Lagrangian versus Eulerian Methods for Toroidally-Magnetized Isothermal Disks

This paper demonstrates that Lagrangian methods reproduce high-resolution Eulerian results for toroidally-magnetized isothermal disks and exhibit superior convergence at low resolutions by following flows to thin midplane layers, suggesting that sustained midplane toroidal fields in recent simulations are physical rather than numerical artifacts.

The Gist

The Big Picture: A Battle of "Digital Simulations"

Imagine you are trying to predict how a giant, swirling whirlpool of gas and magnetic fields behaves around a supermassive black hole. Scientists use supercomputers to run these simulations. But here's the catch: how you build the computer model matters just as much as the physics you put inside it.

This paper is about a disagreement between two different ways of building these models:

1. The "Static Grid" (Eulerian): Imagine a giant, invisible chessboard fixed in space. The gas moves through the squares.
2. The "Moving Swarm" (Lagrangian): Imagine a swarm of bees. The bees (representing chunks of gas) move around, and the computer tracks the bees themselves, not the space they occupy.

The Problem: The "Magnetic Squeeze"

Recently, a team of scientists (Guo et al., 2025) ran a test using the Static Grid method. They found something surprising:

• Low Resolution (Big Chess Squares): If the chess squares were too big, the magnetic field stayed strong and the gas disk stayed thick. Nothing happened.
• High Resolution (Tiny Chess Squares): If they made the squares tiny, the gas suddenly "collapsed." The magnetic field got crushed, the gas got incredibly dense in the middle, and the disk became a razor-thin sheet.

They concluded: "To see the real physics, you need a super-fine grid. If your grid is too coarse, you're missing the collapse."

The New Experiment: Let's Try the "Swarm"

The authors of this paper (Tomar & Hopkins) said, "Wait a minute. Many of the most advanced simulations in the world use the Moving Swarm method. Do they see the same thing? Or is the 'collapse' just an illusion caused by the Static Grid?"

They ran the exact same test using the Moving Swarm method (Lagrangian).

The Results: A Tale of Two Behaviors

Here is what they found, broken down simply:

1. When the Grid is Fine (High Resolution):
Both methods agreed. The gas collapsed, the magnetic field weakened, and the disk got thin. The "Swarm" and the "Chessboard" saw the same movie.

2. When the Grid is Coarse (Low Resolution):
This is where it gets weird.

• The Chessboard (Static Grid): If the squares were too big, the simulation got "stuck." The gas refused to collapse. It was like trying to pour water through a sieve with holes the size of buckets; the water just sits there. The magnetic field stayed strong forever.
• The Swarm (Moving): Even with a "coarse" swarm (fewer bees), the gas still collapsed.
  • The Analogy: Imagine you have a few heavy people (the gas) trying to squeeze into a tiny room. Even if you don't have enough people to fill the room perfectly, they will still huddle together as tightly as possible. The "Swarm" method naturally follows the gas as it piles up. It doesn't care if the "room" (the grid) is big; it just squeezes the gas until it hits the limit of how small a single "bee" can be.

The "Jeans Fragmentation" Connection

The authors compare this to a famous problem in astronomy called Jeans Fragmentation (how clouds of gas break apart to form stars).

• The Old Way (Static Grid): If your grid is too big, the computer thinks the gas is too smooth to break apart. It creates fake, broken stars that shouldn't exist.
• The New Way (Moving Swarm): The gas only breaks apart if it actually has enough mass to do so. It doesn't create fake stars just because of a bad grid.

The authors argue that the "Magnetic Collapse" in this paper is the same thing. The Static Grid method is "numbing" the collapse because its grid is too rigid. The Moving Swarm method is "honest"—it collapses the gas as much as it physically can, even if the resolution isn't perfect.

Why Does This Matter?

This is a huge deal for understanding Active Galactic Nuclei (AGN)—the super-bright centers of galaxies.

