Unique Interaction between Layered Black Phosphorus and Nitrogen Dioxide

Air pollution caused by acid gases (NO2, SO2) or greenhouse gases (CO2) is an urgent environmental problem. Two-dimensional nanomaterials exhibit exciting application potential in air pollution control, among which layered black phosphorus (LBP) has superior performance and is environmentally friendly. However, the current interaction mechanism of LBP with hazardous gases is contradictory to experimental observations, largely impeding development of LBP-based air pollution control nanotechnologies. Here, interaction mechanisms between LBP and hazardous gases are unveiled based on density functional theory and experiments. Results show that NO2 is different from other gases, as it can react with unsaturated defects of LBP, resulting in oxidation of LBP and reduction of NO2. Computational results indicate that the redox is initiated by p orbital hybridization between one oxygen atom of NO2 and the phosphorus atom carrying a dangling single electron in a defect’s center. For NO, the interaction mechanism is chemisorption on unsaturated LBP defects, whereas for SO2, NH3, CO2 or CO, the interaction is dominated by van der Waals forces (57–82% of the total interaction). Experiments confirmed that NO2 can oxidize LBP, yet other gases such as CO2 cannot. This study provides mechanistic understanding in advance for developing novel nanotechnologies for selectively monitoring or treating gas pollutants containing NO2.


Introduction
Hazardous gases emitted from industries, traffic and other causes cause a series of climate, environmental and health problems, including global warming, acid rain and photochemical smog [1][2][3]. According to International Energy Agency's report in 2016, 6.5 million premature deaths can be attributed to air pollution worldwide annually [4]. Exposure during pregnancy and early postnatal periods to nitrogen dioxide (NO 2 ) or sulfur dioxide (SO 2 ) is associated with childhood allergy diseases [5]. Adsorbents are increasingly used in eliminating hazardous gases, such as molecular sieves [6], porous carbon [7], metallic clusters [8], graphene [9] and MXenes [10]. Two-dimensional (2D) nanomaterials have great advantages in gas adsorption over other materials, owing to quantum size effects, large surface areas, excellent electronic/photoelectronic performance, etc. [11,12]. A comprehensive understanding of the interaction mechanisms between nanomaterials and

Adsorbent Model
A 3 × 1 × 3 supercell containing 36 phosphorus atoms (P 1 -P 36 ) was constructed to simulate perfect SLBP. Each P atom of SLBP is covalently bonded to three adjoining P atoms, forming a wrinkled honeycomb structure ( Figure S2). This wrinkled honeycomb structure lets SLBP contain two atomic layers, each including 50% of the atoms. The size of the periodic box was a = 9.845 Å, b = 17.097 Å and c = 13.972 Å along zigzag, vertical and armchair directions, respectively. The vacuum region in the vertical (b) direction was set to be 15 Å to avoid interactions of gas molecule with SLBP in the adjacent periodic box. Similar box size was used for simulating the LBP-based gas sensor [31]. Defective SLBP (d-SLBP) was constructed on the basis of optimized geometry of SLBP. The initial configurations of SLBP and d-SLBP are shown in Text S1. Additional information is included in the Support Information (Figures S1-S27, Table S1-S4, Text S1-S10). Two in-plane SW1 and SW2 [39,43] defects were built by rotating a vertical bond P 18 -P 20 or a horizontal bond P 17 -P 18 by 90 • . SV and DV1-DV3 defects were built by deleting phosphorus atoms. Theoretically, deleting one phosphorus atom creates three neighboring unsaturated phosphorus atoms, each carrying one dangling unpaired electron. SV was constructed by removing atom P 18 . DV1 and DV2 were built by removing two bonded phosphorus atoms, P 18 -P 20 and P 17 -P 18 , respectively. DV3 was obtained by removing two separate atoms, P 18 and P 29 . Edge defects in either zigzag (EGz1 and EGz2) or armchair (EGa1 and EGa2) direction were constructed, considering the anisotropy of LBP [44]. The periodic box was enlarged in the armchair (c) or the zigzag (a) direction to 20 Å for creating edge defects. Box enlargement inevitably results in unsaturated phosphorus atoms on edges. By saturating phosphorus atoms on one edge with hydrogen atoms, EGz1 and EGa1 were built. By deleting one phosphorus atom from EGz1 (P 31 ) or EGa1 (P 21 ), EGz2 or EGa2 was constructed. Fully optimized geometries of SLBP and d-SLBP are shown in Figure 1.

