A UBV / variability / proper motion QSO survey from ...aa.springer.de/papers/7325002/2300457.pdf · A UBV/ variability / proper motion QSO survey from Schmidt plates I. Method and

Astron. Astrophys. 325, 457–472 (1997) ASTRONOMYAND

ASTROPHYSICS

A UBV / variability / proper motion QSO surveyfrom Schmidt plates

I. Method and success rate

R.-D. Scholz1,2, H. Meusinger3, and M. Irwin4

1 WIP Astronomie/Universitat Potsdam, Sternwarte Babelsberg, An der Sternwarte 16, D-14482 Potsdam, Germany2 Astrophysikalisches Institut Potsdam, An der Sternwarte 16, D-14482 Potsdam, Germany,3 Thuringer Landessternwarte Tautenburg, D-07778 Tautenburg, Germany4 Royal Greenwich Observatory, Madingley Road, Cambridge CB3 OEZ, UK

Received 5 February 1997 / Accepted 26 March 1997

Abstract. We describe the first combined variability-propermotion survey for QSOs on Schmidt plates. 85 TautenburgSchmidt plates covering about 9 square degrees near the northGalactic pole were digitised by means of the APM facility inCambridge. Variability indices have been derived from U andB data, both with a time-baseline of three decades. Structurefunction analysis has been applied to search for long-term vari-ability. An almost complete set of proper motions were obtainedto a mean B ≈ 20 with a mean error of about 2.0 mas/yr forB < 19.5. Adopting a limiting (completeness) magnitude ofBlim = 19.7 the success rate of the survey is estimated to > 40per cent. The completeness of the QSO sample detected by thissurvey is expected to about 90 per cent. The final sample con-tains 168 QSO candidates; only one has a radio counterpart. Thevariability-proper motion survey is compared with the selectionbased on the two-colour index diagram.

Key words: quasars: general – galaxies: photometry – stars:variables: other – astrometry – methods: statistical – surveys

1. Introduction

Since the early detections of QSOs it has been realised thatvariability of flux densities and stationarity of their positions arebasic features of this class of objects. In several studies either thevariability or the zero proper motion constraint has been usedto select QSOs (e.g. Sandage & Luyten 1967; van den Berghet al. 1973; Kron & Chiu 1981; Koo, Kron & Cudworth 1986;Veron & Hawkins 1993, 1995). Majewski et al. (1991) obtainedaccurate proper motions of 1185 objects in a 0.1 square degreesfield at the NGP (SA57) from 31 KPNO Mayall 4 m plates

Send offprint requests to: R.-D. Scholz

spanning a 16 years baseline and combined the proper motioncriteria with a variability index. As pointed out by Veron (1993),“the search for objects which are both variable and stationaryis a powerful technique for efficiently finding QSOs with noselection bias with regard to colour, redshift, spectral index, oremission line equivalent width.”

QSOs are known to vary on long timescales. According toSmith & Nair (1995) the average observed base-level time scalefor radio quiet QSOs is about 10 yr. As suggested by Hawkins(e.g. 1983), the contamination of a QSO variability search dueto variable stars (Sanitt 1975) can be suppressed by search-ing for objects which are variable on a timescale of about 1yr or longer. As the time-lag between two observations in theQSOs frame scales with (1+z)−1 a time-baseline of at least twodecades is desirable for QSOs with z = 0...4 assuming a typicaltimescale of a few years. Nevertheless, a remaining contamina-tion is expected due to long-term variability in late-type stars.Such long-term variability has been detected by Weis (1994)in the broadband photometric data of nearly 50 per cent of 43dwarf M stars over the period of one decade. Assuming a scaleheight of 300 pc, about half the late M dwarfs with B < 19.7in a field near the Galactic pole have distances less than 100 pc.Accurate proper motions are expected, therefore, to eliminate asignificant fraction of such contaminating M dwarfs.

Following the work of Majewski et al. and Veron & Hawkinswe started a wide field QSO survey based on zero proper mo-tion and photometric variability measured on Schmidt plateswith a limiting magnitude of about B = 21. In a pilot project(Meusinger et al. 1994a,b,c), the long-term variability of a smallnumber of known QSOs had been investigated on a large num-ber of Tautenburg Schmidt plates. As a main result a detectablevariability fraction of about 80 per cent has been estimated.Now we have undertaken a combined variability-proper motionsurvey (in the following the abbreviation VPM survey is used)in a field of about 9 square degrees. The observational database

458 R.-D. Scholz et al.: A UBV / variability / proper motion QSO survey from Schmidt plates. I

is provided by 85 Tautenburg Schmidt plates in the U , B andV passbands digitised by means of the APM facility in Cam-bridge. The first plate was taken in 1964, the last one in 1994.The time-baseline of three decades for these observations is thelongest for a QSO variability survey so far. This survey is ex-pected to be very efficient due to the combination of variabilityand proper motion data from a large number of plates with sucha long time-baseline. A particular aim is, thanks to the avail-ability of unconstrained colours from time-averaged U,B andV magnitudes, to look for QSOs with red colours, perhaps dueto strong dust-reddening (Webster et al. 1995).

Besides the large number of available plates the survey fieldat α(2000) = 13h41m58s, δ(2000) = 28◦24′ is interesting forseveral other reasons: (1.) At high Galactic latitude, interstel-lar extinction and stellar contamination are minimised. (2.) Thesouthern half of our survey field was part of another optical sur-vey using a blue grism at the prime focus of the CFHT (Cramp-ton et al. 1988, 1990, 1992). The grism survey is sensitive toQSOs with 0.2 < z < 3.4 up to a limiting magnitude com-parable to our plate limit. The sample of grism-detected QSOsprovides a useful database to estimate a priori the complete-ness and the success-rate of our VPM survey. The comparisonof the results of these two surveys are expected to provide in-sights into their selection functions. (3.) For the northern halfof the field positions of many radio sources from the FIRSTVLA survey (Becker et al. 1995) are available. (4.) The field iscentered on the globular cluster M 3, which is one of the best-studied globular clusters with respect to the content of variablestars.

The present paper is concerned with the method and es-timated success rate of this VPM survey. Preliminary re-sults already published (Meusinger et al. 1995, 1996; Scholz,Meusinger & Irwin 1995) were based on the proper motion de-termination by Scholz, Odenkirchen & Irwin (1993). Now, wehave significantly improved the proper motion data due to theuse of accurate position measurements on 81 plates. Also the in-ternal photometric calibration was improved by including a two-dimensional systematic error treatment of the magnitudes withrespect to the position on the plates. In addition to variabilityindices based on either 57 B plates or 22 U plates, respectively,we have computed a long-term variability index from structurefunction analysis. The combination of these variability indiceswith the proper motion index provides a powerful tool in thesearch for QSO candidates.

2. The observational basis

The basis for the observational material of the present study isprovided by Schmidt plates taken with the Tautenburg 2 m tele-scope which has a free aperture of 1.34 m in its Schmidt version.With a plate size of 24 cm and a scale of 51.4 arcsec/mm eachplate covers an unvignetted field of about 3.2◦ × 3.2◦. Thanksto its large focal length, compared to other large Schmidt tele-scopes, the Tautenburg Schmidt has less problems with the dis-tortions caused by plate bending and a better scale for astromet-ric work.

Fig. 1. Distribution of the plate limits of 57 B plates (marked by ×),22 U plates (+) and 6 V plates (◦) as a function of the plate epoch.

The distribution of plates taken with the Tautenburg Schmidttelescope over the last three and a half decades shows that thereare some fields with more than 50 plates taken at differentepochs. This material is very useful for long-term variabilitystudies of faint objects. The region around the globular clusterM 3 is one of the most frequently observed fields. More than 200plates were taken in different passbands mainly between 1964and 1976 and after 1985. For the present study we selected 57B plates (Table 1), 22U plates and 6 V plates (Table 2). Exceptfor 6 of the B plates all plates are centered on M 3. The earliestplate taken originates from 1964, the last one used from 1994.With nearly 3 decades, the time-baseline is the longest for a QSOvariability survey so far. The limiting magnitude of the platesare B = 19.5...21.3, U = 18.6...20.2 and V = 17.9...20.4. Thedistribution of the plate limits of theB plates compared to theUand V plates as a function of the plate epoch is shown in Fig. 1.

3. Measurements, classification and selection of objects

All plates were measured with the automatic plate measuringfacility (APM) in Cambridge (Kibblewhite et al. 1984). Usingthe standard APM software the measured objects on each platewere classified into stars, nonstellar objects, noise images andmerged objects. In order to avoid variability caused by dirt oremulsion defects on the plates only the measurements with stel-lar classification were used in the photometric study, i.e. for theinternal and external magnitude calibration and for the determi-nation of variability indices. The plate edges and areas occupiedby density wedges (on second epoch plates) were excluded inboth the photometric and astrometric reduction. This reducedthe investigated field from 10 to about 9 square degrees. Only0.7 per cent of the overall area were occupied by image pix-els, i.e. the sky fraction occupied by foreground objects (brightstars with large image diameters) can be neglected in our QSOsearch.

