Arxiv:1901.06902V1 [Physics.Plasm-Ph] 21 Jan 2019 Equivalent to ’PDRK’), the ﬂuid Version PASS-F Succeed from PDRF [16], and Possibly More

A Uniﬁed Numerically Solvable Framework for Complicated Kinetic Plasma Dispersion Relations

Hua-sheng XIEa,b

aHebei Key Laboratory of Compact Fusion, Langfang 065001, China bENN Science and Technology Development Co., Ltd., Langfang 065001, China

Abstract A unified numerically solvable framework for dispersion relations with arbitrary number of species drifting at arbitrary directions and with Krook collision is derived for linear uniform/homogenous kinetic plasma, which largely extended the standard one [say, T. Stix, Waves in Plasmas, AIP Press, 1992]. The purpose of this work is to provide a kinetic plasma dispersion relation tool not only the physical model but also the numerical approach be as general/powerful as possible. As a very general application example, we give the final dispersion relations which assume further the equilibrium distribution function be bi-Maxwellian and including parallel drift, two directions of perpendicular drift (i.e., drift across magnetic field), ring beam and loss-cone. Both electromagnetic and electrostatic versions are provided, with also the Darwin (a.k.a., magnetoinductive or magnetostatic) version. The species can be treated either magnetized or unmagnetized. Later, the equations are transformed to the matrix form be solvable by using the powerful matrix algorithm [H. S. Xie and Y. Xiao, Plasma Science and Technology, 18, 2, 97, 2016], which is the first approach can give all the important solutions of a linear kinetic plasma system without requiring initial guess for root finding and thus can be extremely useful to the community. To our knowledge, the present model would be the most comprehensive one in literature for the distribution function constructed bases on Maxwellian, which thus can be applied widely for study waves and instabilities in space, astrophysics, fusion and laser plasma. We limit the present work to non-relativistic case. Keywords: Plasma physics, Kinetic dispersion relation, Waves and instabilities, Matrix eigenvalue

1. Introduction

Due to the complicated evolution of charged particles and electromagnetic field, one of the most important feature of plasma is the numerous waves and instabilities. The fundamental features of linear waves and instabilities in uniform/homogenous plasma can be described by dispersion relation and are discussed by many authors in monographes [3–5] and textbooks (cf. [6]). In standard treatment of kinetic plasma dispersion relation, the velocity space equilibrium distribution function is assumed to be fs0 = fs0(vk, v⊥), thus can not treat the cases with drift across magnetic field. To including the arbitrary directions of drift and collision would largely extend the application range of the dispersion relation, with the instabilities in the non-uniform shock [11, 14] be one of numerous of them. In this work, we try to provide a most comprehensive linear kinetic plasma dispersion relation tool to the community, which includes lots of new capabilities in both the physical model and algorithm. The tool is largely benefit from the powerful matrix approach of PDRK solver [1], which is the first algorithm can yield all the important kinetic solutions without initial guess for root finding. The new tool is named as PASS (Plasma wAve and inStability analySis), which includes the present work be kinetic version PASS-K succeed from PDRK [1] (hereafter ’PASS-K’ is refered arXiv:1901.06902v1 [physics.plasm-ph] 21 Jan 2019 equivalent to ’PDRK’), the fluid version PASS-F succeed from PDRF [16], and possibly more. In the following sections, we firstly derive the most general kinetic dispersion relation equation with drift across the magnetic field and Krook collision in section 2. In section 3, we assume a very general extended Maxwellian

Email address: [email protected], [email protected] (Hua-sheng XIE)

Preprint submitted to XXX January 23, 2019 equilibrium distribution function for all species and derive the corresponding ﬁnal dispersion relation. In section 4, the corresponding equations suitable be solved by PASS-K matrix algorithm are derived. In sections 5, we give some benchmark results. In section 6, we give a summary and some discussions.

2. The General Non-relativistic Dispersion Relation Considering that the widely used magnetized kinetic dispersion relation in literature such as in Ref.[3] does not including the drift across magnetic ﬁeld, we ﬁrstly derive our new models. A similar magnetized model is derived only recently in Ref.[11], but which is still not as general as the present one. We limit our study to non-relativistic model. To help the reader, we will give detailed steps of our derivations.

2.1. Starting equations We consider only the collisionless case or with a Krook collision, the kinetic equation for each species is the Vlasov equation with Krook collision at the right hand side

∂Fs ∂Fs h qs i ∂Fs + v · + as + (E + v × B) · = −νs(Fs − Fs0), (1) ∂t ∂r ms ∂v where the distribution function Fs(r, v, t) = ns0 fs(r, v, t), and qs, ms and ns0 are the charge, mass and the number density of species s, respectively. When the collision frequency νs = 0, the equation reduces to a collisionless case. The Maxwell equations for ﬁelds can be either 2 ∂t E = c ∇ × B − J/0, (2)

∂t B = −∇ × E, (3) for electromagnetic case, or

∇ · E = ρ/0, (4) E = −∇Φ, (5) for electrostatic case, where Z X X 3 J = Js = qsns0 dv v fs, (6) s Z X X 3 ρ = ρs = qsns0 dv fs, (7) s 3 √ dv = dv = dvxdvydvz and c = 1/ 0µ0 is the speed of light. The accelerate term as can be caused by other external forces such as the gravity and other (magneto-)hydro-dynamic forces due to spatial inhomogeneity, which then would cause drift motions across the magnetic field. A typical example is the low hybrid drift instability (LHDI) in space (e.g., in current sheet and shock cases) and fusion [9] plasma (e.g., in mirror and field-reversed configuration). To extend the application range, we will also given the Darwin model [16] version. In Darwin mode, all field variables can be divided into two parts: the transverse (T, divergence free) part and the longitudinal (L, curl free) part, i.e.,

E = EL + ET , ∇ · ET = 0, ∇ × EL = 0, (8)

B = BT , ∇ · BT = 0, (9)

J = JL + JT , ∇ · JT = 0, ∇ × JL = 0. (10) The corresponding ﬁeld equations are

∇ · EL = ρ/0, (11) ∇ · B = 0, (12) ¨¨ 2 ∂t EL + ¨∂t ET = c ∇ × B − J/0, (13) |{z} droped in Darwin model

∂t B = −∇ × E (or, ∂t B = −∇ × EL), (14) 2 where for our usage we need only the last two of them. The only diﬀerence from full electromagnetic model is that the term ∂t ET is dropped in the Darwin model. The Darwin model which eliminates the high frequency electromagnetic wave ω2 ∼ k2c2, is particular interesting in theoretical study and kinetic particle-in-cell and Vlasov simulations, i.e., which can use large time steps in simulation and saves the computation resource a lot [10]. R 3 We assume the zero-order term be a homogenous system with fs(v) = fs0(v) + fs1(v), dv fs0 = 1, as = as0, E = E0 + E1 and B = B0 + B1. The conventional derivation is further assume as0 = 0 and E0 = 0, whereas we treat also as0 , 0 and E0 , 0 in this work. Without loss of generality, we assume the background magnetic ﬁeld in z direction, i.e., B0 = B0zˆ = (0, 0, B0). The zero-order equations

h qs i ∂ fs0 as0 + (E0 + v × B0) · = 0, (15) ms ∂v X J0 = qsns0vds = 0, (16) s X ρ0 = qsns0 = 0, (17) s

R 3 1 0 0 0 0 p 2 2 where vds = dv v fs0. We introduce a cylindrical velocity coordinates (v⊥, φ , vk) with v⊥ = (vx − vdsx) + (vy − vdsy) 0 ms ms 0 0 and v = vz, where vdsx = ( as0y + E0y)/B0 and vdsy = −( as0x + E0x)/B0. We have vx = v cos φ + vdsx, k qs qs ⊥ 0 0 0 0 0 ∂vx 0 0 ∂vy 0 0 ∂v⊥ 0 ∂v⊥ 0 vy = v sin φ + vdsy and vz = v . We have also 0 = −v sin φ , 0 = v cos φ , = cos φ and = sin φ . Eq.(15) ⊥ k ∂φ ⊥ ∂φ ⊥ ∂vx ∂vy is

h qs i ∂ fs0 0 = as0 + (E0 + v × B0) · ms ∂v qs hms ∂ fs0 ms ∂ fs0 ms ∂ fs0 i = as0x + E0x + vyB0 + as0y + E0y − vxB0 + as0z + E0z ms qs ∂vx qs ∂vy qs ∂vz

qsB0 h ∂ fs0 ∂ fs0 i qsB0 0 0 ∂ fs0 0 0 ∂ fs0 = (−vdsy + vy) + (vdsx − vx) = v⊥ sin φ − v⊥ cos φ ms ∂vx ∂vy ms ∂vx ∂vy q B ∂v ∂ f ∂v ∂ f q B ∂ f − s 0 x s0 y s0 − s 0 s0 = 0 + 0 = 0 . (18) ms ∂φ ∂vx ∂φ ∂vy ms ∂φ

∂ fs0 0 0 Thus we have ∂φ0 = 0 if B0 , 0, i.e., fs0 = fs0(v⊥, vk). And, we have assume the force balance in the z-direction, i.e., ms as0z + E0z = 0. qs The ﬁrst order kinetic equation is (we have dropped the ’1’ subscript)

∂ fs ∂ fs qs ∂ fs qs ∂ fs0 + v · + (v × B0) · + (E + v × B) · = −νs fs, (19) ∂t ∂r ms ∂v ms ∂v and Fourier transform the equation using ∂/∂r → ik, ∂/∂t → −iω and also ∂t B = −∇ × E, we obtain ∂ f q h k × Ei ∂ f − · s s × · s0 i(ω k v + iνs) fs + ωcs 0 = E + v , (20) ∂φ ms ω ∂v

qs B0 2 where ωcs = is the cyclone frequency. Without loss of generality, we can assume the wave vector k = (kx, 0, kz) = ms (k sin θ, 0, k cos θ), which gives k⊥ = kx and kk = kz. We study four cases:

1. Electromagnetic or Darwin, magnetized: B0 , 0, B1 , 0.

1 When vds , 0, we have assumed the system to be Galilean invariant. It is not sure how much influence to the result yet. This issue is subtle, because the Vlasov equation is Galilean invariant and the electromagnetic field equations are Lorentz invariant and the total system is neither Galilean nor Lorentz invariant. 2 Ref.[11] assumed the across magnetic field drift vds at x direction but k = (kx, ky, kz), which will limit that all species can only drift at x − z direction. Later, they limit their discussion also to only ky = 0. And thus their result is our result at vdsy = 0 and νs = 0. 3 2. Electromagnetic or Darwin, unmagnetized: B0 = 0 (ωcs = 0), B1 , 0. 3. Electrostatic, magnetized: B0 , 0, B1 = 0 (k × E1 = 0). 4. Electrostatic, unmagnetized: B0 = 0 (ωcs = 0), B1 = 0 (k × E1 = 0). For convenient to theoretical study, we let the user to choose whether a species is magnetized or unmagnetized3, i.e., say for a electromagnetic case, the different species can be either magnetized (labeled as ’m’) or unmagnetized (labeled as ’u’). For unmagnetized case, Eq.(20) reduces to

u qs h k × Ei ∂ fs0 i(ω − k · v + iνs) fs = E + v × · , (21) ms ω ∂v

0 0 0 u where k · v = kxvx + kzvz or k · v = kzvk + kxv⊥ cos φ + kxvdsx, which gives the solution for fs is h i qs E + v × k×E · ∂ fs0 u ms ω ∂v fs = . (22) i(ω − k · v + iνs) For magnetized case, Eq.(20) reduces to ∂ f q h k × Ei ∂ f s − 0 s × · s0 0 i(xs + ys cos φ ) fs = E + v , (23) ∂φ msωcs ω ∂v

ω−k v0 −k v +iν 0 k k x dsx s k⊥v⊥ m where xs = − and ys = . The solution for magnetized f is much complicated. Instead of the ωcs ωcs s approaches using the method of characteristics in Refs.[3] and [11], we using the approach similar to Ref.[6], which solves the differential equation directly. That is to say, Eq.(23) is a first order linear differential equation of the form d f + P(x) f = Q(x), (24) dx which has a general solution Z x 0 R x 0 0 h R x 00 00 i f = e− P(x )dx Q(x0)e P(x )dx dx0 , (25) R where the integration factor term P(x)dx in our case is

Z φ0 Z φ0 0 00 00 00 0 0 P(φ )dφ = −i (xs + ys cos φ )dφ = −i(xsφ + ys sin φ ), (26) and thus Z φ0 0 0 h 00 00 i m i(xsφ +ys sin φ ) 00 −i(xsφ +ys sin φ ) 00 fs = e Q(φ )e dφ , (27) with q h k × Ei ∂ f Q = s E + v × · s0 . (28) msωcs ω ∂v

We notice that Eq.(27) can not reduce to (22) by set ωcs = 0, i.e., the magnetized version can not reduce to unmagnetized version by set ωcs = 0. Thus we should treat them separately.

3A typical case is in shock study, say Ref.[14], where electron are magnetized and drift ions are treated unmagnetized.

4 2.2. Electrostatic case For electrostatic case, k × E = 0, E = −ikΦ, and only Poisson equation for ﬁeld is required. We have

2 k Φ = ρ/0, (29) Z u m X u X m X 3 ρ = ρ + ρ = ρs + ρs = qsns0 dv fs. (30) s=u,m

The unmagnetized species − qsΦ k · ∂ fs0 u ms ∂v fs = , (31) (ω − k · v + iνs)

∂ fs0 ∂ fs0 ∂ fs0 ∂ fs0 0 ∂ fs0 ∂ fs0 where k · = kx + kz or k · = kx cos φ 0 + kz 0 . The corresponding charge density is ∂v ∂vx ∂vz ∂v ∂v⊥ ∂vk

Z 2 Z · ∂ fs0 u 3 u qs ns0Φ 3 k ∂v ρs = qsns0 dv fs = − dv . (32) ms (ω − k · v + iνs)

The velocity space integral may not be easy for complicated fs0, and probably better calculate at (vx, vy, vz) coordinates 0 0 0 instead of (v⊥, φ , vk). For magnetized species

Z φ0 −iΦq 0 0 h ∂ f 00 00 i m s i(xsφ +ys sin φ ) s0 −i(xsφ +ys sin φ ) 00 fs = e k · e dφ , (33) msωcs ∂v and

Z 2 Z Z φ0 −iq n Φ 0 0 h ∂ f 00 00 i m 3 m s s0 i(xsφ +ys sin φ ) s0 −i(xsφ +ys sin φ ) 00 0 0 0 0 ρs = qsns0 dv fs = e k · e dφ dφ v⊥dv⊥dvk, (34) msωcs ∂v which is not yet in a useful form. Let us do some further calculations step by step. We have used cylindrical coordi- 0 0 0 3 0 0 0 0 0 0 iφ0 −iφ0 nates (v⊥, φ , vk) with dv = v⊥dφ dv⊥dvk, and note ∂ fs0/∂φ = 0. Use cos φ = (e + e )/2, we have

Z φ0 Z φ0 ∂ f 00 00 ∂ f ∂ f 00 00 s0 −i(xsφ +ys sin φ ) 00 00 s0 s0 −i(xsφ +ys sin φ ) 00 k · e dφ = kx cos φ 0 + kz 0 e dφ (35) ∂v ∂v⊥ ∂vk Z φ0 Z φ0 ∂ f 00 00 ∂ f 1 00 00 00 00 s0 −i(xsφ +ys sin φ ) 00 s0 iφ −iφ −i(xsφ +ys sin φ ) 00 = kz 0 e dφ + kx 0 (e + e )e dφ . ∂vk ∂v⊥ 2

Now, we use the following expansion ∞ 00 X 00 −iys sin φ −inφ e = Jn(ys)e , (36) n=−∞ where Jn(ys) it the nth order Bessel function. And we have

Z φ0 Z φ0 00 00 X 00 X Jn(ys) 0 e−i(xsφ +ys sin φ )dφ00 = J (y ) e−i(xs+n)φ dφ00 = i e−i(xs+n)φ , (37) n s x n n n s + and

Z φ0 Z φ0 00 00 00 00 X 00 00 iφ −iφ −i(xsφ +ys sin φ ) 00 −i(xs−1+n)φ −i(xs+1+n)φ 00 (e + e )e dφ = Jn(ys) [e + e ]dφ (38) n 0 0 X he−i(xs−1+n)φ e−i(xs+1+n)φ i = i J (y ) + . (39) n s x − n x n n s 1 + s + 1 +

5 We further use ∞ 0 0 X 0 i(xsφ +ys sin φ ) i(xs+m)φ e = Jm(ys)e , (40) m=−∞ and thus in Eq.(34)

Z φ0 0 0 h i ei(xsφ +ys sin φ ) ··· dφ00 (41)

i(m−n+1)φ0 i(m−n−1)φ0 ∂ f X J J 0 i ∂ f X h e e i = ik s0 n m ei(m−n)φ + k s0 J J + z ∂v0 x + n 2 x ∂v0 n m x + n − 1 x + n + 1 k m,n s ⊥ m,n s s i(m−n)φ0 i(m−n)φ0 ∂ f X J J 0 i ∂ f X h e e i = ik s0 n m ei(m−n)φ + k s0 J J + J z ∂v0 x + n 2 x ∂v0 m n+1 x + n n−1 x + n k m,n s ⊥ m,n s s X ∂ f nω ∂ f ei(m−n)φ0 = i J J k s0 + cs s0 , m n z ∂v0 v0 ∂v0 x + n m,n k ⊥ ⊥ s

R 0 where we have used (Jn+1 + Jn−1) = (2n/ys)Jn. Further integral out dφ in Eq.(34) gives

Z Z φ0 Z i(m−n)φ0 0 0 h i X ∂ fs0 nωcs ∂ fs0 e ei(xsφ +ys sin φ ) ··· dφ00 dφ0 = i J J k + dφ0 (42) m n z ∂v0 v0 ∂v0 x + n m,n k ⊥ ⊥ s X J2 ∂ f nω ∂ f = i2π n k s0 + cs s0 x n z ∂v0 v0 ∂v0 n s + k ⊥ ⊥ X ω J2(y ) ∂ f nω ∂ f = −i2π cs n s k s0 + cs s0 , ω − k v0 iν − k v − nω z ∂v0 v0 ∂v0 n k k + s x dsx cs k ⊥ ⊥

R 2π 0 ei(m−n)φ dφ0 πδ δ where because 0 = 2 m,n, and m,n is Kronecker delta. Thus we find the final form is very similar to the standard Harris dispersion relation form, except the term xs. Combine the magnetized and unmagnetized species ρs and substitute them to the Poisson equation (29), we obtain the final electrostatic dispersion relation

∂ f ∂ f ∞ 2 s0 nωcs s0 2 Z ∞ Z ∞ Jn (ys) kk 0 + 0 0 X ωps X ∂vk v⊥ ∂v⊥ D(ω, k) = 1 + 2πv0 dv0 dv0 k2 ω − k v0 + iν − k v − nω ⊥ ⊥ k s=m −∞ 0 n=−∞ k k s x dsx cs 2 Z ∂ fs0 X ωps k · + dv3 ∂v = 0, (43) k2 (ω − k · v + iν ) s=u s

2 where ω2 = ns0qs . ps 0ms

2.3. Electromagnetic case For electromagnetic case, B = (k × E)/ω, the ﬁeld equations we needed are k × E −iωE = ic2 k × − J/ , (44) ω 0 Z u m X u X m X 3 J = J + J = Js + Js = qsns0 dv v fs. (45) s=u,m

If we assume the relation between current J and electric ﬁeld E be

J = σ · E. (46)

6 which will be calculated later via the kinetic equation, we obtain from Eq.(44) D(ω, k) · E = 0, (47) where D can be expressed in terms of the dielectric tensor K(ω, k) and gives the dispersion relation c2 |D(ω, k)| = |K(ω, k) + (kk − k2 I) | = 0, (48) ω2 where I is the unit tensor, and we have used k × (k × E) = k(k · E) − (k · k)E = (kk − k2 I) · E. And the relation to conductivity tensor σ is K = I + Q = I − σ/(iω0), (49) with Q = −σ/(iω0). Now, we calculate J. For unmagnetized species h i Z Z E + v × k×E · ∂ fs0 u 3 u 2 3 ω ∂v Js = qsns0 dv v fs = 0ωps dv v (50) i(ω − k · v + iνs) h i 2 Z − · − · · ∂ fs0 0ωps (ω k v + iνs)E iνs E + k(v E) ∂v = dv3v iω (ω − k · v + iνs) h i 2 2 Z − ∂ fs0 · ∂ fs0 0ωps 0ωps iνs(v ∂v ) + (k ∂v )(vv) = − I · E + dv3 · E, iω iω (ω − k · v + iνs)

