betaFeedback

WorksPhysics and Astronomy

Temporal polarization properties of accretion flows into supermassive black holes: INSANE as an alternative

Abstract

Temporal variability and polarization provide unique diagnostics of the magnetic field structure and plasma dynamics in accretion flows around supermassive black holes. We investigate the temporal polarization properties of horizon-scale emission using general relativistic magnetohydrodynamic (GRMHD) simulations coupled with polarized general relativistic radiative transfer (GRRT) for three accretion states: Standard and Normal Evolution (SANE), Magnetically Arrested Disk (MAD), and an intermediate regime (INSANE). Fourier polarization-mode decomposition, through the ratio ∣βm∣/∣β2∣|β_m|/|β_2|, shows that MAD produces the most rotationally symmetric EVPA patterns, SANE the most asymmetric, and INSANE lies in between. The multi-scale temporal behaviour of the intensity is investigated using the Continuous Wavelet Transform and Hilbert-Huang Transform. Both methods identify localized oscillatory features associated with magnetic-flux eruption events in the MAD simulations. In the shorter interval examined, INSANE distributes its variability more evenly among several time-scales, while the SANE behaviour depends more strongly on the electron-temperature prescription. To quantify the spectral distribution, we define a fast power fraction Ffast,IF_{\mathrm{fast},I} for periods of 80-200 MM within a broad band of 80-1500 MM. Both methods give the smallest fractions for MAD; the ordering of state medians, INSANE > SANE > MAD, persists when the upper fast-band boundary is varied from 200 to 500 MM. Our analysis demonstrates that combining polarization morphology with multi-scale temporal analysis provides a useful means of distinguishing magnetization states, supporting INSANE as a distinct observational alternative to conventional MAD/SANE for interpreting horizon-scale observations of M87∗^{*} and related systems.

Introduction

Black holes accreting at low Eddington ratios power some of the most energetic relativistic jets in the Universe, yet the physical state of the plasma in the immediate vicinity of the event horizon remains uncertain (Yuan & Narayan 2014). A key unresolved question concerns the magnetic flux threading the black hole and inner accretion flow properties, which determine whether the system operates in a Standard And Normal Evolution (SANE) configuration, a Magnetically Arrested Disk (MAD) state, or an intermediate regime with partial magnetic flux accumulation (Narayan et al. 2012b). This distinction has profound implications for jet launching, particle acceleration, and the temporal variability of the emitted radiation.

The nearby galaxy M87 provides an exceptional laboratory for studying these processes. With horizon-scale imaging achieved by the Event Horizon Telescope (EHT) (Event Horizon Telescope Collaboration et al. [7] (2022a)), combined with multi-wavelength monitoring and polarization measurements, M87 offers direct observational access to plasma dynamics occurring on horizon-scale variability (Event Horizon Telescope Collaboration et al. [11]). A complementary perspective is provided by observations of Sgr A*, the supermassive black hole at the centre of the Milky Way, where rapid intrinsic variability and lower jet power suggest a different accretion environment and potentially a lower net magnetic flux state (Event Horizon Telescope Collaboration et al. 2022b,c). While several EHT modelling efforts for M87 indicate that MAD configurations reproduce many observed features, particularly jet energetics and polarized structure, current observational constraints do not uniquely exclude SANE or intermediate-magnetization (“weak-MAD”) flows (Event Horizon Telescope Collaboration et al. 2021a,b, 2025a,b). In contrast, for Sgr A*, present observational and modelling studies generally favour SANE or moderately magnetized configurations, with strongly arrested MAD states appearing less consistent with the observed variability and polarization properties (Event Horizon Telescope Collaboration et al. 2022b,c, 2024a). Together, these systems highlight the importance of variability diagnostics for distinguishing among different magnetic flux regimes in low-luminosity accreting black holes.

General Relativistic Magnetohydrodynamic (GRMHD) simulations predict distinct variability signatures associated with different magnetization states. SANE disks typically exhibit turbulence-dominated, small-scale stochastic fluctuations, whereas MAD systems display more coherent large-scale magnetic structures that can modulate emission on longer and often correlated timescales (Narayan et al. 2003; Igumenshchev et al. 2003; Igumenshchev 2008; Begelman et al. 2022). Intermediate or partially arrested disks, sometimes referred to as INSANE or semi-MAD states, may produce variability characteristics that lie between these two extremes, complicating the interpretation of observational data (Raha et al. 2025). Consequently, systematic comparisons between variability predictions of SANE, intermediate, and MAD simulations are essential for constraining the magnetic state of the M87 accretion flow.

Accretion flows forming the horizon-scale environments of black holes are intrinsically time-dependent, driven by magnetohydrodynamic turbulence, magnetic reconnection, and dynamical plasma instabilities. At millimeter wavelengths, particularly near 1.3 mm, the observed emission is dominated by synchrotron radiation produced by relativistic electrons energized within these turbulent magnetized flows (Event Horizon Telescope Collaboration et al. 2019c). Because the emitting plasma evolves on orbital and inflow timescales close to the innermost stable circular orbit (ISCO), the brightness distribution of the source is expected to vary on characteristic dynamical timescales that depend strongly on the black hole mass and spin (Bardeen et al. 1972).

For the Galactic center black hole Sgr A*, these characteristic timescales are on the order of minutes to tens of minutes, implying that appreciable structural variability can occur even during a single observing night, complicating the interpretation of time-averaged images. In contrast, the substantially larger mass of the M87 black hole shifts the relevant dynamical timescales to days or weeks, suggesting that short-duration observing campaigns should capture comparatively modest structural evolution, while longer baselines may reveal measurable changes in the morphology and azimuthal brightness distribution of the emission ring (Wielgus et al. 2020). Indeed, multi-epoch very-long-baseline interferometry (VLBI) observations spanning several years have already indicated the presence of long-term structural variability in M87, consistent with expectations from turbulent accretion dynamics.

The 2017 EHT observing campaign of M87, which covered multiple nights separated by several days, sampled only a few characteristic dynamical times near the ISCO. Consequently, only moderate day-to-day variations in ring thickness and brightness asymmetry were inferred, highlighting the importance of time-dependent modeling for interpreting interferometric observables. Because the EHT measures complex visibilities corresponding to the Fourier components of the sky brightness distribution, temporal evolution in the accretion flow is encoded directly in the variability of visibility amplitudes, phases, and polarization quantities rather than solely in reconstructed images [8].

In this work, we employ time-dependent general relativistic magnetohydrodynamic (GRMHD) simulations spanning SANE, intermediate (INSANE), and MAD accretion states, coupled with polarized general relativistic radiative transfer (GRRT) calculations tailored to the physical parameters of M87. Using these simulations, we investigate how short- and intermediate-timescale variability in both total intensity and polarized emission manifests in interferometric observables and assess the extent to which variability diagnostics can distinguish between different magnetization regimes of the M87 accretion flow.

