Steps toward Determination of the Size and Structure of the Broad#Line Region in Active Galactic Nuclei. IX. Ultraviolet Observations of Fairall 9

An 8 month monitoring campaign on the Seyfert 1 galaxy Fairall 9 has been conducted with the International Ultraviolet Explorer in an attempt to obtain reliable estimates of continuum-continuum and continuum È emission-line delays for a high-luminosity active galactic nucleus (AGN). While the results of this campaign are more ambiguous than those of previous monitoring campaigns on lower luminosity sources, we Ðnd general agreement with the earlier results: (1) there is no measurable lag between ultraviolet continuum bands, and (2) the measured emission-line time lags are very short. It is especially notable that the Ly a ] N V emission-line lag is about 1 order of magnitude smaller than determined from a previous campaign by Clavel, Wamsteker, & Glass (1989) when Fairall 9 was in a more luminous state. In other well-monitored sources, speciÐcally NGC 5548 and NGC 3783, the highest ionization lines are found to respond to continuum variations more rapidly than the lower ionization lines, which suggests a radially ionization-stratiÐed broad-line region. In this case, the results are less certain, since none of the emission-line lags are very well determined. The best-determined emission line lag is Ly a ] N V , for which we Ðnd that the centroid of the continuum È emission-line cross-correlation function is days. We measure a lag days for He II j 1640; this result is q cent B 14 È 20 q cent [ 4 consistent with the ionization-stratiÐcation pattern seen in lower luminosity sources, but the relatively large uncertainties in the emission-line lags measured here cannot rule out similar lags for Ly a ] N V and He II j 1640 at a high level of signiÐcance. We are unable to determine a reliable lag for C IV j 1550, but we note that the proÐles of the variable parts of Ly a and C IV j 1550 are not the same, which does not support the hypothesis that the strongest variations in these two lines arise in the same region.


INTRODUCTION
With the discovery of coordinated continuum and emission-line variability in active galactic nuclei (AGNs), it was realized by many authors that variability might provide a very valuable tool for unveiling the structure of the innermost regions of these extremely luminous objects. The rapid response of the emission lines to continuum variations strongly argues that the line emission is driven by photoionization by the AGN central source, which is generally supposed to be an accretion disk surrounding a supermassive black hole. Thus, by measuring the time-delayed response of the emission lines to continuum variations, it is in principle possible to infer the distribution and kinematics of the line-emitting gas through a process known as "" reverberation mapping ÏÏ (e.g., & McKee Blandford 1982). Multiwavelength measurements of continuum and emission-line variability also provide important observational constraints on the central source itself by providing a probe (1) of the unobservable ionizing continuum, through the response of the emission lines to changes in the observable continuum, and (2) of the structure of the accretion disk itself, since in any model with a radial temperature gradient (such as a thin accretion disk), di †erent continuum wavelengths arise principally at di †erent disk radii.
On account of the rapid and irregular nature of the continuum variations, it was also realized early that enormous observational e †orts were needed to exploit the potential of variability studies. The techniques to obtain the necessary information (reviewed in Horne, & Peterson Gondhalekar, require long series of high-quality spectroscopic data, 1994) preferably taken at nearly regular intervals. The International Ultraviolet Explorer (IUE) is extremely well suited to this task, which motivated the establishment of international consortia in the late 1980s to conduct a series of major observing campaigns using both IUE and ground-based facilities to obtain the necessary spectroscopic observations. The largest of these groups, the International Edelson 1996). monitoring campaigns can be summarized as follows : 1. The UV and optical continua vary simultaneously within the accuracy of the data and the sampling frequency (typically^2 days or less).
2. The emission-line lags (or mean response times to continuum variations) are found to be quite small, typically ranging between a few days to several weeks.
