Detection of a Millimeter Flare from Proxima Centauri

Figure 1 (leftmost panel) shows the natural weight continuum image produced by combining both ALMA 12 m array observations. With natural weighting, the rms noise in the image is 9.5 μJy beam⁻¹, and the beam size is 099 × 087 (1.3 × 1.1 au, position angle =56°). An unresolved source is detected at the center of the image with a significance of 10σ. Fitting a point source model to the visibilities using the CASA task uvmodelfit yields a total flux density of 101 ± 9 μJy, consistent with the result of 106 ± 12 μJy reported by Anglada et al. (2017). Any difference in flux density likely results from the fact that we fit to the millimeter visibilities, while Anglada et al. (2017) performed all of their fits in the image domain. The two middle panels of Figure 1 show the images that result from imaging each observation individually. The rms noise is 13 μJy beam⁻¹ and 14 μJy beam⁻¹ for SB 1 and SB 2, respectively. Both images show a central point source with measured flux density of 110 ± 19 μJy and 94.1 ± 8.0 μJy, respectively, consistent with each other within the uncertainties. For comparison, the expected photospheric flux at 1.3 mm is 74 ± 4 μJy, obtained by extrapolating the emission model derived by Ribas et al. (2017) through a fit to the stellar spectral energy distribution (SED).

Zoom In Zoom Out Reset image size

Both ALMA 12 m SBs were executed on 2017 April 25 between 04:32:31.2 and 06:50:33.4 UTC. A total of 16 (eight in each SB) on-source integrations, or "scans," were performed during this period, each with a duration of 6.58 minutes. We fit point source models to the millimeter visibilities in each scan independently to constrain any variability in the point source flux density. The resulting light curve is shown in Figure 1 (right panel). We note that the final scan in each SB was only 1.01 minutes in duration, resulting in substantially larger uncertainty. There is some variation in flux density between successive scans, ranging between 157 ± 19 μJy at maximum to 50.4 ± 48.2 μJy at minimum. The standard deviation for these data is σ = 29.1 μJy, and none of the individual scans have a flux density that differs from the mean by more than 3σ. We calculate the normalized "excess variance," ${\sigma }_{\mathrm{rms}}^{2}\,=\tfrac{1}{N{\mu }^{2}}{\sum }_{i=1}^{N}[{({X}_{i}-\mu )}^{2}-{\sigma }_{i}^{2}]$ following Nadra et al. (1997), where N is the number of data points, μ is the unweighted mean, and X_i are the measured flux densities for each scan with errors σ_i. For these data, σ_rms² = 0.017 ± 0.030.

The observations include two linear polarizations (XX and YY) and span ∼18 GHz in total frequency between 224 and 242 GHz, which allows us to determine a lower limit on the fractional linear polarization and a spectral index for the detected point source. Combining both observations together, but imaging the XX and YY polarizations separately, yields total flux densities of F_XX = 105 ± 7 μJy and F_YY = 96.2 ±10.0 μJy. Given this, we can calculate the magnitude of the Stokes $Q=\langle {E}_{X}^{2}\rangle -\langle {E}_{Y}^{2}\rangle$ parameter as a fraction of the Stokes $I=\langle {E}_{X}^{2}\rangle +\langle {E}_{Y}^{2}\rangle$ to be $| Q/I| =0.09\pm 0.12$, consistent with zero. As there are no polarization calibration and cross products available for these observations, $| Q/I|$ provides only a lower limit on the fractional linear polarization. To calculate the spectral index, we determine the flux density of the point source detected in the combined image for the 225+227 GHz and 239+241 GHz spectral windows separately, yielding values of 100 ± 13 μJy and 122 ± 14 μJy, respectively. The resulting spectral index is α = 2.58 ± 2.05, where F_ν ∝ ν^α. Within the significant errors, this result is consistent with blackbody emission in the Rayleigh–Jeans regime as expected for a quiescent stellar photosphere.

