## Introduction

The removal of anthropogenic and natural pollutants from aqueous systems is a major environmental concern1,2,3. Several technologies have been developed for water treatment over the past 170 years1,4,5,6,7,8. Technologies based on adsorption processes have been proven to be the most affordable and suitable for water treatment in low-income communities1,5. During the past three decades, metallic iron (Fe0) has been intensively used for in-situ environmental remediation9,10,11 and ex-situ water treatment12,13,14,15. However, controversy still exists on whether Fe0 acts as a reducing agent for several pollutants (e.g. selected chemicals)11 or a generator of contaminant scavengers (iron corrosion products–FeCPs) for all classes of pollutants (e.g. chemicals and pathogens)16,17,18. It is certain that, while undergoing oxidative dissolution, Fe0 induces contaminant removal in aqueous systems19,20,21,22,23, and this process can last for decades24,25,26,27. This ability to remove contaminants for prolonged periods has prompted the use of granular Fe0 for decentralized water treatment12,14,28,29,30.

Metallic iron (Fe0) and iron sulfide (FeS)-based materials (including pyrite–FeS2) are two important components of Fe0-based water treatment technology11,22,31,32,33. Both materials are reported to be stand-alone reducing agents that effectively degrade several aqueous contaminants34,35. In this context, Henderson and Demond34,36 have explicitly compared the suitability of both materials, and observed the superiority of FeS (including FeS2) over Fe0 with respect to the sustainability in terms of loss of permeability. During the past two decades, Fe0 and FeS2 have been often mixed in an effort to increase the efficiency of single-Fe0 systems22,33,37,38,39,40,41. However, FeS2 is mostly added to avoid the formation of a passive oxide scale (oxide film) which can hinder further reactions between the Fe0 and pollutants41,42. This application contradicts the successful use of FeS2 to improve the removal of non-reducible contaminants (e.g. As) in Fe0/H2O systems22. Thus, there is a need to understand the real mechanism by which FeS2 improves the efficiency of Fe0/H2O systems, irrespective of any redox transformation. The oxidative dissolution of both Fe0 (Eq. 1) and FeS2 (Eq. 2) typically releases Fe2+, which is also a stand-alone reducing agent for several contaminants35. Fe2+ from Eq. (1) and/or Eq. (2) can be further oxidized to Fe3+ (Eq. 3).

$$Fe^{0} + 2H^{+} \Rightarrow Fe^{2+}+H_{2}$$
(1)
$$FeS_{2} + 7/2 O_{2} + H_{2} O \Rightarrow Fe^{{2 + }} + 2SO_{4}^{{2-}} +2H^{+}$$
(2)
$$4Fe^{{2 + }}+O_{2} + 2H^{+} \Rightarrow 4Fe^{{3+}}+ 2HO^{-}$$
(3)

It is evident that both reactions depicted by Eqs. (1 and 3) consume acidity (H+), while reaction in Eq. 2 produces H+. According to the Le Chatelier’s principle, the reactions consuming H+ (Eq. 1 and 3) are accelerated by pyrite oxidation (Eq. 2). Here, the forward Fe0 dissolution is given priority as it is the main reactant, while less attention is paid to the possible inhibitory effect of FeS2 oxidation on Fe0 dissolution (production of Fe2+). In fact, using FeS2 to enhance iron corrosion is consistent with scientific principles (“Background to the experimental methodology”). However, this understanding provides no insights on the mechanisms of decontamination (adsorption, co-precipitation) and/or induced redox transformation of contaminants (degradation, precipitation). The stoichiometry of the reaction in Eq. (1) shows that 1 mol of Fe0 generates one mole of Fe2+ and one mole of H2, while 1 mol of FeS2 generates only one mole of Fe2+ (Eq. 2). Given that Fe2+ and H2 are stand-alone reducing agents, it is clear that there are more reducing agents (electron donors) in Fe0/H2O than in FeS2/H2O systems. The actual reductive characteristics of each system (i.e., Fe0/H2O, FeS2/H2O, and Fe0/FeS2/H2O) primarily depend on the relative dissolution kinetics of Fe0 and FeS2. According to their relative electrode potentials (-0.44 V for FeII/Fe0 versus 0.25 V for SIII/S-I), Fe0 should be transforming SIII species back to S-I ones. This would correspond to the blocking of oxidative dissolution of Fe0. Although this reaction is thermodynamically feasible, it is masked by the more kinetically favourable FeS2 oxidation by dissolved O2. In the long-term (i.e., over the 41 d investigated herein), this process could contribute to the FeII cycle within a Fe0/FeS2 /H2O system.

