betaFeedback

WorksPhysics and Astronomy

First-principles calculations of internal conversion processes in spin defects

Abstract

Optically active spin defects are foundational for quantum technologies, yet common approximations underestimate their internal conversion (IC) rates by orders of magnitude. We propose a broad, predictive framework to compute IC rates that incorporates multi-configurational effects via many-body wavefunctions in TDDFT, and includes all-phonon-mode contributions via analytical non-adiabatic couplings. Our approach resolves discrepancies with experiment, achieving quantitative agreement for the NV−^- center in diamond, and identifying a previously overlooked non-radiative channel in the divacancy triplet lifetime in SiC.

Introduction

Understanding and manipulating the properties of defects in solids has driven technological innovation for decades. Recently, optically active spin defects, e.g., the negatively charged nitrogen-vacancy (NV⁻) center in diamond and the neutral divacancy (VV⁰) center in silicon carbide (SiC) have emerged as promising platforms for the development of quantum technologies[1, 2, 3, 4, 5, 6, 7, 8].

Both experiments and first-principles, theoretical studies have shed light on many aspects of the NV⁻ and VV⁰ optical cycle[9, 10, 11] [12, 13, 14, 15, 16, 17], illustrated in Figure 1. Radiative transitions and inter-system crossing rates have been investigated using various first-principles frameworks[9] [12, 13, 14, 15, 18, 19, 20]; however, internal conversion (IC) processes remain much less explored, thus hampering a full description of the optical cycle from first principles. These non-radiative processes, arising from thermal excitations due to the coupling of electronic and nuclear degrees of freedom, are key to understanding the effect of temperature on the optical cycle[11] and to predict excited-state lifetimes[11] [21, 22]. A detailed description of the lifetimes of excited states, and of the competition between radiative and non-radiative processes, is necessary to determine the efficiency of the readout process[23] in an optical cycle, and to interpret and predict optically-detected magnetic resonance (ODMR) spectra[24].

Schematic representation of the electronic energy levels of the NV⁻ center in diamond and the VV⁰ center in silicon carbide

Figure 1. Schematic representation of the electronic energy levels of the NV⁻ center in diamond and of the VV⁰ center in silicon carbide. Both systems exhibit a triplet ground state (3A2{}^{3}A_{2}) and triplet excited state (3E{}^{3}E), and two singlet shelving states, 1A1{}^{1}A_{1} and 1E{}^{1}E, all labeled according to the C3vC_{3v} point group. In an optical cycle, after a laser excitation from 3A2{}^{3}A_{2} to 3E{}^{3}E, the system can either return to the ground state or transition to 1A1{}^{1}A_{1} through inter-system crossing (ISC); from 1A1{}^{1}A_{1}, the electron then relaxes to 1E{}^{1}E, through non-radiative spin-conserving transitions (internal conversion, IC), and finally to the ground state through a second ISC process. The magnetic sublevels of the ground state represent an effective qubit system[9]. Radiative and non-radiative transitions are indicated by solid and dashed lines, respectively.

The central challenges in describing IC processes are the accurate description of the many-body excited states of the defects and of their electron-phonon (e–ph) coupling. Ab-initio methods based on one-dimensional effective modes and Kohn-Sham electronic orbitals have been successfully used in the literature to describe non-radiative capture rates of charged carriers in several semiconductors and insulators in the strong-coupling regime[25–29]. However, their application to neutral excitations in spin-defective solids resulted in inaccurate lifetimes’ estimates[24, 30].

In this Letter, we present an efficient and accurate ab-initio framework for the calculation of non-radiative rates of internal conversion processes in spin defects. We evalu-

ate the non-adiabatic coupling (NAC) between electronic excited states using linear-response time-dependent density functional theory (LR-TDDFT) with hybrid functionals to capture the multi-configurational character of the excited states, missing in single-orbital descriptions. In addition, we compute analytical derivatives of non-adiabatic couplings, thus enabling the inclusion of all phonon modes in the calculations of rates. We apply the framework to study IC processes in the optical cycles of the NV⁻ center in diamond and the VV⁰ center in SiC. For the first time, we obtain results in excellent agreement with experiments for the NV−^{-} and we identify a substantial non-radiative contribution to the decay of the triplet excited state of the VV0^{0} center, resolving a long-standing discrepancy with experiments.

**Theoretical framework—**We describe the internal conversion processes as non-radiative transitions between two many-body electronic states with the same spin, occurring via multi-phonon exchange[31–36]. We define a temperature-dependent non-radiative transition rate (NRTR) ΓNR,IF(T)\Gamma_{\mathrm{NR},IF}(T) between an initial (II) and a final (FF) state, using the Fermi’s golden rule at first-order in the perturbing e–ph interaction[32, 37–39]:

ΓNR,IF(T)=2πℏ∑kNmodes∣MIFk∣2XIFk(T).(1)\Gamma_{\mathrm{NR},IF}(T)=\frac{2\pi}{\hbar}\sum_{k}^{N_{\mathrm{modes}}}\lvert M_{IF}^{k}\rvert^{2}X_{IF}^{k}(T). \tag*{(1)}

(1) accounts for the contribution of all the NmodesN_{\mathrm{modes}} phonon modes of the solid. The term MIFkM_{IF}^{k} defines the e–ph matrix element between the initial and final electronic states and measures the strength of the coupling with mode kk. The term XIFk(T)X_{IF}^{k}(T) is the phonon-relaxation contribution of the kk-th mode due to the rearrangement of the atoms from the relaxed geometry RIR_{I} of state II to that of state FF (RFR_{F}); this term reads:

XIFk(T)=∑n,mρn(T)∣⟨ΦI,n∣Qk∣ΦF,m⟩∣2×δ(EI,ntot−EF,mtot),(2)X_{IF}^{k}(T)=\sum_{n,m}\rho_{n}(T)\lvert\langle\Phi_{I,n}|Q_{k}|\Phi_{F,m}\rangle\rvert^{2}\times\delta(E_{I,n}^{\mathrm{tot}}-E_{F,m}^{\mathrm{tot}}), \tag*{(2)}

where ∣ΦI,n⟩\lvert\Phi_{I,n}\rangle (∣ΦF,m⟩\lvert\Phi_{F,m}\rangle) is the multi-phonon wave-function of the vibrational configuration nn (mm) of the electronic state II (FF). The initial state nn is assumed to be the equilibrium distribution ρn(T)\rho_{n}(T) at temperature TT, and we consider all possible final vibrational states compatible with the conservation of the total energy, nuclear plus electronic. This requirement is enforced through the delta-function in (2), where:

