WorksPhysics and Astronomy
Analytic Model for Transient Plateau Filling in Vibrational Distributions of Periodically Driven Plasmas
Abstract
In periodically driven molecular plasmas, the vibrational distribution function is governed by the competition between vibration–vibration (V-V) up-pumping along the anharmonic ladder and vibration–translation (V-T) relaxation, producing the Treanor–Gordiets distribution. When vibrational energy accumulates across cycles faster than V-V exchange can populate the upper levels, the Gordiets plateau remains transiently depleted. Existing algebraic models assume a quasi-steady plateau, whereas exact transient solutions of the nonlinear V-V diffusion equation require the elapsed time since excitation and the source strength, which are unavailable in multidimensional simulations that track only local macroscopic variables. Here, elapsed time is replaced by a local vibrational energy timescale formed from the vibrational energy per molecule and its period-averaged rate of change. This reduces the diffusion equation in level space to an ordinary differential equation solved in closed form without linearization, yielding a level-dependent correction to the steady plateau governed by a single dimensionless group: the number of V-V collisions per energy timescale weighted by the plateau amplitude at the Treanor minimum. The distribution front advances as the square root of this group, and the steady distribution is recovered as the group grows large. Compared with state-to-state master-equation calculations for N under nanosecond pulse trains spanning 300-1500 K, 0.041-1 atm, and 1-100 kHz (1440 sampled states), the model reduces the 90th-percentile error in level populations from 1.88 decades for the steady model and 1.32 decades for an adapted self-similar solution to 0.53 decades, at essentially the evaluation cost of the steady model.
Introduction
ON-EQUILIBRIUM vibrational distributions in molecular gases arise in periodically driven plasmas where vibrational energy accumulates from one cycle to the next, including plasma-assisted combustion, [1, 2, 3, 4, 5] laser-induced plasmas for flame anchoring, [6, 7] sinusoidal discharges in semiconductor processing, [8, 9, 10] and atmospheric-pressure plasma sources. [11, 12, 13, 14] In all of these settings, predicting energy coupling and chemical reaction rates requires the molecular vibrational distribution function (VDF). Although the excitation mechanism varies across applications, whether via electron impact or laser absorption, the subsequent redistribution of vibrational quanta is governed by the same two competing heavy-particle processes: vibration–vibration (V–V) exchange pumps quanta upward along the anharmonic vibrational ladder, while vibration–translation (V–T) relaxation drains them into translational energy. The competition between these processes produces the characteristic two-branch Treanor–Gordiets distribution, [15, 16] in which V–V exchange sustains a Treanor distribution at low levels and V–T relaxation terminates a plateau extending across the upper levels.
Because the Treanor–Gordiets distribution governs both heavy-particle dissociation rates and electron kinetics—such as a superelastic electron–vibrational heating, for which a unified analytic closure was recently derived[17]—a substantial body of work has sought closed-form VDF descriptions that can be evaluated directly within multidimensional flow and plasma simulations. The most complete treatment, the state-to-state (StS) method, integrates a coupled rate equation for every vibrational level to capture the full VDF across all regimes. However, its computational cost scales as per grid point, where is the number of tracked vibrational states, making
it prohibitive in multidimensional simulations with or more grid points. The analytic model of Adamovich, Macheret, and Rich[18] (AMR) overcomes this barrier by extending the foundational theory of Gordiets et al.[16] into a fully algebraic framework. Rather than integrating differential rate equations, AMR reconstructs both the Treanor branch and the Gordiets plateau directly from three local macroscopic quantities: gas temperature, vibrational temperature, and molecular number density. This combination of physical fidelity and negligible evaluation cost makes the AMR model well suited for coupled simulations, and it serves here as the steady-state foundation of the present work.
The AMR model, however, assumes that the VDF has reached a quasi-steady state. In the transient filling regime of a periodically driven plasma, this assumption breaks down: vibrational energy accumulates across cycles faster than V–V exchange can relay quanta to the upper levels, leaving the Gordiets plateau depleted relative to its steady-state profile when the next pulse arrives. Prior work has approached this transient regime through either numerical continuum formulations or analytical solutions. Continuum Fokker–Planck formulations, such as the modified approach of Altin et al.,[19] replace the discrete master equation with a partial differential equation (PDE) in level space, but still require numerical integration and do not meaningfully reduce the computational cost relative to StS. Conversely, the analytical theory of Zhdanov, Napartovich, and Starostin[20] (ZNS) and the related analyses of Gordiets and Zhdanok[21] yield closed-form transient solutions of the nonlinear V–V diffusion equation, with ZNS giving self-similar solutions for a source switched on or off and for a single excitation pulse. Yet each of these analytical solutions requires the elapsed time since the excitation event as an explicit input, together with either the source power or the deposited pulse energy. Neither quantity is available in a multidimensional flow or plasma simulation, which transports only the local macroscopic state at each grid point (such as gas temperature, density, and vibrational energy) and retains no record of the excitation history. Thus, to our knowledge, no analytic model currently exists for the transient filling regime that can be evaluated solely from local macroscopic variables.
This work derives such a model and validates it against StS calculations of N in repetitively pulsed nanosecond discharges. Building on the steady AMR foundation, the present model reconstructs the full transient VDF algebraically from four local macroscopic inputs—gas temperature, molecular number density, vibrational energy per molecule , and its period-averaged rate of change —without ordinary differential equation (ODE) integration. Specifically, elapsed time in the ZNS nonlinear diffusion equation is replaced by a local vibrational energy timescale . This substitution reduces the governing PDE in level space to an ODE that is solved in closed form while retaining the full nonlinearity of V–V transport. The resulting solution shows that transient plateau filling is governed by a single dimensionless group and that the distribution front advances as its square root, recovering the steady AMR plateau as this group grows large. Because is a period-averaged rate available from the vibrational energy equation of a flow or plasma solver (in a gas at rest, simply by storing from the previous cycle) and is defined identically regardless of the excitation mechanism, the model is expected to apply broadly to periodically driven plasmas in which vibrational energy builds across cycles, although this remains to be tested beyond the N cases considered here.
Model Derivation and Implementation
The present model is developed in three steps: the steady-state vibrational distribution is established first (Sec. II.A), then extended to the transient filling regime (Sec. II.B), and finally assembled into an algorithm that computes the distribution from local macroscopic variables (Sec. II.C).
