betaFeedback

WorksPhysics and Astronomy

A full-wave ICRH solver with full kinetic accuracy using only first order differential equations

Abstract

Accurate full-wave computation of ion cyclotron resonance heating (ICRH) typically requires solving the computationally demanding integro-differential dielectric tensor of the hot plasma. Based on the method in [Phys. Plasmas 32, 082110 (2025)], we develop BO-FW1D, a 1D frequency domain finite difference full-wave solver which generates full kinetic, accurate and efficient simulations utilizing a fluid plasma model. To accurately convert the conductivity tensor in phase space into a first-order differential equation, we introduce a mathematically rigorous framework using pseudodifferential operator theory based on Kohn-Nirenberg ordering. Benchmarks with an all-orders 1D full-wave solver show high consistency in critical kinetic effects and electric-field distributions, while substantial computational acceleration is achieved. This work could lay the foundation for developing full-wave solvers that combine full kinetic accuracy with high computational efficiency.

Introduction

Ion cyclotron resonance heating (ICRH) is one of the core methods for heating, sustaining, and controlling high-temperature plasmas, and accurately simulating the propagation and absorption of radio-frequency (RF) waves in the plasma is a key component of ICRH modeling. To include critical kinetic effects such as Landau damping [1], Bernstein modes [2], and mode conversion process between fast waves and ion Bernstein waves (IBWs) [3, 4, 5], full-wave solvers must evaluate the plasma dielectric tensor incorporating the full kinetic response, which is computationally expensive.

Traditional spectral full-wave codes, such as AORSA [5], describe the kinetic dielectric tensor as an infinite series of truncated modified Bessel functions of the first kind InI_n. Besides its high computational cost, when treating high-k⊥k_\perp IBWs, the parameter λ=k⊥2ρ2\lambda= k_\perp^2\rho^2 becomes large, the function Γn(λ)=e−λIn(λ)\Gamma_n(\lambda) = e^{-\lambda}I_n(\lambda) decays rapidly in numerical evaluation, making the accuracy difficult to guarantee. TORIC expands the integro-differential plasma current function by using the small Larmor radius approximation, and retains the main finite Larmor radius (FLR) effects through the Swanson-Colestock-Kashuba (SCK) approximation [3, 4]. However, this method is valid only for k⊥≪1k_\perp\ll1, so it underestimates the damping of short-wavelength IBWs during propagation. Dirk Van Eester and E. A. Lerche approximated the integro-differential kinetic dielectric tensor using high-order polynomials of k⊥k_\perp, retaining a higher-order differential equation [6, 7]. While retaining the full FLR response, the fitting method is only valid for k⊥k_\perp inside the fitting range. On the other hand, although conventional multi-fluid [8] and gyro-Landau fluid (GLF) [9] models are computationally efficient, they cannot retain the complete FLR effects and thus miss key kinetic wave modes such as IBWs. Self-consistently incorporating kinetic effects while avoiding explicit kinetic integrals constitutes a long-standing core challenge in the field of ICRH full-wave simulation.

To break this impasse, Xie recently proposed a novel fluid-form treatment of the plasma dielectric tensor [10]. This method applies Padé approximations and pole expansions to rationalize the computationally expensive functions, effectively circumventing the numerical instability while preserving an algebraic rational-fraction form. A related approach has recently been pursued by Shiraiwa et al. [11], who approximate the non-local dielectric response by rational functions of k⊥2k_{\perp}^{2}. However, such an expansion is intrinsically symmetric in k⊥k_{\perp} and can therefore represent only even responses. In Xie’s model, the pole expansion is first order in k⊥k_{\perp}, so that the rational form carries no such symmetry constraint and contains the full response. The “fluid-like” model incorporates Landau damping and finite Larmor radius effects with only rational functions based on frequency ω\omega, parallel wavenumber k∥k_{\parallel} and perpendicular wavenumber k⊥k_{\perp}. Meanwhile, it can accurately attain the wave branches of IBWs and electron Bernstein waves (EBWs) [10] when solving the perpendicular wavenumber. Based on its fluid-like form and its ability to accurately capture Bernstein waves, the model is referred to as the “fluid BW model”.

Nevertheless, this model has so far been validated only for local, linear, electromagnetic kinetic plasma waves in an infinite, homogeneous system [10]. Owing to its ability to incorporate accurate kinetic effects while remaining valid for arbitrary k⊥k_{\perp}, the fluid BW dielectric tensor is expected to enable faster and more accurate full-wave simulation for ICRH. Extending this fluid BW dielectric tensor to inhomogeneous full-wave simulation constitutes the starting point of the present work. The compact rational-fraction algebraic structure of the fluid BW model makes it possible to convert the local dispersion relation into a first-order local differential equation in real space, turning the system into a sparse banded matrix that can be solved at a cost linear in the number of grid points.

