WorksPhysics and Astronomy
The Galactic Center Reference Frame: A 30-Year Solution
Abstract
Stellar orbits probe the metric of and mass around the Sgr A* massive black hole. To enable study of the black hole, the surrounding mass, and tests of gravity, a precise reference frame must be established and maintained. We present an updated stellar maser-based reference frame for the inner parsec of the Galactic Center, referenced to the radio continuum of Sgr A*. This work synthesizes 30 years of astrometric observations of 15 SiO stellar masers and the Sgr A* continuum at 43 GHz. The uncertainty in the mean offset of the maser ensemble from Sgr A* is 0.08 mas in the current epoch and will decay to 0.10 mas by 2030. At a distance of 8.2 kpc, these correspond to position uncertainties of 0.6 AU and 0.8 AU, respectively, which is similar to or smaller than the stellar SiO-emitting region. The mean maser proper motion uncertainties are 0.05 mas yr in RA and 0.08 mas yr in Dec, corresponding to 1.8 km s and 3.2 km s. The reference frame stability between IR and maser astrometry is 10~as yr (0.40 km s) in RA and 15 as yr (0.57 km s) in Dec. The masers show substantial variability in flux density and velocity structure, most notably IRS~7, which shows a 30-fold brightening in a blueshifted component since 2022. This variability may set a floor on future astrometry improvements and especially measurements of radial acceleration.
Introduction
Near-infrared observations of the orbits of stars close to Sgr A* conclusively demonstrated the presence of a massive black hole in the center of the Galaxy [20, 19]. This work relied on an astrometric reference frame enabled by simultaneous radio observations of stellar masers and the Sgr A* radio continuum [36, 44, 45]. Since the black hole was not historically visible in the near-IR imaging, measuring the offsets between IR-bright stars and Sgr A* required an intermediate step, relying on the simultaneous radio detection of Sgr A* and masers associated with a subset of the IR-detected stars. The angular offsets of IR-bright SiO maser-emitting stars from the 1 Jy Sgr A* continuum, along with the masers’ proper motions, provide an astrometric reference frame for the orbital solutions of stars much closer to the massive black hole than the maser-emitting stars [36, 44, 45, 47, 13, 27] (Yelda et al. 2015). This reference frame places Sgr A* at rest and at the coordinate origin, which is also the phase center for the radio interferometry.
While the GRAVITY interferometer operating at ESO’s VLTI can now detect Sgr A* in the near-IR continuum [21], it is still important to be able to refer IR positions to Sgr A* via the maser astrometry. The GRAVITY science fiber FWHM is 74 mas in K-band [23]; stars at larger separations from Sgr A* thus continue to rely on imaging data obtained at the adaptive optics resolution and referencing positions via the maser stars. It is now also possible to address how well the positions of the mass (as measured by orbits), the near-IR emission of Sgr A*, and the radio emission of Sgr A* coincide. The Event Horizon Telescope map of Sgr A* shows an asymmetry, and one assumes that the mass is located near the center (and not the centroid) of the luminous ring (Event Horizon Telescope Collaboration et al. [24]). Further, it is not obvious that the near-IR emission – either in the quiescent state or during flares – is centered on the mass. It is also likely that the unresolved 43 GHz radio continuum does not coincide at some level with the IR emission or the mass; the question is at what scale the standard assumptions break down.
As the reference frame continues to improve over time, both by way of better proper motion solutions from longer time baselines and from improved astrometric methods, the astrometry enables new science. The stars close to Sgr A* probe the black hole metric. So far, only effects of order (), such as Schwarzschild precession (GRAVITY Collaboration et al. [21]), have been detected. These effects rely solely on the black hole mass, but in the next order in , the equations of motion contain a term that is sensitive to black hole spin, commonly referred to as Lense-Thirring precession. Measuring this effect requires stars on orbits with very small pericenter distances. The recent discovery of such a star, S-301, whose pericenter distance of 1.4 mas is roughly an order of magnitude smaller than that of S-02, could enable spin measurements of Sgr A* in the coming decades ([1, 40, 48]). It is therefore important to be able to pinpoint the black hole mass location for such close flybys. At the fourth order, , a quadrupole term enters the equations of motion. In general relativity, this term is fully determined by the black hole spin (assuming no charge), a consequence of the no-hair theorem (e.g., Will [51]). Hence, measuring both spin and quadrupole moment constitutes a test of fundamental physics, but it requires proper motion measurements below 10 as yr (Will [51]). Applying analytical estimates from Will [51] to the orbit of S-301, we estimate Lense-Thirring precession rates of roughly 6 as yr and quadrupolar precession rates of 0.2 as yr assuming a black hole spin parameter (Daly et al. [10]).
The stellar orbits are also sensitive to other masses in the region (GRAVITY Collaboration et al. [24]), and hence offer a possibility to probe the “dark sector.” The standard assumption is that this is dominated by stellar remnants, by far dominant over dark matter unless dark matter forms spikes. The presence of a cusp of stellar-mass black holes would inform science with the future gravitational wave experiment LISA because gravita-
tional waves from extreme-mass ratio inspirals would be produced from exactly such a cusp around a massive black hole (e.g., [26, 2, 3, 6]. A cusp of pulsars would enable significant new tests of general relativity ([34] (Hu & Shao 2024)). Conversely, a dark matter cusp would support the detection of axionic dark matter (e.g., (Hook et al. 2018; Darling 2020a,b; Foster et al. 2022; Battye et al. 2022)).
