Understanding the Effect of Single Atom Cationic Defect Sites in an Al2O3 (012) Surface on Altering Selenate and Sulfate Adsorption: An Ab Initio Study

Abstract

Adsorption is a promising under-the-sink selenate remediation technique for distributed water systems. Recently it was shown that adsorption induced water network re-arraignment control adsorption energetics on the $\text{α}-{\text{Al}}_2{\text{O}}_3$ (012) surface. Here, we aim to elucidate the relative importance of the water network effects and surface cation identity on controlling selenate and sulfate adsorption energy using density functional theory calculations. Density functional theory (DFT) calculations predicted the adsorption energies of selenate and sulfate on nine transition metal cations (Sc-Cu) and two alkali metal cations (Ga and In) in the $\text{α}-{\text{Al}}_2{\text{O}}_3$ (012) surface under simulated acidic and neutral pH conditions. We find that the water network effects had larger impact on the adsorption energy than the cationic identity. However, cation identity secondarily controlled adsorption. Most cations decreased the adsorption energy weakening the overall performance, the larger Sc and In cations enabled inner-sphere adsorption in acidic conditions because they relaxed outward from the surface providing more space for adsorption. Additionally, only Ti induced Se selectivity over S by reducing the adsorbing selenate to selenite but not reducing the sulfate. Overall, this study indicates that tuning water network structure will likely have a larger impact than tuning cation-selenate interactions for increasing adsorbate effectiveness.

Graphical Abstract

1. Introduction

Selenium is an essential micronutrient for all living beings; however, there is a narrow range between deficiency and toxicity.^1–3 Consumption of selenium above 400 μg/day causes nausea, diarrhea, tachycardia and gastrointestinal disturbances, and induces long-term effects like selenosis, hair loss, abnormal functioning of the nervous system, hepatotoxicity, etc.^4–6 Leaching of selenium from anthropogenic practices such as agriculture irrigation, run-off from mining sites, e-waste landfills and natural reservoirs can cause high Se concentrations in drinking water.^7–9 Once in the water, selenium oxidizes to its oxo-anion forms – selenate (Se(VI) or ${{\text{SeO}}_4}^{2-}$) or selenite (Se(IV) or ${\text{HSeO}}_3^{-}$), which are equally toxic to humans.^10–12 Therefore, water treatment methods are required to decrease Se levels to below the EPA limit of 0.05 mg/L in water ¹³.

Selenium removal methods include ion exchange, photocatalysis, coagulation, electrocoagulation, and bioremediation.^{8, 14} These methods, however, require expensive treatment plants, catalysts, chemical addition, or membranes and continuous management, which are not easily or economically accessible to small communities or private wells.¹⁴ Conversely, adsorption based removal of selenium oxo-anions on low-cost materials can be easily implemented using under-the-sink installations for less than $0.06/m³ treated water.¹⁵ Low-cost metal-oxide nanoparticles like hematite and alumina are preferred candidates as adsorbent materials because they are widely available in nature, safe to use in water remediation, and can be recovered again using chemical-, or thermal-regeneration.^16–19

Aluminum oxide (${\text{Al}}_2{\text{O}}_3$) in activated alumina and nanocrystalline particles forms are EPA approved for use as selenium adsorbents.^{20, 21} However, the adsorption capacity of alumina for selenate is 0.16–4.45 times lower than selenite,^{22, 23} which is problematic since selenate (Se(VI)) exists at higher concentrations than selenite in fresh water.^{15, 24, 25} Moreover, the adsorption of oxo-anions on the alumina surface is sensitive to pH.^{26, 27} At pH 3, 1g of alumina in 200 ml solution adsorbs 100 % of 1.25 mM selenate from solution; however, the adsorption decreases to almost 0% when the pH is 9.²⁸ In-home operators cannot decrease the pH of drinking water below 6 for remediation purposes as it renders the water undrinkable;²⁹ additionally, although acidification, treatment, then neutralizing schemes are possible, they are too complex and expensive to implement in under-the-sink applications. Further, selenate adsorption must compete with less toxic sulphate species in the water, which occupies adsorption sites and thus decreases adsorption capacities.¹⁸ For example, the adsorption capacity of aluminum oxide particles embedded in chitosan beads decreases from adsorbing 80% of the Se in water with 1 ppm selenate to less than 10% when 10 ppm of sulphate is present.²³ The decrease in performance arises because selenate and sulphate have similar structures and chemical properties, like pKa, and water typically has 2–1000 times higher S concentrations than Se^{15, 18}. Therefore, materials design strategies are needed to increase the affinity of Se for adsorption across a range of pH.^{17, 30–32}

In our previous study, we examined the effects of solution pH and oxo-anion adsorption configuration on the adsorption of selenate on (012) $\text{α}-{\text{Al}}_2{\text{O}}_3$ surface using density functional theory calculations.³³ We found that the interaction between selenate and the surface water network is the major controller of selenate adsorption. At low pH, when the surface is covered with excess ${\text{H}}^{+}$, the selenate induces the formation of additional hydrogen bonds within the water network, and hence has more exothermic adsorption energy. At high pH, the number of hydrogen bonds decreases when selenate adsorbs, which leads to endothermic adsorption. Moreover, selenate preferentially adsorbed in an outer-sphere configuration at low pH,³³ which explains the low selectivity of ${\text{Al}}_2{\text{O}}_3$ for selenate adsorption.^{18, 23} These findings beg two questions: 1) is the surface water network the sole controller of selenate adsorption on metal-oxide adsorbents?³⁴ and 2) can interactions between the oxo-anion and surface cations overcome the water network effects?^35–37 The importance of these questions is further highlighted by the recent work of Johnston and Chrysochoou who found that the adsorption mechanism of selenate and sulfate changes from inner-sphere to outer-sphere on naturally occurring Al substituted Fe (hydro)oxides at higher Al concentrations.³⁸ However, they did not discuss the cause of the changing motif or the adsorption energies when aluminum ions are substituted into ferrihydrites. Thus, if we are to design more selective and effective adsorbents, a delineation of the effects of water layers and cation identity effects is required.³⁹

To isolate the role of the adsorbent cation, i.e., ${\text{Al}}^{3+}$ in ${\text{Al}}_2{\text{O}}_3$, on adsorption without the confounding effects of various water networks, we used density functional theory (DFT) to calculate the adsorption energies of selenate on single atom substitutions into the (012) ${\text{Al}}_2{\text{O}}_3$ (X -${\text{Al}}_2{\text{O}}_3$) (Figure 1). Specifically, we examine period 4 transition metals and group IIIA metal substitutions (X: Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Ga and In). Past literature provides evidence that some of the transition metal cations (Fe, Co, Ni and Mn) can be substituted as dispersed ions in the $\text{α}$-phase ${\text{Al}}_2{\text{O}}_3$ at the Al:X ratio of 1.9993:0.0007 without changing the mechanical properties of the materials.⁴⁰ However, we ignored the morphological variations that might occur due to the cationic substitutions to maintain a uniform water network on the surface. To understand the role of interaction between metal-ion and selenate on adsorption energy and mechanism, we compare the adsorption energy on the protonated and neutral surfaces in inner and outer sphere configuration to predict if the interaction between oxo-anion and cation overcomes the controlling effect of water network on the ${\text{Al}}_2{\text{O}}_3$ surface. We also compare the adsorption of selenate with sulphate to explore if $\text{X}-{\text{Al}}_2{\text{O}}_3$ can enable selective adsorption. Finally, we investigate the effect of atomic properties of adsorbent material on the adsorption of selenate and sulphate to delineate material properties that have the most significant correlations with the adsorption energy.