For the combination of the magnitudes and coordinates fromall plates we selected one of the deepest B plates with a min-imum of noise images and/or dirt on it as the reference plate.

R.-D. Scholz et al.: A UBV / variability / proper motion QSO survey from Schmidt plates. I 459

Table 1. Plate material - B plates

Plate No. epoch ∆texp emulsion+filter[d/m/y] [min]

B 1478 21 1 1964 30 103 a-O + GG 13B 2001 22 5 1965 20 103 a-O + GG 13B 2167 20 3 1966 30 103 a-O + GG 13B 2174 20 3 1966 30 103 a-O + GG 13B 2175 20 3 1966 30 103 a-O + GG 13B 2176 20 3 1966 30 103 a-O + GG 13B 2222 10 6 1966 20 AS + GG 13B 2382 15 2 1967 45 AS + GG 13B 2384 15 2 1967 45 AS + GG 13B 2395 16 2 1967 45 AS + GG 13B 2396 16 2 1967 45 AS + GG 13B 2397 16 2 1967 45 ZU 2 + GG 13B 2403 17 2 1967 45 ZU 2 + GG 13B 2732 19 4 1968 30 ZU 2 + GG 13B 2734 19 4 1968 30 ZU 2 + GG 13B 2735 19 4 1968 30 ZU 2 + GG 13B 2873 6 4 1969 40 ZU 2 + GG 13B 2879 7 4 1969 40 ZU 2 + GG 13B 3011 5 2 1970 30 ZU 2 + GG 13B 3022 7 2 1970 30 ZU 2 + GG 13B 3076 31 3 1970 17 ZU 2 + GG 13B 3309 16 1 1972 30 ZU 2 + GG 13B 3809 29 3 1973 40 ZU 2 + GG 13B 3817 30 3 1973 25 ZU 2 + GG 13B 4136 19 5 1974 25 ZU 2 + GG 13B 4685 26 2 1976 20 ZU 2 + GG 13B 4776 26 4 1976 20 ZU 2 + GG 13B 6223 6 5 1986 24 ZU 21 + GG 13B 6224 6 5 1986 22 ZU 21 + GG 13B 6226 11 5 1986 27 ZU 21 + GG 13B 6232 9 6 1986 16 ZU 21 + GG 13B 6709 13 5 1988 25 ZU 21 + GG 13B 6999 6 3 1989 28 ZU 21 + GG 13B 7000 6 3 1989 28 ZU 21 + GG 13B 7002 6 3 1989 28 ZU 21 + GG 13B 7013 6 4 1989 25 ZU 21 + GG 13B 7014 6 4 1989 26 ZU 21 + GG 13B 7036 4 6 1989 21 ZU 21 + GG 13B 7872 9 2 1992 20 ZU 21 + GG 13B 7936 9 3 1992 34 ZU 21 + GG 13B 7945 31 3 1992 34 ZU 21 + GG 13B 7977 1 5 1992 43 ZU 21 + GG 13B 8003 22 5 1992 35 ZU 21 + GG 13B 8017 26 5 1992 34 ZU 21 + GG 13B 8284 29 3 1993 36 ZU 21 + GG 13B 8288 30 3 1993 33 ZU 21 + GG 13B 8304 21 4 1993 44 ZU 21 + GG 13B 8310 21 4 1993 41 ZU 21 + GG 13B 8319 24 4 1993 40 ZU 21 + GG 13B 8329 27 4 1993 45 ZU 21 + GG 13B 8336 28 4 1993 42 ZU 21 + GG 13B 8344 16 5 1993 30 ZU 21 + GG 13B 8347 17 5 1993 38 ZU 21 + GG 13B 8366 15 6 1993 25 ZU 21 + GG 13B 8534 16 2 1994 23 ZU 21 + GG 13B 8587 7 4 1994 38 ZU 21 + GG 13B 8603 14 4 1994 36 ZU + GG 13

Table 2. Plate material - U and V plates

Plate No. epoch ∆texp emulsion+filter[d/m/y] [min]

U 2160 24 2 1966 90 103 a-O + UG 2U 2161 24 2 1966 56 103 a-O + UG 2U 2166 19 3 1966 90 103 a-O + UG 2V 2181 21 3 1966 30 103 a-G + GG 11V 2481 29 4 1967 35 103 a-D + GG 11V 2504 8 5 1967 45 103 a-D + GG 11U 2740 20 5 1968 90 103 a-O + UG 2U 2741 21 5 1968 90 103 a-O + UG 2V 2883 5 5 1969 30 103 a-D + GG 11U 3023 7 2 1970 60 ZU 2 + UG 2U 3070 10 3 1970 40 ZU 2 + UG 2U 3071 11 3 1970 40 ZU 2 + UG 2U 3078 31 3 1970 33 103 a-O + UG 2U 3095 6 4 1970 40 103 a-O + UG 2U 3426 13 4 1972 90 103 a-O + UG 2U 3427 13 4 1972 90 103 a-O + UG 2V 7016 6 4 1989 23 103 a-D + GG 11V 7019 7 4 1989 32 103 a-D + GG 11U 8360 25 5 1993 50 ZU 21 + UG 2U 8489 15 1 1994 45 ZU 21 + UG 2U 8499 18 1 1994 44 ZU 21 + UG 2U 8521 15 2 1994 50 ZU 21 + UG 2U 8533 16 2 1994 52 ZU 21 + UG 2U 8563 3 4 1994 68 ZU 21 + UG 2U 8594 8 4 1994 51 ZU 21 + UG 2U 8604 15 4 1994 75 ZU + UG 2U 8643 12 5 1994 48 ZU + UG 2U 8645 13 5 1994 92 ZU + UG 2

This plate (B 2176) had already been used as reference plate inthe determination of the absolute proper motion of the globu-lar cluster M 3 (Scholz, Odenkirchen & Irwin 1993) and in thecomparison of the colour-magnitude-diagram (using the platesB 2176 and V 2504) with astrometric cluster membership crite-ria of faint stars (see Scholz & Kharchenko 1994). In the platematching of all comparison plates with the reference plate weused a search radius of four pixels, corresponding to about 1.5arcsec. From about 32700 images measured on the referenceplate only about 24600 objects identified on at least two otherplates were included in further reduction. In the photometriccalibration using only the stellar objects on each plate we hadabout 21500 stellar images on the reference plate from whichwe selected nearly 12800 objects matched with stellar objectson at least 6 other B plates for the determination of variabilityindices inB. For the determination of variability indices inU weselected all stellar objects on the blue reference plate matchedwith at least 5 stellar measurements on the U plates. The imageclassification near the plate limits is problematic. The minimumnumber of 7 B plates and 5 U plates for the variability studyof images with stellar appearance was selected considering theoverall number of plates per colour and their different limitingmagnitudes.


Fig. 2. All objects with stellar appearance on at least seven B platesare shown. The globular cluster M 3 is seen in the field centre. Theinnermost cluster region is not well represented due to strong imagecrowding on the Schmidt plates. 90 known QSOs with measured zvalues included in our sample (i.e., with stellar classification on atleast seven B plates) are drawn as crosses. The QSOs were identifiedin the catalogues of Hewitt & Burbidge (1993), Veron-Cetty & Veron(1996b) and the NASA/IPAC extragalactic database (NED) allowingfor coordinate differences of up to 10 arcsec.

For the selection of the galaxies defining the absolute ref-erence frame in the proper motion determination we looked atthe objects classified as galaxies on the reference plate and on 5other deepest plates of our sample. In general, there is a largeruncertainty in the automated image classification near the platelimits. Therefore, we also compared all objects measured onthe reference plate with the APM measurements (and objectclassification) on the POSS 1 plates covering our field. Mostof the field is overlapped by the POSS 1 plate pair eo131. Forthe southern part of our field we used another POSS 1 plate pair(eo86), too. Both Palomar plate pairs go deeper than the Tauten-burg plates. Especially the red Palomar plates were very usefulfor the classification of galaxies near the Tautenburg plate lim-its. The region of the globular cluster M 3 was excluded in thesearch for reference galaxies because of possible overlappingimages of cluster stars masquerading as galaxies. By this waywe selected more than 1600 reference galaxies on the referenceplate B 2176 outside a cluster radius of 25 arcmin.