Astrometric SiO maser observations have been ongoing for nearly three decades. While proper motions and the reference frame stability improve over time as , assuming regular sampling, the flip side of this benefit is that astrometric solutions that are not updated will decay as time passes because the uncertainty cone grows over time. In this work, we update the reference frame and estimate its future decay. § 2 describes the data sources, observations, and reduction used to form a 30-year astrometric solution for 15 SiO stellar masers. § 3 outlines the astrometric methods that are described in detail in Darling et al. [13]. § 4 presents the improved astrometry and highlights the substantial variability in the maser spectra. We discuss the implications of the new reference frame and its decay over time in § 5, and we compare the maser-based reference frame to the Hosek et al. [27] **References—**R03 = Reid et al. [44]; R07 = Reid et al. [45]; R22 = Reid (2022, private communication); P22 = Paine & Darling [39]; D23 = Darling et al. [13]; D26 = this work.
HST-Gaia reference frame in § 6. § 7 summarizes this work and discusses the future outlook for precise astrometry in the Galactic Center.
When physical transverse distances or transverse velocities are quoted, we assume the distance to Sgr A* is 8.2 kpc (e.g., [46, 23, 30] (Do et al. 2019)).
Observations and Data Reduction
This work adds a 2026 Karl G. Jansky Very Large Array (VLA)3 epoch to the 1996–2022 SiO stellar maser time series to form a 30-year baseline. Darling et al. [13] describe the legacy data sources in detail, and their properties are summarized in Table 1. As before, we focus on the subset of the stellar masers and the maser transitions that have legacy astrometry (, and ). Table 2 lists the masers, and Figure 1 shows their locations and motions obtained in § 3.

Figure 1. Epoch 2026.208 offsets and proper motions of the observed stellar SiO masers (Table 2 and Figure A1). Redshifted masers are red, and blueshifted masers are blue. Sgr A* is located at the grey circle (not to scale). Projected distances assume a Galactic Center distance of 8.2 kpc.
| Mean Epoch | Program | Beam | Ref. | |
| 1996.413 | VLBA BM060 | 1 | 1.2 × 0.9 | R03 |
| 1998.410 | VLA AM592 | 1 | 70 × 30 | R03 |
| 2000.850 | VLA AR451 | 1 | 80 × 40 | R03 |
| 2006.200 | VLA AR588 | 1 | 86 × 33 | R07 |
| 2008.860 | VLA AR678 | 1 | 82 × 35 | R22 |
| 2011.470 | VLA 11A-101 | 1 | 97 × 42 | R22 |
| 2014.249 | VLA 14A-168 | 1,2 | 66 × 30 | R22 |
| 2020.988 | VLA 19A-310 | 1,2 | 93 × 36 | P22,D23 |
| 2022.227 | VLA 22A-328 | 1,2 | 82 × 39 | D23 |
| 2026.208 | VLA 26A-049 | 1,2 | 81 × 38 | D26 |
Table 1. SiO Maser Data and Observations
| Name | RA Offset | Dec Offset | PM RA | PM Dec | Ref. Epoch | |||
| (km s) | (arcsec) | (arcsec) | (mas yr) | (mas yr) | ||||
| IRS 9 | 0.8 | 2019.021 | 9 | |||||
| IRS 7 | 2.0 | 2017.836 | 7 | |||||
| SiO-14 | 3.3 | 2021.121 | 9 | |||||
| IRS 12N | 1.9 | 2021.167 | 10 | |||||
| IRS 28 | 1.3 | 2015.418 | 9 | |||||
| SiO-15 | 1.4 | 2020.150 | 7 | |||||
| IRS 10EE | 2.7 | 2021.531 | 10 | |||||
| IRS 15NE | 1.0 | 2011.314 | 10 | |||||
| SiO-16 | 13.4 | 2017.293 | 8 | |||||
| SiO-6 | 8.0 | 2012.939 | 9 | |||||
| SiO-17 | 4.7 | 2017.666 | 8 | |||||
| SiO-11 | 23.3 | 2014.625 | 9 | |||||
| IRS 17 | 2.8 | 2024.842 | 7 | |||||
| SiO-12 | 4.1 | 2016.751 | 8 | |||||
| IRS 19NW | 8.4 | 2020.946 | 9 |
Table 2. SiO Maser Angular Offsets and Proper Motions. Note—Coordinate offsets are with respect to the Sgr A* radio centroid at the reference epoch, which is the position variance-weighted date of the time series (see § 3). The LSR velocity is approximate; the masers show variability in their spectral peaks and velocity centroids ([44, 45, 39]). The statistic characterizes the joint weighted least-squares proper motion fit in both coordinates.
VLA program 26A-049 executed in three 5-hour sessions on 09, 20, and 23 March 2026. This brings the time baseline to 29.795 years for a subset of the masers. We observed the field of view centered on Sgr A* in the and 43 GHz SiO maser transitions. The observations were identical to previ-
ous program 22A-328 in order to minimize systematics and harmonize the time series. Two SiO masers, the 43.1221 GHz and the 42.8206 GHz transitions, were observed simultaneously in 128 MHz bandwidths with 100 kHz (0.7 km s) channels, two circular polarizations, and 3 s integrations using the largest VLA configuration. Sgr A* served as the 1 Jy in-beam complex gain calibration source, 3C286 provided flux calibration, and J1924−2914 was used for bandpass and delay calibration. Calibration, imaging, and astrometric measurements replicate those presented in Darling et al. [13]. In short, we use CASA (McMullin et al. [35]) for all steps. We average two channels and two integrations prior to calibration, and the calibration process is standard for Q-band A configuration. We rely on per-integration complex gain calibration to track phase wrapping.
