betaFeedback

WorksPhysics and Astronomy

Steady states of the Parker instability

Abstract

We study the linear properties, nonlinear saturation and a steady, strongly nonlinear state of the Parker instability in galaxies. We consider magnetic buoyancy and its consequences with and without cosmic rays. Cosmic rays are described using the fluid approximation with anisotropic, non-Fickian diffusion. To avoid unphysical constraints on the instability (such as boundary conditions often used to specify an unstable background state), nonideal MHD equations are solved for deviations from a background state representing an unstable magnetohydrostatic equilibrium. We consider isothermal gas and neglect rotation. The linear evolution of the instability is in broad agreement with earlier analytical and numerical models; but we show that most of the simplifying assumptions of the earlier work do not hold, such that they provide only a qualitative rather than quantitative picture. In its nonlinear stage the instability has significantly altered the background state from its initial state. Vertical distributions of both magnetic field and cosmic rays are much wider, the gas layer is thinner, and the energy densities of both magnetic field and cosmic rays are much reduced. The spatial structure of the nonlinear state differs from that of any linear modes. A transient gas outflow is driven by the weakly nonlinear instability as it approaches saturation.

Introduction

The magnetic buoyancy (or magnetic Rayleigh–Taylor) instability (Newcomb 1961), modified and enhanced by cosmic rays, is known as the Parker instability (Parker 1958, 1966, 1979). The horizontal magnetic field in a gas layer confined by gravity can be unstable with respect to undular modes which grow exponentially on a timescale comparable to the sound or Alfvén crossing time over the gas density scale height. For the observed scale height 0.5 kpc for the warm interstellar gas in the Solar neighbourhood, with the sound and Alfvén speeds both at about 10 km s−110\ \mathrm{km}\,\mathrm{s}^{-1}, the time scale is of order 5×107 yr5 \times10^{7}\ \mathrm{yr}. It is much shorter than the galactic life-time, and several effects have been explored that can make the galactic discs less unstable including the roles of cosmic ray diffusion (Kuznetsov & Ptuskin 1983; Kuznetsov 1987; Heintz & Zweibel 2018) and rotation (Zweibel & Kulsrud 1975; Foglizzo & Tagger 1994, 1995). We note in this connection that widely used heuristic models of the stratified interstellar medium, informed by observations of the interstellar medium, are likely to be unstable (e.g., Lachièze-Rey et al. 1980). The Parker instability can hardly be completely suppressed in spiral galaxies, so it is important to explore its nonlinear, steady states in order to understand why the gas distributions observed in spiral galaxies are not destroyed by the instability.

Aside from its effect on the vertical distributions of the interstellar gas, magnetic field and cosmic rays, the Parker instability plays a significant role in the evolution of galaxies. It contributes to driving galactic outflows (winds and fountains) and thereby to what is known as the ‘star formation feedback’ on the evolving galactic disc.

The linear stage of the instability has been thoroughly studied and the dispersion relation has been obtained for a wide range of physical models and parameter regimes (e.g., Giz & Shu 1993; Foglizzo & Tagger 1994, 1995; Kim et al. 1997; Shukurov & Subramanian 2021, and references therein).

However, the nonlinear state of the Parker instability is much less understood, in particular because it can only be studied numerically (Kim et al. 2001). Two-dimensional (2D) simulations of the magnetic buoyancy instability of Matsumoto et al. (1990) and Horiuchi et al. (1988) show how the instability saturate differently across different parameter ranges. We compare our three-dimensional (3D) results with these studies, and find that the solutions may be rather different from any 2D linear modes and their superpositions.

Matsumoto et al. (1990) find two types of nonlinear 2D solutions; oscillatory for the azimuthal (along the unstable magnetic field) wave number kyk_y exceeding some critical value, or shock-wave dominated for kyk_y less than critical. When the initial magnetic pressure is much smaller than the thermal pressure (Pm/Pth<0.3P_{\rm m}/P_{\rm th} < 0.3), the nonlinear oscillations couple to form long waves which eventually decay due to shocks. 3D simulations described here do not reproduce such oscillations or shocks at scales comparable to the instability scales, and the proportions of magnetic and cosmic ray energy density and pressure decrease over time near the midplane. The final nonlinear stages of our models show the magnetic field loops to have less ordered, more small-scale structure in the xyxy-plane than these 2D simulations. Hanasz & Lesch (2000) and Hanasz et al. (2002) study the nonlinear Parker instability with rotation in three dimensions, focusing on the evolution of the magnetic field structure and the mean-field dynamo driven by the instability, adopting a very weak initial magnetic field. Hanasz et al. (2002) observe a decrease in the magnetic field strength at later times (following its growth due to the dynamo action) with an increase in its scale height. Machida et al. (2013) study the 3D nonlinear magnetic buoyancy instability (in the absence of cosmic rays) in toroidal geometry. These authors also observe a gradual reduction in the magnetic field strength accompanied by an increase in its scale, possibly slowed down by the dynamo action. Both Hanasz et al. (2002) and Machida et al. (2013) report the development of vertical current filaments in the late stages of the instability, which could be due to a strongly nonlinear magnetic dissipation in the former case and the ideal magnetohydrodynamic approximation in the latter case. We do not observe any such features.

We explore the instability systematically, starting with the magnetic buoyancy instability and then, to clarify their role, adding cosmic rays in the fluid approximation (based on the advection–diffusion equation). Using our linear stage results where the perturbations remain weak, we test various approximations used in the linear stability analyses (which rely on widely diverse assumptions and simplifications). In order to explore the nonlinear states of the instability, we consider imposed and fixed background gas, magnetic field and cosmic ray distributions and solve fully nonlinear equations for the deviations from that state.

The paper is structured as follows. In § 2, we set out the basic equations (§ 2.1) and specify the physical and numerical parameters used in the simulations. In § 2.2, we provide a brief description of the gravitational field models used, and in § 2.3, the boundary conditions are discussed. The results are presented in § 3.1 and § 3.2 for the linear and nonlinear stages of the instability, respectively (§ 3.1 also contains a discussion of the assumptions on which analytical studies rely). § 4 contains a discussion of the implications of our results with emphasis on the nonlinear, statistically steady state, including the relative distributions of the gas, magnetic fields and cosmic rays (§ 4.1), systematic vertical flows (§ 4.2 and the force balance (§ 4.3). § 5 summarises our conclusions. Appendix A justifies our approach to the implementation of the unstable background state, while the parameter space is explored in § 4.4 where we discuss the influence on the results of the scale heights of the gas and non-thermal pressure components in the background state, of the gravity profile and the role of the gas viscosity and magnetic diffusivity.

Experimental design

As the initial state, we consider a plane-parallel magnetohydrostatic equilibrium in the galactic gravitational field, i.e., a stratified layer of thermal gas, horizontal magnetic field and cosmic rays. The instability is caused by the magnetic buoyancy, which depends on the vertical gradient of the magnetic field strength (Hughes & Cattaneo (1987)), and so leads to the removal of the initial magnetic field from the layer’s midplane, tending to reduce its gradient. As a result, the initial magnetic field is rapidly lost from the system. In galaxies and accretion discs, the large-scale magnetic field is continuously replenished by the dynamo action near the disc’s midplane (Shukurov & Subramanian (2021)). However, most simulations of the Parker instability do not include the dynamo action as the source of the unstable magnetic field. Therefore, an equilibrium state introduced as an initial condition is rapidly destroyed by the instability in such simulations. This has prevented Rodrigues et al. [34] from analysing the steady, strongly nonlinear state of the system. Alternatively, boundary conditions can be used to impose a steady background state. However, this would constrain unphysically the evolution of the system as the fixed boundary conditions would require that the deviations from the background state vanish at the boundaries.

Therefore, our approach is to derive and solve (fully nonlinear) equations for deviations from the background state. In fact, this is the standard approach to explore the linear Parker (or any other) instability analytically, but we extend it to capture a fully nonlinear evolution of the perturbations when their magnitude is no longer small. The boundary conditions for the deviations are not restrictive (we use periodic boundary conditions in the horizontal planes), so that the perturbations can evolve freely. In the nonlinear state of the instability, the magnitude of the deviations from the background state is comparable to that of the background state, altering it fundamentally; so it is important to make the model fully flexible to allow for the possibility of such a strong modification.

In this paper, we consider an isothermal gas and neglect rotation to establish a reference model to allow us to identify the effects of radiative cooling, rotation, dynamo action and supernova activity with the associated multi-phase gas structure, which will be discussed elsewhere.

Basic equations

We solve numerically the non-ideal MHD equations for the gas density ρ\rho, its velocity U\boldsymbol{U}, total pressure PP (which includes the thermal, magnetic and cosmic-ray contributions), magnetic field B=∇×A\boldsymbol{B} = \nabla\times\boldsymbol{A}, its vector potential A\boldsymbol{A} (with the advective gauge Φ=η∇⋅A\Phi= \eta\nabla\cdot\boldsymbol{A}) and the energy density of cosmic rays ϵcr\epsilon_{\rm cr}:

DρDt=−ρ∇⋅U,(1)\frac{D\rho}{Dt} = -\rho\nabla\cdot\boldsymbol{U}, \tag*{(1)}
DUDt=−∇Pρ+g+(∇×B)×B4πρ+∇⋅τρ,(2)\frac{D\boldsymbol{U}}{Dt} = -\frac{\nabla P}{\rho} + \boldsymbol{g} + \frac{(\nabla\times\boldsymbol{B}) \times\boldsymbol{B}}{4\pi\rho} + \frac{\nabla\cdot\boldsymbol{\tau}}{\rho}, \tag*{(2)}
∂A∂t=U×(∇×A)−η∇×∇×A,(3)\frac{\partial\boldsymbol{A}}{\partial t} = \boldsymbol{U} \times(\nabla\times\boldsymbol{A}) - \eta\nabla\times\nabla\times\boldsymbol{A}, \tag*{(3)}
∂ϵcr∂t=−∇⋅(ϵcrU)−Pcr∇⋅U−∇⋅F,(4)\frac{\partial\epsilon_{\rm cr}}{\partial t} = -\nabla\cdot(\epsilon_{\rm cr}\boldsymbol{U}) - P_{\rm cr}\nabla\cdot\boldsymbol{U} - \nabla\cdot\boldsymbol{F}, \tag*{(4)}

where D/Dt=∂/∂t+U⋅∇\mathrm{D}/\mathrm{D}t = \partial/\partial t + \mathbf{U} \cdot\nabla is the Lagrangian derivative, g\mathbf{g} is the gravitational acceleration, Pcr=(γcr−1)ϵcrP_{\rm cr} = (\gamma_{\rm cr} - 1)\epsilon_{\rm cr}, with γcr=4/3\gamma_{\rm cr} = 4/3 for the ultra-relativistic cosmic rays, and F\mathbf{F} is the diffusive flux of cosmic rays discussed below. The viscous term in equation (2) has the form ρ−1∇⋅τ=ν[∇2U+13∇(∇⋅U)+2W⋅∇ln⁡ρ]\rho^{-1}\nabla\cdot\boldsymbol{\tau} = \nu[\nabla^2\mathbf{U} + \frac{1}{3}\nabla(\nabla\cdot\mathbf{U}) + 2\mathbf{W}\cdot\nabla\ln\rho], where W\mathbf{W} is the traceless rate of strain tensor, and we use the cgs system with cc the speed of light. The advection–diffusion equation for cosmic rays (4) is derived by, e.g., Skilling (1975) and Drury & Völk (1981) and used in this form by, e.g., Gupta et al. (2021) and Rodrigues et al. [34].

