betaFeedback

WorksPhysics and Astronomy

Anisotropic Magnetized Asteroseismic Waves

Abstract

We solve for waves in a polytropic, stratified plasmas with a spatially varying background magnetic field that points along a horizontal xx-direction, and with gravity that is directed along the vertical zz-direction. Force balance determines the magnitude of the background magnetic field, B02∼zn+1B_0^2 \sim z^{n+1}, where nn is the polytropic index. Using numerical and asymptotic methods, we deduce an explicit dispersion relation for fast pressure-driven waves: Ω2∼K(2m+n)[1+(1/MA)2(4−2γ+cos⁡2θ−3cos⁡4θ)/4]Ω^2 \sim K\left(2m+n\right) \left[1 + (1/M_\mathrm{A})^2 (4-2γ+\cos^2θ-3\cos^4θ)/4 \right]. Here, ΩΩ is the frequency, KK the wavenumber, θθ the angle the wave-vector makes with the background magnetic field, MAM_\mathrm{A} the Alfvénic Mach number, and mm an integer representing the eigenstate. We discuss roles of such an explicit formula in asteroseismology.

Introduction

The strengths of magnetic fields buried below the surface of stars are not known, though they are vital for improved understanding of the magnetic behaviour of stars. This challenge has impeded progress in understanding stellar magnetism and the evolution of magnetized stellar interiors. To estimate the magnetic field strength, linear asteroseismology is a promising technique [2]. The key idea is to calculate the dispersion relations of surface waves taking into account the presence of subsurface magnetic fields. Then inversion techniques are used, leveraging the observed power spectrum of oscillations, to infer the subsurface magnetic field. Thus, simple analytical dispersion relations are insightful, but obtaining such relations remains a challenge. This challenge also impacts the development of nonlinear asteroseismology [17, 43, 44], which requires linear dispersion relations to evaluate mode resonances. Consequently, the progress in magnetoseismology has been slow.

Observational studies report travel-time perturbations of acoustic waves to be a critical signature of strong magnetic fields in the stellar interior [31]

Ilonidis et al. [19]}. Numerical simulations of asteroseismic waves also suggest the possibility of detecting subsurface fields, before they emerge on the surface [33, 32, 34, 35, 14, 13]. To bolster such findings, a thorough understanding of the impact of magnetic fields on asteroseismic waves is essential [28, 1, 37, 10, 7, 11, 41].

Waves in an unmagnetized polytropic atmosphere were exactly solved analytically by Lamb [20] who derived the relation

Ω22K−(n+1)(n+1−γn)K2γ2Ω2=m+n2,(1)\frac{\Omega^2}{2K}-\frac{(n+1)(n+1-\gamma n)K}{2\gamma^2\Omega^2}=m+\frac{n}{2}, \tag*{(1)}

where Ω\Omega is the frequency, KK the wavenumber, nn the polytropic index, γ\gamma the adiabatic index, and mm the eigenstate index, with m=0,1,2,…m=0,1,2,\ldots. This advancement led to a series of newer and significant understanding of hydrodynamic waves. Under fast-wave approximation, Ω2/K≫1\Omega^2/K \gg1, the leading-order dispersion relation becomes

Ω2∼K(2m+n).(2)\Omega^2 \sim K(2m+n). \tag*{(2)}

A similar closed-form analytical expression for waves in a magnetized polytrope, as of yet, is unknown. Earlier attempts (e.g., Gough & Thompson [16]; Spruit & Bogdan [36]; Cally & Bogdan [9]; Bogdan & Cally [4]; Cally & Bogdan [8] have ended with deriving, analytically, approximate integral expression in certain perturbative limits and then solving them numerically. Hence, they leave out the critical step of obtaining a straightforward analytical understanding of the effect of magnetic fields on linear asteroseismic waves. For example, Spruit & Bogdan [36] consider a plane-parallel polytropic atmosphere threaded by a vertical magnetic field that is uniform throughout the whole domain. They perturbatively derive integrals for changes in wave frequency and then solve the integrals numerically. Cally & Bogdan [9] pursue the model of Spruit & Bogdan [36], but with a different perturbative approach. The effect of horizontal fields on waves is studied by several authors (e.g., [28, 1, 37, 10, 7, 11, 41]), but in simple and unrealistic isothermal atmosphere, which has very different properties of waves than in a more realistic polytropic atmosphere. Notably, Gough & Thompson [16] treat a global problem (in spherical coordinates); there, the computation of eigenfrequencies requires evaluation of integral expressions using numerical methods [e.g., Eqs. (4.11)–(4.13) in Gough & Thompson [16]]. A closed-form analytical formula is currently unavailable. The limitations of purely numerical approach and lack of a closed-form expression were succinctly expressed by Bogdan & Cally [4]:

“Ideally, we would wish to proceed by writing down an equation analogous to Lamb’s formula for the magnetized polytrope. Unfortunately, this approach is not feasible and for the most part one must instead be content with a numerically derived visual comparison of how the allowed oscillation frequencies depend upon the choice of the horizontal wavenumber kk.”

Analytic dispersion relations are also essential for developing wave turbulence theory in the presence of both gravity and magnetic fields. In wave turbulence theory, calculations of mode resonances require simple, explicit dispersion relations that accurately capture the magnetic effect on observed linear waves. We note that the Lamb’s dispersion relation (2), Ω∼K\Omega\sim\sqrt{K}, is similar to that of surface gravity waves in oceans (Hasselmann [18]). However, there is a critical difference: The surface gravity waves do not couple via three-wave resonance, thus requiring a weaker four-wave coupling (Nazarenko & Lukaschuk [26]). The Lamb waves, on the other hand, can couple via three-wave resonance, because there are infinitely many such waves (eigen-

states) at a given wavenumber, unlike only one pair of surface gravity waves at a given wavenumber. Thus, the infinitely many Lamb waves at a given wavenumber have distinct wave frequencies, which allow the sum of three frequencies at three wavenumbers to become null. While a wave turbulence theory for surface gravity waves has been well established, it is yet to be developed for the asteroseismic waves, whose dispersion relation in fully analytic form is a basic requirement for such a theory. The value that an analytic dispersion relation offers in resonant-coupling theory cannot be overstated when magnetic fields make the wave dispersion relation anisotropic and complicated. Motivated by these reasons, we seek here an accurate and simple formulae for the effect of magnetic fields on the Lamb waves.

Introducing magnetic fields, aligned orthogonal to a vertical gravity in Figure 1, we find, as did Bogdan & Cally [4], that the linearized magnetohydrodynamic (MHD) equations are too cumbersome to obtain a closed-form expression for the dispersion relation, even with the fast-wave approximation, Ω2/K≫1\Omega^2/K \gg1. Here, we overcome this difficulty by using both numerical simulations and extensive use of Mathematica, followed by a variant of the Jeffereys–Wentzel–Kramers–Brillouin (JWKB) approximation we devise to deduce

An inhomogeneous magnetic field $B_0(z)$ oriented orthogonal to a constant vertical gravity $\mathbf{g}$

Figure 1. An inhomogeneous magnetic field B0(z)B_0(z), oriented orthogonal to a constant vertical gravity g\mathbf{g}, is considered where the wave is allowed to propagate in an arbitrary direction, shown with K\mathbf{K}, making an angle θ\theta with B0B_0. The gradient in the colormap of the box schematically represents increasing functions with depth zz of the magnetic field, fluid density, pressure, and sound speed.

Ω2∼K (2m+n)[1+ϵ2(4−2γ+cos⁡2θ−3cos⁡4θ)4](3)\Omega^2 \sim K\,(2m+n)\left[1+\frac{\epsilon^2(4-2\gamma+\cos^2\theta-3\cos^4\theta)}{4}\right] \tag*{(3)}

in the limit Ω2/K≫1\Omega^2/K \gg1, with m=0,1,2,…m=0,1,2,\ldots. The parameter ϵ\epsilon is the inverse of the Alfvénic Mach number, and θ\theta is the angle the wave vector makes with the background magnetic field. Equation (3) is the principal result of this paper.

This paper is organized in the following way. In § 2, we describe our model and present the linearized compressible MHD equations. Such equations are then numerically solved in § 3. To obtain analytical understanding of the numerical results, the linearized equations are reduced to a wave equation in § 4. The normal-form wave equation is then perturbatively solved using a variant of the JWKB theory we construct; analytical understanding is gained in § 5. With astrophysical implications and roles of our results, we conclude in § 6.

