Professional Documents
Culture Documents
Hybrid Vlasov-MHD Models: Hamiltonian vs. Non-Hamiltonian
Hybrid Vlasov-MHD Models: Hamiltonian vs. Non-Hamiltonian
This content has been downloaded from IOPscience. Please scroll down to see the full text.
(http://iopscience.iop.org/0741-3335/56/9/095008)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 150.244.199.44
This content was downloaded on 09/07/2014 at 06:38
Abstract
This paper investigates hybrid kinetic-magnetohydrodynamic (MHD) models, where a hot
plasma (governed by a kinetic theory) interacts with a fluid bulk (governed by MHD).
Different nonlinear coupling schemes are reviewed, including the pressure-coupling scheme
(PCS) used in modern hybrid simulations. This latter scheme suffers from being
non-Hamiltonian and is unable to exactly conserve total energy. Upon adopting the Vlasov
description for the hot component, the non-Hamiltonian PCS and a Hamiltonian variant are
compared. Special emphasis is given to the linear stability of Alfvén waves, for which it is
shown that a spurious instability appears at high frequency in the non-Hamiltonian version.
This instability is removed in the Hamiltonian version.
Another more important question emerges in the consequences. Indeed, we discover a spurious instability in the
formulation of hybrid kinetic-MHD models, viz., the particular non-Hamiltonian model.
type of coupling scheme that should be used in the model. Two The remainder of the paper is organized as follows. In
coupling schemes are present in the literature: the current- section 2 we review the two hybrid coupling schemes: the CCS
coupling scheme (CCS), found for example in [1, 3, 30] and and PCS models are derived from first principles and general
the pressure-coupling scheme (PCS), examples of which are comments about their structure are made. This is followed
used in [4, 7, 12]. While the CCS involves the hot momentum in section 3 by a general treatment of the linear problem for
and density the incompressible PCS models expanded about homogeneous
K= p f (x, p) d3 p, (1) isotropic equilibria in a uniform external magnetic field, by
integration over orbits. This is followed, in section 4, by
a study of the dispersion relation for transverse disturbances
n= f (x, p) d3 p, (2)
parallel to the magnetic field. It is in this special case that we
the PCS involves the following tensor: compare the Hamiltonian and non-Hamiltonian PCS models
and discover the spurious instability. For completeness we also
1 compare them to the CCS models. The dispersion relation is
P= pp f (x, p) d3 p, (3)
mh analyzed numerically and analytically and is shown to have a
crossover to instability at high frequencies. Next, in section 5,
which is the pressure tensor of the hot component, calculated comments are made about the behavior of perpendicular
with respect to a zero mean velocity. All these quantities are disturbances. Finally, in section 6 we summarize and conclude.
defined as above in terms of moments of the kinetic probability The paper contains two appendices that are included for
density f (x, p) on phase space. Here, p denotes the kinetic completeness. In appendix A the noncanonical Poisson
momentum p = mh v , while mh denotes the hot particle mass. brackets for the Hamiltonian hybrid models are given, while
Normally, the PCS is derived from the CCS [24, 25], under appendix B records some details of our calculations leading to
the assumption that the hot component is rarefied, so that the dispersion relation used in section 4.
its density levels are much lower than those of the MHD
component. Also, the PCS often appears in two different
versions depending on whether the definition of the pressure 2. Hybrid coupling schemes
tensor involves the absolute [4, 7] or relative [12, 29] velocity.
All of the nonlinear PCS models commonly found in Turning now to the two coupling schemes, we first consider
plasma physics literature suffer from the defect that they do not the CCS, then the PCS.
exactly conserve energy. Indeed, exact energy conservation
is lost when the assumption of a rarefied hot component is
inserted as an approximation in the equations of motion of the 2.1. Current-coupling schemes
model. Consequently, these models are not Hamiltonian field
In order to derive the hybrid CCS model [25], one starts with
theories, ones that are expected to have noncanonical Poisson
the equations of motion for a multifluid plasma in the presence
brackets akin to those introduced into plasma physics in [21]
of an energetic component. Upon formally neglecting the
for MHD and [19, 20] for the Vlasov equation. Since such
a Hamiltonian structure occurs for all good plasma models, vacuum permittivity (see e.g. [5]), one writes
in their non-dissipative limit, (see [16, 17, 19]), this would ∂ us
suggest there should be a Hamiltonian model for the PCS, ρs + us · ∇ us = −∇ ps
∂t
and indeed this was shown to be the case in recent literature +ρs as (E + us × B ) , (4)
[10, 23, 31]. This new model not only conserves energy, but
∂ρs
also conserves the cross-helicity invariants (which are also lost + ∇ · (ρs us ) = 0, (5)
in the non-Hamiltonian case). ∂t
∂f p ∂f p ∂f
A main goal of the present paper is to compare the + · + qh E + ×B · = 0,
Hamiltonian and non-Hamiltonian PCS models, with emphasis ∂t mh ∂ x mh ∂p
on linear stability analyses. It is important to make clear ∇ × B = µ0 J
that we are not arguing that dissipation is unimportant and = µ0 as ρs us + µ0 ah K , (6)
that the Hamiltonian description is the most apt description s
of hybrid plasmas; clearly this is not always the case— ∂B
collisional effects, albeit small, can give rise to important = −∇ × E , (7)
∂t
consequences, as is evident, for instance, from the massive
0= as ρs + qh n, ∇ · B = 0, (8)
body of reconnection studies in the literature. Rather, the
s
goal here is to investigate some consequences of nonphysical
dissipation (or drive) that exists in hybrid models when all the where as = qs /ms is the charge-to-mass ratio for the fluid
clearly identifiable physical dissipative terms are set to zero. species s, while ρs and us are its mass density and velocity,
This fake dissipation may also be small, as is often the case for respectively. The symbol ps , on the other hand, indicates the
physical dissipation, but could lead to substantial yet erroneous partial pressure of the fluid species s, which is assumed to be
2
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
a function of ρs , through the relation ps = ρs2 ∂ Us /∂ρs , with is governed by the Vlasov equation rather than gyrokinetic or
Us (ρs ) indicating the corresponding specific internal energy. drift-kinetic counterparts.
