MTF againgreenspun.com : LUSENET : Large format photography : One Thread
The statement that prompted the response from Rodenstock:
" MTF curves are theoretic only (at least as published by the manufacturers) and represent the 'expected' output of their designs. I don't know of a correlation between these simulations and reality."
The response from Rodenstock:
"The MTF curves are not theoretic but they are calcuted. The reality is when you measure MTF curves on the MTF machine with a lens it could have a tolerance of -10 % at the most. In other words there is a difference when you measure the MTF together with the lens. "
Editor: This is the followup to http://hv.greenspun.com/bboard/q-and-a-fetch-msg.tcl?msg_id=004ZDQ
-- Bob Salomon (email@example.com), February 08, 2001
Once again you've done nothing except shoot yourself in the foot. "Calculated" refers to mathematical equations - it does not infer actual measurements of production lenses being taken. Read the statement as "The MTF curves are not theoretic, BUT they ARE calculated. And once again they speak of the "reality" of actually "measuring" a lens and the "-10%" tolerance. The published graphs are not from measurements derived from production lenses. At least you didn't repeat the snotty, condescending tone of your first post. But the Germans reading comprehension skills, even in a foreign tongue, still surpass yours.
-- Wayne DeWitt (firstname.lastname@example.org), February 08, 2001.
Thanks for supplying the rest of that first sentence. It definitely clears things up and puts this to bed as far as I'm concerned.
From the Miriam-Webster online dictionary:
Main Entry: cal·cu·late 1 a : to determine by mathematical processes b : to reckon by exercise of practical judgment : ESTIMATE
So, either the MTF curves are arrived at by mathematical processes or they are estimated approximations. And when actual lenses are measured, they are within -10% of these calculated curves.
Based on my background in the semiconductor industry, this makes perfect sense. We did all of our design work using complex computer models and simulations representative of real world operating conditions. When the actual silicon arrived, we took extensive measurements in the lab to correlate the simulation results with the actual measurements.
Theoretic values are considered to represent ideal conditions not attainable in a real world environment. Although our simulated results were still arrived at via mathematical calculations, they also took into account the real world variables that would cause deviation from the theoretical ideal. Our simulation results, by including real world losses, offered much better correlation with the actual measured results (real world performance).
This is EXACTLY what is said in the first sentence in the complete quote provided above. To expand slightly: The MTF curves are not theoretic (not ideal, pie-in-the-sky, unattainable in the real world) but they are calculated (arrived at through complex computer models that accurately simulate real world conditions). Even if you ignore my parenthetical comments, and just read the complete original sentence, there can be no other way to interpret what is being said.
Thanks for clearing this up once and for all.
-- Kerry Thalmann (email@example.com), February 08, 2001.
I think that we're splitting hairs here, however, I contend that the original statement that the curves are theoretic is absolutely correct.
The American Heratige Dictionary (dictionary.com) lists the definition of theoretic as:
1.Of, relating to, or based on theory. 2.Restricted to theory; not practical: theoretical physics. 3.Given to theorizing; speculative.
Note that the first definition is 'based on theory'. They list the applicable definition of theory as:
1. a.Systematically organized knowledge applicable in a relatively wide variety of circumstances, especially a system of assumptions, accepted principles, and rules of procedure devised to analyze, predict, or otherwise explain the nature or behavior of a specified set of phenomena. b.Such knowledge or such a system.
Now, note that the original unattributed quote from Bob says "MTF curves are theoretic only... and represent the 'expected' output of their designs..." That's exactly the definition that the original quote is referring to. I don't believe that the word 'theoretic'is referring to the 'restricted to theory' definition that you refer to at all.
The last part of the original unattributed quote from Bob says, "I don't know of a correlation between these simulations and reality." Well, that's what Bob attempted to answer, and the information about -10% can tell you what is reasonable to expect in a real lens sample.
However, this raises a question in my mind about the legitimacy of data, and the ability of a person or organization to manipulate the numbers to achieve a desired outcome. I contend that these curves are in fact the very best possible performance for the lens design, even if they do include discounts for imperfect application of an ideal optical design.
And, I contend that the only way that these curves would be of real applicable value is if you reduce the performance by approximately 5%, so that the curve falls in the middle of the expected performance range, and not at the very top. What I'm getting at here is that I don't believe that these curves are a fair representation of the performance of a typical lens, because the curve ultimately represents the _upper limit_ of the potential performance, not the _reasonable expectation_ of a typical lens.