Adsorption Complex Models
Generally, there are three widely accepted adsorption sites in LBP, including top, bridge and hollow types [29]. In this study, the three adsorption sites were further differentiated between SLBP and d-SLBP to be three top (T, T1, T2), 4 bridge (B, B1-B3) and 5 hollow (H, H1-H4) sites ( Figure 1). Gas molecules were placed at one of these sites as initials. Due to the influences of spatial structures, the adsorption site for gas molecules in the fully relaxed configuration of the adsorption complex was classified according to the closest type. Different initial distances (ca. 2.14 to 4.15 Å) between gas molecule and SLBP or d-SLBP were tested in the computation. No direct bindings between gas molecules and any phosphorus atoms existed in initial configurations.
Orientations of gas molecules were comprehensively considered in the computations, as previous computational results for orientations of gas molecules adsorbed on LBP were obscure [15,[31][32][33][34]39]. For nonlinear molecules (NO 2 , SO 2 , NH 3 ), vertical orientations, including single atom pointing (i.e., 1Ov, Nv, Sv, 1Hv) or multiple atoms pointing (2Ov, 3Hv) to SLBP plane, and parallel orientations (p), were included. For linear molecules (CO 2 , NO, CO), vertical single atom pointing (Ov, Cv, Nv) and parallel (p) orientations were calculated. Considering the anisotropy of LBP, orientations were further differentiated to be along zigzag (z) or armchair (a) directions. Taking 2Ovz as an example, it represents a configuration where two oxygen atoms of a dioxide point to the SLBP plane, and the orientation of the dioxide is along the zigzag direction. The initial configurations of gases with different orientations on SLBP are shown in Text S1.

Computational Methods
All computations were performed with a CASTEP program in the Material Studio (Biovia, San Diego, CA) software [45]. The GGA-PBE [46] rather than LDA (Local Density Approximation) [47] was used to describe exchange-correlation functional, as LDA was regarded to usually underestimate bonding distance and overestimate binding energy. In addition, van der Waals (vdW) correction was performed within the empirical correction scheme of Grimme [48] method. The Brillouin zone was sampled using a 4 × 1 × 3 Monkhorst-Pack k-point grid. The Kinetic energy cutoff of 500 eV was used in geometry optimization process. Convergence tests for energy and k-point are shown in Text S2. Initial configurations were fully optimized until the force on each atom was less than 0.05 eV/Å and the energy tolerance was less than 1 × 10 −5 eV/atom. Spin polarization was included for computing NO, NO 2 or unsaturated d-SLBP (SV, DV1, DV3 and edge defects). Band structure and Mulliken population were analyzed based on fully optimized geometries. A large k-point (10 × 1 × 8) was used to achieve high accuracy in density of states (DOS) computations. The calculation methods of adsorption energies including (E ad ) and not including (E 0 ) vdW interaction energy (E vdW ) corrections and deformation energies (E def ) of d-SLBP after adsorbing gas molecules are described in Text S2.

