The entry says that two instruments disagreed with the one that was trusted and that their readings were set aside as the fault of the instruments. That is the whole story in a sentence, and it is worth slowing down, because the three devices were not alike, the engineers’ ranking of them was not foolish, and the measurements the board made in 1990 were ingenious in a way the entry can only summarise. What follows comes from the board’s report of November 1990, its Appendix D on the interferograms and Appendix E on the mirror as built, and from the quality report that George Rodney’s office wrote beside it.

What a null corrector is for

A spherical mirror is easy to make and easy to test, and it makes a poor image, because rays from its edge come to a focus nearer the mirror than rays from its centre. The cure is to take glass away from the edge until every ray meets at one point: a parabola for a single mirror, a hyperboloid for each of the two mirrors of the Ritchey-Chrétien telescope that Hubble is. The amount to take away grows with distance from the centre, and at Hubble’s edge it is a matter of micrometres. The difficulty is that a hyperboloid, unlike a sphere, cannot be compared with a reference surface of its own kind, because there is none. So opticians put a small optical system between the light source and the mirror that undoes the asphericity on the way out and back: through it, a perfect hyperboloid returns a flat wavefront and straight interference fringes, and every departure from the hyperboloid shows as a bend in the fringes, half a wavelength of surface error per fringe. The board called such a system “an optical template of the desired aspheric shape that can be designed to be accurate to better than 25 nanometers.” A template is only as good as its own making, and “carrying out an unambiguous and accurate test to determine whether a null corrector is producing the correct optical template is a known difficulty.”

Perkin-Elmer had three of them in the tower.

The refractive corrector

The first was the ordinary kind, two lenses, precisely made and precisely spaced, of the type that had tested the primaries of earlier telescopes. Perkin-Elmer had concluded that a refractive corrector “would not yield sufficient precision for testing the figure of the primary mirror”, and used it for the coarse work. From December 1978 at Wilton the mirror was ground and polished against it to about one wavelength of the right shape, and then moved to Danbury. It came back for one more job in May 1981, after the final polishing: the measurement of the mirror’s vertex radius, its basic curvature at the centre, which the reflective corrector could not make because the holes in its two mirrors hid the vertex. A calibrated spherical test plate was set in the mirror’s central hole, white-light fringes were used to align it, a process “extremely sensitive to vibration” whose fringes “could not be captured on film because of the short duration and faintness of the images”, so that “several observers were required to witness that the fringes were seen”; then a helium-neon laser replaced the white light and a photograph was taken.

The photograph, Figure D-2 in the report, is grainy and speckled, with a washed-out ring near the hole. Its purpose was the inner zone, where the mirror’s fringes could be compared with the plate’s. Across the rest of the aperture the fringes curve. “Although the spherical aberration is clearly evident in Figure D-2, the intention in the test procedure was only to provide a straight fringe reference in the inner portion of the mirror, and no attention was to be paid to the curve fringes.” In 1990 several people evaluated them by computer and got 0.4 of a wave of third-order spherical aberration, “to a reasonable level of certainty the error existing in the flight telescope.” The refractive corrector had been believed to have errors of its own. The board had its as-built design analysed and found it accurate to 0.02 of a wave, with one small design slip that put a little fifth-order aberration into it, within its error budget. Rodney’s investigators found something else: of the three instruments it was the only one whose assembly the Wilton quality organisation had witnessed and signed for.

The reflective corrector

The second was the company’s own invention and the reason, in part, that it had won the contract: a reflective null corrector of two spherical mirrors and a small field lens, where the common design has one mirror and a lens. Its virtue was that its template could be calculated exactly “simply by knowing the manufactured dimensions of the two mirrors and the lens, including the lens material, and the spacings of the three optical elements”, and that it had been built so that the spacings could be measured at any time, interferometrically, with metering rods between the centres of curvature. A memorandum of November 1974 on its design, for the 1.5-metre demonstration mirror, had put the philosophy plainly: “Precision of optical performance must be given the highest priority”; and, since “optical performance of the complete assembly cannot be periodically verified by a single test”, it “must be certified by evaluation of separate optical components and monitoring critical alignment procedures under laboratory conditions.” That corrector was built with the company’s research money and, Rodney found, with “detailed build records” showing “a standard QA process”.

For Hubble it was modified: a new field lens, new spacings between the interferometer and the upper mirror, between the two mirrors, and between the field lens and the lower mirror. No formally released drawings of the modification were ever found; Rodney’s investigators thought it “may have been done via redlining the 60-inch drawings.” There were three Invar rods, A, B and C, with rounded, polished ends, located by auto-reflecting a focused beam off the rod end and watching the returning pattern; and, because the beam had to sit on the centre of the rounded end, each rod wore a field cap with a small aperture, its top coated with non-reflecting material. The rods were lengthened for the larger mirror, the longest so that it “could only be inserted in one piece”, which is why, once the corrector was assembled, the field-lens spacing could never be measured again.