The presentation above demonstrates the extreme complexity of the Fe0/FeS2/H2O and accounts for the controversies in the literature on the role of FeS2 in enhancing the efficiency of Fe0/H2O systems22,33,41. Thus, a detailed investigation of the efficiency of the Fe0/H2O system as influenced by the presence of FeS2 is warranted. A critical evaluation of the Fe0 literature suggests that one major limitation has been to test the efficiency of the Fe0/H2O system on individual contaminants or groups of contaminants (e.g. As, dyes, halogenated carbons)22,28,33,41,43. The net result is that increased adsorption, co-precipitation, degradation or precipitation have been reported as the supposed removal mechanisms. An innovative approach was introduced by Miyajima and colleagues using methylene blue (MB) as an indicator of reactivity for the Fe0/H2O system (MB method)44,45. The MB method exploits the differential adsorptive behaviour of MB onto sand and iron oxides or iron-coated sand44,45,46,47. Accordingly, in parallel experiments using constant amounts of sand and different Fe0 specimens, the most reactive system has been the one exhibiting the least MB discoloration, or the one producing the largest amount of iron oxides47,48,49. Note that the exact nature of the oxide is not very important, but its coating activity is the key aspect for the MB method. For example, Banerji and Chaudhari12 did not add any sand in their Fe0 bed to avoid oxide loss by sand coating. The objective of the current study was to characterize the impact of FeS2 addition on the reactivity of the Fe0/H2O system using the MB method. Thus, six different FeS2 mass loadings were used to achieve different final pH values (Eq. 2) (4.5 ≤ pHfinal ≤ 5.2).

## Material and methods

The present research is based on the chemistry of the Fe0/FeS2/sand/H2O system. Therefore, the operating mode of the system will be first discussed. In this study, various amounts of a reactive FeS2 mineral were added to a Fe0/sand mixture to investigate their effects on pH shifts and dye removal.

### Background to the experimental methodology

At neutral pH values, immersed reactive Fe0 corrodes and generates solid iron corrosion products (FeCPs), which progressively coat the surface of sand. The process of iron corrosion causes a pH shift to higher values (Eq. 1). The extent of sand coating depends among other factors on: (i) the Fe0 intrinsic reactivity, (ii) the volume of the solution, (iii) the initial pH value of the solution, (iv) the Fe0/sand ratio, and (v) the duration of the experiment. Under given experimental conditions, the removal efficiency of the system for individual contaminants depends on the final pH value, the extent of sand coating, and the availability of “free” FeCPs. The final pH value determines the speciation of the contaminant and the surface charges of sand and FeCPs50.

When a FeS2 mineral is added to a Fe0/sand system (at a given Fe0:sand ratio) a pH shift to lower values occurs. The extent of pH shift depends on the FeS2 intrinsic reactivity and the amounts added. Lower pH values avoid or delay sand coating and modify the speciation of dissolved contaminants. It then follows that, when FeS2 is added to a Fe0/sand mixture, there are two counteracting processes controlling the pH value of the system37,38,39. Previous results observed with the FeS2 mineral used in the current study41 suggest that pyrite dissolution occurs with much rapid kinetics than Fe0 corrosion. Consequently, the system will not achieve a steady state before the initial pH of the FeS2-free system is achieved. The larger the pH shift the larger the amount of FeCPs generated, which will in turn precipitate at pH > 4.5 and induce contaminant removal by adsorption and co-precipitation.

The methodology used for characterizing the impact of FeS2 on the efficiency of Fe0/H2O systems comprises monitoring the discoloration of a methylene blue solution (MB method) by Fe0/sand systems amended with various FeS2 amounts. Clearly, the availability of FeCPs and their reactivity is modified by lowering the initial pH value to various extents while observing MB discoloration in systems having a final pH value between 4.0 and 5.0. The discoloration of methyl orange (MO) in parallel experiments is used to support the findings based on the MB method. This approach is radically different from the conventional approach testing dyes as model contaminants33,51. For example, Chen et al.33 recently investigated the removal of three different azo dyes (Orange II, Reactive Red X-3B and Amido Black 10B) in the Fe0/FeS2/H2O system. All the three dyes are negatively charged, and were explicitly reported to be removed via reductive transformations. Following the conventional approach, Chen et al.33 monitored the concentrations of dyes, iron and protons (pH value), and performed solid phase characterizations using scanning electron microscopy (SEM), energy dispersive X-ray spectroscopy (EDS) and X-ray photoelectron spectroscopy (XPS). On the contrary, the MB method does not imply such solid phase characterizations because all FeCPs are positively charged50 and the extent to which they cover sand is reflected in the extent of MB discoloration.