The straightforward route is to treat the wavenumber as a differential operator rather than a number, kx→−i∂xk_x \to-i\partial_x, and to rewrite the local dispersion relation as a differential equation in real space. However, in an inhomogeneous plasma the coefficients depend on xx and do not commute with −i∂x-i\partial_x, so that operator compositions cannot be treated as ordinary algebraic products. If so, a non-physical amplification of the IBW amplitude will occur. To this end, we introduce the Kohn-Nirenberg pseudodifferential operator theory [12], which provides a rigorous mathematical framework for the conversion from phase-space symbols to real-space operators. The correction term due to operator non-commutativity is derived under this framework, and the non-physical amplification is eliminated. As a result, the finite-difference solver built on the corrected equations, BO-FW1D, achieves high consistency with the all-orders benchmark code under the JET parameters, demonstrating both the necessity and the effectiveness of the conversion framework.

The remainder of this paper is organized as follows. Section § 2 presents the fluid BW dielectric tensor and derives the corrected response equation in real space. Section § 3 benchmarks the resulting solver in two steps. First, the overall agreement with the all-orders benchmark code together with the necessity of the operator composition correction term. Second, the computational scaling and grid convergence of the solver. Section § 4 summarizes the main findings and discusses the limitations of the present one-dimensional treatment together with possible extensions and future works.

Model and Equations

The Fluid BW Model

The core of fluid BW model [10] is to apply rational approximations to the plasma dispersion function Z(ζ)Z(\zeta) and to Γn(λ)=e−λIn(λ)\Gamma_n(\lambda)=e^{-\lambda}I_n(\lambda), thereby transforming the kinetic dielectric tensor from its integral form into an algebraic fractional form. The plasma dispersion function Z(ζ)Z(\zeta) is expanded into a sum of partial fractions via a J-pole approximation, whose coefficients bjb_j and cjc_j have been derived in Ref. [13]:

Z(ζ)≃∑j=1Jbjζ−cj(1)Z(\zeta) \simeq\sum_{j=1}^{J} \frac{b_j}{\zeta-c_j} \tag*{(1)}

The function Γn(z2)\Gamma_n(z^2) is expressed via an LL-pole expansion (n≥0n \ge0, with Γ−n=Γn\Gamma_{-n}=\Gamma_n):

Γn(z2)=∑ln=12n+Lrlnz−pln(2)\Gamma_n(z^2) = \sum_{l_n=1}^{2n+L} \frac{r_{l_n}}{z-p_{l_n}} \tag*{(2)}

where z=k⊥∣ρcs∣z=k_{\perp}|\rho_{cs}|, ρcs\rho_{cs} is the Larmor radius of species ss. The poles plnp_{l_n} and residues rlnr_{l_n} are fixed by the rational-expansion procedure of Ref. [10]. Under these rational approximations, the full kinetic conductivity tensor in a uniform, Maxwellian plasma is rewritten as a sum of rational fractions:

σ=ϵ0∑l[blxxkx−clblxykx−clblxzkx−clblyxkx−clblyykx−clblyzkx−clblzxkx−clblzykx−clblzzkx−cl](3)\sigma= \epsilon_0 \sum_l \begin{bmatrix} \frac{b_{lxx}}{k_x-c_l} & \frac{b_{lxy}}{k_x-c_l} & \frac{b_{lxz}}{k_x-c_l} \\ \frac{b_{lyx}}{k_x-c_l} & \frac{b_{lyy}}{k_x-c_l} & \frac{b_{lyz}}{k_x-c_l} \\ \frac{b_{lzx}}{k_x-c_l} & \frac{b_{lzy}}{k_x-c_l} & \frac{b_{lzz}}{k_x-c_l} \end{bmatrix} \tag*{(3)}

where the summation index ll runs over the particle species ss, the truncated Bessel harmonic order n=0,1,…,Nn = 0, 1, \ldots, N, and the partial-fraction terms lnl_n of each Γn\Gamma_n:

∑l≡∑s∑n=0N∑ln(4)\sum_l \equiv\sum_s \sum_{n=0}^{N} \sum_{l_n} \tag*{(4)}

The coefficients blαβb_{l\alpha\beta} and clc_l are fully determined by the plasma parameters (density, temperature, and magnetic field) and the expansion coefficients. Their complete expressions are given in Ref. [10]. Expanding the perturbed plasma current with the same index ll as Jα=∑lvlαJ_\alpha= \sum_l v_{l\alpha} (α=x,y,z\alpha= x, y, z) and using the conductivity relation J=σ⋅E\mathbf{J} = \boldsymbol{\sigma} \cdot\mathbf{E}, the algebraic relation in rational-fraction form for each level ll can be constructed. For notational simplicity, the vacuum permittivity ϵ0\epsilon_0 is hereafter absorbed into the coefficients b⃗l\vec{b}_l:

v⃗l=b⃗lkx−cl⋅E(5)\vec{v}_l = \frac{\vec{b}_l}{k_x-c_l}\cdot\mathbf{E} \tag*{(5)}

Equation (5) constitutes the basic unit of the fluid BW dielectric tensor. In Xie’s work, this equation, together with Maxwell’s equations, is cast into the matrix eigenvalue problem M⋅X=kxX\mathbf{M}\cdot\mathbf{X}=k_x\mathbf{X}, which accurately calculates the dispersion relations of IBWs and EBWs in uniform media.