Each variable is represented as the sum of a background equilibrium value, identified with the subscript zero, and a deviation from it, denoted with the prime or a lower-case symbol,

ρ=ρ0+ρ′,U=U0+u,P=P0+P′,B=B0+b,A=A0+a,ϵcr=ϵcr0+ϵcr′,F=F0+F′,U0=0.(5)\begin{aligned} \rho&= \rho_0 + \rho', & \mathbf{U} &= \mathbf{U}_0 + \mathbf{u}, & P &= P_0 + P', \\ \mathbf{B} &= \mathbf{B}_0 + \mathbf{b}, & \mathbf{A} &= \mathbf{A}_0 + \mathbf{a}, & \epsilon_{\rm cr} &= \epsilon_{\rm cr0} + \epsilon'_{\rm cr}, \\ \mathbf{F} &= \mathbf{F}_0 + \mathbf{F}', & \mathbf{U}_0 &= 0 . \tag*{(5)} \end{aligned}

The background state and how it is supported in a steady state is described in detail in § 2.2. It is convenient to include only the deviations in the gas and cosmic ray pressures into P′P', whereas the magnetic field deviations contribute to both the magnetic pressure and magnetic tension and thus are included as a part of the Lorentz force. However, the background pressure P0P_0 includes all three pressure components. We stress again that the deviations from the equilibrium state are not assumed to be weak: the nonlinear governing equation for them are obtained by subtracting those for the background state from equations (1)–(4) as discussed in Appendix A, where we also present equations governing the background state which ensure that it is static, ∂/∂t=0\partial/\partial t = 0. The background magnetic field can be thought of as being maintained by the dynamo action in the disc; in this case, the assumption of its time-invariance is justified if the dynamo time scale is much shorter than the growth time of the Parker instability. The development of the Parker instability when the dynamo time scale is comparable to or longer than the instability time scale will be discussed elsewhere.

In the absence of a background velocity U0\mathbf{U}_0, equations (1)–(3), written for the deviations from the background state, reduce to (Appendix A)

∂ρ′∂t=−u⋅∇ρ0−ρ0∇⋅u−u⋅∇ρ′−ρ′∇⋅u,(6)\frac{\partial\rho'}{\partial t} = -\mathbf{u}\cdot\nabla\rho_0 - \rho_0\nabla\cdot\mathbf{u} - \mathbf{u}\cdot\nabla\rho' - \rho'\nabla\cdot\mathbf{u}, \tag*{(6)}
ρDuDt=ρ′g−∇P′+(∇×B)×B−(∇×B0)×B04π+ρν[∇2u+13∇(∇⋅u)+2W⋅∇ln⁡ρ],(7)\rho\frac{\mathrm{D}\mathbf{u}}{\mathrm{D}t} = \rho'\mathbf{g} - \nabla P' + \frac{(\nabla\times\mathbf{B})\times\mathbf{B} - (\nabla\times\mathbf{B}_0)\times\mathbf{B}_0}{4\pi} + \rho\nu\left[\nabla^2\mathbf{u} + \frac{1}{3}\nabla(\nabla\cdot\mathbf{u}) + 2\mathbf{W}\cdot\nabla\ln\rho\right], \tag*{(7)}
∂a∂t=u×∇×A−η∇×∇×a,(8)\frac{\partial\mathbf{a}}{\partial t} = \mathbf{u}\times\nabla\times\mathbf{A} - \eta\nabla\times\nabla\times\mathbf{a}, \tag*{(8)}

where we recall that the pressure perturbation P′P' contains only the gas and cosmic ray pressure perturbations (when the latter is included); the magnetic pressure perturbation is accounted for within the Lorentz force.

Equation (4) leads to the following equation for the deviation of the cosmic ray energy density from the background state:

∂ϵcr′∂t=−∇⋅(ϵcru)−Pcr∇⋅u−∇⋅F′,(9)\frac{\partial\epsilon'_{\rm cr}}{\partial t} = -\nabla\cdot(\epsilon_{\rm cr}\mathbf{u}) - P_{\rm cr}\nabla\cdot\mathbf{u} - \nabla\cdot\mathbf{F}', \tag*{(9)}

where F′\mathbf{F}' is the evolving cosmic ray flux perturbation derived from the non-Fickian diffusion model for cosmic rays ([34] (Snodin et al. 2006)) (as see Appendix A for the derivation)

τ∂Fi′∂t=−(κij∂ϵcr∂xj−κ0ij∂ϵcr0∂xj)−Fi′,(10)\tau\frac{\partial F'_i}{\partial t} = -\left(\kappa_{ij}\frac{\partial\epsilon_{\rm cr}}{\partial x_j} - \kappa_{0ij}\frac{\partial\epsilon_{\rm cr0}}{\partial x_j}\right) - F'_i, \tag*{(10)}

where κij\kappa_{ij} and κ0ij\kappa_{0ij} are, respectively, the total and background diffusion tensors,

κij=κ⊥δij+(κ∥−κ⊥)B^iB^j,(11)\kappa_{ij} = \kappa_\perp\delta_{ij} + (\kappa_\parallel- \kappa_\perp)\hat{B}_i\hat{B}_j, \tag*{(11)}
κ0ij=κ⊥δij+(κ∥−κ⊥)B^0iB^0j,(12)\kappa_{0ij} = \kappa_\perp\delta_{ij} + (\kappa_\parallel- \kappa_\perp)\hat{B}_{0i}\hat{B}_{0j}, \tag*{(12)}

where the summation convention is understood, B^i=Bi/∣B∣\hat{B}_i = B_i/|\mathbf{B}| and B^0i=B0i/∣B0∣\hat{B}_{0i} = B_{0i}/|\mathbf{B}_0| are the components of the unit vectors in the direction of B\mathbf{B} and B0\mathbf{B}_0, respectively, and the tensors κij\kappa_{ij} and κ0ij\kappa_{0ij} have the same constant diffusivities κ∥\kappa_\parallel and κ⊥\kappa_\perp, respectively parallel and perpendicular to the relevant magnetic field. The non-Fickian description of the cosmic ray diffusion ensures their finite propagation speed controlled by the parameter τ\tau (this may be important given the very large cosmic ray diffusivity), and prevents numerical singularities in the cosmic ray propagation near the magnetic null points (Snodin et al. 2006). The parameters that control the diffusion cosmic rays parameters are adopted as τ=0.01 Gyr\tau= 0.01\,\mathrm{Gyr}, κ⊥=3.16×1025 cm2 s−1\kappa_\perp= 3.16\times10^{25}\,\mathrm{cm}^2\,\mathrm{s}^{-1} and κ∥=1.58×1028 cm2 s−1\kappa_\parallel= 1.58\times10^{28}\,\mathrm{cm}^2\,\mathrm{s}^{-1} as given in Rodrigues et al. [34]; Ryu et al. (2003, and references therein). We also include a weak isotropic component of the diffusion tensor κ0=0.5κ⊥\kappa_0 = 0.5\kappa_\perp to allow for unresolved random magnetic fields (in fact, this part of the diffusion tensor hardly affects the solutions). Since U0=0\mathbf{U}_0 = 0, equation (9) follows as the difference between equations (A2) and (A5) given in Appendix A.

The background state

The background state represents the magnetohydrostatic equilibrium in the galactic gravitational field,

∇P0=ρ0g,(13)\nabla P_0 = \rho_0\mathbf{g}, \tag*{(13)}

where the total pressure consists of the thermal gas pressure Pth0P_{\rm th0}, magnetic pressure Pm0P_{\rm m0} and cosmic ray pressure Pcr0P_{\rm cr0}:

P0=Pth0+Pm0+Pcr0,(14)P_0 = P_{\rm th0} + P_{\rm m0} + P_{\rm cr0}, \tag*{(14)}

and we note that the magnetic tension vanishes in this state because the background magnetic field is unidirectional, B0=(0,B0(z),0)\mathbf{B}_0 = (0,B_0(z),0) in the Cartesian frame (x,y,z)(x,y,z) with g\mathbf{g} aligned with the zz-axis.

It is convenient to introduce the pressure parameters βm\beta_{\rm m} and βcr\beta_{\rm cr} as the ratios of magnetic PmP_{\rm m} and cosmic ray PcrP_{\rm cr} pressures to the thermal pressure PthP_{\rm th}:

βm=PmPthandβcr=PcrPth,(15)\beta_{\rm m} = \frac{P_{\rm m}}{P_{\rm th}} \quad\text{and} \quad\beta_{\rm cr} = \frac{P_{\rm cr}}{P_{\rm th}}, \tag*{(15)}

where

Pth=cs2ρ,Pm=B28π,Pcr=(γcr−1)ϵcr.(16)P_{\rm th} = c_s^2\rho,\quad P_{\rm m} = \frac{B^2}{8\pi},\quad P_{\rm cr} = (\gamma_{\rm cr}-1)\epsilon_{\rm cr}. \tag*{(16)}

We consider isothermal gas. In many of our simulations (as listed in Table 1), the sound speed csc_s is 18 km s−118\,\mathrm{km\,s}^{-1} corresponding to the temperature of T=3.2×104 KT = 3.2\times10^4\,\mathrm{K}, which is within the range for the warm interstellar gas, but we also consider cs∈[9,29] km s−1c_s \in[9,29]\,\mathrm{km\,s}^{-1} to explore the effect of the gas ionization (through the analogue of sound speed) on the instability. In the background state, βm\beta_{\mathrm{m}} and βcr\beta_{\mathrm{cr}} are constant as listed in Table 1, but these ratios vary in space and time for the deviations from the background state as the instability develops. We mainly consider two combinations of these parameters; βm=1\beta_{\mathrm{m}}=1 and βcr=0\beta_{\mathrm{cr}}=0, or βm=βcr=0.5\beta_{\mathrm{m}}=\beta_{\mathrm{cr}}=0.5.

g(z)g(z)hh [kpc]βm\beta_{\mathrm{m}}βcr\beta_{\mathrm{cr}}B0(0)B_0(0) [µG]ν\nu [kpc km s−1^{-1}]η\eta [kpc km s−1^{-1}]csc_{\mathrm{s}} [km s−1^{-1}](Δx,Δy,Δz)(\Delta x,\Delta y,\Delta z) [pc]
Sim1Eq. (17)0.51.0070.10.0318(15, 7, 13)
Sim2Eq. (17)0.50.50.550.10.0318(15, 7, 13)
Sim3Eq. (17)1.00.50.570.10.0325(31, 15, 27)
Sim4Eq. (17)1.00.250.2540.10.0329(31,15, 27)
Sim5Eq. (18)0.41.0070.10.0318(15, 7, 13)
Sim6Eq. (18)0.40.50.550.10.0318(15, 7, 13)
Sim7Eq. (18)0.40.50.551.00.318(31, 15, 27)
Sim8Eq. (18)0.31.0030.10.039(31, 15, 27)
Sim9Eq. (18)0.40.50.550.10.0318(7, 7, 7)

Table 1. The list of simulations with the details of the gravitational field profile and parameters used, including the magnetic field strength B0(0)B_0(0) and cosmic ray energy density ϵcr0(0)\epsilon_{\mathrm{cr}0}(0) in the background state at z=0z = 0. The pressure ratios in the background state βm\beta_{\mathrm{m}} and βcr\beta_{\mathrm{cr}} are defined in equation (15) and hh is the effective exponential height of the gas in the background state. The viscosity ν\nu and magnetic diffusivity η\eta are given as the multiples of 1 kpc km s−1=3×1026 cm2 s−11\ {\rm kpc\ km\ s^{-1}} = 3 \times10^{26}\ {\rm cm^2\ s^{-1}}, and csc_{\mathrm{s}} is the sound speed. The grid spacing along the xx, yy and zz-axes are given in the penultimate column, with the computational box size of (4×4×3.5) kpc3(4 \times4 \times3.5)\ {\rm kpc^3}.