The Sgr A* continuum provides the phase center to which all masers are referenced. Sgr A* shows a significant proper motion with respect to background quasars due to the barycenter motion about the Galactic Center ([43, 31] (Xu et al. 2022)), so only relative astrometry is obtained from this method. All maser positions and subsequent proper motions are tied to the Sgr A* radio continuum, and it is assumed that this continuum is fixed with respect to the massive black hole.
In order to retain the relative in-beam astrometry, no continuum subtraction was performed, and we simultaneously imaged Sgr A* and all of the masers as outlier fields. All three 26A-049 sessions were combined for imaging. We used CASA tclean to clean and image down to 15 mJy, five times the per-channel rms noise. We measure Sgr A* and maser positions in the image plane at every channel in each bandpass from 2D Gaussian fits using CASA imfit. A variance-weighted average of peak fluxes forms the final position and its uncertainty. Offsets from the Sgr A* continuum centroid are calculated for each maser and added to the time series. Maser coordinate uncertainties are generally less than 1 mas (8 AU at 8.2 kpc).
Astrometric Methods
This work replicates the methods presented in Darling et al. [13]. However, we do not apply time-dependent differential offsets to the astrometry in each epoch, which leads to lower scatter in the time series. Table A1 lists the coordinate offsets from the Sgr A* continuum for all epochs.
| Name | Epoch | RA Offset | Dec Offset |
| (arcsec) | (arcsec) | ||
| IRS 9 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| IRS 7 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| SiO-14 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| IRS 12N | 1996.413 | ||
| 1998.410 | |||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| IRS 28 | 1998.410 | ||
| Name | Epoch | RA Offset (arcsec) | Dec Offset (arcsec) |
| 2000.850 | 10.4698 0.0010 | ||
| 2006.200 | 10.4798 0.0007 | ||
| 2008.860 | 10.4843 0.0005 | ||
| 2011.470 | 10.4861 0.0005 | ||
| 2014.249 | 10.4918 0.0003 | ||
| 2020.988 | 10.5013 0.0014 | ||
| 2022.227 | 10.5032 0.0003 | ||
| 2026.208 | 10.5086 0.0005 | ||
| SiO-15 | 2000.850 | ||
| 2006.200 | |||
| 2008.860 | |||
| 2014.240 | |||
| 2020.980 | |||
| 2022.220 | |||
| 2026.200 | |||
| IRS 10EE | 1996.413 | 7.6837 0.0005 | 4.2194 0.0005 |
| 1998.410 | 7.6831 0.0005 | 4.2159 0.0005 | |
| 2000.850 | 7.6852 0.0005 | 4.2099 0.0005 | |
| 2006.200 | 7.6832 0.0005 | 4.1986 0.0005 | |
| 2008.860 | 7.6844 0.0005 | 4.1938 0.0005 | |
| 2011.470 | 7.6844 0.0006 | 4.1906 0.0006 | |
| 2014.249 | 7.6840 0.0003 | 4.1839 0.0004 | |
| 2020.988 | 7.6851 0.0001 | 4.1697 0.0002 | |
| 2022.227 | 7.6849 0.0001 | 4.1676 0.0002 | |
| 2026.208 | 7.6851 0.0001 | 4.1590 0.0001 | |
| IRS 15NE | 1996.413 | 1.2302 0.0005 | 11.3315 0.0005 |
| 1998.410 | 1.2264 0.0004 | 11.3197 0.0006 | |
| 2000.850 | 1.2233 0.0011 | 11.3031 0.0025 | |
| 2006.200 | 1.2111 0.0005 | 11.2750 0.0010 | |
| 2008.860 | 1.2068 0.0006 | 11.2602 0.0007 | |
| 2011.470 | 1.2023 0.0005 | 11.2448 0.0005 | |
| 2014.249 | 1.1969 0.0005 | 11.2284 0.0012 | |
| 2020.988 | 1.1834 0.0003 | 11.1892 0.0017 | |
| 2022.227 | 1.1814 0.0003 | 11.1807 0.0004 | |
| 2026.208 | 1.1733 0.0005 | 11.1581 0.0009 | |
| SiO-16 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| Name | Epoch | RA Offset{sup}a` | Dec Offset |
| 2014.249 | |||
| 2022.227 | |||
| 2026.208 | |||
| SiO-6 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| SiO-17 | 1998.410 | ||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| SiO-11 | 1998.410 | ||
| 2000.850 | |||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| IRS 17 | 2000.850 | ||
| 2006.200 | |||
| 2008.860 | |||
| 2014.249 | |||
| 2020.988 | |||
| 2022.227 | |||
| 2026.208 | |||
| SiO-12 | 2000.850 | ||
| 2006.200 | |||
| 2008.860 | |||
| 2011.470 | |||
| Name | Epoch | RA Offset | Dec Offset |
| (arcsec) | (arcsec) | ||
| 2014.249 | −18.8101 ± 0.0005 | 42.4777 ± 0.0011 | |
| 2020.988 | −18.8004 ± 0.0007 | 42.4854 ± 0.0027 | |
| 2022.227 | −18.7993 ± 0.0008 | 42.4867 ± 0.0012 | |
| 2026.208 | −18.7962 ± 0.0011 | 42.4911 ± 0.0021 | |
| IRS 19NW | 1998.410 | 14.5521 ± 0.0011 | −18.4604 ± 0.0012 |
| 2000.850 | 14.5535 ± 0.0015 | −18.4701 ± 0.0031 | |
| 2006.200 | 14.5591 ± 0.0005 | −18.4638 ± 0.0009 | |
| 2008.860 | 14.5621 ± 0.0010 | −18.4713 ± 0.0013 | |
| 2011.470 | 14.5636 ± 0.0004 | −18.4631 ± 0.0008 | |
| 2014.249 | 14.5695 ± 0.0003 | −18.4709 ± 0.0007 | |
| 2020.988 | 14.5781 ± 0.0003 | −18.4758 ± 0.0008 | |
| 2022.227 | 14.5802 ± 0.0001 | −18.4749 ± 0.0002 | |
| 2026.208 | 14.5858 ± 0.0002 | −18.4781 ± 0.0004 |
Table A1. SiO Maser Coordinate Time Series. Table A1 continued
Proper motions () are obtained from linear fits to the offset-corrected coordinate time series centered at a reference epoch :
