betaFeedback

WorksPhysics and Astronomy

Kinetic modifications to parallel transport in a low-recycling scrape-off layer: A one dimensional gyrokinetic study

Abstract

A liquid lithium boundary offers a route to improved energy confinement via reduced recycling, but the resulting low-density, high-temperature edge can become sufficiently collisionless such that standard fluid models are no longer applicable. Experimental measurements of scrape-off-layer (SOL) parameters, including density decay length (λnλ_n) estimates presented here can be used for simple analytic estimates of kinetic effects, but need more complete kinetic treatment. We present gyrokinetic simulations, with one configuration-space dimension and two velocity-space dimensions, of low and higher-recycling LTX-ββ SOL conditions using Gkeyll, an open-source continuum gyrokinetic code. The simulations follow a single open field line, self-consistently evolving kinetic ion and electron distribution functions, electrostatic potential, collisions, sources, and sheath losses constrained by experimental midplane measurements. The hotter low-recycling case develops a larger parallel potential variation, and a density response not captured by an adiabatic Boltzmann-electron closure of the field equations, demonstrating the importance of kinetic electrons and the ambipolar potential in reducing electron end losses. Collisionality scans show that, as ion neoclassical collisionality falls below unity, the ion distribution develops pronounced non-Maxwellian features. The corresponding SOL particle and energy confinement times increase at low collisionality, particularly for low recycling, because ions become mirror-trapped and electron losses are throttled by the ambipolar potential. These results indicate that low-recycling lithium boundaries in LTX-ββ access a regime where mirror trapping and electrostatic confinement fundamentally alter parallel residence times relative to conventional fluid expectations, an effect expected to be more pronounced in larger devices.

Introduction

Fusion pilot plants must address the divertor power-exhaust challenge due to narrow scrape-off-layer heat flux widths (λq\lambda_q) and correspondingly large parallel heat fluxes [3, 4, 5, 6, 7]. Under such extreme exhaust conditions, even the best solid (tungsten) plasma-facing components (PFCs) face several well-known challenges [8, 9, 10, 11]. Liquid metal plasma facing surfaces/components are seen as an attractive alternative, with flow and/or evaporation offering possible ways to resiliently handle heat fluxes [36, 37].

Beyond general liquid metal effects on power handling, lithium can have a strong effect on SOL physics and heat fluxes due to its unique ability to greatly reduce the recycling of hydrogen plasma incident on the PFCs. In steady state, solid high-Z PFCs recycle all incoming plasma and build up a neutral population that readily charge exchanges with the incoming SOL plasma and acts as an energy sink, clamping edge temperatures [15, 16, 17]. By contrast, lithium can retain up to 85%\% of the incoming ions [14, 13]. Hydrogen reflection and retention also depend on the composition and chemical state of lithium-containing surfaces [20]. With reduced recycling, the neutral source and associated charge-exchange cooling are strongly reduced, so the edge is expected to run hotter than in a conventional high-recycling boundary. In the limiting case where edge cooling mechanisms are removed, low-recycling theory predicts that the temperature at the edge of the confined plasma can approach the core temperature, producing flat or nearly flat temperature profiles [38, 39, 40]. This low-recycling, high-temperature edge regime was anticipated and observed in LTX in the electron channel, where lithium plasma-facing surfaces remained compatible with elevated edge temperatures and reduced edge collisionality [1]. Later LTX-β\beta  [21] experiments extended this flat-temperature, low-recycling operating regime with ohmic and neutral-beam heating [23, 73, 74, 75, 76, 77, 78].

Increasing lithium wall coatings in LTX-β\beta has been associated with improved neutral and plasma density control [24]; edge neutrals also affect tearing-mode activity and access to flat temperature profiles [19]. Vacuum-transfer methods enable characterization of plasma-exposed PFCs [22], and surface measurements in LTX-β\beta have correlated lithium PFC composition and oxidation with plasma conditions and performance [25, 26].

The resulting combination of high edge temperature and low density naturally pushes the SOL toward weak collisionality, where kinetic parallel dynamics and mirror trapping can become important. The kinetic parallel dynamics of a low-recycling SOL have not been explored in detail, and the implications for SOL profiles and heat fluxes are potentially very important but not yet well understood. Experimental SOL decay widths have also not previously been reported in the low-recycling regime. Experimental data from LTX-β\beta indicates that low-recycling discharges typically exhibit densities and temperatures consistent with a kinetic scrape-off layer [12], with ion and electron collisionalities between 1/10 and 1/100 at the last closed flux surface (LCFS).

In this paper we present results from 1D full-ff gyrokinetic simulations of the scrape-off layer of LTX-β\beta using the Gkeyll code [18]. The simulations are performed along a field line that connects the low field side mid-plane to the high field side limiter. The simulations are constrained by density and temperature measurements from representative discharges in a low-recycling ensemble (R=0.61–0.67R=0.61\text{--}0.67) and a high-recycling ensemble (R=0.83–0.87R=0.83\text{--}0.87) [12]. Because the plasma shapes of the two shots are not perfectly identical, we use the equilibrium of the high-recycling shot for a more direct comparison, allowing us to isolate how the low and high-recycling plasma regimes differ in along-the-field-line particle and energy transport.

It is seen from collisionality scans of the open field line that the ion distribution at the low field side mid-plane has a large non-Maxwellian fraction. LTX-β\beta is a low aspect ratio tokamak with limiting surfaces at the high field side. The plasma, therefore, sees a modest mirror ratio of ∼\sim 4 from the low field side midplane to high field side limiter along the field line. At lower collisionality this effect likely gives rise to mirror trapping along the open field line, since trapped particle trajectories are not destroyed collisionally. To test this hypothesis we also simulated the same field line but with homogeneous magnetic field magnitude, effectively removing the mirror force. The simulations with constant magnetic field do not exhibit poloidal density peaking (§ 8).

The manuscript is organized as follows. The low and high-recycling discharges are described in section 2, including Thomson and probe measurements of core and SOL profiles along with estimates of SOL density decay lengths. Section 3 details the full-ff gyrokinetic model (one-dimensional in configuration space and two-dimensional in velocity space) implemented in Gkeyll , describing the field-aligned coordinate geometry, the collision operator, the source terms, and two different electrostatic field models. Section 4 presents the results of these simulations, comparing the high and low-recycling cases and the different electrostatic models, followed by an ion collisionality scan to explore mirror trapping and the effect on SOL confinement times. Section 5 offers concluding remarks, using simple particle-balance arguments to discuss the simulated changes in confinement in the context of the experimentally observed density decay length trends. Finally, the appendices detail convergence studies and comparison simulations without the mirror force.

LTX-β\beta low and high-recycling discharges

The experiments in this work were performed in LTX-β\beta , with solid lithium coatings applied to the stainless steel plasma facing surfaces of its internal shell. The shell conforms to a limited plasma with a major radius of 0.4 m and a minor radius of 0.26 meters [1, 12]. Two lithium evaporators, placed diametrically opposite to each other, coat >> 96%\% of the plasma-facing shell surface. The average coating thickness is greater than the maximum plasma interaction depth [24].

Figure 1

Figure 1. a) Plasma current and line averaged density for two representative shots, one with high recycling and one with low-recycling walls. The time point of investigation is indicated with a vertical line at t = 0.468 sec. b) Thomson scattering data for the two shots at t = 0.468 s. c) Kinetically (Thomson pressure profile) constrained equilibrium reconstruction of the two shots at the same time point.

We compare two shots, 103795 and 103955 [12]. These are the two median shots from ensembles of repeated identical shots from two run days, each with a different state of lithium on the walls. Throughout this paper, the labels “low-recycling case” and “high-recycling case” are shorthand for the simulation parameter sets constrained by these experimentally characterized discharge ensembles; they do not imply that recycling is evolved by the model. Here, “high recycling” denotes the higher of the two LTX-β\beta recycling conditions rather than a conventional R≃1R\simeq 1 high-recycling regime. Shot 103795 will be referred to as the high-recycling case and is from a partially passivated lithium run day, where the previous lithium evaporation is 5 weeks old. It is understood from previous surface science analysis of the LTX-β\beta walls [26, 25] that at least the top several nanometers of lithium is oxidized with a combination of lithium oxide and lithium hydroxide. Oxidized lithium is expected to have a higher hydrogen reflectivity and consequently lead to higher recycling [14]. A global recycling coefficient of R=0.83–0.87R=0.83\text{--}0.87 was inferred for the passivated-lithium ensemble containing this shot [12]. Shot 103955 is the low-recycling case and has a fresh, nearly pure lithium evaporation of ∼\sim 863 mg, with an average coating thickness of >> 100 nm and belongs to the ensemble for which a global recycling coefficient of R=0.61–0.67R=0.61\text{--}0.67 was inferred [12].

The plasma current and line averaged density waveforms are shown in Figure 1(a), while electron temperature and density profiles from the core Thomson scattering system [41] are plotted in Figure 1(b). The two shots have identical gas fueling, but strong pumping by the lithium walls leads to a much lower line averaged density in the low-recycling shot. The core temperature is also higher in the low-recycling shot, likely due to improved confinement and higher plasma current from reduced recycling and reduced radiative losses compared to oxidized lithium. The PSI-TRI [28] equilibrium reconstructions of the two shots appear in Figure 1(c). The open field line geometry is similar between the two shots, but the low-recycling shot has a larger plasma minor radius and elongation and therefore a slightly larger connection length along the SOL field line. We see that the edge density is similar between the two shots, but the edge temperature is higher in the low-recycling shot, consistent with reduced cooling from charge exchange with neutrals and a lower secondary emission coefficient of nearly pure lithium compared to oxidized lithium [42]. For purposes of the Gkeyll simulations in section 3, the midplane density of both cases was taken to be 2×10182\times10^{18} m−3^{-3}, but the temperature was set to the last measured point within the reconstructed boundary, 30 eV for the high-recycling case and 175 eV for the low-recycling case. Using the neoclassical collisionality expression (equation (16)), the electron collisionality of the two cases is 3.4 and 0.12; and the ion collisionality is 13.7 and 0.49.

For comparison, the conventional upstream SOL electron collisionality is νe,SOL∗=L∥/λe≃10−16ne,uL∥/Te,u2\nu_{e,\mathrm{SOL}}^*=L_\parallel/\lambda_e\simeq 10^{-16}n_{e,u}L_\parallel/T_{e,u}^2, with density in m−3^{-3}, length in m, and temperature in eV [32]. Using the representative limiter-to-limiter connection length of 11 m [12] and the midplane parameters above gives νe,SOL∗≃0.072\nu_{e,\mathrm{SOL}}^*\simeq 0.072 and 2.442.44 for the low and high-recycling cases, respectively. An analogous ion measure, νi,SOL∗=L∥νii/vti=L∥/λii\nu_{i,\mathrm{SOL}}^*=L_\parallel\nu_{ii}/v_{ti}=L_\parallel/\lambda_{ii}, gives 0.330.33 and 12.912.9, using vti=Ti0/miv_{ti}=\sqrt{T_{i0}/m_i}, the ion self-collision frequency in equation (8), and reference ion temperatures of 70 and 10 eV, respectively. If the connection length is instead taken from the upstream midplane to one limiter, L∥=5.5L_\parallel=5.5 m, the electron values are 0.0360.036 and 1.221.22, and the ion values are 0.1640.164 and 6.476.47. These SOL measures supplement the neoclassical collisionalities quoted above.

We also examined the evolution of the low-field-side density decay length during the low-recycling discharge. Figure 2 shows the inferred λn,SOL\lambda_{n,\mathrm{SOL}}, or radial density decay length in the SOL, as a function of time through the shot at five Thomson-scattering time points. The density decay length should be related to radial transport, and its evolution as a function of recycling/collisionality can provide indications about kinetic radial transport modifications. For the first three time points, the shot is being fueled by a supersonic gas injector (SGI). The SGI puff terminates about 200 microseconds before 468 msec, which is the time point corresponding to the temperature and density profiles in Figure 1. The estimate was constructed by stitching together several constraints at each time slice: the last-closed-flux-surface location from the kinetic equilibrium reconstruction, edge Thomson-scattering density points outside the LCFS, an LCFS anchor density from TRANSP, the two far-SOL Langmuir-probe density points marked by x symbols near the wall in Figure 2(b,c), projected back to the outboard midplane along open field lines. The fit is performed with a simple exponential function, ne(r)=ne0exp⁡[−(r−rLCFS)/λn,SOL]n_e(r)=n_{e0}\exp\left[-(r-r_{\mathrm{LCFS}})/\lambda_{n,\mathrm{SOL}}\right], where ne0n_{e0} is the density at the LCFS and r−rLCFSr-r_{\mathrm{LCFS}} is the radial distance outside the LCFS at the outboard midplane. The resulting λn,SOL\lambda_{n,\mathrm{SOL}} is shown on the right axis of Figure 2(a) as a function of time, and the fits for two time points are given in Figure 2(b) and (c). We see that λn,SOL\lambda_{n,\mathrm{SOL}} increases as the SGI puff terminates and the SOL warms up.

Figure 2

Figure 2. (a) Line-averaged density for the low-recycling shot, with the inferred low-field-side SOL density decay length λn,SOL\lambda_{n,\mathrm{SOL}} overlaid on the right axis as a function of time. (b) Thomson scattering density measurement near peak line-averaged density, t = 465 msec, with the fitted profile and the LCFS anchor point from TRANSP and low field side probe points mapped to the midplane. (c) At t = 471 msec, after the SGI puff has terminated, with the same fit and anchor point. The inferred λn,SOL\lambda_{n,\mathrm{SOL}} increases as the SOL warms up after the puff terminates.