Steady-State Vibrational Distribution
We begin with the steady-state vibrational distribution model of AMR, [18] which divides the vibrational manifold into two physically distinct regions separated by the Treanor minimum . Below , near-resonant V–V exchange dominates and drives the populations toward the Treanor distribution;[15] above , the competition between V–V up-pumping and V–T relaxation produces the Gordiets plateau.[16, 21, 22]
There, the populations follow the Treanor distribution[15, 17]
where is the ground-state number density, is the number density of vibrational level , is the Boltzmann constant, is the translational temperature, and is the vibrational temperature defined by Treanor[15] as , with denoting the characteristic vibrational temperature of the fundamental transition.[17] The vibrational energy levels are given by the second-order Dunham expansion[17]
where is the Planck constant, is the speed of light, is the harmonic vibrational frequency, and and are the dimensionless first- and second-order anharmonicity constants obtained from the spectroscopic constants and in .[23] Evaluating Eq. (2) at and relates the characteristic vibrational temperature to the spectroscopic constants via . Equation (2) also defines the first unbound vibrational level (the dissociation wall) as the smallest integer satisfying , where is the dissociation energy; for N, eV[23] gives , corresponding to 48 bound vibrational levels ( to 47).
The Treanor minimum is found by setting the derivative of the Treanor exponent to zero using the second-order Dunham expansion and selecting the physical root as outlined by Parent.[17]
For , as for N, is always real and positive. For , the square root can become imaginary (e.g., for O when exceeds about 0.63), in which case the linear Treanor minimum, , obtained from Eq. (3) as ,[17] is used instead. When exceeds a threshold, surpasses and the entire bound manifold follows the Treanor distribution. In practice, is not rounded, and the Treanor branch is evaluated at the integer levels .
Above , the steady distribution follows from Adamovich et al.,[18] with the radiative term absent for a homonuclear molecule and the long-range V–V correction neglected (the latter arises from multipole interactions and is significant only for polar species):
where is the total number density of the molecule. The first term is the classical Gordiets plateau sustained by V–V exchange and the second is the V–T relaxation drain. Because the V–T term grows exponentially with and eventually exceeds the plateau term, all levels above this cutoff point are set to zero.[18] The plateau amplitude is set by matching to the Treanor branch at .[18]
where is evaluated from Eq. (1). A small discontinuity between the Treanor and plateau branches may appear at because the two expressions are matched in amplitude through but not in slope. Because Eq. (1) gives the populations relative to whereas Eqs. (4)–(5) are normalized by , the normalization constant couples the two branches and is obtained by summing the distribution over all bound levels and enforcing conservation of molecule number. Following Macheret,[24] the wall factor accounts for the loss of molecules pumped past the dissociation boundary by V–V exchange,[18] assuming that molecules reaching the boundary are removed faster than resonant V–V exchange can supply them (Eq. 21 of Ref. 18):
The V–T coefficient follows from Eq. (16) of Adamovich et al.[18] for a pure diatomic gas without the long-range V–V correction (i.e., with both the diatomic mole fraction and the long-range correction factor set to unity):
where and are rate-defect parameters defined below.
Evaluating Eq. (7) requires the fundamental rate coefficients and . A V–V collision involves two molecules exchanging vibrational quanta with rate coefficient , where , and , are the initial and final levels of the colliding pair, whereas a V–T collision transfers vibrational quanta from a single molecule to translational motion with rate coefficient . Adamovich et al. treated both fundamental rates as empirical inputs inferred from CO experiments and provided no closed-form expressions. Here, is evaluated from the forced harmonic oscillator–free rotation (FHO–FR) theory of Adamovich, [25] the most complete first-principles V–V rate model available. The accuracy of the present model is therefore tied to the domain of validity of FHO–FR, which has been validated across a range of diatomic species. [25] The fundamental V–T rate is obtained from the Millikan–White correlation, [26] which is widely used and requires only the reduced mass and characteristic vibrational temperature of the collision pair, allowing it to be applied to any diatomic species without species-specific fitting.
For the fundamental resonant exchange, Adamovich gives the thermally averaged FHO–FR rate coefficient in closed form (Eq. 23 of Ref. 25 with , where is the number of quanta exchanged):
where is the FHO maximum interaction distance, with the intermolecular potential steepness and the repulsion coefficient (4 Å and 1730 eV for N); is the mean relative collision speed for reduced mass ; and is the fundamental vibrational transition frequency, with the reduced Planck constant.
Next, for , Landau–Teller theory[27, 28] relates the relaxation time to the fundamental rate coefficient via . Solving for and substituting , where is the pressure, yields
in which cancels between and , leaving a function of alone. The Millikan–White correlation[26] gives in atm·s as
which is multiplied by 101 325 Pa/atm to convert to Pa · s for use in Eq. (9). Here, (Eq. 3a of Ref. 26), where the characteristic oscillator temperature is taken, as in Millikan and White, as the harmonic value (3395 K for N in their Table I), and is the reduced mass of the collision pair in atomic mass units (14 amu for N–N).
It remains to define the rate-defect parameters and in Eqs. (4) and (7). The parameter enters the AMR V–V rate model,[18] which gives the single-quantum exchange rate between a molecule at level and a partner at level as
When and are far apart, the energy mismatch of the non-resonant exchange must be absorbed by translation, suppressing the rate exponentially with increasing . Following Gordiets and Zhdanok,[21] we use
with in Å, in amu, and and in K.
Likewise, the parameter governs the exponential growth of the V–T rate with vibrational level, {sup}18 $k_{\mathrm{VT}}^{n\to n-1}=n\cdot k_{\mathrm{VT}}^{1\to0}\cdot e^{\delta_{\mathrm{VT}}(n-1)}$, where the prefactor $n$ is the harmonic level scaling.{citep}27 Because Adamovich et al. give no prescribed formula for $\delta_{\mathrm{VT}}$, it is derived here from the Landau–Teller law,{citep}27 $k_{\mathrm{VT}}^{1\to0}\propto\exp[-(T_{\mathrm{LT}}/T_{\mathrm{g}})^{1/3}]$, where the Landau–Teller temperature scales with the square of the transition frequency as $T_{\mathrm{LT}}=2\pi^2c\omega_{\mathrm{e}}^2\mu/(\alpha^2k_{\mathrm{B}})\propto\omega_{\mathrm{e}}^2$. Anharmonicity reduces the transition frequency in the $n\to n-1$ transition to $\omega_n=\omega_{\mathrm{e}}(1-2x_{\mathrm{e}}n)$,{citep}17` yielding a level-dependent . Substituting and expanding the exponent to first order in gives
The first term on the right-hand side is already contained in , while the second multiplies the rate by . Combining this factor with the harmonic scaling[27] gives
from which
where has been expressed through the Millikan–White correlation as . For N, Eq. (15) gives at 300 K and 0.16 at 1500 K, which is of the same order as } ( and 0.21) and consistent with the approximation adopted by ZNS for a pure diatomic gas;[20] the two parameters nonetheless differ in their temperature scaling ( versus ), and the Landau–Teller form is retained here. Equations (1)–(15) constitute the complete quasi-steady adjusted AMR model, which we now extend to the transient filling regime.