Figure 1:

Bond distances used to calculate atomic radii. The blue, gray, and red spheres represent X, Al, and O respectively.

2. Methods

2.1. First-principle calculations

We performed Density Functional Theory (DFT) calculations in Vienna Ab initio Simulation Program (VASP)^{41, 42} using spin-polarized generalized gradient approximation functional (Perdew-Burke Ernzerhof - PBE)⁴³. Our previous work has shown that this method predicts energies that are in good agreement with the much more expensive HSE06 hybrid density functional, therefore the less expensive PBE functional is used.³³ Implicit solvent effects and non-local van der Waals (vdW) effects are accounted for by adding water dielectric constant in the polarizable continuum model (VASPsol module)⁴⁴ and DFT-D3 correction, based on the method of Grimme et. al⁴⁵. The projected-augmented wave (PAW)^{41, 42} pseudopotentials used in calculations explicitly describe only valence orbitals 1s (H), 2s,2p (O), 3p,3d (Al, S), 3d,4s (Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu), 3d,4s,4p (Ga, Se) and 4d,5s,5p (In) to reduce the computational cost associated with describing the core electrons.

The $\text{α}-{\text{Al}}_2{\text{O}}_3$ crystal was represented by the Al terminated (012) facet ⁴⁶ slab supercell containing 2×2×1 primitive unit cells ($\text{α}-{\text{Al}}_{12}{\text{O}}_{18}$) separated by 45 Å polarized implicit solvent created using pymatgen^{47, 48}. The (012) facet is the largest constituent of gibbsite particles (47%) and has a low surface energy. ^{49, 50} We obtained the crystal lattice constants for relaxed $\text{α}-{\text{Al}}_2{\text{O}}_3$ (a=b= 4.80Å, c=13.09Å, $\text{α}=\text{β}=90°$ and $\text{γ}=120°$), which closely match previously published experimentally results (a=b= 4.76Å, c=12.99Å, $\text{α}=\text{β}=90°$ and $\text{γ}=120°$).⁵¹ The relaxed structure was used to construct the surface. The slab studied here is 13.5 Å thick and is comprised of 8 ${\text{Al}}^{3+}$ layers, containing 4 cations (Figure S1). We fixed the position of bottom 4 ${\text{Al}}^{3+}$ layers along with the oxygen to reduce the computational cost and preserve the bulk crystalline effects of ${\text{Al}}_2{\text{O}}_3$ nanoparticle. The adsorption energy difference of an example ${{\text{SeO}}_4}^{2-}$ on an 8-layer thick slab and 10-layer thick slab is only 0.005 eV, which is well within chemical accuracy; thus, the less expensive 8-layer slab was used.

The surface was modeled by a hybrid explicit/implicit solvation. An explicit monolayer of dissociated water is added on the top surface of ${\text{Al}}_2{\text{O}}_3$ (one ${\text{H}}_2\text{O}$ molecule per surface cation), as shown in figure 2(a), to mimic experimentally investigated interfacial water layer⁴⁶ in neutral environment to develop micro solvation model.⁵² To model the surface below the point of zero charge, an additional proton was added to the dissociated hydroxyl group. In the vacuum above the explicit water, we included implicit solvation to complete the solvation model. Our previous results showed that this hybrid model reproduced a full four layers of explicit water. ⁵² Implicit continuum solvent space in our supercell was partially filled with desorbed selenate, water species and hydronium ions to retain charge parity upon adsorption. Intermolecular distances of at least 5 Å were maintained between desorbed species and surface to eliminate Van der Waals interactions between desorbed ions.⁵³

Figure 2:

(012) $\text{X}-{\text{Al}}_2{\text{O}}_3$ where X (blue) is the impregnated single Atom (Ga- in figure) on surface in (a-b) neutral environment and (c-d) protonated environment. (a) and (c) are side view and (b) and (d) are top view of the neutral and protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface respectively. Side view of selenate (${{\text{SeO}}_4}^{2-}$) adsorbed in (e) monodentate monomolecular (MM), (f) bidentate monomolecular (BM) (g) bidentate bimolecular (BB) and (h) outer sphere configuration (OS) on (012) $\text{α-Ga-}{\text{Al}}_2{\text{O}}_3$. The grey, orange, blue, red, white spheres represent Al, Se, Ga(X), O and H atoms, respectively.

Adsorption energies were calculated between supercells containing the same number of electrons and atoms using the hybrid solvent method (HSM), as outlined in our previous work.³³ In the HSM, we include an explicit water monolayer on the ${\text{Al}}_2{\text{O}}_3$ surface and implicit water dielectric effects between $\text{α}-{\text{Al}}_2{\text{O}}_3$ slabs to mimic the aqueous phase. Extra protons were added to the water monolayer as appropriate to represent the surface below the point of zero charge. The HSM is necessary to (1) include solvation effects on the surface, (2) compare adsorption energies of inner vs. outer sphere configurations, and (3) achieve experimental parity.

To avoid the high computational cost of large unit cells with implicit solvent and Van der Waals interactions, we initially calculated the geometries and energies of 11 different adsorption configurations (5 inner-sphere and 6 outer-sphere) for each surface, i.e., substitution and protonation state, and sorbent, i.e., Se or S, at 350 eV cut off energy and 1×1×1 gamma point centered Monkhorst-Pack k-point mesh. For all in-depth analysis, we examined only the most exothermic geometries of the inner-sphere adsorption (IS) and outer-sphere adsorption (OS) configurations. These configurations were re-optimized at a higher 550 eV cut of energy and a finer 2×2×1 gamma point centered Monkhorst-Pack k-point mesh. The difference in adsorption energy between 350eV, 1×1×1 optimization and 550 eV, 2×2×1 was less than 0.05 eV, which is small compared to the adsorption energy difference in different configurations (> 0.4 eV). We found that there is only a 0.004 eV difference in adsorption energies calculated at 550 eV and 600 eV and only a 0.0005 eV difference between the 2×2×1 gamma-point centered Monkhorst-Pack k-point mesh and a finer 4×4×1 mesh grid; therefore, the less computationally expensive 550 eV and 2×2×1 mesh parameter were used. Atomic geometries were relaxed until total energies between two ionic steps were less than 0.001 eV.

