DRAFTVERSION Preprint typeset using LATEX style emulateapj v. 12/16/11

THE INITIAL CONDITIONS FOR PLANET FORMATION : TURBULENCE DRIVEN BY HYDRODYNAMICAL INSTABILITIES IN DISKS AROUND YOUNG STARS.

WLADIMIR LYRA1,2 AND ORKAN M.UMURHAN3,4,F (Dated: Received ; Accepted) Draft version

ABSTRACT This review examines recent theoretical developments in our understanding of turbulence in cold, non-magnetically active, planetesimal forming regions of protoplanetary disks which we refer to throughout as “Ohmic zones". We give a brief background introduction to the subject of disk turbu- lence followed by a terse pedagogical review of the phenomenology of hydrodynamic turbulence. The equations governing the dynamics of cold astrophysical disks are given and basic flow states are described. We discuss the Solberg-Høiland conditions required for stability, and the three recently identified turbulence generating mechanisms possibly active in protoplanetary disk Ohmic zones, namely, (i) the Vertical Shear Instability, (ii) The Convective Overstability and (iii) the Zombie Vortex Instability. We summarize the properties of these processes, identify their limitations and discuss where and under what conditions these processes are active in protoplanetary disk Ohmic zones.

1. INTRODUCTION huge sample space, some of the educated guesses of the Planet formation is simultaneously one of the old- time are bound to contain some truth. est and one of the newest concerns of human inquiry. By the 18th century, Newtonian gravity and the orbits “How did the Earth come to be?” is a question that al- of the planets were understood in enough detail to re- most invariably appears in the cosmogonies of the an- alize that the low inclinations of the orbits implied that cients. They not always had a clear idea of what “Earth” the easiest way to attain that configuration was if the meant, but this is a question that, in one form or another, planets have formed in a disk that orbited the proto-Sun virtually every society in recorded history has at some (Kant 1755). Because Jupiter and Saturn are gas giant point asked itself. Particularly interesting are the ideas planets, this disk must have been a disk of gas. Early of Leucippus (480-420? B.C.E.) who, according to testi- mathematical considerations by Laplace (1796) applied monial, is to have said (Diels & Kranz 1961) Newton’s theory of universal gravitation and laws of motion to a slowly rotating spherical cloud, implying The worlds come into being as follows: many that it should collapse under its own weight. Due to bodies of all sorts and shapes move from the in- conservation of angular momentum, the gas settles into finite into a great void; they come together there a flat disk orbiting the condensing proto-Sun in the cen- and produce a single whirl, in which, colliding ter. In this solar nebula, planets are taking shape. with one another and revolving in all manner of C. F. von Weizsäcker extended these fundamental no- ways, they begin to separate like to like. tions and pointed out that eddies in the forming solar nebula ought to increase with distance from the Sun Diogenes Laertius IX, 31 6 prefiguring the role these may have in the formation of planetesimals out of dust (Gamow & Hynek 1945). This vision strikes surprisingly modern, and not with- Rather presciently, Weizsäcker also reasoned that many out foundation within the modern theory of planet other star systems should harbor similar kinds of nebu- formation. Substitute “many bodies of all sorts and lae like our own Sun did after its birth. shapes” by gas and dust, then “single whirl” by protoplan- Modern searches have now revealed such disks etary disk and finally “revolving in all manner of ways” around young stars (Elsasser & Staude 1978; Rucinski by turbulence, and it could have figured in the introduc- 1985; Aumann 1985; Sargent & Beckwith 1987; Strom arXiv:1808.08681v3 [astro-ph.EP] 24 Nov 2018 tion of a paper in the latest issue of a major astronomy et al. 1989; Beckwith et al. 1990; O’dell & Wen 1994; Mc- journal. This attests not to clairvoyance of the ancient Caughrean & O’dell 1996; Ricci et al. 2008), now called Greeks, but to the antiquity of the question. Given the protoplanetary disks or, in symmetry with the solar neb- ula, extrasolar nebulae or exonebulae. In general, proto- 1 California State University, Northridge. Department of Physics and Astronomy 18111 Nordhoff St, Northridge, CA 91330, planetary disks range in radius from 10s to 100s of AU [email protected]. (e.g. Ansdell et al. 2018), in density from 1013 to 1015 2 Jet Propulsion Laboratory, California Institute of Technology, cm−3 (for mean molecular weight µ = 2.5), in mass from 4800 Oak Grove Drive, Pasadena, CA, 91109, [email protected]. −3 −1 3 NASA Ames Research Center, Space Sciences Division, Plane- 10 to 10 M , and from 1000 K near the star to 10 K in tary Sciences Branch, Moffatt Field, CA 94035 the outskirts of the disk. These disks form while the star 4 SETI, Carl Sagan Center, 190 Bernardo Way, Mountain View, CA is being formed, as a consequence of the cloud’s grav- 94043 itational collapse, and here we already find one of the F Corresponding Author: [email protected] first formidable problems of star and planet formation: 6 Scholars of the classical period note that nothing but third person accounts survive of Leucippus’ words. as interstellar clouds are huge in size, even the slightest 2 Lyra & Umurhan rotation means far too much angular momentum (Mes- lar value of 10−2. It was then hypothesized (Safronov tel 1965a,b). Indeed, if a proto-star was to accommodate 1972) that due to the high densities of this midplane all the angular momentum in a ring at 1AU containing layer, the solids could collectively achieve critical num- about 10% of its mass, it would achieve break-up ve- ber density and undergo direct gravitational collapse. locity, a problem that also exists in similar form for the Such a scenario has the advantage of occurring on very gas giant planets in the Solar System (Takata & Steven- rapid timescales, thus avoiding the radial drift barrier. son 1996; Bryan et al. 2018; Batygin 2018). In order to This picture was nonetheless shown to be simplis- accrete, the gas must somehow find a way to transfer tic, in the view that even low levels of turbulence in its angular momentum and if a gas parcel does this in the disk preclude the midplane layer of solids from nearly circular orbits, it must simultaneously find a way achieving densities high enough to trigger the gravi- to lose energy in the process. tational instability (Weidenschilling 1980). Even in the Because gravity is a radial force, and an axisymmet- absence of self-sustained turbulence such as the one ric disk has no pressure forces in the azimuthal direc- generated by the MRI, the solids themselves can gen- tion, the only process that produces change in angu- erate turbulence due to the backreaction of the drag lar momentum is viscous stresses. Angular momentum force onto the gas. Such turbulence can be brought transport is mediated by viscosity; without it, gas will about by Kelvin-Helmholtz instabilities due to the ver- simply orbit the star, not changing its radial position, tical shear the dense layer of solids induces on the gas and star formation will not complete. Although molec- (Weidenschilling 1980; Weidenschilling & Cuzzi 1993; ular viscosity is much too small to account for the ob- Sekiya 1998; Johansen et al. 2006), or by streaming in- served mass accretion rate, it was recognized (Shakura stabilities induced by the radial migration of solids par- & Sunyaev 1973; Lynden-Bell & Pringle 1974) that the ticles (Youdin & Goodman 2005; Paardekooper 2006; stochastic behavior of turbulence would lead to diffu- Youdin & Johansen 2007; Johansen & Youdin 2007). In sion of momentum, similarly to viscosity. Moreover, the turbulent motion, the solids are stirred up by the while molecular viscosity acts in the small scales of the gas, forming a vertically extended layer where the stel- flow, turbulence acts all the way up to the integral scale, lar gravity is balanced by turbulent diffusion (Dubrulle generating much more powerful stresses. The question et al. 1995; Garaud & Lin 2004). But if turbulence pre- of angular momentum transport is thus a question of cludes direct gravitational collapse through sedimenta- how the disk becomes turbulent. tion, it was also shown that it allows for it in an indirect The 1970s and 1980s saw several possible candidates way. As solid particles concentrate in high pressure re- to generate turbulence and angular momentum trans- gions (Haghighipour & Boss 2003), the solids-to-gas ra- port in disks, such as convection (Cameron 1978; Lin & tio can be enhanced in the transient turbulent gas pres- Papaloizou 1980) or nonlinear instability (Shakura et al. sure maxima, potentially reaching values high enough 1978). Yet, it was only with the re-discovery of the mag- to trigger gravitational collapse. Numerical calculations netorotational instability (MRI; Balbus & Hawley 1991) by Johansen et al. (2007) show that this is indeed the that the question seemed settled. The MRI entails a com- case, with the particles trapped in the pressure max- bination of a weak (subthermal) magnetic field and the ima generated by the MRI collapsing into dwarf plan- shear present in the Keplerian rotation the gas, that de- ets when the gravitational interaction between particles stabilizes the flow. The instability is powerful, seeming is considered. While the MRI is not strictly necessary, to explain the required accretion rates. the streaming instability does not seem to operate ef- The MRI also has a positive effect on planet forma- fectively for solar and subsolar metallicities, requiring tion. Starting with micron-sized dust grains, coagu- more solids than initially present in the Solar Nebula. lation models (Brauer et al. 2007) predict growth to However, the conditions for the magnetorotational in- centimeter size by electromagnetic hit-and-stick mech- stability are not met in a huge portion of the disk (Blaes anisms (mostly van der Walls forces). However, growth & Balbus 1994). Being a mechanism that depends on beyond this size is halted, for two reasons. First, col- magnetization, it also depends on ionization; as a result, lisions between pebbles lead to destruction rather than if the ionization fraction is such that resistive effects be- growth (Benz 2000). Second, because of the balance be- come important, the MRI may shut down, defining a tween pressure, rotation and gravity, the gas orbits the MRI-dead zone (Blaes & Balbus 1994; Gammie 1996). star slightly slower than an independent body at the The partially ionized gas can be seen as a three-fluid same distance would. Consequently, pebbles tend to system, composed of the ion fluid, the electron fluid, outpace the gas. The resulting headwind drains their and the neutral fluid. In ideal MHD, where density and angular momentum, leading them into spiral trajecto- ionization fraction are sufficiently high, ions and elec- ries towards the star, in timescales as short as a hundred trons are tied to the magnetic field and they drag the years at 1AU (Weidenschilling 1977). A distinct possi- neutrals along with them; deviations from ideal condi- bility to solve these problems is gravitational instability tions occur when these fluids start to decouple. In the of the layer of solids (Safronov 1972; Lyttleton 1972; Gol- ambipolar diffusion limit the magnetic field is tied to dreich & Ward 1973; Youdin & Shu 2002). When the dust the electrons and ions, and drifts with them through the aggregates had grown to centimeter size, the gas drag neutrals (Blaes & Balbus 1994; Mac Low et al. 1995). In is reduced and the solids are pushed to the midplane of the Ohmic limit collisions with neutrals are efficient in the disk due to the stellar gravity. Although such bodies dragging the ions and electrons, and the magnetic field do not have enough mass to attract each other individ- is not frozen in any fluid. In between these limits lies ually, the sedimentation increases the solids-to-gas ratio the domain of Hall MHD, where the electrons are cou- by orders of magnitude when compared to the interstel- pled to the magnetic field, but collisions with neutrals decouple the ions (Wardle 1999; Balbus & Terquem 2001; Hydrodynamical instabilities in protoplanetary disks 3

Salmeron & Wardle 2005; Pandey & Wardle 2006; War- we give an overview of turbulence, introducing the hy- dle 2007; Pandey & Wardle 2008; Wardle & Salmeron drodynamical instabilities on Sect. 4. A synthesis, the 2012). At constant magnetic field, this is a progres- main goal of this review, is given on Sect. 5. Following sion as density increases, that first the neutrals decou- it, Sect. 6 presents our opinion for the directions of the ple from magnetic effects (ambipolar diffusion), then field in the next few years, finally concluding on Sect. 7. the ions (Hall MHD), and finally the electrons (Ohmic resistivity). 2. ASTROPHYSICAL DISKS: EQUATIONS AND We focus on the Ohmic zone, deep into the resis- STEADY STATES tive regime and any magnetic effect is irrelevant. We Disks are objects in steady state. They are gaseous ob- could have used “hydrodynamical” zone, yet we keep jects, so we need to solve the equations of hydrodynam- the name Ohmic for juxtaposition to the other non-ideal ics in a central potential. The equations to solve are MHD effects. It is in this non-magnetic regime that the hydrodynamical instabilities we review exist. ∂ρ The physics of the Hall-dominated and ambipolar- = −(u · ∇) ρ − ρ∇ · u, (1) ∂ dominated zones is not covered in this review, but we t ∂u point out their importance in providing boundary con- ρ = −ρ (u · ∇) u − ∇p − ρ∇Φ + ∇ · T + F, (2) ditions for the Ohmic zone. The upper atmosphere of ∂t the disk will be dominated by ambipolar diffusion, with 2 p = ρcs /γ (3) the emission of a magnetocentrifugal wind (Bai & Stone GM? 2011; Gressel et al. 2015). As such, it will provide a “lid” Φ = − , (4) to the instabilities here discussed. Until their interaction r with the ambipolar zone is defined, we cannot assume where ρ is density, u is the velocity, p is the pressure that these instabilities are felt in the disk atmosphere; (assumed to be that of an ideal gas law), Φ is the grav- if they are not, they should not be seen in infrared ob- 2 itational potential, cs is the sound speed in which cs ≡ servations. The Hall-dominated zone has a dynami- γRT/µ where T is the gas temperature, µ the gas mean cal instability, the Hall shear instability, if the angular molecular weight, R is universal gas constant, and γ is momentum vector and the magnetic field are aligned the adiabatic index given by the ratio of specific heats at (Kunz 2008; Kunz & Lesur 2013; Lesur et al. 2014; Bai constant pressure and volume respectively. Additional 2015). The reader should keep in mind that our knowl- body forces are represented by the vector F. Finally, G edge of the physics is these zones is still taking shape, is the gravitational constant, M? is the stellar mass, and and it is unclear how the hydrodynamical instabilities r the astrocentric distance. The mathematical symbols we review behave in the ambipolar and Hall zones. used in this work are listed in Table 1. The Ohmic zone is extended. In the inner disk, ioniza- The symmetric viscous stress tensor is T, and in Carte- tion is provided by thermal collisions. Down to ≈900K, sian coordinates it is expressed according to the Einstein thermal velocities have enough energy to ionize the al- index convention form kali metals. This temperature corresponds to a distance    of 0.1 AU, depending on the underlying disk model, so ∂ui ∂uk 2 ∂um ∂um T ↔ Tik ≡ ς + − δik + ζδik , only inwards of it will the disk be sufficiently ionized ∂xk ∂xi 3 ∂xm ∂xm for the MRI. Concurrently, outward of ≈30 AU (again (5) depending on the underlying disk model), even though and recalling that according to this convention, repeated the temperatures are low, the column density is low indices are summed over from 1 to 3. Thus, enough that stellar X-rays can ionize the disk through- ∂T ∂T ∂T ∂T out. Between these limits, a large dead zone exists. ∇ · T ↔ ik = i1 + i2 + i3 . (6) This, incidentally, is squarely where we expect planets ∂xk ∂x1 ∂x2 ∂x3 to form, at least in the Solar System. If we need turbu- The quantities ς and ζ are the dynamic shear and bulk lence to generate accretion and form planets, we have molecular viscosities whose values are rooted in the to look beyond the magnetorotational instability. If star mean free paths of the molecules of the gas (e.g., of formation necessitates turbulent transport through the molecular hydrogen). From a topological point of view, accretion disk, and if planet formation needs turbulent there are two kinds of material motions, either purely pressure maxima to trigger the streaming instability in compressive or purely incompressible. Each type of mo- solar metallicity, we need to look beyond the MRI. tion will have different properties with respect to the The past few years have seen a number of processes transfer or momentum at the molecular level wherein being proposed for purely hydrodynamical instabilities compressible motions will have a corresponding bulk in disks: two linear instabilities, the Vertical Shear Insta- viscosity while incompressible motions will have a cor- bility (Nelson et al. 2013a; Stoll & Kley 2014) and the Con- responding shear viscosity. Since most motions of inter- vective Overstability (Klahr & Hubbard 2014; Lyra 2014), est for disks have a primarily incompressible character and a non-linear instability the Zombie Vortex Instability it is usual practice to ignore the bulk viscosity. As such, (Marcus et al. 2015, 2016). These processes were linked then we may define a viscous diffusion ν ≡ ς/ρ. If fur- to different regimes of opacities by Malygin et al. (2017). ther U and L respectively represent scales typifying un- It is the purpose of this review to frame these disparate steady flow speeds and length scales of the fluid system, processes into a coherent picture of hydrodynamical then one can characterise the relative importance of vis- processes in Ohmic dead zones of accretion disks. This cosity through the definition of the Reynolds number review is organized as follows. In the next section we briefly review the physics of accretion disks. On Sect. 3 Re ≡ ULν. (7) 4 Lyra & Umurhan

TABLE 1 SYMBOLSUSEDINTHISWORK.

Symbol Definition Description Symbol Definition Description R cylindrical radial coordinate Nz Eq. (80) vertical Brunt-Väisälä frequency φ azimuth NR Eq. (79) radial Brunt-Väisälä frequency z vertical coordinate F body force r spherical radial coordinate ` ≈ 2π/k vortex eddy length scale λ wavelength u` typical eddy speed k = 2π/λ wavenumber t` ≈ `/u` eddy overturn time m azimuthal wavenumber `0 energy injection scale t time `diss dissipation length scale ρ density ε erg s−1 g−1 energy dissipation rate per unit mass u velocity Ek energy per mass in eddies between k and k + δk T temperature E = Ek/k energy per mass density γ adiabatic index $ = ∇ × u vorticity 2 cp specific heat at constant pressure Z = $ enstrophy

cV = cp/γ specific heat at constant volume Ψ u = ∇ × Ψzˆ streamfunction  1/2 −1 cs = T cp(γ − 1) sound speed ΩPV = ρ $ · ∇s potential vorticity ∂Ω ≡ PV p = cV (γ − 1)ρT pressure βPV ∂q potential vorticity gradient γ s =cV ln(p/ρ ) specific entropy l latitude Φ Ω = −GM?/r gravitational potential GPV Eq. (127) generalized potential vorticity 0 0 G gravitational constant uiuj spatial correlation of component velocity fluctuations M? stellar mass σSB Stefan-Boltzmann constant 0 0  2 µ mean molecular weight α = uruφ cs measure of turbulent intensity R gas constant νt = αcs H effective turbulent viscosity T viscous stress tensor L = R2Ω angular momentum ς dynamic shear molecular viscosity K spring constant ζ dynamic bulk molecular viscosity q generalized coordinate ν = ς/ρ kinematic viscosity ξ Lagrangian displacement Re = UL/ν Reynolds number B magnetic field p U representative velocity vA = B/ 4πρ Alfvén velocity L representative length β = 2c2/v2 plasma beta s A Q heat source η resistivity p Ω = GM?/R3 Keplerian angular frequency J current H = cs/Ω disk scale height c speed of light h = H/R disk aspect ratio ReM = UL/η Magnetic Reynolds number Σ ∝ ρH column density Λ = v2 /Ωη Elssaser number A q = −dlnΩ/dln R shear parameter Z charge multiplicity qρ = −dlnρ/dln R radial density gradient γi f ≡ γiρiρ(ui − u) ion-neutral drag coefficient qT = −dln T/dln R radial temperature gradient σcoll collisional cross section qΣ = −dlnΣ/dlnr column density gradient n number density ω complex eigenfrequency x ≡ ne/n ionization fraction σ = Im(ω¯ ) growth rate κR Rossland mean opacity −1 Ma = U/cs Mach number `r = (κrρ) photon mean free path p κep = Ω 2(2 − q) epicyclic frequency τr Eq. (129) thermal relaxation time Ak = cs ∂k lnρ normalized density gradient arad = 4σSB/c Radiation constant −1 Gk = γ cs ∂k ln p normalized pressure gradient χ Coefficient of thermal diffusion Sk = Gk − Ak normalized entropy gradient Ro Rossby number

A very large value of Re indicates that viscosity is rel- replace these stress terms with an equivalent turbulent atively weak. For protoplanetary disks, values of Re viscosity model in order to account for the flow of tur- can be as high as 1014 which is based on (i) the values bulent energy (see discussion in Sect. 3). of H2’s viscosity, (ii) the assumption that local sound- Lastly, these equations must be supplemented with speeds give an upper limit of the fluctuating velocity one describing the energy/entropy content of the flow. scales and, (iii) the local disk pressure scale height de- Typically, we start with the evolution equation for the scribes relevant length scales. In practice then, owing to entropy, the small values of these viscosities and the large scales   of interest, protoplanetary disk flows are treated as be- ∂s ρT + u · ∇s = Q. (8) ing inviscid and these stress terms are dropped. How- ∂t ever we will eventually come full-circle: owing to the strong linear instabilities found in these flows, we must where the specific entropy s is defined based on the spe- cific heat at constant volume Hydrodynamical instabilities566 in protoplanetary disks 9 Turbulence 5

γ s ≡ cV ln(p/ρ ). (9) The heat source term Q will usually have a viscous heating and either a radiative cooling or thermal diffu- sion term. For locally isothermal disks, one simply sets γ = 1 in Eq. (3) and not solve Eq. (8). Writing Eq. (1) and Eq. (2) in cylindrical coordinates (R,φ, z) we find, without the viscous terms

∂ρ ∂ρ uφ ∂ρ ∂ρ = −u − − u ∂t R ∂R R ∂φ z ∂z   ∂u u 1 ∂uφ ∂u −ρ R + R + + z (10) ∂R R R ∂φ ∂z 2 ∂u ∂u uφ ∂u ∂u uφ 1 ∂p GM R = −u R − R − u R + − − R ∂t R ∂R R ∂φ z ∂z R ρ ∂R r3 (11) Fig. 9.6 The Richardson cascade. The space-filling eddies decrease in size downwards, in the directionFIG. 1.— of the constant A cartoon energy image flux of turbulent cascade as cast into verse ∂uφ ∂uφ uφ ∂uφ ∂uφ uφuR 1 ∂p by L.F. Richardson (1881-1953) “Big whirls have little whirls/That feed on = −uR − − uz − − (12) their velocity,/And little whirls have lesser whirls/And so on to viscosity." ∂t ∂R R ∂φ ∂z R ρR ∂φ where Eq. (9.176)hasbeenused.Itfollowsthatthesmallestshearisatthetopand Each daughter generation is scaled by its parent size by ξ. ηK is the the largest at the bottom of the cascade.` The shearing of an eddy by eddies very ∂uz ∂uz uφ ∂uz ∂uz 1 ∂p GM same as the dissipation scale diss, sometimes also known as the Kol- = −u − − u − − z (13) mogorovmuch larger microscale. or smaller is Throughout ineffective (can this you inertial see why?). range, So there the main is no shearing build R z 3 action on an eddy of scale ℓ is done by eddies of scale ℓ ℓ. ∂t ∂R R ∂φ ∂z ρ ∂z r up of power – only a flux of ε shuttling energy′ ∼ from larger to smaller scales via nonlinear interactions. [Figure from Regev et al. (2016), by We look for steady state solutions, so ∂t = 0. We also permission.] 9.4.5.2 Reynolds Numbers and Numerical Consequences assume azimuthal symmetry, setting ∂φ = 0. Consider also that the disk is in vertical hydrostatic equilibrium, It is trivial to express, using the phenomenological relations, the velocity and so uz = 0. Plus, for the moment we assume that the cen- turnover time at scale ℓ,thatis,uℓ and τℓ,intermsofthesevariablesattheintegral trifugal balance is maintained, so ur = 0. With these con- scale, u0 and τ0,whichthemselvesarerelatedthroughtheintegrallengthscale strains, the continuity and angular momentum equa- ℓ0 = u0τ0;thus, tions admit only the trivial solution, and we are left with u ℓ 1/3 τ ℓ 2/3 ℓ , and ℓ . (9.181) quite little u ∼ ℓ τ ∼ ℓ 0 ! 0 " 0 ! 0 " Usually when one speaks of the Reynolds number of a turbulent flow, it is this GM 1 ∂p quantity for the integral scale, that is, − Ω2 3 R = + R, (14) r ρ ∂R ℓ u Re 0 0 . (9.182) GM 1 ∂p ∼ ν z = − . (15) r3 ρ ∂z The first equation above is what is left of the radial mo- mentum equation, and the second one from the vertical momentum equation. These equations highlight why disks are flat. It is the influence of the centrifugal force: while pressure counterbalancing gravity makes things spherical, the centrifugal force makes things cylindri- cal. The relative strength of the pressure and the cen- trifugal terms defines if the object will appear more on the spherical side, in hydrostatic equilibrium, or more on the flat side, in centrifugal steady state. A pressure- supported object is spherical, whereas a centrifugally- supported object is flat. Eq. (15) gives the condition of vertical hydrostatic equilibrium, which we turn to now. 2.1. Vertical hydrostatic equilibrium If we assume an isothermal equation of state, or sim- FIG. 2.— Roll-up of 2D vortex strip. Top panel shows vortex strip ply that the sound speed does not depend on z, so we (shear layer) soon after initiation with vorticity shown in side inten- sity panel. Overlain is the sense of the mean velocity field u(y). Bot- can write cs = cs(R), we can solve for the vertical struc- tom panel shows the result of the roll-up of this shear layer. The fi- ture nal resulting structure is larger in scale compared to the initial shear layer indicating an inverse cascade of energy. However, the roll-up ∂lnρ GM z has also generated small scale vortex filaments – indicating the gener- − ation of vorticity on scales much smaller than the original strip. [Fig- = 2 3 (16) ∂z cs r ure adapted from Biancofiore & Umurhan (2018), by permission.] which is integrated to yield 6 Lyra & Umurhan

2    GM c GM qT GM 1 1 Ω2 = − s (q + q ) − z2 . (27) √ − 3 2 ρ T 3 ρ(R,z) = ρ(R)exp 2 . (17) R R R 2R cs R2 + z2 R Finally, substituting the definition of Ω , we find √ p K We can write R2 + z2 = R 1 + (z/R)2, and expand in Taylor series to first order to find (  2  ) 1 H qT  z 2 Ω = ΩK 1 − qρ + qT + (28)  GM  2 R 2 H − 2 ρ(R,z) = ρ(R)exp 2 3 z . (18) 2cs R Notice that if qρ = qT = 0, i.e., no pressure gradient, 2 3 The quantity cs R /GM has dimension of length the angular frequency returns to Keplerian. Also, be- (squared) and thus defines a scale height. GM/R3 is the cause H/R is small, the deviations from Keplerian are square of the Keplerian angular frequency also small. Before we address how this steady-state configuration r GM transitions via linear and nonlinear instabilities into full- ΩK ≡ (19) blown protoplanetary disk turbulence, we must give a R3 brief overview of how turbulence is confronted in this So astrophysical context. c 2.3. Ertel’s theorem and Rossby Waves H ≡ s , (20) ΩK A theorem and quantity of singular importance emerges from an analysis of the equations of motion. and we can write the density stratification compactly If we assume both no heat gain or losses in the flow 2 (Q = 0, which means that the gas specific entropy s is − z ρ(R,z) = ρ(R)e 2H2 . (21) materially conserved), and we neglect gas viscosity by setting T = 0, then a manipulation of the resulting equa- Note that cs = ΩK H and uK = ΩK R, so tions of motion reveals a conserved quantity

H cs 1  ∂  $ · ∇s = ≡ (22) + u · ∇ = 0, (29) R uK Ma ∂t ρ

At the position of Jupiter, T ≈ 180K and uK ≈ 10 km/s, where $ ≡ ∇ × u is the vorticity. The conserved quan- so cs ≈ 500 m/s and thus H/R ≈ 0.05, rendering the tity is called potential vorticity disk thin. This quantity, h ≡ H/R, is usually called the disk aspect ratio. Another feature to notice is that given $ · ∇s Ω ≡ , (30) Ω ∼ R−3/2, and if temperature dependence is described PV ρ −q by T ∼ R T , where qT is an index (see below), then for values of q < 3/2 we see that the scale height H in- and Eq. (29) is known as Ertel’s theorem; a succinct T derivation may be found in Regev et al. (2016). The creases with distance. For qT = 1 it follows that the as- pect ratio h = H/R is constant. dynamical importance of the conservation of potential vorticity (often known as “PV") and its use as a tool to 2.2. Radial Centrifugal balance interpret flows – especially those that are strongly ro- tating and/or ones with one spatial dimension severely Having solved for the disk vertical structure, we turn limited like in planetary atmospheres and disks – can- to the radial equation, Eq. (14). We can write the pres- not be overstated (Pedlosky 1982). sure gradient as While a thorough survey of its consequences is not possible here (for a good introductory view see Val- 1 ∂p  ∂lnρ ∂lnc2  = c2 + s (23) lis 2006), it is important to discuss a dynamical phe- ρ ∂R s ∂R ∂R nomenon that emerges as a consequence of perturba- tions of Eq. (29), namely, the ubiquity of the Rossby wave. and given the hydrostatic solution for density, Eq. (17) To best illustrate the Rossby wave, we consider a very shallow constant density fluid layer on a rotating sphere 1 ∂p c2 GM GM GM q with constant radially directed gravity. If the rotation = − s (q + q ) + − R − z2 T (24) vector of the spinning sphere of radius R is Ωzˆ, where ρ ∂R R ρ T R2 r3 R3 2R 0 zˆ is the spin axis and the constant Ω0 > 0, and if the where we also assumed radial power laws for density fluid layer depth is H (and constant for our purpose), and temperature then the entropy gradient is a delta function with radial unit vectorr ˆ. Since for a given observer at latitude l, the

inner productr ˆ · zˆ = cosl, it follows that ΩPV is non-zero ρ(R) ∝ R−qρ , (25) at the location of the entropy gradient which is taken T(R) ∝ R−qT . (26) to be the surface of the sphere for the sake of simplic- ity. The PV then has the form 2Ω cosl. It can be seen Substituting the pressure gradient on Eq. (14) and that global PV increases with increasing latitude. Intro- solving for Ω duction of perturbations to an otherwise static rotating Hydrodynamical instabilities in protoplanetary disks 7 atmosphere will result in a wave response known as the Rossby wave. To see this consider a small latitudinal 3. TURBULENCE: A TERSE OVERVIEW. zone centered at some given latitude l0. If we expand ΩPV Fluid turbulence remains one of last unsolved prob- = 2Ω0 + β (y − y0), (31) lems of classical physics (Davidson 2004) and despite PV over 100 years of research by thousands of scientists and where engineers, very little can be said. Many who work in the field, when asked “What is turbulence?”, respond with “I 1 ∂Ω 2Ω = 2Ω cosl β ≡ PV , (32) do not really know, but I know it when I see it.” Fluid flows 0 0 PV R ∂l l=l0 that undergo turbulent transition can be said to display widespread structural disorder and chaos. They are in which y has units of length, then we can perform inherently unpredictable and highly nonlinear. Unsta- an approximate linear perturbation analysis of Eq. (29) ble fluid structures in the grips of turbulence can com- assuming strictly two-dimensional incompressible flow pletely fill up the totality of the fluid volume containing whose details can be found in Pedlosky (1982). Pertur-
the turbulence (like what happens in the bounded re- bations of the form exp iωt + ikxx + ikyy in which ω gions containing unstable jet flows or wakes) or display is the frequency response, x is the longitudinal distance sublimity like spatiotemporal intermittency, i.e., regions and where kx and ky are the corresponding wavevector and pockets in space and/or time where the flow ex- of the disturbance results in a Rossby wave frequency hibits laminar behavior surrounded by regions of high response given by intensity vortex tube twisting and attendant chaos. There are several possible routes to turbulence in any β k PV x fluid setting. In the context of protoplanetary disk ωRW = 2 2 . (33) kx + ky Ohmic zones, we focus on the picture of supercritical transition wherein the system of interest, generally in We might imagine a circumplanetary level-set of con- some steady state, will experience a linear instability if stant PV being perturbed with a wavelength kx and ky. physical conditions in the fluid are met – we refer to this The frequency response indicates that the longitudinal as the primary instability. If the physical criteria control- wave pattern propagates in the negative longitudinal ling the onset are only weakly surpassed by the physical (westerly) direction. This is a general feature of systems state of the fluid system, then often times the flow will supporting Rossby waves: in a right-handed coordinate reconfigure itself into some new nonlinear state – often system, positive latitudinal gradients in PV results in a time-dependent with an identifiable set of periods – but negative pattern speed of the resulting wave. nevertheless orderly structured and predictable. If the Rossby waves are also prevalent in protoplanetary physical conditions greatly surpass the criterion for on- disks and this has been extensively written about set, then even this nonlinear state may experience an in- (Lovelace et al. 1999; Li et al. 2000; Umurhan 2010). stability and when this happens we refer to it as the sec- There are notable differences, the main one being: ondary instability. This transition might result in another, while accretion disks exhibit gradients in their mean even more complicated, nonlinear state to emerge – or – PV simply owing to the mean Keplerian flow, Rossby this transition could lead to a catastrophic cascade and waves are mostly supported where there are anoma- a complete breakdown of any ordered flow structure. lous changes in the mean Keplerian flow, like loca- Of many consequences of such chaotic situations, an tions where the disk material supports pressure ex- important one is that fluid vortical structures emerging trema (Lovelace et al. 1999; Li et al. 2000; Meheut et al. on every scale will vigorously exchange angular mo- 2010; Lin 2014) or where there are sharp edges like near mentum with one another. Another feature to keep in disk gaps (Koller et al. 2003; de Val-Borro et al. 2007; mind is that every time a transition occurs in the flow, Lyra et al. 2009; Yellin-Bergovoy et al. 2017), or poten- the system generally loses some symmetry it once pos- tially in places where the disk goes from being Ohmic sessed before the instability. A fully turbulent state has to significantly magnetized (Varnière & Tagger 2006; lost all semblance of symmetries that are permissible in Lyra & Mac Low 2012). However, the Rossby wave the governing equations of motion. Ironically, these lost frequency response estimated based on our simplified symmetries return when the flow is viewed from a sta- rotating sphere model, Eq. (33), may be also a useful tistical and/or spatiotemporally averaged point of view. guide for disks as well (Sheehan et al. 1999). At a lo- For example, in dissipative MRI turbulence the pri- cal radial position R0 in a Keplerian disk with rotation mary instability leads to the emergence of a radial com- 3/2 Ω = Ω0(R0/R) , in which the disk material moves ponent to the basic Keplerian state which periodically only in the azimuthal-radial directions and the fluid is reverses sign as a function of distance away from the idealized to be of constant density of finite disk thick- disk midplane. Because of the emergence of a strong ness, then we find that vertical shear in this radial velocity component, the new flow state undergoes a secondary instability via the ex- 1 ∂(ΩR) Ω Ω = = , (34) citation of parasitic instabilities (Goodman & Xu 1994) PV R ∂R 2 which is an element of the wider class of shear instabili- ties. This result is a dynamical cascade into widespread and following Sheehan et al. (1999), adopting the same and very strong turbulence that effectively transports line of thinking we can estimate the β -parameter to be PV angular momentum.

∂Ω 3 Ω From this general perspective and for our purposes β ∼ PV = − . (35) here we assume that (i) there is a continuous energy PV ∂R 4 R R=R0 9.5 Two-Dimensional8 Turbulence Lyra & 575 Umurhan

where ` is the spatial scale of the fluctuating quantity. Let us define Ek as being the total specific energy in the flow contained in fluid structures whose wavenumber lies between k and k + δk. The specific energy is under- stood to be the energy per unit mass (cm2/s2) – for the purposes of discussion we assume the flow has constant density and we will consider the corresponding specific kinetic energy. The power in this specific energy is for- mally defined through the relationship expressing the total specific energy contained in the fluctuating flow field whose wavenumbers lie between k and k + δk, i.e., Z k+δk ∂Ek Ek = E(k)δk ≡ Edk, lim E(k) → . (36) k δk→0 ∂k E(k) has units of cm3/s2. Evidently, the system inte- grated total specific energy, E, is the integral of E over Fig. 9.8 Sketch of the energy spectra in the two inertial regimes of the Kraichnan–Batchelor all wavenumbers, i.e., doubleF cascadeIG. 3.— model. The Injection double at cascade the forcing spectrum wavenumber of thek0 separates Kraichnan-Batchelor energy moving towards Z ∞ low wavenumbers ( 5/3 scaling), while enstrophy moving towards high wavenumberk leads= to a theory of 2D− steady driven turbulence. The injection scale is 0 E = Edk. (37) 3 scaling` and is ultimately dissipated by viscosity at the dissipation wavenumber kν 0 − 2π/ 0. The two cascade regimes are indicated. The inclusion of linear dissipation acting only at the largest scales accounts for the k3 charac- ter of the energy spectrum in that regime. Simulations of Boffetta & When Ek behaves as a power law (at least locally in k), a aresultthatisthetwo-dimensionalanalogoftheKolmogorov4Musacchio (2010) largely bear out the character of this theory./5law,exceptthat [Figure quick estimate for E is ≈ Ek/k. 3/2replacesfrom Regev4 / et5, al. the (2016), sign change by permission.] implying a reversal of the direction of spectral E(k) is a diagnostic quantity reflecting the state of the − transport: to large scales. fluid at any given instant. If the turbulence is in a statis- Whatsource is the for physical the turbulence, picture of the above, (ii) it mainly enters mathematical, the system description at tically steady state – where injected power is ultimately of a process occurring in spectral space? Observations of approximately two- some integral scale of the system `0 and, (iii) it steadily drained from the system through either by molecular dimensionalinjects turbulence energy on provide these a consistent scales at picture, a given in which rate, aε patch– a quan- of vorticity viscosity on the smallest scales, or by Ekman pumping is distorted—extended in one direction, to a scale D.Asaresultofmaterial tity typically expressed in units of energy per unit time and/or radiative losses to space on the largest scales – conservation of vorticity, the patch is squeezed to a small scale din the transverse −1 −1 2 3 E(k) direction,per unit with d massdecreasing (erg as s D grows.g , or Eventually, cm /s d).’s scale will shrink to the point then will be steady over some suitable time aver- that diffusionFrom will the begin vantage to act. In point the schematic of astrophysical Fig. 9.9,weseeboththeinverse flows like age. Because fluid systems like the Navier-Stokes equa- cascade,protoplanetary to large D,andthedirectcascade,toadissipationscale disks, the presence or absence of turbu-d.Onepoint tions are highly nonlinear, energy contained in one nar- of viewlence associates has several the growth profound of D with effects. mergers Of between importance like-signed to vortices.us row band of wavenumbers can be traded to other fluid However,(and while as intimated regions of above) like-signed is the vorticity ability do aggregate, of turbulent there struc- is mounting structures of differing wavenumber through the pro- evidencetures that to the transport merger process angular contributes momentum. little to the inverse In regions cascade cen- in forced cess of nonlinear interactions. Sometimes the energy ex- two-dimensionaltered on the turbulence. midplane at 5AU in a disk surrounding a 1 change is local in wavenumber space, but other times solar mass star, the Reynolds number Re = UL/ν with the exchange can occur between structures of vastly dif- velocities U of the order of the sound speed and length fering length scales. 9.5.2.1L of Stationary the order State of and the Condensates local scale height H is roughly 13−14 3.2. 3D isotropic turbulence: Kolmogorov-Obukhov theory If the10 inverse, cascade based carries on molecular energy to large hydrogen’s scales, effectively viscous hiding diffu- it from 3 2 −1 and Richardson cascade viscoussivity diffusion,ν ∼ then10 cm theres is no. mechanism The corresponding to dissipate the viscous accumulating dis- large- What little that can be said about freely evolv- sipation scale (Regev et al. 2016) is ` ∼ Re−3/4 H ≈ diss ing turbulence of the 3D incompressible Navier-Stokes 0.1 − 1km. This means that the character of the un- equations relate to its observed statistical properties steady flow between the scales that drive the turbulence E from the generation scales (an order unity fraction of expressed as some signature quality of (k). The Kolmogorov-Obukhov theory of 3D isotropic turbu- H) down to the dissipation scale `diss must somehow be characterized. Currently, only a crude mixing-length lence offers an explanation predicated on certain as- theory characterization exists for disks (see Sect. 3.5). sumptions about the flow, namely that it is (i) homoge- nous and isotropic, (ii) self-similar and, (iii) the dissipa- 3.1. Kinetic energy spectra tion rate is constant. A detailed discussion about this theory can be found in Davidson (2004). We offer a Flows which exhibit turbulence can be identified (at crude sketch to convey the essence of the theory as de- least in part) by their statistical properties. The flow scribing statistically steady state turbulence. The no- is presumed to have a mean state plus a spatiotempo- tion that the fully developed state is homogenous and 0 rally fluctuating field, respectively given byu ¯ and u . isotropic is to be understood from this time-averaged In most instances, one examines the kinetic energy con- statistical point of view. tained in fluctuating quantities by constructing its en- In this light we consider what happens to structures ergy spectrum. This is most readily seen if we assume u0 whose size ` is much smaller than the injection scale to be periodic in each spatial dimension. Provided the `0 but much larger than the dissipation scales `diss. We dimensions containing the fluid, L, is sufficiently larger call this size range the inertial regime. Thus, the first two than the injection scales, then we can identify all compo- of these assumptions give rise to the so-called Richard- nents of the fluctuating field exhibiting spatial structure son cascade (see Fig. 1) governing what happens in the with absolute value of the wavenumber |k| = k ≡ 2π/`, inertial regime, in which a given 3D twisting vortex of Hydrodynamical instabilities in protoplanetary disks 9 some size `, with rotation speed u`, undergoes a nonlin- much larger than the dissipative scales `diss, will exhibit ear bifurcation that generates smaller-sized (scaled on ` a direct cascade of enstrophy and an inverse cascade of en- by 0 < ξ < 1) daughter 3D vortex structures, also see ergy. With the vorticity as defined $, strictly 2D flow discussion also found in Frisch (1995). We can think results in   of the vortical structure on scale ` as "living" only for ∂u ∂uy $ = $ zˆ ≡ x − zˆ. (40) one rotation time before undergoing "destruction", i.e., z ∂y ∂x t` = `/u`. In this steady state picture, the rate of en- ergy entering and exiting every length scale must be the Enstrophy is defined to be the square of the vorticity, i.e., 2 same, i.e., ε is independent of `. Thus, we say that this Z ≡ $ . In the inverse cascade regime (L ≥ ` ≥ `0) the energy rate is given by its energy divided by this rota- energy spectra has the familiar form, E ∼ k−5/3. Unlike tion time. In terms of the specific energy this means to the 3D case discussed above, energy shifts from smaller say, scales to larger scales. Often times this can be identified 2 3 by eye as the numerical experiment of an unstable 2D u`/2 u` −1/3 1/3 −1/3 ε = ∼ , −→ u` ∼ (ε`) ∼ ε k . (38) shear layer in Fig. 2 shows: a thin vortex strip rolls up t` ` under its own induced flow and eventually turns into It immediately follows that the energy contained on the a larger tumbling structure containing a jumble of fine length scale ` scale vortex filaments even smaller than the original fil- ament. These developed finer scale filamentary struc- 2 2/3 −2/3 tures qualitatively exhibits the forward cascade of en- Ek = u`/2 ∼ ε k . (39) strophy. The emergent large scale vortex structure ex- This is sometimes known as the Kolmogorov two-thirds hibits the inverse cascade of energy. rule. The amount of power in the small wavelength In the forward enstrophy cascade regime of steady bin containing k is therefore expressed by E ≈ Ek/k ∼ driven 2D turbulence, the wavenumber dependence of ε2/3k−5/3. This is the famous "universal k−5/3" shape the energy spectrum is much steeper, in fact E ∼ k−3. characterizing the inertial spectrum of freely evolving These properties are summarized in Fig. 3. The trends 3D turbulence. This property has been assessed in nu- predicted in the Kraichnan-Batchelor theory of steady merous experiments of 3D turbulence including wind driven turbulence have yet to be disproven (Boffetta & tunnel experiments and others (Frisch 1995). Ecke 2012): the features of the theory summarized in Identifying a fluid system as being turbulent gener- Fig. 3 are asymptotically approached as recent highly ally requires resolving its inertial spectrum. Reproduc- resolved DNS indicate (Boffetta & Musacchio 2010). ing this k−5/3 shape is now considered a standard bench- The direct enstrophy cascade can be rationalized by mark for for validating most direct numerical simula- the following argument. For 2D incompressible Navier tions (DNS) which purport to model turbulence. For Stokes flow, the equation governing its evolution is numerical experiments, in turn, this means capturing greatly simplified to at least 1−2 decades of the inertial spectrum, i.e., there − d$z 2 ought to be a factor of 10 100 scale difference between = ν∇ $z, (41) the injection and dissipation scales. We shall return to dt this matter and how it bears upon simulations relating in which the incompressibility constraint means that all to protoplanetary disk turbulence in Sect. 6. quantities may be expressed in terms of a single scalar, called the streamfunction Ψ, in which 3.3. 2D turbulence ∂Ψ ∂Ψ 2 Freely decaying 3D turbulence can be considered a ux = − , uy = ,−→ $z = ∇ Ψ. possible end-member state contained in the idealized ∂y ∂x abstract space of turbulent flows. By its construction, there are no body forces acting upon the fluid. Body If the dissipation scale is indeed very small, then Eq. (41) forces, like gravity, often act upon the fluid by constrain- states that in the inertial range viscosity is negligible ing one dimension or enforcing a symmetry on the flow. and that, consequently, the vorticity is a materially con- One might expect the presence of a body force adds served quantity. In this case we can follow the same more complexity to the problem and that, in comparison physically motivated logic administered in our discus- to freely evolving 3D turbulence, even less might be said sion of freely evolving 3D turbulence. By assuming about the statistical quality of any emergent state. In the isotropy, homogeneity and self-similarity we state that idealized case of 2D turbulence, the opposite is true. 2D patches of vorticity can undergo a Richardson type of turbulence is the case of the Navier-Stokes equations in cascade by preserving the vorticity with which it was $ which one dimension is restricted by fiat – henceforth endowed. Calling this vorticity 0, we find there- we assume it is the z-direction. Atmospheric flow (con- fore that mother-daughter structures must preserve this strained by both gravity and rotation) and the motion quantity all the way down to the dissipation scales, along bubble surfaces (constrained by surface tension) therefore u are close, but not exact, physical instances of idealized $ ∼ ` = constant, −→ u = $ ` ∼ $ k−1. (42) 2D turbulence. 0 ` ` 0 0 Such flows are known to support a double cascade. The This means that the energy of the structure is Ek ∼ Kraichnan-Batchelor theory of steady driven 2D turbu- 2 −2 lence says that flows with power injected on some scale $0k , which means that in the forward enstrophy cas- 2 −3 `0, which is much less than scale of the system L and cade zone E(k) ∼ Ek/k ∼ $0k . 10 Lyra & Umurhan

3.4. The quality of turbulence in strongly rotating and zonal flows through the process of wave-mean flow in- stratified flows and Rhines scales teractions. For a deeper and expanded discussion re- Turbulence in strongly rotating and stratified flows is garding wave-mean flow processes see Pedlosky (1982) an unsettled topic. A case in point is the Earth’s atmo- and Vallis (2006). The consequences are clear: systems sphere which one might consider to be a close approx- that are nearly 2D and turbulent, that exhibit both an in- imation to 2D flow. Nastrom & Gage (1985) analyzed verse cascade and other features like strong rotation, can wind data based measurements made on thousands of result in the organization of such disorder into global commercial airplane flights, and they found that the at- scale organized motions like jets and waves. If proto- −3 planetary disk dynamics indeed exhibits the qualities mosphere indeed exhibits a k spectrum on scales rang- of 2D turbulence, then there will arise a similarly mo- ing from the planetary (∼ 10,000km) through the syn- ∼ tivated Rhines scale (Sheehan et al. 1999). We reflect on optic scales ( 1000km). However, in the mesoscales this further below. range, which starts at about 500km and goes down to − about a few km, the spectral shape ∼ k 5/3. While the 3.5. The alpha disk model large scale energy injection mechanism is known (baro- clinic instability ∼1000 km scales), a robust physical ex- Despite the great abundance of observations and data −5/3 available for the Earth’s atmosphere, an understanding planation for the k spectrum, including what are the of the nature of its turbulence is still cloudy. The situ- energy sources and sinks, remains a subject of much de- ation with regards to accretion disk turbulence remains bate with several candidate mechanisms, like gravity much worse – although the developments of the last five wave breaking, moist convection and others (Tulloch & years raise hope that inroads toward its characterization Smith 2006; Marino et al. 2014; Sun et al. 2017). While is on the near term horizon. the large scale trend may be rationalized in terms of In lieu of this, and for a great many number of years, the inverse cascade picture of driven 2D turbulence, the the astrophysical community has relied on a mixing- switch over into what ostensibly exhibits 3D statistical length model to represent accretion disk turbulence. character occurs on length scales that are much larger The so-called α-disk model (Shakura & Sunyaev 1973; than the vertical scales of the atmosphere (∼ 10km, Lynden-Bell & Pringle 1974) takes a heuristic approach corresponding to the height of the tropopause). Sim- toward handling this situation. It assumes that there ex- ilar trends have recently been identified for Jupiter’s ists a dynamically sustained process that generates dis- weather layer based on analysis of imaging data ac- ordered velocity fluctuations about the mean Keplerian quired during the Cassini spacecraft’s Jupiter gravity as- state. If these disordered fluctuations are indeed turbu- sist maneuver (Young & Read 2017). lent then they will be engaged in some kind of nonlin- The inverse cascade is also arrested on the large Rhines ear energy transfer which translates into them exhibit- scales. Rhines (1979) demonstrated that for 2D turbu- ing non-zero spatial correlations. These correlations are lence on a rotating sphere there exists a scale λb span- assessed by calculating spatial averages of the corre- ning the north-south direction on which the differences sponding Reynolds stresses. Following the original ex- of the planetary vorticity across this length scale equals ample, we consider u0 and u0 respectively as the fluctu- the overturn time of the upscale cascading turbulent ed- r φ ating radial and azimuthal velocities about the basic Ke- dies denoted by $eddy ∼ Ub/λb. Disturbances in the −1/2 planetary PV give rise to Rossby waves (as described plerian (azimuthal) velocity UK ∼ R . Denoting spa- in Sect. 2.3). If the frequency response of the planetary tial averages with overbars, the negative of the averaged scale Rossby wave equals that of the turbulent eddy radial-azimuthal component of the Reynolds stress is overturn time, then instead of these eddies sending equated with an effective linear momentum diffusion their kinetic energy to larger length scales as demanded through the following identification: by the inverse cascade process, these λb-scale turbu- ∂U lent eddies will transfer their energy into large scale ν K ≡ −u0 u0 . (44) t ∂R r φ Rossby waves7. If we assume the turbulent eddies are of longitudinally-latitudinally equi-dimensional, then by The left hand side expression of the above is moti- setting kx = ky = kb ≡ 2π/λb in Eq. (33), we find that vated from the definition of viscosity from the molecu- lar dynamics point of view which says that neighboring setting ω = $0 implies RW parcels of gas will exchange momentum in proportion β q to the gradient of their localized bulk (mean) velocities. PV = U k −→ k = β  U b b, b PV 2 b. (43) In this case the mean/bulk velocity is taken as the Ke- 2kb plerian one. But the key ingredient in this characteriza- These special scales are observed in simulations and in tion is the measured correlation u0 u0 . Thus, the effective data of planetary atmospheres. These large scale waves r φ 2 may themselves nonlinearly develop and/or play a role turbulent diffusion (νt, in units of cm /s) is expressed in in generating large scale circulation in the form of global terms of the product of the local sound speed and scale height: νt ≡ αcs H. Thus α is the measure of the turbulent 7 Note that a steady rotating global atmosphere sitting atop flat to- intensity and we have pography will have variations in its potential vorticity as a function of latitude. Potential vorticity will also show variations if there is some 0 0 uruφ amount of significant topography. Thus, large scale inverse-cascading α ∼ , (45) turbulent eddies encountering topography will also result in some of c2 their energy being transferred into Rossby waves. This latter feature s is relevant to planetary atmospheres. which follows from the definition of H in Eq. (20). In this Hydrodynamical instabilities in protoplanetary disks 11 form, α measures the square of the local Mach number Finally we can estimate the energy cascade rate toward of the fluctuating disk velocities which are (presumably) the Rhines scale following the same line of reasoning dominated by the velocities of the injection scales. leading to Eq. (46) Quoting this value of α is the main characterization r used to assess the degree of turbulence in protoplanetary α5/4 3h c3 e ≈ e = s . (48) disks. It says nothing about the statistics of the actual α π 8 H small scale motions driven downscale by the larger scale driven dynamics – i.e., to date E(k) is still not known. 4. THE PRIME MOVER: LINEAR INSTABILITY This has direct consequences for the accumulation of disk dust. In this section we review the Rayleigh crite- Nonetheless one may make estimates. Assuming the rion (Sect. 4.1), the magneto-rotational instability turbulence is steady and the energy cascade is 3D, as in (Sect. 4.3.1), and introduce the hydrodynamical instabil- the sense of Kolmogorov-Obukhov theory, then a cas- ities (Sect. 4.4). cade rate εα based on the measured value of α may be es- 4.1. Rayleigh criterion timated by following the physical reasoning in Sect. 3.2 to be approximately, The Rayleigh criterion is paramount to understand- ing what instabilities can be present in the Ohmic zone. 3 3/2 3 ε ≈ εα ≡ δv /Hoturn = α cs /H. (46) Starting from the momentum equations, Eq. (11) and Eq. (12), without the pressure gradient, and restricting where we have identified the square of the fluctua- ourselves to the midplane 2 ≡ 0 0 tion velocities with the Reynolds stress: δu uruφ . Implicit in this assumption is that the overturn scales u2 ∂uR ∂uR ∂uR φ 2 Hoturn are approximately that of a scale height. The + uR + uφ − = Ω R (49) three large-scale eddy producing linear instabilities de- ∂t ∂R ∂φ R scribed in the upcoming sections generate structures ∂uφ ∂uφ ∂uφ uφuR + u + u + = 0. (50) whose fundamental length scales are generally a fixed ∂t R ∂R φ ∂φ R fraction of the local H. Interestingly, as noted in Sect. 2.1, if the mean temperature profile in a protoplanetary disk We can decompose the flow into base (time- Ohmic zone has a value of qT < 3/2, then the scale independent) and perturbation in height H increases with distance and, therefore, the 0 overturn scales similarly enlarge with radial distance u(x,t) = u¯ (x) + u (x,t), (51) from the central star, much inline with von Weiszkäker’s Ω  0 predictions (Gamow & Hynek 1945). with the base flow u¯R = 0 and u¯φ = R uφ, the per- If, on the other hand, the protoplanetary disk turbu- turbation equations become lence exhibits a 2D energy cascade at the largest scales, then energy will cascade until it encounters its corre- ∂u0 sponding Rhines scale. 3D simulations of buoyant con- R − 2Ωu = 0 (52) ∂t φ vection in disks performed by Cabot (1996) indeed show 0 that the small scale forcing sends power from the small ∂uφ + (2 − q)Ωu = 0 (53) scales and drives into place 2D large-scale zonal jets. ∂t R The length scale of the emergent jets from these simu- where lations Ljet were shown by Sheehan et al. (1999) to be consistent with the corresponding Rhines scale appro- ∂lnΩ priate to the local disk section. Specifically, character- q ≡ − (54) ∂ln R izing the turbulent large scale eddies by the previously motivated α prescription, then the Rhines scale vortices Now consider the perturbation as a Fourier mode 1/2 0 ˆ −i(ωt−k·x) will have circulation speeds of Ub ∼ α cs. Thus, tak- ψ = ψ0e , so that ing this together with the definition of the Rhines scale found in Eq. (43) and the disk appropriate value of the −iωuˆR − 2Ωuˆφ = 0 (55) βPV -parameter found in Eq. (35), then we find that the corresponding predicted value of kb to be −iωuˆφ + (2 − q)ΩuR = 0 (56) s s which leads to the solution for the complex eigenfre- π 8α1/2 8α1/2 λb ∼ = π HR = π H. (47) quency kb 3 3h 2 Ω2 − The convective turbulence simulated in Cabot (1996) ω = 2 (2 q) (57) predicted values of α in the range 10−2 to 10−4. Assum- This frequency is called epicyclic frequency, h ∼ − ing disk aspect ratios 0.05 0.1, plugging in these q values into Eq. (47) show that λ is predicted to be an b κep ≡ Ω 2(2 − q) (58) order one fraction of the local pressure scale height H, which is in rough agreement with the measured values if q < 2 then the flow is stable. Since the Keplerian flow of the jet widths Ljet observed to emerge, presumably has q = 1.5, then undisturbed Keplerian flow is uncondi- as the end-state of the inverse cascade of Cabot (1996). tionally stable. Attempts at turning the Keplerian flow 12 Lyra & Umurhan

Vertical Rotation Axis: Z 2 2 (R + l ,Z + l ) 1 R 1 Z

1 1 (R ,Z ) (R + l ,Z) 1 1 1 R Radius R

FIG. 4.— Rayleigh instability. Consider two annuli 1 and 2, with the first at radial-vertical position R1, Z1 and the second at R2, Z2. The energy argument, as originally found in Kippenhahn et al. (2012) and discussed recently in Lin & Youdin (2015), goes as follows: one interchanges the positions of the two circular annuli by preserving their individual angular momentum. If the total energy of the new system is less upon interchange then setup is regarded as unstable. In the diagram, the fluid ring 1 moves to radial position R1 + λR = R2 and then onto vertical position Z1 + λz = Z2 while ring 2 executes the opposite motion. A circular final state requires energy dissipation either in the form of heat or radiative losses – a feature that is implicit in this energy argument. In the case of purely barotropic Keplerian flow, which is stable according to this analysis, the interchange in the vertical direction is superfluous as the rotational speeds are the same. The vertical interchange matters when analyzing the conditions for the VSI as the vertical shear in the Keplerian flow figures prominently in driving the instability. Figure adapted from Umurhan et al. (2013). unstable have to focus on how the Rayleigh criterion is 4.2. The Solberg-Høiland criteria violated. In the presence of buoyancy, the Rayleigh criterion It is an equivalent statement to the Rayleigh crite- is substituted for the Solberg-Høiland criteria. These rion formulated above that angular momentum must criteria have been derived elsewhere (Tassoul 1978; increase outward. Given the angular momentum L = Abramowicz et al. 1984; Ogilvie 2016), but we repeat it 2 R Ω, then its derivative is here for completeness largely in part because these cri- teria figure prominently in the processes discussed in this review. If we include the adiabatic energy equation, dL = ΩR(2 − q) (59) stated in terms of pressure, dr ∂ρ and thus the condition q < 2 is equivalent to dL/dR > 0, + (u · ∇) ρ = −ρ∇ · u, (60) i.e., angular momentum should increase with distance. ∂t ∂u 1 The classical way to understand this criterion is by ap- + (u · ∇) u = − ∇p + g, (61) pealing to an annular ring interchange argument as ∂t ρ summarized in the cartoon Fig. 4. Consider two annuli ∂p one at an inner radius and disk height, R , Z respec- + (u · ∇) p = −γp∇ · u, (62) 1 1 ∂ tively, and the second at an outer radial and disk height t 0 0 position, R2 = R1 + λR, Z2 = Z1 + λZ respectively. We and linearize the system uR = uR, uφ = Ω(R,z)R + uφ, posit the following dynamical action: we interchange 0 0 0 uz = u , p = p (R,z) + p , and ρ = ρ (R,z) + ρ , then the positions of the two annuli under the condition that z 0 0 their angular momenta are conserved. We place the in- 0 0 0 terchanged annuli into circular orbits and we assess the −iωρ + uR(∂Rρ0 + ikRρ0) + uz(∂zρ0 + ikzρ0) = 0, new total energy of the configuration. If the energy in (63) this final state is lower than the original arrangement ik ρ0 then this new configuration is preferable and the sys- − 0 − Ω 0 R 0 − ∂ iωuR 2 uφ + p 2 R p0 = 0, tem is unstable. If upon interchange the total energy ρ0 ρ0 is higher, then this system is not preferred and the sys- (64) tem is stable. The key feature here is that we posit that 0 0 0 the annuli are placed into circular orbits and this implic- −iωuφ + Ω(2 − q)uR + ∂z(Ωr)uz = 0, itly means that energy must be either gained or lost by (65) the system as a whole. For barotropic Keplerian flows, ik ρ0 the vertical interchange has no effect as the angular mo- − 0 z 0 − ∂ iωuz + p 2 z p0 = 0, mentum is constant on cylinders (i.e., dL/dZ = 0) and, ρ0 ρ0 thus, the criterion for a lower energy state is equivalent (66) to dL/dR > 0. The details of this relatively straightfor- 0 0 2 0 2 ward analysis may be found in Tassoul (1978) or Kip- −iωp + uR(∂R p0 + ρ0cs ikR) + uz(∂z p0 + ρ0cs ikz) = 0, penhahn et al. (2012). (67) Hydrodynamical instabilities in protoplanetary disks 13

2 2 2 2 where we have also substituted p0 = ρ0cs /γ. We define so A = Nz and D = κ + NR. As for B and C, there is freedom to choose them from the kRkz terms in Eq. (74)

Ak ≡ cs ∂k lnρ, (68) −3 2 B + C = −r ∂z L + GRSz + GzSR. (81) G ≡ γ−1c ∂ ln p, (69) k s k Notice that we can construct P as symmetric by setting S ≡ G − A k k k, (70) B = C = (B + C)/2. The eigenvectors of symmetric P as normalizations of the density, pressure, and entropy form an orthogonal basis, and k can be represented from gradients in units of frequency. Solving the system linear combinations of the eigenvectors. Eq. (77) thus yields the dispersion relation means that ω2 is an eigenvalue of P. The characteristic equation Det(P − ω2I) = 0 is h i ω4 − ω2 2(2 − q)Ω2 + A G + A G + c2(k2 + k2 ) z z R R s z R ω4 − (A + D)ω2 + (AD − B2) = 0 (82) 2 2 2 2 2 + cs 2(2 − q)Ω kz − GRSRkz − GzSzkR i.e., a biquadratic with a = 1, b = −Tr(P), and c = Det(P).   −kRkz 2Ω∂z(Ωr) − GzSR − GRSz The eigenvalues are thus

+ C1 + iC2 = 0, (71) q 2ω2 = Tr(P) ± Tr2(P) − 4Det(P) (83) with they will be positive if (i) Tr(P) > 0; and if the square 2 root is always greater than Tr(P), that is, if (ii) Det(P) > C1 = 2(2 − q)AzGzΩ − 2ARGzΩ∂z(Ωr), (72) 0. Conditions (i) and (ii) are the necessary and suffi- G − G Ω∂ Ω A G − A G C2 = cs(kz R kR z)[2 z( r) + z R R z].(73) cient conditions for stability. The trace condition gives 2 the first Solberg-Høiland criterion Keeping only the leading order terms in cs, i.e, the cs terms, we filter out the acoustic modes. The resulting 2 1 anelastic dispersion relation is κeq − ∇p · ∇s > 0 (84) ρcp  2  2 2 2 2 1 ∂L 2 For the determinant condition, we need to define the ω (k + k ) = k − GRSR − k GzSz R z z r3 ∂r R term B = C. We split the kzkR terms in Eq. (74) as  1 ∂L2  + k k − + G S + G S , (74) 2 R z r3 ∂z R z z R B = −κz + GRSz, (85) C = GzSR. (86) where we substituted L = Ωr2 for the angular momen- tum and recognize That B = C follows from the thermal wind equation which can be proven by the following: in equilibrium we have 1 ∂L2 κ2 ≡ 2(2 − q)Ω2 = , (75) eq 3 ∂r 1 r Ω2rrˆ = ∇p + ∇Φ, (87) 2 2 1 ∂L ρ κz ≡ 2Ω∂z(Ωr) = , (76) r3 ∂z and taking the curl of Eq. (87) produces as the squares of the epicyclic frequency κ and a ver- eq   1 tical oscillation frequency κz. Notice that we can cast ∇ Ω2r × rˆ = − ∇ρ × ∇p. (88) the dispersion relation Eq. (74) into the following ma- ρ2 trix form, The term in the RHS is the baroclinic term. From the definition of entropy we write ρ in terms of p and s kTPk ω2 = , (77) T 1 1 k k ∇lnρ = ∇ln p + ∇s, (89) where k is the disturbance wavevector, and the matrix γ cp P is a 2×2 matrix so the baroclinic term is   1 1 AB 2 ∇ρ × ∇p = ∇s × ∇p P = , A = −GzSz, D = κ − GRSR. (78) ρ2 ρc CD p = S × G. (90) The partial derivatives of pressure (G) and entropy (S) combine into the Brunt-Väisälä frequencies Substituting Eq. (90) into Eq. (88), we recover for the φˆ direction

2 1 ∂p ∂s 2 NR ≡ −GRSR = − , (79) −κz + GRSz = GzSR, (91) ρcp ∂R ∂R proving that the matrix P is symmetric. The determi- 2 1 ∂p ∂s Nz ≡ −GzSz = − , (80) nant condition Det(P) > 0 becomes ρcp ∂z ∂z 14 Lyra & Umurhan

and reach pressure equilibrium with its new environ- 2 2 ment. As a result of the expansion, the fluid rings now −GzSz(κeq − GRSR) + GzSR(κz − GRSz) > 0. (92) exhibit different densities which are entirely dictated by the entropy gradient of the fluid, as indicated in Eq. (74). Factoring out Gz from both terms and simplifying re- This interchanged state is unstable if the total resulting veals energy – which is now a sum of both the kinetic energy and the gravitational potential energy – is less than what  2 2 −Gz Szκeq − SRκz > 0. (93) it was in the undisturbed configuration. With respect to Fig. 4, if the gravity vector were pointed down and to- Substituting for the gradients leads us to the second con- ward the star (indicated by ∇p), then the interchange dition would be destabilizing if the adjusted density of parcel 1, having been translated to parcel 2’s original position, ∂p  ∂s ∂L2 ∂s ∂L2  is less than parcel 2’s original density before the inter- − − > 0. (94) change. Similarly, the interchange would signal poten- ∂ ∂ ∂ ∂ ∂ z z r r z tial instability if the adjusted density of parcel 2, having Eq. (84) and Eq. (94) are the Solberg-Høiland criteria been translated to parcel 1’s original position, is greater for stability. If either criterion is violated, then it means than parcel 1’s original density before the interchange. that there are sets of disturbances – characterized by Such a resulting density change means that the total some range wavevectors k – for which the system must gravitational potential energy after the interchange is be linearly unstable. less than with which it began. We are reminded that the We call attention to an often overlooked consequence interchange physics ultimately rests on the work done of the Solberg-Høiland criteria. The determinant condi- by the system upon the interchanged fluid rings and this work is ultimately measured by the terms ∂ p∂ s found tion AD − B2 > 0 implies that A and D must have the k k > in Eq. (74). Thus, while an adverse entropy gradient is same sign. Since the trace condition A + D 0, then A destabilizing according to this energy principle and is and D must each be positive, which implies the essence of the Schwarzschild criterion, instability of the disk fluid requires that the total energy of the in- 2 terchanged fluid elements to be less than in the undis- Nz > 0, (95) 2 2 turbed state. As such, the Solberg-Høiland criteria, em- κeq + NR > 0, (96) bodied in Eq. (84) and Eq. (94), dictate the properties of the disk gas required to insure that the resulting energy have to be satisfied independently for stability. In Eq. (84) of every possible swapped fluid-ring state is higher than the dot product notation is a mathematically compact with which it began – hence, predicting a sufficient con- 2 2 way to write the condition; physically, the κep + NR and dition for stability. N2 z terms are working on different modes. If either of 4.3. A magnetic interlude them are negative, then the trace can still be positive but the second condition (Eq. 94) is negated. Before introducing the hydrodynamical instabilities, Eq. (95) is the usual Schwarzschild criterion assess- it is instructive to recall the basic physics of the MRI and ing the buoyant stability of a vertically stratified fluid. how it violates the Rayleigh criterion. Notice that it follows directly from Eq. (74) by setting 4.3.1. Magnetorotational instability kz = 0. Eq. (96), which follows from Eq. (74) by setting kR = 0, Consider that the gas parcels pertaining to the rings states that kz oscillations that are stable to epicyclic mo- A and B in a Rayleigh-stable situation (upper panel of tions can be de-stabilized by buoyancy, and vice-versa. Fig. 4) are connected by a tether with a restoring force Notice that the criterion is derived for adiabatic motion, −Kx, where K is a spring constant. The equations of mo- so it will be modified in the presence of finite thermal tion are diffusion or cooling, in principle allowing for unstable growth even when the adiabatic criterion is obeyed. In- deed, that is the driving force of the convective oversta- x¨ − 2Ωy + Kx = 0 (97) bility (Sect. 4.4.2). y¨ + (2 − q)Ωx + Ky = 0 (98) Eq. (94) pertains to the stability of the mixed kRkz mode. Violation of Eq. (94) is the driving mechanism Consider the Lagrangian displacement for the vertical shear instability, (Sect. 4.4.1). q(t) = q¯ + ξ(t), Physical Interpretation: The Solberg-Høiland criteria iωt with q¯ = 0 and ξ = q0e , where q is a generalized co- may be rationalized in the same way the Rayleigh crite- ordinate. The derivatives are rion is understood by appealing to the fluid-ring/parcel interchange analysis motivated in Sect. 4.1. In partic- q˙ = iωξ and q¨ = −ω2ξ (99) ular, the contribution to the criterion by the buoyancy term is often referred to as the Schwarzschild criterion: and the equations of motion are Referring once again to Fig. 4, we consider the angu- lar momentum conserving interchange of the two fluid −ξ ω2 − 2Ωξ iω + Kξ = 0, (100) rings as depicted. In their new interchanged circular or- x y x 2 bits, we allow each fluid ring to adiabatically expand −ξyω + (2 − q)Ωξxiω + Kξy = 0. (101) Hydrodynamical instabilities in protoplanetary disks 15

This system is solved to yield the dispersion relation frozen on the electron fluid, being these the most mobile charge carrier. In the case of a weakly ionized gas, the 4  2 2  2 most probable collision a electron would suffer is with ω − 2K + κ ω + K K − 2qΩ = 0 (102) a neutral. These processes enter the induction equation as the electromotive force and Ohmic diffusion, respec- which is a biquadratic equation. The solution is tively

2
2 p 2 2 2K + κ ± 4K (K − 2qΩ ) ∂B ω = . (103) = ∇ × (ue × B − η∇ × B), (109) 2 ∂t

If where ue is the velocity of the electron fluid and η the Ohmic resistivity. The ratio of the two terms is the mag- K − 2qΩ2 < 0 (104) netic Reynolds number then ω2 will have an imaginary component, and thus UL we have at least one growing root. Eq. (104) is the con- ReM = , (110) η dition for instability. We see that if K = 0 (no tether), the condition for stability is simply q > 0, i.e., the angular defined on dimensional grounds. Notice the symmetry velocity decreasing outward. This is satisfied in Keple- with the usual (hydro) Reynolds number, with viscos- rian disks. ity replaced by resistivity. At any length scale L, mag- The situation with the tether is identical to how mag- netic effects will be suppressed if this quantity falls be- netic fields operate. Let us include the Lorentz force in low unity. The velocity associated with magnetic fields Eq. (2), i.e., is the Alfvén velocity vA , and the length of interest is −1 −1 the MRI scale, given by Eq. (108), k ≈ vA Ω , so the 1 MRI F → F = (∇ × B) × B condition for the operation of the MRI in the presence of Lorentz 4π resistivity is  B2  (B · ∇) B = −∇ + (105) v2 8π 4π Λ ≡ Re (v ,k ) = A > 1 (111) M A MRI Ωη where B is the magnetic field. The second line above de- composed the Lorentz force into magnetic pressure and where we defined the Elssaser number, Λ, the magnetic magnetic tension, respectively. Consider now the mag- Reynolds number associated with the MRI. netic tension with a uniform base magnetic field B0eˆ, Of course, Ohmic resistivity, the drag between elec- wheree ˆ is an arbitrary unit vector, upon which we su- trons and neutrals, is not the only impediment to mag- 0 perimpose a Lagrangian field displacement B = B0ikξ. netic coupling. As also stated in the introduction, the The resulting acceleration is drift between electrons and ions when ions are coupled to the neutrals (Hall effect) and ions and neutrals when B ∂B0 k2B2ξ ions are coupled to the electrons (ambipolar diffusion) 0 = − 0 = −(k · v )2 ξ, (106) 4πρ 4πρ A are the two other non-ideal effects. We can decompose the electron velocity u in terms of the ion and neutral p e where we used ∂ = −ik and substituted vA = B/ 4πρ velocities ui and u as (Wardle & Königl 1993; Balbus & for the Alfvén velocity. We see from Eq. (106) that the Terquem 2001) magnetic tension on the perturbation depends on the negative of the displacement, behaving exactly like a ue = u + (ue − ui) + (ui − u) (112) spring, with effective spring constant the first term is the reference frame velocity (following K ≡ (k · v )2. (107) the neutrals). The second is the Hall effect and the third A is ambipolar diffusion. If we consider a singly ionized Therefore, the instability condition is species, the current J = ∑j njZjeuj, where e is the ele- mentary charge and nj and Zje are respectively the num- (k · v )2 − 2qΩ2 < 0 (108) A ber density and charge of species j, becomes Because the magnetic field appears multiplied by the wavevector, even weak fields generate significant ten- J = nie(ui − ue) (113) sion at small enough scales. Indeed the MRI is a weak field instability, and should be generally present when As for the ambipolar term, in collision equilibrium the gas is magnetized. the drift between ions and neutrals is set by, in the ion equation of motion, equating the Lorentz force with the 4.3.2. Suppression of the MRI neutral-ion collisional drag (Balbus & Terquem 2001; Bai & Stone 2017), so that While full ionization is the state of the hot disks around black holes, protoplanetary disks are cold and J × B poorly ionized, as noted in the introduction. In this case, = γiρiρ(ui − u) (114) c neutrals compose the bulk of the flow, and ions and elec- trons drift through them. The magnetic flux is mostly and Eq. (115) can be cast as 16 Lyra & Umurhan

−13 where x13 means ionization fraction in units of 10 , J J × B β the plasma beta parameter (β ≡ 2cs2/v2 ) in units of ue = u − + , (115) 4 A n e cγ ρ ρ 4 −9 3 i i i 10 , ρ9 the gas volume density in units of 10 g/cm , where c is the speed of light. The induction equation T300 the temperature in units of 300 K, and RAU the dis- then becomes tance in AU. Under the approximations made, at 1AU in the MMSN, Ohmic resistivity will be the dominant   impedance, but the problem is very model-dependent, ∂B 4πηJ J × B (J × B) × B and different combinations will yield different Elsasser = ∇ × u × B − − + ∂t c nee cγiρρi numbers at different radii. The resistivities do not con- (116) sider either the presence of grains, which will modify We can cast the Hall and ambipolar terms in terms them (Turner & Drake 2009, but see Simon et al. (2018) of an effective resistivity and a current, similarly to the for a different opinion). Ohmic term We assume that the disk will have a region that is Ohmic-dominated, and concentrate from now on our analysis on this region. ∂B  = ∇ × u × B − ∂t 4.3.3. An interlude within an interlude: Pure Hall MHD and Generalized Potential Vorticity 4π  J × B B × (J × B)   ηO J + ηH + ηA (117) Under conditions in which both the Ohmic and Am- c B 2 B bipolar coefficients are greatly dominated by the Hall term, i.e. when η  η , and η  η , we might con- where we now switched η → ηO for symmetry. Com- O H A H paring Eq. (116) and Eq. (117), the Hall and ambipolar sider the induction equation (Eq. (117)) in the identical resistivities are limit ηO = ηA = 0 and find that it reduces to ∂B  m J × B  = ∇ × u × B − e , (126) Bc ∂ ηH = (118) t xe ρ 4πnee 2 where we have restored the terms defining ηH. If we as- B x Q = ηA = (119) sume the following, (i) is a constant, (ii) 0 which 4πγiρρi means that the gas entropy is materially conserved, and (iii) we neglect gas viscosity by setting T = 0, then Holm We quote the Ohmic resistivity for completeness (1987) showed that combining Eqs. (1-2) and Eq.(8) with (Krall & Trivelpiece 1973; Balbus & Terquem 2001; War- Eq. (126) reveals that the Hall MHD equation set pos- dle 2007; Lesur et al. 2014) sesses a magnetic generalization of Ertel’s theorem. In particular, there exists a magnetic generalization of po- m c2 n tential vorticity, denoted by Ω , which is materially e < > ≈ −1 1/2 2 −1 GPV ηO = 2 σcollu e 234 x T cm s (120) conserved by the flow. That is to say the scalar 4πe ne 1  xe  where me is the electron mass, x is the ionization frac- ΩGPV ≡ $ + B · ∇s, (127) tion, σcoll is the cross section of electron-neutral colli- ρ 4πme sions, and the angled brackets represent averaging over obeys the conservation equation the Maxwellian velocity distribution of the electrons. The resistivities allow for the definition of Hall and am-  ∂  + u · ∇ Ω = 0. (128) bipolar Elsasser numbers, in symmetry with the Ohmic ∂t GPV Elsasser number. In the limit where the ionization fraction goes to zero, v2 Ω A eneB Eq. (127) indicates that GPV limits to the classical con- ΛH ≡ = (121) −1 ΩηH cρΩ servation of fluid vorticity, ΩPV ≡ ρ $ · ∇s as dis- cussed in Sect. 2.3. The magnetic generalization of the v2 γ ρ potential vorticity quantity has been used to better un- Λ ≡ A = i i (122) A Ω Ω derstand the emergence and stability of flow structures ηA in confined plasmas. Similar explorations of the role this In general, the MRI will be present if all Elsassers quantity plays in the emergence of zonal flows in accre- numbers are larger than unity. The dominant resistive tion disks subject to Hall MHD have yet to be fully ex- effect is that of lowest Elsasser number. For the Mini- plored. mum Mass Solar Nebula model (MMSN, Hayashi 1981) and uniform ionization fraction we can write 4.4. Hydrodynamical instabilities Since the ionization level required to couple the gas to the ambient field is not always met (Blaes & Balbus Λ × −6 −1 −1 −1/2 5/4 O = 3.2 10 x13 β4 ρ9 T300 RAU (123) 1994), it leads to zones that are “dead” to the MRI (Gam- −5 −1/2 −1 −1/8 Λ = 3.2 × 10 x β (µ/2) R (124) mie 1996; Turner & Drake 2009). So, the quest for hy- H 13 4 AU drodynamical sources of turbulence continues, if only −2 −5/4 ΛA = 1.4 × 10 x13 ρ9 RAU . (125) to provide accretion through this dead zone. In this Vertical shear instability in discs 2621 Hydrodynamical instabilities in protoplanetary disks 17 section, we describe the three hydrodynamical insta- bilities discovered in the past few years and that may play a role in disk dynamics in the Ohmic zone. In the next subsection we present the Vertical Shear Instability (VSI), followed by the Convective Overstability (COV), concluding with the Zombie Vortex Instability (ZVI). Key to all of the following instabilities is the ques- tion of the thermal cooling times of the primary per- turbations that lead to the widespread instability. For disks this is controlled by radiative processes. Radiation driven thermal relaxation will have two natural limits depending upon the photon mean free path. We define 2 κR to be an effective material Rosseland opacity (cm /g), in other words, a frequency integrated absorption cross section that accounts for the cumulative absorption by both gas and dust per gram. Letting ρ be the mass den- Downloaded from sity of the local disk material, then the relaxation times −1 will depend upon how greatly `rad ≡ (κRρ) exceeds the length scale of interest, `, which is the length scale of a fluid disturbance. When `rad  ` then the thermal relaxation is optically thin while when `rad  ` then the http://mnras.oxfordjournals.org/ thermalFigure 11. relaxationEdge-on contours is optically of the perturbed thick. vertical The velocity thermal as a function relax- Figure 12. Time evolution of the sum of the (normalized) perturbed radial of R, Z and time for the model with H/R 0.05 initiated with random andFIG meridional. 5.— Time kinetic evolution energies of in the discs sum where of the the (normalized) temperature was perturbed initially ation times τr are then given6 by the= following approxi- perturbations of amplitude 10− cs. Note that for clarity, the grey-scale of radialconstant and on meridional cylinders, as kinetic a function energies of the thermalin disks relaxation where the time. temperature Note that mate form, based on the exact formulation found in0.35 Lin was initially constant on cylinders, as a function of the thermal re- the image has been stretched by plotting the quantity sign(vZ) vZ . only the τ Relax 0 and 0.01 cases show growth. & Youdin (2015) based on the derived expression× | found| laxation time (optically= thin cooling law), in units of orbital period, in Spiegel (1957), 2πΩ−1. Note that only the cooling times of 0 and 0.01 cases show corrugation modes. A more quantitative analysis of the growth rates growth.5.7 Thermal Reproduced relaxation from Nelsonin models et al. with (2013a).Ks(R)  2 2 3κRρ c ` associated with the individual modesV and` their contributions` to the We now consider models in which the initial entropy function,  3 ; rad ; overall perturbed kinetic energy16σSBT budget will be presented in future onlyKs(R), a follows function a strict of power-law density, functionp = p( ofρ), radius as is given the by case equa- in work. τr = (129) tion (6), and the mid-plane density, ρmid(R), follows a radial power-  c adiabatic flows in which the fluid temperature may be at University of California, Merced on March 25, 2015  V `  ` law given by equation (3). For all models except one, we adopt the  3 ; rad ; written as a function of density. Under strong stellar 16κRσSBT irradiation,values s 1 like and inp many0, leading parts tothe of protoplanetarymid-plane Mach number, disks, 5.6 Thermal relaxation in models with T(R) the temperature= − field= is mainly a function of the (cylin- where σSB is the Stefan-Boltzmann constant. Obviously, mid,beingconstantatallradii.OurnormalizationofKs(R0) sets drical)M radius from the central star, exhibiting very lit- theWe` now2 dependence consider the evolutionin the optically of models thick where regime we relax reflects the lo- mid 20. We also consider a single model with p 1.5 and tleM variation= in the disk-vertical extent containing= − most thecally diffusive isothermal nature assumption of the associated optically with thick the limit. fluid response to s 0, so that there is no initial radial entropy gradient. The entropy of= its inertia. Therefore, because temperature and den- perturbations.These two limits We evolve are captured the energy in equation the two in extreme (1) andther- intro- in this case is normalized so that mid 20 at R R0 1. These sity isolines do not coincide,M the= flow configuration= = is malduce relaxation thermal relaxation models. by The integrating optically equation thin (18).limit We is often adopt runs are listed in Table 1 as K1R–0 to K7R–0.01. said to be baroclinic which means ∇p × ∇ρ 6= 0. Under referredthe equation to as of Newton’sstate P (γ Law1) ofe and Cooling set γ in1.4. which The the gas We impose reflecting conditions at the meridional boundaries and = − = such baroclinic circumstances, a small amount of verti- sourceis assumed term toQ bein inviscid. Eq. (8) Power-law is given by profiles for the initial tem- consider inviscid evolution. As described in Section 2.1.1, these cal shear can be sustained. peratures, T(R), and mid-plane density, ρmid(R), are adopted with models are convenient to implement numerically because analytic 1 As evident in Eq. (28), the shear is proportional to the q 1andp 1.5Q in= equationsT − (2)T
and, (3). The aim of (130) these solutions can be obtained for the equilibrium density and velocity = − = − τ d radial temperature gradient, in a way that the angular models is to examine the robustnessr of the vertical shear instability speedfields through and momentum equations (3) is and lowest (16). As at such,the midplane these models and allow in- as a function of the thermal relaxation time, τ relax, defined in equa- us to explore the vertical shear instability as a function of the ther- where Td is some spatially specified forcing temperature 8 tion (18) and expressed as a fixed multiple or fraction of the local creasesmal relaxation in the time, diskτ normal, in discs directions. where the initial distribution of representing how the disk gas is forced by the radiation So long as somerelax physical agent maintains this baro- orbital period. These runs are labelled T10R-0.01 to T14R- in temperature no longer follows a power-law function of the cylin- field from the star, where T in τr gets replaced with Td. clinic configuration (like strong thermal forcing with Table 1. ∞ drical radius, but instead varies with both and . We note that In the optically thick limit, the heat source/loss term in gravity), then there may exist a conditionR Z in which an Eq.The (8) evolution is expressed of the as normalized a radiative perturbed diffusion kinetic expression energies for angularequation (16) momenta also demonstrates conserving that interchange a radial power of law two in annu-Ks(R) commonlymodels with knownrelaxation as times the inEddington the range 0 Approximation,τ relax is plotted with s 1 gives rise to an equilibrium vφ that varies with height Z ≤ ≤∞ lar gas= rings− of infinitesimal cross-section can result in in Fig. 12. It is immediately obvious that instability only occurs in at each radius R,butadoptings 0givesvφ independent of Z.The  a c  a lower total energy state= (Tassoul 1978; Kippenhahn either the locally isothermalQ = ∇ caserad (τ relax∇T40), or when τ relax (131)0.01 etnormalized al. 2012). sum This of the isradial the and physical meridional rationalisation kinetic energies isof plot- the orbits. All other simulations result3κ in the= perturbed kinetic= energy ted in Fig. 13 for thermal relaxation times in the range 0 τ R Rayleigh stability of barotropic Keplerian flows,≤ e.g.,relax as≤ contained in the initial seed noise decaying with time. We note discussedlocal orbits. in Sect. The model 4.1:as with thes vertical0andp uniformity1.5 employed of the a where σSB ≡ aradc/4, and arad is the radiation density ∞ = = − that the case of τ relax is directly comparable to a previous meanrelaxation flow time stateτ relax means0.01 only orbits evaluating and is labelled purely ‘Trelax radial0.01, in- constant. = ∞ = = study on the adiabatic evolution of a stratified disc by Rudiger¨ et al. terchanges.s 0’. As it happens to be the case for Keplerian = (2002). The authors considered the hydrodynamic stability under Interestingly, for the−3/2 model with s 1andp 0, all values 4.4.1. Vertical Shear Instability flows with Ω ∼ R , this act results= − in a higher= energy the Solberg–Høiland criterion and also find stability in this case. configurationof τ relax result in and,growth thus, of the is perturbed not preferable meridional state and radial and theen- The vertical shear instability, as the name suggests, Our results indicate that the vertical shear instability requires that systemergies, and will only want the strictlyto return adiabatic to its simulation original configuration.with τ relax drawsthe initial on temperature the free profileenergy of of the vertical fluid is re-established shear in disks. rather shows eventual decay over long time-scales of 100 orbits.= It∞ is ∼ Accordingrapidly during to dynamical the Taylor-Proudman evolution, at least theorem, for the equilibrium rotating noteworthy8 This is because that this the adiabatic temperature model decreases is unstable with according radial distance. to the barotropictemperatureflows and density under profiles the adoptedactionof in these a central particular conserva- models. Ifsecond the radial Solberg–Høiland temperature profiles criterion were discussed to increase, in Section then the 3, situationand this would be reversed with angular speeds and momenta decreasing with tiveWe discuss force the must implications exhibit of constant this for applicability rotation onof the cylinders. instability is probably the cause of the initial growth in perturbation energy. By barotropic we mean to say that the fluid pressure is increases away from the mid-plane in the disk normal direction. to protoplanetary discs in Section 7.1. The subsequent decay may arise because the instability causes the 18 Lyra & Umurhan Vertical shear instability and vortices 3579

Figure 9. Vorticity profile in the mid-plane and meridional plane after 201, 317 and 401 orbital periods, with h 0.2 and τ 0.5. FIG. 6.— Midplane and meridional plane vorticity profiles of the VSI in models utilizing the simple Newton’s= Law of Cooling= formalism. Snapshots shown: after 201, 317 and 401 orbital periods. In these models h ≡ H/R = 0.2 and cooling time τ = 0.5orb. Figures are reproduced from TableRichard 1. Results et al. (2016). for the h 0.2, q 2, p 1.5 simulations. First col- The simulation with a negative mid-plane entropy gradient that = = − = − umn shows cooling time, second column shows growth rate of the VSI, third was run to compare with the previous run with positive entropy column shows peak vorticity perturbation (in units of the local orbital an- gradient had identical parameters (h 0.2, q 1, τ 0.1) except gular velocity), and fourth column shows mean aspect ratio of the emerging for the mid-plane density power law= which= was− set= to p 2. vortices. Nelson et al. (2013) showed that the density power-law index= makes− essentially no difference to the growth rate of the VSI, so we expect Cooling time Growth rate Vorticity Vortex aspect ratio the VSI to develop in much the same way during the early stages (orbits) (orbits) 1 ω /# − z of these two runs. This is exactly what we observe, and in fact 0.5 0.31 0.3 6 the long-term outcome of the simulation is very similar to the one − 0.1 0.9 1.1 2 with positive entropy gradient: vortices with aspect ratios χ 8are − 0.05 1.7 3 1.5 ∼ − formed, and these live for approximately 15 orbital periods before dissolving into that background flow. This result indicates that for this particular set of disc parameters, the SBI has essentially no lived vortices in the flow if appropriate conditions are present in the influence on the formation or amplification of vortices. disc. The results obtained in this section are somewhat puzzling as it In an attempt to clarify this point, we present two simulations is to be expected that the SBI should cause the amplification of the that have either a positive or negative mid-plane entropy gradient, vortices that are formed by the VSI. Furthermore, the run described respectively. The model with a positive entropy gradient has disc earlier in Section 5.3 with h 0.2, p 1.5, q 2andτ 0.5 = = − = − = parameters h 0.2, q 1, p 3andτ 0.1 (note the shallower showed evidence that vortices formed by the VSI were maintained temperature profile).= Fig.= −12 shows= − the evolution= of the vorticity in against decay by the action of the SBI. In order to explore this issue thisFIG simulation.. 7.— Simulations The results of the of VSI this with run are a realistic very similar radiative to those transfer de- model (Stollfurther, & Kley we allowed 2014). Left: the Velocitysimulation in the with meridional negative direction, entropy gradientuθ, in units of local Keplerian velocity for an isothermal simulation without viscosity. The vertical velocities follow a similar pattern. Right: the vertical scribedvelocity earlier in the in midplane the paper of (which the disk all for had a negative 3D model entropy after 4000 gradients orbits. The nearlydescribed axisymmetric earlier in property this section of the to saturated evolve instability, until it had even reached in the a inostensibly the mid-plane): statistically the VSI steady develops state, is axisymmetrically, clearly visible inthese and vortices and in simulationssaturated of Nelson state, et and al. we (2013b). then restarted Figure reproduced it with longer from coolingStoll & Kley times are(2014). formed from the band of perturbed vorticity that is created by (both τ 0.5 and τ 1). In both cases, we observed that the = = the VSI. We find that these vortices have quite large aspect ratios vortices decayed and the VSI stopped operating, such that the disc (χ 8), but the vortex lifetimes measured in the simulation are evolved towards a laminar state. This suggests that either the cooling typically∼ in the range 15–20 orbital periods, considerably shorter time-scale needs to be fine tuned in order to observe the SBI in this than those measured in the runs discussed earlier in this paper where model, or perhaps a more likely explanation is that the disc model large aspect ratio vortices were formed. with h 0.2, p 1.5, q 2andτ 0.5 has a steeper entropy = = − = − =

Downloaded from https://academic.oup.com/mnras/article-abstract/456/4/3571/1032344 MNRAS 456, 3571–3584 (2016) by NASA Ames Research Center user on 15 March 2018 Hydrodynamical instabilities in protoplanetary disks 19

This is one way toward understanding both the Solberg- the bulk inertia of the disk mass. A linear analyses Høiland criteria as well as the physics of axisymmetric in the infinitely fast relaxation time limit (i.e., τr → 0) disk inertial oscillations (which play a further role be- found in Barker & Latter (2015a) confirm the growth low). rate trends reported earlier while raising concerns about When thermal and viscous diffusion are an impor- the appearance of surface modes 10, while the analysis tant feature of the physics, then the flow is now baro- in Umurhan et al. (2016a) verified that wavemode with clinic and the stability properties of rotating flows are the fastest growth rates should occur for finite values of no longer controlled by the Solberg-Høiland criteria. Be- the radial disturbance wavelength, i.e., cause the mean-flow now exhibits some vertical varia- √ tion, a physical interchange of the type just described λR(max) = π|q|hH0, σ(max) = m|q|hΩ0/4, could lead to a configuration with a lower total energy, (134) see once again Fig. 4. With the restriction of an every- where m order the sequence of vertical nodes in the where isentropic and incompressible gas subject to wave- responses. All disturbances with odd values of m are like perturbations of the form exp(ikRR + ikzz) in which called corrugation modes, since the disturbance vertical kR and kz are respectively wavenumbers in the radial velocities are symmetric about the midplane. Responses and disk-vertical directions, a modified stability crite- with even powers of m are termed the breathing modes rion, obtained originally by Goldreich & Schubert (1967) in which the vertical velocities are symmetric with re- and Fricke (1968), in application to the radiative zones of spect to the midplane. The m = 1 and m = 2 mode differentially rotating stars, says that instability is possi- are the respective fundamental responses – these con- ble when tain the majority of the energy in developing VSI. All 2 2 ∂L kR ∂L normal modes with m > 2 (the overtones) increasingly − < 0, (132) ∂R kz ∂z exhibit power concentrated further away from the disk midplane. An immediate consequence is that a majority where L is the angular momentum. Because for baro- of the disk motions are projected onto the two funda- clinic disks the rotation rate is a function of both R and z, mentals with very little energy associated with higher unstable modes are guaranteed to exist since wavevec- overtones. Nonetheless, the linear theory demonstrates tors with ratios kR/kz that satisfy equation Eq. (132) can that modes show increased growth rates with increas- always be found. For a thin quasi-Keplerian disk, ac- ing node number m and this, in turn, raises some con- cording to Eq. (132) instability occurs for modes satis- cerns. Nonlinear numerical simulations show that only −1 fying kR/kz > h , i.e λR < hλz, meaning that unsta- the two fundamental modes get expressed as the pro- ble modes will have radial wavelengths that are much cess grows. It is possible that the faster growing high shorter than vertical ones. The instability can be under- vertical node modes, with increased vertical structure stood through the mathematical lens of the second of with distance from the disk midplane – saturate at low the Solberg-Høiland criteria, Eq. (96). Furthermore, the amplitudes are are suppressed by actual and numerical instability can be physically seen as arising directly as viscosity. For a further discussion on this mathematical a consequence of the ring-parcel interchange diagnostic matter see Barker & Latter (2015) and Umurhan et al. described in Sect. 4.2. While a purely barotropic Keple- (2016b). rian flow is Rayleigh stable, the interchange procedure The nonlinear simulations of the VSI reported in Nel- can result in a lower energy state after both a radial and son et al. (2013b) showed several features of the insta- a vertical translation of the fluid rings. bility as it develops into a nonlinear saturated state. Under conditions in which the thermal relaxation They show that early stages of growth is dominated by times are extremely short (τrΩ  1), the above crite- the fundamental breathing mode but eventually growth rion determines the onset of instability. The possibility switches over to the fundamental corrugation mode. As that this process is possible for protoplanetary disks was such, these two fundamental modes probably constitute first suggested in the analyses of Urpin & Brandenburg the primary injection scales of the resulting turbulence. (1998) and Urpin (2003), followed by the study of Arlt & When the system reaches a quasi-steady state, the ra- Urpin (2004a) who gave the first estimate of the growth dial scales of the resulting structures are much larger rate to be than the radial scales of the fastest growing modes, the σ ∼ |q|hΩ0. (133) latter being approximately H or slightly larger, and be- ing highly suggestive of some amount of inverse energy The first unambiguous identification of the instability cascade happening on these scales – but this has yet to in a realistic disk model came in the study of Nelson 9 be quantified. The simulations reported in both Nelson et al. (2013a) . In this work it was found that oscillat- et al. (2013b) and Richard et al. (2016) utilize the simple ing inertial modes become unstable (overstable) and the thermal "Newton’s Law of Cooling" model in which the fastest growing body modes have a corresponding radial gas is locally forced to a prescribed radial temperature wavelength λR ≈ πhH0. Body modes are inertial modes whose power is concentrated around the midplane of 10 Nelson et al. (2013b) and Barker & Latter (2015b) identified two the disk and, therefore, constitutes responses affecting classes of modes. In addition body modes, a second class of so-called surface modes were also identified. These are structures both attached 9 Arlt & Urpin (2004b) also performed nonlinear simulations but and concentrated on the vertical boundaries of simulations, and they the disk model they adopted assumed a Boussinesq equation of state. have fast growth rates. However, owing to the very low densities at The instability, though technically present, is difficult to identify. Also, these locations they hardly affect the emerging dynamics in the bulk using the Boussinesq equation of state predicts that onset to instability of the disk since they contain very little energy and, as such, are con- occurs at zero frequency, failing to capture the actual oscillatory nature sidered artefacts of imposing artificial boundary conditions (Umurhan of the VSI at onset. et al. 2016a). 20 Lyra & Umurhan profile on a cooling time τc, usually measured in units of local orbit times (orb= 2π/Ω0). Nelson et al. (2013b), who run simulations with h = 0.05, find that the VSI is most robust for τc → 0 and will operate with decreas- ing vigor up to a maximum cooling time τc = 0.1orb. Richard et al. (2016) show that the instability survives for up to larger values of τc for correspondingly larger value of h, e.g., see Fig. 7. When non-isothermal effects are introduced (i.e., τr > 0), buoyancy stabilizes otherwise unstable oscillations, according to the Solberg-Høiland criterion. For this rea- son, it is the exactly-isothermal response (τ = 0) that is most unstable one, and the instability ceases to ex- ist when even moderate levels of adiabaticity are intro- duced, as shown in Fig. 5. Lin & Youdin (2015) quantify the cooling time requirement for vertically isothermal stratification as h|q| Ω τ < (135) k c (γ − 1) where γ is the ratio of specific heats. The growth rates rapidly diminish and shutoff entirely once cooling times get much longer than this longer cooling times because FIG. 8.— A cartoon in four panels indicating the convective over- the unstable wavelength is shifted deep into the viscous stability mechanism. In panel (a) a fluid blob is embedded in a ra- dial entropy gradient. In panel (b) it undergoes half an epicycle and range. Furthermore, we observe that this condition is returns to its original radius with a smaller entropy than when it be- valid for the locally isothermal background state. For gun s1 < s0. It hence feels a buoyancy acceleration inwards and the vertical adiabatic structures, cooling requirements are epicycle is amplified. The process occurs in reverse once the epicycle less stringent. is complete, shown in panel (c), where now s2 > s0. The oscillations hence grow larger and larger. The impression of two-dimensionality Noting that disks are likely optically thick on the is deceptive. To conserve mass, a similar motion must happen in an- length scales on which the VSI operates, Nelson et al. other horizontal layer, in the opposite direction. This requires kz 6= 0. (2013b) equated the critical cooling time (τc ∼= 0.1orb) Reproduced from Latter (2016). to a corresponding radiative (diffusive) relaxation time predicting that the VSI ought to be operative for effec- midplane is attributable to the corresponding decrease 2 tive disk opacities κ < κc ≈ 1cm /gm. Subsequent real- of the mean densities there. istic simulations handling the thermal response of the Lastly, there are apparently two co-acting secondary gas using a radiative diffusion model have been per- transition routes into unsteady activity - one at- formed by Stoll and collaborators (Stoll & Kley 2014, tributable to the roll-up of vortices due to vertical mo- 2016; Stoll et al. 2017a,b). In particular, the impor- tions of the disk and to azimuthal motions in the plane tant study of Stoll & Kley (2014) demonstrated that the of the disk: Latter & Papaloizou (2018) show in a VSI remains robust in such realistic models and, fur- Boussinesq model that the saturation of the VSI into az- thermore, they show that its turbulent intensity is of imuthally oriented vortices occurs via parasitic Kelvin- medium strength (see further below), arguing also that Helmholtz modes (Goodman & Xu 1994), while Richard it likely plays a significant role in the planet formation et al. (2016) postulate that a second secondary transition process as parts of disks in which dust growth occurs occurs via the Rossby wave instability (Lovelace et al. likely support the conditions in which this process op- 1999; Li et al. 2001), leads to continual generation of ver- erates (see also Flock et al. 2017). tically oriented unsteady vortices. Indeed, Manger & The 3D modeling of the VSI (Nelson et al. 2013b; Stoll Klahr (2018) have recently found that long lived vor- & Kley 2014; Richard et al. 2016; Flock et al. 2017) shows tices can be seeded via RWI in a VSI-unstable disk, pro- that it develops into long-lived unsteady activity. Is it vided that the azimuthal domain is large enough (≥ turbulence? At this stage it is not clear. Large scale un- 180◦). And finally, with respect to the secondary tran- steady vortices are typical of the final emergent struc- sition by the Rossby wave instability, it should be noted tures. Despite the successes of these simulations, an in- that the manner in which the basic zonal flows emerge – ertial range has yet to be resolved since in most simu- upon which this secondary instability operates – has yet lations the injection scales are resolved up to 10-20 grid to be fully understood but it is possible that an inverse- points. The need to flesh out the inertial range of this cascade like the Rhines mechanism might be partly re- process forms the basis of future directions on the sub- sponsible. ject. Nonetheless, the global high-resolution modeling, in particular of Richard et al. (2016); Flock et al. (2017) 4.4.2. Convective Overstability have quantified the resultant turbulent speeds as being The second hydrodynamical instability discovered is a few percent of the local sound speed – meaning to the convective overstability, which, as the name sug- −5 −3 say that the midplane (α ≈ 4 × 10 – 10 ), increasing to gests, is a convective process. It contrasts with classical −2 −1 about 20% (α ≈ 10 – 10 ) in the corona (i.e., |z| > 4H0). convection because the direction it occurs is the radial, Note that the rise in the apparent values of α far from the not the vertical one. It manifests itself as epicyclic oscil- Hydrodynamical instabilities in protoplanetary disks 21

FIG. 9.— Linear and nonlinear evolution of the convective overstability. Upper panels: Convergence study with resolution (left) and initial amplitude of perturbation (right). Resolution of 64 points per scale height is enough to resolve the overstability, and initial amplitudes as small −10 as 10 times the sound speed (c0) are enough to lead to growth, demonstrating the linear nature of the process. The linear growth rate (black dashed curve) is well reproduced in all cases. Lower panels: With the linear overstability raising the amplitude of the initial fluctuations to nonlinear levels, a large-scale vortex is generated. The panels show the vertical vorticity. The Reynolds stress saturates at α ≈ 10−3. Adapted from Lyra (2014).

lations that are amplified by buoyancy, hence the name. The COV shares many similarities with the subcritical baroclinic instability (SBI), a process discovered earlier, when Klahr & Bodenheimer (2003), modeling disks with negative entropy gradients, found baroclinic growth of vortices. No linear growth was found in linear analy- sis (Klahr 2004), hinting at nonlinear origin. The nature of the instability was clarified in the works of Petersen et al. (2007a,b), who highlighted the importance of finite thermal inertia. When the thermal time is comparable to the eddy turnover time, the vortex is able to establish an entropy gradient around itself that compensates the large-scale entropy gradient that created it. The mecha- nism was elucidated by Lesur & Papaloizou (2010): the cycle starts with a fluid parcel buoyantly lifted to an outer orbit. As it travels outwards, it finds itself sur- rounded by gas of lower entropy. The resulting buoy- ancy lifts it even further outwards. Deflected by the Coriolis force and following the vortex streamline, the FIG. 10.— The sign of the square of the Brunt-Väisälä frequency de- fines the stability criterion, here shown as a function of the density and fluid now travels along the azimuthal direction. In this temperature power-law indices. The plot shows the lines for two val- leg, the fluid thermalizes, cooling down. As it cools, it ues of γ. Above (below) the respective line the system is unstable (sta- gets denser, and sinks back inwards. The last azimuthal ble). Adapted from Lyra (2014). The dots represent some disk data for leg will thermalize the fluid, heating it back to the tem- which temperature and density gradients have been estimated. The SBI may be an important process for the majority of disks in the sam- perature it had in the beginning, closing the cycle. The ple. original radial entropy gradient and the azimuthal tem- perature gradient established by the thermalization lead 22 Lyra & Umurhan ∇ × ∇ to a non-zero baroclinic term p ρ, generating vor- " # ticity locally and amplifying the vortex. A respectable 1 µ2βN2 σ = − R (137) amount of angular momentum is transported by spiral 2 β2 + µ2 κ2 + N2
density waves excited by the vortex (Heinemann & Pa- R paloizou 2009, 2012), leading to α values in the vicinity where β = 1/γτ is the inverse of the cooling time and −3 2 of 10 . µ = kz/k, with kz the vertical wavenumber, and k = Notice that the process requires the radial entropy 2 2 kR + kz, where kR is the radial wavenumber. gradient and the radial pressure gradient to be aligned The most unstable cooling time is (to give buoyancy) and thermal relaxation or diffusion

(to establish the azimuthal gradient). If the gas is lo- 1 k 1 cally isothermal a gas parcel has always the same tem- τmax = q (138) γ kz 2 2 perature as the surroundings and buoyancy cannot act κ + NR to amplify the motion. If the gas is adiabatic a gas par- cel cannot exchange heat so an outward/inward motion with associated growth rate cannot thermalize and return to its equilibrium position. 2 1 kz N The SBI operates optimally for cooling times similar to σ = − R (139) the dynamical time. max q 4 k 2 + N2 The convective overstability obeys practically the κ R same mechanism. Indeed, Klahr (2004) had already The fastest growth rates occur for kz  kR, i.e., relatively noted that the SBI could be regarded as a convective in- 11 stability in the radial direction, but modified by rotation flat and radially elongated epicycles . If we also as- and shear. The difference is the unstable mode. While sume Keplerian disks (κ = Ω  |NR|), the SBI uses a large scale vortex, in the COV the fluid executes a small epicycle. This unveils also a crucial dif- 1 τ = (140) ference between the two mechanisms: while the SBI is max γΩ nonlinear, the COV is linear. The cartoon of Fig. 8 illus- N2 trates the growth of a particular mode. As a gas parcel σ = − R . (141) moves into areas colder (hotter) gas, it loses (gains) heat, max 4Ω compresses (expands), and because of the extra buoy- Diffusion on the other hand is characterized by a ran- ancy, overshoots the original amplitude of oscillation. dom walk of photons, with a characteristic length scale The cartoon is illustrative and does not exactly repre- set by its Laplacian scaling. Latter (2016) finds the same sent the 3D axisymmetric most unstable mode. Being growth rate in the diffusion regime, now with a length- axisymmetric, the motion is 1D, not necessitating the 2 azimuthal direction. The need for the vertical direction, dependent cooling time β = χkz, where χ is the thermal diffusivity. Maximum growth occurs for β = Ω, setting shown by the kz dependency in the dispersion relation p comes about because of mass conservation: if a gas par- the wavenumber of maximum growth kz = χ/Ω, with cel moves inwards, an equal amount of gas must move same growth rate as in the optically thin cooling law. outwards. In order to conserve mass, the motion shown We show in Fig. 9 (left panel) the evolution of a 3D in the cartoon must be matched by the same motion ex- simulation of convective overstability. We set a box of ecuted in another horizontal layer, in the opposite direc- size 4H × 16H × 2H, with resolution 256 × 256 × 128 in tion. x, y, and z, respectively, including the linearized pres- The overstability exists both if the cooling is provided sure gradient. When the initial kz mode saturates (at by thermal relaxation (Lyra 2014) or by thermal diffu- 50 orbits), a sharp rise in enstrophy occurs, developing sion (Latter 2016). In the case that the cooling time τr into a large scale vortex. The saturated state is identified is finite, the Rayleigh and the Solberg-Høiland criteria with the SBI. are but the isothermal and adiabatic limits of the more Some words are warranted about the robustness of general anelastic dispersion relation (Klahr & Hubbard the COV as an astrophysical process relevant to disks. 2 2014; Lyra 2014; Latter 2016) The condition that NR < 0 requires (for negative radial pressure gradient)

qT qρ − < 0 (142) 3 i 2 2 2 i 2 (γ − 1) ω + ω − ω(κ + NR) − κ = 0. (136) γτr γτr where qρ is the power law index of the density gradient and qT the power law index of the temperature gradient. Thermal relaxation and thermal diffusion should For a column density following a power law mimic the limits of cooling in optically thin and in the 11 In the local box we could set kR = 0 and obtain a channel flow. optically thick regimes. The difference is that thermal Yet, while insightful, channel flows are an artifact of local boxes, not relaxation happens at all scales at the same time, as ex- possible in global disks where κ is a function of radius. Also, setting pected if photons are simply allowed to escape. The kR = 0 violates the short wave approximation for obtaining the disper- sion relation. It suffices to set kz  kR to obtain the maximum growth mode is therefore ballistic, with all scales of the disk ex- rate. Interestingly, this contrasts with the dominant VSI mode, which ecuting the same oscillation (in the local box only; in a is the very opposite, kR  kz, a tall slender mode. This should help global disk the epicyclic frequency is a function of ra- future simulations prone to both instabilities to be able to distinguish dius and height). The growth rate is between the processes. Hydrodynamical instabilities in protoplanetary disks 23

2 and the condition of N < 0 now becomes simply qT = −qΣ Σ ∝ ρH ∝ r (143) (γ˜ − 1)qΣ. We overplot in Fig. 10 densities and temper- ature gradients for a sample of disks from Andrews & we have for a Keplerian disk Williams (2007). According to this criterion we see that most disks in the sample should be susceptible to the qT 3 qΣ = qρ + − (144) SBI. 2 2 Finally, we caution that although the COV exists in The requirement of Eq. (142) is then the local box, its existence has not yet been established in global disks. The different between these models is (γ + 1) tied to what we just discussed previously. In a global 2q < q − 3 (145) Σ T (γ − 1) disk the entropy gradient is allowed to adjust to the flow, whereas in the local box it is hardcoded and en- which, for γ = 7/5 means forced. Global simulations of the COV would be use-   ful to resolve the issue. If the thermal diffusion is set 1 by the turbulence itself, with χ = αΩH2, then the maxi- qΣ < 3 qT − (146) √ 2 mally growing wavelength of the instability is αH, or roughly 0.07H. For H/R = 0.05 − 0.1, between 1500 and For qT = 1/2 the column density has to be flat or increas- ing with distance in order to lead to instability, which is 3000 points in the radial direction are needed to resolve the instability in global models. not reasonable. For qT = 3/4 the onset of instability in the midplane corresponds to qΣ = 3/4 (also for γ = 7/5). 4.4.3. Finite Amplitude Initiated Turbulence: the Zombie Vortex While this is consistent with the range of qΣ ≈ [0.4,1.0] Instability found in the observations of Andrews et al. (2009), it lies in the middle of the range and thus the COV may not be Prior to the discovery that the MRI is a viable process robust in the midplane. to drive turbulence in protoplanetary disks, researchers That does not mean that the COV is not present in long struggled (in vain) to identify a purely hydrody- general. The VSI does not formally exist in the mid- namic supercritical route to turbulence. During the hey- plane either, yet the unstable layers above and below are day of the MRI, many adherents of the hydrodynamic enough to drive the whole disk into a turbulent state as route began examining the possibility that a subcritical shown in numerical simulations. Indeed, for the COV, turbulence transition may be possible in protoplanetary there exists the possibility of unstable radial entropy disks (Balbus 2003). In this picture it is argued that gradients away from the midplane, even when the mid- while a Keplerian disk initiated with a low-amplitude plane is stable. The entropy gradient is velocity field should quickly shear away incipient shear driven dynamic activity, a relatively long-lived turbu- lent state could take root if the initial flow field were ∂ s(R,z) ∂ ln T(R,z) ∂ lnρ(R,z) = − (γ − 1) (147) either of sufficiently high-amplitude (i.e., with speeds ∂ ln R ∂ ln R ∂ ln R approximately that of the local cs) or seeded with per- turbations that result in strong transient growth which which, for the isothermal stratification given by Eq. (21), might trigger could trigger a secondary transition in the becomes flow (Chagelishvili et al. 2003; Yecko 2004). 12 Richard & Zahn (1999) examined the earliest Taylor-  2  ∂ s(r, Z) z Couette experiments – those of Wendt’s 1933 experi- = −q + (γ − 1) qρ − (3 − q ) (148) ∂ ln R T 2H2 T ments report and repeated by G.I. Taylor (Taylor 1936) – that are both stable to the Taylor-Couette instability and For the MMSN model (Hayashi 1981), with qT = 0.5 support a flow profile exhibiting an increase of angu- and qΣ = 1.5, then for γ = 1.4 the entropy gradient, lar momentum with axial distance. Interpreting this as which is 0.6 at the√ midplane, flips sign and becomes reasonably mimicking a Keplerian flow and extrapolat- negative at z = 1.2H ∼ H. Whether this is near the ing the measured laboratory torques (and correspond- midplane enough to render the whole disk column tur- ing effective turbulent viscosity) onto flows with the as- bulent remains to be shown by numerical simulations. tronomical values of Re appropriate to protoplanetary Notice also that yet another entropy gradient is to be disks, they argue that such disks should easily support considered for the nonlinear sibling of the COV, the SBI. a turbulent state owing to their small viscosities even Being essentially a vertically global COV with kz = 0, the with the smallest of initial triggers. Follow-up labora- relevant quantities to be checked for radial buoyancy tory work to test this proposition has been inconclusive are the column density gradient, and the vertically inte- with contradictory results reported (Ji et al. 2006; Pao- grated pressure and entropy. In this case, the vertical in- tegration of the polytropic equation of state passes from 12 The idea of transient growth is that certain small amplitude per- 1+1/n 1+1/n˜ turbations can result in large amplitude fluctuations in the perturba- p ∝ ρ to P ∝ Σ , with (Goldreich et al. 1986) tion energy in a strongly sheared flow. This notion has been shown to explain the transition to turbulence in viscous three-dimensional flows which are otherwise linearly stable (Butler & Farrell 1992). For n˜ = n + 1/2 (149) Keplerian disks, the amplification factor in the perturbation energy, which is equivalent to an effective adiabatic index given by Gmax, was shown in Yecko (2004) and Maretzke et al. (2014) to have a Re2/3 dependence. For a comprehensive review of the ideas 3γ − 1 of transient growth in strongly sheared flows see Schmid & Henning- γ˜ = . (150) son (2001). γ + 1 24 Lyra & Umurhan letti et al. 2012) and theoretical analysis of the labora- tory setting identify the source of instability as emanat- ing from the axial boundaries of the experiment caus- ing the laboratory measured outward transport of an- gular momentum (Avila 2012). Subsequent higher Re laboratory studies (Edlund & Ji 2014, 2015) and numeri- cal experiment (Lopez & Avila 2017) has shown that the observed turbulence increasingly recedes and becomes localized to the axial caps of the system and is entirely the result of triggered Ekman boundary layer flow phe- nomenon. These authors also show that as Re is in- creased, the flow increasingly appears laminar and self- similar within the bulk interior of chamber. The results of these efforts showcase the importance of doggedly assessing the character of these systems at increasingly higher values of Re as well as highlighting the dan- FIG. 11.— Cascade of baroclinic critical layers resulting in turbu- gers of deriving conclusions regarding the behavior of lence. Color coded is the Rossby number (gas vorticity over Keplerian protoplanetary disks based on relatively low Re quasi- vorticity) in an xy plane off the midplane (z = 0.4∆). The simulation was initialized with a vortex in the midplane, not visible at this height. Keplerian laboratory flows and their numerical experi- The vortex layers associated with baroclinic critical layers are seen at ments (Balbus 2017). the locations x/∆ = m for non-zero integer m. Anticyclonic vorticity is Nevertheless, the idea that finite amplitude pertur- indicated by blue, while cyclonic vorticity is red; the darkest red/blue bations may play a role in the life of a protoplane- colors correspond to Ro = ±0.10, while green indicates Ro = 0. (a) t = 64 orbits, (b) t = 256 orbits, (c) t = 576 orbits, and (d) t = 2240 orbits. tary disk has lead to the discovery of several processes Reproduced from Marcus et al. (2013). that can lead to dynamically interesting end-states like the Rossby Wave Instability (Lovelace et al. 1999) and the so-called Subcritical Baroclinic Instability (Klahr & buoyant resonances further on and, thereby, repeating Bodenheimer 2003; Petersen et al. 2007a; Lesur & Pa- the process and setting the whole disk into a turbulent paloizou 2010; Lyra & Klahr 2011). Each of these pro- state (Marcus et al. 2013; Umurhan et al. 2016b) 13. cesses appear to generally lead to the emergence of co- There are minimum criteria for filaments to cause this herent, quasi-steady, and relatively stable vortex struc- kind of eruption. The first of these is that the Rossby tures and while these are interesting in their own right – number (Ro) of the anomalous vorticity must exceed a e.g., in their ability to attract and accumulate dust grains critical threshold. The Ro is assessed based on the ve- required for planet assembly – they themselves do not locity fluctuation, v, around the base Keplerian state appear to play a role in driving hydrodynamic turbu- v. We are reminded that in a fluid shearing box rotat- lence in the sense we envision. ing around the central star with angular frequency Ω, the Keplerian shear is v = −(3/2)Ωx, where x is the ra- The zombie vortex instability (ZVI) is a recently iden- dial coordinate with respect to r0. If the radial extent of tified third hydrodynamical process that has the poten- the filament is `0 and we take v to be the deviation az- tial to initiate widespread turbulence in protoplanetary imuthal velocity characterized by some speed scale δv0, disks. Unlike the VSI and the COV, the ZVI is a finite then an estimate for Ro is amplitude instability. It differs from the aforementioned ∂v . δv finite amplitude examinations in that it requires there to Ro ≡ 2Ω0 ≈ . (151) be a sufficiently large vorticity field as opposed to a rel- ∂x 2Ω`0 atively large amplitude velocity perturbation. Typically, the nature of the filament will be identified Its mechanism of how a vortex column or sheet can as “cyclonic" if sgn(∂v/∂x) > 0 and “anti-cyclonic" oth- render a flow unstable and induce spreading can be ra- erwise (not including zero). Marcus et al. (2013) and- tionalized in a simple model, as shown in Fig. 11, which Marcus et al. (2015) showed that filaments must have is based on the original findings of Marcus et al. (2013). Ro > Ro ≈ 0.2 in order for the ZVI to be initiated. How- In this study, a tubular vortex filament is oriented in the c ever, Umurhan et al. (2016b) also showed that the ZVI azimuthal direction in a constantly vertically stratified may be instigated for lower values of Ro approaching fluid in a local shearing box setting with a locally ap- c even zero. What is important to note here is that mini- plied Keplerian shear. Azimuthal vibrations of the vor- mum requirement for the filament’s vorticity must have tex filament (Rossby waves) induce a response in the its Ro > Ro , so it implies that there is no minimum per- fluid around it with an associated frequency. At some c turbation velocity needed to get the instability to go. Only radial distance from the filament, this Rossby wave fre- the filament’s vorticity matters since a radially-thin az- quency will appear Doppler shifted due to the Keple- rian shear. This Doppler shifted frequency will find res- 13 The name of “zombie”, a fictional undead creature resulting from onance with a buoyancy frequency at some radial dis- the reanimation of a corpse, is given because these vortices happen in tance away in the disk, amplifying the localized buoy- the Ohmic "dead" zones of protoplanetary disks. Because the process ant perturbations. The location and very narrow radial gives rise to new vortices by the excitation of critical layers – that in turn replicate the process – the effect reminds one of the spread of an extent where this resonance operates is called a "criti- infection. Another analogy would be that of tuning a guitar. Strike a cal layer". The amplified perturbation drives the gen- note in a string that another string is tuned to, and the resonance will eration of new filaments which also turn into vortices. make the unstruck string vibrate. An attempt to rename the process In doing so, the newly generated filaments excite other “Guitar String Instability” will not be undertaken. 8 Hydrodynamical instabilitiesZVI III in protoplanetary disks 25 The Astrophysical Journal, 833:148 (14pp), 2016 December 20 Marcus et al.

(a) (d)

FIG. 12.— Anomalous vertical vorticity profile from fully developed ZVI simulations. Notice the banded structure in vertical vorticity, implying weak undulations atop the basic Keplerian profile. Repro- duced from Barranco et al. (2018). Figure 7. Energy spectra E(k ) of the non-Keplerian component of kinetic energy of late-time zombie turbulence for the same five runs in Figure 6. The right panel is a blow-up of the left panel. The figure shows the late-time flows are all attracted to the same energy spectrum, regardless of the initial conditions. Dissipation from FIG. 13.— Energy spectra of the non-Keplerian component of ki- imuthallyhyperviscosity elongated is responsible filament for the rapid with downturn velocity in energy scale at kδv030can0. The Kolmogorovnetic energy spectral in index ZVI of turbulence, 5/3 is indicated based by the on slope simulations of the thin solid in the black shear- lines. satisfyFor 80 thek minimum 300, the spectra vorticity are approximately requirement Kolmogorov, solong but for k as 80, thereing are box departures whose boxfrom size the −=5/H3 slope.. The The different peaks and lines valleys represent in this partdifferent of the spectra indicate the presence of large-scale structures in the flow. The vertical dotted line is at k =D2p , the wavenumber of the approximate cross-stream spacing of δv0/`0 remains finite as `0 → 0 (see also discussion in initial conditions: Kolmogorov noise with initial rms Mach number the large anticyclonic vortices and of the cyclonic layers. of 0.01 (black); Kolmogorov noise with initial rms Mach number of Sect. 3.3). 0.007 (green); Kolmogorov noise with initial rms Mach number of Because the ZVI relies on excitation of baroclinic crit- 0.004 (blue);fi noise with an energy spectral index of 1.0 and initial rmsfi icalThe layers, autocorrelation its growth time times of the are large-scale relatively cyclonic long layers once is of Machstrati numbercation, of and0.007/or (red); no rotation. and a run It initialized has been withfound an that isolated this an-rst the same order as autocorrelation time of the−1 large-scale ticyclonicfamily ofvortex critical (yellow). layers At is dif late-timeficult to flows excite, are which all attracted is consistent to the Ro > Roc, with growth rates of < 0.05 − 0.1Ω . The anticyclones, so they too are long-lived. samewith energy our own spectrum, numerical regardless experiments of the in initialMPJH13 conditions.. However, Vertical the ZVI operates most effectively in the limit of very long lines correspond to scales ≈ 0.1H which marks the beginning of the cooling times as demonstrated by Lesur & Latter (2016), k−5/3second type, baroclinic< k < critical layers, is easily excited by 4.2. Memory of the Critical Layer Eigenmodes spectrum. For 80 300, the spectra are approximatelyfl Kol- who examined the ZVI under a variety of thermal cool- mogorov.perturbations; Based onMPJH13 reportedshowed values that of α when∼ 4 × the10−4 owand ish perturbed= 0.05, a ing timesWenow (whether show that driven the patterns by optically of the persistent thin or optically anticyclones nominallocally value via a of small the Rhines vortex, scale then of theλb excited= πH, based critical on layers Eq. (47),(when al- thickand cooling) cyclonic and vortex showed layers that in late-time the critical zombie cooling turbulence time, are thoughobserved see text in forthe caveats. Galilean Reproduced frame in which from Marcus the perturbing et al. (2016). vortex (b) (e) τr, toset be by longer the “memory than 10” orbitsof the in neutrally order for stable the critical ZVI to layers, be is stationary) have a temporal frequency s=0. The equations dynamicallyrather than significant. initial conditions This or any corresponds other properties to approxi- of the flow. ingof (see motion below). and The boundary emergence conditions of large-scale are invariant organized under In MPJH13, we analyzed the linearized> equations2 for a rotating, zonaltranslation flows in (see the below) cross-stream suggests direction thatx aby subtle any interplaydistance δ, mately a minimumfi gasfl opacity of 50cm /g. This can be rationalizeduniformly strati sinceed theuid, ZVI assuming resonance eigenmodes mechanism proportional re- to ofif both there processes is also a may Galilean be at shift play. to a frame moving in the exp[(ikyyz+- kz st )] (i.e., periodic in streamwise y and The injection scales are those of the critical layers. Un- quires buoyancy oscillations to be maintained over long streamwise direction y with velocity -()32W d (this is the vertical z directions). The eigenequation obtained from this like in purely hydrodynamic shear flows, where0 the size timescales. For this reason, the second criterion is that same invariance that is exploited when one uses shearing box the processlinearization requires(a generalization the gas to beof the close Rayleigh to adiabatic, equation with for the of the critical layer is set by the viscosity of the gas and inviscid stability of shear flows (Drazin & Reid 1981)) is a boundary conditions−1/3 ). Due to this invariance, we are free to very long cooling times. scales like Re , the size of the critical layers (∆c) driv- second-order ordinary differential equation in which the choose an arbitrary origin of the x coordinate. So how then do The spawned vorticity will have an anomalous vortic- ing the ZVI is set by the intrinsic linear growth rate coefficient of the highest-derivative term is we interpret Equation (17) with respect to the location√ of the ity that is jet-like (Umurhan et al. 2016b), i.e., a side-by- √σ withcritical an layers? approximate This conundrum scaling is given solved by by∆ notingc ≈ thatN/Ω the· side pairAx() of=- vortex [vv() x filaments sk ]{[ of() x alternating -- sk ]2 ( signed N k )}2 vor-.16 ( ) −1 ¯yyyyy¯ 0 σonly/Ω physicalk where processkz is that the breaks vertical the invariance wavenumber of the offlow the is ticity. The radial size of this anomalous jet is the size of z While this was derived under the assumption of the Boussinesq fastestthe location growing of the mode perturbing expressed vortex. in The inverse vortex units is implicitly of H the critical layer and, furthermore, it is the anticyclonic (Umurhan et al. 2016b). As the growth rate depends partapproximation, of this jet that we undergoes demonstrated secondary its validity transition. for the anelastic assumed to be at the origin, and the cross-stream distance x approximation in MPJBHL15. For N > 0, it can be shown uponbetween the a vorticity localized perturbation of the triggering and the s filament=0 critical and layer the it Marcus et al. (2015, 2016) showed0 that the ZVI leads Brunt-Väisälä frequency, it is difficult to assess general- that the flow is neutrally stable (i.e., s is real and eigenmodes excites is to widespread turbulence in realistic shearing box mod- ized expected length scale for ∆. Results of published els inneither which grow the nor vertical decay). Critical component layers ofare gravity special locations shows in simulations, together with these analytical considera- fl fi N LyybL D the properthe ow vertical in which dependence the coef cientg of∼ the−Ω highest-derivative2z. These studies term * 0 (c) tions, suggestxm that()º∆c ∼(f)0.1H=ºor less. .18 ( ) showof that the eigenequation the ZVI appears is zero at and locations the eigenmodes away from are singular the Lesur & Latter (2016)W0 also33pm findp thatmm the saturated state midplane(Drazin where & Reid the1981 stratification). For vy ()xx=- is ( strong32 ) W0 and, Equation subse-(16) Figure 3. Late-time evolution¯ for profile Run Brunt Step. Theof left the column ZVI is (subfigures similar to a-c) that shows of the the COV vertical and componen VSI, witht quentlyvanishes spreads for twodown families toward of the critical midplane, layers; suggesting one family is anticyclonicIn plots of vortices vertical velocity,and cyclonic it is sheetsquite easy in the to pinpointlargest of the relative vorticity ωz,verticalvelocityvz ,andfractionalpotentialtemperatureanomalyθ˜/θ¯ during an active turbulent thatlocated the process at xsLm* =- can operatey ()3 p even W0 , andwhen a second the stratification family is at scales.the baroclinic These authors critical concludelayers because that the the vertical small scales velocity ap- is burstis weak, phase which (orbit 2770.6). is consistent The right with column predictions (subfigures made d-f)showthesamevariablesduringalaminarzonalflowphase(or in generally very small except in the immediate vicinity of criticalbit 2797.7). xsNLm* =-()() 3,p W () 17pear characterized by more isotropic turbulence. Bar- analytical work (Umurhan et00 al.y 2016b). Marcus et al. rancolayers et(as al. in (2018) Figure examine1). Vertical similar vorticity behavior is also an but excellent they (2016)where alsom showis a thatnon-zero the saturated integer. The statefirst has family, a turbulent barotropic alsosignature show something because baroclinic quite remarkable. critical layers They generate show vortex that spectrumcriticalE layers, ∼ k−5/3 hassuggesting already been that well-studied a downscale because direct it also long-timelayers with simulations relatively large of vorticity. the ZVI The leads vertical to component the emer- of energyoccurs cascade in shear hasfl beenows with resolved no gravity, in these and simulations/or no density gencethe curl of of large the momentum scale, axisymmetric, equation Equation radially(2b) periodicyields the (Fig. 13). We note that the starting point of the inertial zonal flow. Each successive zonal flow sheet exhibits spectrum appears to correspond to the critical layer scal- 10periodic cyclonic and anti-cyclonic behavior with a ra- 26 Lyra & Umurhan dial wavelength of about 1 scale height. Moreover, this nized activity. While the Re for disks easily satisfies this periodic flow exhibits temporal intermittency in which criterion, it still remains to be understood how the res- the anti-cyclonic portion of the zonal flow periodically onance mechanism can sustainably extract energy from breaks-down after 50-100 orbit times, but eventually the the critical layer and what might be the overall structure flow reorganizes into the zonal flow again – presum- of the disk after ZVI turbulence has acted on it. ably driven by the small scale ZVI itself (Fig. 14). Might Where might the ZVI be active? There are several this be an indication of an inverse cascade of energy and locations in the disk that, if seeded with suitable ini- its transfer onto large scale zonal flows as might be ex- tial vorticity field, may erupt into ZVI turbulence (see pected by the Rhines mechanism? This remains to be further in Sect. 5). Another idea we conjecture is that fully explored. ZVI may emerge perhaps near the boundary of magnet- Is this to be expected for protoplanetary disks? As ically active regions or near disk zones with pressure we stated in the introduction of this section, the ZVI is bumps, i.e., locations that can support radially local- not a linear instability in the usual sense, but rather a ized Rossby waves. For example, simulations of proto- finite-amplitude process. It requires a seed vortex fila- planetary disks indicate that strong zonal flow features ment (or a spatially distributed spectrum) to launch the appear at the boundary separating MHD active zones process with the vorticity of the filament being an or- from purely hydrodynamic zones (Varnière & Tagger der 1 fraction of the Keplerian flow itself. Generating 2006; Lyra & Mac Low 2012; Lyra et al. 2015). Such zonal such an initial finite amplitude state in order to spawn flows support Rossby waves. Since the ZVI mechanism the ZVI cascade/spread was raised by Umurhan et al. requires there to be a Rossby wave to resonate with a (2016b) and Lesur & Latter (2016) as a concern that lim- buoyancy oscillation somewhere in the hydrodynamic its the applicability of the ZVI in actual disks. Mar- regions of the fluid, the zonal flow characterizing this cus et al. (2016) address this matter stating an initial transition region might act like a large amplitude loud- flow state structured with an energy spectrum ∼ k−a, speaker by providing the energy – in the form of the with 1 < a < 3, will always support a small scale ex- parent Rossby wave – to set off the eruption that could hibiting anomalous vorticity satisfying the minimum lead to ZVI turbulence spreading deep into the Ohmic requirement. For example, following our discussion zone of the disk. in Sect. 3.2, a Kolmogorov spectrum (a = 5/3) has an 1/3 −1/3 anomalous velocity δv ∼ e k . Its corresponding 5. SYNTHESIS : A BUTCHER DIAGRAM FOR vorticity scale would be ω ∼ δv/` = δv · k ∼ e1/3k2/3. HYDRO INSTABILITIES Thus, even for a weak initial Kolmogorov energy spec- The three hydrodynamical instabilities operate opti- trum E0(k), – whose total integrated energy is controlled mally in different cooling times, thus existing in very 1/3 by e – there always exists a large enough value of k0 different regimes of opacity. In the adiabatic case (Ωτ  in which ω0 satisfies the minimum criterion – provided, 1) the ZVI dominates. In the isothermal case (Ωτ  1), 14 of course, k0 < kdiss = 2π/`diss. the VSI dominates. In between these extrema, the COV Whether or not this process can be self-sustaining and operates (Ωτ ∼ 1). lead to a turbulent state under a wide umbrella of con- For what material conditions do these cooling con- ditions appropriate to the Ohmic zone remains to be de- straints correspond? Reviewing the expression for the termined. It would seem that its self-sustainability de- optically thick limit for thermal relaxation, i.e., the pends centrally on what way the original disturbances `rad  ` expression of Eq. (129), it is evident that it are structured and how they arrive. Although Marcus depends upon the effective material opacities, density, et al. (2016) find that the critical finite amplitude in the temperature and length scale of perturbation, `. Nelson velocity field to be very small (Ma ≈ 10−6), sustaining a et al. (2013b) examined cooling times to assess these con- ` ∼ coherent perturbation for long enough times may still be ditions assuming values of 0.1H (being typical of the of concern. However keeping in mind the foregoing dis- fastest growing mode of the VSI) together with values of cussion, the minimum velocity scale quoted in Marcus the expected midplane densities and radial temperature et al. (2016) is a value inferred based on the limited res- profiles based on the canonical flared minimum mass olution of their simulations and is probably indicative solar nebula disk models of Chiang & Goldreich (1997). of a maximum lower limit because higher resolved sim- Such disk models predict midplane densities and radial ulations would permit the introduction of radially nar- temperature profiles that are rower filaments that both satisfy the requisite anoma- −9 −39/14 3 lous vorticity criterion while exhibiting even smaller ρmid ≈ 2.7 × 10 F(R/AU) g/cm , (152) values of Ma. Also, Lesur & Latter (2016) and Marcus T = 120(R/AU)−3/7 K, (153) and co-workers find the instability numerically when hyperdiffusion is used - hyperdiffusion is an instance of respectively, where F is the relative mass of the disk in Large Eddy Simulation (LES, see further below). When units of minimum solar mass models. The correspond- Laplacian viscosity is used, Lesur & Latter (2016) find ing cooling times are that the Reynolds numbers must significantly exceed 7 10 in order for the process to lead to sustained disorga-  κ   ` 2  R −53/14 Ωτ/2π = 168F2 R . cm2/g R 20AU 14 We note that a spectrum of the sort envisioned contains distur- bances of every shape for any given value of k. It is that part of the (154) flow field that gives rise to filaments with little azimuthal structure The onset of the VSI for regions at around 10 AU that would trigger the instability. around a 1 M star with F = 1 requires opacities of HydrodynamicalZombie Vortex instabilities Instability in protoplanetary III disks 27 13

Figure 8.TurbulentburstcycleinRunBrunt Step. Shown are horizontal slices at z =2H0 of the vertical component of relative vorticity,FIG with. 14.— same ZVI burst ccolor cycle scale shown as as Fig. radial-azimuthal3.Firstframeisatorbit2750.2andlastframeisatorbit2787 cuts at 2 scale heights of the anomalous vertical vorticity. Each successive.5 with panel corresponds other frames at to 3.5 orbit times. Banded structure undergoes a breakdown, triggering critical layer excitations along the way. Banded structure restored after 50 equal intervalsorbit times. of Complete 3.4 orbits. cycle is approximately 100 orbit times. Reproduced from Barranco et al. (2018).

TABLE 2 HYDRODYNAMICALINSTABILITIESSUMMARYCHARACTERISTICS.

Instability Violation of Mechanism Type Linear growth Length scale Opacity Thermal α 2 cm g Ω Rayleigh criterion√ rate of linear growth κ ( ⁄ ) time ( τ) Vertical Shear dΩ/dz 6= 0 Angular momentum exchange m|q|hΩ/4 π|q|hH < 1  1 10−4 − 10−3 between adjacent elements. 2 2 p −4 −3 Convective NR < 0 Buoyant amplification |N |/4Ω χ/Ω 1 − 50 ∼ 1 10 − 10 of epicyclic oscillations. 2 −4 −3 Zombie Vortex Nz > 0 Resonance between Rossby – – > 50  1 10 − 10 and buoyancy frequency.

κ < 1cm2/g.15 Since the ZVI also has growth on simi- also used in Dzyurkevich et al. (2013), lar scales the corresponding requirements suggest κ > " # 50cm2/g for the ZVI to be minimally operative. Even  r −qΣ  r 2−qΣ though the COV operates at larger scales, a reasonable Σ = Σc exp − (155) rc rc estimate of the opacity is 1 < κ < 50cm2/g. The cooling time requirement means that the ZVI will with be more important in the very inner disk (inwards of 1 AU) , where the gas is more adiabatic; the COV in the Mdisk Σc = (2 − q ) (156) region 1-10 AU; and the VSI in the outer disk beyond Σ 2πr2 10 AU which tends to be more isothermal. As we go fur- c ther outwards, the very outer disk beyond 100 AU starts The stellar mass is M? = 1M , the reference column −2 to become too optically thin and the cooling time goes density is Σc = 1700 g cm and the power laws is taken up again, being unstable to COV. Eventually, if the disk as the median of Andrews et al. (2009), qΣ = 0.9. The cut- is too optically thin and the cooling time long enough, off radius is rc = 40 AU. This profile yields a good match the disk will be unstable to ZVI. That invites a mapping to both CO and continuum observations (Hughes et al. of instabilities that we call a “butcher diagram” 16 , in 2008), and matches the minimum mass solar nebula up Fig. 15. to 100 AU. The temperature chosen is that of a passive The actual mapping of the instabilities depends on disk (Chiang & Goldreich 1997) translating gas densities, dust densities, and tempera- tures, to opacities and thence to optical thicknesses and  1AU 1/2 T = T (157) cooling times. As such, it will depend heavily on the un- 0 R derlying disk model assumed. The specific map shown in Fig. 15 is taken from Malygin et al. (2017), and uses with T0 = 280 K. Vertical temperature gradients are ig- the axisymmetric disk model of Hartmann et al. (1998), nored, and a lower limit of 10 K set beyond 250 AU. Us- ing the dust opacities from Semenov et al. (2003) and −2 15 Note that the opacity is the effective opacity of a fluid mixture of a well-mixed dust at the dust-to-gas ratio of 10 , one gas and dust. Typically minimum mass solar nebula models assume a obtains the mapping in Fig. 15 for the appropriate cool- 1 percent content of dust grains. ing times described in the text. Ignoring vertical tem- 16 Inspired by diagrams delineating various cuts of beef. perature gradients is valid for passive-disks (Chiang & 28 Lyra & Umurhan

FIG. 15.— A “butcher diagram” for hydrodynamical instabilities. Disk opacities define cooling times that, in turn, determine which hydrody- namical instability process is dominant. The ZVI dominates for long cooling times, happening at the optically thick inner disk. The COV requires moderate cooling times, happening in the 10 AU region. The VSI happens where the disk is more isothermal. As one moves outwards, the disk becomes more optically thin and thus the cooling time increases again, leading to a new zone of COV and eventually another adiabatic region for the ZVI. The regions where ZVI dominate may be too ionized and thus prone to MRI (in the inner disk) or magnetocentrifugal winds. Adapted from Malygin et al. (2017). The different instabilities have different saturation mechanisms: the VSI saturates via RWI, the COV via SBI, and the ZVI via itself and RWI.

FIG. 16.— Thermal properties of global alpha-disk models. Left: Rosseland mean opacities after 2 × 105 yr. Opacities quoted in units of cm2/g. 5 Middle: Derived thermal relaxation times after 2 × 10 yr. Cooling time τ in unit of local orbit times around a 1 M star. Right: A butcher diagram indicating locations where various instabilities are operating based on output of Estrada et al. (2016): 2 × 105 years, α = 4 × 10−4 and dust porosities of zero. Adapted from Umurhan et al. (2017).

Goldreich 1997). However, disks in which there is some Rosseland mean and Planck mean opacities, the temper- amount of viscous heating will show potentially strong ature profiles are largely locally isothermal (with mid- deviations from nearly uniform and this is expected to plane to lid-temperature variations of less than 1 K). The be the case if the turbulence is relatively strong α > 10−2 radial temperature dependence is given by T ∼ R−3/7 like, for example, the global disk evolution models con- for disk radii > 3 AU, typical of flared disks and con- sidered in Bitsch et al. (2015). For the kinds of hydro- sistent with theoretical predictions (Chiang & Goldreich dynamic instabilities discussed here, wherein α < 10−3, 1997). Umurhan et al. (2017) displays the effective dust viscous heating is expected to be a secondary effect. opacities in a disk experiencing some amount of alpha- Using the coagulation/fragmentation and dust evolu- disk turbulence according to the disk evolution model tion model of Estrada et al. (2016), and assuming equal of Estrada et al. (2016). The model compiled distributes Hydrodynamical instabilities in protoplanetary disks 29 dust particles in vertical height according to the turbu- tial range is indeed self-similar, then upon its resolution lent intensity (α ∼ 4 × 10−4) and the dust size which to- the inertial spectrum could be extrapolated down to the gether yield a turbulent Stokes number. Estrada et al. smaller scales where dust collection is thought to be ac- (2016) determine the size of particles as a function of tively taking place. disk radius and epoch. The left panel of Fig. 16 shows A case that illustrates the path forward is the state of a profile of the internal opacities at a model after t = our understanding of the VSI. Simulations conducted 2 × 105 years of dust evolution. The thermal relaxation thus far have been only moderately resolved. For ex- times can be very roughly represented as either dom- ample, in the original simulations conducted by Nelson inated by radiative diffusion (optically thick) or radia- et al. (2013b) where the radial domain was a radial an- tive cooling (optically thin) regimes and are given by nulus from R = R0 to R = 2R0, the conditions of the disk Eq. (129). For optically thick conditions, the middle were such that the fastest growing mode had a wave- ` panel of Fig. 16 shows τ for fluid structures with scales length of 0 = 0.008R0. Note that on these scales, the ∼ ` = 0.05H, typical of the VSI. The right panel of Fig. 16 pressure scale height is 0.05R0. These simulations displays a butcher diagram of the instabilities consider- resolved this fastest growing mode with about 15 grid ` ing these thermal relaxation times. The VSI dominates points. With 0 being nominally the injection scale, in or- the turbulence in the range 3-90 AU while the COV oc- der to begin seeing the emergence of an inertial regime curs generally in the inner disk (<3 AU). The ZVI is requires at least 10-100 times the resolution achieved in prominent in outer solar nebula (>100 AU). This partic- those simulations. For the kinds of global scale simula- ular combination of gas and dust model never becomes tions currently feasible, such high resolution studies is optically thick enough for the ZVI in the inner disk. still quite prohibitive. We stress that at the time of this review no model One way around this is to consider very narrow global combining all the three (or even two) of the instabili- simulations or appeal to the often maligned shearing ties has been attempted. Therefore, the boundaries be- box setting (Goldreich & Lynden-Bell 1965; Umurhan tween the instabilities in the butcher diagram are ill- & Regev 2004; Latter & Papaloizou 2017), which repre- defined. As long as the entropy gradient is negative, sents a model facsimile of a box section of a disk orbit- the COV has the least stringent thermal constrains, ef- ing the central star at the reference radius R = R0. The fectively existing for all finite cooling times. The VSI, on spatial scales of the box are usually in units of the local the other hand, seems to be the most robust in respect scale height so, in principle, higher resolution simula- to the violation of the Solberg-Høiland criterion, as stel- tions are possible. For example, simulations of the ZVI lar irradiation should always enforce some amount of are performed in this model setup and, as the results of vertical shear. The ZVI seems to be less robust, as it Marcus et al. (2016) shows, they were able to resolve a requires a “prime mover” in the form of a sufficiently Kolmogorov like spectrum. In the context of the VSI, for large finite-amplitude disturbances to be emplaced in- example, adjustments to the formulation of the shearing side the dead zone. However, we conjecture that this box equations that are suitable for the mechanism may initial finite-amplitude perturbation should be present be needed, like the composite local-global model sug- in the RWI-unstable transition between the MRI zone gested by McNally & Pessah (2015). If a shearing box and the Ohmic zone in the inner disk, where the ther- model is used then the global scale variations that give mal conditions are also adiabatic. We note, although, rise to each of the three instabilities must be carefully that the ZVI, being present in more adiabatic conditions, represented within the confines of these models. For ex- will probably be self-regulated at small α values. At the ample, the global scale variation of the temperature field high densities of the optically thick depths of the disk, must be faithfully represented on the small scales in just modest amounts of accretional stresses, perhaps as lit- the right way so that the instability is triggered. Careful thought must also be given to the choice of boundary ∼ −5 tle as α 10 are sufficient to generate enough heat to conditions adopted, making sure that they do not con- bring the stratification to a profile that shuts down the taminate the results obtained. ZVI. Table 2 summarizes the parameters of the instabil- Because the molecular viscosity of H2 is so small, it is ities. practically impossible at this time to conduct direct nu- merical simulations (DNS) that resolve the flow down 6. NEAR FUTURE GOALS FOR NUMERICAL to the dissipation scales. Since energy cascades toward SIMULATIONS OF DISK TURBULENCE small scales in fully turbulent supercritical 3D flows 17, The near and long term goals of numerical model- unless the natural dissipation scales are resolved in flow ing of turbulence in the protoplanetary disk is to es- modeling, then power can build up at the grid scales of tablish the conditions on which dust particles grow into a simulation leading to unrealistic and, often times, nu- planetesimals on a 0.5Ma to 2.3Ma time frame after the merically unstable situations. Large Eddy Simulation appearance of Calcium Aluminium Inclusions (CAI’s). (LES, Davidson (2004)) is one way to confront this prob- This means properly characterizing the turbulent ve- lem. An LES simulation will introduce a high order dif- locity field on scales where the dust particles accumu- 4 8 16 fusion operator, like ν4∇ or ν8∇ or even ν16∇ , into late. From our standpoint it is incumbent on the fluid the equations of momentum conservation. The over- dynamicists and theorists to establish, with some con- all strength of the dissipation is controlled by the "hy- fidence, the quality of the energy spectrum at the rel- perdiffusion" coefficient νn, where n is the order of the evant scales and the first step toward this is identify- Laplacian operator. With a suitably chosen value of the ing and quantifying the shape of the turbulent energy spectrum in the inertial regime. If the resulting turbu- lent state is both steady and the turbulence in the iner- 17 ...and if 2D flow, then enstrophy cascades toward small scales. 30 Lyra & Umurhan coefficient νn a hyperdiffusion operator permits the nor- recently discovered (Simon et al. 2016). The very high mal cascade of energy through the inertial regime of a up region of the butcher diagram (Fig. 15) are probably turbulent flow but will increasingly dissipate power as prone to ambipolar diffusion, providing a “lid” for all the grid scales are approached in a simulation. This is instabilities; if the ionization fraction is high enough the intended to mimic the continued transference of power MRI can be reactivated. The outer ZVI region may thus to the unresolved dissipation scales by artificially dissi- be completely quenched, by ambipolar diffusion or the pating the energy flux at the resolved scale of the sim- MRI itself. The outer COV region is low density enough ulation. This is usually a delicate procedure requiring for ambipolar diffusion to operate. The VSI region may careful tuning of νn. LES, or any other equivalent small- also operate optimally within the region where the Hall scale energy dissipation procedure, is an essential tool effect is the dominant source of resistivity. It remains to for numerical modeling of protoplanetary disk turbu- be shown how the instabilities here summarized behave lence that will be necessary for the foreseeable future. in connection with these non-ideal terms. The importance of characterizing the shape and qual- 7. CONCLUSION ity of turbulence and its influence upon the growth of The past decade of disk research have seen the emer- planetesimals, especially at the smallest scales, is one of gency of disk instability mechanisms that provide hy- the main messages of this review. While we have not drodynamical turbulence in the Ohmic dead zone. They delved into how the turbulence mitigates the growth of violate the Rayleigh criterion in different ways. The ver- planetesimals, we have emphasized that the dynami- tical shear instability (VSI) by having dΩ/dz 6= 0, the cal state of the protoplanetary disk is the fundamental convective overstability (COV) by the presence of en- canvas on which the particle growth narrative is etched. 2 < With this in mind, we conclude with another quote from tropy gradients with NR 0, and the zombie vortex in- the ancients, this time of Hesiod, who writes of the birth 2 stability (ZVI) with Nz > 0. Because of these require- of the Earth (“Gaia"), ments, they exist in very different regimes of opacity: the ZVI in adiabatic regions, the VSI in isothermal ones, In the beginning there was Chaos; but then there and the COV in between. These can be mapped into dif- came to be Gaia, the broad-breasted, the ever- ferent locations, depending on the particular disk model secure seat of all immortals who dwell in the (Fig. 15 and Fig. 16). The instabilities have different peaks of snowy Olympos. −3 turbulent responses, with α values around 10 . Their 18 properties are summarized in Table 2. Hesiod Theogony 116-118 Of the three processes, the most robust seems to be This vision strikes as modern as well: if one substitutes the VSI, as the unstable entropy gradients necessary for “chaos" with turbulence and “Gaia" with planetesimals, the COV are not necessarily realized in all disks, and then we have a picture describing elements of today’s the first perturbation necessary for the ZVI is difficult scientific paradigm of planet formation. to maintain over long times. The three instabilities sat- urate into three-dimensional vortices. The VSI because of RWI and Kelvin-Helmholtz instability of the thin tall sheets. The COV because it builds up the finite differ- Acknowledgements. We thank Joe Barranco, Kon- ence perturbations that lead to subcritical baroclinic in- stantin Batygin, Luca Biancofiore, Til Birnstiel, Omer stability. The ZVI also via RWI and Kelvin-Helmholtz Blaes, Axel Brandenburg, Simon Casassus, Thayne Cur- instability. All vortices decay due to elliptical instability rie, Jeff Cuzzi, Kees Dullemond, Natalia Dzyurke- in the cores (Pierrehumbert 1986; Bayly 1986; Lesur & vich, Paul Estrada, Mario Flock, Sebastien Fromang, Papaloizou 2009), that leads to a direct enstrophy cas- Oliver Gressel, Nili Harnik, Eyal Heifetz, Tobias cade. In saturation, vortex strength should be de- Heinemann, Thomas Henning, Anders Johansen, Hu- fined by a balance between the saturation mechanism, bert Klahr, Willy Kley, Min-Kai Lin, Mordecai Mac that feeds the vortices (RWI for VSI/ZVI, SBI for COV), Low, Philip Marcus, Frédéric Masset, Colin McNally, and the rate of decay via elliptic instability (Lyra 2013). Heloïse Méhéut, Farzana Meru, Alessandro Morbidelli, Weaker vortices (of larger aspect ratio) linger for longer Krzyzstof Mizerski, Richard Nelson, Chris Ormel, Se- because the growth rates of elliptic instability decrease bastián Pérez, Oded Regev, Luca Ricci, Giora Shaviv, with vortex aspect ratio. Jake Simon, Neal Turner, Nienke van der Marel, Peggy In this review, we have specifically used the term Varnière, and Andrew Youdin, for their support and Ohmic zone instead of the traditional “dead” zone. We valuable scientific discussions over the years. We thank could have used “hydrodynamical” zone since Ohmic Willy Kley and Henrik Latter for permission to use their effects, in the sense of magnetic Reynolds numbers close figures, and for comments on the draft. We thank the to 1, are not at play. The Ohmic zone is deep into the referee, Hubert Klahr, for substantial suggestions that resistive regime and any magnetic effect is irrelevant. vastly improved the quality of this review. We also ap- Yet, we keep the name Ohmic for juxtaposition to the preciate comments from Hantao Ji and Andrew Youdin other non-ideal MHD effects: the Hall effect and am- correcting several omissions from an earlier version of bipolar diffusion, that may be at play in very low den- the work. We express gratitude to Professor Osman sity regions of the disk (Wardle 1999, 2007; Lesur et al. 18 Translation by Professor Osman S. Umurhan who notes that 2014). The Hall effect leads to the Hall shear instability “χαoσ" or “Chaos" represents not disorder in the modern sense, but if the magnetic field and the angular momentum vec- a chasm or abyss, whose properties appear to be a dark, gaping space. tors are aligned. Ambipolar diffusion leads to magne- Later in Hesiod’s account this very chasm appears to have the appro- tocentrifugal winds, evidence for which may have been priate material “to catch fire" from Zeus’ thunderbolt. Hydrodynamical instabilities in protoplanetary disks 31

Umurhan for help with classical period translations. We ence Institute through grant HST-AR-14572 and the are especially indebted to James Y.-K. Cho for illuminat- NASA Exoplanet Research Program through grant 16- ing subtleties characterizing the nature of atmospheric XRP16_2-0065. O.M.U. acknowledges support from turbulence. the NASA Astrophysics Theory Program through grant W.L. acknowledges support of Space Telescope Sci- NNX17AK59G.

REFERENCES Abramowicz, M. A., Livio, M., Piran, T., & Wiita, P. J. 1984, ApJ, 279, Gressel, O., Turner, N. J., Nelson, R. P., & McNally, C. P. 2015, ApJ, 367 801, 84 Andrews, S. M., & Williams, J. P. 2007, ApJ, 659, 705 Haghighipour, N., & Boss, A. P. 2003, ApJ, 583, 996 Andrews, S. M., Wilner, D. J., Hughes, A. M., Qi, C., & Dullemond, Hartmann, L., Calvet, N., Gullbring, E., & D’Alessio, P. 1998, ApJ, C. P. 2009, ApJ, 700, 1502 495, 385 Ansdell, M. et al. 2018, ApJ, 859, 21 Hayashi, C. 1981, Progress of Theoretical Physics Supplement, 70, 35 Arlt, R., & Urpin, V. 2004a, A&A, 426, 755 Heinemann, T., & Papaloizou, J. C. B. 2009, MNRAS, 397, 52 —. 2004b, A&A, 426, 755 —. 2012, MNRAS, 419, 1085 Aumann, H. H. 1985, PASP, 97, 885 Hughes, A. M., Wilner, D. J., Qi, C., & Hogerheijde, M. R. 2008, ApJ, Avila, M. 2012, Physical Review Letters, 108, 124501 678, 1119 Bai, X.-N. 2015, ApJ, 798, 84 Ji, H., Burin, M., Schartman, E., & Goodman, J. 2006, Nature, 444, 343 Bai, X.-N., & Stone, J. M. 2011, ApJ, 736, 144 Johansen, A., Henning, T., & Klahr, H. 2006, ApJ, 643, 1219 —. 2017, ApJ, 836, 46 Johansen, A., Oishi, J. S., Mac Low, M.-M., Klahr, H., Henning, T., & Balbus, S. A. 2003, ARA&A, 41, 555 Youdin, A. 2007, Nature, 448, 1022 —. 2017, Journal of Fluid Mechanics, 824, 1 Johansen, A., & Youdin, A. 2007, ApJ, 662, 627 Balbus, S. A., & Hawley, J. F. 1991, ApJ, 376, 214 Kippenhahn, R., Weigert, A., & Weiss, A. 2012, Stellar Structure and Balbus, S. A., & Terquem, C. 2001, ApJ, 552, 235 Evolution (Springer-Verlag) Barker, A. J., & Latter, H. N. 2015a, MNRAS, 450, 21 Klahr, H. 2004, ApJ, 606, 1070 —. 2015b, MNRAS, 450, 21 Klahr, H., & Hubbard, A. 2014, ApJ, 788, 21 Barranco, J., Pei, S., & Marcus, P. 2018, ArXiv e-prints Klahr, H. H., & Bodenheimer, P. 2003, ApJ, 582, 869 Batygin, K. 2018, AJ, 155, 178 Koller, J., Li, H., & Lin, D. 2003, ApJ, 596, L91 Bayly, B. J. 1986, Physical Review Letters, 57, 2160 Krall, N. A., & Trivelpiece, A. W. 1973, Principles of plasma physics Beckwith, S. V. W., Sargent, A. I., Chini, R. S., & Guesten, R. 1990, AJ, Kunz, M. W. 2008, MNRAS, 385, 1494 99, 924 Kunz, M. W., & Lesur, G. 2013, MNRAS, 434, 2295 Benz, W. 2000, Space Sci. Rev., 92, 279 Latter, H. N. 2016, MNRAS, 455, 2608 Biancofiore, L., & Umurhan, O. M. 2018, ArXiv e-prints Latter, H. N., & Papaloizou, J. 2017, MNRAS, 472, 1432 Bitsch, B., Johansen, A., Lambrechts, M., & Morbidelli, A. 2015, A&A, —. 2018, MNRAS, 474, 3110 575, A28 Lesur, G., Kunz, M. W., & Fromang, S. 2014, A&A, 566, A56 Blaes, O. M., & Balbus, S. A. 1994, ApJ, 421, 163 Lesur, G., & Papaloizou, J. C. B. 2009, A&A, 498, 1 Boffetta, G., & Ecke, R. E. 2012, Annual Review of Fluid Mechanics, —. 2010, A&A, 513, A60 44, 427 Lesur, G. R. J., & Latter, H. 2016, MNRAS, 462, 4549 Boffetta, G., & Musacchio, S. 2010, Phys. Rev. E, 82, 016307 Li, H., Colgate, S. A., Wendroff, B., & Liska, R. 2001, ApJ, 551, 874 Brauer, F., Dullemond, C. P., Johansen, A., Henning, T., Klahr, H., & Li, H., Finn, J. M., Lovelace, R. V. E., & Colgate, S. A. 2000, ApJ, 533, Natta, A. 2007, A&A, 469, 1169 1023 Bryan, M. L., Benneke, B., Knutson, H. A., Batygin, K., & Bowler, B. P. Lin, D. N. C., & Papaloizou, J. 1980, MNRAS, 191, 37 2018, Nature Astronomy, 2, 138 Lin, M.-K. 2014, MNRAS, 437, 575 Butler, K. M., & Farrell, B. F. 1992, Physics of Fluids A, 4, 1637 Lin, M.-K., & Youdin, A. N. 2015, ApJ, 811, 17 Cabot, W. 1996, ApJ, 465, 874 Lopez, J. M., & Avila, M. 2017, Journal of Fluid Mechanics, 817, 21 Cameron, A. G. W. 1978, Moon and Planets, 18, 5 Lovelace, R., Li, H., Colgate, S., & Nelson, A. 1999, ApJ, 513, 805 Chagelishvili, G. D., Zahn, J.-P., Tevzadze, A. G., & Lominadze, J. G. Lynden-Bell, D., & Pringle, J. E. 1974, MNRAS, 168, 603 2003, A&A, 402, 401 Lyra, W. 2013, in European Physical Journal Web of Conferences, Chiang, E. I., & Goldreich, P. 1997, ApJ, 490, 368 Vol. 46, European Physical Journal Web of Conferences, 04003 Davidson, P. A. 2004, Turbulence : an introduction for scientists and Lyra, W. 2014, ApJ, 789, 77 engineers (Oxford) Lyra, W., Johansen, A., Klahr, H., & Piskunov, N. 2009, A&A, 493, de Val-Borro, M., Artymowicz, P., D’Angelo, G., & Peplinski, A. 2007, 1125 A&A, 471, 1043 Lyra, W., & Klahr, H. 2011, A&A, 527, A138 Diels, H., & Kranz, W. 1961, Die Fragmente der Vorsokratiker, vols. I Lyra, W., & Mac Low, M.-M. 2012, ApJ, 756, 62 and II. (Berlin Weidmann) Lyra, W., Turner, N. J., & McNally, C. P. 2015, A&A, 574, A10 Dubrulle, B., Morfill, G., & Sterzik, M. 1995, Icarus, 114, 237 Dzyurkevich, N., Turner, N. J., Henning, T., & Kley, W. 2013, ApJ, Lyttleton, R. A. 1972, MNRAS, 158, 463 765, 114 Mac Low, M.-M., Norman, M. L., Konigl, A., & Wardle, M. 1995, ApJ, Edlund, E. M., & Ji, H. 2014, Phys. Rev. E, 89, 021004 442, 726 —. 2015, Phys. Rev. E, 92, 043005 Malygin, M. G., Klahr, H., Semenov, D., Henning, T., & Dullemond, C. P. 2017, A&A, 605, A30 Elsasser, H., & Staude, H. J. 1978, A&A, 70, L3 Manger, N., & Klahr, H. 2018, MNRAS Estrada, P. R., Cuzzi, J. N., & Morgan, D. A. 2016, ApJ, 818, 200 Marcus, P. S., Pei, S., Jiang, C.-H., & Barranco, J. A. 2016, ApJ, 833, 148 Flock, M., Nelson, R. P., Turner, N. J., Bertrang, G. H.-M., Marcus, P. S., Pei, S., Jiang, C.-H., Barranco, J. A., Hassanzadeh, P., & Carrasco-González, C., Henning, T., Lyra, W., & Teague, R. 2017, Lecoanet, D. 2015, ApJ, 808, 87 ApJ, 850, 131 Marcus, P. S., Pei, S., Jiang, C.-H., & Hassanzadeh, P. 2013, Physical Fricke, K. 1968, ZAp, 68, 317 Review Letters, 111, 084501 Frisch, U. 1995, Turbulence (Cambridge University Press) Maretzke, S., Hof, B., & Avila, M. 2014, Journal of Fluid Mechanics, Gammie, C. F. 1996, ApJ, 457, 355 742, 254 Gamow, G., & Hynek, J. A. 1945, ApJ, 101, 249 Marino, R., Mininni, P. D., Rosenberg, D. L., & Pouquet, A. 2014, Garaud, P., & Lin, D. N. C. 2004, ApJ, 608, 1050 Phys. Rev. E, 90, 023018 Goldreich, P., Goodman, J., & Narayan, R. 1986, MNRAS, 221, 339 McCaughrean, M. J., & O’dell, C. R. 1996, AJ, 111, 1977 Goldreich, P., & Lynden-Bell, D. 1965, MNRAS, 130, 125 McNally, C. P., & Pessah, M. E. 2015, ApJ, 811, 121 Goldreich, P., & Schubert, G. 1967, ApJ, 150, 571 Meheut, H., Casse, F., Varniere, P., & Tagger, M. 2010, A&A, 516, A31 Goldreich, P., & Ward, W. R. 1973, ApJ, 183, 1051 Mestel, L. 1965a, QJRAS, 6, 161 Goodman, J., & Xu, G. 1994, ApJ, 432, 213 —. 1965b, QJRAS, 6, 265 32 Lyra & Umurhan

Nastrom, G. D., & Gage, K. S. 1985, Journal of Atmospheric Sciences, Stoll, M. H. R., Kley, W., & Picogna, G. 2017a, A&A, 599, L6 42, 950 Stoll, M. H. R., Picogna, G., & Kley, W. 2017b, A&A, 604, A28 Nelson, R. P., Gressel, O., & Umurhan, O. M. 2013a, MNRAS, 435, Strom, K. M., Strom, S. E., Edwards, S., Cabrit, S., & Skrutskie, M. F. 2610 1989, AJ, 97, 1451 —. 2013b, MNRAS, 435, 2610 Sun, Y. Q., Rotunno, R., & Zhang, F. 2017, Journal of Atmospheric O’dell, C. R., & Wen, Z. 1994, ApJ, 436, 194 Sciences, 74, 185 Ogilvie, G. I. 2016, Journal of Plasma Physics, 82, 205820301 Takata, T., & Stevenson, D. J. 1996, Icarus, 123, 404 Paardekooper, S.-J. 2006, PhD thesis, Leiden Observatory, Leiden Tassoul, J.-L. 1978, Theory of rotating stars (Princeton) University, P.O. Box 9513, 2300 RA Leiden, The Netherlands Taylor, G. I. 1936, Proceedings of the Royal Society of London Series Pandey, B. P., & Wardle, M. 2006, MNRAS, 371, 1014 A, 157, 546 —. 2008, MNRAS, 385, 2269 Tulloch, R., & Smith, K. S. 2006, Proceedings of the National Paoletti, M. S., van Gils, D. P. M., Dubrulle, B., Sun, C., Lohse, D., & Academy of Science, 103, 14690 Lathrop, D. P. 2012, A&A, 547, A64 Turner, N. J., & Drake, J. F. 2009, ApJ, 703, 2152 Pedlosky, J. 1982, Geophysical fluid dynamics Umurhan, O. M. 2010, A&A, 521, A25 Petersen, M. R., Julien, K., & Stewart, G. R. 2007a, ApJ, 658, 1236 Umurhan, O. M., Estrada, P. R., & Cuzzi, J. N. 2017, in Lunar and Petersen, M. R., Stewart, G. R., & Julien, K. 2007b, ApJ, 658, 1252 Planetary Inst. Technical Report, Vol. 48, Lunar and Planetary Pierrehumbert, R. T. 1986, Physical Review Letters, 57, 2157 Science Conference, 2616 Regev, O., Umurhan, O. M., & Yecko, P. A. 2016, Modern Fluid Umurhan, O. M., Nelson, R. P., & Gressel, O. 2013, in European Dynamics for Physics and Astrophysics (Springer) Physical Journal Web of Conferences, Vol. 46, European Physical Ricci, L., Robberto, M., & Soderblom, D. R. 2008, AJ, 136, 2136 Journal Web of Conferences, 03003 Richard, D., & Zahn, J.-P. 1999, A&A, 347, 734 Umurhan, O. M., Nelson, R. P., & Gressel, O. 2016a, A&A, 586, A33 Richard, S., Nelson, R. P., & Umurhan, O. M. 2016, MNRAS, 456, 3571 Umurhan, O. M., & Regev, O. 2004, A&A, 427, 855 Rucinski, S. M. 1985, AJ, 90, 2321 Umurhan, O. M., Shariff, K., & Cuzzi, J. N. 2016b, ApJ, 830, 95 Safronov, V. S. 1972, Evolution of the protoplanetary cloud and Urpin, V. 2003, A&A, 404, 397 formation of the earth and planets. Urpin, V., & Brandenburg, A. 1998, MNRAS, 294, 399 Salmeron, R., & Wardle, M. 2005, MNRAS, 361, 45 Vallis, G. K. 2006, Atmospheric and Oceanic Fluid Dynamics Sargent, A. I., & Beckwith, S. 1987, ApJ, 323, 294 (Cambridge University Press), 770 Schmid, P. J., & Henningson, D. S. 2001, Stability and Transition in Varnière, P., & Tagger, M. 2006, A&A, 446, L13 Shear Flows (Springer), 558 Wardle, M. 1999, MNRAS, 307, 849 Sekiya, M. 1998, Icarus, 133, 298 —. 2007, Ap&SS, 311, 35 Semenov, D., Henning, T., Helling, C., Ilgner, M., & Sedlmayr, E. Wardle, M., & Königl, A. 1993, ApJ, 410, 218 2003, A&A, 410, 611 Wardle, M., & Salmeron, R. 2012, MNRAS, 422, 2737 Shakura, N. I., & Sunyaev, R. A. 1973, A&A, 24, 337 Weidenschilling, S. J. 1977, MNRAS, 180, 57 Shakura, N. I., Sunyaev, R. A., & Zilitinkevich, S. S. 1978, A&A, 62, —. 1980, Icarus, 44, 172 179 Weidenschilling, S. J., & Cuzzi, J. N. 1993, in Protostars and Planets Sheehan, D. P., Davis, S. S., Cuzzi, J. N., & Estberg, G. N. 1999, Icarus, III, ed. E. H. Levy & J. I. Lunine, 1031–1060 142, 238 Yecko, P. A. 2004, A&A, 425, 385 Simon, J. B., Bai, X.-N., Flaherty, K. M., & Hughes, A. M. 2018, ApJ, Yellin-Bergovoy, R., Heifetz, E., & Umurhan, O. M. 2017, ArXiv 865, 10 e-prints Simon, M. N., Pascucci, I., Edwards, S., Feng, W., Gorti, U., Youdin, A., & Johansen, A. 2007, ApJ, 662, 613 Hollenbach, D., Rigliaco, E., & Keane, J. T. 2016, ApJ, 831, 169 Youdin, A. N., & Goodman, J. 2005, ApJ, 620, 459 Spiegel, E. A. 1957, ApJ, 126, 202 Youdin, A. N., & Shu, F. H. 2002, ApJ, 580, 494 Stoll, M. H. R., & Kley, W. 2014, A&A, 572, A77 Young, R. M. B., & Read, P. L. 2017, Nature Physics, 13, 1135 —. 2016, A&A, 594, A57