Differential Response Equation

To exploit the algebraic structure of Eq. (5) for solving the wave field in inhomogeneous media, the wavenumber kxk_x must be replaced by the differential operator −i∂x-i\partial_x, while the coefficients blαβb_{l\alpha\beta} and clc_l become functions of xx. If the denominator in Eq. (5) is first cleared algebraically before making this substitution, one obtains the first-order differential equation:

(−i∂x−cl(x))vl(x)=b⃗l(x)⋅E(x)(6)\left(-i\partial_x-c_l(x)\right)v_l(x)=\vec{b}_l(x)\cdot\mathbf{E}(x) \tag*{(6)}

However, when Eq. (6) is directly discretized with a central-difference scheme and solved, the computed IBW amplitude exhibits a significant non-physical amplification, which will be demonstrated in § 3. The root of this problem lies in the fact that, in the kxk_x matrix eigenvalue problem, kxk_x in Eq. (5) is an algebraic variable, so that (kx−cl)(k_x-c_l) commutes freely with the other coefficients and the clearing operation is equivalent to simple algebraic multiplication and division. In the full-wave problem with inhomogeneous media, however, −i∂x-i\partial_x is a differentiation rather than an algebraic multiplication: it acts on every xx-dependent quantity to its right, including the field E(x)\mathbf{E}(x), the current vl(x)v_l(x), and the coefficients blαβ(x)b_{l\alpha\beta}(x). The algebraic simplification rules that hold in uniform media lose their validity once differentiation is introduced, because the order of multiplication can no longer be freely interchanged. Hence, the local rational-fraction dispersion relation cannot be extended to an inhomogeneous differential equation by simple algebraic means.

It is therefore necessary to introduce an appropriate mathematical framework. To this end, we adopt the pseudodifferential operator theory based on Kohn–Nirenberg (KN) ordering [12]. The reason why we choose this framework is that it links an arbitrary phase space function σ(x,k)\sigma(x,k) with a real-space operator σ(X,D)\sigma(X,D) through the following equation:

σ(X,D)f(x)=∫−∞+∞eikxσ(x,k)f^(k) dk(7)\sigma(X,D)f(x)=\int_{-\infty}^{+\infty}e^{ikx}\sigma(x,k)\hat{f}(k)\,dk \tag*{(7)}

where f(x)f(x) is an arbitrary function and f^(k)\hat{f}(k) denotes its Fourier transform. The capital letters XX and DD denote the coordinate and differentiation operators, respectively,

Xf(x)=xf(x),Df(x)=−i∂xf(x),Xf(x)=xf(x),\qquad Df(x)=-i\partial_xf(x),

so that σ(X,D)\sigma(X,D) is an operator built from these two operators. A detailed introduction to this theory is given in Appendix § 5.

This theory provides a rigorous mathematical framework for computing the plasma current. In spectral full-wave codes, the real-space plasma current is obtained by taking the product of the conductivity-tensor spectrum at position xx, σ(x,kx)\sigma(x,k_x), with the electric-field spectrum, E(kx)E(k_x), and applying the inverse Fourier transform [5].

J(x)=σ^E(x)=∫−∞+∞eikxxσ(x,kx)E^(k) dk(8)J(x)=\hat{\sigma}E(x)=\int_{-\infty}^{+\infty}e^{ik_xx}\sigma(x,k_x)\hat{E}(k)\,dk \tag*{(8)}

This is exactly the operator action defined by Eq. (7).

Within this framework, the clearing of the denominator that relates Eq. (5) to Eq. (6) cannot be regarded as an algebraic multiplication of kx−clk_x-c_l on both sides, but is instead the application of the operator kx−cl^=−i∂x−cl(x)\widehat{k_x-c_l}=-i\partial_x-c_l(x) to both sides of the equation.