Since the fit combines measurements with different spatial coverage, uncertainties, and model dependence, the resulting λn,SOL\lambda_{n,\mathrm{SOL}} inherits those systematic uncertainties. The probe densities are not used at their raw measurement location, but are mapped to the midplane using the reconstructed open field lines. Finally, the LCFS value used to anchor the fit is taken from TRANSP rather than measured directly by the edge Thomson system. These choices make the inferred decay length useful as a trend indicator, and only the temporal trend is interpreted here.

A similar analysis was not retained for the high-recycling case because the Thomson data and/or the fit quality was not sufficient for a comparable interpretation of the Thomson points past the LCFS. For the low-recycling case, the temporal trend should also be interpreted in the context of the gas puff. While the puff is active, the SOL is expected to remain relatively cold even with fresh lithium on the walls because the SGI floods the SOL with a cold neutral source. After the puff is terminated, the edge density falls and the colder population can be flushed out as the lithium walls continue to pump particles, allowing the SOL to warm back up [12]. The increase in λn,SOL\lambda_{n,\mathrm{SOL}} later in time likely reflects both the warming of a low-recycling SOL and a change in the balance between local fueling, wall pumping, and radial profile evolution. These experimental constraints motivate the field-line kinetic model described next.

Kinetic studies of SOL parallel transport have examined non-Maxwellian distributions and sheath-connected losses across a wide range of collisionalities and recycling conditions [65, 66, 67, 68]. Reduced electron conductive heat flux has also been obtained without magnetic mirroring [69, 70]. Experimentally constrained kinetic SOL modeling has been performed for the high-recycling COMPASS divertor [71]. Those conditions differ from the hot-edge regime expected when wall recycling and the associated neutral cooling are strongly reduced. This low-recycling regime, anticipated theoretically and demonstrated experimentally in LTX and LTX-β\beta, provides the physical basis for the present study [38, 39, 72, 12]. Building on preliminary kinetic modeling of these conditions, this study examines how mirror-trapped ion populations and the self-consistent kinetic-electron response jointly affect parallel density structure and effective SOL confinement. In addition to distribution functions and parallel profiles, we evaluate particle and energy residence times from the corresponding inventories and source rates. These diagnostics quantify how much plasma and thermal energy the SOL retains for a given fueling and heating rate, complementing the heat-flux and sheath diagnostics emphasized in earlier studies. Normalization to a sound-transit time distinguishes changes in effective confinement from the direct effect of temperature on parallel transit. Systematic collisionality scans, together with comparisons between kinetic and Boltzmann electrons and between varying and constant magnetic fields, connect changes in velocity-space structure to changes in these residence times. The combined analysis therefore examines how electron kinetics and ion mirror trapping modify particle and energy retention in experimentally accessible low-recycling lithium-wall conditions, with relevance to comparable regimes beyond LTX-β\beta.

The Gkeyll gyrokinetic model

Gkeyll can solve a full-ff gyrokinetic model [31] that obtains the time and phase-space evolution of the guiding center distribution function fs(R,v∥,μ)f_s({\mathbf{R}},v_\parallel,\mu) of each species labeled ss. Phase space here refers to the high-dimensional space spanned by the guiding center position (R{\mathbf{R}}), the particle velocity (v∥v_\parallel) parallel to the equilibrium magnetic field B=B b^{\mathbf{B}}=B\,{\mathbf{\hat{b}}}, and the magnetic moment (μ=msv⊥2/(2B)\mu=m_sv_\perp^2/(2 B)) related to the perpendicular particle velocity v⊥v_\perp. Gkeyll can solve this equation in a variety of geometries and dimensionalities, and for this work we employ a field-aligned coordinate system [2] where the guiding-center position R(x){\mathbf{R}}({\mathbf{x}}) is mapped to the computational coordinates x=(x,y,z){\mathbf{x}}=(x,y,z):

x=r−r0,y=−r0q0(φ+α(r,θ,φ=0)),z=θ,\begin{equation} x = r-r_0, \qquad y = -\frac{r_0}{q_0}\left(\varphi+\alpha(r,\theta,\varphi=0)\right), \qquad z = \theta, \tag*{(1)} \end{equation}

where θ\theta is the poloidal angle, r0r_0 is the minor radius at the center of the simulation domain, and q0q_0 is the safety factor at r0r_0 assuming an analytic profile q(r)=qLCFS/[1−sLCFS(r−aOMP)/aOMP]q(r)=q_{LCFS}/[1-s_{LCFS}(r-a_{OMP})/a_{OMP}] tailored to approximate and extend the qq-profile inside the LCFS; here aOMPa_{OMP} is the minor radius at the outboard mid-plane (OMP), and qLCFSq_{LCFS} and sLCFSs_{LCFS} are the safety factor and magnetic shear at the LCFS. The field-line label α=α(r,θ,φ)\alpha=\alpha(r,\theta,\varphi) is calculated via

α=φ−I(Ψ)∫0θdθ′ Jr(r,θ′)R(r,θ′)2,I(Ψ)=2πq(∫02πdθ Jr(r,θ)R(r,θ)2)−1.\begin{equation} \alpha = \varphi - I(\Psi)\int_0^\theta\mathrm{d}\theta'\,\frac{J_r(r,\theta')}{R(r,\theta')^2}, \qquad I(\Psi) = 2\pi q\left(\int_0^{2\pi}\mathrm{d}\theta\,\frac{J_r(r,\theta)}{R(r,\theta)^2}\right)^{-1}. \tag*{(2)} \end{equation}

The integrands contain the Jacobian Jr=R[(∂rR)∂θZ−(∂θR)∂rZ]J_r = R\left[\left(\partial_r R\right)\partial_\theta Z-\left(\partial_\theta R\right)\partial_r Z\right] of the (r,θ,φ)(r,\theta,\varphi) coordinates, which can be computed for a Miller model of the flux surfaces [27]

R(r,θ)=Raxis+rcos⁡(θ+arcsin⁡(δ)sin⁡(θ)),Z(r,θ)=κrsin⁡(θ)\begin{align} R(r,\theta) &=& R_{\mathrm{axis}}+ r\cos\left(\theta + \arcsin(\delta)\sin(\theta)\right), \tag*{(3)} \\ Z(r,\theta) &=& \kappa r\sin(\theta) \tag*{(4)} \end{align}

where δ\delta and κ\kappa are the triangularity and elongation, respectively. We adjusted this analytic equilibrium, using Raxis=0.4R_{\mathrm{axis}}=0.4, κ=1.57\kappa=1.57, δ=0.6\delta=0.6, to give an open-flux surface that resembles that inferred by a kinetically constrained PSI-TRI [28] equilibrium (see Figure 3(b)). Also informed by equilibrium reconstruction is our choice of qLCFS=3.6955q_{\mathrm{LCFS}}=3.6955, sLCFS=2.2798s_{\mathrm{LCFS}}=2.2798 and Baxis=0.2244B_{\mathrm{axis}}=0.2244 T.

Figure 3

Figure 3. Field-line geometry used in the Gkeyll simulations. The computational domain follows a single open field line that resembles the PSI-TRI reconstruction of the high-recycling discharge. The figure shows the location of the simulated field line relative to the LCFS for the high-recycling case, and the position of the LTX-β\beta shell. Reproduced from [12] under a Creative Commons Attribution (CC BY) license.

Since we are interested in characterizing parallel dynamics of the LTX-β\beta  SOL, we use Gkeyll in the electrostatic limit in one position-space dimension along a single SOL field line crossing the OMP at r0=R0−Raxisr_0=R_0-R_{\mathrm{axis}} or R0=RLCFS,OMP+0.025 mR_0=R_{\mathrm{LCFS},\mathrm{OMP}}+0.025\,\mathrm{m} (i.e. 2.5 cm outside the LCFS whose major radius at the OMP is RLCFS,OMP=0.5948R_{\mathrm{LCFS},\mathrm{OMP}}=0.5948 m). Gkeyll is thus used here to evolve (in time) the guiding center distribution, fs(z,v∥,μ)f_s(z,v_\parallel,\mu), of a species ss with mass msm_s and charge qsq_s as a function of the along-the-field-line coordinate zz (as well as v∥v_\parallel and μ\mu). Specifically, we solve

∂Jfs∂t+1Jx∂∂z(Jxz˙ Jfs)+∂∂v∥(v∥˙ Jfs)=JCs[fs]+JSs.\begin{equation} \frac{\partial \mathcal{J}f_s}{\partial t} + \frac{1}{J_x}\frac{\partial }{\partial z}\left(J_x\dot{z}\,\mathcal{J}f_s\right) + \frac{\partial }{\partial v_\parallel}\left(\dot{v_\parallel}\,\mathcal{J}f_s\right) = \mathcal{J}\mathcal{C}_s[f_s] + \mathcal{J}S_s. \tag*{(5)} \end{equation}

The time derivatives of zz and v∥v_\parallel are

z˙=CJxB∥∗v∥,v∥˙=−CJxB∥∗1ms(μ∂B∂z+qs∂ϕ∂z),\begin{equation} \dot{z} = \frac{\mathcal{C}}{J_xB^*_\parallel}v_\parallel, \qquad \dot{v_\parallel} = -\frac{\mathcal{C}}{J_xB^*_\parallel}\frac{1}{m_s}\left(\mu\frac{\partial B}{\partial z}+q_s\frac{\partial \phi}{\partial z}\right), \tag*{(6)} \end{equation}

where J=B∥∗=b^⋅B∗≈B\mathcal{J}=B^*_\parallel={\mathbf{\hat{b}}}\cdot{\mathbf{B^*}}\approx B is the Jacobian of the guiding-center transformation, and ϕ\phi is the electrostatic potential. The function Jx=det(gij)J_x=\sqrt{\mathrm{det}\left(g_{ij}\right)} corresponds to the Jacobian of the R(x){\mathbf{R}}({\mathbf{x}}) transformation with the covariant metric tensor gijg_{ij}, such that C=JxB/gzz\mathcal{C}=J_xB/\sqrt{g_{zz}}. The magnetic field magnitude B=(Bφ2+Bθ2)1/2B=\left(B_{\varphi}^2+B_{\theta}^2\right)^{1/2} is composed of the toroidal Bφ(R)=BaxisRaxis/RB_{\varphi}(R)=B_{\mathrm{axis}}R_{\mathrm{axis}}/R and poloidal Bθ=∣∇φ×∇Ψ∣=∣∇r∣Ψ′/RB_\theta=\left| \nabla \varphi\times\nabla \Psi \right|=\left| \nabla r \right|\Psi'/R components, where Ψ′=Bφ(R0)Raxis/I(Ψ)\Psi'=B_\varphi(R_0)R_{\mathrm{axis}}/I(\Psi) and ∣∇r∣2=(R/Jr)2[(∂θR)2+(∂θZ)2]\left| \nabla r \right|^2=\left(R/J_r\right)^2\left[\left(\partial_\theta R\right)^2+\left(\partial_\theta Z\right)^2\right].

On the right-hand side of equation (5), Cs[fs]\mathcal{C}_s[f_s] is the multispecies gyroaveraged Dougherty collision operator,

Cs[fs]=∑rνsr{∂∂v∥[(v∥−u∥,sr)fs+vt,sr2∂fs∂v∥]+∂∂μ[2μfs+2msvt,sr2Bμ∂fs∂μ]}.\begin{equation} \mathcal{C}_s[f_s] = \sum_r \nu_{sr}\left\{ \frac{\partial }{\partial v_\parallel}\left[\left(v_\parallel-u_{\parallel,sr}\right)f_s+v_{t,sr}^2\frac{\partial f_s}{\partial v_\parallel}\right] +\frac{\partial }{\partial \mu}\left[2\mu f_s+2\frac{m_s v_{t,sr}^2}{B}\mu\frac{\partial f_s}{\partial \mu}\right]\right\}. \tag*{(7)} \end{equation}

Here rr denotes the collision partner, and the sum includes both self- and unlike-species collisions. For r=sr=s, u∥,ss=u∥su_{\parallel,ss}=u_{\parallel s} and vt,ss2=vts2v_{t,ss}^2=v_{ts}^2. For r≠sr\ne s, u∥,sru_{\parallel,sr} and vt,sr2v_{t,sr}^2 are the conservative cross-species primitive moments. These moments are calculated from the evolving distributions using prescriptions that conserve momentum and energy to machine precision [53, 54]. Electron–ion and ion–electron collisions were enabled in all simulations with kinetic electrons. The self-collision frequencies  [56] are calculated using

log⁡Λs=6.6−0.5log⁡(ns0/1020)+1.5log⁡(Ts0/e),νss=νfrace4ns0log⁡Λs3(2π)3/2ϵ02dsms Ts03/2,\begin{equation} \begin{aligned} \log\Lambda_s &= 6.6 - 0.5 \log\left(n_{s0}/10^{20}\right) + 1.5 \log\left(T_{s0}/e\right), \\ %\nu_{ee} &= \nu_{\mathrm{frac}} \frac{\log\Lambda_{e} e^4 n_{e0}}{3 \left(2\pi\right)^{3/2} \epsilon_0^2 \sqrt{m_e} T_{e0}^{3/2}}, \\ %\nu_{ii} &= \nu_{\mathrm{frac}} \frac{\log\Lambda_i e^4 n_{s0}}{3 \left(2\pi\right)^{3/2} \epsilon_0^2\sqrt{2m_i}T_{i0}^{3/2}}. \nu_{ss} &= \nu_{\mathrm{frac}} \frac{e^4 n_{s0}\log\Lambda_{s}}{3 \left(2\pi\right)^{3/2} \epsilon_0^2 \sqrt{d_sm_s}~T_{s0}^{3/2}}, \\ \end{aligned} \tag*{(8)} \end{equation}

where ns0n_{s0} and Ts0T_{s0} are the fixed reference number density and temperature (in energy units) of species ss, νfrac=1\nu_{\mathrm{frac}}=1, ds=1d_s=1 for electrons and ds=2d_s=2 for ions. For the Gkeyll version used here, the cross-species frequencies are νei=2νee\nu_{ei}=\sqrt{2}\nu_{ee} and νie=(me/mi)νei\nu_{ie}=(m_e/m_i)\nu_{ei}. Note that the Dougherty operator in equation (7) lacks the 1/v3v^3 dependence in the collision frequency characteristic of the full Fokker-Planck operator. This simplification removes physics, like the thermal force, that can be important in SOL transport [60], and reduces the accuracy with which we model heat-carrying particles that are important in setting divertor heat loads, especially in collisionless reactor conditions [64].

