Is a Black Hole's Accretion Flow Fractal? Simulations Hint at a Magnetic Answer

Black hole GRS 1915+105 is one of the Galaxy's most restless X-ray sources. It flares, dips, and switches among patterns that last seconds or months. Astronomers have catalogued 12 of these temporal classes, but what governs the flickering remains an open question. A new arXiv preprint asks whether part of the answer lies in magnetic structure, and whether the resulting variability has a fractal character.

The work, posted to arXiv on April 17, 2026, combines general relativistic magnetohydrodynamic simulations—supercomputer models of magnetized plasma falling onto a black hole in curved spacetime—with nonlinear timeseries analysis, then compares simulated patterns with archival X-ray observations of GRS 1915+105. The authors focus on two accretion regimes: MAD and SANE.

MAD and SANE: two magnetic ways to feed a black hole

A black hole's accretion flow is the stream of gas and plasma spiraling inward. Magnetic fields can regulate how that material moves, how energy is dissipated, and how powerful any jets become.

In a MAD, or magnetically arrested disk, magnetic flux piles up near the black hole until it intermittently erupts. These systems tend to produce powerful jets. In a SANE flow, or standard and normal evolution, the magnetic field is weaker and more disorganized. Turbulence driven by the magnetorotational instability dominates, and jets or winds are generally weaker.

The new study asks whether those two regimes leave different fingerprints in the timing of their variability.

HARMPI and BHAC simulations: MAD variability looks more complex

The researchers ran GRMHD simulations with two codes, HARMPI and BHAC, and varied black hole spin from a Kerr parameter of -0.9375 (retrograde) to 0.9375 (prograde). They tracked two quantities at the event horizon: outflow efficiency, a proxy for jet activity, and normalized magnetic flux, a proxy for disk dynamics. They analyzed the late, stable part of each simulated timeseries, from 25,000 to 30,000 rg/c, where rg/c is a light-crossing time over the gravitational radius.

Three nonlinear diagnostics were used. The Higuchi fractal dimension, or HFD, ranges between 1 and 2; higher values indicate a more irregular, self-similar signal. The Hurst index measures long-range memory and correlations. The spectral slope describes how variability power is distributed across frequencies. The authors also used surrogate data tests to check whether the signals were genuinely nonlinear rather than linear stochastic signals.

Across most spins, simulated MAD flows had higher HFD, lower Hurst index, and flatter spectral slopes than SANE flows. That pattern appeared in both jet and disk proxies. In the non-spinning case, the jet proxy gave an HFD of 1.89 for MAD and 1.50 for SANE.

The interpretation is that MAD variability is more complex and less correlated over long timescales. Its intermittent flux eruptions break up memory in the signal. SANE flows, by contrast, are more continuously turbulent and more correlated, producing a smoother, lower-HFD pattern.

Spin changes the picture. As the magnitude of black hole spin increases, MAD HFD generally decreases. The authors attribute this to increasingly collimated, ordered jets, which make the flow more coherent. In SANE simulations, HFD rises for prograde spins, where winds and jets interact. The result is not a simple function of spin alone; magnetic structure matters too.

The analysis also comes with internal tensions. At the highest prograde spin, a=0.9375, in the HARMPI runs, SANE HFD overtakes MAD, but the spectral slope does not show the same crossover, and the nonlinearity test is weak. The BHAC runs do not reproduce that crossover and show some linear behavior at high spin. The authors therefore treat the extreme-spin comparison cautiously.

GRS 1915+105: ten nonlinear X-ray classes split into MAD-like and SANE-like clusters

The second half of the paper moves from models to observations. GRS 1915+105 is a well-studied black hole X-ray binary. Its RXTE lightcurves have been sorted into 12 temporal classes. The researchers filtered those lightcurves and computed HFD for each class. Ten classes had z-scores above 2, the paper's threshold for nonlinearity; two classes were classified as stochastic.

They then used k-means clustering, an unsupervised machine-learning method, to group the ten nonlinear classes by X-ray spectral and related properties. One cluster was dominated by a power-law component, about 80 percent on average. The other had a large disk-blackbody component, about 51 percent, alongside a power-law part of about 48 percent.

The authors interpret the power-law-dominated cluster as MAD-like and the disk-blackbody-rich cluster as SANE-like. The mean HFD was higher for the MAD-like cluster, 1.651 versus 1.466 for the SANE-like cluster. A t-test gave p=0.038. HFD also correlated positively with the power-law component (r=0.78, p=0.008) and negatively with the disk-blackbody component (r=-0.8, p=0.005).

If real, that connection would be valuable. It would mean X-ray flickering could help reveal the magnetic state of an accretion flow, complementing traditional spectral fitting.

Extreme spins, 2.5D simulations, and inferred labels keep the claim cautious

The study does not prove that GRS 1915+105 switches between MAD and SANE states. The labels are inferred by analogy between observed spectra and simulated flow properties, not measured directly.

The simulations are also 2.5-dimensional and axisymmetric. They may miss three-dimensional instabilities and turbulent mixing that affect fractal properties. The analysis uses a finite time window, 25,000-30,000 rg/c, chosen for numerical stability and stationarity, and individual simulation realizations can vary even under identical conditions. The two codes differ in how MAD eruptions behave; at high spin, BHAC cannot confirm the nonlinear properties, so some comparisons remain unresolved.

Still, the work offers a new way to read old data. Black hole accretion is often described through spectra, images, or average rates. This paper suggests that the shape of variability over time, its fractal dimension and memory, may carry complementary information about the magnetic fields that regulate the flow.

If that idea holds up in more simulations and more sources, astronomers may gain another tool for connecting the horizon-scale physics of black holes to the X-ray light that reaches Earth. The flickering of GRS 1915+105 would then be more than noise. It would be a fingerprint.