Transient Correction
In the steady plateau of Eq. (4), V–V up-pumping supplies quanta across all levels at a settled rate while V–T relaxation locally drains them and dissociation removes molecules at . During the transient filling regime of a repetitively pulsed discharge, however, V–V exchange must relay quanta upward from level by level, so the upper plateau lags behind its quasi-steady form. Because V–T relaxation acts locally at each level—depending only on the instantaneous population and , not on the time history of the distribution—its drain term is left in its steady form. Likewise, the dissociation wall factor represents removal of molecules at the dissociation boundary, which acts only once the advancing population front reaches . We therefore account for transient plateau filling by introducing a dimensionless unsteady correction factor into the V–V plateau term of Eq. (4):
where is derived from the pure V–V up-pumping problem without the V–T drain or the wall factor (). Combining and multiplicatively in Eq. (16) recovers the steady AMR distribution of Eq. (4) exactly as and is otherwise an approximation, assessed against SS in Section IV.C.
To derive , we note that the steady V–V plateau population varies with level, which complicates the nonlinear algebra of the time-dependent problem. Introducing the scaled variable
simplifies this variation: on the steady V–V plateau, uniformly for every , while an empty level has . The correction factor therefore equals unity on a fully populated level and zero on an empty one. Treating as continuous and retaining only V–V exchange, the time evolution of along the plateau is governed by the nonlinear diffusion equation of ZNS,[20]
which behaves similarly to the nonlinear heat equation but in vibrational level space, diffusing upward along the vibrational ladder at a rate set by the collision frequency .[21]
The diffusion coefficient in Eq. (18) is , with , a dimensionless factor that follows from the V–V rate model of Eq. (11). It matches the coefficient in Eq. 11 of Ref. 18 apart from that equation’s factor ( being Euler’s number). That factor is not part of the V–V rate model: it enters AMR’s expression for the flux of quanta in terms of the population at the Treanor minimum, which is not used here. Expanding Eq. (11) in the diffusion approximation gives a coefficient of about 5.75, consistent with AMR’s 6 without the . ZNS obtain a coefficient of 4 instead of 6 because they assume a purely exponential V–V rate, without the bracketed factor of Eq. (11).[20]
Although ZNS obtained exact solutions of (18), those solutions describe a steady source switched on or off at , or a single excitation pulse, and depend on the elapsed time since that event and on the source power or the energy deposited by the pulse. In a repetitively pulsed discharge, neither the elapsed time nor a single source power is defined, and because (18) is nonlinear, the single-pulse solution cannot be superposed over a train of pulses. Moreover, a flow or plasma solver carries no history of the vibrational distribution itself; it transports , and the rate required here is the period-averaged material rate of change experienced by a gas element. In a gas at rest, this rate is computed from stored at the ends of successive interpulse intervals, at the end of the interpulse of cycle , where is the pulse repetition frequency; its evaluation in a convective flow is described in Sec. II.C.
To express the transient evolution in terms of these local variables, time is eliminated from (18) by replacing the time derivative with , where
is the vibrational energy timescale, i.e., the time it would take for to reach its current value if it had always been growing at the current period-averaged rate . This is justified because every pulse deposits quanta into the low vibrational levels and V–V carries them up the ladder one collision at a time, so the energy grows with each cycle and serves as a record of how long the system has been running. When grows at a steady rate from zero, equals the time elapsed since the first pulse, and as the distribution approaches steady state, V–T losses slow the growth of so that , recovering the steady AMR result. Replacing with is an approximation, as V–V fills the ladder from the bottom upward and different levels fill at different rates. The level-by-level detail is nonetheless retained by the diffusion term on the right of (18), which describes how V–V carries quanta up the ladder, while only the time derivative on the left is replaced.
Since , the same approximation gives . Substituting into (18) and dividing by yields
Recalling that , the product is naturally written as , where
counts the number of fundamental V–V collisions per energy timescale . Substituting into (20) gives the ODE
in which time no longer appears.
(22) is second-order and involves an unknown upper boundary. In a pulsed discharge, V–V exchange pushes quanta upward by level, requiring many collisions and therefore finite time; at any instant, a front level marks the highest level V–V has reached, above which the distribution is empty. The front is part of the solution, introducing a third unknown alongside the two integration constants and requiring three boundary conditions. At the solution matches the Treanor branch, ; at the front the population vanishes, ; and the flux of quanta also vanishes at since no V–V exchange can occur with the unpopulated levels above , meaning . The front is treated as a continuous variable and need not be an integer.
To reduce Eq. (22) from second- to first-order, both sides are multiplied by . The left side becomes by the chain rule. On the right, we use the identity . Integrating from to with as the dummy variable, the left evaluates by the fundamental theorem; applying the zero-flux condition at , the upper limit vanishes, leaving . On the right, the weight is largest at the lower limit and decreases monotonically to zero at the front, where vanishes as , as the resulting solution confirms. The integrand is therefore concentrated near , and is evaluated there and pulled out.[21] Applying then gives . Assembling and multiplying by gives the first-order equation
Since decreases with , we have , so the negative square root is taken. Substituting and dividing by and then by separates the variables:
Integrating both sides from to , the left evaluates as and the right as , giving
With , dividing by 2, solving for , dividing by , and squaring gives
Equation (26) shows that the entire unsteady correction is controlled by the single group . In the limit (when ), the subtracted term is negligible for all plateau levels and uniformly, so Eq. (16) recovers the steady AMR distribution of Eq. (4). For finite , the correction decreases with , meaning high levels are always the most depleted since V–V propagates quanta upward from the bottom of the ladder and the upper levels are the last to fill; when is small, the subtracted term reaches order unity within a few levels of and the plateau collapses from the top down.