EI,ntot−EF,mtot=ΔIF+∑jNmodes(nj+12)ℏωjI−∑jNmodes(mj+12)ℏωjF.(3)E_{I,n}^{\mathrm{tot}}-E_{F,m}^{\mathrm{tot}}=\Delta_{IF}+\sum_{j}^{N_{\mathrm{modes}}}\left(n_{j}+\frac{1}{2}\right)\hbar\omega_{j}^{I}-\sum_{j}^{N_{\mathrm{modes}}}\left(m_{j}+\frac{1}{2}\right)\hbar\omega_{j}^{F}. \tag*{(3)}

with ΔIF=EI(RI)−EF(RF)\Delta_{IF}=E_{I}(R_{I})-E_{F}(R_{F}) the adiabatic energy difference. Hereafter, we adopt the displaced harmonic oscillator approximation and consider ℏωjF=ℏωjI\hbar\omega_{j}^{F}=\hbar\omega_{j}^{I}.

The e–ph matrix element MIFkM_{IF}^{k} in (1) is defined as

MIFk=⟨ΨI∣∂Hel∂Qk∣ΨF⟩,(4)M_{IF}^{k}=\langle\Psi_{I}|\frac{\partial H_{\mathrm{el}}}{\partial Q_{k}}|\Psi_{F}\rangle, \tag*{(4)}

where ∣ΨI⟩\lvert\Psi_{I}\rangle and ∣ΨF⟩\lvert\Psi_{F}\rangle are the many-body wavefunctions of the initial and final electronic states, respectively, HelH_{\mathrm{el}} is the electronic many-body Hamiltonian, and QkQ_{k} is the normal coordinate of mode kk. All quantities are computed at RIR_{I}, using the following relations:

∂Hel∂Qk=∑ANatoms∇RAHel⋅eAkMA,(5)\frac{\partial H_{\mathrm{el}}}{\partial Q_{k}}=\sum_{A}^{N_{\mathrm{atoms}}}\nabla_{R_{A}}H_{\mathrm{el}}\cdot\frac{\mathbf{e}_{A}^{k}}{\sqrt{M_{A}}}, \tag*{(5)}
⟨ΨI∣∇RAHel∣ΨF⟩=(EF−EI)⟨ΨI∣∇RAΨF⟩,(6)\langle\Psi_{I}|\nabla_{R_{A}}H_{\mathrm{el}}|\Psi_{F}\rangle=(E_{F}-E_{I})\langle\Psi_{I}|\nabla_{R_{A}}\Psi_{F}\rangle, \tag*{(6)}

where MAM_{A} is the atomic mass, ek\mathbf{e}^{k} is the kk-th eigenvector of the dynamical matrix, and EIE_{I} and EFE_{F} are the energies of the initial and final electronic states, respectively. We rewrite (4) as

MIFk=(EF−EI)∑ANatomsdIF,A⋅eAkMA,(7)M_{IF}^{k}=(E_{F}-E_{I})\sum_{A}^{N_{\mathrm{atoms}}}\mathbf{d}_{IF,A}\cdot\frac{\mathbf{e}_{A}^{k}}{\sqrt{M_{A}}}, \tag*{(7)}

where we have introduced the non-adiabatic coupling (NAC) vectors dIF,A\mathbf{d}_{IF,A}, defined as

dIF,A=⟨ΨI∣∇RAΨF⟩.(8)\mathbf{d}_{IF,A}=\langle\Psi_{I}|\nabla_{R_{A}}\Psi_{F}\rangle. \tag*{(8)}

The NACs are central quantities in the description of ultra-fast photo-chemical and photo-physical processes, especially when two adiabatic potential energy surfaces are close to each other or cross[40–43]. They also play a key role in non-adiabatic molecular dynamics[44–47]. Here, we use (7) to compute the e–ph matrix elements in terms of the vectors dIF,A\mathbf{d}_{IF,A}. We implemented the calculation of NACs in the LR-TDDFT framework in the WEST code[14, 48, 49] using an analytical formulation based on the extended Lagrangian approach[50–54]. Hence, we avoid the use of finite differences to compute wave-function derivatives, and we can efficiently include all phonon modes while accounting for the many-body nature of the electronic states. Our implementation is optimized for GPU platforms and supports hybrid functionals and spin-flip excitations[14, 49, 54–56].

To compute the NRTRs from (1), we adopt the generating function approach[32, 39]. We define a line-shape function FNR(Δ)F_{\mathrm{NR}}(\Delta) (the subscripts II and FF and the temperature TT are omitted here):

FNR(Δ)=gℏ2∑kNmodes∣MIFk∣2ℜ∫−∞∞Bk(E′)G(Δ−E′) dE′,(9)F_{\mathrm{NR}}(\Delta)=\frac{g}{\hbar^{2}}\sum_{k}^{N_{\mathrm{modes}}}\lvert M_{IF}^{k}\rvert^{2}\Re\int_{-\infty}^{\infty}B_{k}(E')G(\Delta-E')\,\mathrm{d}E', \tag*{(9)}

where gg is the degeneracy of the final state[25], and Bk(E)B_{k}(E) and G(E)G(E) are TT-dependent spectral functions that accounts for multi-phonon processes (See Appendix). Finally, we obtain the value of the NRTR of a given IC process when the energy Δ\Delta equals the adiabatic energy difference ΔIF\Delta_{IF}, namely ΓNR,IF=FNR(ΔIF)\Gamma_{\mathrm{NR},IF}=F_{\mathrm{NR}}(\Delta_{IF}).

