Arxiv:1109.4571V2 [Astro-Ph.CO] 20 Feb 2012 for Virtually All Astrophysical Objects

Home , Mach number, Magnetic Prandtl number, Magnetic Reynolds number

Magnetic Field Ampliﬁcation by Small-Scale Dynamo Action: Dependence on Turbulence Models and Reynolds and Prandtl Numbers

Jennifer Schober∗ Institut fürTheoretische Astrophysik, Zentrum fürAstronomie der UniversitätHeidelberg, Albert-Uberle-Strasse¨ 2, D-69120 Heidelberg, Germany

Dominik Schleicher† Institut fürAstrophysik, Georg-August-UniversitätGöttingen, Institut fürAstrophysik, Friedrich-Hund-Platz, D-37077 Göttingen,Germany

Christoph Federrath‡ Monash Centre for Astrophysics (MoCA), School of Mathematical Sciences, Monash University, Victoria 3800, Australia; CRAL, Ecole Normale Supérieure de Lyon, F-69364 Lyon, France; and Institut fürTheoretische Astrophysik, Zentrum fürAstronomie der UniversitätHeidelberg, Albert-Uberle-Strasse¨ 2, D-69120 Heidelberg, Germany

Ralf Klessen§ Institut fürTheoretische Astrophysik, Zentrum fürAstronomie der UniversitätHeidelberg, Albert-Uberle-Strasse¨ 2, D-69120 Heidelberg, Germany

Robi Banerjee¶ Hamburger Sternwarte, Gojenbergsweg 112, D-21029 Hamburg, Germany (Dated: November 6, 2018) The small-scale dynamo is a process by which turbulent kinetic energy is converted into magnetic energy, and thus it is expected to depend crucially on the nature of the turbulence. In this paper, we present a model for the small-scale dynamo that takes into account the slope of the turbulent velocity spectrum v(`) ∝ `ϑ, where ` and v(`) are the size of a turbulent fluctuation and the typical velocity on that scale. The time evolution of the fluctuation component of the magnetic field, i.e., the small-scale field, is described by the Kazantsev equation. We solve this linear differential equation for its eigenvalues with the quantum-mechanical WKB-approximation. The validity of this method is estimated as a function of the magnetic Prandtl number Pm. We calculate the minimal magnetic Reynolds number for dynamo action, Rmcrit, using our model of the turbulent velocity correlation function. For Kolmogorov turbulence (ϑ = 1/3), we find that the critical magnetic Reynolds number K B is Rmcrit ≈ 110 and for Burgers turbulence (ϑ = 1/2) Rmcrit ≈ 2700. Furthermore, we derive that the growth rate of the small-scale magnetic field for a general type of turbulence is Γ ∝ Re(1−ϑ)/(1+ϑ) in the limit of infinite magnetic Prandtl numbers. For decreasing magnetic Prandtl number (down to Pm & 10), the growth rate of the small-scale dynamo decreases. The details of this drop depend on the WKB-approximation, which becomes invalid for a magnetic Prandtl number of about unity.

