Research on catalytic denitrification by zero-valent iron (Fe0) and Pd-Ag catalyst

This study primarily focused on how to effectively remove nitrate by catalytic denitrification through zero-valent iron (Fe0) and Pd-Ag catalyst. Response surface methodology (RSM), instead of the single factor experiments and orthogonal tests, was firstly applied to optimize the condition parameters of the catalytic process. Results indicated that RSM is accurate and feasible for the condition optimization of catalytic denitrification. Better catalytic performance (71.6% N2 Selectivity) was obtained under the following conditions: 5.1 pH, 127 min reaction time, 3.2 mass ration (Pd: Ag), and 4.2 g/L Fe0, which was higher than the previous study designed by single factor experiments and orthogonal tests, 68.1% and 68.7% of N2 Selectivity, respectively. However, under this optimal conditions, N2 selectivity showed a mild decrease (69.3%), when the real wastewater was used as influent. Further study revealed that cations (K+, Na+, Ca2+, Mg2+, and Al3+) and anions (Cl-, HCO3-, and SO42-) exist in wastewater could have distinctive influence on N2 selectivity. Finally, the reaction mechanism and kinetic model of catalytic denitrification were further studied.


Introduction
Contamination with nitrate (NO 3 -) in water resource has attracted increasing public concern. Nitrate detected in water body is a common contaminant that may cause severe health risks, such as blue baby syndrome, cancer, as well as the eutrophication of water bodies [1]. Agricultural activities (mainly the over-fertilization of nitrogenous fertilizers), atmospheric deposition, and sewage discharges mainly contribute to nitrate pollution [2].
Several technologies have been developed for treatment of nitrate-contaminated water, including physico-chemical denitrification (such as ion exchange, reverse osmosis, chemical precipitation, and electrocoagulation), biological treatment, and chemical reduction [3]. Among these approaches, biological denitrification, and catalytic hydrogenation enable to selectively reduce nitrate to nontoxic nitrogen (N 2 ) [4,5]. However, the biological method requires intensive maintenance, excessive biomass disposal, and constant addition of carbon resources [6]. In recent years, the technology of chemical catalytic reduction of nitrate attracts  [7]. In this catalytic process, catalyst plays the indispensable role, while H 2 has been regarded as the reductant, which provides the active H that can participate in the deoxidation process of the nitrate reduction. However, the low solubility of H 2 in aqueous media and the operational complexity (appropriate H 2 flow rate, pressure) have been the big problem [8]. Several researchers replaced H 2 with organic acid (e.g., HCOOH) or its salt (e.g., NaCOOH) to convert nitrate to N 2 [9]. However, the incomplete decomposition of acid or its salt and the threat to human health greatly restricts its wide application. Based on these, the novel synergistic effect of zero-valent iron (Fe 0 ) and bimetallic catalyst for nitrate reduction was proposed. The experimental design for evaluating and optimizing experimental parameters can minimize costs and maximize desired responses [10,11]. For most researchers, the single factor experiments and orthogonal tests have been widely used for experimental design. However, these two methods are incapable of getting true optimal conditions due to ignoring the interactions among influential variables [12]. Therefore, instead of these two methods, Response Surface Methodology (RSM) was utilized for the optimization of catalytic denitrification conditions in this paper. RSM is a particular set of mathematical and statistical approach that develops for building models, evaluating the effects of variables, and determining the optimal conditions of variables [13]. This method contributes to completing the comprehensive design with a minimum number of experiments, analyzing the interaction between the parameters, and more directly and accurately obtaining the optimal operation parameters [14].
Actually, until now, RSM has not been used as an optimization tool for catalytic reduction of nitrate. Hence, in this research, as a design framework in RSM, Box-Behnken Design (BBD) was used to model and optimize the processes of catalytic denitrification achieved by zerovalent iron (Fe 0 ) and Pd-Ag catalyst. Finally, the reaction mechanism of catalytic denitrification was comprehensively illustrated.

Experimental design
Batch experiments were completed to investigate the potential factors that may impact catalytic performance. All tests were performed in a 1 L plexiglas reactor (Fig 1). Certain amounts of Fe 0 and catalysts were added to the reactor prior to the experiments. To guarantee the better mass transfer effect for catalytic denitrification, the reactor was placed on an magnetic stirrer under 450 rpm at room temperature (20±5˚C). 1 mol/L HCl was added to reactor by one automatic titrator to remain needed solution pH during catalytic process.
Samples were periodically collected to determine the concentration of nitrate-nitrogen ( The N 2 selectivity was calculated as: Where C 0 is the initial nitrate concentration (mg/L), C t is the nitrate concentration (mg/L) at time t (min), C N 2 is the content of N 2 (mg/L).

