ARTICLE IN PRESS

Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370

Manufacturing the Schmidt corrector lens for the Pierre Auger Observatory
M.A.L. de Oliveira*, V. de Souza, H.C. Reis, R. Sato
! Departamento de Raios Cosmicos e Cronologia (DRCC), Instituto de F! sica Gleb Wataghin (IFGW), Universidade Estadual de ! Campinas (UNICAMP), Observatorio Pierre Auger, Cx.P. 6165, CEP 13084-971, Campinas-SP, Brazil Received 14 August 2003; received in revised form 24 November 2003; accepted 26 November 2003

Abstract The Pierre Auger Observatory is designed to provide measurements of Extensive Air Showers initiated in the upper atmosphere by cosmic rays with energies greater than 1018 eV: One of the employed techniques is air uorescence detectors with UV lters and a corrector lens. We describe in this article the production process of the corrector lens in Brazil, starting from the design of the machines up to the nal tests in the prototype telescope. We have used diamond grinding tools to shape a ring of corrector lens with inner radius of 85 cm and outer radius of 110 cm; divided in 24 segments and with aspherical shape. After the production of a rst complete set of segments, we measured a segment shape scanning with a laser. The lens was then installed at the telescope and we measured its overall inuence on the spot size, taking pictures of the Vega star. r 2003 Elsevier B.V. All rights reserved.
PACS: 96.40.z; 96.40.Pq Keywords: Cosmic rays showers; Fluorescence detectors; Corrector lenses

1. Introduction The Pierre Auger Observatory [1] is a large international effort to try to solve one of the most puzzling mysteries of nature that has appeared in the last times: the Ultra High Energy Cosmic Rays (UHECRs). These cosmic rays are subatomic particles going through the universe and reaching Earth with macroscopic energy B1020 eVB16 J: How can these particles gain such tremendous

*Corresponding author. Tel.: +55-1937885276; fax: +551937885512. E-mail address: leigui@i.unicamp.br (M.A.L. de Oliveira).

energy? This is one of the questions that the Pierre Auger Observatory is aiming to answer. Cosmic rays with energies above 5 1019 eV were rst observed in 1962 by Linsley [2] in the Volcano Ranch array. Since then, several events with energy above 1020 eV have been observed [35]. However, the present arrays, with collecting area o100 km2 ; do not provide enough statistics, due to the extremely low ux (1 event century1 km2 ) in this energy region [6,7]. The explanation of these events face theoretical difculties, since it is believed to exist a cutoff in the spectra of cosmic rays around 5 1019 eV due to the interaction with 2:7 K primordial photons, the GZK cutoff [8,9].

0168-9002/$ - see front matter r 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.nima.2003.11.409

ARTICLE IN PRESS
M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 361

The observatory is a hybrid giant detector using two different techniques: water Cherenkov tanks and uorescence telescopes. The rst technique consists of an array of 1600 tanks separated by 1:5 km spread over an area of 3000 km2 : The UHECRs are so rare that even with such a large area, extrapolating the AGASA spectrum [10] below 1020 eV; we will have approximately 103 events/year. Some advantages of the Cherenkov tanks are their low cost and the fact that this kind of detector can operate continuously with any kind of weather. The ground array samples the lateral distribution of the shower generated by the primary UHECRs. The second technique consists of 24 telescopes overseeing the array of tanks. These telescopes observe the uorescence light emitted by the air nitrogen molecules when they are excited by the particles of an Extensive Air Showers (EAS). They operate with clean weather in moonless nights. This condition corresponds to only 10% of the available time, but this disadvantage is compensated by the crucial information that the uorescence detector gives to us: the longitudinal distribution prole of the EAS. This information is used in order to determine the energy, the arrival direction and to estimate the composition of the UHECR that originated the showers. Each one of the 24 telescopes is composed by a spherical mirror with 30 30 eld of view reecting the uorescence light on an array of 440 photomultiplier tubes (PMTs). The system adopts Schmidt optics in order to reduce coma aberration. The basic setup was thought to have an aperture box consisting of a 85 cm diaphragm radius to exclude distant off-axis rays and a band pass lter passing the 300 nmolo410 nm band, which contains the main emission peaks of the nitrogen uorescence. However, it was proposed that the inclusion of a corrector ring in the outer radius part of the diaphragm would increase the collection area of the telescopes, increasing the signal to noise ratio without signicant loss in resolution. The logic for such a design is spelled out in the next section. Simulations have shown [11] that the use of a corrector ring would increase the collection area signicantly, still keeping the aberration within the specied value of 0:5 (which is one-third of the

PMT aperture for the Auger uorescence telescope). Indeed, it was possible to determine the most suitable shape of the corrector ring lenses, as discussed in Ref. [11]. Here, we start from the results described in this paper and explain the steps given for the construction of the rst prototype ring, from the machine conception to the nal tests of the sample lenses.