(kx−cl)^ vl=(kx−cl)^(b^lkx−cl^)⋅E(9)\widehat{(k_x-c_l)}\,\mathbf{v}_l=\widehat{(k_x-c_l)}\left(\frac{\widehat{\mathbf{b}}_l}{\widehat{k_x-c_l}}\right)\cdot\mathbf{E} \tag*{(9)}

Cast into an ordinary differential equation, the left-hand side of Eq. (9) follows directly from kx→−i∂xk_x\to-i\partial_x, whereas its right-hand side involves the composition of the two operators. Under the present framework this composition is not equivalent to the formal algebraic product: in addition to the term obtained by ordinary multiplication, a correction term arises from the ordering of the operators[12]. The detailed derivation is given in Appendix § 5. Including this correction term, the corrected fluid response equation becomes

(−i∂x−cl)vl=b^l⋅E−i(b^l⋅b^l′−1)⋅vl(10)(-i\partial_x-c_l)\mathbf{v}_l=\widehat{\mathbf{b}}_l\cdot\mathbf{E}-i\left(\widehat{\mathbf{b}}_l\cdot\widehat{\mathbf{b}}_l'^{-1}\right)\cdot\mathbf{v}_l \tag*{(10)}

Numerical Implementation

Equation (10) completes the plasma response. To obtain the wave field it must be solved together with the frequency-domain Maxwell equations:

∇×(∇×E)−ω2c2E=iωμ0∑lvl(11)\nabla\times(\nabla\times\mathbf{E})-\frac{\omega^2}{c^2}\mathbf{E}=i\omega\mu_0\sum_l\mathbf{v}_l \tag*{(11)}

With ∂y=0\partial_y=0 and ∂z=ikz\partial_z=ik_z, the operator ∇×(∇×E)\nabla\times(\nabla\times\mathbf{E}) in Eq. (11) retains the toroidal-curvature terms 1/r1/r and 1/r21/r^2 of the cylindrical geometry at r=R0+xr=R_0+x, its radial derivative entering only through the combination d2/dx2+r−1d/dx=r−1ddx(rddx)d^2/dx^2+r^{-1}d/dx=r^{-1}\frac{d}{dx}\left(r\frac{d}{dx}\right).

These coupled equations are discretized by finite differences on a uniform grid in xx, with grid coordinates xi=−a+(i−1)Δxx_i=-a+(i-1)\Delta x (i=1,…,N)(i=1,\ldots,N) and grid spacing Δx=2a/(N−1)\Delta x=2a/(N-1). All derivatives are approximated by second-order central differences:

dfdx∣i≃fi+1−fi−12Δx,d2fdx2∣i≃fi−1−2fi+fi+1Δx2(12)\left.\frac{df}{dx}\right|_i\simeq\frac{f_{i+1}-f_{i-1}}{2\Delta x},\qquad \left.\frac{d^2f}{dx^2}\right|_i\simeq\frac{f_{i-1}-2f_i+f_{i+1}}{\Delta x^2} \tag*{(12)}

Applying Eq. (12) to every interior grid point of Eqs. (10) and (11), together with the boundary conditions in Appendix B, the equations are assembled into a large sparse linear system, which will be solved by the Intel MKL PARDISO solver:

AX=b(13)A X=b \tag*{(13)}

The solution vector XX arranges all field quantities in grid-point order:

X=[Ex1,Ey1,Ez1,vlx1,vly1,vlz1,…,ExN,EyN,EzN,vlxN,vlyN,vlzN]TX=[E_x^1,E_y^1,E_z^1,v_{lx}^1,v_{ly}^1,v_{lz}^1,\ldots,E_x^N,E_y^N,E_z^N,v_{lx}^N,v_{ly}^N,v_{lz}^N]^{T}

where the superscript denotes the grid-point index. For SS particle species, a Bessel harmonic truncation NbesselN_{\mathrm{bessel}}, and expansion orders lnl_n for each Γn\Gamma_n, the total number of field components and current expansion components at a single grid point is 3+3Nl3+3N_l, where Nl=∑s=1S∑n=0NbessellnN_l=\sum_{s=1}^{S}\sum_{n=0}^{N_{\mathrm{bessel}}}l_n. The total matrix dimension is N×(3+3Nl)N\times(3+3N_l). Under typical JET D–H parameters, NN is of the order of several thousand and Nl≈200N_l\approx200, giving a total dimension of roughly 10610^6–10710^7.

This completes the full-wave solver BO-FW1D, which yields wave fields that retain the key kinetic effects, including fast-wave propagation, mode conversion, and the propagation and damping of the IBW. These results will be presented and benchmarked in detail in the next section against the all-orders spectral benchmark code[7]. The simulation geometry and plasma profiles are given in Appendix B.