• The Fear: Some scientists worried that the strong magnetic fields seen in complex, real-world simulations were just a "glitch" caused by not having enough computer power (resolution). They thought, "If we just make the grid finer, the magnetic fields would disappear, and the disk would collapse."
• The Reality: This paper says, "No."
  • The complex simulations that show strong magnetic fields use the Moving Swarm method.
  • Our test shows that the Moving Swarm method wants to collapse the gas, even at low resolution.
  • Therefore, if the complex simulations don't collapse, it's not because of a computer error. It's because real physics (like turbulence, star formation, or different types of magnetic fields) is keeping the disk open.

The Bottom Line

Think of it like trying to fold a blanket.

• The Static Grid is like trying to fold the blanket on a rigid table. If the table is too rough (low resolution), the blanket just sits there flat. You think it can't be folded.
• The Moving Swarm is like having a person fold the blanket. Even if they are clumsy (low resolution), they will still try to fold it as tight as they can.

The authors conclude that the strong magnetic fields seen in recent, complex galaxy simulations are real. They aren't a mistake caused by low computer resolution. The difference between the simple test and the complex reality is likely due to actual physical factors (like turbulence or gravity) that the simple test didn't include.

In short: The "collapse" seen in simple tests is real, but the "stability" seen in complex, real-world simulations is also real. The simple test was just missing some of the ingredients that keep the disk from collapsing in the real universe.

Technical Summary

1. Problem Statement

The paper addresses a discrepancy in numerical simulations of accretion disks around Active Galactic Nuclei (AGN). Recent multi-physics simulations using Lagrangian methods (e.g., GIZMO) have shown the emergence of strongly toroidally-magnetized disks where the midplane magnetic pressure remains low ($\beta_{th} \ll 1$) indefinitely.

However, a recent idealized test problem by Guo et al. (2025) (G25) using Eulerian static-mesh methods (ATHENA) suggested that this behavior is a numerical artifact dependent on resolution. G25 found that:

• Low Resolution ($\Delta x \gtrsim H_{thermal}$): The disk maintains its initial magnetization indefinitely.
• High Resolution ($\Delta x < H_{thermal}$): The disk undergoes a "vertical collapse," forming a dense midplane layer where $\beta_{th} \sim 1$, accompanied by a significant loss of toroidal magnetic flux.

G25 argued that the sustained magnetization seen in Lagrangian simulations might be due to insufficient resolution to resolve the thermal scale-height ($H_{thermal}$). Tomar and Hopkins aim to test whether this convergence behavior is intrinsic to the physics or a fundamental difference between Eulerian and Lagrangian numerical schemes.

2. Methodology

The authors re-ran the G25 idealized test problem using the GIZMO code, which supports multiple numerical methods.

• Test Setup:
  • Physics: Ideal MHD, strictly locally-isothermal equation of state ($T \propto R$), no self-gravity, in an analytic Keplerian potential.
  • Initial Conditions: Collapse of a homogeneous, uniformly rotating, and uniformly magnetized sphere.
  • Resolution Tests: Simulations were run at various resolutions relative to the thermal scale-height ($H_{thermal} = c_s/\Omega$).
    • Resolved Case: $H_{thermal} = 0.05 R$.
    • Unresolved Case: $H_{thermal} = 10^{-4} R$ (intentionally unresolvable).
• Numerical Methods Compared:
  1. Meshless Finite-Mass (MFM): The default Lagrangian method used in GIZMO.
  2. Meshless Finite-Volume (MFV): A variant Lagrangian method.
  3. Fixed Cartesian Mesh: Low-resolution Eulerian runs for comparison.
  4. MHD Schemes: Tests included both Constrained-Gradient (CG-MHD) and standard Dedner divergence-suppression schemes.
• Metrics: The authors tracked the time evolution of the density scale-height ($H_\rho$) and Alfvén scale-height ($H_B$) to define "collapse" and magnetic flux loss. They defined three resolution metrics:
  • $\langle \Delta x \rangle_C$: Equivalent Cartesian resolution (Volume/$N$)$^{1/3}$.
  • $\langle \Delta x \rangle_{mid}$: Median actual resolution in the midplane.
  • $\langle \Delta x \rangle_{2d}$: The limit where the disk becomes effectively "one cell thick."