System setup and linearized perturbation equations

To study waves in a magnetized, stratified medium, we consider the ideal three-dimensional compressible magnetohydrodynamic (MHD) equations (Chandrasekhar (1961)

∂tρ+∇⋅(ρU)=0,(4)\partial_t \rho+ \nabla\cdot(\rho\mathbf{U}) = 0, \tag*{(4)}
ρ[∂tU+(U⋅∇)U]=−∇P+ρg+J×B,(5)\rho\left[\partial_t \mathbf{U} + (\mathbf{U} \cdot\nabla)\mathbf{U}\right] = -\nabla P + \rho\mathbf{g} + \mathbf{J} \times\mathbf{B}, \tag*{(5)}
∂tB=∇×(U×B),(6)\partial_t \mathbf{B} = \nabla\times(\mathbf{U} \times\mathbf{B}), \tag*{(6)}
DρDt=c2DPDt,(7)\frac{D\rho}{Dt} = c^2 \frac{DP}{Dt}, \tag*{(7)}

where ρ\rho, U\mathbf{U}, PP, and B\mathbf{B} represent the density, the velocity, the pressure, and the magnetic field, respectively. The current density is J=∇×B/μ0\mathbf{J} = \nabla\times\mathbf{B}/\mu_0, where μ0\mu_0 represents the magnetic permeability of the vacuum. The magnetic field is additionally constrained to be divergence-less

∇⋅B=0.(8)\nabla\cdot\mathbf{B} = 0. \tag*{(8)}

We consider a local Cartesian domain where the equilibrium density ρ0\rho_0 and pressure P0P_0 satisfy the polytropic relation

P0∼ρ01+1/n,(9)P_0 \sim\rho_0^{1+1/n}, \tag*{(9)}

where nn is the polytropic index of the gaseous atmosphere. In an unmagnetized atmosphere, the force balance gives ∂zP0(z)=ρ0(z)g\partial_z P_0(z) = \rho_0(z)g, where g=ge^z\mathbf{g} = g\hat{\mathbf{e}}_z, with gg as the constant acceleration due to gravity along the vertical zz-axis, as depicted in Figure 1. The force balance requires

ρ0(z)∼zn;P0(z)∼zn+1.(10)\rho_0(z) \sim z^n; \quad P_0(z) \sim z^{n+1}. \tag*{(10)}

Background magnetic field and sound speed

We introduce an inhomogeneous background magnetic field, B=B0(z)e^x\mathbf{B} = B_0(z)\hat{\mathbf{e}}_x. The force-balance relation with the Lorentz force, (∇×B/μ0)×B(\nabla\times\mathbf{B}/\mu_0) \times\mathbf{B}, then becomes

∂zP0=ρ0g−12μ0∂zB02.(11)\partial_z P_0 = \rho_0 g - \frac{1}{2\mu_0}\partial_z B_0^2. \tag*{(11)}

The solution

B02(z)∼zn+1(12)B_0^2(z) \sim z^{n+1} \tag*{(12)}

also satisfies Eq. (7). The plasma β\beta is then a constant throughout the domain

β≡PgasPmagnetic=2μ0P0B02.(13)\beta\equiv\frac{P_{\mathrm{gas}}}{P_{\mathrm{magnetic}}} = \frac{2\mu_0P_0}{B_0^2}. \tag*{(13)}

The sound speed (cc) may now be deduced from Eq. (8), using P0=c2ρ0/γP_0 = c^2\rho_0/\gamma, where γ\gamma is the adiabatic index of the gas

∂zc2=−nc2z+γg1+β−1.(14)\partial_z c^2 = -\frac{nc^2}{z} + \frac{\gamma g}{1+\beta^{-1}}. \tag*{(14)}

The solution to Eq. (11) is c2=c02zc^2=c_0^2z, where c02c_0^2 is a constant given by

c02=γg(n+1)(1+β−1).(15)c_0^2 = \frac{\gamma g}{(n+1)(1+\beta^{-1})}. \tag*{(15)}

It is useful to express β−1\beta^{-1} in terms of the Alfvénic Mach number MAM_A, which is the ratio of the sound speed to the Alfvén speed,

β−1=B02μ0ρ0c2⋅ρ0γρ0⋅γ2=γ2MA2,(16)\beta^{-1} = \frac{B_0^2}{\mu_0\rho_0c^2}\cdot\frac{\rho_0}{\gamma\rho_0}\cdot\frac{\gamma}{2} = \frac{\gamma}{2M_A^2}, \tag*{(16)}

using which the sound speed becomes

c2=γgz(n+1)(1+γMA−2/2).(17)c^2 = \frac{\gamma gz}{(n+1)(1+\gamma M_A^{-2}/2)}. \tag*{(17)}

Linearized perturbation equations

We now linearize the MHD equations around the background profiles [ρ0,0,B0,P0][\rho_0, 0, \mathbf{B}_0, P_0], introduced in Eqs. (7) and (9). Such a linearization yields evolution equations for infinitesimal perturbations [ρ~,u~,b~,p~][\widetilde{\rho}, \widetilde{\mathbf{u}}, \widetilde{\mathbf{b}}, \widetilde{p}] as

∂tρ~=−(u~⋅∇)ρ0−ρ0(∇⋅u~),(18)\partial_t\widetilde{\rho} = -(\widetilde{\mathbf{u}}\cdot\nabla)\rho_0-\rho_0(\nabla\cdot\widetilde{\mathbf{u}}), \tag*{(18)}
∂tu~x=−∂xp~ρ0+b~z∂zB0μ0ρ0,(19)\partial_t\widetilde{u}_x = -\frac{\partial_x\widetilde{p}}{\rho_0}+\frac{\widetilde{b}_z\partial_zB_0}{\mu_0\rho_0}, \tag*{(19)}
∂tu~y=−∂yp~ρ0+B0(∂xb~y−∂yb~x)μ0ρ0,(20)\partial_t\widetilde{u}_y = -\frac{\partial_y\widetilde{p}}{\rho_0}+\frac{B_0(\partial_x\widetilde{b}_y-\partial_y\widetilde{b}_x)}{\mu_0\rho_0}, \tag*{(20)}
∂tu~z=−∂zp~ρ0+ρ~gρ0+B0(∂xb~z−∂zb~x)−b~x∂zB0μ0ρ0,(21)\partial_t\widetilde{u}_z = -\frac{\partial_z\widetilde{p}}{\rho_0}+\frac{\widetilde{\rho}g}{\rho_0}+\frac{B_0(\partial_x\widetilde{b}_z-\partial_z\widetilde{b}_x)-\widetilde{b}_x\partial_zB_0}{\mu_0\rho_0}, \tag*{(21)}
∂tb~x=−B0(∂yu~y+∂zu~z)−u~z∂zB0,(22)\partial_t\widetilde{b}_x = -B_0(\partial_y\widetilde{u}_y+\partial_z\widetilde{u}_z)-\widetilde{u}_z\partial_zB_0, \tag*{(22)}
∂tb~y=−B0∂xu~y,(23)\partial_t\widetilde{b}_y = -B_0\partial_x\widetilde{u}_y, \tag*{(23)}
∂tb~z=−B0∂xu~z,(24)\partial_t\widetilde{b}_z = -B_0\partial_x\widetilde{u}_z, \tag*{(24)}
∂tp~=−(u~⋅∇)P0−c2ρ0(∇⋅u~).(25)\partial_t\widetilde{p} = -(\widetilde{\mathbf{u}}\cdot\nabla)P_0-c^2\rho_0(\nabla\cdot\widetilde{\mathbf{u}}). \tag*{(25)}

We shall use

χ~=∇⋅u~(26)\tilde{\chi}=\nabla\cdot\tilde{\boldsymbol{u}} \tag*{(26)}

in the rest of the article, where helpful.

Equations (15a)–(15h) can be simplified to derive a set of fewer (closed) equations

∂t2u~x=c2∂xχ~+g∂xu~z,(27)\partial_t^2\tilde{u}_x=c^2\partial_x\tilde{\chi}+g\partial_x\tilde{u}_z, \tag*{(27)}
∂t2u~y=c2∂yχ~+g∂yu~z+c2MA2(∂xxu~y+∂yχ~−∂xyu~x),(28)\partial_t^2\tilde{u}_y=c^2\partial_y\tilde{\chi}+g\partial_y\tilde{u}_z+\frac{c^2}{M_{\mathrm{A}}^2}\left(\partial_{xx}\tilde{u}_y+\partial_y\tilde{\chi}-\partial_{xy}\tilde{u}_x\right), \tag*{(28)}
∂t2u~z=c2∂zχ~−g∂xu~x[1+γMA2(1+γMA−2/2)]−g∂yu~y+γ(1+MA−2)gχ~1+γMA−2/2+c2MA2(∂xxu~z+∂zχ~−∂xzu~x).(29)\begin{aligned} \partial_t^2\tilde{u}_z={}&c^2\partial_z\tilde{\chi}-g\partial_x\tilde{u}_x\left[1+\frac{\gamma}{M_{\mathrm{A}}^2(1+\gamma M_{\mathrm{A}}^{-2}/2)}\right]-g\partial_y\tilde{u}_y\\ &+\frac{\gamma(1+M_{\mathrm{A}}^{-2})g\tilde{\chi}}{1+\gamma M_{\mathrm{A}}^{-2}/2}+\frac{c^2}{M_{\mathrm{A}}^2}\left(\partial_{xx}\tilde{u}_z+\partial_z\tilde{\chi}-\partial_{xz}\tilde{u}_x\right). \tag*{(29)} \end{aligned}

We note that the appearance of MA2M_{\mathrm{A}}^2 in Eq. (17c) in certain terms may seem non-trivial at first sight; however, upon inspection, we understand them as terms emerging from the effect of the Lorentz force on the background states, via terms like ∂zc(z)(z)\partial_z c^{(z)}(z) and (∂zP0)/ρn(\partial_z P_0)/\rho_n, while processing Eqs. (15a)–(15h). Admittedly, Eqs. (17a)–(17c) are somewhat challenging to proceed with clarity. Thus, we now non-dimensionalize the equations to make them reasonably transparent.