For simplicity, we consider from now on, the case in which Note, we always assume that the mean velocity V =
the bulk plasma is composed by two species, one consisting of m−1
h K /n of the energetic component is either very low or
ions and the second one of electrons. It is customary to reduce at most comparable with the MHD fluid velocity u. This
the two-fluid system by neglecting the inertia of the electron is consistent with the hypothesis of energetic particles, since
species (taking the limit m2 → 0), thereby obtaining a one- the latter hypothesis involves the temperature rather than the
fluid momentum equation. With this assumption, summation mean velocity. Denoting the temperatures of the hot and fluid
of equations (4) for s = 1, 2 produces components by Th and Tf , respectively, we have Th Tf
(see [4]). With the definition of the temperature Th =
∂ u1
ρ1 + u1 · ∇ u1 = (a1 ρ1 + a2 ρ2 ) E (mh /3nkB ) |v − V |2 f d3 v (where kB denotes Boltzmann’s
∂t
constant), the assumption on the energetic component amounts
+ (a1 ρ1 u1 + a2 ρ2 u2 ) × B − ∇ p , (9)
to an assumption on the trace of the second-order moment of
where p = p1 + p2 . Then, upon using Ampère’s law (6) and the Vlasov density with no assumption on the mean velocity,
the quasineutrality relation of (8), equation (9) becomes which is actually low for hot particles close to isotropic
equilibria [29].
∂ u1
ρ1 + u1 · ∇ u1 = −qh nE Notice that equation (14) involves the Lorentz force term
∂t
qh nu×B , which should normally be negligible for consistency
+ ( J − ah K ) × B − ∇ p , (10)
with the approximation mh n ρ, which yields equation (13)
while equation (4) for the second species yields from equation (11). A variant of the above CCS also exists [30],
1 which, by virtue of the approximation mh n ρ neglects the
E = − u2 × B + ∇ p2 term qh nu × B , but on the other hand retains the term K × B
a2 ρ2
1 1 even though the two terms are in principle of the same order
= (a1 ρ1 u1 + ah K − J ) × B + ∇ p2 . (as long as V is comparable with u).
a2 ρ2 a2 ρ2
One can check directly that the hybrid CCS model of (14)–
Next, one imitates the derivation of ideal MHD [5] and assumes
(17) exactly conserves the following total energy:
that J ×B and ∇ p2 are both negligible compared to the Lorentz
force a1 ρ1 u1 × B . This step leads to an Ohm’s law of the form 1 1
E= ρ|u|2 d3 x + f |p|2 d3 x d3 p
2
2mh
a1 ρ1 u1 + qh nV 1
E=− × B, (11) + ρ U (ρ) d x +3
|B |2 d3 x, (18)
a1 ρ1 + q h n 2µ0
where V = m−1 (see [30]) where U (ρ) is the internal energy per unit mass, from
h K /n is the hot mean velocity. Note, this
means magnetic flux is frozen-in at a velocity given by which the pressure is determined by p = ρ 2 ∂ U /∂ρ. Moreover,
this system is Hamiltonian, with a noncanonical Poisson
a1 ρ1 u1 + qh nV bracket [31] (recorded for completeness in appendix A) and
W = . (12)
a1 ρ1 + q h n it conserves the usual cross-helicity invariant u · B d3 x [10].
However, if V and u1 are comparable and a1 ρ1 qh nh , then
one can replace (11) by the Ohm’s law of ideal MHD, 2.2. Pressure-coupling schemes
Let us consider now the two models that use the PCS—first
E = −u1 × B , (13)
the non-Hamiltonian version then the Hamiltonian one.
and the magnetic flux is then frozen into the MHD bulk flow.
Finally, inserting (13) into equations (10), (5), and (7) yields 2.2.1. Non-Hamiltonian PCS. Once the CCS has been
the Hamiltonian CCS: obtained, the PCS can be derived by computing the evolution
∂u of the total momentum
ρ + u · ∇ u = −∇ p
∂t
+ (qh nu − ah K + J ) × B , (14) M := ρ u + K =: ρ U , (19)
3
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
Then, upon writing U ∼ u, we obtain the system with the following set of equations for the Hamiltonian PCS:
non-Hamiltonian PCS:
∂U
ρ + U · ∇ U = −∇ p − ∇ · P + J × B (27)
∂U ∂t
ρ + U · ∇ U = −∇ p − ∇ · P + J × B (22)
∂t ∂f p ∂f
+ +U · (28)
∂ρ ∂t mh ∂x
+ ∇ · (ρ U ) = 0 (23)
∂f
∂t + p × ah B − ∇ × U − p · ∇ U · = 0,
∂f p ∂f p ∂f ∂p
+ · + qh −U ×B· =0 (24)
∂t mh ∂ x mh ∂p ∂ρ ∂B
+ ∇ · (ρ U ) = 0, = ∇ × (U × B ) . (29)
∂B ∂t ∂t
= ∇ × (U × B ) , (25)
∂t We see that the fluid equation (27) is identical to the
corresponding equation (22) of the non-Hamiltonian model.
Notice that equation (22) is identical to the bulk momentum
However, in the Hamiltonian model, hot particles move with
equation of the hybrid PCS of Fu and Park (see equation (1)
the relative velocity U + p/mh so that the term ∇ U · p =
in [7, 8]), which also includes (23) and (25) (while p ·∇ U + p ×(∇× U ) appears as an inertial force. Consequently,
replacing Vlasov dynamics by its gyrokinetic approximation). both Hamiltonian and non-Hamiltonian PCS’ possess the same
Analogous PCS models with the same fluid equation (22) static equilibria, although the dynamics in the vicinity of these
have been formulated by Cheng [4] (see equation (1) therein) equilibria may be very different, depending on the particular
and Park et al [25] (see equation (3) therein). In some situation under consideration. Also, we notice that, unlike the
situations, the tensor (3) in (22) is replaced by the relative CCS of (14)–(17) and the non-Hamiltonian PCS of (22)–(25),
pressure tensor P = m−1 h (p − mh V )(p − mh V )f d3 p the Hamiltonian equations of (27)–(29) involve a nontrivial
(e.g., the PCS model proposed by Kim, Sovinec and kinetic-fluid coupling, even in the absence of magnetic
Parker [11, 12, 29]). fields.
All the above mentioned PCS models suffer from not
exactly conserving the total energy. Indeed, if we assume 3. Linearized incompressible PCS
that the total
energy is still given by (18), equations (22)–(25)
give Ė = U · ∂t K d3 x, so that the total energy would only In this section, we consider the linearized equations of
be nearly conserved if ∂t K is small. Under this assumption, motion for the incompressible limit (e.g. ∇ · U = 0) of
the CCS and the PCS are completely equivalent, since (16) both Hamiltonian and non-Hamiltonian PCS’. For the sake
yields of simplicity, we shall set all physical constants to unity
(including mh , so that p = v ), although we shall restore them
∂K at a later time.
= −∇ · P + ah K × B − qh nu × B . (26)
∂t
3.1. Equations of motion
However, the assumption that ∂t K is negligible is not
compatible with (16), since the time variation of K may indeed Upon following the standard procedure, we linearize each
play a role in the general case. variable as A = A0 + A1 , so that the subscripts ‘0’ and ‘1’
denote an equilibrium and its perturbation, respectively. As a
result, we obtain
2.2.2. Hamiltonian PCS. The issue of exact energy ∂ U1
= −∇ p1 − ∇ · vv f1 d3 v + (∇ × B1 ) × B0 , (30)
conservation was raised in [31], where an alternative ∂t
Hamiltonian version of the PCS was presented. Besides ∂f1 ∂f1 ∂f1
conserving the energy (18) exactly, this model possesses +v· + v × B0 · (31)
∂t ∂x ∂v
a Poisson bracket structure, which was derived by using = f0 (α vv : ∇ U 1 + β v · U1 × B0 ) ,
well established Hamiltonian techniques in geometric plasma ∂ B1
dynamics [14–21, 27, 28]. = ∇ × (U 1 × B 0 ) , ∇ · U1 = 0, (32)
∂t
In order to derive the Hamiltonian PCS model of [31], where the parameters α and β = 1 − α are inserted so that
one expresses the Hamiltonian structure of the CCS (14)–(17) α = 1 gives the linearized Hamiltonian model, while α = 0
in terms of the total momentum M in (19). Then, instead gives the linearized non-Hamiltonian model. In this way,
of replacing U ∼ u (arising from the original assumption it is clear that the α-terms identify the Hamiltonian model,
mh n ρ) in the equations of motion, one replaces U ∼ u while the β-terms identify its non-Hamiltonian counterpart.
directly in (18) and derives the equations of motion from In equations (30)–(32) we assumed a static equilibrium so that
the Poisson bracket structure written in terms of M (see U0 ≡ 0 and p0 ≡ 0. Also, we consider a uniform magnetic
appendix A). This procedure ensures that the chosen energy field B0 (aligned with the z-axis) and an isotropic equilibrium
functional is always preserved, as long as no approximations for the energetic component, so that f0 = f0 (v 2 /2) with
are made on the Poisson bracket. At this point, one obtains the v 2 = |v |2 . Notice that the special case B0 = 0 yields
4
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
free transport for the hot particles, in the case of the non- about the
equations of motion for the kinetic moments
Hamiltonian model (α = 0). Conversely, the Hamiltonian Ak(1) = v k f1 d3 v :
model (α = 1) retains the fluid velocity terms in the
kinetic equation, even in the absence of the magnetic field • the α-term (of the Hamiltonian model) contributes only
to moments of even order 2n + 2 (e.g. the pressure tensor
(B0 = 0).
P1 ), i.e., the α-term does not contribute to the dynamics
Notice that, although here we have chosen a Vlasov
of odd-order moments;
equilibrium of the form f0 = f0 (v 2 /2), other more
realistic choices are also available. For example, in actual • the β-term (of the non-Hamiltonian model) contributes
only to moments of odd order 2n + 1 (e.g. the averaged
hybrid simulations of the non-Hamiltonian PCS in fusion
momentum K1 ), i.e., the β-term does not contribute to the
devices, toroidal symmetry is involved and the use of
dynamics of even-order moments;
special equilibrium profiles becomes necessary [29]. Another
example arises in reversed field pinch plasmas, in which finite • the first three moments obey the equations
Larmor radius effects allow for equilibria of the type f0 = ∂ t n1 + ∇ · K 1 = 0
f0 (x, v 2 )−ωc−1 ∇f0 · v × B (where ωc = qh B0 /mh ) (see [11]).
∂t K1 + ∇ · P1 − K1 × B0 = −βn0 U1 × B0
On the other hand, the aim here is not to enter into detailed
∂t P1 + ∇ · A3(1) + B0 , P1 = −2α (P0 · ∇)U1
features of particular fusion devices, but to provide insight
into model differences. Therefore, we focus on distributions +((P0 · ∇)U1 )T ,
of the type f0 = f0 (v 2 /2).
1 ei(k·x−ωt) where [·, ·] denotes matrix commutator and we defined the
Assuming perturbations varying as A1 = A
hat operator by wa := w × a (for any two vectors w and
gives
ih = −ihk wk is an antisymmetric matrix).
a, so that w
ωU 1 = k p1 + (k · v )v f1 d3 v + B0 × (k × B1 ), (33) Thus, there is no contribution of the α-term to the linearized
fluid moments (i.e. zeroth and first-order moments) of the
−ωB 1 ,
1 = (k · B0 )U 1 = 0.
k·U (34) kinetic component: indeed, the α-term disappears in the
Dotting equation (33) by k and using equations (34) yields linearized dynamics of the fluid closure for the hot particles.
The α-term contributes only to the dynamics of perturbed
1 1 moments at even order (e.g. the pressure tensor). On the other
p1 = − (k · v )2 f1 dv + (k · B0 )(U
1 · B0 ),
hand, the β-term contributes only to the dynamics of perturbed
|k | 2 ω
moments with odd order, e.g. it has a non-zero contribution to
and the velocity equation becomes the first-order moment (which plays a crucial role in the CCS).
(k · v )2
ωU1 = − k − ( k · v ) v f1 dv
|k|2 3.3. Solution of the linearized Vlasov equation
1 1
+ (k · B0 )2 U (35) Now we solve the linearized kinetic equation (31) in terms
ω
of the fluid velocity U1 . This is done by invoking the
or, upon rearranging the various terms, method of characteristics (integrating over orbits) as is standard
in plasma physics texts (e.g. [13]) for the Maxwell–Vlasov
ω kk
U1 = 1− (k ·
P1 ), (36) system. Following this standard method yields the solution of
(ω2 − k · B0 )2 |k | 2 the Vlasov equation in the form
t
where P1 = vv f1 d3 v and 1 − kk/|k|2 projects transverse
f1 = f0 α∇ U1∗ : v ∗ v ∗ − β U1∗ · v ∗ × B0 dt ∗
to k. −∞
It remains to express f1 in terms of U
1 in order to obtain +f1 (x∗ (−∞), v ∗ (−∞), −∞),
the dispersion relation. This step is performed in section 3.3,
where f0 means derivative of f0 with respect to its argument
but first the next section contains some relevant properties of
v ∗ 2 /2 and the variables (x∗ (t ∗ ), v ∗ (t ∗ ), t ∗ ) satisfy
the linearized equation of Vlasov kinetic moments.
ẋ∗ (t ∗ ) = v ∗ (t ∗ ), v̇ ∗ (t ∗ ) = v ∗ (t ∗ ) × B0 ,
3.2. Remarks on linearized moment dynamics
with x∗ (t) = x, v ∗ (t) = v and the dot indicating the derivative
Before analyzing the linear dynamics, it is of interest to explore with respect to the evolution parameter t ∗ .
some roles played by the α and β terms. To this end it is useful Then, upon introducing the notation ωB0 = |B0 | (i.e. the
to introduce equilibrium moments cyclotron frequency, upon restoring physical quantities) and
the planar rotation
(0)
An i1 i2 ...in = vi1 . . . vin f0 d3 v .
cos(ωB0τ ) − sin(ωB0τ ) 0
(0)
since f0 = f0 (v 2 /2), we notice that A2n+1 = 0. 0 ) = sin(ωB τ ) cos(ωB τ ) 0
R(τ ) = exp(τ B 0 0
5
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
and its antiderivative AR(τ ), we have by taking the first-order moment of (37). Upon integrating by
parts and recalling RT k = Rk = k, we have
x∗ = AR(τ )v + ωB−2 B0 × v + x,
1 = v f1 (v ) dv
0
K
together with v ∗ = R(τ )v and τ = t ∗ − t. Upon Fourier- 0
transforming in the spatial variable, we obtain ∂f0 1
= v · iαkz vz RT (τ )U
0 ∂ v⊥ −∞
f1 = 1 · v∗ ) + β U
f0 iα(k · v ∗ )(U 1 · B0 × v ∗ −β B0 × RT (τ )U 1 ei(kz vz τ −ωτ ) dτ dv⊥ dvz
−∞
i(k·X −ωτ )
0
×e dτ 1
=− f0 iαkz vz RT (τ )U
0 −∞
= 1 − β B0 × U
f0 iα(k · R(τ )v )U 1 · R(τ )v
−∞ −β B0 × RT (τ )U 1 ei(kz vz τ −ωτ ) dτ dv
i(k·X −ωτ )
×e dτ, 0
=− f¯0 (vz2 /2) iαkz vz RT (τ )U 1
where X := x − x and recall vz∗ = vz and z∗ (τ ) = vz τ + z,
∗
−∞
so that RB0 = B0 and (RU 1 )z = U 1z . Thus, since R is a
−β B0 × RT (τ )U 1 ei(kz vz τ −ωτ ) dτ dvz .
rotation,
0
∂f0 Hence,
f1 = · iα(k · R(τ )v )RT (τ )U 1 1z = 0.
K
−∞ ∂ v
1 ei(k·X −ωτ ) dτ.
−β B0 × RT (τ )U (37) The above relation means that the momentum perturbation K 1
is coplanar with U1 , i.e., K1 × U1 = 0 and therefore the
Finally, upon recalling the definition w a := w × a of hat density perturbation vanishes, since 1 /ω ≡ 0.
n 1 = −k · K
operator, the velocity equation becomes ¯
Here, we have introduced the notation f0 (vz /2) = f0 dv⊥ .
2
2
ω − ( k · B 0 )2 U1 Notice that, by proceeding analogously, we have
0 0
kk ∂f0 1
=ω 1− ( k · v ) R iα(k · Rv ) k · P1 = − f¯0 (vz2 /2) kz vz iαkz vz RT U
|k| 2
−∞ ∂v −∞
0 ∂f0 · U 1 v ei(k·X −ωτ ) dτ dv . −β B0 × RT U 1 ei(kz vz −ω)τ dτ dvz .
+β B (38)
∂v
At this point one needs to compute the matrix integral
and the dispersion relation is
0
A= ei(kz vz −ω)τ R(τ ) dτ,
det k 2 (k · B0 )2 − ω2 1 −∞
0 whose components are
−ω
k2 0
(k · v ) v iα(k · Rv )1 + β B
−∞ k z vz − ω
A11 = A22 = −i , (40)
∂f0 i(k·X −ωτ ) (kz vz − ω)2 − ωB20
×R e dτ dv = 0. (39)
∂v ωc
A12 = −A21 = , (41)
At this point, one may write the general dispersion relation (kz vz − ω)2 − ωB20
explicitly. However, we study the linearized system in two i
A33 =− .
particular cases where k⊥ = 0 and kz = 0, which we turn to kz v z − ω
in the next sections. In the above equations, the integrations are defined for
Im(ω) > 0; following the standard procedure the solution
4. Disturbances with k⊥ = 0 is extended to the lower complex plane by analytical
continuation. In conclusion, we have
Now we specialize the preceding results to the spe-
K 0 AT U
1 = − f¯0 (vz2 /2) iαkz vz 1 − β B 1 dvz ,
cial case k⊥ = 0, thus giving the dispersion rela-
tion for parallel propagating transverse disturbances with 0ij = −ij k B0k and (AU1 )z = 0, so that
1 = −k · U1 × B0 = 0. where we recall B
k·E
K1z = 0. Similarly, we obtain
4.1. Dispersion relation k · 0 AT U
P1 = − kz vz f¯0 iαkz vz 1 − β B 1 dvz . (42)
After setting k⊥ = 0 in equation (38) it is useful to compute
the perturbed moment quantities K1 and k · P1 . Notice that Once the moment quantities K 1 and k · P1 are written
k⊥ = 0 implies kz U1z = 0 ⇒ U1z = 0. We compute K 1 explicitly, one is ready to write the dispersion relation for the
6
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
γ/ωc
At this point, direct algebraic computations on the above −3
relation give the following dispersion relation:
−4 κ =1
±
D (ω, kz ) := ω − kz vA + ω(αω ∓ ωc )n0 1 +
2 2 2
κ = 10
∞ −5 κ = 50
F
(ω ∓ ωc ) dvz = 0, (44)
−∞ kz vz − ω ± ωc −6
where all physical constants have been restored: B0 = B0 ez , 0 1 2 3 4
kzvA/ωc
ωc = qh B0 /mh is the cyclotron frequency of the energetic
√
component, vA = B0 / µ0 ρ indicates the Alfvén speed based −3
on the constant equilibrium magnetic field B0 and the constant x 10
∞ 2
bulk mass density ρ, and F := f¯0 /n0 with n0 = −∞ f¯0 dvz , so (b)
that n0 is a dimensionless number indicating the ratio between 1.5
the equilibrium mass density of the energetic component and
that of the bulk component. (For details of this calculation see 1
appendix B.) 0.5
c
γ/ω
7
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
However, upon examination of figure 1 for larger values which indicates damping for both polarizations, except for
of kz or ωr ≈ kz vA , we see that figures 1(a) and (b) differ as the upper sign at ωr = kz vA = ωc where the damping
ωr approaches and exceeds ωc . Most significantly, we see that vanishes. However, upon setting α = 0 we obtain for the
the non-Hamiltonian PCS possess an instability for frequencies non-Hamiltonian PCS, the following:
greater than ωc , as is clearly evident in figure 1(b). Since the
± ωr ω r ± ωc
equilibrium we are considering has no available free energy, Di (ωr , kz ) = π n0 ωc (ωc ∓ ωr ) F , (51)
in either the bulk or in the hot particles, this instability must k kz
be nonphysical and reflects the lack of energy conservation in which reveals the strange nonphysical crossover to instability
non-Hamiltonian PCS. For the Hamiltonian PCS displayed in for one of the polarizations when ωr > ωc .
figure 1(a), the system damps as expected. The hot particles For the record, a calculation similar to that for the PCS
provide Landau damping, in much the way one expects for gives for the CCS the dispersion relation
electron Landau damping of Alfvén and whistler modes, with D ± (kz , ω) = ω2 − kz2 vA2 (52)
the mode at ωc being undamped for one of the polarizations. +∞
F
For κ = 1, it is easily shown by residue calculus that + ωωc n0 ωc dvz ∓ 1
−∞ kz vz − ω ± ωc
+∞
F 1 whence we obtain for the Hamiltonian CCS, the following
dvz = − ,
−∞ kz vz − ω ± ωc ikz v0 + ω ∓ ωc ωr ωr ± ωc
Di± (ωr , kz ) = π n0 ωc2 F . (53)
where we recall F = f¯0 (vz2 )/n0 and f¯0 (vz2 ) is obtained from kz kz
equation (44) for κ = 1 (i.e. f0(1) (v 2 )), upon integrating out the Although (53) indicates a damping rate that is different from
perpendicular components of the velocity. Thus (44) becomes that of the Hamiltonian PCS, it does not possess the spurious
kz v0 (αω ∓ ωc ) instability possessed by the non-Hamiltonian PCS.
ω2 − kz2 vA2 + in0 ω = 0. The damping rates indicated by (50), (51), and (53)
ikz v0 + ω ∓ ωc
have several features in common. First, for low frequencies,
From which one obtains for n0 1, by expanding about ωr ωc , their intended regime, they all agree. Next, they all
ω = kz vA + iγ + δωr , the perturbed frequency scale with F (as opposed to its derivative) which is appropriate
n0 kz v0 (αkz vA ∓ ωc )(kz vA ∓ ωc ) for parallel propagating transverse waves for all isotropic
γ =− (46) equilibrium distribution functions (not just Maxwellians) [13].
2 kz2 v02 + (kz vA ∓ ωc )2
For higher frequencies, (50), (51), and (53) disagree so it is
n0 kz2 v02 (αkz vA ∓ ωc ) useful to compare with a full kinetic theory with electrons,
δωr = − . (47)
2 kz2 v02 + (kz vA ∓ ωc )2 ions, and hot particle components. For cold electron and ion
temperatures, only hot species contributes to the damping, and
From (46) we see explicitly the spurious crossover to instability it is an elementary exercise to show that Di for this case behaves
at ωc observed in figure 1(b) that occurs for α = 0. Similar, precisely as (53), the result for the CCS. Thus, in this frequency
although progressively more complicated, formulae exist for range the CCS gives the best answer, although the Hamiltonian
higher values of κ (see, e.g., relation (117) in [26]), but we will PCS may be reasonable. Clearly, the non-Hamiltonian result
not present these here. is unsatisfactory.