**Results—**We first consider the IC processes between the singlet states of the NV−^{-} center in diamond. Using a recently developed spin-flip TDDFT framework[49], we are able to capture the multi-configurational character of these states, absent in Kohn-Sham orbital descriptions. With a zero-phonon line (ZPL) in the infra-red (IR) at ∼1.19\sim1.19 eV[57, 58], the 1A1→1E^{1}A_{1}\to{}^{1}E transition is predominantly non-radiative, as indicated by the measured integrated IR emission intensities, which are three orders of magnitude lower than those of the visible 3E→3A2^3E \to^3A_2 emission[57]. To shed light on the role of IC processes in the non-radiative decay channel, we computed the NRTRs presented in Figure 2(a). We used a 215-atom supercell, and the adiabatic energy difference ΔIF=1.397\Delta_{IF} = 1.397 eV obtained with TDDFT and the DDH functional[49]. Calculations performed with the PBE functional yield qualitatively similar results (See SM). In the range T=0∼400T = 0 \sim400 K we find rates of the order of ≈101\approx10^{1} GHz. Since the predicted radiative rate is ΓR≃0.53\Gamma_{\mathrm{R}} \simeq0.53 MHz[59], our results indicate a predominant role of IC processes in the decay of the 1A1^1A_1 state.

Non-radiative transition rates, lifetimes, electron–phonon spectral density, and line-shape functions

Figure 2. (a): Non-radiative transition rates between the singlet states in the NV−^- center in diamond (from Eq. (1)). (b): Computed lifetimes of the 1A1^1A_1 state compared to the experimental results of Ulbricht et al.[58] [a], and Acosta et al.[57] [b]. (c): Computed electron-phonon spectral density function Eq. (10), exhibiting a peak at the energy E≃170E \simeq170 meV of the vibrational mode localized around the defect (shown in the inset). (d): Comparison between line-shape functions obtained within different approximations; the black dashed line indicates the TDDFT adiabatic energy difference ∼1.397\sim1.397 eV.

We compared the lifetime τ1A1\tau^{^1A_1} of the upper singlet state with experiments from Refs. [57] and [58], as shown in Figure 2(b). Acosta et al. carried out fluorescence lifetime measurements using the phase-shift technique, where a sinusoidal modulation of the intensity of an optical pump beam results in oscillations of the time-dependent IR fluorescence. From the relative phase shift between the pump and the response, they estimated τ1A1=0.9(5)\tau^{^1A_1} = 0.9(5) ns at T=4T = 4 K, an order of magnitude smaller than the one evaluated here. We note, however, the large relative error of the phase-shift estimate in the experiments of Acosta et al., which might be responsible for an inaccurate result. Ulbricht et al. performed femtosecond transient absorption pump-probe spectroscopy, using a femtosecond optical pump to activate the optical cycle and a femtosecond probe resonant with the transition between the singlet states to study the population dynamics. From the time-dependent normalized differential transmission measured in two different samples, they inferred lifetimes of τ1A1=102.3(3)\tau^{^1A_1} = 102.3(3) ps and τ1A1=92.1(5)\tau^{^1A_1} = 92.1(5) ps, at T=78T = 78 K, in accord with our results (τ1A1=ΓNR−1=117.4\tau^{^1A_1} = \Gamma_{\mathrm{NR}}^{-1} = 117.4 ps at T=78T = 78 K). This good agreement further confirms that IC processes are the main cause of non-radiative decays for the 1A1→1E^1A_1 \to^1E transition.

It is interesting to define and analyze the e–ph spectral density function

D(E)=12πℏ∑kNmodes∣MIFk∣2Bk(E)(10)D(E) = \frac{1}{2\pi\hbar} \sum_{k}^{N_{\mathrm{modes}}} \left|M^{k}_{IF}\right|^2 B_k(E) \tag*{(10)}

to identify the role played by various phonon modes in the IC process. The vibrational mode localized around the defect at E≃170E \simeq170 meV, shown in Figure 2(c), has a strong non-adiabatic character, and it might be expected to be the main driver of IC processes. However, when computing the NRTRs by considering only this mode in the sum of Eq. (10), we obtain ΓNR,loc=4.37\Gamma_{\mathrm{NR,loc}} = 4.37 GHz at T=78T = 78 K, which is roughly half the rate computed with the inclusion of all the modes (ΓNR=8.52\Gamma_{\mathrm{NR}} = 8.52 GHz). This comparison highlights the importance of including all phonon modes to obtain accurate results, and not only those localized around the defect.

Our framework represents a substantial improvement over commonly used methods in which, to reduce the computational cost of Eq. (1) two approximations to XIFkX_{IF}^{k} are usually adopted: the single-mode approximation[30], where only one of the NmodesN_{\mathrm{modes}} modes is considered, or the accepting-mode approximation[26], where an effective mode QQ is constructed from the mass-weighted displacement between the initial RIR_I and the final RFR_F atomic configurations and the vibrational energies are extracted from the one-dimensional configuration-coordinate diagrams of the electronic energies with respect to QQ. In Table I we compare the NRTRs computed with different approaches, and in Figure 2(d) we show the line-shape functions obtained with the different approximations. Compared to the all-modes line-shape function, the single-mode approximation describes better the tail of the line-shape, whereas the accepting-mode approximation performs better at lower energies. We also compare the NRTR at T=300T = 300 K computed in the accepting-mode approximation ΓNR,acc.=2.49\Gamma_{\mathrm{NR,acc.}} = 2.49 GHz against the value ΓNR,acc.=0.048\Gamma_{\mathrm{NR,acc.}} = 0.048 GHz obtained using Kohn-Sham electronic orbitals in Ref. [24]. The orders of magnitude difference between the two approaches highlights the importance of an accurate description of the many-body character of the electronic states. (See Appendix).

ΓNR\Gamma_{\mathrm{NR}} [1A1→1E^{1}A_{1} \rightarrow{}^{1}E] (GHz)
Acc. modeLoc. modesAll modes
T=4T = 4 K
TDDFT3.274.378.51
Exp.1––1.1(6)
T=78T = 78 K
TDDFT3.274.378.52
Exp.2––10.858(59)
9.775(29)

Table I. Comparison of non-radiative transition rates between the singlet states of the NV−^{-} center in diamond obtained from first-principles with different approximations, compared with experimental results.