is the the variance-weighted mean date in the time series, where the variance is the sum in quadrature of the coordinate uncertainties of each epoch. This reference epoch selection method minimizes correlation between the proper motion and the Sgr A* offsets at , . There is one reference epoch per maser, and both coordinate time series are fit simultaneously to assess cross-coordinate correlations. A quadratic term in Equations (1) and (2) could be added to assess transverse acceleration (Paine & Darling [39]).
We re-examined the various time series fitting methods described in Darling et al. [13]. As before, a variance-weighted least-squares linear fit in each coordinate is generally consistent with the other more elaborate methods. A few examples are depicted in Figure B2 and Figure B3. We employ joint RA and Dec weighted linear least-squares fits for all subsequent results and analysis. The reference epoch for each maser is the Sgr A* offset variance-weighted mean date in the time series, , listed in Table 2; joint coordinate fits provide a single reference epoch rather than one for each coordinate solution. This can induce a minor correlation between the slope and intercept of single-coordinate fits, but the correlation between RA and Dec solutions is generally negligible.

Figure B2. Example residual proper motion time series showing the three fitting methods (LS = least-squares; CF = conservative formulation; GB = good-and-bad data). The residual is with respect to the weighted least-squares fit, and the shaded regions indicate AU. Top: SiO-6 shows differing proper motion solutions in RA for all three methods. Bottom: IRS 19NW, showing proper motion solutions for the conservative formulation and good-and-bad methods that differ from the canonical least-squares fit. For most of the masers, the two less conventional fitting methods do not improve upon or differ significantly from the least-squares method.

Figure B3. Example residual proper motion time series showing the result of expanded uncertainties (EU) that are added in quadrature to the measurement uncertainties. The residual is with respect to the weighted least-squares (LS) fit, and the shaded regions indicate AU. The EU proper motions do not significantly differ from the LS proper motions (). Top: SiO-6 has expanded uncertainties of 3.5 and 4.3 mas, respectively. Bottom: The additional uncertainties are 1.3 mas in RA and 3.2 mas in Dec for IRS 19NW.
We no longer assign an uncertainty floor to the IRS 7 offset from Sgr A*, which was done by Reid et al. [45] and Paine & Darling [39] owing to its maser variability and supergiant luminosity. As described in § 4 and shown in Figure 3 and A1 and Table 2, its astrometric time series shows less scatter than many other masers. We do, however, continue to restrict the time series to 2006.200 and later (Darling et al. [13]).

Figure 3. Maser position uncertainty vs. time. This is the uncertainty in the offset from Sgr A* as a function of time based on the 30-year solution (Table 2), combining Equations 1 and 2. The black locus shows the uncertainty in the variance-weighted mean offset of the maser ensemble, and its minimum value is indicated by the white line in the colorbar. The grey region indicates the 30-year span of observations of a subset of the masers. The increasing uncertainties beyond the observation window show how the astrometry decays into the future (and the past).
Li et al. (2010) and Paine & Darling [39] have identified additional masers in the inner few parsecs around Sgr A* that could in principle contribute to the Galactic Center reference frame. While observations in 2019, 2022, and 2026 provide an adequate time series to incorporate new masers into the reference frame solution, these masers are more distant in projection from Sgr A* than the primary sample and are typically of low flux density. As a result, they would not significantly improve the signal-to-noise-weighted astrometry. They are therefore omitted.
Results
All masers were detected in the 2026 epoch, but substantial spectral variation is apparent. Some masers were significantly brighter than in previous sessions, some were fainter, and many changed their shapes and velocity centroids. Figure 2 compares the 2026 epoch with the 2022 observations of the , 43.1221 GHz transition. Most notable is the 30-fold flare in the very blue component of IRS 7: this component was not detected in 2022 but was luminous in pre-2000 observations ([44, 45]). While the maser light centroid, which is dominated by the blue component in 2026, is not anomalous compared to recent epochs (see § 6.2 and Figure A1), the maser velocity and line shape changes will ultimately limit line of sight acceleration measurements compared to transverse accelerations ([39]).

