2 Observations and data reduction

We carried out VLBA observations of the quasar 3C395 switching between frequencies at 8.4 and 15.4 GHz in November 27th 1995 (epoch 1995.91) and at 15.4 and 22.2 GHz in June 30th 1998 (epoch 1998.50). Left and right circular polarizations were recorded during both observing runs. The synthesized bandwidths per circular polarization were 16 MHz and 32 MHz in 1995 and 1998, respectively.

The correlation of the data was done in absentia by the staff of the VLBA correlator in Socorro (NM, USA). We used the NRAO AIPS package to correct for instrumental phase and delay offsets between the separate baseband converters in each antenna, and to determine antenna-based fringe corrections. The visibility amplitudes were calibrated using the system temperatures and gain information provided for each telescope. We estimate amplitude calibration errors to be smaller than 5% at all observing frequencies.

In both epochs, the determination of the feed responses to the polarized signal at each antenna was done using the feed solution algorithm of Leppänen et al. (1995), which calculates the so called D-terms (that is, the terms describing the "leakage'' of the orthogonal polarization into each feed). Independent determinations of the instrumental polarization were done for each observed source, including 3C395, obtaining a good agreement in the results at all frequencies and epochs. This allowed us to apply the corrections derived from a given source in the polarization mapping of that source. During the calibration, we assumed that the circularly polarized emission from all the sources was negligible, as suggested by the data. In 1995, the number of known sources suitable for polarization calibration of VLBI observations at frequencies higher than 15 GHz was very small. We took snapshots of the sources 1656+053, 2145+067, 3C84 and OQ208, hoping to find a good calibrator of the absolute orientation of the electric field vector (electric vector position angle - EVPA) among them. The final EVPA was derived from the direct comparison of Very Large Array (VLA; close-in-time data were requested from the VLA public archive) and our VLBA polarization images of 1656+053, which presents a compact structure at the angular resolutions provided by both instruments. From the VLA data we determined an EVPA = $-15^{\circ}$ for this source. This orientation was consistent with that derived from the D-terms determined by Leppänen from a close-in-time VLBA experiment at 8.4 GHz (private communication). In 1998 the number of known polarization calibrators had increased considerably. We observed 3C279, 1611+343 and 0420-014, all with known polarization properties, and we could obtain a consistent calibration for all of them. The absolute EVPA was derived from 0420-014, comparing our VLBA observations with the UMRAO database, and from the outer component in 3C279 (Taylor & Myers 2000). The EVPA were determined with an error that we estimate to be within $10^{\circ}$ at all epochs and frequencies. Data imaging in total intensity was performed with the Difmap package (Shepherd et al. 1994). Maps of the Stokes parameters Qand U were made and combined in AIPS to finally obtain maps of the linearly polarized emission of 3C395.

Figure 1: a-d), from top to bottom, VLBA maps of 3C395 made at 8.4 GHz and 15.4 GHz in 1995 (X-95 and U-95, respectively), and at 15.4 GHz and 22.2 GHz in 1998 (U-98 and K-98, respectively). In all maps contours are spaced by factors of 2 in brightness. The vectors represent the polarization position angle ($\vec{E}$-vector), with length proportional to the polarized flux. The maps have been obtained applying natural weighting to the uv data. Figure 1d shows an enlargement of the innermost jet, obtained after applying uniform weighting to the 22.2 GHz data, with components labeled according to Sect. 3.1. For each map we list the Gaussian beam size (in mas), the first contour level (mJybeam-1), the peak of brightness (Jybeam-1) and the polarized flux corresponding to 1 mas $\vec{E}$-vector length (mJybeam-1). a) beam = 1.6$\times$1.1 PA $-5.3^{\circ }$; 1st cntr = 0.80; peak = 1.017; 1 mas $\equiv$ 4.0, b) beam = 0.85$\times$0.58 PA $-6.6^{\circ }$; 1st cntr = 0.80; peak = 0.683; 1 mas $\equiv$ 3.33, c) beam = 0.84$\times$0.62 PA $-1.8^{\circ }$; 1st cntr = 0.80; peak = 0.612; 1 mas $\equiv$ 2.5, d) beam = 0.59$\times$0.42 PA $-3.9^{\circ }$; 1st cntr = 0.94; peak = 0.548; 1 mas $\equiv$ 2.5, d (zoom): beam = 0.46$\times$0.32 PA $-6.4^{\circ }$; 1st cntr = 1.50; peak = 0.53