For completeness, we also considered the 3E→3A2{}^{3}E \rightarrow{}^{3}A_{2} transition in NV−^{-}, which is known to be predominantly radiative, with an experimental ZPL ∼1.95\sim1.95 eV. Recent first-principles estimates of the lifetime τ3E\tau^{{}^{3}E} of the triplet excited state including radiative and non-radiative ISC contributions, agree well with experimental measurements[15, 49], suggesting a negligible role of IC processes. Consistent with this interpretation, here we predict ΓNR≲5\Gamma_{\mathrm{NR}} \lesssim5 MHz in the range T=0∼400T = 0 \sim400 K, a negligible contribution compared to the radiative rate ΓR≃83.3\Gamma_{\mathrm{R}} \simeq83.3 MHz[15]. This finding is not surprising since the ground and excited triplet states are well separated in energy. Our results were obtained considering the adiabatic energy difference ΔIF=2.112\Delta_{\mathrm{IF}} = 2.112 eV, as obtained with TDDFT, and the DDH functional. In Figure 3(a) we show the e–ph spectral function D(E)D(E). We do not observe any preferential contribution from any specific mode, suggesting that single-mode approximations may not be valid. Indeed, the line-shape function obtained within the accepting-mode approximation, Figure 3(b), underestimates by several order of magnitude the rates obtained when including all phonons.

Computed electron-phonon spectral density function and comparison of line-shape functions

Figure 3. (a): Computed electron-phonon spectral density function D(E)D(E) [Eq. (10)] for the transition between the triplet states of the NV−^{-} center in diamond. (b): Comparison between line-shape functions obtained within different approximations. The black dashed line indicates the TDDFT adiabatic energy difference ΔIF=2.112\Delta_{\mathrm{IF}} = 2.112 eV.

We now turn to IC processes occurring in the optical

cycle of the kkkk-VV0\mathrm{V}_{\mathrm{V}}^{0} center in the 4H-SiC polytype[60]. Although non-radiative decays are expected to play an important role[59], a thorough understanding of the contribution of IC processes has been hampered by the lack of an adequate computational framework. We first consider the 3E→3A2{}^{3}E \rightarrow{}^{3}A_{2} transition. The energy separation between the triplet ground and excited states (with a ZPL ∼1.096\sim1.096 eV) has often been considered to be sufficiently large to render non-radiative contributions from IC processes negligible. Nevertheless, state-of-the-art first-principles estimates of the lifetime of the 3E{}^{3}E state obtained by considering only the radiative contribution overestimate the experimental measurements of τ3E≃15\tau^{{}^{3}E} \simeq15 ns[61, 62], when using either DFT (τR3E≃36.8\tau_{\mathrm{R}}^{{}^{3}E} \simeq36.8 ns)[13] or constrained DFT (τR3E≃28\tau_{\mathrm{R}}^{{}^{3}E} \simeq28 ns)[11]. Here, we include the contributions of IC processes using the relation

τ3E=[1τR3E+ΓNR[3E→3A2]]−1.(11)\tau^{{}^{3}E} = \left[ \frac{1}{\tau_{\mathrm{R}}^{{}^{3}E}} + \Gamma_{\mathrm{NR}}[{}^{3}E \rightarrow{}^{3}A_{2}] \right]^{-1}. \tag*{(11)}

We considered a 398-atom supercell, we used the DDH functional and computed the adiabatic energy difference ΔIF=1.351\Delta_{\mathrm{IF}} = 1.351 eV[49] with TDDFT. (A comparison with PBE results is given in the SM). We present our results in Figure 4(a), where we show that the inclusion of IC processes improves the agreement with experiments. In Figure 4(b) we display the e–ph spectral function D(E)D(E), and in Figure 4(c) we analyze the line-shape functions, comparing with the results with the accepting-mode approximation. Similar to the decay between triplet states in the NV−^{-}, we observe no predominant phonon mode contribution; indeed, the effective mode approximation severely underestimates the NRTRs, highlighting once again the importance of an approach that includes all-phonon-modes.

Three panels showing radiative lifetime, electron-phonon spectral density, and line-shape functions

Figure 4. (a): Radiative lifetime of the 3E^{3}\mathrm{E} excited state of the kkkk-VV0\mathrm{VV}^{0} center in 4H-SiC [13] (gray dashed line), computed at the DFT level of theory with the PBE functional. The red squares show the total computed lifetime, including non-radiative decays, and the blue and purple dots indicate experimental measurements from Ref.s [62] [a] and [61] [b]. (b): Computed electron-phonon spectral density function D(E)D(E) [Eq. (10)]. (c): Comparison between line-shape functions obtained within different approximations. The black dashed line indicates the TDDFT adiabatic energy difference ΔIF=1.351 eV\Delta_{IF}=1.351\ \mathrm{eV}.

Finally, we analyze the 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition, which has not yet been investigated experimentally and for which neither a ZPL estimate nor singlet-state lifetime measurements are available. Hence our results serve as predictions. We computed the NRTRs using the adiabatic energy difference ΔIF=1.132 eV\Delta_{IF}=1.132\ \mathrm{eV} obtained from TDDFT [49]. (See a comparison with PBE results in the SM). We obtain rates of the order of ≈101 GHz\approx10^{1}\ \mathrm{GHz}, similar to those computed for the NV−^{-}; comparing our results with the first-principles value of the radiative rate ΓR=0.32 MHz\Gamma_{\mathrm{R}}=0.32\ \mathrm{MHz} [59], we conclude that the 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition is dominated by IC processes.

