Draft version January 4, 2018

Typeset using LATEX twocolumn style in AASTeX61

NONLINEAR TRANSVERSE CASCADE AND BALANCES IN MRI-TURBULENCE IN KEPLERIAN DISKS

WITH A NET VERTICAL MAGNETIC FIELD

D. Gogichaishvili,1 G. Mamatsashvili,2, 3, 4 W. Horton,5, 6 and G. Chagelishvili4, 7

1 Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA
2 Helmholtz-Zentrum Dresden-Rossendorf, Dresden D-01328, Germany
3 Department of Physics, Faculty of Exact and Natural Sciences, Tbilisi State University, Tbilisi 0179, Georgia
4 Abastumani Astrophysical Observatory, Ilia State University, Tbilisi 0162, Georgia
5 Institute for Fusion Studies, The University of Texas at Austin, Austin, Texas 78712, USA
6 Space and Geophysics Laboratory, The University of Texas at Austin, 10000 Burnet Rd, Austin, Texas 78758, USA
7 Institute of Geophysics, Tbilisi State University, Tbilisi 0193, Georgia

ABSTRACT

Corresponding author: G. Mamatsashvili

g.mamatsashvili@hzdr.de
2
1. INTRODUCTION
Anisotropic turbulence offers a mechanism to ac-
count for enhanced transport of angular momen-
tum in astrophysical disks (Shakura & Sunyaev 1973;
Lynden-Bell & Pringle 1974) - isotropic turbulence is
unable to ensure such a transport. It is not surpris-
ing that from the 1980s, research in astrophysical disks
focused on identifying sources of turbulence and un-
derstanding its statistical characteristics. The turn-
ing point was the beginning of the 1990s, when a
linear instability mediated by a weak vertical mag-
netic field in differentially rotating conducting fluids
(Velikhov 1959; Chandrasekhar 1960) was rediscov-
ered for sufficiently ionized astrophysical disk config-
uration by Balbus & Hawley (1991). This instability,
referred to as magnetorotational instability (MRI), leads
to exponential growth of axisymmetric perturbations,
gives rise to and steadily supplies with energy mag-
netohydrodynamic (MHD) turbulence, as was demon-
strated in earlier numerical simulations shortly after
the importance of linear MRI in disks had been real-
ized (e.g., Hawley et al. 1995; Brandenburg et al. 1995;
Balbus & Hawley 1998).
The nonnormality of strongly sheared Keplerian flow
of astrophysical disks results in an additional lin-
ear energy supply to the turbulence - transient, or
nonmodal growth of non-axisymmetric perturbations
(Johnson 2007; Shtemler et al. 2011; Pessah & Chan
2012; Salhi et al. 2012; Mamatsashvili et al. 2013;
Squire & Bhattacharjee 2014; Zhuravlev & Razdoburdin
2014). Although the latter is generally less power-
ful than the classical (exponentially growing) MRI, it
is nevertheless capable of driving quite robust MHD
turbulence in the absence of the classic MRI, for in-
stance, in disks with purely azimuthal/toroidal magnetic
fields (e.g., Hawley et al. 1995; Fromang & Nelson 2006;
Simon & Hawley 2009; Guan et al. 2009; Guan & Gammie
2011; Flock et al. 2012; Nauman & Blackman 2014;
Meheut et al. 2015; Gogichaishvili et al. 2017). In such
spectrally stable magnetized disk flows lacking expo-
nentially growing MRI modes, nonlinear processes are
vital for turbulence sustenance. In this situation, the
canonical – direct and inverse – cascade processes (ac-
cepted in classical Kolmogorov or Iroshnikov-Kraichnan
theories of turbulence) are not capable of sustaining the
turbulence – this role is taken over by a new type of
the cascade processes, so-called, nonlinear transverse
cascade (Mamatsashvili et al. 2014). What is the physi-
cal reason for the emergence of the nonlinear transverse
cascade in shear flows and what is its specific nature?
The thing is, the anisotropy of the nonnormality/shear-
induced linear dynamics entails the anisotropy of the
nonlinear processes: transverse/angular redistribution
of perturbation harmonics in Fourier (wavenumber)
space. In spectrally stable Keplerian shear flows, in
which the only mechanism for the energy supply to
perturbations is the linear transient growth process,
the nonlinear transverse cascade continually repopu-
lates with perturbation harmonics those areas of Fourier
space, where they can undergo transient amplification,
and in this way ensures a long-term sustenance of the
turbulence. In the absence of such a feedback, tran-
siently growing non-axisymmetric modes get sheared
away and eventually decay. This process has been stud-
ied in detail in our recent paper (Gogichaishvili et al.
2017) for the case of Keplerian flow with a net az-
imuthal magnetic field by combining direct numerical
simulations and, based on them, the subsequent analy-
sis of the dynamical processes in Fourier space. It has
been shown that the nonlinear transverse cascade does
ensure the permanent regeneration of transiently grow-
ing harmonics, thereby yielding the sustained robust
subcritical turbulence in the flow. Thus, in spectrally
stable MHD shear flows, the existence of turbulence is
due to a suitable interplay of linear transient growth
and the nonlinear transverse cascade processes.
In this paper, we consider a local model of a weakly
ionized Keplerian disk threaded by a net vertical
magnetic field, where the classical MRI with perma-
nent exponential growth of axisymmetric modes takes
place (Balbus & Hawley 1991; Hawley & Balbus 1992;
Wardle 1999; Pessah et al. 2006; Lesur & Longaretti
2007; Pessah & Chan 2008; Shakura & Postnov 2015;
Latter et al. 2015; Mohandas & Pessah 2017). As a re-
sult, there is no deficit in energy supply and the role
of nonlinearity in the sustenance of perturbations is
not vital, as opposed to the azimuthal field case. How-
ever, MRI relies on the shear of disk flow and, con-
sequently, the underlying nonlinear processes are nec-
essarily anisotropic in Fourier space. This anisotropy
above all gives rise to the nonlinear transverse cascade
that is an important player in forming the overall dy-
namical picture of MHD turbulence driven by vertical
field MRI. The demonstration of this is one of the main
goals of the present paper.
Recently, Murphy & Pessah (2015) have investigated
the saturation of vertical field MRI in disks and the
properties of the resulting MHD turbulence both in
physical and Fourier space. Their study was mainly
devoted to characterizing the anisotropic nature of this
turbulence. They also noted a general lack of analy-
sis of the anisotropy in the existing studies of MRI-
turbulence: “Although there have been many stud-
ies of the linear phase of the MRI and its nonlinear

