We used the spectral fitting program XSPEC ([Shafer, Haberl, & Arnaud 1989])
to model the extracted spectra. To permit the use of 
statistics
we grouped the data to have a minimum of 25 counts per spectral bin.
This affected only bins above 6.5 keV.
To avoid the worst uncertainties in the detector response, we restricted
our fits to bins with energies 
.
The SIS0 and SIS1 data were fit jointly.
Although spectral features are obvious in the raw spectrum, we first fit
a simple power law with cold absorption to the data to draw a comparison
with earlier X-ray observations. This model yields an energy index of 0.62,
a flux at 1 keV of 
,
an equivalent neutral hydrogen column of
and 
.
The spectral shape and intensity is significantly different from that found
in earlier Ginga observations ([Kolman et al. 1993]).
The intensity as observed with ASCA is about twice as high,
the spectral index is steeper (
compared to a mean of
in the Ginga data), and the absorbing column
lower than the value of
found by Kolman et al. (1993)Kolman93.
As Ginga is sensitive to column densities of
, there is little doubt that the
low energy absorption has changed significantly in character since 1989.
Such striking differences in column density were seen previously in the
Einstein observations of NGC 3516 ([Halpern 1982]), and
they are typical of variations seen in other sources with intrinsic
absorption such as NGC 4151 ([Yaqoob, Warwick, & Pounds 1989]; [Yaqoob et al. 1993])
and MR2251-178 ([Halpern 1984]; [Otani et al. (1996)]).
The spectral index, however, is sensitive to the modeling of
the absorption and the iron emission in the spectrum.
Using the same Ginga data, Nandra & Pounds (1994)NP94
find a spectral index of 0.65--0.74 and absorption columns of
.
The ratio of the SIS0 data to the simple power-law model shown
in Figure 2
immediately reveals absorption features below 1 keV due to
highly ionized oxygen and strong fluorescent emission from neutral iron
around 6.3 keV.
To characterize these spectral features empirically, we added a succession
of Gaussian-profile emission lines and absorption edges (with opacity
proportional to 
) until we obtained an acceptable fit.
An acceptable description of the spectrum requires two low
energy absorption edges that we attribute to O VII and O VIII,
a narrow unresolved Fe K
emission line,
and a broad base to the Fe K
line.
Table 1 gives the best fit values and 90% confidence
error bars for the parameters of this empirical model.
Figure 2:
To illustrate the features present in the X-ray spectrum of
NGC 3516, the data were divided by a simple model consisting of a power law with
low energy absorption by neutral gas. The ratio of the data
to the model shows a prominent absorption dip around the photoionization
edges of O VII and O VIII as well as complex structure
around the iron K
line.
The broad and narrow Fe K
features have a combined equivalent width (EW)
of 
eV, comparable to the 
eV Fe K
feature in the Ginga
spectrum of Kolman et al. (1993)Kolman93.
The energies of the oxygen edges and the narrow Fe K
line are redshifted
relative to the systemic redshift of NGC 3516 (
, [Vrtilek & Carleton 1985]).
The O VII, O VIII, and Fe K
features have redshifts of
, 
, and 
,
respectively.
Inclusion of the two oxygen edges in the fit broadens the distribution of
opacity at low energies.
This more accurate modeling of the shape of the low energy absorption
leads to a spectral index of 0.78 that is steeper than that in the simple
power-law fit and is more comparable to the mean power law index of 0.73
found for Ginga observations of Seyferts ([Nandra & Pounds 1994]).
The residual absorption by cold gas of
, however, is still higher than
the expected Galactic column of 
([Stark et al. 1992]).
At first glance it might seem natural to attribute this
excess above the Galactic column to cold gas intrinsic to NGC 3516.
However, the far-UV spectrum obtained with HUT
limits the intrinsic neutral hydrogen column in NGC 3516 to
([Kriss et al. 1996])
based on the strengths of the observed Lyman lines and Lyman limit.
The most likely explanation for the observed excess cold column is the
uncertainty in the ASCA calibration below 1 keV.
These uncertainties can lead to excess column
densities of up to 
.
