The comparison of the photoionization models with the observations will of course also require a detailed analysis of the emission lines over the large range of variability displayed by F-9. Especially the large amplitude of the variability offers a good opportunity to evaluate the possible existence of different response characteristics over the extent of the broad lines. The line analysis has been done in three stages following the same procedures as applied by Wamsteker et al.\ (1990) for NGC 5548: first we have removed the various continuum contributions, since especially the small blue bump, mainly composed of Balmer continuum and the many blended FeII lines, form a semi-continuum, which could affect the final results of the line analysis. After the continuum has been properly allowed for, we are left with a pure emission line spectrum in which one can reliably measure the integrated line intensities. The third part is the detailed line decomposition from essentially pure emission line spectra, and derive the details of the line response for the different velocity domains over the broad line profile. In this last part the actual line variability has been used through the differencing method to derive a line model which will allow the component decomposition for all lines in an internally consistent way for all brightness levels with a minimum number of components.
The nuclear continuum emission, as derived above in Sect. 3.1.2,
has been
applied to the UV spectra, i.e. a variable power law spectrum, scaled to
, with constant spectral index 
has
been subtracted from each spectrum. The Balmer continuum (BaC) and the UV FeII
semi-continuum have been allowed for using the synthetic models by Wills
et al. (1985) as applied by Wamsteker et al. (1985) to
the optical, in the region of 
, with a gaussian velocity smoothing of
4000 
. Although no physical conclusions can be derived from the specific
model chosen, the BaC model is characterized by an effective temperature
of
, and an effective optical depth in the Balmer limit of 
. The FeII models used are characterized by a turbulent velocity of
, and an optical depth in the UV3 multiplet at 2343 Å of
. With these theoretical models a fit to the observed
semi-continuum is made between 
(i.e. the Small Blue
Bump), allowing to isolate the broad lines and to measure their total intensity
for each spectrum with some reliability. To establish the intensity of the BaC we
used the photometry of Lub & De Ruiter (1992) as an intermediate
step. After subtracting the stellar contribution and the power law continuum we
matched the BaC shape using the photometry in the L, U, and W bands at
respectively 3837 Å, 3623 Å\
and 3235 Å. At the redshift of F-9, the L band falls at a rest wavelength of
3668 Å(3562 - 3774 Å), essentially free of FeII and centered on the Balmer
limit at 3646 Å. Therefore, we have used the measured flux at this band to
determine the intensity of the BaC. Subtracting those contributions also from the
other two bands, we have fitted the intensity to the observed one at 3837 Å.
From this we choose from the 12 synthetic models the model which gave the minimum
residual flux in the U and W bands, because FeII emission in these bands is
minimum. To determine the BaC intensity to all UV spectra we used the relation
found between the BC(3837 Å) and the interpolated 
(Table 2 (click here) and Fig. 4 (click here)a):
with r=0.76.
This procedure to determine the BaC from the observations worked very well except
when the UV flux was high, larger than 13-15 
. Above this
brightness it was necessary to decrease the model intensity, because it would
rise above the observations. This suggests that the correlation between BaC and
(Eq. 8) does not persist for UV high levels (see also
Fig. 4 (click here)a, where the three last points (
) are somewhat indeterminate in this respect). Such dependence of the
BaC intensity with the UV brightness is similar to
that shown by the hard X-rays (Figs. 4 (click here)a, b). After subtracting the
BaC we fit the selected
optimum FeII spectrum to the region between 
and 
, adjusting its intensity
to each observed spectrum. After subtracting these continuum components a pure
emission line spectrum remains, where the continuum slope is not any longer a free
parameter in the line fitting. This process to isolate the emission lines is illustrated in
Figs. 6 (click here)a-c for three different levels of the UV brightness. When
the UV flux is high (Fig. 6 (click here)a) the BaC models matched using Eq.
(8) are shown too. It is clear that the observed 
intensity becomes too large. The BaC and the FeII (UV) total intensities
are given in Table 6 (click here) and their relation with the UV continuum is
shown in Fig. 8 (click here). The errors in the BaC intensity have been
calculated from the linear relation (8) and the UV continuum errors. We find
a mean relative error of 13%, similar to the error of 8% obtained for
those epochs when more than one spectrum is available. For the FeII
uncertainty this second method has been used and a mean relative error of
17% was assumed for the case when only one observation was available.