4. Magnitude calibration

The APM output image parameters include the isophotal inten-sity, the second order moments and the pixel area at eight differ-ent thresholds. An internal magnitude calibration was appliedas described by Bunclark and Irwin (1983). The resulting APMmagnitudes of the stellar images are almost linearly related tothe photoelectric magnitudes. The instrumental magnitude sys-tem was externally calibrated by photoelectric standards takenfrom Sandage (1970).

Fig. 3. Internal calibration of the instrumental magnitudes of a compar-ison plate by the instrumental magnitudes on the reference plate. Onlystellar images were used in the procedure of binning the magnitudedifferences along the comparison plate magnitude axis and computinga shift for each object by cubic spline interpolation. The line representsmagcomp.plate = magref.plate.

The most frequently used emulsion for both theB and theUband is ZU from ORWO Wolfen, which has been constant overthe years with respect to its spectral characteristics, althoughthe name has repeatedly changed (see Tab. 1). Both zero pointshifts and colour terms in the transformation relations betweenU, B, V magnitudes of the Tautenburg system and the stan-dard system are very small (van den Bergh 1964; Borngen &Chatschikjan 1967). The largest correction that occurs is a shiftof BT − B ≈ −0.1 for U − B > 1.5 (the index T meansthe Tautenburg system). In general, the corrections are muchsmaller and were therefore neglected.

4.1. B plates

Using all objects with stellar classification the instrumentalAPM magnitudes of each plate were transformed to the instru-mental magnitudes measured on the deeper reference plate. Thiswas done by binning the magnitude differences over the compar-ison plate magnitudes and computing the individual correctionto the magnitude of each star by a cubic spline interpolation.Only stellar images were used in this procedure carried out foreach comparison plate. Fig. 3 shows the relationship betweenthe instrumental magnitudes from plate B 6232 (before and af-ter the correction, respectively) and the instrumental magnitudesfrom the reference plate B 2176. In most cases the two distri-butions were nearly parallel with a somewhat smaller shift be-tween them. The tilt in this particular case is due to a changeof the APM output characteristics over the years but does notpose any problem for the internal magnitude calibration. Thecomputation of the correction to the instrumental magnitudeswas less accurate for the brightest stars. Their small numbermade it difficult to calibrate the magnitudes larger than 22 (in


Fig. 4. External magnitude calibration with photoelectricBmagnitudesfrom Sandage (1970): + images with stellar classification, × imageswith nonstellar classification. The line shows the result of the fit witha second-order polynomial using the stellar images only.

the instrumental system of the reference plate), correspondingto B < 12.

After transforming the APM magnitudes of all plates tothose of the reference plate the instrumental magnitude sys-tem was calibrated by photoelectric standards from Sandage(1970). From 52 standards with 11 < B < 22 identified, onlythe star-like objects (all with B < 20.2) were used in the fitwith a second order polynomial (Fig. 4).

In order to minimise photometric field effects from eachplate in comparison to the reference plate the whole field wasdivided into 11× 11 subareas (about 15 arcmin on a side) and asecond correction to the magnitudes of a given plate (i.e. simplya mean shift magref −magcomp) was computed and applied,separately for each subarea. Whereas this procedure removessystematic deviations between the photometric system of thecomparison plate with respect to the reference plate dependenton the plate position, it does not influence the magnitude dis-persion as a function of the position in the field.

4.2. U plates

Due to the use of a blue reference plate in the plate matching thecalibration of the U plates was performed in a somewhat differ-ent manner. First all plates were calibrated independently withthe photoelectric standards from Sandage (1970), using only theobjects with stellar classification by linear fitting the instrumen-tal APM magnitudes to the U magnitudes. The results for eachstar were averaged over the plates. After that the differencesbetween the U magnitudes on a particular plate and the meanU magnitudes were investigated and corrected for systematic

effects with respect to the magnitude and to the position in thefield. By this way the U magnitudes of all plates were trans-formed to a common system defined as the mean magnitudesystem of all U plates, whereas the B magnitude calibrationwas carried out in the system of one reference plate. As for theB plates our aim was to bring all plates to one magnitude systemsuited for an investigation of variability.

4.3. V plates

Due to the small number of V plates we applied only the exter-nal magnitude calibration by a linear fitting of the instrumentalmagnitudes of stellar objects with the photoelectric magnitudesfrom Sandage (1970), followed by an averaging of the V mag-nitudes over all plates an object was measured on.

5. Determination of proper motions

For a first determination of proper motion indices we had usedthe proper motions obtained from five pairs of plates by Scholz,Odenkirchen & Irwin (1993). All these B plates are also in-cluded in the present study (cf. Table 1) but this first sample ofproper motion objects was restricted by the plate with the low-est limiting magnitude among these plates. Therefore, in thefirst step of our variability-proper motion survey (Meusinger etal. 1995, 1996; Scholz, Meusinger & Irwin 1995) the numberof objects with available proper motion data was smaller thanthe number of objects with variability indices determined. Herewe describe a new proper motion reduction based on almost allplates listed in Table 1 and Table 2.

For this new proper motion determination we did not com-bine the plates in pairs but preferred to build up a time seriesof coordinates for each object. For this purpose the measuringcoordinate frame of each plate was transformed to that of thereference plate (No. 2176) by using a quadratic 12 plate con-stant polynomial relationship. In this manner we transformed the(x, y)-frames of 80 plates to that of plate B 2176. Four plates(B 1478, B 2001, B 6223 and B 6224) with only 25 per centoverlap, i.e. with the globular cluster in the plate corner, werenot used for the proper motion study.

After the quadratic 12 plate constant transformation wechecked the differences between the transformed coordinateson each comparison plate with respect to the coordinates on thereference plate as a function of the position in the field. Thiswas done using a two-dimensional binning and smoothing ofthe coordinate differences with a bin size of 20 arcmin × 20arcmin, separately for all matched objects and for stellar ob-jects only. No systematic effects have been found so that noerror reduction technique was applied to the residual coordi-nate differences. We concluded that the differential geometrybetween all plates with respect to the reference plate was wellrepresented by the quadratic polynomial relationship. We havealso experimented before that with a simpler linear (6 constants)coordinate transformation after which, however, we found formost of the plates systematic effects near the plate corners andedges.


Fig. 5. Proper motion errors for different magnitude classes: solid line- B < 17.5, dashed line - 17.5 < B < 19.5, dotted line - B > 19.5.

In former proper motion studies with a maximum of 10Schmidt plates measured with the APM facility in Cambridge(e.g. Scholz, Odenkirchen & Irwin 1993; Scholz et al. 1996) weapplied an additional one-dimensional error removal in order toexclude the known periodic errors of the measuring machine.Here we decided that this kind of correction, i.e. binning thecoordinate differences along the coordinate axes with bin sizesof about 3 arcmin and removing all systematic effects on thissmaller scale iteratively, was not necessary due to the large num-ber of plates by which these errors are simply smoothed. In fact,later we have not found any periodicity in plots of the resultingproper motions, based on more than 20 plates, over the coordi-nates. A more detailed discussion of the necessity for an errorreduction technique in this special case of a proper motion de-termination will be undertaken in a new study of the absoluteproper motion of the globular cluster M 3 in the centre of ourfield and in a new Galactic structure analysis based on the fieldstars data. But this will be subject of a different paper. For themoment we just note that we do not need any assumption ona constant motion of the field stars which is the basis of theerror reduction technique. This may be an advantage in a studyof the outer cluster regions as well as in a Galactic structureinvestigation.

After the plate matching and transformation of the coordi-nates of all measured objects on each comparison plate to themeasuring coordinate system of the reference plate we shiftedthe transformed coordinates by the negative mean coordinatedifferences of all available reference galaxies. The number ofgalaxies found on each comparison plate varied in dependenceon the limiting magnitude of the plate between 100 and 1500with an average of 750 (only 5 plates had less than 300 galaxies).

The absolute proper motion of each object was then deter-mined from the linear regression of the coordinates (x, y)j overtheir epochs Epj , with j = 1...npl, where the number of platesnpl for a given object was dependent on its magnitude, variabil-ity and location in the field. No weights were given to the plates.The error of transforming the coordinates of a plate to an abso-

Fig. 6. Standard deviation of B magnitude with respect to the mean Bmagnitude for all stellar objects (outside 25 arcmin from the centre ofthe globular cluster M 3) measured on at least 7 plates (max. 57 plates)with a typical number of 35 plates

lute zero point by use of the reference galaxies varied between 5and 30 mas. Only objects measured on at least three plates wereincluded in the proper motion determination. Here we used notonly the images with stellar classification but also those withnonstellar classification. The results of the proper motion deter-mination with all images of a given object were not significantlydifferent from those with the stellar images only.

Fig. 5 shows the proper motion errors for three differentmagnitude classes for the sample of about 12800 objects forwhich B variability indices were later determined on the basisof the photometry using only stellar images on at least seven Bplates. The mean number of plates used in the B photometryand in the astrometry for the different magnitude classes were47 and 75 (for B < 17.5), 42 and 70 (for 17.5 < B < 19.5),18 and 39 (for B > 19.5), respectively.