### Solutions

#### Dyes

Methylene blue (MB) was used as a tracer of reactivity47, while methyl orange (MO) was a model organic contaminant52. Both dyes are widely used to characterize the suitability of various systems for water treatment46,52,53,54,55. The used dyes were of analytical grade. MB was supplied by Sinopharm Chemical Reagent Co. Ltd, Shanghai (China) and MO by Tianjin Chemical Reagent Research Institution Co. Ltd, Tianjin (China). The dyes were selected due to: (i) similarity in their molecular size, and (ii) differences in their affinity to positively charged iron oxides (Table 1)55. The initial dye concentration used was 10 mg L−1, equivalent to 31.5 μM for MB and 30.7 μM for MO. The working solutions were prepared by diluting concentrated stock solutions (3150 μM for MB and 3070 μM for MO) using deionized water. The pH values of the initial solutions were 6.5 (MB) and 7.0 (MO).

#### Iron

A standard iron solution (1000 mg L−1) from General Research Institute for Nonferrous Metals was used to calibrate the UV/VIS spectrophotometer used for analysis. In preparation for spectrophotometric analysis, ascorbic acid was used to reduce FeIII in solution to FeII. 1,10 orthophenanthroline was used as reagent for FeII complexation48,49,55. Other chemicals used in this study included L( +)-ascorbic acid and L-ascorbic acid sodium salt. Ascorbic acid also degrades dyes (in particular MO) and eliminates interference during iron determination.

### Solid materials

#### Metallic iron (Fe0)

The Fe0 material was purchased from Shanghai Institute of Fine Technology (China). The material is available as scrap iron with a particle size between 0.05 and 5 mm. Its elemental composition as specified by the supplier was: Fe: > 99.99%; C: < 0.1%; N: < 0.1%; O: < 0.1%. Its kPhen value is 13 mg h−156. The kPhen value is the kinetic constant of Fe0 dissolution in a 2 mM 1,10 orthophenanthroline solution, and characterizes the material’s intrinsic reactivity57. The material was used without any further pre-treatment. Fe0 was proven as a powerful discoloration agent for MB specifically because the discoloration agents are progressively generated in-situ45,55. Therefore, the discoloration capacity of the used Fe0 cannot be exhausted within the experimental duration used in the current study (41 d).

#### Sand

The sand conformed to the China ISO standard, and was used as received without any further pre-treatment or characterization. The particle size was between 1.25 and 2.00 mm. Sand was used because it is cheap and readily available and is widely used as admixing agent to prevent rapid permeability loss in Fe0/H2O systems58.

#### Pyrite (FeS2)

The FeS2 mineral was from Tongling City, Anhui province, China. The particle size was between 38 and 48 μm. Its weight composition was 46.0% Fe and 52.2% S, which is equivalent to a purity of 98.2%41. FeS2 was used because of its demonstrated suitability as a pH shifting agent in Fe0/H2O systems37,41,59.

### Dye discoloration experiments

Quiescent batch experiments were conducted in glass test tubes for an experimental duration of 41 d. Dye discoloration was initiated by adding 20.0 mL of the dye solution to a test tube containing 0.1 g of Fe0, 0.0 to 0.6 g of FeS2, 0.0 or 0.5 g of sand, and Fe0/FeS2/sand mixtures containing varying FeS2 loadings. Table 2 summarizes the aggregate content of the 8 Fe0/FeS2/sand systems and one operational reference (blank experiment), giving a total of 9 experimental treatments. Note that the pure Fe0 system (0.1 g of Fe0) is regarded as a Fe0/FeS2/sand system without FeS2 nor sand. The addition of sand was meant to avoid the compaction of the materials by gelatinous FeCPs via cementation55.

The efficiency of individual Fe0 systems for dye discoloration was characterized at laboratory temperature (about 25 ± 2 °C). The final pH value, the iron concentrations and the residual dye concentrations were recorded. All experiments were carried out in triplicates under laboratory (oxic) conditions. The test tubes were protected from direct sunlight.

