betaFeedback

WorksPhysics and Astronomy

Angular momentum transport in a newborn protostar–disc system: magnetic braking, spiral arms and the magneto-rotational instability

Abstract

Angular momentum transport in young protostar–disc systems involves magnetic braking, magneto-rotational turbulence and gravitational instability, mechanisms that are usually studied separately and on different scales. We present a three-dimensional non-ideal magnetohydrodynamic simulation, including Ohmic and ambipolar diffusion, that follows the collapse of a 1 M⊙1~M_\odot dense core continuously from the cloud scale to the formation of the protostar, and then the resulting protostar–disc system for ∼64 yr\sim 64~\mathrm{yr} at a resolution of ∼10−4 AU\sim 10^{-4}~\mathrm{AU} at the protostellar surface. We decompose the azimuthally averaged angular momentum budget into Reynolds, Maxwell and gravitational torques, each split into a radial and a meridional component, over a quasi-steady phase of the evolution. The collapse produces a thick disc an order of magnitude more massive than the protostar, whose specific angular momentum decreases as it accretes. Accretion through the disc is vertically layered, with accreting and decreting layers at the same radius, a pattern set layer by layer by the torques. The transport separates into three regimes delimited by the coupling of the magnetic field to the gas. Magnetic braking operates in the envelope. Maxwell stresses dominate inside ∼0.1 AU\sim 0.1~\mathrm{AU}, where the field recouples and the magneto-rotational instability develops. In the ambipolar dead zone that makes up the bulk of the disc, the Reynolds stress dominates, carried by m=2m=2 spiral arms in a marginally unstable disc, with the gravitational stress as a secondary contributor. Meridional transport is as important as radial transport, so that a one-dimensional viscous model fitted to the net mass flux would require a negative αα over part of the disc.

Introduction

Star formation is in large part a problem of angular momentum disposal. A dense core inherits rotation from its parent cloud, and as it contracts it spins up, so that by the time material reaches the scale of the forming star it carries far more specific angular momentum than the star itself can hold. The gas therefore cannot fall directly to the centre: deflected by rotation, it settles into a rotationally supported circumstellar disc. The disc, however, only defers the difficulty. For the protostar to grow, the material orbiting it must give its angular momentum to something else — accretion is less a matter of drawing gas inwards than of removing the quantity that keeps it out. Whatever performs that removal sets the rate at which the star is assembled, and the mass, size and structure of the disc through which it is assembled. Such discs are now observed around Class 0 protostars (Tobin et al. 2012; Ohashi et al. 2014), and surveys find most of them to be compact, with radii of a few tens of astronomical units [29] (Tobin et al. 2020). The protostar itself, however, is only a few solar radii across, so the angular momentum must ultimately be removed down to scales of ∼0.01\sim0.01 AU, well below what these observations resolve.

Three mechanisms are generally invoked to remove it. In the magnetised envelope, the field couples the rapidly rotating inner regions to the slowly rotating outer ones and brakes them [35, 21]. Under ideal magnetohydrodynamics this braking is so efficient that no rotationally supported disc survives the collapse [32]; discs form because Ohmic and ambipolar diffusion decouple the field from the weakly ionised gas [28] (Zhao et al. 2016; Tsukamoto et al. 2023), leaving extended dead zones in which magnetic stresses cannot operate. Where the field remains well coupled to a disc, the magneto-rotational instability (MRI) provides turbulent transport [5, 19]. Finally, in the massive discs that emerge from a collapse, gravitational instability drives spiral arms that redistribute angular momentum [22] (Toomre 1964).

These three are usually studied apart from one another. The division is visible in the structure of the field itself: the lectures collected from the school Role and mechanisms of angular momentum transport in the formation and early evolution of stars treat angular momentum in collapsing dense cores [20] and in protoplanetary discs [16] separately, and largely from separate literatures. The reason is methodological. Collapse calculations resolve the envelope and the forming disc, but either stop at the first hydrostatic core or, when they reach the protostar, follow it only briefly. Disc calculations begin instead from an already-formed disc of prescribed structure, without the collapse history that fixes its mass and its magnetic field. Most simulations that follow the collapse down to the protostar [6, 1, 3] (Vaytet et al. 2018) integrate for months to a few years after its formation. A notable exception is Machida & Basu [27], who followed the system for ∼2000\sim2000 yr and found a disc in which, once the field has dissipated, gravitational torques dominate the transport and drive episodic accretion onto the protostar. Their calculation, however, included Ohmic dissipation only, and a resolution of 5.6×10−35.6 \times10^{-3} AU at the protostar, more than an order of magnitude coarser than here. The relative contribution of the different mechanisms has therefore not yet been measured from the full angular momentum budget of a system in which ambipolar diffusion is included and the protostar–disc interface is resolved.

This is what the present paper does. Using the finite-volume code Idefix (Lesur et al. [26]), we follow the collapse of a dense core with Ohmic and ambipolar diffusion continuously from the cloud scale to the formation of the protostar, with a relative resolution Δr/r≃2%\Delta r/r \simeq2\% that gives cells of ∼10−4\sim10^{-4} AU at the protostellar surface, and then integrate the resulting protostar–disc system for some tens of years, long enough for the inner disc to reach a quasi-steady state. We then analyse the angular momentum budget of that system directly, decomposing it into the Reynolds, Maxwell and gravitational stresses and splitting each into a radial and a meridional component. The decomposition is carried out in spherical geometry, and deliberately retains the vertical direction: the disc that emerges from the collapse is geometrically thick and an order of magnitude more massive than the protostar it surrounds, so neither the thin-disc approximation nor the assumption of a dominant central mass applies, and the vertical transport turns out to be as important as the radial transport in setting where material goes.

The paper is organised as follows. § 2 describes the physical model, the numerical setup and the restart strategy that makes a single calculation spanning seven decades in radius possible, together with the diagnostics used to quantify the transport. § 3 presents the resulting protostar–disc system, Sect. 4 establishes that its accretion pattern is controlled by angular momentum transport, and Sect. 5 identifies which stress dominates in which region, expressing the result as an equivalent α\alpha parameter. Section 6 examines the origin of the magnetic and gravitational stresses and of the mean vertical motion in the disc, Sect. 7 discusses the limitations of the models and summarises our conclusions.

Method

Our goal is to follow the gravitational collapse of a dense core continuously, from the initial cloud down to the formation of the protostar, and then to integrate the resulting protostar–disc system for as long as possible. Simulating the whole collapse provides self-consistent initial conditions for the protostar–disc system, at the price of covering seven orders of magnitude in spatial scale within a single run. In this section we describe the physical model (Sect. 2.1), the initial conditions (Sect. 2.2) and the grid (Sect. 2.3), the restart strategy used to follow the successive scales of the collapse (Sect. 2.4), the boundary conditions (Sect. 2.5), the internal limiters required to keep the integration tractable (Sect. 2.6), and finally the diagnostics used to quantify angular momentum transport (Sect. 2.8).

Physical model

Equations solved

We solve the non-ideal magnetohydrodynamics (MHD) equations with self-gravity, including Ohmic and ambipolar diffusion:

diffusion:

∂tρ+∇⋅(ρv)=0,∂t(ρv)+∇⋅(ρv⊗v−B⊗B4π+PtotI)=−ρ∇Φ,∂tB=∇×(v×B−ηOJ+ηA(J×b)×b),∇2Φ=4πGρ,(1)\begin{aligned} \partial_t \rho+ \nabla\cdot(\rho\mathbf{v}) &= 0,\\ \partial_t(\rho\mathbf{v}) + \nabla\cdot\left(\rho\mathbf{v}\otimes\mathbf{v} - \frac{\mathbf{B}\otimes\mathbf{B}}{4\pi} + P_{\rm tot}\mathbf{I}\right) &= -\rho\nabla\Phi,\\ \partial_t\mathbf{B} &= \nabla\times\left(\mathbf{v}\times\mathbf{B} - \eta_{\rm O}\mathbf{J} + \eta_{\rm A}(\mathbf{J}\times\mathbf{b})\times\mathbf{b}\right),\\ \nabla^2\Phi&= 4\pi G\rho, \tag*{(1)} \end{aligned}

where ρ\rho is the gas density, v\mathbf{v} the velocity, B\mathbf{B} the magnetic field, Φ\Phi the gravitational potential and I\mathbf{I} the identity tensor. We have written Ptot=P(ρ)+B2/8πP_{\rm tot}=P(\rho)+B^2/8\pi for the total (thermal plus magnetic) pressure, J=∇×B\mathbf{J}=\nabla\times\mathbf{B}, and b=B/B\mathbf{b}=\mathbf{B}/B. The quantities ηO\eta_{\rm O} and ηA\eta_{\rm A} are the Ohmic and ambipolar diffusivities. No energy equation is solved, since isothermal conditions is closed by a barotropic equation of state P(ρ)P(\rho) (Sect. 2.1.2). No explicit viscosity is included: the molecular viscosity of the gas is entirely negligible under star-formation conditions, and we deliberately introduce no artificial shear viscosity, so that the angular momentum transport analysed in this paper is produced by the flow itself.

Equation of state

We close the system with a barotropic equation of state, P=ρkBT(ρ)/(μmolmH)P=\rho k_{\rm B}T(\rho)/(\mu_{\rm mol}m_{\rm H}), with μmol=2.34\mu_{\rm mol}=2.34, and with the temperature–density relation of Marchand et al. [28],

T(ρ)=T01+(ρρ1)δ1(1+ρρ2)δ2(1+ρρ3)δ3,(2)T(\rho)=T_0\sqrt{1+\left(\frac{\rho}{\rho_1}\right)^{\delta_1}}\left(1+\frac{\rho}{\rho_2}\right)^{\delta_2}\left(1+\frac{\rho}{\rho_3}\right)^{\delta_3}, \tag*{(2)}

with T0=10T_0=10 K, δ1=0.4\delta_1=0.4, δ2=−0.3\delta_2=-0.3 and δ3=0.56667\delta_3=0.56667. This functional form reproduces the four successive phases of the collapse: the isothermal phase (ρ≪ρ1\rho\ll\rho_1), the first adiabatic core (ρ1≪ρ≪ρ2\rho_1\ll\rho\ll\rho_2, effective γeff=7/5\gamma_{\rm eff}=7/5), the second collapse driven by H2_2 dissociation (ρ2≪ρ≪ρ3\rho_2\ll\rho\ll\rho_3, γeff=1.1\gamma_{\rm eff}=1.1), and the second Larson core (ρ3≪ρ\rho_3\ll\rho, γeff=5/3\gamma_{\rm eff}=5/3).

The critical densities of Marchand et al. [28] (ρ1=3.9×10−13\rho_1=3.9\times10^{-13}, ρ2=3.9×10−8\rho_2=3.9\times10^{-8} and ρ3=3.9×10−3\rho_3=3.9\times10^{-3} g cm−3^{-3}) accurately reproduce the central temperature during the first collapse, but underestimate it during the second collapse, and they always underestimate the temperature in the outer layers of the protostar, where accretion shocks and radiative transport deposit energy (Figure 1). Using them yields a protostar smaller than 0.1 R⊙R_{\odot}, which would be prohibitively expensive to resolve. We therefore keep the functional form of Eq. (2) but reduce the three critical densities by a factor of 100, i.e. ρ1=3.9×10−15\rho_1 = 3.9 \times10^{-15}, ρ2=3.9×10−10\rho_2 = 3.9 \times10^{-10} and ρ3=3.9×10−5 g cm−3\rho_3 = 3.9 \times10^{-5}\ \mathrm{g\ cm^{-3}}, which yields a protostar of ∼1 R⊙\sim1\ R_{\odot}. Although the rescaled law no longer reproduces the evolution of the central temperature, it gives a more faithful description of the temperature of the surrounding gas after second core formation (shaded areas in Figure 1). This rescaling has no thermodynamical justification beyond matching the protostellar radius and the general temperature–density behaviour found in radiation-hydrodynamics calculations; our simulations are consequently not predictive regarding the thermal evolution of the system.

Temperature–density relation adopted in this work

Figure 1. Temperature–density relation adopted in this work. The dashed lines show the evolution of the temperature at the centre of an MHD collapse simulation with radiative transfer, and the shaded areas the final state of the medium after second core formation, both from Vaytet et al. (2018). The purple solid line is the barotropic law of Eq. (2) with the critical densities of Marchand et al. [28]; the black solid line is the same law with the critical densities reduced by a factor of 100, which is the prescription used here.

Non-ideal MHD resistivities

The Ohmic and ambipolar diffusivities are read from the pre-computed tables of Marchand et al. [28], as functions of the density, the temperature and the magnetic field strength, assuming an MRN grain size distribution. This standard approach assumes that the microphysics of the gas–dust mixture relaxes on a timescale much shorter than the dynamical timescale, so that the ionisation state is always in equilibrium with the local conditions. This assumption is questionable in the innermost regions, in particular regarding dust evolution (Borderies et al. [8]), and we return to its consequences in Sect. 7.

Numerical scheme

The system (1) is integrated with the finite-volume code Idefix (Lesur et al. [26]), which is built on the Kokkos performance-portability library (Trott et al. 2022) and runs on GPUs. We use the HLLD approximate Riemann solver (Miyoshi & Kusano [34]), second-order piecewise-linear reconstruction with the van Leer slope limiter (van Leer 1979), and a second-order Runge–Kutta time integrator. The choice of Riemann solver is not innocuous for the problem at hand: the more diffusive HLL solver (Harten et al. [18]; Einfeldt et al. [13]) suppresses the magneto-rotational instability at the resolutions we can afford (see also Flock et al. [15]). We therefore adopt HLLD throughout the main text, and report the corresponding HLL run in Appendix B.

The solenoidality of the magnetic field is enforced by constrained transport (Evans & Hawley [14]), in which we evolve the vector potential AA at the cell edges rather than the face-centred field itself, and recover B\mathbf{B} from its circulation. The Poisson equation is solved with a BiCGSTAB iterative solver (Saad 2003), as implemented in Idefix by Mauxion [30], and the gravitational acceleration is obtained from a fourth-order finite-difference estimate of ∇Φ\nabla\Phi. The parabolic (diffusive) terms are treated explicitly with a second-order Runge–Kutta–Legendre super-time-stepping scheme (Meyer et al. [33]; Vaidya et al. 2017), which covers a hyperbolic time step Δthyp\Delta t_{\mathrm{hyp}} in ∼2ζ\sim2\sqrt{\zeta} sub-steps instead of ∼2ζ\sim2\zeta, where ζ=Δthyp/Δtpar\zeta= \Delta t_{\mathrm{hyp}}/\Delta t_{\mathrm{par}} is the ratio of the hyperbolic to the parabolic time step. Even so, ζ\zeta can exceed 10610^6 in a protostellar collapse, which is why an additional diffusivity limiter is required (Sect. 2.6.1).

