Global simulations of accretion flows onto perturbers embedded in magnetized disks-I. MRI and jet formation in ideal MHD

We present the highest resolution global MHD simulations to date of gas flow around a low mass a perturber with mass ratio q ∈ [10−4, 10−3], embedded in an accretion disk around a massive central object. We find that gas flow onto the secondary self-consistently forms a turbulent, magnetized mini-accretion disk. The mini-accretion disk sustains a large-scale magnetic field generated by the dynamo effect of the MRI and the accretion flow into the perturber. Simultaneously, a bipolar, collimated, magnetized outflow is launched, extending beyond the perturber’s Hill sphere. The bipolar outflows are driven by the combined action of magnetic pressure, in the innermost regions of the mini-accretion disk, and the magnetocentrifugal acceleration of gas, which may attain speeds comparable to the escape velocity from the massive central object. Our results establish an important conceptual connection in accretion disk physics across a wide range of astrophysical systems—from mini-accretion disks to circumstellar and black hole accretion disks—by demonstrating that no fine-tuning is required for small-scale disks to naturally enter an outflow-launching regime. Beyond identifying the physical mechanism responsible for launching small-scale outflows, our framework lays the groundwork for developing more sophisticated physical models of mini-accretion disks around embedded low-mass perturbers.

Improvements to AI Systems:

1. Multiscale Magnetohydrodynamic (MHD) Emulation

• Improvement: Train a neural operator or transformer-based surrogate to predict the emergence of MRI-driven dynamos and bipolar outflows from local disk parameters (e.g., mass ratio, magnetic Prandtl number, initial field topology) without running full 3D global simulations.

• Capability: The AI can instantly forecast whether a given embedded perturber will launch outflows, enabling rapid parameter sweeps for protoplanetary or AGN disk models.

2. Self-Consistent Mini-Disk Boundary Condition Generator

• Improvement: Use a generative model (e.g., diffusion or GAN) conditioned on the central object’s accretion rate and perturber mass to produce realistic turbulent, magnetized inflow/outflow boundary conditions for sub-grid models in larger simulations.

• Capability: The AI can replace ad-hoc sink particle prescriptions with physically consistent mini-disk states, improving fidelity of planet-disk or black hole binary simulations.

3. Outflow Launching Regime Classifier

• Improvement: Build a classifier that takes local dimensionless quantities (e.g., plasma beta, magnetic field pitch angle, accretion Mach number) and predicts whether the system is in the outflow-launching regime identified by the paper, using the simulation data as ground truth.

• Capability: The AI can flag regions in large-scale disk simulations where mini-disks will spontaneously launch jets, allowing adaptive mesh refinement to focus computational resources there.

4. Physics-Informed Reduced-Order Model for Magnetic Dynamo Saturation

• Improvement: Develop a reduced-order model (e.g., a recurrent neural network or Koopman operator) that learns the time evolution of the large-scale magnetic field in the mini-disk from the paper’s high-resolution data, capturing dynamo cycles and saturation amplitudes.

• Capability: The AI can predict long-term magnetic field evolution in embedded disks at a fraction of the computational cost, enabling studies of disk–perturber interactions over viscous timescales.

5. Transferable Accretion Disk Physics Encoder

• Improvement: Pre-train a self-supervised encoder on the paper’s simulation outputs (density, velocity, magnetic field, current density) to learn a latent representation of universal accretion disk dynamics (mini, circumstellar, black hole).

• Capability: The AI can transfer this latent knowledge to unseen disk regimes (e.g., different mass ratios or equation of state) for zero-shot prediction of outflow properties, accelerating discovery in astrophysics.

6. Automated Detection of Magnetocentrifugal Acceleration Signatures

• Improvement: Train a computer vision model (e.g., 3D CNN) on the simulation’s velocity and magnetic field data to identify regions where gas is accelerated to escape velocity via the magnetocentrifugal mechanism.

• Capability: The AI can automatically locate and quantify outflow-driving zones in any new MHD simulation, replacing manual analysis and enabling large-scale comparative studies.

What the Improved AI System Can Do:

• Run real-time, high-fidelity predictions of mini-accretion disk behavior for arbitrary perturber masses in protoplanetary or AGN disks.

• Generate physically consistent sub-grid models for planet formation and binary black hole merger codes.

• Automatically classify and extract outflow properties from massive simulation archives, enabling statistical studies of jet launching across all disk scales.

• Provide a transferable foundation for modeling magnetic dynamos and outflows in systems ranging from circumplanetary disks to supermassive black hole accretion.