Figure 2. Composite spatially-integrated SiO , 43.1221 GHz maser spectra from the 2026.208 (top) and 2022.227 (bottom) image cubes. The colors indicate the projected distance from Sgr A*, assuming a Galactic Center distance of 8.2 kpc. The same flux density scale is used in both plots to show the variation between epochs.
Table 2 lists the new Sgr A* offsets, proper motions, reference epochs, and number of epochs used in the proper motion fits. Figure A1 shows the maser proper motion fits and residuals, and Table A1 shows all of the data that are depicted in Figure A1.
The aggregate statistics of the astrometric solutions have improved compared to the Darling et al. [13] results (see § 6). The uncertainties in the variance-weighted means of the maser coordinate offsets from Sgr A* are (0.06, 0.09) mas in the current epoch. These correspond to (0.5, 0.7) AU, which may be smaller than the SiO-emitting region in the host stars. The maser emission is often asymmetric around the stellar center of mass and variable in time, so the size of the unresolved maser-emitting region may be a floor on the astrometric accuracy for tracking the systemic stellar position. The mean maser proper motion uncertainties are (0.05, 0.08) mas yr in RA and Dec, corresponding to 1.8 km s and 3.2 km s. The reference frame stability, the uncertainty in the variance-weighted mean proper motion difference between the IR and maser observations, is (10.2, 14.7) as yr or (0.40, 0.57) km s. These ensemble characterizations of the reference frame5 have improved on the Darling et al. [13] results with the addition of the 2026 epoch (§ 6.2).
Discussion
Given the astrometric solutions in Table 2, one can compute the coordinate offsets of each maser from Sgr A* as a function of time as well as the uncertainty in those offsets. The uncertainties provide an estimate of the quality of the offsets versus observing epoch and a prediction for the growth of the uncertainty cones as the astrometric solution decays. Figure 3 shows the uncertainties in the radial offsets of the masers from Sgr A*. The best-measured masers, IRS 10EE and IRS 12N, reach a minimum uncertainty of mas. The uncertainty in the error-weighted mean offset of the maser ensemble shows a minimum value of 0.06 mas in year 2020.9 and rises to 0.08 mas in the current epoch.
The IR-maser reference frame stability has steadily declined (improved) over time.6 Figure 4 shows this trend in RA since 2009 and in Dec since 2022. The stability fell below a fiducial 15 as yr in RA in 2022 (Darling et al. [13]) and in Dec in 2026 (this work). This threshold corresponds roughly to the proper motion uncertainty needed to use the S0-2/S2 stellar orbit to probe down to a mass of 1000 lying between its apocenter (0.01 pc) and Sgr A*, assuming a 13 yr observation span (Hosek et al. [27]). Figure 4 also shows the reference frame stability between the masers and the Hosek et al. [27] HST-Gaia astrometry discussed in § 6.

Figure 4. Reference frame stability vs. time. The reference frame stability is the uncertainty in the variance-weighted mean of the proper motion differences between the IR and radio measurements of the stellar masers. The blue and green symbols are referenced to the IR astrometry of seven maser-emitting stars reported in Sakai et al. [47]. Filled points show the trend in RA, and the open symbols show Dec. The red triangles indicate the projected reference frame stability decay in the absence of continued astrometric monitoring. The horizontal dotted line indicates the stability threshold needed to detect 1000 M of mass external to Sgr A* but within the orbit of S0-2/S2 (Hosek et al. [27]).
The results of the latest epoch of observations, combined with those spanning the last 30 years, show that: (1) the last 3–4 epochs have very stable astrometry compared to the full history (e.g., IRS 12N, IRS 10EE, SiO-12, and IRS 19NW in Figure A1); (2) there are, however, masers with persistent significant astrometric variations such as SiO-16, SiO-6, SiO-11, and IRS 19NW, indicated by large values in Table 2; (3) the proper motion uncertainties and reference frame stability continue to improve over time (no precision floor has yet been reached); and (4) the reference frame should be durable into the early 2030s (Figure 4). To maintain the stability and precision of the reference frame, the SiO maser observations should be repeated in roughly five years.
There are six masers that show significant variance far in excess of the measurement uncertainties (). Interestingly, these all show positive radial velocities, and
the only positive radial velocity maser with low variance is IRS 17, which is also closest in projection to Sgr A* among this subset. SiO-6 shows significant departures from the proper motion trendline in both coordinates with excursions as large as 10 mas. It is not obviously periodic, and the angular scale of the offsets is likely too large to be due to scintillation. It may be noteworthy that the significant spectral change shown in Figure 2 is accompanied by a significant astrometric change (Figure A1 and Table A1). The per-channel uncertainty in each coordinate spans 1–8 mas in recent epochs, and variation in the number and weighting of spectral channels could therefore alter the measured position. Underestimated uncertainties in the time series of SiO-6 are explored in Appendix § 10. The large variance in SiO-16, SiO-17, SiO-12, and IRS 19NW is dominated by early epochs; all show astrometric convergence during the last 3–4 epochs, since 2014, where the scatter is consistent with the measurement error or within the expected SiO-emitting region of the star. The SiO-11 astrometry is highly suggestive of periodicity, but a rough estimate suggests that the binary companion would be at least 20 . Oyama et al. [38] identified a similar trend over smaller timescales and smaller astrometric excursions for IRS 10EE, claiming a 13–16 companion based on VLBA observations, but the VLA data lack the resolution to test this claim.