Figure 6: a) Fit to the UV continuum with a power law, Balmer
continuum and the FeII model for a spectrum at high UV level.
: Observed spectrum with the power-law continuum,
corrected for extinction.
: Spectrum after subtraction of the power law continuum;
the Balmer continuum model is also shown. The dotted line shows the model obtained
from the application of Eq. (8) at high levels; the actually applied model
for the high UV levels is shown as a full drawn line.
: Spectrum after subtraction of the power-law continuum
and the Balmer continuum, together with the FeII model used (solid line)
5
Figure 6: b) Fit to the UV continuum with a power law, Balmer
continuum and the FeII model for a spectrum at intermediate UV level.
: Observed spectrum with the power-law continuum,
corrected for extinction.
: Spectrum after subtraction of the power law continuum;
the Balmer continuum model is also shown. The full drawn line shows the model
obtained from the application of Eq. (8).
: Spectrum after subtraction of the power-law continuum
and the Balmer continuum, together with the FeII model used (solid line)
5
Figure 6: c) Fit to the UV continuum with a power law, Balmer
continuum and the FeII model for a spectrum at low UV level.
: Observed spectrum with the power-law continuum,
corrected for extinction.
: Spectrum after subtraction of the power law continuum;
the Balmer continuum model is also shown.
: Spectrum after subtraction of the power-law continuum
and the Balmer continuum, together with the FeII model used (solid line)
Table 5: Total intensities of the UV lines (1)
Table 6: Total intensities of the UV lines (2)
Our main purpose with respect to the lines is the study of the line variability over
the profile structure. Some UV lines, such as 
and 
, can not be easily
analyzed in this way due to low S/N. Since there are still only very few
quantitative
results available addressing the detailed broad line structure (NGC 3783,
Pelat et al.
1981; 3C 390.3, Zheng et al. 1991; NGC 5548, Wamsteker
et al. 1990; Fairall-9, Wamsteker et al. 1985), we also report
here the total broad line intensities in Tables 5 (click here) and 6 (click here).
For the lines which have been decomposed, as is the case for 
, 
,
, 
\
and 
, we have used the results of the next section and give in Table
5 (click here) the sum of the intensities of the different components. This
allows us furthermore to measure strongly blended lines such as 
and
. We have used the error measure of Rodríguez-Pascual
(1989) and Wamsteker et al. (1990). This method assumes
that the component errors are perfectly correlated, and thus gives an upper
limit for the error. The relations between the UV continuum and the total
line intensities are shown in Figs. 7 (click here) and 8 (click here). For 
\
and 
, we have measured the intensity by direct integration over the
line width and the errors correspond to the product of the integration
interval with the rms error of the UV continuum (because the continuum
determination is the major source of error).
The relations shown in Figs. 7 (click here) and 8 (click here) display some interesting characteristics:
Figure 7: Total intensities of the UV lines versus 
(1)
Figure 8: Total intensities of the UV lines versus 
(2)
, 
, 
, 
, 
and 
) are strongly
correlated with the continuum, increasing in intensity with the continuum. The
simultaneous variation of the continuum with these many lines of different excitation
potentials confirms that photoionization, by the UV continuum between 13.6 and
77.5 eV, is the main ionization mechanism and that the continuum at 1400 Å is a
direct extension of it. For 
this relation breaks down at large continuum
brightness (the Wamsteker-Colina effect; Shields et al. 1995). When
the continuum level is high (>16 
), 
does not increase any
further. Wamsteker & Colina (1986) suggested this to be caused
by the fact that the emission takes place under matter limited conditions, i.e.
optically thin conditions, which have been discussed recently in more detail by
Shields et al. (1995). The amplitude of the line intensity
variations (Table 7 (click here)) does not show significant differences between
the lines. Clavel et al.\
(1989) found from the CCFs for 
and 
delays of 
and 
, respectively, very similar within the errors,
suggesting only limited stratification of the BLR. On the other hand the
difference in line response characteristics seen in Fig. 7 (click here) and the
Wamsteker-Colina effect may hide a real stratification. Koratkar &
Gaskell (1989) obtain for 
\
a delay of 
and Lub & De Ruiter (1992)
estimate for 
\
and 
a delay of 
with respect to the optical
continuum in the B band (4298 Å). Attempts to determine the transfer
function (TF) through detailed Echo-mapping using both MEMECHO (Horne et
al. 1991) as well as the Regularization method (Vio et al.
1994), do indicate the presence of two peaks in the TF for 
: a
primary peak at 100 days and a secondary peak at 380 days. No response is
indicated at zero delay. These results are, however, rather unstable and could
easily be the consequence of non-linear response of both 
and 
to the
variations in the ionizing flux. This appears to be confirmed by the
fact that the TF for 
and 
are essentially mutually exclusive, and no
acceptable
continuum lightcurve can reproduce the line lightcurves unless the linear response
condition is released.
Table 7: Variations amplitudes of the lines and components
intensities
and FeII show, if any,
a weak correlation with the continuum for
. At higher continuum levels they are
constant. This can be understood if these lines originate in the inner BLR, where
only hard X-rays with energy 
can penetrate. Although the
sampling of the X-ray data is limited, the results shown in
Fig. 4 (click here)b suggest the existence of a similar cutoff in the
vs. 
relation, strongly suggesting that
direct X-ray heating is indeed the dominant mechanism for these lines.
Clavel et al.\
(1989) and Koratkar & Gaskell (1989) found from CCF
analysis delays of 
for 
and 
for 
\
and 
\
with respect to the UV continuum. Lub & De Ruiter (1992) estimate
the MgII delay at 
with respect to the optical
continuum.
, at continuum brightness above
. This could be understood in similar terms as for the low
ionization lines except for the fact that a substancial part of the BaC
emitting region is presumably heated by the lower energy part of the continuum just beyond the Lyman limit.
As a consequence, the BaC can continue to respond to the increase in the UV
brightness beyond levels shown by the low ionization lines.
)/
, similar to
our result of FeII 
(in our case the optical
multiplets have not been summmed; see Table 8 (click here)). These authors
explain this large value by an overabundance of Fe, by a factor of 
with respect to the cosmic abundance.
Table 8: Observed lines ratios and comparation with the
theoretical prediction by Kwan & Krolik (1981)
We have applied the difference method to this homogeneous set of spectra over a
long temporal interval and large variability to isolate the variable components in the
emission lines. This method has been first applied by Wamsteker et al.\
(1990) to the UV and optical spectra of NGC 5548 and has here been used
for 
, 
, 
and 
. With only four components, all line profiles
of F-9 could be described over the whole brightness range. These four basic
components are identified in Table 9 (click here). From a large set of
subtracted spectra we determine the average of the FWHM, position
(= central wavelength) and the height of each component. These are introduced
as input parameters in the total profile fit. Similarly to Wamsteker
et al. (1990), we represent the components as gaussians. This
procedure makes it also possible to measure some of the highly blended
lines by introducing the blends as individual components in an overall line
fit. For example the 
-
blend must be fitted together in the 
\
region, to account for the 
presence (one narrow and one red component). In
, a very weak component at 
is most likely associated with NIV]
.
The fitting with 
minimization (the ESO-IHAP procedures have been used)
proceeded in two phases:
first, the width and the position are fixed, allowing only the height to vary. After this
fit has converged, position and FWHM are released to assure that the solution
remains stable in these two parameters as well. To illustrate the results of the gaussian
fitting to the data with the line model of Table 9 (click here), we show in
Fig. 9 (click here) the results for
and 
at three different levels of brightness. The broad emission lines show
the following components structure: a narrow component with
FWHM 
(unresolved), a central broad component at the same
velocity as the narrow line with 
, a broad red
component with FWHM 
and relative velocity of
, and a broad blue component with FWHM 
and a relative velocity of 
. The intensities in the
different line components are given in Tables 10, 11, 12 and 13 for 
(with 
),
, 
and 
, respectively. The errors are calculated as described by
Rodríguez-Pascual (1989) and Wamsteker et al.\
(1990) using the errors in the total line intensity. The relative errors
in the total intensity are for 
, 
, 
, 
and 
8%, 26%,
44%, 63% and 85%, respectively.
Figure 9: This figure shows the components and total fit to 
and 
for
three different brightness levels of the UV continuum (high = 28-8-78;
intermediate = 30-8-81; and low = 29-10-84)
Table 9: Profiles components of the UV lines
The width of the narrow lines is defined by the resolution of IUE at a FWHM of
(Table 9 (click here)). The fractional contribution of the
narrow line to the
total line intensity is at 23, 17, 11, 52 and 1% for 
, 
,
, 
and 
, respectively. Although the narrow lines do not show a clear
relation with 
, the results do not suggest a pure scatter
diagram, but over the time interval of the UV observations no systematic behavior
could be identified (see Fig. 10 (click here)). The average intensity ratios
for the different narrow components of the
lines are 
/
, 
/
, 
/
and 
/
(with large errors due to the large dispersion in the
data).