The luminosity range for which these results have been obtained ergs s~1 is still [1039 [ L j (1450 A ) [ 1040 A ~1] too small to test critically the expected R P L 1@2 relationship between the broad-line region (BLR) size R and continuum luminosity L predicted by simple photoionization equilibrium arguments (e.g., and extension Peterson 1993), of variability studies to higher luminosity is therefore necessary. In this regard, the luminous Seyfert galaxy Fairall 9 ergs s~1 z \ 0.047] is of interest, [L j (1450 A ) [ 1041 A ~1; as it is one of the few high-luminosity sources that have been monitored with IUE. IUE observations with an average sampling interval of about 90 days suggested emission-line lags of about 115^70 days for Lya and about 200^80 days for C IV j1550 Wamsteker, & (Clavel, Glass see also & Gaskell & de 1989 ;Koratkar 1989 ;Lub Ruiter Ferland, & Peterson 1992 ;Shields, 1995 ;Recondoet al. relative to the UV continuum during a Gonza lez 1997) time when Fairall 9 underwent a dramatic change in luminosity. Between 1978 and1985, the UV Ñux of Fairall 9 decreased nearly monotonically by a factor of about 30.
In view of its well-documented UV variability history and its importance in the R-L relationship, we decided to undertake a more intensive monitoring program on Fairall 9. Our goal was to obtain observations with the same sampling frequency (about one observation every 4 days) and dura-   Sonneborn 1987).
The Ðne error sensor (FES) anomaly, which is attributed to scattered light entering the IUE telescope tube, has not been reported to contaminate any SWP image, since it was found for the Ðrst time in 1991 & Carini (Weinstein 1992). However, the solar-like spectrum of this scattered light a †ects the FES, a photomultiplier with a bandpass from 3900 to 9000 (S-20 cathode), with a sensitivity strongly A A peaked at 4600 & Rice The FES is used for A (Holm 1981). target acquisition and Ðne guiding during exposures ; hence, these operations are signiÐcantly inÑuenced by the presence of the scattered light. For most IUE observations of Fairall 9, there was no star within the FES Ðeld of view suitable for tracking. In these cases, the total integration time was achieved by adding up a number of shorter exposures, typically 15È30 minutes long. After each of these segments, the IUE telescope was moved to point to a nearby bright star to check the pointing stability.
As a consequence of the scattered light, the photometric capabilities of the FES could not be used to obtain a simultaneous optical light curve for Fairall 9. However, an optical light curve has been obtained from the nearly simultaneous ground-based optical campaign, and these data will be presented separately et al. (Santos-Lleo 1997). The log of the IUE observations includes, for (Table 1)  The major advantage of this software is the 1993). method of the raw science data registration, which signiÐcantly reduces the Ðxed pattern noise in the IUE images and improves the photometric correction. NEWSIPS also includes a weighted slit extraction method Bohlin, (Kinney, & Neill and rederived absolute Ñux calibrations 1991) Cassatella, & de la Fuente All of (Gonza lez-Riestra, 1992). these improvements result in higher photometric accuracy and higher signal-to-noise ratio spectra, when compared to the data processed with the old software (IUESIPS ; & Thompson Turnrose 1984). Because of the acquisition and tracking problems mentioned in we have checked every image to identify those°2, in which Fairall 9 may have gone out of the aperture for some time during the total exposure. A Ðrst test was made by checking the FES recentering errors after those exposures that were not guided ; a benchmark test made with an IUE standard star has shown that up to^25 units in the X-axis and^7 units in the Y -axis can be accepted as tolerance limits to ensure that the target did remain in the aperture Another check is based on (Rodr• guez-Pascual 1997). the position within the IUE SWP camera at which the spectrum was recorded. In the spatially resolved spectrum (SILO Ðle of NEWSIPS processing software), we have measured the shift in the spatial direction with respect to the center of the aperture. The wavelength of the peak of the strongest emission lines (Lya and C IV j1550) has been used as reference for the shift in the spectral direction. The 12 spectra recorded more than 4 pixels away from the expected position in either spatial or spectral directions have been rejected and are not considered in this paper.

Continuum and Emission-L ine Measurements
As shown in the continuum Ñux density has Figure 1, been measured in four bands that are free of obvious emission or absorption features in the mean spectrum : 1380È 1400 1500È1520 1780È1810 and 1860È1900 A , A , A , A (observed wavelengths). The corresponding Ñux densities with their associated uncertainties are listed in and Table 2 plotted in Note that the integration times have Figure 2. been adjusted to reach an optimal exposure in the peaks of the strongest emission lines (Lya, C IV j1550) at the cost of underexposing the continuum, where the Ñux density is 5 times weaker ; this results in relatively large uncertainties in the continuum Ñux-density measurements.
With regard to the emission lines, no attempt has been made to deconvolve the observed proÐles into components. Consequently, the Lya Ñux includes the Ñux from the N V j1240 line as well, and we will refer to the measured Ñux as The bands in which the continuum has been measured are marked above the spectrum, and the dotted lines show power-law Ðts to the Ñuxes in these bands in both the mean and rms spectra. The Lya ] N V, C IV j1550, and He II j1640 integration ranges are marked below the mean spectrum.
Lya ] N V. The measurements of C IV j1550 and He II j1640 are to some extent (\2%) a †ected by the overlap of the red wing of C IV j1550 with the blue wing of He II j1640. The observed wavelength ranges used for integration of the Lya ] N V, Si IV j1400, C IV j1550, and He II j1640 line Ñuxes are, respectively, 1229È1326 1400È1500 1540È A , A , 1680 and 1680È1800 as marked in A , A , Figure 1. The continuum underneath the emission lines has been estimated by linear interpolation between the four continuum bands described above.   been removed from the emission-line Ñuxes, since the strongest absorption features (C II, Si II, C IV) seen in the average spectrum are difficult to identify even in the bestexposed single spectrum. The Ñuxes and their associated uncertainties are listed in and the emission-line Table 3, light curves are shown in Figure 3.
Initially, the error estimates for the line and continuum measurements are based on the errors assigned by the NEWSIPS extraction software to the individual pixels in the relevant integration range for each feature. However, it became clear from the resulting light curves that the uncertainties estimated in this fashion are too large. We will make the assumption that the relative magnitudes of the errors obtained in the initial estimate are correct, but that all of the uncertainties should be scaled down by a constant multiplicative factor.
The uncertainties in the line and continuum Ñuxes can be estimated by measurements closely spaced in time, where di †erences should be attributable primarily to random errors rather than intrinsic variations. We consider the ratio of two measurements of the same Ñux and that are f i f j separated by some short time interval *t ij sufficiently small, and are essentially independent meaf i f j surements of the same quantity, and thus the uncertainties in their ratio ought to be where is the p ij 2 B p i 2 ] p j 2, p n uncertainty associated with measurement Averaged over f n . many pairs of observations, the variance of the distribution of will be where S f 2T is the mean square Ñux, that there is no intrinsic variability on timescales shorter than the largest value of Thus, for closely *t ij . spaced points (i.e., those with less than some small *t ij  value), we can compute the variance of the distribution in from the actual Ñux measurements and compare this f i /f j directly with the mean value of computed from the p ij 2/S f 2T error estimates. We Ðnd that the root mean square (rms) of the distribution is in all cases smaller than expected f i /f j from the quoted error estimates and thus reduce all of the error estimates by a constant multiplicative factor (which ranges from 1.03 in the case of Lya ] N V to 3.4 for the 1510 continuum) to bring the two error estimates into agree-A ment. This is a conservative approach : if there is intrinsic variability on short timescales, then the rms of the disf i /f j tribution ought to be larger than the mean value of contrary to what is found. The true uncertainties p ij 2/S f 2T, may be even smaller than the rescaled error estimates, since any low-level variability that occurs on short timescales is attributed to random error in this estimate. Operationally, the factors we use for rescaling the NEWSIPS uncertainties are determined by considering pairs of observations separated by less than 1 day. We also Ðnd that if we increase the maximum time interval to 4 days, which greatly increases the number of pairs of observations in the calculation, then the rescaling factors increase (as expected), but only slightly, indicating that there is little intrinsic variability on such short timescales. The uncertainties quoted in Tables and and illus-2 3 trated in the Ðgures are the rescaled values computed as described.