For a given gravitational profile g(z)g(z) we solve equation (13) to yield the corresponding density, magnetic field and cosmic ray profiles, given the constants βm\beta_{\mathrm{m}} and βcr\beta_{\mathrm{cr}}. Following some studies of the Parker instability (e.g., Giz & Shu 1993; Rodrigues et al. [34]) we use the gravitational acceleration profile of the form

g(z)=−2πGΣtanh⁡(z/H),(17)g(z)=-2\pi G\Sigma\tanh(z/H), \tag*{(17)}

with H=500 pcH=500\,\mathrm{pc}, Σ=102 M⊙ pc−2\Sigma=10^{2}\,\mathrm{M}_{\odot}\,\mathrm{pc}^{-2} for the surface mass density and GG Newton’s gravitational constant, which applies to the self-gravitating stellar disc. This gravity profile is used for the sake of comparison with the earlier results but most of our results are based on the gravity field appropriate for the Solar vicinity of the Milky Way, which also includes the contribution from the dark matter halo (Kuijken & Gilmore 1989),

g(z)=−a1zz12+z2−a2zz22,(18)g(z)=-\frac{a_{1}z}{\sqrt{z_{1}^{2}+z^{2}}}-\frac{a_{2}z}{z_{2}^{2}}, \tag*{(18)}

where a1=4.4×10−9 cm s−2a_{1}=4.4\times10^{-9}\,\mathrm{cm}\,\mathrm{s}^{-2}, a2=1.7×10−9 cm s−2a_{2}=1.7\times10^{-9}\,\mathrm{cm}\,\mathrm{s}^{-2}, z1=0.2 kpcz_{1}=0.2\,\mathrm{kpc} and z2=1 kpcz_{2}=1\,\mathrm{kpc}. Li et al. (2017) propose an alternative form for the gravitational acceleration better applicable at ∣z∣≳2 kpc|z|\gtrsim2\,\mathrm{kpc}, while Ferrière (1998, equation 35) suggests an extension of equation (18) to other galactocentric distances. For the gravity profile (18), the isothermal gas density, the strength of the magnetic field (directed along the yy-axis) and cosmic ray distribution in the magnetohydrostatic equilibrium follow as (see Appendix A)

ρ0(z)=ρ0(0)exp⁡(a1z1−z2+z12−a2z2/(2a1z2)cs2(1+βm+βcr)),(19)\rho_{0}(z)=\rho_{0}(0)\exp\left(\frac{a_{1}z_{1}-\sqrt{z^{2}+z_{1}^{2}}-a_{2}z^{2}/(2a_{1}z_{2})}{c_{\mathrm{s}}^{2}(1+\beta_{\mathrm{m}}+\beta_{\mathrm{cr}})}\right), \tag*{(19)}
B0(z)=∂A0x∂z=[8πβmcs2ρ0(z)]1/2,(20)B_{0}(z)=\frac{\partial A_{0x}}{\partial z}=\left[8\pi\beta_{\mathrm{m}}c_{\mathrm{s}}^{2}\rho_{0}(z)\right]^{1/2}, \tag*{(20)}
ϵcr0(z)=13βcrcs2ρ0(z),(21)\epsilon_{\mathrm{cr}0}(z)=\frac{1}{3}\beta_{\mathrm{cr}}c_{\mathrm{s}}^{2}\rho_{0}(z), \tag*{(21)}

where we adopt ρ0(0)=7×10−25 g cm−3\rho_{0}(0)=7\times10^{-25}\,\mathrm{g}\,\mathrm{cm}^{-3}. Although the density distribution in zz deviates from an exponential, it is useful to characterise it using the fitted exponential scale height hh given in Table 1, where we summarise the model parameters used in the simulations.

Similar expressions for the gravity profile (17) can be found in Rodrigues et al. [34]. Although the qualitative picture of the development and saturation of the instability is the same under both gravity profiles (17) and (18), there are important quantitative differences in both the linear eigenmodes and some features of the nonlinear state: in particular, the outflow speed.

The gravity profile (17) has been adjusted to provide similar values of gzg_z to those of (18) within the computational domain used here, ∣z∣≤Z0=1.75 kpc|z|\leq Z_{0}=1.75\,\mathrm{kpc} (Figure 1), so we do not expect any strong differences in the solutions obtained under the two profiles. However, some rather subtle differences may still occur.

Figure 1. The magnitudes of the gravitational acceleration, Eq. (17) (solid) and Eq. (18) (dotted), normalized to their magnitude at z=Z0z=Z_{0}.

The background magnetohydrostatic equilibrium is unstable. A weak initial velocity perturbation is introduced to launch the instability and the perturbations are then evolved using equations (6)–(9).

Boundary conditions and numerical implementation

We use the PENCIL CODE (Pencil Code Collaboration et al. [33]) to simulate a local Cartesian box within the gas layer; here we neglect rotation (for effects of rotation see Tharakkal et al. 2022). We simulate magnetised interstellar gas and cosmic rays in a region of the size 4×4×3.5 kpc34\times4\times3.5\,\mathrm{kpc}^{3} (symmetric about the midplane located at z=0z=0) along the xx, yy and zz-axes, respectively, and use periodic boundary conditions in xx and yy. The extent of the domain along the background magnetic field (yy) is chosen to be large enough to accommodate the dominant modes of the instability (see § 3.1).

The boundary conditions at the top and bottom of the domain, z=±Z0z=\pm Z_{0} with Z0=1.75 kpcZ_{0}=1.75\,\mathrm{kpc}, are given in equations (22)–(24) and chosen to minimise their effect on the interior of the domain while remaining physically justifiable. Their effect on the processes within the domain is weak or negligible also because Z0Z_{0} exceeds by a factor 2–3 the scales of the background gas, magnetic field and cosmic ray distributions.

The magnetic field perturbations satisfy the boundary conditions

∂ax∂z=∂ay∂z=0,az=0.(22)\frac{\partial a_{x}}{\partial z}=\frac{\partial a_{y}}{\partial z}=0,\qquad a_{z}=0. \tag*{(22)}

This allows Bz≠0B_{z}\ne0 at ∣z∣=Z0|z|=Z_{0}, and the open magnetic lines can support the gas and cosmic ray flows across the boundary.

The boundary condition for ρ′\rho' at ∣z∣=Z0|z|=Z_{0} imposes an exponential decrease at ∣z∣>Z0|z|>Z_{0} (i.e., within the ghost zones used to impose the boundary condition), at the scale that corresponds to the evolving vertical thermal pressure gradient at ∣z∣=Z0|z|=Z_{0}.

The horizontal velocity perturbations vanish at the top and bottom of the domain,

ux=uy=0at ∣z∣=Z0.(23)u_{x}=u_{y}=0\quad\text{at }|z|=Z_{0}. \tag*{(23)}

The vertical velocity at ∣z∣=Z0|z|=Z_{0} should be handled carefully to allow unrestricted gas outflow, while constraining the gas inflow to within numerically stable limits. To suppress numerical instabilities (associated with spurious strong advection of gas from the outside of the domain), any negative uzu_{z} at z=Z0z=Z_{0} and positive uzu_{z} at z=−Z0z=-Z_{0} (an inflow, [uz(x,y,z)sign⁡z]∣z∣=Z0<0[u_{z}(x,y,z)\operatorname{sign}z]_{|z|=Z_{0}}<0) are quadratically quenched to zero within the three ghost points at ∣z∣>Z0|z|>Z_{0}. At those horizontal positions on the top and bottom boundaries where the gas flows out of the domain ([uz(x,y,z)sign⁡z]∣z∣=Z0>0)([u_{z}(x,y,z)\operatorname{sign}z]_{|z|=Z_{0}}>0), the vertical derivatives of uzu_{z} of all relevant orders are assumed to vanish in the ghost zone. Thus, the gas that flows out through ∣z∣=Z0|z| = Z_0 retains its speed in the ghost zone. These conditions, imposed within the ghost zone rather than on the boundary, can be written as

inflow: uz→0u_z \to0

outflow: ∂nuz∂zn=0\frac{\partial^n u_z}{\partial z^n} = 0 for any nn } at ∣z∣>Z0|z| > Z_0,

{#eq:24}

where the higher-order derivatives are involved when hyperdiffusion is used.

The top and bottom boundaries are open for the cosmic ray fluid, allowing it to escape along open magnetic lines at ∣z∣=±Z0|z| = \pm Z_0,

∂2ϵcr∂z2=0,∂Fx∂z=∂Fy∂z=0,Fz=0.(24)\frac{\partial^2 \epsilon_{\mathrm{cr}}}{\partial z^2} = 0,\qquad\frac{\partial F_x}{\partial z} = \frac{\partial F_y}{\partial z} = 0,\qquad F_z = 0. \tag*{(24)}

The viscosity and magnetic diffusivity are global constants in our simulations, and we considered a selection of their values shown in Table 1; in all cases, the magnetic Prandtl number is Pm=ν/η≈3P_{\mathrm{m}} = \nu/\eta\approx3, as in Rodrigues et al. [34]. The magnitudes of the transport coefficients are close to (but, in all models except for Sim7, smaller than) their turbulent values in the ISM; this is appropriate since our simulations do not not include the turbulence driven by supernovae. As the instability saturates, small-scale fluctuations get stronger making the system more susceptible to numerical instabilities. The fluctuations at the resolution scale are regularised using sixth-order hyperdiffusion with the diffusivity ν6=4×10−11 kpc6 Gyr−1\nu_6 = 4 \times10^{-11}\ {\rm kpc^6\ Gyr^{-1}} (e.g., Brandenburg & Sarson (2002); Haugen & Brandenburg [13]), constraining the maximum mesh Reynolds number to ReΔ=csΔx5/ν6\mathrm{Re}_{\Delta} = c_{\mathrm{s}}\Delta x^5/\nu_6. The magnitude of ν6\nu_6 was chosen to ensure that ReΔ\mathrm{Re}_{\Delta} is close to unity along the lowest-resolution dimension, xx. Shocks arising in the system are regularised using the second-order shock diffusion (for implementation, see Gent et al. (2020)). To avoid negative gas density values that may arise in intense divergent flows, we apply a density floor and impose the restriction ρ0+ρ′≥ρmin⁡\rho_0+\rho' \geq\rho_{\min} with ρmin⁡=0.01 M⊙ kpc−3\rho_{\min} = 0.01\,M_\odot\ {\rm kpc^{-3}}. The magnitude of ρmin⁡\rho_{\min} chosen is about one order of magnitude lower than the initial minimum density in the simulation box.

Table 1 presents a summary of the simulations which are based on the two forms of the gravity acceleration, given in equations (17) and (18), and a selection of the pressure ratios of the magnetic field, βm\beta_{\mathrm{m}}, and cosmic rays, βcr\beta_{\mathrm{cr}}, to the thermal pressure in the background state. These factors control the intensity of the instability and the contribution of cosmic rays to it; the case βcr=0\beta_{\mathrm{cr}} = 0 represents the magnetic buoyancy instability. We list the sound speed csc_{\mathrm{s}} in each model. We compare a simulation with the fiducial values (15,7,13) pc(15, 7, 13)\ {\rm pc} for the grid spacing (Δx,Δy,Δz)(\Delta x, \Delta y, \Delta z) along the corresponding axes, to a simulation with (7,7,7) pc(7, 7, 7)\ {\rm pc}.

Results

Our main conclusions are drawn from the numerical models Sim5 and Sim6 in Table 1. They differ in the contribution of cosmic rays described using equations (9) and (10): Sim5 is the case of the magnetic buoyancy instability, enhanced by cosmic rays in Sim6. The total pressure in the background state remains unchanged since βm+βcr=1\beta_{\mathrm{m}}+\beta_{\mathrm{cr}}=1 in each case.

The linear instability

The exponential growth in the perturbations of magnetic field and gas speed in the linear phase of the instability at t≲0.4 Gyrt \lesssim0.4\ {\rm Gyr} (and after the decay of the initial perturbations down to the leading eigenfunction) is clearly visible in Figure 2a,Figure 2b. As shown in Figure 2c, the total magnetic and cosmic ray energy densities vary little with time as long as the perturbations remain much weaker than the background fields; their subsequent nonlinear evolution is discussed in § 3.2. The growth rate of the instability in the models of Table 1 is quantified in Figure 3, which presents both the dimensionless growth rate Γh/v\Gamma h/v and its unit v/hv/h for various choices of the characteristic speed vv.