Setting gives the level of the distribution front,
with capped at . The fraction of the plateau that has filled follows directly from Eq. (27): the front advances by , and the plateau is completely filled when this advance reaches , the distance from to the dissociation level. Defining the fill fraction as the ratio of the actual front advance to this filled value,
where is an empty plateau and is fully filled. Substituting Eq. (27) into Eq. (28) and capping at unity gives
Setting shows that the plateau reaches its filled, quasi-steady regime when
and remains in the transient filling phase for smaller values.
Model Implementation
In a flow or plasma solver, the present model reconstructs the local vibrational distribution algebraically from four macroscopic inputs: the translational gas temperature , the total molecular number density , the vibrational energy per molecule measured from the ground level,
and its period-averaged rate of change . This rate is the period-averaged material (Lagrangian) rate of change of . In a gas at rest it equals the cycle-to-cycle difference of § 2.B, but in a convective flow at a periodic steady state that difference vanishes at a fixed grid point even though gas elements passing through the discharge are still accumulating vibrational energy. There, is instead taken as the period-averaged net source per molecule in the solver’s transport equation (excitation minus V–T and other losses), which equals when diffusion is negligible.
Before solving for the level populations, all quantities that do not depend on are evaluated once for the given state. Specifically, the fundamental rate coefficients and are computed from Eqs. (8)–(10), the rate defect parameters and from Eqs. (12) and (15), and the V–T drain coefficient from Eq. (7), along with the diffusion prefactor . When the vibrational energy is increasing (), the energy timescale and collision parameter are computed from Eqs. (19) and (21). Conversely, when , the gas is outside the transient filling regime; the unsteady correction is bypassed by setting across all plateau levels, which recovers the steady AMR distribution of Eq. (4).
With these state parameters fixed, assembling the distribution requires determining two remaining unknowns: the vibrational temperature and the ground-state number density . These are obtained through a nested bisection procedure in which an outer loop updates to match the input vibrational energy , and an inner loop updates to conserve the total molecular number density . In the outer loop, each trial value of fixes the Treanor minimum via Eq. (3) and determines the normalized Treanor population ratios from Eq. (1) for all levels . If , the Treanor branch covers every bound level and no plateau is built. Otherwise, because the plateau parameters , , and depend on the absolute population at while the V–T drain term in Eq. (16) is independent of , the ground-state density cannot be factored out as a simple linear normalization constant and must instead be found iteratively in the inner bisection loop. For each trial , the plateau amplitude at is evaluated from Eqs. (1) and (5), which in turn sets the active distribution front via Eq. (27), capped at . Across the active plateau levels , the dissociation wall factor and the unsteady correction are then computed from Eqs. (6) and (26). The complete distribution is assembled by combining the Treanor branch of Eq. (1) for with the unsteady plateau of Eq. (16) for , while setting above and at every level from the first one at which the exponential V–T drain exceeds the first term of Eq. (16). The inner bisection loop adjusts until the sum of over all bound levels () equals . Once has converged for the given trial , the corresponding vibrational energy is evaluated from Eq. (31), and the outer bisection loop adjusts until this calculated energy matches the input .
Validation and Baseline Models
We assess the present model by comparing its predicted vibrational distributions against state-to-state calculations of N under repetitive nanosecond pulsed excitation. This section outlines the StS master equation and test matrix (Sec. III.A), followed by the baseline models and the evaluation protocol used in the comparisons (Sec. III.B), and the use of artificial intelligence (AI) tools (Sec. III.C).
State-to-State Reference and Test Matrix
The reference solutions are obtained by integrating the StS master equation for the 48 bound vibrational levels of N (–47) and the N-atom density at constant gas temperature and total number density . The heavy-particle kinetic mechanism includes multi-quantum V–T relaxation and V–V exchange with up to three quanta transferred per collision partner (), together with V–V dissociation from the three bound levels (–47) and reverse three-body recombination obtained by detailed balance. All V–V and V–T rate coefficients are taken from the FHO–FR theory of Adamovich et al. [25]. The rates are tabulated at 1 K intervals from 300 to 3000 K and evaluated at the nearest tabulated ; because each V–V channel is tabulated in both directions, the master equation sums over the table with a factor 1/2. Electron-impact vibrational excitation and superelastic de-excitation processes with are modeled using the Maxwellian rate fits of Laporta et al. [29] at an electron temperature of eV, with de-excitation rates determined by detailed balance. Each simulation models a train of 20 rectangular 100 ns pulses repeated at frequency , initialized from a Treanor distribution (Eq. (1)) at K and the prescribed . To ensure a controlled energy input per cycle, each 100 ns pulse deposits a prescribed net vibrational energy per molecule; the constant electron density during each pulse is adjusted by bisection so that the net electron-impact vibrational power (excitation minus superelastic de-excitation) integrates to over the pulse duration. Electron processes are switched off between pulses, and the vibrational distribution is sampled at the end of each interpulse interval so that all analyzed states reflect purely heavy-particle relaxation. The representative distribution comparisons in Figure 1 and Figure 2 fix atm and eV, whereas the parametric analyses in § 4.B and IV.C span all combinations of three gas temperatures (300, 750, and 1500 K), two pressures (0.041 and 1 atm), four repetition frequencies (1, 10, 50, and 100 kHz), and three pulse energies (0.01, 0.02, and 0.05 eV per molecule), providing 72 operating conditions and 1440 sampled interpulse states across the 20-pulse trains.

Figure 1. Vibrational distributions of N₂ at atm, sampled at the end of the interpulse of cycles 1, 5, 10, and 20, for 100 ns pulses repeated at kHz across gas temperatures , 500, and 750 K ( eV per molecule per pulse). All three analytic models are evaluated from the same StS inputs and ; is the value obtained by the present model.

Figure 2. Vibrational distributions of N at K and atm, sampled at the end of the interpulse of cycles 1, 5, 10, and 20, for 100 ns pulses repeated at , 50, and 100 kHz ( eV per molecule per pulse). All three analytic models are evaluated from the same StS inputs and ; is the value obtained by the present model.
Baseline Models and Adapted ZNS Correction
Three algebraic models are compared against the StS reference: the quasi-steady AMR model, [18] an adapted version of the ZNS self-similar transient theory[20] (denoted “AMR + ZNS”), and the present model. Whereas a coupled flow or plasma solver would obtain and from its vibrational energy transport equation (§ 2.C), here we evaluate the analytic models a posteriori by supplying and the cycle-to-cycle rate directly from the StS solution at the end of each cycle (with taken from the initial distribution). Feeding the exact StS vibrational energy into all three models (steady AMR uses only ) isolates errors in the predicted distribution shape from any cumulative drift in total vibrational energy.