V ariability
The variability parameters of the continuum and emission lines are given in The entries in are the Table 4. Table 4 unweighted mean Ñux S f T (col. [2]) and rms Ñux p (col. [3]) for the entire campaign. These are deÐned in the usual way, i.e., for N individual observations the mean is f i and p is the square root of the variance Also in are two common measures of the variability  random errors, i.e., where the quantity *2 is the mean square value of the uncertainties associated with the Ñuxes i.e., * i f i , As found in previous studies of other AGNs et al. (Clavel et al. et al. et 1991 ;Reichert 1994 ;Korista 1995 ;Crenshaw al. the amplitude of the continuum variations in 1996), Fairall 9 decreases at longer UV wavelengths. Whether this is due to an intrinsic hardening of the featureless continuum as the source becomes more luminous or to the presence of an intrinsically less variable feature, whose contribution increases with wavelength, cannot be addressed with the limited wavelength range covered by the SWP camera. respectively. Subtraction of these values from the A ~1, mean Ñux densities in results in all the continuum Table 4 bands varying with the same amplitude (F var \ 0.34^0.03). Another test that can be done to determine whether or not the AGN continuum hardens as it becomes brighter is to compare on a linear plot simultaneously measured Ñuxes in two well-separated wave bands. In the case of NGC 5548, for example, a plot of the optical Ñux (at 5100 versus the A ) short-wavelength UV Ñux (at 1350 shows (1) that the best A ) Ðt to these requires some curvature, and (2) that the optical axis intercept is nonzero (Fig. 3 of The sense Peterson 1991). of the curvature in this relationship means that the amplitude of the UV variations is greater than the amplitude of the optical variations, i.e., the variable part of the spectrum becomes harder as the continuum gets brighter. The nonzero intercept of the optical axis gives a measure of the constant starlight contribution to the optical Ñuxes, which in this case is in good agreement with the estimate obtained by image decomposition et al. (Romanishin 1995). In we plot versus to test Figure 4, F j (1880 A ) F j (1390 A ) for curvature that would indicate a nonlinear relationship between these two continuum bands. A linear Ðt to these data (as shown in provides a good Ðt Fig. 4) (s l 2 \ 3.08) ; indeed, a quadratic gives a slightly worse Ðt and (s l 2 \ 3.11) the curvature coefficient is statistically insigniÐcant. We conclude that, unlike the case of NGC 5548, and as previously concluded on the basis of existing data on this object, in the case of Fairall 9 there is no evidence that the AGN continuum shape changes as the luminosity changes. To test more directly the hypothesis that the relationship between the two continuum bands is linear, we Ðt the data to the equation The dotted line shows the best linear A , Table 2. Ðt to these data. There is no evidence for curvature in this relationship, which would indicate that the spectral index of the continuum depends on luminosity.
where C is a constant, which we assume to be the yintercept of (i.e., C \ 6.6 ] 10~15 ergs s~1 cm~2 Figure 4 This yields c \ 0.992^0.031, consistent with a linear A ~1).
relationship. This will be tested more critically over a wider wavelength range using the combined UV and optical data on this object et al. (Santos-Lleo 1997). Among the emission lines, the largest variations occur in Lya ] N V and He II j1640, and C IV j1550 varies only weakly. The light curves we measure for Si IV j1400 show no statistically signiÐcant variability, and therefore we do not include these measurements in the tables or Ðgures. We must point out, however, that the Si IV j1400 feature appears, albeit weakly, in the rms spectrum shown in Figure  Weak variations did in fact occur, but the measured light 1. curve is dominated by random noise.