Evolution of root-mean-square magnetic-field perturbation, gas speed, and magnetic and cosmic-ray energy densities

Figure 2. The evolution of the root-mean-square magnitudes at the midplane z=0z = 0 of (a) the magnetic field perturbation ∣b∣|b|, normalised to B0(0)B_0(0), and (b) gas speed in the models Sim5 (solid) and Sim6 (dotted) at z=0z = 0. The linear stage of the instability ends at about t≃0.4t \simeq0.4 Gyr. Panels (c) and (d) show the normalised total magnetic and cosmic ray energy densities at the midplane for the same models, ⟨B(0)/B0(0)⟩xy\langle B(0)/B_0(0)\rangle_{xy} (solid for Sim5 and dotted for Sim6) and ⟨ϵcr(0)/ϵcr0(0)⟩xy\langle\epsilon_{\rm cr}(0)/\epsilon_{\rm cr0}(0)\rangle_{xy} (dotted, for Sim6) respectively.

Figure 3. The dimensionless growth rate Γh/v\Gamma h/v of the r.m.s. magnetic field perturbation is shown for various models, normalized with various crossing times over the density scale height hh: v=csv = c_{\rm s} (diamonds), v=VAv = V_{\rm A} (circles) and v=ceffv = c_{\rm eff} of equation (25) (squares). The horizontal axis presents the corresponding inverse time scale v/hv/h. The model codes are shown within the symbols as in Table 1, e.g. S5 represents Model Sim5.

In agreement with earlier analytical and numerical models, cosmic rays make the system more unstable and the growth rate is smaller in Model Sim5 which has no cosmic rays, than in Sim6 where the cosmic rays and magnetic field have equal pressures in the background state. The growth rate Γ\Gamma of the root-mean-square (r.m.s.) velocity and magnetic fields in these models increases by approximately 20% due to the cosmic rays from 19 Gyr−1^{-1} to 25 Gyr−1^{-1}, similar to the increase in Γ\Gamma in the analytical models of Giz & Shu (1993) and Ryu et al. (2003) and the simulations of Rodrigues et al. [34] (the latter authors provide a detailed comparison of the growth rates in several models).

The growth rate of the Parker instability is often assumed to scale with (and, often, to be of the order of) the inverse sound or Alfvén crossing time over the density scale height, cs/hc_{\rm s}/h or VA/hV_{\rm A}/h, respectively. Our results show that this scaling is rather inaccurate as the growth rate depends significantly on other parameters of the system. Figure 3 shows the dimensionless growth rates for models from Table 1 normalised by various crossing times, Γh/v\Gamma h/v, where v=csv = c_{\rm s}, v=VAv = V_{\rm A} and v=ceffv = c_{\rm eff} with

ceff2=cs2+VA2+βcrcs2.(25)c_{\rm eff}^{2} = c_{\rm s}^{2} + V_{\rm A}^{2} + \beta_{\rm cr}c_{\rm s}^{2}. \tag*{(25)}

We note that the horizontal spread of the data points indicates a variation of Γh/v\Gamma h/v by a factor of two between various models for any choice of vv. Secondly, if the scaling of Γ\Gamma with

v/hv/h were perfect, the data points would be form a horizontal line for the most appropriate choice of vv. This does not occur for any choice of vv, but the scaling with ceffc_{\rm eff} (shown with squares) is marginally better, especially if the Model Sim1 is disregarded. The models that are not included in Figure 3 have similar growth rates and do not provide additional information.

Analytical and numerical studies of the Parker instability rely on a wide diversity of assumptions and, perhaps not surprisingly, different models claim different properties of the most rapidly growing mode (we note that many studies are based on ideal magnetohydrodynamics). For example, Parker & Lerche (1969) assume that the perturbations are only weakly dependent on zz (kz→0k_z \to0). The most rapidly growing mode of Lachèze-Rey et al. (1980) has kx=0k_x = 0, (kyh)2=0.5(k_y h)^2 = 0.5–0.60.6. Giz & Shu (1993) assume that the perturbations have finite kxk_x and kyk_y and the amplitudes of both bzb_z and uzu_z vary similarly with zz while ∣uz∣∝ρ0−1/2(z)|u_z| \propto\rho_0^{-1/2}(z) at large ∣z∣|z|; the most rapidly growing mode has a very large kxk_x. Foglizzo & Tagger (1994) adopt kx≠0k_x \ne0 and ky≠0k_y \ne0 and the magnitude of the perturbations is assumed to be proportional to ρ0−1/2(z)\rho_0^{-1/2}(z). In the model of Parker & Lerche (1969), the system is the most unstable at the largest possible kxk_x, and Giz & Shu (1993) assume that the most unstable mode has a large radial wave number, kxh=103k_xh = 10^3 (their Fig. 2), whereas Heintz & Zweibel (2018) assume that kx→0k_x \to0. Kim & Ryu (2001) argue that kx=kz=0k_x = k_z = 0 in the most rapidly growing mode. Shukurov & Subramanian (2021, Section 2.8.2) assume that the perturbation amplitudes of the density, thermal pressure and Lagrangian displacement scale with the ρ0−1/2(z)\rho_0^{-1/2}(z) while the amplitude of bb is independent of zz, and that the perturbations have kx=0k_x = 0 but finite kyk_y. The wave numbers of the unstable modes along the background magnetic field, kyk_y, are limited from above (Foglizzo & Tagger (1994)),

ky2≤kP2=(1+βm+βcr)βm+βcr2βmh−2.(26)k_y^2 \leq k_{\rm P}^2 = (1+\beta_{\rm m}+\beta_{\rm cr})\frac{\beta_{\rm m}+\beta_{\rm cr}}{2\beta_{\rm m}}h^{-2}. \tag*{(26)}

For a finite magnetic diffusivity η\eta, the radial wave number must be smaller than (Foglizzo & Tagger (1994))

kˉx≃cshη.(27)\bar{k}_x \simeq\sqrt{\frac{c_{\rm s}}{h\eta}}. \tag*{(27)}

Our simulations are consistent with some of these assumptions and conclusions. The spatial structure of the most unstable linear mode is characterised by Figure 4 where we present the 2D power spectra of uzu_z and bzb_z in the (kx,ky)(k_x,k_y)-plane, averaged over ∣z∣<1.75\lvert z\rvert< 1.75 kpc, for the Models Sim5 (without cosmic rays) and Sim6 (with cosmic rays). The spectra of

2D power spectra in the $(k_x,k_y)$-plane

Figure 4. 2D power spectra in the (kx,ky)(k_x,k_y)-plane at t=0.3t = 0.3 Gyr (the linear stage of the instability) at ∣z∣≤1.75\lvert z\rvert\le1.75 kpc for the perturbations as indicated in the colour bar labels of vertical velocity uzu_z in kpc2^2 km2^2 s−2^{-2} and the vertical magnetic field bzb_z in kpc2^2 μG2^2. Panels (a) and (c) depict Model Sim5 and (b) and (d) present Model Sim6.

these variables are nearly identical. The presence of cosmic rays does not change kxk_x much but increases kyk_y significantly (by a factor of two). The energy corresponding to the most unstable mode is also higher by a factor of 4 in Sim6 when compared to Sim5. This confirms that cosmic rays make the system more unstable. The values of kxk_x are well within the upper limit kˉx≈40\bar{k}_x \approx40 kpc−1^{-1} of equation (27).

In Model Sim6, kP=3.5k_P = 3.5 kpc−1^{-1}, corresponding to the wavelength λP=1.8\lambda_P = 1.8 kpc (and kP=2.5k_P = 2.5 kpc−1^{-1} in Model Sim5).

The size of the computational domain along the yy-axis, 4 kpc, is larger than that, so it can accommodate the most unstable modes (see also § 4.6 where we also discuss the wave numbers of the most unstable mode). The spectrum of the perturbations in this model, shown in Figure 4b, has the maximum at ky≃3.1k_y \simeq3.1 kpc−1^{-1}, in agreement with equation (26). As discussed in § 4.6, the power spectrum of the perturbations has a pronounced maximum at ky≈3k_y \approx3 kpc−1^{-1} but a broad range of modes at kx≲3k_x \lesssim3 kpc−1^{-1} carry similar energies for the parameter values of Model Sim6.

Giz & Shu (1993) discuss continuous modes assuming that z=constz = \mathrm{const}. Figure 5 shows the variation with zz of the horizontal average ρ0⟨uz2⟩xy\rho_0\langle u_z^2\rangle_{xy} (normalized to its mean value across the computational domain; we note that ⟨uz2⟩xy\langle u_z^2\rangle_{xy} grows exponentially with time).2 The variation is by an order of magnitude across the full range of zz and is not monotonic. Shukurov & Subramanian (2021) consider solutions where bb is independent of zz and the perturbations in the other variables decrease as exp⁡[−z/(2h)]\exp[-z/(2h)] (in particular, ρ0uz2=const\rho_0u_z^2 = \mathrm{const}). While the perturbations in our simulations do not satisfy these assumptions closely, the one-dimensional perturbation power spectra in zz show that the most unstable modes of uzu_z and bzb_z have kz≈3.5k_z \approx3.5 kpc−1^{-1} in Models Sim5, Sim 6 and Sim9, corresponding to the scale 2π/kz=1.82\pi/k_z = 1.8 kpc. As shown in Figure 6, the magnitude of perturbations in different quantities vary differently with zz, in contradiction to the assumption of the standard stability analysis. In particular, ρ\rho and ecre_{\rm cr} have oscillations in zz superimposed on a large-scale variation. The perturbations in the velocity and magnetic field components uxu_x and bxb_x (which have relatively small amplitudes) and uzu_z and bzb_z (having large amplitudes) are oscillatory in zz. Meanwhile, uyu_y and byb_y do not have pronounced oscillations in zz. The values of kzk_z presented characterise mainly the oscillatory parts of the variation and should be interpreted with caution.

Figure 5. For Sim6 during the linear and weakly nonlinear stages of the instability at times indicated in the legend the vertical variation of ρ0⟨uz2⟩xy\rho_0\langle u_z^2\rangle_{xy} normalized by its volume averaged energy density ⟨ρ0uz2⟩\langle\rho_0u_z^2\rangle.

Vertical profiles of perturbation magnitudes

Figure 6. The vertical profiles of the perturbation magnitudes normalised by their volume averaged r.m.s. values throughout the domain in Model Sim6 for different variables as denoted in the legends at (x,y)=(−1.2,−1.8) pc(x,y)=(-1.2,-1.8)\ {\rm pc} and t=0.3 Gyrt=0.3\ {\rm Gyr}. The profiles at fixed (x,y)(x,y) are shown because the horizontal averages of the perturbations are negligible during the linear phase of the instability.

The stratification of the system affects the spatial structure of the linearly unstable modes; in particular, their wave numbers vary with position and time. For example, Figure 7 shows that the wave number kyk_y of the dominant mode changes with both zz and tt, generally increasing with time at a fixed zz. During the late stages of the exponential growth phase, t≃0.3 Gyrt\simeq0.3\ {\rm Gyr}, kyk_y increases from 0 to 3 kpc−13\ {\rm kpc}^{-1} with distance from the midplane and approaches ky≈3 kpc−1k_y\approx3\ {\rm kpc}^{-1} in the nonlinear phase at all distances (see also Figure 9). Correspondingly, the growth rate of the instability is no longer a constant but becomes a function of zz. As shown in Figure 8, the growth rate of the vertical velocity increases with ∣z∣|z| from 24 Gyr−124\ {\rm Gyr}^{-1} at z=0z=0 to 32 Gyr−132\ {\rm Gyr}^{-1} at z=1.5 kpcz=1.5\ {\rm kpc}. This makes the linear evolution more complicated than the linear analytical models suggests but the properties of the dominant mode are broadly consistent with certain caveats discussed here.