Initial conditions

We adopt initial conditions similar to those of Vaytet et al. (2018): a uniform-density sphere of mass M0M_0 and radius R0R_0, in solid-body rotation at angular velocity Ω0\Omega_0, threaded by a uniform magnetic field B0B_0 aligned with the rotation axis. For a given M0M_0, the three parameters R0R_0, Ω0\Omega_0 and B0B_0 are set by the dimensionless ratios of thermal, rotational and magnetic energy to the gravitational energy of the cloud,

α=52kBT0R0μmolmHGM0,β=13Ω02R03GM0,μ=M0πR02B0/(M/ΦB)crit(3)\alpha= \frac{5}{2}\frac{k_{\mathrm{B}}T_0R_0}{\mu_{\mathrm{mol}}m_{\mathrm{H}}GM_0}, \qquad\beta= \frac{1}{3}\frac{\Omega_0^2R_0^3}{GM_0}, \qquad\mu= \frac{M_0}{\pi R_0^2B_0/(M/\Phi_{\mathrm{B}})_{\mathrm{crit}}} \tag*{(3)}

where T0=10 KT_0 = 10\ \mathrm{K} is the initial cloud temperature and (M/ΦB)crit=0.533π(5G)1/2(M/\Phi_{\mathrm{B}})_{\mathrm{crit}} = \frac{0.53}{3\pi}\left(\frac{5}{G}\right)^{1/2} is the critical mass-to-flux ratio (Mouschovias & Spitzer [36]). We adopt M0=1 M⊙M_0 = 1\ M_{\odot}, α=0.4\alpha= 0.4, β=0.01\beta= 0.01 and μ=4\mu= 4, which correspond to a gravitationally unstable, moderately magnetised and slowly rotating core.

The corresponding initial fields are

ρ=ρ0=3M04πR03if r<R0,ρ0/100 otherwise,vr=vθ=0,vφ=Ω0rsin⁡θif r<R0,0 otherwise,Aφ=B02rsin⁡θif rsin⁡θ<R0,Aφ=B02(rsin⁡θ−R01002/3+R0)otherwise.(4)\begin{aligned} \rho&= \rho_0 = \frac{3M_0}{4\pi R_0^3} && \text{if } r<R_0,\quad\rho_0/100\ \text{otherwise},\\ v_r &= v_\theta= 0,\\ v_\varphi&= \Omega_0r\sin\theta&& \text{if } r<R_0,\quad0\ \text{otherwise},\\ A_\varphi&= \frac{B_0}{2}r\sin\theta&& \text{if } r\sin\theta<R_0,\\ A_\varphi&= \frac{B_0}{2}\left(\frac{r\sin\theta-R_0}{100^{2/3}}+R_0\right) && \text{otherwise}. \tag*{(4)} \end{aligned}

The ambient medium outside the cloud is set to ρ0/100\rho_0/100 to avoid excessively low densities, and the vector potential is scaled down outside the cloud following the B∝ρ2/3B\propto\rho^{2/3} relation expected for spherical collapse. No azimuthal perturbation is added to the system at this stage.

Grid

Idefix does not implement adaptive mesh refinement, so the resolution must be concentrated on the region of interest by the choice of geometry. We use a spherical grid (r,θ,φ)(r,\theta,\varphi) with logarithmic spacing in the radial direction, which gives a constant relative resolution Δr/r\Delta r/r across the seven decades covered by the collapse. The cell interfaces are placed at

ri−1/2=rinαri,withαr=(routrin)1/Nr,(5)r_{i-1/2}=r_{\mathrm{in}}\alpha_r^i,\qquad\text{with}\qquad\alpha_r=\left(\frac{r_{\mathrm{out}}}{r_{\mathrm{in}}}\right)^{1/N_r}, \tag*{(5)}

so that Δri=ri−1/2(αr−1)\Delta r_i=r_{i-1/2}(\alpha_r-1). We set Nr=128log⁡10(rout/rin)N_r=128\log_{10}(r_{\mathrm{out}}/r_{\mathrm{in}}), i.e. 128 cells per decade, giving αr−1≃1.8%\alpha_r-1\simeq1.8\%.

The outer radius is rout=104 AUr_{\mathrm{out}}=10^4\ \mathrm{AU}, large enough to enclose the initial cloud and to keep the outer boundary dynamically irrelevant. The final inner radius is rin=10−3 AUr_{\mathrm{in}}=10^{-3}\ \mathrm{AU}, well inside the expected protostellar radius (Vaytet et al. 2018; Ahmad et al. [3]); it is not used from the start, but reached progressively through the radial restarts described in Sect. 2.4. In the polar direction we use Nθ=128N_\theta=128 uniform cells over [0,π][0,\pi], giving Δθ≃0.025\Delta\theta\simeq0.025, comparable to the radial resolution. In the azimuthal direction we use Nφ=64N_\varphi=64 cells over [0,π][0,\pi] rather than 128 cells over [0,2π][0,2\pi], which gives the same spatial resolution rsin⁡θ Δϕ≃0.05 rsin⁡θr\sin\theta\,\Delta\phi\simeq0.05\,r\sin\theta while suppressing odd azimuthal modes, for the reason explained in Sect. 2.4. The grid is summarised in Table 1.

DirectionLowerUpperResolution
rr10−3 AU10^{-3}\ \mathrm{AU}104 AU10^{4}\ \mathrm{AU}128 cells per decade
θ\theta0π\pi128 cells
φ\varphi02π2\pi64 cells

Table 1. Final grid extent and resolution, after all restarts.

The sin⁡θ\sin\theta factor in the azimuthal resolution makes the cells vanishingly thin close to the polar axis, which would cripple the time step in a region of no physical interest. We therefore use the azimuthal grid coarsening of Idefix ([26]), merging cells in ϕ\phi by a factor

fcoarse=2n,n=max⁡(0,⌊−log⁡2(sin⁡θ)⌋),(6)f_{\mathrm{coarse}} = 2^n,\qquad n = \max\left(0,\left\lfloor-\log_2(\sin\theta)\right\rfloor\right), \tag*{(6)}

No coarsening is applied for π/6≤θ≤5π/6\pi/6 \le\theta\le5\pi/6; closer to the axis, the factor doubles each time sin⁡θ\sin\theta is halved (a factor of 2 for θ<π/6\theta< \pi/6, 4 for θ≲14.5∘\theta\lesssim14.5^\circ, and so on, symmetrically in the southern hemisphere). Since 2nsin⁡θ>1/22^n\sin\theta> 1/2 by construction, the azimuthal spacing fcoarsersin⁡θ Δϕf_{\mathrm{coarse}}r\sin\theta\,\Delta\phi never falls below rΔϕ/2=rΔθr\Delta\phi/2 = r\Delta\theta anywhere in the grid.

Restart procedure

Following the collapse over seven decades in radius, and from axisymmetry to a fully three-dimensional protostar–disc system, is made tractable by a sequence of restarts in which the extent of the domain — but not its relative resolution — is changed. They are detailed in Appendix A, and we only summarise them here.

The inner radius starts at 1 AU and is reduced by one decade at a time down to 10−310^{-3} AU, the new cells being filled by an extrapolation that preserves the solenoidal constraint: the first reduction is performed as the first Larson core is about to form, the second at the onset of the second collapse, and the third as the second Larson core is about to form. The first collapse is computed in axisymmetry, and the simulation is restarted in three dimensions shortly before the formation of the first Larson core, with an m=2m=2 density perturbation to seed non-axisymmetric structures. Until just before the second collapse it is moreover restricted to the northern hemisphere, which enforces equatorial symmetry and keeps the first core at the centre of the grid. Finally, the azimuthal domain is restricted to ϕ∈[0,π]\phi\in[0,\pi], which suppresses odd azimuthal modes: in a full azimuthal domain, an m=1m=1 mode systematically develops and ejects the protostar from the centre of the grid. The higher-resolution run of Table 2 is obtained by restarting the reference run on a refined grid.

NameResolutionSolver
1MsunHLLDTable 1HLLD
1MsunHLLDx2×2 in θ\theta, φ\varphiHLLD
1MsunHLLTable 1HLL

Table 2. Simulations discussed in this paper. 1MsunHLLD is the reference run.

Boundary conditions

Boundary conditions are imposed through two layers of ghost cells. In MHD only the hydrodynamical variables and the magnetic field components tangential to the boundary need to be specified; the normal component follows from the divergence-free constraint.

  • Azimuthal direction. Periodic. In addition, the normal component BϕB_\phi must be explicitly forced to take the same value on both sides, otherwise round-off errors accumulate over long integrations and generate a spurious toroidal field.

  • Polar axis. We use the regularised axis boundary condition of Zhu & Stone (2018), which assigns a single value to fields defined multiple times on the singular axis. Since our domain spans ϕ∈[0,π]\phi\in[0,\pi], it reduces to a reflection with ρ\rho, vrv_r and BrB_r copied and vθv_\theta, vϕv_\phi, BθB_\theta, BϕB_\phi reversed.

  • Mid-plane. During the half-domain phase (Appendix A.§ 2), a reflective condition is applied at θ=π/2\theta= \pi/2: vθv_\theta, BrB_r and BϕB_\phi are reversed in the ghost cells, the other fields copied.

  • Outer radius. Outflow: zero-gradient for all fields, except the radial velocity, which is set to zero if it points inwards, so that no material can be accreted through the outer boundary.

  • Inner radius. A plain outflow condition would impose a vanishing pressure gradient at rinr_{\mathrm{in}} and could not support a hydrostatic core. We therefore use a modified outflow condition, identical to the outer one except for the density, which is set in the ghost cells to the constant value ρin\rho_{\mathrm{in}} obtained by integrating the mass flux through the inner boundary over time. The sphere r<rinr < r_{\mathrm{in}} then behaves as a homogeneous body of density ρin\rho_{\mathrm{in}}, against which a hydrostatic core can build up.

The inner boundary nevertheless remains the most delicate part of the setup, since the zero-gradient condition on BθB_\theta and BϕB_\phi generates artificial currents. We assess its consequences in Sect. 7.1.

Internal limiters

Three limiters are required to keep the integration tractable, or physically meaningful, in the low-density polar cavities that open above the protostar. All three act in regions that we show below to be dynamically disconnected from the disc and the protostar, but we describe them in full since they are not innocuous for the polar regions.

Diffusivity limiter

Even with super-time-stepping, the ratio ζ\zeta of the hyperbolic to the parabolic time step can exceed 10610^6 in a protostellar collapse. Following Vaytet et al. (2018) and Mauxion et al. [31], we cap each diffusivity at

ηmax⁡=CdiffΔl22Δthyp,(7)\eta_{\max} = C_{\mathrm{diff}}\frac{\Delta l^2}{2\Delta t_{\mathrm{hyp}}}, \tag*{(7)}

with Δl\Delta l the local cell size and Cdiff=100C_{\mathrm{diff}} = 100, so that a hyperbolic time step is covered in ∼2Cdiff=20\sim2\sqrt{C_{\mathrm{diff}}} = 20 sub-steps instead of ∼200\sim200. Where the cap is active, diffusion still dominates the local dynamics, so the physical behaviour is preserved. On a logarithmic spherical grid ηmax⁡∝r2\eta_{\max} \propto r^2, which is most restrictive at small radii; two effects mitigate this, namely the reduction of the hyperbolic time step once the protostar forms, and the drop of the diffusivities caused by dust vaporisation at high temperature. As Fig. 2 shows, the cap does not affect the Ohmic diffusivity at all at this stage, and reduces the ambipolar diffusivity only in the polar regions, outside the disc and the protostar.

Alfvén speed and mean-free-path limiters

Polar cavities open above the protostar after the formation of the first Larson core, with densities falling below 10−20 g cm−310^{-20}\ \mathrm{g\,cm^{-3}}. Two problems follow. First, the Alfvén speed vA=B/4πρv_{\mathrm{A}} = B/\sqrt{4\pi\rho} reaches several 103 km s−110^3\ \mathrm{km\,s^{-1}} there. Through the CFL condition, these few cells set a very restrictive time step over the whole domain, and the cost is paid at every step of the simulation in order to follow wave propagation in a region that holds a negligible fraction of the mass, and that is not the object of this study. We therefore cap it at

Figure 2. Ohmic (top) and ambipolar (bottom) diffusivities as a function of spherical radius, after the protostar formation. Colours show ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|, the angle to the equatorial plane. The dashed line is the diffusivity limiter ηmax⁡\eta_{\max}, computed for Δthyp∼3×10−6\Delta t_{\mathrm{hyp}} \sim3 \times10^{-6} yr.

vA,max=vA,refr1 AU,(8)v_{\mathrm{A,max}} = v_{\mathrm{A,ref}} \frac{r}{1\ \mathrm{AU}}, \tag*{(8)}

which is enforced by imposing the density floor ρfloor,A=B2/(4πvA,max2)\rho_{\mathrm{floor,A}} = B^2/(4\pi v_{\mathrm{A,max}}^2). The linear radial dependence has the same origin as for the diffusivity cap: on a logarithmic grid the CFL condition scales as Δt∝r/vA,max\Delta t \propto r/v_{\mathrm{A,max}}, so a larger Alfvén speed can be tolerated at larger radii for the same time-step restriction. We use vA,ref∼90v_{\mathrm{A,ref}} \sim90 AU yr−1^{-1} during the first-core phase, raised to ∼4000\sim4000 AU yr−1^{-1} once the protostar has formed and the sound speed has reduced the hyperbolic time step by two orders of magnitude.

Second, at such densities the particle mean free path exceeds the cell size and the fluid description breaks down. We impose a second density floor,

ρfloor,MFP=μmolmH2σlmax⁡,withlmax⁡=Δl100,(9)\rho_{\mathrm{floor,MFP}} = \frac{\mu_{\mathrm{mol}}m_{\mathrm{H}}}{\sqrt{2}\sigma l_{\max}}, \qquad\text{with}\qquad l_{\max} = \frac{\Delta l}{100}, \tag*{(9)}

where σ=4.58×10−15\sigma= 4.58 \times10^{-15} cm2^2 is the collisional cross section estimated from the van der Waals radius of H2_2 ((Weast 1981)). This floor scales as r−1r^{-1}, whereas ρfloor,A\rho_{\mathrm{floor,A}} follows the magnetic field, so their relative importance changes as the collapse proceeds and depending on the local radius: the two are comparable during the first-core phase, while once the second core has formed, the Alfvén speed floor becomes the dominant one in the inner regions and the mean-free-path floor becomes the dominant one beyond ∼0.1\sim0.1 AU. Figure 3 shows both at that stage. They are active only in the polar cavities, well outside the disc and the protostar; both are applied at fixed momentum and angular momentum, and the total mass they add over the whole simulation stays below 10−6 M⊙10^{-6}\ M_\odot, entirely negligible compared with the mass of the protostar and the disc.

Figure 3. Density as a function of spherical radius after the protostar formation, with colours showing ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|, illustrating the two density floors. The dashed line is the floor ρfloor,A\rho_{\mathrm{floor,A}} associated with the Alfvén speed limiter, here for vA,ref=4000v_{\mathrm{A,ref}} = 4000 AU yr−1^{-1}; the dotted line is the floor ρfloor,MFP\rho_{\mathrm{floor,MFP}} associated with the mean-free-path limiter. Both are active only in the polar cavities.