Numerical setup

General relativistic magnetohydrodynamic setup

We conducted a set of three-dimensional (3D) GRMHD simulations of a magnetized torus around a spinning black hole using the H-AMR code [23]. Each simulation is initialized from a Fishbone–Moncrief hydrodynamic equilibrium torus (Fishbone & Moncrief 1976; Uniyal et al. 2024) with parameters rin=20rgr_{\mathrm{in}} = 20r_{\mathrm{g}} and pressure maximum location rmax=41rgr_{\mathrm{max}} = 41r_{\mathrm{g}}, where rg≡GM/c2r_{\mathrm{g}} \equiv GM/c^{2} is the gravitational radius of the black hole and MM its mass for the MAD, INSANE and SANE cases, while oddities runs have also been performed with rin=6rgr_{\mathrm{in}} = 6r_{\mathrm{g}} and rmax=12rgr_{\mathrm{max}} = 12r_{\mathrm{g}} for the SANE and INSANE cases which are considered standard choices in the literature. We use an ideal-gas equation of state with the adiabatic index Γg=4/3\Gamma_{\mathrm{g}} = 4/3 (Rezzolla & Zanotti 2013). Furthermore, the simulations employ a fixed mesh refinement strategy corresponding to an effective resolution of 384×192×192384 \times192 \times192 cells. We summarize all GRMHD and radiative transfer models in Table 1. The details of the initial configuration setup can be found in Appendix § 8.

a∗a_{*}IDrinr_{\mathrm{in}} (RgR_{\mathrm{g}})rmaxr_{\mathrm{max}} (RgR_{\mathrm{g}})RlowRhighii (deg)Munit/102510^{25} (g)tst_{\mathrm{s}} (M)tft_{\mathrm{f}} (M)Δt\Delta t (M)Resolution
0.9375MAD_2020411(10, 80)163(1.01, 2.62)150003000010384 × 192 × 192
0.9375INSANE_2020411(10, 80)163(9.42, 25.1)150003000010384 × 192 × 192
0.9375INSANE_66121(10, 80)163(1.94 × 10310^{3}, 5.41 × 10310^{3})150003000010384 × 192 × 192
0.9375SANE_2020411(10, 80)163(30.0, 86.1)150003000010384 × 192 × 192
0.9375SANE_66121(10, 80)163(2.47 × 10310^{3}, 1.17 × 10410^{4})150003000010384 × 192 × 192
0.5MAD_0.520411(10, 80)163(1.29, 3.27)150003000010384 × 192 × 192
0.5INSANE_0.520411(10, 80)163(17.0, 41.9)150003000010384 × 192 × 192
0.5SANE_0.520411(10, 80)163(85.6, 190.8)150003000010384 × 192 × 192

Table 1. Overview of GRMHD and radiative transfer models employed to produce various time series of synthetic images.

Figure 1 shows the horizon mass-accretion rate M˙\dot{M}, the horizon magnetic flux ΦBH\Phi_{\mathrm{BH}}, the normalized magnetic flux ϕBH≡ΦBH/M˙\phi_{\mathrm{BH}} \equiv\Phi_{\mathrm{BH}}/\sqrt{\dot{M}}, and the total outflow efficiency ηout\eta_{\mathrm{out}}. At fixed large-torus geometry, MAD generally has the highest characteristic horizon flux and outflow efficiency, with INSANE typically between MAD and SANE, although individual curves can overlap. MAD remains in a saturated state throughout the evolution, with ϕBH\phi_{\mathrm{BH}} and ηout\eta_{\mathrm{out}} fluctuating around approximately constant values. Its episodic decreases of ΦBH\Phi_{\mathrm{BH}} and subsequent recovery are consistent with magnetic-flux eruptions followed by renewed inward advection.

Figure 1. Time evolution of the horizon mass-accretion rate M˙\dot{M}, horizon magnetic flux ϕ\phi, normalized horizon magnetic flux ϕBH=ΦBH/M˙\phi_{\rm BH} = \Phi_{\rm BH}/\sqrt{\dot{M}}, and total outflow efficiency ηout\eta_{\rm out}. The left column shows models with a∗=0.9375a_* = 0.9375 and the right column models with a∗=0.5a_* = 0.5. Model suffixes 20 and 6 denote (rin,rmax)=(20,41) Rg(r_{\rm in}, r_{\rm max}) = (20, 41)\,R_g and (6,12) Rg(6, 12)\,R_g, respectively. Only the large-torus configuration is available for a∗=0.5a_* = 0.5. The curves illustrate that the ordering of the dynamical quantities depends on both accretion state and initial torus configuration.

The SANE and INSANE models show opposite trends for the two torus sizes. In the small-torus models (SANE_6, INSANE_6), M˙\dot{M} decreases with time while ϕBH\phi_{\mathrm{BH}} increases; in the large-torus models (SANE_20, INSANE_20), M˙\dot{M} increases or remains approximately constant while ϕBH\phi_{\mathrm{BH}} decreases. The absolute horizon flux does not follow these trends. In the small-torus models, ΦBH\Phi_{\mathrm{BH}} remains approximately constant (INSANE_6) or decreases only slowly (SANE_6), so the increase of ϕBH\phi_{\mathrm{BH}}, by a factor of ∼4\sim4–5 in INSANE_6, results almost entirely from the decrease of M˙\dot{M} and cannot be interpreted as a growth of the absolute horizon flux. In the large-torus models, ΦBH\Phi_{\mathrm{BH}} decreases by a factor of ∼3\sim3–4 after t≈1.3t \approx1.3–1.5×104 M1.5 \times10^{4}\,M while M˙\dot{M} is maintained, so the decline of ϕBH\phi_{\mathrm{BH}} reflects a genuine loss of horizon flux, with an additional contribution from the rising M˙\dot{M} in SANE_20.

These differences arise because mass and magnetic flux are accreted differently. The accretion rate traces the instantaneous supply of gas, whereas ΦBH\Phi_{\mathrm{BH}} is an accumulated quantity that changes only when magnetic flux of either polarity is advected inward or is removed by reconnection or escape. The small tori have their inner edge and pressure maximum close to the black hole and hence a small mass reservoir with a short inflow time; the torus is progressively depleted and M˙\dot{M} decreases, while the flux accumulated on horizon at early times is largely retained. The large tori have a more extended reservoir and a longer inflow time, which sustain the accretion rate. Since the initial magnetic field follows the torus density distribution, this sustained inflow eventually delivers the outer part of the initial magnetic loop, which carries flux of opposite polarity; we attribute the decrease of ΦBH\Phi_{\mathrm{BH}} in these models to the partial cancellation of the previously accumulated horizon flux by this material (see also Beckwith et al. 2008).

The outflow efficiency follows the evolution of ϕBH\phi_{\mathrm{BH}}, as expected for magnetically driven outflows. It is generally highest in MAD, increases from ∼0.04\sim0.04 to ∼0.15\sim0.15–0.20.2 in INSANE_6, and decreases by about an order of magnitude in SANE_20 and INSANE_20, such that INSANE_6 exceeds INSANE_20 in both ϕBH\phi_{\mathrm{BH}} and ηout\eta_{\mathrm{out}} at t≳2.4×104 Mt \gtrsim2.4 \times10^{4}\,M. At late times, ηout\eta_{\mathrm{out}} is higher in INSANE_6 than in SANE_6. The high-spin MAD model occasionally reaches ηout>1\eta_{\mathrm{out}} > 1, indicating net extraction of black-hole rotational energy (Tchekhovskoy et al. 2011).

Polarised general relativistic radiative transfer setup

In this work, we carry out GRRT calculations for the 3D accretion models using the RAPTOR code (Bronzwaer et al. 2018, 2020). The simulated source is assumed to represent M87∗M87^{*}, with a black hole mass of M=6.5×109M⊙M = 6.5 \times10^{9} M_{\odot} and a distance of 16.8 Mpc (Event Horizon Telescope Collaboration et al. [10]). Images are generated on a grid with a pixel resolution of 500 × 500, while the total field of view in the image plane is set to 130 μ\muas (= 34 RgR_{\mathrm{g}}). As commonly adopted in modeling horizon-scale emission from supermassive black holes, we consider synchrotron radiation from a purely thermal electron population in all simulations. The electron temperature is determined using the standard RR–β\beta prescription for the ion-to-electron temperature ratio (Mościbrodzka et al. 2009),

TiTe=Rlow+Rhighβ21+β2,(1)\frac{T_{\mathrm{i}}}{T_{\mathrm{e}}} = \frac{R_{\mathrm{low}} + R_{\mathrm{high}}\beta^{2}}{1 + \beta^{2}}, \tag*{(1)}

where β=pg/pm\beta= p_{\mathrm{g}}/p_{\mathrm{m}} denotes the plasma beta, defined as the ratio of gas pressure to magnetic pressure. We fix Rlow=1R_{\mathrm{low}} = 1 and use Rhigh=10R_{\mathrm{high}} = 10 and 80, giving two electron-temperature prescriptions for each GRMHD simulation. The viewing inclination is fixed at i=163∘i = 163^{\circ}.