Constructing the AMR + ZNS baseline requires adapting the ZNS transient theory so that it uses the same local macroscopic inputs as the present model. Because Eq. (18) is nonlinear, the ZNS single-pulse solution cannot be superposed across a train of pulses; we instead start from the ZNS switch-on solution for a sustained source turned on at (Eq. 16 of Ref. 20), which represents continuous net vibrational energy addition over multiple cycles. In our notation, this self-similar plateau profile reads
where is the steady plateau amplitude set by the external source power. To express Eq. (32) in terms of local macroscopic variables and embed it within the AMR framework, we map to the Treanor branch at by setting and shifting the linear decay to originate at (replacing with ), and we replace the cumulative collision integral with the local collision count via Eq. (19). Dividing the resulting profile by and flooring at zero gives the adapted ZNS unsteady correction factor,
Replacing with in Eq. (16)—while retaining the same V–T drain and dissociation wall factor —ensures that the present model and the AMR + ZNS baseline differ solely in the functional form of the unsteady plateau correction, with the ZNS front advancing linearly in rather than as the square root of (Eq. (27)).
Use of AI Tools
Claude Opus 5.5 (Anthropic) was used to reduce manual derivation and implementation effort, assisting the authors with parts of the derivations of § 2.A and II.B, parts of the StS solver of Sec. III.A, and the scripts that produce the figures and Table 1. The authors verified each step of the derivations and compared the closed-form solution with a numerical solution of Eq. (22).
| , K | , atm | , kHz | States | Steady AMR | AMR + ZNS | Present model | |
| 300 | all | all | all | 480 | 2.21 | 1.91 | 0.46 |
| 750 | all | all | all | 480 | 1.44 | 1.00 | 0.55 |
| 1500 | all | all | all | 480 | 0.70 | 0.75 | 0.68 |
| all | 0.041 | all | all | 720 | 2.31 | 1.79 | 0.74 |
| all | 1 | all | all | 720 | 0.82 | 0.71 | 0.33 |
| all | all | 1 | all | 360 | 0.35 | 0.45 | 0.33 |
| all | all | 10 | all | 360 | 1.50 | 1.11 | 0.41 |
| all | all | 50 | all | 360 | 2.15 | 1.69 | 0.65 |
| all | all | 100 | all | 360 | 2.31 | 1.87 | 0.82 |
| all | all | all | 821 | 2.29 | 1.67 | 0.72 | |
| all | all | all | 619 | 0.37 | 0.82 | 0.34 | |
| all | all | all | all | 1440 | 1.88 | 1.32 | 0.53 |
Table 1. 90th percentile of the absolute error in vibrational level populations relative to the StS solution, in decades (), pooled over all populated levels and cycles of the states meeting the conditions of each row.
Results and Analysis
We evaluate the model in four steps: first by comparing its predicted vibrational distributions directly against state-to-state calculations at representative conditions (§ 4.A), next by testing how the dimensionless group controls plateau filling (§ 4.B), then by quantifying population errors across the full parametric sweep (§ 4.C), and finally by examining the computational savings for multidimensional solvers (§ 4.D).
Transient Plateau Filling
Figure 1 compares the three algebraic models against the StS reference across gas temperatures of 300, 500, and 750 K at a pulse repetition frequency of 100 kHz. At 300 K, the plateau spans most of the vibrational ladder, and during early cycles vibrational energy is pumped into the low levels much faster than V–V collisions can relay quanta upward. In this transient regime ( small), the three models behave quite differently: steady AMR assumes a fully settled plateau at every cycle and severely overpopulates the upper levels; the adapted ZNS model cuts off prematurely just above the Treanor minimum ; and the present model captures the gradual, level-by-level advance of the distribution through . As additional pulses accumulate—or as the gas temperature rises, which shifts upward and shortens the plateau—the distribution catches up to its quasi-steady profile and all three models converge.
Figure 2 fixes the gas temperature at 300 K and steps the pulse repetition frequency across 10, 50, and 100 kHz. After the first pulse (top row), the plateau has barely begun to form at any frequency because the amplitude at the Treanor minimum is still vanishingly small. Over the next several pulses, V–T losses remain negligible at 300 K, so the accumulated vibrational energy—and hence —depends mainly on how many pulses have fired rather than on the repetition rate. What changes with frequency is the time available between pulses for V–V collisions to do their work: pulsing at 100 kHz leaves ten times less interpulse time for V–V up-pumping than pulsing at 10 kHz, directly reducing . As a result, at 50 and 100 kHz the plateau remains starved of quanta for roughly the first ten pulses, accentuating both the severe overprediction of steady AMR and the premature cutoff of the adapted ZNS model.
This early overpopulation matters most for processes that draw on the top of the ladder, such as V–V dissociation from –47. At 300 K, while the plateau is still filling, StS places at most of the molecules in these levels, whereas steady AMR places – there, an overprediction of 13 to 25 orders of magnitude. The present model, like AMR + ZNS, keeps its front below in these states and leaves the dissociating levels empty. Once the plateau has filled, the present model and steady AMR both give the population of these levels to within about a factor of two of StS.
These contrasting behaviors trace directly back to the nonlinearity of V–V transport in Eq. (18). Because an upward V–V step at level requires two excited molecules to collide, the upward flux of quanta scales with the square of the local population (). Near the advancing front, where the population becomes sparse, the upward flux chokes off and causes the true distribution to taper gradually rather than ending in a sharp cliff.
Steady AMR ignores this finite propagation speed altogether, assuming V–V exchange has already equilibrated the entire plateau after every pulse. The adapted ZNS model accounts for finite propagation, but its self-similar ansatz forces the plateau correction to drop off at a constant slope (Eq. (33)) regardless of how depleted the local population has become. Because that slope never flattens out, the ZNS distribution plunges straight to zero just above , and its front advances only linearly with .
By solving the time-free ODE of Eq. (22) without imposing a self-similar shape, the present model preserves the nonlinear tapering of the V–V flux near the front. As the population thins out at higher levels, the slope of flattens (Eq. (26)), allowing the distribution to feather smoothly into zero rather than cutting off abruptly. Just as importantly, the front advances as the square root of rather than linearly (Eq. (27)). During the first few pulses—when is much smaller than unity—its square root is much larger than the parameter itself, placing the front many levels higher up the ladder than the linear ZNS prediction and capturing the gradual filling seen in the state-to-state curves.
3a as the Controlling Parameter
Because the distribution front advances as (Eq. (27)), whether the plateau has had time to fill depends on the combined product rather than on the collision count alone. Weighting by the plateau amplitude reflects the nonlinear nature of V–V up-pumping: a denser base at supplies more collision partners and drives the front upward much faster for the same number of fundamental collision times.
Figure 3 tests this scaling across all plateau-bearing states in our parametric sweep, plotting the fraction of the plateau that has filled () against the analytic prediction of Eq. (29). In panel (a), where is plotted against directly, the state-to-state data span several orders of magnitude on the horizontal axis yet organize along a single rising trend. This confirms that is an essential part of the controlling group: two conditions at different temperatures may experience a similar number of V–V collisions , but the cooler condition has a lower Treanor minimum and an orders-of-magnitude larger plateau amplitude , driving its front upward much faster.
Because the plateau spans from to , the threshold value of required to reach the dissociation wall—given by in Eq. (30)—depends on . A higher gas temperature pushes upward, leaving a shorter plateau that takes fewer V–V collisions to fill. Consequently, panel (a) exhibits a family of analytic curves that saturate at smaller as increases. Dividing by this filling threshold in panel (b) removes the dependence on plateau length, collapsing the analytic curves onto a single universal profile where full filling occurs at unity on the horizontal axis.
While the rescaled StS data follow the shape of this universal curve in panel (b), two departures stand out. First, during the filling phase (), the StS points sit systematically to the right of the analytic curve, meaning that the model’s front runs slightly ahead of the StS solution—as was visible at intermediate cycles in Figure 1(c) and Figure 2(k). This offset arises because the model assumes the Treanor branch below is already fully equilibrated at the instantaneous . In reality, V–V exchange must also build the upper Treanor branch from the bottom up, so the population right at lags behind the equilibrium Treanor curve during rapid excitation. Reading from the equilibrium Treanor formula therefore overstates the true population at the base of the plateau and pushes the predicted front ahead of the StS reference. Because a higher gas temperature shifts further up the ladder onto levels that take even longer to populate, this lag—and the resulting rightward shift in panel (b)—grows with temperature. Eliminating this offset would require extending the unsteady treatment to the Treanor branch itself, which is beyond the scope of the present work. Second, the small cluster of points lying above the curve at the far left of Figure 3 corresponds to initial pulses where is nearly zero; there, local V–V collisions between the upper Treanor levels and the base of the plateau nudge the StS front slightly past , a cross-branch interaction not included in the plateau-only diffusion equation.

Figure 3. Plateau fill fraction versus (a) and (b) across all plateau-bearing states in the parametric sweep (one symbol per interpulse interval, colored by ). For the StS points, the horizontal coordinate is evaluated from the present model using the StS values of and , while the vertical coordinate is obtained from Eq. (28) with defined as the first level above where the StS population drops below 0.1 of the steady AMR distribution at the same . Lines show the analytic prediction of Eq. (29), which collapses from an -dependent family in panel (a) onto a single universal curve in panel (b).
Accuracy Over the Parameter Space
Table 1 summarizes the 90th-percentile (P90) population error across all 1440 states of the parametric sweep (considering all levels where either distribution exceeds ). Pooled across the entire dataset (bottom row), the steady AMR model errs by nearly two orders of magnitude (1.88 decades) and the adapted ZNS model by more than one order of magnitude (1.32 decades), whereas the present model brings the P90 error down to 0.53 decades—roughly a factor of three in level population.
Looking across the individual operating parameters in Table 1, the largest improvements consistently appear where V–V up-pumping has the furthest to climb or the least time to get there. At 300 K, where the plateau is longest and V–T relaxation is slow, the present model cuts the error from over two decades for steady AMR to under half a decade (0.46). As the gas warms to 1500 K, rises and rapid V–T relaxation drives the distribution to steady state within a few pulses, so all three models converge to similar accuracy. Lowering the pressure from 1 to 0.041 atm slows down the V–V collision rate and keeps almost three-quarters of the sampled states in the transient filling regime, while raising the repetition frequency toward 100 kHz leaves too little time between pulses for the plateau to equilibrate. In both of these strongly transient limits, steady AMR errors exceed two orders of magnitude, whereas the present model remains below a single decade.
Splitting the dataset by the filling threshold of Eq. (30) confirms
that these gains come entirely from the transient regime. Across the 821 states that are still actively filling, the present model reduces the error from 2.29 decades (steady AMR) and 1.67 decades (AMR + ZNS) to 0.72 decades. Once the plateau has filled (the remaining 619 states), the front has reached and the present model naturally matches steady AMR at about a third of a decade. Notably, the adapted ZNS model actually degrades accuracy on these filled states—as well as at 1500 K and 1 kHz—because its slow, linear front advance continues to artificially deplete the upper plateau even when the true distribution has already settled.
Computational Cost
While reducing the population error to half a decade is essential for reliable kinetics, practical use in a multidimensional flow or plasma solver also requires minimal per-cell overhead. A coupled simulation evaluates the vibrational distribution at every grid point on every iteration; across or more cells and thousands of time steps, even modest per-cell costs quickly dominate the runtime. In a state-to-state calculation, transporting a separate species continuity equation for every vibrational level adds 48 transport equations, and hence 48 unknowns per grid point, for pure and roughly 300 for a state-resolved N/O/NO/CO mixture. Worse still, implicit time integration requires assembling and factoring a chemical Jacobian across all of those coupled levels at every step, with a factorization cost that grows as the cube of the number of tracked vibrational states.
The present model avoids transporting individual vibrational populations altogether. It reconstructs the distribution from four local quantities—gas temperature, number density, vibrational energy per molecule, and its period-averaged rate of change (obtained in a gas at rest simply by storing from the end of the previous pulse cycle, or in a convective flow from the period-averaged net source of the equation; Sec. II.C)—all of which are already available in a standard reacting flow or plasma solver carrying 20 to 30 macroscopic scalars. In a multi-species plasma simulation, once the model is extended to mixtures (Sec. V), replacing roughly 300 transported level populations with a single vibrational energy equation per diatomic gas would cut the number of unknowns per cell by an order of magnitude, shrink the Jacobian memory footprint by two orders of magnitude, and reduce matrix factorization cost by three orders of magnitude. At each grid point, the level populations are then recovered locally through the simple scalar bisection loop of Sec. II.C with no differential equations or matrix operations. Because steady AMR, AMR + ZNS, and the present model all run this exact same bisection loop and differ only in the closed-form factor multiplying each plateau level, the present model achieves its threefold-to-fourfold reduction in logarithmic error at essentially zero additional computational cost.
where the AMR approximation holds. Extending it to mixtures with inert diluents requires only AMR’s mole-fraction scaling of the V–V rate and partner-averaged V–T rates, while polar diatomics such as CO require retaining the long-range V–V and radiative terms; mixtures containing multiple V–V-active diatomics would additionally require inter-species V–V′ exchange, which is not treated here. Likewise, because the period-averaged rate reflects only the net accumulation of vibrational energy rather than how that energy is deposited, the model should also apply to sinusoidal discharges and repetitive laser-induced plasmas, although this remains to be tested.
Finally, several modeling approximations and validation bounds point to natural directions for future work. On the theoretical side, replacing the time derivative with condenses the multi-level filling history into a single bulk energy timescale, while the V–T drain and dissociation wall factor are retained in their quasi-steady forms. The model also assumes that the Treanor branch below is already equilibrated at the instantaneous vibrational temperature; as discussed in Sec. IV.B, finite V–V relaxation near causes the upper Treanor levels to lag slightly during rapid excitation, overstating the plateau base amplitude and pushing the predicted front modestly ahead of the state-to-state reference. On the validation side, the model has thus far been evaluated a posteriori in a stationary gas during energy build-up, where the material and cycle-to-cycle rates of change of coincide. Embedding the model directly into a coupled multidimensional flow and plasma solver—evaluating the material rate from the net vibrational energy source in convective flows and extending the treatment to decaying-energy regimes—will provide a complete assessment of its performance in predictive simulations.
Conclusions
This work has derived and validated an analytic model that extends the quasi-steady AMR vibrational distribution to the transient plateau-filling regime of periodically driven molecular plasmas. By replacing elapsed time in the ZNS nonlinear V–V diffusion equation with a local vibrational energy timescale , the governing PDE in level space is reduced to a time-free ODE that can be solved in closed form without linearizing the V–V flux. The resulting unsteady correction factor multiplies the steady plateau and is controlled by a single dimensionless group, , which weights the number of V–V collisions per energy timescale by the plateau amplitude at the Treanor minimum. Because the full nonlinearity of V–V up-pumping is retained, the distribution front advances as the square root of rather than linearly as in adapted self-similar theory—reaching the upper ladder far faster during early pulses, avoiding the many-order-of-magnitude overprediction of the dissociating levels by the steady model, filling the plateau once the front reaches the dissociation limit, and recovering the steady AMR plateau as grows further. Across 1440 sampled states of state-to-state calculations for repetitively pulsed N discharges spanning 300–1500 K, 0.041–1 atm, and 1–100 kHz, the model reduces the 90th-percentile population error from 1.88 decades for steady AMR and 1.32 decades for the adapted ZNS baseline to 0.53 decades. Reconstructing the distribution requires only four local macroscopic quantities—gas temperature, number density, vibrational energy per molecule, and its period-averaged rate of change—avoiding transported level populations altogether at essentially the same evaluation cost as the steady AMR model.
Although benchmarked here for pure N under nanosecond pulse trains, the formulation is general with respect to both gas composition and the periodic excitation source. Because species properties enter only through molecular, spectroscopic, and FHO constants, the model is expected to apply directly to other nonpolar diatomic gases
Acknowledgments
The authors are thankful to the National Science Foundation Transport Phenomena program NSF 26-518 for supporting this research under grant #2607765. The authors are also thankful to Prof. Igor Adamovich for fruitful discussions that helped improve the quality of the manuscript.
Data Availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- [1]Bak, M. S., Kim, W., and Cappelli, M. A., “On the quenching of excited electronic states of molecular nitrogen in nanosecond pulsed discharges in atmospheric pressure air,” Applied Physics Letters, Vol. 98, No. 1, 2011, pp. 011502. https://doi.org/10.1063/1.3535986.
- [2]Adamovich, I. V., Li, T., and Lempert, W. R., “Kinetic mechanism of molecular energy transfer and chemical reactions in low-temperature air-fuel plasmas,” Philosophical Transactions of the Royal Society A, Vol. 373, 2015, pp. 20140336. https://doi.org/10.1098/rsta.2014.0336.
- [3]Chen, B.-S., Garner, A. L., and Bane, S. P. M., “Numerical analysis of a nanosecond repetitively pulsed plasma-assisted counterflow diffusion flame,” Journal of Applied Physics, Vol. 133, 2023, pp. 203302. https://doi.org/10.1063/5.0147305.
- [4]Castela, M., Fiorina, B., Coussement, A., Gicquel, O., Darabiha, N., and Laux, C. O., “Modelling the impact of non-equilibrium discharges on reactive mixtures for simulations of plasma-assisted ignition in turbulent flows,” Combustion and Flame, Vol. 166, 2016, pp. 133–147. https://doi.org/10.1016/j.combustflame.2016.01.009.
- [5]Djoud, R. J., Laws, N., and Guerra-Garcia, C., “Mapping the performance envelope and energy pathways of plasma-assisted ignition across combustion environments,” Combustion and Flame, Vol. 271, 2025, pp. 113793. https://doi.org/10.1016/j.combustflame.2024.113793.
- [6]Dumitrache, C., VanOsdol, R., Limbach, C. M., and Yalin, A. P., “Control of Early Flame Kernel Growth by Multi-Wavelength Laser Pulses for Enhanced Ignition,” Scientific Reports, Vol. 7, 2017, pp. 10239. https://doi.org/10.1038/s41598-017-10457-0.
- [7]Yu, Y., Li, X., An, X., Yu, X., Fan, R., Chen, D., and Sun, R., “Stabilization of a premixed methane-air flame with a high repetition nanosecond laser-induced plasma,” Optics & Laser Technology, Vol. 92, 2017, pp. 24–31. https://doi.org/10.1016/j.optlastec.2017.01.001.
- [8]Tinck, S. and Bogaerts, A., “Role of vibrationally excited HBr in a HBr/He inductively coupled plasma used for etching of silicon,” Journal of Physics D: Applied Physics, Vol. 49, 2016, pp. 245204. https://doi.org/10.1088/0022-3727/49/24/245204.
