Main

The proximity of Sgr A* makes the characteristic angular size scale of the Schwarzschild radius (RSch = 2GM/c2) larger than for any other black hole candidate. At a distance of ∼8 kpc (ref. 8), the Sgr A* Schwarzschild radius is 10 μas, or 0.1 astronomical unit (au). Multi-wavelength monitoring campaigns9,10,11 indicate that activity on scales of a few RSch in Sgr A* is responsible for observed short-term variability and flaring from radio to X-rays, but direct observations of structure on these scales by any astronomical technique has not been possible. Very-long-baseline interferometry (VLBI) at 7 mm and 3.5 mm wavelength shows the intrinsic size of Sgr A* to have a wavelength dependence, which yields an extrapolated size at 1.3 mm of 20–40 μas (refs 6, 7). VLBI images at wavelengths longer than 1.3 mm, however, are dominated by interstellar scattering effects that broaden images of Sgr A*. Our group has been working to extend VLBI arrays to 1.3 mm wavelength, to reduce the effects of interstellar scattering, and to utilize long baselines to increase angular resolution with a goal of studying the structure of Sgr A* on scales commensurate with the putative event horizon of the black hole. Previous pioneering VLBI work at 1.4 mm wavelength detected Sgr A* on 980-km projected baselines, but calibration uncertainties resulted in a range for the derived size of 50–170 μas (ref. 12).

On 10 and 11 April 2007, we observed Sgr A* at 1.3 mm wavelength with a three-station VLBI array consisting of the Arizona Radio Observatory 10-m Submillimetre Telescope (ARO/SMT) on Mount Graham in Arizona, one 10-m element of the Combined Array for Research in Millimeter-wave Astronomy (CARMA) in Eastern California, and the 15-m James Clerk Maxwell Telescope (JCMT) near the summit of Mauna Kea in Hawaii. A hydrogen maser time standard and high-speed VLBI recording system were installed at both the ARO/SMT and CARMA sites to support the observation. The JCMT partnered with the Submillimetre Array (SMA) on Mauna Kea, which housed the maser and the VLBI recording system and provided a maser-locked receiver reference to the JCMT. Two 480-MHz passbands sampled to two-bit precision were recorded at each site, an aggregate recording rate of 3.84 × 109 bits per second (Gbit s-1). Standard VLBI practice is to search for detections over a range of interferometer delay and delay rate. Six bright quasars were detected with high signal to noise on all three baselines allowing array geometry, instrumental delays and frequency offsets to be accurately calibrated. This calibration greatly reduced the search space for detections of Sgr A*. All data were processed on the Mark4 correlator at the MIT Haystack Observatory in Massachusetts.

On both 10 and 11 April 2007, Sgr A* was robustly detected on the short ARO/SMT–CARMA baseline and the long ARO/SMT–JCMT baseline. On neither day was Sgr A* detected on the CARMA–JCMT baseline, which is attributable to the sensitivity of the CARMA station being about a third that of the ARO/SMT (owing to weather, receiver temperature and aperture efficiency). Table 1 lists the Sgr A* detections on the ARO/SMT–JCMT baseline. The high signal to noise ratio, coupled with the tight grouping of residual delays and delay rates, makes the detections robust and unambiguous.

Table 1 VLBI detections of Sgr A* on the ARO/SMT–JCMT baseline at 1.3 mm wavelength
Full size table

There are too few visibility measurements to form an image by the usual Fourier transform techniques; hence, we fit models to the visibilities (shown in Fig. 1). We first modelled Sgr A* as a circular Gaussian brightness distribution, for which one expects a Gaussian relationship between correlated flux density and projected baseline length. The weighted least-squares best-fit model (Fig. 1) corresponds to a Gaussian with total flux density of 2.4 ± 0.5 Jy and full width at half maximum (FWHM) of  μas where errors are 3σ. On the assumption of a Gaussian profile, the intrinsic size of Sgr A* can be extracted from our measurement assuming that the scatter broadening adds in quadrature with the intrinsic size. At a wavelength of 1.3 mm the scattering size extrapolated from previous longer-wavelength VLBI13 is 22 μas along a position angle 80° degrees east of north on the sky, closely aligned with the orientation of the ARO/SMT–JCMT baseline. Removing the scattering effects results in a 3σ range for the intrinsic size of Sgr A* equal to  μas. The 3σ intrinsic size upper limit at 1.3 mm, combined with a lower limit to the mass of Sgr A* of 4 × 105 solar masses, , from proper-motion work14,15, yields a lower limit for the mass density of 9.3 × 1022  pc-3. This limit is an order of magnitude larger than previous estimates7, and two orders of magnitude below the critical density required for a black hole of 4 × 106 . This density lower limit and central mass would rule out most alternatives to a black hole for Sgr A* because other concentrations of matter would have collapsed or evaporated on timescales that are short compared with the age of the Milky Way16.

Figure 1: Fitting the size of Sgr A* with 1.3 mm wavelength VLBI.
figure 1

Shown are the correlated flux density data on the ARO/SMT–CARMA and ARO/SMT–JCMT baselines plotted against projected baseline length (errors are 1σ). Squares show ARO/SMT–CARMA baseline data and triangles show ARO/SMT–JCMT data, with open symbols for 10 April and filled symbols for 11 April. The solid line shows the weighted least-squares best fit to a circular Gaussian brightness distribution, with FWHM size of 43.0 μas. The dotted line shows a uniform thick-ring model with an inner diameter of 35 μas and an outer diameter of 80 μas convolved with scattering effects due to the interstellar medium. The total flux density measurement made with the CARMA array over both days of observing (2.4 ± 0.25 Jy : 1σ) is shown as a filled circle. An upper limit for flux density of 0.6 Jy, derived from non-detections on the JCMT–CARMA baseline, is represented with an arrow near a baseline length of 3,075 × 106λ.