Figure 7. Time-latitude diagram of the azimuthal wave number kyk_y for yy during the linear instability of Model Sim6.

Figure 8. Model Sim6 evolution of ⟨uzrms⟩xy\langle u_{z{\rm rms}}\rangle_{xy} at altitudes indicated in the legend. Corresponding growth rate Γ\Gamma of the instability listed in the legend applies for 0.4≲t≲0.5 Gyr0.4 \lesssim t \lesssim0.5\ {\rm Gyr}. Evolution is similar for z<0z<0.

Gas density and cosmic ray energy density distributions with magnetic field lines in Model Sim6

Figure 9. Upper row: the gas density distribution (colour) and magnetic lines, with the local field direction indicated, in the cross-section at x=800x = 800 pc in Model Sim6 at (a) t=0.30t = 0.30 Gyr and (b) t=0.45t = 0.45 Gyr. Lower row: as above but for the cosmic ray energy density (colour) at (c) t=0.30t = 0.30 Gyr and (d) t=0.45t = 0.45 Gyr. The separation of the magnetic lines does not represent the field strength.

Figure 9 illustrates the development of the spatial structures of the magnetic field, gas density and cosmic rays in

the linear stage of the instability in panels (a) and (c), where the perturbations are weak at all ∣z∣|z|, and at the end of the linear stage in panels (b) and (d), where the perturbations are quite strong at ∣z∣≳h=0.4 kpc|z|\gtrsim h=0.4\ {\rm kpc} but still are weaker closer to the midplane. The magnetic field lines are aligned with the yy-axis initially, and the instability excites the undular modes which deform the magnetic field lines and cause the gas and cosmic rays to be redistributed in zz. As the system evolves into the nonlinear stage, magnetic fluctuations develop at progressively smaller scales.

Model Sim9 has a somewhat higher spatial resolution than Model Sim6 (with all other parameters being similar). The wave numbers and growth rates obtained from the two models are quite similar confirming that model Sim6 resolves well the spatial structures developing in the system. Model Sim8 is similar to Model Sim5 but has a lower sound speed (thus, a lower thermal pressure) and, consequently, a lower gas scale height. This strengthens the instability and makes the gas flow much more irregular at greater heights in the nonlinear state.

The nonlinear instability

The evolution of the r.m.s. velocity and magnetic fields shown in Figure 2 marks three distinct stages of the instability development. The initial stage of exponential growth is followed by a short transitional (weakly nonlinear) period, which starts at t=0.4 Gyrt=0.4\ {\rm Gyr} and lasts for around 0.1 Gyr0.1\ {\rm Gyr} during which the growth slows down and the instability saturates to the final steady state. The total magnetic and cosmic ray energy densities, which remain nearly constant throughout the linear phase, start decaying during the transitional phase. In the statistical steady state they retain only a few percent of the initial values. Meanwhile, the thermal pressure in the system remains almost constant throughout the simulation, except for a small decrease (by 4%) in the nonlinear state.

The changes in the system as the instability evolves are illustrated in Figure 10 which shows the horizontal cross-section of the vertical velocity. The pattern of the perturbations at the linear stage is quite regular and, of course, the perturbations average to zero in the horizontal planes (Figure 10a). However, this pattern undergoes substantial modification during the transitional stage when a systematic transient inflow develops at this altitude (Figure 10b). An outflow at higher altitudes (visible in Figure 11g and discussed in § 4.2) is

Figure 10. The vertical velocity perturbation uzu_z from Model Sim6 at z=0.4z = 0.4 kpc during (a) the linear stage of the instability (t=0.3t = 0.3 Gyr), (b) the transitional stage (t=0.6t = 0.6 Gyr) and (c) the nonlinear state (t=0.9t = 0.9 Gyr).

Figure 11. From Model Sim6 the horizontally averaged vertical profiles of (left column (a)–(c)) the normalised perturbations and (centre column (d)–(f)) the corresponding total profiles. The solid, dotted, dashed and dash-dotted lines respectively. Timing of each profile is as indicated in the legend of (f). Panel (g) shows the horizontally averaged vertical velocity normalised to the sound speed ⟨uz⟩xy/cs\langle u_z\rangle_{xy}/c_s.

titudes (visible in Figure 11g and discussed in § 4.2) is followed by a rather chaotic pattern in the nonlinear state (Figure 10c). The rather regular magnetic loops characteristic of the linear instability (Figure 9) evolve into an ostensibly random pattern. The model of Rodrigues et al. [34], which is evolved as far as the very early nonlinear stage, has a similar transition to a chaotic state.

Figure 12 shows the 2D power spectra of the perturbations in the vertical velocity and magnetic field which are useful to compare with Figure 4. As might be expected, nonlinear effects generally broaden the power spectra, with some notable differences between Models Sim5 and Sim6. In Model Sim5, uzu_z develops modes with small kyk_y but large kxk_x while the linear mode remains prominent. The maximum in the Fourier spectrum of bzb_z shifts to higher values of kxk_x at the same kyk_y as in the linear stage. Meanwhile, the maxima in the power spectra of both uzu_z and bzb_z in Model Sim6 shift to lower kyk_y. However, the spectrum of uzu_z is dominated by very small kxk_x and kyk_y whereas the spectrum of bzb_z is maximum at (kx,ky)=(3.1,1.5) kpc−1(k_x,k_y)=(3.1,1.5)\ {\rm kpc}^{-1}. Altogether, the similarity between the spatial structures of the linear modes and the nonlinear solution is limited and not straightforward.

Four panels of perturbation power spectra

Figure 12. As Figure 4 at ∣z∣≤1.75|z| \le1.75 kpc, but t=0.9t = 0.9 Gyr (the nonlinear stage of the instability) for the perturbations as indicated in the colour bar labels. Panels (a) and (c) depict Model Sim5 and (b) and (d) represent Model Sim6.

The vertical redistribution of the gas, magnetic field and cosmic rays is shown in Figure 11, which presents the horizontally averaged deviations from the background distributions in Panels (a)–(c) and (g) as well as the distributions of the total gas density, magnetic field and cosmic rays in the linear, transitional and nonlinear phases of the instability. In the linear stage the perturbations, periodic in horizontal planes, have vanishing horizontal averages. Non-vanishing horizontal averages arise only due to nonlinear effects. The perturbation in the gas density in the transitional and nonlinear stages is positive near the midplane and negative away from the disc: the nonlinear instability leads to a reduction in the density scale height, so the gas disc becomes thinner. As a result, the mean gas density at the midplane increases from 7×10−25 g cm−37 \times10^{-25}\ {\rm g\,cm}^{-3} to 1.08×10−24 g cm−31.08 \times10^{-24}\ {\rm g\,cm}^{-3} as the instability develops. The average energy densities of the total magnetic field and cosmic rays at the midplane are reduced by more than 75%, expanding their vertical profiles. Similar behaviour occurs in the simulations of Heintz et al. (2020) (e.g., their Fig. 10) which capture the weakly nonlinear, transitional stage of the instability.

Figure 11g shows the vertical velocity planar averages ⟨uz⟩xy\langle u_z\rangle_{xy} at different times. The nonlinear effects drive a systematic inflow at ∣z∣≲1 kpc|z| \lesssim1\ {\rm kpc} and an outflow at ∣z∣≳1 kpc|z| \gtrsim1\ {\rm kpc} in the transitional stage which, however, is transformed into a weaker inflow at t=0.9 Gyrt=0.9\ {\rm Gyr} while the system still adjusts towards the steady state. These features are further discussed in § 4.2. At t=1.6 Gyrt=1.6\ {\rm Gyr}, the system of Model Sim6 reaches a statistically steady state with a residual inflow with ∣⟨uz⟩xy∣/cs≃0.5|\langle u_z\rangle_{xy}|/c_s \simeq0.5 at ∣z∣≳1 kpc|z| \gtrsim1\ {\rm kpc} (which, however, carries little mass). The magnetic field and cosmic ray energy density continue to decrease, saturating at the midplane values ⟨B(0)⟩xy/B0(0)≈0.16\langle B(0)\rangle_{xy}/B_0(0)\approx0.16 and ⟨ϵcr(0)⟩xy/ϵcr,0(0)≈0.02\langle\epsilon_{\rm cr}(0)\rangle_{xy}/\epsilon_{\rm cr,0}(0)\approx0.02, while the gas density increases to ⟨ρ(0)⟩xy/ρ0(0)≈1.6\langle\rho(0)\rangle_{xy}/\rho_0(0)\approx1.6.

The magnetic buoyancy instability is driven by the vertical gradient of the magnetic field strength, and it might be expected that it would saturate via reducing the gradient to a marginal value. However, the system follows a much more dramatic path, getting rid of the magnetic field altogether. As the instability produces strong vertical magnetic perturbations, the cosmic rays are channelled out from the disc and diffuse along the magnetic field at a high rate. The instability results in wide distributions of both magnetic field and cosmic rays enveloping a relatively thin thermal gas disc. A similar kind of evolution also occurs in the simulations of Heintz et al. (2020) and Girichidis et al. (2022). However, the form of the nonlinear state is sensitive to such features of the system as rotation, and to the mechanism maintaining the unstable magnetic field in the disc. As we show in Tharakkal et al. (2022), the instability in a rotating system can lead even to the reversal of the magnetic field in the disc.

Discussion

In this section we discuss the implications of our results for the dynamics of the interstellar medium and describe the force balance in the system.

Cross-correlation between energy densities

Understanding the relative spatial distributions of the gas, magnetic field and cosmic rays in the interstellar medium is crucial for the interpretation of the observations in the radio and other wavelength ranges. In particular, the assumption of a tight correlation between cosmic rays and magnetic fields (e.g., the energy equipartition) is routinely used in the interpretations of synchrotron observations (Seta & Beck 2019, and references therein). The correlation or anti-correlation between the thermal electron density and magnetic field can affect significantly the Faraday rotation (Beck et al. 2003).

Table 2 presents the Pearson cross-correlation coefficient rr between the fluctuations in the energy densities ϵi′=ϵi−⟨ϵi⟩xy\epsilon'_i = \epsilon_i - \langle\epsilon_i\rangle_{xy}, with ⟨⋯ ⟩xy\langle\cdots\rangle_{xy} for the horizontal average and i=th, cr, m, ki = \mathrm{th},\ \mathrm{cr},\ \mathrm{m},\ \mathrm{k} for the thermal, cosmic-ray, magnetic and kinetic (ϵk=12ρu2\epsilon_{\mathrm{k}} = \frac{1}{2}\rho u^2) energy densities, in the nonlinear stage:

ϵth′\epsilon'_{\mathrm{th}}ϵcr′\epsilon'_{\mathrm{cr}}ϵm′\epsilon'_{\mathrm{m}}ϵk′\epsilon'_{\mathrm{k}}
ϵth′\epsilon'_{\mathrm{th}}1,11,\quad1−0.70, −0.64-0.70,\ -0.64−0.80, −0.17-0.80,\ -0.170.17, −0.220.17,\ -0.22
ϵcr′\epsilon'_{\mathrm{cr}}1,11,\quad10.58, −0.460.58,\ -0.46−0.2, −0.47-0.2,\ -0.47
ϵm′\epsilon'_{\mathrm{m}}1,11,\quad10.01, 0.270.01,\ 0.27
ϵk′\epsilon'_{\mathrm{k}}1,11,\quad1

Table 2. The Pearson correlation coefficients rr between various energy densities in the statistically steady state at t=1.6t = 1.6 Gyr presented as a,ba,b, where aa and bb refer to the altitudes z=0.5z = 0.5 and 1.51.5 kpc, respectively.