evolution, only a fraction have explored the mecha-
nism responsible for its saturation in detail, and none
have focused explicitly on the evolution of the degree
of anisotropy exhibited by the magnetized flow as it
evolves from the linear regime of the instability to the
ensuing turbulent state”. The main reason for overlook-
ing the anisotropic nature of MRI-driven turbulence in
most previous works, which focused on its spectral dy-
namics (e.g., Fromang & Papaloizou 2007; Simon et al.
2009; Davis et al. 2010; Lesur & Longaretti 2011) was
a somewhat misleading mathematical treatment, specif-
ically, spherical shell-averaging procedure in Fourier
space (borrowed from forced MHD turbulence studies
without shear flow, see e.g., Verma 2004; Alexakis et al.
2007), employed to extract statistical information about
the properties of MRI-driven turbulence. Obviously,
the use of shell-averaging, which, in fact, wipes out
the transverse cascade, is, strictly speaking, suitable
for isotropic turbulence, but by no means for shear
flow turbulence and therefore for its special case MRI-
turbulence. Thus, the shell-averaging is not an optimal
tool for analyzing spectral energy distributions as well
as dynamical processes in MRI-driven turbulence that
is far from isotropic (see also Nauman & Blackman
2014; Meheut et al. 2015; Murphy & Pessah 2015;
Gogichaishvili et al. 2017). This fact also calls into
question the investigation of power-law character and
corresponding spectral indices of energy spectrum of the
turbulence, which is usually carried out based on the
shell-averaging, because it also wipes out the anisotropy
of the spectral energy distribution that is a direct con-
sequence of the transverse cascade. Apparently, this is
the reason why the kinetic and magnetic energy spectra
do not generally display a good/well-defined power-law
behavior in all the turbulent states of disk flows with a
net vertical field (Simon et al. 2009; Lesur & Longaretti
2011; Walker et al. 2016).
In view of the above, in this work we focus on the
dynamics and balances of MRI-turbulence in disks with
a net vertical/poloidal magnetic field. We adopt the
local shearing-box model of the disk with constant
vertical thermal stratification and perform the anal-
ysis in three-dimensional (3D) Fourier space in full,
i.e., without doing the above-mentioned averaging of
spectral quantities over spherical shells of constant
wavenumber magnitude k = |k|. This allows us to
capture the spectral anisotropy of MRI-turbulence due
to shear and the resulting nonlinear angular redistri-
bution of perturbation harmonics in Fourier space, i.e.,
transverse cascade, thereby getting a deeper under-
standing of spectral and statistical properties of the
turbulence. In previous relevant studies on vertical
3
field MRI (e.g., Goodman & Xu 1994; Hawley et al.
1995; Sano & Inutsuka 2001; Lesur & Longaretti 2007;
Bodo et al. 2008; Latter et al. 2009, 2010; Simon et al.
2009; Pessah & Goodman 2009; Longaretti & Lesur
2010; Pessah 2010; Bai & Stone 2014; Murphy & Pessah
2015), this angular redistribution (scatter) of modes in
Fourier space was attributed to secondary, or parasitic
instabilities. Specifically, the most unstable, exponen-
tially growing axisymmetric MRI modes (channels) are
subject to secondary instabilities of non-axisymmetric
modes with growth rates proportional to the amplitude
of these channel solutions. In this way, the parasitic
instabilities redistribute the energy from the primary
axisymmetric channel modes to non-axisymmetric par-
asitic ones, halting the exponential growth of the former
and leading to the saturation of MRI. Several approx-
imations are made in this description: 1. the large
amplitude channel mode is a time-independent back-
ground off which the small-amplitude parasitic modes
feed and 2. the effects of the imposed vertical field,
the Coriolis force, and the basic Keplerian shear are all
usually neglected. These assumptions clearly simplify
the analysis of the excitation and dynamics of the non-
axisymmetric modes, but, most importantly, because
of neglecting the basic flow shear, omit independent
from the primary MRI modes source of their generation
- the transient/nonmodal growth of non-axisymmetric
modes, which, as discussed above, is an inevitable linear
process in shear flows. The importance of linear tran-
sient growth phenomenon itself should not be underesti-
mated – it is able to ensure onset of MHD turbulence in
the absence of purely exponential growing MRI in disk
flows with an azimuthal field (e.g., Hawley et al. 1995;
Simon & Hawley 2009; Gogichaishvili et al. 2017).
To keep the analysis general and self-consistent, to-
gether with the nonmodal growth, we employ the con-
cept of the nonlinear transverse cascade in the present
problem of MRI with vertical field that naturally encom-
passes the secondary instabilities. This unifying frame-
work enables us to correctly describe the interaction
between channel and non-axisymmetric parasitic modes
when the above assumptions break down – the ampli-
tudes of these two mode types become comparable, so
that one can no longer separate the channel as a primary
background and parasites as small perturbations on top
of that. This is the case in the developed turbulent
state of vertical field MRI, where channels undergo re-
current amplifications (bursts) and decays (disruptions)
due to the parasitic modes (e.g., Sano & Inutsuka 2001;
Lesur & Longaretti 2007; Bodo et al. 2008; Simon et al.
2009; Bai & Stone 2014; Murphy & Pessah 2015). As
it will be shown, the transverse cascade accounts for
4
the transfer of energy from the axisymmetric channel
modes to non-axisymmetric ones, forming a broad spec-
trum of the latter. The nonlinear transverse cascade
may not be the vital source of the energy supply to tur-
bulence in the presence of purely exponentially growing
MRI, but still it shapes the dynamics, sets the satura-
tion level/strength and determines the overall “design”
of the turbulence. In the spirit of our recent works
(Horton et al. 2010; Mamatsashvili et al. 2014, 2016;
Gogichaishvili et al. 2017), here we investigate in detail
the roles of different linear and nonlinear (transverse
cascade) dynamical processes underlying/shaping the
vertical field MRI-turbulence. Namely, we first per-
form direct numerical simulations of the turbulence and
then, using the simulation data, explicitly calculate
and visualize individual linear and nonlinear terms in
spectral equations and analyze them and their inter-
play in detail. The present study is a generalization of
the works by Simon et al. (2009); Lesur & Longaretti
(2011), where the dynamics of vertical field MRI-
turbulence – the spectra of energy, injection and non-
linear transfers – were analyzed in Fourier space, how-
ever, using a restrictive approach of shell-averaging,
which misses out the shear-induced anisotropy of the
turbulence and hence the interaction of axisymmetric
channel and non-axisymmetric modes. It also extends
the study of Murphy & Pessah (2015), who focused on
the anisotropy of vertical field MRI-turbulence in both
physical and Fourier space and examined energy spec-
tra and Maxwell stresses during the linear growth stage
of the channel solutions and after saturation, but not
the action of linear and nonlinear terms governing their
evolution.
The paper is organized as follows. The physical model
and derivation of dynamical equations in Fourier space
is given in Section 2. Numerical simulations of the MRI-
turbulence with vertical field and its general character-
istics, such as time-development, energy spectra and dy-
namically active modes are given in Section 4. The anal-
ysis of the the linear and nonlinear processes in Fourier
space and their interplay is described in Section 5. A
summary and discussion are given in Section 6.
2. PHYSICAL MODEL AND BASIC EQUATIONS
We use a shearing-box model to investigate the essence
of MHD turbulence driven by the classical MRI in
Keplerian disks. The shearing-box represents a local
Cartesian coordinate frame centered at a fiducial point
that co-rotates with the disk at angular velocity Ω
(Hawley et al. 1995). In this frame, we consider the
flow of an incompressible conducting fluid with constant
kinematic viscosity ν, thermal diffusivity χ and Ohmic
resistivity η. The flow is thermally stratified in the ver-
tical direction. Adopting the Boussinesq approximation
for the vertical stratification (Balbus & Hawley 1991;
Lesur & Ogilvie 2010), the basic equations of non-ideal
MHD become
∂U 1 (B · ∇) B
+ (U · ∇)U = − ∇P + − 2Ω × U
∂t ρ 4πρ
+ 2qΩ2 xex − ΛN 2 θ ez + ν∇2 U, (1)
∂θ uz
+ U · ∇θ = + χ∇2 θ, (2)
∂t Λ
∂B
= ∇ × (U × B) + η∇2 B, (3)
∂t
∇ · U = 0, (4)
∇ · B = 0, (5)
where ex , ey , ez are the unit vectors along the carte-
sian coordinates (x, y, z), respectively, in the radial, az-
imuthal and vertical directions, ρ is the density, U is the
velocity, B is the magnetic field, P is the total pressure,
equal to the sum of the thermal and magnetic pressures,
θ ≡ δρ/ρ is the perturbation of the density logarithm
(or entropy, since pressure perturbations are neglected
in the Boussinesq approximation). The shear parame-
ter q = −d ln Ω/d ln r is set to q = 3/2 for a Keplerian
disk. N 2 is the Brunt-Väisälä frequency squared that
controls the stratification in the Boussinesq approxima-
tion. The affects of thermal stratification are taken into
account here in a simple manner, that is, N 2 is assumed
to be positive and spatially constant, equal to 0.25Ω2 ,
in the considered stably stratified local model along the
z-axis (Lesur & Ogilvie 2010). We define a stratification
length Λ ≡ g/N 2 , where g is the vertical component of
the gravity. This length cancels out from the main equa-
tions if we normalize the density logarithm by Λθ → θ,
which will be used henceforth.
Equations (1)-(5) have a stationary equilibrium so-
lution – an azimuthal flow along the y-direction with
linear shear of velocity in the the radial x-direction,
U0 = −qΩxey , with the total pressure P0 , density
ρ0 and threaded by a vertical uniform magnetic field,
B0 = B0z ez , B0z = const > 0. Consider perturbations
of the velocity, total pressure and magnetic field about
the equilibrium, u = U − U0 , p = P − P0 , b = B − B0 .
Substituting them into Equations (1)-(5) and rearrang-
ing the nonlinear terms with the help of divergence-free
conditions (4) and (5), we arrive to the system (A1)-(A9)
governing the dynamics of perturbations with arbitrary
 5