2. Lens construction The Fluorescence Detector telescopes uses Schmidt optics following the proposal in [12] and are sets of spherical mirrors of 340 cm of curvature radius with a diaphragm limiting the coma aberration. Due to the telescope geometry, the spots at the focal surface have spherical aberrations. Their intensity is proportional to the effective collecting area A and the noise is proportional to the square root of this quantity, in such a way that the signal to noise ratio S=N is given by r S A p 1 N O where O is the solid angle covered by a PMT pixel. The diaphragm is introduced to cut rays far away from the optical axis, avoiding the coma aberration [13] and allowing a wider eld of view in comparison with a normal telescope. As the S=N ratio can change with both mirror and PMTs diameter, one can increase the mirror diameter or decrease the PMTs pixel diameter to increase the S =N ratio, but anyone of such solutions would raise too much the costs. Thus, for the designed mirror values (340 cm of curvature radius and 30 30 of aperture) the spots achieve the required resolution (0:5 diameter in the focal surface) for a 85 cm diaphragm radius. The idea of a corrector lens appeared as a way to increase the telescope area, increasing the spot luminosity, without increasing the spot size in excess. The shape of a ring has been chosen instead of a lled lens to reduce the nal cost and because the simulations showed that it was needed to correct only the rays beyond 85 cm from the main optical axis. It has been decided to increase the

ARTICLE IN PRESS
362 M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370

diaphragm radius in 25 cm: Taking into account the glass transmittance and losses through the segment borders, we calculated an overall light loss of about 20% within the corrector ring. Then, including the shadow of the PMT camera, the effective collection is 86% larger for an incident beam parallel to the telescope optical axis 0 and 63% larger for an incident beam inclined of 20 with respect to the telescope optical axis. For further details, see the reference [11]. In Fig. 1 is shown the geometry of one segment of the corrector ring, the studied unit vector normal to the curved surface ~ n and its projections along cylindrical symmetry axes. The graph in Fig. 2 shows the limits within which the radial component of the normal vector ~ nr should lie [12], taking into account the chromatic aberrations, which is discussed in the Appendix A. Some proles were suggested by the collaboration [14,15] in an attempt to simplify the production (see Fig. 3). The comparison of these proles with the previous studied limits is also given in Fig. 2. For the circular approximation, case A, the lens surface is a toroid which goes through itself, in this plot is a straight line passing through the point 0:8 m; 0: We can see that the lens prole cannot be approximated by a spherical surface in the whole segment extension. However, in the case B, a part of the prole is out of the allowed region, then the spot size that the telescope will produce using this type of lens is larger than the required 0:5 0:5 as is written in Ref. [15]. In the case A, it seems to be good enough though a small part of the prole is out of the allowed region: this

Fig. 2. Radial projection of the normal unitary vector, nr ; to the curved surface as a function of the distance, r; to the telescope axis. We compare the limits to maintain the spot resolution with the two proposed proles (as discussed in Fig. 3) and the curve produced by our machine.

Fig. 3. Proposed lens proles for the corrector ring: case A by Palatka [14] and case B by Cordero [15].

Fig. 1. Illustration of the lens surface normal and its projections.

happens because we considered the spot size as the circle that contains 100% of the light [11], while in the Ref. [14] it was taken for 90% of the light. However, our calculation (using 100% of the light) is more restrictive. One way to produce lenses for the Schmidt telescopes is by a vacuum method [16] in which one deforms a parallel plate and grinds a spherical surface while the glass is bent. However, we could not use this method because the deformation would break the plate that we were using, made of optical glass. Other difculty in manufacturing the lens is its size (220 cm in external diameter).

ARTICLE IN PRESS
M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 363

Fig. 4. The project of the corrector ring segments.