ϵm′=B28π−⟨B28π⟩xy,ϵcr′=ϵcr−⟨ϵcr⟩xy,ϵth′=cs2(ρ−⟨ρ⟩xy),ϵk′=12ρu2−⟨12ρu2⟩xy.(28)\begin{aligned} \epsilon'_{\mathrm{m}} &= \frac{B^2}{8\pi} - \left\langle\frac{B^2}{8\pi}\right\rangle_{xy}, \qquad&\epsilon'_{\mathrm{cr}} &= \epsilon_{\mathrm{cr}} - \langle\epsilon_{\mathrm{cr}}\rangle_{xy}, \\ \epsilon'_{\mathrm{th}} &= c_s^2\left(\rho- \langle\rho\rangle_{xy}\right), &\epsilon'_{\mathrm{k}} &= \frac{1}{2}\rho u^2 - \left\langle\frac{1}{2}\rho u^2\right\rangle_{xy}. \tag*{(28)} \end{aligned}

The relations between these variables are different near the midplane and at a higher altitude, so both are presented (the results at z<0z < 0 are similar). The fluctuations in thermal energy density are significantly anti-correlated with those in the magnetic and cosmic ray energy densities near the midplane and with the cosmic ray energy fluctuations at the higher altitude. This appears to reflect the horizontal pressure balance where the total average pressure is independent of xx and yy. The kinetic energy density shows no significant correlation with any other variable. The distribution of the cosmic rays is only weakly correlated with that of the magnetic field near the midplane (r=0.58r = 0.58) but not away from it (r=−0.46r = -0.46).

Figure 13 shows the cross-correlation coefficient of the fluctuations in the magnetic field and cosmic ray energy densities at various heights in the nonlinear stage of Model Sim6. The cross-correlation coefficient of these two quantities increases with ∣z∣|z| reaching a maximum at ∣z∣≃0.5|z| \simeq0.5 kpc and then decreases to become negative (anti-correlation) at ∣z∣≳1|z| \gtrsim1 kpc.

Figure 13. The vertical variation of the Pearson cross-correlation coefficient between the magnetic (ϵm\epsilon_{\mathrm{m}}) and cosmic ray (ϵcr\epsilon_{\mathrm{cr}}) energy densities in Model Sim6 averaged over the time interval 1.2≤t≤1.71.2 \le t \le1.7 Gyr with the cadence Δt=0.1\Delta t = 0.1 Gyr.

To clarify the cause of the correlations, consider the magnetic field and cosmic rays. The magnetic field and cosmic ray energy density can be decomposed into their horizontal averages and fluctuations, B=B‾+b\boldsymbol{B} = \overline{\boldsymbol{B}} + \boldsymbol{b} and ϵcr=ϵ‾+ϵ′\epsilon_{\rm cr} = \overline{\epsilon} + \epsilon', where B‾=⟨B⟩xy\overline{\boldsymbol{B}} = \langle\boldsymbol{B}\rangle_{xy} and ϵ‾=⟨ϵcr⟩xy\overline{\epsilon} = \langle\epsilon_{\rm cr}\rangle_{xy}, such that ⟨B‾⟩xy=B‾\langle\overline{\boldsymbol{B}}\rangle_{xy} = \overline{\boldsymbol{B}}, ⟨b⟩xy=0\langle\boldsymbol{b}\rangle_{xy} = 0, ⟨ϵ‾⟩xy=ϵ‾\langle\overline{\epsilon}\rangle_{xy} = \overline{\epsilon} and ⟨ϵ′⟩xy=0\langle\epsilon'\rangle_{xy} = 0. The correlation coefficient of the total magnetic and cosmic ray energies follows as

r(ϵm,ϵcr)=⟨(B2−⟨B2⟩xy)(ϵcr−ϵ‾)⟩xyσmσϵ,(29)r(\epsilon_{\rm m},\epsilon_{\rm cr}) = \frac{\left\langle\left(B^2-\langle B^2\rangle_{xy}\right)\left(\epsilon_{\rm cr}-\overline{\epsilon}\right)\right\rangle_{xy}}{\sigma_{\rm m}\sigma_{\epsilon}}, \tag*{(29)}

where σm2=⟨B4⟩xy−⟨B2⟩xy2\sigma_{\rm m}^2 = \langle B^4\rangle_{xy} - \langle B^2\rangle_{xy}^2 and σϵ2=⟨(ϵ′)2⟩xy\sigma_{\epsilon}^2 = \langle(\epsilon')^2\rangle_{xy} are the variances of the magnetic (the factor 8π8\pi can be omitted here) and cosmic ray energy densities. Since B2=B‾2+2B‾⋅b+b2B^2 = \overline{B}^2 + 2\overline{\boldsymbol{B}}\cdot\boldsymbol{b}+b^2, ⟨B2⟩xy=B‾2+⟨b2⟩xy\langle B^2\rangle_{xy} = \overline{B}^2 + \langle b^2\rangle_{xy} and ⟨⟨b2⟩xyϵ′⟩xy=⟨b2⟩xy⟨ϵ′⟩xy=0\langle\langle b^2\rangle_{xy}\epsilon'\rangle_{xy} = \langle b^2\rangle_{xy}\langle\epsilon'\rangle_{xy} = 0, we have

r(ϵm,ϵcr)=⟨(2B‾⋅b+b2)ϵ′⟩xyσmσϵ.(30)r(\epsilon_{\rm m},\epsilon_{\rm cr}) = \frac{\left\langle\left(2\overline{\boldsymbol{B}}\cdot\boldsymbol{b}+b^2\right)\epsilon'\right\rangle_{xy}}{\sigma_{\rm m}\sigma_{\epsilon}}. \tag*{(30)}

Both terms contribute similarly to the correlation coefficient: r(b2,ϵcr)=0.23r(b^2,\epsilon_{\rm cr}) = 0.23 and r(B⋅b,ϵcr)=0.43r(\boldsymbol{B}\cdot\boldsymbol{b},\epsilon_{\rm cr}) = 0.43 at z=0.5z = 0.5 kpc and −0.48-0.48 and −0.18-0.18 at z=1.5z = 1.5 kpc, respectively.

The variance of the magnetic energy density can be represented as

σm2=4⟨(B‾⋅b)2⟩xy+⟨b4⟩xy−⟨b2⟩xy2,(31)\sigma_{\rm m}^2 = 4\left\langle\left(\overline{\boldsymbol{B}}\cdot\boldsymbol{b}\right)^2\right\rangle_{xy} + \langle b^4\rangle_{xy} - \langle b^2\rangle_{xy}^2, \tag*{(31)}

since ⟨b2B‾⋅b⟩xy=0\langle b^2\overline{\boldsymbol{B}}\cdot\boldsymbol{b}\rangle_{xy}=0 for symmetric probability distributions of the Cartesian components of b\boldsymbol{b}.

To provide an illustration, if the magnetic field fluctuations are isotropic and the Cartesian component bib_i of b\boldsymbol{b} are statistically independent, each having the standard deviation σi\sigma_i, we have ⟨b2⟩xy=3σi2\langle b^2\rangle_{xy}=3\sigma_i^2 and ⟨bi4⟩xy=3σi4\langle b_i^4\rangle_{xy}=3\sigma_i^4, so that ⟨b4⟩xy=⟨(∑i=13bi2)2⟩xy=15σi4\langle b^4\rangle_{xy}=\left\langle\left(\sum_{i=1}^{3}b_i^2\right)^2\right\rangle_{xy}=15\sigma_i^4 and ⟨b4⟩xy−⟨b2⟩xy2=6σi4\langle b^4\rangle_{xy}-\langle b^2\rangle_{xy}^2=6\sigma_i^4.

As discussed by Beck et al. (2003) (see also Sect. 13.2 of Shukurov & Subramanian 2021, where typographical errors of the original publication are corrected), the anti-correlation of the total magnetic field strength and thermal gas density can significantly affect the interpretation of the Faraday rotation observations in terms of the magnetic field strength, leading to underestimated field strength when the two variables are assumed to be uncorrelated. The weak correlation between the total magnetic and cosmic ray energy densities near the midplane, r≈0.6r \approx0.6, and a similarly weak anti-correlation at a higher altitude, r≈−0.5r \approx-0.5, are inconsistent with the assumption of the local energy equipartition between magnetic fields and cosmic rays at the scales of the fluctuations produced by the Parker instability, of order 1 kpc and less.

The energy equipartition between cosmic rays and magnetic fields is often justified by arguing that cosmic rays can accumulate in the galactic disc until they achieve the energy density comparable with that of the magnetic field. Then their pressure can open up magnetic lines locally and they can be released from the galaxy, after which their pressure decreases down to a level controlled by the magnetic field. Such a self-regulation is argued to lead to their energy equipartition. Our model contains all the elements of these processes in their full nonlinear implementation, and yet no equipartition occurs. In particular, vertical magnetic fields that facilitate the vertical diffusive transport of the cosmic rays are produced by the magnetic buoyancy without any need for the cosmic ray pressure to affect the structure of the magnetic field (even though cosmic rays can enhance the magnetic buoyancy instability).

Vertical flows

The redistribution of the gas, magnetic field and cosmic rays in the unstable system involves systematic vertical flows. As discussed above, the vertical velocity averaged over a horizontal plane necessarily vanishes in the linear stage of the instability but, remarkably, systematic flows emerge in the transitional stage as a nonlinear effect. Figure 14 clarifies the balance between the inflow (uz<0u_z<0) and outflow (uz>0u_z>0) as the instability develops presenting, for 0≤z≤1.750\leq z\leq1.75 kpc (the picture at z<0z<0 is similar), separately, the vertical velocity averaged over those regions in the (x,y)(x,y)-planes where uz>0u_z>0 and uz<0u_z<0, denoted ⟨uz⟩+\langle u_z\rangle_+ and ⟨uz⟩−\langle u_z\rangle_-, respectively. In the linear stage, t=0.3t=0.3 Gyr, the outflows balance the inflows and their horizontally averaged vertical velocity ⟨uz⟩xy=⟨uz⟩−+⟨uz⟩+\langle u_z\rangle_{xy}=\langle u_z\rangle_-+\langle u_z\rangle_+ is negligible. As the nonlinear effects become stronger, at t=0.6t=0.6 Gyr, a systematic inflow develops around z=1z=1 kpc and outflow with the maximum outflow speed of 77 km s−1^{-1} is maintained at higher altitudes. At a later time, t=0.9t=0.9 Gyr, when the nonlinear effects are still stronger but the system continues evolving, an inflow dominates with −4-4 km s−1^{-1} at z=1.5z=1.5 kpc. In the advanced nonlinear stage at t=1.6t=1.6 Gyr, the r.m.s. inflow speed saturates at −8-8 km s−1^{-1} at z=1.5z=1.5 kpc.

Figure 14. The variation of the horizontally averaged vertical velocity component, uzu_z (solid lines) at z>0z > 0 in Model Sim6 at (a) t=0.3t = 0.3 Gyr, (b) 0.60.6 Gyr and (c) 1.61.6 Gyr. The horizontally averaged outflow (inflow) speed is shown dotted (dash-dotted) while the horizontally averaged velocity, their sum, is shown solid.

The fact that the Parker instability can drive systematic vertical flows can be of significance for the galactic evolution and regulation of star formation in the disc. Figure 15 shows the evolution of the mass flux at different heights. In the transitional phase (0.4<t<0.50.4<t<0.5 Gyr), the system develops a strong inflow near the midplane. During this period, the gas is redistributed into a thinner layer. As the system evolves into the nonlinear phase (t>0.5t>0.5 Gyr), the mass flux reduces, through decaying oscillations to become negligible in the late nonlinear stage. The outflow at higher altitudes discussed above involves dilute gas and does not transport any significant gas mass in this model.

Figure 15. The evolution of the mass flux in Model Sim6 at various heights above the midplane as specified in the legend.

Force balance

Figure 16 shows the horizontally averaged values of the thermal, magnetic, cosmic ray pressure gradients (the averaged magnetic tension force is negligible at all times) as well as the gravitational force at three different stages of the evolution. During the linear stage (Figure 16a) the horizontal averages of the magnetic field and cosmic rays vanish, and the average force balance is exactly the same as in the background distributions. From the transitional phase to the nonlinear state (Figure 16b,Figure 16c, respectively), the contributions of the magnetic and cosmic ray pressures decrease systematically as their scale heights increase, whereas the gradient of the thermal pressure increases because the gas scale height decreases (see Figure 11).

Figure 16. The vertical variation of the horizontally averaged forces normalised to the maximum magnitude of the gravitational force (dashed) at different stages of the development of the instability in Model Sim6: (a) t=0.3t = 0.3 Gyr (linear instability), (b) 0.50.5 Gyr (transitional; the magnetic field and cosmic rays have identical mean pressure distributions) and (c) 1.61.6 Gyr (nonlinear state; the mean pressure gradients of the magnetic field and cosmic rays are negligible). The gas pressure gradient (solid) is stronger than the pressure gradients of the magnetic field (dash-dotted) and cosmic rays (dotted), and this dominance is extreme in the nonlinear state.

To clarify the nature of the gas outflow prominent during the transitional stage of the instability, we show in Figure 17 the evolution of the magnitude of the horizontally averaged Lorentz force per unit volume at different heights. During the linear phase of the instability, the average magnetic force remains equal to that in the background (initial) state but it increases sharply during the transitional phase as significant vertical magnetic fields emerge. This increase is stronger at high altitudes (by a factor of 30). Since the gas density decreases with ∣z∣|z|, the increase in the force per unit mass is even stronger, and the magnetic force is clearly the driver of the systematic gas outflow in the transitional state (§ 4.2). As the instability saturates, the Lorentz force decays leaving for the thermal pressure gradient alone to balance the gravitational force.

Figure 17. Horizontally averaged r.m.s. Lorentz force of Model Sim6 (normalised by its midplane magnitude) at distance above the midplane as listed in the legend. J\boldsymbol{J} is electric current density.

In the initial state, the magnetic and cosmic ray pressures vary with zz exactly as the thermal pressure, so both βm\beta_{\rm m} and βcr\beta_{\rm cr} of Eq. (16) are independent of zz. Analytical analyses of the linear Parker instability usually rely on the assumption that both βm\beta_{\rm m} and βcr\beta_{\rm cr} remain independent of zz as the instability develops. Figure 18 shows the variation with zz of the horizontally averaged pressure ratios. In the linear stage, both ratios remain unchanged at ∣z∣≲1|z| \lesssim1 kpc, which justifies the assumption of the analytical studies, but increase significantly at ∣z∣≳1|z| \gtrsim1 kpc, especially βm\beta_{\rm m}. As the system evolves into the linear and nonlinear stages, both ratios are reduced to negligible values at z≲1z \lesssim1 kpc due to the increase in the thermal pressure and the reduction in magnetic and cosmic rays pressures near the midplane caused by the redistribution of the gas towards the midplane and escape of the magnetic field and cosmic rays to greater heights. As a result, the outer layers, ∣z∣≳1|z| \gtrsim1 kpc, are strongly dominated by the magnetic field and cosmic rays – and yet the magnetic force and the gradient of the cosmic pressure are both negligible in comparison with the thermal pressure gradient in the nonlinear state.

Figure 18. From Model Sim6 the horizontally averaged ratios to the thermal pressure, see Eq. (16), of (a) magnetic pressure βm\beta_{\mathrm{m}} and (b) cosmic ray pressure βcr\beta_{\mathrm{cr}}. Vertical distributions are shown at times specified in the legend, which represent the initial state, linear, transitional and nonlinear stages of the instability, respectively.

Sensitivity to parameters

The instability is sensitive to a wide range of parameters including the scale height of the unstable magnetic field, the ratios of the magnetic and cosmic ray pressures to the thermal pressure, βm\beta_{\rm m} and βcr\beta_{\rm cr}, and the form of the gravity profile g(z)g(z). In this section we discuss the effects of the key parameters and assumptions, with emphasis on the nonlinear states. The models Sim1–Sim4 have the gravity profile (17) while Sim5–Sim9 use g(z)g(z) from equation (18). To assess the role of the cosmic ray propagation governed by equation (4), we also consider a model where the only effect of the cosmic rays is to contribute to the total pressure, assuming that this contribution is equal to the magnetic pressure (thus, just doubling the magnetic pressure in such a model). Parameters of the models discussed below can be found in Table 1.

Gas scale height and nonthermal pressures

The purpose of Models Sim1–Sim4 is to explore the role of the gas (and magnetic field) scale heights and nonthermal pressures in the background state. The initial (background) state in Models Sim3 and Sim4 has a relatively large gas scale height h=1h = 1 kpc (and a correspondingly larger scale height of the background magnetic field) and lower magnetic and cosmic ray pressures than in the reference model Sim2 (with the same gravity profile) while Model Sim1 represents the case of the pure magnetic buoyancy instability (βcr=0\beta_{\rm cr} = 0). The adopted values of βm\beta_{\rm m}, βcr\beta_{\rm cr} and hh require a higher sound speed in these models (see equation (9)) of Rodrigues et al. [34]. In this sense, these models allow for the presence of the hot interstellar gas.

Figure 19 shows the evolution of the magnetic and cosmic ray energy densities for these two systems – it is useful to compare it with Figure 2. As expected, the instability is weaker (lower growth rate of the perturbations) in Models Sim3 and Sim4 than in Sim2. The system Sim3 (βm=βcr=0.5\beta_{\rm m} = \beta_{\rm cr} = 0.5) has a higher growth rate than Sim4 (βm=βcr=0.25\beta_{\rm m} = \beta_{\rm cr} = 0.25) and, in the transitional stage (1≲t≲21 \lesssim t \lesssim2 Gyr) loses the magnetic field and cosmic rays faster. The amount of the magnetic field and cosmic ray energies lost in the transitional stage is also slightly smaller in the system with weaker instability.

Four time-evolution plots for Models Sim 3 and Sim 4

Figure 19. As Figure 2 but for Models Sim3 (solid) and Sim4 (dotted) where the effective density scale height is h=1h = 1 kpc in the background state.

Gravity profile

The difference between the gravity profiles (17) and (18) is not strong and yet it can be of a physical significance since the gravity fields vary in magnitude and form between different galaxies and between different locations within a specific galaxy. In the linear stage, the development of the instability mostly depends on the system properties at ∣z∣≲h|z| \lesssim h, where h=0.3h = 0.3–0.50.5 kpc in most of the models considered here, so it might by reasonable to expect the modes of the linear instability to be broadly similar under both gravity profiles but it is useful to identify any subtler changes. In Models Sim2 and Sim6, which only differ in the gravity profiles, the growth rate of the instability is about 25 Gyr−125\ \mathrm{Gyr}^{-1}.

In Figure 20, we compare the 2D power spectra of the vertical velocity uzu_z and magnetic field bzb_z in the linear instability phase for Models Sim2 and Sim6; the higher value of hh in Model

Four two-dimensional power spectra panels

Figure 20. As Figure 4, but at t=0.3t = 0.3 Gyr (the linear stage of the instability) and for ∣z∣≤1.75\lvert z\rvert\le1.75 kpc for the perturbations as indicated in the colour bar labels. However, panels (a) and (c) here depict Model Sim2, while (b) and (d) again depict Model Sim6.

phase for Models Sim2 and Sim6; the higher value of hh in Model Sim2 reflects the somewhat weaker gravity. As also shown in Table 1, both models have similar wave number kyk_y of the most unstable mode although the spectrum in kyk_y around the maximum at ky≈3 kpc−1k_y \approx3\ \mathrm{kpc}^{-1} is wider in Model Sim2 (suggesting that this mode is less dominant). Furthermore, the mode structures in xx and zz differ significantly: (17) leads to significantly larger values of kxk_x and a weaker variation of the solution with zz. The nonlinear state can be expected to be more sensitive to the system properties at greater distances from the midplane where the two gravity profiles differ stronger (see also Li et al. 2017). However, this difference, within our simulation domain is not strong enough to affect, in particular, the vertical flows. In Models Sim2 and Sim6, the outflow speeds during the transitional stage at t=0.6t = 0.6 Gyr are ⟨uz⟩xy=7 km s−1\langle u_z\rangle_{xy} = 7\ \mathrm{km}\,\mathrm{s}^{-1} and 12 km s−112\ \mathrm{km}\,\mathrm{s}^{-1} at z=1.5z = 1.5 kpc. Similarly, the maximum magnitude of the downward velocity in the late nonlinear stage at t=1.25t = 1.25 Gyr is 5 km s−15\ \mathrm{km}\,\mathrm{s}^{-1} and 8 km s−18\ \mathrm{km}\,\mathrm{s}^{-1}, respectively.

The size of the computational domain

The size of the computational domain needs to be large enough to accommodate the most rapidly growing mode. Because of the periodic boundary conditions in xx and yy, only a discrete set of modes can be excited in the numerical model, with kx,yk_{x,y} being multiples of 2π/Lx,y2\pi/L_{x,y}, where LxL_x and LyL_y are the domain sizes along the xx- and yy-axes, respectively. In the models discussed above, the smallest admissible horizontal wave number is kmin⁡=2π/Lx=2π/Ly=12π kpc−1k_{\min} = 2\pi/L_x = 2\pi/L_y = \frac{1}{2}\pi\ \mathrm{kpc}^{-1}. The smallest wave number along the zz-axis (along which the boundary conditions are non-periodic) is 2π/Lz≈1.8 kpc−12\pi/L_z \approx1.8\ \mathrm{kpc}^{-1}.

In the nonlinear state, the solution is not a perfectly periodic function of (x,y)(x,y) within the domain, so that the constraint related to the periodic boundary conditions is much less important.

In the reference model Sim6, the most unstable linear mode has (kx,ky,kz)≈(1.6,3.1,1.8) kpc−1(k_x,k_y,k_z) \approx(1.6,3.1,1.8)\ \mathrm{kpc}^{-1}. This mode has the largest admissible wavelengths in the xx- and zz-directions and two complete wavelengths fit along the yy-direction. In order to confirm that the size of the domain does not affect excessively the parameters of the most unstable mode, we explored the linear phase of the instability using lower-resolution simulations with the same parameters as Model Sim6 but with bigger domains in xx and yy. The results are presented in Table 3. Figure 22 shows how the power spectra of the linear fluctuations depend on the horizontal size of the computational domain. The power spectra in xx (Figure 22a) do not vary much between the cases Lx=8L_x = 8 and 20 kpc and suggest that a wide range of modes with kx≲2 kpc−1k_x \lesssim2\ \mathrm{kpc}^{-1} are excited by the linear instability. The energy at short wavelengths, kx≳2 kpc−1k_x \gtrsim2\ \mathrm{kpc}^{-1}, is overestimated in the smaller domain with Lx=4L_x = 4 kpc but a broad maximum at kx≈1.6 kpc−1k_x \approx1.6\ \mathrm{kpc}^{-1} occurs for all values of LxL_x. The power spectra in yy (Figure 22b) have a prominent maximum at ky≈3 kpc−1k_y \approx3\ \mathrm{kpc}^{-1} in all cases, corresponding to a single most unstable mode. The power spectra in zz shown in Figure 22c have a strong maximum at the smallest available wave number kz≈3.6 kpc−1k_z \approx3.6\ \mathrm{kpc}^{-1} for the two smaller domain sizes (representing the large-scale vertical variations shown in Figure 6) but modes with kz≲3 kpc−1k_z \lesssim3\ \mathrm{kpc}^{-1} dominate when (Lx,Ly)=(20,24)(L_x, L_y) = (20, 24) kpc: the solution develops a nearly uniform component in zz. Remarkably, the horizontal size of the domain affects the vertical structure of the solution. The modes at large wave numbers in all three directions that have an approximately power-law spectrum apparently result from the instability since they are sensitive to the domain size but nonlinear effects, however weak at t≲0.4t \lesssim0.4 Gyr, may also contribute. We conclude that the domain size of the models presented in Table 1 is sufficient to capture the properties of the most rapidly growing modes.