PowerPoint slide

Full size image

Figure 2 shows both observed and intrinsic sizes for Sgr A* over a wide range of wavelengths along with the scattering model13 and the weighted least-squares power-law fit to the intrinsic size measurements. At 1.3 mm wavelength the interstellar scattering size is less than the intrinsic size, demonstrating that VLBI at this wavelength can directly detect structures in Sgr A* on event-horizon scales. The intrinsic size dependence on wavelength, λα (α = 1.44 ± 0.07, 1σ), confirms that the Sgr A* emission region is stratified, with different wavelengths probing spatially distinct layers. The λα fit also provides an improved extrapolation to intrinsic sizes at submillimetre wavelengths consistent with emission models that produce X-ray emission from inverse Compton scattering of longer-wavelength photons9,10,11. The minimum intrinsic brightness temperature derived from our 1.3-mm results is 2 × 1010 K.

Figure 2: Observed and intrinsic size of Sgr A* as a function of wavelength.
figure 2

Red circles show major-axis observed sizes of Sgr A* from VLBI observations (all errors 3σ). Data from wavelengths of 6 cm to 7 mm are from ref. 13, data at 3.5 mm are from ref. 7, and data at 1.3 mm are from the observations reported here. The solid line is the best-fit λ2 scattering law from ref. 13, and is derived from measurements made at λ > 17 cm. Below this line, measurements of the intrinsic size of Sgr A* are dominated by scattering effects, while measurements that fall above the line indicate intrinsic structures that are larger than the scattering size (a ‘source-dominated’ regime). Green points show derived major-axis intrinsic sizes from 2 cm < λ < 1.3 mm and are fitted with a λα power law (α = 1.44 ± 0.07, 1σ) shown as a dotted line. When the 1.3-mm point is removed from the fit, the power-law exponent becomes α = 1.56 ± 0.11 (1σ).

PowerPoint slide

Full size image

The data presented here confirm structure in Sgr A* on linear scales of ∼4RSch, but the exact nature of this structure is not well determined. The assumption of a Gaussian model above is motivated by simplicity, but the increased angular resolution of VLBI at 1.3 mm will soon allow consideration and testing of more complex structures. As an example, the 1.3-mm VLBI data are also well fitted by a uniform thick ring of inner diameter 35 μas and outer diameter 80 μas that is convolved with the expected scattering in the interstellar medium (Fig. 1). Such structures are motivated by simulations of the Sgr A* accretion region that use full general relativistic ray tracing17,18 and magneto-hydrodynamic effects19, and which predict a ‘shadow’ or null in emission in front of the black hole position, especially in the case of face-on accretion disks. The upper limits on correlated flux density from the JCMT–CARMA baseline (Fig. 1) cannot currently discriminate between Gaussian and ring models, but expected and planned increases in both VLBI sensitivity and baseline coverage over the next five years will allow such detailed comparisons.

At present, Sgr A* has been shown to be coincident with the position of the unseen central mass only at the ∼10 mas level3. It is an open question whether or not the Sgr A* source is centred on the black hole. Indeed, several models predict an offset between Sgr A* and the black hole position. In jet models of Sgr A* (ref. 20), for example, millimetre and submillimetre emission arises at a point in the relativistic plasma stream where the optical depth is close to unity, and the peak in emission can be spatially separated from the black hole. Simulations of accretion disks that are inclined to our line of sight show kinematic (Doppler) brightening on the approaching section of the disk, which also results in an emission peak that is off to one side of the black hole17,18,19. Even for modest accretion disk inclinations, this emission peak can be asymmetric and compact with a morphology dependent on a number of factors including black hole spin, underlying magnetic field structure and inner disk radius.

The intrinsic size derived in this work by fitting the circular Gaussian model can be used to argue that Sgr A* is not a spherically symmetric photosphere centred on the central dark mass. This is because radiation originating from a spherical surface at a given radius from a black hole is strongly lensed by gravity, and presents a larger apparent size to observers on the Earth. Such a surface of radius R centred on a non-rotating black hole will have an apparent radius, Ra, given by21,22

This has the important consequence that distant observers will measure a minimum apparent diameter of ∼5.2 RSch for all objects centred on the black hole that have radii less than 1.5 RSch (the minimum circular orbit for photons). In the case of Sgr A*, this corresponds to a minimum apparent diameter of 52 μas. This size is only marginally consistent with the 3σ upper limit on the intrinsic size derived from our 1.3 mm VLBI observations, and suggests that Sgr A* arises in a region offset from the black hole, presumably in a compact portion of an accretion disk or jet that is Doppler-enhanced by its velocity along our line of sight. This lensing argument also holds in the case of a maximally rotating black hole of the same mass, for which the minimum apparent size in the equatorial plane would be 45 μas (ref. 22), which is also larger than the intrinsic size derived here. The intrinsic sizes of Sgr A* measured with VLBI at 3.5 mm and 7 mm exceed the minimum apparent size, and thus cannot similarly be used to constrain the location of Sgr A* relative to the black hole.

Detection of the event-horizon-scale structure reported here indicates that future VLBI observations at λ ≤ 1.3 mm will open a new window onto fundamental black hole physics through observations of our Galactic Centre. Plans to increase the sensitivity of the VLBI array described here by factors of up to 10 are under way, and the addition of more VLBI stations will increase baseline coverage and the ability to model increasingly complex structures. At projected VLBI array sensitivities, Sgr A* will be detected on multiple baselines within 10-s timescales, enabling sensitive tests for time-variable structures such as those suggested by orbiting hotspot18 and flaring models9,10,11.