2.2. Adsorption energy calculation

Supercells containing the single atom alloy substituted (X) metal-oxide surface $(X-{Al}_2{O}_3)$, adsorbed water layer^{54, 55} ($\left(8-y\right)O{H}^{-}.\left(n-z\right)H^{+}$, selenate ($SeO{}_4^{2-}$) and desorbed species ($2H_3{O}^{+}+yO{H}^{-}.z{H}_{aq}^{+}$) were used to calculate adsorption energies. The adsorption energy is computed by subtracting the energy of desorbed_cell ($E_{desorbed\_cell})$ from the adsorbed_cell ($E_{adsorbed\_cell}$). The desorbed cell and adsorbed cell have the same number of atoms and electrons, as all desorbed species are explicitly included in solution regions of the supercell above the metal-oxide surface. Thus, all calculated energies arise from super cells containing exactly the same number of atoms and electrons. Example adsorbed_cell and desorbed_cell geometries are illustrated in supplementary information Figure S1. In equations 1–3, the (.) between terms indicate bonded/adsorbed species and (+) indicates species that are separated by at least by 5 Å in the same supercell. We have previously validated this method for calculating adsorption energies using the bare ${\text{Al}}_2{\text{O}}_3$ surface by accurately reproducing adsorption isotherms.³³

Details for all species present in adsorbed_cell and desorbed_cell for all configurations and protonation are listed in Table S1. We only computed adsorption energies of fully deprotonated oxo-anions (${{\text{SeO}}_4}^{2-}$ and ${{\text{SO}}_4}^{2-}$) as they are only species present at pH > 2, i.e., all environmentally relevant pHs.^{56, 57}

for a given configuration was calculated using eq. 1:

$$\Delta E=E_{adsorbed\_cell}-E_{desorbed\_cell}$$ #1

Here $E_{adsorbed\_cell}$ is the energy of simulated cell containing adsorbed selenate.

$${E_{adsorbed\_cell}=E}_{{X-Al}_2{O}_3.\left(8-y\right)O{H}^{-}.\left(n-z\right)H^{+}.Se{O}_4^{2-}+2H_3{O}_{aq}^{+}+yO{H}^{-}.z{H}_{aq}^{+}}$$ #2

And $E_{dedsorbed\_cell}$ is the energy of simulated cell containing desorbed selenate.

$${E_{desorbed\_cell}=E}_{{X-Al}_2{O}_3.8O{H}^{-}.n{H}^{+}+Se{O_4^{2-}}_{aq}+2H_3{O}_{aq}^{+}}$$ #3

The variable used in eq 2 and 3 change upon the adsorption configuration (y,z), surface protonation (n) and single atom (X) and values are listed in Table 1.

Table 1:

Values of variables used in eq 2 and 3:

Factors		Variables	Values
Single atom impregnated on [012] $\text{α}-{\text{Al}}_2{\text{O}}_3$		X	Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Ga, In
Surface protonation	Protonated surface (PS)	n	12
	Neutral Surface (NS)		8
Adsorption configuration	Bidentate Bimolecular (BB)	y	2
		z	2(PS), 0(NS)
	Monodentate Monomolecular (MM)	y	1
		z	1(PS),0(NS)
	Bidentate Monomolecular (BM)	y	1
		z	1(PS)
	Outer sphere (OS)	y	0
		z	0

In this work we examine adsorption thermodynamics through adsorption energy lens, and neglect entropic effects. However, given a drastic difference in enthalpic adsorption energy (on the order of −2 to 1 eV with ~0.5 eV difference between configurations), we expect the entropic effects will only slightly modify the thermodynamics as the entropy of water solvation is only ~ 0.1 eV.⁵⁸ Moreover, the substitution of one cation for another in the surface should not significantly affect the entropy of water on the surface, water desorption, or oxo-anion adsorption. Therefore, because this study focuses on understanding the effect of cation on adsorption, we did not perform the computationally expensive entropic calculations.

Density of States (DOS), and charge density differences were calculated using vaspkit⁵⁹, and Crystal Orbital Overlap Populations (COOP) was calculated using LOBSTER^{60, 61} software. Crystal structure representations were obtained using VESTA.⁶²

2.3. Ionic radius calculations

We calculated the atomic radii of the single atoms embedded in the ${\text{Al}}_2{\text{O}}_3$ surface by averaging the inter-atomic distance of the single atom with its coordinating surface O atoms (O1-O4) (Figure 1) and subtracting the ionic radii of ${\text{O}}^{2-}$ (1.4 Å).⁶³ The atomic distance with sub-surface oxygen (O5) is not used to calculate atomic radii as this bond partially dissociated when the dispersed atom was much larger than Al and protruded out of the ${\text{Al}}_2{\text{O}}_3$ plane to reduce steric hindrance. We chose to use our calculated atomic radii rather than using literature values as ionic radii changes with their oxidation state.⁶³

3. Results

We modeled the adsorption of selenate and sulphate on single atom substituted [012] ${\text{Al}}_2{\text{O}}_3$ ($\text{X}-{\text{Al}}_2{\text{O}}_3$) where we substituted one in every 8 Al atoms present on the surface to elucidate the effects of the cation on adsorption. Substitutions included Period 4 transition metal (Sc-Cu) and Group IIIA metals (Ga, In). We found that Ti, V, Mn, Fe, Co, Ni and Cu are spin polarized while Sc, Cr, Ga and In only have paired electrons. Only the lowest energy spin configurations are reported in this work.

We assume that the all substitutions are in the dilute limit, where secondary phase formation will not occur, to elucidate the relative effects of cation-selenate interaction and water network-selenate interactions. Experimental work has previously shown that Fe, Co, Ni and Mn can stability substituted at low Al: X ratios (1.9993:0.0007).⁴⁰ The fact that the cations are fully relaxed demonstrates that they are at least in a local minima. Further, COOP analysis shows that the substitutions are strongly bond to their neighboring O anions (Figure S6). We do, however, note that In, Ga, Sc, Ti, V and Cr are expected to be more stable than the Mn, Fe, Co, Ni and Cu substitutions, as the former do not have anti-bonding character, while the later set does have some filled anti-bonding states.