LxL_x [kpc]Δx\Delta x [pc]2π/Lx2\pi/L_x [kpc−1^{-1}]kxk_x [kpc−1^{-1}]LyL_y [kpc]Δy\Delta y [pc]2π/Ly2\pi/L_y [kpc−1^{-1}]kyk_y [kpc−1^{-1}]LzL_z [kpc]Δz\Delta z [pc]2π/Lz2\pi/L_z [kpc−1^{-1}]kzk_z [kpc−1^{-1}]
8310.81.67130.92.83.5131.83.6
20520.31.624460.32.83.5131.83.6

Table 3. The wave numbers kxk_x, kyk_y and kzk_z of the mode that grows most rapidly in simulations with the same parameters as Model Sim6 but with the computational domain of the lengths LxL_x, LyL_y and LzL_z along the xx-, yy- and zz-axes, and the corresponding numerical resolutions Δx\Delta x, Δy\Delta y and Δz\Delta z, where the smallest admissible wave numbers are 2π/Lx2\pi/L_x, 2π/Ly2\pi/L_y and 2π/Lz2\pi/L_z and the largest are 2π/Δx2\pi/\Delta x, 2π/Δy2\pi/\Delta y and 2π/Δz2\pi/\Delta z, respectively. Each row presents a separate simulation.

Image not extracted

Figure 22. The normalised power spectra (a) E(kx)E(k_x), (b) E(ky)E(k_y) and (c) E(kz)E(k_z) of uzu_z (in kpc2^2 km2^2 s−2^{-2}) for models of with domain sizes listed in the legend. Otherwise parameters are as in Model Sim6. Spectra are time-averaged, sampled every 10 Myr for 0.25≤t≤0.300.25 \le t \le0.30 Gyr. The vertical lines of matching colour for each model mark the minimum (solid) and maximum (dotted) admissible wave numbers.

The role of dissipation

The only difference of Model Sim7 from the reference Model Sim6 is the higher (by a factor of 10) gas viscosity ν\nu and magnetic diffusivity η\eta (apart from a lower spatial resolution justified by the fact that stronger dissipation suppresses small-scale structures). The magnitudes of the transport coefficients in Model Sim7 are close to their turbulent values in galaxies. The instability growth rates in the two model are compared in Figure 21; the growth rate is hardly affected by the change. However, stronger dissipation leads to a noticeably larger magnetic field strength and cosmic ray energy density in the steady state; the effects of dissipation in the nonlinear state of the instability perhaps deserve further analysis.

Panels showing the evolution of magnetic and velocity quantities for Sim 6 and Sim 7

Figure 21. As Figure 2 but for Models Sim6 (solid) and Sim7 (dotted); both the viscosity and magnetic diffusivity are larger in the latter model.

Conclusions

The nonlinear, saturated state of the Parker instability is different from what might be expected. It is very different from its linear modes. The instability is driven by the gradient of the magnetic field energy density along the gravitational acceleration, and it might be expected that the saturated state would have a reduced gradient of the magnetic field strength. However, the models described here behave differently: not only is the magnetic field gradient reduced, but its strength also reduces throughout the system. The instability produces ubiquitous local vertical magnetic fields which facilitate rapid removal of the cosmic rays from the system by their rapid diffusion along the magnetic field. As a result, the statistically steady state is notable for a small scale height of the gas and very greater scale heights of the magnetic field and cosmic rays. The system settles into a state close to hydrostatic equilibrium where the gas is supported mainly by the thermal pressure. The regular pattern of magnetic loops and the associated variations in the gas and cosmic rays typical of the linear instability evolves into a rather chaotic nonlinear state where such regular features are hardly noticeable.

The thermal, magnetic and cosmic ray energy densities exhibit significant anti-correlation throughout the system reflecting the pressure balance in horizontal planes. Meanwhile, the magnetic and cosmic ray energy densities are weakly correlated closer to the midplane and anti-correlated further away from it. This behaviour is inconsistent with the energy equipartition between cosmic rays and magnetic field at the scales of the instability of order 1 kpc.

Apart from gas inflow associated with the reduction of the gas scale height, the instability drives a transient systematic outflow for about 0.3 Gyr with the mass flux of order 5×10−2 M⊙ kpc−2 Gyr−15 \times10^{-2}\ M_{\odot}\ \mathrm{kpc}^{-2}\ \mathrm{Gyr}^{-1} while it evolves through a weakly nonlinear regime. The flow is driven by the magnetic force and involves the vertical magnetic field.

We have also tested the simplifying assumptions used in various analytical (and some numerical) models of the Parker instability to conclude that very few – if any – of them are confirmed by the simulations. It appears that analytical analyses of the linear instability provide a reasonable qualitative picture but they should be used with caution when quantitative estimates are required. We find that that the instability growth rate scales more accurately with the scale height crossing time based on the effective sound speed (25) rather than the sound or Alfvén speeds.

The models discussed here neglect rotation. A more realistic model would include (differential) rotation and the dynamo action that supports the unstable background state. Their effects will be discussed elsewhere, and we note that rotation can affect the nonlinear state of the system.

DATA AVAILABILITY

The raw data for this work were obtained from numerical simulations using the open-source PENCIL-CODE available at https://github.com/pencil-code/pencil-code.git). The derived data used for the analysis presented in the paper are available on request from the corresponding author.

REFERENCES

A The background state and deviations

In order to implement the background distributions of the magnetic field and cosmic rays, that can feed the instability without being changed through its linear and nonlinear stages, we formally introduce source terms SAS_A and ScrS_{\rm cr} in equations (3) and (4) for the magnetic vector potential and cosmic rays, respectively:

∂A∂t=U×B+η∇×B+SA,(32)\frac{\partial A}{\partial t}=U\times B+\eta\nabla\times B+S_A, \tag*{(32)}
∂ϵcr∂t=−∇⋅(ϵcrU)−Pcr(∇⋅U)−∇⋅F+Scr.(33)\frac{\partial\epsilon_{\rm cr}}{\partial t}=-\nabla\cdot(\epsilon_{\rm cr}U)-P_{\rm cr}(\nabla\cdot U)-\nabla\cdot F+S_{\rm cr}. \tag*{(33)}

Since the background velocity vanishes, U0=0U_0=0, the background continuity equation is satisfied automatically whereas the background momentum equation reduces to the magnetohydrostatic equilibrium equation (13) where P0P_0 includes the background magnetic pressure Pm0P_{\rm m0}, such that ∇Pm0=(4π)−1(∇×B0)×B0\nabla P_{\rm m0}=(4\pi)^{-1}(\nabla\times B_0)\times B_0.

The background magnetic field B0B_0 corresponds to

SA=−η∇×B0.(34)S_A=-\eta\nabla\times B_0. \tag*{(34)}

In terms of the magnetic field source this corresponds to SB=∇×SAS_B=\nabla\times S_A or, equivalently, SB=η∇2B0S_B=\eta\nabla^2B_0. The desired magnetic field is aligned with the yy-axis and only varies with zz, so that SB=−y^η d2By0/dz2S_B=-\hat{y}\eta\,\mathrm{d}^2B_{y0}/\mathrm{d}z^2. The magnetic source term models the dynamo action responsible for maintaining a stationary, azimuthal field in the disc.

The source term for the energy density of cosmic rays is given by

Scr=∇⋅F0,(35)S_{\rm cr}=\nabla\cdot F_0, \tag*{(35)}

with the background cosmic ray energy density flux given by F0i=−κ0ij ∂ϵcr0/∂xjF_{0i}=-\kappa_{0ij}\,\partial\epsilon_{\rm cr0}/\partial x_j. The divergence of F0F_0 represents the Fickian diffusion of the background energy density, and the source maintains the background state against the diffusion. For the background states described in § 2.2, we have F0=−z^κ⊥ dϵcr0/dzF_0=-\hat{z}\kappa_\perp\,\mathrm{d}\epsilon_{\rm cr0}/\mathrm{d}z, and Scr=−κ⊥ d2ϵcr0/dz2S_{\rm cr}=-\kappa_\perp\,\mathrm{d}^2\epsilon_{\rm cr0}/\mathrm{d}z^2. The background state is then stationary, as required:

∂ϵcr0∂t=−∇⋅F0+Scr=0,(36)\frac{\partial\epsilon_{\rm cr0}}{\partial t}=-\nabla\cdot F_0+S_{\rm cr}=0, \tag*{(36)}
τ∂F0i∂t=−κ0ij∂ϵcr0∂xj−F0i=0.(37)\tau\frac{\partial F_{0i}}{\partial t}=-\kappa_{0ij}\frac{\partial\epsilon_{\rm cr0}}{\partial x_j}-F_{0i}=0. \tag*{(37)}

Note that, in the cosmic ray energy flux perturbation equation (10), the term −(κij∂ϵcr/∂xj−κ0ij∂ϵcr0/∂xj)-(\kappa_{ij}\partial\epsilon_{\rm cr}/\partial x_j-\kappa_{0ij}\partial\epsilon_{\rm cr0}/\partial x_j) represents the relevant term for the perturbation: here the background energy flux is subtracted from the total flux. In practice, there is no need to solve equations (A1) and (A2) for the background fields, which can be specified explicitly in their desired forms presented in § 2.2 in equations (8)–(10). The only purpose of the discussion above is to show that the implementation of the background state adopted here is physically and mathematically sound. The vertical distributions of the background gas density, magnetic field and cosmic rays

can be derived from the magnetohydrostatic equilibrium for a fixed gravitational profile. In terms of ceffc_{\rm eff} of equation (25), with ceff=constc_{\rm eff}=\mathrm{const} in this case, the total pressure is given by P0=Pth0+Pm0+Pcr0=ceff2ρ0P_0=P_{\rm th0}+P_{\rm m0}+P_{\rm cr0}=c_{\rm eff}^2\rho_0. For a chosen gravity profile and in the magnetohydrostatic equilibrium, we have

ceff2dρ0dz=−ρ0g,ln⁡ρ0=−1ceff2∫0zg(z) dz.(38)c_{\rm eff}^2\frac{\mathrm{d}\rho_0}{\mathrm{d}z}=-\rho_0 g,\qquad\ln\rho_0=-\frac{1}{c_{\rm eff}^2}\int_0^z g(z)\,\mathrm{d}z. \tag*{(38)}

For g(z)g(z) of equation (18) this gives

ln⁡ρ0ρ0(0)=a1ceff2(z1−z2+z12−a2z22a1z2).(39)\ln\frac{\rho_0}{\rho_0(0)}=\frac{a_1}{c_{\rm eff}^2}\left(z_1-\sqrt{z^2+z_1^2}-\frac{a_2z^2}{2a_1z_2}\right). \tag*{(39)}

The only non-vanishing component of the magnetic vector potential in the background state then follows from equation (20) as Ax0=∫0zB0 dzA_{x0}=\int_0^z B_0\,\mathrm{d}z.

References

  1. [13]Haugen N. E. L., Brandenburg A., 2004, Phys. Rev. E, 70, 036408
  2. [33]Pencil Code Collaboration et al., 2021, J. Open Source Software, 6, 2807
  3. [34]Rodrigues L. F. S., Sarson G. R., Shukurov A., Bushby P. J., Fletcher A., 2016, ApJ, 816, 2

Paper details

vlnc:W_ebt5nvpc
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 4

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents