Band Structure–Driven Design of a α-CsPbI₃ Ammonia Sensor for Industrial Applications

1. Introduction

Given the current stability challenges [1,2,3,4,5,6] presented by the lead-based perovskite solar cells (PSC), it may seem counterintuitive to propose to use such metastable materials as gas sensors. However, it should be noted that while the stability of lead-based perovskite materials is a concern, they also demonstrate a propensity for self-healing in some unusual conditions, which appears to be reproducible at standard operating conditions if the interfaces and grains are appropriately passivated by a stabilizing species. The exact conditions required to stabilize these materials remain under active investigation. Nevertheless, just as they have shown remarkable improvements in power conversion efficiency (PCE) in photovoltaics [7], they have also demonstrated great promise as gas sensors. Previously, we investigated the exceptional properties of an FAPbCl₃-based resistor-type sensor, which demonstrated both high selectivity and sensitivity towards ammonia, a combination that is known to be quite challenging to find in chemical gas sensors [8,9]. DFT calculations led to the conclusion that this is due to the unique properties of the material associated with the defect chemistry, where certain deep-trap defects can be temporarily saturated by ammonia gas, thus decreasing the electrical resistance of the material. Because of size and polarity, only the ammonia molecule could properly fit into the defect and saturate the dangling bond. We termed this effect the “lock-and-key” mechanism akin to substrate-enzyme interactions in cellular biology. According to Shockley-Read-Hall (SRH) theory [10,11], such deep traps may serve as nonradiative recombination centers; however, due to the long carrier lifetime of the material, it becomes important to account for electron-phonon coupling [12].

A key challenge in sensor design is balancing material reactivity towards analytes with the risk of degradation, which is often accelerated at the higher temperatures required by many industrial process streams. In chemical process design, it is often crucial to have sensors as close to the reaction stream as possible to minimize dead-time between the actuation and response, which, for some fast processes, indicates the need for a high-temperature sensor. This has led to the proliferation of the ABO₃ perovskites for the sensing of combustion products such as CO_x and NO_x.[13,14,15,16,17,18,19]

Black-phase CsPbI₃ under certain conditions exhibits superior stability relative to its organic-inorganic hybrid perovskite (OIHP) counterparts due to the lack of an organic A-site cation. These cations contribute some instability to the material and provide additional avenues of degradation [20,21]. However, it should also be noted that under some exotic conditions, these organic cations can induce self-healing effects [22]. For CsPbI_3, the photoactive black cubic (α) phase is stable in pure form above temperatures of around 573 K to 633 K [23,24], tending to degrade at lower temperatures as a result of Cs being too small to occupy the interstitial space between octahedra, which is quantifiable by the low Goldschmidt tolerance factor of 0.81 [25]. This can be ameliorated by the incorporation of larger A-site cations [25] or smaller X-site halogens [26,27], lowering the stable temperature range while creating other stability issues related to phase segregation [28,29,30]. This high-temperature stability of all-inorganic α-CsPbI₃ indicates that it can be useful for gas sensing in harsh environments such as those encountered in conventional ammonia production.

In the present work, the defect-dependent electronic structure of α-CsPbI₃ is characterized with and without several gas analytes. Gas analytes were selected based on industrial relevance, with a size restriction to allow for diffusion throughout the large interstitial spaces between the BX₆ octahedra, created by the large A-site cation. Studied gas molecules include CO, CO₂, NO, NO₂, H₂, CH₃NH₂ (as a representative volatile organic compound (VOC)), and NH₃. Among all of the analyte’s ammonia exhibits a significant influence on the perovskite’s electronic structure, which may be exploited as the operating mechanism in a novel chemical sensor. We note that in the case of a reactor-based sensor, the selectivity of the sensor is likely of less relevance than sensitivity; thus, we focus on the most promising analyte: NH₃.

One pair of frequently encountered surface-based defects (labeled using Kröger–Vink[31] notation) in black phase perovskites are the antisite defects I_Pb and Pb_I, [32,33] both of which may be related to the destructive Pb²⁺/Pb⁰ and I⁻/I₃⁻ surface-based redox chemistry [34]. Recent DFT calculations have demonstrated that within bulk, the Pb_I defect becomes favorable for n-type doped conditions; however, the most dominant defects in bulk are shallow [35], indicating that our proposed sensor mechanism will be surface and grain-boundary-based. This agrees with the experiment, which typically shows surface-based recombination (τ₁) to be an order of magnitude faster than bulk-based recombination (τ₂) [36].