Figure 2 (leftmost panel) shows the natural weight continuum image resulting from combining all 13 observations taken with the ACA between 2017 January 21 and 2017 March 24. The rms noise is 47 μJy beam⁻¹, and the beam size is 73 × 55 (9.5 × 7.2 au, position angle =−79°). An unresolved point source with flux density 334 ± 48 μJy (7σ) is detected, in agreement with the value of 340 ± 60 μJy from Anglada et al. (2017), with the small difference again likely stemming from the choice in fitting to the visibilities rather than the image. As with the 12 m data, we examined all 13 observations individually to explore the possibility of variability in the point source flux. No sources above 3σ were detected in any of the first 12 observations (even when binned in 30 s intervals that should reveal short flares), and we show the natural weight image produced by averaging these data sets in Figure 2 (left middle panel). The image is essentially noise-like; however, the ∼2σ central peak could indicate the occurrence of small flares below the detection threshold of these observations. Given the rms noise in this image, 68 μJy beam⁻¹, a point source with the measured ALMA 12 m flux density of ∼100 μJy, would not have been detected. Figure 2 (right middle panel) shows an image of only the final observations executed on 2017 March 24 from 06:36:30.1 to 08:10:27.2 UTC with an rms noise level of 150 μJy beam⁻¹. In this final image, a bright central source is clearly seen at the stellar position with a flux density of 1.17 ± 0.10 mJy (9σ).

Zoom In Zoom Out Reset image size

It is clear from this analysis that the point source seen in the combined ACA image is dominated by a transient event that occurred during the final observation on 2017 March 24. During this complete observation, a total of seven on-source scans of 6.58 minutes each were executed. The central point source is only visible in the second to last scan. In order to better constrain the time variable nature of this observed source, we fit models to the visibilities in this scan in 2 s intervals using uvmodelfit in CASA. These fits are point-like with respect to the size of the synthesized beam, 73 × 55 (9. 5 × 7.2 au). The resulting light curve is shown in Figure 2 (right panel). Several small peaks in flux density occur starting at JD 2457836.8349 followed by a steep rise in brightness at JD 2457836.8355 and an exponential fall-off. This steep rise and more gradual fall-off is characteristic of flares from Proxima Centauri observed at other wavelengths (e.g., Kowalski et al. 2016). At peak brightness, the flux density reaches 100 ± 4 mJy, or a luminosity of 2.04 ± 0.15 ×10¹⁴ erg s⁻¹ Hz⁻¹ at 1.3 pc, a factor of nearly 1000 times brighter than during quiescence as measured in the ALMA 12 m observations. The complete transient event lasts for approximately 1 minute in total (FWHM of 30 s) before the source fades away below the noise level. We do not detect any variability in the calibrator flux densities when imaged on similar timescales.

As with the 12 m data, we can examine whether there is any change in the spectral index and polarization during the detected flare. Figure 3 shows the flux density, spectral index as a function of frequency (α), and lower limit on the fractional linear polarization ($| Q/I|$) during the course of the flare binned in 5 s intervals to improve signal-to-noise. As above, we calculate the spectral index between the 225+227 GHz and 239+241 GHz spectral windows (plotted separately in Figure 3 as the green and purple thick lines, respectively). During the flare, the spectral index becomes negative, F_ν ∝ ν^α with a minimum value of α = −1.77 ± 0.45 and an average value of α = −1.64 ± 0.21. During the initial steep rise in flux density, the lower limit on the fractional linear polarization given by $| Q/I| =-0.26\pm 0.02$. At the peak, the sign flips to positive values, $| Q/I| =0.19\pm 0.02$, before returning to zero as the flux density declines. Both α and $| Q/I|$ differ significantly from the "quiescent" values determined from the 12 m ALMA data by >5σ, indicating a change in the source of the emission during the flaring event. As we only have a lower limit, the origin of the changing $| Q/I|$ value is unclear from these observations. Slee et al. (2003) detected significant right circular polarization during a flare from Proxima Centauri at 1.38 GHz, which they attribute to either plasma emission or electron–cyclotron maser emission. The signal-to-noise of the "pre-flare" at JD 2457836.8349 is not high enough to allow us to do the same analysis for this smaller event.