Non-dimensionalized linear equations

We define LL as the characteristic length scale over which the sound speed varies (and, for that matter, pressure, density, and temperature also vary) appreciably. Then we find that the characteristic sound speed in an unmagnetized polytrope is, using Eq. (17), cL=γgL/(n+1)c_L=\sqrt{\gamma gL/(n+1)}. Using LL and L/cLL/c_L as the dimensional length and time units, we non-dimensionalize all variables henceforth, starting with the sound speed

C2=c2cL2=Z1+γMA−2/2,(30)C^2=\frac{c^2}{c_L^2}=\frac{Z}{1+\gamma M_{\mathrm{A}}^{-2}/2}, \tag*{(30)}

where the lowercase dimensional variables (cc and zz) are cast as uppercase non-dimensional variables (CC and ZZ). Thus we replace all the dimensional variables in Eqs. (17a)–(17c) using

(x,y,z)=(LX,LY,LZ),(31)(x,y,z)=(LX,LY,LZ), \tag*{(31)}
c2=cL2Z1+γMA−2/2,(32)c^2=c_L^2\frac{Z}{1+\gamma M_{\mathrm{A}}^{-2}/2}, \tag*{(32)}
∂t≡cLL∂T,(33)\partial_t\equiv\frac{c_L}{L}\partial_T, \tag*{(33)}

where the uppercase characters XX, YY, and TT represent non-dimensional variables.

We analyze perturbations by Fourier-transforming in the (x,y)(x,y)-plane, viz.,

u~x(Z)=∫dKXdKYdΩ u^xexp⁡[i(KXX+KYY+ΩT)],(34)\tilde{u}_x(Z)=\int dK_XdK_Yd\Omega\,\hat{u}_x\exp\left[i(K_XX+K_YY+\Omega T)\right], \tag*{(34)}

where the uppercase characters represent non-dimensional quantities, e.g., K≡(KX,KY)\boldsymbol{K}\equiv(K_X,K_Y) is the non-dimensional wavevector in the (x,y)(x,y)-plane, and Ω\Omega is the nondimensional frequency. Fourier analyzing Eqs. (17a)–(17c) and representing MA−1M_{\mathrm{A}}^{-1} by ϵ\epsilon henceforth, we write

[−Ω2(1+γϵ2/2)iKXZ]u^x+[−(n+1)(1+γϵ2/2)γZ]u^z=χ^,(35)\left[\frac{-\Omega^2(1+\gamma\epsilon^2/2)}{iK_XZ}\right]\hat{u}_x+\left[\frac{-(n+1)(1+\gamma\epsilon^2/2)}{\gamma Z}\right]\hat{u}_z=\hat{\chi}, \tag*{(35)}
[−iϵ2KX]u^x+[−Ω2(1+γϵ2/2)iKYZ+ϵ2KX2iKY]u^y+[−(n+1)(1+γϵ2/2)γZ]u^z=χ^(1+ϵ2),(36)\left[-i\epsilon^2K_X\right]\hat{u}_x+\left[\frac{-\Omega^2(1+\gamma\epsilon^2/2)}{iK_YZ}+\frac{\epsilon^2K_X^2}{iK_Y}\right]\hat{u}_y+\left[\frac{-(n+1)(1+\gamma\epsilon^2/2)}{\gamma Z}\right]\hat{u}_z=\hat{\chi}(1+\epsilon^2), \tag*{(36)}
[iKX{1+γϵ2/2γ+ϵ2(1+Z∂Zn+1)}]u^x+[iKY(1+γϵ2/2)γ]u^y+[−Ω2(1+γϵ2/2)+ϵ2ZKX2n+1]u^z=(χ^+Z∂Zχ^n+1)(1+ϵ2).(37)\begin{aligned} \left[iK_X\left\{\frac{1+\gamma\epsilon^2/2}{\gamma}+\epsilon^2\left(1+\frac{Z\partial_Z}{n+1}\right)\right\}\right]\hat{u}_x +\left[\frac{iK_Y(1+\gamma\epsilon^2/2)}{\gamma}\right]\hat{u}_y +\left[\frac{-\Omega^2(1+\gamma\epsilon^2/2)+\epsilon^2ZK_X^2}{n+1}\right]\hat{u}_z\\ =\left(\hat{\chi}+\frac{Z\partial_Z\hat{\chi}}{n+1}\right)(1+\epsilon^2). \tag*{(37)} \end{aligned}

Equations (21a)–(21c) are general, without any approximation, and host all three families of MHD waves: the fast and slow magnetoacoustic waves, and the Alfvén waves. The Alfvén wave dispersion relation is reproduced by substituting KY=0K_Y=0 and KX=KK_X=K in Eq. (21b), which renders it to: either u^y=0\hat{u}_y=0 (the trivial solution), or Ω2=ϵ2K2/(1+γϵ2)\Omega^2=\epsilon^2K^2/(1+\gamma\epsilon^2). This is indeed the dispersion relation of the Alfvén waves, which is confirmed by writing Ω2=K2vA2\Omega^2=K^2v_{\mathrm{A}}^2, where vA2=B02/(μ0ρ0)v_{\mathrm{A}}^2=B_0^2/(\mu_0\rho_0) is the squared Alfvén speed. This squared Alfvén speed is determined from Eq. (16) to be γP0/(ρ0MA2)\gamma P_0/(\rho_0M_{\mathrm{A}}^2). Recognizing that γP0/ρ0\gamma P_0/\rho_0 is the squared sound speed, given in Eq. (30) [derived from Eq. (14)], the squared Alfvén speed is vA2=ϵ2Z/(1+γϵ2)v_{\mathrm{A}}^2=\epsilon^2Z/(1+\gamma\epsilon^2).

Exact numerical solution

We obtain fully converged numerical solutions to Eqs. (21a)–(21c) by employing the spectral “Dedalus” framework Burns et al. [6]. Referring to the Dedalus methods paper Burns et al. [6] for more details, we briefly outline the numerical procedures employed in Dedalus for eigenvalue problems. At each horizontal wavenumber (KX,KY)(K_X,K_Y), the state variables—the three components of velocity—are expanded in the Chebyshev polynomials along the inhomogeneous zz-axis. Because of the background inhomogeneity, different Chebyshev coefficients couple, creating a dense linear operator. Sparsification is provided by a change of basis from the Chebyshev polynomials of the first kind to those of the second kind. To impose boundary conditions and keep the matrix sparse, Dirichlet preconditioning is applied. Efficient solution of the resulting matrices is then found by passing the matrices L\mathcal{L} and M\mathcal{M} in the eigenvalue (σ\sigma) problem, LX=σMX\mathcal{L}\mathbf{X}=\sigma\mathcal{M}\mathbf{X}, to the “scipy” linear algebra packages. For a given spectral resolution along the inhomogeneous zz-axis, we solve for all the eigenvalues of the matrices. Such a non-targeted, general solution produces a complete eigenspectrum and eigenmodes, corresponding to all families of linear MHD waves: the (fast) pressure-driven, (slow) gravity-driven modes, and the Alfvén modes. We focus on the high-frequency modes to assess the effect of magnetic fields on such fast pressure-driven waves.

For the boundary condition, at the lower boundary z=Lzz=L_z, we require u~z=0\widetilde{u}_z=0. At the upper boundary, z=0z=0, where the atmosphere ceases, we enforce, following Lamb [20],

DpDt=0,(38)\frac{Dp}{Dt}=0, \tag*{(38)}

which implies

c2DρDt=−c2ρ0χ~=0=zn+1χ~.(39)c^2\frac{D\rho}{Dt}=-c^2\rho_0\widetilde{\chi}=0=z^{n+1}\widetilde{\chi}. \tag*{(39)}

We first validate that, in the absence of the magnetic field, our solver successfully reproduces two branches of Lamb’s exact analytical dispersion relation [Eq. (1)]. We identify the pressure-driven modes by comparing their eigenfrequencies with those predicted by Lamb’s solution. We then impose a very weak background magnetic field and systematically increase its strength, and track changes to the eigenfrequencies. A case of such magnetic modification of pressure-driven waves is shown in Figure 2. Although, in Figure 2, we display two surface plots of Ω2(KX,KY)\Omega^2(K_X,K_Y)—one for the hydrodynamic and other for the magnetized polytrope—the two plots are visually indistinguishable. The difference between the two surface plots is shown in Figure 3. For KY=0K_Y=0, ΔΩ2\Delta\Omega^2 is negative, and for KX=0K_X=0, ΔΩ2\Delta\Omega^2 is positive and relatively large. We also note a minor decrease in positive value of ΔΩ2\Delta\Omega^2 in going from KX≈0K_X\approx0 to KX=0K_X=0.

Surface plot of $\Omega^2(K_X,K_Y)$ for hydrodynamic and magnetized polytropes

Figure 2. Ω2(KX,KY)\Omega^2(K_X,K_Y) for hydrodynamic and magnetized polytropes. The two plotted surfaces are visually indifferentiable because the difference (ΔΩ2\Delta\Omega^2) between them is much smaller than Ω2\Omega^2; see Figure 3. The parameters used are ϵ=0.1\epsilon=0.1, m=20m=20, n=2.5n=2.5, and γ=5/3\gamma=5/3. For variations in these parameters, the surface plot of Ω2(KX,KY)\Omega^2(K_X,K_Y) remains the same qualitatively.

Surface plot of $\Delta\Omega^2$ for a magnetized polytrope

Figure 3. ΔΩ2=Ω2−Ωhydro2\Delta\Omega^2=\Omega^2-\Omega_{\mathrm{hydro}}^2 is plotted for a magnetized polytrope with ϵ=0.1\epsilon=0.1, m=20m=20, n=2.5n=2.5, and γ=5/3\gamma=5/3. This surface plot shows the difference between the two surface plots in Figure 2.

It turns out that ΔΩ2\Delta\Omega^2 is related to the hydrodynamic squared-frequency Ωhydro2\Omega_{\mathrm{hydro}}^2; and ΔΩ2/Ωhydro2\Delta\Omega^2/\Omega_{\mathrm{hydro}}^2 is almost entirely independent of the wavevector magnitude (Figure 4). Only angular dependence is observed.

Relative difference of squared-frequency depends on the wavevector propagation angle

Figure 4. Relative difference of squared-frequency depends on the wavevector propagation angle (spanned by the black circle), but the relative difference is insensitive to the wavevector magnitude. The parameters chosen are ϵ=0.1\epsilon= 0.1, m=20m = 20, n=2.5n = 2.5, and γ=5/3\gamma= 5/3.

The extremely low wavenumbers in Figure 4 correspond to very large scale waves that cannot be captured in a finite box in numerical calculation. To capture lower wavenumbers, we significantly extend the domain size along the vertical zz-axis, which allow us to obtain fully converged numerical results for other wavenumbers.

Reduction to wave equation

To analytically determine the dispersion relation, we solve the set of equations (35)–(37) perturbatively in the limit of a weak magnetic field, i.e., ϵ≪1\epsilon\ll1. By setting ϵ=0\epsilon= 0, we recover the Lamb’s equations for the unmagnetized polytrope [20]. The Lamb’s equations reduce to a second-order differential equation for χ^\hat{\chi}. With the same goal, we proceed in the following manner. We rewrite Eqs. (21a)–(21c) as

(M11M12M13M21M22M23M31M32M33)(u^xu^yu^z)=(hx(χ^)hy(χ^)hz(χ^,∂Zχ^))(40)\begin{pmatrix} M_{11} & M_{12} & M_{13} \\ M_{21} & M_{22} & M_{23} \\ M_{31} & M_{32} & M_{33} \end{pmatrix} \begin{pmatrix} \hat{u}_x \\ \hat{u}_y \\ \hat{u}_z \end{pmatrix} = \begin{pmatrix} h_x(\hat{\chi}) \\ h_y(\hat{\chi}) \\ h_z(\hat{\chi},\partial_Z\hat{\chi}) \tag*{(40)} \end{pmatrix}

where the matrix elements MijM_{ij} are independent of ZZ-derivatives, and are functions of KX,KY,Ω,n,γ,ϵK_X,K_Y,\Omega,n,\gamma,\epsilon, and ZZ only (for their complete expressions, see Appendix A). In arriving at Eq. (40), we have replaced ∂Zu^x\partial_Z\hat{u}_x in the first term on the left-hand side of Eq. (21c) with its exact expression obtained by differentiating Eq. (21a) with respect to ZZ. We then substitute ∂Z\partial_Z operation in ∂Zu^z\partial_Z\hat{u}_z by writing it as χ^−iKXu^x−iKYu^y\hat{\chi}-iK_X\hat{u}_x-iK_Y\hat{u}_y. Such a process removes ∂Z\partial_Z operation from the matrix MM. The functions hνh_\nu are linear in χ^\hat{\chi} and ∂Zχ^\partial_Z\hat{\chi}.

Straightforward inversion of the matrix MM expresses all components of the velocity in terms of χ^\hat{\chi} and ∂Zχ^\partial_Z\hat{\chi}:

u^ν=fν(Z)χ^+gν(Z)∂Zχ^,(41)\hat{u}_\nu=f_\nu(Z)\hat{\chi}+g_\nu(Z)\partial_Z\hat{\chi}, \tag*{(41)}

where ν\nu can be either x,yx,y, or zz. We recognize that fνf_\nu and gνg_\nu are functions of ZZ, but do not involve ∂Z\partial_Z.

The three velocity components of Eq. (41) can now be subsumed into a single second-order differential equation for χ^\hat{\chi}:

∂Z2χ^+P(Z,ϵ)∂Zχ^+R(Z,ϵ)χ^=0,(42)\partial_Z^2\hat{\chi}+P(Z,\epsilon)\partial_Z\hat{\chi}+R(Z,\epsilon)\hat{\chi}=0, \tag*{(42)}

which can be recast into the normal form of the second-order differential equation by changing variable as

χ^(Z)=ψ^(Z)exp⁡[−12∫ZdZ P(Z)],(43)\hat{\chi}(Z)=\hat{\psi}(Z)\exp\left[-\frac{1}{2}\int_Z dZ\,P(Z)\right], \tag*{(43)}

which reduces Eq. (42) to the wave equation

∂Z2ψ^+Γ2(Z,ϵ,δ)ψ^=0.(44)\partial_Z^2\hat{\psi}+\Gamma^2(Z,\epsilon,\delta)\hat{\psi}=0. \tag*{(44)}

In (44), we use the explicit notation Γ2(Z,ϵ,δ)\Gamma^2(Z,\epsilon,\delta) to remind us that the two small parameters ϵ\epsilon and δ\delta are implicit in Γ\Gamma, where δ=K/Ω2\delta=K/\Omega^2, with K=KX2+KY2K=\sqrt{K_X^2+K_Y^2} representing the magnitude of the wavevector and Ω\Omega representing the eigenfrequency.

The procedure outlined above appears straightforward. However, the analytical manipulations in arriving at Eq. (44)—a magnetized version of Lamb’s equation—require laborious and careful calculations, as Γ2(Z,ϵ,δ)\Gamma^2(Z,\epsilon,\delta) alone conceals an expression of exhaustive length—tens of pages of this article. In the absence of the magnetic field, the expression for Γ2(Z,ϵ=0,δ)=Γ02(Z,δ)\Gamma^2(Z,\epsilon=0,\delta)=\Gamma_0^2(Z,\delta) is beautifully short, Γ02(Z,δ)=K2(Z−α)(β−Z)/Z2\Gamma_0^2(Z,\delta)=K^2(Z-\alpha)(\beta-Z)/Z^2, where α\alpha and β\beta are the two turning points—two zeros of Γ02(Z,δ)−thatdependon\Gamma_0^2(Z,\delta)-that depend on\delta$, the property of the eigenfrequency.

Perturbative solution for anisotropic magnetic effect

The presence of a weak magnetic field, characterized by small ϵ\epsilon, may be considered as a perturbation to the Lamb’s two-turning-point eigenvalue problem. Hence, the magnetic field changes both the locations of the turning points and the form of the potential Γ(Z,ϵ,δ)\Gamma(Z,\epsilon,\delta).

Perturbative calculations

A formal asymptotic solution to Eq. (44) is constructed by introducing a book-keeping, small parameter 1ζ^1\zeta in (44) (see e.g., [3, 38]):

ζ2∂Z2ψ^+Γ2(Z,ϵ,δ)ψ^=0.(45)\zeta^2\partial_Z^2\hat{\psi}+\Gamma^2(Z,\epsilon,\delta)\hat{\psi}=0. \tag*{(45)}

To find the magnetically modified eigenfrequencies, we develop below an asymptotic theory starting from Eq. (45), by using perturbative expansions in powers of ζ\zeta, ϵ\epsilon, and δ\delta, in that order.

Following the standard procedure of the ζ\zeta-expansion of the state vector ψ^\hat{\psi}, we write the JWKB quantization condition (see e.g., [3, 38]),

1π∫Z1(ϵ,δ)Z2(ϵ,δ)Γ(Z,ϵ,δ) dZ∼(m+12);m=0,1,2,…,(46)\frac{1}{\pi}\int_{Z_1(\epsilon,\delta)}^{Z_2(\epsilon,\delta)}\Gamma(Z,\epsilon,\delta)\,dZ\sim\left(m+\frac{1}{2}\right); \quad m=0,1,2,\ldots, \tag*{(46)}

where mm refers to the eigenstate index, Γ(Z,ϵ,δ)\Gamma(Z,\epsilon,\delta) is the magnetically modified wavenumber, and Z1(ϵ,δ)Z_1(\epsilon,\delta) and Z2(ϵ,δ)Z_2(\epsilon,\delta) are the magnetically shifted turning points. Equation (46) is the leading-order contribution in the conventional JWKB asymptotic series, where the terms corresponding to the first two orders in ζ\zeta are retained. The first-order term yields 1/21/2, shown on the right. Higher-order terms in ζ\zeta can be computed, see, e.g., Eqs. (7) to (10) of Tripathi [38].

Next, we expand the wavenumber Γ(Z,ϵ,δ)\Gamma(Z,\epsilon,\delta) as

Γ(Z,ϵ,δ)=Γ0(Z,δ)+ϵΓ1(Z,δ)+ϵ2Γ2(Z,δ)+ϵ3Γ3(Z,δ)+O(ϵ4).(47)\Gamma(Z,\epsilon,\delta) = \Gamma_0(Z,\delta) + \epsilon\Gamma_1(Z,\delta) + \epsilon^2\Gamma_2(Z,\delta) + \epsilon^3\Gamma_3(Z,\delta) + \mathcal{O}(\epsilon^4). \tag*{(47)}

Note that the leading-order effect of the Lorentz force on the wavefrequency appears only at the second order O(ϵ2)\mathcal{O}(\epsilon^2) in the expansion, i.e., Γ1(Z,δ)\Gamma_1(Z,\delta) above is zero. The expressions for Γ0(Z,δ)\Gamma_0(Z,\delta) and Γ2(Z,δ)\Gamma_2(Z,\delta) are

Γ0(Z,δ)=K(Z−α)(β−Z)Z,(48)\Gamma_0(Z,\delta) = \frac{K\sqrt{(Z-\alpha)(\beta-Z)}}{Z}, \tag*{(48)}
Γ2(Z,δ)=b2Z2+b1Z+b0(Z−α)(β−Z),(49)\Gamma_2(Z,\delta) = \frac{b_2Z^2+b_1Z+b_0}{\sqrt{(Z-\alpha)(\beta-Z)}}, \tag*{(49)}

where we have used KX=Kcos⁡θK_X=K\cos\theta and KY=Ksin⁡θK_Y=K\sin\theta. The parameters α\alpha and β\beta depend on δ=K/Ω2\delta=K/\Omega^2 and satisfy the following properties

αβ=n(n+2)4K2,(50)\alpha\beta=\frac{n(n+2)}{4K^2}, \tag*{(50)}
α+β=[Ω2K−(n+1)(n+1−nγ)γ2KΩ2]1K.(51)\alpha+\beta=\left[\frac{\Omega^2}{K}-\frac{(n+1)(n+1-n\gamma)}{\gamma^2}\frac{K}{\Omega^2}\right]\frac{1}{K}. \tag*{(51)}

The lengthy expressions of b0b_0, b1b_1, and b2b_2 are presented in Appendix B. We note that these parameters depend on Ω\Omega, KK, θ\theta, nn, and γ\gamma only.

Now we expand the turning points Z1(ϵ,δ)Z_1(\epsilon,\delta) and Z2(ϵ,δ)Z_2(\epsilon,\delta) around the turning points of the unmagnetized polytrope, Z1(0)=αZ_1^{(0)}=\alpha and Z2(0)=βZ_2^{(0)}=\beta in powers of ϵ\epsilon. For clarity we now suppress the implicit δ\delta–dependence of the turning points. For example, we write Zj(ϵ,δ)Z_j(\epsilon,\delta) as Zj(ϵ)Z_j(\epsilon), which we expand in powers of ϵ\epsilon as

Zj(ϵ)=Zj(0)+ϵZj(1)+ϵ2Zj(2)+O(ϵ3),(52)Z_j(\epsilon)=Z_j^{(0)}+\epsilon Z_j^{(1)}+\epsilon^2 Z_j^{(2)}+\mathcal{O}(\epsilon^3), \tag*{(52)}

where j=1j=1 and j=2j=2 refer to the left and the right turning points, respectively (i.e., α<β\alpha<\beta). We note that, in Eq. (52), the correction term at the first order in ϵ\epsilon is zero, i.e., Zj(1)=0Z_j^{(1)}=0. We find this result by substituting the expression for Zj(ϵ)Z_j(\epsilon) from Eq. (52) in Γ(Z,ϵ)=0\Gamma(Z,\epsilon)=0, and by employing Eq. (47). Solving the resulting equation order-by-order in ϵ\epsilon produces Zj(1)=0Z_j^{(1)}=0. We note, however, that, at the second order in ϵ\epsilon (which is where the effect of the Lorentz force comes in action), Zj(2)Z_j^{(2)} becomes non-zero. The expressions for Zj(2)Z_j^{(2)} are given in Appendix C.

Because the term Zj(2)Z_j^{(2)} appears at the second order in ϵ\epsilon, it may be tempting to assume that the term contributes to second order in ϵ\epsilon itself in the JWKB integral in (46). This, however, is not the case. The term contributes to third and higher orders in ϵ\epsilon as we show next. Expanding Eq. (46),

1π∫Z1(0)+ϵ2Z1(2)+O(ϵ3)Z2(0)+ϵ2Z2(2)+O(ϵ3)[Γ0(Z)+ϵ2Γ2(Z)+O(ϵ3)] dZ∼(m+12);m=0,1,2,…(53)\frac{1}{\pi}\int_{Z_1^{(0)}+\epsilon^2Z_1^{(2)}+\mathcal{O}(\epsilon^3)}^{Z_2^{(0)}+\epsilon^2Z_2^{(2)}+\mathcal{O}(\epsilon^3)} \left[\Gamma_0(Z)+\epsilon^2\Gamma_2(Z)+\mathcal{O}(\epsilon^3)\right]\,\mathrm{d}Z \sim\left(m+\frac{1}{2}\right); \quad m=0,1,2,\ldots \tag*{(53)}

We now integrate Γ0(Z)\Gamma_0(Z) as

∫Z1(0)+ϵ2Z1(2)+O(ϵ3)Z2(0)+ϵ2Z2(2)+O(ϵ3)Γ0(Z) dZ=(∫α+ϵ2Z1(2)+O(ϵ3)α+∫αβ+∫ββ+ϵ2Z2(2)+O(ϵ3))Γ0(Z) dZ=−2ϵ3K(β−α)1/23α[Z1(2)]3/2+∫αβΓ0(Z) dZ+2ϵ3K(β−α)1/23β[−Z2(2)]3/2+O(ϵ4).(54)\begin{aligned} \int_{Z_1^{(0)}+\epsilon^2Z_1^{(2)}+\mathcal{O}(\epsilon^3)}^{Z_2^{(0)}+\epsilon^2Z_2^{(2)}+\mathcal{O}(\epsilon^3)} \Gamma_0(Z)\,\mathrm{d}Z \\ &=\left(\int_{\alpha+\epsilon^2Z_1^{(2)}+\mathcal{O}(\epsilon^3)}^\alpha+\int_\alpha^\beta+\int_\beta^{\beta+\epsilon^2Z_2^{(2)}+\mathcal{O}(\epsilon^3)}\right)\Gamma_0(Z)\,\mathrm{d}Z \\ &=-\frac{2\epsilon^3K(\beta-\alpha)^{1/2}}{3\alpha}\left[Z_1^{(2)}\right]^{3/2}+\int_\alpha^\beta\Gamma_0(Z)\,\mathrm{d}Z \\ &+\frac{2\epsilon^3K(\beta-\alpha)^{1/2}}{3\beta}\left[-Z_2^{(2)}\right]^{3/2}+\mathcal{O}(\epsilon^4). \tag*{(54)} \end{aligned}

where we notice terms with ϵ3\epsilon^3 arising from the second-order shifts in the turning points, ϵ2Z1(2)\epsilon^2Z_1^{(2)} and ϵ2Z2(2)\epsilon^2Z_2^{(2)}. The additional power of ϵ\epsilon emerges from the integrand Γ0(Z)\Gamma_0(Z), which has a term Z−Zj(0)\sqrt{Z-Z_j^{(0)}}. This term when expanded around Zj(0)Z_j^{(0)} in powers of ϵ\epsilon contributes an ϵ\epsilon to the integral.

In Eq. (54), the term ∫αβΓ0(Z) dZ\int_\alpha^\beta\Gamma_0(Z)\,\mathrm{d}Z is the integral that one finds in Lamb’s calculations. Since here we seek the effect of magnetic fields on Lamb waves, we replace ∫αβΓ0(Z) dZ\int_\alpha^\beta\Gamma_0(Z)\,\mathrm{d}Z with Lamb’s exact dispersion relation for hydrodynamic waves [Eq. (1)]:

1π∫αβΓ0(Z) dZ≡ILamb=(n+1)2[Ω2K(n+1)+(γn−n−1)Kγ2Ω2−1]+1.(55)\frac{1}{\pi}\int_\alpha^\beta\Gamma_0(Z)\,\mathrm{d}Z\equiv I_{\mathrm{Lamb}}=\frac{(n+1)}{2}\left[\frac{\Omega^2}{K(n+1)}+\frac{(\gamma n-n-1)K}{\gamma^2\Omega^2}-1\right]+1. \tag*{(55)}

Thus, we replace the integral ∫αβΓ0(Z) dZ\int_\alpha^\beta\Gamma_0(Z)\,\mathrm{d}Z, appearing in Eq. (54), with ILambI_{\mathrm{Lamb}} from Eq. (55) to obtain