Additionally, we note that several of the cations investigated here can adopt multiple oxidation states or have preferred oxidation states which differ from the +3 charge of Al. DOS analysis, see figure S8, show that Ti, V, Cr Mn, Fe, Co, and Ni induce mid-gap states localized on the cations, while Sc, Cu, Ga and In show slightly unfilled O 2p states. The missing O 2p state arise from the covalent nature of the metal-O bond, and the partial localization of the bonded electron on the cation. The charge localization is further quantified by Bader analysis in Table S6 and shows that the charge density on the neighboring Os’ are modified by ~1.2–0.7 ${\text{e}}^{-}$ depending on the substitution. However, even though the oxidation state of single atom substitution is not exactly +3, the electrons are still localized near single atoms on the neighboring O, and not transferred to the surface water network to form +4 or other oxidation state. This additional charge on the cations relative to Al, or the missing charge on the neighboring O atoms may alter the local point of zero charge, and thus protonation extent at a given system pH. Therefore, we examine both the protonated and neutral surface; we leave prediction of local protonation extents due to substitutions to future work.

We calculated the adsorption energy of selenate on the neutral (pH >7.8) and protonated (pH <~5.5–6) ${\text{Al}}_2{\text{O}}_3$ surface, which are shown in Figure 2a-d. The neutral surface is covered by 8 dissociated water molecules (${\text{Al}}_2{\text{O}}_3.{\text{8OH}}^{-}.{\text{8H}}^{+}$) before adsorption, while the protonated surface has 8 water molecules and 4 protons (${\text{Al}}_2{\text{O}}_3.{\text{4H}}_2{\text{O.4OH}}^{-}.{\text{8H}}^{+}$). The 4 extra protons on the protonated surface bond with the original surface ${\text{OH}}^{-}$ to form ${\text{H}}_2\text{O}$ molecules. The protonated surface represents the system below the point of zero charge, 7.8²⁸ for the (012) ${\text{Al}}_2{\text{O}}_3$ surface; at environmentally relevant conditions the surface may or may not be protonated, but the selenate and sulfate species are fully deprotonated. All charged supercells are compensated with background jellium charge; the large unit cells used here (~5600 ų) minimize the jellium charge in any location. The size effects of the jellium were validated in our previous study.³³ We initially screened one mono-dentate monomolecular (MM), one bidentate molecular (BM) and three bi-dentate bimolecular (BB) inner sphere configurations, and six outer-sphere configurations (two each of monodentate, bidentate and tridentate to the adsorbed waters). Exemplar MM, BM, BB, and OS adsorption configuration on Ga-${\text{Al}}_2{\text{O}}_3$ are shown in Figure 2. We do not discuss BB adsorption energy as it was always significantly more endothermic than MM and BM adsorption energies (Table S2 and S3). The same preference for MM, BM and OS adsorption configurations exist for sulfate.

3.1. Adsorption of selenate on single cation embedded in the [012] $\text{X}-{\text{Al}}_2{\text{O}}_3$

We calculated the adsorption energy of selenate on $\text{X}-{\text{Al}}_2{\text{O}}_3$ as a function of adsorption motif and protonation state of the surface. The results, as summarized in Figure 3, show only the most exothermic configuration identified for the given system. We found that the adsorption of selenate was exothermic on protonated surface and endothermic on neutral surface for all $\text{X}-{\text{Al}}_2{\text{O}}_3$ surfaces. On the neutral surface, the adsorption energies remained similar in magnitude to the unsubstituted ${\text{Al}}_2{\text{O}}_3$ surface, and thus in most cases did not result in a reordering the preferred binding motif. Some cations substitutions, protonated In, Sc, Fe, and neutral Ti, altered the preferred binding from outer sphere to inner-sphere adsorption. However, the adsorption in most situations was less exothermic than on the unsubstituted ${\text{Al}}_2{\text{O}}_3$ surface, thus we saw no significant positive influence by adding dispersed single atoms. The monomolecular adsorption configurations (MM and BM), where selenate replaced one water molecule, was always more exothermic than bimolecular adsorption (BB), where two water molecules are replaced by single ${{\text{SeO}}_4}^{2-}$ molecule. However, the adsorption energy did change in some cases by < ± 0.5 eV from the unsubstituted [012] $\text{α}-{\text{Al}}_2{\text{O}}_3$ surface, usually destabilizing adsorption. In rest of this section, we describe the behavior of adsorption trend across period 4 transition metal substituted and group IIIA metal substituted ${\text{Al}}_2{\text{O}}_3$ surface in (a) protonated and (b) neutral environments.

Figure 3:

Adsorption of ${{\text{SeO}}_4}^{2-}$ on (a) Protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface and (b) Neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface. The adsorption of selenate on [012] $\text{α}-{\text{Al}}_2{\text{O}}_3$ in MM, BM and outer sphere are depicted by dashed lines. BM on neutral surface relaxes in MM configuration and hence not added in (b).

3.1.1. ${{\text{SeO}}_4}^{2-}$ adsorption on the protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ (012) surface

On protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$, we calculated that outer sphere adsorption energies range between −1.82 eV (Ti-${\text{Al}}_2{\text{O}}_3$) and −1.32 eV (Cu-${\text{Al}}_2{\text{O}}_3$) and monomolecular inner-sphere configurations range from −1.83 eV (In-${\text{Al}}_2{\text{O}}_3$) to −0.96 eV (Cu-${\text{Al}}_2{\text{O}}_3$). We categorize the substitutions elements into those which prefer the inner-sphere configuration (In, Sc and Fe) and those which prefer outer-sphere adsorption (Ti, V, Cr, Mn, Fe, Co, Cu, Ni and Ga). The inner-sphere configuration is the energetically favored structure on In-${\text{Al}}_2{\text{O}}_3$ (BM, −1.83 eV), Sc-${\text{Al}}_2{\text{O}}_3$ (BM, −1.77 eV), and Fe-${\text{Al}}_2{\text{O}}_3$ (MM, −1.66 eV) as compared to outer-sphere adsorption.

Adsorption of ${{\text{SeO}}_4}^{2-}$ in the BM configuration on protonated In-${\text{Al}}_2{\text{O}}_3$ (−1.83 eV) has the most exothermic energy across all configurations and surfaces, including unsubstituted ${\text{Al}}_2{\text{O}}_3$. We attribute the BM preference of In and Sc to distortions in its coordination at the surface. The ionic radii of the embedded ${\text{In}}^{3+}$ (0.76 Å) and ${\text{Sc}}^{3+}$ (0.68 Å) are substantially larger than that of their adjacent ${\text{Al}}^{3+}$ (0.51 Å); and thus, In and Sc atoms protrude from the [012] ${\text{Al}}_2{\text{O}}_3$ surface, as shown in Figure 4, to minimize steric overlap with the in-plane O anions. We note that the ionic radii are calculated here using method described in section 2.3 and not from literature values. The extension outward also weakens the sub-surface X-O bond. From COOP analysis, we found that for Sc and In the O5-X bond strength (0.02 eV/bond for Sc and 0.03 eV/bond for In, labeled in Figure 1) is ~50% less than the average bond strength between surface O-X (O1-O4)(0.07 eV/bond for Sc and 0.09 eV/bond for In) and 6% less (0.15 eV/bond between O5-Al) than the average bond strength for Al (0.16 eV/bond). The resulting undercoordination enables bidentate adsorption of selenate on Sc and In cations. The radius of In is so large, and thus so extended from the surface, that without the adsorbed water it is undercoordinated; consequently the MM configuration on In-${\text{Al}}_2{\text{O}}_3$ is unstable and relaxes into BM configuration, providing a more complete coordination. The formation of the BM bond was verified through COOP analysis, which predicts two InO-Se bonds with strengths of 0.06 eV/bond.