Iron edges are also important diagnostics of the ionization state of the
absorbing medium. No iron edge feature is apparent in the residuals from
our fit, and this is not surprising given the order-of-magnitude
lower columns we are seeing relative to earlier X-ray observations of
NGC 3516. Adding an additional sharp edge with its energy constrained to be
greater than 7.1 keV gives no improvement in 
.
The optical depth at the edge must be less than 0.3 at the 90%
confidence level.
The moderately strong edges of highly ionized oxygen in our spectrum
naturally suggest an origin in photoionized gas.
The dominant strength of the O VII edge indicates gas of much lower
ionization and temperature than that modeled by
Krolik & Kriss (1995)KK95,
and the absence of an Fe K edge indicates a much smaller column.
Accordingly we have computed new models that span the potential range
of ionization parameters and column densities with some slight modifications
to the procedure described by Krolik & Kriss (1995)KK95.
First, we used an ionizing spectrum appropriate for NGC 3516 at this epoch.
The best-fit UV power law of 
([Kriss et al. 1996]) was
extrapolated to higher frequencies, and the best-fit X-ray power law found here
was extrapolated to lower frequencies;
the two meet at 51 eV.
Second, the lower temperatures and ionization state place the gas in a
regime where thermal equilibrium is more likely because the cooling time
is rather shorter.
We therefore compute our models in thermal equilibrium.
Finally, for ease of comparison to warm absorber models fit to the X-ray
spectra of other AGN, we assume constant density clouds and use the ionization
parameter 
, where 
is the number density of
ionizing photons between 13.6 eV and 13.6 keV illuminating the cloud and
is the density of hydrogen atoms.
The transmission of each model is computed exactly as described by
Krolik & Kriss (1995)KK95, taking into account resonant line
scattering and electron scattering as well as continuum opacity.
The resulting models are a two parameter family in total column density N and
ionization parameter. These are read into XSPEC as a FITS table for
fitting to the X-ray spectrum.
As in Krolik & Kriss (1995)KK95 we assume low density gas
(
), but there are no density-dependent effects in our
calculations or results, provided one considers densities lower than
.
Although the UV continuum in our photoionizing spectrum is steep, we note
that there is no lack of high energy photons. The spectrum flattens
just below the He II edge to 
,
and the overall 
,
a value typical of Seyfert 1 galaxies (Kriss & Canizares 1985)KC85.
In fact, tests show that our results are rather insensitive to the exact
shape of the ionizing spectrum apart from changes in the deduced ionization
parameter.
(This is a general property of photoionization models with broad distributions
of ionizing flux noticed in even the
earliest calculations [Tarter, Tucker, & Salpeter 1969]TTS69.)
For comparison we computed alternative models assuming either an extremely hard
spectrum with 
from 2500 Å through
the UV and X-ray up to 100 keV,
or the spectral shape of NGC 5548 as used by
Mathur et al. (1995)Mathur95, which contains a soft X-ray excess.
Neither of these match the observed broad-band spectral shape, yet they
both provide equally good descriptions of the opacity of the warm absorber.
The Fe K
lines in our empirical fit to the ASCA spectrum
are indicative of fluorescent emission.
As a better model for the X-ray continuum shape we therefore
use that predicted by Lightman & White (1988)LW88
for a disk illuminated by a power law.
Our data do not constrain the inclination or the solid angle
subtended by the disk, so we fix these parameters at 30 and 
,
respectively. The source is assumed to radiate isotropically, and we impose
a high energy cutoff of 300 keV on the intrinsic power law.
Using the optical depths given by our best empirical fit in
Table 1 and
the threshold photoionization cross sections of Verner & Yakovlev
(1995)VY95, we infer column densities for O VII and
O VIII of 
and
, respectively. Assuming these represent
all the oxygen atoms, the equivalent total hydrogen column for a solar
abundance of oxygen is 
.
Replacing the two oxygen edges in our empirical model with the grid of warm
absorber models, for the best fit we obtain a column density and ionization
parameter that qualitatively matches our expectations,
as shown in the center column of Table 2.
This fit is only slightly worse than our empirical model.
Warm absorber models computed with our alternative ionizing spectra give
identical best fit values for the total column density
(
). For the hard 
spectrum the best-fit ionization
parameter is 
with 
, and
for the spectrum like NGC 5548, 
with 
.