In the present work, we calculated the nonradiative dynamics of the deep I_Pb and Pb_I defects, as well as other defects, in α-CsPbI₃ and compared them to the pristine material as well as with their NH₃-absorbed counterparts. NH₃ is then selected for further exploration due to the global significance of the energy-intensive Haber-Bosch process in the production of fertilizer. This was done with the express purpose of investigating whether this material may be an effective photoluminescence (PL) –based detector of NH₃. Ammonia’s interaction with the deep traps may enable carrier recombination and thus is predicted to allow α-CsPbI₃ to function as an ammonia sensor.

2. Theoretical Methods

For all the DFT calculations performed, we utilized the Vienna Ab initio Simulation Package (VASP) [37]. Initial structural optimizations employed the Perdew-Burke-Ernzerhof (PBE) [38] generalized gradient approximation (GGA) functional and projector augmented-wave (PAW) pseudopotentials [39] and Grimme D3 dispersion corrections [40] were applied throughout. Both the atomic positions and lattice constants of the cubic α-CsPbI₃ unit cell were relaxed; a 4×4×4 supercell was used to mitigate interaction of virtual images of the defects. This large supercell allows sufficient sampling of k-space with a single Γ point, particularly as for an even-numbered cubic perovskite cell, the band-gap folds to the Γ-point for bandgaps at the Γ, R, M, or X point. Each defect was introduced to the cell and relaxed, after which band structures were characterized by single-point (SP) calculations using the Strongly Constrained and Appropriately Normed (SCAN) [41,42] mGGA functional. The results of our calculations were analyzed with VASPKIT [43] for band-unfolding according to the methodology proposed by Ku et al [44]. The defects considered are interstitials, vacancies, antisites (mostly neglected in perovskite literature until relatively recently) and ‘switches,’ defined as pairs of mutually compensating antisites. The cubic lattice vectors were fixed for the subsequent step. Defect formation energy (DFE) as a function of the Fermi level was calculated by the equation below:

$$\Delta H_f\left(X^q\right)=E\left(X^q\right)-E\left(\mathrm{b}\mathrm{u}\mathrm{l}\mathrm{k}\right)-\sum _j{n}_j{\mu }_j+q\left(E_F+V\right)+E^q$$

(1)

where $\Delta H_f\left(X^q\right)$ is the DFE of defect $X$ in charge state $q$, $E\left(\mathrm{b}\mathrm{u}\mathrm{l}\mathrm{k}\right)$ is the energy of the pristine crystal, $n$ and $\mu$ refer to the number and chemical potential of the species removed from or added to the crystal to form a defect $X$, $E_F$ is the Fermi level relative to the top of the valence band, $V$, and $E^q$ is the first-order Makov-Payne correction [45]:

$$E^q={\displaystyle \frac{q^2\alpha }{2\epsilon L}}$$

(2)

where $\alpha$ is the Madelung constant, $\epsilon$ is the dielectric constant, and $L$ is the characteristic length of the cell. Defects with transition energy levels (TELs) near mid-gap were selected for further study, with the lattice vectors allowed to vary.

With the deep traps identified, analytes of interest were introduced to each, and the cells were re-relaxed using PBE, and their electronic structures were re-characterized with SCAN SP calculations and unfolded with VASPKIT. The results were used to identify the analyte with the strongest interaction, which motivated further excited-state study using the NONRAD package, developed for the study of nonradiative recombination dynamics [46,47]. Because NONRAD requires identical lattices for both charge states, and each defect has a set of three consecutive charges, the middle charge state’s lattice was used. NONRAD employs a one-dimensional approximation of phonon modes, which has shown to be sufficiently accurate [48]. The generalized configuration coordinate is defined as:

$$Q=\sqrt{\sum _i{m}_i{\left(R_i-R_{f,i}\right)}^2}$$

(3)

where $i$ is an index running over the atoms in the supercell, $m_i$ is each atom’s mass, $R_i$ is the Cartesian coordinate associated with the mid-gap state of the atom for the transition being considered, and $R_{f,i}$ the same but for the state corresponding to the fully geometrically-relaxed VBM state or electronically-excited geometrically-relaxed CBM state (with h⁺ and e^- carriers). The energy of these states along the paths is shifted relative to the energy of its associated minimum, and also by the carrier energies, i.e., the band gap magnitude for the CBM + h⁺ + e^- state, the associated TEL for the mid-gap states, and zero in the case of the VBM state.