### Analytical methods

Aqueous dye and iron concentrations were determined by a 752 UV/VIS spectrophotometer (automatic) (Shanghai Jing Hua Technology Instrument Co. LTD). The working wavelengths for MB, MO and iron were 664.5, 464.0 and 510.0 nm, respectively. Cuvettes with a 1.0 cm light path were used. The iron determination followed the 1,10 orthophenanthroline method60. The spectrophotometer was calibrated for dye concentrations ≤ 15.0 mg L−1 and iron concentration ≤ 10.0 mg L−1. The pH value was measured by combined glass electrodes (INESA Scientific Instrument Co. China).

### Presentation of experimental results

In order to characterize the magnitude of the tested systems for dye discoloration, the discoloration efficiency (E) was calculated (Eq. (4)). After the determination of the residual dye concentration (Ct), the corresponding percent dye discoloration efficiency (E value) was calculated as:

$$E = [1 - (Ct/C0)] \times 100\%$$
(4)

where C0 is the initial aqueous dye concentration (about 10.0 mg L−1), while Ct gives the final dye concentration at sampling time (t). The operational initial concentration (C0) for each case was acquired from a triplicate control experiment without additive materials (blank). This procedure was mainly meant to account for experimental errors due to dye adsorption onto the walls of the test tubes.

## Results and discussion

### Dye discoloration in single-aggregate and ternary-aggregate systems

Figure 1a compares the extent of dye discoloration in the four investigated systems: (i) single-Fe0, (ii) single-FeS2, (iii) single-sand, and (iv) Fe0/FeS2/sand. Figure 1b shows the final pH variation with varying FeS2 doses in the ternary Fe0/FeS2/sand systems. Figure 1a clearly shows that there was no MO discoloration in the pure sand system (E = 0%). The E values for both dyes in all other systems were larger than 30%. The uniqueness of the single-sand system (100% sand) relative to the other three systems is that it contains no in-situ generated FeCPs (Table 3). Therefore, only sand with its negatively charged surface46,50 is available for dye discoloration via pure electrostatic interactions. Strong surface interactions with positively charged species is thus responsible for the observed MB discoloration, but no MO discoloration occurs in the single-sand system44,61. As quiescent batch experiments were performed (no advection), diffusive mass transfer in the bulk solution and/or in the pores of generated FeCPs are the rate-limiting steps for the discoloration process.

The absence of MO discoloration in the single-sand system is the most important observation from these experiments. MO discoloration is observed in all other systems and is consistently more intensive than MB discoloration in single-Fe0 and single-FeS2 systems (“Dye discoloration in Fe0/sand/H2O systems”). For the Fe0/FeS2/sand system there is no pronounced difference in the discoloration of both dyes. It is recalled that the Fe0/FeS2/sand system contains 5 g L−1 Fe0, 25 g L−1 of sand and 20 g L−1 of FeS2. Thus, according to Table 2, the extent of dye discoloration depends on: (i) the availability of adsorption sites on inert sand (adsorption on sand), (ii) the extent to which sand is covered by in-situ generated FeCPs (selective adsorption on sand and/or FeCPs), and (iii) the extent to which excess FeCPs can be freely precipitated in the bulk solution (co-precipitation with FeCPs) (see Table 3). In this case, free precipitation herein means FeCPs that are not coating the sand surface (“Background to the experimental methodology”)37,38,44,61.

The merit of the experimental design is to demonstrate dye discoloration in a Fe0/sand/H2O systems as the FeS2 mass loadings vary from 0 to 30 g L−1. Figure 1b clearly shows that varying the FeS2 mass loading under the experimental conditions has resulted in various final pH values, ranging from 4.5 to 7.0 (Table 4). The results summarized in Table 4 also show lower pH values in all FeS2-containing systems with the single-FeS2 having a pH value of 3.3 ± 0.2 for both dyes. Notably, comparison of MO versus MB showed no profound difference in the final pH value (4.7 vs 4.8), the iron concentration (63 vs 65) and the E values (71 vs 73) in the ternary system with a FeS2 mass loading of 20 g L−1. An exception was the Fe0/H2O system without FeS2 (i.e., [FeS2] = 0 g L−1) for which a significant difference in the pH value was observed (6.5 for MB vs. 7.0 for MO). This corresponds to the pH value of the respective initial dye solutions (“Solutions”). The ternary-aggregate system used in this case comprised of 20 g L−1 FeS2. For a better illustration of the role of the FeS2 mineral in the process of dye discoloration in Fe0/H2O systems, three lower (5, 12 and 18 g L−1) and two higher (24 and 30 g L−1) FeS2 doses were also used.

