XRISM Spectroscopy of Variable Accretion-driven Disk Winds in NGC 4151: When, Where, and How Fast Outflows are Launched

The paper reports on a study of variable accretion-driven disk winds in the Seyfert-1 AGN NGC 4151, using 14 XRISM/Resolve observations totaling 0.9 Ms of exposure over 395 days. The analysis is organized around three questions: when, where, and how the winds are launched.

When winds are launched: The spectra were grouped into seven windows based on local variability phases (Flares, Postflares, Afters, Hard-dips) and global hardness-intensity states (High-Soft, Low-Soft, Low-Hard). The results show that two persistent warm absorber (WA) components are always present, with outflow velocities of 100–1000 km/s. In contrast, very fast outflows (VFOs; v 10 3–10 4 km/s) and ultra-fast outflows (UFOs; v 0.033–0.33c) are transient. The Postflare spectrum shows the richest structure, containing two WAs, one VFO, and two UFOs, while the High-Soft spectrum lacks significant fast winds. The fast winds are strongest approximately 10 ks after the peak of flares and during low-flux periods. This 10 ks timescale is among the shortest flare–wind response timescales reported in an AGN.

Where winds are located: The characteristic radius and volume filling factor constraints suggest that WAs are unlikely to be simple freely escaping outflows; they are more consistent with slow, stratified, partially failed or circulating material at radii of 10 4–10 5 GM/c squared, within the inner wall of the torus. For the Postflare UFO and VFO components, the 10 ks response corresponds to a light-crossing distance of 60 GM/c squared, interpreted as the most compact causal limiting scale. If the Postflare VFOs and UFOs are associated with this compact scale, their ionization parameters imply local densities of n 5.8 × 10 10 cm-3 for the VFO, n 1.8 × 10 11 cm-3 for the slower UFO, and n 3.4 × 10 11 cm-3 for the faster UFO, requiring a clumpy or filamentary absorber.

How winds are launched: The absorption measure distribution (AMD) slope of the fastest outflows and their extreme outflow momentum rates are broadly consistent with a magnetically driven disk-wind structure. The AMD implies a density profile of n(r) ∝ r-1.5 in the large-scale interpretation, consistent with a magnetocentrifugal disk-wind structure. The VFOs and UFOs often carry momentum rates larger than the available radiative momentum flux, indicating that radiation pressure alone is unlikely to provide the main driving force. The flattened velocity–ionization relations, transient re-emission, and blue-shifted Fe-K emission point to a wind that is locally clumpy, geometrically complex, and strongly phase dependent.

Key quantitative results:

• The intrinsic hot obscurer column density is anti-correlated with CINDICITY (a scalar combining count rate and hardness), being lowest in the bright-soft High-Soft state (NH 1.29 × 10 23 cm-2) and highest in the Low-Hard state (NH 2.01 × 10 23 cm-2).

• The UFOs dominate the kinetic power of the outflow. In nominal estimates, all UFO phases exceed the 0.5% L Edd feedback threshold, while under the most conservative filling-factor corrections they remain marginally close to this level, at values of order 0.3% L Edd.

• The WAs carry the lowest mass fluxes and kinetic powers, staying far below the canonical feedback threshold of 0.5% L Edd.

• The Postflare faster UFO reaches E k ≲ 0.2% L Edd under compact-scale filling factors.

The paper concludes that feedback-level winds in sub-Eddington NGC 4151 are intermittent rather than continuous, and that the fast outflows are most plausibly triggered or made observable by flare-driven magnetic activity, possibly through magnetic reconnection or related coronal processes that lift dense gas from the inner disk.

Improvements to AI Systems Based on This Paper:

1. Temporal Variability-Aware Spectral Classification

• Improvement: Train an AI model to classify AGN spectra not just by instantaneous flux or hardness, but by local variability phase (e.g., Flare, Postflare, Aft, Hard-dip) and global state (High-Soft, Low-Soft, Low-Hard), using time-series features (e.g., flare peak timing, decay slope, count-rate derivatives).

• Capability: The improved AI can predict when fast outflows (VFOs/UFOs) are likely to appear or disappear in future observations, enabling dynamic scheduling of X-ray telescopes to catch transient winds.

2. Causal Time-Lag Inference for Outflow Launch Sites

• Improvement: Implement a neural network that learns the cross-correlation between X-ray flare light curves and absorption-line variability, explicitly modeling a 10 ks lag as a physical parameter (light-crossing radius).

• Capability: The AI can automatically estimate the launch radius of outflows from time-series data alone, without requiring detailed spectral fitting, and flag sources where the lag implies compact (<100 GM/c2) origins.

3. Multi-Component Ionization-Density Solver

• Improvement: Build a differentiable model that, given observed ionization parameters and inferred radii (from time lags), solves for local gas densities (n) and volume filling factors, incorporating clumpiness priors (e.g., log-normal density distributions).