Comparison to Previous Reference Frames
We compare the current maser astrometry to previous work using the subset of masers that overlap each sample. In each case, we quantify the proper motion differences and the Sgr A* offset differences for a specific common reference epoch. The 2026 maser epoch positions and proper motions are subtracted from those of each comparison reference frame – not ordered by epoch – in order to compare to the Hosek et al. [27] results as well as for consistency across samples.
IR AO Astrometry
Sakai et al. [47] present IR AO astrometry for seven maser-emitting IR-bright stars: IRS 9, IRS 7, IRS 12N, IRS 28, IRS 10EE, IRS 15NE, IRS 17. These are the stars used in Figure 4. This IR astrometry relies on earlier maser astrometric epochs for the offsets from Sgr A* (Reid et al. [45]). The optimal common epoch is 2017.3339. The error-weighted mean proper motion differences are (+0.004 ± 0.010, +0.041 ± 0.015) mas yr in RA and Dec; there is no significant (> 3) difference between the two reference frames. The weighted mean differences in offsets from Sgr A* are (+0.043±0.079, +0.450±0.118) mas. The Declination difference is significant, at 3.8. This is likely attributable
to the small sample whereby a single maser can significantly alter the mean. Indeed, a bootstrap resampling of the stars produces much larger uncertainties in the mean values. For the position offset differences, the uncertainties are estimated to be (±0.25, ±0.30) mas and the differences are not significant. Nonetheless, it is clear that a new IR-radio registration is needed.
2022 Maser Astrometry
Comparison to the Darling et al. [13] astrometry includes all 15 masers, and the 2026 measurements are effectively a refinement of previous work. Nevertheless, it is useful to assess the changes and to identify any masers that have significantly different positions or proper motions. The weighted mean proper motion differences are (+1.8±10.3, +0.3±14.6) μas yr, i.e., the proper motions are consistent, and the differences are extremely small. The weighted mean Sgr A* offset differences in the common epoch 2014.7250 are (−0.149 ± 0.081, 0.126 ± 0.129) mas and are not significant. Individual masers do not show significantly different proper motions or Sgr A* offsets in the common epoch, except for IRS 9. IRS 9 shows a 2.98 difference in RA offset, but it has the lowest position uncertainty in the sample, equivalent to 1.1 AU. We may be measuring changes in the maser light centroid at these scales, which is supported by the roughly factor of two change in brightness observed between 2022 and 2026 (Figure 2). This phenomenon may set a future astrometric precision floor on additional masers.
HST-Gaia Astrometry
Following the comparison of the Hosek et al. [27] HST-Gaia sample to the Darling et al. [13] stellar maser reference frame for 13 of the masers in this work, we re-examine the differences in proper motions and positions of the overlap sample.7 Unlike the Sakai et al. [47] astrometry, the HST-Gaia astrometry is not referenced to any maser measurements and represents an independent reference frame tied to the International Celestial Reference System (ICRS; Arias et al. [4]). We use the Hosek et al. [27] “Sgr A* at-rest” coordinates and proper motions to match the maser measurements. Based on Equations 10–12 of Hosek et al. [27], the optimal common epoch is 2017.8145, the error-weighted difference in proper motions is mas yr in RA and Dec, and the error-weighted difference in coordinate offsets from Sgr A* is mas. This preserves the sign convention (subtracting the HST-Gaia values from the maser values) but omits the systemic reference frame uncertainties between HST-Gaia and the maser reference frame. If these uncertainties are included in quadrature, the differences become mas yr and mas. These are consistent with the Hosek et al. [27] results, but with smaller uncertainties.
The differences in proper motion and offset between the stellar maser reference frame and the HST-Gaia frame are all significant (), especially the proper motion in RA () and Declination offset (). Hosek et al. [27] found tension in the RA proper motion and position; these are now confirmed. However, bootstrap resampling of the stars produces larger uncertainty estimates: mas yr in proper motion differences and mas in offset differences. The bootstrap bias is negligible. Based on the bootstrap, only the Dec offset remains significant, at . Figure 5 plots the proper motion and offset differences.

Figure 5. Difference between the Hosek et al. [27] HST-Gaia proper motions (left) and Sgr A* offsets (right) and the proper motions and offsets measured in this work. The red ellipse indicates the 3 contour around the error-weighted mean based on bootstrap resampling of the 13 stars in the comparison.
We note that the difference in RA coordinate offset shows a significant () gradient with RA across the field of view: mas per arcsec. No other differential quantity shows a significant gradient in either coordinate direction. Given the small mean proper motion and offset statistical uncertainties in the maser sample described in § 5 and depicted in Figure 3 and Figure 4, it is important to identify the systematic source of the differences between reference frames. The small number of maser-emitting stars is likely a factor, as demonstrated by the bootstrap uncertainty estimates. Another factor is likely to be the systematic biases in the Gaia astrometric solution (e.g., Lindegren et al. [33]), potentially exacerbated by the dearth of observable quasars near the Galactic Center.