### Iron release in Fe0/sand/H2O systems

Figure 2 compares the iron concentration in the Fe0/sand/H2O systems as the FeS2 loadings vary from 0 to 30 g L−1. Figure 1b and Table 4 have shown a monotonous, but non-linear decrease of the pH value with increasing FeS2 mass loadings. Contrary, Fig. 2a shows a monotonous and linear increase of the iron concentration with increasing FeS2 mass loadings. For a constant FeS2 loading, the final pH values (Fig. 1b) and the [Fe] values (Fig. 2a) were almost the same for both dyes. This observation suggests that the differential behavior of MB and MO in interacting with the involved aggregates, particularly Fe0 and sand, is not reflected in changes of the pH value. Thus, the final pH value arises from the pseudo-steady state equilibrium between two antagonistic processes: (i) Fe0 corrosion which increases the pH value (Eq. 1), and (ii) FeS2 dissolution which decreases the pH value (“Background to the experimental methodology”). Figure 2b clearly shows that lower pH values correspond to higher [Fe] values. Specifically, the [Fe] values dropped from 100 mg L−1 for pH 4.7 to 0.7 mg L−1 for pH 7.0. A sharp decrease of the iron concentration between pH 4.7 and 5.5 is observed and corresponds to the solubility behavior of Fe under oxic conditions62,63. Again, there is no significant difference evident between both dyes.

The fact that iron dissolution from any reactive material increases with decreasing pH value is intuitive (Fig. 2a). However, the extent to which iron is dissolved under any given operational conditions should be characterized (Fig. 2b), and their impact on the investigated process (dye discoloration in this case) discussed. Past efforts to characterize the Fe0/FeS2 system have not properly considered these issues as the final pH value was not always recorded and/or not used in discussing the results22,41. Moreover, in earlier efforts it was commonplace to vary both the initial pH value and the FeS2 loading (Table 5), thereby making it difficult to determine the effect of each parameter. By using only various FeS2 loadings to shift the pH value of a Fe0/sand/H2O system, the present study is an extension of earlier efforts from the early 2000s37,38,39. Moreover, the current study applied a recently developed tool using MB as an indicator of Fe0 reactivity44,45,47.

### Dye discoloration in Fe0/sand/H2O systems

Figure 3a shows the extent of dye discoloration (E values) by the ternary system as the FeS2 loadings increase from 0 to 30 g L−1. Figure 3b depicts the variation of E values as a function of the final pH value. It is evident that there is a general linear decrease in E value with increasing FeS2 loading or decreasing pH values (Fig. 2a). However, three important issues have to be considered: (i) the highest E value for each system corresponds to [FeS2] = 0 g L−1 (Issue 1); (ii) for [FeS2] = 5 g L−1 and [FeS2] > 20 g L−1, there is a significant difference in the E values for both dyes (Issue 2), and (iii) for [FeS2] = 10, 15 and 20 g L−1, there is no profound difference in the E values for both dyes (Issue 3). It is evident that such results would be considered controversial if coming from different independent studies. That is why considering the experimental conditions is crucial in discussing experimental result from independent studies18,23. For the so-called bottle-point technique used herein64, relevant operational variables include: Fe0 pre-treatment, Fe0 particle size (mm, mm, nm), Fe0 mass loading, volume of used vessels, volume of solution, buffer application, mixing type (e.g. stirring, shaking), mixing intensities (e.g. 200 rpm), and experimental duration43. A more detailed discussion of the three issues is given below.

Issue 1 implies that FeS2 addition inhibits the efficiency of Fe0/H2O systems for dye discoloration. Similar results were reported by Noubactep et al.37,38,39 while investigating UVI removal in Fe0/H2O systems. Note that the single-FeS2S2 discolored both dyes (Fig. 1a). This observation raises questions about the assertion that FeS2 increases contaminant removal in Fe0/H2O systems22,33,41.