The continuum light curves show three "" events ÏÏ of similar duration (D70 days), but with noticeably di †erent amplitudes The Ðrst two events show variations of (Fig. 2). up to about 50%, while during the last event the continuum varied by more than a factor of 2. The Lya ] N V light curve is qualitatively similar to that of the continuum, with two low-amplitude events (D20%) followed by a stronger (D50%) third one. The most remarkable di †erence is that the fading of the continuum by the end of the campaign is not clearly seen in the Lya ] N V light curve. Although still smaller than that of the continuum, the amplitude of the He II j1640 variations is the largest among the emission lines. The detection of the Ðrst two events in the He II j1640 and C IV j1550 light curves is quite marginal, but the onset of the third event is clear even in these noisy light curves.

Cross-Correlation Analysis
The emission-line time delays, or lags, relative to the continuum variations can be quantiÐed by cross-correlation of the continuum and emission-line light curves. Before computing the cross-correlation functions (CCFs), multiple measurements obtained during single observing shifts were averaged, weighted in each case by the inverse square of their assigned uncertainties. The CCFs have been computed by using three di †erent algorithms : (1) the interpolation cross-correlation function (ICCF) of & Sparke Gaskell (2) the discrete correlation function (DCF) method of (1986), & Krolik and (3) the z-transformed discrete Edelson (1988), correlation function (ZDCF) method of Alexander (1996). The ICCF and DCF algorithms are as described by White & Peterson All three methods give similar, although (1994). not identical, results. The CCFs between the continuum at 1390 and the other continuum bands show that A (Fig. 5) they are all highly correlated, with no measurable delay among the di †erent bands. In we summarize the Table 5 characteristics of the CCFs, where is the location of the q peak ICCF peak, which has value is the centroid for r max ; q cent ICCF values greater than and FWHM is the full 0.8r max ; width of the ICCF at Also in are the ZDCF 0.5r max . Table 5  results ; is the maximum likelihood estimate of the loca-q ML tion of the true peak in the CCF, and the errors associated with it represent a 68% conÐdence interval (see Alexander and et al.
The maximum likelihood 1996 Edelson 1996). uncertainties quoted in are more conservative error Table 5 estimates than we obtain using either the analytic formula of & Peterson which yields estimated errors Gaskell (1987), *q B 4È7 days for all ICCFs computed in this paper, or error estimates based on Monte Carlo simulations (e.g., & Peterson which yield days for all White 1994), *q [ 2 ICCFs presented here.
The results of cross-correlating the continuum at 1390 A and the three emission-line light curves of and Table 3 are shown in and summarized in Figure 3  99.9% conÐdence. However, the CCF is very Ñat-topped and the width of the CCF is somewhat smaller than that of the continuum autocorrelation function (ACF). We therefore attach no signiÐcance to the measured C IV j1550 lag.

L ine ProÐle V ariations
The low amplitude of the line variations limits the depth of the analysis of the proÐle variability. However, inspection of Figures and shows that, during this observing cam-2 3 paign, Fairall 9 went through two di †erent levels of brightness, and a simple comparison can be made of the proÐles in the high and low states. For the high state, we have averaged all the spectra from JD 2,449,660 to the end of the campaign, and for the low state, the spectra taken between JD 2,449,530 and JD 2,449,580 have been averaged. The Lya, C IV j1550, and He II j1640 line proÐles for the two states are shown in after continuum subtraction. Figure 7, Although the error bars are rather large, the di †erence proÐle of C IV j1550 appears to be narrower than the Lya and He II j1640 di †erence proÐles. In an attempt to quantify roughly the changes in the line proÐles, each di †erence proÐle has been Ðtted with a Gaussian (shown as a thick line in the bottom panels of We Ðnd that the width Fig. 7). of the C IV j1550 di †erence proÐle (4180^340 km s~1) is less than half the width of the Lya (10600^460 km s~1) and He II j1640 (13890^900 km s~1) di †erence proÐles. Also, the center of the Gaussian Ðtted to C IV j1550 is more redshifted relative to the galaxy redshift (c *z \ 1570^170 km s~1) than are either Lya (880^130 km s~1) or He II j1640 (270^400 km s~1). However, given the low level of emission-line variability detected in this experiment, we are reluctant to draw strong conclusions based on these limited data. We draw attention to these di †erence because they are of potential importance should they be substantiated by better data.

DISCUSSION
The selection of Fairall 9 for an intensive monitoring e †ort with IUE was intended to extend the very limited sample of intensively monitored AGNs to higher luminosity and thus test critically the predicted BLR radius-luminosity relationship R P L 1@2. A previous e †ort by et al. Clavel (see also & Gaskell & de Ruiter (1989) Koratkar 1989Lub Ferland, & Peterson 1992 ;Shields, 1995 ;Recondoet al. yielded lags of about 115^70 days Gonza lez 1997) for Lya and about 200^80 days for C IV j1550. These lags were measured during a factor of 30 decrease in luminosity and are based on data that are not well sampled on short timescales. Indeed, cross-correlation lags are strongly inÑuenced by inÑection points in light curves, and the previously published Fairall 9 light curves are not well sampled at the time the continuum and emission lines passed through their respective minima. Moreover, a plot of the observed C IV j1550 time lag versus UV continuum for reasonably well observed AGNs (e.g., Fig. 3  assuming the L 1@2 luminosity dependence, then we 1991) would expect an Lya C IV j1550 lag of between 13 and 73 days for Fairall 9 for its low and high historical states, respectively. These considerations led us speciÐcally to look for more rapid BLR response than could have been detected in the original et al. experiment. Clavel (1989) For Lya ] N V we Ðnd days for this experi-q cent B 14È20 ment, which is not much smaller than the value of 28 days we would expect by scaling the NGC 5548 results to the mean luminosity of Fairall 9 during this campaign. Unfortunately, relatively weak variability during the Ðrst several months of the campaign precluded an accurate measurement of the C IV j1550 response time. Formally, we obtain days for C IV j1550, although we place no con-q cent B 30 Ðdence in this measurement on account of the low signal-tonoise ratio of the C IV j1550 light curve.
The reason for the dramatic di †erence between the Clavel et al.
Lya ] N V lag and that reported here is not (1989) obvious. Just as the previous data were not sensitive to short-timescale emission-line response, the present campaign was not sensitive to long-timescale days) (R/c Z 100 emission-line lags. Among likely possibilities are the following : 1. The BLR did not change between the two experiments. On account of their very di †erent sampling characteristics, the two monitoring programs sampled completely di †erent scales in a very extended BLR. The lack of an obvious turnover in the Lya ] N V light curve at the end of the campaign may be consistent with this interpretation. 2. As a consequence of the signiÐcant change in mean luminosity between the two campaigns (a factor of D3È6, depending on how it is measured), there has been a change in the optical depth in BLR clouds in the sense that the more recent experiment samples clouds that are currently optically thick (and thus produced highly variable emission lines), but which were optically thin (and producing less variable emission lines) in the previous campaign. Such dramatic changes are possible on account of the short recombination time for high-density BLR clouds, q rec B hr. This could occur if much of the line 109/n e (cm~3) response we see in this campaign originates in clouds that have column densities only somewhat deeper than the ionized column ; if the continuum Ñux increased by a signiÐcant factor, then these clouds would be completely ionized and thus show no recombination-line variability with further increases in continuum Ñux. Similar clouds at larger distances would remain partially neutral and responsive to continuum changes.
3. There has been a signiÐcant rearrangement of the BLR gas between the two experiments. The crossing time for an emission-line cloud in the BLR is where q dyn B cq cent /*V , *V is the emission-line width. Taking *V \ 104 km s~1 and days yields yr, which is approx-q cent B 150 q dyn B 10 imately the interval of time between the observations reported here and those made as Fairall 9 passed through the minimum state recorded by et al. it is Clavel (1989) ; certainly plausible that a signiÐcant rearrangement of the BLR gas has occurred between the two experiments.
Detailed modeling beyond the scope of this paper will be required to support further or to eliminate these and other possibilities. Unfortunately, a longer campaign that might have discriminated among the various possibilities is simply not possible on account of the termination of IUE operations.
The ICCF and DCF indicate that He II j1640, with days, responds more rapidly than does Lya ] N V, q cent [ 4 which is consistent with what has been found in other sources and is consistent with a radially ionization-stratiÐed BLR. However, the uncertainties in all of these measurements are sufficiently large that we cannot rule out the possibility that Lya ] N V and He II j1640 actually respond with the same lag.
The C IV j1550 results, however, remain puzzling. In other well-studied AGNs, the response times and line pro-Ðles for Lya and C IV j1550 are similar. While these data do not provide a strong constraint on the C IV j1550 response time, the variable part of the C IV j1550 line proÐle appears to be signiÐcantly narrower and redshifted relative to the Lya and He II j1640 lines. It appears to be rather doubtful that the peak response of C IV j1550 and Lya occurs in the same gas, as might be more easily inferred for NGC 5548, for example. Given (1) the possible similarity of the Lya ] N V and He II j1640 responses, (2) our expectation from other experiments that Lya and C IV j1550 ought to respond on the same timescale, and (3) some, albeit unreliable, indications that the C IV j1550 response is longer than Lya ] N V in Fairall 9, we might well ask if the response we are seeing in Lya is in fact not Lya at all, but N V j1240, which, again based on other sources, we expect to respond like He II j1640. The contribution of the N V j1240 emission line to the Lya ] N V light curve may cause the relatively short lag found if the variations in this high-ionization line are sufficiently large. We have considered this explanation and reject it for the following reasons. First, there is no structure in the di †erence proÐle shown in that may suggest Figure 7 that N V j1240 signiÐcantly contributes to the variations. Second, N V j1240 appears only weakly in the rms spectrum again arguing that the amplitude of N V j1240 (Fig. 1), is insufficient to contribute in a signiÐcant way to the Lya ] N V variability. Third, as a Ðnal check, we remeasured the Lya Ñux as before, but we extended the measurement only to 1228 (in the rest frame) to minimize A contamination by N V j1240 ; this yielded Lya light curves and CCFs that did not di †er in any substantive way from the Lya ] N V results reported here.

SUMMARY
The results of the IUE monitoring campaign on a bright Seyfert 1 nucleus (Fairall 9) qualitatively support the general conclusions drawn from previous campaigns on the less luminous AGNs NGC 3783, NGC 4151, and NGC 5548. In particular, the various continuum bands respond simultaneously, within our ability to measure them, and the emission lines respond to continuum variations on timescales that are very short (i.e., days), although the relatively noisy emission-line light curves measured in this experiment result in large uncertainties in the emission-line time lags. The principal conclusions of this study are as follows : 1. Large-amplitude (factor of 2) UV continuum variations occur in Fairall 9 on a timescale (D70 days) similar to the fastest variations detected (at a similar amplitude) in NGC 5548, which is about 6 times less luminous.
2. No signiÐcant lags have been found among the four UV continuum bands measured, although the amplitude of the variations decreases with increasing wavelength. Unlike NGC 5548, we do not Ðnd evidence that the AGN continuum changes shape (i.e., spectral index) as the luminosity changes. The wavelength baseline over which this has been tested will be extended to the optical region in a forthcoming paper et al. (Santos-Lleo 1997). 3. Among the emission lines, He II j1640 shows the largest fractional variations. The variations in Lya ] N V and C IV j1550 are much smaller.
4. The lags between the emission lines and the UV (q cent ) continuum are days for He II j1640 and about 14È20 [4 days for Lya ] N V. The uncertainties of these determinations are large (D5È10 days), so it cannot be concluded with any conÐdence that the lags for these two lines are not the same, although this is much clearer in other sources. The small variations in C IV j1550 make it difficult to estimate a lag, but it seems to be greater than for Lya ] N V and He II j1640.
5. The line proÐle variations in Lya and He II j1640 di †er from those in C IV j1550. The variable part of the C IV j1550 line appears to be narrower and more redshifted than the variable parts of Lya and He II j1640.
This work has been supported in part by NASA through grant NAG 5-3233 to Ohio State University.