Temperature floor

In the same cavities, the barotropic law returns temperatures below ∼100\sim100 K, whereas the gas there would be directly irradiated by the protostar. This does not create numerical problems, but it does feed the resistivity tables with temperatures at which dust has not been vaporised, and therefore with spuriously high resistivities. We therefore impose, once the protostar has formed and only for the computation of the resistivities, a temperature floor

Tfloor=Tfloor,ref(r0.01 AU)−1/2,(10)T_{\mathrm{floor}} = T_{\mathrm{floor,ref}}\left(\frac{r}{0.01\ \mathrm{AU}}\right)^{-1/2}, \tag*{(10)}

with Tfloor,ref=4000T_{\mathrm{floor,ref}} = 4000 K, close to the protostellar surface temperature. The r−1/2r^{-1/2} scaling follows from radiative equilibrium with the protostar, F∝r−2F \propto r^{-2} and F∝T4F \propto T^4. Since the resistivities are insensitive to temperature below ∼1000\sim1000 K, the floor is not applied where TfloorT_{\mathrm{floor}} falls below that value, i.e. beyond 0.16 AU. This prescription assumes an optically thin cavity and is therefore optimistic; its impact on the magnetic field evolution is discussed in Sect. 7.1.

Simulations

All the simulations discussed here share the initial conditions of Sect. 2.2 and the grid of Table 1, and differ only in

@page 5/26 starts_mid=0 ends_mid=1

Figure 2 (PDF p. 5). Ohmic (top) and ambipolar (bottom) diffusivities as a function of spherical radius, after the protostar formation. Colours show ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|, the angle to the equatorial plane. The dashed line is the diffusivity limiter ηmax⁡\eta_{\max}, computed for Δthyp∼3×10−6\Delta t_{\mathrm{hyp}} \sim3 \times10^{-6} yr.

that holds a negligible fraction of the mass, and that is not the object of this study. We therefore cap it at

vA,max=vA,refr1 AU,(11)v_{\mathrm{A,max}} = v_{\mathrm{A,ref}} \frac{r}{1\ \mathrm{AU}}, \tag*{(11)}

which is enforced by imposing the density floor ρfloor,A=B2/(4πvA,max2)\rho_{\mathrm{floor,A}} = B^2/(4\pi v_{\mathrm{A,max}}^2). The linear radial dependence has the same origin as for the diffusivity cap: on a logarithmic grid the CFL condition scales as Δt∝r/vA,max\Delta t \propto r/v_{\mathrm{A,max}}, so a larger Alfvén speed can be tolerated at larger radii for the same time-step restriction. We use vA,ref∼90v_{\mathrm{A,ref}} \sim90 AU yr−1^{-1} during the first-core phase, raised to ∼4000\sim4000 AU yr−1^{-1} once the protostar has formed and the sound speed has reduced the hyperbolic time step by two orders of magnitude.

Second, at such densities the particle mean free path exceeds the cell size and the fluid description breaks down. We impose a second density floor,

ρfloor,MFP=μmolmH2σlmax⁡,withlmax⁡=Δl100,(12)\rho_{\mathrm{floor,MFP}} = \frac{\mu_{\mathrm{mol}}m_{\mathrm{H}}}{\sqrt{2}\sigma l_{\max}}, \qquad\text{with}\qquad l_{\max} = \frac{\Delta l}{100}, \tag*{(12)}

where σ=4.58×10−15\sigma= 4.58 \times10^{-15} cm2^2 is the collisional cross section estimated from the van der Waals radius of H2_2 ((Weast 1981)). This floor scales as r−1r^{-1}, whereas ρfloor,A\rho_{\mathrm{floor,A}} follows the magnetic field, so their relative importance changes as the collapse proceeds and depending on the local radius: the two are comparable during the first-core phase, while once the second core has formed, the Alfvén speed floor becomes the dominant one in the inner regions and the mean-free-path floor becomes the dominant one beyond ∼0.1\sim0.1 AU. Figure 3 shows both at that stage. They are active only in the polar cavities, well outside the disc and the protostar; both are applied at fixed momentum and angular momentum, and the total mass they add over the whole simulation stays below 10−6 M⊙10^{-6}\ M_\odot, entirely negligible compared with the mass of the protostar and the disc.

Figure 3 (PDF p. 5). Density as a function of spherical radius after the protostar formation, with colours showing ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|, illustrating the two density floors. The dashed line is the floor ρfloor,A\rho_{\mathrm{floor,A}} associated with the Alfvén speed limiter, here for vA,ref=4000v_{\mathrm{A,ref}} = 4000 AU yr−1^{-1}; the dotted line is the floor ρfloor,MFP\rho_{\mathrm{floor,MFP}} associated with the mean-free-path limiter. Both are active only in the polar cavities.

Temperature floor

In the same cavities, the barotropic law returns temperatures below ∼100\sim100 K, whereas the gas there would be directly irradiated by the protostar. This does not create numerical problems, but it does feed the resistivity tables with temperatures at which dust has not been vaporised, and therefore with spuriously high resistivities. We therefore impose, once the protostar has formed and only for the computation of the resistivities, a temperature floor

Tfloor=Tfloor,ref(r0.01 AU)−1/2,(13)T_{\mathrm{floor}} = T_{\mathrm{floor,ref}}\left(\frac{r}{0.01\ \mathrm{AU}}\right)^{-1/2}, \tag*{(13)}

with Tfloor,ref=4000T_{\mathrm{floor,ref}} = 4000 K, close to the protostellar surface temperature. The r−1/2r^{-1/2} scaling follows from radiative equilibrium with the protostar, F∝r−2F \propto r^{-2} and F∝T4F \propto T^4. Since the resistivities are insensitive to temperature below ∼1000\sim1000 K, the floor is not applied where TfloorT_{\mathrm{floor}} falls below that value, i.e. beyond 0.16 AU. This prescription assumes an optically thin cavity and is therefore optimistic; its impact on the magnetic field evolution is discussed in Sect. 7.1.

Simulations

All the simulations discussed here share the initial conditions of Sect. 2.2 and the grid of Table 1, and differ only in their angular resolution and in the Riemann solver, as listed in Table 2. The reference run, 1MsunHLLD, is followed for ∼64\sim64 yr after the formation of the protostar. The convergence run 1MsunHLLDx2 is about an order of magnitude more expensive and is followed for ∼17\sim17 yr only. The HLL run, followed for ∼70\sim70 yr, is presented in Appendix B.

Angular momentum budget

The analysis presented in Sects. 4 and 5 relies on a decomposition of the angular momentum budget, which we define here. Throughout, ⟨X⟩\langle X\rangle denotes the azimuthal average of a field XX over the azimuthal domain φ∈[0,π]\varphi\in[0,\pi],

⟨X⟩(r,θ)=1π∫0πX(r,θ,φ) dφ,(14)\langle X\rangle(r,\theta)=\frac{1}{\pi}\int_{0}^{\pi}X(r,\theta,\varphi)\,\mathrm{d}\varphi, \tag*{(14)}

and δX=X−⟨X⟩\delta X=X-\langle X\rangle its non-axisymmetric part. We write s=rsin⁡θs=r\sin\theta for the cylindrical radius, Ω=vφ/s\Omega=v_{\varphi}/s for the angular velocity, and use the subscript p for the poloidal part of a vector field.

Taking the azimuthal average of the conservation equation for the axial angular momentum, and assuming a steady state, gives

⟨ρvp⟩⋅∇(s2Ω)=−∇⋅(s⟨ρδvφvp⟩)−s4π⟨BφBp⟩+s4πG⟨gφgp⟩,(15)\langle\rho\mathbf{v}_{p}\rangle\cdot\nabla(s^{2}\Omega)=-\nabla\cdot\left(s\langle\rho\delta v_{\varphi}\mathbf{v}_{p}\rangle\right)-\frac{s}{4\pi}\langle B_{\varphi}\mathbf{B}_{p}\rangle+\frac{s}{4\pi G}\langle g_{\varphi}\mathbf{g}_{p}\rangle, \tag*{(15)}

where g=−∇Φ\mathbf{g}=-\nabla\Phi. The left-hand side is the advection of the fluid along the specific angular momentum gradient, i.e. the accretion or decretion of mass; the right-hand side gathers the three stresses that transport angular momentum, namely the Reynolds stress, which measures the correlation of the velocity fluctuations; the Maxwell stress, which requires a toroidal field correlated with a poloidal one; and the gravitational stress, which requires non-axisymmetric density structures.

Because our disc is thick and our grid is spherical, we evaluate the divergence in spherical rather than in the cylindrical coordinates more commonly used in the disc literature. The right-hand side then splits into six local torques, one radial and one meridional for each stress:

Hr=−1r2∂r(r3⟨ρδvφvr⟩),Hθ=−1sin⁡θ∂θ(sin⁡2θ⟨ρδvφvθ⟩),Mr=14πr2∂r(r3⟨BφBr⟩),Mθ=14πsin⁡θ∂θ(sin⁡2θ⟨BφBθ⟩),Gr=−14πGr2∂r(r3⟨gφgr⟩),Gθ=−14πGsin⁡θ∂θ(sin⁡2θ⟨gφgθ⟩),(16)\begin{aligned} \mathcal{H}_{r}&=-\frac{1}{r^{2}}\partial_{r}\left(r^{3}\langle\rho\delta v_{\varphi}v_{r}\rangle\right),\\ \mathcal{H}_{\theta}&=-\frac{1}{\sin\theta}\partial_{\theta}\left(\sin^{2}\theta\langle\rho\delta v_{\varphi}v_{\theta}\rangle\right),\\ \mathcal{M}_{r}&=\frac{1}{4\pi r^{2}}\partial_{r}\left(r^{3}\langle B_{\varphi}B_{r}\rangle\right),\\ \mathcal{M}_{\theta}&=\frac{1}{4\pi\sin\theta}\partial_{\theta}\left(\sin^{2}\theta\langle B_{\varphi}B_{\theta}\rangle\right),\\ \mathcal{G}_{r}&=-\frac{1}{4\pi G r^{2}}\partial_{r}\left(r^{3}\langle g_{\varphi}g_{r}\rangle\right),\\ \mathcal{G}_{\theta}&=-\frac{1}{4\pi G\sin\theta}\partial_{\theta}\left(\sin^{2}\theta\langle g_{\varphi}g_{\theta}\rangle\right), \tag*{(16)} \end{aligned}

with H=Hr+Hθ\mathcal{H}=\mathcal{H}_{r}+\mathcal{H}_{\theta} the total Reynolds torque, and similarly for the Maxwell torque M\mathcal{M} and the gravitational torque G\mathcal{G}. Separating the radial from the meridional component matters here: in a thick, self-gravitating disc, angular momentum is transported vertically as well as radially, and this distinction is lost in the vertically integrated, one-dimensional α\alpha-disc description (Shakura & Sunyaev 1973; Pringle 1981).

Overview of the protostar–disc system

All the results presented in this and the following sections are drawn from the reference run 1MsunHLLD (Table 2), which follows the protostar–disc system for its first ∼64\sim64 yr of evolution. We take the formation of the protostar, defined as the time at which the peak density first exceeds 3×10−5 g cm−33\times10^{-5}\ {\rm g\,cm^{-3}}, as the origin of time throughout, and use the azimuthal average ⟨X⟩\langle X\rangle and the non-axisymmetric deviation δX\delta X introduced in Sect. 2.8.

Before turning to angular momentum transport, we describe the system that the collapse produces. We distinguish three structures throughout. The protostar is the region where the density exceeds 3×10−5 g cm−33\times10^{-5}\ {\rm g\,cm^{-3}}, the threshold also used to define the origin of time. The disc is the rotationally supported region that surrounds it, defined as the region where gravity is balanced by the centrifugal and pressure forces; its surface is drawn as a dashed or dotted contour in the maps that follow. The envelope is the remaining, infalling gas. We then

describe the rotation of the protostar and the disc, the flow that feeds them, the regimes of the magnetic field, the spiral structure that develops in the disc, and the way the protostar itself evolves.

Protostar and disc

The collapse produces two distinct structures, which the rotation profile separates cleanly. Figure 4 shows the mid-plane angular velocity and the enclosed mass as a function of radius. Inside ∼10−2\sim10^{-2} AU, ⟨Ω⟩\langle\Omega\rangle is constant: the protostar is in solid-body rotation, at Ωprotostar∼200 rad yr−1\Omega_{\mathrm{protostar}}\sim200\ \mathrm{rad\ yr}^{-1}, i.e. with a rotation period of ∼11\sim11 days. The enclosed mass flattens at the same radius, giving a protostellar mass of ∼8×10−3 M⊙\sim8\times10^{-3}\ M_{\odot}.

Figure 4. Azimuthally averaged angular velocity at the mid-plane (toptop) and enclosed mass (bottombottom) as a function of radius, 10 yr after the protostar formation. The red dashed line is the equilibrium angular velocity Ωeq=GMenc/r3\Omega_{\rm eq}=\sqrt{GM_{\rm enc}/r^{3}} computed from the enclosed mass, and the blue dotted line a Keplerian Ω∝r−3/2\Omega\propto r^{-3/2} profile.

Outside this radius, and out to ∼2\sim2 AU, the angular velocity follows the equilibrium profile Ωeq=GMenc/r3\Omega_{\mathrm{eq}}=\sqrt{GM_{\mathrm{enc}}/r^{3}}: this is a rotationally supported circumstellar disc, which formed from the first Larson core and is therefore already in place when the protostar appears. The profile is noticeably shallower than Keplerian in the outer part of the disc, because the mass of the disc is not negligible compared with that of the central object: the enclosed mass reaches ∼8×10−2 M⊙\sim8\times10^{-2}\ M_{\odot} by the disc outer edge, so the disc is an order of magnitude more massive than the protostar it surrounds, and it is largely self-gravitating. This has a direct consequence for what follows — gravitational torques cannot be dismissed a priori in such a disc. The disc outer edge, where the rotation profile departs from Ωeq\Omega_{\mathrm{eq}} and where the enclosed mass profile flattens, lies at ∼2−4\sim2-4 AU.

Density structure and flow

Figure 5 shows the density and the poloidal velocity field at three successive scales. At the largest scale (Figure 5a), the disc is embedded in the infalling envelope and fed through a pseudo-disc, a flattened structure supported vertically by thermal and magnetic pressure but not radially ([17]), along which material is funnelled towards the centre. The disc itself is hot and geometrically thick, with an aspect ratio H/R∼1H/R\sim1: it is supported vertically by its thermal pressure, as also found by Ahmad et al. [2]. The thin-disc approximation therefore does not apply, and vertical and radial transport are a priori comparable.

Figure 5. Density in the ss–zz plane at three successive scales, ∼12\sim12 yr after the protostar formation. In each panel, the vertical ticks have the same spacing as the radial ones. Colours show the density and white lines the poloidal velocity field, both averaged in the azimuthal direction. The dashed lines indicate the boundaries of the disc. The black contour in panel (c) marks the protostar surface, defined as the region where the density exceeds 3×10−5 g cm−33 \times10^{-5}\ \mathrm{g\ cm^{-3}}.

Two features of the flow are worth noting at this stage, since both will reappear in the angular momentum budget. First, at the AU scale (Figure 5b) the poloidal velocity field is organised into large axisymmetric meridional eddies, corresponding to a systematic vertical motion of the gas rather than to purely radial accretion. Second, at the protostellar scale (Figure 5c) the mean flow is disordered, with small eddies, and the non-axisymmetric fluctuations are strong (Sect. 3.4); low-density polar cavities have opened above and below the disc, from which material is ejected. Over the duration of our simulation, the mass carried by these ejections is negligible; they may nonetheless carry away a non-negligible amount of angular momentum, and their mass is likely to become significant on longer timescales ([27]). Since they are launched from precisely the regions where the density floors and the Alfvén speed limiter are active (Sect. 2.6), we do not analyse them further here.