Finally, we further explore the differences between
the Hamiltonian and non-Hamiltonian models for arbitrary
5. Disturbances with kz = 0
isotropic equilibria by examining the so-called small-γ
approximation for each. Thus, we assume the resonant This Section presents the dispersion relation for certain
denominator of (44) gives rise to weak damping, and write linear waves propagating transversely to the magnetic field.
ω = ωr + iγ , D = Dr± + iDi± , and then expand as usual to These modes are allowed by the Hamiltonian PCS model
obtain (27)–(29) (with ∇ · U = 0), while they are forbidden by
Di± (ωr , kz ) the (incompressible) non-Hamiltonian variant (22)–(17). In
Dr± (ωr , kz ) = 0, γ =− . (48) particular, we study the special case U 1⊥ = 0, which is
∂Dr± (ωr , kz )/∂ωr 1 = 0.
consistent with the incompressibility relation k · U
For n0 1, ∂Dr± /∂ωr ≈ 2kz vA , and, thus, γ ≈ In order to find the dispersion relation, we specialize
−Di± (kvA , kz )/(2kvA ). Using the Plemelj relations we obtain equation (36) by setting kz = 0. In turn, this affects the Vlasov
perturbation (37). Since U1 = U1z ez and
the following form (44):
1z (B0 × RT (τ )ez ) = U
1 = U
B0 × RT (τ )U 1z (B0 × ez ) = 0,
Di± (ωr , kz ) = (49)
ωr ωr ± ω c
π n0 (αωr ∓ ωc )(ωr ∓ ωc ) F . the Vlasov perturbation (37) becomes
kz kz 0
∂f0
For the Hamiltonian PCS, α = 1 and f1 = iα (k · R(τ )v ) · R (τ )U 1 ei(k·X −ωτ ) dτ
T
−∞ ∂v
0
ωr ωr ± ω c 1z ∂f0
Di± = π n0 (ωr ∓ ωc )2 F , (50) = iα U v⊥ · RT (τ )k ei(k⊥ ·X⊥ −ωτ ) dτ,
kz kz ∂vz −∞
8
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
which shows how the non-Hamiltonian model (α = 0) Department of Energy Contract No. DE-FG05-80ET-53088.
precludes the existence of transversal modes such that ET acknowledges financial support from the Agence Nationale
1⊥ = 0. In what follows, we consider the Hamiltonian case
U de la Recherche (ANR GYPSI) and from the CNRS PEPS
by setting α = 1. project GEOPLASMA.
Notice that, since X⊥ does not depend on vz , the above
n1 = f1 d3 v = 0. Therefore, the relation
expression yields
n1 = k · K1 allows the case K 1 = K 1z ez , where Appendix A. Poisson brackets for Hamiltonian
0 hybrid MHD models
1z = iα U1z ∂f0
K vz v⊥ · RT (τ )k ei(k⊥ ·X⊥ −ωτ )
∂vz −∞
×dτ dv⊥ dvz A Hamiltonian system is a dynamical system generated by a
0 given Hamiltonian (total energy) and a Poisson bracket in the
= −iα U 1z k⊥ · f0 R(τ )v ei(k⊥ ·X⊥ −ωτ ) dτ d3 v. form ∂
/∂t = {
, H }, where
denotes the set of dynamical
−∞ variables. The Poisson bracket {·, ·} must be a bilinear,
Then, combined with the velocity relation (36) and making use antisymmetric operator defined on the space of function(al)s.
of the moment equation for K1 in section 3.2, the special case In addition it must satisfy he Leibniz property
1z = K
1⊥ = 0 yields U
K 1z along with the dispersion relation
0 {F G, H } = G{F, H } + F {G, H }
1+i (k⊥ · R(τ )v ) f0 ei(k⊥ ·X⊥ −ωτ ) dτ d3 v = 0.
−∞ as well as the Jacobi identity
This is an expected Bessel function type of dispersion relation
and its detailed study is left for future work. {{F, G}, H } + {{H, F }, G} + {{G, H }, F } = 0.
6. Summary and conclusions The Leibniz property, bilinearity and antisymmetry are easily
built into the generic form of the Poisson bracket, but the
After a review of hybrid kinetic-MHD models, we presented a proof of the Jacobi identity may require some effort. (See
comparative study of Hamiltonian and non-Hamiltonian PCSs, [16–19] for review and the appendix of [22] for a particularly
where the latter suffer by not exactly conserving energy. In onerous direct proof.) Such Poisson brackets need not have the
particular, the two models were compared from the point of canonical form of conventional field theories and may possess
view of linear stability and their dispersion relations were degeneracy—because of this they were called noncanonical
presented and analyzed. The special cases of pure parallel in [21].
and perpendicular wave propagation were considered.