Equation (5) also contains a source term, SsS_s, that stands as a proxy for radial transport from the core into the SOL and for particle sources due to recycling. Such a source has a Maxwellian shape in velocity space,

Ss=n˙src(2πTsrc,s/ms)3/2exp⁡[−v∥2+2μB/ms2Tsrc,s/ms],\begin{equation} \mathcal{S}_s = \frac{\dot{n}_{\mathrm{src}}}{\left(2\pi T_{\mathrm{src},s}/m_s\right)^{3/2}}\exp\left[-\frac{v_\parallel^2+2\mu B/m_s}{2T_{\mathrm{src},s}/m_s}\right], \tag*{(9)} \end{equation}

with the density source rate (n˙src\dot{n}_{\mathrm{src}}) and source temperature (Tsrc,sT_{\mathrm{src},s}) as in Figure 4. Since these rates and temperatures influence the final density and temperature, we pick these approximately so as to produce densities and temperatures for each of the low and high RR cases informed by the measurements presented in Figure 1-Figure 2. These final profiles are a result of evolving the non-drifting Maxwellian initial conditions (ICs), with the moments in Figure 5, according to equation (5), using zero-flux boundary conditions (BCs) in velocity space and conducting sheath BCs at the ends of the field line [55], until little to no change is observed.

Figure 4

Figure 4. (a) Density source rate, (b) electron source temperature, and (c) ion source temperature for low (blue) and high (orange) RR simulations.

Figure 5

Figure 5. (a) Density, (b) electron temperature, and (c) ion temperature of the non-drifting Maxwellian initial condition (IC) for low (blue) and high (orange) RR simulations.

In order to close the system of equations we need a field equation that determines the evolution of the electrostatic potential. In this work we employed two models used in previous 1D SOL modeling, an approximation to the gyrokinetic Poisson equation, and an Boltzmann electron model.

One-dimensional approximation to the Gyrokinetic Poisson equation

In the long wavelength limit the gyrokinetic Poisson equation is [29]

−∇⋅∑smsn0sB02∇⊥ϕ=∑sqs2πms∫dv∥ dμ Jfs,\begin{equation} -\nabla \cdot \sum_s\frac{m_sn_{0s}}{B_0^2}{{\nabla}_{\perp}} {\phi} = \sum_sq_s\frac{2\pi}{m_s}\int \mathrm{d}v_\parallel\,\mathrm{d}\mu~\mathcal{J}f_s, \tag*{(10)} \end{equation}

where we have made use of a Boussinesq approximation to use the reference density n0sn_{0s} and magnetic field amplitude B0B_0 on the left side of this equation, and ∇⊥{{\nabla}_{\perp}} {} is a gradient perpendicular to the magnetic field. The Boussinesq approximation has of course been identified as a source of error in 2D blobby transport [61, 63], although that error seems to become small in 3D [61, 62]. In this work, however, we pursue one-dimensional simulations along the field line and the perpendicular direction is assumed ignorable. Yet, in equation (10) the variation of the potential ϕ\phi along the field line is given by the distribution of charge on a given perpendicular plane. Although there are other candidates for field equations in 1D models, due to its simplicity, we opt for an approximation to the above Poisson equation in which we assume a Fourier transformation in the perpendicular plane and replace the ∇⋅∇⊥\nabla \cdot {{\nabla}_{\perp}} {} operator with an assumed perpendicular wavenumber k⊥k_\perp [30, 43]. Our one-dimensional field equation thus becomes

k⊥2∑smsn0sB02ϕ=∑sqs2πms∫dv∥ dμ Jfs.\begin{equation} k_\perp^2\sum_s\frac{m_sn_{0s}}{B_0^2}\phi = \sum_sq_s\frac{2\pi}{m_s}\int \mathrm{d}v_\parallel\,\mathrm{d}\mu~\mathcal{J}f_s. \tag*{(11)} \end{equation}

Formally in a 1D model k⊥k_\perp should go to zero, but in order to use a limit of equation (10) to obtain the electric field along the field line we need to keep an approximation to the perpendicular operators on the left side. This model requires us to specify a k⊥k_\perp (or, more conveniently, k⊥ρs0k_\perp\rho_{s0}, for the reference ion sound gyroradius ρs0=miTe0/(qiB0)\rho_{s0}=\sqrt{m_iT_{e0}}/(q_iB_0) in terms of the reference electron temperature Te0T_{e0}). We will evaluate the sensitivity of our results to this free parameter and ensure that we have a sufficiently low k⊥ρsk_\perp\rho_s to approximate the asymptotic limit.

Boltzmann electrons with ambipolar sheath fluxes

As a point of reference alternative to the gyrokinetic Poisson model in § 3.1, we also pursued simulations with a framework that assumes Boltzmann electrons and ambipolar sheath fluxes [30, 44]. The equality of electron (Γes\Gamma_{es}) and ion (Γis\Gamma_{is}) particle fluxes at the sheath entrance (z=zsz=z_s) means

Γes=2πme∫dv∥dμ v∥ J{ne(2πvte02)3/2exp⁡[−v∥2+2μB/me2vte02−eϕsTe0]}∣z=zs=Γis,\begin{equation} \Gamma_{es}=\frac{2\pi}{m_e}\int\mathrm{d}v_\parallel\mathrm{d}\mu\,v_\parallel\,\mathcal{J}\left\{\frac{n_e}{\left(2\pi v_{te0}^2\right)^{3/2}}\exp\left[-\frac{v_\parallel^2+2\mu B/m_e}{2v_{te0}^2}-\frac{e\phi_s}{T_{e0}}\right]\right\}\Bigg|_{z=z_s} = \Gamma_{is}, \tag*{(12)} \end{equation}

where vte0=Te0/mev_{te0}=\sqrt{T_{e0}/m_e} is the thermal speed of isothermal electrons. Carrying out this integral and assuming quasineutrality leads to a prescription of the sheath potential:

ϕs=Te0eln⁡(nisvte02π Γis);\begin{equation} \phi_s = \frac{T_{e0}}{e}\ln\left(\frac{n_{is}v_{te0}}{\sqrt{2\pi}\,\Gamma_{is}}\right); \tag*{(13)} \end{equation}

here we denoted the ion density at the sheath entrance as nisn_{is}. The sheath potential is employed in the Boltzmann electron response

ne(z)=nesexp⁡[e(ϕ(z)−ϕs)Te0],\begin{equation} n_e(z) = n_{es}\exp\left[\frac{e\left(\phi(z)-\phi_s\right)}{T_{e0}}\right], \tag*{(14)} \end{equation}

in terms of the electron density at the sheath entrance nesn_{es}. From equation (14) we can obtain the variation of the potential along the field line:

ϕ(z)=ϕs+Te0eln⁡(ni(z)nis),\begin{equation} \phi(z) = \phi_s+\frac{T_{e0}}{e}\ln\left(\frac{n_i(z)}{n_{is}}\right), \tag*{(15)} \end{equation}

where we once again employed quasineutrality to replace nes=nis=ni(z=zs)n_{es}=n_{is}=n_i(z=z_s). In a time-dependent simulation we can thus solve for the ion dynamics, compute the ion density (nin_i) and particle flux at the sheath entrance (Γis\Gamma_{is}) and use equations (13) and (15) to obtain the potential along the field line. Since this model does not evolve kinetic electrons, it provides an alternative approximation at significantly lower computational cost (around 42X cheaper for hydrogen ions).

Simulation results

The simulations presented here can be reproduced with the open-source code Gkeyll and the provided input files, see § 6. We begin by running reference simulations based on the experimental data described in section 2 and comparing the high-recycling and low-recycling cases. To assess the importance of kinetic-electron effects, we run the simulations with both kinetic-electron and Boltzmann-electron models. These runs establish the baseline density, flow, and temperature structure along the modeled SOL field line. The remainder of the section uses these reference cases (unless otherwise noted) to motivate the collisionality and magnetic-field scans discussed later. We then analyze the kinetic effects by examining the ion distribution function, the non-Maxwellian component of the ion distribution function, and the trapped particle fraction. Finally, we present the results of collisionality scans for both cases, showing how the density profiles and SOL particle and energy confinement times change as a function of collisionality.

The equilibrium geometry, source profile, and numerical resolution are prescribed as run inputs. Gkeyll is initialized with edge density and temperature values of approximately ne=2×1018 m−3n_e = 2\times10^{18}\,\mathrm{m}^{-3} for both the high-recycling and low-recycling cases, Te=175T_e = 175 eV for the low-recycling case, and Te=30T_e = 30 eV for the high-recycling case. An initial ion temperature of Ti=0.4 TeT_i = 0.4\, T_e eV is used. Gkeyll then evolves the initially Maxwellian ion and electron distribution functions self-consistently with the electrostatic potential along a single open field line. The resulting dynamics therefore reflect parallel kinetic transport in a fixed magnetic geometry, together with the effects of sources, sheath losses, and collisions. The sources are localized near the midplane and are designed to match the experimental midplane density and temperature. The sheath boundary conditions at the limiter end of the field line allow for kinetic losses of both ions and electrons, and the limiter is assumed to be perfectly absorbing. The numerical resolution is chosen based on the convergence studies summarized in  § 7.

Reference simulations

The reference simulations were obtained by solving equation (5) with the Gkeyll code [18], using a discontinuous Galerkin discretization with (Nz,Nv∥,Nμ)=(64,16,12)(N_{z},N_{v_\parallel},N_{\mu})=(64,16,12) cells along (z,v∥,μ)(z,v_\parallel,\mu), respectively, and a piecewise linear basis along zz and μ\mu but a piecewise quadratic basis along v∥v_\parallel [31]. The length of the domain along zz is simply the connection length for the selected field line (see Figure 3), while in velocity space the grid extents are v∥∈[−4vts0,4vts0]v_\parallel\in\left[-4v_{ts0},4v_{ts0}\right] and μ∈[0,ms(6vts0)2/(2B0)]\mu\in\left[0,m_s\left(6v_{ts0}\right)^2/(2B_0)\right], where vts0=Ts0/msv_{ts0}=\sqrt{T_{s0}/m_s} is the reference thermal speed of species ss. Note that we use a nonuniform velocity discretization, using a quadratic grid in μ\mu and a linear-quadratic grid in v∥v_\parallel [31]. A resolution convergence study is shown in § 7.

The results show that the low-recycling case sustains a hotter temperature than the high-recycling case. The midplane density evolves to similar values, with the high-recycling case showing a slightly higher value than the initial choice of ne=2×1018 m−3n_e = 2\times10^{18}\,\mathrm{m}^{-3}, as shown in Figure 13. However, the hotter low-recycling case develops a steeper density gradient along the modeled field line. The ratio of midplane to limiter density is also modified for the low-recycling case compared to the high-recycling case, as shown in Figure 6.

Figure 6

Figure 6. (a) Ion number density, (b) parallel speed (normalized to cs=(T∥e+3T∥i)/mic_s=\sqrt{(T_{\parallel e}+3T_{\parallel i})/m_i}), (c) ion temperature, and (d) electron temperature for the low-recycling case. Panels (e)–(h) show the same quantities for the high-recycling case. The kinetic-electron results are shown in solid lines, while the Boltzmann-electron results are shown in dashed lines.

The conducting-sheath boundary condition reflects electrons below the sheath-energy cutoff and absorbs outgoing ions, but does not resolve the Debye sheath or explicitly impose a Bohm flow speed [55]. The subsonic boundary flows have not been verified against the kinetic Bohm criterion. Numerical sensitivity of the sheath-entrance flow is discussed further in § 8.

The simulations were allowed to evolve toward a quasi-steady state. Time traces of the moments at the midplane, an intermediate location, and the sheath entrance are shown in  § 7. Numerical convergence checks for velocity-space resolution, velocity-space extent, and the prescribed k⊥ρsk_\perp\rho_s used in the 1D field model are also shown there. The electrostatic potential is somewhat more sensitive to the assumed k⊥ρsk_\perp\rho_s than the plasma moments, so we show that scan explicitly here. Reference simulations used a value of k⊥ρs=0.15k_\perp\rho_s=0.15, at which point the electrostatic potential at the midplane, normalized to the electron temperature, is within 1% of the asymptotic value (at k⊥ρs=0.025k_\perp\rho_s=0.025).