Discussion—We have introduced a general, efficient first-principles framework for computing non-radiative spin-conserving transition rates in spin defects, addressing a longstanding gap in the theoretical description of the optical cycle of these systems. Two advances are central to our approach. First, we treat the electronic states of defects as many-body wavefunctions computed within LR-TDDFT with hybrid functionals; hence we capture multi-configurational effects that single-particle Kohn-Sham orbitals cannot describe; indeed, using Kohn-Sham single-particle wavefunctions underestimates rates by orders of magnitude. Second, we compute non-adiabatic coupling vectors analytically using an extended Lagrangian formulation; we thus eliminate the bottleneck of finite-difference calculations of wave-function derivatives, and enable the inclusion of all phonon modes for supercells with hundreds of atoms. Applying our framework to the NV−^{-} center in diamond, we find IC rates for the singlet 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition in quantitative agreement with femtosecond transient absorption measurements, and confirm the negligible role of IC in the 3E→3A2^{3}\mathrm{E} \rightarrow{}^{3}\mathrm{A}_{2} decay. For the VV0^{0} center in 4H-SiC, we identify a significant non-radiative contribution to the 3E^{3}\mathrm{E} lifetime that has been previously overlooked, resolving a discrepancy between computed and measured lifetimes present in the literature. For the singlet 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition in VV0^{0}, where no experimental data exist, our calculations serve as quantitative predictions. For defects in both diamond and SiC, we demonstrate that effective single-mode and accepting-mode approximations underestimate non-radiative rates by orders of magnitude.

Together with first-principles methods for inter-system crossing, our framework enables parameter-free predictions of complete optical cycles and ODMR spectra of spin defects. Our approach is broadly applicable to non-adiabatic processes in heterogeneous solids and molecules, and the analytical NAC implementation paves the way for ab initio non-adiabatic molecular dynamics in extended systems and the study of polaron and exciton dynamics in solids. Work in progress to further reduce computational cost using machine-learned dynamical matrices and NAC surrogate models.

Acknowledgments—We gratefully acknowledge fruitful discussions with Chris Anderson, Enrico Drigo, and Michael Nevins. The computational work was supported by the Midwest Integrated Center for Computational Materials (MICCoM) as part of the Computational Materials Sciences Program funded by the U.S. Department of Energy and the study of spin defects was supported by AFOSR grant # FA95502210370. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. ERCAP0036175 and resources of the University of Chicago Research Computing Center (RCC). The Flatiron Institute is a division of the Simons Foundation.

Data availability—The code, inputs and data that support the findings of this article are openly available through the Qresp platform [63].

† [email protected]

References

Generating function approach

Here, we provide the expressions for the generating functions G(E)G(E) and Bk(E)B_k(E) entering Eq. (1). They are derived from the phonon-relaxation term, Eq. (2), in the displaced harmonic oscillator approximation, and they describe the multi-phonon processes at temperature TT, that account for the conservation of the total energy and for the overlap between Franck-Condon-shifted phononic wavefunctions[39]. The generating function G(E)G(E) reads

G(E)=12πℏ∫−∞∞G(t)eiEt−γ∣t∣ℏ dt(12)G(E) = \frac{1}{2\pi\hbar}\int_{-\infty}^{\infty}G(t)e^{\frac{iEt-\gamma\lvert t\rvert}{\hbar}}\,dt \tag*{(12)}

where γ\gamma is a Lorentzian smearing factor accounting for the lifetime broadening of the electronic levels[39], and it holds

G(t)=∏k=1NmodesGk(t)=exp⁡{∑kNmodesSk[(nˉk+1)eiωkt+nˉke−iωkt−(2nˉk+1)]}(13)G(t) = \prod_{k=1}^{N_{\mathrm{modes}}}G_k(t) = \exp\left\{\sum_{k}^{N_{\mathrm{modes}}}S_k\left[(\bar{n}_k+1)e^{i\omega_k t}+\bar{n}_k e^{-i\omega_k t}-(2\bar{n}_k+1)\right]\right\} \tag*{(13)}

where ℏωk\hbar\omega_k is the energy of the kk-th mode, nˉk=(eℏωk/kBT−1)−1\bar{n}_k=(e^{\hbar\omega_k/k_BT}-1)^{-1} is the thermal occupation at temperature TT, kBk_B is the Boltzmann constant, and SkS_k is the partial Huang-Rhys factor[31], namely

Sk=ωkΔQk22ℏ(14)S_k=\frac{\omega_k\Delta Q_k^2}{2\hbar} \tag*{(14)}

with

ΔQk=1ωk2∑A=1NatomsFAMA⋅eAk(15)\Delta Q_k=\frac{1}{\omega_k^2}\sum_{A=1}^{N_{\mathrm{atoms}}}\frac{F_A}{\sqrt{M_A}}\cdot e_A^k \tag*{(15)}

where F\mathbf{F} is the final-state (excited-state) force evaluated at the initial state relaxed geometry. The function Bk(E)B_k(E), instead, reads:

Bk(E)=∫−∞∞Bk(t)eiEtℏ dt(16)B_k(E)=\int_{-\infty}^{\infty}B_k(t)e^{\frac{iEt}{\hbar}}\,dt \tag*{(16)}

where

Bk(t)=ℏ2ωk{eiωkt+2nˉkcos⁡(ωkt)+Sk[−eiωkt−2inˉksin⁡(ωkt)+1]2}.(17)B_k(t)=\frac{\hbar}{2\omega_k}\left\{e^{i\omega_k t}+2\bar{n}_k\cos(\omega_k t)+S_k\left[-e^{i\omega_k t}-2i\bar{n}_k\sin(\omega_k t)+1\right]^2\right\}. \tag*{(17)}

Effective-modes approximations

We provide below details on our calculations of NRTRs in the accepting-mode approximation. Following a standard procedure, we defined an effective mode QQ as the mass-weighted atomic displacement from the relaxed geometry of the initial state RIR_I to that of the final state RFR_F, namely

Q2=∑ANatoms∑i=x,y,zMA(Ri,A−RI;iA)2,(18)Q^2=\sum_{A}^{N_{\mathrm{atoms}}}\sum_{i=x,y,z}M_A(R_{i,A}-R_{I;iA})^2, \tag*{(18)}

where the displacement Ri,AR_{i,A} of atom AA along the Cartesian direction ii is proportional to RI;iA−RF;iAR_{I;iA}-R_{F;iA}. We defined the non-radiative rate ΓNR,acc.(T)\Gamma_{\mathrm{NR,acc.}}(T) as

ΓNR,acc.(T)=2πℏ∣MIF∣2XIF(T).(19)\Gamma_{\mathrm{NR,acc.}}(T)=\frac{2\pi}{\hbar}\lvert M_{IF}\rvert^2X_{IF}(T). \tag*{(19)}