Issue 2 can be regarded as a striking feature as there is either a larger extent of MO discoloration relative to that of MB ([FeS2] = 5 g L−1) or the opposite ([FeS2] > 20 g L−1). Note that the ion-selectivity principle of the Fe0/H2O system implies that in the presence of FeCPs in aqueous systems, the anionic MO is better discolored than the cationic MB. This is obviously the case at [FeS2] = 5 g L−1 where enough FeCPs is generated to cover the surface of sand, thereby inducing a larger E value for MO than for MB. The observation that for Fe0/H2O systems without FeS2 (i.e., at [FeS2] = 0 g L−1), MO discoloration was only slightly higher than that of MB is also noteworthy. This indicates that for the experimental duration used in the current study (41 d), enough FeCPs were generated to co-precipitate both dyes. For further clarification of this issue, a binary Fe0/sand system should have been investigated, but this was beyond the scope of the current study. The higher MB discoloration relative to MO observed at [FeS2] > 20 g L−1 is explained by the formation of complexes between Fe and MO which delay co-precipitation55,61,65.

Finally, Issue 3 can also be regarded as a striking observation because despite all differences (solubility, affinity), there is no difference in the E values for the two dyes. It can be assumed that, under the experimental conditions, MB and MO which have almost the same molecular size (Table 1) are both discolored by co-precipitation66. This assumption is corroborated by results in Fig. 3b showing clearly that there is no quantitative dye discoloration (E > 60%) at pH < 4.5. This corresponds to the observations of Noubactep et al.37,38,39 for the Fe0/UVI/H2O system. The fact that MB, MO and UVI exhibited very similar behaviors in the Fe0/FeS2/H2O system is an indication that contaminant removal might be a pure positive side effect of aqueous iron corrosion. The most tangible proof for this assertion is the kinetics of Fe2+ oxidation by dissolved oxygen (O2). According to Langmuir67, the kinetics of this reaction increases by a factor 65 between pH 4.0 and 5.0. Thus, quantitative dye discoloration is observed only in systems where Fe2+ oxidation to Fe3+ was quantitative for the 41-d experimental period. The in-situ generated FeIII precipitates are good contaminant scavengers.

### Mechanisms of contaminant removal in Fe0/H2O systems

This study has investigated the effect of FeS2 addition on the efficiency of Fe0/sand systems for MB and MO discoloration. No enhanced dye discoloration could be attributed to FeS2 addition at mass loading of 0 to 30 g L−1 for 41 d. Two questions arise: First, why is there no increased dye discoloration in a context where the expected pH shift and increased iron dissolution are evident? (Question 1). It is noteworthy that each individual aggregate (Fe0, FeS2, sand) tested herein can achieve MB discoloration as depicted in Fig. 1a. Second, why did the ternary system perform far lower that single- Fe0 systems? (Question 2). By applying a known experimental approach consisting of varying individual operational parameters to better understand complex systems17,68,69,70, and accounting for the relative slow kinetics of Fe0 and FeS2 dissolution37,38,39, this study has adopted a novel approach to answer Questions 1 and 2. Specifically, the current study assessed the role of FeS2 in enhancing contaminant removal in Fe0/H2O system. MB is used herein as an operational reactivity tracer (“Introduction”) and the achieved results corroborated earlier reports on U(VI) removal37,38,39, and account for discrepancies and inconsistencies reported in literature33,41,49,70.

The evidence that FeS2 oxidation produces acidity (Eq. 2) is corroborated in the current study (Fig. 2a). By consuming acidity, Fe0 (Eq. 1) and Fe2+ (Eq. 3) oxidation are accelerated by Eq. (2) (Le Chatelier’s principle). Fe3+ from Eq. (2) catalyses FeS2 oxidation and produced less soluble Fe(OH)3. Thus, mixing Fe0 and FeS2 can be regarded as continuously generating less soluble Fe(OH)3, until one of the reactants is depleted or until a pseudo-steady state is established. This work posits that Fe(OH)3 discolors the dye solutions mainly by co-precipitation66,70. Thus, dye discoloration is only quantitative when Fe(OH)3 precipitation is intensive (pH > 4.5). Having used quiescent systems, various final pH values could be achieved, thereby confirming the pH shifting function of FeS237,41,59. However, the extent of dye discoloration depends on the amount of free in-situ generated Fe(OH)355,70 which is determined by the kinetics of Fe2+ oxidation by dissolved O267. As expected, for a longer experimental duration (t > 41 d), the efficiency of the ternary mixture will surpass that of the single- Fe0 systems37,38,39,70. This answers Question 1, and demonstrates that enhanced dye discoloration needs more time to occur under quiescent conditions37,38,39. Accordingly, the documented delay of quantitative dye discoloration is not a negation of the view that FeS2 addition enhances the efficiency of Fe0/H2O system33,40,41,70. This study aims to better understand why Fe0/H2O systems are more efficient upon the addition of pyrite (FeS2) relative to those without pyrite.