∂ fs0 where we have used v × (k × E) = k(v · E) − (v · k)E, and vv, v ∂v are dyadic product tensors. The result is similar to ∂ fs0 ∂ fs0 the one in Ref.[14], except our new iνs term. Note also that the tensor v ∂v , ∂v v. For magnetized species Z m 3 m Js = qsns0 dv v fs (51)

2 Z Z φ0 q n 0 0 n h k × Ei ∂ f 00 00 o s s0 i(xsφ +ys sin φ ) s0 −i(xsφ +ys sin φ ) 00 0 0 0 0 = e E + v × · e dφ dφ v⊥dv⊥dvk msωcs ω ∂v 2 Z Z φ0 q n 0 0 n h i ∂ f 00 00 o s s0 i(xsφ +ys sin φ ) s0 −i(xsφ +ys sin φ ) 00 0 0 0 0 = e (ω − k · v)E + k(v · E) · e dφ dφ v⊥dv⊥dvk. msωcsω ∂v We calculate ﬁrstly h i ∂ f h i ∂ f (ω − k · v)E + k(v · E) · s0 = (ω − k v0 − k v0 cos φ0 − k v )E + k(v · E) · s0 ∂v z k x ⊥ x dsx ∂v 0 0 0 ∂ fs0 0 0 ∂ fs0 = (ω − kzvk − kxv⊥ cos φ − kxvdsx)[ 0 (Ex cos φ + Ey sin φ ) + 0 Ez] ∂v⊥ ∂vk

0 0 0 0 0 ∂ fs0 ∂ fs0 +[v⊥(Ex cos φ + Ey sin φ ) + vkEz + vdsxEx + vdsyEy](kx cos φ 0 + kz 0 ) ∂v⊥ ∂vk

n 0 0 0 0 0o∂ fs0 = (ω − kxvdsx)Ez + kzvdsxEx + kzvdsyEy + (kzv⊥Ex − kxv⊥Ez) cos φ + kzv⊥Ey sin φ 0 ∂vk

n 0 0 0 0 0o∂ fs0 + (ω − kzvk − kxvdsx)Ey sin φ + [(ω − kzvk)Ex + kxvdsyEy + kxvkEz] cos φ 0 ∂v⊥ n 0 0 ∂ fs0 0 0 ∂ fs0 o 0 = (kzv⊥Ex − kxv⊥Ez) 0 + [(ω − kzvk)Ex + kxvdsyEy + kxvkEz] 0 cos φ ∂vk ∂v⊥

n 0 ∂ fs0 0 ∂ fs0 o 0 ∂ fs0 + (ω − kzvk − kxvdsx)Ey 0 + kzv⊥Ey 0 sin φ + [(ω − kxvdsx)Ez + kzvdsxEx + kzvdsyEy] 0 ∂v⊥ ∂vk ∂vk 0 0 = (Us1Ex + Us2Ey + Us3Ez) cos φ + Us4Ey sin φ + (Us5Ex + Us6Ey + Us7Ez), (52) 7 with

0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 ∂ fs0 0 ∂ fs0 Us1 = [(ω − kzvk) 0 + kzv⊥ 0 ], Us2 = kxvdsy 0 , Us3 = [kxvk 0 − kxv⊥ 0 ], ∂v⊥ ∂vk ∂v⊥ ∂v⊥ ∂vk

0 ∂ fs0 0 ∂ fs0 ∂ fs0 Us4 = [(ω − kzvk − kxvdsx) 0 + kzv⊥ 0 ], Us5 = kzvdsx 0 , ∂v⊥ ∂vk ∂vk

∂ fs0 ∂ fs0 Us6 = kzvdsy 0 , Us7 = (ω − kxvdsx) 0 , (53) ∂vk ∂vk

∂ fs0 ∂ fs0 ∂ fs0 ∂ fs0 ∂ fs0 0 0 ∂ fs0 where we have used E· = Ex + Ey + Ez = 0 (Ex cos φ + Ey sin φ )+ 0 Ez, v· E = vxEx +vyEy +vzEz = ∂v ∂vx ∂vy ∂vz ∂v⊥ ∂vk 0 0 0 0 ∂ fs0 ∂ fs0 ∂ fs0 0 ∂ fs0 ∂ fs0 v⊥(Ex cos φ + Ey sin φ ) + v Ez + vdsxEx + vdsyEy, k · = kx + kz = kx cos φ 0 + kz 0 . k ∂v ∂vx ∂vz ∂v⊥ ∂vk ∂ fs0 0 0 0 0 0 0 0 0 0 0 terms: (ω − kzv − kxv⊥ cos φ − kxvdsx)(Ex cos φ + Ey sin φ ) + [v⊥(Ex cos φ + Ey sin φ ) + v Ez + vdsxEx + ∂v⊥ k k 0 0 0 0 0 0 0 0 vdsyEy]kx cos φ = (ω−kzvk−kxvdsx)(Ex cos φ +Ey sin φ )+(vkEz+vdsxEx+vdsyEy)kx cos φ = (ω−kzvk−kxvdsx)Ey sin φ + 0 0 0 [(ω − kzvk)Ex + kxvdsyEy + kxvkEz] cos φ . ∂ fs0 0 0 0 0 0 0 0 0 terms: (ω − kzv − kxv⊥ cos φ − kxvdsx)Ez + [v⊥(Ex cos φ + Ey sin φ ) + v Ez + vdsxEx + vdsyEy]kz = (ω − ∂vk k k 0 0 0 0 0 kxvdsx)Ez + kzvdsxEx + kzvdsyEy + (kzv⊥Ex − kxv⊥Ez) cos φ + kzv⊥Ey sin φ . Our Eq.(52) is the same as in Ref.[11] when set our vdsy = 0 and νs = 0, and their ky = 0. iφ −iφ 0 Thus, using sin φ = −i(e − e )/2 and Jn+1 − Jn−1 = −2Jn,

Z φ0 q 0 0 n h 1 00 00 −i 00 00 m s i(xsφ +ys sin φ ) iφ −iφ iφ −iφ fs = e (Us1Ex + Us2Ey + Us3Ez) (e + e ) + Us4Ey (e − e ) msωcsω 2 2 i 00 00 o −i(xsφ +ys sin φ ) 00 +(Us5Ex + Us6Ey + Us7Ez) e dφ Z φ0 qs X 0 n h 1 00 00 −i 00 00 = J ei(xs+m)φ (U E + U E + U E ) (eiφ + e−iφ ) + U E (eiφ − e−iφ ) m ω ω m s1 x s2 y s3 z s4 y s cs m 2 2 i X 00 o −i(xs+n)φ 00 +(Us5Ex + Us6Ey + Us7Ez) Jne dφ n −i(x +n−1)φ0 −i(x +n+1)φ0 qs X 0 X n 1hie s ie s i = J ei(xs+m)φ J (U E + U E + U E ) + m ω ω m n s1 x s2 y s3 z x n − x n s cs m n 2 ( s + 1) ( s + + 1) 0 0 0 −ihie−i(xs+n−1)φ ie−i(xs+n+1)φ i ie−i(xs+n)φ o +Us4Ey − + (Us5Ex + Us6Ey + Us7Ez) 2 (xs + n − 1) (xs + n + 1) (xs + n) iq X e−i(n−m)φ0 n n = s J (U E + U E + U E ) J + U E iJ0 + m ω ω m (x + n) s1 x s2 y s3 z y n s4 y n s cs m,n s s o (Us5Ex + Us6Ey + Us7Ez)Jn ,

8 And thus we have

 0 iφ0 −iφ0  Z Z v⊥(e + e )/2 + vdsx  0 0  m 3 m 3  − 0 iφ − −iφ  m Js = qsns0 dv v fs = qsns0 dv  iv⊥(e e )/2 + vdsy  fs  0  vk iq2n X Z e−i(n−m)φ0 n n = s s0 dv3 (U E + U E + U E ) J + U E iJ0 + m ω ω (x + n) s1 x s2 y s3 z y n s4 y n s cs m,n s s  v0 mJ /y + v J  o  ⊥ m s dsx m   − 0 0  (Us5Ex + Us6Ey + Us7Ez)Jn  iv⊥ Jm + vdsy Jm   0  vk Jm 2 Z ∞ Z ∞ 0 0 0 iq n X 2πv⊥dv⊥dvk n n n = s s0 ( U + U )J E + ( J U + iJ0 U + J U )E + m ω ω x n y s1 s5 n x y n s2 n s4 n s6 y s cs n −∞ 0 ( s + ) s s  0 n   (v⊥ + vdsx)Jn  n o  ys   − 0 0  ( Us3 + Us7)JnEz  ( iv⊥ Jn + vdsy Jn)  ys  0  vk Jn 2 Z ∞ Z ∞ 0 0 0 iq n X 2πv⊥dv⊥dvk = s s0 Π · E, (54) m ω ω x n s s cs n −∞ 0 ( s + ) with  ( n U + U )(v0 n + v )J2 ( n J U + iJ0 U + J U )(v0 n + v )J ( n U + U )(v0 n + v )J2   ys s1 s5 ⊥ ys dsx n ys n s2 n s4 n s6 ⊥ ys dsx n ys s3 s7 ⊥ ys dsx n   n 0 0 n 0 0 0 n 0 0  Π =  ( Us1 + Us5)(−iv⊥ J + vdsy Jn)Jn ( JnUs2 + iJ Us4 + JnUs6)(−iv⊥ J + vdsy Jn)( Us3 + Us7)(−iv⊥ J + vdsy Jn)Jn  . (55) s  ys n ys n n ys n   ( n U + U )v0 J2 ( n J U + iJ0 U + J U )v0 J ( n U + U )v0 J2  ys s1 s5 k n ys n s2 n s4 n s6 k n ys s3 s7 k n Write out each terms:

n 0 n 2 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 n 2 • Π11 = ( Us1 + Us5)(v⊥ + vdsx)Jn = ( [(ω − kzv ) 0 + kzv⊥ 0 ] + kzvdsx 0 )(v⊥ + vdsx)Jn ys ys ys k ∂v⊥ ∂vk ∂vk ys 2 nωcs 0 ∂ fs0 nωcs ∂ fs0 nωcs = Jn [ 0 (ω − kzv ) 0 + kz( + vdsx) 0 ]( + vdsx). [agree with Umeda18, except that his four terms can kxv⊥ k ∂v⊥ kx ∂vk kx cancel]

n 0 0 n n ∂ fs0 0 0 ∂ fs0 0 ∂ fs0 • Π12 = ( JnUs2 + iJnUs4 + JnUs6)(v⊥ + vdsx)Jn = ( Jnkxvdsy 0 + iJn[(ω − kzv − kxvdsx) 0 + kzv⊥ 0 ] + ys ys ys ∂v⊥ k ∂v⊥ ∂vk ∂ fs0 0 n Jnkzvdsy 0 )(v⊥ + vdsx)Jn ∂vk ys 2 nωcs nωcs ∂ fs0 ∂ fs0 0 nωcs 0 ∂ fs0 0 ∂ fs0 = Jn vdsy( +vdsx)( 0 0 +kz 0 )+iJn Jn( +vdsx)[(ω−kzv −kxvdsx) 0 +kzv⊥ 0 ]. [agree with Umeda18] kx v⊥ ∂v⊥ ∂vk kx k ∂v⊥ ∂vk

n 0 0 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 0 • Π21 = ( Us1 + Us5)(−iv⊥ Jn + vdsy Jn)Jn = ( [(ω − kzv ) 0 + kzv⊥ 0 ] + kzvdsx 0 )(−iv⊥ Jn + vdsy Jn)Jn ys ys k ∂v⊥ ∂vk ∂vk 0 0 2 nωcs 0 ∂ fs0 nωcs ∂ fs0 ∂ fs0 = (−iv⊥ Jn Jn + vdsy Jn )[ 0 (ω − kzv ) 0 + kz 0 + kzvdsx 0 ]. [agree with Umeda18, except that his two kxv⊥ k ∂v⊥ kx ∂vk ∂vk terms can cancel]

n 0 0 0 n ∂ fs0 0 0 ∂ fs0 0 ∂ fs0 • Π22 = ( JnUs2 + iJnUs4 + JnUs6)(−iv⊥ Jn + vdsy Jn) = ( Jnkxvdsy 0 + iJn[(ω − kzv − kxvdsx) 0 + kzv⊥ 0 ] + ys ys ∂v⊥ k ∂v⊥ ∂vk ∂ fs0 0 0 Jnkzvdsy 0 )(−iv⊥ Jn + vdsy Jn) ∂vk 0 0 nωcs ∂ fs0 ∂ fs0 0 0 0 0 ∂ fs0 0 ∂ fs0 = Jnvdsy(−iv⊥ Jn + vdsy Jn)[ 0 0 + kz 0 ] + iJn(−iv⊥ Jn + vdsy Jn)[(ω − kzv − kxvdsx) 0 + kzv⊥ 0 ]. [agree with v⊥ ∂v⊥ ∂vk k ∂v⊥ ∂vk Umeda18]

n 0 n 2 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 n 2 • Π13 = ( Us3 + Us7)(v⊥ + vdsx)Jn = ( [kxv 0 − kxv⊥ 0 ] + (ω − kxvdsx) 0 )(v⊥ + vdsx)Jn ys ys ys k ∂v⊥ ∂vk ∂vk ys v0 2 nωcs k ∂ fs0 ∂ fs0 = Jn ( + vdsx)[nωcs 0 0 + (ω − nωcs − kxvdsx) 0 ]. [agree with Umeda18] kx v⊥ ∂v⊥ ∂vk

n 0 2 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 2 • Π31 = ( Us1 + Us5)v Jn = ( [(ω − kzv ) 0 + kzv⊥ 0 ] + kzvdsx 0 )v Jn ys k ys k ∂v⊥ ∂vk ∂vk k 0 2 nωcs 0 ∂ fs0 nωcs ∂ fs0 ∂ fs0 = v Jn [ 0 (ω − kzv ) 0 + kz 0 + kzvdsx 0 ]. [agree with Umeda18, except that his two terms can cancel] k kxv⊥ k ∂v⊥ kx ∂vk ∂vk 9 n 0 0 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 0 • Π23 = ( Us3 + Us7)Jn(−iv⊥ Jn + vdsy Jn) = ( [kxv 0 − kxv⊥ 0 ] + (ω − kxvdsx) 0 )Jn(−iv⊥ Jn + vdsy Jn) ys ys k ∂v⊥ ∂vk ∂vk v0 2 0 0 k ∂ fs0 ∂ fs0 = (vdsy Jn − iv⊥ Jn Jn)[nωcs 0 0 + (ω − nωcs − kxvdsx) 0 ]. [agree with Umeda18] v⊥ ∂v⊥ ∂vk

n 0 0 n ∂ fs0 0 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 • Π32 = ( JnUs2 +iJnUs4 + JnUs6)v Jn = ( Jnkxvdsy 0 +iJn[(ω−kzv −kxvdsx) 0 +kzv⊥ 0 ]+ Jnkzvdsy 0 )v Jn ys k ys ∂v⊥ k ∂v⊥ ∂vk ∂vk k 0 2 nωcs ∂ fs0 ∂ fs0 0 0 0 ∂ fs0 0 ∂ fs0 = v Jn ( 0 0 + kz 0 )vdsy + iv Jn Jn[(ω − kzv − kxvdsx) 0 + kzv⊥ 0 ]. [agree with Umeda18] k v⊥ ∂v⊥ ∂vk k k ∂v⊥ ∂vk

n 0 2 n 0 ∂ fs0 0 ∂ fs0 ∂ fs0 0 2 • Π33 = ( Us3 + Us7)v Jn = ( [kxv 0 − kxv⊥ 0 ] + (ω − kxvdsx) 0 )v Jn ys k ys k ∂v⊥ ∂vk ∂vk k v0 0 2 k ∂ fs0 ∂ fs0 = v Jn [nωcs 0 0 + (ω − nωcs − kxvdsx) 0 ]. [agree with Umeda18] k v⊥ ∂v⊥ ∂vk

ω−k v0 −k v iν 0 k k x dsx+ s kxv⊥ Note: xs = − and ys = . ωcs ωcs At vdsy = 0 and νs = 0 limit, our result reduces to exactly the same result as the Eq.(25) of Ref.[11]. By further set vdsx = 0, our result reduces to the standard without across magnetic ﬁeld drifts one in Ref.[6] and the non-relativistic case in Ref.[3]. We should also note that when vdsx,y , 0, the matrix elements Πi j are not symmetric or antisymmetric any more. The Q in electromagnetic dispersion relation is σ Q = − iω0 h i 2 2 Z − ∂ fs0 · ∂ fs0 X n ωps ωps iνs(v ∂v ) + (k ∂v )(vv) o = − I + dv3 ω2 ω2 (ω − k · v + iν ) s=u s 2 ∞ Z ∞ Z ∞ 0 0 0 X ωps X 2πv⊥dv⊥dvk + Π . (56) ω2 (ω − k v0−k v + iν − nω ) s s=m n=−∞ −∞ 0 k k x dsx s cs

2.4. Darwin model case For Darwin model, we still have B = (k × E)/ω, and thus the calculation of current J = Ju + Jm is exactly the same as in Eq.(45), i.e., the kinetic solutions to the distribution function fs and current J do not need change as in the above electromagnetic model, which simplify our derivation a lot. We discuss the change of the ﬁeld equation here. In Fourier space, we have

k(k · E) kk k(k · E) kk E = E − = (I − ) · E, E = = ( ) · E, (57) T k2 k2 L k2 k2 which can satisﬁed k · ET = 0, k × EL = 0 and E = ET + EL. We have used the tensor relation (UV) · W = U(V · W) and W · (UV) = (W · U)V, where U, V and W are vectors. The ﬁeld equations we needed are k × E −iωE = ic2 k × − J/ , (58) L ω 0 i.e.,

kk c2 ( ) · E + (kk − k2 I) · E + Q · E = 0, (59) k2 ω2 | {z } I·E in electromagnetic model or

k2c2 kk I · E + + 1 − I · E + Q · E = 0, (60) ω2 k2 | {z } 2 2 k c in electromagnetic model ω2

10 · kk · k2c2 where the only change from the electromagnetic model is that the I E term is changed to be ( k2 ) E, or ω2 term is k2c2 changed to be ω2 + 1. And thus the Darwin dispersion relation is

k2c2 kk |D(ω, k)| = |K(ω, k) + + 1 − I | = 0, (61) ω2 k2 | {z } 2 2 k c in electromagnetic model ω2

Eqs.(43), (48) and (61) with Q in (56) are our starting electrostatic, electromagnetic and Darwin dispersion relations with drift across magnetic ﬁeld. The above dispersion relations are valid for arbitrary non-relativistic distribution functions. Later, we will limit our study to treat a special case of the distribution function fs0.

3. The Dispersion Relation for Extend Maxwellian Distribution

The extend Maxwellian distribution here means bi-Maxwellian distribution with loss cone, parallel and perpendicular drifts and ring beam. We model it use the following equilibrium distribution function.