RSM analysis
(1) Box-Behnken design (BBD) BBD was used for experimental design. The levels of BBD were shown in Table 1.
(2) Regression equation fitting and analysis of variance (ANOVA) Minitab 19 was applied to the multiple regression fitting. The experiments were conducted and the quadratic multinomial regression equation was listed as follows, and the regression equation coefficients and T test can be seen in Table 2:
As exhibited in Table 3, P-value = 0.000 <0.01, R 2 = 90.47, which prove the model built above accurately and the regression equation obtained has been better fitted [17]. Therefore, it comes to the conclusion that this model can be used to continuously analyze and predict experimental data.
In addition, in order to validate the model proposed above, the residual plots were checked, listed in Fig 2. It's believed that randomness and unpredictability are essential components for any valid regression model. Through the residual plots analyses, whether the observed error (residuals) is consistent with stochastic error can be accurately assessed. The residuals should be centered on zero throughout the range of fitted values indicated in Fig 2B and 2D. Random errors assumed to produce residuals should be normally distributed. In other words, the residuals should fall in a symmetrical pattern and have a constant spread throughout the range which can be proved in Fig 2A and 2C.

PLOS ONE
(3) 3D response surface analyses 3D response surface analyses were further conducted for four factors, including pH, time, Pd:Ag mass ratio, and Fe 0 dosage, which can be seen in Fig 3. Response surface and contour plots have been applied to intuitively indicate the influence of various factors on N 2 selectivity, so as to find out the optimal parameters and the interaction between the factors [18]. In the contour plots, the central point of the minimum ellipse is the highest point of the response surface. Additionally, the shape of the contour line can reflect the strength of the interaction, and the oval indicates that the interaction between the two factors is significant, while the circle reflects the opposite meaning.
As depicted in Fig 3A, compare with others, response surface and contour plots of X 1 and X 2 on N 2 selectivity show the significant influence trend, which is consistent with the data in Table 4. In order to obtain the predicted maximum value through the model we build, the canonical analysis of response surface was conducted, which was listed in Table 4.
As indicated in Table 4, the predicted maximum value is 69.8%. The actual values of the four factors (X 1 , X 2 , X 3 , and X 4 ) obtained from the coded value are: 5.1 pH, 127 min time, 3.2 Pd: Ag, and 4.2 g/L Fe 0 , respectively, which are the predicted optimal parameters.

PLOS ONE
Removing nitrate in water body by zero-valent iron (Fe 0 ) and Pd-Ag catalyst (4) Validation test The validation experiments were conducted under the predicted optimal parameters: 5.1 pH, 127 min reaction time, 3.2 mass ration (Pd: Ag), and 4.2 g/L Fe 0 . Results showed that the N 2 selectivity of catalytic denitrification reached 71.6%, higher than the study designed by the single factor experiments (68.1%) and orthogonal test (68.7%) in Table 5, which proves that the model used in this research is accurate and can get the true optimal conditions for the catalytic reduction of nitrate.

Simulation experiments of real wastewater
To test the effect of water quality on N 2 selectivity, real wastewater obtained from the secondary effluent of a municipal wastewater treatment plant in Beijing, China, was adopted for batch experiments. The properties of water samples were: concentration of NO As described in Table 6, compared to the artificial solution (NaNO 3 ) as influent, N 2 selectivity showed a mild decrease as the real wastewater was used as the influent. This phenomenon may be due to the ions that exist in wastewater. Therefore, the effect of the ions on catalytic performance was further investigated. A series of experiments with various nitrate salts as the source of nitrate ions revealed that the catalytic performance increased in the following order: K + < Na + < Ca 2+ < Mg 2+ < Al 3+ . It has been reported that these cations have different influence on the migration rate of NO 3 and OHin solution [19]. Cations with high valence or small radius seem more likely to have a strong ability to bond with NO 3 -, preventing NO 3 from catalytic reduction. Similarly, the cations in solution tend to strongly adsorb the formed OHthat may have a negative impact on catalytic denitrification, enhancing the separation of OHfrom bimetallic active sites on surface of the catalyst and offering suitable space and conditions for a the catalytic reaction [20]. As depicted in Fig 4B,   to adverse influence on nitrate reduction [20]. In contrast, due to the different structure, Cland SO 4 2both had little to do with catalytic nitrate reduction [21].