To elucidate the effect of the NH₃ addition to α-CsPbI₃, we performed ab initio molecular dynamics (AIMD) calculations. Equilibration was performed for 2000 fs in the canonical ensemble (NVT). Velocity rescaling was applied to maintain the target temperature during equilibration. We chose two temperature regimes: 300 K and 600 K. The latter is stable, while the former may be stabilized through novel passivation strategies [49]. Production runs were then performed for 10 ps each in the microcanonical ensemble (NVE), allowing the temperature to vary throughout the trajectory. A 1 fs time step was utilized for all AIMD runs, and in this case, we used 3 x 3 x 3 superslabs to ensure that the calculations were tractable to perform.

To run the nonadiabatic molecular dynamics (NAMD) analysis, we cut the last 2 ps of the run and prepare them as snapshots. These snapshots were calculated using SCF convergence with a tightened criterion of 1x10^-6 eV at the R-point, which is where the direct band gap is located for an odd-numbered α-CsPbI₃ supercell. To enable quick evaluation of the nonadiabatic coupling (NAC) elements, we used the Concentric Approximation Nonadiabatic Coupling (CA-NAC) code, which was specifically written to be compatible with the PAW formalism in VASP [50]. The NACs, eigenvalues, and dephasing time generated by CA-NAC were then used to perform semiclassical decoherence-induced surface hopping (DiSH) simulations [51,52] using the parallelized Hefei-NAMD program [53,54]. The NAMD simulations were performed with 1000 independent trajectories and 100 stochastic samples. Each trajectory was allowed to evolve for 10 ns with a timestep of 1.0 fs. In all cases, an extra 10 bands were included above and below the band edges for the DiSH simulations. This approach was used to determine the nonradiative recombination lifetimes (_nr) of charge carriers for the pristine, NH₃·Pristine, I_Pb , NH₃·I_Pb , Pb_I and NH₃·Pb_I systems. For the pristine and NH₃·Pristine systems, $\mathsf{\tau }$_nr was extracted by fitting the decaying CBM population from the NAMD simulation to an exponential decay function:

$$\mathrm{f}\left(\mathrm{t}\right)={\mathrm{e}}^{-\mathrm{t}/\mathsf{\tau }\mathrm{nr}}$$

(4)

For the other four cases, there were defects inside of the bandgap, and thus three-term (in case of NH₃·Pb_I and NH₃·I_Pb) and four-term (in case of Pb_I and I_Pb) differential equation models were required to fit them. In the case of the 4-State Ladder Model (CBM → D1 → D2 → VBM) the following equations were solved:

$${\displaystyle \frac{\mathrm{d}{N}_{\mathrm{C}\mathrm{B}\mathrm{M}}}{\mathrm{d}t}}=-k_1{N}_{\mathrm{C}\mathrm{B}\mathrm{M}}$$

(6)

$${\displaystyle \frac{\mathrm{d}{N}_{\mathrm{D}1}}{\mathrm{d}t}}=k_1{N}_{\mathrm{C}\mathrm{B}\mathrm{M}}-k_2{N}_{\mathrm{D}1}$$

(6)

$${\displaystyle \frac{\mathrm{d}{N}_{D2}}{\mathrm{d}t}}=k_2{N}_{\mathrm{D}1}-k_3{N}_{\mathrm{D}2}$$

(7)

$${\displaystyle \frac{\mathrm{d}{N}_{\mathrm{V}\mathrm{B}\mathrm{M}}}{\mathrm{d}t}}=-k_3{N}_{\mathrm{D}2}$$

(8)

In the three-term case, where $N_{\mathrm{C}\mathrm{B}\mathrm{M}}$, $N_{\mathrm{D}1}$, $N_{\mathrm{D}2}$ and $N_{\mathrm{V}\mathrm{B}\mathrm{M}}$ represent the carrier populations in the respective states, whereas the k₁, k_2, and k₃ constants are the rate constants corresponding to CBM-to-D1 trapping, D1 → D2 relaxation, and final D2 → VBM recombination, respectively. In the case of the 5-State Ladder Model (CBM → D1 → D2 → D3 →VBM), the equations were the same as the equations above, but with equation (5) replaced with the following two equations:

$${\displaystyle \frac{\mathrm{d}{N}_{\mathrm{D}3}}{\mathrm{d}t}}=k_3{N}_{\mathrm{D}2}-k_4{N}_{\mathrm{D}3}$$