3.1. Equilibrium distribution function 0 0 0 0 0 0 p 2 2 We assume equilibrium distribution function Fs0(vk, v⊥) = ns0 fs0(vk, v⊥), with vk = vz, v⊥ = (vx − vdsx) + (vy − vdsy) , and

0 0 0 0 fs0(vk, v⊥) = fs0z(vk) fs0⊥(v⊥) (62) 0 2 0 2 0 2 1 h (vk − vdsz) in r h (v − v ) i r h (v − v ) io = exp − sa exp − ⊥ dsr + sb exp − ⊥ dsr , 3/2 2 2 2 2 π vztsv⊥ts vzts Asa v⊥ts αsAsb αsv⊥ts

1−αs∆s −αs+αs∆s where rsa = and rsb = , and 1−αs 1−αs √ v2 πv v A − dsr dsr − dsr , σ a, b, sσ = exp 2 + erfc for = (63) v⊥tsσ v⊥tsσ v⊥tsσ √ with v⊥tsa = v⊥ts and v⊥tsb = αsv⊥ts. P That is, the equilibrium distribution function fs0 is separated to two sub-distributions: fs0 = σ=a,b rsσ fs0σ = 0 P 0 0 P fs0z(vk) σ=a,b rsσ fs0⊥σ, fs0σ = fs0z(vk) fs0⊥σ(v⊥), fs0⊥ = σ=a,b rsσ fs0⊥σ,

0 2 2 1 h (vk − vdsz) i 1 h (v − v ) i f v0 − − z dsz , s0z( k) = 1/2 exp 2 = 1/2 exp 2 (64) π vzts vzts π vzts vzts and 1 h (v0 − v )2 i f v0 − ⊥ dsr s0⊥σ( ⊥) = 2 exp 2 πAsσv⊥tsσ v⊥tsσ p 2 2 2 1 h ( (vx − vdsx) + (vy − vdsy) − vdsr) i − , = 2 exp 2 (65) πAsσv⊥tsσ v⊥tsσ where vdsx, vdsy, vdsz are the drift velocities in x (perpendicular 1), y (perpendicular 2) and z (parallel) directions, respectively. And, vdsr is the perpendicular ring beam velocity, and erfc(−x) = 1 − erf(−x) = 1 + erf(x) is the comple- R ∞ 2 R x 2 x ≡ √2 e−t dt x ≡ √2 e−t dt v v mentary error function, with erfc( ) π x and erf( ) π 0 . The zts and ⊥tsσ are the parallel and 1 2 1 2 perpendicular thermal velocities and corresponding temperatures are Tzs = 2 kBmsvzts and T⊥sσ = 2 kBmsv⊥tsσ. We deﬁne the temperature anisotropic λT sσ = Tzs/T⊥sσ. The parameters ∆s and αs determine the depth and size of the 11 loss-cone4. Here, ∆ ∈ [0, 1], for max loss cone and no loss cone. If ∆ = 1 or α = 1, i.e., r = 1 and r = 0, the s R R sR s R saR sb above equation reduced to no loss cone case. Note fs0dv = fs0zdv = fs0⊥dv = fs0σdv = fs0⊥σdv = 1. We will use the same distribution in magnetized and unmagnetized versions to simpliﬁed the notations. For unmagnetized version, to remove the trouble of integral, we study only vdsr = 0. How to justify the assumed distribution to the realistic physics problem is leave to the user. For example, the linearized equation (i.e., the dispersion relations) can also run when the zero order current and charge density are not zero. Thus, the user should justify whether it is reasonable. Several ∂ fs0/∂v terms are particular useful for further discussions:

0 ∂ f 2(vk − vdsz) X r f s0 = − f = −2(v0 − v ) sσ s0σ , ∂v0 2 s0 k dsz 2 k vzts σ=a,b vzts ∂ f X r f s0 = −2(v0 − v ) sσ s0σ , ∂v0 ⊥ dsr 2 ⊥ σ=a,b v⊥tsσ

∂ fs0 h vdsr i X rsσ fs0σ = −2 1 − (vx − vdsx) , ∂v p 2 2 2 x (vx − vdsx) + (vy − vdsy) σ=a,b v⊥tsσ

∂ fs0 h vdsr i X rsσ fs0σ = −2 1 − (vy − vdsx) , ∂v p 2 2 2 y (vx − vdsx) + (vy − vdsy) σ=a,b v⊥tsσ ∂ f 2(v − v ) X r f s0 = − z dsz f = −2(v − v ) sσ s0σ , ∂v 2 s0 z dsz 2 z vzts σ=a,b vzts P where we have used that fs0 = σ=a,b rsσ fs0σ.

3.2. Notations q q 2 2kBTs kBTs 2 ns0qs Note the definition of vts, i.e., vts = m , not vts = m as in Ref.[12]. Other notations: ωps = m , s s q 0 s q B k T √ 2 P 2 s 0 2 0 B zs vdsr kBTs⊥σ 1 √v⊥tsσ ωp = s ωps, Ωs = , λ = 2 , asσ = 2k⊥ρcsσ, bsσ = , ρcsσ = = . Note: Ωs < 0 ms Ds ns0qs v⊥tsσ ms Ωs 2Ωs for electron (q < 0), and thus also ρ < 0 and a < 0. [Note the definition in previous bi-Maxwellian version: s q csσ sσ 2 2 kBTs⊥ 1 √v⊥ts −b as = k⊥ρcs, bs = k⊥ρcs, ρcs = = , Γn(b) = In(b)e , In is the modified Bessel function.] ms Ωs 2Ωs We use ’ES3D’ or ’ES’, ’EM3D’ or ’EM’, ’Darwin’ to represent the electrostatic, electromagnetic and Darwin version, respectively. And with suffix ’-U’ and ’-M’ to represent the corresponding unmagnetized and magnetized species.

3.3. Some integrals and functions

Here, we also clarify the correctly treat of the plasma dispersion function for kz ≤ 0. The standard deﬁnition of plasma dispersion function Z(ζ) is Z 1 1 2 Z(ζ) = √ e−x dx, π C x − ζ dZ = −2(1 + ζZ), dζ where C is the Landau contour to analytic continuation from Im(ζ) > 0 to Im(ζ) ≤ 0. However, for our usage the ω−kcs above deﬁnition to plasma dispersion function with ζs = is only correct when kts > 0, where for example kts kts = kkvzts. To correctly capture the physics, one should be careful of the analytic continuation for both kts > 0 and

4Actually, the user can also use two diﬀerent specieses to represent the loss cone distribution, with one of them has negative density. Then, they can also have diﬀerent vdsx, vdsy and vdsr. Many complicated distribution function can be constructed based on our model, which we leave it to the user. For example, Ref.[17] constructed the shell distribution based on ring beam model.

12 kts ≤ 0. From standard derivation of the dispersion relations based on Laplacian transformation instead of Fourier transformation which considered the causality, say Ref.[6], we obtain the correct analytic continuation one should be √  −ζ2 1 R ∞ 1 −x2  i πe s + √ P e dx, kts > 0,  π −∞ x−ζs  1 Z(ζs) = − , kts = 0, (66)  √ ζs  −ζ2 1 R ∞ 1 −x2  −i πe + √ P e dx, kts < 0,  π −∞ x−ζs where P refers to the principal value integral. The corresponding J-pole expansion should also be modified for kts ≤ 0, which will be discussed later. We can find easily that for kts < 0, one can use Z(ζs) = −Z(−ζs), which will simplify our later usage. To simplify the notation, we define also the function Zp(ζs, kts), and have Z p 1 x −x2 Zp(ζs, kts) = √ e dx, π C x − ζs Z0(ζs, kts) = Z(ζs)

Z1(ζs, kts) = [1 + ζsZ0(ζs)],

Z2(ζs, kts) = ζs[1 + ζsZ0(ζs)], 1 Z (ζ , k ) = + ζ2[1 + ζ Z (ζ )], 3 s ts 2 s s 0 s √ R ∞ 2 −x2 π where we have used −∞ x e dx = 2 . Later, we will find by using Z0,1,2,3 will simplify the notations a lot, and also will make the matrix linear transformation quite straightforward, say, Z ∞ fsz Z(ζsn) Z0(ζsn) dvk = − = − , (67) −∞ ωsn − kz(vk − vdsz) kzvzts kzvzts Z ∞ kz(vk − vdsz) fsz dvk = −[1 + ζsnZ(ζsn)] = −Z1(ζsn), (68) −∞ ωsn − kz(vk − vdsz) Z ∞ 2 [kz(vk − vdsz)] fsz dvk = −ωsn[1 + ζsnZ(ζsn)] = −kzvztsZ2(ζsn), (69) −∞ ωsn − kz(vk − vdsz) Z ∞ 3 2 2 [kz(vk − vdsz)] fsz kz vzts 2 2 2 dvk = − − ωsn[1 + ζsnZ(ζsn)] = −kz vztsZ3(ζsn). (70) −∞ ωsn − kz(vk − vdsz) 2 0 For Bessel function, we have (p256 of Ref.[3]): Jn(x) = [Jn−1(x) − Jn+1(x)]/2, nJn(x)/x = [Jn−1(x) + Jn+1(x)]/2, 2 2 P∞ 2 P∞ 0 P∞ 2 P∞ 0 2 1 P∞ n Jn (x) 1 P∞ 0 n=−∞ Jn = 1, n=−∞ Jn Jn = 0, n=−∞ nJn = 0, n=−∞(Jn) = 2 , n=−∞ x2 = 2 , n=−∞ nJn Jn = 0, − − n P∞ Jn( x) = J−n(x) = ( 1) Jn(x), and m,0,n=−∞ Jn Jn+m = 0. We define Z ∞ 2 −(y−b)2 An(a, b, c) ≡ Jn (ay)e (y − c)dy, 0 Z ∞ 0 −(y−b)2 Bn(a, b, c) ≡ Jn(ay)Jn(ay)e y(y − c)dy, 0 Z ∞ 0 2 −(y−b)2 2 Cn(a, b, c) ≡ Jn (ay)e y (y − c)dy. 0 √ vdsr For our usage a = 2k⊥ρcs, b = v and c = b or 0. Here, we calculate An, Bn and Cn using numerical integral. When ⊥ts 0 b = c = 0, the above integrals reduce to the conventional Maxwellian form with modified Bessel function In and In, i.e.,

2 2 2 1 − a a 1 a A (a, 0, 0) = e 2 I ( ) = Γ ( ), n 2 n 2 2 n 2 2 2 2 2 a − a 0 a a a 0 a B (a, 0, 0) = e 2 [I ( ) − I ( )] = Γ ( ), n 4 n 2 n 2 4 n 2 2 2 2 2 2 2 2 2 2 2 2 − a h n a a 0 a a a i n a a 0 a C (a, 0, 0) = e 2 I ( ) − I ( ) + I ( ) = Γ ( ) − Γ ( ). n 2a2 n 2 4 n 2 4 n 2 2a2 n 2 4 n 2 13 −b 0 0 −b 0 where Γn(b) = e In(b). And Γn(b) = (In − In)e , In(b) = (In+1 + In−1)/2, I−n = In. Thus, for species with vdsr = 0, we 0 will still use the Bessel function form Γn and Γn. Note, we have Z ∞ 2 0 0 1 Jn (ys) fs0⊥σv⊥dv⊥ = An(asσ, bsσ, 0), (71) 0 πAsσ Z ∞ 2 0 0 1 Jn (ys) fs0⊥σ(v⊥ − vdsr)dv⊥ = An(asσ, bsσ, bsσ), (72) 0 πAsσ Z ∞ 0 02 0 v⊥tsσ Jn(ys)Jn(ys) fs0⊥σv⊥ dv⊥ = Bn(asσ, bsσ, 0), (73) 0 πAsσ Z ∞ 0 0 0 0 v⊥tsσ Jn(ys)Jn(ys) fs0⊥σv⊥(v⊥ − vdsr)dv⊥ = Bn(asσ, bsσ, bsσ), (74) 0 πAsσ Z ∞ 2 0 0 03 0 v⊥tsσ Jn(ys)Jn(ys) fs0⊥σv⊥ dv⊥ = Cn(asσ, bsσ, 0), (75) 0 πAsσ Z ∞ 2 0 0 02 0 0 v⊥tsσ Jn(ys)Jn(ys) fs0⊥σv⊥ (v⊥ − vdsr)dv⊥ = Cn(asσ, bsσ, bsσ). (76) 0 πAsσ We have checked in Matlab, the speed of numerical integral is also fast, compared to using the Bessel function. To short the notation, we would also use such as An(0) = An(asσ, bsσ, 0) and An(bsσ) = An(asσ, bsσ, bsσ). 4 4 4 For short the notations using Anbsσ = An(asσ, bsσ, bsσ), Bnbsσ = Bn(asσ, bsσ, bsσ), Cnbsσ = Cn(asσ, bsσ, bsσ), Asσ Asσ Asσ 4 4 4 An0σ = An(asσ, bsσ, 0), Bn0σ = Bn(asσ, bsσ, 0) and Cn0σ = Cn(asσ, bsσ, 0). Asσ Asσ Asσ 2 R ∞ 2 −(y−b )2 2 R ∞ −(y−b )2 Note also: y e s (y − bs)dy = 1 and ye s dy = 1. As 0 As 0 P∞ (−1)m y 2m+n Expansion An and Bn at a → 0 would be useful. For |y| 1 and n ≥ 0, Jn(y) = m=0 m!Γ(m+n+1) ( 2 ) , with 1 ay n 1 ay n+2 0 1 ay n−1 Γ(n + 1) = n! be the Euler Γ function. Thus, we have Jn(ay) ∼ n! ( 2 ) − (n+1)! ( 2 ) and Jn≥1(ay) ∼ 2(n−1)! ( 2 ) − (n+2) ay n+1 2 2(n+1)! ( 2 ) . To O(a ), we have

 Bs As 2 Z ∞  2 − 4 a , n = 0, 2  A (a, b, b) = J2(ay)e−(y−b) (y − b)dy ∼  As a2, n = ±1, n n  2 0  0, others,

 As Z ∞  − 4 a, n = 0, 2  B (a, b, b) ∼ J (ay)J0 (ay)e−(y−b) y(y − b)dy ∼  As a, n = ±1, n n n  8 0  0, others,  1 2b2+3 2 1 2 Z ∞  As − a As + a Bs, n = 0, 2  2 8 8 2 −(y−b)  2b2+3 2 1 2 An(a, b, 0) = J (ay)e ydy ∼ a A − a B , n = ±1, n  16 s 16 s 0  0, others,  2  − 2b +3 1 Z ∞  8 aAs + 8 aBs, n = 0, 0 −(y−b)2 2  2b2+3 1 Bn(a, b, 0) ∼ Jn(ay)J (ay)e y dy ∼ aA − aB , n = ±1, n  16 s 16 s 0  0, others,

ay 2 ay 1 ay 2 0 ay 0 1 3 ay 2 where we have used J0(ay) ∼ 1 − ( 2 ) , J1(ay) ∼ 2 , J2(ay) ∼ 2 ( 2 ) , Jn≥3 ∼ 0, J0(ay) ∼ −( 2 ), J1(ay) ∼ 2 − 4 ( 2 ) , R ∞ 2 2 √ J0 ay ∼ 1 ay J0 ay ∼ 1 ay 2 J0 ∼ A e−(y−b) ydy e−b πb b B 2( ) 2 ( 2 ), 3( ) 4 ( 2 ) , n≥4 0, and deﬁned s = 2 0 = + [erf( ) + 1], s = R ∞ 2 2 e−(y−b) y − b dy e−b 2 0 ( ) = . 3.4. Electrostatic dispersion relation We derive the electrostatic dispersion relation in this subsection base on Eq.(43) and the distribution function (62). The term ∂ f nω ∂ f X r f nω X r f k s0 + cs s0 = −2(v0 − v )k sσ s0σ − 2(v0 − v ) cs sσ s0σ , k ∂v0 v0 ∂v0 k dsz z 2 ⊥ dsr v0 2 k ⊥ ⊥ σ=a,b vzts ⊥ σ=a,b v⊥tsb 14 and thus 0 ∂ fs0 nωcs ∂ fs0 0 0 0 v⊥ kk 0 + 0 0 − − − − ∂vk v⊥ ∂v⊥ X 2r f (vk vdsz)kzv⊥ (v⊥ vdsr)nωcsλT sσ sσ s0σ , 0 = 2 0 ω − kkv + iν − k v − nω ω − k (v − v ) − k v + iν − k v − nω k s x dsx cs σ=a,b vzts z k dsz z dsz s x dsx cs P and thus the magnetized s=m term in Eq.(43) is

2 Z ∞ Z ∞ ∞ 0 0 0 0 X X ωps X X 4πrsσ f −vkkzv⊥ − (v⊥ − vdsr)nωcsλT sσ ··· = J2(y ) s0σ dv0 dv0 k2 n s 2 ω − k v0 − k v + iν − k v − nω ⊥ k s=m s=m −∞ 0 n=−∞ σ=a,b vzts z k z dsz s x dsx cs 2 Z ∞ Z ∞ ∞ 0 0 0 0 X ωps X X 4πrsσ f vkkzv⊥ + (v⊥ − vdsr)nωcsλT sσ = − J2(y ) s0σ dv0 dv0 k2 n s 2 ω − k v0 ⊥ k s=m −∞ 0 n=−∞ σ=a,b vzts sn z k 2 Z ∞ ∞ X ωps X X 4πr f n Z(ζ )o = J2(y ) sσ s0⊥σ v0 [1 + ζ Z(ζ )] + (v0 − v )nω λ sn dv0 k2 n s 2 ⊥ sn sn ⊥ dsr cs T sσ k v ⊥ s=m 0 n=−∞ σ=a,b vzts z zts 2 ∞ X ωps X X 4r n nω λ o = sσ Z (ζ )A (a , b , 0) + cs T sσ Z (ζ )A (a , b , b ) 2 2 A 1 sn n sσ sσ k v 0 sn n sσ sσ sσ s=m k vzts n=−∞ σ=a,b sσ z zts 2 ∞ X ωps X X n nω λ o = r Z (ζ )A + cs T sσ Z (ζ )A , (77) 2 2 sσ 1 sn n0σ k v 0 sn nbsσ s=m k vzts n=−∞ σ=a,b z zts

We ﬁnd the above result is the same as in Ref.[13] ring-beam case, except that our new (1) ωsn = ω − kzvdsz − nωcs−kxvdsx + iνs, and (2) the summation for loss cone. For the unmagnetized species

∂ fs0 h vdsr i X rsσ fs0σ X rsσ fs0σ k · = −2 1 − kx(vx − vdsx) − 2kz(vz − vdsz) ∂v p 2 2 2 2 (vx − vdsx) + (vy − vdsy) σ=a,b v⊥tsσ σ=a,b vzts 0 0 X h bsσ kx x kzz i = −2 1 − p + rsσ fs0σ 02 02 v⊥ v σ=a,b x + y tsσ zts 0 0 √ X h bsσ kx x kzz i rsσ 1 02 02 02 2 − − e−z −( x +y −bsσ) , = 2 1 p + 2 02 02 v⊥ v A 3/2 σ=a,b x + y tsσ zts sσ π vztsv⊥tsσ where we have chosen the transformation

 0 (v −v )  x = x dsx  0  v⊥tsσ  vx = v⊥tsσ x + vdsx  0 (vy−vdsy)  0 y = , vy = v⊥tsσy + vdsy . (78)  v⊥   tsσ  0  0 (vz−vdsz)   z = vz = vztsz + vdsz vzts Further consider

ω − k · v + iνs = ω − kxvx − kzvz + iνs 0 0 = ω − kxv⊥tsσ x − kzvztsz − kxvdsx − kzvdsz + iνs

= ω − kvtsσ x − kxvdsx − kzvdsz + iνs, (79)

0 0 we need a new transformation to let the two variables kxv⊥tsσ x + kzvztsz change to a single variable kvtsσ x, which is    x = kxv⊥tsσ x0 + kzvzts z0  x0 = kxv⊥tsσ x + kzvzts z  kvtsσ kvtsσ  kvtsσ kvtsσ  0  0  y = y ,  y = y , (80)  k v k v  k v k v  z = z zts x0 − x ⊥tsσ z0  z0 = z zts x − x ⊥tsσ z kvtsσ kvtsσ kvtsσ kvtsσ

15 i.e.,  kxv⊥tsσ kzvzts  vx = v⊥tsσ x + v⊥tsσ z + vdsx  kvtsσ kvtsσ  v = v y + v , dv dv dv = −v2 v dxdydz, (81)  y ⊥tsσ dsy x y z ⊥tsσ zts  kzvzts kxv⊥tsσ  vz = vzts x − vzts z + vdsz kvtsσ kvtsσ q q 2 2 2 2 2 2 where kvtsσ = kxv⊥tsσ + kz vzts = vzts kx/λT sσ + kz . To transform the integral to be able to use the Z function, we −x2 need e term be separate. It is not easy to do so, except only when we set vdsr = 0, i.e., bsσ = 0 and Asσ = 1. At 02 p 02 02 2 02 02 02 2 2 2 vdsr = 0 case, we can have z + ( x + y − bsσ) = x + y + z = x + y + z . Thus, to make life easy, we set 5 vdsr = 0 for unmagnetized study . ω−kxvdsx−kzvdsz+iνs P We deﬁne ζsσ = , and thus ω − kxvx − kzvz + iνs = kvtsσ(ζsσ − x). The unmagnetized term in kvtsσ s=u Eq.(43) is

h 0 0 i 2 2 2 2 P kx x kzz rsσ −(x +y +z ) Z −2 + 2 e , X X ωps σ=a,b v⊥tsσ vzts π3/2v v ··· = dv3 zts ⊥tsσ k2 (ω − k · v + iν ) s=u s=u s h 2 i 2 2 2 2 Z ∞ k kxkz v⊥tsσ − rsσ −(x +y +z ) X ωps X kv x + kv v (λT sσ 1)z π3/2 e , = 2 dxdydz tsσ tsσ zts k2 kv (ζ − x) s=u σ=a,b −∞ tsσ σ 2 Z ∞ −(x2+y2+z2) X ωps X 2r 1 xe , = sσ dxdydz k2 2 π3/2 (ζ − x) s=u σ=a,b vtsσ −∞ σ 2 2 X ωps X 2r X ωps X 2r = sσ [1 + ζ Z(ζ )] = sσ Z (ζ ), (82) k2 2 sσ sσ k2 2 1 sσ s=u σ=a,b vtsσ s=u σ=a,b vtsσ

0 0 2 where we have used kx x + kzz = kx [ kxv⊥tsσ x + kzvzts z] + kz [ kzvzts x − kxv⊥tsσ z] = k x + kxkz vzts z − kxkz v⊥tsσ z = v⊥tsσ vzts v⊥tsσ kvtsσ kvtsσ vzts kvtsσ kvtsσ kvtsσ kvtsσ v⊥tsσ kvtsσ vzts 2 k kxkz v⊥tsσ kxkz v⊥tsσ x + (λT sσ − 1)z. We note that the (λT sσ − 1)z term is integralled to be zero which makes the result kvtsσ kvtsσ vzts kvtsσ vzts simpliﬁed a lot. Thus we obtain the ﬁnal electrostatic dispersion relation

2 ∞ X ωps X X n nω λ o D(ω, k) = 1 + r Z (ζ )A + cs T sσ Z (ζ )A 2 2 sσ 1 sn n0σ k v 0 sn nbsσ s=m k vzts n=−∞ σ=a,b z zts 2 X ωps X 2r + sσ Z (ζ ) = 0. (83) k2 2 1 sσ s=u σ=a,b vtsσ

The above dispersion relation Eq.(83) is very general, except that required vdsr = 0 for unmagnetized species.

