Actually... we call that one the "little lab collimator" The big one is in vacuum and could swallow all of Hubble without even a hiccough! We of course absolutely characterize all test set wavefront mappings and back them out or measurements. Typical test residuals are down around λ/1000, no doubt less... but I don't like to brag. Tom
DISCLAIMER: These claims are for novelty purposes only. Tom
Lambda\1000....?! my friend has a noise floor around 10pm, on a good day when there aren’t too many people jumping around.
Got to be aware of systematic uncertainty terms... don’t want to making another “Hubble”?!
OK, I spotted the disclaimer but will analyze your statement anyway.
Assume it is a mirror designed to work in the optical at λ = 500nm (if it is a decametric radio reflector achieving cm smoothness is fairly easy), so λ/1000 is 0.5nm. A quick rummage around in Wikipedia informs me that the Si-O bond length in silica is 0.16nm. You are measuring your surface to an accuracy of about three atoms.
I knew this would get some scrutiny. The biggest challenge is that it's on large telescopes final certification, confirming that the operational wavefront across all fields shall/will meet requirements. That involves all sorts of test set and test config vs ops config differentials. The other ones are test set noise floor and any other systematic or random uncertainties. The unknown systematics are of greatest concern... what condemned Hubble to catastrophic anomalous failure (which event was subsequently repaired).
Bringing the noise floor precision (relative measurement noise) and accuracy (absolute vs ideal noise) down just involves ~best practices~ ... some heroic. Things like support of the (very large) Primary Mirror (1g vector vs 0g ops), system alignment differentials, vibration suppression/isolation, thermal control, vacuum environment, pupil distortion mapping (oft neglected by some shops) etc. etc. Under the best practices, it is possible to push the unknown systematics down below the residual random test noise. And that is where the (couple of thousandths) single pass wavefront RMS noise floor is achieved. The only way one can ironclad confirm that this has truly been achieved (vs simple brag 'n' boast) is when the system goes operational, and delivered product (imagery and on-board wavefront sensor suite)... confirm that it is indeed exceeding stringent requirements. At that point, what you see is what you got. If the system is one of many, one can (only then) use the history of success to credibly assert that the subsequent delivered product will indeed satisfy.
Optical metrology, in general, is Catch-22 difficult in that we use light to form the operational images ... and the same flavour of light for metrologies.
PS: My claim of λ/1000 test accuracy may be closer to λ/500. For large systems... the cost of testing alone can be a significant fraction of the entire build budget! But in the context of "pay me now or pay me later" --- insurance that's well worth it! Tom