Experimental Methods
LBP was prepared by the liquid exfoliation method described in a previous study [19]. Firstly, the bulk black phosphorus was ground into powders with an agate mortar, and transferred into oxygen-free Millipore ultrapure water in the glovebox. Then, the powders were sonicated with a probe for 12 h. The LBP suspension was transferred to centrifuge tubes in a glovebox, and then centrifuged at 10,000 rpm for 30 min. The supernatant was transferred to the glovebox and filtered through a 0.22 µm cellulose membrane to collect LBP. After drying overnight in the glovebox, the LBP was transferred into three anaerobic bottles. Two of the bottles were continuously pumped with NO 2 (200 ppm NO 2 in N 2 , Beijing Lvyuan Dade Biological Tech Co., Ltd., Beijing, China) and CO 2 (>99.5%, Shenyang Shuntai Special Gas Co., Ltd., Shenyang, China) at a flow rate of 10 mL/min for 3 h. Then, the two bottles were sealed, allowing thorough interaction of LBP with the gases. After 24 h, each of the two bottles was injected with pure N 2 for 1 h to remove excess NO 2 or CO 2 . LBP in the third bottle was filled with pure N 2 (>99.99%, Shenyang Shuntai Special Gas Co., Ltd.) for 3 h and placed in the glovebox for over 24 h. X-ray photoelectron spectroscopy (XPS, ESCALAB250, Thermo VG, Waltham, MA, USA) was used to identify phosphorus components of LBP after treatment with N 2 , NO 2 or CO 2 .

Adsorption of Gas Molecules on Perfect LBP
A perfect single-layer LBP (SLBP) contained in a periodic box was used in computations ( Figure 1a). Initial configurations of adsorption complexes were constructed by varying orientation, adsorption sites or distances of gas molecules on the plane of SLBP, in order to solve the inconsistencies in E ad values and in favorable adsorption configurations among previous computational studies. By comprehensively screening hundreds of adsorption configurations, the most stable adsorption configuration (adsorption complexes possessing the most negative E ad values) of each gas molecule was obtained ( Figures S3-S5).
Computational results indicate that orientations, together with distances of gas molecule toward the LBP plane, in the initially constructed adsorption configurations were key factors influencing the computational results on favorable adsorption configurations and E ad (Text S3). These two factors were the primary reasons for contrary computational results in previous studies [15,[31][32][33][34]37] (Text S4). In comparison with initial orientations of gas molecules, impacts of initial adsorption sites on favorable configurations or E ad are not remarkable.
According to E ad values (Table 1), the two oxygen atoms in pointing adsorption configuration (E ad = −0.225 eV, Figure S5a) are more stable than the nitrogen atom in pointing configuration (E ad = −0.211, −0.202 eV, ( Figure S5b,c) in NO 2 . This is different from the adsorption of NO 2 on other 2D nanomaterials, such as graphene or indium nitride, where NO 2 is bound to the sheet's surface with the nitrogen end [49]. In the case of SO 2 , two oxygen atoms in pointing configuration are less favorable ( Figure S5d Figure S5d,e). Three linear molecules, CO 2 , NO and CO, prefer parallel configurations, as indicated by E ad values. In the most stable adsorption configurations, CO 2 is almost completely parallel to the SLBP plane with a negligible dihedral angle ( Figure S5g), whereas for NO or CO, the dihedral angle to the SLBP plane is 9-17 • (Figure S5h-j). This small dihedral angle is mainly attributed to the asymmetry of monoxide molecules. Differently from NO 2 , the nitrogen atom pointing configuration is the most stable configuration for NH 3 ( Figure S5k). The small E ad values (−0.310 to -0.138 eV), low charge transfer amounts (−0.15 to 0.04 e) and large distances to the SLBP plane (2.47-3.44 Å) indicate weak interaction between gas molecules and SLBP ( Table 1). The distance of the N atom in NO to SLBP (2.47 Å) is the shortest in the most stable adsorption configurations of the six gas molecules, but is still much longer than a P-N covalent bond length (1.89 Å [50,51]). The adsorption of gas molecules on perfect SLBP plane is dominated by vdW forces, as indicated by the ratio of 57-82% relative to E ad .
Based on the first guess, adsorption of NO 2 and NO carrying odd numbers of valence electrons on LBP should be stronger than that of other gas molecules. However, the E ad value of SO 2 is the largest among all gas molecules. The computed E ad value of NH 3 is between that of NO and NO 2 . This is not in accordance with the first guess or experimental results, which indicate that the sensitivity of LBP sensors toward NH 3 (10 ppm) is about 3-5 orders of magnitude lower than that toward NO 2 (0.4-20 ppb) in a dry environment [16,22,23]. Accordingly, there must be other mechanisms responsible to the distinct interaction between LBP and NO 2 .  Interestingly, in comparison with SV, EGz1 or EGa2 (E ad = −1.543 to -1.304 eV), DV3 possess a larger E ad value (−2.584 eV), not only interacting with NO 2 , but also with SO 2 or CO 2 (E ad = −1.378 to -1.211 eV), as shown in Figure 2d. However, no new bonds are formed between DV3 and SO 2 or CO 2 (d shortest = 2.89-3.66 Å), indicating the negative E ad values are unusually large. This is because non-negligible deformation happens to DV3 while interacting with dioxides, resulting in inauthentic interaction energies. To obtain real interaction energies between dioxides and d-SLBP, deformation energies (E def ) were computed for each d-SLBP (Text S8). Taking NO 2 as an example, the E def values of DV3 (−0.718 eV) are higher than those of other d-SLBP (−0.013 to −0.254 eV), implying that the deformation of other d-SLBP is slight. After deducting E def from E ad , the modified interaction energy (E ad-def ) of NO 2 and DV3 is −1.866 eV, which is comparable to those of SV, EGz1 and EGa2 (−1.543 to -1.304 eV) ( Figure S6). Accordingly, the four d-SLBP show similar interaction strengths with NO 2 and significantly improve the adsorption of NO 2 compared with SLBP. The E ad-def values of SO 2 and CO 2 interacting with DV3 are −0.304 and −0.137 eV, respectively, comparable to those computed for interacting with SLBP ( Figure S6). Therefore, d-SLBP enhances only the interaction of LBP with NO 2 , but not with SO 2 or CO 2 .