We considered the data between t=15000t = 15000 M and 30000 M, with a cadence of 10 M, unless otherwise specified. The density normalization is specified by the mass unit MunitM_{\mathrm{unit}} in Table 1. The analysed light curves are normalized to a median 230 GHz flux density of 0.5 Jy, consistent with the intensity observed for M87∗M87^{*}.

Asymmetry of polarization morphology

Palumbo et al. [31] proposed a way to quantify the polarization morphology using the Fourier mode coefficients βm\beta_{m}:

βm=1Iann∫ρminρmaxdρ ρ∫02πdφ P(ρ,φ)e−imφ,(2)\beta_{m} = \frac{1}{I_{\mathrm{ann}}}\int_{\rho_{\mathrm{min}}}^{\rho_{\mathrm{max}}} d\rho\,\rho\int_{0}^{2\pi} d\varphi\,P(\rho,\varphi)e^{-im\varphi}, \tag*{(2)}
Iann=∫ρminρmaxdρ ρ∫02πdφ I(ρ,φ),(3)I_{\mathrm{ann}} = \int_{\rho_{\mathrm{min}}}^{\rho_{\mathrm{max}}} d\rho\,\rho\int_{0}^{2\pi} d\varphi\,I(\rho,\varphi), \tag*{(3)}

where P(ρ,φ)P(\rho,\varphi) and I(ρ,φ)I(\rho,\varphi) are the complex linear polarization and intensity located at the polar image-plane coordinates (ρ,φ)(\rho,\varphi). The magnitude ∣β2∣|\beta_{2}| quantifies the strength of the rotationally symmetric polarization, while the phase ∠β2\angle\beta_{2} specifies the orientation of this symmetry relative to a purely radial EVPA pattern. For example, ∠β2=0\angle\beta_{2} = 0 corresponds to a purely radial polarization field, ∠β2=π\angle\beta_{2} = \pi to a purely azimuthal field, and negative or positive ∠β2\angle\beta_{2} values indicate spirals with left- or right-handed helicity. Other modes βm\beta_{m}, m≠2m \ne2 represent the rotation asymmetry of the EVPA patterns.

Figure 2 presents the time-averaged polarized images between t=15000t = 15000 M and 30000 M for the three accretion states, two black-hole spins, and two electron-heating prescriptions. We note that the torus sizes have comparable distributions of polarization properties, as shown in Figure B1, therefore, we focus on comparing the simulations using a large torus with Rin=20RgR_{\mathrm{in}} = 20R_{\mathrm{g}} for time-averaged properties. The EVPA morphology immediately reveals systematic differences between the magnetic configurations. As expected, the MAD models exhibit the most ordered polarization structure, with predominantly azimuthal EVPA patterns characteristic of highly organized magnetic fields. In contrast, the SANE models display noticeably more distorted EVPA structures, reflecting a greater degree of rotational asymmetry. The INSANE models occupy an intermediate regime, producing polarization patterns that are more ordered than those of SANE while remaining less symmetric than those of MAD. These qualitative differences are quantified by the quadrupolar polarization mode β2\beta_{2}, whose amplitude measures the strength of the rotationally symmetric polarization component. Across all combinations of black-hole spin and RhighR_{\mathrm{high}}, the INSANE models yield intermediate ∣β2∣|\beta_{2}| values between the MAD and SANE limits, confirming that their average polarization morphology is likewise intermediate. The dependence on the electron-heating prescription is comparatively weak for the INSANE and SANE models, whereas the MAD models exhibit a stronger response to changes in RhighR_{\mathrm{high}}. Increasing the black-hole spin generally produces more symmetric EVPA patterns, although the overall ordering of the three accretion states remains unchanged.

Averaged intensity images with EVPA patterns

Figure 2. Averaged intensity images with EVPA patterns between t=15000 Mt = 15000\,M and t=30000 Mt = 30000\,M at 230 GHz. In each panel, the GRRT images are produced by the post-processing of various GRMHD simulations at a=0.94a = 0.94 and a=0.5a = 0.5, and Rhigh=10R_{\mathrm{high}} = 10 and Rhigh=80R_{\mathrm{high}} = 80. Polarization ticks are displayed only in regions where I>0.1 Imax⁡I > 0.1\,I_{\max}. The eDF used here are thermal distributions, and all images are produced from the total region of GRMHD simulations.

Figure B1. Histograms of polarization and intensity quantities for the MAD, INSANE, and SANE models at 230 GHz. Results for Rhigh=10R_{\mathrm{high}} = 10 and Rhigh=80R_{\mathrm{high}} = 80 are shown in red and blue, respectively. From left to right, the columns correspond to MAD, INSANE, and SANE configurations. The shaded bands (light blue, light orange, and light green) indicate the ranges inferred from different years of EHT observations of M87 (2017, 2018, and 2021). The histograms between simulations using a large torus and a small torus at spin 0.9375 are similar.

The average polarization morphology can be quantified further by examining the complete spectrum of Fourier modes. Figure 3 shows the amplitudes of the polarization coefficients averaged over t=15000t = 15000 M and 30000 M. The upper panel presents the absolute amplitudes ∣βm∣|\beta_{m}|. The overall polarization strength decreases systematically from MAD to INSANE and finally to SANE, consistent with the increasing level of magnetic disorder seen in the polariza tion images. To isolate the degree of rotational asymmetry from the overall polarization amplitude, the lower panel shows the normalized mode ratios ∣βm∣/∣β2∣|\beta_m|/|\beta_2|, following Long et al. (2026). Here, ∣β2∣|\beta_2| represents the dominant rotationally symmetric component expected from ring-like emission, while modes with m≠2m \ne2 quantify deviations from this symmetry. The SANE models consistently exhibit the largest relative power in the asymmetric modes, demonstrating that their polarization morphology is intrinsically less axisymmetric than that of the other accretion states. For the high spin a=0.9375a = 0.9375 case, the m=3m = 3 component is comparable to or even exceeds the quadrupolar mode (∣β3∣/∣β2∣≃0.93|\beta_3|/|\beta_2| \simeq0.93 and 1.05 for Rhigh=10R_{\rm high} = 10 and 80, respectively), indicating that asymmetric structures dominate the polarization morphology. At the opposite extreme, the MAD models retain only weak higher-order modes, consistent with their highly ordered magnetic fields. The INSANE models again occupy an intermediate regime, with asymmetric mode amplitudes lying between the MAD and SANE limits.

Figure 3. Different beta modes of MAD, INSANE, and SANE models take the median value from t=15000t = 15000 M to 3000030000 M. The top panel and bottom panel show βm\beta_m and the ratio of ∣βm∣/∣β2∣|\beta_m|/|\beta_2| at different modes for total emissions.

We define A32=median⁡t(∣β3(t)∣/∣β2(t)∣)A_{32} = \operatorname{median}_t(|\beta_3(t)|/|\beta_2(t)|), as a characteristic of polarization asymmetry. Figure 4 shows that MAD models consistently yield the lowest values, A32≃0.28A_{32} \simeq0.28–0.400.40, compared with 0.640.64–0.870.87 for INSANE and 0.750.75–1.081.08 for SANE. The systematically smaller MAD ratios indicate a stronger relative contribution from the rotationally symmetric component across the spins and electron-heating prescriptions explored here. At matched spin, initial torus size, and RhighR_{\rm high}, INSANE generally lies between MAD and SANE, although the separation between INSANE and SANE depends on the model parameters. In particular, their ratios nearly coincide at a=0.5a = 0.5 and Rhigh=10R_{\rm high} = 10, with A32≃0.75A_{32} \simeq0.75 for both states. By contrast, the high-spin SANE models approach or exceed unity, indicating that the third-mode amplitude is typically comparable to or larger than that of the second mode. Thus, A32A_{32} captures a systematic change in the polarization morphology across the accretion states and is particularly useful for distinguishing MAD from INSANE and SANE within our model set.