amplitude that is given in Appendix. These equations (Horton et al. 2010; Mamatsashvili et al. 2014, 2016;
form the basis for our numerical simulations to get a Gogichaishvili et al. 2017), we analyze in Fourier (k-
complete data set of the perturbation evolution in the ) space in order to understand the interplay of linear
turbulent state. and nonlinear processes underlying the turbulence dy-
We normalize the variables by taking Ω−1 as the unit namics. Obtaining the simulation data is just the first,
of time, the disk thickness H as the unit of length, ΩH preparatory stage of the main part of our study that
√
as the unit of velocity, ΩH 4πρ0 as the unit of magnetic focuses on the dynamics in Fourier space. This sec-
field and ρ0 Ω2 H 2 as the unit of pressure and energy. Vis- ond stage involves derivation of evolution equations for
cosity, thermal diffusivity and resistivity are measured, physical quantities (for the amplitudes of velocity and
respectively, by Reynolds number, Re, Péclet number, magnetic field) in Fourier space and subsequent analysis
Pe, and magnetic Reynolds number, Rm, defined as of the right hand side terms of these spectral dynamical
ΩH 2 ΩH 2 ΩH 2 equations.
Re = , Pe = , Rm = , We start the analysis of the spectral dynamics by de-
ν χ η
composing the perturbations f ≡ (u, p, θ, b) into spatial
which are taken to be equal Re = Pe = Rm = 3000
Fourier harmonics/modes
(i.e., the magnetic Prandtl number Pm = Rm/Re = 1).
The strength of the background uniform vertical mag-
Z
f (r, t) = f¯(k, t) exp (ik · r) d3 k (6)
netic field is measured by the parameter β = 2Ω2 H 2 /vA 2
,
3 1/2
which we fix to β = 10 , where vA = B0z /(4πρ0 ) is
the corresponding Alfvén speed. In the incompressible where f¯ ≡ (ū, p̄, θ̄, b̄) denotes the corresponding Fourier
case, this parameter is a proxy of the usual plasma β pa- transforms. Substituting decomposition (6) into per-
rameter (Longaretti & Lesur 2010), since sound speed in turbation Equations (A1)-(A9), we obtain the evolution
thin disks is cs ∼ ΩH. p In this non-dimensional units, equations for the spectral velocity (Equations B25-B27),
the mean field is B0z = 2/β = 0.0447. logarithmic density (entropy, Equation B13) and mag-
We carry out numerical simulations of the main Equa- netic field (Equations B14-B16). The derivation of these
tions (A1)-(A9) using the publicly available pseudo- spectral equations is a technical task and presented in
spectral code SNOOPY (Lesur & Longaretti 2007) with Appendix. Below we give the final set of the equa-
implemented standard shearing-box boundary condi- tions for the amplitudes of these quantities. Multiplying
tions: shearing-periodic in x and periodic in y and Equations (B25)-(B27), respectively, by ū∗x , ū∗y , ū∗z , and
z (Hawley et al. 1995). Our numerical box has sizes adding up with their complex conjugates, we obtain
(Lx , Ly , Lz ) = (4, 4, 1) and resolutions (Nx , Ny , Nz ) =
∂ |ūx |2 ∂ |ūx |2
(256, 256, 128), respectively, in the x, y, z-directions. = −qky +Hx +Ix(uθ) +Ix(ub) +Dx(u) +Nx(u) ,
This chosen aspect ratio of the box is most preferable, ∂t 2 ∂kx 2
(7)
as it is itself isotropic in the (x, y)-plane and avoids “nu-
merical deformation” of the generic anisotropic dynam- ∂ |ūy |2 ∂ |ūy |2
ics of MRI-turbulence. (Such a numerical deformation = −qky +Hy +Iy(uθ) +Iy(ub) +Dy(u) +Ny(u) ,
∂t 2 ∂kx 2
at different aspect ratios, Ly /Lx , is analyzed in de- (8)
tailed in Gogichaishvili et al. (2017) for Keplerian disk
flows with an azimuthal field). The same aspect ratio ∂ |ūz |2 ∂ |ūz |2
= −qky +Hz +Iz(uθ) +Iz(ub) +Dz(u) +Nz(u) ,
was adopted also in other related studies (Bodo et al. ∂t 2 ∂kx 2
2008; Longaretti & Lesur 2010; Lesur & Longaretti (9)
2011; Bai & Stone 2014; Meheut et al. 2015) to di- where the terms of linear origin are
minish the recurrent bursts of the channel modes in
k2
 
kx ky
MRI-turbulence with vertical field, however, as we Hx = 1 − x2 (ūx ū∗y + ū∗x ūy ) + 2(1 − q) 2 |ūx |2 ,
k k
show below, they still play an important role in the (10)
turbulence dynamics even in this extended box. The
initial conditions consist of small amplitude random
" #
1 ky2 kx ky
noise perturbations of velocity on top of the Keple- Hy = q − 2 − 2(q − 1) 2 (ūx ū∗y +ū∗x ūy )−2 2 |ūy |2
2 k k
rian shear flow. A subsequent evolution is followed up
(11)
to tf = 630 (about 100 orbits). The data accumu-
lated from the simulations, in fact, represents complete ky kz kx kz
information about the MRI-turbulence in the consid- Hz = (1 − q) (ūx ū∗z + ū∗x ūz ) − 2 (ūy ū∗z + ū∗y ūz ),
k2 k
ered flow system, which, following our previous works (12)
6

  where M is the Maxwell stress spectrum multiplied by

(uθ) 2 ki kz θ̄ū∗i + θ̄∗ ūi q,
Ii =N − δiz , (13) q
k2 2 M = − (b̄x b̄∗y + b̄∗x b̄y ), (24)
2
(ub) i
Ii = kz B0z (ū∗i b̄i − ūi b̄∗i ), (14) (bu) (ub) i
2 Ii = −Ii = kz B0z (ūi b̄∗i − ū∗i b̄i ) (25)
2
k2(u)
Di |ūi |2 ,
=− (15) (b)k2
Re Di |b̄i |2
=− (26)
Rm
and the modified nonlinear transfer functions for the
and the modified nonlinear terms for the quadratic forms
quadratic forms of the velocity components are
of the magnetic field components are
(u) 1
Ni = (ūi Q∗i + ū∗i Qi ). (16) i ∗
2 Nx(b) = b̄ [ky F̄z − kz F̄y ] + c.c., (27)
2 x
Here the index i = x, y, z henceforth, δiz is the Kro- i
Ny(b) = b̄∗y [kz F̄x − kx F̄z ] + c.c., (28)
necker delta and Qi (given by Equation B28) describes 2
the nonlinear transfers via triad interactions for the i
Nz(b) = b̄∗z [kx F̄y − ky F̄x ] + c.c., (29)
spectral velocities ūi in Equations (B25)-(B27). It 2
is readily shown that the sum of Hi is equal to the
where F̄x , F̄y , F̄z (given by Equations B21-B23) are the
Reynolds stress spectrum multiplied by the shear pa-
fourier transforms of the respective components of the
rameter q, H = Hx + Hy + Hz = q(ūx ū∗y + ū∗x ūy )/2.
perturbed electromotive force and describe nonlinear
Next, multiplying Equation (B13) by θ̄∗ and adding
transfers for the spectral magnetic field components in
up with its complex conjugate, we get
Equations (B14)-(B16).
∂ |θ̄|2 ∂ |θ̄|2 Our analysis is based on Equations (7) -(9), (17) and
= −qky + I (θu) + D(θ) + N (θ) , (17) (21) -(23), which describe processes of linear (Hi (k, t),
∂t 2 ∂kx 2 (uθ) (ub) (bu)
Ii (k, t), I (θu) (k, t), Ii (k, t), Ii (k, t), M(k, t))
where the terms of linear origin are (u) (b)
and nonlinear ( Ni (k, t), N (θ) (k, t), Ni (k, t)) ori-
(u) (b)
1 gin. Di (k, t), D(θ) (k, t), Di (k, t) describe, respec-
I (θu) = (ūz θ̄∗ + ū∗z θ̄), (18) tively, the effects of viscous, thermal and resistive dis-
2
sipation as a function of wavenumber and are neg-
k2 2 ative definite. These basic dynamical equations in
D(θ) = −
|θ̄| (19) Fourier space and, hence the above terms, formally
Pe
coincide with the corresponding ones in the case of
and the modified nonlinear transfer function for the
a net azimuthal magnetic field (see Equations A17-
quadratic form of the entropy is
A37 of Gogichaishvili et al. 2017, and the descrip-
i ∗ tion of each term in those spectral equations). The
N (θ) = θ̄ (kx Nx(θ) + ky Ny(θ) + kz Nz(θ) ) + c.c., (20) only difference is in the expressions for the kinetic-
2
(ub) (bu)
magnetic cross terms Ii (k, t) = −Ii (k, t), how-
(θ)
where Ni (given by Equation B20) describes the non- ever, this difference drastically changes the dynami-
linear transfers for the spectral entropy θ̄ in Equation cal processes. This cross terms describe, respectively,
(B13). the influence of the i-component of the magnetic field
Finally, multiplying Equations (B14)-(B16), respec- (b̄i ) on the same component of the velocity (ūi ) and
tively, by b̄∗x , b̄∗y , b̄∗z , and adding up with their complex vice versa for each mode. These terms are of linear
conjugates, we obtain origin and in the present case with vertical magnetic
(ub)
field, Ii (k, t) ∝ kz B0z , while in the case with az-
∂ |b̄x |2 ∂ |b̄x |2 (ub)
= −qky + Ix(bu) + Dx(b) + Nx(b) (21) imuthal magnetic field, Ii (k, t) ∝ ky B0y . Conse-
∂t 2 ∂kx 2
quently, in the presence of vertical field, this mutual
influence is nonzero for axisymmetric (with ky = 0)
∂ |b̄y |2 ∂ |b̄y |2 modes, which are not sheared by the flow, leading to
= −qky +M+Iy(bu) +Dy(b) +Ny(b) (22)
∂t 2 ∂kx 2 their exponential growth, i.e., to the classical MRI. It
is important to note that, in addition to this expo-
∂ |b̄z |2 ∂ |b̄z |2
= −qky + Iz(bu) + Dz(b) + Nz(b) , (23) nential growth of axisymmetric modes, there exists a
∂t 2 ∂kx 2
 7

linear mechanism of transient, or nonmodal growth of

X (u) 2k 2 X (b)
D(u) = Di =− EM , N (b) = Ni .
non-axisymmetric (with ky 6= 0) modes in the given i
Rm i
magnetized shear flow (Johnson 2007; Pessah & Chan
2012; Salhi et al. 2012; Mamatsashvili et al. 2013;
Squire & Bhattacharjee 2014; Zhuravlev & Razdoburdin
2014). Assessment of relative roles of the exponential
growth of axisymmetric modes due to MRI and the tran-
sient growth of non-axisymmetric modes is a key ques-
tion of our study. We note in advance that despite the
transient nature, the linear growth of non-axisymmetric
modes is not of secondary importance in the mainte-
nance of the turbulence and in the forming its spectral
characteristics. In the case of azimuthal magnetic field,
the course of events is somewhat different: the cross
terms for axisymmetric modes vanish that leads to the
disappearance of the classical MRI and the only source Figure 1. Optimal nonmodal growth factor, G, in (kx , ky )-
of energy for the turbulence maintenance remains linear plane at td = 1.33 and different kz = 1, 2, 3 calculated with
transient growth of non-axisymmetric modes . the linearized version of main Equations (A1)-(A9).
Although our analysis is based on the above dynami-
cal equations for quadratic forms of physical quantities The equation of the thermal energy of modes, Eth =
in Fourier space, to get general feeling, as in Simon et al. N 2 |θ|2 /2, is straightforward to derive by multiplying
(2009); Lesur & Longaretti (2011); Gogichaishvili et al. Equation (17) just by N 2 , but we do not give it here,
(2017), we also present equation for the spectral ki- as the thermal processes have a minor influence on the
netic energy of modes/harmonics, EK = (|ūx |2 + |ūy |2 + strength and the dynamics of the turbulence (see below).
|ūz |2 )/2: One can easily write also the equation for the total
∂EK ∂EK spectral energy of modes, E = EK + Eth + EM ,
= −qky +H+I (uθ) +I (ub) +D(u) +N (u) , (30)
∂t ∂kx
which is obtained by summing of Equations (7)-(9). ∂E ∂E
Right hand side terms of the above equation is the sum = −qky + H + M + D(u) + N 2 D(th)
∂t ∂kx
of the related terms:
+ D(b) + N (u) + N 2 N (θ) + N (b) . (32)
X
H= Hi = q(ūx ū∗y + ū∗x ūy )/2,
i

(uθ) (ub)
X X
I (uθ) = Ii , I (ub) = Ii ,
i i
X (u) 2k 2 X (u)
D(u) = Di =− EK , N (u) = Ni . This work is funded in part by the US Department
i
Re i of Energy under grant DE-FG02-04ER54742, the Space
Similarly, we can get equation for the spectral magnetic and Geophysics Laboratory at the University of Texas
energy of modes, EM = (|b̄x |2 + |b̄y |2 + |b̄z |2 )/2, by sum- at Austin and by the Rustaveli National Science Foun-
ming Equations (21)-(23), dation grant ?? (Georgia). G.M. is supported by the
Georg Forster Research Fellowship from the Alexan-
∂EM ∂EM
= −qky + M + I (bu) + D(b) + N (b) , (31) der von Humboldt Foundation (Germany). The sim-
∂t ∂kx ulations were performed in Texas Advanced Comput-
where ing Center in Austin (TX) and on the high-performance
Linux cluster Hydra at the Helmholtz-Zentrum Dresden-
(bu)
X
M = −q(b̄x b̄∗y + b̄∗x b̄y )/2, I (bu) = Ii = −I (ub) , Rossendorf (Germany).
i
8