Figure 7

Figure 7. Electrostatic potential profiles for simulations of the low-recycling case (a) and high-recycling case (b) as a function of the parameter k⊥ρsk_\perp\rho_s, and the central potential normalized to the central electron temperature for each value of k⊥ρsk_\perp\rho_s (c).

Figure 7 shows that the electrostatic potential differs substantially between the two recycling states. The low-recycling case develops a much larger potential and a much larger variation of that potential along the field line than the high-recycling case. In a mirror geometry, particles are trapped when conservation of total energy and magnetic moment causes their parallel velocity to vanish before they can reach the high-field region, producing a trapped-passing boundary in velocity space [34, 33]. In the LTX-β\beta SOL, once the collisionality falls well below unity, the magnetic-field variation along the field line provides that mirror-trapping geometry. In the high-recycling case, this distinction is less important because the collisionality is large enough that trapped ion orbits are expected to be disrupted by pitch angle scattering before mirror trapping can strongly structure the distribution function. In that regime, the dynamics remain closer to the conventional high-recycling SOL picture, where collisional transport, parallel flow, and sheath-connected losses dominate.

In the low-recycling case, however, the larger potential implies a stronger parallel electric field, directed from the midplane toward the limiter on each side of the domain, since the potential peaks near the midplane and falls toward the sheath entrance. That field exerts a force on the electrons back toward the midplane and therefore acts to reduce direct electron end loss, while ions are more readily accelerated toward the limiter. This is the ambipolar response required to help restore quasineutrality in a system where electrons would otherwise escape too efficiently along the open field line. At the same time, the low-recycling case is also hotter, so although the potential is much larger in volts, the normalized quantity eϕ/Tee\phi/T_e is not larger than in the high-recycling case. Taken together, these results suggest that the low-recycling LTX-β\beta SOL is not well described by standard collisional SOL fluid descriptions alone. Instead, it enters a regime in which weak collisionality, magnetic mirror trapping, and the ambipolar potential combine to produce a more complex SOL structure with stronger parallel gradients and non-Maxwellian features in the distribution functions. The rest of this section explores these kinetic effects in more detail and shows how they evolve as the collisionality is varied.

The boundary potentials represent the unresolved sheath drops relative to the wall. Using the local sheath-entrance electron temperatures of approximately 145 and 30 eV gives eϕs/Te,s≃1.9e\phi_s/T_{e,s}\simeq 1.9 and 2.82.8 for the low and high-recycling cases, respectively. We interpret the smaller normalized sheath drop in the low-recycling case in terms of the combined action of the ambipolar potential along the field line and the sheath in regulating electron losses. An extended Pastukhov-type ambipolar potential was previously anticipated for the weakly collisional, mirror-trapped LTX SOL [1]. Depletion of the higher-energy electron tail can further reduce the sheath potential required to balance electron and ion losses.

Conventional SOL fluid models [32] provide a useful reference point for interpreting the present simulations. In the sheath-limited regime, parallel losses to the material boundary dominate and the plasma can often be approximated as nearly isothermal along the field line, with the sheath setting the loss condition at the wall. In the conduction-limited or high-recycling regime, stronger parallel temperature gradients develop and thermal conduction plays a larger role in setting the upstream-to-target connection. These fluid descriptions capture the basic balance between parallel transport, cross-field source terms, and sheath losses, but they rely on closure assumptions that become less reliable as the SOL becomes weakly collisional and non-Maxwellian. The 1x2v Gkeyll model used here retains that same open-field-line geometry while allowing the particle distributions and electrostatic response to evolve kinetically, making it better suited for assessing departures from standard fluid behavior in the low-recycling LTX-β\beta SOL.

A Boltzmann-electron closure was explored to isolate the role of kinetic electron dynamics in these simulations. In this limit, the electrons respond adiabatically to the electrostatic potential and do not develop the full non-Maxwellian structure allowed in the kinetic model. Comparing the Boltzmann and kinetic-electron cases therefore helps identify which features of the density, flow, and temperature profiles arise from ion kinetics alone and which depend on kinetic electron effects. Figure 6 shows that, for the low-recycling case, the density profile is underpredicted when adiabatic Boltzmann electrons are used. The trend suggests that an adiabatic electron response is not sufficient to capture the parallel force balance in the low-recycling SOL. Instead, the kinetic-electron result is consistent with a Pastukhov-type ambipolar potential [34] developing along the field line, together with the associated parallel electric field needed to reduce electron end loss and maintain quasineutrality. In the high-recycling case, where the kinetic and Boltzmann-electron profiles nearly overlap, those additional kinetic-electron effects appear to be much less important.

Changes with collisionality

The reference simulations show that in the low-collisionality limit, the SOL departs most strongly from standard fluid-like behavior. To examine that regime more closely, we next vary the collisionality numerically and track how the phase-space structure responds. We begin with the ion distribution function and the non-Maxwellian component of the ion distribution at the midplane, shown in Figure 8 and Figure 9, which show stronger non-Maxwellian structure in the low-recycling case. We then quantify the trapped fraction and connect it to the low-collisionality behavior of the SOL poloidal density profiles, poloidal density peaking, and the SOL particle and energy confinement times. The collisionality scan is used to identify how mirror trapping and kinetic effects emerge and become dynamically important in the weakly collisional low-recycling SOL.

Figure 8

Figure 8. Ion distribution function at the outboard midplane, θ=0\theta=0, for simulations of the low-recycling case (a) and high-recycling case (b). This is a local velocity-space slice, not an average along the field line. White dotted lines indicate the miv2/2=Ti(θ=0)m_iv^2/2=T_i(\theta=0) contour in order to compare to an equivalent Maxwellian distribution of the same temperature.

Figure 9

Figure 9. Non-Maxwellian component of the ion distribution function for the low-recycling case (a) and high-recycling case (b), at the outboard midplane (θ=0\theta=0). The dotted line marks the trapped–passing boundary defined by equation (17).

The distributions in Figure 8 and Figure 9 are sampled locally at the outboard midplane. Away from the midplane, parallel electric and mirror forces shift and reshape the distribution in v∥v_\parallel, so its departure from a Maxwellian varies along the field line. Spatial variation between mirror-trapped upstream distributions and skewed downstream distributions has also been demonstrated in other Gkeyll SOL simulations [57].

We define the collisionality used in the scan as the standard neoclassical collisionality [35].

νi∗=νii/ϵωbi=νiiq0R0ϵ3/2vti,\begin{equation} \nu_i^* = \frac{\nu_{ii}/\epsilon}{\omega_{bi}} = \frac{\nu_{ii} q_0 R_0}{\epsilon^{3/2} v_{ti}}, \tag*{(16)} \end{equation}

where νii\nu_{ii} is the ion-ion collision frequency, vtiv_{ti} is the ion thermal speed, and ωbi=ϵ vti/(q0R0)\omega_{bi}=\sqrt{\epsilon}\,v_{ti}/(q_0R_0) is the ion bounce frequency. Here ϵ=r0/R0\epsilon=r_0/R_0 is the inverse aspect ratio evaluated at the outboard midplane location of the simulated field line, R0R_0 is the major radius of that field line at the outboard midplane, r0=R0−Raxisr_0=R_0-R_{\rm axis} is the corresponding minor radius, and q0q_0 is the safety factor at r0r_0, from the kinetic reconstruction of the shot at that time point. In the collisionality scan, the collision frequencies are varied by a multiplicative factor νfrac\nu_{\mathrm{frac}} defined in equation (8), so that the plotted values of νi∗\nu_i^* correspond to the value νfracνi∗\nu_{\mathrm{frac}}\nu_i^*.

Figure 10

Figure 10. Trapped fraction as a function of ion collisionality for simulations of the low-recycling case (solid blue) and high-recycling case (dashed orange). The base cases, with unscaled collision frequencies, are indicated by stars.

μlc(v∥)=1Bmin 12msv∥2+qs(ϕ(θ=0)−ϕ(θmax⁡))Rm−1=msv∥2+2qs(ϕ(θ=0)−ϕ(θmax⁡))2Bmin(Rm−1)\begin{equation} \mu_{\mathrm{lc}}(v_{\parallel}) = \frac{1}{B_{min}}\,\frac{\frac{1}{2} m_s v_{\parallel}^2+q_s\left(\phi(\theta=0) - \phi(\theta_{\max})\right)}{R_m - 1}=\frac{m_s v_{\parallel}^2+2q_s\left(\phi(\theta=0) - \phi(\theta_{\max})\right)}{2 B_{min} (R_m - 1)} \tag*{(17)} \end{equation}

The trapped fraction is calculated from the midplane ion distribution function by using equation (17) to separate trapped and passing regions in (v∥,μ)(v_{\parallel},\mu) space. For each simulation in the collisionality scan, the magnetic-field magnitude along the field line is used to define Bmin=min⁡BB_{min}=\min B, Bmax=max⁡BB_{max}=\max B, and the mirror ratio Rm=Bmax/BminR_m=B_{max}/B_{min}. The distribution function is then sampled at the midplane, and all phase-space cells with μ>μlc(v∥)\mu>\mu_{\mathrm{lc}}(v_{\parallel}) are counted as trapped. The trapped fraction plotted in Figure 10 is the ratio of the distribution-function weight in this trapped region to the total distribution-function weight at the midplane. This diagnostic therefore measures how much of the local ion distribution occupies mirror-trapped phase space as the ion neoclassical collisionality is varied.

In the high-collisionality limit, as νi∗\nu_i^* approaches unity and beyond, collisions scatter ions across the trapped-passing boundary on a bounce time or less, progressively destroying the mirror-confined orbit structure. In that limit the distribution at the midplane relaxes toward an approximately Maxwellian shape. The higher-recycling case has a ion and electron collisionality of 13.7 and 3.4. Since Gkeyll is a full-ff solver it could be that the tail of the ion distribution function is moderately mirror trapped, but not the bulk, that might explain the modest poloidal potential the high-recycling case develops, Figure 7 and the modest trapped fraction seen in Figure 10. However, at lower collisionality, a significant fraction of the distribution function can end up in the trapped region, for the low recycling base case, ∼\sim two thirds of the distribution function is in the trapped region, as seen in Figure 10.

Figure 11

Figure 11. SOL plasma poloidal density profiles in simulations of the low-recycling case (a) and high-recycling case (b) with varying collisionality νi∗\nu_i^*, obtained by scaling the collision frequencies. (c) Poloidal density peaking as a function of collisionality measured as the ratio of the midplane to the sheath entrance ion density; cases with unscaled collision frequencies are indicated by the stars.

The source parameters in the 1x2v simulations are adjusted quantities: the electron source temperature, ion source temperature, and particle fueling rate are chosen so that the simulations reproduce the measured or estimated midplane density and temperatures for each recycling state. Because these source parameters differ between the low-recycling and high-recycling cases, and because simulations are on a single field line neglecting cross field transport, direct comparison of end-plate particle fluxes has limited physical relevance. In steady state, the integrated heat and particle fluxes to the sheath must balance the imposed particle and power sources, so they mostly reflect the chosen source strength rather than the intrinsic parallel transport response. A more useful diagnostic is therefore the effective SOL residence time: how much particle or thermal energy content is maintained in the field-line domain for a given source rate.

Figure 12
Figure 12

Figure 12. SOL particle (a) and energy (b) confinement times (normalized to the sound transit time) as a function of ion collisionality νi∗\nu_i^* for simulations of the low-recycling (blue) and high-recycling (orange) cases, for ions (solid, dashed) and electrons (dotted, dash-dot).

Figure 12 shows the SOL particle and energy confinement times inferred from this balance as a function of ion neoclassical collisionality. The particle confinement time is estimated from the ratio of the total number of particles in the SOL to the integrated particle source, while the energy confinement time is estimated from the ratio of the total thermal energy in the SOL to the imposed source power. Both are normalized to a simple sheath-limited transit time, τ∥0=Lz/(2cs)\tau_{\parallel 0}=L_z/(2c_s), where cs=(Te0+5Ti0/3)/mic_s=\sqrt{(T_{e0}+5T_{i0}/3)/m_i} is evaluated using the midplane temperatures. This normalization removes the trivial change in parallel sound speed between the hotter low-recycling case and the colder high-recycling case, allowing the remaining variation to be interpreted as a change in effective parallel confinement.

The main result in Figure 12 is that the normalized SOL confinement times increase strongly as collisionality is reduced, especially for the low-recycling case. This is consistent with the interpretation developed above: at low νi∗\nu_i^*, mirror trapping and kinetic phase-space structure allow particles and energy to remain in the SOL longer than expected from a simple sheath-limited fluid transit estimate. At higher collisionality, collisions more efficiently destroy the mirror-confined orbit structure and push the system back toward the more conventional fluid-like regime. Thus Figure 12 connects the phase-space evidence for trapping to a macroscopic transport consequence: enhanced parallel particle and energy residence times along the modeled field line in the low-recycling, weakly collisional SOL. We hypothesize that this enhanced residence time may contribute to the observed poloidal density peaking in the low-recycling LTX-β\beta SOL. A quantitative connection to cross-field transport is therefore left for future work.