Figure 4. Relative amplitudes of the third and second azimuthal polarization modes for the MAD (red), INSANE (gold), and SANE (blue) models. Bars show A32=median⁡t(∣β3(t)∣/∣β2(t)∣)A_{32} = \operatorname{median}_t(|\beta_3(t)|/|\beta_2(t)|), evaluated over 15000–30000 M. Filled and outlined bars represent Rhigh=10R_{\mathrm{high}} = 10 and 80, respectively. Larger A32A_{32} indicates a stronger third-mode contribution relative to the second mode; MAD models exhibit systematically lower ratios across the models.

Multi-scale variability of intensity

In this section, we investigate the variability of the intensity time series. Figure 1 shows the total intensity light curves, which exhibit variability on both short and long timescales. The variability amplitude and secular evolution depend on the accretion state and initial torus configuration: models initialized with a larger torus (rin=20 Rgr_{\rm in} = 20\,R_g) display pronounced long-term trends, whereas the small-torus models show weaker secular evolution.

The fractional variability, σ/μ\sigma/\mu (Event Horizon Telescope Collaboration et al. (2022); Wielgus et al. (2022)), summarizes the overall fluctuation amplitude but does not distinguish between timescales or reveal how variability evolves with time. Moreover, its measured value depends on the duration of the light curve (Wielgus et al. (2022)) and secular trends can dominate the variance and obscure the value. We therefore use time-frequency analysis to characterize the evolving variability of the intensity, together with the ratio of high- to low-frequency power as a diagnostic of changes in the relative contributions of rapid and slow fluctuations.

Time–frequency analysis

The small-torus INSANE_6 and SANE_6 simulations exhibit substantially weaker long-term evolution and therefore provide a more representative basis for time-series analysis. We therefore focus on the analysis of the small torus in this subsection, and similar analysis for other models can be found in the Appendix C.

To characterize the variability as a function of both time and frequency, we analyse the 230 GHz total intensity II with two complementary methods, the continuous wavelet transform (CWT) and the Hilbert–Huang transform (HHT). The CWT projects the light curve onto a fixed basis of scaled wavelets, whereas the HHT decomposes it adaptively into intrinsic mode functions (IMFs), each with a time-dependent instantaneous frequency and amplitude, and is well suited to non-linear and non-stationary signals. Features recovered by both methods are therefore unlikely to be artefacts of either decomposition. To limit edge effects, we mask the part of each scaleogram that lies below the cone of influence (COI) and discard the first and last 10 per cent of each Hilbert spectrum. The COI also sets the lowest accessible frequency, which is ≃1.9×10−4 M−1\simeq1.9 \times10^{-4}\,M^{-1} for the full interval and ≃5.7×10−4 M−1\simeq5.7 \times10^{-4}\,M^{-1} for the shorter interval considered below.

Figure 5 shows the analysis over t=15000t = 15000–30000 M30000\,M. In the MAD model, the wavelet amplitude is concentrated in two episodes, at t≈17000t \approx17000–18500 M18500\,M and t≈23000t \approx23000–25000 M25000\,M, with characteristic frequencies f≈(5f \approx(5–8)×10−4 M−18) \times10^{-4}\,M^{-1}, i.e. time-scales of ≈1300\approx1300–1800 M1800\,M. Each is associated with a large flare of the light curve that ends in a rapid decline. In the second episode II rises from ≈0.6\approx0.6 to ≈1.5\approx1.5 Jy over ≈1000 M\approx1000\,M and then falls by an order of magnitude within ≈400 M\approx400\,M, at the time of the deepest drop of the horizon magnetic flux in this simulation (t≈24400t \approx24400–24900 M24900\,M in Figure 1). This supports an association with a magnetic-flux eruption. The Hilbert spectra show their brightest ridges at the same times and at f≈(0.3f \approx(0.3–1)×10−3 M−11) \times10^{-3}\,M^{-1}, and the instantaneous frequency changes by a factor of a few across each event, as expected for a transient rather than for a coherent oscillation. The MAD panels for Rhigh=10R_{\mathrm{high}} = 10 and Rhigh=80R_{\mathrm{high}} = 80 are almost indistinguishable.

Figure 5. Time–frequency analysis of the computed 230 GHz intensity light curves, over 15 000–30 000 MM for the MAD, INSANE and SANE models (columns) with Rhigh=10R_{\mathrm{high}} = 10 and Rhigh=80R_{\mathrm{high}} = 80 (rows). Top: continuous wavelet transform scaleograms, computed with a complex Morlet wavelet. The colour scale shows the wavelet amplitude ∣CWT∣|\mathrm{CWT}|. The dashed curve marks the cone of influence; the grey region below it is affected by edge effects and is excluded. Bottom: Hilbert–Huang spectra of the same light curves, obtained by empirical mode decomposition followed by the Hilbert transform of each intrinsic mode function (IMF). The colour scale shows the instantaneous Hilbert amplitude in Jy. The grey strips mask the first and last 10 per cent of the time range, where the decomposition is unreliable. The number of IMFs and the index of orthogonality (IO) are given in each panel. In both sets of panels, time is in units of 103 M10^{3}\,M and frequency in M−1M^{-1}, and each panel has its own colour scale.

The INSANE model shows no comparably localized episode. Its largest wavelet amplitudes lie at lower frequency, f≈3×10−4 M−1f \approx3 \times10^{-4}\,M^{-1} (a time-scale of ≈3000\approx3000 M), which is also where the time-averaged wavelet spectrum peaks for both values of RhighR_{\mathrm{high}}. This component is strongest at t≈19 500t \approx19\,500–21 00021\,000 M. It remains visible over most of the interval inside the COI for Rhigh=10R_{\mathrm{high}} = 10 and fades after t≈22 000t \approx22\,000 M for Rhigh=80R_{\mathrm{high}} = 80. The HHT assigns it to one or two low-frequency IMFs whose instantaneous frequencies drift by a factor of 2–3 around this value. It reflects the broad maximum of the light curve at t≈17 000t \approx17\,000–19 00019\,000 M and the subsequent slow decline. Only about five cycles of this period fit within the analysed interval, and the temporal resolution of the wavelet at this frequency is several thousand M, so we interpret it as a slow modulation and not as a quasi-periodic oscillation. At f≳10−3f \gtrsim10^{-3} M−1^{-1} the INSANE scaleograms contain many weak, short-lived features that are distributed evenly in time.

The SANE model is the most sensitive to the electron-heating prescription. For Rhigh=10R_{\mathrm{high}} = 10 the scaleogram is dominated by one feature at t≈26 000t \approx26\,000–27 50027\,500 M and f≈5×10−4f \approx5 \times10^{-4} M−1^{-1} (a time-scale of ≈2000\approx2000 M), produced by a dip of II to ≈0.2\approx0.2 Jy followed by a flare to ≈1\approx1 Jy at t≈26 400t \approx26\,400 M. The Hilbert amplitudes of four of the seven IMFs peak within this interval. For Rhigh=80R_{\mathrm{high}} = 80 the same flare reaches only ≈0.6\approx0.6 Jy, and the scaleogram instead shows a faint band at f≈(5f \approx(5–6)×10−46) \times10^{-4} M−1^{-1} of nearly constant amplitude between t≈17 000t \approx17\,000 and 27 00027\,000 M. The HHT recovers this band as an IMF with a median frequency of 6×10−46 \times10^{-4} M−1^{-1} and an amplitude of ≈0.08\approx0.08 Jy, superposed on a secular decline of the flux that accounts for the power at the lowest resolved frequencies. On time-scales longer than ∼1000\sim1000 M, the intensity variability of the MAD model and of the SANE model with Rhigh=10R_{\mathrm{high}} = 10 is therefore dominated by a few discrete, large-amplitude events, whereas that of the INSANE model is a slow modulation without such events.