The Poisson brackets for the Hamiltonian models of the
Upon considering isotropic equilibria for the hot
present paper were given in [31], where it was also shown how
component, it was shown that the non-Hamiltonian PCS
possesses an instability absent in its Hamiltonian variant and they may be used to formulate new hybrid MHD models that
in the CCS, which is also Hamiltonian. We argued that conserve energy exactly. Indeed, while exact conservation
the instability emerging in the non-Hamiltonian model is not of (18) is guaranteed for the CCS model (14)–(17) by its
physically viable. Extensive investigation of the dispersion noncanonical Poisson bracket
relation will be considered in future work.
δF δG 3
Although the unstable mode is of large frequency and thus {F, G}CCS = m · , d x (A1)
δm δm
outside the original intent of the PCS models, which were
δF δG δG δF
developed to describe low-frequency behavior, their presence − ρ ·∇ − ·∇ d3 x
would suggest results obtained from non-Hamiltonian PCS δm δρ δm δρ
models extended into this regime should be viewed with δF δG
+qh f B · ×
caution. Even if some artifice, numerical or other, were used to δm δm
suppress the unphysical linear instability, nonlinear coupling
δF ∂ δG δG ∂ δF 3 3
could give rise to differences in their turbulent transport − × + × d xd p
δ m ∂ p δf δ m ∂ p δf
behavior.
δF δG
+ f ,
δf δf
Acknowledgments
∂ δF ∂ δG 3 3
+qh B · × d xd p
CT is indebted with J Carrillo, P Degond, D D Holm, ∂ p δf ∂ p δf
G Lapenta and C Sovinec for several interesting discussions. δF δG δG δF
Partial support by the Institute of Mathematics and its + B· ×∇ × − ×∇ × d3 x,
δm δB δm δB
Applications Grant No. SGS27/13 is greatly acknowledged.
P J M gratefully acknowledges stimulating conversations the PCS models available in the literature fail to conserve
with B Breizman, C-Z Cheng, C W Horton, F Pegoraro energy exactly. For the Hamiltonian PCS (HPCS) (27)–(29),
and F Waelbroeck. His research was supported by the US exact conservation of (14)–(17) follows from the Poisson
9
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
10
Plasma Phys. Control. Fusion 56 (2014) 095008 C Tronci et al
[17] Morrison P J 1998 Hamiltonian description of the ideal fluid [24] Park W, Belova E V, Fu G Y, Tang X Z, Strauss H R and
Rev. Mod. Phys. 70 467–521 Sugiyama L E 1999 Plasma simulation studies using
[18] Morrison P J 2009 On Hamiltonian and action principle multilevel physics models Phys. Plasmas 6 1796–803
formulations of plasma dynamics AIP Conf. Proc. [25] Park W et al 1992 Three-dimensional hybrid
1188 329–44 gyrokinetic-magnetohydrodynamics simulation Phys.
[19] Morrison P J 1982 Poisson brackets for fluids and plasmas AIP Fluids B 4 2033–7
Conf. Proc. 88 13–46 [26] Podesta J J 2004 Plasma dispersion function for the kappa
[20] Morrison P J 1986 The Maxwell–Vlasov equations as a distribution NASA Report NASA/CR–2004-212770
continuous Hamiltonian system Phys. Lett. A 80 383–6 [27] Spencer R G 1982 The Hamiltonian structure of multi-species
[21] Morrison P J and Greene J M 1980 Noncanonical Hamiltonian fluid electrodynamics AIP Conf. Proc. 88 121–6
density formulation of hydrodynamics and ideal [28] Spencer R G and Kaufman A N 1982 Hamiltonian structure of
magnetohydrodynamics Phys. Rev. Lett. two-fluid plasma dynamics Phys. Rev. A 25 2437–9
45 790–4 [29] Takahashi R, Brennan D P and Kim C C 2009 Kinetic effects
[22] Morrison P J 2013 A general theory for gauge-free lifting of energetic particles on resistive MHD stability Phys. Rev.
Phys. Plasmas 20 012104 Lett. 102 135001
[23] Morrison P J, Tassi E and Tronci C 2014 Energy stability [30] Todo Y, Sato T, Watanabe K, Watanabe T H and Horiuchi R
analysis for a hybrid fluid-kinetic plasma model Nonlinear 1995 Magnetohydrodynamic Vlasov simulation of the
Physical Systems—Spectral Analysis, Stability and toroidal Alfvén eigenmode Phys. Plasmas 2 2711–6
Bifurcations ed O Kirillov and D Pelinovsky (New York: [31] Tronci C 2010 Hamiltonian approach to hybrid plasma models
Wiley) (doi:10.1002/9781118577608.ch14) J. Phys. A: Math. Theor. 43 375501
11