0.8 qh−bx by i qhux uy i

0.6
0.4
0.2
0
0 100 200 300 400 500 600
t
1.2
1 hEM i hEK i hEth i
0.8
0.6
0.4
0.2
0
0 100 200 300 400 500 600
t

1.2
1 hu2x i1/2 hu2y i1/2 hu2z i1/2
0.8
0.6
0.4
0.2
0
0 100 200 300 400 500 600
t
1.6
1.4 hb2x i1/2 hb2y i1/2 hb2z i1/2
1.2
1
0.8
0.6
0.4
0.2
0
0 100 200 300 400 500 600
t
Figure 2. Evolution of volume-averaged Reynolds and
Maxwell stresses (top row), kinetic, thermal and magnetic
energy densities (second row), rms of velocity (third row)
and magnetic field (bottom row) components. Turbulence
sets in after several orbits with recurrent bursts of these
quantities. Each burst starts with the amplification of uy ,
followed by amplifications of ux and uz . Dashed line in the
bottomppanel corresponds to the imposed mean vertical field
B0z = 2/β = 0.0447.

APPENDIX
A. PERTURBATION EQUATIONS IN PHYSICAL SPACE
Equations governing the evolution of the velocity, total pressure and magnetic field perturbations, u, p, b, about the
equilibrium Keplerian flow U0 = −qΩxey with a net vertical field B0 = B0z ez are obtained from the basic Equations
(1)-(5) and componentwise have the form:
 2     
Dux 1 ∂p B0z ∂bx ∂ bx 2 ∂ bx by ∂ bx bz
= 2Ωuy − + + − ux + − ux uy + − ux uz + ν∇2 ux , (A1)
Dt ρ0 ∂x 4πρ0 ∂z ∂x 4πρ0 ∂y 4πρ0 ∂z 4πρ0
!
b2y
   
Duy 1 ∂p B0z ∂by ∂ bx by ∂ 2 ∂ bz by
= (q −2)Ωux − + + − ux uy + − uy + − uz uy +ν∇2 uy (A2)
Dt ρ0 ∂y 4πρ0 ∂z ∂x 4πρ0 ∂y 4πρ0 ∂z 4πρ0

b2z
     
Duz 1 ∂p B0z ∂bz ∂ bx bz ∂ by bz ∂
=− − N 2θ + + − ux uz + − uy uz + − u2z + ν∇2 uz (A3)
Dt ρ0 ∂z 4πρ0 ∂z ∂x 4πρ0 ∂y 4πρ0 ∂z 4πρ0
 9
0.1
0.08 ÊK
0.06 ÊM
0.04 Êth
0.02
0
−10−9 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10
kz
0.08
0.06 Ĥ
0.04 M̂
0.02
0
−0.02
−10−9 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10
kz

Figure 3. Integrated in (kx , ky )-plane time-averaged ki-

netic, ÊK , magnetic, ÊM and thermal Êth energy spectra (up-
per panel) as well as Reynolds, Ĥ, and Maxwell, M̂, stresses
(lower panel) as a function of kz .

Figure 4. Logarithm of the spectral magnetic energy, log10 EM , in (kx , ky )-plane at kz = 0, 1, 2. The spectra is strongly
anisotropic due to the shear, having larger power on the kx /ky > 0 side at a given ky .

Dθ ∂ ∂ ∂
= uz − (ux θ) − (uy θ) − (uz θ) + χ∇2 θ (A4)
Dt ∂x ∂y ∂z

Dbx ∂ux ∂ ∂
= B0z + (ux by − uy bx ) − (uz bx − ux bz ) + η∇2 bx , (A5)
Dt ∂z ∂y ∂z

Dby ∂uy ∂ ∂
= −qΩbx + B0z − (ux by − uy bx ) + (uy bz − uz by ) + η∇2 by , (A6)
Dt ∂z ∂x ∂z

Dbz ∂uz ∂ ∂
= B0z + (uz bx − ux bz ) − (uy bz − uz by ) + η∇2 bz , (A7)
Dt ∂z ∂x ∂y

∂ux ∂uy ∂uz

+ + = 0, (A8)
∂x ∂y ∂z

∂bx ∂by ∂bz

+ + = 0, (A9)
∂x ∂y ∂z
where D/Dt = ∂/∂t − qΩx∂/∂y is the total derivative along the background flow.
10
3
2 kz=0
0.7
0.6
1 0.5
0 0.4
−1 0.3
−2 0.2
0.1
−3
3 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8
2
kz=1 0.7
0.6
1 0.5
k 0 0.4
y 0.3
−1
−2 0.2
0.1
−3
3 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8
2 kz=2
0.7
0.6
1 0.5
0 0.4
−1 0.3
−2 0.2
0.1
−3
−8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8
kx

Figure 5. Distribution of the active modes in k-space at

kz = 0, 1, 2. The color dots represent the modes whose mag-
netic energy, EM , grows more than 50% of the maximum
spectral magnetic energy, EM,max , and the colors indicate
the fraction of time each mode contains this higher energy
during the quasi-steady state until the end of the simulation.
The basic mode with kx = ky = 0, kz = 1 (bigger red dot)
corresponds to the maximum of the spectral energy most of
the time.

B. DERIVATION OF SPECTRAL EQUATIONS

Here we derive evolution equations for velocity, entropy and magnetic field perturbations in Fourier space. Substi-
tuting decomposition (6) into Equations (A1)-(A9) and taking into account the normalization in the text, we arrive
at the following equations governing the dynamics of perturbation modes in Fourier space

k2
 
∂ ∂ (u) (u) (u)
+ qky ūx = 2ūy − ikx p̄ + ikz B0z b̄x − ūx + ikx Nxx + iky Nxy + ikz Nxz , (B10)
∂t ∂kx Re

k2
 
∂ ∂ (u) (u) (u)
+ qky ūy = (q − 2)ūx − iky p̄ + ikz B0z b̄y − ūy + ikx Nxy + iky Nyy + ikz Nyz , (B11)
∂t ∂kx Re

k2
 
∂ ∂
+ qky ūz = −ikz p̄ − N 2 θ̄ + ikz B0z b̄z − (u)
ūz + ikx Nxz (u)
+ iky Nyz (u)
+ ikz Nzz , (B12)
∂t ∂kx Re

k2
 
∂ ∂
+ qky θ̄ = ūz − θ̄ + ikx Nx(θ) + iky Ny(θ) + ikz Nz(θ) , (B13)
∂t ∂kx Pe

k2
 
∂ ∂
+ qky b̄x = ikz B0z ūx − b̄x + iky F̄z − ikz F̄y , (B14)
∂t ∂kx Rm

k2
 
∂ ∂
+ qky b̄y = −q b̄x + ikz B0z ūy − b̄y + ikz F̄x − ikx F̄z (B15)
∂t ∂kx Rm

k2
 
∂ ∂
+ qky b̄z = ikz B0z ūz − b̄z + ikx F̄y − iky F̄x (B16)
∂t ∂kx Rm

kx ūx + ky ūy + kz ūz = 0, (B17)

kx b̄x + ky b̄y + kz b̄z = 0, (B18)

 11

Figure 6. Spectrum of the radial field, |b̄x |, and the maps of the corresponding linear and nonlinear terms governing its
dynamics (Equation ??), in (kx , ky )-plane at kz = 0(lef t), 1(middle), 2(right). The spectrum as well as the action of these
terms are highly anisotropic, (i.e., depend on the wavevector azimuthal angle) due to the shear. These terms are appreciable
and primarily operate in the vital area |kx | . 12, |ky | . 3. The red and yellow (blue and dark blue) regions in each panel
correspond to wavenumbers where respective dynamical terms are positive (negative) and hence act as a source/production
(sink) for |b̄x |2 . In light green regions, outside the vital area, these terms are small, although, as we checked, preserve the same
(b)
anisotropic shape. In particular, the nonlinear transfer term Nx transversely redistributes |b̄x |2 from the blue and dark blue
(b) (b)
regions, where Nx < 0, to the red and yellow regions, where Nx > 0. These regions exhibit considerable variations with the
azimuthal angle of the wavevector and also depend on kz . The basic mode kb = (0, 0, 1) corresponds to the maximum of |b̄x |;
(bu) (b)
it is supplied by positive Ix (k0 ) > 0 due to MRI and drained by nonlinear transfers Nx < 0.

p
where k 2 = kx2 + ky2 + kz2 and B0z = 2/β is the normalized background vertical field. These spectral equations contain
(u) (θ)
the linear as well as the nonlinear (Nij (k, t), Ni (k, t), F̄i (k, t), i, j = x, y, z) terms that are the Fourier transforms of
the corresponding linear and nonlinear terms in the original Equations (A1)-(A9). The latter are given by convolutions
Z
(u)
Nij (k, t) = d3 k′ b̄i (k′ , t)b̄j (k − k′ , t) − ūi (k′ , t)ūj (k − k′ , t) ,
 
(B19)

Z
(θ)
Ni (k, t) = − d3 k′ ūi (k′ , t)θ̄(k − k′ , t) (B20)
12

Figure 7. Same as in Figure 6, but for the azimuthal field b̄y with the corresponding dynamical terms from Equation (??).
The dynamics of this component is primarily determined by the combined action of the drift, the Maxwell stress M, which is
(b)
positive, ensuring energy injection due to MRI, and the nonlinear term Ny , which is negative and acts as a sink in the vital
area. These two processes are most intensive for the basic mode k0 = (0, 0, 1), which corresponds to the maximum of |b̄y |. The
(bu)
linear exchange term Iy is negligible compared with the above terms and is not shown here.

where i, j = x, y, z and F̄x , F̄y , F̄z are the fourier transforms of the respective components of the perturbed electromotive
force F = u × b,
Z
F̄x (k, t) = d3 k′ ūy (k′ , t)b̄z (k − k′ , t) − ūz (k′ , t)b̄y (k − k′ , t)
 
(B21)

Z
d3 k′ ūz (k′ , t)b̄x (k − k′ , t) − ūx (k′ , t)b̄z (k − k′ , t)
 
F̄y (k, t) = (B22)

Z
d3 k′ ūx (k′ , t)b̄y (k − k′ , t) − ūy (k′ , t)b̄x (k − k′ , t)
 
F̄z (k, t) = (B23)

and describe the contribution from nonlinearity to the magnetic field perturbations. In the case of classical forced
MHD turbulence without background shear flow, these nonlinear transfer terms in k-space were also derived in Verma
 13

Figure 8. Same as in Figure 6, but for b̄z with the corresponding dynamical terms from Equation (??). The transverse
(b)
character of the nonlinear redistribution, Nz , is also evident. |b̄z | is small in comparison with |b̄x | and |b̄y |.

(2004). From Equations (B10)-(B12) and the divergence-free conditions (B17) and (B18) we can eliminate pressure
ky kx kz X ki kj (u)
p̄ = 2i(1 − q) 2 ūx − 2i 2 ūy + iN 2 2 θ̄ + N (B24)
k k k k 2 ij
(i,j)=(x,y,z)

Substituting it back into Equations (B10)-(B12) we get

k2 k2
   
∂ ∂ kx ky kx kz
+ qky ūx = 2 1 − x2 ūy + 2(1 − q) 2 ūx + N 2 2 θ̄ + ikz B0z b̄x − ūx + Qx , (B25)
∂t ∂kx k k k Re
" #
ky2 k2
 
∂ ∂ kx ky ky kz
+ qky ūy = q − 2 − 2(q − 1) 2 ūx − 2 2 ūy + N 2 2 θ̄ + ikz B0z b̄y − ūy + Qy , (B26)
∂t ∂kx k k k Re
kz2 k2
   
∂ ∂ ky kz kx kz 2
+ qky ūz = 2(1 − q) 2 ūx − 2 2 ūy − N 1 − 2 θ̄ + ikz B0z b̄z − ūz + Qz , (B27)
∂t ∂kx k k k Re
where
X (u)
X km kn
(u)
Qi = i kj Nij − iki 2
Nmn , i, j, m, n = x, y, z. (B28)
j m,n
k
14

Figure 9. Spectra of |ūx | and the maps of the corresponding linear and nonlinear terms governing its dynamics (Equation ??)
in (kx , ky )-plane at kz = 0(lef t), 1(middle), 2(right). The dynamics of this velocity component is primarily determined by Hx ,
(u)
which acts as source/production and Nx , which acts as a sink (although with some production at kz = 0); the linear exchange
(uθ) (ub)
terms, Ix and Ix , are negligible compared with the above terms and are not shown here. Again, these two processes are
most intensive for the basic mode.