To examine shorter time-scales, Figure 6 repeats the analysis of the same light curves over t=22 000t = 22\,000–27 000 M27\,000\,M. Restricting the interval removes the slow components described above, and the three accretion states then differ more clearly. The MAD panels contain a single feature, centred on the flux collapse at t≈24 100t \approx24\,100–24 400 M24\,400\,M, which appears in the scaleogram as a narrow streak extending from the lowest accessible frequency to f∼10−2 M−1f \sim10^{-2}\,M^{-1}. The Hilbert amplitudes of nearly all IMFs peak at this time, and one IMF with a median frequency of ≈10−3 M−1\approx10^{-3}\,M^{-1} carries 50–63 per cent of the total IMF energy in the unmasked part of the interval.

Figure 6. As Figure 5, but for the interval 22 000–27 000 MM of the same light curves.

The INSANE light curve contains no dominant event in this interval. Its fractional rms variability is 14–21 per cent, compared with ≈58\approx58 per cent for the MAD model and 21–33 per cent for the SANE model. Its wavelet amplitude is instead spread over most of a decade in frequency. It exceeds half of its maximum up to f≈(3f \approx(3–4)×10−3 M−14) \times10^{-3}\,M^{-1}, compared with ≈10−3 M−1\approx10^{-3}\,M^{-1} for the MAD model and (2(2–3)×10−3 M−13) \times10^{-3}\,M^{-1} for the SANE model, and it consists of several local maxima of comparable strength, with periods between ≈300\approx300 and ≈1800 M\approx1800\,M, that are spread over the whole interval. The HHT gives the same picture. The energy is shared among several IMFs, and the most energetic one in the unmasked part has a median frequency of (2.4(2.4–3.1)×10−3 M−13.1) \times10^{-3}\,M^{-1} (a time-scale of ≈320\approx320–410 M410\,M), a factor of ≈3\approx3 higher than the (6(6–13)×10−4 M−113) \times10^{-4}\,M^{-1} found for the MAD and SANE models. This behaviour is consistent with stochastic, broad-band variability without a preferred time-scale. In absolute terms, however, the INSANE wavelet amplitude at f≳10−3 M−1f \gtrsim10^{-3}\,M^{-1} is comparable to that of the SANE model and lower than that of the MAD model. The INSANE model is distinguished by the absence of a dominant low-frequency event, and not by an excess of short-time-scale power.

The SANE model is intermediate. Its amplitude is concentrated at the lowest accessible frequencies, with weak and fragmented power above 10−3 M−110^{-3}\,M^{-1}. For Rhigh=10R_{\rm high} = 10 the amplitude increases towards the end of the interval, at the onset of the flare at t≈26 400 Mt \approx26\,400\,M identified above, which lies close to the region affected by edge effects. The CWT and the HHT recover the same ordering of the three accretion states for both values of RhighR_{\rm high}.

Fast power fraction

The time–frequency maps of the previous subsection suggest that the accretion states differ in how their variability is distributed over time-scales, and not only in its amplitude. To quantify this for all models, we define the fast power fraction,

Ffast,I=E‾I,[P0,Pf]E‾I,[P0,Pb],(4)F_{\mathrm{fast},I} = \frac{\overline{E}_{I,[P_0,P_f]}}{\overline{E}_{I,[P_0,P_b]}} , \tag*{(4)}

where EI,[P1,P2](t)E_{I,[P_1,P_2]}(t) is the instantaneous variability energy of the total-intensity light curve in the period band P1P_1–P2P_2, the overline denotes a time average, and [P0,Pf][P_0,P_f] and [P0,Pb][P_0,P_b] are the fast and the broad band, respectively, with Pf<PbP_f < P_b. The numerator and the denominator are averaged over the same time samples. Ffast,IF_{\mathrm{fast},I} is therefore the fraction of the variability energy at periods P0P_0–PfP_f that is carried by periods shorter than PfP_f. It describes the shape of the variability spectrum, is independent of the flux normalization of the light curve, and is not a measure of the variability amplitude.

We evaluate the instantaneous variability energy with both methods. For the CWT,

EI,[P1,P2]CWT(t)=∫ln⁡P1ln⁡P2∣W(s,t)∣2s dln⁡P,(5)E^{\mathrm{CWT}}_{I,[P_1,P_2]}(t) = \int_{\ln P_1}^{\ln P_2} \frac{|W(s,t)|^2}{s}\,\mathrm{d}\ln P , \tag*{(5)}

where W(s,t)W(s,t) is the wavelet coefficient of the mean-subtracted light curve at scale ss, and P≃1.03 sP \simeq1.03\,s is the equivalent Fourier period of the Morlet wavelet with ω0=6\omega_0 = 6. This is the usual scale-averaged wavelet power, up to a constant that cancels in equation (4). For the HHT,

EI,[P1,P2]HHT(t)=12∑kak2(t) Θk(t),(6)E^{\mathrm{HHT}}_{I,[P_1,P_2]}(t) = \frac{1}{2}\sum_k a_k^2(t)\,\Theta_k(t) , \tag*{(6)}

where ak(t)a_k(t) and fk(t)f_k(t) are the instantaneous amplitude and frequency of the kkth IMF, and Θk(t)=1\Theta_k(t) = 1 if 1/P2≤fk(t)≤1/P11/P_2 \leq f_k(t) \leq1/P_1 and Θk(t)=0\Theta_k(t) = 0 otherwise. The factor 1/21/2 makes the contribution of a sinusoidal mode equal to its variance. Times at which no IMF lies in the band contribute zero energy to the average. The two estimators weight the signal differently, so their values of Ffast,IF_{\mathrm{fast},I} need not coincide.

We adopt P0=80 MP_0 = 80\,M and Pb=1500 MP_b = 1500\,M. The lower limit corresponds to eight samples per period at the cadence of 10 M10\,M. The upper limit is set by edge effects: after removing 2730 M2730\,M at each end of the light curves, a conservative allowance for the cone of influence at P=1500 MP = 1500\,M, the retained interval 17 73017\,730–27 270 M27\,270\,M still covers 6.4 periods of the slowest component. The boundary PfP_f is the only free choice. For a circular Keplerian orbit, P=2π(r3/2+a) MP = 2\pi(r^{3/2} + a)\,M, periods of 80, 200 and 1500 M\,M correspond to radii of r≈5r \approx5, 10 and 38 rg38\,r_g for both spins considered here. A boundary of a few hundred MM therefore separates the orbital time-scales of the inner ≈10 rg\approx10\,r_g from those of the flow at larger radii, and it is also of the order of the time-scales expected for turbulent fluctuations in that region. We do not attempt to identify PfP_f with a specific physical timescale in this work. We present results for Pf=200 MP_f = 200\,M as an illustrative choice, and we have repeated the measurement for every boundary between 200 and 500 M\,M in steps of 1 M1\,M.

Figure 7 shows Ffast,IF_{\mathrm{fast},I} for all 16 models. The MAD models have the lowest values, 3.9–7.6 per cent with the CWT and 3.2–6.4 per cent with the HHT, compared with 10.8–25.1 and 9.7–28.2 per cent for the INSANE models, and 7.9–19.0 and 9.0–15.3 per cent for the SANE models. The MAD range does not overlap that of either of the other states. At fixed spin, torus size and RhighR_{\rm high}, the MAD value is lower than the INSANE value by a factor of 1.7–8.9 and lower than the SANE value by a factor of 1.7–4.9. The two methods agree to within 5 percentage points for every model, with a median difference of 1.3 percentage points. A low Ffast,IF_{\mathrm{fast},I} does not imply weak variability. In the previous subsection, the MAD light curve was found to be dominated by a few large-amplitude events with time-scales ≳1000 M\gtrsim1000\,M, which we associated with magnetic-flux eruptions. Such events place most of the energy at the long-period end of the broad band and thereby lower the fast fraction. The MAD values also depend little on the electron-heating prescription, changing by at most 4 percentage points between Rhigh=10R_{\rm high} = 10 and 80.