The e–ph coupling term MIFM_{IF} reads

MIF=⟨ΨI∣∂Hel∂Q∣ΨF⟩,(20)M_{IF}=\left\langle\Psi_I\left\lvert\frac{\partial H_{\mathrm{el}}}{\partial Q}\right\rvert\Psi_F\right\rangle, \tag*{(20)}
=(EF−EI)⟨ΨI∣∂ΨF∂Q⟩,(21)=(E_F-E_I)\left\langle\Psi_I\left\lvert\frac{\partial\Psi_F}{\partial Q}\right.\right\rangle, \tag*{(21)}
=(EF−EI)∑ANatomsdIF,A⋅r^IF,AMA(22)=(E_F-E_I)\sum_A^{N_{\mathrm{atoms}}}\mathbf{d}_{IF,A}\cdot\frac{\hat{\mathbf{r}}_{IF,A}}{\sqrt{M_A}} \tag*{(22)}

where we projected the NAC vectors along the direction r^IF\hat{\mathbf{r}}_{IF} of the atomic displacement from RIR_I to RFR_F, namely

r^IF=RI−RF∣RI−RF∣.(23)\hat{\mathbf{r}}_{IF}=\frac{R_I-R_F}{\lvert R_I-R_F\rvert}. \tag*{(23)}

The phonon-relaxation term XIF(T)X_{IF}(T) reads

XIF(T)=∑m,nρn(T)∣⟨ΦI,n∣Q∣ΦF,m⟩∣2×δ(EI,ntot−EF,mtot),(24)X_{IF}(T)=\sum_{m,n}\rho_n(T)\lvert\langle\Phi_{I,n}\lvert Q\rvert\Phi_{F,m}\rangle\rvert^2\times\delta(E^{\mathrm{tot}}_{I,n}-E^{\mathrm{tot}}_{F,m}), \tag*{(24)}

where nn and mm are the occupation numbers of the initial and final effective-mode vibrational states, respectively, and

EI,ntot−EF,mtot=ΔIF+(n+12)ℏωI−(m+12)ℏωF,(25)E^{\mathrm{tot}}_{I,n}-E^{\mathrm{tot}}_{F,m}=\Delta_{IF}+\left(n+\frac{1}{2}\right)\hbar\omega_I-\left(m+\frac{1}{2}\right)\hbar\omega_F, \tag*{(25)}

The vibrational energies ℏωI\hbar\omega_I and ℏωF\hbar\omega_F are extracted from the configuration-coordinate diagram of the electronic energies EIE_I and EFE_F as a function of QQ, respectively.

We used the accepting-mode approximation to compute the NRTRs of the IC processes occurring in the NV−^- in diamond and the VV0^0 in 4H-SiC, with the parameters given in Table II, which were obtained with TDDFT and the DDH functional. For the 3E→3A2{}^3E\to{}^3A_2 transitions, we extracted the values of ℏωI\hbar\omega_I and ℏωF\hbar\omega_F from the configuration-coordinate energy diagrams obtained with the DDH functional. For the 1A1→1E{}^1A_1\to{}^1E transition of the NV−^- in diamond, we used the same values

ΔQ\Delta Q (Å amu1/2^{1/2})ℏωI\hbar\omega_I (meV)ℏωF\hbar\omega_F (meV)
3E→3A2^3\mathrm{E} \rightarrow^3\mathrm{A}_2 in NV−^-0.6372.9666.54
3E→3A2^3\mathrm{E} \rightarrow^3\mathrm{A}_2 in kkkk-VV0^00.7836.4138.70
1A1→1E^1\mathrm{A}_1 \rightarrow^1\mathrm{E} in NV−^-0.4274.0787.34

Table II. Parameters used to compute the NRTRs in the accepting-mode approximation.

adopted in Ref. [24] to ensure a consistent comparison of the rates; the important difference is that our rates are obtained using the NACs to compute the e–ph coupling term instead of finite differences of Kohn-Sham orbitals.

We emphasize that the use of NACs in Eq. (22) already constitutes a fundamental advancement over common practices, even in the effective-mode approximation. Indeed, it eliminates the need to compute wave-function derivatives with finite differences, and allows for the inclusion of multi-configurational effects in the description of the electronic states, which is missing in single-particle approaches. In Table III we compare the values of MIFM_{IF} computed with NACs against those obtained using Kohn-Sham electronic orbitals. The latter are obtained by approximating the expression in Eq. (21) as follows

MIFM_{IF} (meV/Å amu1/2^{1/2})
w/ NACsw/ KS
3E→3A2^3\mathrm{E} \rightarrow^3\mathrm{A}_2 in NV−^-7.2740.387
3E→3A2^3\mathrm{E} \rightarrow^3\mathrm{A}_2 in kkkk-VV0^00.2970.142
1A1→1E^1\mathrm{A}_1 \rightarrow^1\mathrm{E} in NV−^-239.210.5a^a

Table III. Values of the electron-phonon coupling term MIFM_{IF} computed in the accepting-mode approximation using NACs compared with values obtained using Kohn-Sham (KS) electronic orbitals.

(EF−EI)⟨ΨI|∂ΨF∂Q⟩≈(εF−εI)⟨φI|∂φF∂Q⟩(26)(E_F-E_I)\left\langle\Psi_I\middle|\frac{\partial\Psi_F}{\partial Q}\right\rangle\approx(\varepsilon_F-\varepsilon_I)\left\langle\varphi_I\middle|\frac{\partial\varphi_F}{\partial Q}\right\rangle \tag*{(26)}

where φI\varphi_I (φF\varphi_F) and εI\varepsilon_I (εF\varepsilon_F) indicate the Kohn-Sham electronic orbitals and energies of the initial (final) state, respectively. The overlap derivatives are computed by linear interpolation of the term ⟨φI(0)∣φF(Q)⟩\langle\varphi_I(0)|\varphi_F(Q)\rangle with respect to QQ. To approximate the electronic excited states with Kohn-Sham orbitals, we adopted the standard C3v_{3v} point group minimal model for the defect’s orbitals[9].