Results

Validation against the All-Orders Kinetic Benchmark

BO-FW1D is compared directly with the spectral full-wave code of Ref. [7] (all-orders kinetic equations, spectral method), the two codes being run independently under the JET D–H parameters of Table 2, with BO-FW1D using N=5000N = 5000 grid points. We apply the Root Relative Squared Error (RRSE) to quantitatively measure the difference between the two results:

SymbolDensityTemperature
G0G_0n0=7×1019 m−3n_0=7\times10^{19}\ \mathrm{m}^{-3}T0=5 keVT_0=5\ \mathrm{keV}
GsG_sns=2×1019 m−3n_s=2\times10^{19}\ \mathrm{m}^{-3}Ts=100 eVT_s=100\ \mathrm{eV}
α\alphaαn=1\alpha_n=1αT=1.5\alpha_T=1.5
λ\lambdaλn=0.05 m\lambda_n=0.05\ \mathrm{m}λT=0.05 m\lambda_T=0.05\ \mathrm{m}
apa_pap=0.95 ma_p=0.95\ \mathrm{m}ap=0.95 ma_p=0.95\ \mathrm{m}

Table 2. Simulation parameters of the JET D–H plasma

RRSEα=∑i∣Eαsim(xi)−Eαref(xi)∣2∑i∣Eαref(xi)−Eαref‾∣2,α=x,y,z\mathrm{RRSE}_\alpha= \sqrt{\frac{\sum_i \left|E_\alpha^{\mathrm{sim}}(x_i)-E_\alpha^{\mathrm{ref}}(x_i)\right|^2}{\sum_i \left|E_\alpha^{\mathrm{ref}}(x_i)-\overline{E_\alpha^{\mathrm{ref}}}\right|^2}}, \quad\alpha=x,y,z

where Eαref‾\overline{E_\alpha^{\mathrm{ref}}} is the mean of the reference electric field over the simulation domain. Following Ref. [7], the sums run over the core plasma only, ∣x∣<ap=0.95 m|x| < a_p = 0.95\ \mathrm{m}, and the RRSE is evaluated separately for each component, since their amplitudes differ by orders of magnitude.

From Figure 1, the wave amplitudes and phases of the two codes nearly coincide in the fast-wave propagation region (x>0 mx > 0\ \mathrm{m}). However, in the IBW region on the x<0x < 0 side, the amplitude of the uncorrected BO-FW1D is clearly higher than the all-orders baseline. After the correction term is introduced, the BO-FW1D result recovers to nearly coincide with the benchmark: near the magnetic axis both consistently capture the mode conversion from the fast wave to the IBW, the short-wavelength IBW oscillations are clearly distinguishable on the x<0x < 0 side and gradually decaying with propagation, and the amplitude and decay envelope of the IBW are identical.

Real parts of the electric-field components of the JET D–H case

Figure 1. Real parts of the electric-field components of the JET D–H case of Table 2: (a) BO-FW1D without and (b) BO-FW1D with the operator-composition correction term. In each block the upper row shows ExE_x, EyE_y and EzE_z over the full simulation domain, RRSE with the benchmark are quoted in the title. The lower row enlarges the short-wavelength IBW region −0.7 m≤x≤−0.1 m-0.7\ \mathrm{m} \le x \le-0.1\ \mathrm{m}.

Quantitatively measuring the correction effect, the RRSEs of all three electric-field components decrease with respect to the all-orders kinetic benchmark, among which the RRSE of ExE_x drops from 0.194 to 0.090 (about 54%). This result fully demonstrates that the operator composition correction term is necessary for the finite-difference code to obtain correct results, and the mathematical treatment of the equations is self-consistent and reliable when solving the fluid BW dielectric tensor.

Computational Scaling and Grid Convergence

The comparisons above were all performed at a single grid resolution, so what remains to be established is whether the finite-difference formulation is convergent under grid refinement and whether the cost of the solve is indeed only linear in the number of grid points NN, as anticipated from the sparse banded structure of Eq. (13). To this end, the JET D–H case of Table 2 is repeated from N=1500N = 1500 to 7000 in steps of ΔN=100\Delta N = 100, with all remaining numerical settings and the boundary treatment kept unchanged. The reference solution is the all-orders kinetic spectrum of Ref. [7], reconstructed separately on the grid of each run so that every NN is compared at its own resolution.

Figure 2a compares the total computation time of the two codes, and Figure 2b solely compares the time for solving the matrix. The fitted power laws y=aNby = aN^b and typical computation times are recorded in Table 1. These measurements show that BO-FW1D attains close to linear scaling in NN both in the matrix solve and in the total computation time. For the all-orders code, the total computational cost approximately follows O(N2)O(N^2), and the matrix solving process follows the O(N3)O(N^3) law reported in Ref. [7]. Consequently, the speed advantage of BO-FW1D over the all-orders code grows with the number of grid points.

Grid-refinement study plots

Figure 2. Grid-refinement study of BO-FW1D under the JET D–H parameters: (a) total computation time (b) matrix solve time as functions of the number of grid points NN, together with the power-law fit aNbaN^b (solid curve) (c) RRSE of the three electric-field components with respect to the all-orders reference. (Environment: a laptop with Intel i9-14900HX and 32 GB memory. All runs use identical compiler settings.)

BO-FW1DAll-orders
Total cost∝N0.97\propto N^{0.97}∝N2.09\propto N^{2.09}
Matrix solve∝N0.98\propto N^{0.98}∝N3.01\propto N^{3.01}
Matrix structurebandeddense
N=1500N = 15002.58 s8.96 s
N=7000N = 700011.2 s157 s

Table 1. Computational cost of the two codes.

Figure 2c turns to the accuracy side of the same scan. Since the IBW is, to leading order, a quasi-electrostatic mode that propagates nearly perpendicular to the magnetic field with k⊥ρi≳1k_\perp\rho_i \gtrsim1 [4], its electric field is dominated by the component along k⊥k_\perp; in the present slab geometry (B=Bz^\mathbf{B} = B\hat{\mathbf{z}}, ky=0k_y = 0) this is ExE_x, so that the RRSE of ExE_x is the quantity of interest. It decreases with NN overall, from 0.112 at N=1500N = 1500 to 0.086 at N=7000N = 7000, a reduction of about 23% over the window. The two remaining components are of secondary interest only. EzE_z has an amplitude far smaller than those of ExE_x and EyE_y, since the very large conductivity parallel to the magnetic field strongly suppresses the parallel electric field, so that its relative error has little bearing on the accuracy of the wave field; moreover, for the present JET case EyE_y does not carry the mode-converted IBW [7]. Consistent with this, RRSE⁡(Ey)\operatorname{RRSE}(E_y) is essentially independent of NN and stays close to 0.025 throughout the whole window. The NN-dependence of the error is therefore confined to a small residual decrease, and the solution is already close to convergence.

Taken together, these results show that the fluid BW formulation retains the full kinetic response of the plasma–the mode conversion, the short-wavelength IBW and its damping–while requiring only a sparse banded solve whose cost grows almost linearly with the number of grid points. Calculating full kinetic wave physics therefore no longer demands a dense, globally coupled response matrix.

Conclusion and Discussion

Building on the fluid BW dielectric tensor proposed by Xie [10], this work develops a novel way of including kinetic effects in full-wave simulations with a fluid model. It establishes, and validates, a computational paradigm that extends the local kinetic dispersion relation to the full-wave solution in inhomogeneous media in the form of real-space differential equations, develops the one-dimensional frequency-domain finite-difference code BO-FW1D, and carries out a systematic validation under JET D–H plasma parameters.

The main findings and conclusions of this work are as follows. The local dispersion relation of the fluid BW model cannot be straightforwardly extended to differential equations by algebraic manipulation; if so, the non-commutativity of the multiplication order would give rise to a non-physical amplification of the IBW amplitude. The pseudodifferential operator theory based on Kohn-Nirenberg ordering offers a systematic treatment of this problem, and the operator-composition correction term restores the finite-difference results to agreement with the all-orders benchmark. In the resulting code, full-wave simulation using the fluid BW dielectric tensor is able to capture all key kinetic effects, with central differencing yielding a sparse banded matrix system whose solution cost grows only linearly with NN (b≃0.97b \simeq0.97), and grid refinement driving the solution towards convergence.

There is still much to be done. As the matrix structure is fixed, tailoring the solver according to the fixed structure of the matrix would bring a substantial speed-up. Meanwhile, a systematic scan of the L-pole expansion order LL and truncated Bessel harmonic order NN would quantify the accuracy–cost trade-off. Moreover, the fluid BW model’s application in other frequency regimes is also worth exploring. The simulation results are currently restricted to one-dimensional slab geometry, the ky=0k_y = 0 assumption, and a simplified antenna model. Because the speed advantage over the all-orders code grows with NN, extending the formulation to the more grid-intensive two-dimensional toroidal geometry is a particularly promising next step.

Pseudodifferential Operator Correction Based on Kohn-Nirenberg Ordering

The core of pseudodifferential operator theory is the quantization problem: associating to each function σ(x,k)\sigma(x,k) on phase space an operator σ(X,D)\sigma(X,D) acting on the function space, such that the coordinate function xx corresponds to the coordinate operator XX and the wavenumber function kk corresponds to the differential operator D=−i∂xD = -i\partial_x. Conversely, the phase-space function σ(x,k)\sigma(x,k) associated with a given operator is called its symbol. When the symbol is a polynomial in kk, the corresponding operator reduces to an ordinary differential operator; when the symbol is not a polynomial—e.g., a rational fraction—the corresponding operator is called a pseudodifferential operator. Since XX and DD do not commute, the operator associated with a given symbol depends on the order in which xx and kk act. The Kohn-Nirenberg ordering places the coordinate operator before the differential operator, so that the kk-dependence of the symbol acts on the field before its xx-dependence.