Figure 7. Fraction of the 230-GHz intensity variability, in terms of fast power fraction Ffast,IF_{\mathrm{fast},I} (equation (4)), for all 16 models, computed from continuous wavelet transform (CWT; top) and Hilbert–Huang transform (HHT; bottom). The fast band is 80–200 MM and the broad band is 80–1500 MM. The variability energies are averaged over the time interval 17 730–27 270 MM, which excludes the parts of the 15 000–30 000 MM light curves affected by edge effects at a period of 1500 MM. Colours denote the accretion state (red: MAD; yellow: INSANE; blue: SANE), and filled and open bars show Rhigh=10R_{\mathrm{high}} = 10 and Rhigh=80R_{\mathrm{high}} = 80, respectively. The R20 and Rin6 models have black hole spin a=0.9375a = 0.9375 and the 0.5 models have a=0.5a = 0.5; no MAD_6 run is available. Note that the values are fractions of variability energy, not fractional rms amplitudes. Both panels share the same vertical scale.

The INSANE and SANE models are less clearly separated. The medians over the models of each state follow the hierarchy INSANE >> SANE >> MAD, with 15, 13 and 6 per cent for the CWT and 15, 14 and 5 per cent for the HHT, and the highest fractions of all, 19–28 per cent, are reached by INSANE_20. For the R20 models the INSANE value exceeds the SANE value by 6–13 percentage points. The ranges of the two states nevertheless overlap, and the hierarchy is not satisfied by every matched pair. For the a=0.5a = 0.5 models the two states are reversed with the CWT for both values of RhighR_{\rm high} (by 1.4 and 2.5 percentage points) and with the HHT for Rhigh=80R_{\rm high} = 80 (10 per cent for INSANE against 15 per cent for SANE), and for the Rin6 models with Rhigh=80R_{\rm high} = 80 they are reversed with the CWT (15 per cent for INSANE against 19 per cent for SANE) but not with the HHT (15 against 14 per cent). The dependence on RhighR_{\rm high} is also larger than in the MAD case for some of these models. The strongest example is SANE_6, for which Ffast,IF_{\mathrm{fast},I} increases from 8–9 per cent for $R_{\rm high} = 10 to 14–19 per cent for Rhigh=80R_{\mathrm{high}} = 80. This is consistent with the previous subsection, where the Rhigh=10R_{\mathrm{high}} = 10 light curve of this model was found to be dominated by a single slow dip and flare that is much weaker for Rhigh=80R_{\mathrm{high}} = 80.

These results are not specific to Pf=200 MP_{\mathrm{f}} = 200\,M. All fractions increase with PfP_{\mathrm{f}}, the state medians rising from 5–6 (MAD), 15 (INSANE) and 13–14 per cent (SANE) at Pf=200 MP_{\mathrm{f}} = 200\,M to 24–28, 44–48 and 40–43 per cent at 500 MM. The MAD value is the lowest in every matched comparison, with both methods, over the whole range 200–500 MM, and the hierarchy of the state medians is preserved throughout, although the INSANE and SANE medians come within 1 percentage point of each other at Pf≈290P_{\mathrm{f}} \approx290–340 MM with the CWT and at ≈360\approx360–410 MM with the HHT. The exceptions concern only the ordering of the INSANE and SANE models. INSANE_0.5 falls below SANE_0.5 at all boundaries with the CWT for both values of RhighR_{\mathrm{high}} and with the HHT for Rhigh=80R_{\mathrm{high}} = 80, and at Pf≥307 MP_{\mathrm{f}} \ge307\,M with the HHT for Rhigh=10R_{\mathrm{high}} = 10. INSANE_6 falls below SANE_6 for Rhigh=80R_{\mathrm{high}} = 80 at Pf≤319 MP_{\mathrm{f}} \le319\,M with the CWT, and INSANE_20 falls below SANE_20 for Rhigh=80R_{\mathrm{high}} = 80 at Pf>361 MP_{\mathrm{f}} > 361\,M with the HHT. The reversals reach at most 22 percentage points.

Summary and Discussion

By systematically comparing MAD, SANE, and INSANE GRMHD simulations coupled with polarized radiative transfer calculations, we identify two diagnostics that may help distinguish between different magnetization states and may provide useful probes of horizon-scale accretion flows such as M87*.

Our analysis shows that

  1. MAD model produces the most symmetric EVPA patterns, and SANE produces the least symmetric pattern, whereas INSANE lies between them.

  1. The multi-scale temporal behaviour and its variability distribution of the intensity differs qualitatively between the three states. While MAD is characterized by time-localized oscillatory variability associated with magnetic-flux eruptions and SANE shows weaker oscillatory behaviour, INSANE exhibits persistent broadband multi-frequency variability.

Together, these results demonstrate that combining polarization morphology with temporal analysis provides more information than static imaging alone.

Recent EHT observations of M87* and Sgr A* have demonstrated the importance of polarimetric measurements for distinguishing accretion-flow models [17] (Event Horizon Telescope Collaboration et al. 2025b). The Fourier decomposition provides a framework for quantifying the polarization morphology [31, 30, 32], typically using the quadrupolar mode β2\beta_2, which characterizes the dominant rotationally symmetric component of the EVPA field. Our analysis shows the importance of considering modes βm\beta_m (m≠2m \ne2) and the ratio of ∣βm∣/∣β2∣|\beta_m|/|\beta_2| (Long et al. 2026) as measures of deviations from the rotational symmetry. The retention of the βm\beta_m in the Fourier decomposition for quantifying polarization morphology is essential for probing the magnetic field structure in the accretion flow. More specifically, we have identified the ratio ∣β3∣/∣β2∣|\beta_3|/|\beta_2| showing the clearest separation between the three accretion states, whereas solely considering ∣β1∣|\beta_1|, ∣β2∣|\beta_2| (see e.g., in Event Horizon Telescope Collaboration et al. [19]) and their ratio is inadequate to fully capture the magnetic structure information. Our results, therefore, suggest that future multi-epoch EHT observations could use the temporal ∣β3∣/∣β2∣|\beta_3|/|\beta_2| as one of the sensitive observational probes of the magnetic structure of horizon-scale accretion flows.

The temporal and polarization properties of horizon-scale emission carry important information about the magnetic flux properties of the accretion flow. When future EHT-like observations improve in cadence and sensitivity, temporal measurements will become an increasingly important observational diagnostic. The different temporal signatures identified in this work likely reflect the distinct dynamical processes governing the accretion states in the three models. In the timing analysis of the MAD simulations, both the Continuous Wavelet Transform and Hilbert–Huang Transform identify localized oscillatory features associated with magnetic-flux eruption events consistently. The eruptions are a characteristic of strongly magnetized accretion flows, in which the accumulation of magnetic flux periodically disrupts the inflow before the excess magnetic flux is expelled. The localized oscillatory signatures, therefore, likely trace the dynamical reconfiguration of the inner accretion flow during these eruptive episodes. The SANE simulations exhibit weaker localized oscillatory behaviour, and the presence of identifiable localized oscillatory features depends on the adopted electron-heating prescription. This sensitivity suggests that, in weakly magnetized disks, the observed temporal variability depends not only on the underlying MHD turbulence but also on the emission region of synchrotron radiation. The INSANE simulations display a qualitatively different form of variability. Instead of isolated oscillatory events, they exhibit persistent broadband multi-frequency behaviour spanning a wide range of timescales. This behaviour is consistent with variability arising from fluctuations over multiple spatial and temporal scales, without a single dominant oscillatory component. A quantifiable measurement of the variability distribution across temporal scales, Ffast,IF_{\mathrm{fast},I}, identifies a hierarchy of INSANE > SANE > MAD for each radiative model. The agreement between the Continuous Wavelet Transform and Hilbert–Huang Transform suggests that the broadband variability is unlikely to be an artifact of a particular time–frequency decomposition. Although both methods remain susceptible to finite-duration effects and non-stationarity, they rely on fundamentally different approaches to signal decomposition, making it less likely that the observed broadband structure is produced by the analysis itself.