PACS numbers: 47.65.Md,95.30.Qd,47.27.-i,84.60.Lw

I. INTRODUCTION et al. [1]. Recent studies using γ-ray observations of the −8 intergalactic medium find a lower limit of B0 10 – −6 ≈ Magnetic fields are observed in the whole Universe on 10 nG [2–4], indicative of an early generation scenario. different length scales and they play an important role For the very early Universe, we have only upper limits, but no direct observations showing the presence of mag- arXiv:1109.4571v2 [astro-ph.CO] 20 Feb 2012 for virtually all astrophysical objects. To probe the origin of these fields, it is crucial to gain better theoretical netic fields. Yamazaki et al. [5] derive an upper limit of insight into their dynamical properties. the magnetic field strength from the cosmic microwave Observations suggest that magnetic fields were already background temperature anisotropy. They predict that space filling in the early Universe. Hints toward high- the magnetic field strength at the present scale of 1 Mpc redshift magnetic fields come from the study of Bernet [6] is B0 . 4.7 nG. Moreover, Schleicher and Miniati [7] present a method to derive the magnetic field strength in the reionization epoch (redshift z > 6). They show that the upper halo mass is set by the magnetic pressure ∗ [email protected] and conclude, by using data from the reionization epoch, † [email protected] ‡ [email protected] that the co-moving magnetic field strength is B0 . 3 nG § [email protected] (see also [8]). ¶ [email protected] Besides these observational constraints, there are differ- 2 ent theories that describe the origin of these primordial lent velocity field, but used a small-scale approximation magnetic fields. The first seed fields could already have of this tensor. There are also alternative approaches for been produced during inflation. Turner and Widrow [9] the limit of small Prandtl number based on the separa- −25 −1 find that B0 10 –10 nG on a scale of 1 Mpc can tion of scales [22–24], which find similar solutions of the be produced≈ when the conformal invariance is broken. Kazantsev equation. Further ways to solve this equation Following Sigl et al. [10], there is also the possibility of exist, but go beyond the scope of this work (see for ex- creating a magnetic field during the phase transitions in ample [25, 26]). the very early Universe. They predict a field strength In this paper, we aim to explore dynamo amplification for −20 ϑ B0 10 nG from the electroweak phase transition turbulence models with v(`) ` in the inertial range, ≈ −11 ∝ and B0 10 nG from the quantum chromodynamics i.e., the range of the kinetic energy cascade. Here, ` is (QCD) phase≈ transition. the size of a turbulent fluctuation and v(`) the typical ve- The observed magnetic field strengths in high-redshift as locity on this scale. The range of ϑ goes from 1/3, which well as in present-day galaxies are, however, typically or- describes the incompressible Kolmogorov turbulence [27], ders of magnitude higher than predicted by the theories. to 1/2 for Burgers turbulence [28]. In astrophysical ob- For example the Milky Way has locally a magnetic field jects we expect that the exponent ϑ lies between those of (6 2) 103 nG [11]. Hence there must be additional extrema [29–33]. Burgers turbulence is often referred to and strong± × amplification. The simplest way of amplify- as highly compressible turbulence, obtained in the limit ing a magnetic field by gravitational compression under of very high Mach numbers. the condition of flux freezing, when the magnetic field is Federrath et al. [34] explore the Mach number depen- perfectly coupled to the gas. With the conservation of dence of the small-scale dynamo and show, that the dy- magnetic flux, one can directly show that B ρ2/3 for namo works for both, subsonic and highly supersonic spherical collapse, where ρ is the mass density∝ of the gas. turbulence. Furthermore, these numerical experiments However, magnetohydrodynamical dynamos can amplify indicate that the small-scale magnetic field saturates af- small initial magnetic fields exponentially in time, lead- ter a certain time. They find that the saturation levels ing to a growth of B much faster than ρ2/3 [12–14]. A and growth rates depend strongly on the nature of the dynamo, in general, is a process by which kinetic energy turbulence. However, most simulations typically have a is converted into magnetic energy. One has to distinguish magnetic Prandtl number of Pm 1, requiring further two different types of dynamo: the large-scale dynamo, efforts and understanding for Pm ≈ 1 and Pm 1. which is excited by large-scale motions, and the small- Here, we present an improved analytical model for the scale dynamo, which is excited by turbulence on very small-scale dynamo that takes into account highly com- small scales. Dynamos amplify the magnetic energy ex- pressible turbulence also. In Sec. II, we outline the main ponentially. The small-scale dynamo is important in par- ideas and equations behind the Kazantsev theory and de- ticular during the formation of the first stars and galax- scribe the small-scale dynamo mathematically. We use ies, where turbulence is released from the gravitational the WKB approximation as a solution of the Kazant- potential well, providing the means to amplify magnetic sev equation, which is valid in the limit of large mag- fields [14]. The same processes may also occur during netic Prandtl number. However, the evolution equation the formation of large-scale structure [15]. In most as- of the magnetic energy turns out to depend on the cor- trophysical objects, the growth rate of the small-scale relation function of the turbulent velocity field. There- dynamo is very high and that can explain the high mag- fore, we present a general model for this correlation func- netic field strengths observed in the primordial as well as tion for turbulence of arbitrary type with v(`) `ϑ in in the present-day Universe. the inertial range. In Sec. III, we calculate the∝ critical The small-scale dynamo was first suggested by Batchelor magnetic Reynolds number for the small-scale dynamo, in 1950 [16]. In 1967, Kazantsev developed a spectral the- Rmcrit. Furthermore, we determine the growth rate Γ ory of this process [17, 18]. He derived a differential equa- of the magnetic energy. The latter is calculated in the tion, the Kazantsev equation, which describes the evolu- limit Pm and for finite magnetic Prandtl num- → ∞ tion of the magnetic energy of the small-scale field. This bers (Pm & 10). In Sec. IV, we discuss our results and equation has been solved for different special cases. For compare them to the literature. We close our work with example, Subramanian [19] used this theory to describe conclusions in Sec. V. the small-scale dynamo in the case of incompressible turbulence, i.e., Kolmogorov turbulence. In the limit of infinite magnetic Prandtl number Pm, which is the ratio of II. ANALYTIC FRAMEWORK the magnetic to the hydrodynamical Reynolds numbers, Rm and Re, he found a growth rate of the small-scale A. Kazantsev theory magnetic energy of (15V )/(4L)√Re [20]. Schekochihin et al. [21] analyzed the Kazantsev theory for an arbitrary compressibility (and spatial dimension), also in the In 1967 Kazantsev set up a theory for describing the limit of large Prandtl number. However, they did not time evolution of the magnetic energy that grows due to give a model for the correlation function of the turbu- turbulent motions of a conducting fluid [18]. The mecha- nism of converting kinetic energy into magnetic energy in 3 this way is known as the turbulent or small-scale dynamo. with the two-point correlation function In this section we describe the Kazantsev theory follow- r r r r ing mainly the formalism proposed by Brandenburg and i j i j Mij(r, t) = δij 2 MN(r, t) + 2 ML(r, t). (8) Subramanian [20] and Subramanian [19]. − r r We will omit the dependencies for a better overview in most of the following equations. 1. The turbulent velocity field As the magnetic field is always divergence-free, i.e., ∂/∂r1i Mij(r, t) = ∂/∂r1j Mij(r, t) = 0, we can derive The theoretical description of turbulence starts with a relation between the transverse and the longitudinal decomposition of the velocity field v into a mean field correlation function similar to (4): v and a turbulent component δv: h i 1 d 2 v = v + δv. (1) MN = r ML , (9) h i 2r dr 2 Following the work of Taylor [35] we model the spatial where we have used that (rirj/r )Mij = ML and appearance of turbulence via the two-point correlation (ri/rj)Mij = MN. The time derivative of δBiδBj is function. The correlation of two turbulent velocity com- h i ponents at the positions r1 and r2 at the times t and s for ∂Mij ∂ = ( δBiδBj ) a Gaussian random velocity field with zero mean, which ∂t ∂t h i is isotropic, homogeneous and δ-correlated in time, is ∂ = ( BiBj Bi Bj ) ∂t h i − h i h i δvi(r1, t)δvj(r2, s) = Tij(r)δ(t s) (2) h i − ∂Bi ∂Bj = Bj + Bi with the two-point correlation function Tij(r), where ∂t ∂t r r1 r2 . It was shown by Batchelor [36] that the ∂ ≡ | − | ( Bi Bj ) . (10) correlation function can be divided into a transverse part −∂t h i h i TN and a longitudinal part TL in the following way: In the upper equation we can substitute the induction rirj rirj equation Tij(r) = δij TN(r) + TL(r). (3) − r2 r2 ∂B We neglect here the effect of helicity, which would appear = ∇ v B η∇ ∇ B, (11) ∂t × × − × × as an additional term in Tij. In the special case of a divergence-free turbulent veloc- where η c2/(4πσ) is the magnetic diffusivity with the ity field (div δv = 0), characteristic of incompressible speed of≡ light c and the electrical conductivity σ, and the fluids, we find that the transverse correlation function is evolution equation of the magnetic mean field connected to the longitudinal one by ∂ B h i = ∇ [ v B η ∇ B ] (12) 1 d 2 eff TN(r) = r TL(r) . (4) ∂t × h i × h i − × h i 2r dr with the effective parameter ηeff = η + TL(0). After a For the other extreme case, an irrotational turbulent ve- lengthy derivation [20] this leads to locity field (rot δv = 0), as expected for purely shock- dominated flows, we find the relation ∂ML 00 4κdiff 0 0 = 2κdiffM + 2 + κ M ∂t L r diff L dT (r) T (r) = r N + T (r). (5) L N 4 TN TL 0 0 dr + T T ML (13) r r − r − N − L

2. Kazantsev equation with

κdiff(r) = η + TL(0) TL(r). (14) Like the velocity field, the magnetic field can be sepa- − rated into a mean field B and a fluctuation part δB: h i The prime denotes differentiation with respect to r. The diffusion of the magnetic correlations, κ , contains in B = B + δB. (6) diff h i addition to the magnetic diffusivity η the scale-dependent B turbulent diffusion TL(0) TL(r). Now let us assume that the fluctuating component δ , − like the velocity field, is a homogeneous, isotropic Gaus- With the solution of Eq. (13) we can calculate MN also sian random field with zero mean. Then we can write the by using the relation (9) and so find the total correlation correlation function as function of the magnetic field fluctuations Mij. We note that this quantity is proportional to the energy density 2 δBi(r1, t)δBj(r2, t) = Mij(r, t) (7) of the magnetic field, B /(8π). h i 4

In order to separate the time from the spatial coordinates The boundary conditions for ψ(r) and θ(x) are we use the ansatz r→0,∞ ψ(r) 0 −−−−−→ 1 2Γt r→±∞ ML(r, t) 2 ψ(r)e . (15) θ(x) 0. (23) ≡ r √κdiff ⇒ −−−−−→ We can make some predictions about the shape of the Substitution of this ansatz in Eq. (13) gives us function θ(x). For very small x (x ), p(x) goes to 1/4 < 0, which leads to exponentially→ −∞ growing and 2 d ψ(r) decaying− solutions of θ. In the other limit (x ), κdiff(r) + U(r)ψ(r) = Γψ(r). (16) 2x → ∞ − d2r − p(x) Γ/η e , we have growing mode solutions only for positive→ − Γ. The boundary conditions require that This is the Kazantsev equation. It formally looks like θ needs to grow exponentially for x and decay the quantum-mechanical Schrödinger equation with a → −∞ 2 exponentially at x . In order to arrange this, p(x) “mass” ~ /(2κdiff) and the “potential” must go through zero,→ ∞ so U(x) needs to become negative 00 0 2 0 for some r. From now on we label the roots of U(x) as κ (κ ) 2κdiff 2T 2(TL TN + κdiff) diff diff N x1 and x2 > x1. As U(r) becomes negative for some r, U(r) + 2 + + − 2 . ≡ 2 − 4κdiff r r r p(r) becomes positive for certain values of r. This means (17) that we have oscillatory solutions for x1 < x < x2. The It describes the kinematic limit, because U is indepen- condition for the eigenvalues Γ in this case is [38] dent of the time. x Z 2 p 2n + 1 p(x0)dx0 = π (24) x1 2

B. WKB approximation for different excitation levels n N. In this work we concentrate on the lowest mode n∈= 0. We can use common methods from quantum mechanics, like the WKB approximation, to solve the Kazantsev 2. Validity of the WKB approximation in general equation (16). WKB stands for Wentzel, Kramers, and Brillouin, who developed this method in 1926 [37]. In order to find the limits in which the WKB method leads to valid solutions of the Kazantsev equation, we derive the differential equation that is solved exactly by 1. Solution of the Kazantsev equation in the WKB approximation Z x2 1 p 0 0 θ(x) = 1/4 exp i p(x )dx . (25) p ± x1 In order to use the standard WKB method, we have The second derivative of θ(x) with respect to x can be to make some substitutions. Let us first introduce a new written as radial coordinate x by defining r ex. This leads to ≡ 00 0 2 00 p 3 (p ) κ (x) d 1 dψ(x) θ (x) + 1 + pθ(x) = 0, (26) diff (Γ + U(x)) ψ(x) = 0. (18) 4p2 − 16 p3 ex dx ex dx − where now the prime denotes d/dx. This equation results Next we eliminate the first-derivative terms through the in the Kazantsev equation (20) if substitution f(x) 1, (27) | | ψ(x) ex/2θ(x), (19) ≡ with p00 3 (p0)2 to obtain f(x) . (28) ≡ 4p2 − 16 p3 d2θ(x) 2 + p(x)θ(x) = 0 (20) We use this result in the Appendix to check the range of dx parameters in which the WKB method produces accurate with the definition solutions of the Kazantsev equation. [Γ + U(x)]e2x 1 p(x) . (21) C. Modeling the turbulent correlation function ≡ − κdiff(x) − 4

The WKB solutions of Eq. (20) are linear combinations We analyze the case of general types of turbulence, of which can be described by the relation between the ve- Z locity v(`) and the size ` of a turbulent ﬂuctuation, 1 p 0 0 θ(x) = exp i p(x )dx . (22) ϑ p1/4 ± v(`) ` . (29) ∝ 5

The power-law index ϑ varies for the different types. It In order to find a general expression for TN we make the attains its minimal value of ϑ = 1/3 for Kolmogorov ansatz theory [27], i.e., incompressible turbulence. For Burgers VL r ϑ+1 turbulence [28], i.e., highly compressible turbulence, ϑ TN(r) = 1 t(ϑ) , (36) gets its maximal value of 1/2 [39]. 3 − L Motivated by the definition of the scale-dependent tur- where t(ϑ) = a bϑ. With Eqs. (34) and (35) we find that bulent diffusion coefficient in the last section, a = 21/5 and b−= 38/5. Furthermore, we find the small- scale transverse correlation (i.e., 0 < r < `c) by steady ηturb(r) = TL(0) TL(r), (30) − continuation. So we end up with the following model for we construct a model for the longitudinal correlation the transverse correlation function for the general slope function of the turbulent velocity field TL(r). The dif- of the turbulent velocity spectrum: fusion coefficient is calculated from the power law (29) in  VL (1−ϑ)/(1+ϑ) r 2 the following way:  1 t(ϑ)Re 0 < r < `c  3 − L  ϑ+1 ϑ ϑ+1 TN(r) = VL r ηturb(r) v`` ` ` = ` . (31) 3 1 t(ϑ) L `c < r < L ∝ ∝  − 0 L < r, So we assume the correlation function in the inertial (37) range to be [19, 40] with t(ϑ) = (21 38ϑ)/5. − VL ϑ+1 The longitudinal and transverse correlation functions de- TL(r) = 1 (r/L) . (32) 3 − pend on the dimensionless parameter y r/L as shown in Fig. 1 for Kolmogorov and Burgers≡ turbulence. We The pre factor VL fixes the units, which should be the choose here a fixed hydrodynamical Reynolds number of same as for diffusivity. V and L are the velocity and the 105. In the inset of Fig. 1 we show a zoom into the dis- length scale of the largest eddies. On the diffusive scale sipative range (0 < r < `c). Furthermore, in Fig. 2 we the correlation function should be steadily continued and plot the potential of the Kazantsev equation, resulting 0 satisfy the condition that its derivative TL(0) vanishes from our model of TL for Kolmogorov and Burgers tur- at r = 0. This is accomplished, for example, for TL bulence. We choose two different values for the Reynolds 2 ∝ r . The exact form of TL in the diffusive range does not number Re and different magnetic Prandtl numbers Pm. affect the results crucially [19]. Furthermore, we expect The magnetic Prandtl number is defined as the ratio of no correlation on scales larger than the largest eddies, so the magnetic Reynolds number Rm (VL)/η and the TL should vanish there. hydrodynamical Reynolds number Re.≡ The potential at Taken all together, we can set up a general turbulence fixed Re and Pm is deeper in the small-scale range in the model for the longitudinal correlation function on the Kolmogorov case than in the Burgers case. For higher different length scales as follows: Re the potential gets deeper and the cutoff scale `c decreases. For higher Pm the potential in the small-scale  VL (1−ϑ)/(1+ϑ) r 2  1 Re 0 < r < `c  3 − L range gets broader.  ϑ+1 TL(r) = VL r 3 1 L `c < r < L  − 0 L < r, III. SOLUTION OF THE KAZANTSEV (33) EQUATION −1/(ϑ+1) where `c = L Re denotes the cutoff scale of the turbulence and L the length of the largest eddies. The In this section we use our model of the turbulent veloc- hydrodynamic Reynolds number Re is defined as (VL)/ν ity correlation function (33) and (37) as the input for the with the typical velocity of the largest eddies V and the Kazantsev theory. We solve the Kazantsev equation in viscosity of the gas ν. order to obtain the characteristic properties of the small- The transverse correlation functions TN for a divergence- scale dynamo. We use the WKB method, which gives free (i.e., Kolmogorov turbulence) and for an irrotational a good approximate solution for large magnetic Prandtl (i.e., Burgers turbulence) turbulent velocity field can be number. In fact, in the limit of infinite magnetic Prandtl derived from the relations (4) and (5). Notice, however, number the WKB approximation is an exact solution of that a turbulent velocity field that is divergence free or the Kazantsev equation. For more details about the va- irrotational in the inertial range does not have to be this lidity of this approximation, see the Appendix. in the diffusive range. We find for the extreme cases in the inertial range (`c < r < L) A. Critical magnetic Reynolds number for VL 5 r 4/3 T K(r) = 1 , (34) small-scale dynamo action N 3 − 3 L VL 2 r 3/2 Intuitively, one expects that the high magnetic diffu- T B(r) = 1 . (35) N 3 − 5 L sivity for very low magnetic Reynolds numbers prevents 6

1.4 500 Kolmogorov turbulence: 3TL/(VL) Kolmogorov turbulence 1.2 Kolmogorov turbulence: 3TN /(VL) 0 Burgers turbulence: 3TL/(VL) Burgers turbulence: 3TN /(VL) -500 (7) (5) 1 `c `c L K B lc lc L -1000 0.8 1 -1500 0.999995 0.6 UL/V -2000 Re = 105, Pm = 102 0.99999 -2500 5 4 0.4 Re = 10 , Pm = 10 correlation function 5 8 0.999985 lK lB Re = 10 , Pm = 10 c c -3000 Re = 107, Pm = 102 0.2 0.99998 Re = 107, Pm = 104 0.0001 -3500 Re = 107, Pm = 108 0 10 0.0001 0.001 0.01 0.1 1 Burgers turbulence y 0

FIG. 1. (Color online) Dependence of the longitudinal and -10 transverse correlation functions T and T on the dimension- L N -20 (7) (5) L less parameter y ≡ r/L for Kolmogorov (ϑ = 1/3) and Burg- `c `c ers (ϑ = 1/2) turbulence. We choose a ﬁxed Reynolds number -30 5

of 10 . The vertical lines indicate the cutoff scale of the tur- UL/V 5 2 bulence `c and the largest scale of the eddies L. Notice that -40 Re = 10 , Pm = 10 the cutoff scale for Kolmogorov turbulence (`K = Re−3/4L) is Re = 105, Pm = 104 c 5 8 B −2/3 -50 Re = 10 , Pm = 10 different from the one for Burgers turbulence (`c = Re L). Re = 107, Pm = 102 The inset shows a zoom of the dissipative range. 7 4 -60 Re = 10 , Pm = 10 Re = 107, Pm = 108 -70 8 6 4 2 0 2 4 amplification of the magnetic field. Even higher diffu- 10− 10− 10− 10− 10 10 10 sivity eventually results in a net decrease of the field y strength. In this section we calculate the critical magnetic Reynolds number Rmcrit for small-scale dynamo FIG. 2. (Color online) Dependence of the potential on the action. To accomplish this we set the growth in our equa- dimensionless parameter y ≡ r/L for Kolmogorov (ϑ = 1/3) tions at zero. and Burgers (ϑ = 1/2) turbulence. We choose two different Reynolds numbers Re = 105 and Re = 107, and differ- It should be noted that we use the inertial range (`c < ent Prandtl numbers Pm = 102, Pm = 104 and Pm = 108. r < L) for determining Rmcrit as the potential in this The cutoff scale `c depends on the turbulence model and the range is always negative and for that we have a positive −3/4 Reynolds number. For Kolmogorov turbulence `c = Re L; growth rate (see Fig. 2). In this range with our turbu- −2/3 for Burgers turbulence `c = Re L. A Reynolds number lence spectrum and Γ = 0 we get for the p-function (21) x (x) 10 is indicated in the cutoff scale as `c . ϑ+1 2 2(ϑ+1) 9/4 a(ϑ)Rmcrity + b(ϑ)Rm y p(y) = − − crit 1 ϑ+12 1 + 3 Rmcrity which means that p(y) needs to go through zero during (38) this transition. So we have our second root at roughly r L and y2 = 1. with a(ϑ) 5/6 (79/30) ϑ + (157/30) ϑ2 and b(ϑ) We≈ can solve Eq. (39) numerically for the critical mag- ≡ − 2 ≡ (14/15) ϑ (103/60) ϑ . netic Reynolds number Rm if we put in a fixed value − crit Now we can evaluate the eigenvalue condition (24) for of ϑ. Recall that ϑ was defined in the inertial range this p(y) in the “ground state” n = 0: of the turbulence via the relation v(`) `ϑ. Results ∝ y for common models in the literature can be found in Z 02 p dy π p(y) = , (39) Table I. In Fig. 3 we show how the critical magnetic y 2 y01 Reynolds number depends on ϑ. Here one can see that the critical magnetic Reynolds number increases rapidly in which the additional y comes from the substitution as ϑ gets closer to its maximum value of 1/2. An em- y = r/L = ex/L. The limits of the integral are the roots pirical fit Rm (ϑ) through these data in the range of p(y). There is only one real and positive root of p(y), crit,fit 0.33 < ϑ < 0.5 is which we label y1. For the upper limit we have to realize 2 that the potential (17) changes for y > 1 to 2η/(yL) , Rmcrit,fit(ϑ) = 88 (tan(2.68ϑ + 0.2) 1) . (40) which is clearly always positive. Furthermore, also the − diffusion coefficient κdiff = η +TL(0) > 0 for y > 1. With Furthermore, we collect the results for common turbu- U and κdiff being positive p(y) is negative in this range, lence models in the literature in Table I. 7

We find that the small-scale dynamo is more easily ex- eral turbulence spectrum (33) in the dissipative range is cited in the case of a purely rotational turbulent veloc- A z4 B z2 45Re(3+7ϑ)/(2+2ϑ) ity field, i.e., for Kolmogorov turbulence, where we find p(z) = 0 0 − − 2 Rmcrit 110. The critical magnetic Reynolds number 20Re1/2 Re(1+3ϑ)/(2+2ϑ) + Re1/(1+ϑ)z2 for a turbulent≈ field with a vanishing rotational component, i.e., Burgers turbulence, is roughly 2700. (42) The results are discussed in Sec. IV. with the definitions

(5+ϑ)/(2+2ϑ) 20 5/2 A0 = Re (163 304ϑ) Re Γ¯, (43) 3000 − − 3 Result 2 20 (2+8ϑ)/(1+ϑ) Rmcrit,ﬁt(ϑ) B0 = (304ϑ 98) Re + Re Γ¯, (44) 2500 − 3 and the normalized growth rate 2000

) L

ϑ ¯

( Γ Γ. (45) 1500 ≡ V crit

Rm In the limit of large Prandtl number z is large, too, and 1000 we can neglect the constant terms. We obtain

−(5+ϑ)/(2+2ϑ) 2 Re A0z B0 500 p(z) = − . (46) 20 z2

0 The one real and positive root of this function is z1 = 0.35 0.4 0.45 0.5 p B0/A0. At the cutoff scale of the turbulence the p ϑ function changes its sign. We take this as our second p (3ϑ−3)/(4ϑ+4) root and so have z2 = Pm/3 Re . So we FIG. 3. (Color online) Dependence of the critical magnetic get for the general eigenvalue condition Reynolds number Rmcrit from the slope of the turbulent ve- r locity spectrum ϑ. The dashed line is an empirical fit through Re−(5+ϑ)/(4+4ϑ) Z z2 A z2 B π our results. 0 0 4− dz = , (47) 2√5 z1 z 2 resulting in the analytical solution of the integral −(5+ϑ)/(4+4ϑ) Re hp p A0ln 2 A0z+ 2√5z B. Growth rate of the small-scale magnetic field i z2 π p 2 p 2 A0z B0 A0z B0 = . (48) − − − z1 2

1. Growth Rate in the Limit Pm → ∞ For z2 1 this becomes −(5+ϑ)/(4+4ϑ) Re p h p In this section we derive a general analytical solution A0 1 ln 4 A0z2 2√5 − for the growth rate Γ for an arbitrary slope of the tur- 1 π bulent velocity spectrum, in the limit of inﬁnite Prandtl + ln (4B0) = . (49) number. 2 2 As the potential has its minimum in the small-scale A zero-order iterative solution for Γ¯ gives us range, i.e., the dissipative range of the turbulence (see 163 304ϑ Fig. 2), the growth rate, which is the eigenvalue of the Γ¯ = − Re(1−ϑ)/(1+ϑ) (50) Kazantsev equation, takes its maximal value there. So 60 − !2 we expect the fastest growing mode to be in the small- π√5 scale range. −(5+ϑ)/(4+4ϑ) , Re 1 ln 4√A0z2 + 1/2 ln (4B0) In order to have scale-independent equations, we intro- − duce the substitution which becomes for large Prandtl number

!1/2 !1/2 ¯ 163 304ϑ (1−ϑ)/(1+ϑ) V √Re Re3/2Pm Γ = − Re . (51) z r = y, (41) 60 ≡ 3Lη 3 As a result we get for the absolute growth rate Γ for a general slope of the turbulent velocity spectrum where the magnetic Prandtl number is Pm=Rm/Re. (163 304ϑ) V Γ = − Re(1−ϑ)/(1+ϑ) (52) The p function in z space [see (21) and (41)] for the gen- 60 L 8

100 105

104 Re = 108 10 103

102 ¯ ¯ Γ Γ Re = 104 101 1 K41: ϑ = 1/3 SL94: ϑ = 0.35 Re = 102 BNP02: ϑ = 0.37 100 L81: ϑ = 0.38 FRKSM10(sol): ϑ = 0.43 1 FRKSM10(comp),OM02: ϑ = 0.47 10− 0.1 B48: ϑ = 1/2 2 10− 1 0 1 2 3 4 5 6 1 2 3 4 10− 10 10 10 10 10 10 10 10 10 10 10 Re Pm

FIG. 4. (Color online) The normalized growth rate of the FIG. 5. (Color online) Dependence of the normalized growth small-scale dynamo in the limit of infinite magnetic Prandtl rate of the small-scale dynamo on the magnetic Prandtl num- number, depending on the Reynolds number Re. For the ber Pm for Kolmogorov turbulence. We choose different val- slopes of the turbulent velocity spectrum ϑ we choose com- ues of the Reynolds number. Notice that in the limit Pm → ∞ mon values from the literature: K41 [27], SL94 [29], BNP02 the normalized growth rates are Γ¯ = 10.28 for Re = 102, [30], L81 [31], FRKSM10 [32] (sol: solenoidal forcing; comp: Γ¯ = 102.78 for Re = 104, and Γ¯ = 10277.78 for Re = 108. compressive forcing), OM02 [33] and B48 [28]. These limits are indicated in the plot as horizontal lines. in the limit Pm . In Fig. 4 we show the dependency → ∞ ¯ of the normalized growth rate Γ on the Reynolds num- 100 ber for different types of turbulence. One extreme case is incompressible turbulence, i.e., Kolmogorov turbulence, Re = 108 1/2 with Γ¯ Re . In the other extreme case, highly com- 10 pressible∝ turbulence, i.e., Burgers turbulence, the growth ¯

1/3 Γ rate increases only as Re . Altogether we ﬁnd that the Re = 104 growth rate increases faster with the Reynolds number 1 when the compressibility is lower. Re = 102 0.1

2. Growth rate as a function of the Prandtl number 0.01 3 4 5 6 7 8 9 10 11 12 In this section we discard the assumption of infinite 10 10 10 10 10 10 10 10 10 10 Prandtl number. In this case we have to solve the full Pm equation resulting from the WKB method (24), FIG. 6. (Color online) Dependence of the normalized growth p rate of the small-scale dynamo on the magnetic Prandtl num- Z p(z) π dz = , (53) ber Pm for Burgers turbulence. We choose different values of z 2 the Reynolds number. Notice that in the limit Pm → ∞ the normalized growth rates are Γ¯ = 0.85 for Re = 102, Γ¯ = 3.95 with p(z) from (42). There is no analytical solution of for Re = 104, and Γ¯ = 85.1 for Re = 108. These limits are this integral equation. indicated in the plot as horizontal lines. The numerical results of the normalized growth rate are shown in Fig. 5 for Kolmogorov turbulence and in Fig. 6 for Burgers turbulence. We plot the normalized growth small-scale dynamo given by Fig. 3 and (52). In Table I, rate depending on the Prandtl number for different values we collect our results for typical turbulence models from of the Reynolds number. the literature, defined via the relation v(`) `ϑ. The two extrema of turbulence are Kolmogorov turbulence∝ [27], i.e., incompressible turbulence, with ϑ = 1/3 and Burg- IV. DISCUSSION ers turbulence with ϑ = 1/2, which describes highly compressible turbulence with vanishing rotational component The main results of this work are the critical magnetic [28]. Between those extreme values we choose values of Reynolds number Rmcrit and the growth rate Γ of the ϑ from observations of molecular clouds, like ϑ 0.38 ≈ 9 from Larson [31] and ϑ 0.47 from Ossenkopf and Mac simulation [43]. In resolution studies, Sur et al. [12] and Low [33]. Furthermore,≈ we give ϑ 0.35 as an example Federrath et al. [13] found that the typical length of a for a theoretical model of intermittency≈ [29]. Numerical turbulent fluctuation needs to be resolved with at least experiments give ϑ 0.37 for driven supersonic magne- 30 grid cells in magnetohydrodynamic simulations of a tohydrodynamical (MHD)≈ turbulence [30] and ϑ 0.43 self-gravitating gas. Only then is the magnetic field am- and ϑ 0.47 for solenoidal and compressive forcing≈ of plified exponentially, which is explained by the action of the turbulence≈ [32]. Notice, however that the mean val- the small-scale dynamo. ues of ϑ from observations and simulations have a typical For a physical interpretation of this result it is useful to uncertainty of 10%. take the stretch-twist-fold-dynamo as a toy-model of the turbulent dynamo. In this picture we think of a magnetic flux rope that gets stretched due to turbulent motions. A. Critical magnetic Reynolds number This motion decreases the cross section of the flux rope. If we assume that the magnetic flux is frozen in the gas, then the magnetic field increases during this process, be- We have evaluated Eq. (39) for different slopes of the cause the magnetic flux is a conserved quantity [20, 41]. turbulent velocity spectrum ϑ. The results for common This process works best in a purely rotational turbulent types of turbulence from the literature are listed in Table velocity field. Therefore, we expect the dynamo to be I and pictured in Fig. 3. more easily excited in Kolmogorov turbulence. In order From these results we see that for all types of turbulence to see this process in simulations, one needs to resolve the a high magnetic Reynolds number needs to be exceeded stretching, twisting, and folding of the field lines, which for small-scale dynamo action. In astrophysical objects explains the required high resolution. we often find very high magnetic Reynolds numbers (see The determination of the critical magnetic Reynolds the compilation in [41]). The core of Jupiter, for example, number is, moreover, the first step to understanding the has Rm 106, the solar convection zone has Rm 108, ≈ 12 ≈ saturation of the small-scale dynamo. For if the magnetic and the solar corona already Rm 10 . In the interstel- field in a system increases, back reactions from the gas lar medium we find Rm 1017 ≈and in a typical galaxy 19 ≈ become more important. Then processes like ambipo- about 10 . Consequently, the critical magnetic Reynolds lar diffusion can change the properties of the gas and the number is exceeded by far in nature, and we expect that magnetic Reynolds number can decrease. If the magnetic a small-scale dynamo is excited in typical astrophysical Reynolds number becomes smaller than the critical mag- objects. netic Reynolds number, the magnetic field stops growing There are different ways to obtain approximate solutions and the small-scale dynamo is saturated. of the Kazantsev equation. In addition to the WKB method there is also an asymptotic solution, which uses the separation of scales [24]. The potential U(r) and the mass m(r) are estimated on different scales, such that a B. Growth rate solution of the Kazantsev equation can be found. Ro- gachevskii and Kleeorin [22] used this method to deter- In Fig. 4 as well as in Table I, we present our results mine the critical magnetic Reynolds number for differ- for the growth rate of the small-scale dynamo in the limit ent compressibilities of turbulence. They found that, in of infinite magnetic Prandtl number. Our results show the limit of small magnetic Prandtl number Rm needs that the growth rate is proportional to the velocity V of to be larger than roughly 400 for excitation of a small- the largest eddy divided by its length L. The ratio V/L scale dynamo in the case of Kolmogorov turbulence. For is the reciprovcal of the turnover time of an eddy. Thus, a larger compressibility, i.e., toward Burgers turbulence, the growth rate increases with decreasing turnover time, they found that Rmcrit increases sharply. We see the and the smallest modes grow at the highest rate. This is same trend of increasing Rmcrit for higher compressibil- expected, because smaller turnover times lead to a faster ity. However, our result for Kolomogorov turbulence dif- tangling of the magnetic field lines. fers by a factor of roughly 4, which might be caused Furthermore, the growth rate increases with increasing by the fact that we analyze the limit of large magnetic hydrodynamical Reynolds number for all types of turbu- Prandtl number. lence, characterized by v(`) `ϑ. In order to achieve Recent high-resolution numerical studies confirm the ex- the same growth rate for Kolmogorov∝ and Burgers tur- istence of a critical magnetic Reynolds number for small- bulence we have to provide a larger Reynolds number scale dynamo action. Haugen et al. [42] found Rmcrit in the latter case. Assuming a fixed Reynolds numbers ≈ K B 35 for subsonic turbulence and Rmcrit 70 for super- in both extreme cases, Re and Re , the growth rates sonic turbulence at a magnetic Prandtl number≈ of about of the two different turbulence types are the same for unity. In numerical simulations, the magnetic Reynolds ReK 0.18(ReB)3/2. This fact can again be explained ϑ+1 ≈ number can be estimated by Rm (λ/`c) , where λ is with the stretch-twist-fold model (see Sec. IV A). We ≈ the typical size of turbulent structures and `c is the cut- need solenoidal modes, i.e., divergence-free modes, of off scale of the turbulence. The latter can be estimated the turbulence for this process [34], which explains why by `c 0.5 ∆x with ∆x the minimal resolved size in a incompressible turbulence amplifies the magnetic field ≈ 10

Model and reference ϑ Rmcrit Γ¯ (Pm → ∞)

37 1/2 Kolmogorov [27] 1/3 ≈ 107 36 Re Intermittency of Kolmogorov turbulence (She and Leveque [29]) 0.35 ≈ 118 0.94 Re0.48

Driven supersonic MHD-turbulence (Boldyrev et al. [30]) 0.37 ≈ 137 0.84 Re0.46

Observation in molecular clouds (Larson [31]) 0.38 ≈ 149 0.79 Re0.45

Solenoidal forcing of the turbulence (Federrath et al. [32]) 0.43 ≈ 227 0.54 Re0.40

Compressive forcing of the turbulence (Federrath et al. [32]), 0.47 ≈ 697 0.34 Re0.36 Observations in molecular clouds (Ossenkopf and Mac Low [33]) 11 1/3 Burgers [28] 1/2 ≈ 2718 60 Re

TABLE I. The critical magnetic Reynolds number Rmcrit and the normalized growth rate of the small-scale dynamo Γ¯ in the limit of infinite magnetic Prandtl number. We show our results for different types of turbulence, which are characterized by the exponent ϑ of the slope of the turbulent velocity spectrum, v(`) ∝ `ϑ. The extreme values of ϑ are 1/3 for Kolmogorov turbulence and 1/2 for Burgers turbulence. more effectively. plicable (see the Appendix). However, the trend is that With the asymptotic solution of the Kazantsev equation, the growth rate decreases for lower Prandtl numbers and Rogachevskii and Kleeorin [22] found in the limit of small this can explain the lower growth rates from the simu- magnetic Prandtl number Γ ln(Rm/Rmcrit). The con- lations. Yet the ratio of the growth rate of turbulence stant of proportionality depends∝ on the amount of com- driven by solenoidal and compressive forcing is in both pressibility. In a later work Kleeorin and Rogachevskii cases about 2 (our model: Γ¯sol/Γ¯comp 2.1; Waagan et [23] found that this logarithmic scaling of the growth al. [44]: Γ¯sol/Γ¯comp 2.1), which shows≈ that incompress- rate is valid only in the vicinity of the threshold of small- ible turbulence is more≈ efficient in amplifying a magnetic scale dynamo excitation. For magnetic Reynolds num- field via the small-scale dynamo. bers much larger than Rmcrit, they found in the limit Furthermore, Federrath et al. [34] have presented a study of small magnetic Prandtl number Γ Rm1/2 for Kol- of the Mach number dependency of the growth rate of mogorov turbulence. As for a constant∝ magnetic Prandtl the small-scale dynamo, where they compared solenoidal number Rm Re, this agrees with our result. with compressive forcing of the turbulence. They found There are recent∝ high-resolution numerical simulations that, for low Mach numbers, the ratio of the growth rate that model the turbulent dynamo. The two limiting ways of turbulence driven by solenoidal and compressive forc- of driving the turbulence are solenoidal, i.e., divergence- ing is about 30 (for Mach number M = 0.1: Γ¯sol 1.2, comp ≈ free, forcing and compressive, i.e., rotation-free, forc- Γ¯ 0.04). However, for higher Mach numbers their ≈ sol comp ing. These simulations show, in agreement with our calculations also result in a ratio of Γ¯ /Γ¯ 2 (for sol comp ≈ study, that solenoidally driven turbulence leads to larger Mach number M = 10: Γ¯ 0.7, Γ¯ 0.3), which ≈ ≈ growth rates of the small-scale dynamo. Waagan et al. is in agreement with our results. The lower growth rates [44] found, using a Reynolds number of about 1500 [45], in the simulation again may come from low Prandtl num- and a magnetic Prandtl number of about 1, for totally bers in the simulations, of the order of unity. However, solenoidal forcing of the turbulence Γ¯sol = 0.60 and for a great uncertainty is the Reynolds number in numerical totally compressive forcing Γ¯comp = 0.28. These values of simulations, which is only a crude estimate. the growth rate are about a factor of 17 lower than those from our model (with Re = 1500), Γ¯sol = Γ¯ϑ=0.43 10.07 and Γ¯comp = Γ¯ϑ=0.47 4.73. This can be explained≈ by V. SUMMARY AND CONCLUSIONS the fact that the simulations≈ had a very low magnetic Prandtl number of about 1. However, our result for the growth rate in Table I was derived with the assumption We have presented an analytical treatment of the of infinite Prandtl number. We have also explored the small-scale dynamo, using the Kazantsev theory. For range of lower Prandtl numbers. The result is presented that we have modeled the correlation function of the in Fig. 5 for Kolmogorov turbulence and in Fig. 6 for turbulent velocity field, depending on the slope of the ϑ Burgers turbulence. But with our model, we can make turbulent velocity spectrum, ϑ, in v(`) ` . With this ∝ no predictions for Prandtl numbers around unity, because model, we solved the Kazantsev equation in the WKB in this range the WKB approximation is no longer ap- approximation and tested the validity of this approximation. We determined the critical magnetic Reynolds 11 number for the small-scale dynamo and its growth rate in under Grant No. KL 1358/14-1. D.R.G.S. acknowl- the case of infinite and finite magnetic Prandtl numbers. edges funding through the SPP 1573 (Project No. The main results of our work are as follows: SCHL 1964/1-1) and the SFB 963 “Astrophysical Flow Instabilities and Turbulence”. C.F. acknowl- The critical magnetic Reynolds number Rmcrit for • edges funding from the Australian Research Council the small-scale dynamo increases as the exponent ϑ (Grant No. DP110102191) and from the European Re- increases (see Fig. 3). For Kolmogorov turbulence search Council (FP7 Grant Agreement No. 247060). K B Rmcrit 110 and for Burgers turbulence Rmcrit C.F., R.B., and R.S.K. acknowledge subsidies from the ≈ ≈ 2700. Baden-Württemberg-Stiftung under Research Contract No. P-LS-SPII/18 and from the German Bundesminis- The growth rate of the magnetic field energy in the terium fürBildung und Forschung via the ASTRONET • limit of infinite magnetic Prandtl number is project STAR FORMAT (Grant No. 05A09VHA). R.S.K (163 304ϑ) V also thanks the DFG for financial support via Grants Γ = − Re(1−ϑ)/(1+ϑ) (54) No. KL1358/10 and No. KL1358/11, as well as via the 60 L SFB 881 “The Milky Way System”. R.B. acknowl- (see also Figure 4). edges funding by the Emmy-Noether Grant No. (DFG) BA 3706 and the DFG via the Grands No. BA 3706/1-1 For decreasing magnetic Prandtl number the and No. BA 3706/3-1. • growth rate decreases. The details of this drop depend on the type of turbulence (see Figs. 5 and 6). APPENDIX: VALIDITY OF THE WKB APPROXIMATION A validity test shows that the WKB-approximation • gives exact solutions in the limit of infinite mag- The WKB method is only an approximate solution of netic Prandtl number. The approximation breaks the Kazantsev equation. We have derived condition (27), down at a Prandtl number of around unity (see f 1, for which the WKB-method is valid in order to Figs. 7 and 8). find| | solutions. In z space, f reads

With these results we are able to make predictions about z2p00(z) + 2zp0(z) 3 [zp0(z)]2 the first magnetic fields in the Universe. As the hydro- f(z) 2 3 . (55) dynamical Reynolds numbers are typically very high in ≡ 4p(z) − 16 p(z) astrophysical objects, the growth rates of the small-scale However, we have seen that the magnetic field is ampli- p dynamo are very high, too. We thus expect that the fied most strongly on the scale `c(z) = Pm/3, as here time until saturation is shorter than the collapse time of the potential U has its minimum. So we analyze f(z, Γ) a halo [13]. For that reason we might already have high on this scale and get a dependency on the Prandtl num- magnetic field strengths even before the formation of the ber Pm. Hence we label f(`c, Γ) f(Pm, Γ). first stars, the first galaxies, and the first galaxy clusters. One can show that f(Pm, Γ) vanishes≡ in the limit of large Turbulence and magnetic fields are key ingredients of Prandtl number for all Γ and all turbulence types, current star formation theory [46, 47]. Magnetic fields drive jets and outflows from young stars. Stellar winds lim f(Pm, Γ) = 0. (56) and supernova explosions, which end the lives of mas- Pm→∞ sive stars, enrich the interstellar medium with heavy el- This means that the WKB method is very good in the ements forged in the stellar interior. These processes are limit of large magnetic Prandtl number. crucial for the chemical composition of the Universe, determining cooling and heating processes in the gas. This, in turn, is very important for the formation of the next generation of stars. The momentum from jets and out- 1. Validity of the WKB approximation for flows around accreting protostars may disperse some of Kolmogorov turbulence the envelope material that otherwise would fall onto the central star. Thus, they are important ingredients for In order to check also lower Prandtl numbers we plot our understanding of the physical origin of the observed f(Pm, Γ) for different normalized growth rates Γ¯ (45) and distribution of stellar masses [48]. Kolmogorov turbulence in Fig. 7. However, one can show that f(Pm, Γ) does not depend on the Reynolds number for Kolmogorov turbulence. So we choose values for Γ¯ be- ACKNOWLEDGMENTS tween 0 and the maximal value Γ¯max for the plot in Fig. 7, where Γ¯max is the value for an infinite Prandtl number J.S. acknowledges financial support by the Deutsche and depends on the Reynolds number. One can see that Forschungsgemeinschaft (DFG) in the Schwerpunktpro- the critical Prandtl number for the WKB approximation gramm SPP 1573 “Physics of the Interstellar Medium” gets larger with increasing normalized growth rate. 12

growth rate: 0.15 2 Pm(Re = 10 ) & 500, (58) 0.1 4 Γ¯ = 0 Pm(Re = 10 ) & 1100, (59) ¯ 8 0.2 Γmax Pm(Re = 10 ) & 5100. (60) 0.05 0.5 Γ¯max 0.9 Γ¯max 0.99 Γ¯ 0 max (Pm) f 0.15 -0.05

0.1 -0.1 Γ¯ = 0 0.2 Γ¯max 0.05 0.5 Γ¯max -0.15 0.9 Γ¯max 0.99 Γ¯max 2 1 0 1 2 3 4 5 6 7 0 − − (Pm) f log10(Pm) -0.05 Re = 102 FIG. 7. (Color online) The function f(Pm, Γ)¯ for different values of the normalized growth rate for Kolmogorov turbu- -0.1 lence. Γ¯max is the normalized growth rate in the limit of in- 1/2 -0.15 finite magnetic Prandtl numbers, Γ¯max = (37/36)Re (see Sec. III for the derivation). The WKB approximation is valid 0.15 within the non-hatched area, i.e., for |f(Pm, Γ)| < 0.1. 0.1 Γ¯ = 0 0.2 Γ¯max ¯ To make a more quantitative estimate of the criti- 0.05 0.5 Γmax 0.9 Γ¯max ¯ cal Prandtl number, we have hatched the area above 0.99 Γmax 0 f(Pm, Γ)¯ = 0.1 and below f(Pm, Γ)¯ = 0.1. When f (Pm) − f is not in this area its absolute value is smaller than 10% -0.05 of 1. We take this as a threshold for our approximation. Re = 104 ¯ We find that our method is applicable in the case of Γ = 0 -0.1 for -0.15

Pm & 13. (57) 0.15

0.1 For higher normalized growth rates the critical Prandtl Γ¯ = 0 ¯ number increases. 0.2 Γmax 0.05 0.5 Γ¯max 0.9 Γ¯max 0.99 Γ¯max 0 (Pm) f 2. Validity of the WKB approximation for Burgers -0.05 8 turbulence Re = 10 -0.1

We can analyze the validity of the WKB solutions for -0.15 Burgers turbulence in the same way as for Kolmogorov turbulence using criterion (27). -2 0 2 4 6 8 10 However, we find that the function f given in (55) now log10(Pm) depends not only on the normalized growth rate Γ¯ and the Prandtl number Pm, but also on the Reynolds num- FIG. 8. (Color online) The function f(Pm, Γ)¯ for fixed ber Re. The result is shown in Fig. 8, where we plot f Reynolds numbers and different values of the normalized against the Prandtl number for different Reynolds num- growth rate for Burgers turbulence. Notice that in the limit bers and different normalized growth rates. Pm → ∞ the normalized growth rates are Γ¯ = 0.85 for 2 ¯ 4 ¯ 8 We again determine the critical Prandtl number for the Re = 10 , Γ = 3.95 for Re = 10 , and Γ = 85.1 for Re = 10 . WKB method for a vanishing normalized growth rate. The WKB approximation is valid within the non-hatched area, i.e., for f(Pm, Γ) < 0.1. For our different values of the Reynolds number we get the following critical Prandtl numbers at vanishing 13

[1] M. L. Bernet, F. Miniati, S. J. Lilly, P. P. Kronberg, [27] A. Kolmogorov, Dokl. Akad. Nauk SSSR 30, 301 (1941). and M. Dessauges-Zavadsky, Nature (London) 454, 302 [28] J. Burgers, in A Mathematical Model Illustrating the The- (2008). ory of Turbulence, edited by R. V. Mises and T. V. [2] F. Tavecchio, G. Ghisellini, L. Foschini, G. Bonnoli, Kármán,Advances in Applied Mechanics, Vol. 1 (Else- G. Ghirlanda, and P. Coppi, Mon. Not. R. Astron. Soc. vier, Amsterdam, 1948) pp. 171–199. 406, L70 (2010). [29] Z.-S. She and E. Leveque, Phys. Rev. Lett. 72, 336 [3] A. Neronov and I. Vovk, Science 328, 73 (2010). (1994). [4] A. M. Taylor, I. Vovk, and A. Neronov, Astron. & As- [30] S. Boldyrev, A.˚ Nordlund, and P. Padoan, Astrophys. J. trophys. 529, A144 (2011). 573, 678 (2002). [5] D. G. Yamazaki, K. Ichiki, T. Kajino, and G. J. Math- [31] R. B. Larson, Mon. Not. R. Astron. Soc. 194, 809 (1981). ews, Astrophys. J. 646, 719 (2006). [32] C. Federrath, J. Roman-Duval, R. S. Klessen, [6] 1 Mpc ≈ 3.1 × 1022 m. W. Schmidt, and M.-M. Mac Low, Astron. & Astrophys. [7] D. R. G. Schleicher and F. Miniati, Mon. Not. R. As- 512, A81 (2010). tron. Soc. (2011), submitted (arXiv:1108.1874). [33] V. Ossenkopf and M.-M. Mac Low, Astron. & Astrophys. [8] D. R. G. Schleicher, D. Galli, S. C. O. Glover, R. Baner- 390, 307 (2002). jee, F. Palla, R. Schneider, and R. S. Klessen, Astrophys. [34] C. Federrath, G. Chabrier, J. Schober, R. Banerjee, R. S. J. 703, 1096 (2009). Klessen, and D. R. G. Schleicher, Phys. Rev. Lett. 107, [9] M. S. Turner and L. M. Widrow, Phys. Rev. D 37, 2743 114504 (2011). (1988). [35] G. I. Taylor, Proc. R. Soc. London, Ser. A 151, 421 [10] G. Sigl, A. V. Olinto, and K. Jedamzik, Phys. Rev. D (1935). 55, 4582 (1997). [36] G. K. Batchelor, The Theory of Homogeneous Turbu- [11] R. Beck, Space Science Rev. 99, 243 (2001). lence, Cambridge Monographs on Mechanics and Applied [12] S. Sur, D. R. G. Schleicher, R. Banerjee, C. Federrath, Mathematics (Cambridge University Press, Cambridge, and R. S. Klessen, Astrophys. J. Lett. 721, L134 (2010). 1953) pp. XI, 197 S. [13] C. Federrath, S. Sur, D. R. G. Schleicher, R. Banerjee, [37] H. A. Kramers, Z. Phys. 39, 828 (1926). and R. S. Klessen, Astrophys. J. 731, 62 (2011). [38] L. Mestel and K. Subramanian, Mon. Not. R. As- [14] D. R. G. Schleicher, R. Banerjee, S. Sur, T. G. Arshakian, tron. Soc. 248, 677 (1991). R. S. Klessen, R. Beck, and M. Spaans, Astron. & As- [39] W. Schmidt, C. Federrath, M. Hupp, S. Kern, and J. C. trophys. 522, A115 (2010). Niemeyer, Astron. & Astrophys. 494, 127 (2009). [15] D. Ryu, H. Kang, J. Cho, and S. Das, Science 320, 909 [40] S. I. Vainshtein, JETP 83, 161 (1982). (2008). [41] S. Childress and A. D. Gilbert, Stretch, Twist, Fold: The [16] G. K. Batchelor, Proc. R. Soc. London, Ser. A 201, Fast Dynamo, Lecture Notes in Physics : New series m, 405 (1950). monographs ; 37 ; Lecture notes in physics / New series [17] A. P. Kazantsev, JETP 53, 1806 (1967). M No. 37 (Springer, Berlin ; Heidelberg [u.a.], 1995) pp. [18] A. P. Kazantsev, JETP 26, 1031 (1968). XI, 406 S., literaturverz. S. 381 - 396. [19] K. Subramanian, arXiv:astro-ph/9708216 (1997). [42] N. E. Haugen, A. Brandenburg, and W. Dobler, Phys. [20] A. Brandenburg and K. Subramanian, Phys. Rep. 417, Rev. E 70, 016308 (2004). 1 (2005). [43] R. Benzi, L. Biferale, R. T. Fisher, L. P. Kadanoff, D. Q. [21] A. A. Schekochihin, S. A. Boldyrev, and R. M. Kulsrud, Lamb, and F. Toschi, Phys. Rev. Lett. 100, 234503 Astrophys. J. 567, 828 (2002). (2008). [22] I. Rogachevskii and N. Kleeorin, Phys. Rev. E 56, 417 [44] K. Waagan, C. Federrath, and C. Klingenberg, Journal (1997). of Computational Physics 230, 3331 (2011). [23] N. Kleeorin and I. Rogachevskii, arXiv:astro- [45] Waagan et al. [44] give a magnetic Reynolds number of ph/1112.3926 (2011). about 200. However, these ideal MHD-simulations were [24] Y. B. Zeldovich, A. A. Ruzmaikin, and D. D. Sokoloff, later calibrated with resistive non-ideal MHD-simulations The Almighty Chance (World Scientific, Singapore, in Reference [34] showing that the Reynolds number is 1990). about 1500. [25] N. Kleeorin, I. Rogachevskii, and D. Sokoloff, Phys. Rev. [46] M.-M. Mac Low and R. S. Klessen, Rev. Mod. Phys. 76, E 65 (2002), 10.1103/PhysRevE.65.036303. 125 (2004). [26] N. Kleeorin and I. Rogachevskii, Phys. Rev. E 77 (2008), [47] C. F. McKee and E. C. Ostriker, Annu. Rev. Astron. As- 10.1103/PhysRevE.77.036307. trophys. 45, 565 (2007). [48] G. Chabrier, Publ. Astron. Soc. Pac. 115, 763 (2003).