Pseudodifferential operator theory provides a rigorous mathematical framework for handling the real-space counterparts of non-polynomial functions on phase space. Under the KN ordering, the real-space operator corresponding to a phase-space function σ(x,k)\sigma(x,k) is the one defined by Eq. (7): the symbol is convolved with the spectrum of f(x)f(x), its kk-dependence being applied before its xx-dependence.

As mentioned in the main text, the so-called denominator-clearing process introduces a composite operator

P=(kx−cl)∘(b1↔kx−cl)(14)\mathcal{P} = (k_x-c_l) \circ\left(\frac{\overleftrightarrow{b_1}}{k_x-c_l}\right) \tag*{(14)}

In the pseudodifferential operator framework, the composition of two operators follows (Ref. [12], Theorem 2.52):

σA∘B=σA⋅σB−i∂σA∂k∂σB∂x+(−i)22!∂2σA∂k2∂2σB∂x2+⋯(15)\sigma_{A\circ B} = \sigma_A \cdot\sigma_B - i\frac{\partial\sigma_A}{\partial k}\frac{\partial\sigma_B}{\partial x} + \frac{(-i)^2}{2!}\frac{\partial^2 \sigma_A}{\partial k^2}\frac{\partial^2 \sigma_B}{\partial x^2} + \cdots \tag*{(15)}

In inhomogeneous media, blb_l and clc_l are functions of xx, so the correction terms cannot be neglected. Since σA=kx−cl(x)\sigma_A = k_x - c_l(x) is merely linear in kk, ∂2σA/∂k2=0\partial^2\sigma_A/\partial k^2 = 0, hence all correction terms of second order and higher in Eq. (15) vanish identically, and the first-order correction already gives the exact composition result:

δP=−i⋅∂∂k(kx−cl)⋅∂∂x(bl↔kx−cl)=−ibl↔′kx−cl−ibl↔cl′(kx−cl)2(16)\begin{aligned} \delta P=-i\cdot\frac{\partial}{\partial k}(k_x-c_l)\cdot\frac{\partial}{\partial x}\left(\frac{\overleftrightarrow{b_l}}{k_x-c_l}\right) \\ &=-i\frac{\overleftrightarrow{b_l}^{\prime}}{k_x-c_l}-i\frac{\overleftrightarrow{b_l}c_l^{\prime}}{(k_x-c_l)^2} \tag*{(16)} \end{aligned}

Here, bl↔′≡∂bl↔/∂x\overleftrightarrow{b_l}^{\prime}\equiv\partial\overleftrightarrow{b_l}/\partial x and cl′≡∂cl/∂xc_l^{\prime}\equiv\partial c_l/\partial x. The second term in Eq. (16) contains cl′c_l^{\prime}; since clc_l depends only on the Larmor radius ρcs\rho_{cs} and thus varies slowly in space, this term is a small quantity under typical tokamak parameters and can be neglected. Using vl=[bl↔/(kx−cl)]⋅E\boldsymbol{v}_l=[\overleftrightarrow{b_l}/(k_x-c_l)]\cdot\boldsymbol{E}, the first term can be written as Eq. (17), which is the correction term in Eq. (10) of the main text:

δPE=−i(bl↔′⋅bl↔−1)⋅vl(17)\delta P\boldsymbol{E}=-i\left(\overleftrightarrow{b_l}^{\prime}\cdot\overleftrightarrow{b_l}^{-1}\right)\cdot\boldsymbol{v}_l \tag*{(17)}

Simulation Geometry and Plasma Profiles

To facilitate benchmarking against existing full-wave codes, the simulation geometry and plasma profiles are chosen to be identical to those of [7], i.e., typical parameters of the JET tokamak. Taking the magnetic axis as the radial coordinate origin (x=R−R0x=R-R_0), the major radius is R0=2.97R_0=2.97 m, the plasma minor radius is a=1.05a=1.05 m, and the simulation domain is x∈[−a,a]x\in[-a,a].

The magnetic field contains only the toroidal component, decaying radially as 1/R1/R:

Bz(x)=B0R0R0+x,B0=3.45 TB_z(x)=B_0\frac{R_0}{R_0+x},\qquad B_0=3.45\ \mathrm{T}

The ICRH antenna operates at frequency f=51f=51 MHz with toroidal mode number ntor=27n_{\mathrm{tor}}=27, corresponding to the JET antenna dipole phasing mode. The parallel wavenumber varies radially as kz(x)=ntor/(R0+x)k_z(x)=n_{\mathrm{tor}}/(R_0+x).

