MF accuracy vs. AF accuracy

[KEVIN BURKE <KBURKE@iterated.com>]

While getting acquainted with the challenges of using a 75/1.4
Summilux at maximum aperture and close distances I decided
to find out if M series rangefinders could really be expected to
focus such a lens with reasonable accuracy.  At the same time
I looked into some of the similar limitations with MF SLR systems.
I was developing an opinion that an SLR might might be a better
choice for me with something like the 80/1.4 Summilux but wanted
to know more before punting the M system.  Here's a little of what
I found out related to the MF/AF debate:

If it can be assumed the mechanical and optical tolerances in a
camera's focusing system are negligible (big hand-wave here),
one can determine the lower limits a particular method of focusing
allows.  I was interested in finding these lower limits, testing to
make sure I had really gotten reasonable answers, and then see
how capable a "real" camera was in acheiving them.  If the theoretical
limits weren't satisfactory for the conditions I was trying to shoot
under, there wouldn't be any hope for that method as I expected
a production system to be a bit worse than the theoretically perfect
one.

With the theoretically perfect systems (both RF & SLR), the
photographer's eye(s) set the lower bound on accuracy but in
different ways.  Depending on who's text you look in, the average
eye can discriminate an "object" to about 1 minute of arc.
Smaller than this, the eye has difficulty seeing the "object" as
actually being distinctly smaller.  When evaluating the image directly
in the viewfinder of an SLR (with a perfectly smooth screen)
the magnification resulting from longer lenses effectively lowers
this limit.  Likewise, shorter focal lengths effectively increase the limit.
As has already been stated in another post, DOF plays a
significant part as well.  It helps to create a narrower band to
either side of perfect focus based on the 1 min arc.

For "vernier seeing" the eye has even lower limits.  I was pleasantly
surprised to find that Leica RF propaganda about this was indeed
correct and the limit was significantly below that for viewing of a
perfect ground glass.  There are a number of different types of
vernier seeing but the two of interest for many focusing systems
are "split image" and "coincident".  Split image systems allow
the average eye to use it's lower limit of 5-10 seconds of arc.
Warren Smith mentions in his optics book that some exceptional
persons get down to 1-2 sec arc.  That's really down there!  The
split image method involves the butt-end alignment of two lines
and can be used by either RF folks or SLR folks who use screens
with the dual-wedges in the middle.  The accuracy of this kind of
SLR focusing is also a function of the focal length of the lens in
use.  The dual-wedge portion of most SLR screens is cut to use
f/4.5-f/5.6 lens aperture as the effective rangefinder base.  (I
understand that some Leicaflex screens were cut for f/2.8 - a
40-100% improvement in accuracy.)  The additional magnification
of long lenses (again lowering the effective limits of eye) helps
even further.  So, the SLR really comes on strong as the focal
length is increased.

The limits for coincident seeing (alignment of one line superimposed
on top of another line) are somewhere between 1 min arc and
5-10 sec arc depending on the individual.  Most of the data I
have says its about twice that of split image seeing.  So for
rangefinder use, the Leica literature is correct: the split image
method is significantly more accurate than the coincident method.

As for AF systems, I did not look into them very much but have
seen several drawings of popular systems that look like they
use a coincident style of alignment on their sensors.  From this
it seems plausible that a sensor could be made with detector sizes
such that it met or exceeded the limits of the human eye in it's ability
to tell when the same images were lined up.  Manufacturing such a
thing could be another story.  The overall precision of the system
might have to be very high to realize the sensor's performance
potential.

As a data point, I have shot regularly with an old 8008 & 85/1.8AF
lens, wide open, near it's near focus limit.  The AF can find near-
perfect focus if allowed to.  When the AF is first activated, it gets
close to optimum focus and locks.  Then, if I released and activate
it again, it makes a minute adjustment from the previously locked
postion which snaps the image into practically perfect focus.  It
seems that the electronic decision making and control of the AF
could be as big a culprit in actual accuracy as the fundamental
imaging system.

Apologies if this is too long or off topic...

 - Kevin
kburke@iterated.com