In a ternary Fe0/FeS2/sand system, sand is non-reactive (inert) and is in-situ coated by iron oxides from the dissolution of the two other aggregates (“Background to the experimental methodology” and Table 2). This in-situ coating of sand delays the availability of free Fe(OH)3 for dye co-precipitation. Initially, MB and Fe2+/Fe3+ compete for adsorptive removal at the negatively charged sand surface70,71. Once the sand surface is completely coated, it will be no longer attractive for MB. This competition for active adsorption sites explains the observations in Fig. 3a. Note that neither Fe0 nor FeS2 are the discoloring agents, but rather the products of their oxidative dissolution which are variably available in the investigated systems (Table 3). To completely answer Question 2, the ternary mixture performed less than the single-aggregate systems because: (i) sand is in-situ coated, thereby retarding the availability of free Fe(OH)3, and (ii) the synergy of Fe0 and FeS2 has not yet produced enough free Fe(OH)3. The latter is the case whenever the pH value of the system has not exceeded 4.5 (Fig. 3b).

The presentation until this point has not addressed the redox properties of MB and MO. The thermodynamics predict MO reduction by Fe055,61. The results reported herein demonstrate that even the ion-selective nature of the individual dyes was not the key factor accounting for dye discoloration when the pH was lower than 4.5. Thus, regardless of any redox properties, the current work has demonstrated that Fe0-based systems are only efficient when the final pH value is larger than 4.5. Unlike the current study, several previous works have mostly failed to record the final pH values of their systems and use them in their discussion (Table 5).

### Significance of the results

Fe0-based systems have been important components of the water treatment industry for the past 170 years8,27,70. Research reported before 1990 is not really considered by current active scientists whose starting point is the advent of in-situ permeable reactive barriers (PRBs), and the premise that Fe0 is an environmental reducing agent18,23,72,73. Conventional PRBs use micro-scale or granular Fe0 specimens (g Fe0). During the past two decades, some tools have been developed to improve the efficiency of g Fe0. In this regard, the following three tools have been introduced: (i) using nano-scale Fe0, (ii) alloying g Fe0 with metals such as Pd or Ni (also at nano-scale), and (iii) admixing another aggregate with g Fe011,24,25. The Fe0/FeS2/sand system investigated herein is part of the third category. It has been reported that in sulfide-containing environments, using g Fe0 results in the formation of iron sulfides which are conductive and sustain electron transfer from Fe0 to the contaminant22,74,75,76. On the other hand, such iron sulfides are stand-alone reducing agents for the reductive transformations of many contaminants70,77,78. Because Fe0 and FeS2 have in common the release of FeII species, it can be assumed that the material containing more Fe will be first passivated by FeIII species. However, when both materials are mixed, FeS2 accelerates Fe0 corrosion and none of both materials is really available for quantitative reductive transformation of other foreign species, including contaminants. Consequently, any observed enhancement of contaminant removal in a Fe0/H2O system by virtue of the presence of FeS2 is an indirect process. This assertion was elegantly demonstrated in the present study by slowing down the process of iron precipitation via addition of various FeS2 doses to the same Fe0/sand system for 41 d. It then follows that, FeS2 is mostly a pH shifting agent for the Fe0/H2O system37,59,70.

Table 5 reveals that all other investigations on the Fe0/FeS2 system were performed under shaken/stirred conditions. However, under such conditions, the target FeS2 intrinsic properties (including semi-conduction) are undermined. For example, how can FeS2 act as a ‘mediator’ for electron transfer’ from Fe0 to contaminants (Fig. 4) when the whole system is mechanically stirred at 400 rpm? Such a high stirring speed was explicitly selected to ensure that both Fe0 and FeS2 could be uniformly dispersed in the reaction solution22. This example clearly shows that using FeS2 to enhance the efficiency of Fe0/H2O systems is a simple tool to design more sustainable Fe0-based systems. However, current rationalization efforts are not really based on scientific principles22,33. Thus, only when the scientific principles are well-understood can better systems be designed8,27,70. A typical design problem is how to cope with the increased Fe0 dissolution specifically in column operations intrinsically prone to clogging79,80. Thus, in solving the enigma of the Fe0/FeS2/H2O system, this work leads to several avenues for sustaining the efficiency of conventional Fe0/H2O remediation systems. This result is especially important as Fe0-based (filtration) systems are an excellent candidate to help the international community to solve the long-lasting issue of universal safe drinking water8,14,15,27,29,30,81,82.