3. Key Results

A. High-Resolution Convergence

When the thermal scale-height is well-resolved ($\Delta x \ll H_{thermal}$), Lagrangian methods (MFM/MFV) reproduce the G25 Eulerian results.

• Both methods show a vertical collapse of the midplane.
• The midplane density increases, and the toroidal magnetic flux decreases until $\beta_{th} \sim 1$.
• This confirms that the physical mechanism of flux loss exists and is captured by both methods when resolution is sufficient.

B. Low-Resolution Divergence (The Core Finding)

The critical difference emerges when the thermal scale-height is poorly resolved ($\Delta x \gg H_{thermal}$):

• Eulerian (Static Mesh): The disk does not evolve. The initial magnetization is maintained indefinitely because the fixed grid cannot resolve vertical gradients below the grid spacing $\Delta x$. The Parker dynamo cannot shut down, and flux cannot be removed from scales $|z| \lesssim \Delta x$.
• Lagrangian (MFM/MFV): The disk still collapses, even when $\Delta x \gg H_{thermal}$.
  • The code evolves "as close as possible" to the converged solution.
  • The midplane collapses until it reaches the effective resolution limit (approximately one cell thick, $\langle \Delta x \rangle_{2d}$).
  • Magnetic flux is lost, and density increases, though the final state is limited by the single-cell thickness rather than the physical $H_{thermal}$.

C. Mechanism of Difference

The authors argue this difference stems from the fundamental nature of the methods:

1. Grid vs. Flow: In Eulerian codes, vertical structure is capped by the fixed grid size. In Lagrangian codes, particles/cells move with the flow. As gas concentrates toward the midplane, the local resolution ($\langle \Delta x \rangle_{mid}$) automatically improves, allowing the code to resolve the collapse even if the global average resolution is poor.
2. Flux Transport: In Lagrangian methods, cells carrying magnetic flux can physically move away from the midplane even when the vertical gradient becomes numerically ill-defined (single-cell thick). In Eulerian methods, flux is trapped within the fixed grid cells.

4. Significance and Implications

A. Resolution of the "Numerical Artifact" Debate

The results strongly suggest that the sustained, strong toroidal fields observed in recent complex, multi-physics simulations (e.g., Hopkins et al. 2024b, 2025; Kaaz et al. 2025) are not a numerical resolution artifact.

• If the lack of collapse in those simulations were due to poor resolution, the Lagrangian methods used in those studies would have collapsed to a single-cell thickness (as seen in the low-resolution G25 test).
• Since they do not collapse, and since Lagrangian methods can reproduce the collapse in the idealized test even at low resolution, the persistence of magnetization in complex simulations must be due to real physical differences.

B. Physical Drivers

The paper posits that the difference between the idealized G25 test and the complex simulations arises from:

• Multi-phase gas and turbulence: Complex simulations include turbulent inflows and multi-phase gas, unlike the smooth, uniform initial conditions of G25.
• Poloidal Magnetic Fields: The complex simulations contain non-negligible poloidal fields, which G25 found aid in maintaining low $\beta$ at high resolution.
• Boundary Conditions: Asymmetric, time-dependent, and turbulent initial conditions in complex simulations likely drive different dynamics (e.g., continuous flux replenishment) compared to the symmetric, static G25 setup.

C. Analogy to Jeans Fragmentation

The authors draw a parallel to the well-known "artificial fragmentation" problem. Just as Eulerian codes suffer from spurious fragmentation when the Jeans length is unresolved (while Lagrangian codes do not), Eulerian codes here fail to allow magnetic flux removal when the thermal scale-height is unresolved. This highlights a systematic bias in static-mesh codes for thin-disk problems.

5. Conclusion

Tomar and Hopkins conclude that the discrepancy between Eulerian and Lagrangian results in the G25 test is a fundamental numerical effect related to how each method handles unresolved vertical structures and flux transport. The fact that Lagrangian methods collapse even at low resolution implies that the lack of collapse in high-resolution, multi-physics Lagrangian simulations is a robust physical result, not a numerical failure. Future work must focus on identifying which specific physical ingredients (turbulence, poloidal fields, feedback) prevent the vertical collapse in realistic AGN disk scenarios.