3.5. Electromagnetic dispersion relation The electromagnetic case is much complicated than the electrostatic case.

3.5.1. Unmagnetized terms We start from the unmagnetized term ﬁrstly. We use the same transformation for v as in the unmagnetized electrostatic case. Again, we assume vdsr = 0 for unmagnetized species, and calculate term by term: 0 0 √ ∂ fs0 X h kx x kzz i rsσ 02 02 02 2 k · − e−z −( x +y −bsσ) = 2 + 2 ∂v v⊥ v 3/2 σ=a,b tsσ zts π vztsv⊥tsσ 2 X h k k k v i r 2 2 2 = −2 x + x z ⊥tsσ (λ − 1)z sσ e−(x +y +z ), kv kv v T sσ 3/2 2 σ=a,b tsσ tsσ zts π vztsv⊥tsσ

5 It is also rare to meet ring beam in unmagnetized case. Thus, using vdsr = 0 for unmagnetized species will not limit too much to the application of present model. Actually, our present model is much general than Ref.[14] and the non-relativistic case in Ref.[15]. 16 ∂ f X r f X h k p k i r 2 2 2 s0 = −2(v − v ) sσ s0σ = −2 x x + λ z z sσ e−(x +y +z ), ∂v x dsx 2 kv T sσ kv 3/2 2 x σ=a,b v⊥tsσ σ=a,b tsσ tsσ π vztsv⊥tsσ

∂ f X r f X 1 r 2 2 2 s0 − v − v sσ s0σ − y sσ e−(x +y +z ), = 2( y dsx) 2 = 2 2 ∂v v⊥ 3/2 y σ=a,b v⊥tsσ σ=a,b tsσ π vztsv⊥tsσ

∂ f X r f X h k k 1 i r 2 2 2 s0 = −2(v − v ) sσ s0σ = −2 z x − x √ z sσ e−(x +y +z ). ∂v z dsz 2 kv kv 3/2 2 z σ=a,b vzts σ=a,b tsσ tsσ λT sσ π vztsv⊥tsσ

2 2 2 rsσ −(x +y +z ) Deﬁne Fsσ ≡ 3/2 2 e (note: Asσ = 1), we have π vztsv⊥tsσ    vx  ∂ fs0   ∂ fs0 ∂ fs0 ∂ fs0 (v ) =  vy  ( , , ) ∂v   ∂vx ∂vy ∂vz vz    vx  X   h kx p kz i 1 h kz kx 1 i = −2 Fsσ  vy  ( x + λT sσ z , y, x − √ z )   kvtsσ kvtsσ v⊥tsσ kvtsσ kvtsσ λT sσ σ=a,b vz  2 2 k2v2 2 2  kxv⊥tsσ 2 kxvdsx z zts 2 kxkzv⊥tsσ 2 kzvdsx − kxkzv⊥tsσ 2  2 2 x + kv x + 2 2 z 0 2 2 x + kv x 2 2 z  X  k vtsσ tsσ k vtsσ k vtsσ tsσ k vtsσ   kxvdsy 2 kzvdsy  → −2 Fsσ  x y x   kvtsσ kvtsσ   k k v2 k v2 k2v2 2 2  σ=a,b  x z zts 2 kxvdsz kz x zts 2 z zts 2 kzvdsz kxv⊥tsσ 2  2 2 x + x − z 0 2 2 x + x + 2 2 z k vtsσ kvtsσ kvtsσ kvtsσ k vtsσ kvtsσ k vtsσ Note also: √  ∂ fs0   h kx kz i     x + λT sσ z  ∂ f  ∂vx  X  kvtsσ kvtsσ  (v s0 )  ∂ fs0  (v , v , v ) = −2 F  1 y  (v , v , v ) ,  ∂vy  x y z sσ  v⊥tsσ  x y z ∂v    h k i   ∂ fs0  σ=a,b  z x − kx √ 1 z  ∂vz kvtsσ kvtsσ λT sσ  2 2 k2v2 k v k k v2 k v2  kxv⊥tsσ 2 kxvdsx z zts 2 x dsy x z zts 2 kxvdsz − kz x zts 2  2 2 x + kv x + 2 2 z kv x 2 2 x + kv x kv kv z  X  k vtsσ tsσ k vtsσ tsσ k vtsσ tsσ tsσ tsσ   2  → −2 Fsσ  0 y 0   2 2 k v k2v2 2 2  σ=a,b  kxkzv⊥tsσ 2 kzvdsx kxkzv⊥tsσ 2 z dsy z zts 2 kzvdsz kxv⊥tsσ 2  2 2 x + x − 2 2 z x 2 2 x + x + 2 2 z k vtsσ kvtsσ k vtsσ kvtsσ k vtsσ kvtsσ k vtsσ The odd function term from y and z can be integralled to vanish, and thus we have omitted them in the above ﬁnal expressions. Note  kxv⊥tsσ kzvzts  vx = v⊥tsσ x + v⊥tsσ z + vdsx  kvtsσ kvtsσ  v = v y + v , (84)  y ⊥tsσ dsy  kzvzts kxv⊥tsσ  vz = vzts x − vzts z + vdsz kvtsσ kvtsσ and we have used: √ 2 2 2 2 2 2 2 kx kz kxv⊥tsσ kzvzts kx kxv⊥tsσ kz vzts 2 kxv⊥tsσ 2 kxvdsx kz vzts 2 • [ x+ λT sσ z](v⊥tsσ x+v⊥tsσ z+vdsx) → x( x+vdsx)+ 2 2 z = 2 2 x + x+ 2 2 z , kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ k vtsσ k vtsσ kvtsσ k vtsσ h √ i 2 2 kx kz kzvzts kxv⊥tsσ kx kzvzts kz kxvzts 2 kxkzvzts 2 kxvdsz • x + λT sσ z (vzts x − vzts z + vdsz) → x(vzts x + vdsz) − z = 2 2 x + x − kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ k vtsσ kvtsσ 2 kz kxvzts z2, kvtsσ kvtsσ

h k i k v k k v2 k v k k v2 z kx √ 1 kxv⊥tsσ z zts x z ⊥tsσ 2 z dsx x z ⊥tsσ 2 • x − z (v⊥tsσ x + v⊥tsσ z + vdsx) → 2 2 x + x − 2 2 z , kvtsσ kvtsσ λT sσ kvtsσ kvtsσ k vtsσ kvtsσ k vtsσ

h k i k v k2v2 k v k2v2 z kx √ 1 z zts kxv⊥tsσ z zts 2 z dsz x ⊥tsσ 2 • x − z (vzts x − vzts z + vdsz) → 2 2 x + x + 2 2 z . kvtsσ kvtsσ λT sσ kvtsσ kvtsσ k vtsσ kvtsσ k vtsσ And   2  vx  ∂ fs0 X h k kxkz v⊥tsσ i   (k · )(vv) = −2 Fsσ x + (λT sσ − 1)z  vy  (vx, vy, vz) ∂v kvtsσ kvtsσ vzts   σ=a,b vz 17 with terms in vv

kxv⊥tsσ kzvzts 2 • vxvx → [v⊥tsσ x + v⊥tsσ z + vdsx] . kvtsσ kvtsσ

kxv⊥tsσ kzvzts • vxvy = vyvx → [v⊥tsσ x + v⊥tsσ z + vdsx]vdsy. kvtsσ kvtsσ

kxv⊥tsσ kzvzts kzvzts kxv⊥tsσ • vxvz = vzvx → [v⊥tsσ x + v⊥tsσ z + vdsx][vzts x − vzts z + vdsz]. kvtsσ kvtsσ kvtsσ kvtsσ

2 2 2 • vyvy → [v⊥tsσy + vdsy].

kzvzts kxv⊥tsσ • vyvz = vzvy → [vzts x − vzts z + vdsz]vdsy. kvtsσ kvtsσ

kzvzts kxv⊥tsσ 2 • vzvz → [vzts x − vzts z + vdsz] . kvtsσ kvtsσ h i ∂ fs0σ ∂ fs0σ 2 −iνs(v )+(k· )(vv) u P ωps u u n R 3 ∂v ∂v o 1 −(x2+y2+z2) Deﬁne Q = rsσ 2 P , with P = − I + dv , and g = 3/2 e . We calculate s σ=a,b ω s s (ω−k·v+iνs) π term by term, similar to h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = −1+ s11 (ω−k·v+iνs) 11 2 2 2 2 h i kxv⊥tsσ 2 kxvdsx kz vzts 2 k2 kxkz v⊥tsσ kxv⊥tsσ kzvzts 2 −iν ( x + x+ z )+ x+ (λ −1)z [v⊥ x+v⊥ z+v ] s 2 2 kvtsσ 2 2 kvtsσ kvtsσ vzts T sσ tsσ kvtsσ tsσ kvtsσ dsx 2 2 R k vtsσ k vtsσ 2 n kxv⊥tsσ 2 dxdydzg = −1+ −iνs( 2 2 Z2+ kvtsσ(ζsσ−x) kvtsσ k vtsσ 2 2 2 2 2 2 kxvdsx kz vzts k kxv⊥tsσ 2 k kxv⊥tsσ 1 k kzvzts 2 k 2 kxkz v⊥tsσ Z1+ 2 2 Z0)+ (v⊥tsσ ) Z3+2 (v⊥tsσ )vdsxZ2+[ (v⊥tsσ ) + v + (λT sσ− kvtsσ 2k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 kvtsσ kvtsσ kvtsσ dsx kvtsσ vzts kxv⊥tsσ kzvzts kxkz v⊥tsσ kzvzts o 1)(v⊥tsσ )v⊥tsσ ]Z1 + (λT sσ − 1)v⊥tsσ vdsxZ0 kvtsσ kvtsσ kvtsσ vzts kvtsσ n 2 2 2 2 2 2 4 2 2 2 2 2 2 2 kxv⊥tsσ kxvdsx kz vzts k kxv⊥tsσ k kxv⊥tsσ 1 2 k kz vzts k 2 = −1+ −iνs( 2 2 Z2 + Z1 + 2 2 Z0)+ 3 3 Z3 +2 2 2 vdsxZ2 +[ v⊥tsσ 3 3 + v +(λT sσ − kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ k vtsσ 2 k vtsσ kvtsσ dsx 2 2 4 2 kxkz v⊥tsσ kxkz 2 o 1) 3 3 ]Z1 + 2 2 (λT sσ − 1)v⊥tsσvdsxZ0 , k vtsσ k vtsσ we have others

h ∂ f ∂ f i −iν (v s0σ )+(k· s0σ )(vv) u n R 3 s ∂v ∂v o • Ps12 = − I + dv (ω−k·v+iν ) = 0+ h s i 12 k2 kxkz v⊥tsσ kxv⊥tsσ kzvzts −iνs0+ kv x+ kv v (λT sσ−1)z [v⊥tsσ kv x+v⊥tsσ kv z+vdsx]vdsy 2 2 2 R tsσ tsσ zts tsσ tsσ 2 n k kxv⊥tsσ k 2 dxdydzg = vdsyZ2 + vdsxvdsyZ1 + kvtsσ(ζsσ−x) kvtsσ kvtsσ kvtsσ kvtsσ 1 kxkz v⊥tsσ kzvzts o (λT sσ − 1)v⊥tsσ vdsyZ0 2 kvtsσ vzts kvtsσ n 2 2 2 2 o 2vdsy k kxv⊥tsσ k 1 kxkz 2 = 2 2 Z2 + vdsxZ1 + 2 2 (λT sσ − 1)v⊥tsσZ0 , kvtsσ k vtsσ kvtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = 0+ s21 (ω−k·v+iνs) 21 k v h 2 i − x dsy k kxkz v⊥tsσ − kxv⊥tsσ kzvzts R iνs kv x+ kv x+ kv v (λT sσ 1)z [v⊥tsσ kv x+v⊥tsσ kv z+vdsx]vdsy 2 dxdydzg tsσ tsσ tsσ zts tsσ tsσ kvtsσ(ζsσ−x) 2 2 2 2 n kxvdsy k kxv⊥tsσ k 1 kxkz v⊥tsσ kzvzts o = − iνs Z1 + vdsyZ2 + vdsxvdsyZ1 + (λT sσ − 1)v⊥tsσ vdsyZ0 kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ n 2 2 2 2 o 2vdsy kx k kxv⊥tsσ k 1 kxkz 2 = − iνs Z1 + 2 2 Z2 + vdsxZ1 + 2 2 (λT sσ − 1)v⊥tsσZ0 , kvtsσ kvtsσ k vtsσ kvtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = −1+ s22 (ω−k·v+iνs) 22 h 2 i − 2 k kxkz v⊥tsσ − 2 2 2 R iνsy + kv x+ kv v (λT sσ 1)z [v⊥tsσy +vdsy] 2 dxdydzg tsσ tsσ zts kvtsσ(ζsσ−x) 2 n 1 1 2 2 k2 o = −1+ − iνs Z0 + ( v + v ) Z1 , kvtsσ 2 2 ⊥tsσ dsy kvtsσ