Reduced upstream parallel electron heat flux relative to the local classical prediction has been obtained in kinetic SOL simulations without magnetic mirroring. In ALLA, preferential escape of higher-energy electrons depletes the upstream heat-carrying population [69], while in KIPP, nonlocal electron transport and the asymmetric loss of electrons through the sheath modify the conductive heat flux [70]. These effects can suppress the heat flux upstream and enhance it near the target relative to the local classical prediction. Both models include electrostatic regulation of electron losses. In the weakly collisional, low-recycling regime considered here, inclusion of the mirror force is expected to further inhibit parallel transport by retaining ions, with the ambipolar potential regulating electron losses to maintain quasineutrality. When the magnetic-field magnitude is held constant, removing the mirror force while retaining the same connection length and source parameters, the density profiles become substantially flatter, with peak densities approximately 3.7 and 2.7 times lower for the low and high-recycling cases, respectively. These comparisons support an additional contribution from magnetic trapping to plasma retention, beyond kinetic heat-flux suppression alone. Further details are given in § 8.

Discussion and Conclusion

We have used gyrokinetic simulations with one configuration-space dimension and two velocity-space dimensions, using Gkeyll to study the open-field-line SOL response associated with high and low-recycling LTX-β\beta conditions. The experimentally constrained low-recycling case is hotter and more weakly collisional than the high-recycling case, and develops a substantially larger electrostatic-potential variation along the modeled field line. The comparison with an adiabatic Boltzmann-electron closure of the field equations shows that the low-recycling density response cannot be recovered without kinetic electrons, indicating that the ambipolar potential and non-Maxwellian electron dynamics are important parts of the parallel force balance in this regime.

Collisionality scans further show that, as the ion neoclassical collisionality is reduced below unity, the ion distribution develops stronger mirror-trapped phase-space structure and the normalized particle and energy confinement times increase, most strongly for the low-recycling case. These results suggest that low-recycling lithium-wall operation in LTX-β\beta accesses a SOL regime in which weak collisionality, mirror trapping, and kinetic-electron effects increase the normalized parallel residence time relative to conventional high-recycling fluid expectations. We also document using experimental data an increase in the estimated SOL density fall-off length λn,SOL\lambda_{n,\mathrm{SOL}} as the low-recycling SOL warms.

The omission of neutral interactions and cross-field drifts is a limitation of the present field-line model. Across the relevant discharge ensembles, estimated SOL neutral densities of 11–5×1016 m−35\times 10^{16}\,\mathrm{m}^{-3} give charge-exchange lifetimes that increase from approximately 0.80.8 ms at higher recycling to 44 ms at low recycling, while curvature-induced guiding-center drift times are estimated at 1.61.6–3.93.9 ms [12]. The low-recycling charge-exchange lifetime is therefore more than an order of magnitude longer than the approximately 0.170.17 ms parallel particle residence time obtained for the low-recycling reference case. Together with the longer drift times, this supports the isolation of parallel kinetic dynamics as a first approximation. This ordering does not establish that all perpendicular transport is negligible: E×BE\times B transport and neutral interactions may affect the radial SOL structure and become more important at higher neutral densities or longer residence times.

Impurity interactions and radiation are also omitted. Impurity sources in LTX-β\beta depend strongly on lithium-conditioning history, with repeated lithium deposition associated with reduced impurity influx and improved recycling control compared with the initial transition from stainless-steel operation [26, 76]. Relatively weak radiative losses are therefore expected for the well-conditioned, low-recycling discharge ensembles motivating the present study, consistent with their previous interpretation [12]. Nevertheless, radiation from the remaining Li, C and O impurities is not included in the present simulations and could modify the quantitative energy balance and parallel transport. Multistep atomic processes can substantially reduce electron heat loads in cold, dense divertor plasmas [79]; however, these effects are expected to be less important in the hot, low-density SOL considered here. Non-Maxwellian electron distributions can also modify impurity ionization and radiation rates, including those of lithium [80]. The consequences for the energy balance depend on impurity concentration as well as the electron distribution. These radiation effects are therefore expected to be more important in high-recycling regimes, such as those in lithium rich vapor-box divertors, which deliberately use a dense lithium vapor for volumetric cooling [37], than in the hot, low-density SOL sought by low-recycling flowing-lithium concepts such as divertorlets, which limit evaporation by removing heat through liquid flow [36]. Radiation would introduce an additional energy-loss channel and could modify collisionality, the ambipolar potential and particle residence times through changes in the electron distribution. The present results therefore characterize parallel transport without radiative losses, rather than the complete SOL energy balance. A recently developed kinetic line-radiation operator in Gkeyll provides a route to assessing these effects self-consistently [81].

The experimentally observed broadening of the SOL density profile can be considered along with a simple particle-balance argument. Particles that diffuse radially across the SOL are also lost along the open field line to material surfaces. If radial transport is represented by an effective cross-field diffusivity D⊥D_\perp, and parallel loss is represented by an effective field-line residence time τ∥\tau_\parallel, a broader radial density profile is expected when particles either diffuse outward more rapidly or remain on the field line longer before being lost. For an exponential SOL density profile, this balance gives

λn∼D⊥ τ∥.\begin{equation} \lambda_n \sim \sqrt{D_\perp \, \tau_\parallel}. \tag*{(18)} \end{equation}

Using the λn,SOL\lambda_{n,\mathrm{SOL}} values from the two profile-fit time points shown in Figure 2 and the effective particle residence times from the simulations as shown in Figure 12 (90 μ\mus and 168 μ\mus), this scaling would imply an effective cross-field diffusivity that increases from D⊥∼3D_\perp \sim 3 m2^2/s when the gas puff is on and flooding the SOL with cold neutrals to D⊥∼25D_\perp \sim 25 m2^2/s after the gas puff terminates and the edge temperature increases. The observed broadening would require enhanced radial transport, increased parallel residence time, or some combination of the two. From the SOL particle-balance perspective, the puff-on phase behaves more like a cold, high-recycling boundary because of the neutral source, while after the puff terminates the SOL returns toward the warmer, lower-collisionality regime in which the simulations find longer normalized parallel residence times. The increase in the implied effective D⊥D_\perp points to perpendicular transport physics not captured in the present 1D model.

The stronger poloidal variation of the electrostatic potential also implies a radial E×BE\times B drift through its interaction with the toroidal magnetic field, providing an additional mechanism for redistribution between field lines. As parallel residence increases, perpendicular redistribution can become important before particles are lost to the limiters, modifying the profiles, trapped population, and loss balance. Comparable parallel-loss and perpendicular-transport times are already encountered in conventional SOL models [5, 82]. The absence of this self-consistent coupling is a central limitation of the present study, particularly when extrapolating toward increasingly low collisionality.

Since particle and energy losses are coupled in this low-recycling, weakly collisional regime, changes in the balance between parallel residence time and cross-field spreading will also affect the radial heat-flux width. This raises an important unanswered question about the nature of particle and energy transport in the low-recycling SOL. Self-consistent multidimensional kinetic simulations incorporating perpendicular drifts, neutral fueling, and wall pumping are therefore the highest-priority extension of this study. Together with experimental measurements of the SOL heat-flux width, these calculations are needed to establish how low-recycling operation modifies the distribution of heat loads on plasma-facing surfaces. We hope to explore some of these effects in future experiments, planned on ST40 with lithium walls [83].

Acknowledgements

We thank Akash Shukla for discussing oscillations in the constant BB case with us. Support for E. Perez was provided by the Department of Energy (DOE) Science Undergraduate Laboratory Internships (SULI) program. Funding for M. Francisquez, A. Hakim and G. W. Hammett was provided by the CEDA project as part of DOE’s Scientific Discovery through Advanced Computing (SciDAC) program, and the Distinguished Science Fellow program under contracts DE-AC02-09CH11466 and DE-FG02-04ER54742, as well as DOE’s Distinguished Scientist program and LDRD grants via DOE contract DE-AC02-09CH11466 for the Princeton Plasma Physics Laboratory. A. Maan, R. Majeski, and D. P. Boyle were supported by DE-AC02-09CH11466, and M. Dorf was supported by DE-AC52-07NA27344. This research used the Stellar cluster supported by Princeton Research Computing, a consortium of groups including the Princeton Institute for Computational Science and Engineering (PICSciE) and the Office of Information Technology’s High Performance Computing Center and Visualization Laboratory at Princeton University. The United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes.

Data Availability

The raw data is available upon request. All input files to reproduce it can be found in https://github.com/ammarhakim/gkyl-paper-inp/tree/master/2026_NF_LTX_1x2v. Other data used in the paper is hosted by Princeton Data Commons [59].

Getting Gkeyll  and reproducing results

Readers may reproduce our results and also use Gkeyll  for their applications. The code and input files used here are available online. Full installation instructions for Gkeyll  are provided on the Gkeyll  website (https://gkeyll.readthedocs.io). The input files used here are under version control and can be obtained from the repository at https://github.com/ammarhakim/gkyl-paper-inp/tree/master/2026_NF_LTX_1x2v.

Convergence studies

In this appendix we show data from convergence studies done to ensure that our findings are obtained with a suitable choice for final simulation time, number of cells, velocity grid extents, and k⊥ρsk_\perp\rho_s in kinetic electron simulations.

In Figure 13 we present time traces of the ion density, parallel drift speed and temperature, as well as the electron temperature at three key locations: the midplane (z=0z=0), half way between the midplane and the sheath entrance (z=zmax/2z=z_{\mathrm{max}}/2), and the sheath entrance (z=zmaxz=z_{\mathrm{max}}). We see that, for both the low and high RR simulations, these quantities are approximately at steady state by the time the simulation is terminated, at t=1t=1 ms. Although not shown here, we also tried running these much longer (t>3t>3 ms), but the results and analysis do not change. This also applies to the simulations with oscillations presented in § 8; specifically, running the low RR simulation with B=B0B=B_0 for much longer times does not make oscillations go away.

Figure 13

Figure 13. Time traces of the moments in simulations of low R (left column) and high R (right column) shots at three different locations: outboard midplane (solid), half-way to the sheath entrance (dashed) and sheath entrance (dotted).

We also ensured that the results presented in the body of this manuscript are representative of what would be obtained at much higher resolution. From the resolution convergence study presented in Figure 14 we draw the following conclusions. Obtaining reasonably converged T⊥iT_{\perp i} (and thus TiT_i) requires at least 12 cells along μ\mu. Likewise, producing a TeT_e close to that obtained at the highest v∥v_\parallel resolution requires Nv∥≥16N_{v_\parallel}\geq16. Lastly, according to Figure 14(a) and (e), the density is reasonably converged with 64 cells in zz. Hence, the resolution chosen for the reference simulations in § 4.1 is 64×16×1264\times16\times12 cells.

Figure 14

Figure 14. Ion number density (a,e), parallel speed (b,f) and temperature (c,g), and electron temperature (d,h), in simulations of low R (left column) and high R (right column) shots, using kinetic electrons and resolutions doubling or halving the number of cells in each direction.

A similar analysis was performed for the extents of the velocity-space grid. In Figure 15 we show that changing the velocity space limits so that the maximum v∥v_\parallel or v⊥v_\perp is between 4vts04v_{ts0} and 9vts09v_{ts0} changes the results relatively little. The biggest change is seen when the v⊥,maxv_{\perp,\mathrm{max}} is too low, as it is for the purple curve. If v⊥,maxv_{\perp,\mathrm{max}} is too low, the densities come out too high, and the temperatures too low. By v⊥,max=6vts0v_{\perp,\mathrm{max}}=6v_{ts0} the simulation is very close to converged.

Figure 15

Figure 15. Ion number density (a,e), parallel speed (b,f) and temperature (c,g), and electron temperature (d,h), in simulations of low R (left column) and high R (right column) shots, using kinetic electrons and varying the velocity extents in v∥v_\parallel and μ∝v⊥2\mu\propto v_\perp^2.

Lastly, we show what happens if smaller and larger values of the k⊥k_\perp parameter in the kinetic electron field model (see equation (11)) are used. Figure 16 presents the final moments of ions and electrons for both low and high RR simulations as a function of the k⊥k_\perp parameter. In general the moments are weakly sensitive to k⊥ρsk_\perp\rho_s, provided that a sufficiently low value is chosen. The reference simulations in § 4.1 used k⊥ρs=0.15k_\perp\rho_s=0.15, for which the moment profiles are relatively close to their asymptotic value at k⊥ρs=0.025k_\perp\rho_s=0.025. Only the electron temperature at high RR (Figure 16(h)) appears to be more sensitive to this free parameter, but note that there is a suppressed zero in that plot, causing the differences to appear larger than they really are.

Figure 16

Figure 16. Moments in simulations of low R (left column) and high R (right column) cases as a function of the k⊥ρsk_\perp\rho_s parameter.

Simulations without mirror force

The mirror force (∝∂z(μB)\propto\partial_z(\mu B)) plays an important role in the formation of edge profiles in LTX-β\beta . The low collisionality (due to high temperatures and low densities) and considerable mirror ratio (∼\sim 4) give the mirror force term in equation (5) greater importance than in the edge of high-recycling scenarios. This effect is typically excluded from edge fluid simulations with codes such as UEDGE [45], SOLPS [46], EDGE2D [47], SOLEDGE [51], BOUT++ [50], GBS [48], GRILLIX [49] and TOKAM3X [52]. Using fluid models that neglect the mirror force would produce very different characterizations or predictions of the SOL in these low collisionality, low-recycling regimes. In order to demonstrate what would happen were one to neglect the mirror force, in this appendix we show simulations with a constant magnetic field amplitude B=B0=constB=B_0=\mathrm{const}. Unless stated otherwise, the connection length, the source parameters, and other inputs are kept the same as they were for the B=B(z)B=B(z) simulations.

Two major changes are observed when the magnetic field amplitude is set to a constant: the density profiles are less peaked, and oscillations arise near the sheath entrance in the low-recycling simulation. Both of these can be appreciated in the space-time evolution of the ion density given in Figure 17. The steady state peak ion density (ni(θ=0,t≈1ms)n_i(\theta=0,t\approx1\mathrm{ms})) is much lower when using constant magnetic fields for both low and high RR. Additionally, in Figure 17(b) we can appreciate oscillations that persist for t>0.3t>0.3 ms.

Figure 17
Figure 17

Figure 17. Ion density as a function of space and time for low (left) and high (right) RR simulations, and for B=B(z)B=B(z) (top) and B=B0B=B_0 (bottom).

A density flattening is more clearly appreciated in the time-averaged, steady state profiles given in Figure 18(a) and (e). We see that the constant BB equilibrium profiles have peak densities lower by factors of approximately 3.7 and 2.7 for the low and high RR scenarios, respectively, compared to their inhomogeneous BB counterparts. The stronger flattening at low RR is attributed to the loss of trapped (banana) particle orbits, which is particularly important for low RR, as exemplified by the larger trapped particle fraction quantified in Figure 10. The lack of mirror trapping is also evident in the reduced off-midplane peaks in TiT_i, which are associated with the increase in T⊥iT_{\perp i} as a result of conservation of μ\mu in a varying magnetic field. Since temperature is a local moment of the distribution function, this mirror-induced redistribution and the spatial variation of the contributing particle populations can modify the temperature profile without violating single-particle energy conservation. Such peaks do not exist in TeT_e because, although subject to the same μ\mu-conservation that causes TiT_i to increase off mid-plane, the electrons have much higher energy (180 eV) at the midplane and manage to escape mirror trapping and transport their energy out of the domain, thus causing the temperature to be peaked in the center.

Figure 18

Figure 18. (a) Ion number density, (b) parallel speed, (c) ion temperature, and (d) electron temperature for the low-recycling case averaged over t∈[0.8 ms, 1 ms]t\in\left[0.8~\mathrm{ms},~1~\mathrm{ms}\right]. Panels (e)–(h) show the same quantities for the high-recycling case. The B=B(z)B=B(z) results are shown in solid lines, while the B=B0B=B_0 results are shown in dashed lines.

As mentioned in this section, the low RR simulation with B=B0B=B_0 develops oscillations near the boundaries. These oscillations propagate both downstream and upstream, and produce quasiperiodic patterns in the potential, the moments and the distribution function. We performed a very extensive study of these oscillations (over several years) in order to determine the origin of these oscillations or find what algorithms or physics could stabilize them. We experimented with the following:

  • Decreasing time step Δt\Delta t.

  • Considerably increasing the resolution in all dimensions.

  • Changing the extents of the grid in v∥v_\parallel and μ\mu.

  • Altering the initial conditions (ICs); e.g. we tried a Gaussian ns(t=0)n_s(t=0) instead of Figure 5(a), or non-Maxwellian ICs.

  • Modifying the source shape or strength.

  • Increasing or decreasing the connection length.

  • Scanning the electron-ion temperature ratio (oscillations are stabilized for small Te0/Ti0T_{e0}/T_{i0}).

  • Scanning the collision frequency (oscillations are stabilized at large νss\nu_{ss}).

  • Several algorithms intended to enforce the Bohm sheath criterion on u∥iu_{\parallel i}.

  • Different ways of constructing the numerical flux and discretizing the geometric factors arising in Gkeyll’s field-aligned coordinate system.

None of the above produced a convincing story that would portray these oscillations as either numerical or physical.

In order to simplify the system and attempt to understand the origin of these oscillations in the constant BB case, we switched to using the Boltzmann electron model (see § 3.2). The Boltzmann electron model also removes the conducting sheath BC [55] in favor of an absorbing BC for ions. Even with Boltzmann electrons, the oscillations emerge at the same location (Figure 19(b)), although they have a different amplitude and are relatively constant in time.

Figure 19

Figure 19. Ion density in low RR simulations with B=B0B=B_0 and kinetic (a) or Boltzmann (b) electrons, as a function of time and space. The first 0.5 ms are omitted to avoid the contrast caused by transients that ICs induce.

Notice from Figure 18 that the low RR simulation with B=B0B=B_0 arrives at equilibrium profiles with very different values. This observation begs the question of whether the oscillations are produced by an instability that only arises in this very different parameter regime. In order to rule out that possibility, we adjusted the density source rate and the source temperature so that the B=B0B=B_0, Boltzmann electron simulation culminates in profiles in the same range as those in the B=B(z)B=B(z), kinetic electron simulation. The result is given in Figure 20. We see now that despite having a density and temperature close to the more realistic simulation, the B=B0B=B_0 simulation with Boltzmann electrons still develops oscillations near the boundaries. For reference we also ran this case with the COGENT gyrokinetic code [58], essentially solving the same equation with the same setup. The result is also shown in the top panel of Figure 20, where we see that for both the first-order (UW1) and third-order (UW3) upwind discretizations COGENT did not develop oscillations near the boundaries. This suggested that either COGENT had enough numerical dissipation (e.g. from upwinding) to suppress these (possibly physical) oscillations, or that they are a numerical artifact produced by Gkeyll’s numerics.

Figure 20

Figure 20. For Gkeyll ’s standard algorithm, the ion density (top), parallel drift speed normalized to cs=(T∥e+3T∥i)/mic_s=\sqrt{(T_{\parallel e}+3T_{\parallel i})/m_i} (middle) and temperature (bottom) in low RR simulations with B=B(z)B=B(z) and kinetic electrons (solid blue), and B=B0B=B_0 and Boltzmann electrons after adjusting the sources so that ni(z=0)n_i(z=0) roughly matches the peak density of the case with inhomogeneous BB and kinetic electrons (dashed orange). The blue line consists of the final steady state, while the orange line is the result of a 1 ms time average.

For lack of time to dig into COGENT’s numerics we explored if Gkeyll’s algorithm could be causing these oscillations. The short answer appears to be yes. Explaining how Gkeyll’s numerics can cause such oscillations, and how to remedy them, in detail will have to be postponed to a future manuscript, but we provide a high-level description here, and a demonstration that an alternative algorithm is more stable. In simple terms, the 1D1V limit of the Cartesian, constant BB gyrokinetic equation (neglecting collisions and sources) reads:

∂fs∂t+∂↑v∥∂z(v∥fs)+∂↑a∥∂v∥(a∥fs)=0,a∥=−qsms∂ϕ∂z.\begin{equation} \frac{\partial f_s}{\partial t} + \frac{\partial ^{\uparrow v_\parallel}}{\partial z}\left(v_\parallel f_s\right) + \frac{\partial ^{\uparrow a_{\parallel}}}{\partial v_\parallel}\left(a_{\parallel} f_s\right) = 0, \qquad a_\parallel = -\frac{q_s}{m_s}\frac{\partial \phi}{\partial z}. \tag*{(19)} \end{equation}

In this equation we have added a superscript to the derivatives to symbolize that these are upwinded derivatives. A weak upwinded derivative of a scalar AA, for example ∂z↑v∥A\partial_z^{\uparrow v_\parallel}A, follows from projecting this derivative onto the kk-th finite element or DG basis ψjk\psi^{k}_j in cell jj, using integration by parts

∫zj−1/2zj+1/2dz ψjk ∂z↑v∥A=ψjk A^∣zj−1/2zj+1/2−∫zj−1/2zj+1/2dz A ∂zψjk,\begin{equation} \int_{z_{j-1/2}}^{z^{j+1/2}}\mathrm{d}z\,\psi^{k}_j\,\partial_z^{\uparrow v_\parallel}A = \psi^{k}_j\,\widehat{A}\Big|_{z_{j-1/2}}^{z^{j+1/2}} - \int_{z_{j-1/2}}^{z^{j+1/2}}\mathrm{d}z\,A\,\partial_z\psi^{k}_j, \tag*{(20)} \end{equation}

and the numerically upwinded AA:

A^j−1/2=12(Aj−1/2−+Aj−1/2+)−sign(v∥)2(Aj−1/2+−Aj−1/2−),\begin{equation} \widehat{A}_{j-1/2} = \frac{1}{2}\left(A^-_{j-1/2}+A^+_{j-1/2}\right)-\frac{\mathrm{sign}\left(v_\parallel\right)}{2}\left(A^+_{j-1/2}-A^-_{j-1/2}\right), \tag*{(21)} \end{equation}

where −-/++ superscripts indicate the left/right value at a cell boundary. The v∥v_\parallel derivative in equation (19) is upwinded similarly, using the sign of a∥a_\parallel. The zz derivative in a∥∝∂zϕa_{\parallel}\propto\partial_z\phi is not upwinded; it is simply computed locally from the intracell high-order representation of ϕ\phi. This is, in a nutshell, the algorithm in Gkeyll. The problem is that if one works out the discrete dispersion relation for this system, assuming that ϕ\phi is computed as a continuous finite element scalar [31], an oscillatory mode can arise for subsonic parameter regimes.

One way to avoid this (we tried a couple) is to ensure that the derivative used to compute the electric field is the matched adjoint of the derivative in the position-space advection term. That is, to define a downwinded derivative ∂z↓v∥(⋅)\partial_z^{\downarrow v_\parallel}\left(\cdot\right) such that

∫dz dv∥ ∂z↑v∥A=−∫dz dv∥ ∂z↓v∥A.\begin{equation} \int\mathrm{d}z\,\mathrm{d}v_\parallel\,\partial_z^{\uparrow v_\parallel}A = - \int\mathrm{d}z\,\mathrm{d}v_\parallel\,\partial_z^{\downarrow v_\parallel}A. \tag*{(22)} \end{equation}

More explicitly, by computing the electric field using

∫zj−1/2zj+1/2dz ψjk ∂z↓v∥ϕ=ψjk ϕˇ∣zj−1/2zj+1/2−∫zj−1/2zj+1/2dz ϕ ∂zψjk,ϕˇj−1/2=12(ϕj−1/2−+ϕj−1/2+)+sign(v∥)2(ϕj−1/2+−ϕj−1/2−).\begin{equation} \begin{aligned} \int_{z_{j-1/2}}^{z^{j+1/2}}\mathrm{d}z\,\psi^{k}_j\,\partial_z^{\downarrow v_\parallel}\phi &= \psi^{k}_j\,\check{\phi}\Big|_{z_{j-1/2}}^{z^{j+1/2}} - \int_{z_{j-1/2}}^{z^{j+1/2}}\mathrm{d}z\,\phi\,\partial_z\psi^{k}_j, \\ % \check{\phi}_{j-1/2} &= \frac{1}{2}\left(\phi^-_{j-1/2}+\phi^+_{j-1/2}\right)+\frac{\mathrm{sign}\left(v_\parallel\right)}{2}\left(\phi^+_{j-1/2}-\phi^-_{j-1/2}\right). \end{aligned} \tag*{(23)} \end{equation}

For this to have an impact, ϕ\phi cannot be represented as a continuous finite element quantity (as has historically been done in Gkeyll). We implemented this new algorithm in Gkeyll and found that the oscillations are indeed stabilized. For example, in Figure 21 we once again show the ion moments for the B(z)B(z) kinetic electron and B=B=const. Boltzmann electron Gkeyll simulations using this new matched adjoint algorithm, compared to the COGENT result. The Gkeyll and COGENT results now lie very close to each other. Additional tests of this new algorithm, with kinetic electrons and in higher dimensions, are deferred to a separate manuscript.

Figure 21

Figure 21. For Gkeyll ’s new matched adjoint algorithm, the ion density (top), parallel drift speed normalized to cs=(T∥e+3T∥i)/mic_s=\sqrt{(T_{\parallel e}+3T_{\parallel i})/m_i} (middle) and temperature (bottom) in low RR simulations with B=B(z)B=B(z) and kinetic electrons (solid blue), and B=B0B=B_0 and Boltzmann electrons after adjusting the sources so that ni(z=0)n_i(z=0) roughly matches the peak density of the case with inhomogeneous BB and kinetic electrons (dashed orange). The blue line consists of the final steady state, while the orange line is the result of a 1 ms time average.

References

References

  1. [1]Majeski, R., Bell, R. E., Boyle, D. P., Kaita, R., Kozub, T., LeBlanc, B. P., Lucia, M., Maingi, R., Merino, E., Raitses, Y., Schmitt, J. C., Allain, J. P., Bedoya, F., Bialek, J., Biewer, T. M., Canik, J. M., Buzi, L., Koel, B. E., Patino, M. I., … Tritz, K. (2017). Compatibility of lithium plasma-facing surfaces with high edge temperatures in the Lithium Tokamak Experiment. Physics of Plasmas, 24(5), 056110. https://doi.org/10.1063/1.4977916
  2. [2]Beer, M. A., Cowley, S. C., & Hammett, G. W. (1995). Field--aligned coordinates for nonlinear simulations of tokamak turbulence. Physics of Plasmas, 2(7), 2687–2700. https://doi.org/10.1063/1.871232
  3. [3]Menard, J. E. (2019). Compact steady-state tokamak performance dependence on magnet and core physics limits. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 377(2141), 20170440. https://doi.org/10.1098/rsta.2017.0440
  4. [4]Menard, J. E., Grierson, B. A., Brown, T., Rana, C., Zhai, Y., Poli, F. M., Maingi, R., Guttenfelder, W., & Snyder, P. B. (2022). Fusion pilot plant performance and the role of a sustained high power density tokamak. Nuclear Fusion, 62(3), 036026. https://doi.org/10.1088/1741-4326/ac49aa
  5. [5]Goldston, R. J. (2012). Heuristic drift-based model of the power scrape-off width in low-gas-puff H-mode tokamaks. Nuclear Fusion, 52(1), 013009. https://doi.org/10.1088/0029-5515/52/1/013009
  6. [6]Eich, T., Leonard, A. W., Pitts, R. A., Fundamenski, W., Goldston, R. J., Gray, T. K., Herrmann, A., Kirk, A., Kallenbach, A., Kardaun, O., Kukushkin, A. S., Labombard, B., Maingi, R., Makowski, M. A., Scarabosio, A., Sieglin, B., Terry, J., & Thornton, A. (2013). Scaling of the tokamak near the scrape-off layer H-mode power width and implications for ITER. Nuclear Fusion, 53(9), 093031. https://doi.org/10.1088/0029-5515/53/9/093031
  7. [7]Kuang, A. Q., Ballinger, S., Brunner, D., Canik, J., Creely, A. J., Gray, T., Greenwald, M., Hughes, J. W., Irby, J., Labombard, B., Lipschultz, B., Lore, J. D., Reinke, M. L., Terry, J. L., Umansky, M., Whyte, D. G., & Wukitch, S. (2020). Divertor heat flux challenge and mitigation in SPARC. Journal of Plasma Physics, 86(5), 865860505. https://doi.org/10.1017/S0022377820001117
  8. [8]Yang, Q., You, Y.-W., Liu, L., Fan, H., Ni, W., Liu, D., Liu, C. S., Benstetter, G., & Wang, Y. (2015). Nanostructured fuzz growth on tungsten under low-energy and high-flux He irradiation. Scientific Reports, 5, 10959. https://doi.org/10.1038/srep10959
  9. [9]Wright, J. A. R. (2022). A review of late-stage tungsten fuzz growth. Tungsten, 4, 184–193. https://doi.org/10.1007/s42864-021-00133-2
  10. [10]Matthews, G. F., & JET EFDA Contributors. (2013). Plasma operation with an all metal first-wall: Comparison of an ITER-like wall with a carbon wall in JET. Journal of Nuclear Materials, 438, S2–S10. https://doi.org/10.1016/j.jnucmat.2013.01.282
  11. [11]Pütterich, T., & JET Contributors. (2012). Tungsten screening and impurity control in JET. In Poster presented at the 24th IAEA Fusion Energy Conference, San Diego, CA. https://scipub.euro-fusion.org/archives/jet-archive/tungsten-screening-and-impurity-control-in-jetscipub.euro-fusion.org/archives/jet-archive/tungsten-screening-and-impurity-control-in-jet
  12. [12]Maan, A., Boyle, D. P., Majeski, R., Wilkie, G. J., Francisquez, M., Banerjee, S., Kaita, R., Maingi, R., LeBlanc, B. P., Abe, S., Jung, E., Perez, E., Capecchi, W., Ostrowski, E. T., Elliott, D. B., Hansen, C., Kubota, S., Soukhanovskii, V., & Zakharov, L. (2024). Estimates of global recycling coefficients for LTX-$ $ discharges. Physics of Plasmas, 31(2), 022505. https://doi.org/10.1063/5.0177604
  13. [13]Baldwin, M. J., Doerner, R. P., Luckhardt, S. C., & Conn, R. W. (2002). Deuterium retention in liquid lithium. Nuclear Fusion, 42(11), 1318–1323. https://doi.org/10.1088/0029-5515/42/11/305
  14. [14]Krstic, P. S., Maan, A., Majeski, R., & Koel, B. E. (2025). Effects of size of an oxide layer on the surface properties of lithium irradiated by deuterium plasma. Nuclear Fusion, 65(7), 074001. https://doi.org/10.1088/1741-4326/addad6
  15. [15]Groth, M., Allen, S. L., Rensink, M. E., & Hill, D. N. (1989). Recycling and neutral transport in the DIII-D tokamak. Journal of Nuclear Materials, 162--164, 80–92. https://doi.org/10.1016/0022-3115(89)90259-6
  16. [16]Wang, H. Q., Hong, R. J., Groebner, R., Jian, X., Rhodes, T., Leonard, A. W., Ma, X., Mordijck, S., Wilks, T., Yan, Z., Shafer, M. W., Scotti, F., Truong, D., Ren, J., Laggner, F., Grierson, B. A., Osborne, T. H., Thomas, D. M., & Watkins, J. G. (2024). Manipulating density pedestal structure to improve core-edge integration towards low collisionality. Nuclear Fusion, 64(12), 126061. https://doi.org/10.1088/1741-4326/ad8585
  17. [17]Kaita, R. (2019). Fusion applications for lithium: wall conditioning in magnetic confinement devices. Plasma Physics and Controlled Fusion, 61(11), 113001. https://doi.org/10.1088/1361-6587/ab4156
  18. [18]Princeton Center for Heliophysics. (2026). Gkeyll 2.0 Documentation Home Page. In Accessed 2026-03-31. https://gkeyll.readthedocs.io/en/latest/gkeyll.readthedocs.io/en/latest
  19. [19]Banerjee, S., Boyle, D. P., Maan, A., Ferraro, N., Wilkie, G., Majeski, R., Podesta, M., Bell, R., Hansen, C., Capecchi, W., & Elliott, D. (2024). Investigating the role of edge neutrals in exciting tearing mode activity and achieving flat temperature profiles in LTX-$ $. Nuclear Fusion, 64(4), 046026. https://doi.org/10.1088/1741-4326/ad2ca7
  20. [20]Krstic, P. S., Ostrowski, E. T., Dwivedi, S., Abe, S., Maan, A., van Duin, A. C. T., & Koel, B. E. (2023). Detailed studies of the processes in low energy H irradiation of Li and Li-compound surfaces. Journal of Applied Physics, 134(10), 100902. https://doi.org/10.1063/5.0149502
  21. [21]Elliott, D., Anderson, J., Bell, R. E., Biewer, T. M., Boyle, D. P., Donovan, D., Hansen, C., Hughes, P., Kaita, R., Koel, B., Kubota, S., Lunsford, R., Maan, A., Majeski, R., Scotti, F., Soukhanovskii, V., & Zakharov, L. (2020). Initial Results from the Newly Upgraded LTX-$ $. IEEE Transactions on Plasma Science, 48(6), 1382–1387. https://doi.org/10.1109/TPS.2020.2983854
  22. [22]Maan, A., Kaita, R., Ostrowski, E. T., Majeski, R., Boyle, D. P., Donovan, D. C., Ellis, R. A., Koel, B. E., & Biewer, T. M. (2020). A simple vacuum suitcase for enabling plasma facing component characterization in fusion devices. Review of Scientific Instruments, 91(2), 026104. https://doi.org/10.1063/1.5119166
  23. [23]Boyle, D. P., Anderson, J., Banerjee, S., Bell, R., Capecchi, W., Elliott, D., Hansen, C., Kubota, S., LeBlanc, B., Maan, A., Maingi, R., Majeski, R., Menard, J., Oliva, S., Rhodes, T., Soukhanovskii, V., & Zakharov, L. (2023). Extending the low-recycling, flat temperature profile regime in the Lithium Tokamak Experiment-$ $ (LTX-$ $) with ohmic and neutral beam heating. Nuclear Fusion, 63(5), 056020. https://doi.org/10.1088/1741-4326/acc4da
  24. [24]Maan, A., Boyle, D. P., Majeski, R., Banerjee, S., Francisquez, M., Kaita, R., Wilkie, G. J., Capecchi, W., Kubota, S., Hansen, C., & Soukhanovskii, V. (2023). Improved neutral and plasma density control with increasing lithium wall coatings in the Lithium Tokamak Experiment-$ $ (LTX-$ $). Nuclear Materials and Energy, 35, 101408. https://doi.org/10.1016/j.nme.2023.101408
  25. [25]Maan, A., Ostrowski, E., Kaita, R., Donovan, D., Majeski, R., Boyle, D., Hughes, P., Merino, E., Kozub, T., Koel, B. E., Elliott, D., Biewer, T., Scotti, F., Soukhanovskii, V., & Lunsford, R. (2020). Plasma Facing Component Characterization and Correlation with Plasma Conditions in Lithium Tokamak Experiment-$ $. IEEE Transactions on Plasma Science, 48(6), 1463–1467. https://doi.org/10.1109/TPS.2020.2969115
  26. [26]Maan, A., Boyle, D. P., Kaita, R., Ostrowski, E. T., Donovan, D. C., Majeski, R. P., Koel, B. E., Biewer, T. M., Hughes, P. E., Hansen, C., Kubota, S., & Soukhanovskii, V. (2021). Oxidation of lithium plasma facing components and its effect on plasma performance in the lithium tokamak experiment-$ $. Plasma Physics and Controlled Fusion, 63(2), 025007. https://doi.org/10.1088/1361-6587/abcd0f
  27. [27]Miller, R. L., Chu, M. S., Greene, J. M., Lin-Liu, Y. R., & Waltz, R. E. (1998). Noncircular, finite aspect ratio, local equilibrium model. Physics of Plasmas, 5(4), 973–978. https://doi.org/10.1063/1.872666
  28. [28]Hansen, C., Boyle, D. P., Schmitt, J. C., & Majeski, R. (2017). Equilibrium reconstruction with 3D eddy currents in the Lithium Tokamak eXperiment. Physics of Plasmas, 24(4), 042513. https://doi.org/10.1063/1.4981214
  29. [29]Shi, E. L., Hammett, G. W., Stoltzfus-Dueck, T., & Hakim, A. (2017). Gyrokinetic continuum simulation of turbulence in a straight open-field-line plasma. Journal of Plasma Physics, 83(3), 905830304. https://doi.org/10.1017/S002237781700037X
  30. [30]Shi, E. L., Hakim, A. H., & Hammett, G. W. (2015). A gyrokinetic one-dimensional scrape-off layer model of an edge-localized mode heat pulse. Physics of Plasmas, 22(2), 022504. https://doi.org/10.1063/1.4907160
  31. [31]Francisquez, M., Cagas, P., Shukla, A., Juno, J., & Hammett, G. W. (2026). Conservative velocity mappings for discontinuous Galerkin kinetics. Journal of Computational Physics, 558, 114852. https://doi.org/10.1016/j.jcp.2026.114852
  32. [32]Stangeby, P. C. (2000). The Plasma Boundary of Magnetic Fusion Devices. Institute of Physics Publishing.
  33. [33]Ochs, I. E., Munirov, V. R., & Fisch, N. J. (2023). Confinement time and ambipolar potential in a relativistic mirror-confined plasma. Physics of Plasmas, 30(5), 052508. https://doi.org/10.1063/5.0147466
  34. [34]Pastukhov, V. P. (1974). Collisional losses of electrons from an adiabatic trap in a plasma with a positive potential. Nuclear Fusion, 14(1), 3–6. https://doi.org/10.1088/0029-5515/14/1/001
  35. [35]Helander, P., & Sigmar, D. J. (2002). Collisional Transport in Magnetized Plasmas. Cambridge University Press. https://doi.org/10.1017/CBO9780511493558
  36. [36]Fisher, A. E., Sun, Z., & Kolemen, E. (2020). Liquid metal ``divertorlets'' concept for fusion reactors. Nuclear Materials and Energy, 25, 100855. https://doi.org/10.1016/j.nme.2020.100855
  37. [37]Goldston, R. J., Myers, R., & Schwartz, J. (2016). The lithium vapor box divertor. Physica Scripta, 2016(T167), 014017. https://doi.org/10.1088/0031-8949/T167/1/014017
  38. [38]Zakharov, L. E., Gorelenkov, N. N., White, R. B., Krasheninnikov, S. I., & Pereverzev, G. V. (2004). Ignited spherical tokamaks and plasma regimes with LiWalls. Fusion Engineering and Design, 72(1--3), 149–168. https://doi.org/10.1016/j.fusengdes.2004.07.015
  39. [39]Catto, P. J., & Hazeltine, R. D. (2006). Isothermal tokamak. Physics of Plasmas, 13(12), 122508. https://doi.org/10.1063/1.2403090
  40. [40]Smirnov, R. D., Pigarov, A. Y., Krasheninnikov, S. I., Rognlien, T. D., Soukhanovskii, V. A., Rensink, M. E., Maingi, R., Skinner, H. C., Stotler, D. P., Bell, R. E., & Kugel, H. W. (2010). Multi-fluid modeling of low-recycling divertor regimes. Contributions to Plasma Physics, 50(3--5), 299–305. https://doi.org/10.1002/ctpp.201010049
  41. [41]Kaur, M., Boyle, D., Elliott, D., Maan, A., & Majeski, R. (2026). Upgrades to the Thomson scattering diagnostic on the Lithium Tokamak eXperiment-beta (LTX-$ $). In Accepted for publication in IEEE Transactions on Plasma Science, Special Issue - Selected Papers from SOFE 2025, manuscript TPS17994.R1.
  42. [42]Capece, A. M., Patino, M. I., Raitses, Y., & Koel, B. E. (2016). Secondary electron emission from lithium and lithium compounds. Applied Physics Letters, 109(1), 011605. https://doi.org/10.1063/1.4955461
  43. [43]Geyko, V. I., Joseph, I., Dorf, M. A., Ghosh, D., & Dorr, M. R. (2024). Drift kinetic electrostatic simulations of the edge localized mode heat pulse. Physics of Plasmas, 31(12), 123903. https://doi.org/10.1063/5.0230913
  44. [44]Francisquez, M., Rosen, M. H., Mandell, N. R., Hakim, A., Forest, C. B., & Hammett, G. W. (2023). Toward continuum gyrokinetic study of high-field mirrors. Physics of Plasmas, 30(10), 102504. https://doi.org/10.1063/5.0152440
  45. [45]Rognlien, T. D., Milovich, J. L., Rensink, M. E., & Porter, G. D. (1992). A fully implicit, time dependent 2-D fluid code for modeling tokamak edge plasmas. Journal of Nuclear Materials, 196-198, 347–351. https://doi.org/10.1016/S0022-3115(06)80058-9
  46. [46]Wiesen, S., Reiter, D., Kotov, V., Baelmans, M., Dekeyser, W., Kukushkin, A. S., Lisgo, S. W., Pitts, R. A., Rozhansky, V., Saibene, G., Veselova, I., & Voskoboynikov, S. (2015). The new SOLPS-ITER code package. Journal of Nuclear Materials, 463, 480–484. https://doi.org/10.1016/j.jnucmat.2014.10.012
  47. [47]Simonini, R., Corrigan, G., Radford, G., Spence, J., & Taroni, A. (1994). Models and Numerics in the Multi-Fluid 2-D Edge Plasma Code EDGE2D/U. Contributions to Plasma Physics, 34(2-3), 368–373. https://doi.org/10.1002/ctpp.2150340242
  48. [48]Halpern, F. D., Ricci, P., Jolliet, S., Loizu, J., Morales, J., Mosetto, A., Musil, F., Riva, F., Tran, T. M., & Wersal, C. (2016). The GBS code for tokamak scrape-off layer simulations. Journal of Computational Physics, 315, 388–408. https://doi.org/10.1016/j.jcp.2016.03.040
  49. [49]Stegmeir, A., Coster, D., Ross, A., Maj, O., Lackner, K., & Poli, E. (2018). GRILLIX: a 3D turbulence code based on the flux-coordinate independent approach. Plasma Physics and Controlled Fusion, 60(3), 035005. https://doi.org/10.1088/1361-6587/aaa373
  50. [50]Dudson, B. D., Allen, A., Breyiannis, G., Brugger, E., Buchanan, J., Easy, L., Farley, S., Joseph, I., Kim, M., McGann, A. D., et al. (2015). BOUT++: Recent and current developments. Journal of Plasma Physics, 81(1), 365810104. https://doi.org/10.1017/S0022377814000816
  51. [51]Bufferand, H., Bucalossi, J., Ciraolo, G., Falchetto, G., Gallo, A., Ghendrih, P., Rivals, N., Tamain, P., Yang, H., Giorgiani, G., Schwander, F., Scotto d’Abusco, M., Serre, E., Marandet, Y., Raghunathan, M., WEST Team, & the JET Team. (2021). Progress in edge plasma turbulence modelling—hierarchy of models from 2D transport application to 3D fluid simulations in realistic tokamak geometry. Nuclear Fusion, 61(11), 116052. https://doi.org/10.1088/1741-4326/ac2873
  52. [52]Tamain, P., Bufferand, H., Ciraolo, G., Colin, C., Galassi, D., Ghendrih, P., Schwander, F., & Serre, E. (2016). The TOKAM3X code for edge turbulence fluid simulations of tokamak plasmas in versatile magnetic geometries. Journal of Computational Physics, 321, 606–623. https://doi.org/10.1016/j.jcp.2016.05.038
  53. [53]Francisquez, M., Bernard, T. N., Mandell, N. R., Hammett, G. W., & Hakim, A. (2020). Conservative discontinuous Galerkin scheme of a gyro-averaged Dougherty collision operator. Nuclear Fusion, 60(9), 096021. https://doi.org/10.1088/1741-4326/aba0c9
  54. [54]Francisquez, M., Juno, J., Hakim, A., Hammett, G. W., & Ernst, D. R. (2022). Improved multispecies Dougherty collisions. Journal of Plasma Physics, 88(3), 905880303. https://doi.org/10.1017/S0022377822000289
  55. [55]Shi, E. L. (2017). Gyrokinetic Continuum Simulation of Turbulence in Open-Field-Line Plasmas. In Princeton University.
  56. [56]Huba, J. D. (2019). NRL Plasma Formulary. https://library.psfc.mit.edu/catalog/online_pubs/NRL_FORMULARY_19.pdflibrary.psfc.mit.edu/catalog/online_pubs/NRL_FORMULARY_19.pdf
  57. [57]Shukla, A., Roeltgen, J., Kotschenreuther, M., Juno, J., Bernard, T. N., Hakim, A., Hammett, G. W., Hatch, D. R., Mahajan, S. M., & Francisquez, M. (2025). Direct comparison of gyrokinetic and fluid scrape-off layer simulations. AIP Advances, 15(7), 075121. https://doi.org/10.1063/5.0268104
  58. [58]Dorf, M. A., Dorr, M. R., Hittinger, J. A., Cohen, R. H., & Rognlien, T. D. (2016). Continuum kinetic modeling of the tokamak plasma edge. Physics of Plasmas, 23(5), 056102. https://doi.org/10.1063/1.4943106
  59. [59]Maan, A. (2026). Data for "One Dimensional Gyrokinetic Study of a Low-Recycling Scrape-Off Layer". https://doi.org/10.34770/q55d-yk29x
  60. [60]Homma, Y., Hoshino, K., Yamoto, S., Tokunaga, S., Asakura, N., Sakamoto, Y., & Joint Special Design Team for Fusion DEMO. (2020). Effect of collisionality dependence of thermal force on impurity transport under a lower collisional condition in DEMO scrape-off layer plasma. Nuclear Fusion, 60(4), 046031. https://doi.org/10.1088/1741-4326/ab7537
  61. [61]Angus, J. R., & Umansky, M. V. (2014). Modeling of large amplitude plasma blobs in three-dimensions. Physics of Plasmas, 21(1), 012514. https://doi.org/10.1063/1.4863503
  62. [62]Ross, A., Stegmeir, A., & Coster, D. (2018). Effect of the Boussinesq approximation: Turbulence studies with GRILLIX in slab geometry. Contributions to Plasma Physics, 58(6-8), 478–483. https://doi.org/10.1002/ctpp.201700154
  63. [63]Held, M., & Wiesenberger, M. (2023). Beyond the Oberbeck–Boussinesq and long wavelength approximation. Nuclear Fusion, 63(2), 026008. https://doi.org/10.1088/1741-4326/aca9e0
  64. [64]Chankin, A. V., & Coster, D. P. (2015). On the locality of parallel transport of heat carrying electrons in the SOL. Journal of Nuclear Materials, 463, 498–501. https://doi.org/10.1016/j.jnucmat.2014.10.057
  65. [65]Procassini, R. J., Birdsall, C. K., & Cohen, B. I. (1990). Particle simulations of collisional transport in a high recycling, diverted tokamak scrape-off layer. Nuclear Fusion, 30(11), 2329–2348. https://doi.org/10.1088/0029-5515/30/11/010
  66. [66]Chodura, R. (1992). Kinetic Effects in the Scrape Off Layer. Contributions to Plasma Physics, 32(3-4), 219–230. https://doi.org/10.1002/ctpp.2150320308
  67. [67]Manfredi, G., Hirstoaga, S., & Devaux, S. (2011). Vlasov modelling of parallel transport in a tokamak scrape-off layer. Plasma Physics and Controlled Fusion, 53(1), 015012. https://doi.org/10.1088/0741-3335/53/1/015012
  68. [68]Takizuka, T. (2017). Kinetic effects in edge plasma: kinetic modeling for edge plasma and detached divertor. Plasma Physics and Controlled Fusion, 59(3), 034008. https://doi.org/10.1088/1361-6587/59/3/034008
  69. [69]Batishchev, O. V., Shoucri, M. M., Batishcheva, A. A., & Shkarofsky, I. P. (1999). Fully kinetic simulation of coupled plasma and neutral particles in scrape-off layer plasmas of fusion devices. Journal of Plasma Physics, 61(2), 347–364. https://doi.org/10.1017/S0022377898007375
  70. [70]Zhao, M., Chankin, A. V., & Coster, D. P. (2017). Kinetic simulations of electron heat flux in the scrape-off layer. Nuclear Materials and Energy, 12, 819–824. https://doi.org/10.1016/j.nme.2017.01.025
  71. [71]Tskhakaya, D., Adamek, J., Dimitrova, M., Hromasova, K., Seidl, J., Sos, M., Vondracek, P., & COMPASS team. (2021). Kinetic model of the COMPASS tokamak SOL. Nuclear Materials and Energy, 26, 100893. https://doi.org/10.1016/j.nme.2020.100893
  72. [72]Boyle, D. P., Majeski, R., Schmitt, J. C., Hansen, C., Kaita, R., Kubota, S., Lucia, M., & Rognlien, T. D. (2017). Observation of flat electron temperature profiles in the Lithium Tokamak Experiment. Physical Review Letters, 119, 015001. https://doi.org/10.1103/PhysRevLett.119.015001
  73. [73]Krstic, P., Maan, A., Koel, B. E., & Majeski, R. (2025). Diffusion of atoms in glassy mixtures of deuterium and lithium. The Journal of Physical Chemistry B, 129(43), 11318–11324.
  74. [74]Pérez, C. L., Maan, A., Abe, S., Boyle, D., Dahanayake, A., Jaramillo-Correa, C., Koel, B. E., Majeski, R., Marin, A., Le, T., et al. (2026). Testing of spark plasma sintered porous tungsten under neon glow discharge cleaning conditions in LTX-$ $. Nuclear Materials and Energy, 47, 102123. https://doi.org/10.1016/j.nme.2026.102123
  75. [75]Le, T. X., Banerjee, S., Maan, A., Majeski, R., Boyle, D. P., Shousha, R., Li, B., Morales, J., López Pérez, C., Gajani, H., et al. (2026). Synthetic modeling of soft x-ray emissivity for magnetic island analysis in LTX-$ $. Plasma Physics and Controlled Fusion, 68(6), 065003.
  76. [76]Maan, A., Le, T., Boyle, D. P., Majeski, R., Banerjee, S., Wilkie, G. J., Lampert, M., López Pérez, C., Shousha, R., Capecchi, W., & Gajani, H. (2026). Photodiode based multi-modal diagnostic for low-energy neutral beam injection in the LTX-$ $ spherical tokamak. Review of Scientific Instruments, 97(9), 093512. https://doi.org/10.1063/5.0344755
  77. [77]Maan, A., Majeski, R., Pérez, C. L., Boyle, D., Le, T., & Lunsford, R. (2026). Development of flash lithium evaporators for NSTX-U. https://arxiv.org/abs/2606.00777
  78. [78]Maan, A., Majeski, R., Pérez, C. L., Boyle, D., Le, T., & Lunsford, R. (2026). A plasma-material interaction design basis for NSTX-U flash lithium evaporators. Nuclear Materials and Energy, 49, 102218. https://doi.org/10.1016/j.nme.2026.102218
  79. [79]Tskhakaya, D. (2023). Implementation of dressed cross-section model into the BIT1 code. The European Physical Journal D, 77, 135. https://doi.org/10.1140/epjd/s10053-023-00682-w
  80. [80]Power, D., Mijin, S., Verhaegh, K., Militello, F., & Kingham, R. J. (2025). The impact of non-local parallel electron transport on plasma-impurity reaction rates in tokamak scrape-off layer plasmas. Plasma Physics and Controlled Fusion, 67, 045030. https://doi.org/10.1088/1361-6587/adc200
  81. [81]Roeltgen, J., Juno, J., Kotschenreuther, M., Bernard, T., Shukla, A., Francisquez, M., Hakim, A., Hammett, G., Power, D., & Hatch, D. (2025). A kinetic line-driven radiation operator and its application to Gyrokinetics. Nuclear Fusion, 65, 106020. https://doi.org/10.1088/1741-4326/adff28
  82. [82]Russell, D. A., Myra, J. R., Militello, F., & Moulton, D. (2021). Reduced-model scrape-off layer turbulence (nSOLT) simulations comparing three fueling scenarios. Physics of Plasmas, 28, 092305. https://doi.org/10.1063/5.0060524
  83. [83]Asunta, O., Abouelazayem, S., Ahmadi, T., Alieva, A., Anastopoulos-Tzanis, M. S., Andrew, Y., Ashton-Key, T., Bertelli, N., Bohlin, H., Brewer, T., et al. (2026). Overview of ST40 results and future: expanding the physics basis of high-field spherical tokamaks. Nuclear Fusion, 66(11), 116004. https://doi.org/10.1088/1741-4326/ae4a3c

Paper details

vlnc:W_4r9em882
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

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