The last photograph through it, in February 1982, after coating, is Figure D-1: straight, equally spaced fringes with minor deviations, three circular obscurations where catch plates covered holes in the mirror, an irregular patch from a chip in the corrector’s lower mirror, and semicircular fiducial marks at the rim. Perkin-Elmer’s analysis gave a residual surface error under 0.014 of a wave, “a surface smoothness of 0.009 [micrometres] rms”, and “this analysis convinced the Project that the goals of the fabrication had been achieved.” One sentence of the appendix goes with it: during these tests the mirror was raised towards the corrector until the fringes straightened, and “the lengthy distance from the RNC to the mirror was not precisely measured.”

The inverse corrector

The third was built to check the second. The inverse null corrector, a positive lens, a negative lens and a concave mirror, was designed “to mimic the reflection from a perfect primary mirror”: swung into place beneath the reflective corrector, it should have returned straight fringes and so proved the template right, at the laboratory bench before the mirror was ever tested and at the top of the tower at any time afterwards. Rodney’s report records that its designer had meant it “as a partial in-situ calibrator of the overall Reflective Null Corrector test system, but never was used for this purpose. It only was used as an alignment monitoring device.” It went in at the beginning and end of each test sequence, and the technicians looked at its fringes “as a qualitative check upon the condition and alignment of the RNC, but the fringe pattern was apparently not fully evaluated at any time during the test period.”

The fringes were not straight. Figure D-3, taken in February 1981, shows “several waves of spherical aberration in the wavefront”, of an amount “comparable to, and of the opposite sign as, the amount observed in the telescope”, which is what a template too flat at the edge would show when tested against a device imitating a correct mirror. The explanation at the time was that the inverse corrector itself was wrong: “It was known that the components of the INC were very sensitive to alignment, and also that there was a probable spacing error in the INC.” Unverifiable testimony spoke of a waiver for a spacing error in it. In 1990 an error was indeed found in its design calculations, producing “a small amount of spherical aberration”, and when the as-built unit was analysed its accuracy came out at 0.14 of a wave. “This indicates that the INC is a factor of three more accurate than the error observed in the INC/RNC interferograms.” It was, the board wrote, “quite accurate enough to detect the gross error, and indeed did so.”

Two more things the inverse corrector proved in 1990. Put under the reflective corrector again on 22 July, it gave fringes that, once a lateral shift in its position was allowed for, correlated excellently with those of 1982 and 1981: “This indicates not only the fundamental error in the RNC, but also the excellent stability of the entire CORI INC assembly.” Whatever had been wrong in 1981 was still exactly as wrong, which meant the instrument at the top of the tower was a true record of the mirror in orbit.

What the board measured

The board’s first rule was that nothing be done to the correctors that might disturb them, since they “were the only direct links by which to determine the actual shape of the primary mirror in orbit.” The reflective corrector therefore stayed in the tower and was examined through the access ports in its sides. A fault-tree analysis of the corrector and the manufacturing data left three possibilities, all longitudinal, because “spherical aberration is a symmetric error and can only be produced by a longitudinal spacing error”: the field lens inserted backwards; the wrong glass in the lens; the elements wrongly spaced, which “seemed highly unlikely because of the method used to set the lens spacings.” The lens was measured in place and was the right way round. Its magnification was measured, which fixes its focal length and so its index of refraction, and the glass was right; two spare lenses from the same lot were found and measured for figure and focal length, and agreed.

That left the spacing, and the rod that had set it no longer fitted. So the engineers shone collimated light up through the field lens from a Zygo interferometer, placed a flat mirror at the lens’s focus, about 0.55 metres above it, finding the position by the interferometer’s best focus, and then measured from the flat down to the vertex of the lower mirror, using a fixture in the mirror’s hole as a reference. The field lens was about 1.3 millimetres too far from the lower mirror. A lightweight spacing rod and a new vertex plug were made and the distance measured mechanically, confirming the error to a tenth of a millimetre. The other spacings were measured and were correct.

Then the arithmetic was run both ways. The field lens position error was applied to the reflective corrector’s data, and the mirror that came out “would account for most of the error observed in the HST images.” The refractive corrector’s old interferograms were reprocessed with its as-built dimensions, and gave “a mirror shape very close in value to the RNC/FLPE data.” Optical Research Associates, consultants engaged by Rodney’s office, reached the same conclusions independently and added two: that “cross checking the reflective and refractive null correctors through tests with inverse null and primary mirror would have shown discrepancies too large to ignore”, and that “inserting windows of different thicknesses into the reflective null corrector may aid in verifying the magnitude of the conic constant error without requiring disassembly.” The result, in Appendix E, is a pair of tables. The design primary had a radius of curvature of 11,040.0 millimetres and a conic constant of minus 1.0022985; the mirror as built had a radius of 11,041.70 and a conic constant of minus 1.013236, “the actual conic constant now on the mirror, due to the null corrector spacing error.” With that one number put back into the design, “the result is the third-order spherical aberration observed in the HST wavefront on orbit”: at best focus, 0.4 of a wave rms at 632.8 nanometres, or 0.253 micrometres, and 70 percent of a star’s light in a circle 0.7 arcseconds in radius instead of 0.1. Had the mirror carried its design conic constant, with the same residual polishing errors, the images “would still have resulted in over 70 percent of the light from a single star falling within a circle of 0.1-arcsecond radius, precisely as the specification called for.” The board marked all of this preliminary and said why: the measurements were made to confirm the source of the error and not to fix its size; the apertures, obstructions and fiducials needed to analyse the old interferograms definitively were uncertain for want of documentation; and the final prescription would have to weigh every source of data, including the images from orbit.