18 h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o R • Pu = − I + dv3 = 0+2 dxdydzg s13 (ω−k·v+iνs) 13 2 2 h i 2 2 kxkzv⊥tsσ 2 kzvdsx kxkzv⊥tsσ 2 k2 kxkz v⊥tsσ kxv⊥tsσ kzvzts kzvzts kxv⊥tsσ −iν ( x + x− z )+ x+ (λ −1)z [ x+v⊥ z+v ][ x−v z+v ] s 2 2 kvtsσ 2 2 kvtsσ kvtsσ vzts T sσ kvtsσ tsσ kvtsσ dsx kvtsσ zts kvtsσ dsz k vtsσ k vtsσ kvtsσ(ζsσ−x) 2 2 2 2 2 2 2 2 2 2 n kxkzv⊥tsσ kzvdsx kxkzv⊥tsσ k kxv⊥tsσ kzvzts k kxv⊥tsσ kzvzts k = − iνs( 2 2 Z2 + Z1 − 2 2 Z0) + Z3 + [ vdsz + vdsx ]Z2 + [ vdsxvdsz − kvtsσ k vtsσ kvtsσ 2k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 1 k kzvzts kxv⊥tsσ 1 kxkz v⊥tsσ kxv⊥tsσ kxv⊥tsσ 1 kxkz v⊥tsσ kzvzts kzvzts 1 kxkz v⊥tsσ v⊥tsσ vzts − (λT sσ−1) vzts + (λT sσ−1)v⊥tsσ ]Z1+[ (λT sσ− 2 kvtsσ kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ kvtsσ 2 kvtsσ vzts kzvzts 1 kxkz v⊥tsσ kxv⊥tsσ o 1)v⊥tsσ vdsz − (λT sσ − 1)vdsxvzts ]Z0 kvtsσ 2 kvtsσ vzts kvtsσ 2 2 2 2 2 2 2 2 2 2 n kxkzv⊥tsσ kzvdsx kxkzv⊥tsσ k kxkzvztsv⊥tsσ k kxv⊥tsσ kzvzts k = − iνs( 2 2 Z2 + Z1 − 2 2 Z0) + 3 3 Z3 + [ vdsz + vdsx ]Z2 + [ vdsxvdsz − kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ −1) 3 3 + (λT sσ −1) 3 3 ]Z1 +[ (λT sσ −1) 2 2 vdsz − (λT sσ −1)vdsx 2 2 ]Z0 , 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o R • Pu = − I + dv3 = 0+2 dxdydzg s31 (ω−k·v+iνs) 31 2 2 h i 2 2 kxkzvzts 2 kxvdsz kxkzvzts 2 k2 kxkz v⊥tsσ kxv⊥tsσ kzvzts kzvzts kxv⊥tsσ −iν ( x + x− z )+ x+ (λ −1)z [ x+v⊥ z+v ][ x−v z+v ] s 2 2 kvtsσ 2 2 kvtsσ kvtsσ vzts T sσ kvtsσ tsσ kvtsσ dsx kvtsσ zts kvtsσ dsz k vtsσ k vtsσ kvtsσ(ζsσ−x) 2 2 2 2 2 2 2 2 2 2 n kxkzvzts kxvdsz kxkzvzts k kxv⊥tsσ kzvzts k kxv⊥tsσ kzvzts k = − iνs( 2 2 Z2 + Z1 − 2 2 Z0) + Z3 + [ vdsz + vdsx ]Z2 + [ vdsxvdsz − kvtsσ k vtsσ kvtsσ 2k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 1 k kzvzts kxv⊥tsσ 1 kxkz v⊥tsσ kxv⊥tsσ kxv⊥tsσ 1 kxkz v⊥tsσ kzvzts kzvzts 1 kxkz v⊥tsσ v⊥tsσ vzts − (λT sσ−1) vzts + (λT sσ−1)v⊥tsσ ]Z1+[ (λT sσ− 2 kvtsσ kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ kvtsσ 2 kvtsσ vzts kzvzts 1 kxkz v⊥tsσ kxv⊥tsσ o 1)v⊥tsσ vdsz − (λT sσ − 1)vdsxvzts ]Z0 kvtsσ 2 kvtsσ vzts kvtsσ 2 2 2 2 2 2 2 2 2 2 n kxkzvzts kxvdsz kxkzvzts k kxkzvztsv⊥tsσ k kxv⊥tsσ kzvzts k = − iνs( 2 2 Z2 + Z1 − 2 2 Z0) + 3 3 Z3 + [ vdsz + vdsx ]Z2 + [ vdsxvdsz − kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ −1) 3 3 + (λT sσ −1) 3 3 ]Z1 +[ (λT sσ −1) 2 2 vdsz − (λT sσ −1)vdsx 2 2 ]Z0 , 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = 0+ s23 (ω−k·v+iνs) 23 k v h 2 i − z dsy k kxkz v⊥tsσ − kzvzts − kxv⊥tsσ R iνs kv x+ kv x+ kv v (λT sσ 1)z [vzts kv x vzts kv z+vdsz]vdsy 2 dxdydzg tsσ tsσ tsσ zts tsσ tsσ kvtsσ(ζsσ−x) 2 2 2 n kzvdsy k kzvzts k 1 kxkz v⊥tsσ kxv⊥tsσ o = − iνs Z1 + vzts vdsyZ2 + vdszvdsyZ1 − (λT sσ − 1)vzts vdsyZ0 kvtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ 2 2 2 2 2 2vdsy n kz k kzvzts k 1 kxkzv⊥tsσ o = − iνs Z1 + 2 2 Z2 + vdszZ1 − (λT sσ − 1) 2 2 Z0 , kvtsσ kvtsσ k vtsσ kvtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = 0+ s32 (ω−k·v+iνs) 32 h 2 i − k kxkz v⊥tsσ − kzvzts − kxv⊥tsσ R iνs0+ kv x+ kv v (λT sσ 1)z [vzts kv x vzts kv z+vdsz]vdsy 2 dxdydzg tsσ tsσ zts tsσ tsσ kvtsσ(ζsσ−x) 2 2 2 n k kzvzts k 1 kxkz v⊥tsσ kxv⊥tsσ o = vzts vdsyZ2 + vdszvdsyZ1 − (λT sσ − 1)vzts vdsyZ0 kvtsσ kvtsσ kvtsσ kvtsσ 2 kvtsσ vzts kvtsσ 2 2 2 2 2 2vdsy n k kzvzts k 1 kxkzv⊥tsσ o = 2 2 Z2 + vdszZ1 − (λT sσ − 1) 2 2 Z0 , kvtsσ k vtsσ kvtsσ 2 k vtsσ h i − ∂ fs0σ · ∂ fs0σ n R iνs(v ∂v )+(k ∂v )(vv) o • Pu = − I + dv3 = −1+ s33 (ω−k·v+iνs) 33 k2v2 k v k2v2 h 2 i −iν ( z zts x2+ z dsz x+ x ⊥tsσ z2)+ k x+ kxkz v⊥tsσ (λ −1)z [v kzvzts x−v kxv⊥tsσ z+v ]2 s 2 2 kvtsσ 2 2 kvtsσ kvtsσ vzts T sσ zts kvtsσ zts kvtsσ dsz 2 2 R k vtsσ k vtsσ 2 n kz vzts 2 dxdydzg = −1 + − iνs( 2 2 Z2 + kvtsσ(ζsσ−x) kvtsσ k vtsσ 2 2 2 2 2 2 kzvdsz kxv⊥tsσ k kzvzts 2 k kzvzts 1 k kxv⊥tsσ 2 k 2 kxkz v⊥tsσ Z1 + 2 2 Z0) + (vzts ) Z3 + 2 vzts vdszZ2 + [ (vzts ) + v − (λT sσ − kvtsσ 2k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 kvtsσ kvtsσ kvtsσ dsz kvtsσ vzts kzvzts kxv⊥tsσ kxkz v⊥tsσ kxv⊥tsσ o 1)vzts vzts ]Z1 − (λT sσ − 1)vzts vdszZ0 kvtsσ kvtsσ kvtsσ vzts kvtsσ n 2 2 2 2 2 2 4 2 2 2 2 2 2 2 2 kz vzts kzvdsz kxv⊥tsσ k kz vzts k kzvzts 1 k kxvztsv⊥tsσ k 2 = −1 + − iνs( 2 2 Z2 + Z1 + 2 2 Z0) + 3 3 Z3 + 2 2 2 vdszZ2 + [ 3 3 + v − (λT sσ − kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ k vtsσ 2 k vtsσ kvtsσ dsz 2 2 2 2 2 2 kxkz vztsv⊥tsσ kxkzv⊥tsσ o 1) 3 3 ]Z1 − (λT sσ − 1) 2 2 vdszZ0 , k vtsσ k vtsσ where the argument for Z0,1,2,3 is ζsσ. 19 3 3 0 0 2 2 1 Note, with the ’→’ means that to integral out the x, y and z terms, say, y, z → 0, y , z → 0, y , z → 1, y , z → 2 , p x → Zp(ζsσ)), we have used • (ax + bz)(cx + dz + e)2 = ac2 x3 + 2acdx2z + 2acex2 + ad2 xz2 + 2adexz + ae2 x + bc2 x2z + 2bcdxz2 + 2bcexz + 2 3 2 2 2 1 2 2 bd z + 2bdez + be z → ac Z3 + 2aceZ2 + [ 2 ad + ae + bcd]Z1 + bdeZ0. 2 2 1 • (ax + bz)(cx + dz + e) f = ac f x + ad f xz + ae f x + bc f xz + bd f z + be f z → ac f Z2 + ae f Z1 + 2 bd f Z0. 2 2 1 • (ax + bz)(cx − dz + e) f = ac f x − ad f xz + ae f x + bc f xz − bd f z + be f z → ac f Z2 + ae f Z1 − 2 bd f Z0. 2 2 2 1 • (ax + bz)(cy + d) = acxy + adx + bcy z + bdz → [ 2 ac + ad]Z1. • (ax + bz)(cx + dz + e)( f x − gz + h) = ac f x3 − acgx2z + achx2 + ad f x2z − adgxz2 + adhxz + ae f x2 − aegxz + 2 2 2 3 2 2 aehx + bc f x z − bcgxz + bchxz + bd f xz − bdgz + bdhz + be f xz − begz + behz → ac f Z3 + [ach + ae f ]Z2 + 1 1 1 1 1 [aeh − 2 adg − 2 bcg + 2 bd f ]Z1 + [ 2 bdh − 2 beg]Z0. • (ax + bz)(cx − dz + e)2 = ac2 x3 − 2acdx2z + 2acex2 + ad2 xz2 − 2adexz + ae2 x + bc2 x2z − 2bcdxz2 + 2bcexz + 2 3 2 2 2 1 2 2 bd z − 2bdez + be z → ac Z3 + 2aceZ2 + [ 2 ad + ae − bcd]Z1 − bdeZ0. Although the above expressions are much complicated and need be carefully, they can be solved easily by PASS-K matrix approach. We have also checked that, when νs = 0, λT sσ = 1 and vdsz = vdsy = 0, the result can reduce to exactly the same form as in Ref.[14] .

3.5.2. Magnetized terms Next, we calculate the magnetized terms. 2 0 0 0 0 0 00 m P ωps m m P∞ R ∞ R ∞ 2πv⊥dv⊥dvk P∞ R ∞ R ∞ 2πv⊥dv⊥dvk Deﬁne Qs = σ=a,b rsσ 2 Ps , with Ps = n=−∞ 0 Πsσ = n=−∞ 00 Πsσ ω −∞ 0 (ω−kkvk−kxvdsx+iνs−nωcs) −∞ 0 (ωsn−kzvk ) and ωsn = ω − kzvdsz − nωcs−kxvdsx + iνs, i.e., ω = ωsn + kzvdsz + nωcs + kxvdsx − iνs. We have also

0 0 00 ∂ f 2(vk − vdsz) (vk − vdsz) vk s0σ − f − f − f , 0 = 2 s0σ = 2 2 s0σ = 2 2 s0σ ∂vk vzts vzts vzts ∂ f (v0 − v ) s0σ − ⊥ dsr f , 0 = 2 2 s0σ ∂v⊥ v⊥tsσ

v002 0 2 00 0 1 h k (v −vdsr)⊥ i 0 00 with vk = vk −vdsz, and fs0σ = 3/2 2 exp − 2 − 2 , i.e., vk = vk +vdsz. Since we have deﬁned An, Bn, Cn, π vztsv⊥tsσ Asσ vzts v⊥ts 0 we do not need transform v⊥. Thus,

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs Anbsσ Z0 nωcs Z1 nωcs • P = n=−∞ 00 Πsσ11 = n=−∞ [ (nωcs + kxvdsx − iνs) 2 + An0σ( + vdsx) 2 ]( + s11 −∞ 0 (ωsn−kzvk ) kx v⊥tsσ kzvzts kx vzts kx o P nωcs nωcs vdsx) − n Anbsσ 2 ( + vdsx). kxv⊥tsσ kx

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs Anbsσ Z0 Z1 nωcs • P = 00 Πsσ12 = vdsy( +vdsx)(nωcs 2 +An0σ 2 )+i( +vdsx)[(nωcs − s12 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kx v kzvzts v kx o k ⊥tsσ zts Bnbsσ Z0 Bn0σv⊥tsσZ1 iνs) + 2 ] . v⊥tsσ kzvzts vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs 1 Z0 • P = 00 Πsσ21 = (nωcs + kxvdsx − iνs)[−iv⊥tsσBnbsσ + vdsyAnbsσ] 2 + s21 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kx v kzvzts k o ⊥tsσ nωcs Z1 [−iv⊥tsσBn0σ + vdsyAn0σ]( + vdsx) 2 . kx vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs Z0 • P = 00 Πsσ22 = vdsy[−iv⊥tsσBnbsσ + vdsyAnbsσ] 2 + vdsy[−iv⊥tsσBn0σ + s22 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ v kzvzts k ⊥tsσ o Z1 1 Z0 Z1 P vdsy vdsyAn0σ] 2 +v⊥tsσ[v⊥tsσCnbsσ+ivdsyBnbsσ](nωcs−iνs) 2 +v⊥tsσ[v⊥tsσCn0σ+ivdsyBn0σ] 2 − n(Cnbsσ + i Bnbsσ). vzts v⊥tsσ kzvzts vzts v⊥tsσ

20 ∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs Z1 Z0 Anbsσ • P = 00 Πsσ13 = ( + vdsx)[nωcs( + vdsz ) 2 + An0σ(vztsZ2 + (kzvdsz − s13 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kx kz kzvzts v o k ⊥tsσ Z1 1 iνs) ) 2 ] . kz vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ nωcs Anbsσ Z1 Z0 nωcs • P = 00 Πsσ31 = (nωcs + kxvdsx − iνs) 2 ( + vdsz ) + An0σ( + s31 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kx v kz kzvzts kx o k ⊥tsσ Z2 Z1 vdsx)( + vdsz 2 ) . vzts vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ Z1 Z0vdsz nωcs • P = 00 Πsσ23 = [vdsyAnbsσ−iv⊥tsσBnbsσ][( + ) 2 ]+[vdsyAn0σ−iv⊥tsσBn0σ][vztsZ2+ s23 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kz kzvtsz v o k ⊥tsσ Z1 1 (kzvdsz − iνs) ] 2 . kz vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ Anbsσ Z1 kzvztsZ2 Anbsσ Z0 Z1 • P = 00 Πsσ32 = [nωcs 2 +An0σ 2 ]vdsy+[nωcs 2 +An0σ 2 ]vdsyvdsz+ s32 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ v kz v v kzvzts v k ⊥tsσ zts o ⊥tsσ zts Bnbsσ Z1 v⊥tsσZ2 Bnbsσ Z0 v⊥tsσZ1 i[(nωcs − iνs) + Bn0σ ] + ivdsz[(nωcs − iνs) + Bn0σ 2 ] . v⊥tsσ kz vzts v⊥tsσ kzvzts vzts

∞ ∞ 2πv0 dv0 dv00 n m P∞ R R ⊥ ⊥ k P∞ vztsZ2 Z1 Anbsσ Z2 • P = 00 Πsσ33 = [nωcs( + vdsz ) 2 + An0σ(Z3 + (kzvdsz − iνs) )] + s33 n=−∞ −∞ 0 (ωsn−kzv ) n=−∞ kz kz v kzvzts k o ⊥tsσ Z1 Z0 Anbsσ Z2 Z1 P 1 vdsz[nωcs( + vdsz ) 2 + An0σ( + (kzvdsz − iνs) 2 )] − n An0σ. kz kzvzts v⊥tsσ vzts kzvzts 2

P∞ 0 P∞ 2 P∞ 0 where the argument for Z0,1,2,3 is ζsn, and we have used n=−∞ Jn Jn = 0, n=−∞ nJn = 0, n=−∞ nJn Jn = 0. If we 2 2 P∞ 2 P∞ 0 2 1 P∞ n Jn (x) 1 P m m m further use , n=−∞ Jn = 1, n=−∞(Jn) = 2 , n=−∞ x2 = 2 , the last n terms in Ps11, Ps22 and Ps33 yields to 1. 0 00 0 00 We derive the above equation base on (deﬁne ωsn = ωsn − kzvk , i.e., ωsn = ωsn + kzvk )

0 0 2 nωcs 0 ∂ fs0σ 0 nωcs ∂ fs0σ nωcs 2 nωcs 0 (v⊥−vdsr) • v⊥Πsσ11 = Jn [ (ω−kzvk) 0 +v⊥kz( +vdsx) 0 ]( +vdsx) → Jn [ (ωsn +nωcs +kxvdsx −iνs) 2 + kx ∂v⊥ kx ∂vk kx kx v⊥tsσ v00 0 k v00 0 nωcs k nωcs 2 nωcs (v⊥−vdsr) 0 nωcs z k nωcs v⊥kz( + vdsx) 2 ]( + vdsx) → Jn [ (nωcs + kxvdsx − iνs) 2 + v⊥( + vdsx) 2 ]( + vdsx) + kx vzts kx kx v⊥tsσ kx vzts kx 0 2 nωcs (v⊥−vdsr) nωcs 0 Jn 2 ( + vdsx)(ωsn). kx v⊥tsσ kx

0 2 nωcs ∂ fs0σ 0 ∂ fs0σ 0 0 nωcs 0 ∂ fs0σ 0 ∂ fs0σ • v⊥Πsσ12 = Jn vdsy( + vdsx)(nωcs 0 + v⊥kz 0 ) + iv⊥ Jn Jn( + vdsx)[(ω − kzv − kxvdsx) 0 + kzv⊥ 0 ] → kx ∂v⊥ ∂vk kx k ∂v⊥ ∂vk 0 v00 0 v00 2 nωcs (v⊥−vdsr) 0 k 0 0 nωcs 0 (v⊥−vdsr) 0 k 2 nωcs Jn vdsy( +vdsx)(nωcs 2 +v⊥kz 2 )+iv⊥ Jn Jn( +vdsx)[(ωsn+nωcs−iνs) 2 +kzv⊥ 2 ] → Jn vdsy( + kx v⊥tsσ vzts kx v⊥tsσ vzts kx 0 k v00 0 k v00 0 (v⊥−vdsr) 0 z k 0 0 nωcs (v⊥−vdsr) 0 z k 0 0 nωcs (v⊥−vdsr) 0 vdsx)(nωcs 2 +v⊥ 2 )+iv⊥ Jn Jn( +vdsx)[(nωcs −iνs) 2 +v⊥ 2 ]+iv⊥ Jn Jn( +vdsx) 2 (ωsn). v⊥tsσ vzts kx v⊥tsσ vzts kx v⊥tsσ

0 0 0 2 nωcs 0 ∂ fs0σ 0 nωcs ∂ fs0σ 0 ∂ fs0σ 0 0 2 nωcs 0 • v⊥Πsσ21 = (−iv⊥ Jn Jn+vdsy Jn )[ (ω−kzv ) 0 +v⊥ kz 0 +v⊥kzvdsx 0 ] → (−iv⊥ Jn Jn+vdsy Jn )[ (ωsn+ kx k ∂v⊥ kx ∂vk ∂vk kx 0 v00 0 (v⊥−vdsr) 0 nωcs k 0 0 2 nωcs (v⊥−vdsr) 0 nωcs nωcs +kxvdsx −iνs) 2 +v⊥kz( +vdsx) 2 ] → (−iv⊥ Jn Jn +vdsy Jn )[ (nωcs +kxvdsx −iνs) 2 +v⊥( + v⊥tsσ kx vzts kx v⊥tsσ kx k v00 0 z k 0 0 2 nωcs (v⊥−vdsr) 0 vdsx) 2 ] + (−iv⊥ Jn Jn + vdsy Jn ) 2 (ωsn). vzts kx v⊥tsσ

0 0 0 ∂ fs0σ 0 ∂ fs0σ 0 0 0 0 0 ∂ fs0σ • v⊥Πsσ22 = Jnvdsy(−iv⊥ Jn + vdsy Jn)(nωcs 0 + v⊥kz 0 ) + iv⊥ Jn(−iv⊥ Jn + vdsy Jn)[(ω − kzv − kxvdsx) 0 + ∂v⊥ ∂vk k ∂v⊥ 0 v00 0 0 ∂ fs0 0 0 (v⊥−vdsr) 0 k 0 0 0 0 0 (v⊥−vdsr) kzv⊥ 0 ] → Jnvdsy(−iv⊥ Jn + vdsy Jn)(nωcs 2 + v⊥kz 2 ) + iv⊥ Jn(−iv⊥ Jn + vdsy Jn)[(ωsn + nωcs − iνs) 2 + ∂vk v⊥tsσ vzts v⊥tsσ v00 0 k v00 0 k v00 0 k 0 0 (v⊥−vdsr) 0 z k 0 0 0 0 (v⊥−vdsr) 0 z k kzv⊥ 2 ] → Jnvdsy(−iv⊥ Jn +vdsy Jn)(nωcs 2 +v⊥ 2 )+iv⊥ Jn(−iv⊥ Jn +vdsy Jn)[(nωcs −iνs) 2 +v⊥ 2 ]+ vzts v⊥tsσ vzts v⊥tsσ vzts 0 0 0 0 0 (v⊥−vdsr) 0 iv⊥ Jn(−iv⊥ Jn + vdsy Jn) 2 (ωsn). v⊥tsσ

0 0 2 nωcs 0 ∂ fs0σ 0 ∂ fs0σ 2 nωcs 00 (v⊥−vdsr) • v⊥Πsσ13 = Jn ( + vdsx)[nωcsvk 0 + v⊥(ω − nωcs − kxvdsx) 0 ] → Jn ( + vdsx)[nωcs(vk + vdsz) 2 + kx ∂v⊥ ∂vk kx v⊥tsσ v00 0 v00 0 0 00 k 2 nωcs 00 (v⊥−vdsr) 0 00 k 2 nωcs v⊥(ωsn + kzvk + kzvdsz − iνs) 2 ] → Jn ( + vdsx)[nωcs(vk + vdsz) 2 + v⊥(kzvk + kzvdsz − iνs) 2 ] + Jn ( + vzts kx v⊥tsσ vzts kx 00 0 vk 0 vdsx)v⊥ 2 (ωsn). vzts

21 0 0 2 nωcs 0 ∂ fs0σ 0 nωcs ∂ fs0σ 0 ∂ fs0σ 00 2 nωcs 0 • v⊥Πsσ31 = v Jn [ (ω − kzv ) 0 + v⊥ kz 0 + v⊥kzvdsx 0 ] → (v + vdsz)Jn [ (ωsn + nωcs + kxvdsx − k kx k ∂v⊥ kx ∂vk ∂vk k kx 0 v00 0 k v00 (v⊥−vdsr) 0 nωcs k 00 2 nωcs (v⊥−vdsr) 0 nωcs z k 00 iνs) 2 + v⊥kz( + vdsx) 2 ] → (vk + vdsz)Jn [ (nωcs + kxvdsx − iνs) 2 + v⊥( + vdsx) 2 ] + (vk + v⊥tsσ kx vzts kx v⊥tsσ kx vzts 0 2 nωcs (v⊥−vdsr) 0 vdsz)Jn 2 (ωsn). kx v⊥tsσ