Zoom In Zoom Out Reset image size

Proxima Centauri is known to undergo frequent X-ray flares at the rate of one large event (10 times quiescent level) every few days (Haisch et al. 1983; Güdel et al. 2002, 2004; Fuhrmeister et al. 2011; Kowalski et al. 2016). Güdel et al. (2004) observed the star for 65 ks with XMM-Newton and caught a large X-ray flare with a peak luminosity of 3.9 × 10²⁸ erg s⁻¹ and total X-ray energy of 1.5 × 10³² erg that lasted for ∼10 minutes. Low-level X-ray variability from small flares is observed continually. Visual wavelength flares occur at a similar rate of several per day (Davenport et al. 2016). By extrapolating the flare energy frequency distribution, Davenport et al. (2016) estimated that Proxima Centauri might display about eight visual flares each year with energy >10³³ erg.

Stellar flares are typically associated with magnetic reconnection events. Late-type stars such as Proxima Centauri are fully convective and have high surface magnetic fields (Reiners & Basri 2008). X-ray observations suggest that Proxima Centauri's coronal flares are similar in temperature and density to those on the Sun, while also exhibiting the Neupert effect (Fuhrmeister et al. 2011). In the standard model of solar and stellar flares, a release of magnetic energy accelerates electrons that then impinge on hot ionized gas in the surrounding plasma. The flare produces emission across the entire electromagnetic spectrum: thermal X-rays from gas at ∼10⁷ K, thermal bremsstrahlung (free–free) radiation at optical to radio wavelengths, and gyrosynchrotron emission in the radio. The observed slow decay depends on the cooling mechanism, and the timescale is typically set by the optical depth of the flare region. In principle, the shape and energy of the flare can be used to measure the emission mechanism, electron density, and size/optical depth of the emitting region.

Flares from M dwarf stars have not previously been well studied at (sub)millimeter wavelengths. However, RS CVn binaries have been known to exhibit large millimeter flares (Beasley & Bastian 1998; Brown & Brown 2006; Massi et al. 2006). Millimeter emission has also been detected in solar flares. These events are associated with extreme X-ray emission (class X6 and above), have durations of minutes, and have positive spectral indices with frequency that range from 0.3 to 5 (Krucker et al. 2013). Even for the Sun, the origin of this emission is unclear because both free–free and gyrosynchrotron emission fail to reproduce the correct slope and X-ray fluxes given reasonable assumptions for the size and optical depth of the emission region. Two possible emission mechanisms are synchrotron emission from ultra-relativistic electrons or an extreme-IR thermal extension of the heated chromosphere at 10,000 K. The maximum (sub)millimeter flux density of a solar flare observed by Krucker et al. (2013) is ∼7 × 10⁸ Jy corresponding to a luminosity of ∼2 × 10¹³ erg s⁻¹ Hz⁻¹. The peak luminosity of the observed Proxima Centauri flare, (2.04 ± 0.15) × 10¹⁴ erg s⁻¹ Hz⁻¹, is a factor of 10 times larger. Furthermore, the observed spectral index, α = −1.77 ±0.45, is quite steep (modulo the significant uncertainty) and contradictory to the observations of solar flares that instead suggest a rising (positive) spectral index (α > 2). This spectral index may indicate that we are seeing the optically thin part of the gyrosynchrotron spectrum, where the power-law index of non-thermal electrons, δ, can be inferred from α: α =1.22–0.9δ (Dulk 1985; Osten et al. 2016). From the observed spectral index, we can determine δ = 3.32 ± 0.83, within the range of 2.2 ≤ δ ≤ 3.9 expected for relatively hard radio spectra. Given the anomalous nature of this flaring event and the fact that observations are only available in a narrow wavelength range, it is difficult to definitively constrain the nature of the emission mechanism. Future observations at longer radio (∼1 cm) wavelengths, near the expected peak of gyrosynchrotron emission, with the frequency resolution possible with the Australia Telescope Compact Array (ATCA) or ACA along with coincident X-ray measurements are needed to better understand this new flaring regime.

Anglada et al. (2017) proposed a system of three dust belts to explain the ALMA 12 m and ACA observations of Proxima Centauri: (1) warm dust at ∼0.4 au, (2) a cold belt from 1 to 4 au, and (3) an outer belt at ∼30 au. They note that the warm dust (1) and outer belt (3) are marginal detections, while only the cold belt (2) is detected with confidence. The warm dust is inferred from the ∼30 μJy difference between the stellar photosphere and the total flux density of the 12 m image. The cold belt is inferred from the ∼200 μJy apparent flux difference between the expected photospheric contribution and the total flux density measured in the ACA image. The outer belt position and geometry are determined by averaging the observed millimeter emission in elliptical annuli centered on the star. Given that our reanalysis of the data reveals that Proxima Centauri experienced a significant flaring event during the ACA observations, the hypothesis of dust emission from at least belts (1) and (2) must be re-examined.