If this were a reasonable expectation of a typical lens, the error would be + or - 5% or so, not -10%. I believe a sampling of a few lenses and a statistical analysis of the performance curves would reveal a much more useful set of curves, and one that could then have an error based on the statistical sample.
As an engineer by training, I understand what having only a negative error means, and that tells me that the curves have a certain amount of 'spin' to them, to pad the performance. This is marketing, and I expect that every optical manufacturer (and just about every other company that sells performance products around the world) does to improve the overall appearence of the performance specs.
Until I see a set of measured performance curves from a reliable, independant source, the MTF curves will always be a little bit suspect. They're not unuseable, but you have to take into consideration the source of the data to make an intellegent analysis of the data.
-- Michael (firstname.lastname@example.org), February 08, 2001.
Sigh, time to add my two cents. I would rather be presented with "theoretical" MTF numbers identified as such rather than "real world" numbers. Since Kerry and Michael are both knowledgeable in this they will understand where I'm coming from. Knowing that the numbers are "theroretical" ideals, and that they do not take into account the variability in materials and manufacturing, does give you a starting point at least. But, if you are presented with "measured" MTF numbers, I believe that you are more likely to read information into them that is not there.
Suppose your favorite lensmaker has their technicians test a "representative" (whatever that is) sample of 20 lenses (still not statistically valid except for very small production runs). Suppose that the resulting averaged graph is presented to the head of marketing or engineering who says "the numbers aren't good enough". What do you think they will do? They will either pretest and hand pick their next "sample" or they'll make sure that Hans and Fritz (their most skilled assemblers) assemble the next batch with elements tested and handpicked by Bertha (their most capable line inspector), using mechanical parts machined by Helmut (their best machinest), and then "cherrypick" the resulting lenses. Just stating that the tests are "representative" of their product doesn't give you any assurance that your sample will fall within the projected distribution curve.
We all have someone in our family who doesn't fit. Whether it's the 5'6" runt brother in a family whose mother is 6', or the curly-haired blonde daughter whose siblings all have straight brown hair. All manufacturers have "off-days", "bad-runs", and "outliers", it doesn't matter what the graphs look like when you get a dog. I just believe that if you know going in that the numbers are theoretical (or "calculated") you will not be subject to the "sleigh-of-hand" that can be used to massage "real numbers". The variables are still present but they are not falsely "accounted for".
-- Wayne DeWitt (email@example.com), February 08, 2001.
Wayne, who you callin' a runt? Come over here and say that to me and my curly-haired sister!
-- Steve Singleton (firstname.lastname@example.org), February 08, 2001.
Send your sister over and I'll apologize to her, and send her back with an apology note for you (eventually).
-- Wayne DeWitt (email@example.com), February 08, 2001.
As a native German speaker I wish to add my two cents: In German, the word "theoretic" does not mean "based on a theory" but it has a very strong connotation of "not achievable in the real world". I therefore believe that Kerry's (and not Michael's) interpretation is the correct one. (Note, Rodenstock states that the curves are NOT theoretic, implying that they are achievable in real lenses).
Still it would be nice if Rodenstock could clarify what their calculations are based on!
Also, what does -10% tolerance mean??? How can they be sure that no lens is below 10%, unless they measure each and every one and discard them in case they are below!?
Another burning question: Lets say the curve shows 40% MTF at a certain frequency. Does -10% tolerance imply that no lens is below 36% or does it imply no lens is below 30%???
-- Andreas Carl (firstname.lastname@example.org), February 08, 2001.
I would propose that the situation is someplace in the middle. I would be very surprised if lens design is accomplished by the trial and error making of lenses! The alternative is to design lenses on the basis of a model. All models are based, at least in part, on a theory; they can't be based entirely on emperical observation. (It's theorectically not possible!) However, I am sure that, before commiting a design to glass, lens designers apply models that, to the extent possible, represent real life. Would we not expect these models to calculate MTF curves as an important standard against which lenses are compared? I would hope so. So, there we go, MTF curves are "calculated" from models that are based, at least in part, on a "theory". See, right smack dab in the middle. How about that.
With all of this said, why would MTF curves NOT be based on data collected from the lenses themselves? Wouldn't this be the most representative form of MTF curves? It's not as if the test were destructive! Sounds mighty suspicious to me.
-- neil poulsen (email@example.com), February 09, 2001.