Interaction Mechanisms of Defective LBP and Gas Molecules
To probe interaction mechanisms between NO 2 and d-SLBP, spin density distributions of d-SLBP were mapped (Figure 3). Spin density can explain the distribution of unpaired electrons or single electrons in space [54]. The results show that SV, DV3, EGz1 and EGa2 possess non-zero spin density, meaning that each of the four d-SLBP has unsaturated phosphorus atoms carrying unpaired dangling single electrons. This is caused by losing one or more neighbor P atoms. Accordingly, the unpaired electron of d-SLBP is inferred to be closely related to the unique interaction with NO 2 . Taking SV as an example, there were three unsaturated atoms, P 13 , P 17 and P 20 , in the initially constructed geometry due to loss of P 18 . After geometry optimization, a new bond P 13 -P 20 , 2.37 Å in length was formed, leaving one dangling atom, P 17 , in SV (Figure 1d). Most of the spin density (0.64) is distributed around atom P 17 in SV (Figure 3c), indicating the dangling single electron mainly presents around atom P 17 . According to E ad values, the most stable adsorption configuration of NO 2 on the top of SV is that in which NO 2 binds with atom P 17 (Figure 2e). For the other three d-SLBP, the most stable adsorption configuration of NO 2 also uses the phosphorus atoms carrying the highest non-zero spin density (Text S6). Unsaturated carbons in defected graphene were found to be active toward NO 2 molecules [55]. Therefore, the results indicate that the unique interaction with NO 2 is attributable to unpaired single electron carried by the four d-SLBP.
Although interacting with d-SLBP carrying dangling single electrons, the adsorption of SO 2 or CO 2 is not enhanced, as indicated by the E ad-def values ( Figure S6). This means that the electron properties of gas molecules are another key factor governing interaction strengths between LBP and gases. To further prove this inference, interactions of the four unsaturated d-SLBP with NO and NH 3 were computed. Similarly to NO 2 , NO has an odd number of outmost valence electrons. NH 3 has an even number of outmost valence electrons, and every two of them are paired.
The E ad-def values (Figure 4a) between NO and SV, DV3 or EGa2 are almost as large as that of NO 2 . The E ad-def values between NO and EGz1 are a little lower, but still more negative than that between NO and SLBP. In the most stable adsorption configurations of the adsorption complex of each d-SLBP, the N atom of NO is near d-SLBP at a distance of 1.78-1.98 Å (Figure 4b). This distance is close to (in the case of SV, DV3, EGa2) or slightly larger (in the case of EGz1) than the length of a typical P-N single bond (commonly accepted value is 1.89 Å [50,51]), but much smaller than the distance (2.47 Å) between NO and SLBP (Table 1), which implies that there is a strong chemisorption between NO and unsaturated d-SLBP. The difference is that d-SLBP binds NO by forming a weak P-N bond but reacts with NO 2 by abstracting one O atom in redox reactions. The original bond of NO (1.16 Å) was weakened but not broken (1.20 Å) by d-SLBP, but one N=O bond of NO 2 was broken due to reaction with d-SLBP. The E ad-def values or distances of NH 3 interacting with unsaturated d-SLBP are comparable to those of NH 3 interacting with SLBP (Figures 4a and S7). This confirms that the outmost valence electron characteristics of gas molecules are responsible for strong interactions with LBP. The above computational results show that NO 2 can oxidize unsaturated LBP defects to produce oxidized phosphorus and NO. This redox mechanism has never been reported for NO 2 and other two-dimensional nanomaterials as far as we know. Previous studies showed that vacancy defects in graphene [56] or tungsten trioxide [57] can improve the adsorption of NO 2 , but did not explain the reason. Yan et al. [42] showed that unsaturated carbon atoms in carbon vacancy defects of graphene can bind the nitrogen atom of NO 2 . Carbon vacancies produced chemisorption sites on graphene for NO 2 [58]. Therefore, the enhanced adsorption of NO 2 on graphene is most likely due to strong chemisorption similar to that of NO and unsaturated LBP defects, but not a redox reaction.