It is clear from our analysis that the entirety of the 1.3 mm emission detected by the ACA results from the short duration stellar flare. No emission is detected above a 3σ level during quiescence (see Figure 2). This flare reaches a peak flux density of 100 ± 4 mJy, but lasts for only ∼1 minute. When averaged together with the remaining ∼19 hr of non-detections, this transient event appeared as the ∼340 μJy source detected by Anglada et al. (2017). Without the detection of a quiescent excess, there is no need to posit the cold belt at 1–4 au.

Given the high level of stellar activity displayed by Proxima Centauri, the ∼30 μJy excess detected by the 12 m observations could also be plausibly explained by stellar emission. The quiescent X-ray luminosity of the star varies by at least a factor of a few, and small X-ray and optical flares are observed frequently (Güdel et al. 2004; Fuhrmeister et al. 2011). Thus, it is reasonable to assume that Proxima Centauri may also undergo many small radio flares. If we assume that small flares have a flat spectrum at millimeter wavelengths, which likely overestimates the luminosity of the emission if it has the ν² spectral index of thermal bremsstrahlung, then the 1.3 mm excess of ∼30 μJy measured by the 12 m array yields a quiescent millimeter luminosity of 10²¹ erg s⁻¹, or less than 10⁻⁹ times the bolometric luminosity of the star. The 2σ peak seen in the first 12 ACA observations (Figure 2) at the expected stellar position could also indicate some weak flare emission during these observations. If the observed 12 m array excess is indeed due to low-level flaring activity, then there is no evidence for any warm dust emission at ∼0.4 au. Coronal emission at millimeter wavelengths has been observed for other M dwarf stars including AU Mic, which hosts a resolved cold debris disk at ∼40 au and exhibits continual quiescent 1.3 mm emission in excess of expected photospheric levels and variable 9 mm emission (MacGregor et al. 2016). Models of time-averaged turbulent coronal heating (which exhibits flare-like bursts in simulations) result in thermal bremsstrahlung sufficient to explain both the 9 mm variability and 1.3 mm emission, as well as the observed X-ray emission (Cranmer et al. 2013; MacGregor et al. 2016). The quiescent millimeter emission from AU Mic is considerably larger, however, with a 1.3 mm flux density six times the photospheric level and 10 times the excess observed from Proxima Centauri (MacGregor et al. 2013). To support the claim of an inner warm disk, Anglada et al. (2017) also noted a slight elongation of the 12 m central source along a position angle of 130°. Our fits to the millimeter visibilities using uvmodelfit in CASA are consistent with a point source, and do not suggest that this source is spatially resolved. Any elongation of the central source is at a 2σ level in our natural weight images and varies between the two observations (see Figure 1).

The outer disk at ∼30 au is posited by Anglada et al. (2017) from several additional ≳3σ peaks in the ACA image that contribute to an azimuthally summed flux density of ∼1.7 mJy. We find that the ACA visibilities do not adequately cover the short spatial frequencies needed to measure the total disk flux density and make it impossible to robustly constrain the disk geometry. Additional observations are needed to test the veracity of this detection. The possibility of contamination must also be considered. Each of the peaks that contribute to the putative disk have a flux density of ∼150–200 μJy. Deep ALMA surveys at millimeter wavelengths have placed strong constraints on the expected number of faint background sources in an image (Carniani et al. 2015) that agree with results from cosmological simulations (Shimizu et al. 2012; Cai et al. 2013). Given the differential source counts at 1.3 mm, the number of sources with flux density >150 μJy that are expected within the ACA primary beam (FWHM ∼ 39'') is ${13}_{-8}^{+10}$, a not insignificant number. In addition, Proxima Centauri sits close to the Galactic plane (latitude −2°) in a region of high background cirrus, which made it impossible for Spitzer to provide an upper limit on the source flux density at 60 μm (Gautier et al. 2007). Between possible extragalactic background and Galactic sources, it will be difficult to detect or image the proposed extended disk.