Reaction mechanism
(1) Role of the reductant-Fe 0 Fe 0 primarily served as electron donor in catalytic process. In general, the catalytic denitrification involved the directional electron transfer from Fe 0 to nitrate, which is then converted into nontoxic N 2 or less toxic species (NO 2 and NH 4 + ) [22]. In practical terms, at the metal active sites at the surface of carrier, the electron that Fe 0 lost could bond with H + in solution and form active H, which took part in the deoxidization process and reduced NO 3 -, as shown in Fig 5A. XRD patterns of Fe 0 before and after catalytic reaction were exhibited in Fig 5B. It's obvious to find that magnetite (Fe 3 O 4 ) and hematite (Fe 2 O 3 ) were detected on surface of Fe 0 , which is consistent with the Schlicker's finding [23]. The possible reaction equations are listed as below:

PLOS ONE
Removing nitrate in water body by zero-valent iron (Fe 0 ) and Pd-Ag catalyst (2) Catalytic denitrification process It's believed that the catalytic reduction of nitrate has been the stepwise processes. As indicated in Eqs (Eq 5-11) [24], H + receives the electron from Fe 0 , forming the active H, which takes part in the deoxidization process, converting NO 3 to N species (NO 2 -, NH 4 + , or N 2 ) [25]. It's worth noting that more N 2 can be produced, only the appropriate H + concentration in solution has been remained. High H + concentration may lead to the generation of undesired NH 4 + , which has to be treated again. Additionally, H + can also reduce the accumulation of OHgenerated with the catalytic processes.
In the catalytic denitrification processes, catalyst composed of the active ingredients and the carrier significantly influences the catalytic performance [26]. The carrier that supports the active ingredients can provide reaction sites for catalytic reaction [27]. In addition, the physico-chemical properties (pore structure, surface area, mechanical strength, and the chemical components) of the carrier determine the dispersion degree of the supported active metal particles (Pd, Ag) that control the processes of adsorption, diffusion, reaction, and desorption of the reactants (mainly NO 3 -, NO 2 -) and the products (mainly NH 4 + , N 2 ) that occurred on the catalyst's surface, which may greatly affect the catalytic reduction of nitrate [28]. Therefore, the

PLOS ONE
Removing nitrate in water body by zero-valent iron (Fe 0 ) and Pd-Ag catalyst materials that possess the porous structure, larger specific surface area, good adsorptive capacity, and stable physico-chemical properties tend to be selected as the carrier of the catalyst.
In addition, the active ingredients can affect the catalytic performance by directly and indirectly participate in the catalytic reaction. Research found that the active ingredients loaded on the carrier should better comprise of a noble metal (such as Pd or Pt) and an auxiliary element (such as Ag, Cu or In) [29]. The bimetallic-Pd and Ag can active the formed H, which involves in the deoxidization process to reduce nitrate. Actually, Ag-H mainly acts with the reactant-NO 3 -, producing NO 2 -. Furthermore, on Pd active sites, the product-NO 2 can be continuously reduced to other N species (NO, NH, N 2 , and NH 4 + ) [30]. The catalytic reaction mechanism is illustrated in Fig 6.

Kinetic study
Currently, significant research focuses on the kinetics of catalytic hydrogenation. Rare research on the catalytic process using Fe 0 and bimetallic catalyst to reduce nitrate was conducted. It can be assumed that the zero-order kinetics and first-order equation of Langmuir-Hinshelwood could be employed to describe this process. According to our previous study, the catalytic denitrification process could be better explained by the first order kinetic model [31]. The kinetic equation could be obtained: y = 247.1x +0.1398, R 2 = 0.9975.
It has been suggested that in the process of catalytic denitrification, the produced intermediates such as NO and NH have been negligible [32]. Based on the first-order equation above, the reaction rates are presented in Eqs 12-15. A kinetic study on catalytic denitrification with different catalysts was further conducted, as listed in Table 7.
Where k 1 , k 2 , and k 3 are the rate constants for reduction of NO 3 to NO 2 -, N 2 and NH 4 + , respectively; k 4 and k 5 are the rate constants for reduction of NO 2 to NH 4 + and N 2 .
Results indicated that different catalysts performed distinct reaction rates in catalytic denitrification, which can be explained by k value in Table 7. According to the calculation, for each catalytic process, the summation of k 1, k 2 , k 3 that stands for the overall reaction rate constant was close to k, which implies the catalytic process is a stepwise process. Results indicated that compared to other catalysts, Pd-Ag/graphene showed a higher catalytic rate, which has been proved by data in Table 2. This may be due to the unique properties of graphene, including the porous structure, active surface area, outstanding electronic properties and promising mechanical and thermal stability [33].

Conclusion
Response surface methodology was used to optimize parameters of catalytic reduction of nitrate. Results indicated that the application of response surface methodology was proved to be feasible. 71.6% of N 2 Selectivity was obtained under the optimum conditions: 5.1 pH, 127 min reaction time, 3.2 mass ration (Pd: Ag), and 4.2 g/L Fe 0 . However, the cations (K + , Na + , Ca 2+ , Mg 2+ , and Al 3+ ) and anions (Cl -, SO 4 2-, and HCO 3 -) in water body performed different influence on catalytic denitrification. Study on reaction mechanism found that the catalytic denitrification can be achieved with deoxidization processes. Additionally, as the components of catalyst, active ingredients (Pd-Ag) and carrier (graphene) played different role in the catalytic denitrification. The catalytic process could be better explained by first order kinetic model.