The mirror that was right

NASA had directed Perkin-Elmer to subcontract a second primary mirror to Eastman Kodak, “because NASA considered the quality of the primary mirror to be a major challenge”, and Kodak made it by other means: a more traditional method of figuring, and both a refractive and a reflective corrector “of a completely different design from the Perkin-Elmer version.” Kodak’s mirror matched both its templates to better than 0.014 of a wave, “and the Board has every reason to believe it is the correct hyperboloidal shape.” It was finished in 1980, before Perkin-Elmer’s; it was never coated; and it sat in its shipping crate at Danbury through the rest of the programme and the launch. The Independent Optical Review Panel, briefed on 5 July 1990, recommended in its sixth finding that the backup primary “should be tested using the HDOS and EK null lens optics”, and Rodney’s report suggested what might have been done years before: “round-robin null testing with a common Inverse Null Corrector between Perkin-Elmer and Eastman Kodak.” The board’s judgement was that NASA “did not utilize the opportunity offered by this directed subcontract to validate, and gain confidence in, the P-E approach to the primary mirror manufacture”, because “a highly competitive environment existed between Perkin-Elmer and the Eastman Kodak subcontractor”, and “there was limited additional technical exchange of experience.” Since 2001 Kodak’s mirror has been in the National Air and Space Museum in Washington.

The tests that were not made

A true end-to-end test of the assembled telescope, to the accuracy the mirrors were made to, would have needed a flat mirror as large as the primary and, to test the flat, a spherical mirror about fifteen percent larger: “the test could have required two additional mirrors as large as or larger than the OTA primary”, at a cost “on the order of what the OTA itself cost.” That is the test everyone argued about in July 1990. The board’s point was that it was not the only one. “In hindsight, a much less severe test could have been done to check for a gross error such as did occur. The belief at the time was that if the two mirrors had each exceeded their individual specifications, only a test at the level of accuracy of the individual mirrors would have been meaningful.”

The lesser tests were these. When the mirror moved from Wilton to Danbury, the refractive corrector’s interferograms were compared with the reflective corrector’s first ones, “and the discrepancy could have been noted”, but “the data and the circumstances of transfer are unclear, and the requirements for transfer appeared to be adequately met; therefore no concern was noted.” In May 1981 an audit team at Perkin-Elmer asked in writing for a Hartmann test or a laser-pattern test of the figure, “to uncover some gross error such as an incorrect null corrector”, and for a simple autocollimation test with a twelve-inch flat once the mirrors were assembled, “to look for gross errors which may have occurred in the very long chain of analyses and tests leading to a complete optical system.” No record shows how that was resolved. After assembly, a focus test with a 0.36-metre collimator was made several times over a small part of the aperture, before and after the telescope went to Lockheed; “careful analysis of the data might have revealed the problem”, but the telescope lay horizontal, gravity bent it, and “only the focus position was verified.” And Kodak’s mirror lay in its crate, a template of a different lineage against which either corrector could have been tried.

The board’s sentence on all of them is short. “A range of feasible tests to verify the shape of the primary mirror were considered, but not carried out. Finally, no end-to-end tests were planned or implemented to verify the performance of the OTA.” And its lesson: the company “discounted evidence of spherical aberration from INC and RvNC measurements on the basis of ‘uncertified’ status.” The less precise instruments could not have certified the mirror. They could have shown, and did show, that something was grossly wrong, which is a different question, and the one that mattered.

Sources

• Lew Allen and others, The Hubble Space Telescope Optical Systems Failure Report, NASA-TM-103443, November 1990: Chapter IV on the three correctors, the vertex radius test and the final tests; Chapter VI on the fault tree, the 1990 measurements and the reprocessed data; Chapter IX on the tests considered and not made; Chapter X on the accuracy of critical measurements; Appendix B for the Independent Optical Review Panel’s findings and the “wire test”; Appendix C’s tutorial on spherical aberration; Appendix D on the four interferograms; Appendix E for the design and as-built tables and their caveats. ntrs.nasa.gov
• George A. Rodney, Hubble Space Telescope: SRM&QA Observations and Lessons Learned, October 1990: the 1974 design memorandum, the 60-inch corrector’s build records, the redlined drawings, the refractive corrector’s witnessed assembly, the inverse corrector’s intended use, the audit team’s recommendation of May 1981, Optical Research Associates’ seven conclusions, and the round-robin suggestion. ntrs.nasa.gov
• National Air and Space Museum, “Mirror, Primary Backup, Hubble Space Telescope,” object A20010288000; and Spaceflight Now, “Hubble’s backup mirror goes on display,” 22 September 2001: Kodak’s mirror, its completion in 1980, its crate and its transfer. airandspace.si.edu, spaceflightnow.com