0 0 2 0 0 0 ∂ fs0σ 0 ∂ fs0σ 2 0 0 00 (v⊥−vdsr) • v⊥Πsσ23 = (vdsy Jn −iv⊥ Jn Jn)[nωcsvk 0 +v⊥(ω−nωcs−kxvdsx) 0 ] → (vdsy Jn −iv⊥ Jn Jn)[nωcs(vk +vdsz) 2 + ∂v⊥ ∂vk v⊥tsσ v00 0 v00 0 0 00 k 2 0 0 00 (v⊥−vdsr) 0 00 k v⊥(ωsn + kzvk + kzvdsz − iνs) 2 ] → (vdsy Jn − iv⊥ Jn Jn)[nωcs(vk + vdsz) 2 + v⊥(kzvk + kzvdsz − iνs) 2 ] + vzts v⊥tsσ vzts 00 2 0 0 0 vk 0 (vdsy Jn − iv⊥ Jn Jn)v⊥ 2 (ωsn). vzts

0 0 0 2 ∂ fs0σ 0 ∂ fs0 0 0 0 0 ∂ fs0 0 ∂ fs0σ 00 2 (v⊥−vdsr) • v⊥Πsσ32 = vk Jn (nωcs 0 +v⊥kz 0 )vdsy+ivkv⊥ Jn Jn[(ω−kzvk−kxvdsx) 0 +kzv⊥ 0 ] → (vk +vdsz)Jn (nωcs 2 + ∂v⊥ ∂vk ∂v⊥ ∂vk v⊥tsσ v00 0 v00 0 k v00 0 k 00 0 0 0 (v⊥−vdsr) 0 k 00 2 (v⊥−vdsr) 0 z k v⊥kz 2 )vdsy +i(vk +vdsz)v⊥ Jn Jn[(ωsn +nωcs −iνs) 2 +kzv⊥ 2 ] → (vk +vdsz)Jn (nωcs 2 +v⊥ 2 )vdsy + vzts v⊥tsσ vzts v⊥tsσ vzts 0 k v00 0 00 0 0 (v⊥−vdsr) 0 z k 00 0 0 (v⊥−vdsr) 0 i(vk + vdsz)v⊥ Jn Jn[(nωcs − iνs) 2 + v⊥ 2 ] + i(vk + vdsz)v⊥ Jn Jn 2 (ωsn). v⊥tsσ vzts v⊥tsσ

0 0 0 2 0 ∂ fs0σ 0 ∂ fs0σ 00 2 00 (v⊥−vdsr) 0 0 • v⊥Πsσ33 = vk Jn [nωcsvk 0 + v⊥(ω − nωcs − kxvdsx) 0 ] → (vk + vdsz)Jn [nωcs(vk + vdsz) 2 + v⊥(ωsn + ∂v⊥ ∂vk v⊥tsσ v00 0 v00 (v00+v )v00 00 k 00 2 00 (v⊥−vdsr) 0 00 k 0 2 k dsz k 0 kzvk + kzvdsz − iνs) 2 ] → (vk + vdsz)Jn [nωcs(vk + vdsz) 2 + v⊥(kzvk + kzvdsz − iνs) 2 ] + v⊥ Jn 2 (ωsn). vzts v⊥tsσ vzts vzts where ’→’ means we have omitted the coefficient ’−2 fs0σ’. m We find by set νs = 0, vdsx = vdsy = 0, these Psi j reduce exactly the same one as in Ref.[12] ring beam case. By set νs = 0, vdsy = vdsr = 0, we have also checked that they can reduce to the one in Ref.[11] for drift across magnetic field case.

3.5.3. Final Form u The ﬁnal form of the electromagnetic dispersion relation is the combine of the above unmagnetized Ps and mag- m netized Ps terms to Q in Eq.(56), and then to Eq.(48).

4. Transform to PASS-K matrix Equation

The conventional root ﬁnding approach to solve the above dispersion relations can only give one solution at one time and heavily depends on initial guess. The Cauchy contour integral approach [22] can locate all the solutions in a selected complex domain, however which still can not give all the important solutions and is also diﬃcult for complicated dispersion relation. To solve the dispersion relation using PASS-K matrix approach [1], which can give all the important solution at one time, we need two further steps: • (1) Do J-pole expansion of Z function, i.e.,

J X b j Z(ζ) ' Z (ζ) = , (85) J ζ − c j=1 j

where b j and c j are constants for given J, as given in Ref.[1] for kts > 0; • (2) Do linear transformation to a equivalent matrix eigenvalue problem. The standard eigenvalue library can solve all the eigenvalues of a matrix. The first step with J = 8 has been used well for more than thirty years in WHAMP [2] code; The second step is firstly developed in the first version of PASS-K/PDRK code [1]. We derive the corresponding equations step by step. P P P 2 PJ 3 We note: j b j = −1, j b jc j = 0, j b jc j = −1/2 and j=1 b jc j = 0 [8]. And also 1 b b 1 1 = − . ω ω − c c ω − c ω

22 ω−kcs For ζs = , we have kts

 PJ b j  Z(ζs) ' kts  j=1 ω−kcs−ktsc j J J  b X b j X ktsb j  − kts PJ j Z0(ζs, kts) =  ω−k = kts j=1 ω−k = kts = ,  cs cs ω − kcs − |kts|c j ω − cs j  PJ b j j=1 j=1  −Z(−ζs) ' kts j=1 ω−kcs+ktsc j which is written to one compact form for both kts > 0 and kts ≤ 0, with cs j = kcs + |kts|c j. And

J J J X b j X h b jc j i X b j|kts|c j Z (ζ ) = 1 + ζ Z (ζ ) ' 1 + ζ = 1 + b + = , 1 s s 0 s s ζ − c j ζ − c ω − c j=1 s j j=1 s j j=1 s j

PJ where we have used j=1 b j = −1. And

J J ω − k X b j|kts|c j |k | X b jc j(ω − kcs) Z (ζ ) = ζ (1 + ζ Z ) ' cs = ts 2 s s s 0 k ω − c k ω − c ts j=1 s j ts j=1 s j J J 2 2 |k | X h b jc j(cs j − kcs)i 1 X b j|kts| c j = ts b c + = , k j j ω − c k ω − c ts j=1 s j ts j=1 s j

PJ where we have used j=1 b jc j = 0. And

J 2 J 2 1 1 ω − k X b jc j 1 X b jc j (ω − kcs) Z (ζ ) = + ζ2(1 + ζ Z ) ' + cs k = + 3 s 2 s s 0 2 k ts ω − c 2 ω − c ts j=1 s j j=1 s j J 2 J 3 3 1 X h b jc j (cs j − kcs)i 1 X b j|kts| c j = + b c2 + = , 2 j j ω − c 2 ω − c j=1 s j kts j=1 s j

PJ 2 where we have used j=1 b jc j = −1/2. In the above Z0,1,2,3 and cs j, we ﬁnd for kts < 0 it could be simply by doing the variables change with: |kts| → kts and c j → −c j. Thus, in the later usage, to simplify the notation, all |kts| is changed to be kts, and the meaning of c j for kts < 0 is the −c j in the default ZJ. Thus for both kts > 0 and kts ≤ 0, we have a single compact form

J p X b jc j Z (ζ ) ' k , p s ts ω − c j=1 s j with cs j = kcs + ktsc j. And the only change is that c j = c j0 for kts ≥ 0 and c j = −c j0 for kts < 0 , with c j0 be the c j for kts > 0. ω−kzvdsz−nωcs−kxvdsx+iνs ω−kxvdsx−kzvdsz+iνs Typically, for magnetized species ζsn = k v and unmagnetized species ζsσ = kv , with q z zts tsσ 2 2 2 2 kvtsσ = kxv⊥tsσ + kz vzts (which always ≥ 0), we have corresponding kts = kzvzts and kvtsσ, respectively. For cs j: csn j = kzvdsz + nωcs + kxvdsx − iνs + kzvztsc j, and cs jσ = kzvdsz + kxvdsx − iνs + kvtsσc j.

23 4.1. The Electrostatic case We solve Eq.(83) for electrostatic case. The corresponding linear transformation is straightforward and simple.

2 ∞ J J X ωps X X n X b j nω λ X b j o D(ω, k) ' 1 + r [1 + ζ ]A + cs T sσ A 2 2 sσ sn ζ − c n0σ k v ζ − c nbsσ s=m k vzts n=−∞ σ=a,b j=1 sn j z zts j=1 sn j 2 J X ωps X 2r X b j + sσ [1 + ζ ] k2 2 sσ ζ − c s=u σ=a,b vtsσ j=1 sσ j

2 ∞ J AnbsσnωcsλT sσ 2 J X ωps X X X An0σb jc j + k v b j X ωps X 2r X b jc j = 1 + r z zts + sσ 2 2 sσ ζ − c k2 2 ζ − c s=m k vzts n=−∞ σ=a,b j=1 sn j s=u σ=a,b vtsσ j=1 sσ j m 2 u 2 X rsσωps An0σkzvztsb jc j + AnbsσnωcsλT sσb j X 2rsσωps b jc j = 1 + + 2 2 ω − c kv ω − c snσ j k vzts sn j s jσ tsσ s jσ m u X bsn j X bs jσ = 1 + + = 0. (86) ω − c ω − c sn j sn j s jσ s jσ q q ω−kzvdsz−nωcs−kxvdsx+iνs ω−kxvdsx−kzvdsz+iνs 2 2 2 2 2 2 Notation: ζsn = , ζsσ = , kvtsσ = k v + k v = vzts k /λT sσ + k . kzvzts kvtsσ x ⊥tsσ z zts x z 2 P rsσωps Deﬁne: csn j = kzvdsz + nωcs + kxvdsx − iνs + kzvztsc j, bsn j = σ 2 2 (An0σkzvztsb jc j + AnbsσnωcsλT sσb j), cs jσ = k vzts 2 2rsσωps kzvdsz + kxvdsx − iνs + kvtsσc j, and bs jσ = b jc j. kvtsσ The equivalent linear system can be

ωns j = cs jns j + bs jE, X E = − ns j, (87) s j or sparse matrix one

ωns j = cs jns j + bs jE, X X ωE = − cs jns j − bs jE, (88) s j s j where bs j and cs j is short notation for both unmagnetized and magnetized species bs jσ,sn j and cs jσ,sn j. We find the only singularity in the above final form occurs at k = 0, which requires kx = kz = 0. And thus the final form can be applied for arbitrary k , 0. Some other solvers in literature may meet singularity for kz = 0 or kx = 0, and may be incorrect for kz ≤ 0. It is also obvious that the magnetized species can not reduce to unmagnetized species by set ωcs = 0. The major difference is kzvzts in magnetized species and kvtsσ in unmagnetized species.

4.2. The Electromagnetic case The electromagnetic case is much complicated. However, the linear transformation for Qm(ω, k) is still similar to the original PASS-K/PDRK derivation. To seek an equivalent linear system, the Maxwells equations

2 ∂t E = c ∇ × B − J/0, (89a)

∂t B = −∇ × E, (89b) do not need to be changed. We only need to seek a new linear system for J = Jm + Ju = (σm + σu) · E = σ · E.

24 4.2.1. The unmagnetized terms 2 u u P ωps u Considering the deﬁnation σs = −i0ωQs = −i0 σ=a,b rsσ ω Psσ, after J-pole expansion, we have

n 2 2 2 2 2 2 4 2 2 2 2 2 2 u 2 kxv⊥tsσ kxvdsx kz vzts k kxv⊥tsσ k kxv⊥tsσ 1 2 k kz vzts k 2 • P = −1 + − iνs( 2 2 Z2 + Z1 + 2 2 Z0) + 3 3 Z3 + 2 2 2 vdsxZ2 + [ v⊥tsσ 3 3 + v + sσ11 kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ k vtsσ 2 k vtsσ kvtsσ dsx 2 2 4 2 o n 2 2 2 2 2 2 4 kxkz v⊥tsσ kxkz 2 PJ b j kxv⊥tsσ 2 kxvdsx kz vzts k kxv⊥tsσ 3 (λT sσ−1) 3 3 ]Z1+ 2 2 (λT sσ−1)v⊥tsσvdsxZ0 ' −1+2 j=1 − −iνs( 2 2 c + c j+ 2 2 )+ 3 3 c + k vtsσ k vtsσ ω cs jσ k vtsσ j kvtsσ 2k vtsσ k vtsσ j 2 2 2 2 2 2 2 2 4 2 o k kxv⊥tsσ 2 1 2 k kz vzts k 2 kxkz v⊥tsσ kxkz 2 PJ p11s jσ 2 2 2 vdsxc + [ v⊥tsσ 3 3 + v + (λT sσ − 1) 3 3 ]c j + 2 2 (λT sσ − 1)v⊥tsσvdsx = −1 + j=1 − , k vtsσ j 2 k vtsσ kvtsσ dsx k vtsσ k vtsσ ω cs jσ

n 2 2 2 2 o n 2 2 2 u 2vdsy k kxv⊥tsσ k 1 kxkz 2 PJ b j k kxv⊥tsσ 2 k vdsx • P = 2 2 Z2 + vdsxZ1 + 2 2 (λT sσ − 1)v⊥tsσZ0 ' 2vdsy j=1 2 2 c + c j + sσ12 kvtsσ k vtsσ kvtsσ 2 k vtsσ ω−cs jσ k vtsσ j kvtsσ 2 1 kxkz 2 o PJ p12s jσ 2 2 (λT sσ − 1)v⊥tsσ = j=1 , 2 k vtsσ ω−cs jσ

n 2 2 2 2 o n u 2vdsy kx k kxv⊥tsσ k 1 kxkz 2 PJ b j kx • P = − iνs Z1 + 2 2 Z2 + vdsxZ1 + 2 2 (λT sσ − 1)v⊥tsσZ0 ' 2vdsy j=1 − iνs c j + sσ21 kvtsσ kvtsσ k vtsσ kvtsσ 2 k vtsσ ω−cs jσ kvtsσ 2 2 2 2 o k kxv⊥tsσ 2 k 1 kxkz 2 PJ p21s jσ 2 2 c + vdsxc j + 2 2 (λT sσ − 1)v⊥tsσ = j=1 , k vtsσ j kvtsσ 2 k vtsσ ω−cs jσ

u 2 n 1 1 2 2 k2 o PJ b j n 1 1 2 2 k2 o • P = −1+ − iνs Z0 + ( v + v ) Z1 ' −1+2 − iνs + ( v + v ) c j = sσ22 kvtsσ 2 2 ⊥tsσ dsy kvtsσ j=1 ω−cs jσ 2 2 ⊥tsσ dsy kvtsσ −1 + PJ p22s jσ , j=1 ω−cs jσ

n 2 2 2 2 2 2 2 2 2 u 2 kxkzv⊥tsσ kzvdsx kxkzv⊥tsσ k kxkzvztsv⊥tsσ k kxv⊥tsσ kzvzts k • P = −iνs( 2 2 Z2 + Z1 − 2 2 Z0)+ 3 3 Z3 + [ vdsz +vdsx ]Z2 +[ vdsxvdsz − sσ13 kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ−1) 3 3 + (λT sσ−1) 3 3 ]Z1+[ (λT sσ−1) 2 2 vdsz− (λT sσ−1)vdsx 2 2 ]Z0 ' 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ n 2 2 2 2 2 2 2 2 2 PJ b j kxkzv⊥tsσ 2 kzvdsx kxkzv⊥tsσ k kxkzvztsv⊥tsσ 3 k kxv⊥tsσ kzvzts 2 k 2 j=1 − − iνs( 2 2 c + c j − 2 2 ) + 3 3 c + [ vdsz + vdsx ]c + [ vdsxvdsz − ω cs jσ k vtsσ j kvtsσ 2k vtsσ k vtsσ j kvtsσ kvtsσ kvtsσ j kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ −1) 3 3 + (λT sσ −1) 3 3 ]c j +[ (λT sσ −1) 2 2 vdsz − (λT sσ −1)vdsx 2 2 ] = 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ PJ p13s jσ , j=1 ω−cs jσ

n 2 2 2 2 2 2 2 2 2 u 2 kxkzvzts kxvdsz kxkzvzts k kxkzvztsv⊥tsσ k kxv⊥tsσ kzvzts k • P = − iνs( 2 2 Z2 + Z1 − 2 2 Z0) + 3 3 Z3 + [ vdsz + vdsx ]Z2 + [ vdsxvdsz − sσ31 kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ kvtsσ kvtsσ kvtsσ kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ−1) 3 3 + (λT sσ−1) 3 3 ]Z1+[ (λT sσ−1) 2 2 vdsz− (λT sσ−1)vdsx 2 2 ]Z0 ' 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ n 2 2 2 2 2 2 2 2 2 PJ b j kxkzvzts 2 kxvdsz kxkzvzts k kxkzvztsv⊥tsσ 3 k kxv⊥tsσ kzvzts 2 k 2 j=1 − − iνs( 2 2 c + c j − 2 2 ) + 3 3 c + [ vdsz + vdsx ]c + [ vdsxvdsz − ω cs jσ k vtsσ j kvtsσ 2k vtsσ k vtsσ j kvtsσ kvtsσ kvtsσ j kvtsσ 2 2 2 3 4 3 2 2 2 2 2 2 1 k kxkzvztsv⊥tsσ 1 kxkzv⊥tsσ 1 kxkz vztsv⊥tsσ 1 kxkz v⊥tsσ 1 kxkzv⊥tsσ o 3 3 − (λT sσ −1) 3 3 + (λT sσ −1) 3 3 ]c j +[ (λT sσ −1) 2 2 vdsz − (λT sσ −1)vdsx 2 2 ] = 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ 2 k vtsσ PJ p31s jσ , j=1 ω−cs jσ

n 2 2 2 2 2 o n u 2vdsy kz k kzvzts k 1 kxkzv⊥tsσ PJ b j kz • P = − iνs Z1 + 2 2 Z2 + vdszZ1 − (λT sσ − 1) 2 2 Z0 ' 2vdsy j=1 − iνs c j + sσ23 kvtsσ kvtsσ k vtsσ kvtsσ 2 k vtsσ ω−cs jσ kvtsσ 2 2 2 2 2 o k kzvzts 2 k 1 kxkzv⊥tsσ PJ p23s jσ 2 2 c + vdszc j − (λT sσ − 1) 2 2 = j=1 , k vtsσ j kvtsσ 2 k vtsσ ω−cs jσ

n 2 2 2 2 2 o n 2 2 2 u 2vdsy k kzvzts k 1 kxkzv⊥tsσ PJ b j k kzvzts 2 k 1 • P = 2 2 Z2 + vdszZ1 − (λT sσ − 1) 2 2 Z0 ' 2vdsy j=1 2 2 c + vdszc j − (λT sσ − sσ32 kvtsσ k vtsσ kvtsσ 2 k vtsσ ω−cs jσ k vtsσ j kvtsσ 2 2 2 kxkzv⊥tsσ o PJ p32s jσ 1) 2 2 = j=1 , k vtsσ ω−cs jσ

n 2 2 2 2 2 2 4 2 2 2 2 2 2 2 u 2 kz vzts kzvdsz kxv⊥tsσ k kz vzts k kzvzts 1 k kxvztsv⊥tsσ k 2 • P = −1+ −iνs( 2 2 Z2 + Z1 + 2 2 Z0)+ 3 3 Z3 +2 2 2 vdszZ2 +[ 3 3 + v −(λT sσ − sσ33 kvtsσ k vtsσ kvtsσ 2k vtsσ k vtsσ k vtsσ 2 k vtsσ kvtsσ dsz 2 2 2 2 2 2 o n 2 2 2 2 2 2 4 kxkz vztsv⊥tsσ kxkzv⊥tsσ PJ b j kz vzts 2 kzvdsz kxv⊥tsσ k kz vzts 3 1) 3 3 ]Z1 − (λT sσ − 1) 2 2 vdszZ0 ' −1 + 2 j=1 − − iνs( 2 2 c + c j + 2 2 ) + 3 3 c + k vtsσ k vtsσ ω cs jσ k vtsσ j kvtsσ 2k vtsσ k vtsσ j 2 2 2 2 2 2 2 2 2 2 2 2 2 o k kzvzts 2 1 k kxvztsv⊥tsσ k 2 kxkz vztsv⊥tsσ kxkzv⊥tsσ PJ p33s jσ 2 2 2 vdszc + [ 3 3 + v − (λT sσ − 1) 3 3 ]c j − (λT sσ − 1) 2 2 vdsz = −1 + j=1 − . k vtsσ j 2 k vtsσ kvtsσ dsz k vtsσ k vtsσ ω cs jσ