• Capability: The AI can directly output density profiles and clumpiness factors for each wind component, distinguishing between continuous outflows and filamentary/cloudy absorbers—useful for classifying wind geometry in large AGN surveys.

4. Momentum-Budget Classifier for Driving Mechanisms

• Improvement: Train a classifier that compares measured outflow momentum rates (Ṗ) to the available radiative momentum flux (L Edd/c) and magnetic energy density proxies, using the ratio as a feature.

• Capability: The AI can automatically label whether a wind is radiatively driven, magnetically driven, or hybrid, based on spectral and timing data, reducing manual astrophysical interpretation.

5. Feedback-Efficiency Estimator with Filling-Factor Uncertainty

• Improvement: Develop a Bayesian neural network that estimates kinetic power (Ė k) as a fraction of L Edd, explicitly propagating uncertainties in filling factor, distance, and ionization—including conservative lower bounds (e.g., compact-scale clumping).

• Capability: The AI can output probabilistic feedback classifications (e.g., above 0.5% L Edd vs. marginally below) for each transient phase, helping prioritize which AGN warrant multi-wavelength follow-up for feedback studies.

6. Absorption Measure Distribution (AMD) Slope Predictor

• Improvement: Use a regression model to infer the AMD slope (e.g., n(r) ∝ r-1.5) from a set of absorption-line equivalent widths and ionization parameters, with a physics-informed loss that penalizes non-physical density profiles.

• Capability: The AI can reconstruct the radial density structure of disk winds from sparse spectral data, enabling comparisons to magnetocentrifugal wind models across many AGN without full radiative transfer simulations.

7. Transient Wind Trigger Detector

• Improvement: Implement a recurrent neural network (e.g., LSTM) on X-ray light curves and hardness ratios to predict the onset of fast winds (VFO/UFO) up to 10 ks before they appear in absorption spectra.

• Capability: The AI can serve as an early-warning system for observatories, triggering high-cadence spectral mode or coordinated multi-wavelength observations during flare-driven wind events.

8. Clumpy Absorber Simulator for Spectral Fitting

• Improvement: Create a generative model that produces synthetic absorption spectra from clumpy, filamentary outflows (with random cloud distributions) and trains a contrastive network to distinguish clumpy vs. smooth outflows.

• Capability: The AI can automatically flag AGN where a smooth outflow model fails and a clumpy model is required, improving the accuracy of mass and energy outflow rate estimates in automated pipelines.

9. State-Dependent Obscuration Predictor

• Improvement: Train a model to predict the intrinsic hot obscurer column density (NH) from global hardness-intensity state and recent flare history, using the observed anti-correlation with CINDICITY.

• Capability: The AI can estimate unobserved NH values for archival data or predict future obscuration, aiding in de-biasing AGN population studies and correcting luminosity functions.

10. Multi-Epoch Outflow Evolution Mapper

• Improvement: Build a transformer-based model that ingests a time series of spectra (e.g., 14 epochs) and outputs a temporal wind state diagram (presence/absence, velocity, ionization of each component) with uncertainty.

• Capability: The AI can automatically produce a complete dynamical history of outflows for any AGN with repeated observations, enabling statistical studies of wind intermittency and its correlation with accretion state—directly applicable to future XRISM/Athena surveys.

X-ray observations probe the inner accretion flow within active galactic nuclei, revealing the highest gas column densities and fastest winds. The most diverse winds yet revealed with the Resolve calorimeter spectrometer aboard XRISM are found in NGC 4151, a nearby Seyfert-1 AGN that may qualify as a ``changing-look'' source (CLAGN). Herein, we report on wind variability in 14 XRISM observations of NGC 4151, summing to 0.9 Ms of exposure over a period of 395 days. We examined the dependence of key wind parameters on hardness and intensity selections, and as a function of time relative to flaring and dip events. The results suggest a globally organized but locally complex wind structure. Slow ``warm absorber'' components (WAs; v about 100-1000) are always observed and likely represent failed winds at radius of 10 4 - 10 5 GM/c squared, within the inner wall of the torus. In contrast, ``very fast'' and ``ultra-fast'' outflows (VFOs and UFOs; v about 10 3-10 4, v about 0.033-0.33 c) are strongest 10 ks after the peak of flares, and during periods with low flux. Ten kiloseconds is among the shortest flare--wind response timescales reported in an AGN, suggesting that the winds are observed close to the launching site. The absorption measure distribution (AMD) and the large outflow momentum rates suggest that the high-velocity flows visible in the Fe K band are magnetically driven, while locally clumpy, likely owing to radiation pressure; one or both of these mechanisms may be enhanced following a flare and most visible during low-flux windows.