The problem is not only the machine to produce this huge lens but also to obtain the appropriate material and transportation to the site. To solve these problems the lens was divided in 24 segments (following the suggestion by Paul Sommers). Fig. 4 shows the layout of each segment. A problem that appeared with segmentation was the fact that the center of symmetry is far away from each piece that could make its production more difcult. One way to solve these problems was suggested by the enterprise Schwantz Ltd1 that proposed a machine with a circular base to hold all the 24 segments. The circular base rotates around its central axis and a circular tool (a disk with small abrasive diamond cylinders) grinds the material. Only one edge of this disk is in contact with the glass. Considering that the position of the tool, the material and the base are rigidly positioned, the prole that such machine can produce is shown in Fig. 2 (see Appendix B). The machine prole depends on the position, orientation and the diameter of the tool. The diameter of the tool was chosen to be 35 cm for two reasons: Schwantz has technical problems to produce larger tools and it is impossible to produce the desired surface prole with a smaller tool (see Fig. 15b and Ref. [11]).
1 ! ! rcio Optico Schwantz Ferramentas Diamantadas e Come Ltda, Indaiatuba, Brazil.

The process starts with rectangular 296 307 12 mm3 BK7 glass plates from Schott2. First, one surface is ground and polished to become as plane as possible using a planning machine. The planning machine is constituted by a needle that controls the position and orientation of a grinding disk with about 20 cm in diameter and the glass plate rotates around its center. With the friction between glass plate and grinding disk, this disk rotates and grinds the plate almost homogeneously. This process does not produce a really plane surface but it is good enough for our purposes: deviation of the normal is around 2 mrad: Then the glass is cut to the shape (see Fig. 4) and is put on the machine to make the prole. The machine (as illustrated in Fig. 5) was constructed at Schwantz considering the described ideas. The base disk can be controlled to spin with a period in the range of 340 min: The grinding disk is connected directly to its motor axis and the motor to a base which allows to control the orientation and the position of the disk by screws. The frequency of the grinding disk is of about 700 rpm: The produced prole was controlled during grinding process measuring the curvature of the surface with a spherometer.

Schott Glaswerke, Mainz, Germany.

ARTICLE IN PRESS
364 M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370

Fig. 6. Left: illustrates the furrows that appeared in the lens using only the griding disk; Right: shows the system used to copy the lens prole to the pseudo-random movement tool.

production of an entire corrector ring, with seven manual interventions between each step. Although approximately 75% of this time is consumed by the grinding machine which is autonomous.

3. Lenses tests After the completion of the lenses production, we have performed tests to measure their surfaces proles. These tests were done with a laser scanning method and with a CCD image of a star, as we will describe below. 3.1. Laser method
Fig. 5. Design of the constructed machine to produce the curved prole of the lens.

However, a problem appeared using this system (probably because the system is not rigid): the produced surface had small furrows (see Fig. 6). Then, another tool was introduced to correct these defects. It is composed of an iron plate which has approximately the shape of one lens segment with grinding pastilles of aluminum oxide. It accommodates under the plate and makes pseudorandom movements on the glass while the base makes a revolution in about 10 min: Before using this tool, the produced surface of the lens is copied putting a sandpaper on the corrector ring surface while the tool performs the pseudo-random movements on the sandpaper (see Fig. 6). After all these steps the segments need to be polished, what is done by both correction tools, but replacing the diamond cylinders by felts. The whole process takes about 180 man-hours for the

When a beam of light impinges on a surface separating two media with different refraction indices, it splits in two beams: the reected and the refracted. Therefore, impinging a beam on the lens segment we can see three beams: two going backward and one going forward, relatively to the incident beam. This happens because the beam that goes in the lens can undergo multiple reections before getting out the lens. Measuring the direction of the incident and of the outgoing beams we can obtain the normal of the lens surfaces and the lens prole. In Fig. 7, we have a layout of the system that was used to study the lens prole. A diode laser was used as the light source, S ; whose wavelength is between 635 and 670 nm: On the screen, A1; we measured the positions, R1 and R2; with a millimeter paper. The scan method consists in varying the impinging position, L1; on the rst surface, S 1; and reading the arriving points, R1

ARTICLE IN PRESS
M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 365

Fig. 7. Scanning the surfaces S1 and S2 of the lens with a laser at the light source S : The reections R1 and R2 on the screen A1; as well as L1 and S; are used in the ray-tracing reconstruction.

and R2; on the screen. To make easier the measurements of the positions L1; we put a paper with holes of 1 mm of diameter along the lens x-axis. Unfortunately, the source has an undesirable angular divergence (of about 0:5 cm after a propagation of 400 cm). This gives an additional uncertainty in the measured positions R1 and R2 and consequently in the lens prole. The way we have treated these uncertainties is described below. We aligned the system, making the beam to turn back on itself, restricting the incident light to propagate on the plane x z: A reference point on screen A1 was adopted averaging two adjacent reections to the laser shadow. With the positions S; L1 and R1 we can obtain the direction of the ~1 and of the reected ~ incident ~ i SL r1 ~1 beams. The direction of the normal to the L1R surface S 1 is ~ ~ i r1 ~ n1 p : 2 ~ jj~ r1 jj jji jj Measuring the normal in some points of the surface S 1 we can obtain its prole. Almost the same approach can be used to get the prole of the plane surface, S2; with S ; L1 and R2; now considering the direction of the incident and reected light inside the lens using the direction ~ n1 ; obtained in the previous step. We adopted a refraction index of 1.514 for the BK7 glass2 to

reconstruct the data, which is the value for the center of the laser wavelengths band. The main uncertainties of the laser method come from the angular divergence of the source. This was treated by a Monte Carlo method in the following way: we took randomly distributed points within the spots, according to a Gaussian distribution with mean equal to the center of gravity of the spot and standard deviation given by the spot divergence, i.e., the points were sorted within S 7DS ; L17DL1; R17DR1 and R27DR2 and we calculated the normal unit vectors of each surface, n1 and n2 : We repeated this process several times and we got a distribution of n1 and n2 for a given measured datum. The results for n1 and n2 and the statistical uncertainties are presented in Fig. 8. The graph of Fig. 8 is a plot in a general reference system dened by the laboratory walls. We can see that the component in y-axis of both surfaces almost vanishes along all segment. It means that the produced prole is almost symmetric, as desirable. The other component (x) seems to follow a straight line. So, as x projection is changing, we can say that these surfaces have approximately circular prole, similar to what was proposed by Palatka (see Fig. 3) where the curvature radius can be obtained as C dnr =dr1 : As the alignment of lens segment has been done with the reected light going back to the source, the plane surface S 2 does not coincide with the plane x y: Consequently, we cannot compare directly these results with Fig. 2. But, we must rotate the results to the frame in which the average normal of the surface S 2 is pointing in the direction of z- and x-axis is pointing along the radial direction of lens segment. The new plane x y of this frame should be very near the plane in which the segment was holding on the machine during the production process. Fig. 9 shows the results for the curved surface of four segments. The full lines show the limits, given by the simulation, within which the data must be and the dotted line is the prole that the machine should produce if everything (disk tool, base and glass) were rigid and assuming the following parameters: R 97:5 cm; rc 17:5 cm; yr 1:10 and yf 1:07 (see Appendix B, Fig. 14, for their denitions). We can see that the data t

ARTICLE IN PRESS
366 M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370

Fig. 8. Normal unit vectors components as a function of the distance to the diaphragm center for a sample segment. On the left side is given the projections of n1 (normal to the curved surface) and on the right side the projections of n2 (normal to the plane surface). In each sub-gure the projections are designated in the upper left box.

very well in the region from 87.5105 cm: We think that the disagreement near the edges of the lens (or outside the region described above) is mainly due to the introduction of the second header (random movement header). But, as we will see further in the next test, this problem does not poses us serious difculties. The curvature radius of the lens can be calculated as the inverse of the inclination of the line which best describe the data: C 1 dnr =dr: Fitting the available data of Fig. 9 in the range 90 cmoro103 cm; we can calculate C 64075 cm:
Fig. 9. Data: resulting proles of four samples shifted by about 90o within the ring (the segments are numbered from 1 to 24 for the full ring); full curves: the allowed limits given by the simulation and; dashed line: the best t for the machine curve. The errors are smaller than 0.44 in the same ordinate scale.

3.2. Star picture at the Auger telescope This test was done using the Auger Fluorescence Telescope in Malargue, . Argentina. The data was

ARTICLE IN PRESS
M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 367

taken during four nights of March 2002. During this period the Vega star was in the eld of view of the telescope. Vega (Alpha Lyrae) is a star of spectral class A0V, with zero magnitude and a relatively at electromagnetic spectrum within the wavelengths 350850 nm: The test consisted on taking pictures with a CCD camera of the star image formed at the focal surface. The light of the star enters the system through the aperture window and is reected by the mirror into the camera surface. The camera is covered with a white UV reecting screen. The CCD camera is placed in front of the mirrors aimed at the screen so that we can get the image of the star (see Fig. 10). Our intention was to take several pictures of the star and to determine the spot size of its image in the camera surface. This measurement was crucial to quantify the quality of the whole telescope setup by establishing the total inuence of the corrector ring on the spot size. In order to increase the S=N ratio avoiding the motion of the star in the camera eld of view, we took 32 pictures of the star in 8 s exposure time each in two telescopes congurations: without corrector ring (85 cm of aperture diaphragm radius) and corrector ring only. For the latter setup, we masked the inner 85 cm of the aperture in order to get only light that passed through the

Fig. 11. Superpositions of Vega pictures without corrector ring (above) and with corrector ring only (below).

Fig. 10. Taking pictures of Vega spot image with a CCD camera.

lenses. Fig. 11 shows the superposition of the 32 pictures of Vega in both cases. The analysis procedure consisted in measuring the light integral as a function of the distance from the star center. This can be seen in Fig. 12 where it is possible to notice the effect of the corrector ring. In each case, the light intensity was normalized to the own total amount. The light that enters the telescope passing through the lenses is concentrated in the center of the spot while the light coming from the inner part of the aperture is spread in a larger area. The full line in the plot shows the concentration of light for the total aperture of the telescope (radius of 1:1 m) with the corrector ring in place. The gure shows the good corrector ring performance, approximately 95% of the light through the corrector ring will end within the 0:75 cm spot radius, which corresponds to the specied value of 0:5 ; or one third of the PMT pixel size.

ARTICLE IN PRESS
368 M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370

work. This work was supported by FAPESP and CNPq (Brazil).

Appendix A. Chromatic aberration Let nr be the radial component of the normal to the curved lens surface and suppose that the plane #-axis, the allowed surface has its normal along the z values for nr r were studied in Ref. [11]. As the lens prole, in that work, was calculated with a specic refraction index, the chromatic aberrations were not taken into account. However, we know that there will be a variation in the refraction index in the allowed UV lter region of wavelengths. In the case of the used glass (BK7), the refraction index ranges from 1:550 to 1.528 (from the shortest to the longest wavelengths). Thus, one needs to calculate the limits for each refraction index value, as shown in Fig. 13. So, our goal is reached if we guarantee that the surface prole is in the region between the continuous upper line and the dashed lower line. These limits are adopted throughout this work.

Fig. 12. Light integral of Vega pictures. This plot determines the spot size of the telescope conguration.

4. Conclusion We described in this article the production of the corrector ring for the aperture system of the Pierre Auger Fluorescence Detector. Starting from its conception and supported by a ray tracing simulation [11], where the limits of the projections of one lens surface normal over the another were established, we detailed the proposals and the implemented process, describing the machine that has been constructed to produce the ring. After the production was completed, we tested some sample segments by direct laser scanning and the whole ring by star picture. In the rst test, we achieved the lens prole of one surface, as produced by the machine, while in the last we evaluated the overall performance of the corrector ring working together with one telescope at the site. The results show that the segments were within the specications and that it was possible to increase the collection area of the detector without increasing the spherical aberrations.

Appendix B. Lens prole produced by a circular tool The surface dened by the revolution around the position O of the circumference with radius rc

Acknowledgements We would like to thank Carlos O. Escobar Paul ! rio Marcon, AntSommers, Hans Klages, Roge # onio C. da Costa and Emerson Schwantz for important contributions on many aspects of this

Fig. 13. Chromatic aberration: the limits within which the lens prole should be produced for the correspondings refraction indices for l 410 nm n1 and l 300 nm n2 :

ARTICLE IN PRESS
M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 369

std R+ RLimits

-10 1000 X nr

-20

-30

-40

-50 0.85

0.9

0.95 r(m)

1.05

1.1

(a)
0

std r+ rLimits

Fig. 14. System to produce lens prole using a disk with radius rc :
1000 X nr

-10

-20

-30

and which is in the plane that pass through the ~ and the normal ~ point R n (see Fig. 14) may be described by the height z and distance r from the position O since the surface has cylindrical symmetry. # we would obtain a toroidal If we had ~ n py surface described by equation is z2 r R2 r2 c; that is the prole proposed by Palatka et al. [14] (see Fig. 3A) with rc 838:3 cm and R 79:527 cm: Indeed, we would like to analyze the surface produced when the normal ~ n is almost parallel to #-axis. In order to do this, we calculated the the z position of the points of the disk writing the component z as function of r to a given ~ n ; R and rc : With these points we calculated numerically dz=dr and then the radial component of the normal, using the expression of Ref. [11] Z
r r0

-40

-50 0.85

0.9

0.95

1.05

1.1

(b)
0

r(m)

std nn+ Limits

-10 1000 X nr

-20

-30

-40

(c)

-50 0.85

0.9

0.95 r(m)

1.05

1.1

std nn+ Limits

-10

1000 X nr

-20

zr z0

nr r 0 q dr0 1 nr r0 2

-30

B:1
-40

(d)

-50 0.85

0.9

0.95

1.05

1.1

In Fig. 15 we have some of these surfaces within the tolerance limits for the lens proles. We can see that the overall surface is moved by WR when we change R to R WR (Fig. 15a), while decreasing (increase) rc the prole becomes more curved (straight) in the graphic nr r (Fig. 15b). The others two graphics show how the direction ~ n changes the prole.

r(m)

Fig. 15. Surface proles produced by the system of Fig. 14 with different parameters. All the graphics show the two limits of the lens proles and std the surface produced with parameters R 0:975 m; rc 0:175 m and normal components of plane of disk nr 0:021 and ny 0:015: Other proles are identied as x7 means that the parameter x was changed x-x7Wx; where WR 0:025 m; Wr 0:025 m; Wnr 0:005 and Wny 0:005:

ARTICLE IN PRESS
370 M.A.L. de Oliveira et al. / Nuclear Instruments and Methods in Physics Research A 522 (2004) 360370 [11] R. Sato, J.A. Bellido, H.C. Reis, Improving the S=N Ratio for the Auger Fluorescence Detector, GAP 2000-009; http://www.auger.org/admin-cgi-bin/woda/gap notes.pl. [12] A. Cordero-Davila, et al., Proposal for the Optical System of the Fluorescence Detectors of the Auger Project, GAP 1996039; http://www.auger.org/admin-cgi-bin/woda/gap notes.pl. [13] M. Born, E. Wolf, Principles of Optics, Cambridge University Press, Cambridge, 1980. [14] M. Palatka, et al., Analysis of Possible Substitution of the Aspheric Surface of the Corrector Annulus by the Shape With Simpler Production Technology, GAP 2000-003; http://www.auger.org/admin-cgi-bin/woda/gap notes.pl. [15] A. Cordero-Davila, et al., Segmented Spherical Corrector Rings 1: Computer Simulations; GAP 2000-018, http:// www.auger.org/admin-cgi-bin/woda/gap notes.pl. [16] E. Everhart, Appl. Opt. 5 (1966) 713.

References
[1] Pierre Auger Observatory Collaboration, Pierre Auger Design Report, Fermilab, 1997; http://www.auger.org/ admin/DesignReport/index.html. [2] J. Linsley, Phys. Rev. Lett. 34 (1963) 146. [3] D.J. Bird, et al., Astrophys. J. 441 (1995) 144. [4] N. Hayashida, et al., Phys. Rev. Lett. 73 (1994) 3491. [5] M.A. Lawrence, R.J.O. Reid, A.A. Watson, J. Phys. G 17 (1991) 733. [6] M. Takeda, et al., Astro/ph-0209422, November 2002. [7] D.J. Bird, et al., Phys. Rev. Lett. 71 (1993) 3401. [8] K. Greisen, Phys. Rev. Lett. 16 (1966) 748. [9] G.T. Zatsepin, V.A. Kuzmin, JETP Lett. 4 (1966) 78. [10] S. Yoshida, et al., Astropart. Phys. 3 (1995) 105.