Hard X-ray Emission in AU Mic Flares: A Minor Contributor to Planetary Atmospheric Escape

TL;DR

This study investigates the contribution of hard X-rays to exoplanetary atmospheric escape by analyzing two major flares on AU Mic with quasi-simultaneous NuSTAR, Swift, and EP observations. It establishes energy-band definitions, derives SXR–EUV and HXR–SXR scaling relations, and performs time-resolved spectral fits to quantify the partition of radiative energy among EUV, SXR, and HXR bands. The results show that HXR contributes only a few percent to the total radiative budget, with flares dominated by thermal emission and a potential high-energy tail in one event. The findings support XUV-driven atmospheric escape as the primary mechanism for AU Mic–like systems, while acknowledging uncertainties in HXR escape efficiencies and the value of extended high-energy observations for future constraints.

Abstract

Stellar flares are potent drivers of atmospheric evolution on orbiting exoplanets, primarily through extreme ultraviolet (EUV) and soft X-ray (XUV) irradiation. However, the contribution of hard X-rays (HXR; 3--20 keV)-which penetrate deeper into planetary atmospheres-to mass loss and particle acceleration has remained poorly understood. To quantify the HXR share of the total radiative budget, we conducted quasi-simultaneous observations of the active M-dwarf AU Mic using NuSTAR, Swift, and the Einstein Probe. Our analysis detected two major flares, and we performed an empirical check by deriving a quiescent-phase soft X-ray (SXR; 0.3--3 keV)-HXR relation and then applying it to the flares. By combining this with the quiescent coronal SXR-EUV relations conversion of J. Sanz-Forcada et al. (2011), we computed the total high-energy flux (EUV + SXR + HXR) and assessed the relative role of HXR in atmospheric escape. We find that HXR accounts for only a few percent of the total radiative energy budget during both quiescent and flaring states. While a high-energy spectral tail is detected in the second flare, time-resolved spectroscopy reveals a dominant chromospheric-evaporation signature, indicating that the flare energetics are primarily thermal.

Figures (9)

• Figure 1: Count rates from Swift, NuSTAR, and EP observations of AU Mic. The main panel shows background-subtracted Swift (0.3–10 keV), NuSTAR FPMA/B average (3–20 keV), and EP FXT-A/B average (0.5–10 keV) count rates, all binned at 3600 s. The 0–0.15 cps range that is expanded in the inset, which displays the NuSTAR data in greater detail. Error bars denote 1$\sigma$ uncertainties.
• Figure 2: Top: Swift-XRT light curve, with the quiescent level (red dashed line) and ±3$\sigma$ bounds (dotted lines) estimated via one iteration of the 2$\sigma$-clipping method. Bottom: NuSTAR light curve with selected flare intervals shaded in gray. The red-shaded regions indicate intervals overlapping with the EP observations. Quiescent level and ±3$\sigma$ band are overlaid using the same sigma-clipping procedure, with seven iterations. Bin size = 3600 s.
• Figure 3: Joint spectral fitting results. Black points: FPM-A data; red points: FPM-B data; green points: FXT-A data; blue points: FXT-B data. The top panel shows the observed spectra and the best-fit folded model, while the bottom panel displays the residuals in units of $\sigma$.
• Figure 4: Correlation between HXR (NuSTAR, 3–20 keV) and SXR (corrected EP, 0.3–3 keV) luminosities. The data are binned with a fixed bin size of 240 s to minimize statistical noise from NuSTAR's relatively low count rates. NuSTAR time bins with zero count rates were excluded to avoid suppression of correlation strength. Additionally, a small number of bins exhibiting anomalously high or low luminosities, often due to edge effects or dominated by upper limits with large uncertainties, were excluded from the regression fits. These excluded points typically had extremely weak or absent hard X-ray signals. The blue line shows the best-fit least squares regression (LSR), while the red line represents the orthogonal distance regression (ODR) which accounts for errors in both axes. The purple line represents the median of the Bayesian linear regression posterior, obtained via Markov Chain Monte Carlo sampling, with the shaded region showing the 90% posterior confidence interval. Corresponding fit parameters are listed in Table \ref{['tab:scaling_laws']}. The SXR–HXR luminosity scaling we derive spans only one order of magnitude in luminosity; expanding the dynamic range is required to generalize the slope.
• Figure 5: Top: NuSTAR spectrum of Flare 1 with the best‐fit absorbed apec model overlaid. Bottom: NuSTAR spectrum of Flare 2 with the best‐fit absorbed apec model overlaid. Black points denote FPM-A data and red points denote FPM-B data.