Figure 4:

Protrusion of the In from the protonated ${\text{Al}}_2{\text{O}}_3$ surface before adsorption. The grey, purple, red, white spheres represent Al, In, O and H atoms, respectively.

Contrary to Sc and In, only the MM configuration of Fe-${\text{Al}}_2{\text{O}}_3$ is more exothermic, by −0.41 eV, than the outer-sphere configuration. This configuration is −0.04 eV more exothermic than inner-sphere adsorption on unsubstituted ${\text{Al}}_2{\text{O}}_3$, but 0.32 eV less stable than outer sphere adsorption on ${\text{Al}}_2{\text{O}}_3$. The ${\text{e}}^{-}$ density charge transfer between ${\text{Al}}_2{\text{O}}_3$ surface and selenate is similar when selenate adsorbs on ${\text{Al}}_2{\text{O}}_3$ and Fe-${\text{Al}}_2{\text{O}}_3$. Therefore, the preferential adsorption of selenate on Fe-${\text{Al}}_2{\text{O}}_3$ in the MM configuration is due to weaker adsorption in the outer sphere configuration, rather than the Fe stabilizing inner-sphere selenate adsorption.

Selenate adsorption at the Ti, V, Cr, Mn, Co, Cu, Ni and Ga sites maintain an outer sphere preference. In all cases except Ti the adsorption is less exothermic than unsubstituted ${\text{Al}}_2{\text{O}}_3$(Figure 3). We note that the strong adsorption energy, which is on Ti-${\text{Al}}_2{\text{O}}_3$, is only slightly (−0.03 eV) stronger than neat ${\text{Al}}_2{\text{O}}_3$. We attribute modification of outer-sphere adsorption energy to changes in the partial charges of the protons of the water layer. For all substitutions, additional charge density is localized near substituted cations as compared to unsubstituted ${\text{Al}}_2{\text{O}}_3$ (Figure 5(a-b)). Mn, Fe, Co, Cu and Ni induce the most charge localization, >0.05 ${\text{e}}^{-}$/Ų, while Sc, Ti, V, Cr, Ga, and In induce the least. This localization weakens the HO-X bond for V, Cr, Mn, Fe, Co, Ni and Cu (0.05–0.02 eV/bond via COOP), as compared to higher bond strength of HO-X bond for unsubstituted ${\text{Al}}_2{\text{O}}_3$ (0.08 eV/bond) and other substitutions (0.10–0.07 eV/bond). Thus, more electron density accumulates on the protons of water molecules near V, Cr, Mn, Co, Cu and Ni substitutions as compared to unsubstituted ${\text{Al}}_2{\text{O}}_3$. The higher electron density, and thus reduced partial positive charge, decreases the coulombic attraction between the protons and the O of the selenate (Figure 5(c)). This weakening resulted in poorer adsorption energy of selenate on $\text{X}-{\text{Al}}_2{\text{O}}_3$ in outer-sphere adsorption. Conversely, in the case of Ti-${\text{Al}}_2{\text{O}}_3$, Bader analysis indicates that the electron density is slightly lower (0.07 ${\text{e}}^{-}$) on the protons thus enabling slightly stronger hydrogen bonding and −0.03 eV stronger adsorption on Ti-${\text{Al}}_2{\text{O}}_3$ as compared to unsubstituted ${\text{Al}}_2{\text{O}}_3$.

Figure 5:

Difference in charge density on protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ relative to protonated unsubstituted ${\text{Al}}_2{\text{O}}_3$ before adsorption of selenate. (a) X: Sc, Ti, V, Cr, Ga, In and (b)X: Mn, Fe, Co, Ni, Cu. (c) Charge density difference between $\text{X}-{\text{Al}}_2{\text{O}}_3$ (X: Ti, V, Cr, Mn, Co, Cu, Ni and Ga) and unsubstituted ${\text{Al}}_2{\text{O}}_3$ vs. adsorption energy of selenate adsorption in outer-sphere.

While outer-sphere adsorption is energetically preferred for Ti, V, Cr, Mn, Co, Cu, Ni and Ga, it was not necessarily the only stable configuration. Selenate adsorption was also dynamically stable in the MM configuration for Cr, Fe Co, Cu. Ni, Ga, while Ti, V, and Mn have stable MM and BM configurations. When the selenate was placed in a BM configuration on Cr, Fe Co, Cu, Ni, it relaxed to a MM configuration. We attribute the ability to form BM configurations to the size of the substituted cation. The three BM binding cations, Mn, Ti and V which have atomic radii larger than 0.6 Å, (0.67 Å, 0.65 Å and 0.62 Å respectively) bind in BM configuration, while those with smaller radii do not (Table 2). In the BM configuration, the O-Se-O bond angle decreases (91°−99°) as compared to desorbed selenate (110°); which increases the strain on the oxo-anion. The adsorption of selenate in BM configuration needs to compensate for both the broken X-OH bond and increased strain of O-Se-O bond. For In and Sc, the net difference in bond strengths is positive (0.041 eV/bond (In) and 0.001 eV/bond (Sc)); while the net difference is negative for Ti (−0.014 eV/bond), V (−0.004 eV/bond) and Mn (−0.017 eV/bond). The altered bond strength enables stable BM configurations for Ti and Sc and less exothermic adsorption of selenate in BM for Mn, Ti, and V. We also observe that the protonation of adsorbed ${{\text{SeO}}_4}^{-}$ changes to ${{\text{HSeO}}_4}^{-}$ (pulling ${\text{H}}^{+}$ from the surface) on Ti-${\text{Al}}_2{\text{O}}_3$ and Ga-${\text{Al}}_2{\text{O}}_3$ in outer-sphere configuration and Sc-${\text{Al}}_2{\text{O}}_3$ and Mn-${\text{Al}}_2{\text{O}}_3$ in monodentate configuration. However, this change does not show any effect on the adsorption energies.

Table 2:

Ionic radii vs. Inner-sphere adsorption of selenate on protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface. The data is arranged in ascending order for ionic radius.

X	Ionic radius (Å)	Inner sphere adsorption energy (eV)	Preferred inner-sphere configuration
Al	0.51	−1.49	MM
Cr	0.58	−1.50	MM
Ga	0.59	−1.54	MM
Cu	0.62	−1.03	MM
V	0.62	−1.51	MM
Ni	0.63	−0.92	MM
Co	0.64	−1.17	MM
Fe	0.65	−1.54	MM
Ti	0.65	−1.58	MM
Mn	0.67	−1.33	BM
Sc	0.68	−1.77	BM
In	0.76	−1.83	BM

3.1.2. ${{\text{SeO}}_4}^{2-}$ adsorption on neutral metals substituted [012] $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface.

We calculate that selenate adsorption is endothermic on all neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ surfaces. The adsorption energies range between 0.24–0.48 eV in outer sphere configurations and 0.04–1.25 eV in MM configurations. Thus adsorption, in either MM or outer-sphere, is at best only −0.21 eV stronger than the unsubstituted $\text{α}-{\text{Al}}_2{\text{O}}_3$ (endothermic by 0.24 eV), but some substitutions destabilize adsorption to 0.50 eV. Ti-${\text{Al}}_2{\text{O}}_3$ has the least endothermic adsorption energy (0.04 eV in its MM configuration) among all the neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ and unsubstituted $\text{α}-{\text{Al}}_2{\text{O}}_3$ surfaces and is the only surface that prefers the MM configuration to outer-sphere adsorption. The adsorption energies in the outer sphere configuration on $\text{X}-{\text{Al}}_2{\text{O}}_3$ are either equal to (Ti, Co) or slightly more endothermic than (Sc, V, Cr, Mn, Fe, Ni, Cu, Ga, In) unsubstituted ${\text{Al}}_2{\text{O}}_3$. The adsorption of selenate in a monodentate configuration to all $\text{X}-{\text{Al}}_2{\text{O}}_3$ except Co-${\text{Al}}_2{\text{O}}_3$ and Ga-${\text{Al}}_2{\text{O}}_3$ is less endothermic than unsubstituted $\text{α}-{\text{Al}}_2{\text{O}}_3$. The Bidentate Monomolecular configuration is unstable on all neutral surfaces, and they relax to the monodentate monomolecular configuration. The selenate remains in fully de-protonated (${{\text{SeO}}_4}^{2-}$) form upon adsorption on neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ in all configurations.

Neutral Ti-${\text{Al}}_2{\text{O}}_3$ is the only surface structure that has a less endothermic adsorption energy in inner-sphere configuration than outer sphere configuration. From Bader analysis, we calculate that the Ti in Ti-${\text{Al}}_2{\text{O}}_3$ transfers 0.18 ${\text{e}}^{-}$ worth of electron density to the bonding O (${\text{Δe}}^{-}$ density = 0.16) of selenate. The remaining 0.02 ${\text{e}}^{-}$ is distributed among the surface ${\text{OH}}^{-}$ ions. The transfer stretches the bond distance between the surface bonded O and Se to 2.5 Å from the ~1.7 Å between selenate’s Se and the other O atoms. These bond changes suggest that the Se is only three-fold, rather than four-fold, quadrated after adsorption. Thus, the Ti cation acts to reduce selenate to selenite upon adsorption in neutral surface. The ability to reduce selenate arise from the instability of Ti in its +3-state which is present on the ${\text{Al}}_2{\text{O}}_3$ surface. Only after selenate adsorption, and the associated electron donation, does the Ti cation adopt its preferred +4 state.

Further electronic analysis supports the finding that Ti reduces the adsorbing selenate to selenite. The density of states (DOS) plot in Figure 6 shows that before adsorption, Ti induces mid-gap states 0.33 eV, 0.58 eV and 0.83 eV above the Fermi level. The highest mid-gap band is thus similar in energy to the valence band of selenate (1.08 eV above the Fermi level). Upon adsorption, the 3p orbital from Se ion is filled with the Ti 3d electron density, which now shifts to 0.04 eV above Fermi level (Figure 6(b)). Lastly, the reduction and Se-O bond cleavage is further verified by Crystal Orbital Overlap Population analysis (COOP), which shows presence of net anti-bonding orbital character (ICOOP = −0.0033 eV/bond) between TiO-Se bond after adsorption. The bond strength of Ti-OSe (ICOOP = 0.1038 eV/bond) after adsorption is also stronger than bond strength of Ti-OH (ICOOP = 0.0583 eV/bond) present before adsorption. This bond strengthening is not true for other surfaces, where the bond strength of X-OSe is always lower than the bond strength of X-OH. Thus, the formation of a strong, reducing Ti-O bond compensates for the water network disruption and ${\text{H}}_2\text{O}$ dissociation upon selenate adsorption.

Figure 6:

Selenate adsorption on neutral (a) ${\text{Al}}_2{\text{O}}_3$ and (b) Ti-${\text{Al}}_2{\text{O}}_3$ in inner-sphere (MM) configuration. Density of State plots of Selenate on Ti-${\text{Al}}_2{\text{O}}_3$ (c) before adsorption and (d) after adsorption. The grey, blue, dark grey, orange, red, white spheres represent Al, Ti, Ni, Se, O and H. atoms, respectively.

The adsorption of selenate in outer-sphere configuration results in adsorption energy on neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ is essentially equal to, or more endothermic than unsubstituted ${\text{Al}}_2{\text{O}}_3$. Additionally, the range in adsorption energies for outer-sphere adsorption on neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface is 0.26 eV which is the smallest of all the configurations. We attribute the narrower range to the indirect interactions between the selenate and the substitutional cations, via the water network which is only slightly modified by the cations. We found that the adsorption energies of selenate in outer sphere in Co-${\text{Al}}_2{\text{O}}_3$ and Ti-${\text{Al}}_2{\text{O}}_3$ (0.24 eV) is essentially the same as adsorption ${\text{Al}}_2{\text{O}}_3$ (0.25 eV). This is because the ionic radius of Co (1.94 Å) is the closest to ionic radii of Al (1.94 Å) on the neutral surface and thus the water network remains mostly undisturbed, while Ti is again unstable and donates 0.04 ${\text{e}}^{-}$ to selenate upon adsorption. For other species, the ionic radius is larger than 1.99 Å which cause slight adjustments in water network, and hence decrease the adsorption energy. Measurements of the water network changes will be discussed in detail in Section 3.3.

Finally, we performed COOP analysis between the single atom substitutions and surface O’s to (O1-O4, figure 1) to analyze the stability of metal substitution to confirm that the adsorption trends we observe are because of changes in surface properties and not due to poor stability of substitution cation (Figure S6). We find that the adsorption energies are not related to the stability of substitution on the ${\text{Al}}_2{\text{O}}_3$ surface. Mn, Fe, Co, Ni and Cu metal substitutions are less likely to be stable than other substitutions (In, Ga, Sc, Ti, V and Cr) as they have some filled anti-bonding orbitals before fermi energy (ICOOP < 0 eV/bond). However, the substitutions are stable enough and do not cause dramatic changes to the adsorption energies investigated here.

3.2. Adsorption energy of sulphate on single cation embedded in the [012] $\text{X}-{\text{Al}}_2{\text{O}}_3$

The adsorption behavior of sulfate on ${\text{Al}}_2{\text{O}}_3$ is very similar to that of selenate due to their similar geometric and electronic structures, and pKa’s. The adsorption energies of ${{\text{SeO}}_4}^{2-}$ and ${{\text{SO}}_4}^{2-}$ on neutral and protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ in their most relaxed (exothermic) configurations are presented in Figure 7. The adsorption of these two species is essentially identical within methodological accuracy on the protonated unsubstituted ${\text{Al}}_2{\text{O}}_3$, having adsorption energies of −1.79 eV (${{\text{SO}}_4}^{2-}$) and −1.79 eV (${{\text{SeO}}_4}^{2-}$) in the favored outer sphere configuration. Their adsorption energies are very similar on neutral $\text{α}-{\text{Al}}_2{\text{O}}_3$ being 0.24 eV (${{\text{SO}}_4}^{2-}$) and 0.25 eV (${{\text{SeO}}_4}^{2-}$) which are both endothermic. Thus, because there is no meaningful thermodynamic preference for one ion or the other for the exothermic outer sphere binding, we do not predict ${\text{Al}}_2{\text{O}}_3$ to provide selenate selectivity.⁶⁴ This similarity in adsorption energy matches the experiments which shows no selectivity for selenate.

Figure 7:

Adsorption of ${{\text{SeO}}_4}^{2-}$ vs. ${{\text{SO}}_4}^{2-}$ on (a) Protonated $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface and (b) Neutral $\text{X}-{\text{Al}}_2{\text{O}}_3$ surface in the most exothermic configuration.

Ti is the only substitution which induces selectivity on the neutral surface. Ti stabilizes ${{\text{SeO}}_4}^{2-}$ adsorption by 0.2 eV over ${{\text{SO}}_4}^{2-}$. All other substitutions only modified the adsorption energy of selenate by ±0.1 eV over sulfate, which is within the resolution of the computational method. The preference of Ti substitutions for Se over S arises from the slight differences in the electronic structures of Se and S and the position of the mid-gap states induced by Ti. On neutral Ti-${\text{Al}}_2{\text{O}}_3$ selenate adsorption in the MM configuration was −1.06 eV more exothermic than unsubstituted ${\text{Al}}_2{\text{O}}_3$, because of the overlapping selenate and Ti orbitals, which cause selenate to reduce into selenite (Figure 6). Reduction does not, however, occur for sulphate as the valence bands of sulfate are 2.06 eV above Fermi level (Figure S2). This finding suggests that selective adsorption of selenate over sulfate is possible when Se(VI) reduces to Se(IV) upon adsorption. However, the structures where Se(VI) reduce to Se(IV) are not the energetically favorable structures, and thus selectivity of Ti modified ${\text{Al}}_2{\text{O}}_3$ is unlikely to provide selectivity.

Both selenate and sulphate adsorb in BM configurations on the protonated Sc, V, Cr, Mn and Ti-${\text{Al}}_2{\text{O}}_3$. Using Sc-${\text{Al}}_2{\text{O}}_3$ as the exemplar, we calculate that it slightly prefers ${{\text{SO}}_4}^{2-}$ adsorption by −0.18 eV. This preference arises because ${{\text{SO}}_4}^{2-}$ has shorter O-S bonds (1.52 Å) than the O-Se of ${{\text{SeO}}_4}^{2-}$(1.70 Å). When the oxo-anions adsorb on Sc in BM configuration, O-Se-O bond angle decreases (93.25°) more than O-S-O bond angle (98.53°) (Figure S3). This leads to lower bond strength of SeO bond (0.11 eV/bond) as compared to SO bond (0.13 eV/bond). Hence, we see a preferential adsorption of ${{\text{SO}}_4}^{2-}$ on protonated Sc-${\text{Al}}_2{\text{O}}_3$ as compared to ${{\text{SeO}}_4}^{2-}$. The same accounted for the performance of the other cations and BM configurations (Table S4 & S5). This finding suggest size selective adsorption is possible, though not for selenate over sulfate, if we identify materials which adsorbs oxo-anions in inner-sphere bidentate configurations.

3.3. Factors affecting the adsorption of ${{\text{SeO}}_4}^{2-}$ and ${{\text{SO}}_4}^{2-}$ on $\text{X}-{\text{Al}}_2{\text{O}}_3$.

Lastly, we investigated mechanisms that control the adsorption energy of selenate and sulphate on $\text{X}-{\text{Al}}_2{\text{O}}_3$. In our previous study, we found that water network controls the adsorption of selenate on unsubstituted $\text{α}-{\text{Al}}_2{\text{O}}_3$.³³ The similar preference in adsorption motifs of the substituted and neat ${\text{Al}}_2{\text{O}}_3$, albeit with slightly modified energetics, suggests that the water network remains the primary director of adsorption thermodynamics, and that the substitutions only secondarily modifies the energetics; we detail more concreate analysis which confirms this observation.

We examined 11 different variables which could control or describe the system as outlined in Table 3. Broadly, they consider the effect of water network, cation size and electron density, and configurations changes upon adsorption of selenate in the most exothermic configurations. We examined the water network effects by quantifying two types of water bonds, the number of intermolecular hydrogen bonds (#HO <1.2 Å)⁶⁵ and the number of intra-molecular H-O covalent bonds (#OH ~ 1.2–2.5 Å)⁶⁵, along with surrogates for the bond strength, namely the average bond length of hydrogen bonds (HO_length) and H-O covalent (OH_length) bonds in the water network. The water network parameters depend on the substitutions because the position and charge of the cation slightly modifies the bond lengths of water networks on the surface. Moreover, on the surface, the selenate and sulfate both interact with surface hydroxyl groups. In some instances, oxo-anion upon adsorption on the surface takes up protons from the surface hydroxyl groups and change the protonation of oxo-anion (Figure S7). This behavior is accounted in our regression model and PCA analysis, where we include the change of surface water properties (including interaction of adsorbed selenate with water) as 4 separate variables. We also investigated the effects of cationic properties on adsorption of selenate such as ionic radii (size of X calculated using method mentioned in Section 2.3) and electron density present on the single atom imbedded in ${\text{Al}}_2{\text{O}}_3$ surface (Electron density(X), which was calculated using Bader analysis).

Table 3:

List of variables considered for linear regression and PCA.

Variables	descriptors	label
Change in H-O covalent bonds	$\text{Δ\#OH}$	a
Change in length of H-O covalent bonds	$\text{ΔOH\_length}$	b
Change in HO bonds	$\text{Δ\#HO}$	c
Change in length of OH bonds	$\text{ΔHO\_length}$	d
Number of H-O covalent bonds	#OH	e
Length of H-O covalent bonds	OH_length	f
Number of HO bonds	#HO	g
Length of HO bonds	HO_length	h
Ionic radii of the single atom	Cation_size(X)	i
Electron density present on the single atom	Electron_den(X)	j
Electron density present on the oxo-anion cation	Electron_density(Oxo)	k

We find that the water network controls the adsorption behavior, as shown by the correlation plots in Figure 8 of the regression predicted energies and the DFT calculated energy. Linear regression of the adsorption energy with each of the 11 variables showed that the water network variables #OH and #HO have the highest correlation, achieving ${\text{r}}^2$ values of 0.92 and 0.90 respectively. Conversely, the cation descriptors only achieve a ${\text{r}}^2$ values of 0.00 and 0.01. We note, however, that the best performance based on ${\text{r}}^2$ show highly clustered data points. In effect, these descriptors essentially determine the protonated and non-protonated surfaces. The water network variables which have a reasonable spread, i.e. OH_length and $\text{ΔOH}$, still present reasonably good correlations (0.63 and 0.54 respectively), particularly as compared to the cation descriptors. Thus, even though the cationic size affects OH_length, it is the water network variable which describes adsorption well, not the size. Only when cation size is combined with water network descriptors do reasonable correlations arise (achieving ${\text{r}}^2$ of 0.92); however, this is still not the best overall descriptor, which comes from the #OH and length (OH_length) of the water network (0.94).

Figure 8:

Performance of the predicted adsorption energy of ${{\text{SeO}}_4}^{2-}$ and ${{\text{SO}}_4}^{2-}$ adsorption on $\text{X}-{\text{Al}}_2{\text{O}}_3$from the linear regressions of one or two descriptors. A 1:1 line demonstrates the ideal prediction behavior. Diagonal elements are single variables (as outlined in Table 3) while off-diagonal plots show two component linear regression. The number in the corner is the ${\text{r}}^2$ value of the linear regression.

To better understand the mechanistic properties that control the adsorption energies, we performed Principal Component Analysis (PCA) of the 11 variables described above to identify orthogonal variables for regression (Figure 9). Principal component (PC) 1 is mostly composed of variables describing the water network and associated changes induced upon adoption (#HO (0.46), #OH (0.44) and $\text{Δ\#HO}$ (0.34)) and bond strength (OH_length (0.43) and $\text{ΔOH\_length}$ (0.33)), as shown in Figure 8 a. We found that regressing the adsorption energy against PC1 achieved an ${\text{r}}^2$ value of 0.8. This ${\text{r}}^2$ value is lower than some of the single components because there is large spread in the data, and the data are no longer just classified by protonation of the surface. Thus, it represents a better fit. The addition of PC2, which is dominated by cation size (0.53), only improves the fit to an ${\text{r}}^2$ of 0.83. There is a modest correlation between cation size and the intra- and inter H-bond lengths (i.e.,0.5–0.1) as shown in Figure 8e. However, these length descriptors are more important in determining the adsorption energy as they appear prominently in PC1, which explains most of the adsorption energy. Thus, we attribute the cationic size effects to be only secondary, where the cation size modifies the water network and the modified water network controls the adsorption energy. Finally, PC 3 is mostly composed of the cation electron density descriptor (0.61) and improves the ${\text{r}}^2$ to 0.89. The minimal correlation between the charge density on the cation and bonding energy suggests that the charge interaction between the cation and selenate is small; additionally, this finding suggest our consideration of non-charged substitutions is valid for examining substitutional cationic effects. Additionally, only in PC 3 does the descriptor for ${{\text{SeO}}_4}^{2-}$ or ${{\text{SeO}}_4}^{2-}$ appear, and even there it contributes similarly (0.31) to other water network descriptors contributing to PC 3 ($\text{ΔHO\_length}$, 0.35). The relegation of oxo-anion descriptor to a minor contributor suggests that finding adsorbates which rely on water networks to selectively remove selenate from water is challenging (Figure 10). This finding is consistent with the difficulty seen experimentally with identifying selective sorbents.

Figure 9:

(a) Strength of each variable in 3 Principal Components. (b-d) Linear regression between Principal components and DFT calculated adsorption energy. (e) correlation plots between different variable specified in Table 3

Figure 10:

Principal Component Analysis of ${{\text{SeO}}_4}^{2-}$ vs. ${{\text{SO}}_4}^{2-}$ adsorption on $\text{X}-{\text{Al}}_2{\text{O}}_3$.

To further examine the relative importance of the water network and cation effects, we repeated the PCA including only the water or cationic descriptors, the results are shown in SI Figures S3 & S4. We find that the PC1–2 without the cationic descriptors fit the calculated values as well as the full set of descriptors, having ${\text{r}}^2$ values of 0.83 & 0.83 (1 & 2 PC’s) and 0.8 &0.83 respectively. Conversely, the ion only descriptors were very poor in predicting the adsorption energies with ${\text{r}}^2$ values of 0.01. These results further support the conclusion that the cation affects adsorption only indirectly via modified water networks.

4. Conclusions

We investigated adsorption of selenate and sulphate on single cations embedded in the surface of (012) ${\text{Al}}_2{\text{O}}_3$ to delineate the effects of cation identity on adsorption without disturbing the water network on the surface. Overall, the addition of substitutional cations did not increase the adsorption of selenate or sulfate to the surface; only Sc, In and Ti increased the binding strength, albeit very slightly. We found that the water network exerts strongest effect on adsorption energy of selenate, far greater than any correlation occurring between oxo-anion and cation. The adsorption of selenate on protonated surface is always exothermic and adsorption of selenate on neutral surface is always endothermic. We find that the size of cation embedded in protonated ${\text{Al}}_2{\text{O}}_3$ surface controls the adsorption configuration of selenate through modifying the water network. Larger cations bind selenate in bidentate configuration as they break and weaken the sublayer X-O bonds. However, the bidentate adsorption comes with the added strain on oxo-anion as the bond angle decreases upon adsorption which destabilizes the adsorbate.

In terms of selectivity, we find that reducing cations, like Ti, favors selenate as its unoccupied bands are slightly lower in energy than those of sulfate, and thus are more easily occupied. Conversely, inner-sphere bidentate adsorption induces a slight preference towards sulphate as the angular strain experienced by sulfate is lower than selenate. We hypothesize that cations with ionic radii larger than In that might reverse the selectivity towards selenate. However, we cannot guarantee if the synthesis of such alloys or embedding of larger cations might be possible without changing the morphology of the adsorbent. Overall, we conclude that modifying and controlling the water network morphology on the surface provides the most promising path for developing highly active selenate sorbents. Therefore, materials that retain extra protons on the surface around neutral pH ranges or materials where the water network is less rigid should be sought to improve adsorption technology for toxic oxo-anion removal.