(9)

$${\displaystyle \frac{\mathrm{d}{N}_{\mathrm{V}\mathrm{B}\mathrm{M}}}{\mathrm{d}t}}=-k_4{N}_{\mathrm{D}3}$$

(10)

For simplicity, the three constants (k₁-k₃) of the former model and the four constants (k₁-k₄) of the latter model represent sequential trapping and recombination rates along the ladder pathway. Since the intermediate ladder steps relax much faster than the final transition, the effective nonradiative recombination lifetime was determined by the slowest step of the ladder. Optical properties were calculated using a summation over states approach [55]; see our previous works for theoretical details [56,57,58,59].

3. Results and Discussions

3.1. Defect Characterization and Band Structure Analysis

Of the defects surveyed (interstitials, vacancies, antisites, and switches, or pairs of superimposed mutually compensating antisites where adjacent atoms swap sites), only Pb_I and I_Pb exhibited TELs within the band gap, (+3/+2) and (+2/+1) for Pb_I and (0/-1) and (-1/-2) for I_Pb. Both defects also have lower energy two-carrier charge transitions [(+1/-1) for I_Pb and (+3/+1) for Pb_I], but the low probability of double carrier capture kinetically limits these transitions. Figure 1 shows the geometries of the middle charges for each defect case. For Pb_I⁺², the I forming the corners of the adjacent PbI₆ octahedra both rotate to bond with the extra Pb, whereas for I_Pb^–1 the smaller (in the sense of ionic radius) antisite I has relatively minimal impact on the local geometry, which is not surprising considering iodine’s ability to form chains [57].

Figure 2 shows the unfolded band diagrams [44] of pristine α-CsPbI₃ (Figure 2a), as well as the antisite Pb_I⁺² (Figure 2b) and I_Pb^–1 (Figure 2c) defects. The pristine band structure matches that of the expected unit cell, with a direct band gap at the R-symmetric point (see Ku et al [44] for the convoluting effect increasing supercell size has on the resulting band structure; the direct band gap should, in theory, allow for both easier carrier excitation/combination through the lack of need for kinetically hindering phonon absorption/emission in order to obey conservation of momentum. This gap is lower than the one which is experimentally observed, as expected for a mGGA functional. The disorder introduced by the defects is readily apparent in the band diagrams of the two defects in the splitting of energy levels as compared to the discrete levels in the pristine case, particularly for Pb_I⁺² for which the higher disorder results in a more continuous valence band. Both defects have energy levels which oscillate along the M-Γ-R path, corresponding to high dispersion, but which are flat along the Γ-X path. Flatter bands indicate higher effective electron masses and reduced kinetic energy,[58] making it harder for electrons in these states to escape.

3.2. Gas Absorption Effects

Next, analytes were introduced to study the effects on the band structure in search of sensing opportunities. It was previously unknown why α-CsPbI₃ has such low recombination rates, even though it has thermodynamically viable deep trap states. This behavior likely arises from the soft, anharmonic nature of the perovskite lattice: carriers persist for nanoseconds if allowed sufficient time to absorb multiple phonons [59]. This suggests that significant geometric contortions and energy barriers must be overcome to complete one or more of the charge transitions.[60] To search for analytes which could alter this delicate dynamic, band structures were examined for strong trapping potential. NH₃ showed the greatest enhancement of recombination rates, indicated by the mid-gap band levels. In contrast to the gas-less cases of Figure 2b and 2c, Figure 3 shows the relaxed geometries of the two deep traps with NH_3, along with the associated band structures.

First, by comparing Figure 3 with Figure 2, one can see that NH₃ has little effect on the band gap of defective CsPbI₃. This is likely due to the coordination of NH₃ with Cs, which contributes little to the structure of the valence and conduction bands as seen by its contribution (red) in Figure 4a. As is typical for perovskites, the X-site antibonding halogen orbitals (with a small amount of Pb(s)) dominate the valence band while the B-site lead metal dominates the conduction band [56] Figure 4b compares the total density of states (DOS) of pristine CsPbI₃ with that of I_Pb with and without NH₃; when comparing pristine CsPbI₃ with I_Pb the introduction of the trap state below the conduction band can be seen, its notable that the addition of NH₃ increases the DOS near the valence band maximum (VBM) and decreases it near the conduction band minimum (CBM).