The analysis above for diagnosing the observables is particularly useful for the ongoing investigation of alternative magnetisation states of accretion flows in GRMHD simulations. Traditionally, black-hole accretion has largely been framed in terms of two limiting magnetisation states in terms of the SANE and MAD models (Narayan et al. 2012a). These two states are distinguished in GRMHD simulations, typically using physical quantities such as the magnetic flux, the mass accretion rate, the jet efficiency, and the dynamical influence of the magnetic field. However, recent numerical studies (e.g. Raha et al. 2026; Wong & Stone 2026) suggested that magnetized accretion flows occupy a broader range of physical states than this simple dichotomy. Intermediate magnetic-flux configurations, alternative magnetic-flux supply histories, and partially magnetically arrested accretion flows have all been shown to exhibit properties that cannot be fully characterized within the conventional framework in terms of the MAD/SANE model (Ressler et al. 2020; Manikantan et al. 2024; Raha et al. 2026). All these developments motivate the search for observable diagnostics capable of distinguishing such accretion states.

Our analysis addresses this need by identifying observable time-resolved polarization quantities that distinguish different magnetised accretion disk states. While the time-averaged polarization morphology places INSANE naturally between the MAD and SANE, its temporal behaviour of intensity and polarization is qualitatively distinct. In particular, the persistent broadband multi-frequency variability of intensity shown in both the Continuous Wavelet Transform and Hilbert–Huang Transform is absent from either MAD or SANE. These results demonstrate that temporal behaviour carries information beyond that contained in static images alone, providing an additional observational dimension for identifying partially arrested accretion flows. Such diagnostics are well-suited to future time-domain EHT-like observations of M87*, where long-term monitoring will allow the temporal evolution of polarization morphology and variability to be measured directly. They may also prove valuable for Sgr A*, whose observed variability has proven difficult to reconcile with either canonical MAD or SANE models alone ([11]). Although comparisons with observations should ultimately include propagation effects such as external Faraday rotation, their effect on the temporal polarization diagnostics explored here is unknown.

Note that in this study, we consider only thermal electron distributions, a limited set of black-hole spins and electron-heating prescriptions, and a viewing geometry appropriate for M87*. A more comprehensive work should consider a broader range of plasma models and magnetic-field configurations, and include non-thermal electron populations and various black-hole spins. Applying the same methodology to simulations for parameters appropriate for Sgr A* will help verify whether partially arrested accretion flows can simultaneously reproduce its observed variability and polarization properties. More generally, as an increasingly diverse range of magnetized accretion states emerges, continued exploration via GRMHD simulations, combined with polarisation morphology analysis and multi-scale temporal analysis, offers a promising route for connecting horizon-scale observations and black-hole accretion physics.

Acknowledgements

We thank Dr Alvina On (NCTS) for valuable discussions on time series analysis and Dr Kaye Jiale Li (UCL) on general aspects of the science. JSL is supported by UCL through a UCL-RES scholarship and by the TDLI visiting student program. AU acknowledges the support from the Research Fund for Excellent International PhD Students (grant No. W2442004). YM is supported by the National Key R&D Program of China (grant No. 2023YFE0101200), the National Natural Science Foundation of China (grant Nos. 12273022, 12511540053, and W2641030), and the Shanghai municipality orientation program of basic research for international scientists (grant No. 22JC1410600). KW is supported by a UKRI-STFC SA grant awarded to UCL-MSSL. JSL and KW acknowledge the support from the UCL Cosmoparticle Initiative. The simulations were conducted on the UCL-MSSL cluster and the Siyuan Mark-I HPC at Shanghai Jiao Tong University. This work has made use of the NASA Astrophysics Data System. We acknowledge using ChatGPT in polishing some of the sentences in the text.

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Details on the GRMHD simulation setups

The initial magnetic configuration is prescribed through a single dipolar loop with vector potential

AϕSANE=max⁡(qS,0),qS=ρ/ρmax⁡−0.2,A_{\phi}^{\mathrm{SANE}}=\max(q_{\mathrm{S}},0),\qquad q_{\mathrm{S}}=\rho/\rho_{\max}-0.2,
AϕINSANE=max⁡(qS2r3,0),A_{\phi}^{\mathrm{INSANE}}=\max(q_{\mathrm{S}}^{2}r^{3},0),
AϕMAD=max⁡(qM,0),A_{\phi}^{\mathrm{MAD}}=\max(q_{\mathrm{M}},0),
qM=ρρmax⁡(rrin)3sin⁡3θ e−r/400−0.2(7)q_{\mathrm{M}}=\frac{\rho}{\rho_{\max}}\left(\frac{r}{r_{\mathrm{in}}}\right)^{3}\sin^{3}\theta\,e^{-r/400}-0.2 \tag*{(7)}

and the four horizon quantities:

M˙=−∫ρur−g dθ dϕ,\dot{M}=-\int\rho u^{r}\sqrt{-g}\,\mathrm{d}\theta\,\mathrm{d}\phi,
ΦBH=12∫∣Br∣−g dθ dϕ,ϕBH=ΦBHM˙,\Phi_{\mathrm{BH}}=\frac{1}{2}\int\lvert B^{r}\rvert\sqrt{-g}\,\mathrm{d}\theta\,\mathrm{d}\phi,\qquad\phi_{\mathrm{BH}}=\frac{\Phi_{\mathrm{BH}}}{\sqrt{\dot{M}}},
E˙=∫T tr−g dθ dϕ,ηout=M˙−E˙M˙(8)\dot{E}=\int T^{r}_{\ t}\sqrt{-g}\,\mathrm{d}\theta\,\mathrm{d}\phi,\qquad\eta_{\mathrm{out}}=\frac{\dot{M}-\dot{E}}{\dot{M}} \tag*{(8)}

We use c=1c=1 and positive M˙\dot{M} and E˙\dot{E} for inward fluxes. The ideal-GRMHD stress–energy tensor is

T νμ=(ρ+ug+p+b2)uμuν+(p+b2/2)δ νμ−bμbν(9)T^{\mu}_{\ \nu}=(\rho+u_{g}+p+b^{2})u^{\mu}u_{\nu}+\left(p+b^{2}/2\right)\delta^{\mu}_{\ \nu}-b^{\mu}b_{\nu} \tag*{(9)}

where ugu_{g} is the internal-energy density, pp the gas pressure and bμb^{\mu} the comoving magnetic-field four-vector. Thus, E˙\dot{E} contains matter and electromagnetic energy fluxes, but no radiative flux because the simulations are non-radiative.

In Figure A1, we show the density profile of the initial torus configuration along with the corresponding magnetic field topologies for the MAD, INSANE, and SANE models. As mentioned previously, to ensure a consistent comparison among the different setups, we keep the size and structure of the initial torus identical for all three models with rin=20 rgr_{\rm in}=20\,r_g and rmax=41 rgr_{\rm max}=41\,r_g. For comparison with the commonly used configurations in the literature for the SANE model, we also consider an additional setup shown in the second row, where the torus parameters are rin=6 rgr_{\rm in}=6\,r_g and rmax=12 rgr_{\rm max}=12\,r_g for the SANE and INSANE cases.

Figure A1. (Updated figure with clarified figure captions) Initial configurations of MAD, INSANE, and SANE models (from left to right). For illustration purposes, we show the configurations for a torus with rin=20Rgr_{\mathrm{in}} = 20R_{\mathrm{g}}.