We have compared the proper motions with other propermotion catalogues in order to confirm the high accuracy of ourcatalogue. The dispersion of the proper motion differences ofcommon stars in two catalogues gives a realistic error estimatewhich represents the error sum

√(σ2

pm1 + σ2pm2), where σpm1

and σpm2 are the proper motion errors of the two catalogues.475 objects (including nonstellar images) of our new proper

motion catalogue were identified with objects in the catalogue ofTucholke, Scholz & Brosche (1994). This catalogue is based on14 plates of the Bonn double refractor (scale 40.44 arcsec/mm)over a time baseline of 85 years and contains 791 stars withB <16.5 in a field of 1.5◦×1.5◦ centered on the globular cluster M 3.The proper motion errors of this catalogue are typically 0.5 to1.2 mas/yr. In dependence on further selection criteria (numberof plates the objects were measured in both catalogues, distancefrom the cluster centre, etc.) the dispersion of the proper motiondifferences varied between 1.5 and 2.0 mas/yr for selected sub-samples of the 475 matched objects.

We have also compared our results with the highly accuratecatalogue of relative proper motions of Cudworth (1979) based


on 9 plates taken with the Yerkes 1-meter refractor (scale 10.655arcsec/mm) over a time baseline of 75 years. This cataloguecontains 266 stars covering only the central cluster region (r <17arcmin, B < 16.8, σpm ∼ 0.2 mas/yr). 170 stars could beidentified with objects in our catalogue. Due to the strong imagecrowding in the cluster region we excluded all stars classifiedas merged objects on the reference plate. The dispersion of theproper motion differences for the remaining 95 stars was foundto be 1.3 mas/yr. After further exclusion of the very centralcluster region (r < 400 arcsec) the dispersion decreased evento 1.0 mas/yr for 54 comparison stars.

From the comparison with two other proper motion cata-logues we conclude that our internal proper motion accuraciesare only slightly underestimated by about 10 per cent. Thatmeans, the formal proper motion errors determined from thelinear regression of the positions over the epochs can be mul-tiplied by a factor of 1.1 in order to get a realistic estimate ofthe errors. Nevertheless, we have achieved a high proper motionaccuracy comparable to the Hipparcos accuracy for a completesample of stars up to B = 18.5 in a field of 9 square degrees.

6. Variability index and proper motion index

6.1. Variability index Ivar

For each object a mean magnitude was calculated over the nplates on which it was measured. Fig. 6 shows the standarddeviation in B as a function of the mean B magnitude for allobjects measured (as stellar) on at least 7 plates. There is aremarkably large interval (12 < B < 18) with mean standarddeviations clearly below 0.1 mag whereas the mean standarddeviations rise up to 0.2 at the faint end (B = 20.5) mainlydue to larger random photometric errors near the plate limitwhich is different for each plate. The increase of the standarddeviations for stars brighter than B = 12 is due to problemsin the internal magnitude calibration (cf. Fig. 3) arising fromthe small numbers of stars above instrumental magnitudes ofabout 22 (in the system of the reference plate), but is of lessimportance in our QSO search.

Hereafter, we describe the definition of the variability indexfor the B plates. The variability index from the U plates wasobtained in analogous manner. An estimate of the photometricrandom errors of eachB plate as a function of magnitude σj(B)was determined from the comparison of the measurements onthat plate to the mean values from all B plates. More precisely,the σj(B) for the j-th plate were defined as the mean absolutedeviations between the mean magnitudes (over all plates) andthe magnitudes on the j-th plate. Objects with large absolutedeviations exceeding the dispersion at a given magnitude bya factor greater than 3 were excluded in the determination ofσj(B).

Following Majewski et al. (1991) we computed a variabilityindex as the average number of σ(B) a given object differed in

Fig. 7. Absolute differences between mean magnitude (over all plates)and the magnitude on a particular plate as a function of magnitude. Asan example twoU plates with different limiting magnitude and differentphotometric random errors with respect to the common magnitudesystem of all U plates are shown. The solid lines show the averageabsolute differences, i.e. the photometric random error of the j-th plateσj(U ).

absolute value from the mean magnitude, Bi, over all plates.For the i-th object measured on nipl plates we have:

I (B)var, i =

1nipl

nipl∑j=1

|Bij −Bi|σj(Bij)

with Bi =1nipl

nipl∑j=1

Bij , (1)

where σj(Bij) is the photometric random error of the j-th plateat the magnitude measured for the object on this plate.

The variability index described above is not only normalisedfor different magnitudes but also takes into account the differentphotometric random errors of the plates. An example of twoplates with different photometric random errors is shown inFig. 7. The distribution of the variability indices over the meanmagnitude is shown in Fig. 8. With regard to the considerationof the “instrumental” contribution to the measured variability,the above-described variability index provides a considerableimprovement compared to our previous study.

6.2. Long-term variability from structure functions

QSOs are expected to be variable on timescales of years andlonger (e.g. Smith & Nair 1995). Searching for objects whichvaried significantly with such long timescales provides an ef-ficient way to minimise contamination due to variable stars(Hawkins 1983; Veron & Hawkins 1995).

Structure function (SF) analysis is a robust method to detectvariability and to estimate roughly the timescale of the dom-inant variability process, even for relatively small numbers ofunevenly sampled data. The first order SF as a function of thetime-lag τ is defined as

SF (τ ) = 〈[m(t + τ )−m(t)]2〉, (2)


Fig. 8a and b. Variability index in B (a) and U (b) over mean Band U magnitude, respectively, computed on the basis of at least 5 Uplates and 7 B plates, respectively. Only objects located outside theglobular cluster region (r > 35 arcmin) are plotted in order to excludethe large number of variable stars in M3 from the sample. The dashedlines indicate the variability level 1.3.

where m(t) denotes the apparent magnitude measured at theepoch t, and the angular brackets indicate the time-average.(For definitions and general properties see e.g. Simonetti et al.1985; Meusinger et al. 1994b; Lewis & Irwin 1996).

For an object randomly variable on a time-scale of tvar ≈0.5 yr, for example, the SF monotonically increases to a maxi-mum at τ ≈ 100 d and remains essentially constant for longertime-lags. Generally the maximum occurs at τ ≈ tvar/2. Inorder to search for objects with variability timescales longerthan a few months, we compare the value of mean SFs fortime-lags longer than 100 d with that for time-lags shorter than100 d. Obviously, the ratio RS100 = SF (τ > 100)/SF (τ <100) is an indicator of long-term variability, so we can uselog10(RS100) as long-term variability index. For objects vari-able with tvar >∼0.5 yr we have log10(RS100) > 0.

We applied SF analysis to all objects with variability in-dices I (B)

var > 1.3 and I (U )var > 1.3, respectively (see Sect. 7.1).

The distributions of log10(RS(B)100 ) and log10(RS(U )

100 ) are shownin Fig. 9. The assumption of long-term variability for the knownQSOs in the field is clearly established. The candidate sample,on the other hand, is strongly contaminated by objects withshorter variability time-scales (or simply relatively large mea-surement errors). The mean values for log10(RS(B)

100 ) are 0.50 forthe QSOs and 0.24 for all variable objects. Fig. 9 can be used to

Fig. 9. Normalised distribution of the frequency of the long-term vari-ability index from structure function analysis for B magnitudes (left)and U magnitudes (right). The solid curves correspond to the samplesof all variable objects with Ivar > 1.3, the dashed curves to the sampleof the known QSOs in the field.

derive a limit of log10(RS100) suited for candidate selection. Weadopted log10(RS100) > 0.15 which yields a strong reductionof the candidate sample while only a few QSOs would remainundetected.

6.3. Proper motion index Ipm

The proper motion index was defined as the total value of theabsolute proper motion of an object in units of its error:

Ipm =µ2x + µ2

y√µ2xσ

2µx + µ2

yσ2µy

, (3)

where (µx, µy) are the two proper motion components and(σµx , σµy ) are their errors.

In our preliminary investigation (Meusinger et al. 1995,1996, Scholz, Meusinger & Irwin 1995) based on the proper mo-tions from only 10 plates (Scholz, Odenkirchen & Irwin 1993)we divided the total value of the absolute proper motion bya normalised proper motion error in dependence on the mag-nitude σµ(B). The magnitude-dependent proper motion errorsobtained from 10 plates were less than 3 mas/yr forB < 18 andincreased rapidly up to about 10 mas/yr at B = 20 (cf. Fig. 5in Scholz & Kharchenko 1994). In the present study, we deter-mined the proper motion index for the objects of our samplemeasured on a minimum of 3 and a maximum of 81 plates byusing the individual proper motion errors.