Magnetic field regimes

The magnetic field enters the angular momentum budget only through the Maxwell stress, which requires a toroidal field correlated with a poloidal one. Such a stress can only

develop where the field is coupled to the gas and wound up by its rotation, and we first identify where this is the case. Figure 6 and Figure 7 reveal two distinct regimes.

Figure 6. Magnetic field structure in the ss–zz plane at three successive scales, ∼12\sim12 yr after the protostar formation. In each panel, the vertical ticks have the same spacing as the radial ones. Colours show the ratio ⟨Bφ⟩/⟨B⟩\langle B_\varphi\rangle/ \langle B\rangle and white lines the poloidal magnetic field, both averaged in the azimuthal direction. The black contour in panel (c) marks the protostar surface.

Figure 7. Azimuthally averaged Elsässer number Λ\Lambda of Eq. (17) (top), plasma beta βp\beta_{\mathrm{p}} (middle) and magnetic field strength (bottom) as a function of density, ∼12\sim12 yr after the protostar formation. Colours show ∣δ∣=∣π/2−θ∣\lvert\delta\rvert=\lvert\pi/2-\theta\rvert, the angle to the equatorial plane, so that red points sample the disc and blue points the polar regions. The dotted lines mark Λ=1\Lambda=1 and βp=1\beta_{\mathrm{p}}=1, and the dashed line is the ideal-MHD scaling B∝ρ2/3B\propto\rho^{2/3}.

Immediately around the disc, out to a few AU, the field is nearly vertical and almost purely poloidal (Figure 6a): ambipolar diffusion has decoupled the field from the gas, which can neither drag nor wind it up. On larger scales (≳7−8\gtrsim7-8 AU), in the envelope, the field is coupled to the gas again, and the toroidal component generated by the rotation of the envelope is no longer negligible. The B−ρB-\rho relation follows the ideal-MHD scaling B∝ρ2/3B \propto\rho^{2/3} at low density, but saturates at ∼0.1\sim0.1 G over several decades in density near the midplane, from ∼10−14\sim10^{-14} to ∼10−7\sim10^{-7} g cm−3^{-3} (Figure 7). This plateau is the signature of ambipolar diffusion decoupling the field from the gas: the field is no longer amplified by compression, and it cannot be wound up by the differential rotation of the disc. The coupling is measured more directly by the Elsässer number,

Λ=vA2(ηΩ+ηA)Ω,(17)\Lambda= \frac{v_{\mathrm{A}}^2}{(\eta_{\Omega}+\eta_{\mathrm{A}})\Omega}, \tag*{(17)}

with vA=B/4πρv_{\mathrm{A}}=B/\sqrt{4\pi\rho} the Alfvén speed, which compares the rate at which the rotation of the gas winds up the field with

the rate at which Ohmic and ambipolar diffusion dissipate it. In the mid-plane, Λ\Lambda falls below unity at ρ≳10−13\rho\gtrsim10^{-13} g cm−3^{-3} and reaches ∼10−8\sim10^{-8} near 10−810^{-8} g cm−3^{-3} (Figure 7, top panel): this region is a magnetic dead zone, in which the Maxwell stress is expected to vanish.

Below ∼0.1\sim0.1 AU, the picture reverses. The temperature is high enough to vaporize the dust grains, causing the resistivities to drop sharply ([28, 8]) and the field to recouple to the gas: Λ\Lambda rises steeply with density and exceeds unity again above ∼10−7\sim10^{-7} g cm−3^{-3}. The differential rotation of the inner disc winds the poloidal field into a strong toroidal component: ∣Bφ∣/B|B_\varphi|/B approaches unity and changes sign across the mid-plane (Figure 6b), while the field strength rises steeply with density to reach 3–10 kG in the protostar region. Closer to the protostar (Figure 6c), the field structure becomes disordered and appears turbulent, a point we return to in Sect. 6.1.

The plasma beta βp\beta_{\mathrm{p}} gives a complementary piece of information. While Λ\Lambda indicates whether the field is coupled to the gas, βp\beta_{\mathrm{p}} indicates how large a stress a coupled field can exert: normalised by the gas pressure, the Maxwell stress is bounded by ∣BrBϕ∣/4πP≤B2/8πP=1/βp\lvert B_rB_\phi\rvert/4\pi P \le B^2/8\pi P = 1/\beta_{\mathrm{p}}. In the mid-plane, ⟨βp⟩\langle\beta_{\mathrm{p}}\rangle reaches ∼107\sim10^7 in the dead zone and drops by several orders of magnitude in the inner disc (Figure 7, middle panel), so that the recoupled field is also strong enough to exert a significant stress there. The system therefore separates into a magnetically dead outer disc and a magnetically active inner disc, with the transition set by the recoupling radius near 0.1 AU. We should consequently expect the Maxwell torque to be confined to the inner region, and some other mechanism to be responsible for the transport further out. § 5 confirms this expectation quantitatively.

The non-axisymmetric structure of the disc

The description so far has treated the disc as axisymmetric. It is not, and the departures are large. Figure 8 shows the mid-plane in the x−yx-y plane, at the scale of the inner disc and at the scale of the disc as a whole.

Figure 8. Non-axisymmetric structure of the disc in the x−yx-y plane at z=0z=0, at two scales: 0.3 AU (top) and 4 AU (bottom). From left to right: the relative density perturbation δρ/⟨ρ⟩\delta\rho/\langle\rho\rangle, the relative azimuthal velocity perturbation δvφ/⟨vφ⟩\delta v_\varphi/\langle v_\varphi\rangle, and the radial velocity normalised by the azimuthal one, vr/⟨vφ⟩v_r/\langle v_\varphi\rangle. Only half of the disc is shown, since the azimuthal domain spans φ∈[0,π]\varphi\in[0,\pi] (Sect. 2.4). The black contour marks the protostar surface.

The disc is threaded by spiral arms, visible in the density across the whole disc and, in the sub-au region, in the azimuthal and radial velocities as well. The dominant azimuthal wavenumber is m=2m = 2, as expected for a self-gravitating disc. Their signature in vrv_r is the one that will matter later: the arms do not merely modulate the density, they drive a radial velocity of order a percent of the orbital velocity, alternating in sign across each arm. Both the Reynolds and the gravitational stresses of Eq. (16) are, by construction, produced by departures from axisymmetry — the first requires correlated velocity fluctuations, the second a non-axisymmetric mass distribution — so these arms are the natural candidates to carry them. We take that possibility up in Sect. 5.

Two further features are worth noting. In the innermost disc, inside ∼0.2\sim0.2 AU, a set of concentric rings appears in the density and the radial velocity but not in the azimuthal velocity, which identifies them as sound waves launched by the protostar–disc interaction rather than as a rotating pattern. In the outer disc, beyond ∼1\sim1 AU, the structure becomes less obviously spiral and more disordered, with strong perturbations concentrated near the outer edge; this is the most variable part of the disc.

Evolution of the protostar

Figure 9 shows the time evolution of the mass, accretion rate, axial angular momentum and magnetic energy of the protostar. The mass evolves in two stages. After a brief adjustment in the first few years it grows only slowly, from ∼8.8×10−3\sim8.8 \times10^{-3} to ∼9.2×10−3 M⊙\sim9.2 \times10^{-3}\ M_\odot over the following thirty years, with an accretion rate — computed as the time derivative of the protostar mass — that drops from its initial value by three orders of magnitude and then fluctuates between 10−610^{-6} and 10−4 M⊙ yr−110^{-4}\ M_\odot\ \mathrm{yr}^{-1}. Between ∼43\sim43 and ∼52\sim52 yr this quiescent growth is interrupted by an accretion burst: the accretion rate rises to a sustained 10−4−3×10−4 M⊙ yr−110^{-4} - 3 \times10^{-4}\ M_\odot\ \mathrm{yr}^{-1}, and the protostar gains ∼8×10−4 M⊙\sim8 \times10^{-4}\ M_\odot, nearly a tenth of its mass, in under a decade, before returning to its earlier rate. Episodic accretion of this kind was also found by [27], where it is driven by gravitational torques in the disc. We do not investigate the origin of this burst further here, and focus on the quasi-steady phase that precedes it (Sect. 4).

Figure 9. Evolution of the protostar properties as a function of time after its formation: mass (top left), accretion rate (top right), axial angular momentum (bottom left) and magnetic energy (bottom right). All quantities are integrated over the region enclosed by the protostar surface, and the accretion rate is the time derivative of the protostar mass.

The angular momentum behaves quite differently. It peaks at ∼13 M⊙R⊙2 yr−1\sim13\,M_\odot R_\odot^2\,\mathrm{yr}^{-1} about two years after the protostar forms, then decreases by roughly a fifth over the following ten years and remains near 10 M⊙R⊙2 yr−110\,M_\odot R_\odot^2\,\mathrm{yr}^{-1} until the burst. During this quiescent phase, the angular momentum of the protostar thus stays nearly constant while its mass slowly increases, so that its specific angular momentum decreases: the angular momentum brought in by the accreted material is extracted from the protostar. The burst reverses this for a time — the angular momentum rises to ∼11.4 M⊙R⊙2 yr−1\sim11.4\,M_\odot R_\odot^2\,\mathrm{yr}^{-1}, the material accreted during the burst bringing in more angular momentum than the stresses remove — after which the decline resumes. The magnetic energy, for its part, rises steeply after formation, peaking at ∼1041 erg\sim10^{41}\,\mathrm{erg} around 4 yr, and then settles into oscillations about 1040 erg10^{40}\,\mathrm{erg} for the rest of the run, without responding appreciably to the burst. That it peaks during the early extraction phase already suggests that magnetic stresses play a part in it.

The protostar loses specific angular momentum while continuing to accrete from a rotationally supported disc. Neither is possible without an efficient mechanism to carry angular momentum outwards, and the nature of that mechanism is what the rest of this paper sets out to determine.

Angular momentum equilibrium

Where the mass actually flows

Section 3 established that the protostar accretes while its specific angular momentum decreases. The natural first question is through which parts of the disc the accreted material travels. Figure 10 shows the azimuthally and time-averaged radial mass flux ⟨ρvr⟩\langle\rho v_r\rangle. Since the cells of the map correspond to surfaces of different sizes depending on their radius and latitude, we multiply the flux by r2sin⁡θr^2\sin\theta: the quantity shown, r2sin⁡θ⟨ρvr⟩r^{2}\sin\theta\langle\rho v_{r}\rangle, is the mass accretion rate per unit polar and azimuthal angle, so that the same colour at different radii and latitudes corresponds to the same mass accretion rate.

Azimuthally averaged radial mass flux

Figure 10. Azimuthally averaged radial mass flux, rescaled by r2sin⁡θr^2\sin\theta to account for the geometry, averaged in time between t=10 yrt=10\ \mathrm{yr} and t=40 yrt=40\ \mathrm{yr}. Blue denotes accretion and red decretion; the arrows show the direction of the poloidal mass flux. The vertical axis is δ=π/2−θ\delta=\pi/2-\theta, the angle to the equatorial plane. The solid black line marks the protostar surface and the dotted contour the disc surface, defined as the region where gravity is balanced by the centrifugal and pressure forces.

The result is not a disc that accretes as a whole. Beyond the disc outer edge, at r≳3r\gtrsim3 AU, the envelope falls in and feeds the system through the pseudo-disc, producing the broad in-flow that dominates the map at those radii. Inside the disc, however, between ∼0.05\sim0.05 and ∼1\sim1 AU, the flux breaks up into vertically stacked layers of alternating sign: accretion and decretion coexist at the same radius, at different heights above the mid-plane. Closer in, below ∼0.05\sim0.05 AU, the pattern becomes irregular, and the mid-plane just outside the protostar surface shows net decretion: material is leaving the protostar there, even though the protostar as a whole is gaining mass (Figure 9).

Accretion and decretion therefore coexist at the same radius, in different layers of the disc. This is not a detail of the flow: the two directions carry comparable fluxes, so a description in terms of a single radial accretion rate — the vertically integrated, one-dimensional α\alpha-disc picture ((Shakura & Sunyaev 1973; Pringle 1981)) — would report a net accretion rate for a region in which material is simultaneously moving both ways. Since a fluid element in a rotationally supported disc can only move radially by exchanging angular momentum, this layered pattern must reflect an equally layered angular momentum transport.

Choice of the averaging window

Equation (15) assumes a steady state, and the maps above are time averages, so we must first establish over which interval such a state holds. Figure 11 shows the amplitude of the three torques as a function of time and radius over the whole run.

Figure 11. Amplitude of the Reynolds (left), Maxwell (middle) and gravitational (right) torques as a function of time and radius, rescaled by r2r^2. Each term is first averaged in azimuth, and its absolute value is then averaged in the polar direction between θ=π/4\theta=\pi/4 and θ=3π/4\theta=3\pi/4, so that the maps measure the total transport activity at each radius.

All three share a strong transient during the first ∼5\sim5 yr, while the protostar and its immediate environment settle after the second collapse; it coincides with the peak of the magnetic energy of the protostar and with a strong decrease of its angular momentum near t∼5t\sim5 yr (Figure 9). After ∼10\sim10 yr the torques fluctuate about well-defined radial profiles without systematic drift, and they continue to do so until ∼40\sim40 yr. Around 43 yr the Reynolds and gravitational torques in the inner disc, between ∼0.01\sim0.01 and ∼0.03\sim0.03 AU, increase sharply, and the Maxwell torque strengthens in the same region: this is the accretion burst described in Sect. 3.5.

We restrict the analysis to t=10t=10–4040 yr, and all time averages below are taken over that window. This window is the interval over which the mass and angular momentum of the protostar evolve slowly and its accretion rate shows no secular trend (Figure 9), which is precisely the condition under which the time derivative neglected in Eq. (15) is small. During the burst that term is not negligible — the angular momentum of the protostar rises by some fifteen per cent in under a decade — and averaging across it would combine two different regimes. A weaker episode of gravitational activity is also visible in the inner disc around 25–30 yr, with a corresponding rise in the accretion rate. We have checked that neither this episode nor the choice of window conditions our results: repeating the analysis separately over t=10t = 10–25 and 25–40 yr gives qualitatively the same results — the same three transport regions and the same ranking of the stresses within each, as presented in Sect. 5 — so that the quasi-steady phase is well characterised by a single average.

Figure 11 also already shows the single most robust feature of the transport, which does not depend on any of the details discussed later: the Maxwell torque is active below ∼0.1\sim0.1 AU, essentially absent between ∼0.1\sim0.1 and ∼7\sim7 AU, and non-zero again at larger radii. The gap is the dead zone identified in Sect. 3, seen now in the transport itself rather than in the field structure; beyond it, in the envelope, the field is still ideal and the Maxwell torque reappears, as discussed in Sect. 5. The Reynolds torque, by contrast, is active at all radii, with maxima near the protostar surface and near the disc outer edge, and the gravitational torque is weaker but present throughout the disc.