- [9]Foucher, M., Marinov, D., Carbone, E., Chabert, P., and Booth, J.-P., “Highly vibrationally excited O₂ molecules in low-pressure inductively-coupled plasmas detected by high sensitivity ultra-broadband optical absorption spectroscopy,” Plasma Sources Science and Technology, Vol. 24, 2015, pp. 042001. https://doi.org/10.1088/0963-0252/24/4/042001.
- [10]Annušová, A., Marinov, D., Booth, J.-P., Sirse, N., da Silva, M. L., Lopez, B., and Guerra, V., “Kinetics of highly vibrationally excited O₂(X) molecules in inductively-coupled oxygen plasmas,” Plasma Sources Science and Technology, Vol. 27, 2018, pp. 045006. https://doi.org/10.1088/1361-6595/aab47d.
- [11]Jiang, J., Richards, C., Adamovich, I., and Bruggeman, P. J., “Molecular beam mass spectrometry measurements of vibrationally excited N₂ in the effluent of an atmospheric plasma jet: A comparison with a state-to-state kinetic model,” Plasma Sources Science and Technology, Vol. 31, No. 10, 2022, pp. 10LT03. https://doi.org/10.1088/1361-6595/ac954c.
- [12]Miyake, A., Shirai, N., and Sasaki, K., “Contribution of vibrationally excited molecular nitrogen to ammonia synthesis using an atmospheric-pressure plasma jet,” Journal of Applied Physics, Vol. 135, No. 21, 2024, pp. 213301. https://doi.org/10.1063/5.0208655.
- [13]Bruggeman, P. J., Kushner, M. J., Locke, B. R., et al., “Plasma-liquid interactions: A review and roadmap,” Plasma Sources Science and Technology, Vol. 25, No. 5, 2016, pp. 053002. https://doi.org/10.1088/0963-0252/25/5/053002.
- [14]Davies, H. L., Guerra, V., van der Woude, M., Gans, T., O’Connell, D., and Gibson, A. R., “Vibrational kinetics in repetitively pulsed atmospheric pressure nitrogen discharges: Average-power-dependent switching behaviour,” Plasma Sources Science and Technology, Vol. 32, No. 1, 2023, pp. 014003. https://doi.org/10.1088/1361-6595/aca9f4.
- [15]Treanor, C. E., Rich, J. W., and Rehm, R. G., “Vibrational Relaxation of Anharmonic Oscillators with Exchange-Dominated Collisions,” Journal of Chemical Physics, Vol. 48, No. 4, 1968, pp. 1798–1807. https://doi.org/10.1063/1.1668914.
- [16]Gordiets, B. F. and Mamedov, S. S., “Vibrational distribution functions and relaxation rate in anharmonic-oscillator systems,” J. Appl. Mech. Tech. Phys., Vol. 15, 1974, pp. 297–304. https://doi.org/10.1007/BF00852216.
- [17]Parent, B., “Superelastic Heating in Treanor-Gordiets Plasmas: A Unified Analytic Closure,” Journal of Chemical Physics, Vol. 164, 2026, pp. 181101. https://doi.org/10.1063/5.0325075.
- [18]Adamovich, I. V., Macheret, S. O., and Rich, J. W., “Spatial non-homogeneity effects in nonequilibrium vibrational kinetics,” Chemical Physics, Vol. 182, 1994, pp. 167–183. https://doi.org/10.1016/0301-0104(94)00004-2.
- [19]Altin, M., Vialetto, L., Longo, S., Viegas, P., and Diomede, P., “A Modified Fokker–Planck Approach for a Complete Description of Vibrational Kinetics in a N₂ Plasma Chemistry Model,” Journal of Physical Chemistry A, Vol. 127, No. 1, 2023, pp. 261–275. https://doi.org/10.1021/acs.jpca.2c06042.
- [20]Zhdanok, S. A., Napartovich, A. P., and Starostin, A. N., “Establishment of the vibrational distribution of diatomic molecules,” Sov. Phys. JETP, Vol. 49, No. 1, 1979, pp. 66–71, Zh. Eksp. Teor. Fiz. 76, 130–139 (1979).
- [21]Gordiets, B. F. and Zhdanok, S., “Analytical Theory of Vibrational Kinetics of Anharmonic Oscillators,” Nonequilibrium Vibrational Kinetics, edited by M. Capitelli, Vol. 39 of Topics in Current Physics, Springer, Berlin, 1986, pp. 47–84. https://doi.org/10.1007/978-3-642-48615-9_3.
- [22]Gordiets, B. F., Osipov, A. I., and Shelepin, L. A., Kinetic Processes in Gases and Molecular Lasers, Gordon and Breach, London, 1988, Russian original: Nauka, Moscow, 1980.
- [23]Linstrom, P. J. and Mallard, W. G., editors, NIST Chemistry WebBook, NIST Standard Reference Database Number 69, National Institute of Standards and Technology, Gaithersburg, MD, 2026. https://doi.org/10.18434/T4D303.
- [24]Macheret, S. O., Ph.D. thesis, I. V. Kurchatov Institute of Atomic Energy, Moscow, 1985, in Russian.
- [25]Adamovich, I. V., “Three-Dimensional Analytic Model of Vibrational Energy Transfer in Molecule-Molecule Collisions,” AIAA Journal, Vol. 39, No. 10, 2001, pp. 1916–1925. https://doi.org/10.2514/2.1181.
- [26]Millikan, R. C. and White, D. R., “Systematics of vibrational relaxation,” The Journal of Chemical Physics, Vol. 39, No. 12, 1963, pp. 3209–3213. https://doi.org/10.1063/1.1734182.
- [27]Nikitin, E. E. and Troe, J., “70 Years of Landau–Teller Theory for Collisional Energy Transfer: Semiclassical Three-Dimensional Generalizations of the Classical Collinear Model,” Physical Chemistry Chemical Physics, Vol. 10, No. 11, 2008, pp. 1483–1501. https://doi.org/10.1039/b715095d.
- [28]Landau, L. and Teller, E., “Zur Theorie der Schalldispersion,” Physikalische Zeitschrift der Sowjetunion, Vol. 10, 1936, pp. 34–43.
- [29]Laporta, V., Little, D. A., Celiberto, R., and Tennyson, J., “Electron-Impact Resonant Vibrational Excitation and Dissociation Processes Involving Vibrationally Excited N₂ Molecules,” Plasma Sources Science and Technology, Vol. 23, No. 6, 2014, pp. 065002. https://doi.org/10.1088/0963-0252/23/6/065002.