The line decomposition of 
by Wamsteker et al. (1985) showed
that the intrinsic narrow line width is 670 
. Even so the ratio 
/
\
is 
, two times higher than the Case B of recombination, but similar
to the obtained one by Ferland & Osterbrock (1986) with their
Seyfert 2 galaxies sample (
/
(intrinsic) 
). They explain
this as due to the fact that the NLR is photo-ionized by a
hard optical-X-rays continuum (
, although also in
their data a resolution effect can not be excluded. Our observations (Sect.
3.2.5)
also suggest a hard UV-X-rays spectral index, 
or -1.38.
Figure: Intensities of the narrow components of the strong emission lines
versus the UV continuum brightness at 1400 Å
This component (Table 9 (click here)) has an average velocity respect to the
narrow line of 
, while its mean FWHM is 
.
Its contribution to the total intensity is respectively 32, 38, 48 and 39%
for 
, 
, 
and 
. Its behavior with respect to the UV
continuum (Fig. 10 (click here)) shows that all the lines correlate with 
, though 
stops increasing at higher levels of continuum
(>10 
). As seen in Fig. 11 (click here), 
and 
\
appear to show a looping behavior
in line vs. continuum flux diagram, strongly suggesting delay with respect to the
continuum (in the 
figure the time sense has been indicated by an arrow). The
amplitudes of the central component for the different lines is given in
Table 7 (click here).
In Fig. 12 (click here) (see also Table 14 (click here)), we present the CCF
(Gaskell & Peterson 1987) and the Discrete Correlation
Function (DCF; Edelson & Krolik 1988) for the central
components with respect to the UV continuum are shown. For the DCF
calculus we have chosen an interval between points of 100 days, according
to the mean interval between two consecutive observations (96 days).
Although with this interval the DCF is noisy, it has the advantage that
the time resolution in the data is not degraded. For the CCF we indicate
the peak (with its error calculated as Gaskell & Peterson
1987) and the centroid of the function, calculated as the average of
the positive and negative HWHM. The peak of the CCF seems to be more
influenced by the response of the gas nearer the source (Robinson &
Pérez 1990; Pérez et al. 1992), while the centroid
seems to indicate the radius where the radiation is larger (Koratkar
& Gaskell 1991). For the DCF the delay of the maximum point is
given, more compatible with the center of the CCF than with the peak.
There are no significant differences between the delays of the CCF
centers for 
, 
, 
and 
, so that an averaged delay is 
(average error). This is consistent with the averaged delay derived from
the DCF of 
and both values are very similar to the
delay found for the region of hot dust emission of 
by
Clavel et al.\
(1989).
The line ratios for the central components for 
/
, 
/
as well as
/
appear to show a slight increase with the UV continuum
(Fig. 18 (click here)). The mean ratios
are 
, 
and 
, respectively. This central component is
also present in the 
profile decomposition (Wamsteker et al.\
1985), with a FWHM=3730 
and a relative velocity with respect to
the narrow line of +130 
.
Interpolating in 
to obtain simultaneous values with the 
central
component a mean ratio 
/
is obtained. Lub & De
Ruiter's (1992) optical study do not decompose the Balmer lines, on the
other hand they find a strong dependence of the delay versus the B continuum
(at 4298 Å), obtaining 
for 
, 
for 
and
for 
. Note that this central component could not
be discriminated in the lines of 
and 
suggesting that this component
is much weaker for these two high ionization lines.
Figure: Intensities of the central components (see Table
9 (click here)) versus the UV continuum at 1400 Å. The arrow in
the box for 
\
(upper left) indicates the direction of time along the points connected
by the dotted lines for 
and 
Figure: Correlation functions for the central component.
: Cross Correlation Function (CCF).
: Discrete Correlation Function (DCF).
: Autocorrelation Function (ACF)
for the UV continuum at 1400 Å
The red component with a FWHM of 
shows a velocity with
respect to the narrow line of 
(Table 9 (click here)). It
contributes to the total intensity with 26, 17, 24, 48 and 34% for 
,
, 
, 
\
and 
, respectively, somewhat less than the fractional contribution of the
central component. The
intensity of the red component is very tightly correlated with the UV continuum
(Fig. 13 (click here)) for most of the lines, except 
, which does not
appear to respond to the UV continuum. At the time resolution in our data
(96 days), no indication of delay is suggested for any of the high
excitation lines. The amplitude of the variations (Table 7 (click here)) are
the same within the errors for all high excitation lines.
The correlation functions between the several red components and the UV
continuum
are given in Table 14 (click here) and shown in Fig. 14 (click here). For all lines
(except 
) the CCF peak indicate a mean delay of 
indicating no delay at the resolution available in the sampling and
consistent with the absence of delay indicated by the DCF. Since for
CIV, SiIV and NV the red component disappears completely when the
continuum is weak, the center of CCF is the only meaningful parameter for
the delay determinations. At very low levels no line component flux can be
determined and the lightcurves of the continuum and the line become very
unequally sampled, completely distorting the extremes of the CCF.
Therefore the centroid of the CCF, at 
, becomes
solely a representation of the incompleteness of the light curve sampling
and has no physical meaning. On the other hand, this component is the only
one for which a statistically significant TF could be determined. The TF
shows an unresolved peak at zero days delay and no additional side lobes at
larger delays as seen in the full line intensities (see also Sect. 4.2).
The resulting Echo Map was fully consistent in the reproduction of the
continuum lightcurve and was statistically quite stable. Although the
errors are large, a weak correlation for the red component line ratios of
/
and 
/
is clearly present at lower levels of the UV
continuum brightness (Fig. 20 (click here)), again appearing to flatten at
levels above >15 
. The mean values are Ly
, 
, 
and 
.
The blue component has a mean FWHM of 
and a velocity
with
respect to the narrow line of 
. Its contribution to the total
intensity is 19, 23, 17 and 26% for 
, 
, 
and 
, respectively,
generally lower than the red component. The intensity correlation with the UV
continuum (Fig. 15 (click here)) is intermediate between the red and the central
component for all lines but 
which is constant. It does not correlate
as directly with the UV continuum as the red component but it does not show
the strong loops shown by the central component. The amplitude of its
variations (Table 7 (click here)) does not vary significantly between 
,
and 
.
The correlation functions for blue component with the UV continuum are shown in
Fig. 16 (click here) and given in Table 14 (click here). The correlation functions
for the blue component are significantly different from those for both the
red and central components and consistent between the CCF, for which the
center measure gives a mean delay of 
, and the DCF,
which results in a delay of 200 days (Fig. 16 (click here)).
For the blue component the line ratios (Fig. 19 (click here)) show a strong
correlation with the continuum for both Ly
/CIV and 
, while 
keeps constant within the errors. The mean values are 
for
, 
for 
and 
for
. The 
decomposition by Wamsteker et al.\
(1985) required a very broad component (of FWHM=9220 
and
velocity respect to the narrow line of -1710 
). Depending on the line
ratios of the red and blue component identified in the UV spectra with respect
to 
it is quite probable that this is essentially the optical
counterpart of these two components in the UV lines. A similar result was also
found for the line decomposition applied to the simultaneous UV and optical
spectra of NGC 5548 (Wamsteker et al. 1990), where a very broad
component clearly present in the UV was only marginally detectable in the 
\
profile. Lub & De Ruiter (1992) also obtain a very broad and
variable component in 
and 
, in order to explain the different
delays shown for these lines with respect to the optical variable continuum of 
for 
and 
for 
. For the blue component also no solution could be obtained for
the transfer function at a reasonable level of significance, most likely also
associated with the non-linearity in the line response as was the case for the
central component.
Figure: Intensities of the red components (see Table
9 (click here)) versus the UV continuum at 1400 Å
Figure: Correlation Functions for the red component (see also Table
14 (click here)). : Cross Correlation Function
(CCF). : Discrete Correlation
Function (DCF).
: Autocorrelation Function (ACF) for the UV
continuum at 1400 Å
Figure: Intensities of the blue components versus the UV
continuum at 1400 Å
Figure: Correlation functions for the blue components (see Table
14 (click here)). : Cross Correlation Function
(CCF). : Discrete Correlation
Function (DCF). : Autocorrelation Function (ACF)
for the UV continuum at 1400 Å