6.4. The Ivar - Ipm diagram

As the main result of our VPM survey we obtained a relation-ship between the variability index (in B and U ) and the propermotion index. Fig. 10 shows these results for all objects withproper motion indices lower than 10 which were measured onat least seven B plates (Fig. 10a) or five U plates (Fig. 10b),


Fig. 10a and b. Variability index inB (a) andU (b) over proper motionindex. Only objects with proper motion index lower than 10 and locatedoutside the globular cluster region (r > 35 arcmin) are plotted in orderto show the region of interest for the QSO survey (the upper left partof the diagram) in more detail and to exclude the large number ofvariable stars in M3 from the sample. The dashed lines indicate thevariability level 1.3 and proper motions exceeding their errors by fourtimes, respectively.

respectively. Here, and in the following, we exclude the imme-diate globular cluster region (r < 35 arcmin) to avoid contam-ination due to the large number of cluster variables and effectsdue to field crowding. The majority of the objects in the fieldhave Ivar < 1.3 (see also Fig. 8). In the following section wewill show that such a limit provides a useful search criterion forvariable objects.

For zero proper motion we allowed a proper motion index of4 corresponding to total proper motions four times exceeding thetotal proper motion error of a given object. Such a limitation doesefficiently reduce the contamination of the variability sampledue to variable nearby late-type stars (Weis 1994). This canbe seen by comparing the mean colours of the sample of zeroproper motion stars with those of the non-zero proper motionstars. Considering only the variable star-like objects (I (B)

var >1.3, B < 19.7) outside the cluster region we have (a) U −B =−0.16 ± 0.03, B − V = 0.59 ± 0.02 for stars with Ipm < 4,and (b) U −B = 0.13 ± 0.02, B − V = 0.72 ± 0.02 for thesample with Ipm > 4.

Table 3. Number of known QSOs with large variability indices for dif-ferent magnitude intervals (the numbers in parentheses give the num-bers of objects for which variability indices were determined)

mag. I (B)var > 1.3 I (U )

var > 1.3 I (B)var > 1.3 IBvar > 1.3

interval and orIUvar > 1.3 IUvar > 1.3

B < 19.7 49 (52) 39 (49) 38 (49) 50 (52)B > 19.7 28 (38) 11 (19) 9 (19) 30 (38)

6.5. Ivar and Ipm for the known QSOs in the field

There are about 140 QSOs known (i.e. with measured redshifts)within the whole field (NED1, Hewitt & Burbidge 1993, Veron-Cetty & Veron 1996b). With only two exceptions, all knownQSOs are located in the southern part of the M 3 field whichwas covered by the CFHT grism survey (Crampton et al. 1990,1992). From these 140 QSOs we identified 124 with measuredobjects on the reference plate (B 2176) using a search radius of10 arcsec. This sample was further reduced by the conditionsof stellar classification and measurement on at least seven Bplates so that we were left with 90 known QSOs included inour sample of about 12800 star-like objects (see Fig. 2). Thevariability properties of this QSO sample will be the subject ofa subsequent paper. Here we will briefly consider some generalproperties relevant to the completeness and success rate of theVPM survey.

For the QSOs with available variability indices there are86% with I (B)

var > 1.3 and 74% with I (U )var > 1.3 (note that

only for 68 of the 90 QSOs an U variability index was com-puted). In the Ivar-Ipm diagrams, the majority of the knownQSOs are located in the region indicative of both variability andzero-proper motion (compare Fig. 11 with Fig. 10). There are15 among the brighter QSOs (B < 19.7) which were alreadyincluded in our previous study (Meusinger et al. 1994) basedon iris-photometry of a larger number of B plates. The variabil-ity indices given there are found to be well-correlated with theI (B)var of the present paper (Spearman rank correlation coefficient

0.88), although both the definitions of the indices and the usedplates are not identical.

From 52 known QSOs with B < 19.7 there are 38 withlarge variability indices in B and U (see Table 3). If we lookfor candidates with IBvar > 1.3 or IUvar > 1.3 we find avariability fraction of 96 per cent for these bright QSOs. Thecorresponding numbers for the 38 faint QSOs (B > 19.7) are9 and 30. Due to the U plates going not as deep as the B plates(see Fig. 1), we were not able to determineU variability indicesfor all objects.

With the exception of one object all known QSOs show zeroproper motions, i.e. their total proper motions do not exceed the

1 The NASA/IPAC extragalactic database (NED) is operated by theJet Propulsion Laboratory, California Institute of Technology, undercontract with the National Aeronautics and Space Administration.


Fig. 11a and b. Known QSOs’ variability index inB (a) andU (b) overproper motion index. The dashed lines indicate the variability level 1.3and proper motions exceeding their errors by four times, respectively.The crosses mark QSOs with B < 19.7 and U < 19.0, respectivelyin a) and b), small circles show the faint QSOs with B > 19.7 andU > 19.0, respectively.

4σ level (see Fig. 11). The exception is 1336.4+2810 with a totalproper motion of 27.9±4.7 mas/yr determined from the coordi-nate measurements on 51 plates. A very similar result was ob-tained when only measurements with stellar classification wereused. The object is rather faint (B = 20.2, U = 19.2, V = 20.0)and was measured as a star-like object on 15 B plates and 7U plates. It was identified within 2 arcsec of the coordinatesgiven in Hewitt & Burbidge (1993). Crampton et al. (1988) givez = 0.5 for this object classified as an AGN by Veron-Cetty &Veron (1996b).

It is known that high redshift QSOs have smaller ampli-tudes of variability (Hook et al. 1994; Veron & Hawkins 1995;Meusinger et al. 1994b,c). However, due to our low threshold forthe variability index this does not necessarily introduce a strongz-dependent bias. Indeed, all of the 12 QSOs with z > 2.2 andB < 19.7 in the field are found to be variable with Ivar > 1.3,10 showing long-term variability. A more detailed discussion ofvariability properties of the known QSOs in our field is reservedto a subsequent paper.

6.6. Variability selection

Variable objects are selected using the indices Ivar andlog10(RS100), both for B and U data, if available. The statis-tics given in the previous sections were based on Ivar > 1.3.

Table 4. Number of known QSOs with large variability indices as afunction of the number of plates the objects were measured on (thenumbers in parentheses give the numbers of objects for which variabil-ity indices were determined)

numberof I (B)

var > 1.3 I (U )var > 1.3

plates> 4 50 (68) ∼73%> 6 77 (90) ∼86% 43 (55) ∼78%> 10 69 (78) ∼88% 32 (35) ∼91%> 15 61 (66) ∼92% 16 (17) ∼94%> 25 40 (42) ∼95%

This value will be derived for the B data from the estimatesof completeness and success rate in Sect. 7.2. For the U in-dex the limit should be larger by about 0.1 as the frequencydistribution of I (U )

var is slightly shallower than that for I (B)var (see

e.g. Fig. 10). The criterion for long-term variability was alreadygiven in Sect. 6.2. A QSO candidate has to show long-term vari-ability at least in that passband where variability is indicated byIvar.

7. Limiting magnitude, completeness, success rate

7.1. Limiting magnitude

Due to the restriction that only objects with stellar image clas-sification were included in the variability analysis the limitingmagnitude of our variability - proper motion survey is aboutB = 20.8. The number - magnitude relation for all star-like ob-jects measured on at least sevenB plates indicates completenessessentially up to B ≈ 20.0.

The existence of a limiting magnitude clearly induces a biasin any variability search: variable objects having mean mag-nitudes near the limit are preferentially measured during theirbrighter stages. The consequences are that (1.) the number ofdetections decreases and the variability index is underestimated,(2.) the sample of variable objects becomes increasingly incom-plete near the limit, and (3.) the sample of objects with strongvariability is biased against the fainter objects. These effects areclearly present in the Ivar vs.mean magnitude diagrams for theknown QSOs (Fig. 12; cf. also Fig. 1). The measured variabil-ity fraction increases with the minimum number of plates theQSOs were measured on (Table 4).

Besides sophisticated methods of survival statistics (e.g.Isobe et al. 1986), the application of Monte-Carlo simulationsis a possibility to overcome difficulties with such censored data.We have done extensive numerical simulations of the situation inour variability survey. The main aim is to estimate the maximummagnitudeBmax up to which such effects are still insignificant.

A sample of 80 objects on 54 plates were simulated usingplate limits Bj

lim distributed as in Fig. 1. The true mean mag-nitudes of the objects are distributed like those of the measuredmean magnitudes of the known QSOs in the field. The simu-


Fig. 12a and b. Variability index of known QSOs in B (a) and U(b) over mean B and U magnitude, respectively. The dashed linesindicate the variability level of 1.3 . The QSOs measured (as star-likeobjects) on more than 15 B plates and on more than 10 U plates aredrawn as crosses, respectively in a) and b), whereas the objects withmeasurements on 7...15 B plates and 5...10 U plates, respectively areplotted as small circles.

lated “measured” magnitudes Bij of an object i on a plate jinclude a Gaussian distributed photometric error according tothe σphot(B) relation from Fig. 6 and a Gaussian distributedmagnitude scatter σivar due to variability. The latter one differsfrom object to object with a dispersion σσ , where 〈σvar〉 and σσare free parameters of the simulations. From these “measured”magnitudes a variability index Iivar is calculated according toeq. (1), taking into account that an object may not be measuredon few particular plates (e.g. due to wrong classification witha probability depending on Bij − Bj

lim). An object is consid-ered as detected if Iivar > Ivar,min. The ratio of the numberof “detected” objects to the total number of 80 objects is takenas the completeness of the sample. The distribution of the sim-ulated “observed” variability indices was compared with thedistribution of the measured indices of the known QSOs to de-termine the values for σvar and σσ; best agreement was foundforσvar ≈ 0.30±0.15. From the inspection of the completenessas a function of Ivar,min and Bmax, respectively, we estimatethat a combination of Ivar,min = 1.3 and Bmax = 19.7 yieldssamples which are about 90 per cent complete (Fig. 13). Thissurvey limit is well below the mean B limit of the plates at 20.5(Fig. 1). Clearly, the use of mean magnitudes for the determina-

Fig. 13. Completeness versus limiting magnitude from simulationsof samples of variable QSOs with I (B)

var > 1.3. The three curves cor-respond to different assumptions on the mean magnitude dispersion ofQSOs 〈σvar = 0.4, 0.3, and 0.2 (from top to bottom).

tion of the survey limit is preferable to magnitudes at one epoch(e.g. reference plate) since Malmquist bias is much less.

7.2. Completeness and success rate

A main aim of our VPM survey is the detection of new QSOs. Forfollow-up spectroscopic observations we need a candidate listwith a high success rate but also with highest possible complete-ness. Both completeness and success rate are strongly depen-dent on the variability criterion, i.e. Ivar,min. A small variabilitylimit yields a large candidate sample with strong contaminationdue to instrumental scatter, i.e. a small success rate, but witha high probability to detect a significant fraction of the QSOs,i.e. a high completeness. Higher limits yield smaller and moreincomplete samples but with higher success rates. The value ofIvar,min for the candidate selection should be derived, there-fore, from the dependence of completeness and success rate onIvar,min.

Completeness and success rate of the VPM survey can beestimated from the QSOs detected by the grism survey in thesouthern subfield. Of course, this is correct only if the grism sur-vey is complete up to the limit of the VPM survey. The grismsurvey is supposedly complete to O = 20.5, i.e. deeper than theVPM limit of B = 19.7. On the other hand, it is worth point-ing out that O = U + B and as such will be about 0.25 magdifferent for UVX objects. The completeness is defined as thefraction of the known QSOs fulfilling the selection criteria ofthe VPM survey. The success rate means the ratio of the num-ber of these QSOs to the total number of VPM candidates. Wecalculate the success rate and the completeness from differentVPM candidate samples created adopting different variabilitythresholds Ivar,min = 0.8...2. In detail the variability selectionis as follows: The primary selection criterion is (see Sect. 6.2)I (B)var > Ivar,min, or I (U )

var > Ivar,min+0.1, respectively. Further-


Fig. 14. Completeness (solid curve) and success rate (dashed curve), re-spectively, of the VPM survey as a function of the minimum variabilityindex (see text).

more, a weak correlation between I (B)var and I (U )

var is demanded,if both indices were measured. This is expressed by the empiri-cally found relation: I (U )

var > 0.8 I0.6var,min if I (B)

var > Ivar,min andvice versa. (The results do not strongly depend on the exact pa-rameters of this relation.) The results, i.e. the completeness andthe success rate as functions of Ivar,min are shown in Fig. 14.For Ivar,min = 1.3 a high completeness of about 90 per centis indicated in combination with an acceptable success rate ofmore than 40 per cent.

8. The VPM candidate sample

8.1. General description

From 6078 stellar objects with B < 19.7 in the field, (clus-ter region excluded; without the known QSOs and any ob-jects known by SIMBAD) 1762 were found to have Ipm < 4.Adding the variability selection for I (B)

var and I (U )var from Sect. 6.6

yields a sample of 374 objects (for comparison: the constraintI (B)var > 1.3 gives 310 candidates). Combined with the long-term

variability constraint the sample is reduced to 168 objects. Thisis considered as our final candidate sample; its B magnitudedistribution is shown in Fig. 15. Adopting a variability limitIvar,min = 1.5 instead of 1.3 would result in a sample of 93candidates.

The sample of all variable zero-proper motion objects showsa homogeneous distribution of positions over the field (withoutcluster region). Due to the exclusion of the already known QSOsfrom our candidate sample, however, we have about 1.7 timesmore candidates in the northern than in the southern subfield. Itcan be shown easily that exactly the same north-south asymme-try is expected if the success rate of 0.42 from the southern halfis true for the whole sample. The other way round, this agree-ment is taken as an independent proof for the estimated successrate. Note that this argument does not exclude the possibility ofdetection of QSOs not found by the grism survey.

Fig. 15. Frequency distribution of the B magnitudes for the 168 QSOcandidates from the VPM survey.

In order to remove known variable stars from the QSO can-didate sample we have conducted a SIMBAD search. With anidentification depth of 10 arcsec 21 variable stars from SIMBADwere identified with objects on the reference plate. Among thesethere are two RR Lyrae stars showing high variability indicesand Ipm = 18.8 and 2.8, respectively, without indications forlong-term variability in the latter case (structure function analy-sis has been applied only to the zero-proper motion candidates).The zero-proper motion RR Lyrae has been, of course, removedfrom the QSO candidate sample. The remaining 19 objects areclassified in the SIMBAD catalogue as suspected variables. Wefound only 6 objects among these fulfilling both the criteria forthe variability indices I (B)

var and I (U )var and for zero proper motion,

three of these are located within the globular cluster region, twoothers have B > 19.7. The remaining one object shows strongindications for long-term variability and is not rejected from ourcandidate sample due to its uncertain nature.

8.2. The colour distribution of the VPM QSO candidates

From our UBV photographic photometry we obtained colour-colour-diagrams. The colours (U − B)i and (B − V )i of anobject i are mean colours based on all plates the object wasmeasured on. The minimum number ofU,B, V plates for colourspecification are 7/5/2. Considering star-like objects with B <19.7 and measured on at least 7 B plates, typical numbers are40/18/5. The U −B vs. B − V diagram for the whole sampleof objects with B < 19.7 is given in Fig. 16 a.

First, we consider briefly the expected location of QSOs ontheU−B vs. B−V plane. A preliminary comparison of colourselection with the VPM selection will be discussed in Sect. 9.The identified known QSOs in the field are shown in Fig. 16 b,the majority being from the grism survey by Crampton et al.Note that there is a selection bias for the fainter QSOs: only 50per cent of the QSOs with B > 19.7 were measured on at least5 U plates, compared to 94 per cent for the brighter subsample.As a consequence, objects with fainterU magnitudes, i.e. largerU −B in general, have a higher probability to be missed in the


Fig. 16a–f. U −B vs B − V colour-colour diagrams for objects with stellar classification and measured on at least 7 B plates, 5 U plates and2 V plates. The diagonal lines represent the relation (U − B) = −0.75 (B − V ) − 0.05 for colour selection (see Sect. 9). a All objects withB < 19.7 outside the region of the globular cluster M 3. b All identified QSOs in the field; open boxes: B > 19.7; filled boxes: B < 19.7;small dots: B < 19.7 but star-like on less than 5 U plates. c QSOs with z ≥ 2.2. Small filled boxes: from the grism survey of Crampton et al.with B < 19.7. Big open boxes: from the variability survey by Veron & Hawkins (1995) with B < 19.7. Small open boxes: the same source,but B ≥ 19.7. d 168 candidates from the VPM survey (B < 19.7). Candidates with strong variability in B (I (B)

var > 2) are marked with openboxes. 29 objects with significant variability in B but not enough measurements for the determination of variability in U (classified as star-likeon less than five U plates) are shown as faint dots. e The candidates from the southern half of the field. f The candidates from the northern halfof the field.

fainter subsample. There are only very few QSOs withU−B >−0.2. Those QSOs from the Crampton et al. survey havingz ≥ 2.2 are plotted again in Fig. 16 c. Multi-colour surveysare known to be very incomplete for such redshifts (e.g. Warrenet al. 1991), but the grism survey is sensitive to QSOs with 0.2 <z < 3.4. For the sake of comparison, we have overplotted theQSOs from the medium redshift sample of the variability surveyby Veron & Hawkins (1995; their table 3). The field of the Veron& Hawkins sample is about 4 times larger than the field of theplotted QSOs from the grism survey; thus the surface densitieson theU−B vs.B−V plane are incompatible. However, thereis also a difference in the distribution of U − B. Consideringonly QSOs with B < 19.7 and z > 2.2, we find mean valuesof U −B = −0.27 ± 0.05 for 12 QSOs from Crampton etal. and +0.16 ± 0.14 for 17 QSOs from Veron & Hawkins,respectively. The differences in B − V are insignificant. Forfainter limiting magnitudes the difference becomes even larger,which may be affected, however, by an increasing number of

fainter QSOs without measured U magnitudes. Note that thereis no such selection bias in the brighter subsample; all QSOswith B < 19.7 and z > 2.2 have been measured on at least fiveU plates. The colour corrections of ourU andB magnitudes areexpected to be much smaller than the difference of more than0.4 in U −B; negligible shifts of +0.05 and 0.00 are derivedadopting the transformation relations between Tautenburg andstandard U −B from Borngen & Chatschikjan (1967) and vanden Bergh (1964), respectively. Thus the data in Figs. 16 b,cseem to indicate that the colour distribution of QSOs detected byvariability surveys may be more extended toward larger U −Bthan those detected by the grism survey.

Fig. 16 d shows the colour-colour diagram for the QSO can-didates. The mean colours are U −B = −0.18 ± 0.04 andB − V = 0.52 ± 0.03 which is for U − B clearly less thanthe mean value for all star-like objects in Fig. 16 a. Candi-dates with strong variability in B (I (B)

var > 2) have coloursclose to those of the identified QSOs in the field. 29 objects


Table 5. Optically identified radio sources in the survey field. ∆ is the position difference, “class.” means the classification of the opticalcounterpart on the reference plate (s: stellar; n: nonstellar). The names are either from the FIRST survey (Becker et al. 1995) or from the 87GBcatalogue (Gregory & Condon 1991).

radio source name ∆ class. B U −B B − V I (B)var I (U )

var

FIRST J134222.2+294934 2.35 n 17.97 -0.28 0.58FIRST J134159.7+294653 0.81 n 15.66 1.49 2.18FIRST J134348.2+294109 1.63 n 20.69 1.61FIRST J134136.0+293810 1.89 n 17.45 0.30 1.08FIRST J134031.8+291538 0.47 n 20.60 1.03FIRST J133954.3+291147 2.11 n 18.61 0.07 1.2187GB 134421.5+291555 0.55 s 20.67 0.40FIRST J134255.0+285648 0.91 n 17.84 0.16 1.08FIRST J134014.9+285637 0.55 s 20.31 -0.81 1.09 1.30 1.08FIRST J134152.8+285238 1.37 n 20.50 -0.63 2.0987GB 134157.9+290727 2.63 s 20.07 -0.54 0.26 2.08FIRST J133809.6+284435 1.04 s 20.30 -1.09 0.81 1.77 2.57FIRST J133858.3+284412 1.56 n 19.31 0.13 1.5987GB 1345+2854 1.58 s 17.84 -0.10 -0.12 3.32 3.26FIRST J134808.6+284008 0.43 n 20.56 1.86FIRST J134400.3+283825 0.74 n 18.22 0.61 1.86FIRST J134444.4+283755 2.06 n 20.78 1.85FIRST J134630.3+283645 0.84 n 17.48 0.89 1.83FIRST J134214.9+282934 0.45 n 20.79 -1.68 2.01FIRST J134730.0+282915 1.48 n 17.83 0.72 1.94FIRST J134329.4+282853 2.95 s 19.63 -0.18 0.97 1.00FIRST J134152.2+282601 2.46 s 20.36 -1.08 0.79 1.20

were measured as star-like on less than five U plates and have,therefore, no I (U )

var. These objects are shown as small dots withU − B taken from all available measurements (including non-stellar image classifications) yieldingU −B = 0.28±0.08 andB − V = 0.90± 0.09, i.e. much redder than the rest. The vari-ability indices are I (B)

var > 1.5 for 12,> 1.8 for three and> 2 forone of these candidates. The reddest objects shown in Fig. 16 dturned out to be preferentially located near the plates edges andhave, therefore, a higher probability to be contaminants.

The distribution of the whole candidate sample is remark-ably different from that of the QSOs in Fig. 16 b. The coloursof the less variable objects are slightly concentrated toward thelocus of the field stars (U − B ≈ 0, B − V ≈ 0.5). Thismay be a hint on stellar contamination. However, the coloursof the intermediate redshift QSOs (Fig. 16 c) from a variabilitysearch (Veron & Hawkins 1995) are distributed over an areawhich is quite similar to that covered by stellar objects. Alsothe simulated colours for QSOs with z > 2.2 from Warren et al.(1991) show a concentration toward the locus of the stars (theirFig. 8). Further restriction of the candidate sample on the basisof colours will certainly improve the success rate of our surveybut clearly at the expense of completeness.

The candidates from the northern half of the field haveU −B = −0.25 ± 0.06 compared to −0.06 ± 0.04 for thesouthern part (Fig. 16 e,f). Such a difference is expected if thenorthern subsample contains the counterparts of the already-detected QSOs in the southern subfield. Indeed, for the com-

bined sample of candidates and QSOs in the southern part wehave U −B = −0.29, in good agreement with the northernsubfield.

8.3. Search for radio counterparts

In our field of 9 square degrees there are 434 radio sourceswhich we found in the NED. The majority of the radio sourcesare from the FIRST survey; these are located in the northernhalf of the field and have typical position uncertainties of about1 arcsec with a maximum of 9 arcsec. About five per cent ofthe NED radio sources have large radio position uncertaintiesbetween 40 and 80 arcsec. Due to these very different radio po-sition errors, as a first step toward an optical identification, 85objects were identified with a large search radius of 100 arcsecas the next neighbour object on the reference plate (measuredon at least two other plates, too). With the exception of onlyfive radio sources with large uncertainties (> 50 arcsec), allidentified sources have radio position error ellipses with largesemi-major axes less than 4 arcsec. From the remaining 80 ra-dio sources we selected owing to the high position accuracy ofthe optical sources from the reference plate only those within afinal search radius of 4 arcsec. For such conditions, 22 opticalcounterparts of radio sources have been identified (Tab. 5), allwith small proper motion indices (only two have 4 ≤ Ipm < 5),however, 15 of these have nonstellar classification. The most ofthe remaining 7 star-like objects are too faint to be included


Fig. 17. Completeness versus success rate for QSO candidate sam-ples from different search methods: (a) variability and zero-proper mo-tion (VPM), (b) ultra-violet excess (UVX), (c) UBV (2 col), and (d)the combination of (a) and (c). The asterisks are for the preferredselection parameters (Ivar,min = 1.3 for variability selection and(U −B) + 0.75 (B − V ) < −0.05 for UBV selection, respectively).

in our candidate sample. Only two optical counterparts withstellar classifications are brighter than B = 19.7, one beingstrongly variable with clear-cut evidence for long-term vari-ability (log10(RS(B)

100 ) = 0.86, log10(RS(U )100 ) = 0.81). This is

the only identified optical counterpart of a radio source in ourcandidate sample. It is contained in the list of blue objects nearthe north Galactic pole (Richter et al. 1968) and is quoted there-fore in the SIMBAD catalogue as AGN, though its extragalacticnature has not been established so far.

9. Comparison of variability-proper motion selection andcolour selection

The VPM selection yielded a sample of 168 QSO candidatesexpected to be complete to about 90 per cent with a success rateof more than 40 per cent. In Fig. 16 b, it is indicated that colourselection provides an effective search method, too. Most of theknown QSOs have U −B < −0.75 (B− V )− 0.05. Adoptingsuch a selection criterion we find a sample of 96 candidates.Completeness and success rate are estimated on the basis ofthe grism QSOs in the southern subfield to about 0.9 and 0.6,respectively.

In Fig. 17, the estimated completeness is shown in depen-dence on the success rate both for the VPM selection and thecolour selection. For comparison also the UVX search is in-cluded. For the VPM survey the free parameter is the mini-mum variability index Ivar,min. For colour selection the crite-rion U − B < 0.75 (B − V ) − c is used, with c being a freeparameter. The parameter of the UVX search is the maximumvalue of U − B (-0.1...-0.8). Obviously, the 2-colour selectionprovides the most clean and successful QSO candidate sample.One should keep in mind, however, that this result is based on(1.) mean colours from, in general, a large number of B and U

Fig. 18. B variability index vs. proper motion index for QSO can-didates selected from UBV colours (i.e. the objects lying above thecolour dividing line (U −B) < −0.75 (B − V )− 0.05 in Fig. 16 a).Long-term variable objects are marked by large boxes.

plates thus strongly reducing the effects of variability on colourselections. (2.) It was assumed that the QSOs from the grismsurvey in the southern subfield provide an essentially completeQSO sample, which may not be correct.

It is one of the main aims of the present study to find aQSO sample free of major colour-induced selection bias. Suchselection bias is expected for redder QSOs of low redshift (z <2.2) and for z > 2.2 in general. Of the five known QSOs belowthe colour dividing line (see Fig. 16 b), four have z > 2.2 andone z = 0.3. The VPM sample is considered as our basic listof candidates. A subsample of considerably improved successrate can be found, however, by combining the VPM with thecolour selection. The colour selected sample consists of QSOsplus some white dwarfs plus a small number of hot stars. TheIvar vs. Ipm diagram (Fig. 18) shows two clearly separatedsamples: (a) objects with low proper motion indices but a widevariety of variability indices and strong indications for long-term variability and (b) a sample of essentially non-variableobjects with a wide range of proper motion indices. The successrate for the combined VPM-UBV subsample of 59 objects isestimated to 0.76. The completeness of 79% is somewhat lowercompared to exclusive VPM selection. The UVX survey (U −B < −0.2), for comparison, provides a large (395) candidatesample of much lower completeness and success rate.

10. Conclusion

The combination of the variability and the zero-proper motionconstraints has been applied to construct a sample of 168 QSOcandidates with B < 19.7 in a field of about 9 square degree.About 40 per cent of the candidates are likely to be QSOs.Follow-up spectroscopy is envisaged for confirmation. In agree-


ment with other studies (Majewski et al. 1991; Hook et al. 1994;Meusinger et al. 1994b; Hawkins & Veron 1995) we find a highdetectable variability fraction of QSOs. The completeness ofthe QSO sample expected to be found from our survey is es-timated to about 90 per cent. The completeness of the VPMsearch is comparable to that of a 2-colour (UBV ) selection. Forred objects (i.e. below the colour deviding line in Fig. 16) thesuccess rate is likely much lower than 40 per cent. The 2-colourselection of blue objects provides a larger success rate thanVPM selection, but also redshift-dependent incompleteness. Acombined VPM-colour selection can be used to find a QSO sub-sample which is nearly clean but somewhat more incomplete.We would like to emphasize that we did not use colours as se-lection criterion but merely as a way of checking on how theVPM survey compares with more traditional methods.

The VPM QSOs combined with QSOs from other meth-ods (grism, colour, radio) will be well suited to investigate thevariability properties of an unbiased sample of QSOs with atime-baseline of three decades. The main goal of this projectis to contribute to a better understanding of the astrophysics ofvariability phenomena in QSOs and to check the hypothesis ofthe microlensing origin of QSO variability. A preliminary in-vestigation of the long-term variability of 52 QSOs in the fieldwith B < 19.7 already identified is in preparation.

Moreover, the extensive data base provided by the presentstudy will be useful for several other projects: (1.) We will usethese much more extensive and more accurate photographicphotometry and absolute proper motion data for a new deter-mination of the mean absolute proper motion of the globularcluster M 3 as well as for its membership study. (2.) The datamaterial is suited for a search for previously undetected variablestars in the outer cluster region. (3.) Thanks to the combinationof accurate proper motions up toB ≈ 20 with mean magnitudesfrom averaging over a considerable number of plates in threebands the data are very useful for a Galactic structure survey.

A similar investigation in another Tautenburg field (M 92)with a time-baseline of at least 33 years is presently in prepara-tion.

Acknowledgements. This research has made use of the NASA/IPACextragalactic database (NED) which is operated by the Jet PropulsionLaboratory, Caltech, under contract with the National Aeronautics andSpace Administration. The research has also made use of the SIM-BAD database, operated at CDS, Strasbourg, France. R. Ziener andF. Borngen are greatly acknowledged for their support of the obser-vation campaign. We thank J. Wambsganss for helpful discussion andfor reading the manuscript. We also thank the referee, P. Veron for hiscomments and suggestions.

References

Becker R.H., White R.L., Helfand D.J., 1995, ApJ 450, 559Borngen F., Chatschikjan E., 1967, AN 289, 19Bunclark P.S., Irwin M.J., 1983, in Rolfe E.J., ed, Proc of Internat.

Colloq. on Statistical Methods in Astronomy, ESA SP-201, p.195Crampton D., Cowley A.P., Schmidtke P.C., Janson T., Durrell P., 1988,

AJ 96, 816Crampton D., Cowley A.P., Hartwick F.D.A., 1990, AJ 100, 47

Crampton D., Cowley A.P., Hartwick F.D.A., Ko P. W. 1992, AJ 104,1706

Cudworth K.M., 1979, AJ, 84, 1312Gregory P. C., Condon J. J., 1991, ApJS 75, 1011Hawkins M.R.S., 1983, MNRAS 202, 571Hawkins M.R.S., Veron P., 1993, MNRAS 260, 202Hewitt A., Burbidge G. 1993, ApJS 87, 451Hook I.M., McMahon R.G., Boyle B.J., Irwin M.J., 1994, MNRAS

268, 305Isobe T., Feigelson E. D., Nelson P. I., 1986, ApJ 306, 490Kibblewhite E., Bridgeland M., Bunclark P., Cawson M., Irwin M.,

1984, in Capaccioli, M., ed, Proc. of Conference on Astronomywith Schmidt-Type Telescopes, Reidel Publ., p. 89

Koo D.C., Kron R.G., Cudworth K.M., 1986, PASP, 98, 225Kron R.G., Chiu L.-T.G., 1981, PASP, 93, 397Lewis G. F., Irwin M. J., 1996, MNRAS 283, 225Majewski S.R., Munn J.A., Kron R.G., Bershady M.A., Smetanka J.J.,

Koo D.C., 1991, ASP Conf. Series, 21, 55Meusinger H., Klose S., Ziener R.: 1994a, In: MacGillivray et al. (eds.),

Astronomy from Wide-Field Imaging, IAU Symp. 161, Kluwer,Dordrecht, p.737

Meusinger H., Klose S., Ziener R., Scholz R.-D., 1994b, A&A, 289,67

Meusinger H., Klose S., Ziener R., Scholz R.-D., 1994c, AG AbstractSeries, 10, 223

Meusinger H., Klose S., Ziener R., Scholz R.-D., 1995, ASP Conf.Series, 84, 486

Meusinger H., Scholz R.-D., Boller Th., Brunner H., Lamer G., Irwin,M., 1996, In: H.U. Zimmermann, J.E. Trumper, H. Yorke (eds.),Rontgenstrahlung from the Universe, Internat. Internat. Conferenceon X-ray astronomy and astrophysics, Wurzburg, 25-29 September1995, MPE Report 263, p. 483

Richter L., Richter N., Schnell A., 1968, Tautenburg Mitt. 38Sandage A., 1970, ApJ, 162, 841Sandage A., Luyten W.G., 1967, ApJ, 148, 767Scholz R.-D., Kharchenko N., 1994, Astron. Nachr. 315, 73Scholz R.-D., Meusinger H., Irwin M.J., 1995, AG Abstract Series, 11,

215Scholz R.-D., Odenkirchen M., Irwin M.J., 1993, MNRAS, 264, 579Scholz R.-D., Odenkirchen M., Hirte S., Irwin M.J., Borngen F., Ziener

R., 1996, MNRAS, 278, 251Simonetti J. H., Cordes J. M., Heeschen D. S., 1985, ApJ 296, 46Smith A. G., Nair A. D., 1995, PASP 107, 863Tucholke H.-J., Scholz R.-D., Brosche P., 1994, A&AS, 104, 161van den Bergh S., 1964 AJ 69, 610Veron P., Hawkins M.R.S., 1993, MNRAS, 260, 202Veron P., Hawkins M.R.S., 1995, A&A, 296, 665Veron-Cetty M.P., Veron P., 1996a, A&AS, 115, 97Veron-Cetty M.P., Veron P., 1996b, ESO Scientific Report, in pressVeron P., 1993, Proceedings of the 31st Liege international astrophys-

ical colloquium, p. 237Warren S., Hewett P. C., Irwin M. J., Osmer P. S., 1991, ApJS 76, 1Webster R. L., Francis P. J., Peterson B. A., Drinkwater M. J., Masci

F. J., 1995, Nature 275, 469Weis E. W., 1994, AJ 107, 1135

This article was processed by the author using Springer-Verlag LaTEXA&A style file L-AA version 3.

Documents

A UBV / variability / proper motion QSO survey from ...aa.springer.de/papers/7325002/2300457.pdf · A UBV/ variability / proper motion QSO survey from Schmidt plates I. Method and