Conclusions
This work presents a 30-year baseline for the stellar SiO maser-based reference frame tied to the continuum of Sgr A*. The weighted mean uncertainty in the maser offsets from Sgr A* is 0.25 mas, which is similar to the size of the SiO-emitting regions (2.0 AU). The mean proper motion uncertainties are mas yr or km s in RA and Dec. The reference frame stability reaches sub-km s levels: km s or mas yr. These represent a new level of precision, and the reference frame decay should be minor for years but should be refreshed in the early 2030s. The statistical uncertainties in the reference frame, however, show that systematic differences between this frame and the Hosek et al. [27] HST-Gaia frame must be reconciled.
At some point, acceleration will be detected in the masers via proper motion curvature and/or changing radial velocities. No credible accelerations have been detected to date, but the best limits are below km s yr (Paine & Darling [39]). Neglecting accelerations will lead to growing errors in the reference frame, but including them will enable new and more precise physical measurements. Future IR stellar tracking will use an accelerating reference frame, and it will be essential to match this to the masers. Acceleration measurement also determines a stellar maser’s physical distance from Sgr A* as well as the star’s orbital parameters. If acceleration is not detected, the acceleration upper limit corresponds to a distance lower limit. The inherent motion of the maser-emitting shell (which is detectable in variable maser spectra and is a source of noise in astrometry) may be confused with systemic acceleration of the star over short time scales, but any acceleration detected over several decades should reflect the stellar kinematics.
Acknowledgments
We dedicate this work to Karl Menten, a pioneer in masers, molecules, astrometry, and radio astronomy in general. We thank Mark Reid for sharing his expertise and guidance and for passing this interesting baton to the next generation, and we thank the anonymous referee for helpful feedback. JD acknowledges support from NSF grant AST-1908122. M.W.H. is supported by the Brinson Prize Fellowship. This research made use of NumPy [50], Matplotlib [29], and Astropy6, a community-developed core Python package for Astronomy [5, 42].
Facility: VLA, VLBA
Software: CASA [35], astropy [5, 42], NumPy [50], Matplotlib [29], lmfit [37], emcee [17], CARTA [9]
Astrometric Offset Time Series
Table A1 lists the Sgr A* coordinate offset time series for each maser used in this analysis. Figure A1 shows the time series in each coordinate, the least-squares proper motion fit residuals, and the time series on the sky.

Figure A1. Time series for each stellar maser, as an offset from Sgr A* (rows). We show the motion on the sky (right column; the arrows indicate the direction of proper motion), the time series in RA (left), and the time series in Dec (center). For each coordinate time series, the orange lines show the least-squares proper motion fits, and the upper panels show the time series residuals. The fit line width indicates AU in all panels. The time series is connected sequentially when it is ambiguous, as in SiO-16. RA offsets are corrected for .

Figure A1 (continued).

Figure A1 (continued, PDF p. 15).
Proper Motion Measurement Methods
Darling et al. [13] describes in detail several methods for obtaining proper motions from astrometric time series and finds that these generally do not provide significantly different outcomes from simple variance-weighted least-squares fits. In this work, we re-examined the methods and again find that the conventional method remains adequate. Nonetheless, sample comparisons are presented here. The fitting methods are:
a variance-weighted least-squares;
a “conservative formulation” of uncertainties that allows for large outliers in the data (Sivia & Skilling [49]; Darling et al. [14]);
a “good-and-bad data” model (Box & Tiao [8]; Sivia & Skilling [49]); and
a variance-weighted least-squares with expanded uncertainties as a free parameter (Foreman-Mackey et al. [17]).
Figure B2 shows examples of the proper motion fit residuals for the first three methods for two masers that show scatter inconsistent with their astrometric uncertainties, IRS 19NW and SiO-6. Even for these extreme cases, the proper motions obtained from the fitting methods (2) and (3) are formally consistent with the simple least-squares fit. Figure B3 shows proper motion fit residuals for methods (1) and (4) for the same masers. In this case, expanded uncertainties in the astrometry are favored, but the proper motions do not significantly differ () in part due to larger uncertainties in the method (4) proper motions. As seen in (Darling et al. [13]), the expanded uncertainties are larger than the expected size of the maser-emitting regions, so the astrometric scatter cannot be explained by changing structure of maser-emitting regions.
For the entire maser ensemble, expanded uncertainties are favored in roughly half of the sample, which is consistent with the distribution of values in Table 2. The proper motions obtained from expanding the uncertainties are never significantly different from the least-squares fits to the astrometric time series, and the reference frame stability changes by only in each coordinate.