I cannot resist jumping in with my 25 cents. There is system resolution and gate resolution. There's not only the Modulation transfer Function and lenses, but also the issue of the cumlative error of the whole system. The lens/camera set-up, the film plane, the film, the films MTF, the enlarger lens & set-up and so on. The greatest lens ever made cannot strut its stuff if its performance is compromised by other variables that are off the mark.
-- Jonathan Brewer (firstname.lastname@example.org), February 09, 2001.
I too, would like a clarification of what the "-10%" figure refers to. Which axis of the MTF curve is involved?
If it's a drop of 10% contrast at a given spatial frequency, then that's significant. If it's a 10% variation of the contrast then it's not very significant.
OTOH, if the -10% refers to the spatial frequency at which a certain low value of contrast is reached, again, that's an important variation. A difference between 'resolving' 80 cycles per millimetre and 72 cycles per millimetre can effectively mean the difference between a high-end lens, and an average one.
If Rodenstock really want to lay this one to rest, they should test a statistically meaningful sample of lenses (>30), and issue the results in the unequivocal form of Mean, Standard deviation, together with the best and worst case figures.
If the MTF figures really are measured results, then the sample size and statistical data should be readily available.
-- Pete Andrews (email@example.com), February 09, 2001.
You're not looking at the correct quote in Bob's original message. My statement was only in reference to the first quote and the use of the word 'theoretic'. If you look at thta statement, you will see that the definition is exactly as I state, the performance based on 'expected' behavior of the lenses.
The German use of the word is not at debate here, since you are clearly correct that they meant 'theoretic' in the 'unachievable' sense.
I just wanted to clear that up for you, because I have never been wrong about anything important, and if I were, it clearly wasn't as important as I thought. ;-)
Ultimately, it would be great to get a statement from Rodenstock, (and Schneider), that clearly annunciates how they derive their curves, and what the curves represent as a performance expectation, so that the task of comparing them will be much easier. This, I suspect is exactly the reason they do it they way they do; so it's harder to compare the two brands directly. Simply the marketing departments working their magic.
-- Michael Mutmansky (firstname.lastname@example.org), February 09, 2001.
I am sure we would all like custom MTF curves for the lens we buy... but lets be real. We now know what the curves are, and we have some clue as to the variation (although it would be nice to know whether the 10% is relative or absolute) in production lenses.
Since there is no rash of people reporting on Rodenstock lemons, and in fact most users regard their Rodenstock lenses as superb, it is clear that Rodenstock exercises appropriate quality control measures to ensure that lenses leaving the manufacturer meet their quality goals.
I still run chart and field tests on new lenses. My latest Rodenstock 55mm Apo-Grandagon resolved nearly 80 lp/mm. Thats pretty impressive, and the images back it up.
I have always assumed that real lenses vary downward from published curves by some amount. The curves are mostly useful for defining the possibilities and priorities of the lens design.
Finally, except for Zeiss, Rodenstock has now provided us with more information than any other manufacturer. Schneider publishes calculated curves, but not information on chromatic aberration or production tolerance. Nikon and Fuji provide nothing.
-- Glenn Kroeger (email@example.com), February 09, 2001.
I agree we are splitting hairs here, and arguing semantics more than science.
A couple of points. I don't believe the first unattributed quote in Bob's post came from anyone at Rodenstock. The first quote is a blanket statement and makes no specific mention of Rodentock (or any other manufacturer). It appears to be generic in that respect. Who made that quoted statement, I have no idea, but I believe it is Bob's contention that the statement is false and hence he asked Rodenstock for a clarification on the derivation of THEIR MTF curves.
If you look only at the second statement, that Bob attributes to some unknown being at Rodenstock, it is very clear that the source of that quote considers "theoretic" and "calculated" to be two different things. The first sentence in that statement states:
"The MTF curves are not theoretic but they are calcuted (sic)."
Even if we disagree on the precise meaning of "theoretic" in this context, I think we both agree that the Rodenstock MTF curves are mathematically derived, NOT based on measured data.
That seems to be the exact opposite of what Bob was trying to prove in the previous thread on this topic. However, with the complete first sentence provided here, it is very clear and unambiguos that the Rodenstock MTF curves are based on calculations, NOT measurements (there is no mention of measuring anything in that first sentence).
Now, the question becomes, how accurate are those calculated MTF curves? There seems to be a lot of confusion surrounding the "tolerance of -10 %" claim. Not necessarily the accuracy of the -10% figure, but the meaning of -10% in the context of MTF curves (absolute or relative).
Based on the first sentence in the Rodenstock quote, I do believe lossy models were used by the Rodenstock engineers in generating these curves. And, therefore, the curves represent something less than the theoretical ideal, or perfect, lens. The accuracy of the curves then becomes a function of the accuracy of the models used. We have no idea how accurate those models are, but can safely assume the engineers at Rodenstock have a pretty good handle on it.
Simulation results can be VERY accurate and correlate VERY closely to real world performance. Again, it is a function of accurate models and a comprehensive set of simulation scenarios. In my former career, I did a LOT of simulations, and the models were constantly refined based on measured data. Also, the simulations were generally run using "best case", "typical" and "worst case" scenarios for all variables involved. This results in a set of curves that bound the performance of the system. Measured data is then used to verify that the real world performance does indeed fall within these bounds.
WRT to the Rodenstock MTF curves, what we don't know is if they represent "best case", "typical" or "worst case" conditions (I won't speculate on which). It would be nice to have a complete set of curves, on the same graph, showing best case, typical and worst case performance, but I doubt if we'll ever see it. Even if it was provided, it's probably overkill for our applications (and besides, we have no way of verifying it - at least I know I don't have my own personal MTF test machine). Unlike semiconductors, where the system designers NEED to know upper and lower performance boundaries for each component to guarantee a functional system, our LF cameras will not completely cease to function if your lens is slightly out of spec (either slightly better, or slightly worse).
Sure, we want the best system performance possible, but in most cases (with modern lenses), it will not be the taking lens limiting that performance. Perhaps we need to run some simulations on the complete image producing system that take into account the best case, typical and worst case scenarios for all variables involved (taking variables - lens performance, film performance, focusing errors, film plane location, film flatness, camera movement, subject movement; developing variables - time, temperature, agitation, developer strength; printing variables - enlarger lens performance, enlarger vibration, paper performance, paper flatness, focusing accuracy, etc.)
NAH, I'd rather just go out and take some pictures.
Seriously, we've beat this to death, and without more data, or at least a clarification from Rodenstock, we can't really take it any further without a lot of assumptions and speculation (not that that has stopped us so far). At least, by providing the entire first sentence of the quote from Rodenstock, Bob has has clarified, beyond any doubt, that the Rodenstock MTF curves are based on calculations, and not measure data. So, we did learn something new. Thanks Bob.
-- Kerry Thalmann (firstname.lastname@example.org), February 09, 2001.
If I wanted to photograph these "split hairs" which lens manufacturer's product would produce the most hairs split per mm.
-- dave bulmer (email@example.com), February 09, 2001.
How much wood would a wood-chuck chuck if a wood-chuck could chuck wood?
-- Carl Whitman (firstname.lastname@example.org), February 14, 2001.
Any lens will do, just fit your lens with a Tiffen filter and all hairs will be gone. Cheers,
-- Geoffrey Chen (DB45TEK@AOL.COM), February 14, 2001.
MTF curves for Zeiss lenses for Hasselblad (medium format) are generated by taking ACTUAL measurements of the lens in the appropriate laboratory testing device for determination of MTF and using the appropriate software (mathematical equations) to convert the measurments to MTF data.
They do not test every lens manufactured. They test a statistically sufficient number of lens of a specific lens type (e.g. 180/4 Sonnar) and with the manufacturing tolerances allowed for the final finished lens as a product, they determine the MTF which is published in graphical form as a technical specification for that particular lens.
Zeiss manufacturing tolerances are so tight, that any lens so tested on an appropriate MTF laboratory system should "fall" on the line of this published graph - the error in reading this "rough" published graph should encompass any tested lens.
I know of no other manufacturer who physically tests lenses for MTF in this way. As Bob stated, other manufacturers calculate the MTF curves.
Your discussion of what is meant by "calculate" and "theoretic" is an area I have not investigated nor discussed with other lens manufacturers. It would be very interesting to know the exact procedure (measurements and mathematics) to arrive at the "calculated" MTF.
At the website www.photodo.com, MTF values and curves for 35mm amd medium format lenses are tested on the exact MTF equipment used by Hasselblad and the tests are performed by the Hasselblad technician who is qualified to determine MTF measurements. There is no information on the number of lenses tested for a specific lens type nor any (if any) statistical methodology to determine the "error" in the stated MTF and published MTF curves.
-- Daniel Salvucci (email@example.com), June 22, 2001.
Do uou think that the MTF equipement for optical imaging have received his theoretical limites. Do you think that there are new areas left for future research?
-- Ani (firstname.lastname@example.org), July 29, 2001.