Orbital Analysis on the Nature of Interaction
Total density of states (TDOS) and partial density of states (PDOS) were computed to further understand interaction mechanism of NO 2 with LBP (Text S9). Taking SLBP and SW1 as examples of saturated LBP, new impurity states that emerged in conduction bands (CB) of the adsorption complexes at around 2 eV were mainly contributed by p orbitals of O and N atoms, and those in valence bands (VB) at −4 to -2 eV were contributed by p orbitals of O atoms (Figure 5a,b). Slight p orbital hybridization of O, N and P atoms near the Fermi level can explain the small amount of charge transfer (−0.11 e, −0.14 e) between NO 2 and neighboring P atoms (Table 1). Since there is no single electron that can interact with these NO 2 in SLBP and SW1, the whole system exhibits spin asymmetry, which is mainly derived from the nitrogen and oxygen of NO 2 .
Taking SV as an example of unsaturated LBP, an individual SV owns an odd number of single electrons, as shown by asymmetric TDOS (Text S9). As NO 2 carries an odd number of valence electrons, the total number of valence electrons of adsorption complexes formed by SV and NO 2 is even. However, the results show that the complex has a spin asymmetric TDOS (Figure 5c), indicating that electrons are unpaired. This is a cue for the presence of single-electron species in the complex of NO 2 and SV, i.e., [NO] (a NO molecule) carrying an unpaired valence electron. Significant orbital overlaps indicting strong hybridization between p orbitals of O and P 17 was observed, indicating formation of a P-O bond. Similar results were observed for other d-SLBP; see Text S9 for detailed discussion. Large and continuous orbital overlaps were also observed for adsorption complexes of NO and unsaturated d-SLBP ( Figure S8), but not for those with SO 2 or CO 2 (Text S9). This explains how NO can form strong chemisorption with d-SLBP, but SO 2 or CO 2 cannot. The orbital overlap of NO with in-plane vacancy defects (SV and DV3) is stronger than that with edge defects (EGa2 and EGz1), in accordance with the order of the E ad values (DV3 > SV > EGa2 > EGz1). Differently from NO 2 , TDOS of the adsorption complex of NO and SV is spin symmetric (Figure S8), indicating that all electrons are paired. As both individual NO and individual SV have unpaired electrons, the spin symmetry of adsorption complex is a clear clue for the formation of a P-N=O moiety ( Figure S7a).