Supplemental Material for: “First-principles calculations of internal conversion processes in spin defects”

Stefano Paolo Villani; Yu Jin; Giulia Galli

supplementary.pdf

Supplemental Material for: ``First-principles calculations of internal conversion processes in spin defects''

Comparison of non-radiative rates obtained with different DFT functionals

Here, we analyze the effects of the DFT functional on the NRTRs by comparing results obtained with the PBE and DDH functionals. We start with the 1A1→1E{}^{1}A_{1} \to{}^{1}E transition of the NV−\mathrm{NV}^{-} in diamond. In Figure S1 we compare the NRTRs obtained with the two functionals and observe that the computed values differ by a factor of ∼2\sim2. By analyzing the e-ph spectral densities and the line-shape functions, displayed, respectively, in Figure S1 and Figure S1, we conclude that this discrepancy mainly arises from the large difference between the adiabatic energy differences computed with the two functionals (ΔIF=0.871\Delta_{IF} = 0.871 eV with PBE and ΔIF=1.397\Delta_{IF} = 1.397 eV with DDH). If we use the experimental ZPL value 1.195 eV instead of the PBE or DDH adiabatic energy differences, the ratio between the values of the NRTRs is reduced to ∼1.5\sim1.5.

Non-radiative rates, e-ph coupling spectral density functions, and line-shape functions

Figure S1. (a): Non-radiative rates of the 1A1→1E^{1}\mathrm{A}_{1} \to{}^{1}\mathrm{E} transition of the NV−\mathrm{NV}^{-} in diamond for temperatures in the range T=0∼400T = 0 \sim400 K obtained using the PBE and DDH functionals. (b): Comparison between the e-ph coupling spectral density functions. (c): Comparison between the line-shape functions. The blue dashed line and the red dotted-dashed line indicate the adiabatic energy differences computed with TDDFT and the PBE and DDH functionals, respectively. The black dotted line indicates the experimental value of the ZPL.

Next, we analyze the 3E→3A2{}^{3}\mathrm{E} \to{}^{3}\mathrm{A}_2 transition of the NV−\mathrm{NV}^{-} in diamond. Similar to the 1A1→1E{}^{1}\mathrm{A}_1 \to{}^{1}\mathrm{E} transition, the values of the NRTRs computed with the two functionals differ by a factor ∼1.5\sim1.5, as can be observed in Figure S2. However, the discrepancies now arise from sizable differences in both the e-ph coupling spectral functions and the adiabatic energy differences (ΔIF=1.894\Delta_{IF} = 1.894 eV with PBE and ΔIF=2.112\Delta_{IF} = 2.112 eV with DDH), as can be observed in Figure S2 and Figure S2, respectively.

Non-radiative rates, e-ph coupling spectral density functions, and line-shape functions

Figure S2. (a): Non-radiative rates of the 3E→3A2{}^{3}\mathrm{E} \to{}^{3}\mathrm{A}_{2} transition of the NV−\mathrm{NV}^{-} in diamond for temperatures in the range T=0∼400T = 0 \sim400 K obtained using the PBE and DDH functionals. (b): Comparison between the e-ph coupling spectral density functions. (c): Comparison between the line-shape functions. The blue dashed line and the red dotted-dashed line indicate the adiabatic energy differences computed with TDDFT and the PBE and DDH functionals, respectively. The black dotted line indicates the experimental value of the ZPL.

Next, we consider the 1A1→1E{}^{1}\mathrm{A}_1 \to{}^{1}\mathrm{E} transition of the kkkk-VV0\mathrm{VV}^{0} in 4H-SiC. We computed the NRTRs with PBE and DDH using the corresponding adiabatic energy differences ΔIF=0.633\Delta_{IF} = 0.633 eV and ΔIF=1.132\Delta_{IF} = 1.132 eV obtained from TDDFT[1], and present our results in Figure S3. We observe noteworthy differences between the results obtained with the two functionals; differences are also found in the e-ph spectral densities and the line-shape functions, shown in Figure S3 and Figure S3, respectively. We attribute this difference to the different adiabatic energies predicted by the two functionals.

Three-panel comparison of non-radiative rates, e-ph coupling spectral density functions, and line-shape functions

Figure S3. (a): Non-radiative rates of the 1A1→1E{}^{1}A_{1} \to{}^{1}E transition of the kkkk-VV0VV^{0} in 4H-SiC for temperatures in the range T=0∼300T = 0 \sim300 K obtained using the PBE and DDH functionals. (b): Comparison between the e-ph coupling spectral density functions. (c): Comparison between the line-shape functions. The blue dashed line and the red dotted-dashed line indicate the adiabatic energy differences computed with TDDFT and the PBE and DDH functionals, respectively. To date, there are no experimental estimates of the ZPL.

Finally, we present results for the 3E→3A2{}^{3}\mathrm{E} \to{}^{3}\mathrm{A}_2 transition of the kkkk-VV0\mathrm{VV}^{0} in 4H-SiC, where we find large differences in the NRTRs computed with the two functionals, as displayed in Figure S4. The strong dependence on the functional is also evident in the e-ph spectral densities, in Figure S4, and in the line-shape functions, in Figure S4. The difference between ΔIF\Delta_{IF} values (0.971 eV and 1.371 eV with PBE and DDH, respectively) is not the only factor responsible for the differences in results found with the two functionals. The two e-ph spectral densities D(E)D(E) show quantitative differences not only in the relative heights of the peaks, but also in the magnitude of the coupling strength. A detailed investigation reveals that this discrepancy arises from the NACs, whose values differ substantially between the two functionals, as shown in Figure S5. We find that at the PBE level, inaccuracies in the description of the 3E{}^{3}\mathrm{E} state and its relaxed geometry lead to a NAC remarkably different from that obtained with DDH. In Figure S6 we compare the defect’s single-particle levels computed using PBE and DDH at the respective relaxed geometrical configurations. We find that PBE predicts the energy of the aˉ1\bar{a}_1 defect orbital to be below the valence-band maximum, in contrast with the expected behavior captured by the DDH functional, which predicts the orbital energy to be in the band gap of the host. Furthermore, we computed the orbital localization factors using the WEST code and found that the aˉ1\bar{a}_1 orbital obtained using PBE and the PBE relaxed geometry has highly delocalized character, in contrast with the expected behavior, which is found in the localized character of the orbital obtained with

Three-panel comparison of non-radiative rates, electron–phonon coupling strength, and line-shape functions

Figure S4. (a): Non-radiative rates of the 3E→3A2^{3}\mathrm{E} \rightarrow{}^{3}\mathrm{A}_{2} transition of the kkkk-VV0\mathrm{VV}^{0} in 4H-SiC for temperatures in the range T=0∼400T = 0 \sim400 K obtained using the PBE and DDH functionals. (b): Comparison between the e-ph coupling spectral density functions. (c): Comparison between the line-shape functions. The blue dashed line and the red dotted-dashed line indicate the adiabatic energy differences computed with TDDFT and the PBE and DDH functionals, respectively. The black dotted line indicates the experimental value of the ZPL.