In contrast to NH₃, most other analytes (Figures S1-S4 for Pb_I and Figures S5-S9 for I_Pb) reduced trapping potential or had no effect on the band gap of pristine and defective CsPbI₃. For example, see Figure 5a, where NO₂·Pb_I results in flat energy levels within the bands near the edges. While CH₃NH₂·Pb_I’s band diagram looks very similar to that of NH₃·Pb_I (as shown in Figure 5b) for the case of CH₃NH₂·I_Pb, in Figure 5c, it resembles NO₂·Pb_I with its absence of band gap TEL trap states. This indicates that, despite the chemically similar R–NH₂ functional group, CH₃NH₂ is unlikely to increase nonradiative recombination rate as NH₃ does, a positive indication for the selectivity of a sensor with this operating mechanism. According to SRH theory, it is likely that the perovskite + gas system would continue to have conductivity similar to its pristine counterpart when exposed to these gases.

With NH₃ identified as a potential enabler of the recombination centers Pb_I and I_Pb, the energetics along the charge transition path must be characterized. The resulting paths are shown in Figure 6. Note that these were calculated with PBE rather than SCAN, and thus the magnitude of the energy shifts corresponds to the PBE calculations (i.e., a band gap shift of 1.48 eV and respective shifts for each TEL based on the point of intersection for the charge states as calculated by eq. 1 using the PBE values). Most significantly, we note the lower difference in ∆Q between charge states than previously reported for the case of (gasless) δ-CsPbI₃.[60] This is likely due to the NH₃ absorbing some of the donated charge and mitigating consequent geometry shifts. For the anionic defects (Figure 6b and 6d), the I_Pb ( – 1/ – 2) transition is almost barrier-less for carrier capture from CBM to mid-gap state, but a sharper energy increase is observed for both charge states in the −Q direction, resulting in a |∆Q| > 20 amu^½ Å shift required for hole capture. Notably, for the I_Pb (0/ – 1) transition ∆Q between the geometries associated with each state is very small, but due to very similar ${\displaystyle \frac{d^{2}E}{d{Q}^2}}$ relatively large contortions for both e⁻ and h⁺ capture are still required, however, at much lower energy barriers. In contrast, in the case of the cationic defects, the results from the calculations along the Pb_I (+3/+2) path again show easy e⁻ capture followed by more difficult h⁺ capture, although for this case it appears only a mild ∆Q and ∆E are required, attributable to the lower ${\displaystyle \frac{d^{2}E}{d{Q}^2}}$ for Pb_I⁺² relative to Pb_I⁺³.

For Pb_I^(+2/+1), there are large barriers present. The origin of this anharmonic scaling of E vs. Q is a consequence of the significant change in the coordination environment of the antisite Pb from +1 to +2/+3; at +1 the antisite Pb remains in the displaced I’s position, while at +2/+3 the extra h⁺ on the Pb induces stronger bonding with nearby I, pulling away from the location the displaced I would occupy in the pristine lattice. This was previously described in Zhang et al.’s work, meaning this defect’s behavior matches their work qualitatively, although the geometry shift is mitigated with the presence of the NH₃.[60] Returning to I_Pb^{(0/ – 1/ – 2)}, the NH₃ almost completely heals the anharmonicity; the reason can be seen in Figure 3b. The NH₃ displaces one of the corner I’s (foreground) in what would make up part of a PbI₆ octahedron in the pristine lattice and coordinates with the Cs via the lone electron pair on the N. Additional e⁻ are apparently delocalized, which minimizes their effect on the geometry around the defect. This makes the local energy environment of the potential energy surface (PES) behave more like a harmonic oscillator, which is particularly visible in the case of the PES for NH₃·I_Pb^{(0/ – 1)} (Figure 6d), where the two curves resemble parabolas even far from their minima. This near-constant curvature results in smaller perturbations needed to overcome the finite geometric and energy barriers between the states. Although NH₃·I_Pb^(–1/–2) does retain some degree of anharmonicity, it is greatly reduced as compared to the steeply accelerating curvature in the −Q direction of e.g. Pb_I^(+2/+1), which strengthens the case for NH₃ enabling I_Pb to function as a recombination center, especially at the higher temperatures relevant to sensor operation.

3.3. Recombination Times via Nonadiabatic Molecular Dynamics

The results of our NAMD calculations are presented in Table 1 (300 K) and 2 (600 K). By comparing these tables, we see that at 300 K, NH₃ shortens the intrinsic recombination time relative to pristine (492 vs. 797 ns), while at 600 K the trend inverts, with NH₃:Pristine exhibiting a slightly longer lifetime than pristine (557 vs. 352 ns). A similar inversion of temperature dependence is observed in antisite systems: for I_Pb and Pb_I recombination slows at higher temperature (23 → 28 ns and 1 → 0 ns, respectively), whereas in their NH₃-complexed counterparts recombination instead accelerates (30 → 22 ns for NH₃:I_Pb ​ and 0 → 1 ns for NH₃:Pb_I). This consistent inversion suggests that NH₃ perturbs the recombination landscape differently from bare defects, suppressing certain scattering pathways at elevated temperature while leaving trap-assisted channels dominant at room temperature.

Table 1. Calculated band gap (E_g), root-mean-square nonadiabatic coupling (NAC) magnitude, pure dephasing time, and recombination time (τ_NR) for the pristine, NH₃·Pristine, I_Pb , NH₃·I_Pb , Pb_I, and NH₃·Pb_I, systems. Simulations at 300 K.

System	Transition	Eg (eV)	NACs (meV)	Dephasing Time (ps)	Τ (ns)
Pristine	VBM-CBM	1.68±0.04	0.77	16.49	797
NH3·Pristine	VBM-CBM	1.64±0.06	0.52	11.08	492
IPb	VBM-D3	1.74±0.04	21.27	24.29	23
	D2-D3		0.49	2.65
	D2-D1		63.62	2.66
	D1-CBM		42.81	42.66
NH3·IPb	VBM-D2	1.70±0.04	30.60	27.71	30
	D2-D1		0.28	3.12
	D1-CBM		82.06	3.17
PbI	VBM-D3	2.23±0.08	1.43	5.57	1
	D2-D3		2.40	6.98
	D2-D1		40.24	17.43
	D1-CBM		44.38	81.69
NH3·PbI	VBM-D2	2.24±0.09	4.09	5.29	0
	D2-D1		4.29	6.55
	D1-CBM		57.00	54.11

Table 1. Calculated band gap (E_g), root-mean-square nonadiabatic coupling (NAC) magnitude, pure dephasing time, and recombination time (τ_NR) for the pristine, NH₃·Pristine, I_Pb , NH₃·I_Pb , Pb_I, and NH₃·Pb_I, systems. Simulations at 300 K.

System	Transition	Eg (eV)	NACs (meV)	Dephasing Time (ps)	Τ (ns)
Pristine	VBM-CBM	1.68±0.04	0.77	16.49	797
NH3·Pristine	VBM-CBM	1.64±0.06	0.52	11.08	492
IPb	VBM-D3	1.74±0.04	21.27	24.29	23
	D2-D3		0.49	2.65
	D2-D1		63.62	2.66
	D1-CBM		42.81	42.66
NH3·IPb	VBM-D2	1.70±0.04	30.60	27.71	30
	D2-D1		0.28	3.12
	D1-CBM		82.06	3.17
PbI	VBM-D3	2.23±0.08	1.43	5.57	1
	D2-D3		2.40	6.98
	D2-D1		40.24	17.43
	D1-CBM		44.38	81.69
NH3·PbI	VBM-D2	2.24±0.09	4.09	5.29	0
	D2-D1		4.29	6.55
	D1-CBM		57.00	54.11

Table 2. Calculated band gap (E_g), root-mean-square nonadiabatic coupling (NAC) magnitude, pure dephasing time, and recombination time (τ_NR) for the pristine, NH₃·Pristine, I_Pb, NH₃·I_Pb, Pb_I, and NH₃·Pb_I, systems. Simulations at 600 K.

System	Transition	Eg (eV)	NACs (meV)	Dephasing Time (ps)	Τ (ns)
Pristine	VBM-CBM	1.68±0.04	0.77	16.49	797
NH3·Pristine	VBM-CBM	1.64±0.06	0.52	11.08	492
IPb	VBM-D3	1.74±0.04	21.27	24.29	23
	D2-D3		0.49	2.65
	D2-D1		63.62	2.66
	D1-CBM		42.81	42.66
NH3·IPb	VBM-D2	1.70±0.04	30.60	27.71	30
	D2-D1		0.28	3.12
	D1-CBM		82.06	3.17
PbI	VBM-D3	2.23±0.08	1.43	5.57	1
	D2-D3		2.40	6.98
	D2-D1		40.24	17.43
	D1-CBM		44.38	81.69
NH3·PbI	VBM-D2	2.24±0.09	4.09	5.29	0
	D2-D1		4.29	6.55
	D1-CBM		57.00	54.11

Table 2. Calculated band gap (E_g), root-mean-square nonadiabatic coupling (NAC) magnitude, pure dephasing time, and recombination time (τ_NR) for the pristine, NH₃·Pristine, I_Pb, NH₃·I_Pb, Pb_I, and NH₃·Pb_I, systems. Simulations at 600 K.

System	Transition	Eg (eV)	NACs (meV)	Dephasing Time (ps)	Τ (ns)
Pristine	VBM-CBM	1.68±0.04	0.77	16.49	797
NH3·Pristine	VBM-CBM	1.64±0.06	0.52	11.08	492
IPb	VBM-D3	1.74±0.04	21.27	24.29	23
	D2-D3		0.49	2.65
	D2-D1		63.62	2.66
	D1-CBM		42.81	42.66
NH3·IPb	VBM-D2	1.70±0.04	30.60	27.71	30
	D2-D1		0.28	3.12
	D1-CBM		82.06	3.17
PbI	VBM-D3	2.23±0.08	1.43	5.57	1
	D2-D3		2.40	6.98
	D2-D1		40.24	17.43
	D1-CBM		44.38	81.69
NH3·PbI	VBM-D2	2.24±0.09	4.09	5.29	0
	D2-D1		4.29	6.55
	D1-CBM		57.00	54.11

3.4. Ammonia-Driven Phase Change and Photoluminescent Detection Pathways

For the gas-less defects, a large supercell reveals that the cubic symmetry is preserved beyond the immediately adjacent unit cells (Figure 7a). While Figure 1a highlights the locally broken symmetry, Figure 7a demonstrates that the distortions do not propagate. In contrast, NH₃ interacts strongly with the Pb atom at the center of the PbI₆ polyhedron, inducing rotations throughout the 4x4x4 supercell and driving a α → δ phase change.

To evaluate photoluminescent sensing potential, optoelectronic calculations were also conducted and are shown in Figure 8, implying again the stronger interaction of NH₃ with I_Pb. The presence of either defect significantly lowers the absorption coefficient, as seen by comparing the solid lines in each subfigure. In Figure 8a, the cationic defect interacts most strongly with CO₂ and NO₂ and least with CO, NO, and H₂. R–NH₂ compounds fall between these extremes. In Figure 8b, the effect is more binary: analytes without an amine group follow a trend of moderate absorption reduction, while CH₃NH₂ and NH₃ show increased absorption. This suggests that a light-absorption-based sensor could effectively distinguish amines from non-amines but may not be suitable for differentiating NH₃ from other small amine-containing compounds. A tandem chemo-optical resistor sensor could be particularly effective, combining spectral differences with the increased resistivity caused by gas interactions with deep carrier traps.

4. Conclusions

The point defect physics and electronic structure of CsPbI₃ were analyzed, confirming previous reports that the I_Pb and Pb_I antisite defects are the only potential nonradiative recombination centers. Additional calculations incorporating gas analytes revealed that NH₃ significantly influences the trap state energy levels by coordinating with the metal cations (Pb for Pb_I, Cs for I_Pb). Since prior studies suggest that anharmonicity can hinder recombination, we examined the energetics of the recombination processes, revealing that in the case of I_Pb the presence of NH₃ on the site reduces the anharmonicity and may allow it to function as a recombination center at higher temperatures. This was explored using nonadiabatic molecular dynamics, revealing potential for ammonia sensing at conditions relevant to industrially relevant processes. Phase change inducement was noted as an alternative sensing mechanism, and optoelectronic calculations were conducted, indicating that light absorption may be sensitive to amines. Taken together, this work contributes a novel band-driven approach to studying gas-solid interactions while also demonstrating an NH₃ sensor ripe for industrial applications.