The plasma consists of fully ionized deuterium ions (95%) and a hydrogen minority (5%), with the electron density determined by the quasi-neutrality condition ne=nD+nHn_e=n_D+n_H. Both the density and temperature profiles adopt a piecewise form with a parabolic-like core and an exponentially decaying edge:

Gcore(x)=(G0−Gs)(1−(x/ap)2)α+GsG_{\mathrm{core}}(x)=(G_0-G_s)\left(1-\left(x/a_p\right)^2\right)^\alpha+G_s
Gedge(x)=Gsexp⁡(−∣x−ap∣λ)G_{\mathrm{edge}}(x)=G_s\exp\left(-\frac{\lvert x-a_p\rvert}{\lambda}\right)

In the above, G0G_0 is the value at the magnetic axis, apa_p is the core–edge transition point, GsG_s is the profile value at that point, and α\alpha and λ\lambda control the shapes of the core and edge regions, respectively. The specific parameters are listed in Table 2.

The electron and ion temperatures use the same profile parameters. The density profiles of the deuterium and hydrogen ions are assigned as nD(x)=0.95n(x)n_D(x)=0.95n(x) and nH(x)=0.05n(x)n_H(x)=0.05n(x).

The boundary conditions are set as follows. At x=±ax=\pm a, EyE_y, EzE_z, and all current expansion components vlv_l satisfy homogeneous Dirichlet conditions; the xx-derivative of ExE_x is zero, implemented via a first-order one-sided difference: Ex1=Ex0E_x^1=E_x^0 and ExN−1=ExNE_x^{N-1}=E_x^N, where x0x_0 and xN+1x_{N+1} lie outside the solution domain and serve only to close the upper boundary difference stencils. The antenna excitation is introduced by modifying the right-hand-side vector b\boldsymbol{b}: the right-hand side of the equation corresponding to EyNE_y^N on the right boundary is set to 1, so that a tangential electric field of unit amplitude models the radio-frequency excitation of the antenna at the plasma edge.

Acknowledgments

The authors would like to thank the discussions and comments from Haojie Ma, Yilin Li, Yuxuan Wang, Jian Bao, Tieshuan Fan, Andong Xu, Beining Chen and Ziheng Zhou.

Funding

This work was supported by the National MCF Energy R&D Program of China (Grant No. 2025YFE03000100)

References

  1. [1]Lev Landau. On the vibrations of the electronic plasma. Zhurnal eksperimentalnoi i teoreticheskoi fiziki, 16(7):574–586, 1946.
  2. [2]Ira B Bernstein. Waves in a plasma in a magnetic field. Physical Review, 109(1):10, 1958.
  3. [3]M Brambilla. Finite larmor radius wave equations in tokamak plasmas in the ion cyclotron frequency range. Plasma physics and controlled fusion, 31(5):723–757, 1989.
  4. [4]Marco Brambilla. Numerical simulation of ion cyclotron waves in tokamak plasmas. Plasma Physics and Controlled Fusion, 41(1):1–34, 1999.DOI
  5. [5]EF Jaeger, LA Berry, E D’azevedo, DB Batchelor, and MD Carter. All-orders spectral calculation of radio-frequency heating in two-dimensional toroidal plasmas. Physics of Plasmas, 8(5):1573–1583, 2001.DOI
  6. [6]Dirk Van Eester and EA Lerche. Solving the all flr icrh integro-differential wave equation as a high-order differential equation for studying combined icrh-nbi heating. Nuclear Fusion, 61(1):016024, 2021.DOI
  7. [7]RHS Budé, D Van Eester, J van Dijk, RJE Jaspers, and AB Smolders. Accelerating simulations of electromagnetic waves in hot, magnetized fusion plasmas. Plasma Physics and Controlled Fusion, 63(3):035014, 2021.DOI
  8. [8]Huasheng Xie, Haojie Ma, and Yukun Bai. Plasma wave propagation conditions analysis using the warm multi-fluid model. Fundamental Plasma Physics, 10:100050, 2024.DOI
  9. [9]Gregory W Hammett and Francis W Perkins. Fluid moment models for landau damping with application to the ion-temperature-gradient instability. Physical review letters, 64(25):3019, 1990.pubmed.ncbi.nlm.nih.gov/10041874
  10. [10]Huasheng Xie. Developing a linear fluid plasma model with accurate kinetic bernstein waves: A first step. Physics of Plasmas, 32(8), 2025.DOI
  11. [11]Syun’ichi Shiraiwa, Nicola Bertelli, Eun-Hwa Kim, John Wright, Maja Verstraeten, and Philippe Lamalle. Full-wave modeling of rf waves in fusion plasmas with finite element method: Progress in past decades and its future role. EPJ Web of Conferences, 346:01001, 2026. ISSN 2100-014X.
  12. [12]GB FOLLAND. Harmonic-analysis in phase-space. ANNALS OF MATHEMATICS STUDIES, (122), 1989.
  13. [13]Huasheng Xie. Rapid computation of the plasma dispersion function: Rational and multi-pole approximation, and improved accuracy. AIP Advances, 14(7), 2024.DOI

Paper details

vlnc:W_r58c3gug
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