As discussed in our companion paper ([Kriss et al. 1996]), the strengths of the
UV lines observed in NGC 3516 require a zone of lower ionization and lower
column density than this warm-absorber model.
Other observations also indicate that the warm absorbing medium may be
complex.
Otani et al. (1996)Otani96 find that the O VIII opacity
in MCG--6-30-15 is variable, while the O VII opacity is not,
suggesting that the absorption arises in at least two different zones.
Prompted by these suggestions of spectral complexity, we experimented with
adding a second warm absorber to the fit.
This significantly improves 
.
Although the level of improvement is not sufficient to be apparent in any
particular feature in the spectrum,
an F test for 3 additional parameters producing 
shows that this second model component is significant at the 99%
confidence level.
The parameters of this best fit are summarized in the last column
of Table 2.
The SIS0 and SIS1 spectra and this best fit model are illustrated in
the top panel of Figure 3.
Figure 3:
Upper Panel: The solid lines are the best-fit empirical model folded
through the ASCA SIS0 and SIS1 detector responses.
The data points are crosses with 1
error bars.
The SIS1 data are offset down by 0.5 in the log for clarity.
The model includes a power law with photon index 1.78, absorption by
neutral gas with an equivalent neutral hydrogen column of
,
a photoionization edge due to O VII at 0.71 keV with an optical
depth at the edge of 0.65,
a photoionization edge due to O VIII at 0.86 keV with optical depth 0.31,
an unresolved iron K
line at 6.29 keV with an equivalent width of 73 eV,
and a broad (FWHM = 1.58 keV) iron K
line at 5.88 keV with an
equivalent width of 180 eV.
Lower Panel:
The contributions to 
of each spectral bin are shown.
The solid line is for SIS0 and the dotted line for SIS1.
As in Krolik & Kriss (1995)KK95
all the warm absorber models above assumed resonant line scattering profiles
with Doppler parameters given by the sound speed in the photoionized gas.
Since the line opacity is sensitive to the assumed profile width, it
influences the transmission computed for each model.
At low velocities the opacity is dominated by continuum absorption;
at high velocities resonant line scattering makes a significant contribution.
To illustrate, Figure 4 shows the computed
transmission for 
divided by that for
.
Physical parameters as determined by the best-fit single absorber model
in Table 2 were used for the ionization parameter and
column density.
The upper panel shows the ratio of the models themselves.
The lower panel shows the ratio after the models have been folded through
the ASCA response function. The error bars on each bin are taken from the
corresponding data points.
Figure 4:
Top Panel:
The ratio of the transmission for a warm absorber model with absorption
lines broadened by a Doppler parameter of 
to
the same model broadened with 
.
Physical parameters as determined by the best-fit single absorber model
in Table 2 were used for the ionization parameter and
column density.
Ion species contributing to the most significant lines are marked.
Lower Panel:
Here the models are folded through the ASCA response function before
computing the ratio.
The error bars on each bin are taken from the corresponding data at that point.
The stronger predicted line absorption at 
is clear. Below 1 keV O VIII and Fe-L transitions dominate the
increased opacity. The only significant features above 1 keV are the resonance
transitions of Mg XI and Si XIII.
At 
the absorption lines have an integrated
equivalent width of 5 eV and make a negligible contribution to the total
opacity, whereas at 
their equivalent width is 42 eV
and 6% of the opacity between 0.6 and 2.0 keV is due to line absorption.
Within the context of our photoionization models the dependence of line opacity
on line width permits us to constrain the Doppler parameter, even though
discrete absorption lines are not
unambiguously present in the observed spectrum.
In essence we find that the absence of significant resonant absorption
line features permits us to set upper limits on the Doppler parameter.
We computed grids of models with Doppler parameters varying from
to 
.
The best fits for both the single and the double warm absorber have
Doppler parameters of 
, approximately the sound
speed for these models.
(For the double warm absorber model we used the same Doppler parameter
for each component.)
At 90% confidence for a single interesting
parameter (
) we constrain the Doppler parameter
to values less than 160 
for the single warm absorber model,
and to less than 120 
for the model with two warm absorbers.