Experimental Verification of Oxidation of LBP by NO 2
According to the computational results, the unique interaction between NO 2 and LBP is redox in nature, as the unsaturated defect in LBP is oxidized and NO 2 is reduced. To verify that LBP interacts with NO 2 differently from other gas pollutants, experiments were performed. LBP was prepared following the method described in Text S10. As defects are inevitably formed during the exfoliation of LBP in water, the as-prepared LBP was exposed to NO 2 in order to check whether oxidation can occur. As a comparison, the as-prepared LBP was exposed to CO 2 , which has no unpaired valence electrons and interacts with LBP mainly through van der Waals interactions based on our computational results. LBP exposed to N 2 was used as a blank control to show the oxidation status of the original LBP.
After 24 h of exposure, LBP nanoflakes were characterized by X-ray photoelectron spectroscopy. Characteristic peaks corresponding to P 2p1/2 and P 2p3/2 [59] were observed at 130.4 and 129.6 eV. No oxidation peaks were found when BP was exposed to N 2 (Figure 6a) or CO 2 (Figure 6b). However, after exposure to NO 2 , a noticeable peak emerged slightly above 133 eV (Figure 6c), corresponding to oxidized phosphorus (PO x ) [60]. The PO x peak around 133 eV accounts for 23.3% of total phosphorus contents in LBP. This confirms that LBP can be rapidly oxidized by NO 2 . The PO x peak (if there is any) in Figure 6b is as negligible at that in Figure 6a, indicating that direct interaction between oxygen atoms of CO 2 with LBP is negligible. These experimental results give solid evidence supporting the computational conclusion that NO 2 can oxidize LBP, whereas CO 2 , being without an unpaired single electron, cannot. . XPS spectra (P 2p) of LBP nanoflakes exposed to N 2 (a), CO 2 (b), or NO 2 (c) for 24h.

Conclusions
In summary, an interaction mechanism of LBP with common gas pollutants was unveiled in this study. LBP can react with NO 2 following a redox mechanism, which is essentially different from the physisorption of SO 2 , NH 3 , CO 2 and CO, or chemisorption of NO. The adsorption mechanism of gas molecules on LBP was first clarified. The unique interaction with NO 2 is owed to unsaturated phosphorus carrying a single electron in LBP. This mechanism not only solves the problems puzzling experimental studies on LBPbased gas sensors, but also provides ideas for the development of environmentally friendly nanotechnology based on LBP to monitor or treat hazardous gas pollutants.
Supplementary Materials: The supporting information can be downloaded at: https://www.mdpi. com/article/10.3390/nano12122011/s1. Figure S1: Outermost electrons of gas molecules; Figure S2: Atomic structural of SLBP; Figures S3 and S4: Variations of E ad with d shortest , orientations of gas molecule or adsorption sites: Figure S5: Thermodynamically favorable adsorption configurations of gas molecules on SLBP. Figure S6: Modified interaction energies of dioxide on d-SLBP; Figure S7: Stable adsorption configurations for NO and NH 3 on d-SLBP; Figure S8: TDOS and PDOS of NO on d-SLBP. Text S1-S2: Model construction and calculation details, it includes Figures S9-S16; Text S3-S4: Computational results on adsorption of gas molecules on SLBP; Text S5-S6: Computational results on individual d-SLBP and gas molecules on d-SLBP, it includes Figures S17-S16, Tables S1 and S2; Text S7: Nature of the [NO] species, it includes Figure S20 and Table S3; Text S8: Deformation of DV3, it includes Figure S21 and Table S4; Text S9: Single electron dominated mechanism for SLBP or d-SLBP absorbing gas molecules, it includes Figures S22-S26; Text S10: Experimental Details, it includes Figure S27.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.

Acknowledgments:
The computational resources were provided by the Computer Network Information Center of Chinese Academy of Science.

Conflicts of Interest:
The authors declare no conflict of interest.