ILamb+ϵ2π∫αβΓ2(Z) dZ+O(ϵ3)∼(m+12);m=0,1,2,…(56)\begin{aligned} I_{\mathrm{Lamb}}+\frac{\epsilon^2}{\pi}\int_\alpha^\beta\Gamma_2(Z)\,\mathrm{d}Z+\mathcal{O}(\epsilon^3) \\ &\sim\left(m+\frac{1}{2}\right); \quad m=0,1,2,\ldots \tag*{(56)} \end{aligned}

which is accurate up to second order in ϵ\epsilon. We substitute Γ2(Z)\Gamma_2(Z) from Eq. (49) and perform the integral in Eq. (56) to arrive at

ILamb−ϵ2[b0+b1(α+β2)+b2(3α2+2αβ+3β28)]∼(m+12);m=0,1,2,…(57)I_{\mathrm{Lamb}}-\epsilon^{2}\left[b_{0}+b_{1}\left(\frac{\alpha+\beta}{2}\right)+b_{2}\left(\frac{3\alpha^{2}+2\alpha\beta+3\beta^{2}}{8}\right)\right]\sim\left(m+\frac{1}{2}\right); \quad m=0,1,2,\ldots \tag*{(57)}

Employing fast-wave approximation, we now expand each term on the left-hand side of Eq. (57) in powers of δ=K/Ω2\delta=K/\Omega^{2} as

ILamb≈1δ[12+δ(1−n)2+O(δ2)],(58)I_{\mathrm{Lamb}}\approx\frac{1}{\delta}\left[\frac{1}{2}+\frac{\delta(1-n)}{2}+\mathcal{O}(\delta^{2})\right], \tag*{(58)}
b0≈1δ[2−γ4+O(δ2)],(59)b_{0}\approx\frac{1}{\delta}\left[\frac{2-\gamma}{4}+\mathcal{O}(\delta^{2})\right], \tag*{(59)}
b1(α+β2)≈1δ[cos⁡2θ8+O(δ2)],(60)b_{1}\left(\frac{\alpha+\beta}{2}\right)\approx\frac{1}{\delta}\left[\frac{\cos^{2}\theta}{8}+\mathcal{O}(\delta^{2})\right], \tag*{(60)}
b2(3α2+2αβ+3β28)≈1δ[−3cos⁡4θ8+O(δ2)].(61)b_{2}\left(\frac{3\alpha^{2}+2\alpha\beta+3\beta^{2}}{8}\right)\approx\frac{1}{\delta}\left[\frac{-3\cos^{4}\theta}{8}+\mathcal{O}(\delta^{2})\right]. \tag*{(61)}

Thus we obtain a simplified dispersion relation from Eq. (57):

Ω2K[12−ϵ24{2−γ+cos⁡2θ−3cos⁡4θ2}]∼(m+n2);m=0,1,2,…(62)\frac{\Omega^{2}}{K}\left[\frac{1}{2}-\frac{\epsilon^{2}}{4}\left\{2-\gamma+\cos^{2}\theta-\frac{3\cos^{4}\theta}{2}\right\}\right]\sim\left(m+\frac{n}{2}\right); \quad m=0,1,2,\ldots \tag*{(62)}

It turns out that Eq. (62) does not agree excellently with our numerical results, see Figure A1 in Appendix D. But replacing cos⁡2θ\cos^{2}\theta with (cos⁡2θ)/2(\cos^{2}\theta)/2 gives excellent agreement.

Figure A1. Comparison of different expressions to assess their individual contribution. The innermost (red) curve is the universal curve on which all numerical data collapse.

Informed in this way, we write the final dispersion relation

Ω2K[12−ϵ24{2−γ+cos⁡2θ2−3cos⁡4θ2}]∼(m+n2);m=0,1,2,…(63)\boxed{\frac{\Omega^{2}}{K}\left[\frac{1}{2}-\frac{\epsilon^{2}}{4}\left\{2-\gamma+\frac{\cos^{2}\theta}{2}-\frac{3\cos^{4}\theta}{2}\right\}\right]\sim\left(m+\frac{n}{2}\right); \quad m=0,1,2,\ldots} \tag*{(63)}

Comparison between theory and numerics

Using the Lamb’s relation, Ωhydro2∼K(2m+n)\Omega_{\mathrm{hydro}}^{2}\sim K(2m+n), we rewrite Eq. (63) as

g(θ)≡γ2+ΔΩ2ϵ2Ωhydro2=1+14(cos⁡2θ−3cos⁡4θ),(64)g(\theta)\equiv\frac{\gamma}{2}+\frac{\Delta\Omega^{2}}{\epsilon^{2}\Omega_{\mathrm{hydro}}^{2}} =1+\frac{1}{4}\left(\cos^{2}\theta-3\cos^{4}\theta\right), \tag*{(64)}

where ΔΩ2≡Ω2−Ωhydro2\Delta\Omega^{2}\equiv\Omega^{2}-\Omega_{\mathrm{hydro}}^{2}. This is a remarkable result. The right-hand-side of Eq. (64) is independent of every other possible parameter other than θ\theta! In Figure 5, we plot the function g(θ)g(\theta) from our numerical determination of the eigenfrequencies for numerous different values of mm and KK, and two different values of ϵ\epsilon, and two different values of γ\gamma. These numerical values are shown with different markers. All the different curves collapse onto each other, creating a universal master curve. We also plot our asymptotic expression, Eq. (63), which agrees very well with the numerical results. This demonstrates that our theory is in excellent agreement with numerics even for ϵ\epsilon as large as ϵ=0.1\epsilon=0.1.

Polar plot comparing asymptotic theory and numerical solutions

Figure 5. Theory and numerics: Our theory predicts that g(θ)=ΔΩ2/(ϵ2Ωhydro2)+γ/2g(\theta)=\Delta\Omega^2/(\epsilon^2\Omega_{\mathrm{hydro}}^2)+\gamma/2 depends only on θ\theta as given in Eq. (64). The function g(θ)g(\theta) is independent of the wavenumber KK, Alfvénic Mach number ϵ−1\epsilon^{-1}, the polytropic index nn, the adiabatic index γ\gamma, and the eigenstate index mm. Here, we plot g(θ)g(\theta) obtained from our numerical solutions for several different values of mm, ranging from 10 to 35, and for several different values of KK (0.6≤∣K∣≤20.6\leq|K|\leq2), and for two values of ϵ\epsilon and two values of γ\gamma (different symbols). All data points collapse on the same universal curve. The function g(θ)g(\theta) from our asymptotic theory, Eq. (64), is also plotted here. The asymptotic curve is indistinguishable from the exact numerical solutions.

It is surprising that, in our attempt to capture the anisotropy brought in by the magnetic field, despite a myriad of unwieldy expressions encountered on the way, an expression as simple as Eq. (63), is obtained. This simple expression is also highly accurate, as demonstrated in Figure 5. The gratifying success of our asymptotic theory is somewhat unexpected, given that the analytical solution is accurate only up to the leading order in ζ\zeta, ϵ\epsilon, and δ=K/Ω2\delta=K/\Omega^{2}.

More accurate solutions may be obtained by using the full expressions of b0b_{0}, b1b_{1}, and b2b_{2} from Appendix B in Eqs. (40d)–(40b), and retaining higher-order terms in ζ\zeta and ϵ\epsilon.

We emphasize that the closed-form expression we obtain is not solely by use of the perturbative calculation. At the last step, we needed guidance from our numerical solutions. This implies that future work should develop a better perturbation theory. We discuss a few possibilities. Let us first revisit our perturbative calculation. Equation (44) is exact. It has two small parameters, ϵ\epsilon and δ\delta. For ϵ=0\epsilon=0, this equation reduces to the Lamb’s problem. We note that our final dispersion relation (63) reproduces the high-frequency Lamb’s dispersion relation (2) if we set ϵ=0\epsilon=0 in Eq. (63). In our perturbative calculation, we also introduce the additional small parameter ζ\zeta to allow us to use the JWKB formalism. The first step then is to retain the first two leading-order terms in ζ\zeta. Thus we obtain Eq. (46). We arrive at a curious JWKB problem where the turning points themselves depend on a small parameter ϵ\epsilon, see Eq. (53). We retain terms up to second order in ϵ\epsilon in this expansion, thereby arriving at Eq. (56). Next, we expand all terms in powers of δ\delta. If we change the order of expansion in ϵ\epsilon and δ\delta, then we no longer obtain the mathematical structure of the two-turning-point JWKB problem. The leading-order term reduces to an expression whose ZZ-dependence is solely Z−1/2Z^{-1/2}. To reformulate that expansion such that we can compare it with the Lamb’s result, we must resum the δ\delta-expanded infinite series which is not straightforward. So, this does not seem to be a useful alternative. Next, we remind the reader that a naive application of the JWKB method to calculate eigenvalues of the Schrodinger equation for either the harmonic oscillator or the Coulomb potential in three dimensions gives a factor of ℓ(ℓ+1)\ell(\ell+1) [ℓ\ell is the quantum number for orbital angular momentum], which must be replaced with (ℓ+1/2)2(\ell+1/2)^2 to obtain the correct expression (Langer [21]). This correction appears naturally when the JWKB series is resummed to all orders ([30, 29, 38]). Hence, there is a possibility that a resummation of the higher-order terms in powers of ζ\zeta may reproduce Eq. (63). Finally, we note the third possibility. We mapped Eq. (42) to a form amenable to the JWKB method. Instead, for ϵ=0\epsilon=0, Eq. (42) can be transformed to the confluent hypergeometric equation—this is how Lamb first solved for the waves in a hydrodynamic polytrope. Thus, for a small ϵ\epsilon, it possible to use the method of multiple scales in a way such that the confluent hypergeometric functions appear as leading-order solutions, allowing to compute perturbed eigenfrequencies.