REFERENCES
Alexakis, A., Mininni, P. D., & Pouquet, A. 2007, New J.
Phys., 9, 298
Bai, X.-N., & Stone, J. M. 2014, ApJ, 796, 31
Balbus, S. A., & Hawley, J. F. 1991, ApJ, 376, 214
—. 1998, Rev. Mod. Phys., 70, 1
Bodo, G., Mignone, A., Cattaneo, F., Rossi, P., & Ferrari,
A. 2008, A&A, 487, 1
Brandenburg, A., Nordlund, A., Stein, R. F., & Torkelsson,
U. 1995, ApJ, 446, 741
Chandrasekhar, S. 1960, Proceedings of the National
Academy of Science, 46, 253
Davis, S. W., Stone, J. M., & Pessah, M. E. 2010, ApJ, 713,
52
Flock, M., Henning, T., & Klahr, H. 2012, ApJ, 761, 95
Fromang, S., & Nelson, R. P. 2006, A&A, 457, 343
Fromang, S., & Papaloizou, J. 2007, A&A, 476, 1113
Gogichaishvili, D., Mamatsashvili, G., Horton, W.,
Chagelishvili, G., & Bodo, G. 2017, ApJ, 845, 70
Goodman, J., & Xu, G. 1994, ApJ, 432, 213
Guan, X., & Gammie, C. F. 2011, ApJ, 728, 130
Guan, X., Gammie, C. F., Simon, J. B., & Johnson, B. M.
2009, ApJ, 694, 1010
 15

Figure 10. Same as in Figure 9, but for ūy with the corresponding dynamical terms from Equation (??). The influence of the
(uθ)
thermal process, Iy , is negligible and not shown here. The spectrum of |ūy | reaches a maximum at kx = ±1, ky = kz = 0,
(u)
which corresponds to the zonal flow in physical space, which as seen here is driven by Ny , rather than at the wavenumber of
the kx = ky = 0, kz = 1 of the basic mode. The nonlinear term is now the source for ūy (k0 ) of the basic mode, which is then
(ub)
amplified by positive Iy , and drained by negative Hy
.
Hawley, J., Gammie, C., & Balbus, S. 1995, ApJ, 440, 742 Johnson, B. M. 2007, ApJ, 660, 1375
Hawley, J. F., & Balbus, S. A. 1992, ApJ, 400, 595
Latter, H. N., Fromang, S., & Faure, J. 2015, MNRAS, 453,
Horton, W., Kim, J.-H., Chagelishvili, G. D., Bowman,
J. C., & Lominadze, J. G. 2010, PhRvE, 81, 066304 3257
16

Figure 11. Same as in Figure 9, but for ūz with the corresponding dynamical terms from Equation (??). The influence of the
(ub) (u)
linear magnetic exchange term Iz is negligible and is not shown here. The nonlinear term Nz transfers |ūz |2 towards small
kz (as it also seen in the corresponding panel of Figure ??).

Latter, H. N., Fromang, S., & Gressel, O. 2010, MNRAS,
406, 848
Latter, H. N., Lesaffre, P., & Balbus, S. A. 2009, MNRAS,
394, 715
Lesur, G., & Longaretti, P.-Y. 2007, MNRAS, 378, 1471
—. 2011, A&A, 528, A17
Lesur, G., & Ogilvie, G. I. 2010, MNRAS, 404, L64
Longaretti, P.-Y., & Lesur, G. 2010, A&A, 516, A51
Lynden-Bell, D., & Pringle, J. E. 1974, MNRAS, 168, 603
Mamatsashvili, G., Khujadze, G., Chagelishvili, G., et al.
2016, PhRvE, 94, 023111
Mamatsashvili, G. R., Chagelishvili, G. D., Bodo, G., &
Rossi, P. 2013, MNRAS, 435, 2552

Mamatsashvili, G. R., Gogichaishvili, D. Z., Chagelishvili,
G. D., & Horton, W. 2014, PhRvE, 89, 043101
Meheut, H., Fromang, S., Lesur, G., Joos, M., &
Longaretti, P.-Y. 2015, A&A, 579, A117
Mohandas, G., & Pessah, M. E. 2017, ApJ, 838, 48
Murphy, G. C., & Pessah, M. E. 2015, ApJ, 802, 139
Nauman, F., & Blackman, E. G. 2014, MNRAS, 441, 1855
Pessah, M. E. 2010, ApJ, 716, 1012
Pessah, M. E., & Chan, C.-k. 2008, ApJ, 684, 498
Pessah, M. E., & Chan, C.-K. 2012, ApJ, 751, 48
Pessah, M. E., Chan, C.-K., & Psaltis, D. 2006, MNRAS,
372, 183
Pessah, M. E., & Goodman, J. 2009, ApJL, 698, L72
Salhi, A., Lehner, T., Godeferd, F., & Cambon, C. 2012,
PhRvE, 85, 026301
17
Sano, T., & Inutsuka, S.-i. 2001, ApJL, 561, L179
Shakura, N., & Postnov, K. 2015, MNRAS, 448, 3697
Shakura, N. I., & Sunyaev, R. A. 1973, A&A, 24, 337
Shtemler, Y. M., Mond, M., & Liverts, E. 2011, MNRAS,
413, 2957
Simon, J. B., & Hawley, J. F. 2009, ApJ, 707, 833
Simon, J. B., Hawley, J. F., & Beckwith, K. 2009, ApJ,
690, 974
Squire, J., & Bhattacharjee, A. 2014, ApJ, 797, 67
Velikhov, E. 1959, Zh. Eksp. Teor. Fiz., 36, 1398
Verma, M. K. 2004, Phys. Rep., 401, 229
Walker, J., Lesur, G., & Boldyrev, S. 2016, MNRAS, 457,
L39
Wardle, M. 1999, MNRAS, 307, 849
Zhuravlev, V. V., & Razdoburdin, D. N. 2014, MNRAS,
442, 870