A further argument against the electrochemical nature of FeS2 in mediating electron transfer from Fe0 to contaminants (Fig. 4) is given by recent investigations in efforts to suppress FeS2 oxidation under environmental conditions83,84,85,86. For example, Seng et al.84 reported that Fe0 is able to stop FeS2 oxidation, and thus remediate acid mine drainage. In essence, Seng et al.84 investigated a contaminant-free Fe0/FeS2/H2O system (Table 6) and concluded that from the intrinsic properties the addition of Fe0 selectively suppress pyrite oxidation. Table 6 shows that the experimental conditions of Seng et al.84 are very close to those of remediation Fe0/FeS2/H2O systems. The only two distinct differences are: (i) the higher FeS2 mass loading (FeS2: Fe0 = 10), and (ii) the longer experimental duration (41 days versus < 10 h). By using an even longer experimental duration (41 days) and quiescent conditions (0 rpm), the present work has demonstrated the essential virtue of working under near-field conditions. In other words, it is fair to state that the Fe0/FeS2 literature is full of possibly reproducible results, but with low practical value. As already shown in Table 5, the variability of the operational conditions is a major issue and the significance of results of solid phase characterization is questionable. In fact, as seen in Table 6, a myriad of characterization tools were used to “confirm” the reducing properties of Fe0 for dyes33. In such studies species like methylene blue70 used herein as a ‘tracer’ of reactivity or arsenic22, and proven to be non-reducible in Fe0/H2O systems are quantitatively removed. The first merit of the MB method is to uncover these controversial views without solid phase analysis.

The conclusion of Seng et al.84 supports the view presented herein that the relative kinetics of Fe0 and FeS2 oxidation determinate the preponderance of processes in Fe0/FeS2/H2O systems70. However, the reported selectivity of the process is questionable as sand and other natural minerals are also covered with FeCPs under similar conditions29,87,88,89. As an example, Song et al.87 reported on increased CrVI reduction in Fe0/sand/H2O systems compared to Fe0/H2O ones. The extent of coating of each aggregate (e.g. gravel, peat, pyrite, sand) depends on both the intrinsic reactivity of used Fe0 and the relative proportion of available materials. In the light of the kinetic arguments given herein, a re-evaluation of published works is possible, for example, the data of Sheba et al.86 discussing the extent of degradation of chlorinated organic compounds (RCl) by Fe0/FeS2/H2O systems and reporting on differential mechanisms at different Fe0:FeS2 ratios. The discussion given herein clearly suggests that if there are differential removal mechanisms, it is due to the differential extent of pH shift. Future research should be designed based on the chemistry of the systems89.

## Concluding remarks

The concept that adsorption and co-precipitation are the fundamental mechanisms of contaminant removal in Fe0/H2O systems is consistent with many experimental observations. In particular, quantitative dye discoloration was only observed for pH values corresponding to iron precipitation (hydroxide formation) (pH > 4.5, Fig. 3b), while selective dye discoloration promoted adsorptive removal. Further, while the role of the redox-mediated reactions in the discoloration of both dyes can only be speculatively discussed based on one’s results, it is established that the role of FeS2 is as follows: (i) shifting the pH to more acidic values, and (ii) enhancing contaminant removal by adsorption and co-precipitation during the subsequent pH increase by virtue of iron corrosion. Finally, negatively charged methyl orange (MO) showed no significant increase in discoloration relative to positively charged methylene blue (MB). Both MB and MO have a similar molecular size. This observation is consistent with the role of Fe0 as a generator of contaminant scavengers, and not as a reducing agent. This observation could explain why various As (AsIII and AsV)90 or Se (SeIV and SeVI)91 species are quantitatively removed in Fe0/H2O systems, but not by aged iron oxides. Further research is needed to investigate the phenomena highlighted in the current study using a wide range of contaminants commonly occurring in drinking water and wastewaters.