SHEVE observations of the radio source associated with GRO J1655-40 were obtained at short notice as a target of opportunity following the first detection of the radio source. Initially the real-time 275 km Parkes-Tidbinbilla interferometer (PTI) was used to determine that rapid changes were taking place in the source [Reynolds & Jauncey 1994]. Based on this information, the SHEVE array of telescopes, supplemented with several other telescopes, observed the source on the four days 1994 August 21, 22, 23, and 24 at a frequency of 2.3 GHz, for the purpose of imaging the source structure and following its evolution. The telescopes which participated were: Tidbinbilla (34 m, 70 m), Parkes, Hobart, Mopra, Narrabri, Hartebeesthoek, Goldstone (DSN, 70 m), and Mauna Kea (VLBA, 22 m). The data were all recorded in Mark II mode, obtained and processed as described in chapter 2.
The data tapes were quickly processed at the Caltech/JPL correlator, the data calibrated, and images produced. The entire experiment was turned around in less than two weeks.
During the correlation and fringe-fitting stages of data reduction it became apparent that the source was not detected on any inter continental baselines. The usable data were restricted to the baselines between the telescopes in Australia. On 1994 August 21 and 22, the source was detected at a useful signal to noise (> 7 
) on all baselines except the longest baselines of the array, those from Mopra and Narrabri to Hobart. The Mopra-Hobart and Narrabri-Hobart baselines were removed due to this reason. Thus, several closure phase triangles were lost and all groups of 4 telescopes involving Hobart were lost. Therefore, during imaging, the Hobart data were used during the initial iterations of phase self-calibration but were not constrained when amplitude self-calibration was attempted, and were automatically removed from the dataset by the imaging software, DIFMAP. On 1994 August 23 and 24, no fringes on any baselines to Hobart were detected at a usable signal to noise.
Figures 6.1 (1994 August 21), 6.2 (1994 August 22), 6.3 (1994 August 23), and 6.4 (1994 August 24) show the images resulting from the data for each observation. The source consists of two components, each elongated along the position angle which joins them, 48 
, and separated by approximately 0.''5. The brightest component in the source is the south-west component, which is also the most compact component. The north-east component appears to have significant sub-structure at each of the four epochs. Over the first two days the north-east component contains three sub-components, but on the last two days only two remain. The middle component appears to have faded more quickly than its neighbours.
A comparison of the relative positions of the two components over the four day period indicates that they were separating at a rate of 
60 mas/day, in general agreement with the results of the PTI observations. At the resolution of these observations the components were moving at a rate greater than one synthesised beam width per 12 hours, the time taken for a full synthesis observation.
Figure: Map peak, 0.3 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, 64% of peak. Beam FWHM, 80.9 
26.8 mas @ -81.9 .
Figure: Map peak, 0.4 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 67.6 24.6 mas @ -84.9 .
Figure: Map peak, 0.1 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 71.4 27.9 mas @ 87.9 .
Figure: Map peak, 0.2 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 74.8 27.1 mas @ -85.4 .
These observations create an unusual problem in that they violate a fundamental assumption of synthesis imaging, that the source structure remains constant over the period of the observation. The effect of a 60 mas/day motion of a source observed with the SHEVE array at 2.3 GHz is illustrated in Figure 6.5.
The data shown in Figure 6.5 were simulated with the program FAKEM, a modified version of the Caltech VLBI program FAKE (A.K. Tzioumis 1994, private communication) which is used to generate the data expected at a given time, for a given source model, and a given array of telescopes. FAKEM accepts source models with two extra parameters. For each model component one can define an angular motion for the component in mas/day, relative to the phase tracking centre, d, and a position angle for the component motion, 
.
Figure 6.5 shows simulated data for a source with a model, similar in structure to GRO J1655-40, as given in Table 6.1, over a period of approximately 12 hours, on 1994 August 21, with the Tidbinbilla, Parkes, Mopra, and Narrabri antennae. The celestial coordinates for the source model are those of GRO J1655-40. The solid line shown in Figure 6.5 corresponds to the source model as it was at the beginning of the observation. The comparison between stationary model and data from the evolving model shows that a large change in the source structure is apparent between the start of the observation and the end of the observation.
Figure 6.5: Stationary model and data from evolving model
Table 6.1: Dynamic model to illustrate motion in GRO J1655-40
Imaging simulations on data produced from FAKEM by A.K. Tzioumis (described in Tingay et al. 1995) revealed that data from sources which change their structure of the order of a beam width during the period of the observation produce significant distortions in the resultant images. Therefore, some strategy was required to ensure that reliable images could be produced from the real GRO J1655-40 data.
As yet, no general imaging or model-fitting algorithm exists to handle data for sources with rapidly varying structures. Therefore, guided by the simulation shown in Figure 6.5 and the data for GRO J1655-40, it was decided that the validity of the features in Figures 6.1 - 6.4 could be confirmed by considering only a subset of the original data. An inspection of Figure 6.5 shows that over the first 5 hours of observation the data and stationary model are a reasonable match. Only after this period do they become seriously discrepant. The first 5 hours of the observation also include the visibilities which contain the majority of the information constraining the source, as seen in the real data shown in Figure 6.10 from 1994 August 21.
Figure: SHEVE GRO J1655-40 data, 1994 Aug. 21
From the 4 original data sets, 4 new data sets were produced which contained only the first 5 hours of data from each observation. From these data, the images in Figures 6.7, 6.8, 6.9, and 6.10 were produced. These new images confirm the overall structure of the source and in particular the sub-structure of the north-east and south-west components. Now, with 5 hours data, the source components separate only 13 mas over the observation period, approximately half of the small dimension of the restoring beam. The dynamic range and resolution of the images has been reduced however.
Figure: Map peak, 0.4 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 123.0 25.7 mas @ 80.1 .
Figure: Map peak, 0.5 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 106.0 23.4 mas @ 76.4
Figure: Map peak, 0.2 Jy/beam. Contours, -2, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 123.0 25.9 mas @ 72.2
Figure: Map peak, 0.2 Jy/beam. Contours, -1, 1, 2, 4, 8, 16, 32, and 64% of peak. Beam FWHM, 106.0 25.6 mas @ 78.4
Finally, the four images resulting from the restricted data sets have been rotated by 42 , shifted vertically, aligned on the brightest component at the south-west end of the source, and plotted in Figure 6.11 on a common flux density scale. The peak flux density in the montage is 0.6 Jy/beam and the flux density contours are 1, 2, 4, 8, 16, 24, 32, 64, 80, and 95% of the peak. The restoring beam for the montage is 123 23 mas at a position angle of 72 . The measurement of the positions of the sub-components in the north-east component, relative to the south-west component, gives a value for the angular speed of expansion of the source of 65 
5 mas/day.
If this expansion is extrapolated back in time to the point at which the two components were coincident, the zero separation date is 1994 August 13.5 
. This date is approximately one day after the rapid rise in radio flux density [Campbell-Wilson & Hunstead 1994] and close to the end of the decline in X-ray luminosity [Wilson et al. 1994].
Figure 6.11: Final SHEVE images of GRO J1655-40