Role in helioseismic inversions

Since our theoretical formula and numerics show in Figure 5 that the expression ΔΩ2/(ϵ2Ωhydro2)+γ/2\Delta\Omega^2/(\epsilon^2\Omega_{\mathrm{hydro}}^2)+\gamma/2 collapses on a single universal curve g(θ)=1+14(cos⁡2θ−3cos⁡4θ)g(\theta)=1+\frac{1}{4}(\cos^2\theta-3\cos^4\theta), it suggests that high-frequency helioseismic oscillation data should collapse on a single curve if one plots the fractional change in observed frequency compared to the Lamb’s hydrodynamic model. Such a plot allows extraction of two parameters: ϵ2\epsilon^2, which encodes the magnetic field strength, and γ\gamma, which is the adiabatic index of the gaseous atmosphere. Since our theory is applicable to high-frequency (m≫1m\gg1) waves, the polytropic index nn cannot be inferred reliably using our method, as the squared-frequency of oscillations in the hydrodynamic model in Eq. (2) scales as 2m+n∼2m2m+n\sim2m.

We also note that we have considered a horizontal magnetic field in local Cartesian coordinates, which corresponds to non-radial magnetic fields in spherical coordinates. Such non-radial fields are found to be the most sensitive to the helioseismic inversion kernels near the solar surface ([13, 14]). This justifies our consideration. Note further that the inversion kernels in general are model-dependent, and the models for stars other than the Sun are not well-constrained. Since our formula has only two fitting parameters, more reliable estimation of those parameters is possible.

The effect of magnetic fields on slow gravity-driven waves is not considered in this paper, as the gravity-driven waves generally penetrate deep in the core of a star, which requires global geometry. Our current analysis in the Cartesian domain is appropriate for the fast pressure-driven waves that are confined in the subsurface layers of a star. We shall nevertheless remark briefly on the potential impact of magnetic fields on slow gravity-driven waves. For the slow waves, the first term on the left-hand side of Eq. (1) may be neglected, which implies: (i) squared time period of such waves is linearly proportional to the eigenstate index, and (ii) such waves in unmagnetized medium become unstable when γ<(1+1/n)\gamma< (1 + 1/n). This instability criterion (ii) is consistent with the energy principle of Newcomb [27]. When a magnetic field is present, by applying the energy principle of Newcomb, we find, for KX≠0,2K_X \ne0, 22 the instability threshold on γ\gamma for the gravity waves is lifted to γ<(1+1/n)(1+1/β)\gamma< (1 + 1/n)(1 + 1/\beta). When the magnetic field is very strong (β≪1\beta\ll1), the regular perturbation series in powers of ϵ\epsilon (∝1/β\propto1/\sqrt{\beta}) is possibly inadequate for unstable gravity-driven waves. Such considerations are clearly beyond the scope of the present paper. We note, however, that thorough understanding of the effect of magnetic fields on mixed gravito-acoustic waves holds a promise to deliver reliable estimation of the strengths and geometries of magnetic fields that are buried in deep layers of stellar interiors ([5, 25, 24]). As a testament to this promise, recent global asteroseismic studies have detected and characterized magnetic fields in the core of red giant stars, opening a wholly new avenue to magnetoseismology (Li et al. 2022, 2023; Deheuvels et al. 2023).

Conclusions

Here we derive, for pressure-driven waves, an accurate and simple analytical formula that captures the effects of magnetic field and five other parameters: adiabatic index, polytropic index, eigenmode state index, wavenumbers, and the angle between the wavevector and the magnetic field [θ=cos⁡−1(K^⋅B^0)\theta= \cos^{-1}(\hat{K} \cdot\hat{B}_0)]. Such a six-parameter-dependent formula is distilled using a perturbative solution to magnetized version of Lamb’s hydrodynamic polytropic waves. Our explicit analytical formula overcomes the limitation of previously attempted formulae for the magnetized polytrope that were presented in general integral forms; such a formulation, for instance, that of Gough & Thompson [ [16]; e.g., Eq. (4.11) and Bogdan & Cally [4], requires numerical evaluation of the eigenfrequencies, and thus leaves out the critical step of obtaining an analytical understanding and expression. We arrive at our result, guided by our numerical solutions and at the cost of extensive use of Mathematica for our perturbative analyses.

The simplicity and accuracy of our formula are encouraging to employ the formula to help solve the inverse problem of magnetoseismology. Our formula provides an explicit, analytical dependence of the observed surface oscillation frequency with the orientation and strength of the subsurface magnetic field. Such an understanding may be able to predict the surface-emergence location and strength of active regions, and to understand the properties of subsurface dynamo, e.g., magnetic activity in the near-surface shear layer ([45]).

Nonlinear asteroseismology can also directly benefit from our analytical work, as weak turbulence theory of asteroseismic waves inevitably requires accurate and simple expressions of linear wave frequencies in resonant triad interactions. The effect of the magnetic field on such waves remains unknown. However, observations now exist that suggest resonant mode interactions are possible and can be a critical element of strongly pulsating stars ([17]). Nonlinear mode coupling of linear eigenmodes (Tripathi et al. 2023a,b; Tripathi et al. 2024) may also need to be analyzed, in addition to mode resonances. Future planned research will directly take advantage of the formula derived here to assess the role of the magnetic field and other parameters in asteroseismic wave turbulence, which is eagerly awaiting to soon enter adulthood from its infancy.

Acknowledgments

We are pleased to thank Ellen G. Zweibel for helpful discussions and for her contribution in applying the energy principle of Newcomb [27] to our system. We are grateful to Srijan B. Das and the anonymous reviewer for their feedback that improved the paper. The inception of this paper took place at Nordic Institute for Theoretical Physics (NORDITA), Sweden, while B.T. was visiting on a research fellowship generously made available by NORDITA.

Data Availability

Mathematica notebook and Dedalus script used in this paper are deposited in https://github.com/BindeshTripathi/polytrope. The notebook is developed to perform lengthy analytical manipulations, and the python script is created for spectral numerical solution of linear MHD waves.

Appendix A

The matrix elements MijM_{ij} and hμh_\mu that appear in Eq. (40) are

M11=−Ω2(1+γϵ2/2),(65)M_{11}=-\Omega^2(1+\gamma\epsilon^2/2), \tag*{(65)}
M12=0,(66)M_{12}=0, \tag*{(66)}
M13=−iKX(n+1)(1+γϵ2/2)γ,(67)M_{13}=-\frac{iK_X(n+1)(1+\gamma\epsilon^2/2)}{\gamma}, \tag*{(67)}
M21=ϵ2KXKYZ,(68)M_{21}=\epsilon^2K_XK_YZ, \tag*{(68)}
M22=−Ω2(1+γϵ2/2)+ϵ2KX2Z,(69)M_{22}=-\Omega^2(1+\gamma\epsilon^2/2)+\epsilon^2K_X^2Z, \tag*{(69)}
M23=−iKY(n+1)(1+γϵ2/2)γ,(70)M_{23}=-\frac{iK_Y(n+1)(1+\gamma\epsilon^2/2)}{\gamma}, \tag*{(70)}
M31=iKXγ[1+ϵ2(3γ2+KX2ZΩ2)],(71)M_{31}=\frac{iK_X}{\gamma}\left[1+\epsilon^2\left(\frac{3\gamma}{2}+\frac{K_X^2Z}{\Omega^2}\right)\right], \tag*{(71)}
M32=iKYγ[1+ϵ2(γ2+KX2ZΩ2)],(72)M_{32}=\frac{iK_Y}{\gamma}\left[1+\epsilon^2\left(\frac{\gamma}{2}+\frac{K_X^2Z}{\Omega^2}\right)\right], \tag*{(72)}
M33=−Ω2(1+γϵ2/2)+ϵ2KX2Zn+1,(73)M_{33}=\frac{-\Omega^2(1+\gamma\epsilon^2/2)+\epsilon^2K_X^2Z}{n+1}, \tag*{(73)}
hx=iKXZχ^,(74)h_x=iK_XZ\hat{\chi}, \tag*{(74)}
hy=iKYZχ^(1+ϵ2),(75)h_y=iK_YZ\hat{\chi}(1+\epsilon^2), \tag*{(75)}
hz=χ^+Z∂Zχ^n+1+ϵ2χ^[1+KX2Z{γ−(n+1)(1+γϵ2/2)}(n+1)γΩ2(1+γϵ2/2)]+ϵ2Z∂Zχ^n+1[1+KX2ZΩ2(1+γϵ2/2)].(76)h_z=\hat{\chi}+\frac{Z\partial_Z\hat{\chi}}{n+1}+\epsilon^2\hat{\chi}\left[1+\frac{K_X^2Z\{\gamma-(n+1)(1+\gamma\epsilon^2/2)\}}{(n+1)\gamma\Omega^2(1+\gamma\epsilon^2/2)}\right]+\epsilon^2\frac{Z\partial_Z\hat{\chi}}{n+1}\left[1+\frac{K_X^2Z}{\Omega^2(1+\gamma\epsilon^2/2)}\right]. \tag*{(76)}

Appendix B

The parameters introduced in Eq. (32b), while writing the expression for Γ2(Z)\Gamma_2(Z), appear below:

b2=−K3cos⁡4θΩ2,(77)b_2=-\frac{K^3\cos^4\theta}{\Omega^2}, \tag*{(77)}
b1=−K[−Ω8γ4+Ω4γ2K2(n+1){(n+1+γ−γn)cos⁡(2θ)−n−1−3γ}+2K4(n+1)4−K4(n+1)3(n+1−γn)cos⁡(2θ)]2Ω4γ2sec⁡2θ[Ω4γ2−K2(n+1)2],(78)b_1=\frac{-K\left[-\Omega^8\gamma^4+\Omega^4\gamma^2K^2(n+1)\{(n+1+\gamma-\gamma n)\cos(2\theta)-n-1-3\gamma\}+2K^4(n+1)^4-K^4(n+1)^3(n+1-\gamma n)\cos(2\theta)\right]}{2\Omega^4\gamma^2\sec^2\theta\left[\Omega^4\gamma^2-K^2(n+1)^2\right]}, \tag*{(78)}
b0=(4Ω8γ4(2−γ)−2Ω4γ2K2{2(n+1)2(4−3γ)+n(2n+3)γ2}+K4(n+1)2{nγ2(2n+3)−8(n+1)2(γ−1)}+2K2γcos⁡(2θ)[K2(n+1)2{n(γ−4+2n(γ−1))−2}+Ω4γ2{2(n+1)2−γn(2n+3)}]+nγ2K4(n+1)2cos⁡(4θ))16Ω2γ2K[Ω4γ2−K2(n+1)2].(79)b_0=\frac{\left(4\Omega^8\gamma^4(2-\gamma)-2\Omega^4\gamma^2K^2\{2(n+1)^2(4-3\gamma)+n(2n+3)\gamma^2\}+K^4(n+1)^2\{n\gamma^2(2n+3)-8(n+1)^2(\gamma-1)\}+2K^2\gamma\cos(2\theta)\left[K^2(n+1)^2\{n(\gamma-4+2n(\gamma-1))-2\}+\Omega^4\gamma^2\{2(n+1)^2-\gamma n(2n+3)\}\right]+n\gamma^2K^4(n+1)^2\cos(4\theta)\right)}{16\Omega^2\gamma^2K\left[\Omega^4\gamma^2-K^2(n+1)^2\right]}. \tag*{(79)}

Appendix C

Due to the magnetic field, the locations of the turning points, Z1Z_1 and Z2Z_2, shift—which to the second order in ϵ\epsilon in Eq. (52) are given by Z1(2)Z_1^{(2)} and Z2(2)Z_2^{(2)}.

Z1(2)=α(b2α2+b1α+b0)K(β−α),(80)Z_1^{(2)}=\frac{\alpha(b_2\alpha^2+b_1\alpha+b_0)}{K(\beta-\alpha)}, \tag*{(80)}
Z2(2)=−β(b2β2+b1β+b0)K(β−α).(81)Z_2^{(2)}=-\frac{\beta(b_2\beta^2+b_1\beta+b_0)}{K(\beta-\alpha)}. \tag*{(81)}

Appendix D

References

  1. [1]Adam, J. 1977, Solar Physics, 52, 293
  2. [2]Aerts, C., Christensen-Dalsgaard, J., & Kurtz, D. W. 2010, Astereoseismology (Springer Science & Business Media)
  3. [3]Bender, C. M., & Orszag, S. A. 1978, Advanced Mathematical Methods for Scientists and Engineers
  4. [4]Bogdan, T. J., & Cally, P. S. 1997, Proceedings of the Royal Society of London Series A, 453, 943, doi: 10.1098/rspa.1997.0052DOI
  5. [5]Bugnet, L., Prat, V., Mathis, S., et al. 2021, A&A, 650, A53, doi: 10.1051/0004-6361/202039159DOI
  6. [6]Burns, K. J., Vasil, G. M., Oishi, J. S., Lecoanet, D., & Brown, B. P. 2020, Physical Review Research, 2, 023068, doi: 10.1103/PhysRevResearch.2.023068DOI
  7. [7]Cally, P. S. 2007, Astronomische Nachrichten, 328, 286, doi: 10.1002/asna.200610731DOI
  8. [8]Cally, P. S., & Bogdan, T. 1997, The Astrophysical Journal, 486, L67
  9. [9]Cally, P. S., & Bogdan, T. J. 1993, ApJ, 402, 721, doi: 10.1086/172172DOI
  10. [10]Campos, L. 1983, Wave Motion, 5, 1
  11. [11]Campos, L., & Marta, A. 2015, Geophysical & Astrophysical Fluid Dynamics, 109, 168
  12. [13]Das, S. B. 2022, The Astrophysical Journal, 940, 92
  13. [14]Das, S. B., Chakraborty, T., Hanasoge, S. M., & Tromp, J. 2020, The Astrophysical Journal, 897, 38
  14. [16]Gough, D., & Thompson, M. 1990, Monthly Notices of the Royal Astronomical Society, 242, 25
  15. [17]Guo, Z. 2020, ApJ, 896, 161, doi: 10.3847/1538-4357/ab911fDOI
  16. [18]Hasselmann, K. 1962, Journal of Fluid Mechanics, 12, 481, doi: 10.1017/S0022112062000373DOI
  17. [19]Ilonidis, S., Zhao, J., & Kosovichev, A. 2011, Science, 333, 993, doi: 10.1126/science.1206253DOI
  18. [20]Lamb, H. 1911, Proceedings of the Royal Society of London Series A, 84, 551, doi: 10.1098/rspa.1911.0008DOI
  19. [21]Langer, R. E. 1937, Physical Review, 51, 669
  20. [24]Mathis, S., & Bugnet, L. 2023, A&A, 676, L9, doi: 10.1051/0004-6361/202346832DOI
  21. [25]Mathis, S., Bugnet, L., Prat, V., et al. 2021, A&A, 647, A122, doi: 10.1051/0004-6361/202039180DOI
  22. [26]Nazarenko, S., & Lukaschuk, S. 2016, Annual Review of Condensed Matter Physics, 7, 61, doi: 10.1146/annurev-conmatphys-071715-102737DOI
  23. [27]Newcomb, W. A. 1961, Physics of Fluids, 4, 391, doi: 10.1063/1.1706342DOI
  24. [28]Nye, A. H., & Thomas, J. H. 1976, ApJ, 204, 573, doi: 10.1086/154205DOI
  25. [29]Robnik, M., & Salasnich, L. 1997, Journal of Physics A: Mathematical and General, 30, 1711, doi: 10.1088/0305-4470/30/5/031DOI
  26. [30]Romanovski, V. G., & Robnik, M. 2000, Journal of Physics A: Mathematical and General, 33, 8549, doi: 10.1088/0305-4470/33/47/315DOI
  27. [31]Schunker, H., Braun, D. C., Cally, P. S., & Lindsey, C. 2005, The Astrophysical Journal, 621, L149
  28. [32]Singh, N. K., Brandenburg, A., Chitre, S., & Rheinhardt, M. 2015, Monthly Notices of the Royal Astronomical Society, 447, 3708
  29. [33]Singh, N. K., Brandenburg, A., & Rheinhardt, M. 2014, The Astrophysical Journal Letters, 795, L8
  30. [34]Singh, N. K., Raichur, H., & Brandenburg, A. 2016, The Astrophysical Journal, 832, 120
  31. [35]Singh, N. K., Raichur, H., Käpylä, M. J., et al. 2020, Geophysical & Astrophysical Fluid Dynamics, 114, 196
  32. [36]Spruit, H. C., & Bogdan, T. J. 1992, ApJL, 391, L109, doi: 10.1086/186409DOI
  33. [37]Thomas, J. H. 1983, Annual Review of Fluid Mechanics, 15, 321
  34. [38]Tripathi, B. 2022, PhRvD, 105, 036010, doi: 10.1103/PhysRevD.105.036010DOI
  35. [41]Tripathi, B., & Mitra, D. 2022, ApJ, 934, 61, doi: 10.3847/1538-4357/ac79b1
  36. [43]Van Beeck, J., Bowman, D. M., Pedersen, M. G., et al. 2021, A&A, 655, A59, doi: 10.1051/0004-6361/202141572DOI
  37. [44]Van Beeck, J., Van Hoolst, T., Aerts, C., & Fuller, J. 2023, arXiv e-prints, arXiv:2311.02972, doi: 10.48550/arXiv.2311.02972DOI
  38. [45]Vasil, G. M., Lecoan et, D., Augustson, K., et al. 2024, Nature, 629, 769, doi: 10.1038/s41586-024-07315-1DOI

Paper details

vlnc:W_9azybbgk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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