Accretion is controlled by angular momentum transport

We first verify that the two sides of Eq. (15) agree. Figure 12 shows the total torque H+M+G\mathcal{H} + \mathcal{M} + \mathcal{G} and the advection term ⟨ρvp⟩⋅∇(s2Ω)\langle\rho\mathbf{v}_{p} \rangle\cdot\nabla(s^{2}\Omega) side by side, on the same scale.

Total torque and advection term

Figure 12. Total torque H+M+G\mathcal{H}+\mathcal{M}+\mathcal{G} (top) and advection term ⟨ρvp⟩⋅∇(s2Ω)\langle\rho\mathbf{v}_{p}\rangle\cdot\nabla(s^{2}\Omega) (bottom), the two sides of Eq. (15), averaged in time between t=10t=10 yr and t=40t=40 yr and rescaled by r2sin⁡θr^{2}\sin\theta. The arrows in the bottom panel show the direction of ∇(s2Ω)\nabla(s^{2}\Omega), onto which the poloidal mass flux is projected in the advection term, i.e. the direction in which the specific angular momentum increases. Contours are as in Figure 10.

The two maps agree closely in sign and in structure over most of the domain, and in particular they reproduce the same layered pattern between 0.05 and 1 AU and the same mid-plane feature just outside the protostar surface. The agreement is best in the inner disc, between ∼0.01\sim0.01 and ∼0.3\sim0.3 AU, where the orbital time is shorter than a year, so that the averaging window covers many orbits and fluctuations average out efficiently. Further out the agreement degrades, for the reason one would expect: near 1 AU the orbital time approaches the length of the averaging window itself, only a few orbits are sampled, and the steady-state assumption becomes correspondingly weaker. The region around 1 AU is also one where the total torque happens to be small, so that residual dynamical evolution is comparatively more visible. Inside the protostar (r≲0.01r \lesssim0.01 AU) the two maps are not expected to match at all, since that region is pressure- rather than centrifugally supported and Eq. (15) does not describe it.

Within those limits, the budget closes. This is a non-trivial check: the six torques of Eq. (16) are computed independently of the mass flux, from second-order correlations of the velocity, magnetic and gravitational fields, so their sum reproducing the advection map validates both the steady-state approximation in the inner disc and our evaluation of the individual terms.

The physical content of Figure 12 lies in the correspondence between the two maps: where the stresses remove angular momentum, the gas accretes; where they deposit it, the gas moves outwards. The correspondence holds layer by layer, so that the vertical structure of the mass flux is set by the vertical structure of the transport, as anticipated in Sect. 4.1.

The mid-plane region near 0.01 AU illustrates this most clearly. There the total torque is strongly positive: the stresses deposit angular momentum, the centrifugal support of the gas increases, and material is expelled from the protostar. The accreting layers, at the mid-plane further out and at the upper and lower disc surfaces, are correspondingly those in which angular momentum is extracted.

Transport mechanisms

§ 4 established that the accretion pattern is set by the angular momentum budget. We now decompose that budget into the six torques of Eq. (16) and ask which of them acts where.

Which stress dominates where

Figure 13 ranks the six torques by amplitude at each location. Rather than plotting six maps, we show the first, second and third most important contributions, using colour for the nature of the stress and shade for its direction; this makes both the identity of the dominant mechanism and the radial-versus-meridional character of the transport readable at once. Four regions emerge.

Figure 13. Relative importance of the six torques of Eq. (16), averaged in time between t=10t = 10 yr and t=40t = 40 yr. At each location, the six torques are ranked by absolute value: the top panel shows the largest one, the middle panel the second largest, and the bottom panel the third largest. In the middle and bottom panels, locations where that torque amounts to less than 5% of the sum of the absolute values of the six torques are shown in grey. Colour encodes the nature of the torque — blue for Reynolds H\mathcal{H}, green for Maxwell M\mathcal{M}, red for gravitational G\mathcal{G} — and shade its direction, light for the meridional component and dark for the radial one. Contours are as in Figure 10.

The envelope and the polar regions. — Outside the disc — beyond ∼3\sim3 AU in the pseudo-disc, and above the disc surface — the Maxwell torque dominates. The envelope is still in the ideal-MHD regime, where the field remains anchored to the infalling gas, and the toroidal component generated at these scales (Sect. 3) carries the associated torque. This is classical magnetic braking of the envelope, which is what delivers low-angular-momentum material to the disc in the first place.

The outer disc, 1–3 AU. — Within the disc the Maxwell torque disappears — this is the dead zone — and the Reynolds torque takes over, with radial and meridional components of comparable importance. This is the most massive and most weakly bound part of the disc, and also the part where the steady-state description is least secure (Sect. 4). The gravitational torque is the second contributor, almost purely radial.

The middle disc, 0.1–1 AU. — The Reynolds torque still dominates, but now overwhelmingly through its meridional component. Angular momentum in this region is not passed outwards from ring to ring, as a one-dimensional viscous model would have it, but carried vertically, and it accumulates at intermediate latitudes, near δ≃±π/8\delta\simeq\pm\pi/8. The centrifugal support rises there and material is expelled, while the mid-plane and the disc surfaces, which lose angular momentum, both accrete. This is the three-layered pattern of Figure 10: accretion at the mid-plane and at the surfaces, decretion in between. The gravitational torque is again the second contributor and again almost purely radial, so that the two act along different directions.

The inner disc, 0.01–0.1 AU. — Inside the recoupling radius the Maxwell torque dominates: the field has recoupled to the gas and is wound up into the strong toroidal component observed in Figure 6b. The Reynolds torque is not far behind as the second contributor, and both are mixed in direction, with comparable radial and meridional components. The transport here is therefore genuinely two-dimensional: angular momentum is carried outwards in radius and upwards towards the disc surface at the same time, which is what allows accretion to proceed along some layers while others decrete.

The gravitational torque is nowhere dominant, but it is a persistent second or third contributor across the whole disc, and it is never negligible, as anticipated in Sect. 3 from the disc mass.

What drives the Reynolds torque

The Reynolds torque dominates the transport over most of the disc, but the label covers two physically distinct processes. To separate them, we split the Reynolds flux into the contribution of the mean flow and that of the non-axisymmetric fluctuations,

⟨ρv δvϕ⟩=⟨v⟩⟨ρ δvϕ⟩+⟨ρ δv δvϕ⟩.(18)\left\langle\rho\mathbf{v}\,\delta v_{\phi}\right\rangle= \langle\mathbf{v}\rangle\left\langle\rho\,\delta v_{\phi}\right\rangle+ \left\langle\rho\,\delta\mathbf{v}\,\delta v_{\phi}\right\rangle. \tag*{(18)}

The first term is the transport of the angular momentum perturbation by the axisymmetric flow — in practice, by the mean meridional motion — while the second is the transport by the non-axisymmetric fluctuations themselves. Figure 14 shows the two terms of the meridional flux, the component that dominates the middle disc.

Figure 14. Decomposition of the meridional Reynolds flux, Eq. (18), at θ=0.4π\theta= 0.4\pi, as a function of time and radius. Left: the contribution of the axisymmetric flow, ⟨vθ⟩⟨ρ δvφ⟩\langle v_\theta\rangle\langle\rho\,\delta v_\varphi\rangle. Right: the contribution of the non-axisymmetric fluctuations, ⟨ρ δvθδvφ⟩\langle\rho\,\delta v_\theta\delta v_\varphi\rangle. Both are normalised by ⟨ρ⟩⟨vφ⟩2\langle\rho\rangle\langle v_\varphi\rangle^2 and shown on the same scale.

The answer is unambiguous: the non-axisymmetric term exceeds the mean-flow term by roughly two orders of magnitude throughout the region where the Reynolds torque dominates. The angular momentum is carried by the non-axisymmetric fluctuations, not by the mean meridional circulation. Both terms show the same pattern of outward-propagating features, but with very different amplitudes.

The non-axisymmetric fluctuations may, however, be either the large-scale spiral arms of Figure 8 or small-scale turbulence. To separate the two, we decompose the fluctuating term, for its meridional (i=θi = \theta) and radial (i=ri = r) components, as

⟨ρ δvi δvϕ⟩=⟨ρ⟩⟨δvi δvϕ⟩+⟨δρ δvi δvϕ⟩=∑mR1,m+R2,(19)\left\langle\rho\,\delta v_i\,\delta v_{\phi}\right\rangle= \langle\rho\rangle\left\langle\delta v_i\,\delta v_{\phi}\right\rangle+ \left\langle\delta\rho\,\delta v_i\,\delta v_{\phi}\right\rangle= \sum_m R_{1,m} + R_2, \tag*{(19)}

where R1,mR_{1,m} is the contribution of the azimuthal Fourier mode mm to ⟨ρ⟩⟨δvi δvφ⟩\langle\rho\rangle\langle\delta v_i\,\delta v_\varphi\rangle, obtained from the Fourier coefficients of δvi\delta v_i and δvφ\delta v_\varphi (Parseval’s theorem), and R2R_2 the contribution of the density fluctuations. Turbulence would spread the flux over many modes, whereas a large-scale spiral pattern concentrates it in the lowest ones. Figure 15 shows that, for both components, the m=2m = 2 mode carries most of the flux, typically between half and all of it, at almost all radii between 0.01 and ∼2\sim2 AU; the modes m≥10m \ge10 and the density term R2R_2 are negligible. Beyond ∼0.1\sim0.1 AU, m=2m = 2 also remains the dominant mode at almost all times for both components, within and outside the averaging window, whereas inside 0.1 AU it dominates the meridional flux most of the time, but the radial flux is often dominated by the m=4m = 4 mode (not shown). In the middle disc, where the Reynolds torque dominates the transport, the fluctuations are therefore carried by the m=2m = 2 spiral arms rather than by turbulence. The picture differs at both ends of the disc. In the inner disc, the higher modes carry a larger share of the flux, mostly in the radial component, where the m=4m = 4 mode contributes ∼20\sim20–30%30\% of the time-averaged flux between ∼0.015\sim0.015 and ∼0.3\sim0.3 AU: smaller-scale fluctuations also contribute to the transport there (Sect. 6.1). Beyond ∼2\sim2 AU, the higher modes become dominant in places, in agreement with the more disordered, turbulent structure of the outer disc seen in Figure 8. We note that, since our azimuthal domain spans φ∈[0,π]\varphi\in[0,\pi], only even modes are present, and m=2m = 2 is the largest-scale pattern the domain allows (Sect. 2.4).

Figure 15. Azimuthal mode decomposition of the fluctuating Reynolds flux, Eq. (19), for its meridional component ⟨ρ δvθδvφ⟩\langle\rho\,\delta v_\theta\delta v_\varphi\rangle (top) and its radial component ⟨ρ δvrδvφ⟩\langle\rho\,\delta v_r\delta v_\varphi\rangle (bottom), at θ=0.4π\theta= 0.4\pi and averaged in time between t=10t = 10 yr and t=40t = 40 yr. Each curve shows the signed share of one contribution — the azimuthal mode mm of R1,mR_{1,m}, or the density term R2R_2 — that is, the contribution divided by ∑m∣R‾1,m∣+∣R‾2∣\sum_m |\overline{R}_{1,m}|+|\overline{R}_2|, where the overline denotes the time average.

The mean-flow term is nonetheless not zero, and it is structured rather than noisy: it organises into coherent stripes drifting outwards, on a timescale much longer than the orbital one. This is the signature of the systematic radial velocity motion already visible in the poloidal velocity field of Figure 5. It contributes little to the angular momentum budget, but it is a genuine feature of the flow, and we discuss its origin in Sect. 6.3.

Transport efficiency and the parametrisation

The torques of Eq. (16) have so far been quoted in code units, which makes them difficult to compare with the disc literature, where transport efficiency is almost invariably expressed as a Shakura–Sunyaev α\alpha (Shakura & Sunyaev 1973). Writing the right-hand side of Eq. (15) as the divergence of a stress, its components are

FiφH=⟨ρ δvφvi⟩,FiφM=−14π⟨BφBi⟩,\mathcal{F}^{\mathcal{H}}_{i\varphi} = \left\langle\rho\,\delta v_\varphi v_i \right\rangle,\qquad\mathcal{F}^{\mathcal{M}}_{i\varphi} = -\frac{1}{4\pi}\left\langle B_\varphi B_i \right\rangle,
FiφG=14π⟨gφgi⟩,(20)\mathcal{F}^{\mathcal{G}}_{i\varphi} = \frac{1}{4\pi}\left\langle g_\varphi g_i \right\rangle, \tag*{(20)}

with i=ri = r or θ\theta, and we define the corresponding dimensionless efficiencies as

αiφX=FiφX⟨P⟩,X∈{H,M,G}.(21)\alpha^X_{i\varphi} = \frac{\mathcal{F}^X_{i\varphi}}{\langle P\rangle},\qquad X \in\{\mathcal{H},\mathcal{M},\mathcal{G}\}. \tag*{(21)}

Two choices in Eq. (21) deserve comment. First, ⟨P⟩\langle P\rangle is the local gas pressure, evaluated at the same position as the flux, rather than a mid-plane value. In a thin disc every point can be associated unambiguously with the mid-plane point directly beneath it; in a disc as thick as ours this association is no longer straightforward, and there is no natural choice of which mid-plane pressure to attach to a given point of the disc. The local pressure avoids that choice. Since it falls to the floor values of Sect. 2.6 in the polar cavities, where the ratio would diverge, we mask the maps outside the disc surface. Second, we normalise by the gas rather than the total pressure; the two differ appreciably only inside ∼0.1\sim0.1 AU, where βp\beta_{\mathrm{p}} approaches unity (Figure 7), and the choice changes α\alpha there by a factor of order unity.

Since Figure 16 is meant to represent the time-averaged torque maps of Figure 12 and Figure 13 in dimensionless form, α\alpha is computed as the ratio of the time-averaged flux to the time-averaged pressure, rather than as the time average of their ratio.

Figure 16. Dimensionless transport efficiencies αiφX\alpha^X_{i\varphi} defined by Eq. (21), averaged in time between t=10t = 10 yr and t=40t = 40 yr. Left column: the radial component i=ri = r, the one to which the standard α\alpha prescription corresponds. Right column: the meridional component i=θi = \theta, which has no counterpart in that prescription. From top to bottom: the Reynolds, Maxwell and gravitational stresses. The fluxes are normalised by the local gas pressure, and the maps are masked outside the disc surface. Contours are as in Figure 10.

For the radial Reynolds and Maxwell components, Eq. (21) reduces to the quantity usually denoted α\alpha, up to the distinction between a local and a vertically integrated normalisation. The meridional components have no counterpart in the standard prescription, which is the point of showing them alongside.

Figure 16 shows the six efficiencies over the disc. Two features stand out.

First, their magnitude. Inside ∼0.1\sim0.1 AU the Maxwell stress reaches αM∼10−2−10−1\alpha^{M}\sim10^{-2}-10^{-1}, concentrated in two lobes above and below the mid-plane at the protostar–disc interface, and falls to zero beyond that radius: the dead zone is as sharply delimited in efficiency as it was in the torque maps. Across the dead zone the Reynolds stress gives αH∼10−3−10−2\alpha^{H}\sim10^{-3}-10^{-2}, and the gravitational stress roughly an order of magnitude less, αG∼10−4−10−3\alpha^{G}\sim10^{-4}-10^{-3}.

Second, the meridional efficiencies cannot be neglected compared with the radial ones: αθϕH\alpha^{H}_{\theta\phi} reaches the same ∼10−2\sim10^{-2} as αrϕH\alpha^{H}_{r\phi} over much of the dead zone. This confirms, in dimensionless form, what the relative importance of the torques already showed (Figure 13): the vertical transport of angular momentum between the disc surfaces and the mid-plane is as significant as the radial transport, and cannot be neglected.

The maps above describe the transport locally. It is worth asking, finally, what a one-dimensional model would infer if it were confronted only with the accretion rate that our disc actually produces. Inverting the steady-state relation M˙=3πνΣ\dot{M}=3\pi\nu\Sigma with ν=αcs2/Ωeq\nu=\alpha c_{\mathrm{s}}^{2}/\Omega_{\mathrm{eq}} gives the equivalent viscosity parameter

αeq=M˙Ωeq3πΣcs2,(22)\alpha_{\mathrm{eq}}=\frac{\dot{M}\Omega_{\mathrm{eq}}}{3\pi\Sigma c_{\mathrm{s}}^{2}}, \tag*{(22)}

that is, the number a viscous model would need in order to reproduce the measured mass flux at each radius. Here Σ(r)\Sigma(r) is the density integrated over θ\theta at constant rr within the disc, csc_{\mathrm{s}} the mid-plane sound speed, and Ωeq\Omega_{\mathrm{eq}} the equilibrium angular velocity of Figure 4. We build the viscosity from Ωeq\Omega_{\mathrm{eq}} rather than from the Keplerian frequency because the disc carries an appreciable fraction of the mass of the system and rotates measurably below the Keplerian rate (Sect. 3); using ΩK\Omega_{\mathrm{K}} would misstate αeq\alpha_{\mathrm{eq}} in the outer disc, where the departure is largest. Because accretion and decretion coexist at the same radius (Sect. 4), we evaluate Eq. (22) separately for the inward and outward parts of the flux and for their difference.

Figure 17 shows the result, and the two panels say different things. Taken individually, the one-way flows are unremarkable: each corresponds to αacc,αdec∼10−3−10−2\alpha_{\mathrm{acc}},\alpha_{\mathrm{dec}}\sim10^{-3}-10^{-2} across the dead zone, in reasonable agreement with the local Reynolds efficiencies of Figure 16. The net flux, however, is the small residual of two much larger opposing flows, and between ∼0.09\sim0.09 and ∼0.35\sim0.35 AU that residual is negative, with αeq\alpha_{\mathrm{eq}} reaching ∼−10−3−10−2\sim-10^{-3}-10^{-2}. Over a third of the disc, a viscous model fitted to our mass flux would require a negative viscosity.

Figure 17. Top: accreting and decreting contributions to the mass flux through the spherical surface of radius rr, and their difference. At each radius M˙acc\dot{M}_{\mathrm{acc}} sums the cells in which the flux is inward and M˙dec\dot{M}_{\mathrm{dec}} those in which it is outward, both counted positively, so that their difference is the net inward flux. Only cells lying inside the disc are included—the dotted contour of Figure 10—so that the polar regions are excluded. Bottom: the equivalent viscosity parameter of Eq. (22) obtained from each. Both panels are averaged in time between t=10t=10 yr and t=40t=40 yr. Note the symmetric logarithmic scale: the net flux, and with it αeq\alpha_{\mathrm{eq}}, is negative between ∼0.09\sim0.09 and ∼0.35\sim0.35 AU.

The negative αeq\alpha_{\mathrm{eq}} has a simple explanation. A positive viscosity can only carry angular momentum outwards, and so cannot make it accumulate at a given radius. The vertical transport does: in the dead zone, αθϕH\alpha^{H}_{\theta\phi} is positive above the mid-plane and negative below it (Figure 16), which, with eθe_{\theta} pointing from north to south, means that angular momentum is carried from both disc surfaces towards the disc interior, where it accumulates. A one-dimensional model, having no vertical direction, can only mimic this accumulation with a negative viscosity.

What this leaves open is the origin of the stresses themselves, which we discuss in the next section.

Origin of the stresses

Section 5 identified where each stress acts, but not what sustains it. We now consider the origin of the two dominant ones: the magnetic stress in the inner disc (Sect. 6.1), and the spiral arms that drive the Reynolds stress in the dead zone (Sect. 6.2). We then turn to the mean vertical motion responsible for the coherent, mean-flow part of the Reynolds stress (Sect. 6.3).

The magneto-rotational instability in the inner disc

The Maxwell torque dominates the transport inside ∼0.1 AU\sim0.1\ \mathrm{AU}, and Sect. 5 showed that it acts radially and meridionally in comparable measure. Simple winding of a poloidal field by differential rotation would not produce this: it generates an ordered toroidal field and a predominantly radial stress. The mixed, disordered character of the stress, together with the turbulent field structure visible in Figure 6c, points instead to the magneto-rotational instability (MRI, [5, 19]).

The conditions for the MRI are met in this region. The rotation profile is differential (Figure 4), and the field is well coupled to the gas: this is precisely the region inside the recoupling radius, where the Elsasser number of Eq. (17) exceeds unity (Figure 7). The MRI also requires the field to be weak enough, βp≳1\beta_{\mathrm{p}} \gtrsim1. Its most unstable wavelength is

λMRI=2πvAΩ,(23)\lambda_{\mathrm{MRI}} = \frac{2\pi v_{\mathrm{A}}}{\Omega}, \tag*{(23)}

and since the disc scale height is H∼cs/ΩH \sim c_{\mathrm{s}}/\Omega, we have λMRI/H∼2πvA/cs∝βp−1/2\lambda_{\mathrm{MRI}}/H \sim2\pi v_{\mathrm{A}}/c_{\mathrm{s}} \propto\beta_{\mathrm{p}}^{-1/2}: requiring βp≳1\beta_{\mathrm{p}} \gtrsim1 is equivalent to requiring that λMRI\lambda_{\mathrm{MRI}} fits within the disc thickness. Finally, for the instability to be captured by our grid, λMRI\lambda_{\mathrm{MRI}} must be resolved by a sufficient number of cells.

Figure 18 shows that both conditions are satisfied over an appreciable part of the inner disc at t=5 yrt = 5\ \mathrm{yr}, with more than ten cells per wavelength in the innermost region. The direct evidence, however, is Figure 19: the azimuthal field at r=0.015 AUr = 0.015\ \mathrm{AU} reverses sign quasi-periodically and the reversals migrate away from the mid-plane, producing the butterfly pattern that is the classical signature of MRI-driven dynamo activity in a stratified disc. The radial and meridional components show the same pattern. We conclude that the MRI operates in the inner disc and is the most likely origin of the Maxwell torque there.

Figure 19. Butterfly diagram of the three magnetic field components at r=0.015 AUr = 0.015\ \mathrm{AU}, averaged in azimuth, as a function of time and of the angle δ\delta to the equatorial plane.

This has two consequences for the transport. First, MRI turbulence saturates the field rather than letting it grow without bound, which is why the field strength saturates, reaching 3 – 10 kG in the protostar region, instead of continuing to be amplified by the shear. This saturation is not set by equipartition: βp\beta_{\mathrm{p}} remains well above unity there (Figure 7), so the field is far from equipartition with the thermal pressure. Second, the same turbulence feeds velocity fluctuations, and therefore contributes to the Reynolds stress. The m=2m=2 spiral pattern still carries most of the Reynolds flux inside 0.1 AU, but the higher azimuthal modes contribute more there than in the middle disc, in particular to the radial flux, which is often dominated by the m=4m=4 mode (Figure 15), consistent with a turbulent contribution from the MRI.

Two caveats must be attached to this conclusion. The first is that the MRI is not numerically converged. Figure 18 shows that by t=20t=20 yr the region where the instability can be contained has grown, but the number of cells per wavelength has fallen below ten over most of it, and correspondingly the butterfly pattern weakens with time. Figure 20 makes the point directly: doubling the angular resolution makes the pattern appear earlier and raises its frequency, exactly as expected when the instability is better resolved. The MRI in our simulations is therefore under-resolved, and the amplitude of the Maxwell torque in the inner disc should be regarded as a lower limit rather than a converged measurement.

Figure 18. Resolution of the most unstable MRI mode in the inner disc, at t=5 yrt = 5\ \mathrm{yr} (left) and t=20 yrt = 20\ \mathrm{yr} (right). Colours show the number of cells per wavelength λMRI/Δr\lambda_{\mathrm{MRI}}/\Delta r, white corresponding to ten cells per wavelength. Regions where λMRI\lambda_{\mathrm{MRI}} exceeds the local disc height, and where the instability therefore cannot develop, are shown in grey.

Figure 20. Butterfly diagram of the azimuthal magnetic field at r=0.015 AUr = 0.015\ \mathrm{AU} in the 1MsunHLLD (top) and 1MsunHLLDx2 (bottom) simulations, over the interval covered by both runs.

The second is that the result depends on the Riemann solver. In the more diffusive HLL run, no butterfly pattern develops at the same resolution, and the field is instead amplified into an ordered toroidal configuration by the shear alone (Appendix B, Figure B1) — a difference already documented for MRI simulations by Flock et al. [15]. This is the main reason we adopt HLLD here. It also means that the quantitative partition between Maxwell and Reynolds stresses in the inner disc is resolution- and solver-dependent. In both runs, however, the Maxwell torque is the dominant contribution inside 0.1 AU: this follows from the recoupling of the field and its winding by the differential rotation, which are set by the non-ideal MHD and not by the Riemann solver.

Figure B1. Magneto-rotational instability in the 1MsunHLL run. Left: butterfly diagram of the azimuthally averaged toroidal field BϕB_\phi at r=0.015r = 0.015 AU, as a function of time and of the angle δ\delta to the equatorial plane, to be compared with Figure 20. Right: number of cells per wavelength of the most unstable MRI mode at t=5t = 5 and 2020 yr, as in Figure 18; regions where λMRI\lambda_{\mathrm{MRI}} exceeds the local disc height are shown in grey.

The spiral arms

Outside the recoupling radius, the field is decoupled and the transport is carried by the Reynolds stress, which Sect. 5 traced to the spiral arms. The question is then why the disc is spiral in the first place, and the natural candidate is gravitational instability — as anticipated in Sect. 3, where the disc was found to be an order of magnitude more massive than the protostar.

The standard diagnostic is the Toomre parameter Q=csκ/(πGΣ)Q=c_{\mathrm{s}}\kappa/(\pi G\Sigma) (Toomre 1964), with κ\kappa the epicyclic frequency and Σ\Sigma the surface density. Our disc is thick, so a vertically integrated criterion is not appropriate, and we use instead the three-dimensional form

Q3D=κ24πGρ(24)Q_{\mathrm{3D}}=\frac{\kappa^2}{4\pi G\rho} \tag*{(24)}

which for a thin Keplerian disc reduces to Q3D=π/8 QQ_{\mathrm{3D}}=\sqrt{\pi/8}\,Q and therefore yields a comparable stability threshold near unity. Figure 21 shows Q3DQ_{\mathrm{3D}} at the mid-plane as a function of time and radius, and the three regions of Sect. 5 reappear in it. Inside ∼0.03\sim0.03 AU, the disc starts marginally unstable and then

Figure 21. Three-dimensional Toomre parameter Q3D=κ2/(4πGρ)Q_{\mathrm{3D}}=\kappa^2/(4\pi G\rho) at the mid-plane, as a function of time and radius. White corresponds to Q3D=1Q_{\mathrm{3D}}=1. The white band near the disc outer edge marks the region where the epicyclic frequency squared is negative and Q3DQ_{\mathrm{3D}} is undefined.

stabilises, with Q3DQ_{\mathrm{3D}} rising well above unity after ∼7\sim7 yr as the protostar grows and comes to dominate the local epicyclic frequency. This matches the early, transient gravitational activity seen in the first years of Figure 11, and explains why the gravitational torque retreats to a secondary role in the inner disc once the system has settled.

Between ∼0.1\sim0.1 and ∼1\sim1 AU — the dead zone, where the Reynolds torque takes over — Q3DQ_{\mathrm{3D}} hovers close to unity throughout the simulation. The disc there is marginally stable: unstable enough to sustain the spiral structure of Figure 8, but not unstable enough to fragment, and indeed no fragmentation occurs in any of our runs. In discs with a full thermal evolution, such a marginal state is usually maintained by a balance between the heating of the gas in the spiral shocks and its radiative cooling (Laughlin et al. [23]; Cossins et al. [10]; Kratter & Lodato [22]). That mechanism is absent here. With a barotropic equation of state the gas does heat as it is compressed, its temperature rising monotonically with density, but only through that compression: there is no dissipative heating in the spiral shocks, and no radiative cooling. The pressure support that this compressional heating provides as the disc becomes denser is likely required so that Q3DQ_{\mathrm{3D}} settles close to unity.

Beyond ∼2\sim2 AU, Q3DQ_{\mathrm{3D}} drops to its lowest values. This is the outer edge, where the disc meets the infalling envelope, and where Figure 8 shows perturbations that are strong but no longer organised into clean spirals. It is also where the Reynolds torque acquires a radial component comparable to its meridional one (Figure 13). The picture is consistent: the most gravitationally unstable part of the disc is also the most dynamically active, and the least well described by a steady state.

Vertical motion in the disc

The decomposition of Sect. 5 left a residue: the mean-flow contribution to the Reynolds stress is two orders of magnitude smaller than the fluctuating one, but it is coherent rather than noisy, and it corresponds to the large meridional eddies already visible in Figure 5. Although it is not important for the angular momentum budget, a systematic vertical circulation matters for other reasons — dust settling and mixing in particular — and it is worth identifying. Figure 22 characterises this motion. The mean meridional velocity reaches a few percent of the sound speed and organises into structures that propagate outwards, slowly: their tracks in the (t,r)(t,r) plane are much shallower than the sound-speed characteristics drawn in black. The middle panel shows the ratio ϵ\epsilon of vertical to radial shear over the same region. Its

Figure 22. Characterisation of the mean vertical motion, as a function of time and radius. Top: azimuthally averaged, mass-weighted meridional velocity normalised by the sound speed, averaged between θ=0.4π\theta= 0.4\pi and θ=0.5π\theta= 0.5\pi. Middle: the local ratio ϵ=∂z(s2Ω)/∂s(s2Ω)\epsilon= \partial_z(s^2\Omega)/\partial_s(s^2\Omega) of vertical to radial shear, over the same angular range. Bottom: the azimuthal density perturbation at φ=0\varphi= 0 and θ=0.4π\theta= 0.4\pi, tracing the spiral pattern. Black lines are sound-speed characteristics, spaced by the orbital time at 1 AU.

spatial and temporal structure closely matches that of the vertical velocity, and the two are anti-correlated: the gas moves against the vertical shear. The amplitude of ϵ\epsilon is a few to ten percent.

This correlation seems to relate this motion to the vertical shear instability [37, 24], and the slow outward propagation is consistent with the recent analysis of Ogilvie et al. (2025). The motion is odd with respect to the mid-plane, i.e. a breathing mode; we find no corrugation mode. The bottom panel rules out the obvious alternative: the spiral pattern propagates along the sound-speed characteristics, on a timescale quite different from that of the vertical motion, so the arms are not driving it.

The origin of the vertical shear itself is less clear, and here our thermodynamic simplification becomes an obstacle. In a real disc, vertical shear is maintained baroclinically, by a radial temperature gradient at fixed height; a barotropic fluid has no baroclinic torque, so the shear in our simulations must be sustained by something else. The most likely candidate is the accretion of envelope material onto the disc surfaces near δ≈±π/4\delta\approx\pm\pi/4, visible in Figure 10: this material arrives with a specific angular momentum different from that of the disc at the same cylindrical radius, and depositing it at the disc surface establishes a vertical gradient of s2Ωs^2\Omega. The spiral arms, the other obvious candidate, are disfavoured by the timescale argument above. Surface accretion would be a mechanism specific to the embedded, still-accreting phase we are simulating, and one that has no counterpart in isolated disc models. Confirming it would require a dedicated study.

Discussion and conclusions

We first discuss the limitations of the present models (Sect. 7.1), and then summarise our conclusions (Sect. 7.2).

Limitations

Thermodynamics. — The barotropic equation of state is the most consequential simplification. It fixes the sound speed as a function of density alone, which affects the Toomre parameter (Sect. 6.2); it removes the baroclinic torque, which is the standard driver of the vertical shear (Sect. 6.3); and it excludes by construction the other thermally driven instabilities that can operate in discs, such as the convective overstability and the zombie vortex instability (see [24]). It also carries the rescaling of the critical densities described in Sect. 2.1.2, adopted to obtain a protostar of realistic size. Simulations including radiative transfer will be needed to establish which of our results survive a proper treatment of the energy budget.

Convergence. — As discussed in Sect. 6.1, the MRI is not resolved well enough for the inner-disc transport to be quantitatively converged, and the higher-resolution run differs measurably from the reference one. The partition between Maxwell and Reynolds stresses inside 0.1 AU should therefore be read as indicative.

The inner boundary. — The zero-gradient condition imposed on the tangential magnetic field at rinr_{\rm in} (Sect. 2.5) generates artificial currents, and hence some magnetic field amplification in the immediate vicinity of the boundary. We have verified that this does not affect our results by repeating the simulation with a strongly enhanced Ohmic resistivity in the innermost cells, which suppresses the artefact without changing either the field structure or the evolution of the protostar; the test is described in Appendix C.

The resistivity model. — The diffusivities are read from equilibrium tables computed for a fixed MRN grain distribution (Sect. 2.1.3), and the temperature used to evaluate them in the polar cavities is set by the floor of Sect. 2.6.3, which assumes an optically thin medium and is therefore optimistic. A run without that floor develops a measurably different magnetic evolution on timescales longer than our analysis window. This matters here because the entire structure of the transport hinges on a single quantity: the radius at which the field recouples to the gas, which separates the magnetically active inner disc from the dead zone. Anything that moves that radius moves the boundary between the two transport regimes. Dust evolution is one such effect: Borderies et al. [8] showed that the dynamical vaporisation of grains in the innermost regions can change the resistivities substantially with respect to the equilibrium tables used here, precisely in the range of temperatures and densities encountered near 0.1 AU.

The azimuthal domain. — Restricting the domain to φ∈[0,π]\varphi\in[0,\pi] (Sect. 2.4) suppresses all odd azimuthal modes. This was necessary to keep the protostar at the centre of the grid, but it excludes m=1m = 1 structures, which can contribute to gravitational torques in massive discs and mediate the displacement of the star from the centre of mass. The m=2m = 2 mode we observe is the dominant one for a self-gravitating disc, so we do not expect this to change our conclusions qualitatively, but the gravitational torque may be underestimated.

Integration time and the steady-state assumption. — The reference run covers ∼64\sim64 yr after the formation of the protostar, of which the analysis uses the thirty years between t=10t = 10 and 4040 yr. This is many orbital times in the inner disc, where the analysis is most secure, but only a few at the outer edge, which is why the steady-state description degrades there (Sect. 4). The quasi-steady phase is also interrupted by an accretion burst around 43 yr (Sect. 3.5), which we excluded from the analysis since the steady-state assumption does not hold during it. Whether the quasi-steady regime persists as the disc grows and the envelope depletes will require longer integrations.

The polar regions. — The density floors, the Alfvén speed limiter and the azimuthal grid coarsening are all active in the low-density cavities above the disc (Sect. 2.6). The polar ejections launched from these regions should not be taken at face value, and the magnetic braking of the envelope reported in Sect. 5, which partly operates in the same region, should be regarded as qualitative.

The Hall effect. — Our model includes Ohmic and ambipolar diffusion but not the Hall effect, which can be comparable to the other two non-ideal effects over part of the density range relevant here ([28] (Tsukamoto et al. 2023)). Including it is a numerical challenge: the Hall term is dispersive, introducing whistler waves whose phase speed increases as their wavelength decreases, which severely restricts the time step. In IDEFIX, moreover, it cannot be integrated with the super-time-stepping scheme used for the Ohmic and ambipolar terms (Sect. 2.1.4), so that this time step restriction applies in full, and it can only be used with the HLL solver (Lesur et al. [26], which suppresses the MRI at our resolution (Sect. 6.1)). It would also complicate the analysis of the effects studied here: since the Hall effect depends on the orientation of the magnetic field with respect to the rotation axis, the evolution differs depending on whether the two are aligned or anti-aligned ((Wurster & Li 2018; Tsukamoto et al. 2023)). It may nonetheless modify the extent of the dead zone and the transport within it (Lesur et al. [25]), and should be included in future studies.

Conclusions

We have performed three-dimensional non-ideal MHD simulations of the collapse of a magnetised, rotating dense core, followed continuously from the cloud scale down to the formation of the protostar and then for ∼64\sim64 yr beyond it, and we have used them to analyse how angular momentum is transported in the resulting protostar–disc system. The analysis rests on a decomposition of the azimuthally averaged angular momentum budget into Reynolds, Maxwell and gravitational torques, each split into a radial and a meridional component, evaluated in spherical geometry so as not to presuppose a thin disc. Our main results are the following.

  1. During the quasi-steady phase that follows its formation, the angular momentum of the protostar stays nearly constant while its mass slowly increases, so that its specific angular momentum decreases: the angular momentum brought in by accretion is extracted from the protostar. This phase is interrupted at ∼43\sim43 yr by an accretion burst, during which the protostar gains nearly a tenth of its mass. The results that follow describe the quiescent phase, and are robust across it.

  1. Accretion through the disc is vertically layered. Between ∼0.05\sim0.05 and ∼1\sim1 AU, accreting and decreting layers coexist at the same cylindrical radius at different heights, and material is expelled from the mid-plane just outside the protostar even as the protostar gains mass overall.

  1. Angular momentum transport controls the accretion pattern: layers from which the stresses remove angular momentum accrete, layers into which they deposit it decrete. Since the meridional transport is comparable to the radial one, this structure is produced by a vertical redistribution of angular momentum that a vertically integrated, one-dimensional description cannot represent.

  1. As a consequence, a one-dimensional viscous model fitted to the net mass flux of the disc would require a negative equivalent α\alpha between ∼0.09\sim0.09 and ∼0.35\sim0.35 AU: the vertical transport makes angular momentum accumulate in the disc interior, which a positive viscosity cannot do.

  1. The transport separates into three regimes, delimited by the two radii at which the magnetic field changes its relationship to the gas. In the envelope, where the field is coupled to the gas and strong, magnetic braking removes angular momentum from the infalling material. Inside ∼0.1\sim0.1 AU, where the field recouples to the gas (Λ>1\Lambda> 1) and is wound into a strong toroidal component, the Maxwell torque dominates, with a comparable Reynolds contribution. Between the two, in the ambipolar dead zone (Λ≪1\Lambda\ll1) that makes up the bulk of the disc, transport is purely hydrodynamical and gravitational; this dead zone itself divides into a middle disc (0.1 – 1 AU), where the transport is predominantly meridional, and an outer disc (1 – 3 AU), where radial and meridional transport are comparable.

  1. In the dead zone the Reynolds torque is driven by the non-axisymmetric fluctuations rather than by the mean flow, the former exceeding the latter by about two orders of magnitude. A decomposition into azimuthal modes shows that these fluctuations are dominated by the m=2m = 2 mode: the agents are the spiral arms of a disc that is an order of magnitude more massive than the protostar and marginally gravitationally unstable, with Q3D∼1Q_{\mathrm{3D}} \sim1 over precisely the region where the Reynolds torque dominates.

  1. In the inner disc, the magneto-rotational instability operates, as shown by the butterfly pattern of the azimuthally averaged field. It both saturates the field, which reaches 3–10 kG in the protostar region instead of being amplified indefinitely by the shear, and feeds the velocity fluctuations that make the Reynolds torque comparable to the Maxwell torque there. The instability is, however, not numerically converged.

  1. A systematic mean vertical motion is present in the disc, anti-correlated with the vertical shear and propagating outwards well below the sound speed, with the properties of a vertical shear instability breathing mode. It contributes negligibly to the angular momentum budget, but may matter for dust settling and mixing.

The broader implication is that no single mechanism transports angular momentum in a young protostar–disc system. The magnetic field controls the two ends of the problem — the envelope and the innermost disc — while the bulk of the disc, magnetically decoupled by ambipolar diffusion, relies on hydrodynamical and gravitational stresses. The boundaries between these regimes are not set by the disc structure but by the microphysics of the gas–dust mixture, through the radius at which the field recouples to the gas. Improving the resistivity model, and in particular accounting for the dynamical vaporisation of dust grains (Borderies et al. [8]), is therefore not a refinement of a detail: it determines where one transport regime gives way to another.

These results also bear on how young discs are modelled. One-dimensional viscous models describe the transport by a single α\alpha parameter (Shakura & Sunyaev 1973); the layered accretion and the negative equivalent α\alpha found here show that such a description cannot capture the transport in a young, thick and embedded disc, in which the vertical redistribution of angular momentum matters as much as the radial one.

The natural extensions of this work follow from the limitations of Sect. 7.1: a proper treatment of the energy budget including radiative transfer, a higher resolution in the inner disc, and a full azimuthal domain. Finally, our analysis has taken the magnetic field of the inner disc as given. The mechanism that amplifies it and sustains its poloidal component against dissipation, and hence the Maxwell torque itself, will be the subject of a forthcoming paper (Borderies et al., in prep.).

This work has received funding from the French Agence Nationale de la Recherche (ANR) through the projects PROMETHEE (ANR-22-CE31-0020) and DISKBUILD (ANR-20-CE49-0006). This work was granted access to the HPC resources of IDRIS (Jean Zay) under the allocation AD010417023 made by GENCI. We gratefully acknowledge the support of the Centre Blaise Pascal’s IT test platform at ENS de Lyon (Lyon, France) for GPU computing facilities. The platform operates the SIDUS solution (Quemener & Corvellec 2013) developed by Emmanuel Quemener.

DATA AVAILABILITY

The simulation data and the analysis scripts underlying this article will be shared on reasonable request to the corresponding author. The Idefix code is publicly available at https://github.com/idefix-code/idefix.

Restart procedure

Radial restart

This appendix details the restarts summarised in Sect. 2.4, in which the extent of the domain — but not its relative resolution — is changed.

The hyperbolic time step scales with the innermost cell size, and hence with rinr_{\mathrm{in}}. Starting the simulation with rin=10−3r_{\mathrm{in}} = 10^{-3} AU would impose the protostellar time step from the very beginning, long before any structure exists at that scale. We therefore start with rin=1r_{\mathrm{in}} = 1 AU and reduce it by one decade at a time, following the characteristic scale of the collapse, down to 10−310^{-3} AU.

Two criteria decide when the inner radius can be reduced. First, the density near rinr_{\mathrm{in}} must be nearly uniform and close to the mean density ρin\rho_{\mathrm{in}} inside rinr_{\mathrm{in}}, which is obtained by integrating the mass flux through the inner boundary; a significant density contrast means that the collapse is already developing at that scale. Second, the Jeans length associated with ρin\rho_{\mathrm{in}} must remain larger than rinr_{\mathrm{in}}, so that no unresolved collapsing structure is hidden inside the inner boundary. In practice both criteria are met with a wide margin at each restart, the density contrast staying at the percent level. Three reductions then fall naturally at the successive stages of the collapse: the first is performed as the first Larson core is about to form, the second at the onset of the second collapse, and the third, which brings the grid to its final inner radius of 10−310^{-3} AU, as the second Larson core is about to form.

The solution must then be extrapolated inwards. The density is set to ρin\rho_{\mathrm{in}} in the new cells, and the velocity is extrapolated linearly so that it vanishes at the centre:

ρextra(r)=ρin,vextra(r)=v(rin,old,θ,φ) rrin,old.(25)\rho_{\mathrm{extra}}(r) = \rho_{\mathrm{in}}, \qquad v_{\mathrm{extra}}(r) = v(r_{\mathrm{in,old}},\theta,\varphi)\,\frac{r}{r_{\mathrm{in,old}}}. \tag*{(25)}

The magnetic field is more delicate, since the solenoidal constraint must be preserved. We extrapolate the vector potential instead, requiring BθB_\theta and BφB_\varphi to remain radially uniform and continuous inside rin,oldr_{\mathrm{in,old}}. This construction leaves an unphysical Br∝r−2B_r \propto r^{-2} singularity at the centre, which we remove by adopting

{Ar,new=Ar,old(rin,old,θ,φ)rrin,old,Aθ,new=aθrrin,old,Aφ,new=aφrrin,old,(26)\begin{cases} A_{r,\mathrm{new}} = A_{r,\mathrm{old}}(r_{\mathrm{in,old}},\theta,\varphi)\dfrac{r}{r_{\mathrm{in,old}}},\\ A_{\theta,\mathrm{new}} = a_\theta\dfrac{r}{r_{\mathrm{in,old}}}, \qquad A_{\varphi,\mathrm{new}} = a_\varphi\dfrac{r}{r_{\mathrm{in,old}}}, \tag*{(26)} \end{cases}

for r<rin,oldr < r_{\mathrm{in,old}}, and Aθ,new=Aθ,old−bθrin,old/rA_{\theta,\mathrm{new}} = A_{\theta,\mathrm{old}} - b_\theta r_{\mathrm{in,old}}/r (and similarly for AφA_\varphi) outside, with

{aθ=12∂r(rAθ)(rin,old,θ,φ),bθ=Aθ(rin,old,θ,φ)−aθ,aφ=12∂r(rAφ)(rin,old,θ,φ),bφ=Aφ(rin,old,θ,φ)−aφ.(27)\begin{cases} a_\theta= \frac{1}{2}\partial_r(rA_\theta)(r_{\mathrm{in,old}},\theta,\varphi), \qquad b_\theta= A_\theta(r_{\mathrm{in,old}},\theta,\varphi)-a_\theta,\\ a_\varphi= \frac{1}{2}\partial_r(rA_\varphi)(r_{\mathrm{in,old}},\theta,\varphi), \qquad b_\varphi= A_\varphi(r_{\mathrm{in,old}},\theta,\varphi)-a_\varphi. \tag*{(27)} \end{cases}

With this choice BθB_\theta and BφB_\varphi are preserved throughout the grid and BrB_r is not amplified near the centre; the residual modification of BrB_r decreases as r−2r^{-2} and leaves the inner field structure unchanged (Figure A1).

Figure A1. Radial magnetic field during a radial restart, from rin,old=10−2 AUr_{\mathrm{in,old}} = 10^{-2}\ \mathrm{AU} (vertical dashed line) to rin,new=10−3 AUr_{\mathrm{in,new}} = 10^{-3}\ \mathrm{AU}. Colours show ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|, the angle to the equatorial plane. Left: field before the restart. Middle: the continuity-preserving extrapolation, which amplifies BrB_r by several orders of magnitude near the centre through the singular bθb_\theta and bϕb_\phi terms. Right: the extrapolation adopted here, Eq. (26), which removes the singularity while preserving the inner field structure.

Polar restart

A spherical grid is only efficient if the protostar forms at its centre. The first Larson core ([7]) is turbulent and accretes asymmetrically ([9, 4]), and the non-symmetric structures developing during the first-core phase can displace the centre of mass through round-off errors and asymmetric interactions with the boundaries, and produce an off-centre protostar — which not only wastes the inner resolution but also disturbs the velocity and magnetic fields near the inner boundary. We therefore run the collapse and the first-core evolution on a half-domain, θ∈[0,π/2]\theta\in[0,\pi/2] with Nθ=64N_\theta= 64 cells, which enforces equatorial symmetry, and restart on the full sphere just before the second collapse. The lower hemisphere is filled by symmetry,

{ρ(r,π−θ,φ)=ρ(r,θ,φ),Xr(r,π−θ,φ)=Xr(r,θ,φ),Xθ(r,π−θ,φ)=−Xθ(r,θ,φ),Xφ(r,π−θ,φ)=Xφ(r,θ,φ),(28)\begin{cases} \rho(r,\pi-\theta,\varphi) = \rho(r,\theta,\varphi),\\ X_r(r,\pi-\theta,\varphi) = X_r(r,\theta,\varphi),\\ X_\theta(r,\pi-\theta,\varphi) = -X_\theta(r,\theta,\varphi),\\ X_\varphi(r,\pi-\theta,\varphi) = X_\varphi(r,\theta,\varphi), \tag*{(28)} \end{cases}

with X=vX = v or AA. Note that the symmetry relations are those of a vector, not of the pseudo-vector BB, precisely because the field is evolved through its potential.

Azimuthal restart

Three-dimensional simulations are considerably more expensive, and no non-axisymmetric structure is expected to play a significant role during the first collapse, such structures developing once the first core and its disc have formed ([9, 4]). We therefore compute that stage in axisymmetry and restart on a three-dimensional grid shortly before the formation of the first Larson core. The axisymmetric fields are simply copied in φ\varphi, with a density perturbation added to seed non-axisymmetric structures:

ρ3D(r,θ,φ)=ρ2D(r,θ)[1+ϵsin⁡(2φ)],(29)\rho_{\mathrm{3D}}(r,\theta,\varphi) = \rho_{\mathrm{2D}}(r,\theta)\left[1+\epsilon\sin(2\varphi)\right], \tag*{(29)}

the velocity components being copied unchanged. We force an m=2m = 2 mode, the most unstable one [12] (Tohline et al. 1985), with ϵ\epsilon between 10−210^{-2} and 10−110^{-1}; the precise value only sets the time needed for the structures to develop and does not affect the results. For the vector potential, only AϕA_\phi is defined in two dimensions; we use the gauge freedom to set Ar=0A_r = 0 and obtain AθA_\theta by integrating 1r∂r(rAθ)=Bϕ\frac{1}{r}\partial_r(rA_\theta) = B_\phi.

We initially used a full azimuthal domain, ϕ∈[0,2π]\phi\in[0, 2\pi], but found that an m=1m = 1 mode systematically develops and ejects the protostar from the centre of the grid, even when the seed perturbation contains no odd mode. Since the centre of mass itself becomes off-centre when this happens, we attribute it to the interaction of the protostar with the inner boundary: once a small m=1m = 1 mode appears, the net force exerted by the inner boundary no longer vanishes and amplifies it. This may be related to the reflex instability described by Crida et al. [11], in which the reflex motion of the star excites the eccentricity of the disc, whose m=1m = 1 density perturbation in turn pulls the star further away; confirming this would require a dedicated study. We therefore restrict the domain to ϕ∈[0,π]\phi\in[0, \pi], which suppresses all odd modes and keeps the protostar centred, at the cost of filtering out m=1m = 1 physics.

Increased-resolution restart

Finally, a restart at higher resolution is used for the convergence run of Table 2. Each parent cell is subdivided by an integer factor in each refined direction (here θ\theta and ϕ\phi); cell-centred variables are transferred by zeroth-order injection, and the magnetic field is refined through its vector potential so that the solenoidal constraint is preserved exactly.

The HLL Riemann solver

All the results presented in the main text use the HLLD Riemann solver [34]. For comparison, we also ran the reference setup with the more diffusive HLL solver [18, 13], all other parameters being identical (Table 2). This appendix collects the differences.

The essential one is the fate of the magneto-rotational instability in the inner disc. Figure B1 shows that, at our resolution, the numerical dissipation of the HLL solver is large enough to suppress it. The region in which the most unstable mode fits within the disc is thinner than with HLLD (Figure 18), and the mode is resolved by fewer than ten cells over most of it. No butterfly pattern develops: the toroidal field keeps the ordered configuration that the HLLD run displays before the onset of the instability (Figure 20), negative above the mid-plane and positive below it. It changes sign only once over the run, in the northern hemisphere between ∼20\sim20 and ∼35\sim35 yr, without the quasi-periodic, migrating reversals of the HLLD run. Without MRI turbulence to saturate it, the field is also stronger than in the HLLD run, by a factor of ∼2\sim2 overall, although of the same order of magnitude.

The consequences for the transport are shown in Figure B2. Inside ∼0.1\sim0.1 AU, the Maxwell torque dominates almost the whole inner disc, whereas in the HLLD run it shares this region with the Reynolds torque (Figure 13). This is consistent with the contribution of MRI turbulence to the velocity fluctuations in the HLLD run (Sect. 6.1). Outside the recoupling radius, where the field is decoupled and the transport is hydrodynamical and gravitational, the two runs show the same regions: the Reynolds torque dominates the disc, predominantly through its meridional component in the middle disc, and the Maxwell torque dominates the envelope.

Figure B2. Dominant torque at each location in the 1MsunHLL run, averaged in time between t=10t = 10 yr and t=40t = 40 yr, as in the top panel of Figure 13.

The inner boundary condition

The magnetic boundary condition at the inner radius imposes a vanishing gradient on BθB_\theta and BϕB_\phi (Sect. 2.5). This is not a physical condition: it forces currents to close through the boundary, and those artificial currents amplify the field in the innermost cells. Since the Maxwell torque dominates the transport in exactly this part of the domain (Sect. 5), it is necessary to establish that the amplification is confined to the boundary layer and does not contaminate the inner disc.

The cleanest way to remove the artefact without altering the physics elsewhere is to decouple the field from the gas in the immediate vicinity of the boundary, so that no current can be sustained there. We therefore repeated the simulation with an Ohmic resistivity enhanced close to rinr_{\mathrm{in}},

ηΩ=12(1−tanh⁡(r−rΩδrΩ))ηΩ,boost+ηΩ,0,(30)\eta_{\Omega} = \frac{1}{2}\left(1-\tanh\left(\frac{r-r_{\Omega}}{\delta r_{\Omega}}\right)\right)\eta_{\Omega,\mathrm{boost}}+\eta_{\Omega,0}, \tag*{(30)}

where ηΩ,0\eta_{\Omega,0} is the resistivity read from the tables of Marchand et al. [28], rΩ=3rinr_{\Omega}=3r_{\mathrm{in}} is the transition radius and δrΩ=rΩ/30\delta r_{\Omega}=r_{\Omega}/30 its width, the tanh ensuring a smooth transition. We use ηΩ,boost=1010ηΩ,0\eta_{\Omega,\mathrm{boost}}=10^{10}\eta_{\Omega,0}. Since the physical resistivity is very small at the centre of the protostar, this amounts to an essentially complete decoupling over the first few cells.

This experiment was carried out with the HLL solver rather than with HLLD. The artefact it addresses is a property of the boundary condition and of the innermost cells, not of the Riemann solver: the magnetic structure at the protostar–disc interface is qualitatively the same in the two runs, with a dominant toroidal component of opposite sign on either side of the mid-plane and a comparable field strength, and both show the same amplification against the boundary. The principal difference between the solvers, the presence of the MRI (Sect. 6.1), develops at radii several times larger than the region in which the resistivity is enhanced here, and so does not bear on what is being tested. We have therefore not repeated the experiment with HLLD.

Figures C1 and C2 compare the two runs. The enhanced resistivity does what it was designed to do: in the B−ρB-\rho relation, the narrow spike at the highest densities — the signature of the field being amplified against the boundary — disappears. Everything else is unchanged. The field retains the same topology at the protostellar scale, with the toroidal component dominant and antisymmetric about the mid-plane, the same poloidal structure, and the same field strength in the inner disc. The evolution of the protostar mass, angular momentum and magnetic energy is likewise indistinguishable from that of the 1MsunHLL run without enhanced resistivity over the interval covered by the test.

We conclude that the inner boundary produces a genuine but strictly local artefact, restricted to a few cells, and that the transport results reported in Sects. 4 and 5 are not affected by it.

This paper was built using the Open Journal of Astrophysics \LaTeX{} template. The OJA is a journal which provides fast and easy peer review for new papers in the astro-ph section of the arXiv, making the reviewing process simpler for authors and referees alike. Learn more at http://astro.theoj.org.

Figure C1. Magnetic field structure at the protostellar scale, without (left) and with (right) the enhanced Ohmic resistivity of Eq. (30). Colours show the azimuthally averaged ratio ⟨Bφ⟩/⟨B⟩\langle B_{\varphi}\rangle/ \langle B\rangle and white lines the poloidal field. The black contour marks the protostar surface.

Figure C2. Azimuthally averaged magnetic field strength as a function of density, without (left) and with (right) the enhanced Ohmic resistivity. Colours show ∣δ∣=∣π/2−θ∣|\delta| = |\pi/2 - \theta|. The narrow spike reaching ∼2×104\sim2 \times10^{4} G at the highest densities in the left panel is the boundary artefact; it is absent on the right.

References

  1. [1]Ahmad, A., González, M., Hennebelle, P., & Commerçon, B. 2023, A&A, 680, A23
  2. [2]Ahmad, A., González, M., Hennebelle, P., & Commerçon, B. 2024, A&A, 687, A90
  3. [3]Ahmad, A., González, M., Hennebelle, P., Lebreuilly, U., & Commerçon, B. 2025, A&A, 696, A238
  4. [4]Ahmad, A. A., Commerçon, B., Lynch, E., et al. 2026, Astronomy and Astrophysics, 707, A97
  5. [5]Balbus, S. A. & Hawley, J. F. 1991, The Astrophysical Journal, 376, 214
  6. [6]Bhandare, A., Kuiper, R., Henning, T., et al. 2020, A&A, 638, A86
  7. [7]Bhandare, A., Kuiper, R., Henning, T., et al. 2018, Astronomy and Astrophysics, 618, A95
  8. [8]Borderies, A., Commerçon, B., & Bourdon, B. 2025, A&A, 694, A89
  9. [9]Commerçon, B., Lovascio, F., Lynch, E., & Ragusa, E. 2024, Astronomy and Astrophysics, 689, L9
  10. [10]Cossins, P., Lodato, G., & Clarke, C. J. 2009, Monthly Notices of the Royal Astronomical Society, 393, 1157
  11. [11]Crida, A., Baruteau, C., Gonzalez, J.-F., et al. 2025, The Open Journal of Astrophysics, 8, 115
  12. [12]Durisen, R. H., Gingold, R. A., Tohline, J. E., & Boss, A. P. 1986, The Astrophysical Journal, 305, 281
  13. [13]Einfeldt, B., Roe, P. L., Munz, C. D., & Sjogreen, B. 1991, Journal of Computational Physics, 92, 273
  14. [14]Evans, C. R. & Hawley, J. F. 1988, The Astrophysical Journal, 332, 659
  15. [15]Flock, M., Dzyurkevich, N., Klahr, H., & Mignone, A. 2010, Astronomy and Astrophysics, 516, A26
  16. [16]Fromang, S. 2013, in EAS Publications Series, Vol. 62 (eprint: arXiv:1305.3416; EDP), 95–142DOI
  17. [17]Galli, D. & Shu, F. H. 1993, The Astrophysical Journal, 417, 243
  18. [18]Harten, A., Lax, P. D., & van Leer, B. 1983, SIAM Rev., 25, 35
  19. [19]Hawley, J. F., Gammie, C. F., & Balbus, S. A. 1995, The Astrophysical Journal, 440, 742
  20. [20]Hennebelle, P. 2013, in EAS Publications Series, Vol. 62 (EDP), 67–94
  21. [21]Hennebelle, P. & Fromang, S. 2008, A&A, 477, 9
  22. [22]Kratter, K. & Lodato, G. 2016, Annual Review of Astronomy and Astrophysics, 54, 271
  23. [23]Laughlin, G., Korchagin, V., & Adams, F. C. 1997, The Astrophysical Journal, 477, 410
  24. [24]Lesur, G., Flock, M., Ercolano, B., et al. 2023a, Hydro-, Magnetohydro-, and Dust-Gas Dynamics of Protoplanetary Disks
  25. [25]Lesur, G., Kunz, M. W., & Fromang, S. 2014, Astronomy and Astrophysics, 566, A56
  26. [26]Lesur, G. R. J., Baghdadi, S., Wafflard-Fernandez, G., et al. 2023b, A&A, 677, A9
  27. [27]Machida, M. N. & Basu, S. 2019, The Astrophysical Journal, 876, 149
  28. [28]Marchand, P., Masson, J., Chabrier, G., et al. 2016, A&A, 592, A18
  29. [29]Maury, A. J., André, Ph., Testi, L., et al. 2019, Astronomy and Astrophysics, 621, A76
  30. [30]Mauxion, J. 2024, These de doctorat, Université Grenoble Alpes
  31. [31]Mauxion, J., Lesur, G., & Maret, S. 2024, Astronomy and Astrophysics, 686, A253
  32. [32]Mellon, R. R. & Li, Z.-Y. 2008, The Astrophysical Journal, 681, 1356
  33. [33]Meyer, C. D., Balsara, D. S., & Aslam, T. D. 2014, Journal of Computational Physics, 257, 594
  34. [34]Miyoshi, T. & Kusano, K. 2005, Journal of Computational Physics, 208, 315
  35. [35]Mouschovias, T. Ch. & Paleologou, E. V. 1980, The Astrophysical Journal, 237, 877
  36. [36]Mouschovias, T. Ch. & Spitzer, Jr., L. 1976, ApJ, 210, 326
  37. [37]Nelson, R. P., Gressel, O., & Umurhan, O. M. 2013, Monthly Notices of the Royal Astronomical Society, 435, 2610

Paper details

vlnc:W_8g8232f2
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

downloads · 1

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents