for decentralized water treatment.

2 Experimental section

2.1 SrTiO₃ synthesis and fabrication of POF-STO

Aluminum-doped SrTiO₃ loaded with RhCrCoO_x cocatalysts (STO) was employed in this study. The synthesis of STO followed established procedures reported in our previous work.^25,26 The resulting photocatalyst was then immobilized onto POFs to form STO-coated POFs (STO-POF), using a spray-coating method. The pristine POF surface was chemically etched using acetone to create a porous anchoring network and enhance side light emission. The STO photocatalyst was then spray-coated onto the fibers using colloidal silica as an inorganic binder to ensure robust adhesion. Detailed descriptions of STO synthesis and STO-POF fabrication procedures are provided in SI (Sections S1 and S2).

2.2 Hydrogen production measurements of POF-STO systems

Photocatalytic experiments were conducted in a sealed 10 mL cylindrical reactor designed to accommodate both batch and continuous-flow configurations (schematic in Fig. 1a and detailed real setup provided in Fig. S4). The 15-run Response Surface Methodology (RSM) experimental matrix, along with the “Rapid Response” and “Fuel-Priority” operational modes, were executed using the vertical batch configuration (Fig. S4a) to enable accurate quantification of the accumulated products. For continuous-flow evaluations (operational mode “Maintenance”), the reactor was adapted into a horizontal flow-through system equipped with defined inlet and outlet ports (Fig. S4b).

Fig. 1 (a) Schematic of the POF-STO reactor configuration. (b) Proposed mechanism for the simultaneous generation of H₂ and HClO over STO on the POF surface. (c–e) SEM characterization of the POF-STO surface evolution: (c) the pristine, unmodified POF; (e) the acetone-etched POF with irregular pores (<1 µm); and (d) the final morphology after STO catalyst deposition.

In all configurations, the POF-STO samples were positioned at the bottom of the reactor. Illumination was supplied by monochromatic UV light-emitting diodes (365 nm, 3–10 W; Shenzhen Yunju Electronics Co., Ltd) coupled to the fibers via SMA 905 optical connectors. The reaction vessel was secured with a tight-fitting elastomeric O-ring, and sealing integrity was rigorously verified across all connections using a Swagelok gas leak detector to preclude any ingress of ambient air. Prior to illumination, the saline reaction medium was vigorously purged with high-purity nitrogen gas for 30 minutes to ensure the complete removal of dissolved oxygen and residual air.

Photocatalytic performance was evaluated in aqueous NaCl solutions with varying salinity. NaCl solutions were prepared by dissolving analytical-grade NaCl (ACS reagent, ≥99.0%) in ultrapure water within a 100 mL standard volumetric flask immediately prior to each experiment. Ultrapure water (18.2 MΩ cm) was produced using a Thermo Scientific™ Barnstead™ GenPure™ water purification system.

The SrTiO₃ loading on the POFs was quantified by measuring the Sr content using inductively coupled plasma mass spectrometry (ICP-MS, PerkinElmer NexION 2000). For each measurement, a 2 cm fiber segment was placed in a 55 mL MARSXpress TFM vessel containing 10 mL of a HNO₃ and H₂O₂ mixture (9 : 1 v/v) for overnight extraction. Microwave digestion was subsequently performed using a CEM MARS 5 system with a programmed ramp to 100 °C over 5 minutes (5-minute hold), followed by a ramp to 160 °C over 5 minutes (20-minute hold) at 1600 W (75% power). The resulting digested samples were diluted to 50 mL with Milli-Q water and filtered through a 0.22 µm membrane (Cole-Parmer). Final dilutions were adjusted with Milli-Q water and nitric acid to ensure analyte concentrations fell within the 1–1000 ppb calibration range with a final matrix acidity of 1.5–2.5% (v/v) HNO₃.

The optical properties of POF-STO under different illumination conditions were evaluated through photon irradiance measurements. These parameters included the total light input at the proximal end (I₀), the side-emitted light due to scattering (I_S) per cm of the POF, and the transmitted light irradiance at the distal end (I_T). Light irradiance (µW cm⁻²) was measured using a spectrophotoradiometer (Avantes AvaSpec-2048L). The calculation of total side-emitted optical power (µW) for different LED sources are summarized in Section S3. The side-emitting light intensity is highest at the proximal end (near the LED) and decays longitudinally (see Fig. S3 for the irradiance profile). Preliminary length-dependency tests showed that increasing the active length from 10 cm to 18 cm yielded only a ∼20% increase in total H₂ and HClO production, despite an 80% increase in catalyst amount/fiber length, indicating a plateau in local light utilization efficiency. The 18 cm active length was intentionally selected for the current reactor prototype to span the full physical depth of the 10 mL water column. Consequently, the generation rates reported represent the macroscopically integrated performance across this entire 18 cm length, rather than a uniform local rate.

Free chlorine concentrations in water were determined using a colorimetric method with a HANNA Instruments free chlorine test kit (HI3831F), following standard analytical procedures of the test kit. Calibration curves were prepared by serial dilution of a commercial 1% hypochlorous acid (HOCl) standard solution to generate free chlorine concentrations ranging from 0 to 100 mg L⁻¹ (Fig. S5a).

Gas samples were periodically taken from the reactor headspace at 30 min intervals and analyzed using a gas chromatograph (Shimadzu GC-2010) equipped with a thermal conductivity detector (TCD) operated at 200 °C and a Carboxen 1010 PLOT column (30 m length, 0.53 mm inner diameter) maintained isothermally at 230 °C. Ultra-high-purity argon (>99.999%) was used as the carrier gas at a flow rate of 40 mL min⁻¹. Hydrogen quantification was calibrated using standard gas mixtures containing 0, 10, 20, 40, 80, and 100% H₂ in N₂ (200 µL injection volume), yielding a linear relationship between integrated peak area and hydrogen concentration (Fig. S5b).

2.3 Experimental design and statistical analysis

A three-factor RSM was employed to quantify the individual and synergistic effects of operational parameters on the reactor's performance.¹³ The experimental matrix was generated using JMP Pro software (SAS Institute Inc., Cary, NC, USA), varying three continuous independent factors: Salinity (X₁), Catalyst Loading (X₂), and LED Light Intensity (X₃), according to the levels defined in Table 1.

Table 1 Experimental design matrix and observed responses

Run	Coded levels			Input factors			Registered responses
	X 1	X 2	X 3	Salinity (M NaCl)	Catalyst loading (mg STO)	Energy input (W)	Free chlorine generation rate (mg L^−1 h^−1)	Energy consumption (kWh kg^−1 H2)	H2 evolution rate (µmol h^−1)
1	−1	−1	0	0.01	15	0.45	0.15	5.98	42.78
2	−1	1	0	0.01	5	0.27	0.40	2.07	67.22
3	1	−1	0	0.1	10	0.9	1.33	7.33	64.27
4	1	1	0	1	15	0.27	5.03	2.67	50.95
5	−1	0	−1	1	5	0.9	0.74	6.79	63.84
6	−1	0	1	1	15	0.45	4.14	4.72	45.69
7	1	0	−1	1	5	0.27	3.92	2.13	64.47
8	1	0	1	1	10	0.45	4.02	3.9	55.68
9	0	−1	−1	1	10	0.9	3.25	7.47	61.79
10	0	−1	1	0.01	10	0.27	0.07	2.06	61.03
11	0	1	−1	0.1	5	0.45	0.47	3.41	63.04
12	0	1	1	0.1	15	0.27	0.72	2.64	49.73
13	0	0	0	1	15	0.9	5.02	8.56	52.62
14	0	0	0	0.01	5	0.9	0.01	6.6	67.67
15	0	0	0	0.01	15	0.9	2.60	8.18	50.48

---

The design consisted of 15 experimental runs. For each run, the reactor was operated for a fixed duration of 2 hours, and three response variables were measured: aqueous free chlorine generation Rate (mg· L⁻¹ h⁻¹), energy consumption (kWh ·kg⁻¹ H₂), and hydrogen evolution rate (µmol h⁻¹). These results, which serve as the basis for statistical modeling, are also summarized in Table 1. Furthermore, the oxygen evolution rate (µmol h⁻¹) is summarized in Table S1. To ensure stoichiometric charge balance, the total rate of electron consumption must equal the total rate of hole consumption:

2 × R_H2(electrons) = 2 × R_HClO(holes) + 4 × R_O2(holes) (1)

As detailed in Section S4 and Table S1, ‘surplus’ holes not utilized for the mass-transport-limited chloride oxidation instead drive the competitive oxygen evolution reaction. This complementary pathway ensures global charge neutrality across all operational regimes, with a minor −6 to +3% deviation which might be attributed to O₂ solubility.

The experimental data were fitted to a second-order polynomial model using the standard least squares method, as expressed in eqn (2):

where Y represents the predicted response; β₀ is the constant intercept; k represents the number of factors (k = 3 in our study), β_i, β_ii, and β_ij correspond to the linear, quadratic, and interaction coefficients, respectively; χ_i and χ_j are the coded independent variables.

The statistical significance of the model terms was evaluated using Analysis of Variance (ANOVA) at a 95% confidence level (α = 0.05). The adequacy of the fitted models was validated through the coefficient of determination (R²) and analysis of residuals. Finally, 2D contour plots and desirability profiling were generated to visualize the design space and identify specific operational modes that optimize the trade-offs between fuel production, disinfection efficacy, and energy efficiency.

3 Results and discussion

3.1 Photocatalytic mechanism and reaction pathways

Fig. 1b shows the mechanism of simultaneous generation of gaseous H₂ and aqueous HClO over STO on the POF surface, which is driven by the precise band engineering of the photocatalyst relative to the redox potentials of the saline solution. Upon irradiation by photons delivered via the side-emitting optical fibers, electron–hole pairs are generated within the lattice. The photogenerated electrons (e⁻) immediately migrate to the conduction band to initiate the reduction half-reaction. Mediated by the co-catalyst sites, these electrons reduce protons, derived from both the bulk solution and the counter-oxidation step, into molecular hydrogen (eqn (3)).

2H₂O + 2e⁻ → H₂ + 2OH⁻ (E⁰ = 0 V vs. RHE) (3)

Simultaneously, the photogenerated holes (h⁺) migrate to the valence band surface to drive the oxidation half-reaction. The valence band potential of STO is sufficiently positive (+1.48 V via RHE) to overcome the standard potential for chloride oxidation. Previous study indicate this proceeds via the initial oxidation of chloride ions to chlorine gas (Cl₂) (eqn (4)), which subsequently undergoes rapid disproportionation (pK_a = 7.5) in the aqueous phase to generate HClO and protons (eqn (4)).^5,30

2Cl⁻ + 2h⁺ → Cl₂ (E⁰ = +1.48 V vs. RHE) (4)

Cl₂ + H₂O ↔ HClO + H⁺ + Cl⁻ (5)

A critical challenge in this dual-production system is the stability of the generated oxidant, as bare reduction co-catalysts such as Rh or Pt are known to actively promote the decomposition of HClO back into volatile Cl₂.^30,31 The chromium oxide (CrO_x) component of the co-catalyst serves as a selective passivation layer to address this issue.³² This selective permeability allows the POF-STO reactor to achieve high steady-state concentrations of chlorine without catalytic self-degradation.

The POF surface throughout the catalyst deposition process was characterized via SEM, as illustrated in Fig. 1c–e. The pristine, unmodified POF exhibits a characteristically smooth, featureless topography with negligible surface roughness (Fig. 1c). To generate the necessary surface area and anchoring sites for catalyst adhesion, the fiber underwent a controlled chemical etching step via acetone. As revealed in Fig. 1d, this treatment drastically alters the morphology, creating a highly porous structure characterized by a dense network of deep, irregular pores with the pore size <1 µm. This corresponds to the fiber turning from transparent to a frosted appearance.

Fig. 1e demonstrates the final morphology following STO deposition. The SEM image confirms the successful formation of a catalyst layer, STO particles interspersed with smaller colloidal silica particles which serve as inorganic binders. Furthermore, the image shows that these particles are not only depositing on the surface but are physically interlocked within the pores created during the etching phase. This mechanical anchoring into the porous POF surface is essential for ensuring the long-term stability of the coating in the saline environment while providing a high surface area for the “inside-out” illumination reaction. Additional SEM images are provided in Fig. S6 to display the lateral surface, cross-section, and STO deposition morphology of the POF-STO. The SEM image confirms the successful deposition of the STO particles interspersed with smaller colloidal silica particles which serve as inorganic binders. Importantly, the image shows that these particles do not form a continuous uniform layer on the surface; instead, they form localized aggregates, or clusters, that are physically interlocked within the deep pores created during the etching phase.

To standardize the system's performance metrics and facilitate direct benchmarking, we evaluated the reactor's efficiency across three distinct figures of merit. First, to evaluate photon utilization, the Apparent Quantum Yield (AQY) for hydrogen evolution was calculated (detailed methodology is provided in Section S5 and the results are summarized in Table S2). The calculated AQY ranges from 1.02% to 4.54%, with the highest efficiencies observed at the lowest light intensity (0.27 W). Achieving an AQY of up to 4.54% without the use of sacrificial electron donors demonstrates the highly efficient H₂ generation facilitated by the POF-STO architecture.

Second, to assess macroscopic production capacity, the absolute yields were normalized to the catalyst mass, the results are also summarized in Table S2. Under optimal conditions (5 mg STO loading), the peak H₂ evolution rate translates to a specific yield of 13.5 mmol g⁻¹ h⁻¹. This robust mass-normalized rate is highly competitive, particularly considering it is achieved during continuous co-generation without any sacrificial reagents.

Third, to determine intrinsic catalytic activity, the Turnover Frequency (TOF) was evaluated based on the specific active sites (detailed methodology is also provided in Section S6 with all values reported in Table S2). The TOF reached ∼2789 h⁻¹ per Rh active site for H₂ evolution, while the HClO generation TOF peaked at ∼15 h⁻¹ per Co site. This stark contrast in specific activity quantitatively confirms that the oxidation half-reaction is severely throttled by the mass transport of trace chloride ions, rather than intrinsic catalytic limitations.

Finally, we benchmarked the POF-STO system against state-of-the-art photocatalytic architectures (Table S3). While specialized UV-driven systems report higher AQYs for standard water splitting, our system targets the complex, dual-function co-generation of H₂ and HClO without any sacrificial electron donors. Under these strict conditions, our mass-normalized yield (13.5 mmol g⁻¹ h⁻¹) and Rh TOF (∼2789 h⁻¹) are highly competitive. This confirms that our ‘inside-out’ optical framework efficiently maximizes photon delivery and charge utilization, overcoming the typical optical scattering losses of conventional slurry reactors.

3.2 Model validation and fit statistics

The statistical reliability of the second-order polynomial models fitted to the free chlorine generation rate, hydrogen evolution rate, and energy consumption was confirmed via ANOVA, as summarized in Table 2. To evaluate the predictive accuracy of the established framework, the relationship between predicted and experimental values for all three responses was analyzed.

Table 2 Summary of ANOVA and Fit statistics for reactor performance models

Response variable	R ^2	RMSE	F-ratio	p-Value
Energy consumption	0.99	0.44	45.48	0.0003
Aqueous free chlorine concentration	0.98	0.98	24.32	0.0014
Hydrogen (H2) evolution	0.90	4.52	4.96	0.0444

---

The regression model for energy consumption demonstrated an exceptional fit with the experimental data (R² = 0.99, p < 0.001). This high predictive accuracy confirms that the system's energy usage is a deterministic function of the LED power input and operational time, with minimal experimental error. Furthermore, this near-perfect correlation confirms the high fidelity of the experimental setup. This serves as an internal control, verifying that the photon delivery system maintained consistent output stability across the entire experimental matrix.

Similarly, the aqueous free chlorine production model exhibited strong predictive capability (R² = 0.98). This confirmation provides critical insight into the limiting regimes of the reactor. The dominance of salinity indicates that the oxidation half-reaction is chemically rate-limited by the mass transport and adsorption of chloride ions onto the active sites. Essentially, the supply of reactants to the surface constrains the disinfection capacity, rather than the intrinsic turnover frequency of the catalyst.

The hydrogen evolution model showed a comparatively higher variance (RMSE = 4.52) and a lower coefficient of determination (R² = 0.90). This behavior is characteristic of heterogeneous photocatalysis. Yet, the model still remains statistically significant (p = 0.044 < 0.05) and effectively identifies catalyst loading as the rate-limiting step for fuel production. This variance might be attributed to the physical instability of gas bubbles at the fiber surface. Unlike free chlorine, which dissolves instantly into the liquid, hydrogen forms microbubbles that cling to the catalyst. Since the reactor relies on the evanescent field, light leaking from the fiber surface, these bubbles interfere with photon delivery.^33,34 The refractive index mismatch between gas bubbles (n = 1.0) and the surrounding water (n = 1.33) causes unpredictable light scattering.^35,36 This creates natural “optical noise” in the reaction rate. Namely, the random growth and detachment of these bubbles cause dynamic fluctuations in local photon delivery, resulting in the observed statistical variance in the hydrogen evolution rate.

To ensure the statistical metrics reflect true physical behavior, we verified the model accuracy using diagnostic plots (Fig. S7). The “Predicted vs. Actual” plots (Fig. S7a, c and e) visually confirm the high determination coefficients (R²). The data points align closely with the diagonal line, indicating that the model predictions match the experimental results well. This is most clear in the energy consumption model (Fig. S7e), where the linear pattern confirms the model captures nearly all the experimental variance. Additionally, the “Residual vs. Predicted” plots (Fig. S7b, d and f) show that the errors are randomly distributed around zero. This randomness is particularly important for the hydrogen model. Although the hydrogen data show more scatter than the chlorine or energy data (Fig. S7b), the lack of any pattern confirms that this is due to normal experimental noise rather than a defect in the model equations.

3.3 Statistical analysis of factor influence

To identify which physical and chemical factors drive these rates, the statistical effects were analyzed. As illustrated in the Pareto chart (Fig. 2), the values shown (labeled as Orthogonal Estimates) simply measure the strength and direction of each factor's impact: the larger the absolute value, the greater its impact on the response. Additionally, a positive number means the factor promotes the reaction, while a negative number indicates that it hinders it.

Fig. 2 Pareto charts showing the orthogonal estimates of experimental parameters on (a) free chlorine concentration, (b) hydrogen evolution rate, and (c) energy consumption. The length of each bar represents the magnitude of the factor's impact on the response. Positive values indicate that the factor promotes the reaction, while negative values indicate a suppression effect.

The analysis revealed that free chlorine generation in water is driven largely by the concentration of sodium chloride (Fig. 2a). With an orthogonal estimate of 1.50, the impact of NaCl is more than double that of the catalyst concentration (0.66), indicating that the reaction is supply-limited by chloride availability rather than photon flux. Consequently, salinity acts as the primary control lever for disinfection capacity.

In contrast, hydrogen evolution (Fig. 2b) shows a much weaker dependence on salinity (estimate of −0.89). Instead, catalyst loading (STO) emerges as the dominant driver with the largest magnitude of effect (−7.09). However, this estimate is negative, confirming that adding catalyst beyond the optimal point significantly hinders performance. This counter-intuitive result is likely due to a “shielding effect,” where excessive particle density blocks light penetration and water molecule delivery, causing outer particles to shade inner ones and rendering the extra catalyst “dead weight” rather than active producers.

To mechanistically confirm the optical interactions and the “shielding effect” within this porous architecture, quantitative side-emission irradiance profiles were measured along the fiber axis at varying catalyst loadings (Fig. S3a). These optical profiles provide direct physical evidence of the shielding phenomenon. Excessive catalyst loading induces a substantially steeper exponential decay of light along the longitudinal axis. The dense proximal catalyst aggregates aggressively absorb the internal photon flux, preventing light propagation to the distal end of the fiber (>5 cm) and effectively creating a downstream optical dead zone.

To functionally validate the macroscopic consequences of this rapid photon depletion, the effective “working length” of the reactor was evaluated (Fig. S3b and c). The catalytic performance of a standard 18 cm POF-STO sample was compared to a 10 cm POF-STO sample. Despite a 44% reduction in the total catalyst-coated surface area, the generation rates for free chlorine and H₂ decreased by only 19% and 22%, respectively. These functional data perfectly corroborate the optical profiles: the aggressive light absorption at the proximal end leaves the distal portion of the fiber optically starved. These quantitative measurements mechanistically validate that optimizing catalyst density is critical to balancing the number of available active sites with sufficient photon penetration length along the reactor axis.

Finally, the analysis of energy consumption (Fig. 2c) confirmed that LED light intensity was the sole dominant factor, exhibiting an estimate of 2.21, which is at least four times larger than that of the other factors, confirming that the energy cost is a direct function of photon input power. This isolation of energy efficiency allows for the minimization of power consumption without altering the fundamental reaction chemistry, provided the minimum photon flux required for activation is met.

3.4 Analysis of experimental factor interaction effects

To further examine specific pairwise dependencies and identify critical thresholds of various factors, interaction prediction profilers were generated (Fig. 3). These profiles plot the predicted response as a function of one factor while holding the others constant. The significance of the interaction terms identified in the Pareto analysis (Fig. 2) is visually confirmed here by the non-parallel behavior of the interaction slopes. Fig. 3 displays the predicted response as a function of one factor while holding others fixed. Crucially, the geometry of the lines reveals the underlying physics: parallel lines indicate that factors operate independently, while converging or non-parallel slopes signal a complex interaction where one factor alters the effect of another.

Fig. 3 Interaction prediction profilers illustrating the dependencies of the reaction responses. The matrices display the predicted response (y-axis) as a function of one factor (x-axis) while holding the second factor at fixed high (blue) or low (red) levels. Parallel lines indicate that the factors operate independently, while non-parallel or converging slopes indicate a significant interaction between the two parameters. (a) Hydrogen evolution rate. (b) Energy consumption. (c) Free chlorine concentration.

The profile for the hydrogen evolution response (Fig. 3a) provides the first evidence of non-ideal behavior. The response to the catalyst loading factor exhibits a characteristic parabolic shape that plateaus and declines at higher concentrations. This non-linearity validates the large negative orthogonal estimate for the catalyst loading factor in the Pareto analysis, confirming that excess catalyst becomes an optical obstruction rather than an active site. Additionally, the starkly crossed lines between salinity and LED intensity visually reinforce the strong interdependence required to balance photon flux with hole scavenging.

In contrast, the energy consumption profiles (Fig. 3b) display rigid, parallel linearity, particularly regarding the LED intensity factor. The absence of crossing or curving lines visually confirms that the chemical factors (NaCl and STO) do not interact with the energy draw. This indicates that the electrical efficiency is independent of the reaction rate; the system consumes a fixed amount of power to generate photons, regardless of whether those photons are successfully converted into fuel or simply wasted as heat.

The free chlorine response in the aqueous phase (Fig. 3c) reveals a regime dominated by reactant starvation. The response to the salinity factor shows a steep, positive slope that does not plateau, confirming that the surface reaction is rapid but throttled strictly by the supply of chloride ions. The positive STO*LED interaction identified in the Pareto analysis (Fig. 2a) is confirmed here, showing that the benefit of increasing light intensity is significantly amplified when higher catalyst loadings provide sufficient surface area to utilize the increased photon flux.

While the interaction profiles identify trends, the inspection of the 3D response surfaces reveals the sophisticated nature of the three-phase system (Fig. 4). These topological maps visualize the reaction landscape, where “elevation” represents the magnitude of the response. A flat plane indicates a simple linear relationship, while a curved “hill” or “valley” confirms a saturation point or physical limit.

Fig. 4 Three-dimensional response surface plots mapping the experimental design space. The vertical axis represents the magnitude of the predicted response, while the base axes represent the interacting experimental factors. The topography of the surface serves as a visual guide to the system behavior: a flat plane indicates a linear dependence, while convex curvature or a “hill” indicates saturation limits and non-linear interactions. (a) Hydrogen evolution rate. (b) Energy consumption. (c) Free chlorine concentration.

The resulting topography for hydrogen evolution (Fig. 4a) exhibits a pronounced twisted saddle shape, visually confirming a strong interaction between photon flux and catalyst mass. At low catalyst loadings (5 mg per POF), hydrogen yield increases sharply, indicating efficient surface utilization where the majority of particles are photo-active. However, as the loading increases toward the upper bound under maximum LED power, the surface exhibits a steep decline. This diminishing return is mechanically attributed to the “shielding effect”. Because light propagates via evanescent waves and scattering from the core outwards, excessive particle density restricts the optical penetration depth. The inner layers of the catalyst coating absorb or scatter the majority of the photons, effectively shading the outer layers. These shielded particles become “dead weight”; they contribute to mass but not to charge carrier generation, thereby reducing the overall specific activity of the photocatalyst.

In contrast, the response surface for aqueous free chlorine (Fig. 4b) displays a steep ascent driven primarily by salinity, which begins to plateau at higher concentrations. The steepness of this slope, combined with the lack of significant quadratic curvature along the LED axis, suggests that the oxidative half-reaction operates in a regime governed strictly by reactant supply (chloride ions) rather than optical limitations. Mechanistically, this indicates that the generation of valence band holes is kinetically faster than the adsorption of chloride ions to the surface. Consequently, the reaction is not limited by the electronic properties of the photocatalyst (charge transfer), but by the mass transfer of Cl⁻ ions from the bulk solution to the active sites. This confirms that increasing salinity is the most direct method to push the equilibrium forward, as the surface is currently starved for hole scavengers.

Next, the energy consumption surface (Fig. 4c) reveals a critical engineering constraint regarding photon-to-electron coupling. While the Pareto chart indicated a linear dependence on LED intensity, the 3D surface reveals a “twisted plane” topography, indicating a subtle interaction between light intensity and chemical composition. The steep ascent along the LED axis confirms that the primary energy cost is the electrical power required to generate photons. However, the flatness of the surface along the NaCl axis implies that the POF-STO system consumes the same amount of power regardless of whether the catalyst is efficiently converting those photons into fuel or wasting them as heat via recombination. This underscores a vital efficiency mechanism: energy minimization cannot be achieved by chemical tuning alone; it requires precise matching of the photon flux (LED power) to the catalyst's saturation point to ensure that all the energy input encounters an active site capable of utilizing it.

In addition, Fig. S8 presents the other sets of 3D response surface plots. These plots confirm that the kinetic regimes and factor interactions are consistent across the entire experimental design space. The plots for hydrogen evolution rate (Fig. S8a and b) provide further evidence of the optical shielding effect. Fig. S8a exhibits a complex saddle topography, demonstrating that maximizing yield requires a precise balance between photon flux and chloride hole scavengers to suppress rapid charge recombination. Furthermore, Fig. S8b isolates the interaction between catalyst loading and salinity. The surface maintains a distinct convex curve along the STO axis, confirming that excessive catalyst loading reduces performance due to optical shielding, regardless of the salt concentration.

The free chlorine concentration surfaces (Fig. S8c and d) reinforce the conclusion that this reaction is dominated by mass transfer limits. Fig. S8c highlights a steep, continuous ascent along the NaCl axis, contrasted with a rapid saturation plateau along the STO axis. In contrast, Fig. S8d maps light intensity against catalyst loading without the salinity variable. While this surface exhibits a concave topography, its overall response magnitude (Z-axis variance) is substantially smaller than that of the reactant-driven plots. This confirms that when the system is mass-transfer limited by chloride diffusion, excess photogenerated holes are instead consumed by the competing oxygen evolution reaction. Hence, supplying additional photons or active sites yields negligible gains in targeted free chlorine production.

Finally, the energy consumption plots (Fig. S8e and f) provide definitive visual proof of the decoupling between chemical parameters and electrical power draw. Fig. S8e reveals a concave ‘valley’ topography, demonstrating that baseline energy demand reaches a localized minimum at intermediate levels of salinity and catalyst loading. However, Fig. S8f confirms that the absolute primary driver of energy consumption remains the LED intensity setting, evidenced by the steep, dominant linear ascent exclusively along the LED axis. Overall, the statistical model validates a key design strategy: minimizing catalyst loading to prevent hydrogen shielding, maximizing salinity to overcome mass-transfer limits, and tuning LED intensity to the precise saturation threshold.

3.5 Tunable operation: the three modes strategy

The statistical independence of the primary drivers, where salinity controls disinfection and catalyst loading limits hydrogen evolution, enables a unique operational advantage: the ability to tune the reactor for specific output priorities. By fixing the catalyst loading at the optimized low level of 5 mg to mitigate the optical shielding effect identified in the ANOVA, the reactor configuration is stabilized. This allows the system to be shifted between three distinct operational modes solely by adjusting the sodium chloride concentration and LED power input. Here, we propose three different working modes, which are summarized in Table 3.

Table 3 Operational parameters and predicted responses/functions for the three-mode strategy

Mode	Factors			Expected main responses and functions
	Salinity	LED input	Catalyst	Free chlorine	Hydrogen fuel	Function
Mode 1 Rapid Response	1.0 M	0.90 W	5 mg	∼0.60 mg L^−1 in 10 min 6.19 mg L^−1 (2-h total)	128.5 µmol (2-h total)	High-HClO stock solution generation
					64.3 µmol h^−1
Mode 2 Maintenance (batch)	0.01 M	0.27 W	5 mg	1.34 mg L^−1 (accumulated)	140.8 µmol (2-h total)	Maintain steady HClO concentration at water outlet
					70.4 µmol h^−1
Mode 2 Maintenance (continuous)	0.01 M	0.27 W	5 mg	0.5–4.0 mg L^−1 (at flow rate 10–45 mL min^−1)	Continuous
Mode 3 Fuel Priority	0.01 M	0.90 W	5 mg	0.47 mg L^−1	153.2 µmol (max yield)	Maximize fuel production
					76.6 µmol h^−1

---

3.5.1 Mode 1: Rapid Response. Designed for immediate water safety and bulk concentrate generation, this mode operates at high salinity (1.0 M) and high LED input (0.90 W). By maximizing the salinity, the system mitigates mass-transfer limitations to deliver potable water in minimal time. The statistical model indicates that the reactor achieves the WHO minimum safety threshold (0.60 mg L⁻¹) within just 10 minutes of operation. Continued operation over the full 2-hour cycle yields a cumulative chlorine output of 6.19 mg L⁻¹, effectively creating a high-potency stock solution for the treatment of larger water volumes. Simultaneously, the system generates 64.3 µmol h⁻¹ of hydrogen, ensuring that fuel recovery is maintained even during high-priority disinfection sequences.

3.5.2 Mode 2: Maintenance (steady HClO generation rate). For daily consumption where water quality is stable, the priority shifts to energy conservation and safety. Operating at Low Salinity (0.01 M) and Low Light Intensity (0.27 W) drastically reduces the energy footprint. In this mode, reducing the LED power directly minimizes consumption to 1.97 mW. This model predicts that even at these minimal settings, the reactor maintains a residual chlorine level of 1.34 mg L⁻¹, falling perfectly within the WHO and EPA safety guidelines (0.5–4 mg L⁻¹) to suppress microbial growth while ensuring potable water safety.^37,38 Remarkably, due to the robust catalyst kinetics, hydrogen production remains high at 70.4 µmol h⁻¹, proving that fuel generation can be sustained even during energy-saving maintenance cycles.

3.5.3 Mode 3: Fuel Priority (maximized H2 evolution). Represents the novel contribution of the POF-STO system: the decoupling of fuel production from disinfection. By maintaining low salinity (0.01 M) but increasing to high light intensity (0.90 W), the system ensures maximum stability in hydrogen evolution (76.6 µmol h⁻¹). As indicated by the Pareto analysis, hydrogen production is effectively independent of salinity, meaning the fuel yield is not affected by the low salt concentration. However, the lower salinity effectively “throttles” the chlorine production down to potable levels (0.47 mg L⁻¹), preventing an overdose despite the high photon flux. This allows the reactor to operate as a dual-function system, harvesting consistent hydrogen fuel over the 2-hour period while simultaneously producing safe drinking water.

To validate the statistical predictions of the RSM, we conducted experiments for each of the three proposed operational modes. The comparison between the model-predicted values and the actual experimental profiles (Fig. 5) reveals a high degree of concordance, confirming the robustness of the tunable design strategy.

Fig. 5 Experimental validation of the tunable three-mode strategy showing aqueous free chlorine generation (red spots) and gaseous hydrogen evolution (blue spots) for: (a) Mode 1 (Rapid Response); (b) Mode 2 (Maintenance – continuous); (c) Mode 3 (Fuel Priority).

For mode 1 (Rapid Response), the experimental data (Fig. 5a) exhibits a strictly linear generation profile for both hydrogen and free chlorine. The accumulated yields after 2 hours reached 6.2 mg L⁻¹ aqueous free chlorine concentration and 128 µmol for hydrogen, which is closed to the RSM predicted values of 7.19 mg L⁻¹ and 71.5 µmol h⁻¹. The system reaches the WHO minimum safety threshold of 0.50 mg L⁻¹ within 20 minutes, validating the model's assertion that high salinity can overcome mass-transfer limitations to deliver potable water almost immediately upon startup.

The validation of mode 2 (Maintenance) required bridging the gap between batch and continuous flow operation. The continuous flow experiment (Fig. 5b) operated at a flow rate of 13 mL min⁻¹. The effluent concentration gradually rose from zero, stabilizing at a steady-state value of approximately 0.50 mg L⁻¹ after 180–200 minutes. This confirms that the continuous stirred-tank reactor design can deliver maximum throughput while strictly adhering to the minimum disinfection standard. However, we also need to highlight that the concurrent hydrogen collection during continuous flow was much lower than we expected. The continuous hydraulic turnover disrupted gas accumulation in the headspace, causing partial loss of microbubbles via the liquid outlet. In the future, the addition of a gas–liquid separation unit is necessary to achieve accurate online hydrogen monitoring.

Finally, the experimental results for mode 3 Fuel Priority (Fig. 5c) demonstrate that hydrogen and chlorine production can be controlled independently. Despite maximizing the photon flux to 0.90 W, the free chlorine generation slope remained extremely shallow, reaching a final concentration of only 0.40 mg L⁻¹, which is virtually identical to the model-predicted throttle limit of 0.47 mg L⁻¹. Conversely, the hydrogen evolution rate remained aggressive, yielding ∼70 µmol h⁻¹, proving that fuel production is indeed optically limited and independent of salinity. This divergence in Fig. 5c confirms the unique mechanism of the POF-STO system: high-energy illumination can be used to harvest fuel without the risk of overdosing the water with chlorine.

For practical water treatment, the operational stability of the photocatalyst and the structural integrity of the cocatalyst are paramount. To evaluate this, the POF-STO reactor was subjected to three consecutive operational cycles. The system exhibited robust reusability, maintaining stable H₂ and HClO generation rates without significant degradation (Fig. S9). Furthermore, to address the safety concern of heavy metal leaching from the RhCrCoO_x cocatalyst, the treated effluent was analyzed via ICP-MS. The concentrations of all cocatalyst species, including potentially toxic Cr ions, were below the instrumental limit of detection. This confirms that the cocatalyst is highly stable in the saline oxidative environment and poses no Cr contamination risk to the treated water.

To bridge the gap between bench-scale validation and field deployment, we propose a 50-fiber bundle assembly driven by the same light source. This configuration distributes the photon flux across a 50× larger catalyst surface area, effectively mitigating the optical shielding effect by relieving local saturation and maximizing photon-to-electron conversion. Applying a conservative 50× scale factor to account for bundle packing and optical distribution, the performance of all three operational modes is significantly enhanced. For example, in the shock disinfection mode, the 50-fiber assembly increases oxidative capacity by an order of magnitude, generating sufficient chlorine in 10 minutes to treat the full 500 mL reactor volume to near WHO standards (>0.5 mg L⁻¹). Alternatively, if used in a static 10 mL reactor, the bundle would generate a true shock concentration of approximately 24 mg L⁻¹ free chlorine in water under 10 minutes, providing the massive disinfectant required to sterilize heavily contaminated medical equipment or surface wounds in field hospitals.

For maintenance operation, the distributed architecture allows the fibers to be driven by the same low-power LED input, drastically reducing the specific energy consumption while maintaining residual disinfection. Finally, in Fuel-Priority mode, the scaled system generates 2.9 mmol of hydrogen per cycle, corresponding to a continuous power supply of 48 mW, which can offset approximately 5.3% of the total energy input (assuming H₂ is consumed by a 50% efficiency fuel cell). Also, the energy recovered via hydrogen can be used to support low-power IoT water quality sensors, such as LoRaWAN nodes, which typically consume less than 1 mW.^39,40

4 Conclusion and future look

This study successfully demonstrated the efficacy of a side-emitting POF reactor immobilized with STO for the simultaneous generation of hydrogen fuel in the gas phase and disinfecting free chlorine in the liquid phase. By employing a three-factor RSM, we effectively decoupled the complex kinetic drivers governing this three-phase system, revealing that the reactor's performance is defined by two distinct limiting regimes: chemical supply and optical architecture. The statistical analysis established that free chlorine evolution in water is a chemically driven process, governed almost exclusively by salinity. The linearity of this response confirms that the oxidation half-reaction is salinity-limited; essentially, the disinfection capacity is throttled not by the catalyst loading, but by the mass transport of chloride ions to the active sites.

Conversely, hydrogen evolution was identified as an optically limited process, governed primarily by the catalyst loading rather than LED light intensity. A critical finding of this work is the identification of the “optical shielding effect” in POF reactors. The negative coefficient observed for catalyst loading indicates that beyond an optimal threshold, additional catalyst particles act as optical obstructions instead of functioning as active sites. These excess particles block the evanescent field from penetrating the outer layers of the coating, effectively creating “dead mass” that scatters photons rather than generating excitons. This underscores a fundamental design rule for internal-illumination reactors: the ratio of photon flux to catalyst density is as critical as the chemical stoichiometry.

The most significant engineering contribution of this work is the translation of these statistical insights into a tunable and demand-responsive operational strategy. By utilizing the independent control of the reaction drivers, where salinity dictates disinfection levels while catalyst loading and light intensity govern fuel production, we developed a protocol that allows a single reactor unit to pivot between three distinct operational modes. These include a “Rapid Response” mode for emergency use, which employs high salinity to achieve critical disinfection levels (above 0.60 mg L⁻¹) in only 10 minutes; a “Maintenance” mode designed for continuous operation, which utilizes a calibrated flow rate (13 mL h⁻¹) to maintain a steady-state free chlorine concentration in water within the WHO-recommended range, effectively preventing microbial regrowth while ensuring potability (above 0.50 mg L⁻¹); and a “Fuel-Priority” mode that maximizes hydrogen evolution. Notably, the study reveals the system's “fuel resilience”; since hydrogen evolution is largely independent of salinity, the reactor can operate in a low-salt regime to produce potable water while maintaining maximum hydrogen/energy recovery, effectively functioning as an energy-harvesting device that produces clean water as a beneficial byproduct.

While this study establishes a robust statistical and operational framework for the POF-STO reactor, bridging the gap between laboratory-scale validation and field deployment requires addressing several critical engineering challenges. First, we need to address the hydrodynamic and optical optimization of the fiber bundle. In Section 3.4, we propose a 50-fiber bundle module, but this requires further practice application. Future engineering efforts must focus on optimizing the specific packing density of the bundle, as there is an inherent trade-off: while higher density increases the active surface area per unit volume, hydrogen microbubbles can become entrapped between fibers, blocking active sites and scattering incident light. This bubble-induced scattering is a key challenge in the three-mode technique; furthermore, entrapped gas can lead to physical fuel loss during continuous-flow operation as microbubbles are swept out with the liquid effluent. To prevent this, future designs must integrate active extraction mechanisms, such as introducing pulsatile flow regimes to mechanically shear bubbles from the catalyst surface or embedding hydrophobic gas-permeable membranes within the reactor housing to continuously siphon the evolved fuel from the liquid phase. Also, integrated computer simulations of water flow and light propagation will be required to design a manifold that eliminates stagnant ‘dead zones' within the fiber bundle.^41,42

Simultaneously, the POF architecture must be improved to reduce the energy loss. Currently, the hydrogen yield recovers only ∼5.3% of the UV-LED power consumption. To bridge this deficit, future efforts must focus on maximizing photon utilization efficiency within the reactor through a formal cost analysis and Techno-Economic Assessment to balance capital costs against operational savings. This requires enhancing the side-emission capability of the POFs through controlled surface engineering, such as laser texturing or embedding scattering nanoparticles, to force a greater percentage of the guided light to interact with the catalyst. Furthermore, installing reflective coatings on the distal fiber ends would recycle unabsorbed photons back through the reactor, effectively minimizing light waste.

Beyond materials and optics, the “smart” functionality of the system requires physical validation on a large schedule. A critical next step is the prototyping of the physical feedback control circuit, integrating the hydrogen fuel cell to verify it can stably drive a LoRaWAN conductivity sensor.^39,43 Validating this “closed-loop” architecture, where the reactor's own fuel powers its decision-making hardware, would represent a paradigm shift, moving the technology from a passive treatment unit to a semi-autonomous utility node.

Finally, the robustness of the “salinity-driven” disinfection model must be stress-tested against complex, real-world water matrices. The current models were validated in synthetic brine, but natural waters contain co-ions, natural organic materials and turbidity, which are known radical scavengers that can dampen the linearity of chloride oxidation. Specifically, physical turbidity competitively scatters the evanescent wave, while organic impurities and specific co-ions consume the generated chlorine, necessitating a quantitative assessment of these barriers. Future research should measure how organic impurities in a real water environment consume the generated chlorine. By establishing a “correction factor,” the system's algorithm can automatically increase production to ensure a sufficient disinfection rate even in challenging disaster-relief environments. Additionally, while laboratory results confirmed zero detectable metal leaching, practical municipal deployment will mandate continuous online monitoring to strictly mitigate the risk of chromium toxicity from the RhCrCoO_x co-catalyst.

Conflicts of interest

The authors have patents on related catalytic optical fiber technologies and partnered with H2Optic Insights LLC on related work.

Data availability

The data supporting this article have been included as part of the supplementary information (SI). Supplementary information: detailed STO synthesis and POF fabrication procedures, optical power estimation models, standard calibration curves, statistical diagnostic plots, and additional response surface methodology (RSM) interaction plots. See DOI: https://doi.org/10.1039/d6ta01486k.

Acknowledgements

This work was partially funded by NASA (80NSSC23PB439 and 80NSSC25C0215). The authors appreciate assistance with experimental design from Prof. Sergi Garcia-Segura and Aksana Atrashkevich. We also thank Michael Serpa for performing the ICP-MS analysis.

References

1. T. Zin, K. D. Mudin, T. Myint, D. K. S. Naing, T. Sein and B. S. Shamsul, Influencing factors for household water quality improvement in reducing diarrhoea in resource-limited areas, WHO South East Asia J. Public Health, 2013, 2(1), 6–11,  DOI:10.4103/2224-3151.115828.
2. M. Peter-Varbanets, C. Zurbrügg, C. Swartz and W. Pronk, Decentralized systems for potable water and the potential of membrane technology, Water Res., 2009, 43(2), 245–265,  DOI:10.1016/j.watres.2008.10.030.
3. S. A. Jones, A. Anya, N. Stacey and L. Weir, A Life-cycle Approach to Improve the Sustainability of Rural Water Systems in Resource-Limited Countries, Challenges, 2012, 3(2), 233–260.
4. . United Nations Educational, S.; Organization, C, The United Nations World Water Development Report 2023, United Nations, 2023,  DOI:10.18356/9789210026208.
5. M. Deborde and U. von Gunten, Reactions of chlorine with inorganic and organic compounds during water treatment—Kinetics and mechanisms: A critical review, Water Res., 2008, 42(1), 13–51,  DOI:10.1016/j.watres.2007.07.025.
6. X. Cheng, H. Dong and Z. Qiang, Formation and transformation of pre-chlorination-formed disinfection byproducts in drinking water treatment process, Sci. Total Environ., 2023, 904, 166241,  DOI:10.1016/j.scitotenv.2023.166241.
7. K. Gopal, S. S. Tripathy, J. L. Bersillon and S. P. Dubey, Chlorination byproducts, their toxicodynamics and removal from drinking water, J. Hazard. Mater., 2007, 140(1), 1–6,  DOI:10.1016/j.jhazmat.2006.10.063.
8. V. P. Olivieri, M. C. Snead, C. W. Krusé and K. Kawata, Stability and effectiveness of chlorine disinfectants in water distribution systems, Environ. Health Perspect., 1986, 69, 15–29,  DOI:10.1289/ehp.866915.
9. M. D. Sobsey, C. E. Stauber, L. M. Casanova, J. M. Brown and M. A. Elliott, Point of Use Household Drinking Water Filtration: A Practical, Effective Solution for Providing Sustained Access to Safe Drinking Water in the Developing World, Environ. Sci. Technol., 2008, 42(12), 4261–4267,  DOI:10.1021/es702746n.
10. M. Lindmark, K. Cherukumilli, Y. S. Crider, P. Marcenac, M. Lozier, L. Voth-Gaeddert, D. S. Lantagne, J. R. Mihelcic, Q. M. Zhang and C. Just, et al., Passive In-Line Chlorination for Drinking Water Disinfection: A Critical Review, Environ. Sci. Technol., 2022, 56(13), 9164–9181,  DOI:10.1021/acs.est.1c08580.
11. A. Atrashkevich, A. Alum, R. Stirling, M. Abbaszadegan and S. Garcia-Segura, Approaching easy water disinfection for all: Can in situ electrochlorination outperform conventional chlorination under realistic conditions?, Water Res., 2024, 250, 121014,  DOI:10.1016/j.watres.2023.121014.
12. E. Mostafa, P. Reinsberg, S. Garcia-Segura and H. Baltruschat, Chlorine species evolution during electrochlorination on boron-doped diamond anodes: In situ electrogeneration of Cl₂, Cl₂O and ClO₂, Electrochim. Acta, 2018, 281, 831–840,  DOI:10.1016/j.electacta.2018.05.099.
13. A. Atrashkevich and S. Garcia-Segura, Engineering decentralized electrodisinfection to sustain consistent chlorine generation under varying drinking water chloride content, Appl. Catal. O: Open, 2024, 195, 207012,  DOI:10.1016/j.apcato.2024.207012.
14. M. A. Alkhadra, X. Su, M. E. Suss, H. Tian, E. N. Guyes, A. N. Shocron, K. M. Conforti, J. P. de Souza, N. Kim and M. Tedesco, et al., Electrochemical Methods for Water Purification, Ion Separations, and Energy Conversion, Chem. Rev., 2022, 122(16), 13547–13635,  DOI:10.1021/acs.chemrev.1c00396.
15. T. Takata, J. Jiang, Y. Sakata, M. Nakabayashi, N. Shibata, V. Nandal, K. Seki, T. Hisatomi and K. Domen, Photocatalytic water splitting with a quantum efficiency of almost unity, Nature, 2020, 581(7809), 411–414,  DOI:10.1038/s41586-020-2278-9.
16. T. Hisatomi and K. Domen, Reaction systems for solar hydrogen production via water splitting with particulate semiconductor photocatalysts, Nat. Catal., 2019, 2(5), 387–399,  DOI:10.1038/s41929-019-0242-6.
17. H. Fu, D. Ioka and Z. Pan, Transition-Metal-Doped Perovskite Oxide Materials: A Key Pathway for Photocatalytic Water Splitting Under Visible-Light Irradiation, ACS Catal., 2026, 16(1), 31–56,  DOI:10.1021/acscatal.5c05792.
18. H. Fu, Z. Pan and J. Wenderott, Engaging with clean power: integrated, hands-on lessons on photocatalytic hydrogen production for high school students, Discov. Educ., 2025, 4(1), 105,  DOI:10.1007/s44217-025-00490-x.
19. A. N. Arias-Sanchez, H. Fu, T. Betoni, P. Westerhoff, C. M. Sánchez-Sánchez, M. Gomez-Mingot and S. Garcia-Segura, Hydrogen Production from Untapped Water Sources through Electro- and Photocatalysis: Key Performance Indicator (KPI) Based Benchmarking Framework, ACS Sustain. Chem. Eng., 2025, 14(1), 38–46,  DOI:10.1021/acssuschemeng.5c10440.
20. A. N. Arias-Sanchez, K. Flores, H. Fu, T. Betoni, P. Westerhoff and S. Garcia-Segura, Perspectives of electrochemical and photocatalytic technologies for the water-energy nexus potential of water splitting of brines, Energy and Climate Change, 2025, vol. 6, p.100176,  DOI:10.1016/j.egycc.2025.100176.
21. S. Okunaka, T. Nakamura, T. Ikeda, K. Tsuruda and H. Tokudome, Solar-driven simultaneous production of hypochlorous acid and hydrogen from saline water over RhCrOx-loaded SrTiO3 photocatalyst systems, Sustainable Energy Fuels, 2024, 8(4), 745–751,  10.1039/D3SE01163A.
22. Z. Zhao, H. Fu, L. Ling and P. Westerhoff, Advancing Light-Driven Reactions with Surface-Modified Optical Fibers, Acc. Chem. Res., 2025, 58(10), 1596–1606,  DOI:10.1021/acs.accounts.5c00022.
23. R. Molinari, C. Lavorato, P. Argurio, K. Szymański, D. Darowna and S. Mozia, Overview of Photocatalytic Membrane Reactors in Organic Synthesis, Energy Storage and Environmental Applications, Catalysts, 2019, 9(3), 239.
24. M. Motegh, J. R. van Ommen, P. W. Appel and M. T. Kreutzer, Scale-Up Study of a Multiphase Photocatalytic Reactor—Degradation of Cyanide in Water over TiO2, Environ. Sci. Technol., 2014, 48(3), 1574–1581,  DOI:10.1021/es403378e.
25. H. Fu, I. Ryoma, Y.-J. S. Lai, J. Ananpattarachai, M. Serpa, N. Shapiro, Z. Zhao, Z. Pan and P. Westerhoff, Bridging Materials Innovation with an Efficient Photocatalyst-Enabled Optical Fiber Reactor for H2O2 Production, Environ. Sci. Technol., 2025, 59(28), 14716–14727,  DOI:10.1021/acs.est.5c04043.
26. H. Fu, Z. Pan, Y.-J. S. Lai, J. Ananpattarachai, M. Serpa, N. Shapiro, Z. Zhao and P. Westerhoff, Green hydrogen production via a photocatalyst-enabled optical fiber system: A promising route to net-zero emissions, Energy and Climate Change, 2025, vol. 6, p. 100175,  DOI:10.1016/j.egycc.2025.100175.
27. H. Fu, T.-H. Wang, R.-a. Doong, Y.-J. S. Lai, S. Garcia-Segura, Z. Zhao and P. Westerhoff, Boosting Hydrogen Production via Water Splitting: An ITO Plus g-C3N4 Nanomaterial Enabled Polymer Optical Fiber Design, ACS Mater. Lett., 2024, 6(6), 2267–2275,  DOI:10.1021/acsmaterialslett.4c00456.
28. T.-H. Wang, Y. S. Lai, C.-K. Tsai, H. Fu, R.-A. Doong, P. Westerhoff and B. E. Rittmann, Efficient CO₂ Conversion through a Novel Dual-Fiber Reactor System, Environ. Sci. Technol., 2024, 58(31), 13717–13725,  DOI:10.1021/acs.est.3c10274.
29. Z. Cheng, J. Wang, P. Westerhoff, A. Tayara, R. Wang, Y. Yang, W. Chunxu, C. Shang and L. Ling, Coupling LED-Irradiated Photocatalyst-Embedded Hierarchically Porous Polymeric Optical Fibers with Residual Chlorine for Enhanced Emerging Contaminants Degradation in Decentralized Water Treatment System, Environ. Sci. Technol., 2025, 60(1), 1378–1391,  DOI:10.1021/acs.est.5c09842.
30. R. K. B. Karlsson and A. Cornell, Selectivity between Oxygen and Chlorine Evolution in the Chlor-Alkali and Chlorate Processes, Chem. Rev., 2016, 116(5), 2982–3028,  DOI:10.1021/acs.chemrev.5b00389.
31. S. Harjanto, Y. Cao, A. Shibayama, I. Naitoh, T. Nanami, K. Kasahara, Y. Okumura, K. Liu and T. Fujita, Leaching of Pt, Pd and Rh from Automotive Catalyst Residue in Various Chloride Based Solutions, Mater. Trans., 2006, 47(1), 129–135,  DOI:10.2320/matertrans.47.129.
32. B. Endrődi, S. Sandin, M. Wildlock, N. Simic and A. Cornell, Suppressed oxygen evolution during chlorate formation from hypochlorite in the presence of chromium(VI), J. Chem. Technol. Biotechnol., 2019, 94(5), 1520–1527,  DOI:10.1002/jctb.5911.
33. Y. Song, L. Ling, P. Westerhoff and C. Shang, Evanescent waves modulate energy efficiency of photocatalysis within TiO2 coated optical fibers illuminated using LEDs, Nat. Commun., 2021, 12(1), 4101,  DOI:10.1038/s41467-021-24370-8.
34. Z. Zhao, M. Lanzarini-Lopes, E. Westerhoff, X. Long, H. Rho, Y. Bi, L. Ling and P. Westerhoff, Evanescent wave interactions with nanoparticles on optical fiber modulate side emission of germicidal ultraviolet light, Environ. Sci.: Nano, 2021, 8(9), 2441–2452,  10.1039/D1EN00199J.
35. I. Tanabe, Y. Y. Tanaka, K. Watari, T. Hanulia, T. Goto, W. Inami, Y. Kawata and Y. Ozaki, Far- and deep-ultraviolet surface plasmon resonance sensors working in aqueous solutions using aluminum thin films, Sci. Rep., 2017, 7(1), 5934,  DOI:10.1038/s41598-017-06403-9.
36. C. L. Zhao, J. Li, S. Zhang, Z. Zhang and S. Jin, Simple Fresnel Reflection-Based Optical Fiber Sensor for Multipoint Refractive Index Measurement Using an AWG, IEEE Photonics Technol. Lett., 2013, 25(6), 606–608,  DOI:10.1109/LPT.2013.2243833.
37. Organization, W. H., Guidelines for Drinking-Water Quality, 1, 3rd ed., 2004.
38. A. M. Dietrich and G. A. Burlingame, Critical Review and Rethinking of USEPA Secondary Standards for Maintaining Organoleptic Quality of Drinking Water, Environ. Sci. Technol., 2015, 49(2), 708–720,  DOI:10.1021/es504403t.
39. T. Basmaji, D. Alamri, O. Habash, R. Salim, M. AlKhedher and M. Ghazal, Hydrogen-Based PEM Fuel Cells for Low-Cost Sustainable Powering of IoT Systems, in 2021 International Conference on Decision Aid Sciences and Application (DASA), 2021, pp. 436–440,  DOI:10.1109/DASA53625.2021.9682258.
40. R. L. Rosa, L. Boulebnane, A. Pagano, F. Giuliano and D. Croce, Towards Mass-Scale IoT with Energy-Autonomous LoRaWAN Sensor Nodes, Sensors, 2024, 24, 4279.
41. F. Denny, J. Scott, V. Pareek, G.-D. Peng and R. Amal, Computational fluid dynamics modelling and optimal configuring of a channelled optical fibre photoreactor, Chem. Eng. Sci., 2010, 65(17), 5029–5040,  DOI:10.1016/j.ces.2010.06.003.
42. G. X. de Oliveira, S. Kuhn, H. G. Riella, C. Soares and N. Padoin, Combining computational fluid dynamics, photon fate simulation and machine learning to optimize continuous-flow photocatalytic systems, React. Chem. Eng., 2023, 8(9), 2119–2133,  10.1039/D3RE00239J.
43. S. Vanmathi, H. Awasthi, A. Pal and S. Goel, IoT enabled carbon cloth-based 3D printed hydrogen fuel cell integrated with supercapacitor for low-power microelectronic devices, Sci. Rep., 2024, 14(1), 16953,  DOI:10.1038/s41598-024-67759-3.

---