References
- [1]Abd El Dayem, K., Abuter, R., Aimar, N., et al. 2026, Nature, 657, 359, doi: 10.1038/s41586-026-10894-w
- [2]Alexander, T., & Hopman, C. 2009, ApJ, 697, 1861, doi: 10.1088/0004-637X/697/2/1861DOI
- [3]Amaro-Seoane, P., & Preto, M. 2011, Classical and Quantum Gravity, 28, 094017, doi: 10.1088/0264-9381/28/9/094017
- [4]Arias, E. F., Charlot, P., Feissel, M., & Lestrade, J.-F. 1995, A&A, 303, 604
- [5]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068
- [6]Babak, S., Gair, J., Sesana, A., et al. 2017, PhRvD, 95, 103012, doi: 10.1103/PhysRevD.95.103012DOI
- [8]Box, G. E. P., & Tiao, G. C. 1968, Journal of the American Statistical Association, 63, 174, doi: 10.1080/01621459.1968.11009232DOI
- [9]Comrie, A., Wang, K.-S., Hsu, S.-C., et al. 2021, CARTA: The Cube Analysis and Rendering Tool for Astronomy, 2.0.0, Zenodo, doi: 10.5281/zenodo.4905459
- [10]Daly, R. A., Donahue, M., O’Dea, C. P., et al. 2024, MNRAS, 527, 428, doi: 10.1093/mnras/stad3228DOI
- [13]Darling, J., Paine, J., Reid, M. J., et al. 2023, ApJ, 955, 117, doi: 10.3847/1538-4357/acec71DOI
- [14]Darling, J., Truebenbach, A. E., & Paine, J. 2018, ApJ, 861, 113, doi: 10.3847/1538-4357/aac772DOI
- [17]Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125, 306, doi: 10.1086/670067
- [19]Genzel, R., Eisenhauer, F., & Gillessen, S. 2010, Reviews of Modern Physics, 82, 3121, doi: 10.1103/RevModPhys.82.3121DOI
- [20]Ghez, A. M., Salim, S., Weinberg, N. N., et al. 2008, ApJ, 689, 1044, doi: 10.1086/592738DOI
- [21]GRAVITY Collaboration, Abuter, R., Amorim, A., et al. 2020, A&A, 636, L5, doi: 10.1051/0004-6361/202037813
- [23]Gravity Collaboration, Abuter, R., Amorim, A., et al. 2021, A&A, 647, A59, doi: 10.1051/0004-6361/202040208
- [24]GRAVITY Collaboration, Abuter, R., Aimar, N., et al. 2022, A&A, 657, L12, doi: 10.1051/0004-6361/202142465
- [26]Hopman, C., & Alexander, T. 2006, ApJL, 645, L133, doi: 10.1086/506273DOI
- [27]Hosek, Jr., M. W., Do, T., Martinez, G. D., et al. 2025, ApJ, 989, 79, doi: 10.3847/1538-4357/ade803DOI
- [29]Hunter, J. D. 2007, Computing in Science Engineering, 9, 90
- [30]Leung, H. W., Bovy, J., Mackereth, J. T., et al. 2023, MNRAS, 519, 948, doi: 10.1093/mnras/stac3529DOI
- [31]Lewis-Merrill, R. A., Do, T., Hosek, Jr., M. W., et al. 2026, ApJ, 1009, 16, doi: 10.3847/1538-4357/ae878c
- [33]Lindegren, L., Klioner, S. A., Hernández, J., et al. 2021, A&A, 649, A2, doi: 10.1051/0004-6361/202039709DOI
- [34]Liu, K., Wex, N., Kramer, M., Cordes, J. M., & Lazio, T. J. W. 2012, ApJ, 747, 1, doi: 10.1088/0004-637X/747/1/1DOI
- [35]McMullin, J. P., Waters, B., Schiebel, D., Young, W., & Golap, K. 2007, in Astronomical Society of the Pacific Conference Series, Vol. 376, Astronomical Data Analysis Software and Systems XVI, ed. R. A. Shaw, F. Hill, & D. J. Bell (Astronomical Society of the Pacific), 127
- [36]Menten, K. M., Reid, M. J., Eckart, A., & Genzel, R. 1997, ApJL, 475, L111, doi: 10.1086/310472DOI
- [37]Newville, M., Otten, R., Nelson, A., et al. 2021, lmfit/lmfit-py: 1.0.3, Zenodo, doi: 10.5281/zenodo.5570790
- [38]Oyama, T., Miyoshi, M., Deguchi, S., Imai, H., & Shen, Z.-Q. 2008, PASJ, 60, 11, doi: 10.1093/pasj/60.1.11DOI
- [39]Paine, J., & Darling, J. 2022, ApJ, 927, 181, doi: 10.3847/1538-4357/ac3a87DOI
- [40]Piran, T., Amaro-Seoane, P., Aytac, B., et al. 2026, arXiv e-prints, arXiv:2607.24931, doi: 10.48550/arXiv.2607.24931DOI
- [42]Price-Whelan, A. M., Sipőcz, B. M., Günther, H. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f
- [43]Reid, M. J., & Brunthaler, A. 2020, ApJ, 892, 39, doi: 10.3847/1538-4357/ab76cdDOI
- [44]Reid, M. J., Menten, K. M., Genzel, R., et al. 2003, ApJ, 587, 208, doi: 10.1086/368074DOI
- [45]Reid, M. J., Menten, K. M., Trippe, S., Ott, T., & Genzel, R. 2007, ApJ, 659, 378, doi: 10.1086/517744
- [46]Reid, M. J., Menten, K. M., Brunthaler, A., et al. 2019, The Astrophysical Journal, 885, 131, doi: 10.3847/1538-4357/ab4a11DOI
- [47]Sakai, S., Lu, J. R., Ghez, A., et al. 2019, ApJ, 873, 65, doi: 10.3847/1538-4357/ab0361DOI
- [48]Silva, R. P., Correia, A. C. M., & Garcia, P. J. V. 2026, arXiv e-prints, arXiv:2610.07123, doi: 10.48550/arXiv.2610.07123DOI
- [49]Sivia, D. S., & Skilling, J. 2006, Data Analysis - A Bayesian Tutorial, 2nd edn., Oxford Science Publications (Oxford University Press)
- [50]van der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science Engineering, 13, 22
- [51]Will, C. M. 2008, ApJL, 674, L25, doi: 10.1086/528847DOI