Absolute values of the NAC vectors between the 3E and 3A2 states

Figure S5. Absolute values of the NAC vectors dIF,Ad_{\mathrm{IF,A}} between the 3E{}^{3}\mathrm{E} and 3A2{}^{3}\mathrm{A}_{2} states of the kkkk-VV0^{0} center in 4H-SiC computed for each of the 398 atoms in the simulation cell using the DDH and PBE functionals on the respective excited-state relaxed geometries.

Position of the single-particle defect levels for the $kk$-$\mathrm{VV}^{0}$ center in 4H-SiC

Figure S6. Position of the single-particle defect levels for the kkkk-VV0\mathrm{VV}^{0} center in 4H-SiC computed using the PBE and the DDH functionals and the relaxed atomic geometries obtained, respectively, minimizing the PBE and DDH TDDFT total energy of the 3E^{3}\mathrm{E} state.

DDH at the DDH relaxed geometry. We argue that this incorrect description of the 3E^{3}\mathrm{E} state character and of the relaxed geometry stems from known inaccuracies of the PBE functional in describing charge-transfer processes in spin-defective systems.

Computational details and workflow description

All the DFT calculations were carried out with the Quantum ESPRESSO code[2]. For both systems, we used a kinetic energy cutoff of 60 Ry for the wavefunctions and optimized norm-conserving Vanderbilt (ONCV) pseudo-potentials[3]. All TDDFT calculations[1] were performed with the WEST code, which implements analytical formulations of excited-state forces[1] and NACs[4]. We generated the dynamical matrices using the phonopy code[5]. To reduce the computational cost required to obtain excited-state phonons with finite displacements in large supercells, we computed all the dynamical matrices at the PBE level of theory, following a practice justified in previous works[6, 7]. We assumed no mode mixing between the vibrational modes of the initial and final electronic states and we employed the displaced harmonic oscillator approximation throughout. Specifically, we considered the excited-state phonons of the initial excited state for all the transitions analyzed except for the 3E→3A2^{3}\mathrm{E} \rightarrow{}^{3}\mathrm{A}_{2} transition in SiC, where instead we used the ground-state phonons. We note that the relaxed atomic configuration of the 3E^{3}\mathrm{E} state in kkkk-VV0\mathrm{VV}^{0} in 4H-SiC possesses a broken-symmetry geometry with C1hC_{1h} point group and computing the dynamical matrix of the excited state in the 398-atom supercell requires thousands of TDDFT-forces calculations, a prohibitively expensive task even with the PBE functional. Line-shape functions were computed using the PyPL code[8]. For each transition, we used values of the Lorentzian smearing factor γ\gamma appearing in the generating function G(E)G(E) estimated from low-temperature photoluminescence linewidth measurements. For the 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition in NV−\mathrm{NV}^{-} we used γ=0.18\gamma= 0.18 meV[9]; for the 3E→3A2^{3}\mathrm{E} \rightarrow{}^{3}\mathrm{A}_{2} transition in NV−\mathrm{NV}^{-} we used γ=2.07×10−4\gamma= 2.07 \times10^{-4} meV[10]; for the 3E→3A2^{3}\mathrm{E} \rightarrow{}^{3}\mathrm{A}_{2} transition in kkkk-VV0\mathrm{VV}^{0} in 4H-SiC we used γ=4.13×10−4\gamma= 4.13 \times10^{-4} meV[11]; for the 1A1→1E^{1}\mathrm{A}_{1} \rightarrow{}^{1}\mathrm{E} transition in kkkk-VV0\mathrm{VV}^{0} in 4H-SiC, where no experimental data are available, we used the same γ=0.18\gamma= 0.18 meV used for the NV−\mathrm{NV}^{-}.

References

References

  1. [1]Y. Jin, V. W.-z. Yu, M. Govoni, A. C. Xu, and G. Galli, Journal of Chemical Theory and Computation 19, 8689 (2023).
  2. [2]P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, M. Cococcioni, et al., Journal of physics: Condensed matter 29, 465901 (2017).
  3. [3]M. Schlipf and F. Gygi, Computer Physics Communications 196, 36 (2015).
  4. [4]S. P. Villani, Y. Jin, and G. Galli, in preparation.
  5. [5]A. Togo and I. Tanaka, Scripta materialia 108, 1 (2015).
  6. [6]Y. Jin, M. Govoni, G. Wolfowicz, S. E. Sullivan, F. J. Hermans, D. D. Awschalom, and G. Galli, Physical Review Materials 5, 084603 (2021).
  7. [7]A. Kundu and G. Galli, The Journal of Physical Chemistry Letters 15, 802 (2024).
  8. [8]https://github.com/jinyuchem/pypl.
  9. [9]T. Biktagirov, A. Smirnov, V. Y. Davydov, M. Doherty, A. Alkauskas, B. C. Gibson, and V. Soltamov, Physical Review B 96, 075205 (2017).
  10. [10]K.-M. C. Fu, C. Santori, P. E. Barclay, L. J. Rogers, N. B. Manson, and R. G. Beausoleil, Physical Review Letters 103, 256404 (2009).
  11. [11]D. J. Christle, P. V. Klimov, C. F. de las Casas, K. Szász, V. Ivády, V. Jakubavicius, J. Ul Hassan, M. Syväjärvi, W. F. Koehl, T. Ohshima, et al., Physical Review X 7, 021046 (2017).

Paper details

vlnc:W_ke8657k9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents