2 Observations

2.1 Radio observations

All radio observations were performed with the Australia Telescope Compact Array (ATCA). The ATCA synthesis telescope is an east-west array consisting of six 22 m antennas. The 8.6 GHz data that we used is from Corbel et al. (2000); however, we re-analaysed all (five) observations for which the radio flux densities were weaker than 1 mJy. Further details concerning the ATCA and its data reduction can be found in Corbel et al. (2000). We also added the result of a series of three new ATCA observations, for a total duration of nearly 20 hours, performed on 2000 September 12, 15 and 18, during the recent off state. These observations provided a strong (99% confidence level) upper limit of 60 $\mu$Jy at 8.6 GHz, which is the best constraint we have for the level of radio emission originating from GX 339-4 during its off state.

2.2 X-ray observations

2.2.1 RXTE

We used the Rossi X-ray Timing Explorer (RXTE) to perform a number of observations of GX 339-4 from 1997-1999, most of which represent the LHS. Analysis of the brightest of these observations was previously presented (Nowak et al. 2002); here we also consider analyses of faint, "off state'' observations from late 1999. All data extractions were performed in an identical manner as that described by Nowak et al. (2002). The flux in various energy bands was determined by fitting a model comprised of neutral hydrogen absorption ($N_{\rm H}$ was fixed to $6\times10^{21}~{\rm cm^{-2}}$), a multi-temperature disk blackbody (e.g., Mitsuda et al. 1984) with peak temperature fixed at 0.25 keV, an exponentially cut-off broken power law with break energy at $\approx$10 keV, and a (potentially) broad Gaussian line with peak energy fixed at 6.4 keV. The faintest observations were fit with a simpler absorbed, single power law, plus (fixed peak energy) line feature. Feng et al. (2001) found that the iron line is shifted to higher energies when GX 339-4 was observed at low X-ray fluxes. However, this shift is not intrinsic to GX 339-4, but is rather due to the Galactic diffuse emission (Wardzinski et al. 2003). Note that due to differences between the two sets of instruments that comprise RXTE, the Proportional Counter Array (PCA, ${\approx} 3{-}20$ keV) and the High Energy X-ray Timing Experiment (HEXTE, $\approx$20-200 keV), a normalization constant between the PCA and HEXTE detectors was used, and all flux values are normalized to the PCA flux levels (for further descriptions of this process, see Wilms et al. 1999). The flux error bars were chosen to be the larger of the statistical error, or 1%, which is a reasonable estimate of the RXTE internal systematic error (e.g., Wilms et al. 1999). Short timescale ($\le$few seconds) X-ray variability is usually observed in the low-hard state of GX 339-4 (Smith & Liang 1999; Nowak et al. 2002). However, on a longer timescale (e.g. 10 min) the radio emission is steady (see Fig. 3 in Corbel et al. 2000) and so also is the X-ray spectrum of GX 339-4 integrated on those timescales (i.e., there is almost no very low frequency power in the power spectral densities, Nowak et al. 2002 and references therein). Therefore the error bars used in Table 1 are likely not affected by the variability of the source (which is quite steady on timescales greater than 10 min). Note that in Table 1 we quote the absorbed flux level; however, as we only consider energies $\ge$3 keV, this is at most a few percent different to the unabsorbed flux level.

RXTE has a broad ${\approx} 1^\circ$ radius field of view, and therefore is potentially subject to contamination from faint background sources (or, in the case of GX 339-4, diffuse emission from the galactic ridge, Wardzinski et al. 2003). Four of the RXTE observations, however, were performed simultaneously with the much narrower field of view ($\approx$4 arcmin radius) Advanced Satellite for Cosmology and Astrophysics (ASCA). Utilizing the same models described above, the 3-9 keV flux of the brightest two simultaneous observations determined by ASCA was 75-81% of that determined by the PCA - consistent with a well-known calibration offset between PCA and ASCA (see the discussion in Nowak et al. 2002). For the faintest two observations, the relative normalizations of the ASCA spectra substantially decreased with decreasing flux. This was taken as evidence for a faint background source or sources that lie within the field of view of the PCA, but not within the field of view of ASCA. The RXTE observation of 1999 July 29 is assumed to be heavily dominated by this contaminating source, and this spectrum, multiplied by 0.78, is subtracted as a "background correction'' before model fitting and flux determination, from all RXTE observations occurring later than the observation of May 14 1999. With this additional background subtracted, the ASCA determined fluxes of the faintest two simultaneous observations become 73% and 83% of the corrected PCA 3-9 keV flux levels. Good agreement also is obtained between the ASCA and the corrected PCA spectra.

2.2.2 BeppoSAX

During the recent off state of GX 339-4 we conducted an $\sim$50 ks X-ray observation with BeppoSAX on September 10 2000. GX 339-4 is detected in the 1-10 keV energy range with both LECS and MECS, and the spectrum can be fitted with a power-law with a photon index 2.22 $\pm$0.24 (90% confidence level) with interstellar absorption fixed to 5.1 $\times\ 10^{21}$ cm^-2 ( $\chi_0^2 = 0.74$ for 35 degrees of freedom). The absorbed flux in the 3-9 keV energy range is $5.7 \times 10^{-13}$ erg cm^-2 s^-1 (relative to the MECS normalization). To within a few percent, the fluxes normalized to MECS are consistent with the ones normalized to PCA (e.g. Della Ceca et al. 2001). We also re-analysed the BeppoSAX observation performed on August 13 1999 by Kong et al. (2000), as it was close to the date of one of our radio observations. All measurements (radio and X-ray) are tabulated in Table 1.

Figure 1: The radio flux density at 8.6 GHz is plotted versus the X-ray flux in the 3-9 keV energy band. The continuous line denotes the fit to the data with the function described in the body of the paper and with the parameters estimated in Table 3, the dotted line represents the one-sigma deviation to those parameters. Upper limits are plotted at the three sigma level. The diamond points are those points that are not strictly simultaneous (1999.08.17) or maybe affected by a small reflare observed in hard X-rays (1999.09.01, see Fig. 15 in Corbel et al. 2000).

Figure 2: Same as Fig. 1, but for the X-ray flux in the 9-20 keV energy band.