n 2 2 2 2 2 2 4 2 2 2 2 2 kxv⊥tsσ 2 kxvdsx kz vzts k kxv⊥tsσ 3 k kxv⊥tsσ 2 1 2 k kz vzts In the above, for example, p11s jσ = 2b j − iνs( 2 2 c + c j + 2 2 ) + 3 3 c + 2 2 2 vdsxc + [ v⊥tsσ 3 3 + k vtsσ j kvtsσ 2k vtsσ k vtsσ j k vtsσ j 2 k vtsσ 2 2 2 4 2 k 2 kxkz v⊥tsσ kxkz 2 o v + (λT sσ − 1) 3 3 ]c j + 2 2 (λT sσ − 1)v⊥tsσvdsx , and others are similar and thus we have not written them kvtsσ dsx k vtsσ k vtsσ out explicitly. 25 We ﬁnd the result is very simple by use Z0,1,2,3, which would also make the EM3D-M case be much simpler than the previous PASS-K [1] derivation. This is also why we use Z0,1 in the ES3D case, which gives a more compact form. u P Thus, we obtain the relations between J and E, which has the following form (with s=u)

u u u  b11 P bs jσ11 b12 P bs jσ12 b13 P bs jσ13   u   + s jσ + s jσ + s jσ    Jx  ω ω−cs jσ ω ω−cs jσ ω ω−cs jσ  Ex    bu bu bu     u   21 P bs jσ21 22 P bs jσ22 23 P bs jσ23     Jy  = −i0  + s jσ + s jσ + s jσ   Ey  , (90)    ω ω−cs jσ ω ω−cs jσ ω ω−cs jσ     u   bu b bu b bu b    Jz  31 + P s jσ31 32 + P s jσ32 33 + P s jσ33  Ez ω s jσ ω−cs jσ ω s jσ ω−cs jσ ω s jσ ω−cs jσ with the coeﬃcients  b = r ω2 p /c , bu = − Ps=u ω2 [1 + P r P p /c ],  s jσ11 sσ ps 11s jσ s jσ 11 s ps σ sσ j 11s jσ s jσ  b = r ω2 p /c , bu = − Ps=u ω2 [P r P p /c ],  s jσ12 sσ ps 12s jσ s jσ 12 s ps σ sσ j 12s jσ s jσ  b = r ω2 p /c , bu = − Ps=u ω2 [P r P p /c ],  s jσ21 sσ ps 21s jσ s jσ 21 s ps σ sσ j 21s jσ s jσ  2 u − Ps=u 2 P P  bs jσ22 = rsσωps p22s jσ/cs jσ, b22 = s ωps[1 + σ rsσ j p22s jσ/cs jσ],  Ps=u P P  b = r ω2 p /c , bu = − ω2 [ r p /c ],  s jσ13 sσ ps 13s jσ s jσ 13 s ps σ sσ j 13s jσ s jσ (91)  b = r ω2 p /c , bu = − Ps=u ω2 [P r P p /c ],  s jσ31 sσ ps 31s jσ s jσ 31 s ps σ sσ j 31s jσ s jσ  b = r ω2 p /c , bu = − Ps=u ω2 [P r P p /c ],  s jσ23 sσ ps 23s jσ s jσ 23 s ps σ sσ j 23s jσ s jσ  b = r ω2 p /c , bu = − Ps=u ω2 [P r P p /c ],  s jσ32 sσ ps 32s jσ s jσ 32 s ps σ sσ j 32s jσ s jσ  2 u Ps=u 2 P P  bs jσ33 = rsσωps p33s jσ/cs jσ, b = − s ωps[1 + σ rsσ j p33s jσ/cs jσ],  33  cs jσ = kzvdsz + kxvdsx − iνs + kvtsσc j,

Considering that c j are complex number and kx, kz and νs are real number, the singularity of the above form is also only k = 0.

4.2.2. The magnetized terms 2 m m P ωps m Similarly to the unmagnetized case, considering the deﬁnation σs = −i0ωQs = −i0 σ=a,b rsσ ω Psσ, after J-pole expansion, we have n o m P∞ PJ kzvztsb j nωcs Anbsσ 1 nωcs c j nωcs • P ' n=−∞ j=1 [ (nωcs + kxvdsx − iνs) 2 + An0σ( + vdsx) 2 ]( + vdsx) − 1 = −1 + sσ11 ω−csn j kx v⊥tsσ kzvzts kx vzts kx P P p11sn j . n j ω−csn j n m P∞ PJ kzvztsb j nωcs Anbsσ 1 c j nωcs Bnbsσ 1 • P ' n=−∞ j=1 vdsy( + vdsx)(nωcs 2 + An0σ 2 ) + i( + vdsx)[(nωcs − iνs) + sσ12 ω−csn j kx v⊥tsσ kzvzts vzts kx v⊥tsσ kzvzts Bn0σv⊥tsσc j o P P p12sn j 2 ] = n j . vzts ω−csn j n m P∞ PJ kzvztsb j nωcs 1 1 nωcs • P ' n=−∞ j=1 (nωcs+kxvdsx−iνs)[−iv⊥tsσBnbsσ+vdsyAnbsσ] 2 +[−iv⊥tsσBn0σ+vdsyAn0σ]( + sσ21 ω−csn j kx v⊥tsσ kzvzts kx c j o P P p21sn j vdsx) 2 = n j . vzts ω−csn j n m P∞ PJ kzvztsb j nωcs 1 c j • P ' n=−∞ j=1 vdsy[−iv⊥tsσBnbsσ+vdsyAnbsσ] 2 +vdsy[−iv⊥tsσBn0σ+vdsyAn0σ] 2 +v⊥tsσ[v⊥tsσCnbsσ+ sσ22 ω−csn j v⊥tsσ kzvzts vzts 1 1 c j o P P p22sn j ivdsyBnbsσ](nωcs − iνs) 2 + v⊥tsσ[v⊥tsσCn0σ + ivdsyBn0σ] 2 − 1 = −1 + n j . v⊥tsσ kzvzts vzts ω−csn j n o m P∞ PJ kzvztsb j nωcs c j 1 Anbsσ 2 c j 1 P P p13sn j • P ' n=−∞ j=1 ( +vdsx)[nωcs( +vdsz ) 2 +An0σ(vztsc +(kzvdsz−iνs) ) 2 ] = n j . sσ13 ω−csn j kx kz kzvzts v⊥tsσ j kz vzts ω−csn j

n c2 o m P∞ PJ kzvztsb j nωcs Anbsσ c j 1 nωcs j c j • P ' n=−∞ j=1 (nωcs + kxvdsx − iνs) 2 ( + vdsz ) + An0σ( + vdsx)( + vdsz 2 ) = sσ31 ω−csn j kx v⊥tsσ kz kzvzts kx vzts vzts P P p31sn j . n j ω−csn j n m P∞ PJ kzvztsb j c j vdsz nωcs 2 • P ' n=−∞ j=1 [vdsyAnbsσ − iv⊥tsσBnbsσ][( + ) 2 ] + [vdsyAn0σ − iv⊥tsσBn0σ][vztsc + (kzvdsz − sσ23 ω−csn j kz kzvtsz v⊥tsσ j c j 1 o P P p23sn j iνs) ] 2 = n j . kz vzts ω−csn j

26 n k v c2 m P∞ PJ kzvztsb j Anbsσ c j z zts j Anbsσ 1 c j Bnbsσ c j • P ' n=−∞ j=1 [nωcs 2 +An0σ 2 ]vdsy+[nωcs 2 +An0σ 2 ]vdsyvdsz+i[(nωcs−iνs) + sσ32 ω−csn j v⊥tsσ kz vzts v⊥tsσ kzvzts vzts v⊥tsσ kz v c2 ⊥tsσ j Bnbsσ 1 v⊥tsσc j o P P p32sn j Bn0σ ] + ivdsz[(nωcs − iνs) + Bn0σ 2 ] = n j . vzts v⊥tsσ kzvzts vzts ω−csn j

n v c2 c2 m P∞ PJ kzvztsb j zts j c j Anbsσ 3 j c j 1 Anbsσ • P ' n=−∞ j=1 [nωcs( +vdsz ) 2 +An0σ(c +(kzvdsz −iνs) )]+vdsz[nωcs( +vdsz ) 2 + sσ33 ω−csn j kz kz v⊥tsσ j kzvzts kz kzvzts v⊥tsσ 2 c j c j o P P p33sn j An0σ( + (kzvdsz − iνs) 2 )] − 1 = −1 + n j . vzts kzvzts ω−csn j

n nωcs Anbsσ 1 nωcs c j nωcs o In the above, say, p11sn j = kzvztsb j [ (nωcs + kxvdsx − iνs) 2 + An0σ( + vdsx) 2 ]( + vdsx) , and other kx v⊥tsσ kzvzts kx vzts kx terms are similar. It is thus easy to ﬁnd that after J-pole expansion, the relations between Jm and E has the following form (with P s=m) m m m  b11 P bsn j11 b12 P bsn j12 b13 P bsn j13   m   + sn j + sn j + sn j    Jx  ω ω−csn j ω ω−csn j ω ω−csn j  Ex    bm bm bm     m   21 P bsn j21 22 P bsn j22 23 P bsn j23     Jy  = −i0  + sn j + sn j + sn j   Ey  . (92)    ω ω−csn j ω ω−csn j ω ω−csn j     m   bm b bm b bm b    Jz  31 + P sn j31 32 + P sn j32 33 + P sn j33  Ez ω sn j ω−csn j ω sn j ω−csn j ω sn j ω−csn j with the coeﬃcients  P 2 m − Ps=m 2 P P nωcs nωcs P  bsn j11 = σ rsσωps p11sn j/csn j, b11 = s ωps σ rsσ[ n Anbsσ 2 ( k + vdsx) + n j p11sn j/csn j],  kxv⊥tsσ x  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j12 σ sσ ps 12sn j sn j 12 s ps σ sσ n j 12sn j sn j  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j21 σ sσ ps 21sn j sn j 21 s ps σ sσ n j 21sn j sn j  P 2 m Ps=m 2 P P vdsy P  bsn j22 = σ rsσωps p22sn j/csn j, b22 = − s ωps σ rsσ[ n(Cnbsσ + i v Bnbsσ) + n j p22sn j/csn j],  ⊥tsσ  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j13 σ sσ ps 13sn j sn j 13 s ps σ sσ n j 13sn j sn j (93)  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j31 σ sσ ps 31sn j sn j 31 s ps σ sσ n j 31sn j sn j  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j23 σ sσ ps 23sn j sn j 23 s ps σ sσ n j 23sn j sn j  b = P r ω2 p /c , bm = − Ps=m ω2 [P r P p /c ],  sn j32 σ sσ ps 32sn j sn j 32 s ps σ sσ n j 32sn j sn j  P 2 m Ps=m 2 P P 1 P  bsn j33 = σ rsσωps p33sn j/csn j, b = − s ωps σ rsσ[ n An0σ + n j p33sn j/csn j],  33 2  csn j = kzvdsz + nωcs + kxvdsx − iνs + kzvztsc j.

P PN In numerical test, considering the cut of summation n, i.e., n = n=−N with N , ∞, we ﬁnd the above original form m m m of b11, b22 and b33, are better than the below form [to undetstand]

 bm = − Ps=m ω2 [1 + P r P p /c ],  11 s ps σ sσ n j 11sn j sn j  m − Ps=m 2 P P  b22 = s ωps[1 + σ rsσ n j p22sn j/csn j], (94)  m Ps=m 2 P P b33 = − s ωps[1 + σ rsσ n j p33sn j/csn j].

1 m nωcs m m It is readily to see that all the singularities from k in Psσ are removable. The k singularities at kx = 0 in Psσ11, Psσ12, m m m z x Psσ21, Psσ13, Psσ31 are also removable. Thus, the overall equations have no singularity and will not meet numerical diﬃculty. In the solver, to short the code, if kxρs < kδ we set kxρs = kδ for magnetized species in EM version. For −30 example, we can set kδ = 10 . Combining Eqs. (89), (90) and (92), the equivalent linear system for electromagnetic dispersion relation can be

27 obtained as  ωvs=m c v b E b E b E ,  sn jx = sn j sn jx + sn j11 x + sn j12 y + sn j13 z  s=u  ωvs jσx = cs jσvs jσx + bs jσ11Ex + bs jσ12Ey + bs jσ13Ez,   ω jx = b11Ex + b12Ey + b13Ez,  Ps=m Ps=u  iJx = jx + vsn jx + vs jσx,  0 sn j s jσ  ωvs=m = c v + b E + b E + b E ,  sn jy sn j sn jy sn j21 x sn j22 y sn j23 z  ωvs=u = c v + b E + b E + b E ,  s jσy s jσ s jσy s jσ21 x s jσ22 y s jσ23 z   ω jy = b21Ex + b22Ey + b23Ez,  Ps=m Ps=u  iJy/0 = jy + sn j vsn jy + s jσ vs jσy,  s=m  ωv = csn jvsn jz + bsn j31Ex + bsn j32Ey + bsn j33Ez,  sn jz (95)  ωvs=u = c v + b E + b E + b E ,  s jσz s jσ s jσz s jσ31 x s jσ32 y s jσ33 z  ω j = b E + b E + b E ,  z 31 x 32 y 33 z  iJ / = j + Ps=m v + Ps=u v ,  z 0 z sn j sn jz s jσ s jσz  ωE = c2k B − iJ / ,  x z y x 0  − 2 2 −  ωEy = c kzBx + c kxBz iJy/0,  2  ωEz = −c kxBy − iJz/0,   ωBx = −kzEy,   ωBy = kzEx − kxEz,   ωBz = kxEy, m u which yields a sparse matrix eigenvalue problem, where b11 = b11 + b11 and so on. The symbols vsn jx, jx,y,z and Jx,y,z used here do not have direct physical meanings but are analogy to the perturbed velocity and current density in the fluid derivations of plasma waves. The elements of the eigenvector (Ex, Ey, Ez, Bx, By, Bz) still represent the original electric and magnetic fields. Thus, the polarization of the solutions can also be obtained in a straightforward manner. The dimension of the matrix is NN = 3 × (NS mNJ + NS uJ + 1) + 6 = 3 × {[S m × (2 × N + 1) + S u × 2] × J + 1} + 6. And another good aspect of the final PASS-K matrix equation is that it is valid for arbitrary real number of kx and kz, i.e., θ ∈ [0, 2π] and the only requirement is k , 0.

4.3. The Darwin case Based on the electromagnetic result, the Darwin model case is straightforward, where the linear system for J = Jm + Ju = (σm + σu) · E = σ · E is the same to electromagnetic case. We only need to modify the the linear system of the Maxwells equations, which are also straightforward kk · − 2 × − ω( k2 ) E = c k B iJ/0, (96a) ωI · B = k × E, (96b) and the matrix eigenvalue problem becomes ωMB · X = MA · X, where MA is still the same as the electromagnetic one from Eq.(95) and MB changes from unit matrix INN ×NN to  I   (NN −6)×(NN −6)   kk  MB =  ( 2 )3×3  . (97)  k  I3×3

kk Though ( k2 )3×3 may not be full rank matrix, the standard eigenvalue library, such as ’eig()’ in Matlab, can solve the eigenvalue problem well.

4.4. The polarizations 6 The matrix solver can obtain (Ex, Ey, Ez, Bx, By, Bz) directly from the matrix eigenvalue problem. Considered that the magnitude of the wave has no meaning for a linear system, we should do normalizations. We set |E| = 1mV/m and Ex = Re(Ex).

6 In principle, PASS-K matrix can also obtain (Ex, Ey, Ez) as in standard 3 × 3 matrix D · E = 0. To obtain group velocity vg = dω/dk or do ray tracing, we may also need ∂D/∂ω and ∂D/∂k. 28 0 ∗ 1 ∗ Some other useful: electric field energy UE = E· E , magnetic field energy UB = B· B , energy flux Poynting 2 2µ0 1 ∗ vector S = E × B . P = Ey/iEx. µ0

5. Benchmark There exists numerous applications of this newly developed updated version PASS-K tool, we only show some typical benchmarks to the reader to get a flavor of it. The first benchmark is the ring beam case in Ref.[12], which is to make sure the function An, Bn and Cn are treated correctly in our model. The results are shown in Fig.1, with very good agreement with Min’s [17] code. The second benchmark shown in Fig.2 is the mixed of magnetized and unmagnetized species in Ref.[14] for the instabilities driven by perpendicular beam in shock, which also show good agreement. And, the treatment of ion to be magnetized species also shows close result to the unmagnetized ion model, which implies that the unmagnetized ion assumption is valid in that case. This also gives us confidence of the validity of our magnetized model, since that the equations are totally different but yield similar solution as should be. The third benchmark shown in Fig.3 is for the Darwin model in Ref.[16], which is the same as the one solved using accurate Z function with conventional iterative root finding in Fig.5 of Ref.[16]. Here, we have also shown other branches and kz < 0 branches. The symmetry between kz < 0 and kz > 0 solutions implies that the Z function is treated correctly for kz < 0 in this new solver.

Benchmark Umeda2012PoP Fig. 1 For ring beam, 3=40° 8 0.5 passk-em-1 7 passk-em-2 passk-em-3 K. Min's-1 0 6 K. Min's-2 passk-es-1

5 -0.5

! !

/ 4

r i

! ! 3 -1

2 -1.5 1

0 -2 0 0.1 0.2 0.3 0 0.1 0.2 0.3 k6 k6 De De

Figure 1: Benchmark the electromagnetic ring beam dispersion relation between PASS-K and Min’s code [17] with the parameters in Ref.[12] Fig.1 for θ = 40◦, which shows very good agreement. PASS-K gives all the three unstable branches, whereas the third branch ’passk-em-3’ could easily be missed by the iterative root ﬁnding approaches used in Min’s [17] or Umeda’s [12] codes. One branch of the electrostatic PASS-K solution is also shown for reference, which is close to the electromagnetic one implies that this branch is essentially an electrostatic mode. The results also agree well with the PIC simulation result in Ref.[12].

The benchmark parameters for above cases are listed below.

• Defaultly we use SI unit: c = 2.99792458e8, 0 = 8.854187817e − 12, kB = 1.38064852e − 23, qe = 1.60217662e − 19, mp = 1.6726219e − 27.

• Parameters for Ref.[12] benchmark: B0 = 96.24E − 9. qs = [−1, −1], ms = [5.447e − 4, 5.447e − 4], ns = [1e5, 9e5], Tzs = T⊥s = [5.1, 5.1], vdsz/c = [0.1, 0], vdsr/c = [0.05, 0].

• Parameters for Ref.[14] benchmark: qs = [1, 1, −1], ms = [1, 1, 1.1111e − 3], ns = [0.8, 0.2, 1.0], Tzs = T⊥s = [0.111, 0.111, 0.01], vdsx/c = [−8.333e − 3, 3.333e − 2, 0]. B0 = 1, c = 300, µ0 = 1.0, kB = 1.0, qe = 1.0, mp = 1.0. The purpose of the present work is to provide the foundation of this new tool. And thus, the applications to new examples would be discussed in other works. 29 ° Benchmark Muschietti17 Fig. 1 For v beam, 3=50 600 15 ? passk-em-u passk-em-m, N=60 500 Muschietti2017 Fig. 1, - =0.02 e 10

400

! /

300 / 5

r i

! !

200 0 100

0 -5 0 20 40 60 80 0 20 40 60 80 kc/! kc/! pi pi

Figure 2: Benchmark the electromagnetic penpendicular beam dispersion relation between PASS-K and Ref.[14] Fig. 1 with the parameters θ = 50◦ and βe = 0.02, which shows very good agreement. Ref.[14] assumes unmagnetized ions and magnetized electron. ’passk-em-u’ uses the same P assumption; whereas ’passk-em-m’ uses also magnetized ions with N = 60 where the n has convergent. The agreement between ’-U’ and ’-M’ versions also implies that the unmagnetized ions is a valid assumption for this case.

Benchmark Xie2014 Fig. 5 For Darwin model, 3=0° 3 passk-darwin-1 passk-darwin-2 0 2.5 passk-darwin-3 passk-em-1 passk-em-2 -0.1 2 passk-em-3 passk-em-4 -0.2

passk-em-5

! !

/ 1.5 -0.3

r i

! ! -0.4 1 -0.5 0.5 -0.6

0 -0.7 -2 -1 0 1 2 -2 -1 0 1 2 k c/! k c/! z ce z ce

◦ Figure 3: Compare√ the electromagnetic model and Darwin model in PASS-K with the parameters in Ref.[16] Fig. 5, θ = 0 and ωce/ωpe = 1.2, c/vzte = 5/ 2, mi/me = 4, Ti = Te. The kz < 0 solutions are also shown, which shows good symmetry to the kz > 0 solutions. The results show that Darwin model is a good approximate model in this case, and the slight deviations agree with the theoretical expectation in Ref.[16].

30 6. Summary and Discussion

In summary, a powerful new kinetic dispersion relation tool is developed, which largely extend both the physical models and numerical capacilities of other works in literature. The advantages of this new version of PASS-K tool is that it contents many new features (anisotropic temperature/loss cone/drift in arbitrary direction/ring beam/collision, unmagnetized/magnetized, electrostatic/electromagnetic/Darwin, etc) and can be widely applied. And compared to some other solvers, the kz = 0 (θ = π/2) and kx = 0 (θ = 0) cases are not singular in PASS-K. What is more, the kz < 0 modes are also correctly treated. The most attractive feature is that it does not require initial guess for root finding and thus will not miss solutions. Thus, we think that this is a unified tool what exactly dreamed of by the plasma community. The limitation of the present PASS-K approach is that it can only used for cases when the 2D velocity integral 0 0 are decoupled, i.e., usually required fs0(v) = fz(vk) f⊥(v⊥). Typically, the present PASS-K approach can not be used to oblique propagation κ-distribution function [19, 20] and relativistic [3] or other arbitrary distribution[18] cases. However, extensions are possible. For example, for κ-distribution, we can use the similar J-pole for corresponding 7 Zκ function of parallel velocity integral and use Gaussian quadrature for perpendicular velocity integral, which will yield a 2D J-pole expansion and then can be transformed to a equivalent linear matrix eigenvalue problem, which is solvable though the corresponding matrix dimension could be much larger than the Maxwellian one, depends on how many nodes are used for perpendicular velocity integral. And thus, with 2D J-pole expansion, the PASS-K approach can also be used for other distributions and non-uniform magnetic field drift wave case such as in gyrokinetic model [21]. The discussion of these extensions are out of the range of the present work. Besides the conventional iterative root finding approaches, it is still required to develop a new powerful algorithm, say, similar to PASS-K approach, to solve the more challenged relativistic kinetic plasma dispersion relation generally.

7. Acknowledgments

The author thanks the many suggestions, discussions and benchmarks from Richard Denton, Xin Tao, Jin-song Zhao, Zhong-wei Yang (helped the benchmark of Fig.2), Chao-jie Zhang (helped the benchmark of unmagnetized version), Can Huang, Wen-ya Li, Liang Wang, Kyungguk Min (provided the benchmark data in Fig.1), Yang Li, and many others. This work, which is particular useful, say for examples to study the loss-cone and LHDI instabilities in magnetic conﬁnement fusion and beam instabilities in inertial fusion, is supported by the compact fusion project in ENN group.

Appendix A. The PASS-K Solver The update version of PASS-K/PDRK code includes more features, but still consists of two parts: the main program and the input data ﬁle. The input ﬁle “passk.in” has the follow structure (the case in Fig.2) qs(e) ms(mp) ns(m^-3) Tzs(eV) Tps(eV) alphas Deltas vdsz/c vdsx/c vdsy/c vdsr/c nu_s m_or_u(1/0) 1 1.0 0.8e0 0.111 0.111 1.0 1.0 0.0 -8.333e-3 0.0 0.0 0.0 0 1 1.0 0.2e0 0.111 0.111 1.0 1.0 0.0 3.333e-2 0.0 0.0 0.0 0 -1 1.1111e-3 1.0e0 0.010 0.010 1.0 1.0 0.0 0.0 0.0 0.0 0.0 1

One can add the corresponding parameters for additional species. Here, ’qs(e)’, ’ ms(mp) ’, ’ns(mˆ-3)’, ’Tzs(eV) ’, ’Tps(eV) ’, ’alphas’, ’Deltas’, ’vdsz/c’, ’vdsx/c ’, ’ vdsy/c’, ’ vdsr/c ’, ’nu s ’, ’m or u(1/0) ’, are the charge in electron charge unit qs/e, mass in proton mass unit ms/mp, density ns0, parallel temperature Tzs, perpendicular temperature T⊥s, loss cone parameters αs and ∆s, drift velocities vdsz/c, vdsx/c, vdsy/c, ring beam velocity vdsr/c, collision frequency νs, and whether the species is magnetized, respectively. Normalizations and other parameters can be set in PASS-K main program (more information can be found at: http://hsxie.me/codes/pdrk/ or https://github.com/hsxie/pdrk/).

7 PJ b j PJ b j The κ-distribution Zκ(ζ) = j , which is slightly diﬀerent from the standard one Z(ζ) = and thus the corresponding j=1 (ζ−c j) j=1 ζ−c j transformation is also slightly diﬀerent.

31 Appendix B. Relation to the dispersion relation of drift instabilities in inhomogeneous plasma

In conventional derivation [3–5] of drift instabilities in inhomogeneous magnetized plasma, the distribution function is assumed to depend on space inhomogeneous. We take the electrostatic density inhomogeneous case as in Chap. 4 of Ref.[5] for example to compare with our model. The conventional derivation, i.e., Eqs.(4.2.1) and (4.2.2) 2 in Ref.[5], which ignores all terms of order n and valids only for n k and nρcs 1, gives

2 2 2 − vz (vy−vgs) +vx 1 h vy vgs i − v − v f √ x e ts2 e ts2 , k , k , k , s0 = 3 1 + n + = (0 y z) (B.1) πvts ωcs 2 ∞ 2 X ωps n X hω − ky(vns + vgs) nω i a o D(ω, k) = 1 + 1 + + n Z(ζ )Γ ( s ) 2 2 |k |v |k |v k sn n 2 s 2k vzts n=−∞ z ts z zts y 2 ∞ 2 X ωps n X ω − k⊥(vns + vgs) a o ' 1 + 1 + Z(ζ )Γ ( s ) = 0, (B.2) 2 2 |k |v sn n 2 s 2k vzts n=−∞ z ts

2 nvts g ω−k⊥vgs−nωcs n with vns = , vgs = and ζsn = , and where we have ignored the 1 term. Here, n is parameter ωcs ωcs |kz|vts ky for space inhomogeneous and g = |g| is for force such as gravity. The corresponding dispersion relation in our model is

2 2 2 vz (vx−vdsx) +vy 1 − v − v f √ e ts2 e ts2 , k k , , k , s0 = 3 = ( x 0 z) (B.3) πvts 2 ∞ 2 X ωps X h nω i a D(ω, k) = 1 + Z (ζ ) + cs Z (ζ ) Γ ( s ) 2 2 1 sn k v 0 sn n 2 s 2k vzts n=−∞ z zts 2 ∞ 2 X ωps n X ω − k v a o = 1 + 1 + ⊥ dsx Z(ζ )Γ ( s ) = 0, (B.4) 2 2 |k |v sn n 2 s 2k vzts n=−∞ z ts

ω−k⊥vdsx−nωcs P∞ with ζsn = , and we have used Γn = 1. If we consider the perpendicular drift in our model is due |kz|vts n=−∞ ∇p to space inhomogeneous − and gravity mg, with p = ns0(y)kBTs0, ns0(y) = ns0(1 − ny) and g = gyˆ, by using force ns0 2 ∇(ns0kBTs0) 2 Fs nvts balance and Fs = ms g − = (msg + nmsv )yˆ, we have vdsx = = vgs + = vgs + vns. We noticed that ns0 ts qs B0 ωcs Eqs.(B.2) and (B.4) are exactly the same, except that in the conventional derivation (B.2) the density inhomogeneous drift velocity vns in not included in the argument of Z function ζsn. That is to say, our model can be used to study the drift modes in inhomogeneous plasma although we assume the derivation is under homogeneous plasma assumption. It is also not easy to say which of the the two derivations, i.e., the conventional one and the present one, is more accurate to describe the drift modes in inhomogeneous plasma, because that in our derivation the inhomogeneity contributes to the drift velocity and all other steps are rigorous, whereas the conventional derivation ignores high order terms in more than one steps to obtain the ﬁnal dispersion relation.

Appendix C. More J-pole coeﬃcients of Z(ζ) function

Ref.[1] has provided the typical J-pole coeﬃcients of Z(ζ) function with J = 4, 8, 12. Here, following Ref.[1], we update them to more accurate data and even beyond the double precision (16 digit number) number with the help of symbolic computation. In the Table C.1, I means we keep I equations for ζ → 0, and K means we keep K equations for ζ → ∞, as described in Ref.[1].

Appendix D. The Possible Numerical Inaccuracy

Since PASS-K solves the dispersion relation use matrix eigenvalue approach, the computational time to obtain all α P solutions for a NN × NN dimensions matrix is O(NN ), with usually α ∼ 2.7. Thus, the numerical n with cut oﬀ to 32 Table C.1: The more accurate coeﬃcients c j and b j for J = 4, J = 8, J = 12, J = 16 and J = 24 under J-pole Pade approximations of Z(ζ), where the asterisk denotes complex conjugation, and δZ = |ZJ (ζ) − Z(ζ)| is the maximum error between ZJ (ζ) and Z(ζ) for ζ = x + iy with x ∈ [0, 50] and y = −0.1.

b1=- 1.0467968598346571444 - 2.1018525680357924343i c1=0.37861161238699661757 - 1.3509435854325440801i −3 J = 4 b2=0.54679685983465714437 + 0.037196505239893094435i c2=1.2358876534356917085 - 1.2149821325576149719i δZ = 3 × 10 (I = 5, K = 3) b(3 : 4)=b∗(1 : 2) c(3 : 4)=−c∗(1 : 2) b1=- 5.5833741816150427087 - 11.208550459628098648i c1=0.27393621805538084727 - 1.9417870375760945628i b2=- 0.7399178112200519477 - 0.83951828462027428396i c2=- 1.4652340919391423883 - 1.7896202996033145873i −6 J = 8 b3=- 0.017340112270400811857 - 0.04630643962629377424i c3=2.2376877251342932158 - 1.6259410241203623422i δZ = 6 × 10 (I = 10, K = 6) b4=5.8406321051054954683 - 0.95360275132203964347i c4=- 0.83925396636792203153 - 1.8919952115314256943i b(5 : 8)=b∗(1 : 4) c(5 : 8)=−c∗(1 : 4) b1=- 47.913598578418315281 - 106.98699311451399461i c1=0.22536708628380726987 - 2.4862558428460328565i b2=- 20.148858425809293248 + 12.874749056250453631i c2=1.1590491549279069691 - 2.4061921257040740764i b3=- 4.5311004339957471789E-3 + 6.3311756354943215316E-4i c3=2.9785703941315209704 - 2.0490809954949754985i −9 J = 12 b4=0.2150040123642351701 + 0.20042340981056393122i c4=2.2568587892309227294 - 2.2080229126485700572i δZ = 8 × 10 (I = 16, K = 8) b5=0.43131038679231352184 - 4.1505366661190555077i c5=1.6738373878120108271 - 2.3235155478934783777i b6=66.920673705505055584 + 20.747375125403268524i c6=0.68229440981712468 - 2.4598334422617114946i b(7 : 12)=b∗(1 : 6) c(7 : 12)=−c∗(1 : 6) b1=- 10.020983259474214017 - 14.728932929429874883i c1=0.22660012611958088508 - 2.0716877594897791206i b2=- 0.58878169153449514493 + 0.19067303610080007359i c2=- 1.7002921516300350075 - 1.882247422161272446i b3=- 0.27475707659732384029 + 3.617920717493884482i c3=1.1713932508560117853 - 1.9772503319208541098i −8 J = 12 b4=4.5713742777499515344E-4 + 2.7155393843737098852E-4i c4=3.0666201126826972102 - 1.5900208259325997176i δZ = 5 × 10 (I = 12, K = 12) b5=0.017940627032508378515 - 0.036436053276701248142i c5=2.3073274904105782764 - 1.7546732543728200654i b6=10.366124263145749629 - 2.5069048649816145967i c6=0.68720052490601906567 - 2.0402885259758440187i b(7 : 12)=b∗(1 : 6) c(7 : 12)=−c∗(1 : 6) b1=- 86.416592794839804566 - 147.57960545984972964i c1=0.19664397441136646085 - 2.5854046363167904821i b2=- 22.962540986214500398 + 46.211318219085729914i c2=1.0004276870893045112 - 2.5277610669350594581i b3=- 8.8757833558787660662 - 11.561957978688249474i c3=1.4263380087098663429 - 2.4694803409658086505i −11 J = 16 b4=- 0.025134802434111256483 + 0.19730442150379382482i c4=2.382753075769737514 - 2.2903917960623787648i δZ = 3 × 10 (I = 18, K = 14) b5=- 5.6462830661756538039E-3 - 2.7884991898011769583E-3i c5=2.9566517643704010427 - 2.1658992556376956217i b6=2.8262945845046458372E-5 + 2.6335348714810255537E-5i c6=- 3.6699741330155866185 - 2.0087276133120462601i b7=2.3290098166119338312 - 0.57238325918028725167i c7=1.8818356204685089975 - 2.3907395820644127768i b8=115.45666014287557906 - 2.8617578808752183449i c8=0.59330036294742852232 - 2.5662607006180515205i b(9 : 16)=b∗(1 : 8) c(9 : 16)=−c∗(1 : 8) b1=- 579.77656932346560644 - 844.01436313629880827i c1=0.16167711630587375808 - 2.9424665391729649011i b2=- 179.52530851977905732 - 86.660002027244731382i c2=1.1509135876493567245 - 2.874554296549015316i b3=- 52.107235029274485215 + 453.3246806707749413i c3=0.81513635269214329287 - 2.9085569383176322447i b4=- 2.1607927691932962178 + 0.63681255371973499384i c4=2.2362950589041724111 - 2.7033607074680388479i b5=- 0.018283386874895507814 - 0.21941582055233427677i c5=2.6403561313404041541 - 2.6228400297078984517i −13 J = 24 b6=- 6.819511737162705016E-5 + 3.2026091897256872621E-4i c6=3.5620497451197056658 - 2.4245607245823420556i δZ < 2 × 10 (I = 24, K = 24) b7=- 2.8986123310445793648E-6 - 9.9510625011385493369E-7i c7=4.1169251257106753931 - 2.3036541720854573609i b8=2.3382228949223867744E-9 - 4.0404517369565098657E-9i c8=4.8034117493360317933 - 2.1592490859689535413i b9=0.01221466589423530596 + 0.00097890737323377354166i c9=3.0778922349246567316 - 2.5301774598854448463i b10=7.3718296773233126912 - 12.575687057120635407i c10=- 1.8572088635240765004 - 2.7720571884094886584i b11=44.078424019374375065 - 46.322124026599601416i c11=1.496988132246689338 - 2.8290855580900544693i b12=761.62579175738689742 + 185.11797721443392707i c12=- 0.48636891219330428093 - 2.9311741817223824196i b(13 : 24)=b∗(1 : 12) c(13 : 24)=−c∗(1 : 12)

33 n ∈ [−N, N] is crucial. For larger N, the matrix dimension could be large and the computation will slower. Usually N ≤ 10 is suﬃcient for many cases and the computation is very fast, which can yield all solutions for a typical single k run in less than 1 second. For larger N, say N > 50, the sparse matrix approach to search solutions around some initial guesses are still fast and thus the PASS-K tool has not limitation on this. However, there exists several possible numerical inaccuracies, which should be careful of:

• (1) The J-pole expansion of Z(ζ) may give some artiﬁcial growing mode with growth rate γ ∼ 10−6 for J = 8, which however can be distinguished and reduced by using J = 12, 16, 24, etc.

P∞ PN • (2) The cut oﬀ in n=−∞ → n=−N , which should be checked by using larger N to make sure the results are convergent. • (3) Rand oﬀ error of eigen solver library, due to that the default numerical data is double precision (16 digit number) in the solver. This in principle can be solved by using high precision digit data.

• (4) Others, such as singularity in Darwin matrix, degree of accuracy of functions In, An, Bn and Cn, etc. The above inaccuracies mainly occur at extremely small k and large k. The third inaccuracy is not easy to distinguish at this moment, and however, we have not met this problem for most benchmarks.

References

[1] H. S. Xie and Y. Xiao, PDRK: A General Kinetic Dispersion Relation Solver for Magnetized Plasma, Plasma Science and Technology, 18, 2, 97 (2016). DOI: 10.1088/1009-0630/18/2/01. Update/bugs fixed at http://hsxie.me/codes/pdrk/ or https://github.com/hsxie/pdrk/. [2] K. Ronnmark, WHAMP - Waves in Homogeneous Anisotropic Multicomponent Magnetized Plasma, KGI Report No. 179, Sweden, 1982. [3] T. Stix, Waves in Plasmas, AIP Press, 1992. [4] D. G. Swanson, Plasma Waves, IOP, 2nd (Ed.), 2003. [5] S P. Gary, Theory of Space Plasma Microinstabilities. Cambridge University Press, New York, 1993. [6] D. A. Gurnett and A. Bhattacharjee, Introduction to plasma physics: with space and laboratory applications, Cambridge, 2005. [7] H. S. Xie, PDRF: A general dispersion relation solver for magnetized multi-fluid plasma, Computer Physics Communications, 185, 670675 (2014). [8] K. Ronnmark, Computation of the dielectric tensor of a Maxwellian plasma, Plasma Phys., 25 (1983) 699. [9] Loren C. Steinhauer, Review of field-reversed configurations, Phys. Plasmas 18, 070501 (2011). [10] J. Busnardo-Neto, P. L. Pritchett, A. T. Lin and J. M Dawson, A self-consistent magnetostatic particle code for numerical simulation of plasmas, J. Comput. Phys. 23 300, 1977. [11] T. Umeda and T. K. M. Nakamura, Electromagnetic linear dispersion relation for plasma with a drift across magnetic field revisited, Physics of Plasmas, 25, 102109 (2018). [12] T. Umeda, et al, A numerical electromagnetic linear dispersion relation for Maxwellian ring-beam velocity distributions, Physics of Plasmas, 19, 072107, (2012). [13] T. Umeda, Numerical study of electrostatic electron cyclotron harmonic waves due to Maxwellian ring velocity distributions, Earth Planets Space, 59, 12051210 (2007). [14] L. Muschietti and B. Lembege, Two-stream instabilities from the lower-hybrid frequency to the electron cyclotron frequency: application to the front of quasi-perpendicular shocks, Ann. Geophys., 35, 10931112, 2017. [15] C. Ruyer, Kinetic instabilities in plasmas: from electromagnetic fluctuations to collisionless shocks, Universite Paris-Sud, PhD thesis, Chap 1, 2014. [16] H. S. Xie, J. Zhu and Z. W. Ma, Darwin model in plasma physics revisited, Phys. Scr. 89 (2014) 105602. [17] K. Min and K. Liu, Fast magnetosonic waves driven by shell velocity distributions, J. Geophys. Res. Space Physics, 120, 27392753, 2015. [18] Daniel Verscharen, Kristopher G. Klein, Benjamin D. G. Chandran, Michael L. Stevens, Chadi S. Salem and Stuart D. Bale, ALPS: The Arbitrary Linear Plasma Solver, arXiv:1803.04697, 2018. [19] D. Summers, S.Xue and Richard M. Thorne, Calculation of the dielectric tensor for a generalized Lorentzian (kappa) distribution function, Phys. Plasmas, 1, 2012 (1994). [20] P. Astfalk, T. Gorler and F. Jenko, DSHARK: A dispersion relation solver for obliquely propagating waves in bi-kappa-distributed plasmas, J. Geophys. Res. Space Physics, 120, 71077120 (2015). [21] H. S. Xie, Y. Y. Li, Z. X. Lu, W. K. Ou, and B. Li, Comparisons and applications of four independent numerical approaches for linear gyrokinetic drift modes, Physics of Plasmas, 24, 072106 (2017). [22] Peter Kravanja and Marc Van Barel, Computing the Zeros of Analytic Functions, Springer, 2000.