To assess whether the radiative transfer calculations are performed on statistically steady accretion flows, Figure A2 shows the time-averaged radial profiles of the mass accretion rate M˙\dot{M}, between 15000 MM and 30000 MM for all simulations. For the large-torus models, the accretion rate becomes approximately radius-independent within r≲70 Rgr\lesssim70\,R_g, indicating that the inner accretion flow has reached a quasi-steady state despite the continuing evolution of the outer disc. For the small-torus models, M˙\dot{M} remains nearly constant within r≲30 Rgr\lesssim30\,R_g, demonstrating that the inflow equilibrium extends throughout the entire simulated domain.

Figure A2. Time-averaged radial profiles of the accretion rate M˙\dot{M} of the MAD, INSANE, and SANE models between 15000 M15000\,M and 30000 M30000\,M. The left panel has a spin of 0.93750.9375, whereas the right has a spin of 0.50.5.

The polarized radiative transfer calculations presented in this work are performed using emission within a radius of 17 RgR_g. This emitting region, therefore, lies well inside the quasi-steady inflow region for both the large- and small-torus simulations, ensuring that the synthetic images and light curves are not significantly affected by the non-equilibrium evolution of the outer accretion flow.

Polarimetric properties

To further characterise the statistical properties of the simulated images, we analyse the polarization observables across the full ensemble of snapshots and present their distributions as histograms in Figure B1. The quantities considered include the amplitude and phase of the quadrupolar polarization mode, ∣β2∣|\beta_2| and arg⁡(β2)\arg(\beta_2); circular and linear polarization measures, ∣vnet∣|v_{\rm net}|, ⟨∣m∣⟩\langle|m|\rangle, and ∣mnet∣|m_{\rm net}|; as well as the total intensity II.

The MAD models generally produce higher linear polarization fractions compared with the INSANE and SANE models. In addition, the MAD configuration exhibits a stronger sensitivity to the electron heating parameter RhighR_{\rm high}. In contrast, the INSANE and SANE models show comparatively weaker dependence of the linear polarization properties on the heating prescription.

For circular polarization, however, the behaviour is reversed. The INSANE and SANE models show noticeable differences in ∣vnet∣|v_{\rm net}| between the two heating prescriptions (Rhigh=10R_{\rm high}=10 and Rhigh=80R_{\rm high}=80), whereas the MAD models produce similar levels of circular polarization for both heating choices. This suggests that circular polarization is more strongly influenced by plasma thermodynamics in weakly magnetized flows, while in strongly magnetized MAD configurations, the magnetic field geometry dominates the polarization structure.

Complete Time-series analysis

The main text focuses on the INSANE_6 and SANE_6 models because their weaker secular evolution makes them better suited for quantitative time-series analysis. For completeness, we present here the corresponding variability and time–frequency analyses for the large-torus simulations (INSANE_20, SANE_20, INSANE_0.5, and SANE_0.5), together with the MAD_0.5 model.

The intensity light curves of all models are shown in Figure C1. All of them are normalized to the same median flux, but the large-torus INSANE and SANE light curves decline throughout the analysed interval. Between the first and the last quarter of the interval, the mean intensity falls by a factor of ≈10\approx10–15 for INSANE_20 and INSANE_0.5, by a factor of ≈6\approx6–9 for SANE_20, and by a factor of ≈3\approx3 for SANE_0.5, whereas it changes by only a factor of ≈1.5\approx1.5 for MAD_0.5. The fractional rms variability of the large-torus INSANE light curves, 88–95 per cent, is therefore dominated by this decline and not by fluctuations about a steady state.

Light curves of the 230 GHz total intensity $I$ for MAD, INSANE, and SANE models

Figure C1. Light curves of the 230 GHz total intensity II. MAD, INSANE, and SANE models are marked as red, purple, and blue. The columns show, from left to right, the small-torus models with a=0.9375a = 0.9375, the large-torus models with a=0.9375a = 0.9375, and the models with a=0.5a = 0.5. The medians of the light curves are plotted as the horizontal dashed lines. The upper panels use the heating prescription of Rlow=1R_{\mathrm{low}} = 1 and Rhigh=10R_{\mathrm{high}} = 10, and the lower panels use Rlow=1R_{\mathrm{low}} = 1 and Rhigh=80R_{\mathrm{high}} = 80.

The corresponding CWT and HHT analyses are shown in Figure C2. The secular decline affects them in two ways. First, it concentrates the amplitude at the lowest frequencies. The scalograms of INSANE_20 and SANE_20 are strongest along the low-frequency boundary of the cone of influence, and their time-averaged wavelet spectra are still rising at the lowest accessible frequency. In the HHT, the trend is carried by the residue together with the IMFs slower than the lowest frequency accessible to the CWT (f<1.9×10−4 M−1f<1.9\times10^{-4}\,M^{-1}), which account for ≳85\gtrsim85 per cent of the variance of the large-torus INSANE light curves and for 70–91 per cent of that of the large-torus SANE light curves. Secondly, the wavelet and Hilbert amplitudes are measured in Jy and therefore scale with the flux level. In the INSANE_20 and INSANE_0.5 panels almost all of the structure lies at t≲21000 Mt\lesssim21000\,M, where the flux is highest, with maxima at f≈(3f\approx(3–6)×10−4 M−16)\times10^{-4}\,M^{-1} and t≈17000t\approx17000–20000 MM, and the panels are nearly empty at later times. This asymmetry follows the decline of the flux and does not by itself indicate a change in the character of the variability.

Figure C2. As Figure 5, but for the large-torus models. The upper two sets of panels show the CWT scaleograms and Hilbert–Huang spectra of the a=0.9375a = 0.9375 models (MAD_20, INSANE_20 and SANE_20; the MAD panels are the same as in Figure 5), and the lower two sets show those of the a=0.5a = 0.5 models (MAD_0.5, INSANE_0.5 and SANE_0.5). Each panel has its own colour scale.

The MAD_0.5 model evolves the least and behaves like the MAD model of the main text. Its wavelet amplitude is dominated by one episode at t≈24000t\approx24000–26000 MM and f≈(3f\approx(3–4)×10−4 M−14)\times10^{-4}\,M^{-1} (time-scales of ≈2300\approx2300–3600 MM), which corresponds to a flare of the light curve to ≈1.2\approx1.2–1.4 Jy at t≈25500 Mt\approx25500\,M followed by a rapid decline. The Hilbert spectra show their brightest ridge at the same time, and the result is insensitive to RhighR_{\rm high}. The characteristic time-scale of this episode is about twice that of the corresponding episodes in the a=0.9375a=0.9375 MAD model.

This paper has been typeset from a \TeX/\LaTeX file prepared by the author.

References

  1. [7]Event Horizon Telescope Collaboration et al., 2019a, ApJ, 875, L1
  2. [8]Event Horizon Telescope Collaboration et al., 2019b, ApJ, 875, L4
  3. [10]Event Horizon Telescope Collaboration et al., 2019d, ApJ, 875, L6
  4. [11]Event Horizon Telescope Collaboration et al., 2021a, ApJ, 910, L12
  5. [17]Event Horizon Telescope Collaboration et al., 2024b, ApJ, 964, L26
  6. [19]Event Horizon Telescope Collaboration et al., 2025b, A&A, 704, A91
  7. [23]Liska M. T. P., et al., 2022, ApJS, 263, 26
  8. [30]Palumbo D. C. M., Wong G. N., 2022, ApJ, 929, 49
  9. [31]Palumbo D. C. M., Johnson M. D., Doeleman S. S., 2020, ApJ, 894, 156
  10. [32]Palumbo D. C. M., Wong G. N., Chael A., Johnson M. D., 2023, ApJ, 952, L31

Paper details

vlnc:W_rtsz8ajv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

ChatGPT

View statement
Looking for related papers…
Contents